]>
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> |
a79316c6 | 23 | #include <linux/kasan.h> |
5a896d9e | 24 | #include <linux/kmemcheck.h> |
81819f0f CL |
25 | #include <linux/cpu.h> |
26 | #include <linux/cpuset.h> | |
27 | #include <linux/mempolicy.h> | |
28 | #include <linux/ctype.h> | |
3ac7fe5a | 29 | #include <linux/debugobjects.h> |
81819f0f | 30 | #include <linux/kallsyms.h> |
b9049e23 | 31 | #include <linux/memory.h> |
f8bd2258 | 32 | #include <linux/math64.h> |
773ff60e | 33 | #include <linux/fault-inject.h> |
bfa71457 | 34 | #include <linux/stacktrace.h> |
4de900b4 | 35 | #include <linux/prefetch.h> |
2633d7a0 | 36 | #include <linux/memcontrol.h> |
81819f0f | 37 | |
4a92379b RK |
38 | #include <trace/events/kmem.h> |
39 | ||
072bb0aa MG |
40 | #include "internal.h" |
41 | ||
81819f0f CL |
42 | /* |
43 | * Lock order: | |
18004c5d | 44 | * 1. slab_mutex (Global Mutex) |
881db7fb CL |
45 | * 2. node->list_lock |
46 | * 3. slab_lock(page) (Only on some arches and for debugging) | |
81819f0f | 47 | * |
18004c5d | 48 | * slab_mutex |
881db7fb | 49 | * |
18004c5d | 50 | * The role of the slab_mutex is to protect the list of all the slabs |
881db7fb CL |
51 | * and to synchronize major metadata changes to slab cache structures. |
52 | * | |
53 | * The slab_lock is only used for debugging and on arches that do not | |
54 | * have the ability to do a cmpxchg_double. It only protects the second | |
55 | * double word in the page struct. Meaning | |
56 | * A. page->freelist -> List of object free in a page | |
57 | * B. page->counters -> Counters of objects | |
58 | * C. page->frozen -> frozen state | |
59 | * | |
60 | * If a slab is frozen then it is exempt from list management. It is not | |
61 | * on any list. The processor that froze the slab is the one who can | |
62 | * perform list operations on the page. Other processors may put objects | |
63 | * onto the freelist but the processor that froze the slab is the only | |
64 | * one that can retrieve the objects from the page's freelist. | |
81819f0f CL |
65 | * |
66 | * The list_lock protects the partial and full list on each node and | |
67 | * the partial slab counter. If taken then no new slabs may be added or | |
68 | * removed from the lists nor make the number of partial slabs be modified. | |
69 | * (Note that the total number of slabs is an atomic value that may be | |
70 | * modified without taking the list lock). | |
71 | * | |
72 | * The list_lock is a centralized lock and thus we avoid taking it as | |
73 | * much as possible. As long as SLUB does not have to handle partial | |
74 | * slabs, operations can continue without any centralized lock. F.e. | |
75 | * allocating a long series of objects that fill up slabs does not require | |
76 | * the list lock. | |
81819f0f CL |
77 | * Interrupts are disabled during allocation and deallocation in order to |
78 | * make the slab allocator safe to use in the context of an irq. In addition | |
79 | * interrupts are disabled to ensure that the processor does not change | |
80 | * while handling per_cpu slabs, due to kernel preemption. | |
81 | * | |
82 | * SLUB assigns one slab for allocation to each processor. | |
83 | * Allocations only occur from these slabs called cpu slabs. | |
84 | * | |
672bba3a CL |
85 | * Slabs with free elements are kept on a partial list and during regular |
86 | * operations no list for full slabs is used. If an object in a full slab is | |
81819f0f | 87 | * freed then the slab will show up again on the partial lists. |
672bba3a CL |
88 | * We track full slabs for debugging purposes though because otherwise we |
89 | * cannot scan all objects. | |
81819f0f CL |
90 | * |
91 | * Slabs are freed when they become empty. Teardown and setup is | |
92 | * minimal so we rely on the page allocators per cpu caches for | |
93 | * fast frees and allocs. | |
94 | * | |
95 | * Overloading of page flags that are otherwise used for LRU management. | |
96 | * | |
4b6f0750 CL |
97 | * PageActive The slab is frozen and exempt from list processing. |
98 | * This means that the slab is dedicated to a purpose | |
99 | * such as satisfying allocations for a specific | |
100 | * processor. Objects may be freed in the slab while | |
101 | * it is frozen but slab_free will then skip the usual | |
102 | * list operations. It is up to the processor holding | |
103 | * the slab to integrate the slab into the slab lists | |
104 | * when the slab is no longer needed. | |
105 | * | |
106 | * One use of this flag is to mark slabs that are | |
107 | * used for allocations. Then such a slab becomes a cpu | |
108 | * slab. The cpu slab may be equipped with an additional | |
dfb4f096 | 109 | * freelist that allows lockless access to |
894b8788 CL |
110 | * free objects in addition to the regular freelist |
111 | * that requires the slab lock. | |
81819f0f CL |
112 | * |
113 | * PageError Slab requires special handling due to debug | |
114 | * options set. This moves slab handling out of | |
894b8788 | 115 | * the fast path and disables lockless freelists. |
81819f0f CL |
116 | */ |
117 | ||
af537b0a CL |
118 | static inline int kmem_cache_debug(struct kmem_cache *s) |
119 | { | |
5577bd8a | 120 | #ifdef CONFIG_SLUB_DEBUG |
af537b0a | 121 | return unlikely(s->flags & SLAB_DEBUG_FLAGS); |
5577bd8a | 122 | #else |
af537b0a | 123 | return 0; |
5577bd8a | 124 | #endif |
af537b0a | 125 | } |
5577bd8a | 126 | |
345c905d JK |
127 | static inline bool kmem_cache_has_cpu_partial(struct kmem_cache *s) |
128 | { | |
129 | #ifdef CONFIG_SLUB_CPU_PARTIAL | |
130 | return !kmem_cache_debug(s); | |
131 | #else | |
132 | return false; | |
133 | #endif | |
134 | } | |
135 | ||
81819f0f CL |
136 | /* |
137 | * Issues still to be resolved: | |
138 | * | |
81819f0f CL |
139 | * - Support PAGE_ALLOC_DEBUG. Should be easy to do. |
140 | * | |
81819f0f CL |
141 | * - Variable sizing of the per node arrays |
142 | */ | |
143 | ||
144 | /* Enable to test recovery from slab corruption on boot */ | |
145 | #undef SLUB_RESILIENCY_TEST | |
146 | ||
b789ef51 CL |
147 | /* Enable to log cmpxchg failures */ |
148 | #undef SLUB_DEBUG_CMPXCHG | |
149 | ||
2086d26a CL |
150 | /* |
151 | * Mininum number of partial slabs. These will be left on the partial | |
152 | * lists even if they are empty. kmem_cache_shrink may reclaim them. | |
153 | */ | |
76be8950 | 154 | #define MIN_PARTIAL 5 |
e95eed57 | 155 | |
2086d26a CL |
156 | /* |
157 | * Maximum number of desirable partial slabs. | |
158 | * The existence of more partial slabs makes kmem_cache_shrink | |
721ae22a | 159 | * sort the partial list by the number of objects in use. |
2086d26a CL |
160 | */ |
161 | #define MAX_PARTIAL 10 | |
162 | ||
81819f0f CL |
163 | #define DEBUG_DEFAULT_FLAGS (SLAB_DEBUG_FREE | SLAB_RED_ZONE | \ |
164 | SLAB_POISON | SLAB_STORE_USER) | |
672bba3a | 165 | |
fa5ec8a1 | 166 | /* |
3de47213 DR |
167 | * Debugging flags that require metadata to be stored in the slab. These get |
168 | * disabled when slub_debug=O is used and a cache's min order increases with | |
169 | * metadata. | |
fa5ec8a1 | 170 | */ |
3de47213 | 171 | #define DEBUG_METADATA_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER) |
fa5ec8a1 | 172 | |
210b5c06 CG |
173 | #define OO_SHIFT 16 |
174 | #define OO_MASK ((1 << OO_SHIFT) - 1) | |
50d5c41c | 175 | #define MAX_OBJS_PER_PAGE 32767 /* since page.objects is u15 */ |
210b5c06 | 176 | |
81819f0f | 177 | /* Internal SLUB flags */ |
f90ec390 | 178 | #define __OBJECT_POISON 0x80000000UL /* Poison object */ |
b789ef51 | 179 | #define __CMPXCHG_DOUBLE 0x40000000UL /* Use cmpxchg_double */ |
81819f0f | 180 | |
81819f0f CL |
181 | #ifdef CONFIG_SMP |
182 | static struct notifier_block slab_notifier; | |
183 | #endif | |
184 | ||
02cbc874 CL |
185 | /* |
186 | * Tracking user of a slab. | |
187 | */ | |
d6543e39 | 188 | #define TRACK_ADDRS_COUNT 16 |
02cbc874 | 189 | struct track { |
ce71e27c | 190 | unsigned long addr; /* Called from address */ |
d6543e39 BG |
191 | #ifdef CONFIG_STACKTRACE |
192 | unsigned long addrs[TRACK_ADDRS_COUNT]; /* Called from address */ | |
193 | #endif | |
02cbc874 CL |
194 | int cpu; /* Was running on cpu */ |
195 | int pid; /* Pid context */ | |
196 | unsigned long when; /* When did the operation occur */ | |
197 | }; | |
198 | ||
199 | enum track_item { TRACK_ALLOC, TRACK_FREE }; | |
200 | ||
ab4d5ed5 | 201 | #ifdef CONFIG_SYSFS |
81819f0f CL |
202 | static int sysfs_slab_add(struct kmem_cache *); |
203 | static int sysfs_slab_alias(struct kmem_cache *, const char *); | |
107dab5c | 204 | static void memcg_propagate_slab_attrs(struct kmem_cache *s); |
81819f0f | 205 | #else |
0c710013 CL |
206 | static inline int sysfs_slab_add(struct kmem_cache *s) { return 0; } |
207 | static inline int sysfs_slab_alias(struct kmem_cache *s, const char *p) | |
208 | { return 0; } | |
107dab5c | 209 | static inline void memcg_propagate_slab_attrs(struct kmem_cache *s) { } |
81819f0f CL |
210 | #endif |
211 | ||
4fdccdfb | 212 | static inline void stat(const struct kmem_cache *s, enum stat_item si) |
8ff12cfc CL |
213 | { |
214 | #ifdef CONFIG_SLUB_STATS | |
88da03a6 CL |
215 | /* |
216 | * The rmw is racy on a preemptible kernel but this is acceptable, so | |
217 | * avoid this_cpu_add()'s irq-disable overhead. | |
218 | */ | |
219 | raw_cpu_inc(s->cpu_slab->stat[si]); | |
8ff12cfc CL |
220 | #endif |
221 | } | |
222 | ||
81819f0f CL |
223 | /******************************************************************** |
224 | * Core slab cache functions | |
225 | *******************************************************************/ | |
226 | ||
6446faa2 | 227 | /* Verify that a pointer has an address that is valid within a slab page */ |
02cbc874 CL |
228 | static inline int check_valid_pointer(struct kmem_cache *s, |
229 | struct page *page, const void *object) | |
230 | { | |
231 | void *base; | |
232 | ||
a973e9dd | 233 | if (!object) |
02cbc874 CL |
234 | return 1; |
235 | ||
a973e9dd | 236 | base = page_address(page); |
39b26464 | 237 | if (object < base || object >= base + page->objects * s->size || |
02cbc874 CL |
238 | (object - base) % s->size) { |
239 | return 0; | |
240 | } | |
241 | ||
242 | return 1; | |
243 | } | |
244 | ||
7656c72b CL |
245 | static inline void *get_freepointer(struct kmem_cache *s, void *object) |
246 | { | |
247 | return *(void **)(object + s->offset); | |
248 | } | |
249 | ||
0ad9500e ED |
250 | static void prefetch_freepointer(const struct kmem_cache *s, void *object) |
251 | { | |
252 | prefetch(object + s->offset); | |
253 | } | |
254 | ||
1393d9a1 CL |
255 | static inline void *get_freepointer_safe(struct kmem_cache *s, void *object) |
256 | { | |
257 | void *p; | |
258 | ||
259 | #ifdef CONFIG_DEBUG_PAGEALLOC | |
260 | probe_kernel_read(&p, (void **)(object + s->offset), sizeof(p)); | |
261 | #else | |
262 | p = get_freepointer(s, object); | |
263 | #endif | |
264 | return p; | |
265 | } | |
266 | ||
7656c72b CL |
267 | static inline void set_freepointer(struct kmem_cache *s, void *object, void *fp) |
268 | { | |
269 | *(void **)(object + s->offset) = fp; | |
270 | } | |
271 | ||
272 | /* Loop over all objects in a slab */ | |
224a88be CL |
273 | #define for_each_object(__p, __s, __addr, __objects) \ |
274 | for (__p = (__addr); __p < (__addr) + (__objects) * (__s)->size;\ | |
7656c72b CL |
275 | __p += (__s)->size) |
276 | ||
54266640 WY |
277 | #define for_each_object_idx(__p, __idx, __s, __addr, __objects) \ |
278 | for (__p = (__addr), __idx = 1; __idx <= __objects;\ | |
279 | __p += (__s)->size, __idx++) | |
280 | ||
7656c72b CL |
281 | /* Determine object index from a given position */ |
282 | static inline int slab_index(void *p, struct kmem_cache *s, void *addr) | |
283 | { | |
284 | return (p - addr) / s->size; | |
285 | } | |
286 | ||
d71f606f MK |
287 | static inline size_t slab_ksize(const struct kmem_cache *s) |
288 | { | |
289 | #ifdef CONFIG_SLUB_DEBUG | |
290 | /* | |
291 | * Debugging requires use of the padding between object | |
292 | * and whatever may come after it. | |
293 | */ | |
294 | if (s->flags & (SLAB_RED_ZONE | SLAB_POISON)) | |
3b0efdfa | 295 | return s->object_size; |
d71f606f MK |
296 | |
297 | #endif | |
298 | /* | |
299 | * If we have the need to store the freelist pointer | |
300 | * back there or track user information then we can | |
301 | * only use the space before that information. | |
302 | */ | |
303 | if (s->flags & (SLAB_DESTROY_BY_RCU | SLAB_STORE_USER)) | |
304 | return s->inuse; | |
305 | /* | |
306 | * Else we can use all the padding etc for the allocation | |
307 | */ | |
308 | return s->size; | |
309 | } | |
310 | ||
ab9a0f19 LJ |
311 | static inline int order_objects(int order, unsigned long size, int reserved) |
312 | { | |
313 | return ((PAGE_SIZE << order) - reserved) / size; | |
314 | } | |
315 | ||
834f3d11 | 316 | static inline struct kmem_cache_order_objects oo_make(int order, |
ab9a0f19 | 317 | unsigned long size, int reserved) |
834f3d11 CL |
318 | { |
319 | struct kmem_cache_order_objects x = { | |
ab9a0f19 | 320 | (order << OO_SHIFT) + order_objects(order, size, reserved) |
834f3d11 CL |
321 | }; |
322 | ||
323 | return x; | |
324 | } | |
325 | ||
326 | static inline int oo_order(struct kmem_cache_order_objects x) | |
327 | { | |
210b5c06 | 328 | return x.x >> OO_SHIFT; |
834f3d11 CL |
329 | } |
330 | ||
331 | static inline int oo_objects(struct kmem_cache_order_objects x) | |
332 | { | |
210b5c06 | 333 | return x.x & OO_MASK; |
834f3d11 CL |
334 | } |
335 | ||
881db7fb CL |
336 | /* |
337 | * Per slab locking using the pagelock | |
338 | */ | |
339 | static __always_inline void slab_lock(struct page *page) | |
340 | { | |
341 | bit_spin_lock(PG_locked, &page->flags); | |
342 | } | |
343 | ||
344 | static __always_inline void slab_unlock(struct page *page) | |
345 | { | |
346 | __bit_spin_unlock(PG_locked, &page->flags); | |
347 | } | |
348 | ||
a0320865 DH |
349 | static inline void set_page_slub_counters(struct page *page, unsigned long counters_new) |
350 | { | |
351 | struct page tmp; | |
352 | tmp.counters = counters_new; | |
353 | /* | |
354 | * page->counters can cover frozen/inuse/objects as well | |
355 | * as page->_count. If we assign to ->counters directly | |
356 | * we run the risk of losing updates to page->_count, so | |
357 | * be careful and only assign to the fields we need. | |
358 | */ | |
359 | page->frozen = tmp.frozen; | |
360 | page->inuse = tmp.inuse; | |
361 | page->objects = tmp.objects; | |
362 | } | |
363 | ||
1d07171c CL |
364 | /* Interrupts must be disabled (for the fallback code to work right) */ |
365 | static inline bool __cmpxchg_double_slab(struct kmem_cache *s, struct page *page, | |
366 | void *freelist_old, unsigned long counters_old, | |
367 | void *freelist_new, unsigned long counters_new, | |
368 | const char *n) | |
369 | { | |
370 | VM_BUG_ON(!irqs_disabled()); | |
2565409f HC |
371 | #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \ |
372 | defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE) | |
1d07171c | 373 | if (s->flags & __CMPXCHG_DOUBLE) { |
cdcd6298 | 374 | if (cmpxchg_double(&page->freelist, &page->counters, |
0aa9a13d DC |
375 | freelist_old, counters_old, |
376 | freelist_new, counters_new)) | |
6f6528a1 | 377 | return true; |
1d07171c CL |
378 | } else |
379 | #endif | |
380 | { | |
381 | slab_lock(page); | |
d0e0ac97 CG |
382 | if (page->freelist == freelist_old && |
383 | page->counters == counters_old) { | |
1d07171c | 384 | page->freelist = freelist_new; |
a0320865 | 385 | set_page_slub_counters(page, counters_new); |
1d07171c | 386 | slab_unlock(page); |
6f6528a1 | 387 | return true; |
1d07171c CL |
388 | } |
389 | slab_unlock(page); | |
390 | } | |
391 | ||
392 | cpu_relax(); | |
393 | stat(s, CMPXCHG_DOUBLE_FAIL); | |
394 | ||
395 | #ifdef SLUB_DEBUG_CMPXCHG | |
f9f58285 | 396 | pr_info("%s %s: cmpxchg double redo ", n, s->name); |
1d07171c CL |
397 | #endif |
398 | ||
6f6528a1 | 399 | return false; |
1d07171c CL |
400 | } |
401 | ||
b789ef51 CL |
402 | static inline bool cmpxchg_double_slab(struct kmem_cache *s, struct page *page, |
403 | void *freelist_old, unsigned long counters_old, | |
404 | void *freelist_new, unsigned long counters_new, | |
405 | const char *n) | |
406 | { | |
2565409f HC |
407 | #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \ |
408 | defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE) | |
b789ef51 | 409 | if (s->flags & __CMPXCHG_DOUBLE) { |
cdcd6298 | 410 | if (cmpxchg_double(&page->freelist, &page->counters, |
0aa9a13d DC |
411 | freelist_old, counters_old, |
412 | freelist_new, counters_new)) | |
6f6528a1 | 413 | return true; |
b789ef51 CL |
414 | } else |
415 | #endif | |
416 | { | |
1d07171c CL |
417 | unsigned long flags; |
418 | ||
419 | local_irq_save(flags); | |
881db7fb | 420 | slab_lock(page); |
d0e0ac97 CG |
421 | if (page->freelist == freelist_old && |
422 | page->counters == counters_old) { | |
b789ef51 | 423 | page->freelist = freelist_new; |
a0320865 | 424 | set_page_slub_counters(page, counters_new); |
881db7fb | 425 | slab_unlock(page); |
1d07171c | 426 | local_irq_restore(flags); |
6f6528a1 | 427 | return true; |
b789ef51 | 428 | } |
881db7fb | 429 | slab_unlock(page); |
1d07171c | 430 | local_irq_restore(flags); |
b789ef51 CL |
431 | } |
432 | ||
433 | cpu_relax(); | |
434 | stat(s, CMPXCHG_DOUBLE_FAIL); | |
435 | ||
436 | #ifdef SLUB_DEBUG_CMPXCHG | |
f9f58285 | 437 | pr_info("%s %s: cmpxchg double redo ", n, s->name); |
b789ef51 CL |
438 | #endif |
439 | ||
6f6528a1 | 440 | return false; |
b789ef51 CL |
441 | } |
442 | ||
41ecc55b | 443 | #ifdef CONFIG_SLUB_DEBUG |
5f80b13a CL |
444 | /* |
445 | * Determine a map of object in use on a page. | |
446 | * | |
881db7fb | 447 | * Node listlock must be held to guarantee that the page does |
5f80b13a CL |
448 | * not vanish from under us. |
449 | */ | |
450 | static void get_map(struct kmem_cache *s, struct page *page, unsigned long *map) | |
451 | { | |
452 | void *p; | |
453 | void *addr = page_address(page); | |
454 | ||
455 | for (p = page->freelist; p; p = get_freepointer(s, p)) | |
456 | set_bit(slab_index(p, s, addr), map); | |
457 | } | |
458 | ||
41ecc55b CL |
459 | /* |
460 | * Debug settings: | |
461 | */ | |
89d3c87e | 462 | #if defined(CONFIG_SLUB_DEBUG_ON) |
f0630fff | 463 | static int slub_debug = DEBUG_DEFAULT_FLAGS; |
89d3c87e AR |
464 | #elif defined(CONFIG_KASAN) |
465 | static int slub_debug = SLAB_STORE_USER; | |
f0630fff | 466 | #else |
41ecc55b | 467 | static int slub_debug; |
f0630fff | 468 | #endif |
41ecc55b CL |
469 | |
470 | static char *slub_debug_slabs; | |
fa5ec8a1 | 471 | static int disable_higher_order_debug; |
41ecc55b | 472 | |
a79316c6 AR |
473 | /* |
474 | * slub is about to manipulate internal object metadata. This memory lies | |
475 | * outside the range of the allocated object, so accessing it would normally | |
476 | * be reported by kasan as a bounds error. metadata_access_enable() is used | |
477 | * to tell kasan that these accesses are OK. | |
478 | */ | |
479 | static inline void metadata_access_enable(void) | |
480 | { | |
481 | kasan_disable_current(); | |
482 | } | |
483 | ||
484 | static inline void metadata_access_disable(void) | |
485 | { | |
486 | kasan_enable_current(); | |
487 | } | |
488 | ||
81819f0f CL |
489 | /* |
490 | * Object debugging | |
491 | */ | |
492 | static void print_section(char *text, u8 *addr, unsigned int length) | |
493 | { | |
a79316c6 | 494 | metadata_access_enable(); |
ffc79d28 SAS |
495 | print_hex_dump(KERN_ERR, text, DUMP_PREFIX_ADDRESS, 16, 1, addr, |
496 | length, 1); | |
a79316c6 | 497 | metadata_access_disable(); |
81819f0f CL |
498 | } |
499 | ||
81819f0f CL |
500 | static struct track *get_track(struct kmem_cache *s, void *object, |
501 | enum track_item alloc) | |
502 | { | |
503 | struct track *p; | |
504 | ||
505 | if (s->offset) | |
506 | p = object + s->offset + sizeof(void *); | |
507 | else | |
508 | p = object + s->inuse; | |
509 | ||
510 | return p + alloc; | |
511 | } | |
512 | ||
513 | static void set_track(struct kmem_cache *s, void *object, | |
ce71e27c | 514 | enum track_item alloc, unsigned long addr) |
81819f0f | 515 | { |
1a00df4a | 516 | struct track *p = get_track(s, object, alloc); |
81819f0f | 517 | |
81819f0f | 518 | if (addr) { |
d6543e39 BG |
519 | #ifdef CONFIG_STACKTRACE |
520 | struct stack_trace trace; | |
521 | int i; | |
522 | ||
523 | trace.nr_entries = 0; | |
524 | trace.max_entries = TRACK_ADDRS_COUNT; | |
525 | trace.entries = p->addrs; | |
526 | trace.skip = 3; | |
a79316c6 | 527 | metadata_access_enable(); |
d6543e39 | 528 | save_stack_trace(&trace); |
a79316c6 | 529 | metadata_access_disable(); |
d6543e39 BG |
530 | |
531 | /* See rant in lockdep.c */ | |
532 | if (trace.nr_entries != 0 && | |
533 | trace.entries[trace.nr_entries - 1] == ULONG_MAX) | |
534 | trace.nr_entries--; | |
535 | ||
536 | for (i = trace.nr_entries; i < TRACK_ADDRS_COUNT; i++) | |
537 | p->addrs[i] = 0; | |
538 | #endif | |
81819f0f CL |
539 | p->addr = addr; |
540 | p->cpu = smp_processor_id(); | |
88e4ccf2 | 541 | p->pid = current->pid; |
81819f0f CL |
542 | p->when = jiffies; |
543 | } else | |
544 | memset(p, 0, sizeof(struct track)); | |
545 | } | |
546 | ||
81819f0f CL |
547 | static void init_tracking(struct kmem_cache *s, void *object) |
548 | { | |
24922684 CL |
549 | if (!(s->flags & SLAB_STORE_USER)) |
550 | return; | |
551 | ||
ce71e27c EGM |
552 | set_track(s, object, TRACK_FREE, 0UL); |
553 | set_track(s, object, TRACK_ALLOC, 0UL); | |
81819f0f CL |
554 | } |
555 | ||
556 | static void print_track(const char *s, struct track *t) | |
557 | { | |
558 | if (!t->addr) | |
559 | return; | |
560 | ||
f9f58285 FF |
561 | pr_err("INFO: %s in %pS age=%lu cpu=%u pid=%d\n", |
562 | s, (void *)t->addr, jiffies - t->when, t->cpu, t->pid); | |
d6543e39 BG |
563 | #ifdef CONFIG_STACKTRACE |
564 | { | |
565 | int i; | |
566 | for (i = 0; i < TRACK_ADDRS_COUNT; i++) | |
567 | if (t->addrs[i]) | |
f9f58285 | 568 | pr_err("\t%pS\n", (void *)t->addrs[i]); |
d6543e39 BG |
569 | else |
570 | break; | |
571 | } | |
572 | #endif | |
24922684 CL |
573 | } |
574 | ||
575 | static void print_tracking(struct kmem_cache *s, void *object) | |
576 | { | |
577 | if (!(s->flags & SLAB_STORE_USER)) | |
578 | return; | |
579 | ||
580 | print_track("Allocated", get_track(s, object, TRACK_ALLOC)); | |
581 | print_track("Freed", get_track(s, object, TRACK_FREE)); | |
582 | } | |
583 | ||
584 | static void print_page_info(struct page *page) | |
585 | { | |
f9f58285 | 586 | pr_err("INFO: Slab 0x%p objects=%u used=%u fp=0x%p flags=0x%04lx\n", |
d0e0ac97 | 587 | page, page->objects, page->inuse, page->freelist, page->flags); |
24922684 CL |
588 | |
589 | } | |
590 | ||
591 | static void slab_bug(struct kmem_cache *s, char *fmt, ...) | |
592 | { | |
ecc42fbe | 593 | struct va_format vaf; |
24922684 | 594 | va_list args; |
24922684 CL |
595 | |
596 | va_start(args, fmt); | |
ecc42fbe FF |
597 | vaf.fmt = fmt; |
598 | vaf.va = &args; | |
f9f58285 | 599 | pr_err("=============================================================================\n"); |
ecc42fbe | 600 | pr_err("BUG %s (%s): %pV\n", s->name, print_tainted(), &vaf); |
f9f58285 | 601 | pr_err("-----------------------------------------------------------------------------\n\n"); |
645df230 | 602 | |
373d4d09 | 603 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); |
ecc42fbe | 604 | va_end(args); |
81819f0f CL |
605 | } |
606 | ||
24922684 CL |
607 | static void slab_fix(struct kmem_cache *s, char *fmt, ...) |
608 | { | |
ecc42fbe | 609 | struct va_format vaf; |
24922684 | 610 | va_list args; |
24922684 CL |
611 | |
612 | va_start(args, fmt); | |
ecc42fbe FF |
613 | vaf.fmt = fmt; |
614 | vaf.va = &args; | |
615 | pr_err("FIX %s: %pV\n", s->name, &vaf); | |
24922684 | 616 | va_end(args); |
24922684 CL |
617 | } |
618 | ||
619 | static void print_trailer(struct kmem_cache *s, struct page *page, u8 *p) | |
81819f0f CL |
620 | { |
621 | unsigned int off; /* Offset of last byte */ | |
a973e9dd | 622 | u8 *addr = page_address(page); |
24922684 CL |
623 | |
624 | print_tracking(s, p); | |
625 | ||
626 | print_page_info(page); | |
627 | ||
f9f58285 FF |
628 | pr_err("INFO: Object 0x%p @offset=%tu fp=0x%p\n\n", |
629 | p, p - addr, get_freepointer(s, p)); | |
24922684 CL |
630 | |
631 | if (p > addr + 16) | |
ffc79d28 | 632 | print_section("Bytes b4 ", p - 16, 16); |
81819f0f | 633 | |
3b0efdfa | 634 | print_section("Object ", p, min_t(unsigned long, s->object_size, |
ffc79d28 | 635 | PAGE_SIZE)); |
81819f0f | 636 | if (s->flags & SLAB_RED_ZONE) |
3b0efdfa CL |
637 | print_section("Redzone ", p + s->object_size, |
638 | s->inuse - s->object_size); | |
81819f0f | 639 | |
81819f0f CL |
640 | if (s->offset) |
641 | off = s->offset + sizeof(void *); | |
642 | else | |
643 | off = s->inuse; | |
644 | ||
24922684 | 645 | if (s->flags & SLAB_STORE_USER) |
81819f0f | 646 | off += 2 * sizeof(struct track); |
81819f0f CL |
647 | |
648 | if (off != s->size) | |
649 | /* Beginning of the filler is the free pointer */ | |
ffc79d28 | 650 | print_section("Padding ", p + off, s->size - off); |
24922684 CL |
651 | |
652 | dump_stack(); | |
81819f0f CL |
653 | } |
654 | ||
75c66def | 655 | void object_err(struct kmem_cache *s, struct page *page, |
81819f0f CL |
656 | u8 *object, char *reason) |
657 | { | |
3dc50637 | 658 | slab_bug(s, "%s", reason); |
24922684 | 659 | print_trailer(s, page, object); |
81819f0f CL |
660 | } |
661 | ||
d0e0ac97 CG |
662 | static void slab_err(struct kmem_cache *s, struct page *page, |
663 | const char *fmt, ...) | |
81819f0f CL |
664 | { |
665 | va_list args; | |
666 | char buf[100]; | |
667 | ||
24922684 CL |
668 | va_start(args, fmt); |
669 | vsnprintf(buf, sizeof(buf), fmt, args); | |
81819f0f | 670 | va_end(args); |
3dc50637 | 671 | slab_bug(s, "%s", buf); |
24922684 | 672 | print_page_info(page); |
81819f0f CL |
673 | dump_stack(); |
674 | } | |
675 | ||
f7cb1933 | 676 | static void init_object(struct kmem_cache *s, void *object, u8 val) |
81819f0f CL |
677 | { |
678 | u8 *p = object; | |
679 | ||
680 | if (s->flags & __OBJECT_POISON) { | |
3b0efdfa CL |
681 | memset(p, POISON_FREE, s->object_size - 1); |
682 | p[s->object_size - 1] = POISON_END; | |
81819f0f CL |
683 | } |
684 | ||
685 | if (s->flags & SLAB_RED_ZONE) | |
3b0efdfa | 686 | memset(p + s->object_size, val, s->inuse - s->object_size); |
81819f0f CL |
687 | } |
688 | ||
24922684 CL |
689 | static void restore_bytes(struct kmem_cache *s, char *message, u8 data, |
690 | void *from, void *to) | |
691 | { | |
692 | slab_fix(s, "Restoring 0x%p-0x%p=0x%x\n", from, to - 1, data); | |
693 | memset(from, data, to - from); | |
694 | } | |
695 | ||
696 | static int check_bytes_and_report(struct kmem_cache *s, struct page *page, | |
697 | u8 *object, char *what, | |
06428780 | 698 | u8 *start, unsigned int value, unsigned int bytes) |
24922684 CL |
699 | { |
700 | u8 *fault; | |
701 | u8 *end; | |
702 | ||
a79316c6 | 703 | metadata_access_enable(); |
79824820 | 704 | fault = memchr_inv(start, value, bytes); |
a79316c6 | 705 | metadata_access_disable(); |
24922684 CL |
706 | if (!fault) |
707 | return 1; | |
708 | ||
709 | end = start + bytes; | |
710 | while (end > fault && end[-1] == value) | |
711 | end--; | |
712 | ||
713 | slab_bug(s, "%s overwritten", what); | |
f9f58285 | 714 | pr_err("INFO: 0x%p-0x%p. First byte 0x%x instead of 0x%x\n", |
24922684 CL |
715 | fault, end - 1, fault[0], value); |
716 | print_trailer(s, page, object); | |
717 | ||
718 | restore_bytes(s, what, value, fault, end); | |
719 | return 0; | |
81819f0f CL |
720 | } |
721 | ||
81819f0f CL |
722 | /* |
723 | * Object layout: | |
724 | * | |
725 | * object address | |
726 | * Bytes of the object to be managed. | |
727 | * If the freepointer may overlay the object then the free | |
728 | * pointer is the first word of the object. | |
672bba3a | 729 | * |
81819f0f CL |
730 | * Poisoning uses 0x6b (POISON_FREE) and the last byte is |
731 | * 0xa5 (POISON_END) | |
732 | * | |
3b0efdfa | 733 | * object + s->object_size |
81819f0f | 734 | * Padding to reach word boundary. This is also used for Redzoning. |
672bba3a | 735 | * Padding is extended by another word if Redzoning is enabled and |
3b0efdfa | 736 | * object_size == inuse. |
672bba3a | 737 | * |
81819f0f CL |
738 | * We fill with 0xbb (RED_INACTIVE) for inactive objects and with |
739 | * 0xcc (RED_ACTIVE) for objects in use. | |
740 | * | |
741 | * object + s->inuse | |
672bba3a CL |
742 | * Meta data starts here. |
743 | * | |
81819f0f CL |
744 | * A. Free pointer (if we cannot overwrite object on free) |
745 | * B. Tracking data for SLAB_STORE_USER | |
672bba3a | 746 | * C. Padding to reach required alignment boundary or at mininum |
6446faa2 | 747 | * one word if debugging is on to be able to detect writes |
672bba3a CL |
748 | * before the word boundary. |
749 | * | |
750 | * Padding is done using 0x5a (POISON_INUSE) | |
81819f0f CL |
751 | * |
752 | * object + s->size | |
672bba3a | 753 | * Nothing is used beyond s->size. |
81819f0f | 754 | * |
3b0efdfa | 755 | * If slabcaches are merged then the object_size and inuse boundaries are mostly |
672bba3a | 756 | * ignored. And therefore no slab options that rely on these boundaries |
81819f0f CL |
757 | * may be used with merged slabcaches. |
758 | */ | |
759 | ||
81819f0f CL |
760 | static int check_pad_bytes(struct kmem_cache *s, struct page *page, u8 *p) |
761 | { | |
762 | unsigned long off = s->inuse; /* The end of info */ | |
763 | ||
764 | if (s->offset) | |
765 | /* Freepointer is placed after the object. */ | |
766 | off += sizeof(void *); | |
767 | ||
768 | if (s->flags & SLAB_STORE_USER) | |
769 | /* We also have user information there */ | |
770 | off += 2 * sizeof(struct track); | |
771 | ||
772 | if (s->size == off) | |
773 | return 1; | |
774 | ||
24922684 CL |
775 | return check_bytes_and_report(s, page, p, "Object padding", |
776 | p + off, POISON_INUSE, s->size - off); | |
81819f0f CL |
777 | } |
778 | ||
39b26464 | 779 | /* Check the pad bytes at the end of a slab page */ |
81819f0f CL |
780 | static int slab_pad_check(struct kmem_cache *s, struct page *page) |
781 | { | |
24922684 CL |
782 | u8 *start; |
783 | u8 *fault; | |
784 | u8 *end; | |
785 | int length; | |
786 | int remainder; | |
81819f0f CL |
787 | |
788 | if (!(s->flags & SLAB_POISON)) | |
789 | return 1; | |
790 | ||
a973e9dd | 791 | start = page_address(page); |
ab9a0f19 | 792 | length = (PAGE_SIZE << compound_order(page)) - s->reserved; |
39b26464 CL |
793 | end = start + length; |
794 | remainder = length % s->size; | |
81819f0f CL |
795 | if (!remainder) |
796 | return 1; | |
797 | ||
a79316c6 | 798 | metadata_access_enable(); |
79824820 | 799 | fault = memchr_inv(end - remainder, POISON_INUSE, remainder); |
a79316c6 | 800 | metadata_access_disable(); |
24922684 CL |
801 | if (!fault) |
802 | return 1; | |
803 | while (end > fault && end[-1] == POISON_INUSE) | |
804 | end--; | |
805 | ||
806 | slab_err(s, page, "Padding overwritten. 0x%p-0x%p", fault, end - 1); | |
ffc79d28 | 807 | print_section("Padding ", end - remainder, remainder); |
24922684 | 808 | |
8a3d271d | 809 | restore_bytes(s, "slab padding", POISON_INUSE, end - remainder, end); |
24922684 | 810 | return 0; |
81819f0f CL |
811 | } |
812 | ||
813 | static int check_object(struct kmem_cache *s, struct page *page, | |
f7cb1933 | 814 | void *object, u8 val) |
81819f0f CL |
815 | { |
816 | u8 *p = object; | |
3b0efdfa | 817 | u8 *endobject = object + s->object_size; |
81819f0f CL |
818 | |
819 | if (s->flags & SLAB_RED_ZONE) { | |
24922684 | 820 | if (!check_bytes_and_report(s, page, object, "Redzone", |
3b0efdfa | 821 | endobject, val, s->inuse - s->object_size)) |
81819f0f | 822 | return 0; |
81819f0f | 823 | } else { |
3b0efdfa | 824 | if ((s->flags & SLAB_POISON) && s->object_size < s->inuse) { |
3adbefee | 825 | check_bytes_and_report(s, page, p, "Alignment padding", |
d0e0ac97 CG |
826 | endobject, POISON_INUSE, |
827 | s->inuse - s->object_size); | |
3adbefee | 828 | } |
81819f0f CL |
829 | } |
830 | ||
831 | if (s->flags & SLAB_POISON) { | |
f7cb1933 | 832 | if (val != SLUB_RED_ACTIVE && (s->flags & __OBJECT_POISON) && |
24922684 | 833 | (!check_bytes_and_report(s, page, p, "Poison", p, |
3b0efdfa | 834 | POISON_FREE, s->object_size - 1) || |
24922684 | 835 | !check_bytes_and_report(s, page, p, "Poison", |
3b0efdfa | 836 | p + s->object_size - 1, POISON_END, 1))) |
81819f0f | 837 | return 0; |
81819f0f CL |
838 | /* |
839 | * check_pad_bytes cleans up on its own. | |
840 | */ | |
841 | check_pad_bytes(s, page, p); | |
842 | } | |
843 | ||
f7cb1933 | 844 | if (!s->offset && val == SLUB_RED_ACTIVE) |
81819f0f CL |
845 | /* |
846 | * Object and freepointer overlap. Cannot check | |
847 | * freepointer while object is allocated. | |
848 | */ | |
849 | return 1; | |
850 | ||
851 | /* Check free pointer validity */ | |
852 | if (!check_valid_pointer(s, page, get_freepointer(s, p))) { | |
853 | object_err(s, page, p, "Freepointer corrupt"); | |
854 | /* | |
9f6c708e | 855 | * No choice but to zap it and thus lose the remainder |
81819f0f | 856 | * of the free objects in this slab. May cause |
672bba3a | 857 | * another error because the object count is now wrong. |
81819f0f | 858 | */ |
a973e9dd | 859 | set_freepointer(s, p, NULL); |
81819f0f CL |
860 | return 0; |
861 | } | |
862 | return 1; | |
863 | } | |
864 | ||
865 | static int check_slab(struct kmem_cache *s, struct page *page) | |
866 | { | |
39b26464 CL |
867 | int maxobj; |
868 | ||
81819f0f CL |
869 | VM_BUG_ON(!irqs_disabled()); |
870 | ||
871 | if (!PageSlab(page)) { | |
24922684 | 872 | slab_err(s, page, "Not a valid slab page"); |
81819f0f CL |
873 | return 0; |
874 | } | |
39b26464 | 875 | |
ab9a0f19 | 876 | maxobj = order_objects(compound_order(page), s->size, s->reserved); |
39b26464 CL |
877 | if (page->objects > maxobj) { |
878 | slab_err(s, page, "objects %u > max %u", | |
f6edde9c | 879 | page->objects, maxobj); |
39b26464 CL |
880 | return 0; |
881 | } | |
882 | if (page->inuse > page->objects) { | |
24922684 | 883 | slab_err(s, page, "inuse %u > max %u", |
f6edde9c | 884 | page->inuse, page->objects); |
81819f0f CL |
885 | return 0; |
886 | } | |
887 | /* Slab_pad_check fixes things up after itself */ | |
888 | slab_pad_check(s, page); | |
889 | return 1; | |
890 | } | |
891 | ||
892 | /* | |
672bba3a CL |
893 | * Determine if a certain object on a page is on the freelist. Must hold the |
894 | * slab lock to guarantee that the chains are in a consistent state. | |
81819f0f CL |
895 | */ |
896 | static int on_freelist(struct kmem_cache *s, struct page *page, void *search) | |
897 | { | |
898 | int nr = 0; | |
881db7fb | 899 | void *fp; |
81819f0f | 900 | void *object = NULL; |
f6edde9c | 901 | int max_objects; |
81819f0f | 902 | |
881db7fb | 903 | fp = page->freelist; |
39b26464 | 904 | while (fp && nr <= page->objects) { |
81819f0f CL |
905 | if (fp == search) |
906 | return 1; | |
907 | if (!check_valid_pointer(s, page, fp)) { | |
908 | if (object) { | |
909 | object_err(s, page, object, | |
910 | "Freechain corrupt"); | |
a973e9dd | 911 | set_freepointer(s, object, NULL); |
81819f0f | 912 | } else { |
24922684 | 913 | slab_err(s, page, "Freepointer corrupt"); |
a973e9dd | 914 | page->freelist = NULL; |
39b26464 | 915 | page->inuse = page->objects; |
24922684 | 916 | slab_fix(s, "Freelist cleared"); |
81819f0f CL |
917 | return 0; |
918 | } | |
919 | break; | |
920 | } | |
921 | object = fp; | |
922 | fp = get_freepointer(s, object); | |
923 | nr++; | |
924 | } | |
925 | ||
ab9a0f19 | 926 | max_objects = order_objects(compound_order(page), s->size, s->reserved); |
210b5c06 CG |
927 | if (max_objects > MAX_OBJS_PER_PAGE) |
928 | max_objects = MAX_OBJS_PER_PAGE; | |
224a88be CL |
929 | |
930 | if (page->objects != max_objects) { | |
931 | slab_err(s, page, "Wrong number of objects. Found %d but " | |
932 | "should be %d", page->objects, max_objects); | |
933 | page->objects = max_objects; | |
934 | slab_fix(s, "Number of objects adjusted."); | |
935 | } | |
39b26464 | 936 | if (page->inuse != page->objects - nr) { |
70d71228 | 937 | slab_err(s, page, "Wrong object count. Counter is %d but " |
39b26464 CL |
938 | "counted were %d", page->inuse, page->objects - nr); |
939 | page->inuse = page->objects - nr; | |
24922684 | 940 | slab_fix(s, "Object count adjusted."); |
81819f0f CL |
941 | } |
942 | return search == NULL; | |
943 | } | |
944 | ||
0121c619 CL |
945 | static void trace(struct kmem_cache *s, struct page *page, void *object, |
946 | int alloc) | |
3ec09742 CL |
947 | { |
948 | if (s->flags & SLAB_TRACE) { | |
f9f58285 | 949 | pr_info("TRACE %s %s 0x%p inuse=%d fp=0x%p\n", |
3ec09742 CL |
950 | s->name, |
951 | alloc ? "alloc" : "free", | |
952 | object, page->inuse, | |
953 | page->freelist); | |
954 | ||
955 | if (!alloc) | |
d0e0ac97 CG |
956 | print_section("Object ", (void *)object, |
957 | s->object_size); | |
3ec09742 CL |
958 | |
959 | dump_stack(); | |
960 | } | |
961 | } | |
962 | ||
643b1138 | 963 | /* |
672bba3a | 964 | * Tracking of fully allocated slabs for debugging purposes. |
643b1138 | 965 | */ |
5cc6eee8 CL |
966 | static void add_full(struct kmem_cache *s, |
967 | struct kmem_cache_node *n, struct page *page) | |
643b1138 | 968 | { |
5cc6eee8 CL |
969 | if (!(s->flags & SLAB_STORE_USER)) |
970 | return; | |
971 | ||
255d0884 | 972 | lockdep_assert_held(&n->list_lock); |
643b1138 | 973 | list_add(&page->lru, &n->full); |
643b1138 CL |
974 | } |
975 | ||
c65c1877 | 976 | static void remove_full(struct kmem_cache *s, struct kmem_cache_node *n, struct page *page) |
643b1138 | 977 | { |
643b1138 CL |
978 | if (!(s->flags & SLAB_STORE_USER)) |
979 | return; | |
980 | ||
255d0884 | 981 | lockdep_assert_held(&n->list_lock); |
643b1138 | 982 | list_del(&page->lru); |
643b1138 CL |
983 | } |
984 | ||
0f389ec6 CL |
985 | /* Tracking of the number of slabs for debugging purposes */ |
986 | static inline unsigned long slabs_node(struct kmem_cache *s, int node) | |
987 | { | |
988 | struct kmem_cache_node *n = get_node(s, node); | |
989 | ||
990 | return atomic_long_read(&n->nr_slabs); | |
991 | } | |
992 | ||
26c02cf0 AB |
993 | static inline unsigned long node_nr_slabs(struct kmem_cache_node *n) |
994 | { | |
995 | return atomic_long_read(&n->nr_slabs); | |
996 | } | |
997 | ||
205ab99d | 998 | static inline void inc_slabs_node(struct kmem_cache *s, int node, int objects) |
0f389ec6 CL |
999 | { |
1000 | struct kmem_cache_node *n = get_node(s, node); | |
1001 | ||
1002 | /* | |
1003 | * May be called early in order to allocate a slab for the | |
1004 | * kmem_cache_node structure. Solve the chicken-egg | |
1005 | * dilemma by deferring the increment of the count during | |
1006 | * bootstrap (see early_kmem_cache_node_alloc). | |
1007 | */ | |
338b2642 | 1008 | if (likely(n)) { |
0f389ec6 | 1009 | atomic_long_inc(&n->nr_slabs); |
205ab99d CL |
1010 | atomic_long_add(objects, &n->total_objects); |
1011 | } | |
0f389ec6 | 1012 | } |
205ab99d | 1013 | static inline void dec_slabs_node(struct kmem_cache *s, int node, int objects) |
0f389ec6 CL |
1014 | { |
1015 | struct kmem_cache_node *n = get_node(s, node); | |
1016 | ||
1017 | atomic_long_dec(&n->nr_slabs); | |
205ab99d | 1018 | atomic_long_sub(objects, &n->total_objects); |
0f389ec6 CL |
1019 | } |
1020 | ||
1021 | /* Object debug checks for alloc/free paths */ | |
3ec09742 CL |
1022 | static void setup_object_debug(struct kmem_cache *s, struct page *page, |
1023 | void *object) | |
1024 | { | |
1025 | if (!(s->flags & (SLAB_STORE_USER|SLAB_RED_ZONE|__OBJECT_POISON))) | |
1026 | return; | |
1027 | ||
f7cb1933 | 1028 | init_object(s, object, SLUB_RED_INACTIVE); |
3ec09742 CL |
1029 | init_tracking(s, object); |
1030 | } | |
1031 | ||
d0e0ac97 CG |
1032 | static noinline int alloc_debug_processing(struct kmem_cache *s, |
1033 | struct page *page, | |
ce71e27c | 1034 | void *object, unsigned long addr) |
81819f0f CL |
1035 | { |
1036 | if (!check_slab(s, page)) | |
1037 | goto bad; | |
1038 | ||
81819f0f CL |
1039 | if (!check_valid_pointer(s, page, object)) { |
1040 | object_err(s, page, object, "Freelist Pointer check fails"); | |
70d71228 | 1041 | goto bad; |
81819f0f CL |
1042 | } |
1043 | ||
f7cb1933 | 1044 | if (!check_object(s, page, object, SLUB_RED_INACTIVE)) |
81819f0f | 1045 | goto bad; |
81819f0f | 1046 | |
3ec09742 CL |
1047 | /* Success perform special debug activities for allocs */ |
1048 | if (s->flags & SLAB_STORE_USER) | |
1049 | set_track(s, object, TRACK_ALLOC, addr); | |
1050 | trace(s, page, object, 1); | |
f7cb1933 | 1051 | init_object(s, object, SLUB_RED_ACTIVE); |
81819f0f | 1052 | return 1; |
3ec09742 | 1053 | |
81819f0f CL |
1054 | bad: |
1055 | if (PageSlab(page)) { | |
1056 | /* | |
1057 | * If this is a slab page then lets do the best we can | |
1058 | * to avoid issues in the future. Marking all objects | |
672bba3a | 1059 | * as used avoids touching the remaining objects. |
81819f0f | 1060 | */ |
24922684 | 1061 | slab_fix(s, "Marking all objects used"); |
39b26464 | 1062 | page->inuse = page->objects; |
a973e9dd | 1063 | page->freelist = NULL; |
81819f0f CL |
1064 | } |
1065 | return 0; | |
1066 | } | |
1067 | ||
19c7ff9e CL |
1068 | static noinline struct kmem_cache_node *free_debug_processing( |
1069 | struct kmem_cache *s, struct page *page, void *object, | |
1070 | unsigned long addr, unsigned long *flags) | |
81819f0f | 1071 | { |
19c7ff9e | 1072 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); |
5c2e4bbb | 1073 | |
19c7ff9e | 1074 | spin_lock_irqsave(&n->list_lock, *flags); |
881db7fb CL |
1075 | slab_lock(page); |
1076 | ||
81819f0f CL |
1077 | if (!check_slab(s, page)) |
1078 | goto fail; | |
1079 | ||
1080 | if (!check_valid_pointer(s, page, object)) { | |
70d71228 | 1081 | slab_err(s, page, "Invalid object pointer 0x%p", object); |
81819f0f CL |
1082 | goto fail; |
1083 | } | |
1084 | ||
1085 | if (on_freelist(s, page, object)) { | |
24922684 | 1086 | object_err(s, page, object, "Object already free"); |
81819f0f CL |
1087 | goto fail; |
1088 | } | |
1089 | ||
f7cb1933 | 1090 | if (!check_object(s, page, object, SLUB_RED_ACTIVE)) |
5c2e4bbb | 1091 | goto out; |
81819f0f | 1092 | |
1b4f59e3 | 1093 | if (unlikely(s != page->slab_cache)) { |
3adbefee | 1094 | if (!PageSlab(page)) { |
70d71228 CL |
1095 | slab_err(s, page, "Attempt to free object(0x%p) " |
1096 | "outside of slab", object); | |
1b4f59e3 | 1097 | } else if (!page->slab_cache) { |
f9f58285 FF |
1098 | pr_err("SLUB <none>: no slab for object 0x%p.\n", |
1099 | object); | |
70d71228 | 1100 | dump_stack(); |
06428780 | 1101 | } else |
24922684 CL |
1102 | object_err(s, page, object, |
1103 | "page slab pointer corrupt."); | |
81819f0f CL |
1104 | goto fail; |
1105 | } | |
3ec09742 | 1106 | |
3ec09742 CL |
1107 | if (s->flags & SLAB_STORE_USER) |
1108 | set_track(s, object, TRACK_FREE, addr); | |
1109 | trace(s, page, object, 0); | |
f7cb1933 | 1110 | init_object(s, object, SLUB_RED_INACTIVE); |
5c2e4bbb | 1111 | out: |
881db7fb | 1112 | slab_unlock(page); |
19c7ff9e CL |
1113 | /* |
1114 | * Keep node_lock to preserve integrity | |
1115 | * until the object is actually freed | |
1116 | */ | |
1117 | return n; | |
3ec09742 | 1118 | |
81819f0f | 1119 | fail: |
19c7ff9e CL |
1120 | slab_unlock(page); |
1121 | spin_unlock_irqrestore(&n->list_lock, *flags); | |
24922684 | 1122 | slab_fix(s, "Object at 0x%p not freed", object); |
19c7ff9e | 1123 | return NULL; |
81819f0f CL |
1124 | } |
1125 | ||
41ecc55b CL |
1126 | static int __init setup_slub_debug(char *str) |
1127 | { | |
f0630fff CL |
1128 | slub_debug = DEBUG_DEFAULT_FLAGS; |
1129 | if (*str++ != '=' || !*str) | |
1130 | /* | |
1131 | * No options specified. Switch on full debugging. | |
1132 | */ | |
1133 | goto out; | |
1134 | ||
1135 | if (*str == ',') | |
1136 | /* | |
1137 | * No options but restriction on slabs. This means full | |
1138 | * debugging for slabs matching a pattern. | |
1139 | */ | |
1140 | goto check_slabs; | |
1141 | ||
1142 | slub_debug = 0; | |
1143 | if (*str == '-') | |
1144 | /* | |
1145 | * Switch off all debugging measures. | |
1146 | */ | |
1147 | goto out; | |
1148 | ||
1149 | /* | |
1150 | * Determine which debug features should be switched on | |
1151 | */ | |
06428780 | 1152 | for (; *str && *str != ','; str++) { |
f0630fff CL |
1153 | switch (tolower(*str)) { |
1154 | case 'f': | |
1155 | slub_debug |= SLAB_DEBUG_FREE; | |
1156 | break; | |
1157 | case 'z': | |
1158 | slub_debug |= SLAB_RED_ZONE; | |
1159 | break; | |
1160 | case 'p': | |
1161 | slub_debug |= SLAB_POISON; | |
1162 | break; | |
1163 | case 'u': | |
1164 | slub_debug |= SLAB_STORE_USER; | |
1165 | break; | |
1166 | case 't': | |
1167 | slub_debug |= SLAB_TRACE; | |
1168 | break; | |
4c13dd3b DM |
1169 | case 'a': |
1170 | slub_debug |= SLAB_FAILSLAB; | |
1171 | break; | |
08303a73 CA |
1172 | case 'o': |
1173 | /* | |
1174 | * Avoid enabling debugging on caches if its minimum | |
1175 | * order would increase as a result. | |
1176 | */ | |
1177 | disable_higher_order_debug = 1; | |
1178 | break; | |
f0630fff | 1179 | default: |
f9f58285 FF |
1180 | pr_err("slub_debug option '%c' unknown. skipped\n", |
1181 | *str); | |
f0630fff | 1182 | } |
41ecc55b CL |
1183 | } |
1184 | ||
f0630fff | 1185 | check_slabs: |
41ecc55b CL |
1186 | if (*str == ',') |
1187 | slub_debug_slabs = str + 1; | |
f0630fff | 1188 | out: |
41ecc55b CL |
1189 | return 1; |
1190 | } | |
1191 | ||
1192 | __setup("slub_debug", setup_slub_debug); | |
1193 | ||
423c929c | 1194 | unsigned long kmem_cache_flags(unsigned long object_size, |
ba0268a8 | 1195 | unsigned long flags, const char *name, |
51cc5068 | 1196 | void (*ctor)(void *)) |
41ecc55b CL |
1197 | { |
1198 | /* | |
e153362a | 1199 | * Enable debugging if selected on the kernel commandline. |
41ecc55b | 1200 | */ |
c6f58d9b CL |
1201 | if (slub_debug && (!slub_debug_slabs || (name && |
1202 | !strncmp(slub_debug_slabs, name, strlen(slub_debug_slabs))))) | |
3de47213 | 1203 | flags |= slub_debug; |
ba0268a8 CL |
1204 | |
1205 | return flags; | |
41ecc55b CL |
1206 | } |
1207 | #else | |
3ec09742 CL |
1208 | static inline void setup_object_debug(struct kmem_cache *s, |
1209 | struct page *page, void *object) {} | |
41ecc55b | 1210 | |
3ec09742 | 1211 | static inline int alloc_debug_processing(struct kmem_cache *s, |
ce71e27c | 1212 | struct page *page, void *object, unsigned long addr) { return 0; } |
41ecc55b | 1213 | |
19c7ff9e CL |
1214 | static inline struct kmem_cache_node *free_debug_processing( |
1215 | struct kmem_cache *s, struct page *page, void *object, | |
1216 | unsigned long addr, unsigned long *flags) { return NULL; } | |
41ecc55b | 1217 | |
41ecc55b CL |
1218 | static inline int slab_pad_check(struct kmem_cache *s, struct page *page) |
1219 | { return 1; } | |
1220 | static inline int check_object(struct kmem_cache *s, struct page *page, | |
f7cb1933 | 1221 | void *object, u8 val) { return 1; } |
5cc6eee8 CL |
1222 | static inline void add_full(struct kmem_cache *s, struct kmem_cache_node *n, |
1223 | struct page *page) {} | |
c65c1877 PZ |
1224 | static inline void remove_full(struct kmem_cache *s, struct kmem_cache_node *n, |
1225 | struct page *page) {} | |
423c929c | 1226 | unsigned long kmem_cache_flags(unsigned long object_size, |
ba0268a8 | 1227 | unsigned long flags, const char *name, |
51cc5068 | 1228 | void (*ctor)(void *)) |
ba0268a8 CL |
1229 | { |
1230 | return flags; | |
1231 | } | |
41ecc55b | 1232 | #define slub_debug 0 |
0f389ec6 | 1233 | |
fdaa45e9 IM |
1234 | #define disable_higher_order_debug 0 |
1235 | ||
0f389ec6 CL |
1236 | static inline unsigned long slabs_node(struct kmem_cache *s, int node) |
1237 | { return 0; } | |
26c02cf0 AB |
1238 | static inline unsigned long node_nr_slabs(struct kmem_cache_node *n) |
1239 | { return 0; } | |
205ab99d CL |
1240 | static inline void inc_slabs_node(struct kmem_cache *s, int node, |
1241 | int objects) {} | |
1242 | static inline void dec_slabs_node(struct kmem_cache *s, int node, | |
1243 | int objects) {} | |
7d550c56 | 1244 | |
02e72cc6 AR |
1245 | #endif /* CONFIG_SLUB_DEBUG */ |
1246 | ||
1247 | /* | |
1248 | * Hooks for other subsystems that check memory allocations. In a typical | |
1249 | * production configuration these hooks all should produce no code at all. | |
1250 | */ | |
d56791b3 RB |
1251 | static inline void kmalloc_large_node_hook(void *ptr, size_t size, gfp_t flags) |
1252 | { | |
1253 | kmemleak_alloc(ptr, size, 1, flags); | |
0316bec2 | 1254 | kasan_kmalloc_large(ptr, size); |
d56791b3 RB |
1255 | } |
1256 | ||
1257 | static inline void kfree_hook(const void *x) | |
1258 | { | |
1259 | kmemleak_free(x); | |
0316bec2 | 1260 | kasan_kfree_large(x); |
d56791b3 RB |
1261 | } |
1262 | ||
8135be5a VD |
1263 | static inline struct kmem_cache *slab_pre_alloc_hook(struct kmem_cache *s, |
1264 | gfp_t flags) | |
02e72cc6 AR |
1265 | { |
1266 | flags &= gfp_allowed_mask; | |
1267 | lockdep_trace_alloc(flags); | |
d0164adc | 1268 | might_sleep_if(gfpflags_allow_blocking(flags)); |
7d550c56 | 1269 | |
8135be5a VD |
1270 | if (should_failslab(s->object_size, flags, s->flags)) |
1271 | return NULL; | |
1272 | ||
1273 | return memcg_kmem_get_cache(s, flags); | |
02e72cc6 AR |
1274 | } |
1275 | ||
1276 | static inline void slab_post_alloc_hook(struct kmem_cache *s, | |
1277 | gfp_t flags, void *object) | |
d56791b3 | 1278 | { |
02e72cc6 AR |
1279 | flags &= gfp_allowed_mask; |
1280 | kmemcheck_slab_alloc(s, flags, object, slab_ksize(s)); | |
1281 | kmemleak_alloc_recursive(object, s->object_size, 1, s->flags, flags); | |
8135be5a | 1282 | memcg_kmem_put_cache(s); |
0316bec2 | 1283 | kasan_slab_alloc(s, object); |
d56791b3 | 1284 | } |
7d550c56 | 1285 | |
d56791b3 RB |
1286 | static inline void slab_free_hook(struct kmem_cache *s, void *x) |
1287 | { | |
1288 | kmemleak_free_recursive(x, s->flags); | |
7d550c56 | 1289 | |
02e72cc6 AR |
1290 | /* |
1291 | * Trouble is that we may no longer disable interrupts in the fast path | |
1292 | * So in order to make the debug calls that expect irqs to be | |
1293 | * disabled we need to disable interrupts temporarily. | |
1294 | */ | |
1295 | #if defined(CONFIG_KMEMCHECK) || defined(CONFIG_LOCKDEP) | |
1296 | { | |
1297 | unsigned long flags; | |
1298 | ||
1299 | local_irq_save(flags); | |
1300 | kmemcheck_slab_free(s, x, s->object_size); | |
1301 | debug_check_no_locks_freed(x, s->object_size); | |
1302 | local_irq_restore(flags); | |
1303 | } | |
1304 | #endif | |
1305 | if (!(s->flags & SLAB_DEBUG_OBJECTS)) | |
1306 | debug_check_no_obj_freed(x, s->object_size); | |
0316bec2 AR |
1307 | |
1308 | kasan_slab_free(s, x); | |
02e72cc6 | 1309 | } |
205ab99d | 1310 | |
588f8ba9 TG |
1311 | static void setup_object(struct kmem_cache *s, struct page *page, |
1312 | void *object) | |
1313 | { | |
1314 | setup_object_debug(s, page, object); | |
1315 | if (unlikely(s->ctor)) { | |
1316 | kasan_unpoison_object_data(s, object); | |
1317 | s->ctor(object); | |
1318 | kasan_poison_object_data(s, object); | |
1319 | } | |
1320 | } | |
1321 | ||
81819f0f CL |
1322 | /* |
1323 | * Slab allocation and freeing | |
1324 | */ | |
5dfb4175 VD |
1325 | static inline struct page *alloc_slab_page(struct kmem_cache *s, |
1326 | gfp_t flags, int node, struct kmem_cache_order_objects oo) | |
65c3376a | 1327 | { |
5dfb4175 | 1328 | struct page *page; |
65c3376a CL |
1329 | int order = oo_order(oo); |
1330 | ||
b1eeab67 VN |
1331 | flags |= __GFP_NOTRACK; |
1332 | ||
2154a336 | 1333 | if (node == NUMA_NO_NODE) |
5dfb4175 | 1334 | page = alloc_pages(flags, order); |
65c3376a | 1335 | else |
96db800f | 1336 | page = __alloc_pages_node(node, flags, order); |
5dfb4175 | 1337 | |
f3ccb2c4 VD |
1338 | if (page && memcg_charge_slab(page, flags, order, s)) { |
1339 | __free_pages(page, order); | |
1340 | page = NULL; | |
1341 | } | |
5dfb4175 VD |
1342 | |
1343 | return page; | |
65c3376a CL |
1344 | } |
1345 | ||
81819f0f CL |
1346 | static struct page *allocate_slab(struct kmem_cache *s, gfp_t flags, int node) |
1347 | { | |
06428780 | 1348 | struct page *page; |
834f3d11 | 1349 | struct kmem_cache_order_objects oo = s->oo; |
ba52270d | 1350 | gfp_t alloc_gfp; |
588f8ba9 TG |
1351 | void *start, *p; |
1352 | int idx, order; | |
81819f0f | 1353 | |
7e0528da CL |
1354 | flags &= gfp_allowed_mask; |
1355 | ||
d0164adc | 1356 | if (gfpflags_allow_blocking(flags)) |
7e0528da CL |
1357 | local_irq_enable(); |
1358 | ||
b7a49f0d | 1359 | flags |= s->allocflags; |
e12ba74d | 1360 | |
ba52270d PE |
1361 | /* |
1362 | * Let the initial higher-order allocation fail under memory pressure | |
1363 | * so we fall-back to the minimum order allocation. | |
1364 | */ | |
1365 | alloc_gfp = (flags | __GFP_NOWARN | __GFP_NORETRY) & ~__GFP_NOFAIL; | |
d0164adc MG |
1366 | if ((alloc_gfp & __GFP_DIRECT_RECLAIM) && oo_order(oo) > oo_order(s->min)) |
1367 | alloc_gfp = (alloc_gfp | __GFP_NOMEMALLOC) & ~__GFP_DIRECT_RECLAIM; | |
ba52270d | 1368 | |
5dfb4175 | 1369 | page = alloc_slab_page(s, alloc_gfp, node, oo); |
65c3376a CL |
1370 | if (unlikely(!page)) { |
1371 | oo = s->min; | |
80c3a998 | 1372 | alloc_gfp = flags; |
65c3376a CL |
1373 | /* |
1374 | * Allocation may have failed due to fragmentation. | |
1375 | * Try a lower order alloc if possible | |
1376 | */ | |
5dfb4175 | 1377 | page = alloc_slab_page(s, alloc_gfp, node, oo); |
588f8ba9 TG |
1378 | if (unlikely(!page)) |
1379 | goto out; | |
1380 | stat(s, ORDER_FALLBACK); | |
65c3376a | 1381 | } |
5a896d9e | 1382 | |
588f8ba9 TG |
1383 | if (kmemcheck_enabled && |
1384 | !(s->flags & (SLAB_NOTRACK | DEBUG_DEFAULT_FLAGS))) { | |
b1eeab67 VN |
1385 | int pages = 1 << oo_order(oo); |
1386 | ||
80c3a998 | 1387 | kmemcheck_alloc_shadow(page, oo_order(oo), alloc_gfp, node); |
b1eeab67 VN |
1388 | |
1389 | /* | |
1390 | * Objects from caches that have a constructor don't get | |
1391 | * cleared when they're allocated, so we need to do it here. | |
1392 | */ | |
1393 | if (s->ctor) | |
1394 | kmemcheck_mark_uninitialized_pages(page, pages); | |
1395 | else | |
1396 | kmemcheck_mark_unallocated_pages(page, pages); | |
5a896d9e VN |
1397 | } |
1398 | ||
834f3d11 | 1399 | page->objects = oo_objects(oo); |
81819f0f | 1400 | |
1f458cbf | 1401 | order = compound_order(page); |
1b4f59e3 | 1402 | page->slab_cache = s; |
c03f94cc | 1403 | __SetPageSlab(page); |
2f064f34 | 1404 | if (page_is_pfmemalloc(page)) |
072bb0aa | 1405 | SetPageSlabPfmemalloc(page); |
81819f0f CL |
1406 | |
1407 | start = page_address(page); | |
81819f0f CL |
1408 | |
1409 | if (unlikely(s->flags & SLAB_POISON)) | |
1f458cbf | 1410 | memset(start, POISON_INUSE, PAGE_SIZE << order); |
81819f0f | 1411 | |
0316bec2 AR |
1412 | kasan_poison_slab(page); |
1413 | ||
54266640 WY |
1414 | for_each_object_idx(p, idx, s, start, page->objects) { |
1415 | setup_object(s, page, p); | |
1416 | if (likely(idx < page->objects)) | |
1417 | set_freepointer(s, p, p + s->size); | |
1418 | else | |
1419 | set_freepointer(s, p, NULL); | |
81819f0f | 1420 | } |
81819f0f CL |
1421 | |
1422 | page->freelist = start; | |
e6e82ea1 | 1423 | page->inuse = page->objects; |
8cb0a506 | 1424 | page->frozen = 1; |
588f8ba9 | 1425 | |
81819f0f | 1426 | out: |
d0164adc | 1427 | if (gfpflags_allow_blocking(flags)) |
588f8ba9 TG |
1428 | local_irq_disable(); |
1429 | if (!page) | |
1430 | return NULL; | |
1431 | ||
1432 | mod_zone_page_state(page_zone(page), | |
1433 | (s->flags & SLAB_RECLAIM_ACCOUNT) ? | |
1434 | NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE, | |
1435 | 1 << oo_order(oo)); | |
1436 | ||
1437 | inc_slabs_node(s, page_to_nid(page), page->objects); | |
1438 | ||
81819f0f CL |
1439 | return page; |
1440 | } | |
1441 | ||
588f8ba9 TG |
1442 | static struct page *new_slab(struct kmem_cache *s, gfp_t flags, int node) |
1443 | { | |
1444 | if (unlikely(flags & GFP_SLAB_BUG_MASK)) { | |
1445 | pr_emerg("gfp: %u\n", flags & GFP_SLAB_BUG_MASK); | |
1446 | BUG(); | |
1447 | } | |
1448 | ||
1449 | return allocate_slab(s, | |
1450 | flags & (GFP_RECLAIM_MASK | GFP_CONSTRAINT_MASK), node); | |
1451 | } | |
1452 | ||
81819f0f CL |
1453 | static void __free_slab(struct kmem_cache *s, struct page *page) |
1454 | { | |
834f3d11 CL |
1455 | int order = compound_order(page); |
1456 | int pages = 1 << order; | |
81819f0f | 1457 | |
af537b0a | 1458 | if (kmem_cache_debug(s)) { |
81819f0f CL |
1459 | void *p; |
1460 | ||
1461 | slab_pad_check(s, page); | |
224a88be CL |
1462 | for_each_object(p, s, page_address(page), |
1463 | page->objects) | |
f7cb1933 | 1464 | check_object(s, page, p, SLUB_RED_INACTIVE); |
81819f0f CL |
1465 | } |
1466 | ||
b1eeab67 | 1467 | kmemcheck_free_shadow(page, compound_order(page)); |
5a896d9e | 1468 | |
81819f0f CL |
1469 | mod_zone_page_state(page_zone(page), |
1470 | (s->flags & SLAB_RECLAIM_ACCOUNT) ? | |
1471 | NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE, | |
06428780 | 1472 | -pages); |
81819f0f | 1473 | |
072bb0aa | 1474 | __ClearPageSlabPfmemalloc(page); |
49bd5221 | 1475 | __ClearPageSlab(page); |
1f458cbf | 1476 | |
22b751c3 | 1477 | page_mapcount_reset(page); |
1eb5ac64 NP |
1478 | if (current->reclaim_state) |
1479 | current->reclaim_state->reclaimed_slab += pages; | |
f3ccb2c4 | 1480 | __free_kmem_pages(page, order); |
81819f0f CL |
1481 | } |
1482 | ||
da9a638c LJ |
1483 | #define need_reserve_slab_rcu \ |
1484 | (sizeof(((struct page *)NULL)->lru) < sizeof(struct rcu_head)) | |
1485 | ||
81819f0f CL |
1486 | static void rcu_free_slab(struct rcu_head *h) |
1487 | { | |
1488 | struct page *page; | |
1489 | ||
da9a638c LJ |
1490 | if (need_reserve_slab_rcu) |
1491 | page = virt_to_head_page(h); | |
1492 | else | |
1493 | page = container_of((struct list_head *)h, struct page, lru); | |
1494 | ||
1b4f59e3 | 1495 | __free_slab(page->slab_cache, page); |
81819f0f CL |
1496 | } |
1497 | ||
1498 | static void free_slab(struct kmem_cache *s, struct page *page) | |
1499 | { | |
1500 | if (unlikely(s->flags & SLAB_DESTROY_BY_RCU)) { | |
da9a638c LJ |
1501 | struct rcu_head *head; |
1502 | ||
1503 | if (need_reserve_slab_rcu) { | |
1504 | int order = compound_order(page); | |
1505 | int offset = (PAGE_SIZE << order) - s->reserved; | |
1506 | ||
1507 | VM_BUG_ON(s->reserved != sizeof(*head)); | |
1508 | head = page_address(page) + offset; | |
1509 | } else { | |
bc4f610d | 1510 | head = &page->rcu_head; |
da9a638c | 1511 | } |
81819f0f CL |
1512 | |
1513 | call_rcu(head, rcu_free_slab); | |
1514 | } else | |
1515 | __free_slab(s, page); | |
1516 | } | |
1517 | ||
1518 | static void discard_slab(struct kmem_cache *s, struct page *page) | |
1519 | { | |
205ab99d | 1520 | dec_slabs_node(s, page_to_nid(page), page->objects); |
81819f0f CL |
1521 | free_slab(s, page); |
1522 | } | |
1523 | ||
1524 | /* | |
5cc6eee8 | 1525 | * Management of partially allocated slabs. |
81819f0f | 1526 | */ |
1e4dd946 SR |
1527 | static inline void |
1528 | __add_partial(struct kmem_cache_node *n, struct page *page, int tail) | |
81819f0f | 1529 | { |
e95eed57 | 1530 | n->nr_partial++; |
136333d1 | 1531 | if (tail == DEACTIVATE_TO_TAIL) |
7c2e132c CL |
1532 | list_add_tail(&page->lru, &n->partial); |
1533 | else | |
1534 | list_add(&page->lru, &n->partial); | |
81819f0f CL |
1535 | } |
1536 | ||
1e4dd946 SR |
1537 | static inline void add_partial(struct kmem_cache_node *n, |
1538 | struct page *page, int tail) | |
62e346a8 | 1539 | { |
c65c1877 | 1540 | lockdep_assert_held(&n->list_lock); |
1e4dd946 SR |
1541 | __add_partial(n, page, tail); |
1542 | } | |
c65c1877 | 1543 | |
1e4dd946 SR |
1544 | static inline void |
1545 | __remove_partial(struct kmem_cache_node *n, struct page *page) | |
1546 | { | |
62e346a8 CL |
1547 | list_del(&page->lru); |
1548 | n->nr_partial--; | |
1549 | } | |
1550 | ||
1e4dd946 SR |
1551 | static inline void remove_partial(struct kmem_cache_node *n, |
1552 | struct page *page) | |
1553 | { | |
1554 | lockdep_assert_held(&n->list_lock); | |
1555 | __remove_partial(n, page); | |
1556 | } | |
1557 | ||
81819f0f | 1558 | /* |
7ced3719 CL |
1559 | * Remove slab from the partial list, freeze it and |
1560 | * return the pointer to the freelist. | |
81819f0f | 1561 | * |
497b66f2 | 1562 | * Returns a list of objects or NULL if it fails. |
81819f0f | 1563 | */ |
497b66f2 | 1564 | static inline void *acquire_slab(struct kmem_cache *s, |
acd19fd1 | 1565 | struct kmem_cache_node *n, struct page *page, |
633b0764 | 1566 | int mode, int *objects) |
81819f0f | 1567 | { |
2cfb7455 CL |
1568 | void *freelist; |
1569 | unsigned long counters; | |
1570 | struct page new; | |
1571 | ||
c65c1877 PZ |
1572 | lockdep_assert_held(&n->list_lock); |
1573 | ||
2cfb7455 CL |
1574 | /* |
1575 | * Zap the freelist and set the frozen bit. | |
1576 | * The old freelist is the list of objects for the | |
1577 | * per cpu allocation list. | |
1578 | */ | |
7ced3719 CL |
1579 | freelist = page->freelist; |
1580 | counters = page->counters; | |
1581 | new.counters = counters; | |
633b0764 | 1582 | *objects = new.objects - new.inuse; |
23910c50 | 1583 | if (mode) { |
7ced3719 | 1584 | new.inuse = page->objects; |
23910c50 PE |
1585 | new.freelist = NULL; |
1586 | } else { | |
1587 | new.freelist = freelist; | |
1588 | } | |
2cfb7455 | 1589 | |
a0132ac0 | 1590 | VM_BUG_ON(new.frozen); |
7ced3719 | 1591 | new.frozen = 1; |
2cfb7455 | 1592 | |
7ced3719 | 1593 | if (!__cmpxchg_double_slab(s, page, |
2cfb7455 | 1594 | freelist, counters, |
02d7633f | 1595 | new.freelist, new.counters, |
7ced3719 | 1596 | "acquire_slab")) |
7ced3719 | 1597 | return NULL; |
2cfb7455 CL |
1598 | |
1599 | remove_partial(n, page); | |
7ced3719 | 1600 | WARN_ON(!freelist); |
49e22585 | 1601 | return freelist; |
81819f0f CL |
1602 | } |
1603 | ||
633b0764 | 1604 | static void put_cpu_partial(struct kmem_cache *s, struct page *page, int drain); |
8ba00bb6 | 1605 | static inline bool pfmemalloc_match(struct page *page, gfp_t gfpflags); |
49e22585 | 1606 | |
81819f0f | 1607 | /* |
672bba3a | 1608 | * Try to allocate a partial slab from a specific node. |
81819f0f | 1609 | */ |
8ba00bb6 JK |
1610 | static void *get_partial_node(struct kmem_cache *s, struct kmem_cache_node *n, |
1611 | struct kmem_cache_cpu *c, gfp_t flags) | |
81819f0f | 1612 | { |
49e22585 CL |
1613 | struct page *page, *page2; |
1614 | void *object = NULL; | |
633b0764 JK |
1615 | int available = 0; |
1616 | int objects; | |
81819f0f CL |
1617 | |
1618 | /* | |
1619 | * Racy check. If we mistakenly see no partial slabs then we | |
1620 | * just allocate an empty slab. If we mistakenly try to get a | |
672bba3a CL |
1621 | * partial slab and there is none available then get_partials() |
1622 | * will return NULL. | |
81819f0f CL |
1623 | */ |
1624 | if (!n || !n->nr_partial) | |
1625 | return NULL; | |
1626 | ||
1627 | spin_lock(&n->list_lock); | |
49e22585 | 1628 | list_for_each_entry_safe(page, page2, &n->partial, lru) { |
8ba00bb6 | 1629 | void *t; |
49e22585 | 1630 | |
8ba00bb6 JK |
1631 | if (!pfmemalloc_match(page, flags)) |
1632 | continue; | |
1633 | ||
633b0764 | 1634 | t = acquire_slab(s, n, page, object == NULL, &objects); |
49e22585 CL |
1635 | if (!t) |
1636 | break; | |
1637 | ||
633b0764 | 1638 | available += objects; |
12d79634 | 1639 | if (!object) { |
49e22585 | 1640 | c->page = page; |
49e22585 | 1641 | stat(s, ALLOC_FROM_PARTIAL); |
49e22585 | 1642 | object = t; |
49e22585 | 1643 | } else { |
633b0764 | 1644 | put_cpu_partial(s, page, 0); |
8028dcea | 1645 | stat(s, CPU_PARTIAL_NODE); |
49e22585 | 1646 | } |
345c905d JK |
1647 | if (!kmem_cache_has_cpu_partial(s) |
1648 | || available > s->cpu_partial / 2) | |
49e22585 CL |
1649 | break; |
1650 | ||
497b66f2 | 1651 | } |
81819f0f | 1652 | spin_unlock(&n->list_lock); |
497b66f2 | 1653 | return object; |
81819f0f CL |
1654 | } |
1655 | ||
1656 | /* | |
672bba3a | 1657 | * Get a page from somewhere. Search in increasing NUMA distances. |
81819f0f | 1658 | */ |
de3ec035 | 1659 | static void *get_any_partial(struct kmem_cache *s, gfp_t flags, |
acd19fd1 | 1660 | struct kmem_cache_cpu *c) |
81819f0f CL |
1661 | { |
1662 | #ifdef CONFIG_NUMA | |
1663 | struct zonelist *zonelist; | |
dd1a239f | 1664 | struct zoneref *z; |
54a6eb5c MG |
1665 | struct zone *zone; |
1666 | enum zone_type high_zoneidx = gfp_zone(flags); | |
497b66f2 | 1667 | void *object; |
cc9a6c87 | 1668 | unsigned int cpuset_mems_cookie; |
81819f0f CL |
1669 | |
1670 | /* | |
672bba3a CL |
1671 | * The defrag ratio allows a configuration of the tradeoffs between |
1672 | * inter node defragmentation and node local allocations. A lower | |
1673 | * defrag_ratio increases the tendency to do local allocations | |
1674 | * instead of attempting to obtain partial slabs from other nodes. | |
81819f0f | 1675 | * |
672bba3a CL |
1676 | * If the defrag_ratio is set to 0 then kmalloc() always |
1677 | * returns node local objects. If the ratio is higher then kmalloc() | |
1678 | * may return off node objects because partial slabs are obtained | |
1679 | * from other nodes and filled up. | |
81819f0f | 1680 | * |
6446faa2 | 1681 | * If /sys/kernel/slab/xx/defrag_ratio is set to 100 (which makes |
672bba3a CL |
1682 | * defrag_ratio = 1000) then every (well almost) allocation will |
1683 | * first attempt to defrag slab caches on other nodes. This means | |
1684 | * scanning over all nodes to look for partial slabs which may be | |
1685 | * expensive if we do it every time we are trying to find a slab | |
1686 | * with available objects. | |
81819f0f | 1687 | */ |
9824601e CL |
1688 | if (!s->remote_node_defrag_ratio || |
1689 | get_cycles() % 1024 > s->remote_node_defrag_ratio) | |
81819f0f CL |
1690 | return NULL; |
1691 | ||
cc9a6c87 | 1692 | do { |
d26914d1 | 1693 | cpuset_mems_cookie = read_mems_allowed_begin(); |
2a389610 | 1694 | zonelist = node_zonelist(mempolicy_slab_node(), flags); |
cc9a6c87 MG |
1695 | for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) { |
1696 | struct kmem_cache_node *n; | |
1697 | ||
1698 | n = get_node(s, zone_to_nid(zone)); | |
1699 | ||
dee2f8aa | 1700 | if (n && cpuset_zone_allowed(zone, flags) && |
cc9a6c87 | 1701 | n->nr_partial > s->min_partial) { |
8ba00bb6 | 1702 | object = get_partial_node(s, n, c, flags); |
cc9a6c87 MG |
1703 | if (object) { |
1704 | /* | |
d26914d1 MG |
1705 | * Don't check read_mems_allowed_retry() |
1706 | * here - if mems_allowed was updated in | |
1707 | * parallel, that was a harmless race | |
1708 | * between allocation and the cpuset | |
1709 | * update | |
cc9a6c87 | 1710 | */ |
cc9a6c87 MG |
1711 | return object; |
1712 | } | |
c0ff7453 | 1713 | } |
81819f0f | 1714 | } |
d26914d1 | 1715 | } while (read_mems_allowed_retry(cpuset_mems_cookie)); |
81819f0f CL |
1716 | #endif |
1717 | return NULL; | |
1718 | } | |
1719 | ||
1720 | /* | |
1721 | * Get a partial page, lock it and return it. | |
1722 | */ | |
497b66f2 | 1723 | static void *get_partial(struct kmem_cache *s, gfp_t flags, int node, |
acd19fd1 | 1724 | struct kmem_cache_cpu *c) |
81819f0f | 1725 | { |
497b66f2 | 1726 | void *object; |
a561ce00 JK |
1727 | int searchnode = node; |
1728 | ||
1729 | if (node == NUMA_NO_NODE) | |
1730 | searchnode = numa_mem_id(); | |
1731 | else if (!node_present_pages(node)) | |
1732 | searchnode = node_to_mem_node(node); | |
81819f0f | 1733 | |
8ba00bb6 | 1734 | object = get_partial_node(s, get_node(s, searchnode), c, flags); |
497b66f2 CL |
1735 | if (object || node != NUMA_NO_NODE) |
1736 | return object; | |
81819f0f | 1737 | |
acd19fd1 | 1738 | return get_any_partial(s, flags, c); |
81819f0f CL |
1739 | } |
1740 | ||
8a5ec0ba CL |
1741 | #ifdef CONFIG_PREEMPT |
1742 | /* | |
1743 | * Calculate the next globally unique transaction for disambiguiation | |
1744 | * during cmpxchg. The transactions start with the cpu number and are then | |
1745 | * incremented by CONFIG_NR_CPUS. | |
1746 | */ | |
1747 | #define TID_STEP roundup_pow_of_two(CONFIG_NR_CPUS) | |
1748 | #else | |
1749 | /* | |
1750 | * No preemption supported therefore also no need to check for | |
1751 | * different cpus. | |
1752 | */ | |
1753 | #define TID_STEP 1 | |
1754 | #endif | |
1755 | ||
1756 | static inline unsigned long next_tid(unsigned long tid) | |
1757 | { | |
1758 | return tid + TID_STEP; | |
1759 | } | |
1760 | ||
1761 | static inline unsigned int tid_to_cpu(unsigned long tid) | |
1762 | { | |
1763 | return tid % TID_STEP; | |
1764 | } | |
1765 | ||
1766 | static inline unsigned long tid_to_event(unsigned long tid) | |
1767 | { | |
1768 | return tid / TID_STEP; | |
1769 | } | |
1770 | ||
1771 | static inline unsigned int init_tid(int cpu) | |
1772 | { | |
1773 | return cpu; | |
1774 | } | |
1775 | ||
1776 | static inline void note_cmpxchg_failure(const char *n, | |
1777 | const struct kmem_cache *s, unsigned long tid) | |
1778 | { | |
1779 | #ifdef SLUB_DEBUG_CMPXCHG | |
1780 | unsigned long actual_tid = __this_cpu_read(s->cpu_slab->tid); | |
1781 | ||
f9f58285 | 1782 | pr_info("%s %s: cmpxchg redo ", n, s->name); |
8a5ec0ba CL |
1783 | |
1784 | #ifdef CONFIG_PREEMPT | |
1785 | if (tid_to_cpu(tid) != tid_to_cpu(actual_tid)) | |
f9f58285 | 1786 | pr_warn("due to cpu change %d -> %d\n", |
8a5ec0ba CL |
1787 | tid_to_cpu(tid), tid_to_cpu(actual_tid)); |
1788 | else | |
1789 | #endif | |
1790 | if (tid_to_event(tid) != tid_to_event(actual_tid)) | |
f9f58285 | 1791 | pr_warn("due to cpu running other code. Event %ld->%ld\n", |
8a5ec0ba CL |
1792 | tid_to_event(tid), tid_to_event(actual_tid)); |
1793 | else | |
f9f58285 | 1794 | pr_warn("for unknown reason: actual=%lx was=%lx target=%lx\n", |
8a5ec0ba CL |
1795 | actual_tid, tid, next_tid(tid)); |
1796 | #endif | |
4fdccdfb | 1797 | stat(s, CMPXCHG_DOUBLE_CPU_FAIL); |
8a5ec0ba CL |
1798 | } |
1799 | ||
788e1aad | 1800 | static void init_kmem_cache_cpus(struct kmem_cache *s) |
8a5ec0ba | 1801 | { |
8a5ec0ba CL |
1802 | int cpu; |
1803 | ||
1804 | for_each_possible_cpu(cpu) | |
1805 | per_cpu_ptr(s->cpu_slab, cpu)->tid = init_tid(cpu); | |
8a5ec0ba | 1806 | } |
2cfb7455 | 1807 | |
81819f0f CL |
1808 | /* |
1809 | * Remove the cpu slab | |
1810 | */ | |
d0e0ac97 CG |
1811 | static void deactivate_slab(struct kmem_cache *s, struct page *page, |
1812 | void *freelist) | |
81819f0f | 1813 | { |
2cfb7455 | 1814 | enum slab_modes { M_NONE, M_PARTIAL, M_FULL, M_FREE }; |
2cfb7455 CL |
1815 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); |
1816 | int lock = 0; | |
1817 | enum slab_modes l = M_NONE, m = M_NONE; | |
2cfb7455 | 1818 | void *nextfree; |
136333d1 | 1819 | int tail = DEACTIVATE_TO_HEAD; |
2cfb7455 CL |
1820 | struct page new; |
1821 | struct page old; | |
1822 | ||
1823 | if (page->freelist) { | |
84e554e6 | 1824 | stat(s, DEACTIVATE_REMOTE_FREES); |
136333d1 | 1825 | tail = DEACTIVATE_TO_TAIL; |
2cfb7455 CL |
1826 | } |
1827 | ||
894b8788 | 1828 | /* |
2cfb7455 CL |
1829 | * Stage one: Free all available per cpu objects back |
1830 | * to the page freelist while it is still frozen. Leave the | |
1831 | * last one. | |
1832 | * | |
1833 | * There is no need to take the list->lock because the page | |
1834 | * is still frozen. | |
1835 | */ | |
1836 | while (freelist && (nextfree = get_freepointer(s, freelist))) { | |
1837 | void *prior; | |
1838 | unsigned long counters; | |
1839 | ||
1840 | do { | |
1841 | prior = page->freelist; | |
1842 | counters = page->counters; | |
1843 | set_freepointer(s, freelist, prior); | |
1844 | new.counters = counters; | |
1845 | new.inuse--; | |
a0132ac0 | 1846 | VM_BUG_ON(!new.frozen); |
2cfb7455 | 1847 | |
1d07171c | 1848 | } while (!__cmpxchg_double_slab(s, page, |
2cfb7455 CL |
1849 | prior, counters, |
1850 | freelist, new.counters, | |
1851 | "drain percpu freelist")); | |
1852 | ||
1853 | freelist = nextfree; | |
1854 | } | |
1855 | ||
894b8788 | 1856 | /* |
2cfb7455 CL |
1857 | * Stage two: Ensure that the page is unfrozen while the |
1858 | * list presence reflects the actual number of objects | |
1859 | * during unfreeze. | |
1860 | * | |
1861 | * We setup the list membership and then perform a cmpxchg | |
1862 | * with the count. If there is a mismatch then the page | |
1863 | * is not unfrozen but the page is on the wrong list. | |
1864 | * | |
1865 | * Then we restart the process which may have to remove | |
1866 | * the page from the list that we just put it on again | |
1867 | * because the number of objects in the slab may have | |
1868 | * changed. | |
894b8788 | 1869 | */ |
2cfb7455 | 1870 | redo: |
894b8788 | 1871 | |
2cfb7455 CL |
1872 | old.freelist = page->freelist; |
1873 | old.counters = page->counters; | |
a0132ac0 | 1874 | VM_BUG_ON(!old.frozen); |
7c2e132c | 1875 | |
2cfb7455 CL |
1876 | /* Determine target state of the slab */ |
1877 | new.counters = old.counters; | |
1878 | if (freelist) { | |
1879 | new.inuse--; | |
1880 | set_freepointer(s, freelist, old.freelist); | |
1881 | new.freelist = freelist; | |
1882 | } else | |
1883 | new.freelist = old.freelist; | |
1884 | ||
1885 | new.frozen = 0; | |
1886 | ||
8a5b20ae | 1887 | if (!new.inuse && n->nr_partial >= s->min_partial) |
2cfb7455 CL |
1888 | m = M_FREE; |
1889 | else if (new.freelist) { | |
1890 | m = M_PARTIAL; | |
1891 | if (!lock) { | |
1892 | lock = 1; | |
1893 | /* | |
1894 | * Taking the spinlock removes the possiblity | |
1895 | * that acquire_slab() will see a slab page that | |
1896 | * is frozen | |
1897 | */ | |
1898 | spin_lock(&n->list_lock); | |
1899 | } | |
1900 | } else { | |
1901 | m = M_FULL; | |
1902 | if (kmem_cache_debug(s) && !lock) { | |
1903 | lock = 1; | |
1904 | /* | |
1905 | * This also ensures that the scanning of full | |
1906 | * slabs from diagnostic functions will not see | |
1907 | * any frozen slabs. | |
1908 | */ | |
1909 | spin_lock(&n->list_lock); | |
1910 | } | |
1911 | } | |
1912 | ||
1913 | if (l != m) { | |
1914 | ||
1915 | if (l == M_PARTIAL) | |
1916 | ||
1917 | remove_partial(n, page); | |
1918 | ||
1919 | else if (l == M_FULL) | |
894b8788 | 1920 | |
c65c1877 | 1921 | remove_full(s, n, page); |
2cfb7455 CL |
1922 | |
1923 | if (m == M_PARTIAL) { | |
1924 | ||
1925 | add_partial(n, page, tail); | |
136333d1 | 1926 | stat(s, tail); |
2cfb7455 CL |
1927 | |
1928 | } else if (m == M_FULL) { | |
894b8788 | 1929 | |
2cfb7455 CL |
1930 | stat(s, DEACTIVATE_FULL); |
1931 | add_full(s, n, page); | |
1932 | ||
1933 | } | |
1934 | } | |
1935 | ||
1936 | l = m; | |
1d07171c | 1937 | if (!__cmpxchg_double_slab(s, page, |
2cfb7455 CL |
1938 | old.freelist, old.counters, |
1939 | new.freelist, new.counters, | |
1940 | "unfreezing slab")) | |
1941 | goto redo; | |
1942 | ||
2cfb7455 CL |
1943 | if (lock) |
1944 | spin_unlock(&n->list_lock); | |
1945 | ||
1946 | if (m == M_FREE) { | |
1947 | stat(s, DEACTIVATE_EMPTY); | |
1948 | discard_slab(s, page); | |
1949 | stat(s, FREE_SLAB); | |
894b8788 | 1950 | } |
81819f0f CL |
1951 | } |
1952 | ||
d24ac77f JK |
1953 | /* |
1954 | * Unfreeze all the cpu partial slabs. | |
1955 | * | |
59a09917 CL |
1956 | * This function must be called with interrupts disabled |
1957 | * for the cpu using c (or some other guarantee must be there | |
1958 | * to guarantee no concurrent accesses). | |
d24ac77f | 1959 | */ |
59a09917 CL |
1960 | static void unfreeze_partials(struct kmem_cache *s, |
1961 | struct kmem_cache_cpu *c) | |
49e22585 | 1962 | { |
345c905d | 1963 | #ifdef CONFIG_SLUB_CPU_PARTIAL |
43d77867 | 1964 | struct kmem_cache_node *n = NULL, *n2 = NULL; |
9ada1934 | 1965 | struct page *page, *discard_page = NULL; |
49e22585 CL |
1966 | |
1967 | while ((page = c->partial)) { | |
49e22585 CL |
1968 | struct page new; |
1969 | struct page old; | |
1970 | ||
1971 | c->partial = page->next; | |
43d77867 JK |
1972 | |
1973 | n2 = get_node(s, page_to_nid(page)); | |
1974 | if (n != n2) { | |
1975 | if (n) | |
1976 | spin_unlock(&n->list_lock); | |
1977 | ||
1978 | n = n2; | |
1979 | spin_lock(&n->list_lock); | |
1980 | } | |
49e22585 CL |
1981 | |
1982 | do { | |
1983 | ||
1984 | old.freelist = page->freelist; | |
1985 | old.counters = page->counters; | |
a0132ac0 | 1986 | VM_BUG_ON(!old.frozen); |
49e22585 CL |
1987 | |
1988 | new.counters = old.counters; | |
1989 | new.freelist = old.freelist; | |
1990 | ||
1991 | new.frozen = 0; | |
1992 | ||
d24ac77f | 1993 | } while (!__cmpxchg_double_slab(s, page, |
49e22585 CL |
1994 | old.freelist, old.counters, |
1995 | new.freelist, new.counters, | |
1996 | "unfreezing slab")); | |
1997 | ||
8a5b20ae | 1998 | if (unlikely(!new.inuse && n->nr_partial >= s->min_partial)) { |
9ada1934 SL |
1999 | page->next = discard_page; |
2000 | discard_page = page; | |
43d77867 JK |
2001 | } else { |
2002 | add_partial(n, page, DEACTIVATE_TO_TAIL); | |
2003 | stat(s, FREE_ADD_PARTIAL); | |
49e22585 CL |
2004 | } |
2005 | } | |
2006 | ||
2007 | if (n) | |
2008 | spin_unlock(&n->list_lock); | |
9ada1934 SL |
2009 | |
2010 | while (discard_page) { | |
2011 | page = discard_page; | |
2012 | discard_page = discard_page->next; | |
2013 | ||
2014 | stat(s, DEACTIVATE_EMPTY); | |
2015 | discard_slab(s, page); | |
2016 | stat(s, FREE_SLAB); | |
2017 | } | |
345c905d | 2018 | #endif |
49e22585 CL |
2019 | } |
2020 | ||
2021 | /* | |
2022 | * Put a page that was just frozen (in __slab_free) into a partial page | |
2023 | * slot if available. This is done without interrupts disabled and without | |
2024 | * preemption disabled. The cmpxchg is racy and may put the partial page | |
2025 | * onto a random cpus partial slot. | |
2026 | * | |
2027 | * If we did not find a slot then simply move all the partials to the | |
2028 | * per node partial list. | |
2029 | */ | |
633b0764 | 2030 | static void put_cpu_partial(struct kmem_cache *s, struct page *page, int drain) |
49e22585 | 2031 | { |
345c905d | 2032 | #ifdef CONFIG_SLUB_CPU_PARTIAL |
49e22585 CL |
2033 | struct page *oldpage; |
2034 | int pages; | |
2035 | int pobjects; | |
2036 | ||
d6e0b7fa | 2037 | preempt_disable(); |
49e22585 CL |
2038 | do { |
2039 | pages = 0; | |
2040 | pobjects = 0; | |
2041 | oldpage = this_cpu_read(s->cpu_slab->partial); | |
2042 | ||
2043 | if (oldpage) { | |
2044 | pobjects = oldpage->pobjects; | |
2045 | pages = oldpage->pages; | |
2046 | if (drain && pobjects > s->cpu_partial) { | |
2047 | unsigned long flags; | |
2048 | /* | |
2049 | * partial array is full. Move the existing | |
2050 | * set to the per node partial list. | |
2051 | */ | |
2052 | local_irq_save(flags); | |
59a09917 | 2053 | unfreeze_partials(s, this_cpu_ptr(s->cpu_slab)); |
49e22585 | 2054 | local_irq_restore(flags); |
e24fc410 | 2055 | oldpage = NULL; |
49e22585 CL |
2056 | pobjects = 0; |
2057 | pages = 0; | |
8028dcea | 2058 | stat(s, CPU_PARTIAL_DRAIN); |
49e22585 CL |
2059 | } |
2060 | } | |
2061 | ||
2062 | pages++; | |
2063 | pobjects += page->objects - page->inuse; | |
2064 | ||
2065 | page->pages = pages; | |
2066 | page->pobjects = pobjects; | |
2067 | page->next = oldpage; | |
2068 | ||
d0e0ac97 CG |
2069 | } while (this_cpu_cmpxchg(s->cpu_slab->partial, oldpage, page) |
2070 | != oldpage); | |
d6e0b7fa VD |
2071 | if (unlikely(!s->cpu_partial)) { |
2072 | unsigned long flags; | |
2073 | ||
2074 | local_irq_save(flags); | |
2075 | unfreeze_partials(s, this_cpu_ptr(s->cpu_slab)); | |
2076 | local_irq_restore(flags); | |
2077 | } | |
2078 | preempt_enable(); | |
345c905d | 2079 | #endif |
49e22585 CL |
2080 | } |
2081 | ||
dfb4f096 | 2082 | static inline void flush_slab(struct kmem_cache *s, struct kmem_cache_cpu *c) |
81819f0f | 2083 | { |
84e554e6 | 2084 | stat(s, CPUSLAB_FLUSH); |
c17dda40 CL |
2085 | deactivate_slab(s, c->page, c->freelist); |
2086 | ||
2087 | c->tid = next_tid(c->tid); | |
2088 | c->page = NULL; | |
2089 | c->freelist = NULL; | |
81819f0f CL |
2090 | } |
2091 | ||
2092 | /* | |
2093 | * Flush cpu slab. | |
6446faa2 | 2094 | * |
81819f0f CL |
2095 | * Called from IPI handler with interrupts disabled. |
2096 | */ | |
0c710013 | 2097 | static inline void __flush_cpu_slab(struct kmem_cache *s, int cpu) |
81819f0f | 2098 | { |
9dfc6e68 | 2099 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu); |
81819f0f | 2100 | |
49e22585 CL |
2101 | if (likely(c)) { |
2102 | if (c->page) | |
2103 | flush_slab(s, c); | |
2104 | ||
59a09917 | 2105 | unfreeze_partials(s, c); |
49e22585 | 2106 | } |
81819f0f CL |
2107 | } |
2108 | ||
2109 | static void flush_cpu_slab(void *d) | |
2110 | { | |
2111 | struct kmem_cache *s = d; | |
81819f0f | 2112 | |
dfb4f096 | 2113 | __flush_cpu_slab(s, smp_processor_id()); |
81819f0f CL |
2114 | } |
2115 | ||
a8364d55 GBY |
2116 | static bool has_cpu_slab(int cpu, void *info) |
2117 | { | |
2118 | struct kmem_cache *s = info; | |
2119 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu); | |
2120 | ||
02e1a9cd | 2121 | return c->page || c->partial; |
a8364d55 GBY |
2122 | } |
2123 | ||
81819f0f CL |
2124 | static void flush_all(struct kmem_cache *s) |
2125 | { | |
a8364d55 | 2126 | on_each_cpu_cond(has_cpu_slab, flush_cpu_slab, s, 1, GFP_ATOMIC); |
81819f0f CL |
2127 | } |
2128 | ||
dfb4f096 CL |
2129 | /* |
2130 | * Check if the objects in a per cpu structure fit numa | |
2131 | * locality expectations. | |
2132 | */ | |
57d437d2 | 2133 | static inline int node_match(struct page *page, int node) |
dfb4f096 CL |
2134 | { |
2135 | #ifdef CONFIG_NUMA | |
4d7868e6 | 2136 | if (!page || (node != NUMA_NO_NODE && page_to_nid(page) != node)) |
dfb4f096 CL |
2137 | return 0; |
2138 | #endif | |
2139 | return 1; | |
2140 | } | |
2141 | ||
9a02d699 | 2142 | #ifdef CONFIG_SLUB_DEBUG |
781b2ba6 PE |
2143 | static int count_free(struct page *page) |
2144 | { | |
2145 | return page->objects - page->inuse; | |
2146 | } | |
2147 | ||
9a02d699 DR |
2148 | static inline unsigned long node_nr_objs(struct kmem_cache_node *n) |
2149 | { | |
2150 | return atomic_long_read(&n->total_objects); | |
2151 | } | |
2152 | #endif /* CONFIG_SLUB_DEBUG */ | |
2153 | ||
2154 | #if defined(CONFIG_SLUB_DEBUG) || defined(CONFIG_SYSFS) | |
781b2ba6 PE |
2155 | static unsigned long count_partial(struct kmem_cache_node *n, |
2156 | int (*get_count)(struct page *)) | |
2157 | { | |
2158 | unsigned long flags; | |
2159 | unsigned long x = 0; | |
2160 | struct page *page; | |
2161 | ||
2162 | spin_lock_irqsave(&n->list_lock, flags); | |
2163 | list_for_each_entry(page, &n->partial, lru) | |
2164 | x += get_count(page); | |
2165 | spin_unlock_irqrestore(&n->list_lock, flags); | |
2166 | return x; | |
2167 | } | |
9a02d699 | 2168 | #endif /* CONFIG_SLUB_DEBUG || CONFIG_SYSFS */ |
26c02cf0 | 2169 | |
781b2ba6 PE |
2170 | static noinline void |
2171 | slab_out_of_memory(struct kmem_cache *s, gfp_t gfpflags, int nid) | |
2172 | { | |
9a02d699 DR |
2173 | #ifdef CONFIG_SLUB_DEBUG |
2174 | static DEFINE_RATELIMIT_STATE(slub_oom_rs, DEFAULT_RATELIMIT_INTERVAL, | |
2175 | DEFAULT_RATELIMIT_BURST); | |
781b2ba6 | 2176 | int node; |
fa45dc25 | 2177 | struct kmem_cache_node *n; |
781b2ba6 | 2178 | |
9a02d699 DR |
2179 | if ((gfpflags & __GFP_NOWARN) || !__ratelimit(&slub_oom_rs)) |
2180 | return; | |
2181 | ||
f9f58285 | 2182 | pr_warn("SLUB: Unable to allocate memory on node %d (gfp=0x%x)\n", |
781b2ba6 | 2183 | nid, gfpflags); |
f9f58285 FF |
2184 | pr_warn(" cache: %s, object size: %d, buffer size: %d, default order: %d, min order: %d\n", |
2185 | s->name, s->object_size, s->size, oo_order(s->oo), | |
2186 | oo_order(s->min)); | |
781b2ba6 | 2187 | |
3b0efdfa | 2188 | if (oo_order(s->min) > get_order(s->object_size)) |
f9f58285 FF |
2189 | pr_warn(" %s debugging increased min order, use slub_debug=O to disable.\n", |
2190 | s->name); | |
fa5ec8a1 | 2191 | |
fa45dc25 | 2192 | for_each_kmem_cache_node(s, node, n) { |
781b2ba6 PE |
2193 | unsigned long nr_slabs; |
2194 | unsigned long nr_objs; | |
2195 | unsigned long nr_free; | |
2196 | ||
26c02cf0 AB |
2197 | nr_free = count_partial(n, count_free); |
2198 | nr_slabs = node_nr_slabs(n); | |
2199 | nr_objs = node_nr_objs(n); | |
781b2ba6 | 2200 | |
f9f58285 | 2201 | pr_warn(" node %d: slabs: %ld, objs: %ld, free: %ld\n", |
781b2ba6 PE |
2202 | node, nr_slabs, nr_objs, nr_free); |
2203 | } | |
9a02d699 | 2204 | #endif |
781b2ba6 PE |
2205 | } |
2206 | ||
497b66f2 CL |
2207 | static inline void *new_slab_objects(struct kmem_cache *s, gfp_t flags, |
2208 | int node, struct kmem_cache_cpu **pc) | |
2209 | { | |
6faa6833 | 2210 | void *freelist; |
188fd063 CL |
2211 | struct kmem_cache_cpu *c = *pc; |
2212 | struct page *page; | |
497b66f2 | 2213 | |
188fd063 | 2214 | freelist = get_partial(s, flags, node, c); |
497b66f2 | 2215 | |
188fd063 CL |
2216 | if (freelist) |
2217 | return freelist; | |
2218 | ||
2219 | page = new_slab(s, flags, node); | |
497b66f2 | 2220 | if (page) { |
7c8e0181 | 2221 | c = raw_cpu_ptr(s->cpu_slab); |
497b66f2 CL |
2222 | if (c->page) |
2223 | flush_slab(s, c); | |
2224 | ||
2225 | /* | |
2226 | * No other reference to the page yet so we can | |
2227 | * muck around with it freely without cmpxchg | |
2228 | */ | |
6faa6833 | 2229 | freelist = page->freelist; |
497b66f2 CL |
2230 | page->freelist = NULL; |
2231 | ||
2232 | stat(s, ALLOC_SLAB); | |
497b66f2 CL |
2233 | c->page = page; |
2234 | *pc = c; | |
2235 | } else | |
6faa6833 | 2236 | freelist = NULL; |
497b66f2 | 2237 | |
6faa6833 | 2238 | return freelist; |
497b66f2 CL |
2239 | } |
2240 | ||
072bb0aa MG |
2241 | static inline bool pfmemalloc_match(struct page *page, gfp_t gfpflags) |
2242 | { | |
2243 | if (unlikely(PageSlabPfmemalloc(page))) | |
2244 | return gfp_pfmemalloc_allowed(gfpflags); | |
2245 | ||
2246 | return true; | |
2247 | } | |
2248 | ||
213eeb9f | 2249 | /* |
d0e0ac97 CG |
2250 | * Check the page->freelist of a page and either transfer the freelist to the |
2251 | * per cpu freelist or deactivate the page. | |
213eeb9f CL |
2252 | * |
2253 | * The page is still frozen if the return value is not NULL. | |
2254 | * | |
2255 | * If this function returns NULL then the page has been unfrozen. | |
d24ac77f JK |
2256 | * |
2257 | * This function must be called with interrupt disabled. | |
213eeb9f CL |
2258 | */ |
2259 | static inline void *get_freelist(struct kmem_cache *s, struct page *page) | |
2260 | { | |
2261 | struct page new; | |
2262 | unsigned long counters; | |
2263 | void *freelist; | |
2264 | ||
2265 | do { | |
2266 | freelist = page->freelist; | |
2267 | counters = page->counters; | |
6faa6833 | 2268 | |
213eeb9f | 2269 | new.counters = counters; |
a0132ac0 | 2270 | VM_BUG_ON(!new.frozen); |
213eeb9f CL |
2271 | |
2272 | new.inuse = page->objects; | |
2273 | new.frozen = freelist != NULL; | |
2274 | ||
d24ac77f | 2275 | } while (!__cmpxchg_double_slab(s, page, |
213eeb9f CL |
2276 | freelist, counters, |
2277 | NULL, new.counters, | |
2278 | "get_freelist")); | |
2279 | ||
2280 | return freelist; | |
2281 | } | |
2282 | ||
81819f0f | 2283 | /* |
894b8788 CL |
2284 | * Slow path. The lockless freelist is empty or we need to perform |
2285 | * debugging duties. | |
2286 | * | |
894b8788 CL |
2287 | * Processing is still very fast if new objects have been freed to the |
2288 | * regular freelist. In that case we simply take over the regular freelist | |
2289 | * as the lockless freelist and zap the regular freelist. | |
81819f0f | 2290 | * |
894b8788 CL |
2291 | * If that is not working then we fall back to the partial lists. We take the |
2292 | * first element of the freelist as the object to allocate now and move the | |
2293 | * rest of the freelist to the lockless freelist. | |
81819f0f | 2294 | * |
894b8788 | 2295 | * And if we were unable to get a new slab from the partial slab lists then |
6446faa2 CL |
2296 | * we need to allocate a new slab. This is the slowest path since it involves |
2297 | * a call to the page allocator and the setup of a new slab. | |
81819f0f | 2298 | */ |
ce71e27c EGM |
2299 | static void *__slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node, |
2300 | unsigned long addr, struct kmem_cache_cpu *c) | |
81819f0f | 2301 | { |
6faa6833 | 2302 | void *freelist; |
f6e7def7 | 2303 | struct page *page; |
8a5ec0ba CL |
2304 | unsigned long flags; |
2305 | ||
2306 | local_irq_save(flags); | |
2307 | #ifdef CONFIG_PREEMPT | |
2308 | /* | |
2309 | * We may have been preempted and rescheduled on a different | |
2310 | * cpu before disabling interrupts. Need to reload cpu area | |
2311 | * pointer. | |
2312 | */ | |
2313 | c = this_cpu_ptr(s->cpu_slab); | |
8a5ec0ba | 2314 | #endif |
81819f0f | 2315 | |
f6e7def7 CL |
2316 | page = c->page; |
2317 | if (!page) | |
81819f0f | 2318 | goto new_slab; |
49e22585 | 2319 | redo: |
6faa6833 | 2320 | |
57d437d2 | 2321 | if (unlikely(!node_match(page, node))) { |
a561ce00 JK |
2322 | int searchnode = node; |
2323 | ||
2324 | if (node != NUMA_NO_NODE && !node_present_pages(node)) | |
2325 | searchnode = node_to_mem_node(node); | |
2326 | ||
2327 | if (unlikely(!node_match(page, searchnode))) { | |
2328 | stat(s, ALLOC_NODE_MISMATCH); | |
2329 | deactivate_slab(s, page, c->freelist); | |
2330 | c->page = NULL; | |
2331 | c->freelist = NULL; | |
2332 | goto new_slab; | |
2333 | } | |
fc59c053 | 2334 | } |
6446faa2 | 2335 | |
072bb0aa MG |
2336 | /* |
2337 | * By rights, we should be searching for a slab page that was | |
2338 | * PFMEMALLOC but right now, we are losing the pfmemalloc | |
2339 | * information when the page leaves the per-cpu allocator | |
2340 | */ | |
2341 | if (unlikely(!pfmemalloc_match(page, gfpflags))) { | |
2342 | deactivate_slab(s, page, c->freelist); | |
2343 | c->page = NULL; | |
2344 | c->freelist = NULL; | |
2345 | goto new_slab; | |
2346 | } | |
2347 | ||
73736e03 | 2348 | /* must check again c->freelist in case of cpu migration or IRQ */ |
6faa6833 CL |
2349 | freelist = c->freelist; |
2350 | if (freelist) | |
73736e03 | 2351 | goto load_freelist; |
03e404af | 2352 | |
f6e7def7 | 2353 | freelist = get_freelist(s, page); |
6446faa2 | 2354 | |
6faa6833 | 2355 | if (!freelist) { |
03e404af CL |
2356 | c->page = NULL; |
2357 | stat(s, DEACTIVATE_BYPASS); | |
fc59c053 | 2358 | goto new_slab; |
03e404af | 2359 | } |
6446faa2 | 2360 | |
84e554e6 | 2361 | stat(s, ALLOC_REFILL); |
6446faa2 | 2362 | |
894b8788 | 2363 | load_freelist: |
507effea CL |
2364 | /* |
2365 | * freelist is pointing to the list of objects to be used. | |
2366 | * page is pointing to the page from which the objects are obtained. | |
2367 | * That page must be frozen for per cpu allocations to work. | |
2368 | */ | |
a0132ac0 | 2369 | VM_BUG_ON(!c->page->frozen); |
6faa6833 | 2370 | c->freelist = get_freepointer(s, freelist); |
8a5ec0ba CL |
2371 | c->tid = next_tid(c->tid); |
2372 | local_irq_restore(flags); | |
6faa6833 | 2373 | return freelist; |
81819f0f | 2374 | |
81819f0f | 2375 | new_slab: |
2cfb7455 | 2376 | |
49e22585 | 2377 | if (c->partial) { |
f6e7def7 CL |
2378 | page = c->page = c->partial; |
2379 | c->partial = page->next; | |
49e22585 CL |
2380 | stat(s, CPU_PARTIAL_ALLOC); |
2381 | c->freelist = NULL; | |
2382 | goto redo; | |
81819f0f CL |
2383 | } |
2384 | ||
188fd063 | 2385 | freelist = new_slab_objects(s, gfpflags, node, &c); |
01ad8a7b | 2386 | |
f4697436 | 2387 | if (unlikely(!freelist)) { |
9a02d699 | 2388 | slab_out_of_memory(s, gfpflags, node); |
f4697436 CL |
2389 | local_irq_restore(flags); |
2390 | return NULL; | |
81819f0f | 2391 | } |
2cfb7455 | 2392 | |
f6e7def7 | 2393 | page = c->page; |
5091b74a | 2394 | if (likely(!kmem_cache_debug(s) && pfmemalloc_match(page, gfpflags))) |
4b6f0750 | 2395 | goto load_freelist; |
2cfb7455 | 2396 | |
497b66f2 | 2397 | /* Only entered in the debug case */ |
d0e0ac97 CG |
2398 | if (kmem_cache_debug(s) && |
2399 | !alloc_debug_processing(s, page, freelist, addr)) | |
497b66f2 | 2400 | goto new_slab; /* Slab failed checks. Next slab needed */ |
894b8788 | 2401 | |
f6e7def7 | 2402 | deactivate_slab(s, page, get_freepointer(s, freelist)); |
c17dda40 CL |
2403 | c->page = NULL; |
2404 | c->freelist = NULL; | |
a71ae47a | 2405 | local_irq_restore(flags); |
6faa6833 | 2406 | return freelist; |
894b8788 CL |
2407 | } |
2408 | ||
2409 | /* | |
2410 | * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc) | |
2411 | * have the fastpath folded into their functions. So no function call | |
2412 | * overhead for requests that can be satisfied on the fastpath. | |
2413 | * | |
2414 | * The fastpath works by first checking if the lockless freelist can be used. | |
2415 | * If not then __slab_alloc is called for slow processing. | |
2416 | * | |
2417 | * Otherwise we can simply pick the next object from the lockless free list. | |
2418 | */ | |
2b847c3c | 2419 | static __always_inline void *slab_alloc_node(struct kmem_cache *s, |
ce71e27c | 2420 | gfp_t gfpflags, int node, unsigned long addr) |
894b8788 | 2421 | { |
894b8788 | 2422 | void **object; |
dfb4f096 | 2423 | struct kmem_cache_cpu *c; |
57d437d2 | 2424 | struct page *page; |
8a5ec0ba | 2425 | unsigned long tid; |
1f84260c | 2426 | |
8135be5a VD |
2427 | s = slab_pre_alloc_hook(s, gfpflags); |
2428 | if (!s) | |
773ff60e | 2429 | return NULL; |
8a5ec0ba | 2430 | redo: |
8a5ec0ba CL |
2431 | /* |
2432 | * Must read kmem_cache cpu data via this cpu ptr. Preemption is | |
2433 | * enabled. We may switch back and forth between cpus while | |
2434 | * reading from one cpu area. That does not matter as long | |
2435 | * as we end up on the original cpu again when doing the cmpxchg. | |
7cccd80b | 2436 | * |
9aabf810 JK |
2437 | * We should guarantee that tid and kmem_cache are retrieved on |
2438 | * the same cpu. It could be different if CONFIG_PREEMPT so we need | |
2439 | * to check if it is matched or not. | |
8a5ec0ba | 2440 | */ |
9aabf810 JK |
2441 | do { |
2442 | tid = this_cpu_read(s->cpu_slab->tid); | |
2443 | c = raw_cpu_ptr(s->cpu_slab); | |
859b7a0e MR |
2444 | } while (IS_ENABLED(CONFIG_PREEMPT) && |
2445 | unlikely(tid != READ_ONCE(c->tid))); | |
9aabf810 JK |
2446 | |
2447 | /* | |
2448 | * Irqless object alloc/free algorithm used here depends on sequence | |
2449 | * of fetching cpu_slab's data. tid should be fetched before anything | |
2450 | * on c to guarantee that object and page associated with previous tid | |
2451 | * won't be used with current tid. If we fetch tid first, object and | |
2452 | * page could be one associated with next tid and our alloc/free | |
2453 | * request will be failed. In this case, we will retry. So, no problem. | |
2454 | */ | |
2455 | barrier(); | |
8a5ec0ba | 2456 | |
8a5ec0ba CL |
2457 | /* |
2458 | * The transaction ids are globally unique per cpu and per operation on | |
2459 | * a per cpu queue. Thus they can be guarantee that the cmpxchg_double | |
2460 | * occurs on the right processor and that there was no operation on the | |
2461 | * linked list in between. | |
2462 | */ | |
8a5ec0ba | 2463 | |
9dfc6e68 | 2464 | object = c->freelist; |
57d437d2 | 2465 | page = c->page; |
8eae1492 | 2466 | if (unlikely(!object || !node_match(page, node))) { |
dfb4f096 | 2467 | object = __slab_alloc(s, gfpflags, node, addr, c); |
8eae1492 DH |
2468 | stat(s, ALLOC_SLOWPATH); |
2469 | } else { | |
0ad9500e ED |
2470 | void *next_object = get_freepointer_safe(s, object); |
2471 | ||
8a5ec0ba | 2472 | /* |
25985edc | 2473 | * The cmpxchg will only match if there was no additional |
8a5ec0ba CL |
2474 | * operation and if we are on the right processor. |
2475 | * | |
d0e0ac97 CG |
2476 | * The cmpxchg does the following atomically (without lock |
2477 | * semantics!) | |
8a5ec0ba CL |
2478 | * 1. Relocate first pointer to the current per cpu area. |
2479 | * 2. Verify that tid and freelist have not been changed | |
2480 | * 3. If they were not changed replace tid and freelist | |
2481 | * | |
d0e0ac97 CG |
2482 | * Since this is without lock semantics the protection is only |
2483 | * against code executing on this cpu *not* from access by | |
2484 | * other cpus. | |
8a5ec0ba | 2485 | */ |
933393f5 | 2486 | if (unlikely(!this_cpu_cmpxchg_double( |
8a5ec0ba CL |
2487 | s->cpu_slab->freelist, s->cpu_slab->tid, |
2488 | object, tid, | |
0ad9500e | 2489 | next_object, next_tid(tid)))) { |
8a5ec0ba CL |
2490 | |
2491 | note_cmpxchg_failure("slab_alloc", s, tid); | |
2492 | goto redo; | |
2493 | } | |
0ad9500e | 2494 | prefetch_freepointer(s, next_object); |
84e554e6 | 2495 | stat(s, ALLOC_FASTPATH); |
894b8788 | 2496 | } |
8a5ec0ba | 2497 | |
74e2134f | 2498 | if (unlikely(gfpflags & __GFP_ZERO) && object) |
3b0efdfa | 2499 | memset(object, 0, s->object_size); |
d07dbea4 | 2500 | |
c016b0bd | 2501 | slab_post_alloc_hook(s, gfpflags, object); |
5a896d9e | 2502 | |
894b8788 | 2503 | return object; |
81819f0f CL |
2504 | } |
2505 | ||
2b847c3c EG |
2506 | static __always_inline void *slab_alloc(struct kmem_cache *s, |
2507 | gfp_t gfpflags, unsigned long addr) | |
2508 | { | |
2509 | return slab_alloc_node(s, gfpflags, NUMA_NO_NODE, addr); | |
2510 | } | |
2511 | ||
81819f0f CL |
2512 | void *kmem_cache_alloc(struct kmem_cache *s, gfp_t gfpflags) |
2513 | { | |
2b847c3c | 2514 | void *ret = slab_alloc(s, gfpflags, _RET_IP_); |
5b882be4 | 2515 | |
d0e0ac97 CG |
2516 | trace_kmem_cache_alloc(_RET_IP_, ret, s->object_size, |
2517 | s->size, gfpflags); | |
5b882be4 EGM |
2518 | |
2519 | return ret; | |
81819f0f CL |
2520 | } |
2521 | EXPORT_SYMBOL(kmem_cache_alloc); | |
2522 | ||
0f24f128 | 2523 | #ifdef CONFIG_TRACING |
4a92379b RK |
2524 | void *kmem_cache_alloc_trace(struct kmem_cache *s, gfp_t gfpflags, size_t size) |
2525 | { | |
2b847c3c | 2526 | void *ret = slab_alloc(s, gfpflags, _RET_IP_); |
4a92379b | 2527 | trace_kmalloc(_RET_IP_, ret, size, s->size, gfpflags); |
0316bec2 | 2528 | kasan_kmalloc(s, ret, size); |
4a92379b RK |
2529 | return ret; |
2530 | } | |
2531 | EXPORT_SYMBOL(kmem_cache_alloc_trace); | |
5b882be4 EGM |
2532 | #endif |
2533 | ||
81819f0f CL |
2534 | #ifdef CONFIG_NUMA |
2535 | void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t gfpflags, int node) | |
2536 | { | |
2b847c3c | 2537 | void *ret = slab_alloc_node(s, gfpflags, node, _RET_IP_); |
5b882be4 | 2538 | |
ca2b84cb | 2539 | trace_kmem_cache_alloc_node(_RET_IP_, ret, |
3b0efdfa | 2540 | s->object_size, s->size, gfpflags, node); |
5b882be4 EGM |
2541 | |
2542 | return ret; | |
81819f0f CL |
2543 | } |
2544 | EXPORT_SYMBOL(kmem_cache_alloc_node); | |
81819f0f | 2545 | |
0f24f128 | 2546 | #ifdef CONFIG_TRACING |
4a92379b | 2547 | void *kmem_cache_alloc_node_trace(struct kmem_cache *s, |
5b882be4 | 2548 | gfp_t gfpflags, |
4a92379b | 2549 | int node, size_t size) |
5b882be4 | 2550 | { |
2b847c3c | 2551 | void *ret = slab_alloc_node(s, gfpflags, node, _RET_IP_); |
4a92379b RK |
2552 | |
2553 | trace_kmalloc_node(_RET_IP_, ret, | |
2554 | size, s->size, gfpflags, node); | |
0316bec2 AR |
2555 | |
2556 | kasan_kmalloc(s, ret, size); | |
4a92379b | 2557 | return ret; |
5b882be4 | 2558 | } |
4a92379b | 2559 | EXPORT_SYMBOL(kmem_cache_alloc_node_trace); |
5b882be4 | 2560 | #endif |
5d1f57e4 | 2561 | #endif |
5b882be4 | 2562 | |
81819f0f | 2563 | /* |
94e4d712 | 2564 | * Slow path handling. This may still be called frequently since objects |
894b8788 | 2565 | * have a longer lifetime than the cpu slabs in most processing loads. |
81819f0f | 2566 | * |
894b8788 CL |
2567 | * So we still attempt to reduce cache line usage. Just take the slab |
2568 | * lock and free the item. If there is no additional partial page | |
2569 | * handling required then we can return immediately. | |
81819f0f | 2570 | */ |
894b8788 | 2571 | static void __slab_free(struct kmem_cache *s, struct page *page, |
ff12059e | 2572 | void *x, unsigned long addr) |
81819f0f CL |
2573 | { |
2574 | void *prior; | |
2575 | void **object = (void *)x; | |
2cfb7455 | 2576 | int was_frozen; |
2cfb7455 CL |
2577 | struct page new; |
2578 | unsigned long counters; | |
2579 | struct kmem_cache_node *n = NULL; | |
61728d1e | 2580 | unsigned long uninitialized_var(flags); |
81819f0f | 2581 | |
8a5ec0ba | 2582 | stat(s, FREE_SLOWPATH); |
81819f0f | 2583 | |
19c7ff9e CL |
2584 | if (kmem_cache_debug(s) && |
2585 | !(n = free_debug_processing(s, page, x, addr, &flags))) | |
80f08c19 | 2586 | return; |
6446faa2 | 2587 | |
2cfb7455 | 2588 | do { |
837d678d JK |
2589 | if (unlikely(n)) { |
2590 | spin_unlock_irqrestore(&n->list_lock, flags); | |
2591 | n = NULL; | |
2592 | } | |
2cfb7455 CL |
2593 | prior = page->freelist; |
2594 | counters = page->counters; | |
2595 | set_freepointer(s, object, prior); | |
2596 | new.counters = counters; | |
2597 | was_frozen = new.frozen; | |
2598 | new.inuse--; | |
837d678d | 2599 | if ((!new.inuse || !prior) && !was_frozen) { |
49e22585 | 2600 | |
c65c1877 | 2601 | if (kmem_cache_has_cpu_partial(s) && !prior) { |
49e22585 CL |
2602 | |
2603 | /* | |
d0e0ac97 CG |
2604 | * Slab was on no list before and will be |
2605 | * partially empty | |
2606 | * We can defer the list move and instead | |
2607 | * freeze it. | |
49e22585 CL |
2608 | */ |
2609 | new.frozen = 1; | |
2610 | ||
c65c1877 | 2611 | } else { /* Needs to be taken off a list */ |
49e22585 | 2612 | |
b455def2 | 2613 | n = get_node(s, page_to_nid(page)); |
49e22585 CL |
2614 | /* |
2615 | * Speculatively acquire the list_lock. | |
2616 | * If the cmpxchg does not succeed then we may | |
2617 | * drop the list_lock without any processing. | |
2618 | * | |
2619 | * Otherwise the list_lock will synchronize with | |
2620 | * other processors updating the list of slabs. | |
2621 | */ | |
2622 | spin_lock_irqsave(&n->list_lock, flags); | |
2623 | ||
2624 | } | |
2cfb7455 | 2625 | } |
81819f0f | 2626 | |
2cfb7455 CL |
2627 | } while (!cmpxchg_double_slab(s, page, |
2628 | prior, counters, | |
2629 | object, new.counters, | |
2630 | "__slab_free")); | |
81819f0f | 2631 | |
2cfb7455 | 2632 | if (likely(!n)) { |
49e22585 CL |
2633 | |
2634 | /* | |
2635 | * If we just froze the page then put it onto the | |
2636 | * per cpu partial list. | |
2637 | */ | |
8028dcea | 2638 | if (new.frozen && !was_frozen) { |
49e22585 | 2639 | put_cpu_partial(s, page, 1); |
8028dcea AS |
2640 | stat(s, CPU_PARTIAL_FREE); |
2641 | } | |
49e22585 | 2642 | /* |
2cfb7455 CL |
2643 | * The list lock was not taken therefore no list |
2644 | * activity can be necessary. | |
2645 | */ | |
b455def2 L |
2646 | if (was_frozen) |
2647 | stat(s, FREE_FROZEN); | |
2648 | return; | |
2649 | } | |
81819f0f | 2650 | |
8a5b20ae | 2651 | if (unlikely(!new.inuse && n->nr_partial >= s->min_partial)) |
837d678d JK |
2652 | goto slab_empty; |
2653 | ||
81819f0f | 2654 | /* |
837d678d JK |
2655 | * Objects left in the slab. If it was not on the partial list before |
2656 | * then add it. | |
81819f0f | 2657 | */ |
345c905d JK |
2658 | if (!kmem_cache_has_cpu_partial(s) && unlikely(!prior)) { |
2659 | if (kmem_cache_debug(s)) | |
c65c1877 | 2660 | remove_full(s, n, page); |
837d678d JK |
2661 | add_partial(n, page, DEACTIVATE_TO_TAIL); |
2662 | stat(s, FREE_ADD_PARTIAL); | |
8ff12cfc | 2663 | } |
80f08c19 | 2664 | spin_unlock_irqrestore(&n->list_lock, flags); |
81819f0f CL |
2665 | return; |
2666 | ||
2667 | slab_empty: | |
a973e9dd | 2668 | if (prior) { |
81819f0f | 2669 | /* |
6fbabb20 | 2670 | * Slab on the partial list. |
81819f0f | 2671 | */ |
5cc6eee8 | 2672 | remove_partial(n, page); |
84e554e6 | 2673 | stat(s, FREE_REMOVE_PARTIAL); |
c65c1877 | 2674 | } else { |
6fbabb20 | 2675 | /* Slab must be on the full list */ |
c65c1877 PZ |
2676 | remove_full(s, n, page); |
2677 | } | |
2cfb7455 | 2678 | |
80f08c19 | 2679 | spin_unlock_irqrestore(&n->list_lock, flags); |
84e554e6 | 2680 | stat(s, FREE_SLAB); |
81819f0f | 2681 | discard_slab(s, page); |
81819f0f CL |
2682 | } |
2683 | ||
894b8788 CL |
2684 | /* |
2685 | * Fastpath with forced inlining to produce a kfree and kmem_cache_free that | |
2686 | * can perform fastpath freeing without additional function calls. | |
2687 | * | |
2688 | * The fastpath is only possible if we are freeing to the current cpu slab | |
2689 | * of this processor. This typically the case if we have just allocated | |
2690 | * the item before. | |
2691 | * | |
2692 | * If fastpath is not possible then fall back to __slab_free where we deal | |
2693 | * with all sorts of special processing. | |
2694 | */ | |
06428780 | 2695 | static __always_inline void slab_free(struct kmem_cache *s, |
ce71e27c | 2696 | struct page *page, void *x, unsigned long addr) |
894b8788 CL |
2697 | { |
2698 | void **object = (void *)x; | |
dfb4f096 | 2699 | struct kmem_cache_cpu *c; |
8a5ec0ba | 2700 | unsigned long tid; |
1f84260c | 2701 | |
c016b0bd CL |
2702 | slab_free_hook(s, x); |
2703 | ||
8a5ec0ba CL |
2704 | redo: |
2705 | /* | |
2706 | * Determine the currently cpus per cpu slab. | |
2707 | * The cpu may change afterward. However that does not matter since | |
2708 | * data is retrieved via this pointer. If we are on the same cpu | |
2ae44005 | 2709 | * during the cmpxchg then the free will succeed. |
8a5ec0ba | 2710 | */ |
9aabf810 JK |
2711 | do { |
2712 | tid = this_cpu_read(s->cpu_slab->tid); | |
2713 | c = raw_cpu_ptr(s->cpu_slab); | |
859b7a0e MR |
2714 | } while (IS_ENABLED(CONFIG_PREEMPT) && |
2715 | unlikely(tid != READ_ONCE(c->tid))); | |
c016b0bd | 2716 | |
9aabf810 JK |
2717 | /* Same with comment on barrier() in slab_alloc_node() */ |
2718 | barrier(); | |
c016b0bd | 2719 | |
442b06bc | 2720 | if (likely(page == c->page)) { |
ff12059e | 2721 | set_freepointer(s, object, c->freelist); |
8a5ec0ba | 2722 | |
933393f5 | 2723 | if (unlikely(!this_cpu_cmpxchg_double( |
8a5ec0ba CL |
2724 | s->cpu_slab->freelist, s->cpu_slab->tid, |
2725 | c->freelist, tid, | |
2726 | object, next_tid(tid)))) { | |
2727 | ||
2728 | note_cmpxchg_failure("slab_free", s, tid); | |
2729 | goto redo; | |
2730 | } | |
84e554e6 | 2731 | stat(s, FREE_FASTPATH); |
894b8788 | 2732 | } else |
ff12059e | 2733 | __slab_free(s, page, x, addr); |
894b8788 | 2734 | |
894b8788 CL |
2735 | } |
2736 | ||
81819f0f CL |
2737 | void kmem_cache_free(struct kmem_cache *s, void *x) |
2738 | { | |
b9ce5ef4 GC |
2739 | s = cache_from_obj(s, x); |
2740 | if (!s) | |
79576102 | 2741 | return; |
b9ce5ef4 | 2742 | slab_free(s, virt_to_head_page(x), x, _RET_IP_); |
ca2b84cb | 2743 | trace_kmem_cache_free(_RET_IP_, x); |
81819f0f CL |
2744 | } |
2745 | EXPORT_SYMBOL(kmem_cache_free); | |
2746 | ||
994eb764 | 2747 | /* Note that interrupts must be enabled when calling this function. */ |
484748f0 CL |
2748 | void kmem_cache_free_bulk(struct kmem_cache *s, size_t size, void **p) |
2749 | { | |
fbd02630 JDB |
2750 | struct kmem_cache_cpu *c; |
2751 | struct page *page; | |
2752 | int i; | |
2753 | ||
fbd02630 JDB |
2754 | local_irq_disable(); |
2755 | c = this_cpu_ptr(s->cpu_slab); | |
2756 | ||
2757 | for (i = 0; i < size; i++) { | |
2758 | void *object = p[i]; | |
2759 | ||
2760 | BUG_ON(!object); | |
3eed034d JDB |
2761 | /* kmem cache debug support */ |
2762 | s = cache_from_obj(s, object); | |
2763 | if (unlikely(!s)) | |
2764 | goto exit; | |
2765 | slab_free_hook(s, object); | |
2766 | ||
fbd02630 | 2767 | page = virt_to_head_page(object); |
fbd02630 JDB |
2768 | |
2769 | if (c->page == page) { | |
2770 | /* Fastpath: local CPU free */ | |
2771 | set_freepointer(s, object, c->freelist); | |
2772 | c->freelist = object; | |
2773 | } else { | |
2774 | c->tid = next_tid(c->tid); | |
2775 | local_irq_enable(); | |
2776 | /* Slowpath: overhead locked cmpxchg_double_slab */ | |
2777 | __slab_free(s, page, object, _RET_IP_); | |
2778 | local_irq_disable(); | |
2779 | c = this_cpu_ptr(s->cpu_slab); | |
2780 | } | |
2781 | } | |
3eed034d | 2782 | exit: |
fbd02630 JDB |
2783 | c->tid = next_tid(c->tid); |
2784 | local_irq_enable(); | |
484748f0 CL |
2785 | } |
2786 | EXPORT_SYMBOL(kmem_cache_free_bulk); | |
2787 | ||
994eb764 | 2788 | /* Note that interrupts must be enabled when calling this function. */ |
484748f0 | 2789 | bool kmem_cache_alloc_bulk(struct kmem_cache *s, gfp_t flags, size_t size, |
994eb764 | 2790 | void **p) |
484748f0 | 2791 | { |
994eb764 JDB |
2792 | struct kmem_cache_cpu *c; |
2793 | int i; | |
2794 | ||
994eb764 JDB |
2795 | /* |
2796 | * Drain objects in the per cpu slab, while disabling local | |
2797 | * IRQs, which protects against PREEMPT and interrupts | |
2798 | * handlers invoking normal fastpath. | |
2799 | */ | |
2800 | local_irq_disable(); | |
2801 | c = this_cpu_ptr(s->cpu_slab); | |
2802 | ||
2803 | for (i = 0; i < size; i++) { | |
2804 | void *object = c->freelist; | |
2805 | ||
ebe909e0 JDB |
2806 | if (unlikely(!object)) { |
2807 | local_irq_enable(); | |
2808 | /* | |
2809 | * Invoking slow path likely have side-effect | |
2810 | * of re-populating per CPU c->freelist | |
2811 | */ | |
2812 | p[i] = __slab_alloc(s, flags, NUMA_NO_NODE, | |
2813 | _RET_IP_, c); | |
2814 | if (unlikely(!p[i])) { | |
2815 | __kmem_cache_free_bulk(s, i, p); | |
2816 | return false; | |
2817 | } | |
2818 | local_irq_disable(); | |
2819 | c = this_cpu_ptr(s->cpu_slab); | |
2820 | continue; /* goto for-loop */ | |
2821 | } | |
994eb764 | 2822 | |
3eed034d JDB |
2823 | /* kmem_cache debug support */ |
2824 | s = slab_pre_alloc_hook(s, flags); | |
2825 | if (unlikely(!s)) { | |
2826 | __kmem_cache_free_bulk(s, i, p); | |
2827 | c->tid = next_tid(c->tid); | |
2828 | local_irq_enable(); | |
2829 | return false; | |
2830 | } | |
2831 | ||
994eb764 JDB |
2832 | c->freelist = get_freepointer(s, object); |
2833 | p[i] = object; | |
3eed034d JDB |
2834 | |
2835 | /* kmem_cache debug support */ | |
2836 | slab_post_alloc_hook(s, flags, object); | |
994eb764 JDB |
2837 | } |
2838 | c->tid = next_tid(c->tid); | |
2839 | local_irq_enable(); | |
2840 | ||
2841 | /* Clear memory outside IRQ disabled fastpath loop */ | |
2842 | if (unlikely(flags & __GFP_ZERO)) { | |
2843 | int j; | |
2844 | ||
2845 | for (j = 0; j < i; j++) | |
2846 | memset(p[j], 0, s->object_size); | |
2847 | } | |
2848 | ||
994eb764 | 2849 | return true; |
484748f0 CL |
2850 | } |
2851 | EXPORT_SYMBOL(kmem_cache_alloc_bulk); | |
2852 | ||
2853 | ||
81819f0f | 2854 | /* |
672bba3a CL |
2855 | * Object placement in a slab is made very easy because we always start at |
2856 | * offset 0. If we tune the size of the object to the alignment then we can | |
2857 | * get the required alignment by putting one properly sized object after | |
2858 | * another. | |
81819f0f CL |
2859 | * |
2860 | * Notice that the allocation order determines the sizes of the per cpu | |
2861 | * caches. Each processor has always one slab available for allocations. | |
2862 | * Increasing the allocation order reduces the number of times that slabs | |
672bba3a | 2863 | * must be moved on and off the partial lists and is therefore a factor in |
81819f0f | 2864 | * locking overhead. |
81819f0f CL |
2865 | */ |
2866 | ||
2867 | /* | |
2868 | * Mininum / Maximum order of slab pages. This influences locking overhead | |
2869 | * and slab fragmentation. A higher order reduces the number of partial slabs | |
2870 | * and increases the number of allocations possible without having to | |
2871 | * take the list_lock. | |
2872 | */ | |
2873 | static int slub_min_order; | |
114e9e89 | 2874 | static int slub_max_order = PAGE_ALLOC_COSTLY_ORDER; |
9b2cd506 | 2875 | static int slub_min_objects; |
81819f0f | 2876 | |
81819f0f CL |
2877 | /* |
2878 | * Calculate the order of allocation given an slab object size. | |
2879 | * | |
672bba3a CL |
2880 | * The order of allocation has significant impact on performance and other |
2881 | * system components. Generally order 0 allocations should be preferred since | |
2882 | * order 0 does not cause fragmentation in the page allocator. Larger objects | |
2883 | * be problematic to put into order 0 slabs because there may be too much | |
c124f5b5 | 2884 | * unused space left. We go to a higher order if more than 1/16th of the slab |
672bba3a CL |
2885 | * would be wasted. |
2886 | * | |
2887 | * In order to reach satisfactory performance we must ensure that a minimum | |
2888 | * number of objects is in one slab. Otherwise we may generate too much | |
2889 | * activity on the partial lists which requires taking the list_lock. This is | |
2890 | * less a concern for large slabs though which are rarely used. | |
81819f0f | 2891 | * |
672bba3a CL |
2892 | * slub_max_order specifies the order where we begin to stop considering the |
2893 | * number of objects in a slab as critical. If we reach slub_max_order then | |
2894 | * we try to keep the page order as low as possible. So we accept more waste | |
2895 | * of space in favor of a small page order. | |
81819f0f | 2896 | * |
672bba3a CL |
2897 | * Higher order allocations also allow the placement of more objects in a |
2898 | * slab and thereby reduce object handling overhead. If the user has | |
2899 | * requested a higher mininum order then we start with that one instead of | |
2900 | * the smallest order which will fit the object. | |
81819f0f | 2901 | */ |
5e6d444e | 2902 | static inline int slab_order(int size, int min_objects, |
ab9a0f19 | 2903 | int max_order, int fract_leftover, int reserved) |
81819f0f CL |
2904 | { |
2905 | int order; | |
2906 | int rem; | |
6300ea75 | 2907 | int min_order = slub_min_order; |
81819f0f | 2908 | |
ab9a0f19 | 2909 | if (order_objects(min_order, size, reserved) > MAX_OBJS_PER_PAGE) |
210b5c06 | 2910 | return get_order(size * MAX_OBJS_PER_PAGE) - 1; |
39b26464 | 2911 | |
9f835703 | 2912 | for (order = max(min_order, get_order(min_objects * size + reserved)); |
5e6d444e | 2913 | order <= max_order; order++) { |
81819f0f | 2914 | |
5e6d444e | 2915 | unsigned long slab_size = PAGE_SIZE << order; |
81819f0f | 2916 | |
ab9a0f19 | 2917 | rem = (slab_size - reserved) % size; |
81819f0f | 2918 | |
5e6d444e | 2919 | if (rem <= slab_size / fract_leftover) |
81819f0f | 2920 | break; |
81819f0f | 2921 | } |
672bba3a | 2922 | |
81819f0f CL |
2923 | return order; |
2924 | } | |
2925 | ||
ab9a0f19 | 2926 | static inline int calculate_order(int size, int reserved) |
5e6d444e CL |
2927 | { |
2928 | int order; | |
2929 | int min_objects; | |
2930 | int fraction; | |
e8120ff1 | 2931 | int max_objects; |
5e6d444e CL |
2932 | |
2933 | /* | |
2934 | * Attempt to find best configuration for a slab. This | |
2935 | * works by first attempting to generate a layout with | |
2936 | * the best configuration and backing off gradually. | |
2937 | * | |
422ff4d7 | 2938 | * First we increase the acceptable waste in a slab. Then |
5e6d444e CL |
2939 | * we reduce the minimum objects required in a slab. |
2940 | */ | |
2941 | min_objects = slub_min_objects; | |
9b2cd506 CL |
2942 | if (!min_objects) |
2943 | min_objects = 4 * (fls(nr_cpu_ids) + 1); | |
ab9a0f19 | 2944 | max_objects = order_objects(slub_max_order, size, reserved); |
e8120ff1 ZY |
2945 | min_objects = min(min_objects, max_objects); |
2946 | ||
5e6d444e | 2947 | while (min_objects > 1) { |
c124f5b5 | 2948 | fraction = 16; |
5e6d444e CL |
2949 | while (fraction >= 4) { |
2950 | order = slab_order(size, min_objects, | |
ab9a0f19 | 2951 | slub_max_order, fraction, reserved); |
5e6d444e CL |
2952 | if (order <= slub_max_order) |
2953 | return order; | |
2954 | fraction /= 2; | |
2955 | } | |
5086c389 | 2956 | min_objects--; |
5e6d444e CL |
2957 | } |
2958 | ||
2959 | /* | |
2960 | * We were unable to place multiple objects in a slab. Now | |
2961 | * lets see if we can place a single object there. | |
2962 | */ | |
ab9a0f19 | 2963 | order = slab_order(size, 1, slub_max_order, 1, reserved); |
5e6d444e CL |
2964 | if (order <= slub_max_order) |
2965 | return order; | |
2966 | ||
2967 | /* | |
2968 | * Doh this slab cannot be placed using slub_max_order. | |
2969 | */ | |
ab9a0f19 | 2970 | order = slab_order(size, 1, MAX_ORDER, 1, reserved); |
818cf590 | 2971 | if (order < MAX_ORDER) |
5e6d444e CL |
2972 | return order; |
2973 | return -ENOSYS; | |
2974 | } | |
2975 | ||
5595cffc | 2976 | static void |
4053497d | 2977 | init_kmem_cache_node(struct kmem_cache_node *n) |
81819f0f CL |
2978 | { |
2979 | n->nr_partial = 0; | |
81819f0f CL |
2980 | spin_lock_init(&n->list_lock); |
2981 | INIT_LIST_HEAD(&n->partial); | |
8ab1372f | 2982 | #ifdef CONFIG_SLUB_DEBUG |
0f389ec6 | 2983 | atomic_long_set(&n->nr_slabs, 0); |
02b71b70 | 2984 | atomic_long_set(&n->total_objects, 0); |
643b1138 | 2985 | INIT_LIST_HEAD(&n->full); |
8ab1372f | 2986 | #endif |
81819f0f CL |
2987 | } |
2988 | ||
55136592 | 2989 | static inline int alloc_kmem_cache_cpus(struct kmem_cache *s) |
4c93c355 | 2990 | { |
6c182dc0 | 2991 | BUILD_BUG_ON(PERCPU_DYNAMIC_EARLY_SIZE < |
95a05b42 | 2992 | KMALLOC_SHIFT_HIGH * sizeof(struct kmem_cache_cpu)); |
4c93c355 | 2993 | |
8a5ec0ba | 2994 | /* |
d4d84fef CM |
2995 | * Must align to double word boundary for the double cmpxchg |
2996 | * instructions to work; see __pcpu_double_call_return_bool(). | |
8a5ec0ba | 2997 | */ |
d4d84fef CM |
2998 | s->cpu_slab = __alloc_percpu(sizeof(struct kmem_cache_cpu), |
2999 | 2 * sizeof(void *)); | |
8a5ec0ba CL |
3000 | |
3001 | if (!s->cpu_slab) | |
3002 | return 0; | |
3003 | ||
3004 | init_kmem_cache_cpus(s); | |
4c93c355 | 3005 | |
8a5ec0ba | 3006 | return 1; |
4c93c355 | 3007 | } |
4c93c355 | 3008 | |
51df1142 CL |
3009 | static struct kmem_cache *kmem_cache_node; |
3010 | ||
81819f0f CL |
3011 | /* |
3012 | * No kmalloc_node yet so do it by hand. We know that this is the first | |
3013 | * slab on the node for this slabcache. There are no concurrent accesses | |
3014 | * possible. | |
3015 | * | |
721ae22a ZYW |
3016 | * Note that this function only works on the kmem_cache_node |
3017 | * when allocating for the kmem_cache_node. This is used for bootstrapping | |
4c93c355 | 3018 | * memory on a fresh node that has no slab structures yet. |
81819f0f | 3019 | */ |
55136592 | 3020 | static void early_kmem_cache_node_alloc(int node) |
81819f0f CL |
3021 | { |
3022 | struct page *page; | |
3023 | struct kmem_cache_node *n; | |
3024 | ||
51df1142 | 3025 | BUG_ON(kmem_cache_node->size < sizeof(struct kmem_cache_node)); |
81819f0f | 3026 | |
51df1142 | 3027 | page = new_slab(kmem_cache_node, GFP_NOWAIT, node); |
81819f0f CL |
3028 | |
3029 | BUG_ON(!page); | |
a2f92ee7 | 3030 | if (page_to_nid(page) != node) { |
f9f58285 FF |
3031 | pr_err("SLUB: Unable to allocate memory from node %d\n", node); |
3032 | pr_err("SLUB: Allocating a useless per node structure in order to be able to continue\n"); | |
a2f92ee7 CL |
3033 | } |
3034 | ||
81819f0f CL |
3035 | n = page->freelist; |
3036 | BUG_ON(!n); | |
51df1142 | 3037 | page->freelist = get_freepointer(kmem_cache_node, n); |
e6e82ea1 | 3038 | page->inuse = 1; |
8cb0a506 | 3039 | page->frozen = 0; |
51df1142 | 3040 | kmem_cache_node->node[node] = n; |
8ab1372f | 3041 | #ifdef CONFIG_SLUB_DEBUG |
f7cb1933 | 3042 | init_object(kmem_cache_node, n, SLUB_RED_ACTIVE); |
51df1142 | 3043 | init_tracking(kmem_cache_node, n); |
8ab1372f | 3044 | #endif |
0316bec2 | 3045 | kasan_kmalloc(kmem_cache_node, n, sizeof(struct kmem_cache_node)); |
4053497d | 3046 | init_kmem_cache_node(n); |
51df1142 | 3047 | inc_slabs_node(kmem_cache_node, node, page->objects); |
6446faa2 | 3048 | |
67b6c900 | 3049 | /* |
1e4dd946 SR |
3050 | * No locks need to be taken here as it has just been |
3051 | * initialized and there is no concurrent access. | |
67b6c900 | 3052 | */ |
1e4dd946 | 3053 | __add_partial(n, page, DEACTIVATE_TO_HEAD); |
81819f0f CL |
3054 | } |
3055 | ||
3056 | static void free_kmem_cache_nodes(struct kmem_cache *s) | |
3057 | { | |
3058 | int node; | |
fa45dc25 | 3059 | struct kmem_cache_node *n; |
81819f0f | 3060 | |
fa45dc25 CL |
3061 | for_each_kmem_cache_node(s, node, n) { |
3062 | kmem_cache_free(kmem_cache_node, n); | |
81819f0f CL |
3063 | s->node[node] = NULL; |
3064 | } | |
3065 | } | |
3066 | ||
55136592 | 3067 | static int init_kmem_cache_nodes(struct kmem_cache *s) |
81819f0f CL |
3068 | { |
3069 | int node; | |
81819f0f | 3070 | |
f64dc58c | 3071 | for_each_node_state(node, N_NORMAL_MEMORY) { |
81819f0f CL |
3072 | struct kmem_cache_node *n; |
3073 | ||
73367bd8 | 3074 | if (slab_state == DOWN) { |
55136592 | 3075 | early_kmem_cache_node_alloc(node); |
73367bd8 AD |
3076 | continue; |
3077 | } | |
51df1142 | 3078 | n = kmem_cache_alloc_node(kmem_cache_node, |
55136592 | 3079 | GFP_KERNEL, node); |
81819f0f | 3080 | |
73367bd8 AD |
3081 | if (!n) { |
3082 | free_kmem_cache_nodes(s); | |
3083 | return 0; | |
81819f0f | 3084 | } |
73367bd8 | 3085 | |
81819f0f | 3086 | s->node[node] = n; |
4053497d | 3087 | init_kmem_cache_node(n); |
81819f0f CL |
3088 | } |
3089 | return 1; | |
3090 | } | |
81819f0f | 3091 | |
c0bdb232 | 3092 | static void set_min_partial(struct kmem_cache *s, unsigned long min) |
3b89d7d8 DR |
3093 | { |
3094 | if (min < MIN_PARTIAL) | |
3095 | min = MIN_PARTIAL; | |
3096 | else if (min > MAX_PARTIAL) | |
3097 | min = MAX_PARTIAL; | |
3098 | s->min_partial = min; | |
3099 | } | |
3100 | ||
81819f0f CL |
3101 | /* |
3102 | * calculate_sizes() determines the order and the distribution of data within | |
3103 | * a slab object. | |
3104 | */ | |
06b285dc | 3105 | static int calculate_sizes(struct kmem_cache *s, int forced_order) |
81819f0f CL |
3106 | { |
3107 | unsigned long flags = s->flags; | |
3b0efdfa | 3108 | unsigned long size = s->object_size; |
834f3d11 | 3109 | int order; |
81819f0f | 3110 | |
d8b42bf5 CL |
3111 | /* |
3112 | * Round up object size to the next word boundary. We can only | |
3113 | * place the free pointer at word boundaries and this determines | |
3114 | * the possible location of the free pointer. | |
3115 | */ | |
3116 | size = ALIGN(size, sizeof(void *)); | |
3117 | ||
3118 | #ifdef CONFIG_SLUB_DEBUG | |
81819f0f CL |
3119 | /* |
3120 | * Determine if we can poison the object itself. If the user of | |
3121 | * the slab may touch the object after free or before allocation | |
3122 | * then we should never poison the object itself. | |
3123 | */ | |
3124 | if ((flags & SLAB_POISON) && !(flags & SLAB_DESTROY_BY_RCU) && | |
c59def9f | 3125 | !s->ctor) |
81819f0f CL |
3126 | s->flags |= __OBJECT_POISON; |
3127 | else | |
3128 | s->flags &= ~__OBJECT_POISON; | |
3129 | ||
81819f0f CL |
3130 | |
3131 | /* | |
672bba3a | 3132 | * If we are Redzoning then check if there is some space between the |
81819f0f | 3133 | * end of the object and the free pointer. If not then add an |
672bba3a | 3134 | * additional word to have some bytes to store Redzone information. |
81819f0f | 3135 | */ |
3b0efdfa | 3136 | if ((flags & SLAB_RED_ZONE) && size == s->object_size) |
81819f0f | 3137 | size += sizeof(void *); |
41ecc55b | 3138 | #endif |
81819f0f CL |
3139 | |
3140 | /* | |
672bba3a CL |
3141 | * With that we have determined the number of bytes in actual use |
3142 | * by the object. This is the potential offset to the free pointer. | |
81819f0f CL |
3143 | */ |
3144 | s->inuse = size; | |
3145 | ||
3146 | if (((flags & (SLAB_DESTROY_BY_RCU | SLAB_POISON)) || | |
c59def9f | 3147 | s->ctor)) { |
81819f0f CL |
3148 | /* |
3149 | * Relocate free pointer after the object if it is not | |
3150 | * permitted to overwrite the first word of the object on | |
3151 | * kmem_cache_free. | |
3152 | * | |
3153 | * This is the case if we do RCU, have a constructor or | |
3154 | * destructor or are poisoning the objects. | |
3155 | */ | |
3156 | s->offset = size; | |
3157 | size += sizeof(void *); | |
3158 | } | |
3159 | ||
c12b3c62 | 3160 | #ifdef CONFIG_SLUB_DEBUG |
81819f0f CL |
3161 | if (flags & SLAB_STORE_USER) |
3162 | /* | |
3163 | * Need to store information about allocs and frees after | |
3164 | * the object. | |
3165 | */ | |
3166 | size += 2 * sizeof(struct track); | |
3167 | ||
be7b3fbc | 3168 | if (flags & SLAB_RED_ZONE) |
81819f0f CL |
3169 | /* |
3170 | * Add some empty padding so that we can catch | |
3171 | * overwrites from earlier objects rather than let | |
3172 | * tracking information or the free pointer be | |
0211a9c8 | 3173 | * corrupted if a user writes before the start |
81819f0f CL |
3174 | * of the object. |
3175 | */ | |
3176 | size += sizeof(void *); | |
41ecc55b | 3177 | #endif |
672bba3a | 3178 | |
81819f0f CL |
3179 | /* |
3180 | * SLUB stores one object immediately after another beginning from | |
3181 | * offset 0. In order to align the objects we have to simply size | |
3182 | * each object to conform to the alignment. | |
3183 | */ | |
45906855 | 3184 | size = ALIGN(size, s->align); |
81819f0f | 3185 | s->size = size; |
06b285dc CL |
3186 | if (forced_order >= 0) |
3187 | order = forced_order; | |
3188 | else | |
ab9a0f19 | 3189 | order = calculate_order(size, s->reserved); |
81819f0f | 3190 | |
834f3d11 | 3191 | if (order < 0) |
81819f0f CL |
3192 | return 0; |
3193 | ||
b7a49f0d | 3194 | s->allocflags = 0; |
834f3d11 | 3195 | if (order) |
b7a49f0d CL |
3196 | s->allocflags |= __GFP_COMP; |
3197 | ||
3198 | if (s->flags & SLAB_CACHE_DMA) | |
2c59dd65 | 3199 | s->allocflags |= GFP_DMA; |
b7a49f0d CL |
3200 | |
3201 | if (s->flags & SLAB_RECLAIM_ACCOUNT) | |
3202 | s->allocflags |= __GFP_RECLAIMABLE; | |
3203 | ||
81819f0f CL |
3204 | /* |
3205 | * Determine the number of objects per slab | |
3206 | */ | |
ab9a0f19 LJ |
3207 | s->oo = oo_make(order, size, s->reserved); |
3208 | s->min = oo_make(get_order(size), size, s->reserved); | |
205ab99d CL |
3209 | if (oo_objects(s->oo) > oo_objects(s->max)) |
3210 | s->max = s->oo; | |
81819f0f | 3211 | |
834f3d11 | 3212 | return !!oo_objects(s->oo); |
81819f0f CL |
3213 | } |
3214 | ||
8a13a4cc | 3215 | static int kmem_cache_open(struct kmem_cache *s, unsigned long flags) |
81819f0f | 3216 | { |
8a13a4cc | 3217 | s->flags = kmem_cache_flags(s->size, flags, s->name, s->ctor); |
ab9a0f19 | 3218 | s->reserved = 0; |
81819f0f | 3219 | |
da9a638c LJ |
3220 | if (need_reserve_slab_rcu && (s->flags & SLAB_DESTROY_BY_RCU)) |
3221 | s->reserved = sizeof(struct rcu_head); | |
81819f0f | 3222 | |
06b285dc | 3223 | if (!calculate_sizes(s, -1)) |
81819f0f | 3224 | goto error; |
3de47213 DR |
3225 | if (disable_higher_order_debug) { |
3226 | /* | |
3227 | * Disable debugging flags that store metadata if the min slab | |
3228 | * order increased. | |
3229 | */ | |
3b0efdfa | 3230 | if (get_order(s->size) > get_order(s->object_size)) { |
3de47213 DR |
3231 | s->flags &= ~DEBUG_METADATA_FLAGS; |
3232 | s->offset = 0; | |
3233 | if (!calculate_sizes(s, -1)) | |
3234 | goto error; | |
3235 | } | |
3236 | } | |
81819f0f | 3237 | |
2565409f HC |
3238 | #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \ |
3239 | defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE) | |
b789ef51 CL |
3240 | if (system_has_cmpxchg_double() && (s->flags & SLAB_DEBUG_FLAGS) == 0) |
3241 | /* Enable fast mode */ | |
3242 | s->flags |= __CMPXCHG_DOUBLE; | |
3243 | #endif | |
3244 | ||
3b89d7d8 DR |
3245 | /* |
3246 | * The larger the object size is, the more pages we want on the partial | |
3247 | * list to avoid pounding the page allocator excessively. | |
3248 | */ | |
49e22585 CL |
3249 | set_min_partial(s, ilog2(s->size) / 2); |
3250 | ||
3251 | /* | |
3252 | * cpu_partial determined the maximum number of objects kept in the | |
3253 | * per cpu partial lists of a processor. | |
3254 | * | |
3255 | * Per cpu partial lists mainly contain slabs that just have one | |
3256 | * object freed. If they are used for allocation then they can be | |
3257 | * filled up again with minimal effort. The slab will never hit the | |
3258 | * per node partial lists and therefore no locking will be required. | |
3259 | * | |
3260 | * This setting also determines | |
3261 | * | |
3262 | * A) The number of objects from per cpu partial slabs dumped to the | |
3263 | * per node list when we reach the limit. | |
9f264904 | 3264 | * B) The number of objects in cpu partial slabs to extract from the |
d0e0ac97 CG |
3265 | * per node list when we run out of per cpu objects. We only fetch |
3266 | * 50% to keep some capacity around for frees. | |
49e22585 | 3267 | */ |
345c905d | 3268 | if (!kmem_cache_has_cpu_partial(s)) |
8f1e33da CL |
3269 | s->cpu_partial = 0; |
3270 | else if (s->size >= PAGE_SIZE) | |
49e22585 CL |
3271 | s->cpu_partial = 2; |
3272 | else if (s->size >= 1024) | |
3273 | s->cpu_partial = 6; | |
3274 | else if (s->size >= 256) | |
3275 | s->cpu_partial = 13; | |
3276 | else | |
3277 | s->cpu_partial = 30; | |
3278 | ||
81819f0f | 3279 | #ifdef CONFIG_NUMA |
e2cb96b7 | 3280 | s->remote_node_defrag_ratio = 1000; |
81819f0f | 3281 | #endif |
55136592 | 3282 | if (!init_kmem_cache_nodes(s)) |
dfb4f096 | 3283 | goto error; |
81819f0f | 3284 | |
55136592 | 3285 | if (alloc_kmem_cache_cpus(s)) |
278b1bb1 | 3286 | return 0; |
ff12059e | 3287 | |
4c93c355 | 3288 | free_kmem_cache_nodes(s); |
81819f0f CL |
3289 | error: |
3290 | if (flags & SLAB_PANIC) | |
3291 | panic("Cannot create slab %s size=%lu realsize=%u " | |
3292 | "order=%u offset=%u flags=%lx\n", | |
d0e0ac97 CG |
3293 | s->name, (unsigned long)s->size, s->size, |
3294 | oo_order(s->oo), s->offset, flags); | |
278b1bb1 | 3295 | return -EINVAL; |
81819f0f | 3296 | } |
81819f0f | 3297 | |
33b12c38 CL |
3298 | static void list_slab_objects(struct kmem_cache *s, struct page *page, |
3299 | const char *text) | |
3300 | { | |
3301 | #ifdef CONFIG_SLUB_DEBUG | |
3302 | void *addr = page_address(page); | |
3303 | void *p; | |
a5dd5c11 NK |
3304 | unsigned long *map = kzalloc(BITS_TO_LONGS(page->objects) * |
3305 | sizeof(long), GFP_ATOMIC); | |
bbd7d57b ED |
3306 | if (!map) |
3307 | return; | |
945cf2b6 | 3308 | slab_err(s, page, text, s->name); |
33b12c38 | 3309 | slab_lock(page); |
33b12c38 | 3310 | |
5f80b13a | 3311 | get_map(s, page, map); |
33b12c38 CL |
3312 | for_each_object(p, s, addr, page->objects) { |
3313 | ||
3314 | if (!test_bit(slab_index(p, s, addr), map)) { | |
f9f58285 | 3315 | pr_err("INFO: Object 0x%p @offset=%tu\n", p, p - addr); |
33b12c38 CL |
3316 | print_tracking(s, p); |
3317 | } | |
3318 | } | |
3319 | slab_unlock(page); | |
bbd7d57b | 3320 | kfree(map); |
33b12c38 CL |
3321 | #endif |
3322 | } | |
3323 | ||
81819f0f | 3324 | /* |
599870b1 | 3325 | * Attempt to free all partial slabs on a node. |
69cb8e6b CL |
3326 | * This is called from kmem_cache_close(). We must be the last thread |
3327 | * using the cache and therefore we do not need to lock anymore. | |
81819f0f | 3328 | */ |
599870b1 | 3329 | static void free_partial(struct kmem_cache *s, struct kmem_cache_node *n) |
81819f0f | 3330 | { |
81819f0f CL |
3331 | struct page *page, *h; |
3332 | ||
33b12c38 | 3333 | list_for_each_entry_safe(page, h, &n->partial, lru) { |
81819f0f | 3334 | if (!page->inuse) { |
1e4dd946 | 3335 | __remove_partial(n, page); |
81819f0f | 3336 | discard_slab(s, page); |
33b12c38 CL |
3337 | } else { |
3338 | list_slab_objects(s, page, | |
945cf2b6 | 3339 | "Objects remaining in %s on kmem_cache_close()"); |
599870b1 | 3340 | } |
33b12c38 | 3341 | } |
81819f0f CL |
3342 | } |
3343 | ||
3344 | /* | |
672bba3a | 3345 | * Release all resources used by a slab cache. |
81819f0f | 3346 | */ |
0c710013 | 3347 | static inline int kmem_cache_close(struct kmem_cache *s) |
81819f0f CL |
3348 | { |
3349 | int node; | |
fa45dc25 | 3350 | struct kmem_cache_node *n; |
81819f0f CL |
3351 | |
3352 | flush_all(s); | |
81819f0f | 3353 | /* Attempt to free all objects */ |
fa45dc25 | 3354 | for_each_kmem_cache_node(s, node, n) { |
599870b1 CL |
3355 | free_partial(s, n); |
3356 | if (n->nr_partial || slabs_node(s, node)) | |
81819f0f CL |
3357 | return 1; |
3358 | } | |
945cf2b6 | 3359 | free_percpu(s->cpu_slab); |
81819f0f CL |
3360 | free_kmem_cache_nodes(s); |
3361 | return 0; | |
3362 | } | |
3363 | ||
945cf2b6 | 3364 | int __kmem_cache_shutdown(struct kmem_cache *s) |
81819f0f | 3365 | { |
41a21285 | 3366 | return kmem_cache_close(s); |
81819f0f | 3367 | } |
81819f0f CL |
3368 | |
3369 | /******************************************************************** | |
3370 | * Kmalloc subsystem | |
3371 | *******************************************************************/ | |
3372 | ||
81819f0f CL |
3373 | static int __init setup_slub_min_order(char *str) |
3374 | { | |
06428780 | 3375 | get_option(&str, &slub_min_order); |
81819f0f CL |
3376 | |
3377 | return 1; | |
3378 | } | |
3379 | ||
3380 | __setup("slub_min_order=", setup_slub_min_order); | |
3381 | ||
3382 | static int __init setup_slub_max_order(char *str) | |
3383 | { | |
06428780 | 3384 | get_option(&str, &slub_max_order); |
818cf590 | 3385 | slub_max_order = min(slub_max_order, MAX_ORDER - 1); |
81819f0f CL |
3386 | |
3387 | return 1; | |
3388 | } | |
3389 | ||
3390 | __setup("slub_max_order=", setup_slub_max_order); | |
3391 | ||
3392 | static int __init setup_slub_min_objects(char *str) | |
3393 | { | |
06428780 | 3394 | get_option(&str, &slub_min_objects); |
81819f0f CL |
3395 | |
3396 | return 1; | |
3397 | } | |
3398 | ||
3399 | __setup("slub_min_objects=", setup_slub_min_objects); | |
3400 | ||
81819f0f CL |
3401 | void *__kmalloc(size_t size, gfp_t flags) |
3402 | { | |
aadb4bc4 | 3403 | struct kmem_cache *s; |
5b882be4 | 3404 | void *ret; |
81819f0f | 3405 | |
95a05b42 | 3406 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) |
eada35ef | 3407 | return kmalloc_large(size, flags); |
aadb4bc4 | 3408 | |
2c59dd65 | 3409 | s = kmalloc_slab(size, flags); |
aadb4bc4 CL |
3410 | |
3411 | if (unlikely(ZERO_OR_NULL_PTR(s))) | |
6cb8f913 CL |
3412 | return s; |
3413 | ||
2b847c3c | 3414 | ret = slab_alloc(s, flags, _RET_IP_); |
5b882be4 | 3415 | |
ca2b84cb | 3416 | trace_kmalloc(_RET_IP_, ret, size, s->size, flags); |
5b882be4 | 3417 | |
0316bec2 AR |
3418 | kasan_kmalloc(s, ret, size); |
3419 | ||
5b882be4 | 3420 | return ret; |
81819f0f CL |
3421 | } |
3422 | EXPORT_SYMBOL(__kmalloc); | |
3423 | ||
5d1f57e4 | 3424 | #ifdef CONFIG_NUMA |
f619cfe1 CL |
3425 | static void *kmalloc_large_node(size_t size, gfp_t flags, int node) |
3426 | { | |
b1eeab67 | 3427 | struct page *page; |
e4f7c0b4 | 3428 | void *ptr = NULL; |
f619cfe1 | 3429 | |
52383431 VD |
3430 | flags |= __GFP_COMP | __GFP_NOTRACK; |
3431 | page = alloc_kmem_pages_node(node, flags, get_order(size)); | |
f619cfe1 | 3432 | if (page) |
e4f7c0b4 CM |
3433 | ptr = page_address(page); |
3434 | ||
d56791b3 | 3435 | kmalloc_large_node_hook(ptr, size, flags); |
e4f7c0b4 | 3436 | return ptr; |
f619cfe1 CL |
3437 | } |
3438 | ||
81819f0f CL |
3439 | void *__kmalloc_node(size_t size, gfp_t flags, int node) |
3440 | { | |
aadb4bc4 | 3441 | struct kmem_cache *s; |
5b882be4 | 3442 | void *ret; |
81819f0f | 3443 | |
95a05b42 | 3444 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) { |
5b882be4 EGM |
3445 | ret = kmalloc_large_node(size, flags, node); |
3446 | ||
ca2b84cb EGM |
3447 | trace_kmalloc_node(_RET_IP_, ret, |
3448 | size, PAGE_SIZE << get_order(size), | |
3449 | flags, node); | |
5b882be4 EGM |
3450 | |
3451 | return ret; | |
3452 | } | |
aadb4bc4 | 3453 | |
2c59dd65 | 3454 | s = kmalloc_slab(size, flags); |
aadb4bc4 CL |
3455 | |
3456 | if (unlikely(ZERO_OR_NULL_PTR(s))) | |
6cb8f913 CL |
3457 | return s; |
3458 | ||
2b847c3c | 3459 | ret = slab_alloc_node(s, flags, node, _RET_IP_); |
5b882be4 | 3460 | |
ca2b84cb | 3461 | trace_kmalloc_node(_RET_IP_, ret, size, s->size, flags, node); |
5b882be4 | 3462 | |
0316bec2 AR |
3463 | kasan_kmalloc(s, ret, size); |
3464 | ||
5b882be4 | 3465 | return ret; |
81819f0f CL |
3466 | } |
3467 | EXPORT_SYMBOL(__kmalloc_node); | |
3468 | #endif | |
3469 | ||
0316bec2 | 3470 | static size_t __ksize(const void *object) |
81819f0f | 3471 | { |
272c1d21 | 3472 | struct page *page; |
81819f0f | 3473 | |
ef8b4520 | 3474 | if (unlikely(object == ZERO_SIZE_PTR)) |
272c1d21 CL |
3475 | return 0; |
3476 | ||
294a80a8 | 3477 | page = virt_to_head_page(object); |
294a80a8 | 3478 | |
76994412 PE |
3479 | if (unlikely(!PageSlab(page))) { |
3480 | WARN_ON(!PageCompound(page)); | |
294a80a8 | 3481 | return PAGE_SIZE << compound_order(page); |
76994412 | 3482 | } |
81819f0f | 3483 | |
1b4f59e3 | 3484 | return slab_ksize(page->slab_cache); |
81819f0f | 3485 | } |
0316bec2 AR |
3486 | |
3487 | size_t ksize(const void *object) | |
3488 | { | |
3489 | size_t size = __ksize(object); | |
3490 | /* We assume that ksize callers could use whole allocated area, | |
3491 | so we need unpoison this area. */ | |
3492 | kasan_krealloc(object, size); | |
3493 | return size; | |
3494 | } | |
b1aabecd | 3495 | EXPORT_SYMBOL(ksize); |
81819f0f CL |
3496 | |
3497 | void kfree(const void *x) | |
3498 | { | |
81819f0f | 3499 | struct page *page; |
5bb983b0 | 3500 | void *object = (void *)x; |
81819f0f | 3501 | |
2121db74 PE |
3502 | trace_kfree(_RET_IP_, x); |
3503 | ||
2408c550 | 3504 | if (unlikely(ZERO_OR_NULL_PTR(x))) |
81819f0f CL |
3505 | return; |
3506 | ||
b49af68f | 3507 | page = virt_to_head_page(x); |
aadb4bc4 | 3508 | if (unlikely(!PageSlab(page))) { |
0937502a | 3509 | BUG_ON(!PageCompound(page)); |
d56791b3 | 3510 | kfree_hook(x); |
52383431 | 3511 | __free_kmem_pages(page, compound_order(page)); |
aadb4bc4 CL |
3512 | return; |
3513 | } | |
1b4f59e3 | 3514 | slab_free(page->slab_cache, page, object, _RET_IP_); |
81819f0f CL |
3515 | } |
3516 | EXPORT_SYMBOL(kfree); | |
3517 | ||
832f37f5 VD |
3518 | #define SHRINK_PROMOTE_MAX 32 |
3519 | ||
2086d26a | 3520 | /* |
832f37f5 VD |
3521 | * kmem_cache_shrink discards empty slabs and promotes the slabs filled |
3522 | * up most to the head of the partial lists. New allocations will then | |
3523 | * fill those up and thus they can be removed from the partial lists. | |
672bba3a CL |
3524 | * |
3525 | * The slabs with the least items are placed last. This results in them | |
3526 | * being allocated from last increasing the chance that the last objects | |
3527 | * are freed in them. | |
2086d26a | 3528 | */ |
d6e0b7fa | 3529 | int __kmem_cache_shrink(struct kmem_cache *s, bool deactivate) |
2086d26a CL |
3530 | { |
3531 | int node; | |
3532 | int i; | |
3533 | struct kmem_cache_node *n; | |
3534 | struct page *page; | |
3535 | struct page *t; | |
832f37f5 VD |
3536 | struct list_head discard; |
3537 | struct list_head promote[SHRINK_PROMOTE_MAX]; | |
2086d26a | 3538 | unsigned long flags; |
ce3712d7 | 3539 | int ret = 0; |
2086d26a | 3540 | |
d6e0b7fa VD |
3541 | if (deactivate) { |
3542 | /* | |
3543 | * Disable empty slabs caching. Used to avoid pinning offline | |
3544 | * memory cgroups by kmem pages that can be freed. | |
3545 | */ | |
3546 | s->cpu_partial = 0; | |
3547 | s->min_partial = 0; | |
3548 | ||
3549 | /* | |
3550 | * s->cpu_partial is checked locklessly (see put_cpu_partial), | |
3551 | * so we have to make sure the change is visible. | |
3552 | */ | |
3553 | kick_all_cpus_sync(); | |
3554 | } | |
3555 | ||
2086d26a | 3556 | flush_all(s); |
fa45dc25 | 3557 | for_each_kmem_cache_node(s, node, n) { |
832f37f5 VD |
3558 | INIT_LIST_HEAD(&discard); |
3559 | for (i = 0; i < SHRINK_PROMOTE_MAX; i++) | |
3560 | INIT_LIST_HEAD(promote + i); | |
2086d26a CL |
3561 | |
3562 | spin_lock_irqsave(&n->list_lock, flags); | |
3563 | ||
3564 | /* | |
832f37f5 | 3565 | * Build lists of slabs to discard or promote. |
2086d26a | 3566 | * |
672bba3a CL |
3567 | * Note that concurrent frees may occur while we hold the |
3568 | * list_lock. page->inuse here is the upper limit. | |
2086d26a CL |
3569 | */ |
3570 | list_for_each_entry_safe(page, t, &n->partial, lru) { | |
832f37f5 VD |
3571 | int free = page->objects - page->inuse; |
3572 | ||
3573 | /* Do not reread page->inuse */ | |
3574 | barrier(); | |
3575 | ||
3576 | /* We do not keep full slabs on the list */ | |
3577 | BUG_ON(free <= 0); | |
3578 | ||
3579 | if (free == page->objects) { | |
3580 | list_move(&page->lru, &discard); | |
69cb8e6b | 3581 | n->nr_partial--; |
832f37f5 VD |
3582 | } else if (free <= SHRINK_PROMOTE_MAX) |
3583 | list_move(&page->lru, promote + free - 1); | |
2086d26a CL |
3584 | } |
3585 | ||
2086d26a | 3586 | /* |
832f37f5 VD |
3587 | * Promote the slabs filled up most to the head of the |
3588 | * partial list. | |
2086d26a | 3589 | */ |
832f37f5 VD |
3590 | for (i = SHRINK_PROMOTE_MAX - 1; i >= 0; i--) |
3591 | list_splice(promote + i, &n->partial); | |
2086d26a | 3592 | |
2086d26a | 3593 | spin_unlock_irqrestore(&n->list_lock, flags); |
69cb8e6b CL |
3594 | |
3595 | /* Release empty slabs */ | |
832f37f5 | 3596 | list_for_each_entry_safe(page, t, &discard, lru) |
69cb8e6b | 3597 | discard_slab(s, page); |
ce3712d7 VD |
3598 | |
3599 | if (slabs_node(s, node)) | |
3600 | ret = 1; | |
2086d26a CL |
3601 | } |
3602 | ||
ce3712d7 | 3603 | return ret; |
2086d26a | 3604 | } |
2086d26a | 3605 | |
b9049e23 YG |
3606 | static int slab_mem_going_offline_callback(void *arg) |
3607 | { | |
3608 | struct kmem_cache *s; | |
3609 | ||
18004c5d | 3610 | mutex_lock(&slab_mutex); |
b9049e23 | 3611 | list_for_each_entry(s, &slab_caches, list) |
d6e0b7fa | 3612 | __kmem_cache_shrink(s, false); |
18004c5d | 3613 | mutex_unlock(&slab_mutex); |
b9049e23 YG |
3614 | |
3615 | return 0; | |
3616 | } | |
3617 | ||
3618 | static void slab_mem_offline_callback(void *arg) | |
3619 | { | |
3620 | struct kmem_cache_node *n; | |
3621 | struct kmem_cache *s; | |
3622 | struct memory_notify *marg = arg; | |
3623 | int offline_node; | |
3624 | ||
b9d5ab25 | 3625 | offline_node = marg->status_change_nid_normal; |
b9049e23 YG |
3626 | |
3627 | /* | |
3628 | * If the node still has available memory. we need kmem_cache_node | |
3629 | * for it yet. | |
3630 | */ | |
3631 | if (offline_node < 0) | |
3632 | return; | |
3633 | ||
18004c5d | 3634 | mutex_lock(&slab_mutex); |
b9049e23 YG |
3635 | list_for_each_entry(s, &slab_caches, list) { |
3636 | n = get_node(s, offline_node); | |
3637 | if (n) { | |
3638 | /* | |
3639 | * if n->nr_slabs > 0, slabs still exist on the node | |
3640 | * that is going down. We were unable to free them, | |
c9404c9c | 3641 | * and offline_pages() function shouldn't call this |
b9049e23 YG |
3642 | * callback. So, we must fail. |
3643 | */ | |
0f389ec6 | 3644 | BUG_ON(slabs_node(s, offline_node)); |
b9049e23 YG |
3645 | |
3646 | s->node[offline_node] = NULL; | |
8de66a0c | 3647 | kmem_cache_free(kmem_cache_node, n); |
b9049e23 YG |
3648 | } |
3649 | } | |
18004c5d | 3650 | mutex_unlock(&slab_mutex); |
b9049e23 YG |
3651 | } |
3652 | ||
3653 | static int slab_mem_going_online_callback(void *arg) | |
3654 | { | |
3655 | struct kmem_cache_node *n; | |
3656 | struct kmem_cache *s; | |
3657 | struct memory_notify *marg = arg; | |
b9d5ab25 | 3658 | int nid = marg->status_change_nid_normal; |
b9049e23 YG |
3659 | int ret = 0; |
3660 | ||
3661 | /* | |
3662 | * If the node's memory is already available, then kmem_cache_node is | |
3663 | * already created. Nothing to do. | |
3664 | */ | |
3665 | if (nid < 0) | |
3666 | return 0; | |
3667 | ||
3668 | /* | |
0121c619 | 3669 | * We are bringing a node online. No memory is available yet. We must |
b9049e23 YG |
3670 | * allocate a kmem_cache_node structure in order to bring the node |
3671 | * online. | |
3672 | */ | |
18004c5d | 3673 | mutex_lock(&slab_mutex); |
b9049e23 YG |
3674 | list_for_each_entry(s, &slab_caches, list) { |
3675 | /* | |
3676 | * XXX: kmem_cache_alloc_node will fallback to other nodes | |
3677 | * since memory is not yet available from the node that | |
3678 | * is brought up. | |
3679 | */ | |
8de66a0c | 3680 | n = kmem_cache_alloc(kmem_cache_node, GFP_KERNEL); |
b9049e23 YG |
3681 | if (!n) { |
3682 | ret = -ENOMEM; | |
3683 | goto out; | |
3684 | } | |
4053497d | 3685 | init_kmem_cache_node(n); |
b9049e23 YG |
3686 | s->node[nid] = n; |
3687 | } | |
3688 | out: | |
18004c5d | 3689 | mutex_unlock(&slab_mutex); |
b9049e23 YG |
3690 | return ret; |
3691 | } | |
3692 | ||
3693 | static int slab_memory_callback(struct notifier_block *self, | |
3694 | unsigned long action, void *arg) | |
3695 | { | |
3696 | int ret = 0; | |
3697 | ||
3698 | switch (action) { | |
3699 | case MEM_GOING_ONLINE: | |
3700 | ret = slab_mem_going_online_callback(arg); | |
3701 | break; | |
3702 | case MEM_GOING_OFFLINE: | |
3703 | ret = slab_mem_going_offline_callback(arg); | |
3704 | break; | |
3705 | case MEM_OFFLINE: | |
3706 | case MEM_CANCEL_ONLINE: | |
3707 | slab_mem_offline_callback(arg); | |
3708 | break; | |
3709 | case MEM_ONLINE: | |
3710 | case MEM_CANCEL_OFFLINE: | |
3711 | break; | |
3712 | } | |
dc19f9db KH |
3713 | if (ret) |
3714 | ret = notifier_from_errno(ret); | |
3715 | else | |
3716 | ret = NOTIFY_OK; | |
b9049e23 YG |
3717 | return ret; |
3718 | } | |
3719 | ||
3ac38faa AM |
3720 | static struct notifier_block slab_memory_callback_nb = { |
3721 | .notifier_call = slab_memory_callback, | |
3722 | .priority = SLAB_CALLBACK_PRI, | |
3723 | }; | |
b9049e23 | 3724 | |
81819f0f CL |
3725 | /******************************************************************** |
3726 | * Basic setup of slabs | |
3727 | *******************************************************************/ | |
3728 | ||
51df1142 CL |
3729 | /* |
3730 | * Used for early kmem_cache structures that were allocated using | |
dffb4d60 CL |
3731 | * the page allocator. Allocate them properly then fix up the pointers |
3732 | * that may be pointing to the wrong kmem_cache structure. | |
51df1142 CL |
3733 | */ |
3734 | ||
dffb4d60 | 3735 | static struct kmem_cache * __init bootstrap(struct kmem_cache *static_cache) |
51df1142 CL |
3736 | { |
3737 | int node; | |
dffb4d60 | 3738 | struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT); |
fa45dc25 | 3739 | struct kmem_cache_node *n; |
51df1142 | 3740 | |
dffb4d60 | 3741 | memcpy(s, static_cache, kmem_cache->object_size); |
51df1142 | 3742 | |
7d557b3c GC |
3743 | /* |
3744 | * This runs very early, and only the boot processor is supposed to be | |
3745 | * up. Even if it weren't true, IRQs are not up so we couldn't fire | |
3746 | * IPIs around. | |
3747 | */ | |
3748 | __flush_cpu_slab(s, smp_processor_id()); | |
fa45dc25 | 3749 | for_each_kmem_cache_node(s, node, n) { |
51df1142 CL |
3750 | struct page *p; |
3751 | ||
fa45dc25 CL |
3752 | list_for_each_entry(p, &n->partial, lru) |
3753 | p->slab_cache = s; | |
51df1142 | 3754 | |
607bf324 | 3755 | #ifdef CONFIG_SLUB_DEBUG |
fa45dc25 CL |
3756 | list_for_each_entry(p, &n->full, lru) |
3757 | p->slab_cache = s; | |
51df1142 | 3758 | #endif |
51df1142 | 3759 | } |
f7ce3190 | 3760 | slab_init_memcg_params(s); |
dffb4d60 CL |
3761 | list_add(&s->list, &slab_caches); |
3762 | return s; | |
51df1142 CL |
3763 | } |
3764 | ||
81819f0f CL |
3765 | void __init kmem_cache_init(void) |
3766 | { | |
dffb4d60 CL |
3767 | static __initdata struct kmem_cache boot_kmem_cache, |
3768 | boot_kmem_cache_node; | |
51df1142 | 3769 | |
fc8d8620 SG |
3770 | if (debug_guardpage_minorder()) |
3771 | slub_max_order = 0; | |
3772 | ||
dffb4d60 CL |
3773 | kmem_cache_node = &boot_kmem_cache_node; |
3774 | kmem_cache = &boot_kmem_cache; | |
51df1142 | 3775 | |
dffb4d60 CL |
3776 | create_boot_cache(kmem_cache_node, "kmem_cache_node", |
3777 | sizeof(struct kmem_cache_node), SLAB_HWCACHE_ALIGN); | |
b9049e23 | 3778 | |
3ac38faa | 3779 | register_hotmemory_notifier(&slab_memory_callback_nb); |
81819f0f CL |
3780 | |
3781 | /* Able to allocate the per node structures */ | |
3782 | slab_state = PARTIAL; | |
3783 | ||
dffb4d60 CL |
3784 | create_boot_cache(kmem_cache, "kmem_cache", |
3785 | offsetof(struct kmem_cache, node) + | |
3786 | nr_node_ids * sizeof(struct kmem_cache_node *), | |
3787 | SLAB_HWCACHE_ALIGN); | |
8a13a4cc | 3788 | |
dffb4d60 | 3789 | kmem_cache = bootstrap(&boot_kmem_cache); |
81819f0f | 3790 | |
51df1142 CL |
3791 | /* |
3792 | * Allocate kmem_cache_node properly from the kmem_cache slab. | |
3793 | * kmem_cache_node is separately allocated so no need to | |
3794 | * update any list pointers. | |
3795 | */ | |
dffb4d60 | 3796 | kmem_cache_node = bootstrap(&boot_kmem_cache_node); |
51df1142 CL |
3797 | |
3798 | /* Now we can use the kmem_cache to allocate kmalloc slabs */ | |
34cc6990 | 3799 | setup_kmalloc_cache_index_table(); |
f97d5f63 | 3800 | create_kmalloc_caches(0); |
81819f0f CL |
3801 | |
3802 | #ifdef CONFIG_SMP | |
3803 | register_cpu_notifier(&slab_notifier); | |
9dfc6e68 | 3804 | #endif |
81819f0f | 3805 | |
f9f58285 | 3806 | pr_info("SLUB: HWalign=%d, Order=%d-%d, MinObjects=%d, CPUs=%d, Nodes=%d\n", |
f97d5f63 | 3807 | cache_line_size(), |
81819f0f CL |
3808 | slub_min_order, slub_max_order, slub_min_objects, |
3809 | nr_cpu_ids, nr_node_ids); | |
3810 | } | |
3811 | ||
7e85ee0c PE |
3812 | void __init kmem_cache_init_late(void) |
3813 | { | |
7e85ee0c PE |
3814 | } |
3815 | ||
2633d7a0 | 3816 | struct kmem_cache * |
a44cb944 VD |
3817 | __kmem_cache_alias(const char *name, size_t size, size_t align, |
3818 | unsigned long flags, void (*ctor)(void *)) | |
81819f0f | 3819 | { |
426589f5 | 3820 | struct kmem_cache *s, *c; |
81819f0f | 3821 | |
a44cb944 | 3822 | s = find_mergeable(size, align, flags, name, ctor); |
81819f0f CL |
3823 | if (s) { |
3824 | s->refcount++; | |
84d0ddd6 | 3825 | |
81819f0f CL |
3826 | /* |
3827 | * Adjust the object sizes so that we clear | |
3828 | * the complete object on kzalloc. | |
3829 | */ | |
3b0efdfa | 3830 | s->object_size = max(s->object_size, (int)size); |
81819f0f | 3831 | s->inuse = max_t(int, s->inuse, ALIGN(size, sizeof(void *))); |
6446faa2 | 3832 | |
426589f5 | 3833 | for_each_memcg_cache(c, s) { |
84d0ddd6 VD |
3834 | c->object_size = s->object_size; |
3835 | c->inuse = max_t(int, c->inuse, | |
3836 | ALIGN(size, sizeof(void *))); | |
3837 | } | |
3838 | ||
7b8f3b66 | 3839 | if (sysfs_slab_alias(s, name)) { |
7b8f3b66 | 3840 | s->refcount--; |
cbb79694 | 3841 | s = NULL; |
7b8f3b66 | 3842 | } |
a0e1d1be | 3843 | } |
6446faa2 | 3844 | |
cbb79694 CL |
3845 | return s; |
3846 | } | |
84c1cf62 | 3847 | |
8a13a4cc | 3848 | int __kmem_cache_create(struct kmem_cache *s, unsigned long flags) |
cbb79694 | 3849 | { |
aac3a166 PE |
3850 | int err; |
3851 | ||
3852 | err = kmem_cache_open(s, flags); | |
3853 | if (err) | |
3854 | return err; | |
20cea968 | 3855 | |
45530c44 CL |
3856 | /* Mutex is not taken during early boot */ |
3857 | if (slab_state <= UP) | |
3858 | return 0; | |
3859 | ||
107dab5c | 3860 | memcg_propagate_slab_attrs(s); |
aac3a166 | 3861 | err = sysfs_slab_add(s); |
aac3a166 PE |
3862 | if (err) |
3863 | kmem_cache_close(s); | |
20cea968 | 3864 | |
aac3a166 | 3865 | return err; |
81819f0f | 3866 | } |
81819f0f | 3867 | |
81819f0f | 3868 | #ifdef CONFIG_SMP |
81819f0f | 3869 | /* |
672bba3a CL |
3870 | * Use the cpu notifier to insure that the cpu slabs are flushed when |
3871 | * necessary. | |
81819f0f | 3872 | */ |
0db0628d | 3873 | static int slab_cpuup_callback(struct notifier_block *nfb, |
81819f0f CL |
3874 | unsigned long action, void *hcpu) |
3875 | { | |
3876 | long cpu = (long)hcpu; | |
5b95a4ac CL |
3877 | struct kmem_cache *s; |
3878 | unsigned long flags; | |
81819f0f CL |
3879 | |
3880 | switch (action) { | |
3881 | case CPU_UP_CANCELED: | |
8bb78442 | 3882 | case CPU_UP_CANCELED_FROZEN: |
81819f0f | 3883 | case CPU_DEAD: |
8bb78442 | 3884 | case CPU_DEAD_FROZEN: |
18004c5d | 3885 | mutex_lock(&slab_mutex); |
5b95a4ac CL |
3886 | list_for_each_entry(s, &slab_caches, list) { |
3887 | local_irq_save(flags); | |
3888 | __flush_cpu_slab(s, cpu); | |
3889 | local_irq_restore(flags); | |
3890 | } | |
18004c5d | 3891 | mutex_unlock(&slab_mutex); |
81819f0f CL |
3892 | break; |
3893 | default: | |
3894 | break; | |
3895 | } | |
3896 | return NOTIFY_OK; | |
3897 | } | |
3898 | ||
0db0628d | 3899 | static struct notifier_block slab_notifier = { |
3adbefee | 3900 | .notifier_call = slab_cpuup_callback |
06428780 | 3901 | }; |
81819f0f CL |
3902 | |
3903 | #endif | |
3904 | ||
ce71e27c | 3905 | void *__kmalloc_track_caller(size_t size, gfp_t gfpflags, unsigned long caller) |
81819f0f | 3906 | { |
aadb4bc4 | 3907 | struct kmem_cache *s; |
94b528d0 | 3908 | void *ret; |
aadb4bc4 | 3909 | |
95a05b42 | 3910 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) |
eada35ef PE |
3911 | return kmalloc_large(size, gfpflags); |
3912 | ||
2c59dd65 | 3913 | s = kmalloc_slab(size, gfpflags); |
81819f0f | 3914 | |
2408c550 | 3915 | if (unlikely(ZERO_OR_NULL_PTR(s))) |
6cb8f913 | 3916 | return s; |
81819f0f | 3917 | |
2b847c3c | 3918 | ret = slab_alloc(s, gfpflags, caller); |
94b528d0 | 3919 | |
25985edc | 3920 | /* Honor the call site pointer we received. */ |
ca2b84cb | 3921 | trace_kmalloc(caller, ret, size, s->size, gfpflags); |
94b528d0 EGM |
3922 | |
3923 | return ret; | |
81819f0f CL |
3924 | } |
3925 | ||
5d1f57e4 | 3926 | #ifdef CONFIG_NUMA |
81819f0f | 3927 | void *__kmalloc_node_track_caller(size_t size, gfp_t gfpflags, |
ce71e27c | 3928 | int node, unsigned long caller) |
81819f0f | 3929 | { |
aadb4bc4 | 3930 | struct kmem_cache *s; |
94b528d0 | 3931 | void *ret; |
aadb4bc4 | 3932 | |
95a05b42 | 3933 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) { |
d3e14aa3 XF |
3934 | ret = kmalloc_large_node(size, gfpflags, node); |
3935 | ||
3936 | trace_kmalloc_node(caller, ret, | |
3937 | size, PAGE_SIZE << get_order(size), | |
3938 | gfpflags, node); | |
3939 | ||
3940 | return ret; | |
3941 | } | |
eada35ef | 3942 | |
2c59dd65 | 3943 | s = kmalloc_slab(size, gfpflags); |
81819f0f | 3944 | |
2408c550 | 3945 | if (unlikely(ZERO_OR_NULL_PTR(s))) |
6cb8f913 | 3946 | return s; |
81819f0f | 3947 | |
2b847c3c | 3948 | ret = slab_alloc_node(s, gfpflags, node, caller); |
94b528d0 | 3949 | |
25985edc | 3950 | /* Honor the call site pointer we received. */ |
ca2b84cb | 3951 | trace_kmalloc_node(caller, ret, size, s->size, gfpflags, node); |
94b528d0 EGM |
3952 | |
3953 | return ret; | |
81819f0f | 3954 | } |
5d1f57e4 | 3955 | #endif |
81819f0f | 3956 | |
ab4d5ed5 | 3957 | #ifdef CONFIG_SYSFS |
205ab99d CL |
3958 | static int count_inuse(struct page *page) |
3959 | { | |
3960 | return page->inuse; | |
3961 | } | |
3962 | ||
3963 | static int count_total(struct page *page) | |
3964 | { | |
3965 | return page->objects; | |
3966 | } | |
ab4d5ed5 | 3967 | #endif |
205ab99d | 3968 | |
ab4d5ed5 | 3969 | #ifdef CONFIG_SLUB_DEBUG |
434e245d CL |
3970 | static int validate_slab(struct kmem_cache *s, struct page *page, |
3971 | unsigned long *map) | |
53e15af0 CL |
3972 | { |
3973 | void *p; | |
a973e9dd | 3974 | void *addr = page_address(page); |
53e15af0 CL |
3975 | |
3976 | if (!check_slab(s, page) || | |
3977 | !on_freelist(s, page, NULL)) | |
3978 | return 0; | |
3979 | ||
3980 | /* Now we know that a valid freelist exists */ | |
39b26464 | 3981 | bitmap_zero(map, page->objects); |
53e15af0 | 3982 | |
5f80b13a CL |
3983 | get_map(s, page, map); |
3984 | for_each_object(p, s, addr, page->objects) { | |
3985 | if (test_bit(slab_index(p, s, addr), map)) | |
3986 | if (!check_object(s, page, p, SLUB_RED_INACTIVE)) | |
3987 | return 0; | |
53e15af0 CL |
3988 | } |
3989 | ||
224a88be | 3990 | for_each_object(p, s, addr, page->objects) |
7656c72b | 3991 | if (!test_bit(slab_index(p, s, addr), map)) |
37d57443 | 3992 | if (!check_object(s, page, p, SLUB_RED_ACTIVE)) |
53e15af0 CL |
3993 | return 0; |
3994 | return 1; | |
3995 | } | |
3996 | ||
434e245d CL |
3997 | static void validate_slab_slab(struct kmem_cache *s, struct page *page, |
3998 | unsigned long *map) | |
53e15af0 | 3999 | { |
881db7fb CL |
4000 | slab_lock(page); |
4001 | validate_slab(s, page, map); | |
4002 | slab_unlock(page); | |
53e15af0 CL |
4003 | } |
4004 | ||
434e245d CL |
4005 | static int validate_slab_node(struct kmem_cache *s, |
4006 | struct kmem_cache_node *n, unsigned long *map) | |
53e15af0 CL |
4007 | { |
4008 | unsigned long count = 0; | |
4009 | struct page *page; | |
4010 | unsigned long flags; | |
4011 | ||
4012 | spin_lock_irqsave(&n->list_lock, flags); | |
4013 | ||
4014 | list_for_each_entry(page, &n->partial, lru) { | |
434e245d | 4015 | validate_slab_slab(s, page, map); |
53e15af0 CL |
4016 | count++; |
4017 | } | |
4018 | if (count != n->nr_partial) | |
f9f58285 FF |
4019 | pr_err("SLUB %s: %ld partial slabs counted but counter=%ld\n", |
4020 | s->name, count, n->nr_partial); | |
53e15af0 CL |
4021 | |
4022 | if (!(s->flags & SLAB_STORE_USER)) | |
4023 | goto out; | |
4024 | ||
4025 | list_for_each_entry(page, &n->full, lru) { | |
434e245d | 4026 | validate_slab_slab(s, page, map); |
53e15af0 CL |
4027 | count++; |
4028 | } | |
4029 | if (count != atomic_long_read(&n->nr_slabs)) | |
f9f58285 FF |
4030 | pr_err("SLUB: %s %ld slabs counted but counter=%ld\n", |
4031 | s->name, count, atomic_long_read(&n->nr_slabs)); | |
53e15af0 CL |
4032 | |
4033 | out: | |
4034 | spin_unlock_irqrestore(&n->list_lock, flags); | |
4035 | return count; | |
4036 | } | |
4037 | ||
434e245d | 4038 | static long validate_slab_cache(struct kmem_cache *s) |
53e15af0 CL |
4039 | { |
4040 | int node; | |
4041 | unsigned long count = 0; | |
205ab99d | 4042 | unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) * |
434e245d | 4043 | sizeof(unsigned long), GFP_KERNEL); |
fa45dc25 | 4044 | struct kmem_cache_node *n; |
434e245d CL |
4045 | |
4046 | if (!map) | |
4047 | return -ENOMEM; | |
53e15af0 CL |
4048 | |
4049 | flush_all(s); | |
fa45dc25 | 4050 | for_each_kmem_cache_node(s, node, n) |
434e245d | 4051 | count += validate_slab_node(s, n, map); |
434e245d | 4052 | kfree(map); |
53e15af0 CL |
4053 | return count; |
4054 | } | |
88a420e4 | 4055 | /* |
672bba3a | 4056 | * Generate lists of code addresses where slabcache objects are allocated |
88a420e4 CL |
4057 | * and freed. |
4058 | */ | |
4059 | ||
4060 | struct location { | |
4061 | unsigned long count; | |
ce71e27c | 4062 | unsigned long addr; |
45edfa58 CL |
4063 | long long sum_time; |
4064 | long min_time; | |
4065 | long max_time; | |
4066 | long min_pid; | |
4067 | long max_pid; | |
174596a0 | 4068 | DECLARE_BITMAP(cpus, NR_CPUS); |
45edfa58 | 4069 | nodemask_t nodes; |
88a420e4 CL |
4070 | }; |
4071 | ||
4072 | struct loc_track { | |
4073 | unsigned long max; | |
4074 | unsigned long count; | |
4075 | struct location *loc; | |
4076 | }; | |
4077 | ||
4078 | static void free_loc_track(struct loc_track *t) | |
4079 | { | |
4080 | if (t->max) | |
4081 | free_pages((unsigned long)t->loc, | |
4082 | get_order(sizeof(struct location) * t->max)); | |
4083 | } | |
4084 | ||
68dff6a9 | 4085 | static int alloc_loc_track(struct loc_track *t, unsigned long max, gfp_t flags) |
88a420e4 CL |
4086 | { |
4087 | struct location *l; | |
4088 | int order; | |
4089 | ||
88a420e4 CL |
4090 | order = get_order(sizeof(struct location) * max); |
4091 | ||
68dff6a9 | 4092 | l = (void *)__get_free_pages(flags, order); |
88a420e4 CL |
4093 | if (!l) |
4094 | return 0; | |
4095 | ||
4096 | if (t->count) { | |
4097 | memcpy(l, t->loc, sizeof(struct location) * t->count); | |
4098 | free_loc_track(t); | |
4099 | } | |
4100 | t->max = max; | |
4101 | t->loc = l; | |
4102 | return 1; | |
4103 | } | |
4104 | ||
4105 | static int add_location(struct loc_track *t, struct kmem_cache *s, | |
45edfa58 | 4106 | const struct track *track) |
88a420e4 CL |
4107 | { |
4108 | long start, end, pos; | |
4109 | struct location *l; | |
ce71e27c | 4110 | unsigned long caddr; |
45edfa58 | 4111 | unsigned long age = jiffies - track->when; |
88a420e4 CL |
4112 | |
4113 | start = -1; | |
4114 | end = t->count; | |
4115 | ||
4116 | for ( ; ; ) { | |
4117 | pos = start + (end - start + 1) / 2; | |
4118 | ||
4119 | /* | |
4120 | * There is nothing at "end". If we end up there | |
4121 | * we need to add something to before end. | |
4122 | */ | |
4123 | if (pos == end) | |
4124 | break; | |
4125 | ||
4126 | caddr = t->loc[pos].addr; | |
45edfa58 CL |
4127 | if (track->addr == caddr) { |
4128 | ||
4129 | l = &t->loc[pos]; | |
4130 | l->count++; | |
4131 | if (track->when) { | |
4132 | l->sum_time += age; | |
4133 | if (age < l->min_time) | |
4134 | l->min_time = age; | |
4135 | if (age > l->max_time) | |
4136 | l->max_time = age; | |
4137 | ||
4138 | if (track->pid < l->min_pid) | |
4139 | l->min_pid = track->pid; | |
4140 | if (track->pid > l->max_pid) | |
4141 | l->max_pid = track->pid; | |
4142 | ||
174596a0 RR |
4143 | cpumask_set_cpu(track->cpu, |
4144 | to_cpumask(l->cpus)); | |
45edfa58 CL |
4145 | } |
4146 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
88a420e4 CL |
4147 | return 1; |
4148 | } | |
4149 | ||
45edfa58 | 4150 | if (track->addr < caddr) |
88a420e4 CL |
4151 | end = pos; |
4152 | else | |
4153 | start = pos; | |
4154 | } | |
4155 | ||
4156 | /* | |
672bba3a | 4157 | * Not found. Insert new tracking element. |
88a420e4 | 4158 | */ |
68dff6a9 | 4159 | if (t->count >= t->max && !alloc_loc_track(t, 2 * t->max, GFP_ATOMIC)) |
88a420e4 CL |
4160 | return 0; |
4161 | ||
4162 | l = t->loc + pos; | |
4163 | if (pos < t->count) | |
4164 | memmove(l + 1, l, | |
4165 | (t->count - pos) * sizeof(struct location)); | |
4166 | t->count++; | |
4167 | l->count = 1; | |
45edfa58 CL |
4168 | l->addr = track->addr; |
4169 | l->sum_time = age; | |
4170 | l->min_time = age; | |
4171 | l->max_time = age; | |
4172 | l->min_pid = track->pid; | |
4173 | l->max_pid = track->pid; | |
174596a0 RR |
4174 | cpumask_clear(to_cpumask(l->cpus)); |
4175 | cpumask_set_cpu(track->cpu, to_cpumask(l->cpus)); | |
45edfa58 CL |
4176 | nodes_clear(l->nodes); |
4177 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
88a420e4 CL |
4178 | return 1; |
4179 | } | |
4180 | ||
4181 | static void process_slab(struct loc_track *t, struct kmem_cache *s, | |
bbd7d57b | 4182 | struct page *page, enum track_item alloc, |
a5dd5c11 | 4183 | unsigned long *map) |
88a420e4 | 4184 | { |
a973e9dd | 4185 | void *addr = page_address(page); |
88a420e4 CL |
4186 | void *p; |
4187 | ||
39b26464 | 4188 | bitmap_zero(map, page->objects); |
5f80b13a | 4189 | get_map(s, page, map); |
88a420e4 | 4190 | |
224a88be | 4191 | for_each_object(p, s, addr, page->objects) |
45edfa58 CL |
4192 | if (!test_bit(slab_index(p, s, addr), map)) |
4193 | add_location(t, s, get_track(s, p, alloc)); | |
88a420e4 CL |
4194 | } |
4195 | ||
4196 | static int list_locations(struct kmem_cache *s, char *buf, | |
4197 | enum track_item alloc) | |
4198 | { | |
e374d483 | 4199 | int len = 0; |
88a420e4 | 4200 | unsigned long i; |
68dff6a9 | 4201 | struct loc_track t = { 0, 0, NULL }; |
88a420e4 | 4202 | int node; |
bbd7d57b ED |
4203 | unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) * |
4204 | sizeof(unsigned long), GFP_KERNEL); | |
fa45dc25 | 4205 | struct kmem_cache_node *n; |
88a420e4 | 4206 | |
bbd7d57b ED |
4207 | if (!map || !alloc_loc_track(&t, PAGE_SIZE / sizeof(struct location), |
4208 | GFP_TEMPORARY)) { | |
4209 | kfree(map); | |
68dff6a9 | 4210 | return sprintf(buf, "Out of memory\n"); |
bbd7d57b | 4211 | } |
88a420e4 CL |
4212 | /* Push back cpu slabs */ |
4213 | flush_all(s); | |
4214 | ||
fa45dc25 | 4215 | for_each_kmem_cache_node(s, node, n) { |
88a420e4 CL |
4216 | unsigned long flags; |
4217 | struct page *page; | |
4218 | ||
9e86943b | 4219 | if (!atomic_long_read(&n->nr_slabs)) |
88a420e4 CL |
4220 | continue; |
4221 | ||
4222 | spin_lock_irqsave(&n->list_lock, flags); | |
4223 | list_for_each_entry(page, &n->partial, lru) | |
bbd7d57b | 4224 | process_slab(&t, s, page, alloc, map); |
88a420e4 | 4225 | list_for_each_entry(page, &n->full, lru) |
bbd7d57b | 4226 | process_slab(&t, s, page, alloc, map); |
88a420e4 CL |
4227 | spin_unlock_irqrestore(&n->list_lock, flags); |
4228 | } | |
4229 | ||
4230 | for (i = 0; i < t.count; i++) { | |
45edfa58 | 4231 | struct location *l = &t.loc[i]; |
88a420e4 | 4232 | |
9c246247 | 4233 | if (len > PAGE_SIZE - KSYM_SYMBOL_LEN - 100) |
88a420e4 | 4234 | break; |
e374d483 | 4235 | len += sprintf(buf + len, "%7ld ", l->count); |
45edfa58 CL |
4236 | |
4237 | if (l->addr) | |
62c70bce | 4238 | len += sprintf(buf + len, "%pS", (void *)l->addr); |
88a420e4 | 4239 | else |
e374d483 | 4240 | len += sprintf(buf + len, "<not-available>"); |
45edfa58 CL |
4241 | |
4242 | if (l->sum_time != l->min_time) { | |
e374d483 | 4243 | len += sprintf(buf + len, " age=%ld/%ld/%ld", |
f8bd2258 RZ |
4244 | l->min_time, |
4245 | (long)div_u64(l->sum_time, l->count), | |
4246 | l->max_time); | |
45edfa58 | 4247 | } else |
e374d483 | 4248 | len += sprintf(buf + len, " age=%ld", |
45edfa58 CL |
4249 | l->min_time); |
4250 | ||
4251 | if (l->min_pid != l->max_pid) | |
e374d483 | 4252 | len += sprintf(buf + len, " pid=%ld-%ld", |
45edfa58 CL |
4253 | l->min_pid, l->max_pid); |
4254 | else | |
e374d483 | 4255 | len += sprintf(buf + len, " pid=%ld", |
45edfa58 CL |
4256 | l->min_pid); |
4257 | ||
174596a0 RR |
4258 | if (num_online_cpus() > 1 && |
4259 | !cpumask_empty(to_cpumask(l->cpus)) && | |
5024c1d7 TH |
4260 | len < PAGE_SIZE - 60) |
4261 | len += scnprintf(buf + len, PAGE_SIZE - len - 50, | |
4262 | " cpus=%*pbl", | |
4263 | cpumask_pr_args(to_cpumask(l->cpus))); | |
45edfa58 | 4264 | |
62bc62a8 | 4265 | if (nr_online_nodes > 1 && !nodes_empty(l->nodes) && |
5024c1d7 TH |
4266 | len < PAGE_SIZE - 60) |
4267 | len += scnprintf(buf + len, PAGE_SIZE - len - 50, | |
4268 | " nodes=%*pbl", | |
4269 | nodemask_pr_args(&l->nodes)); | |
45edfa58 | 4270 | |
e374d483 | 4271 | len += sprintf(buf + len, "\n"); |
88a420e4 CL |
4272 | } |
4273 | ||
4274 | free_loc_track(&t); | |
bbd7d57b | 4275 | kfree(map); |
88a420e4 | 4276 | if (!t.count) |
e374d483 HH |
4277 | len += sprintf(buf, "No data\n"); |
4278 | return len; | |
88a420e4 | 4279 | } |
ab4d5ed5 | 4280 | #endif |
88a420e4 | 4281 | |
a5a84755 | 4282 | #ifdef SLUB_RESILIENCY_TEST |
c07b8183 | 4283 | static void __init resiliency_test(void) |
a5a84755 CL |
4284 | { |
4285 | u8 *p; | |
4286 | ||
95a05b42 | 4287 | BUILD_BUG_ON(KMALLOC_MIN_SIZE > 16 || KMALLOC_SHIFT_HIGH < 10); |
a5a84755 | 4288 | |
f9f58285 FF |
4289 | pr_err("SLUB resiliency testing\n"); |
4290 | pr_err("-----------------------\n"); | |
4291 | pr_err("A. Corruption after allocation\n"); | |
a5a84755 CL |
4292 | |
4293 | p = kzalloc(16, GFP_KERNEL); | |
4294 | p[16] = 0x12; | |
f9f58285 FF |
4295 | pr_err("\n1. kmalloc-16: Clobber Redzone/next pointer 0x12->0x%p\n\n", |
4296 | p + 16); | |
a5a84755 CL |
4297 | |
4298 | validate_slab_cache(kmalloc_caches[4]); | |
4299 | ||
4300 | /* Hmmm... The next two are dangerous */ | |
4301 | p = kzalloc(32, GFP_KERNEL); | |
4302 | p[32 + sizeof(void *)] = 0x34; | |
f9f58285 FF |
4303 | pr_err("\n2. kmalloc-32: Clobber next pointer/next slab 0x34 -> -0x%p\n", |
4304 | p); | |
4305 | pr_err("If allocated object is overwritten then not detectable\n\n"); | |
a5a84755 CL |
4306 | |
4307 | validate_slab_cache(kmalloc_caches[5]); | |
4308 | p = kzalloc(64, GFP_KERNEL); | |
4309 | p += 64 + (get_cycles() & 0xff) * sizeof(void *); | |
4310 | *p = 0x56; | |
f9f58285 FF |
4311 | pr_err("\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n", |
4312 | p); | |
4313 | pr_err("If allocated object is overwritten then not detectable\n\n"); | |
a5a84755 CL |
4314 | validate_slab_cache(kmalloc_caches[6]); |
4315 | ||
f9f58285 | 4316 | pr_err("\nB. Corruption after free\n"); |
a5a84755 CL |
4317 | p = kzalloc(128, GFP_KERNEL); |
4318 | kfree(p); | |
4319 | *p = 0x78; | |
f9f58285 | 4320 | pr_err("1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p); |
a5a84755 CL |
4321 | validate_slab_cache(kmalloc_caches[7]); |
4322 | ||
4323 | p = kzalloc(256, GFP_KERNEL); | |
4324 | kfree(p); | |
4325 | p[50] = 0x9a; | |
f9f58285 | 4326 | pr_err("\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n", p); |
a5a84755 CL |
4327 | validate_slab_cache(kmalloc_caches[8]); |
4328 | ||
4329 | p = kzalloc(512, GFP_KERNEL); | |
4330 | kfree(p); | |
4331 | p[512] = 0xab; | |
f9f58285 | 4332 | pr_err("\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p); |
a5a84755 CL |
4333 | validate_slab_cache(kmalloc_caches[9]); |
4334 | } | |
4335 | #else | |
4336 | #ifdef CONFIG_SYSFS | |
4337 | static void resiliency_test(void) {}; | |
4338 | #endif | |
4339 | #endif | |
4340 | ||
ab4d5ed5 | 4341 | #ifdef CONFIG_SYSFS |
81819f0f | 4342 | enum slab_stat_type { |
205ab99d CL |
4343 | SL_ALL, /* All slabs */ |
4344 | SL_PARTIAL, /* Only partially allocated slabs */ | |
4345 | SL_CPU, /* Only slabs used for cpu caches */ | |
4346 | SL_OBJECTS, /* Determine allocated objects not slabs */ | |
4347 | SL_TOTAL /* Determine object capacity not slabs */ | |
81819f0f CL |
4348 | }; |
4349 | ||
205ab99d | 4350 | #define SO_ALL (1 << SL_ALL) |
81819f0f CL |
4351 | #define SO_PARTIAL (1 << SL_PARTIAL) |
4352 | #define SO_CPU (1 << SL_CPU) | |
4353 | #define SO_OBJECTS (1 << SL_OBJECTS) | |
205ab99d | 4354 | #define SO_TOTAL (1 << SL_TOTAL) |
81819f0f | 4355 | |
62e5c4b4 CG |
4356 | static ssize_t show_slab_objects(struct kmem_cache *s, |
4357 | char *buf, unsigned long flags) | |
81819f0f CL |
4358 | { |
4359 | unsigned long total = 0; | |
81819f0f CL |
4360 | int node; |
4361 | int x; | |
4362 | unsigned long *nodes; | |
81819f0f | 4363 | |
e35e1a97 | 4364 | nodes = kzalloc(sizeof(unsigned long) * nr_node_ids, GFP_KERNEL); |
62e5c4b4 CG |
4365 | if (!nodes) |
4366 | return -ENOMEM; | |
81819f0f | 4367 | |
205ab99d CL |
4368 | if (flags & SO_CPU) { |
4369 | int cpu; | |
81819f0f | 4370 | |
205ab99d | 4371 | for_each_possible_cpu(cpu) { |
d0e0ac97 CG |
4372 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, |
4373 | cpu); | |
ec3ab083 | 4374 | int node; |
49e22585 | 4375 | struct page *page; |
dfb4f096 | 4376 | |
4db0c3c2 | 4377 | page = READ_ONCE(c->page); |
ec3ab083 CL |
4378 | if (!page) |
4379 | continue; | |
205ab99d | 4380 | |
ec3ab083 CL |
4381 | node = page_to_nid(page); |
4382 | if (flags & SO_TOTAL) | |
4383 | x = page->objects; | |
4384 | else if (flags & SO_OBJECTS) | |
4385 | x = page->inuse; | |
4386 | else | |
4387 | x = 1; | |
49e22585 | 4388 | |
ec3ab083 CL |
4389 | total += x; |
4390 | nodes[node] += x; | |
4391 | ||
4db0c3c2 | 4392 | page = READ_ONCE(c->partial); |
49e22585 | 4393 | if (page) { |
8afb1474 LZ |
4394 | node = page_to_nid(page); |
4395 | if (flags & SO_TOTAL) | |
4396 | WARN_ON_ONCE(1); | |
4397 | else if (flags & SO_OBJECTS) | |
4398 | WARN_ON_ONCE(1); | |
4399 | else | |
4400 | x = page->pages; | |
bc6697d8 ED |
4401 | total += x; |
4402 | nodes[node] += x; | |
49e22585 | 4403 | } |
81819f0f CL |
4404 | } |
4405 | } | |
4406 | ||
bfc8c901 | 4407 | get_online_mems(); |
ab4d5ed5 | 4408 | #ifdef CONFIG_SLUB_DEBUG |
205ab99d | 4409 | if (flags & SO_ALL) { |
fa45dc25 CL |
4410 | struct kmem_cache_node *n; |
4411 | ||
4412 | for_each_kmem_cache_node(s, node, n) { | |
205ab99d | 4413 | |
d0e0ac97 CG |
4414 | if (flags & SO_TOTAL) |
4415 | x = atomic_long_read(&n->total_objects); | |
4416 | else if (flags & SO_OBJECTS) | |
4417 | x = atomic_long_read(&n->total_objects) - | |
4418 | count_partial(n, count_free); | |
81819f0f | 4419 | else |
205ab99d | 4420 | x = atomic_long_read(&n->nr_slabs); |
81819f0f CL |
4421 | total += x; |
4422 | nodes[node] += x; | |
4423 | } | |
4424 | ||
ab4d5ed5 CL |
4425 | } else |
4426 | #endif | |
4427 | if (flags & SO_PARTIAL) { | |
fa45dc25 | 4428 | struct kmem_cache_node *n; |
81819f0f | 4429 | |
fa45dc25 | 4430 | for_each_kmem_cache_node(s, node, n) { |
205ab99d CL |
4431 | if (flags & SO_TOTAL) |
4432 | x = count_partial(n, count_total); | |
4433 | else if (flags & SO_OBJECTS) | |
4434 | x = count_partial(n, count_inuse); | |
81819f0f | 4435 | else |
205ab99d | 4436 | x = n->nr_partial; |
81819f0f CL |
4437 | total += x; |
4438 | nodes[node] += x; | |
4439 | } | |
4440 | } | |
81819f0f CL |
4441 | x = sprintf(buf, "%lu", total); |
4442 | #ifdef CONFIG_NUMA | |
fa45dc25 | 4443 | for (node = 0; node < nr_node_ids; node++) |
81819f0f CL |
4444 | if (nodes[node]) |
4445 | x += sprintf(buf + x, " N%d=%lu", | |
4446 | node, nodes[node]); | |
4447 | #endif | |
bfc8c901 | 4448 | put_online_mems(); |
81819f0f CL |
4449 | kfree(nodes); |
4450 | return x + sprintf(buf + x, "\n"); | |
4451 | } | |
4452 | ||
ab4d5ed5 | 4453 | #ifdef CONFIG_SLUB_DEBUG |
81819f0f CL |
4454 | static int any_slab_objects(struct kmem_cache *s) |
4455 | { | |
4456 | int node; | |
fa45dc25 | 4457 | struct kmem_cache_node *n; |
81819f0f | 4458 | |
fa45dc25 | 4459 | for_each_kmem_cache_node(s, node, n) |
4ea33e2d | 4460 | if (atomic_long_read(&n->total_objects)) |
81819f0f | 4461 | return 1; |
fa45dc25 | 4462 | |
81819f0f CL |
4463 | return 0; |
4464 | } | |
ab4d5ed5 | 4465 | #endif |
81819f0f CL |
4466 | |
4467 | #define to_slab_attr(n) container_of(n, struct slab_attribute, attr) | |
497888cf | 4468 | #define to_slab(n) container_of(n, struct kmem_cache, kobj) |
81819f0f CL |
4469 | |
4470 | struct slab_attribute { | |
4471 | struct attribute attr; | |
4472 | ssize_t (*show)(struct kmem_cache *s, char *buf); | |
4473 | ssize_t (*store)(struct kmem_cache *s, const char *x, size_t count); | |
4474 | }; | |
4475 | ||
4476 | #define SLAB_ATTR_RO(_name) \ | |
ab067e99 VK |
4477 | static struct slab_attribute _name##_attr = \ |
4478 | __ATTR(_name, 0400, _name##_show, NULL) | |
81819f0f CL |
4479 | |
4480 | #define SLAB_ATTR(_name) \ | |
4481 | static struct slab_attribute _name##_attr = \ | |
ab067e99 | 4482 | __ATTR(_name, 0600, _name##_show, _name##_store) |
81819f0f | 4483 | |
81819f0f CL |
4484 | static ssize_t slab_size_show(struct kmem_cache *s, char *buf) |
4485 | { | |
4486 | return sprintf(buf, "%d\n", s->size); | |
4487 | } | |
4488 | SLAB_ATTR_RO(slab_size); | |
4489 | ||
4490 | static ssize_t align_show(struct kmem_cache *s, char *buf) | |
4491 | { | |
4492 | return sprintf(buf, "%d\n", s->align); | |
4493 | } | |
4494 | SLAB_ATTR_RO(align); | |
4495 | ||
4496 | static ssize_t object_size_show(struct kmem_cache *s, char *buf) | |
4497 | { | |
3b0efdfa | 4498 | return sprintf(buf, "%d\n", s->object_size); |
81819f0f CL |
4499 | } |
4500 | SLAB_ATTR_RO(object_size); | |
4501 | ||
4502 | static ssize_t objs_per_slab_show(struct kmem_cache *s, char *buf) | |
4503 | { | |
834f3d11 | 4504 | return sprintf(buf, "%d\n", oo_objects(s->oo)); |
81819f0f CL |
4505 | } |
4506 | SLAB_ATTR_RO(objs_per_slab); | |
4507 | ||
06b285dc CL |
4508 | static ssize_t order_store(struct kmem_cache *s, |
4509 | const char *buf, size_t length) | |
4510 | { | |
0121c619 CL |
4511 | unsigned long order; |
4512 | int err; | |
4513 | ||
3dbb95f7 | 4514 | err = kstrtoul(buf, 10, &order); |
0121c619 CL |
4515 | if (err) |
4516 | return err; | |
06b285dc CL |
4517 | |
4518 | if (order > slub_max_order || order < slub_min_order) | |
4519 | return -EINVAL; | |
4520 | ||
4521 | calculate_sizes(s, order); | |
4522 | return length; | |
4523 | } | |
4524 | ||
81819f0f CL |
4525 | static ssize_t order_show(struct kmem_cache *s, char *buf) |
4526 | { | |
834f3d11 | 4527 | return sprintf(buf, "%d\n", oo_order(s->oo)); |
81819f0f | 4528 | } |
06b285dc | 4529 | SLAB_ATTR(order); |
81819f0f | 4530 | |
73d342b1 DR |
4531 | static ssize_t min_partial_show(struct kmem_cache *s, char *buf) |
4532 | { | |
4533 | return sprintf(buf, "%lu\n", s->min_partial); | |
4534 | } | |
4535 | ||
4536 | static ssize_t min_partial_store(struct kmem_cache *s, const char *buf, | |
4537 | size_t length) | |
4538 | { | |
4539 | unsigned long min; | |
4540 | int err; | |
4541 | ||
3dbb95f7 | 4542 | err = kstrtoul(buf, 10, &min); |
73d342b1 DR |
4543 | if (err) |
4544 | return err; | |
4545 | ||
c0bdb232 | 4546 | set_min_partial(s, min); |
73d342b1 DR |
4547 | return length; |
4548 | } | |
4549 | SLAB_ATTR(min_partial); | |
4550 | ||
49e22585 CL |
4551 | static ssize_t cpu_partial_show(struct kmem_cache *s, char *buf) |
4552 | { | |
4553 | return sprintf(buf, "%u\n", s->cpu_partial); | |
4554 | } | |
4555 | ||
4556 | static ssize_t cpu_partial_store(struct kmem_cache *s, const char *buf, | |
4557 | size_t length) | |
4558 | { | |
4559 | unsigned long objects; | |
4560 | int err; | |
4561 | ||
3dbb95f7 | 4562 | err = kstrtoul(buf, 10, &objects); |
49e22585 CL |
4563 | if (err) |
4564 | return err; | |
345c905d | 4565 | if (objects && !kmem_cache_has_cpu_partial(s)) |
74ee4ef1 | 4566 | return -EINVAL; |
49e22585 CL |
4567 | |
4568 | s->cpu_partial = objects; | |
4569 | flush_all(s); | |
4570 | return length; | |
4571 | } | |
4572 | SLAB_ATTR(cpu_partial); | |
4573 | ||
81819f0f CL |
4574 | static ssize_t ctor_show(struct kmem_cache *s, char *buf) |
4575 | { | |
62c70bce JP |
4576 | if (!s->ctor) |
4577 | return 0; | |
4578 | return sprintf(buf, "%pS\n", s->ctor); | |
81819f0f CL |
4579 | } |
4580 | SLAB_ATTR_RO(ctor); | |
4581 | ||
81819f0f CL |
4582 | static ssize_t aliases_show(struct kmem_cache *s, char *buf) |
4583 | { | |
4307c14f | 4584 | return sprintf(buf, "%d\n", s->refcount < 0 ? 0 : s->refcount - 1); |
81819f0f CL |
4585 | } |
4586 | SLAB_ATTR_RO(aliases); | |
4587 | ||
81819f0f CL |
4588 | static ssize_t partial_show(struct kmem_cache *s, char *buf) |
4589 | { | |
d9acf4b7 | 4590 | return show_slab_objects(s, buf, SO_PARTIAL); |
81819f0f CL |
4591 | } |
4592 | SLAB_ATTR_RO(partial); | |
4593 | ||
4594 | static ssize_t cpu_slabs_show(struct kmem_cache *s, char *buf) | |
4595 | { | |
d9acf4b7 | 4596 | return show_slab_objects(s, buf, SO_CPU); |
81819f0f CL |
4597 | } |
4598 | SLAB_ATTR_RO(cpu_slabs); | |
4599 | ||
4600 | static ssize_t objects_show(struct kmem_cache *s, char *buf) | |
4601 | { | |
205ab99d | 4602 | return show_slab_objects(s, buf, SO_ALL|SO_OBJECTS); |
81819f0f CL |
4603 | } |
4604 | SLAB_ATTR_RO(objects); | |
4605 | ||
205ab99d CL |
4606 | static ssize_t objects_partial_show(struct kmem_cache *s, char *buf) |
4607 | { | |
4608 | return show_slab_objects(s, buf, SO_PARTIAL|SO_OBJECTS); | |
4609 | } | |
4610 | SLAB_ATTR_RO(objects_partial); | |
4611 | ||
49e22585 CL |
4612 | static ssize_t slabs_cpu_partial_show(struct kmem_cache *s, char *buf) |
4613 | { | |
4614 | int objects = 0; | |
4615 | int pages = 0; | |
4616 | int cpu; | |
4617 | int len; | |
4618 | ||
4619 | for_each_online_cpu(cpu) { | |
4620 | struct page *page = per_cpu_ptr(s->cpu_slab, cpu)->partial; | |
4621 | ||
4622 | if (page) { | |
4623 | pages += page->pages; | |
4624 | objects += page->pobjects; | |
4625 | } | |
4626 | } | |
4627 | ||
4628 | len = sprintf(buf, "%d(%d)", objects, pages); | |
4629 | ||
4630 | #ifdef CONFIG_SMP | |
4631 | for_each_online_cpu(cpu) { | |
4632 | struct page *page = per_cpu_ptr(s->cpu_slab, cpu) ->partial; | |
4633 | ||
4634 | if (page && len < PAGE_SIZE - 20) | |
4635 | len += sprintf(buf + len, " C%d=%d(%d)", cpu, | |
4636 | page->pobjects, page->pages); | |
4637 | } | |
4638 | #endif | |
4639 | return len + sprintf(buf + len, "\n"); | |
4640 | } | |
4641 | SLAB_ATTR_RO(slabs_cpu_partial); | |
4642 | ||
a5a84755 CL |
4643 | static ssize_t reclaim_account_show(struct kmem_cache *s, char *buf) |
4644 | { | |
4645 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_RECLAIM_ACCOUNT)); | |
4646 | } | |
4647 | ||
4648 | static ssize_t reclaim_account_store(struct kmem_cache *s, | |
4649 | const char *buf, size_t length) | |
4650 | { | |
4651 | s->flags &= ~SLAB_RECLAIM_ACCOUNT; | |
4652 | if (buf[0] == '1') | |
4653 | s->flags |= SLAB_RECLAIM_ACCOUNT; | |
4654 | return length; | |
4655 | } | |
4656 | SLAB_ATTR(reclaim_account); | |
4657 | ||
4658 | static ssize_t hwcache_align_show(struct kmem_cache *s, char *buf) | |
4659 | { | |
4660 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_HWCACHE_ALIGN)); | |
4661 | } | |
4662 | SLAB_ATTR_RO(hwcache_align); | |
4663 | ||
4664 | #ifdef CONFIG_ZONE_DMA | |
4665 | static ssize_t cache_dma_show(struct kmem_cache *s, char *buf) | |
4666 | { | |
4667 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_CACHE_DMA)); | |
4668 | } | |
4669 | SLAB_ATTR_RO(cache_dma); | |
4670 | #endif | |
4671 | ||
4672 | static ssize_t destroy_by_rcu_show(struct kmem_cache *s, char *buf) | |
4673 | { | |
4674 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_DESTROY_BY_RCU)); | |
4675 | } | |
4676 | SLAB_ATTR_RO(destroy_by_rcu); | |
4677 | ||
ab9a0f19 LJ |
4678 | static ssize_t reserved_show(struct kmem_cache *s, char *buf) |
4679 | { | |
4680 | return sprintf(buf, "%d\n", s->reserved); | |
4681 | } | |
4682 | SLAB_ATTR_RO(reserved); | |
4683 | ||
ab4d5ed5 | 4684 | #ifdef CONFIG_SLUB_DEBUG |
a5a84755 CL |
4685 | static ssize_t slabs_show(struct kmem_cache *s, char *buf) |
4686 | { | |
4687 | return show_slab_objects(s, buf, SO_ALL); | |
4688 | } | |
4689 | SLAB_ATTR_RO(slabs); | |
4690 | ||
205ab99d CL |
4691 | static ssize_t total_objects_show(struct kmem_cache *s, char *buf) |
4692 | { | |
4693 | return show_slab_objects(s, buf, SO_ALL|SO_TOTAL); | |
4694 | } | |
4695 | SLAB_ATTR_RO(total_objects); | |
4696 | ||
81819f0f CL |
4697 | static ssize_t sanity_checks_show(struct kmem_cache *s, char *buf) |
4698 | { | |
4699 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_DEBUG_FREE)); | |
4700 | } | |
4701 | ||
4702 | static ssize_t sanity_checks_store(struct kmem_cache *s, | |
4703 | const char *buf, size_t length) | |
4704 | { | |
4705 | s->flags &= ~SLAB_DEBUG_FREE; | |
b789ef51 CL |
4706 | if (buf[0] == '1') { |
4707 | s->flags &= ~__CMPXCHG_DOUBLE; | |
81819f0f | 4708 | s->flags |= SLAB_DEBUG_FREE; |
b789ef51 | 4709 | } |
81819f0f CL |
4710 | return length; |
4711 | } | |
4712 | SLAB_ATTR(sanity_checks); | |
4713 | ||
4714 | static ssize_t trace_show(struct kmem_cache *s, char *buf) | |
4715 | { | |
4716 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_TRACE)); | |
4717 | } | |
4718 | ||
4719 | static ssize_t trace_store(struct kmem_cache *s, const char *buf, | |
4720 | size_t length) | |
4721 | { | |
c9e16131 CL |
4722 | /* |
4723 | * Tracing a merged cache is going to give confusing results | |
4724 | * as well as cause other issues like converting a mergeable | |
4725 | * cache into an umergeable one. | |
4726 | */ | |
4727 | if (s->refcount > 1) | |
4728 | return -EINVAL; | |
4729 | ||
81819f0f | 4730 | s->flags &= ~SLAB_TRACE; |
b789ef51 CL |
4731 | if (buf[0] == '1') { |
4732 | s->flags &= ~__CMPXCHG_DOUBLE; | |
81819f0f | 4733 | s->flags |= SLAB_TRACE; |
b789ef51 | 4734 | } |
81819f0f CL |
4735 | return length; |
4736 | } | |
4737 | SLAB_ATTR(trace); | |
4738 | ||
81819f0f CL |
4739 | static ssize_t red_zone_show(struct kmem_cache *s, char *buf) |
4740 | { | |
4741 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_RED_ZONE)); | |
4742 | } | |
4743 | ||
4744 | static ssize_t red_zone_store(struct kmem_cache *s, | |
4745 | const char *buf, size_t length) | |
4746 | { | |
4747 | if (any_slab_objects(s)) | |
4748 | return -EBUSY; | |
4749 | ||
4750 | s->flags &= ~SLAB_RED_ZONE; | |
b789ef51 CL |
4751 | if (buf[0] == '1') { |
4752 | s->flags &= ~__CMPXCHG_DOUBLE; | |
81819f0f | 4753 | s->flags |= SLAB_RED_ZONE; |
b789ef51 | 4754 | } |
06b285dc | 4755 | calculate_sizes(s, -1); |
81819f0f CL |
4756 | return length; |
4757 | } | |
4758 | SLAB_ATTR(red_zone); | |
4759 | ||
4760 | static ssize_t poison_show(struct kmem_cache *s, char *buf) | |
4761 | { | |
4762 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_POISON)); | |
4763 | } | |
4764 | ||
4765 | static ssize_t poison_store(struct kmem_cache *s, | |
4766 | const char *buf, size_t length) | |
4767 | { | |
4768 | if (any_slab_objects(s)) | |
4769 | return -EBUSY; | |
4770 | ||
4771 | s->flags &= ~SLAB_POISON; | |
b789ef51 CL |
4772 | if (buf[0] == '1') { |
4773 | s->flags &= ~__CMPXCHG_DOUBLE; | |
81819f0f | 4774 | s->flags |= SLAB_POISON; |
b789ef51 | 4775 | } |
06b285dc | 4776 | calculate_sizes(s, -1); |
81819f0f CL |
4777 | return length; |
4778 | } | |
4779 | SLAB_ATTR(poison); | |
4780 | ||
4781 | static ssize_t store_user_show(struct kmem_cache *s, char *buf) | |
4782 | { | |
4783 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_STORE_USER)); | |
4784 | } | |
4785 | ||
4786 | static ssize_t store_user_store(struct kmem_cache *s, | |
4787 | const char *buf, size_t length) | |
4788 | { | |
4789 | if (any_slab_objects(s)) | |
4790 | return -EBUSY; | |
4791 | ||
4792 | s->flags &= ~SLAB_STORE_USER; | |
b789ef51 CL |
4793 | if (buf[0] == '1') { |
4794 | s->flags &= ~__CMPXCHG_DOUBLE; | |
81819f0f | 4795 | s->flags |= SLAB_STORE_USER; |
b789ef51 | 4796 | } |
06b285dc | 4797 | calculate_sizes(s, -1); |
81819f0f CL |
4798 | return length; |
4799 | } | |
4800 | SLAB_ATTR(store_user); | |
4801 | ||
53e15af0 CL |
4802 | static ssize_t validate_show(struct kmem_cache *s, char *buf) |
4803 | { | |
4804 | return 0; | |
4805 | } | |
4806 | ||
4807 | static ssize_t validate_store(struct kmem_cache *s, | |
4808 | const char *buf, size_t length) | |
4809 | { | |
434e245d CL |
4810 | int ret = -EINVAL; |
4811 | ||
4812 | if (buf[0] == '1') { | |
4813 | ret = validate_slab_cache(s); | |
4814 | if (ret >= 0) | |
4815 | ret = length; | |
4816 | } | |
4817 | return ret; | |
53e15af0 CL |
4818 | } |
4819 | SLAB_ATTR(validate); | |
a5a84755 CL |
4820 | |
4821 | static ssize_t alloc_calls_show(struct kmem_cache *s, char *buf) | |
4822 | { | |
4823 | if (!(s->flags & SLAB_STORE_USER)) | |
4824 | return -ENOSYS; | |
4825 | return list_locations(s, buf, TRACK_ALLOC); | |
4826 | } | |
4827 | SLAB_ATTR_RO(alloc_calls); | |
4828 | ||
4829 | static ssize_t free_calls_show(struct kmem_cache *s, char *buf) | |
4830 | { | |
4831 | if (!(s->flags & SLAB_STORE_USER)) | |
4832 | return -ENOSYS; | |
4833 | return list_locations(s, buf, TRACK_FREE); | |
4834 | } | |
4835 | SLAB_ATTR_RO(free_calls); | |
4836 | #endif /* CONFIG_SLUB_DEBUG */ | |
4837 | ||
4838 | #ifdef CONFIG_FAILSLAB | |
4839 | static ssize_t failslab_show(struct kmem_cache *s, char *buf) | |
4840 | { | |
4841 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_FAILSLAB)); | |
4842 | } | |
4843 | ||
4844 | static ssize_t failslab_store(struct kmem_cache *s, const char *buf, | |
4845 | size_t length) | |
4846 | { | |
c9e16131 CL |
4847 | if (s->refcount > 1) |
4848 | return -EINVAL; | |
4849 | ||
a5a84755 CL |
4850 | s->flags &= ~SLAB_FAILSLAB; |
4851 | if (buf[0] == '1') | |
4852 | s->flags |= SLAB_FAILSLAB; | |
4853 | return length; | |
4854 | } | |
4855 | SLAB_ATTR(failslab); | |
ab4d5ed5 | 4856 | #endif |
53e15af0 | 4857 | |
2086d26a CL |
4858 | static ssize_t shrink_show(struct kmem_cache *s, char *buf) |
4859 | { | |
4860 | return 0; | |
4861 | } | |
4862 | ||
4863 | static ssize_t shrink_store(struct kmem_cache *s, | |
4864 | const char *buf, size_t length) | |
4865 | { | |
832f37f5 VD |
4866 | if (buf[0] == '1') |
4867 | kmem_cache_shrink(s); | |
4868 | else | |
2086d26a CL |
4869 | return -EINVAL; |
4870 | return length; | |
4871 | } | |
4872 | SLAB_ATTR(shrink); | |
4873 | ||
81819f0f | 4874 | #ifdef CONFIG_NUMA |
9824601e | 4875 | static ssize_t remote_node_defrag_ratio_show(struct kmem_cache *s, char *buf) |
81819f0f | 4876 | { |
9824601e | 4877 | return sprintf(buf, "%d\n", s->remote_node_defrag_ratio / 10); |
81819f0f CL |
4878 | } |
4879 | ||
9824601e | 4880 | static ssize_t remote_node_defrag_ratio_store(struct kmem_cache *s, |
81819f0f CL |
4881 | const char *buf, size_t length) |
4882 | { | |
0121c619 CL |
4883 | unsigned long ratio; |
4884 | int err; | |
4885 | ||
3dbb95f7 | 4886 | err = kstrtoul(buf, 10, &ratio); |
0121c619 CL |
4887 | if (err) |
4888 | return err; | |
4889 | ||
e2cb96b7 | 4890 | if (ratio <= 100) |
0121c619 | 4891 | s->remote_node_defrag_ratio = ratio * 10; |
81819f0f | 4892 | |
81819f0f CL |
4893 | return length; |
4894 | } | |
9824601e | 4895 | SLAB_ATTR(remote_node_defrag_ratio); |
81819f0f CL |
4896 | #endif |
4897 | ||
8ff12cfc | 4898 | #ifdef CONFIG_SLUB_STATS |
8ff12cfc CL |
4899 | static int show_stat(struct kmem_cache *s, char *buf, enum stat_item si) |
4900 | { | |
4901 | unsigned long sum = 0; | |
4902 | int cpu; | |
4903 | int len; | |
4904 | int *data = kmalloc(nr_cpu_ids * sizeof(int), GFP_KERNEL); | |
4905 | ||
4906 | if (!data) | |
4907 | return -ENOMEM; | |
4908 | ||
4909 | for_each_online_cpu(cpu) { | |
9dfc6e68 | 4910 | unsigned x = per_cpu_ptr(s->cpu_slab, cpu)->stat[si]; |
8ff12cfc CL |
4911 | |
4912 | data[cpu] = x; | |
4913 | sum += x; | |
4914 | } | |
4915 | ||
4916 | len = sprintf(buf, "%lu", sum); | |
4917 | ||
50ef37b9 | 4918 | #ifdef CONFIG_SMP |
8ff12cfc CL |
4919 | for_each_online_cpu(cpu) { |
4920 | if (data[cpu] && len < PAGE_SIZE - 20) | |
50ef37b9 | 4921 | len += sprintf(buf + len, " C%d=%u", cpu, data[cpu]); |
8ff12cfc | 4922 | } |
50ef37b9 | 4923 | #endif |
8ff12cfc CL |
4924 | kfree(data); |
4925 | return len + sprintf(buf + len, "\n"); | |
4926 | } | |
4927 | ||
78eb00cc DR |
4928 | static void clear_stat(struct kmem_cache *s, enum stat_item si) |
4929 | { | |
4930 | int cpu; | |
4931 | ||
4932 | for_each_online_cpu(cpu) | |
9dfc6e68 | 4933 | per_cpu_ptr(s->cpu_slab, cpu)->stat[si] = 0; |
78eb00cc DR |
4934 | } |
4935 | ||
8ff12cfc CL |
4936 | #define STAT_ATTR(si, text) \ |
4937 | static ssize_t text##_show(struct kmem_cache *s, char *buf) \ | |
4938 | { \ | |
4939 | return show_stat(s, buf, si); \ | |
4940 | } \ | |
78eb00cc DR |
4941 | static ssize_t text##_store(struct kmem_cache *s, \ |
4942 | const char *buf, size_t length) \ | |
4943 | { \ | |
4944 | if (buf[0] != '0') \ | |
4945 | return -EINVAL; \ | |
4946 | clear_stat(s, si); \ | |
4947 | return length; \ | |
4948 | } \ | |
4949 | SLAB_ATTR(text); \ | |
8ff12cfc CL |
4950 | |
4951 | STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath); | |
4952 | STAT_ATTR(ALLOC_SLOWPATH, alloc_slowpath); | |
4953 | STAT_ATTR(FREE_FASTPATH, free_fastpath); | |
4954 | STAT_ATTR(FREE_SLOWPATH, free_slowpath); | |
4955 | STAT_ATTR(FREE_FROZEN, free_frozen); | |
4956 | STAT_ATTR(FREE_ADD_PARTIAL, free_add_partial); | |
4957 | STAT_ATTR(FREE_REMOVE_PARTIAL, free_remove_partial); | |
4958 | STAT_ATTR(ALLOC_FROM_PARTIAL, alloc_from_partial); | |
4959 | STAT_ATTR(ALLOC_SLAB, alloc_slab); | |
4960 | STAT_ATTR(ALLOC_REFILL, alloc_refill); | |
e36a2652 | 4961 | STAT_ATTR(ALLOC_NODE_MISMATCH, alloc_node_mismatch); |
8ff12cfc CL |
4962 | STAT_ATTR(FREE_SLAB, free_slab); |
4963 | STAT_ATTR(CPUSLAB_FLUSH, cpuslab_flush); | |
4964 | STAT_ATTR(DEACTIVATE_FULL, deactivate_full); | |
4965 | STAT_ATTR(DEACTIVATE_EMPTY, deactivate_empty); | |
4966 | STAT_ATTR(DEACTIVATE_TO_HEAD, deactivate_to_head); | |
4967 | STAT_ATTR(DEACTIVATE_TO_TAIL, deactivate_to_tail); | |
4968 | STAT_ATTR(DEACTIVATE_REMOTE_FREES, deactivate_remote_frees); | |
03e404af | 4969 | STAT_ATTR(DEACTIVATE_BYPASS, deactivate_bypass); |
65c3376a | 4970 | STAT_ATTR(ORDER_FALLBACK, order_fallback); |
b789ef51 CL |
4971 | STAT_ATTR(CMPXCHG_DOUBLE_CPU_FAIL, cmpxchg_double_cpu_fail); |
4972 | STAT_ATTR(CMPXCHG_DOUBLE_FAIL, cmpxchg_double_fail); | |
49e22585 CL |
4973 | STAT_ATTR(CPU_PARTIAL_ALLOC, cpu_partial_alloc); |
4974 | STAT_ATTR(CPU_PARTIAL_FREE, cpu_partial_free); | |
8028dcea AS |
4975 | STAT_ATTR(CPU_PARTIAL_NODE, cpu_partial_node); |
4976 | STAT_ATTR(CPU_PARTIAL_DRAIN, cpu_partial_drain); | |
8ff12cfc CL |
4977 | #endif |
4978 | ||
06428780 | 4979 | static struct attribute *slab_attrs[] = { |
81819f0f CL |
4980 | &slab_size_attr.attr, |
4981 | &object_size_attr.attr, | |
4982 | &objs_per_slab_attr.attr, | |
4983 | &order_attr.attr, | |
73d342b1 | 4984 | &min_partial_attr.attr, |
49e22585 | 4985 | &cpu_partial_attr.attr, |
81819f0f | 4986 | &objects_attr.attr, |
205ab99d | 4987 | &objects_partial_attr.attr, |
81819f0f CL |
4988 | &partial_attr.attr, |
4989 | &cpu_slabs_attr.attr, | |
4990 | &ctor_attr.attr, | |
81819f0f CL |
4991 | &aliases_attr.attr, |
4992 | &align_attr.attr, | |
81819f0f CL |
4993 | &hwcache_align_attr.attr, |
4994 | &reclaim_account_attr.attr, | |
4995 | &destroy_by_rcu_attr.attr, | |
a5a84755 | 4996 | &shrink_attr.attr, |
ab9a0f19 | 4997 | &reserved_attr.attr, |
49e22585 | 4998 | &slabs_cpu_partial_attr.attr, |
ab4d5ed5 | 4999 | #ifdef CONFIG_SLUB_DEBUG |
a5a84755 CL |
5000 | &total_objects_attr.attr, |
5001 | &slabs_attr.attr, | |
5002 | &sanity_checks_attr.attr, | |
5003 | &trace_attr.attr, | |
81819f0f CL |
5004 | &red_zone_attr.attr, |
5005 | &poison_attr.attr, | |
5006 | &store_user_attr.attr, | |
53e15af0 | 5007 | &validate_attr.attr, |
88a420e4 CL |
5008 | &alloc_calls_attr.attr, |
5009 | &free_calls_attr.attr, | |
ab4d5ed5 | 5010 | #endif |
81819f0f CL |
5011 | #ifdef CONFIG_ZONE_DMA |
5012 | &cache_dma_attr.attr, | |
5013 | #endif | |
5014 | #ifdef CONFIG_NUMA | |
9824601e | 5015 | &remote_node_defrag_ratio_attr.attr, |
8ff12cfc CL |
5016 | #endif |
5017 | #ifdef CONFIG_SLUB_STATS | |
5018 | &alloc_fastpath_attr.attr, | |
5019 | &alloc_slowpath_attr.attr, | |
5020 | &free_fastpath_attr.attr, | |
5021 | &free_slowpath_attr.attr, | |
5022 | &free_frozen_attr.attr, | |
5023 | &free_add_partial_attr.attr, | |
5024 | &free_remove_partial_attr.attr, | |
5025 | &alloc_from_partial_attr.attr, | |
5026 | &alloc_slab_attr.attr, | |
5027 | &alloc_refill_attr.attr, | |
e36a2652 | 5028 | &alloc_node_mismatch_attr.attr, |
8ff12cfc CL |
5029 | &free_slab_attr.attr, |
5030 | &cpuslab_flush_attr.attr, | |
5031 | &deactivate_full_attr.attr, | |
5032 | &deactivate_empty_attr.attr, | |
5033 | &deactivate_to_head_attr.attr, | |
5034 | &deactivate_to_tail_attr.attr, | |
5035 | &deactivate_remote_frees_attr.attr, | |
03e404af | 5036 | &deactivate_bypass_attr.attr, |
65c3376a | 5037 | &order_fallback_attr.attr, |
b789ef51 CL |
5038 | &cmpxchg_double_fail_attr.attr, |
5039 | &cmpxchg_double_cpu_fail_attr.attr, | |
49e22585 CL |
5040 | &cpu_partial_alloc_attr.attr, |
5041 | &cpu_partial_free_attr.attr, | |
8028dcea AS |
5042 | &cpu_partial_node_attr.attr, |
5043 | &cpu_partial_drain_attr.attr, | |
81819f0f | 5044 | #endif |
4c13dd3b DM |
5045 | #ifdef CONFIG_FAILSLAB |
5046 | &failslab_attr.attr, | |
5047 | #endif | |
5048 | ||
81819f0f CL |
5049 | NULL |
5050 | }; | |
5051 | ||
5052 | static struct attribute_group slab_attr_group = { | |
5053 | .attrs = slab_attrs, | |
5054 | }; | |
5055 | ||
5056 | static ssize_t slab_attr_show(struct kobject *kobj, | |
5057 | struct attribute *attr, | |
5058 | char *buf) | |
5059 | { | |
5060 | struct slab_attribute *attribute; | |
5061 | struct kmem_cache *s; | |
5062 | int err; | |
5063 | ||
5064 | attribute = to_slab_attr(attr); | |
5065 | s = to_slab(kobj); | |
5066 | ||
5067 | if (!attribute->show) | |
5068 | return -EIO; | |
5069 | ||
5070 | err = attribute->show(s, buf); | |
5071 | ||
5072 | return err; | |
5073 | } | |
5074 | ||
5075 | static ssize_t slab_attr_store(struct kobject *kobj, | |
5076 | struct attribute *attr, | |
5077 | const char *buf, size_t len) | |
5078 | { | |
5079 | struct slab_attribute *attribute; | |
5080 | struct kmem_cache *s; | |
5081 | int err; | |
5082 | ||
5083 | attribute = to_slab_attr(attr); | |
5084 | s = to_slab(kobj); | |
5085 | ||
5086 | if (!attribute->store) | |
5087 | return -EIO; | |
5088 | ||
5089 | err = attribute->store(s, buf, len); | |
107dab5c GC |
5090 | #ifdef CONFIG_MEMCG_KMEM |
5091 | if (slab_state >= FULL && err >= 0 && is_root_cache(s)) { | |
426589f5 | 5092 | struct kmem_cache *c; |
81819f0f | 5093 | |
107dab5c GC |
5094 | mutex_lock(&slab_mutex); |
5095 | if (s->max_attr_size < len) | |
5096 | s->max_attr_size = len; | |
5097 | ||
ebe945c2 GC |
5098 | /* |
5099 | * This is a best effort propagation, so this function's return | |
5100 | * value will be determined by the parent cache only. This is | |
5101 | * basically because not all attributes will have a well | |
5102 | * defined semantics for rollbacks - most of the actions will | |
5103 | * have permanent effects. | |
5104 | * | |
5105 | * Returning the error value of any of the children that fail | |
5106 | * is not 100 % defined, in the sense that users seeing the | |
5107 | * error code won't be able to know anything about the state of | |
5108 | * the cache. | |
5109 | * | |
5110 | * Only returning the error code for the parent cache at least | |
5111 | * has well defined semantics. The cache being written to | |
5112 | * directly either failed or succeeded, in which case we loop | |
5113 | * through the descendants with best-effort propagation. | |
5114 | */ | |
426589f5 VD |
5115 | for_each_memcg_cache(c, s) |
5116 | attribute->store(c, buf, len); | |
107dab5c GC |
5117 | mutex_unlock(&slab_mutex); |
5118 | } | |
5119 | #endif | |
81819f0f CL |
5120 | return err; |
5121 | } | |
5122 | ||
107dab5c GC |
5123 | static void memcg_propagate_slab_attrs(struct kmem_cache *s) |
5124 | { | |
5125 | #ifdef CONFIG_MEMCG_KMEM | |
5126 | int i; | |
5127 | char *buffer = NULL; | |
93030d83 | 5128 | struct kmem_cache *root_cache; |
107dab5c | 5129 | |
93030d83 | 5130 | if (is_root_cache(s)) |
107dab5c GC |
5131 | return; |
5132 | ||
f7ce3190 | 5133 | root_cache = s->memcg_params.root_cache; |
93030d83 | 5134 | |
107dab5c GC |
5135 | /* |
5136 | * This mean this cache had no attribute written. Therefore, no point | |
5137 | * in copying default values around | |
5138 | */ | |
93030d83 | 5139 | if (!root_cache->max_attr_size) |
107dab5c GC |
5140 | return; |
5141 | ||
5142 | for (i = 0; i < ARRAY_SIZE(slab_attrs); i++) { | |
5143 | char mbuf[64]; | |
5144 | char *buf; | |
5145 | struct slab_attribute *attr = to_slab_attr(slab_attrs[i]); | |
5146 | ||
5147 | if (!attr || !attr->store || !attr->show) | |
5148 | continue; | |
5149 | ||
5150 | /* | |
5151 | * It is really bad that we have to allocate here, so we will | |
5152 | * do it only as a fallback. If we actually allocate, though, | |
5153 | * we can just use the allocated buffer until the end. | |
5154 | * | |
5155 | * Most of the slub attributes will tend to be very small in | |
5156 | * size, but sysfs allows buffers up to a page, so they can | |
5157 | * theoretically happen. | |
5158 | */ | |
5159 | if (buffer) | |
5160 | buf = buffer; | |
93030d83 | 5161 | else if (root_cache->max_attr_size < ARRAY_SIZE(mbuf)) |
107dab5c GC |
5162 | buf = mbuf; |
5163 | else { | |
5164 | buffer = (char *) get_zeroed_page(GFP_KERNEL); | |
5165 | if (WARN_ON(!buffer)) | |
5166 | continue; | |
5167 | buf = buffer; | |
5168 | } | |
5169 | ||
93030d83 | 5170 | attr->show(root_cache, buf); |
107dab5c GC |
5171 | attr->store(s, buf, strlen(buf)); |
5172 | } | |
5173 | ||
5174 | if (buffer) | |
5175 | free_page((unsigned long)buffer); | |
5176 | #endif | |
5177 | } | |
5178 | ||
41a21285 CL |
5179 | static void kmem_cache_release(struct kobject *k) |
5180 | { | |
5181 | slab_kmem_cache_release(to_slab(k)); | |
5182 | } | |
5183 | ||
52cf25d0 | 5184 | static const struct sysfs_ops slab_sysfs_ops = { |
81819f0f CL |
5185 | .show = slab_attr_show, |
5186 | .store = slab_attr_store, | |
5187 | }; | |
5188 | ||
5189 | static struct kobj_type slab_ktype = { | |
5190 | .sysfs_ops = &slab_sysfs_ops, | |
41a21285 | 5191 | .release = kmem_cache_release, |
81819f0f CL |
5192 | }; |
5193 | ||
5194 | static int uevent_filter(struct kset *kset, struct kobject *kobj) | |
5195 | { | |
5196 | struct kobj_type *ktype = get_ktype(kobj); | |
5197 | ||
5198 | if (ktype == &slab_ktype) | |
5199 | return 1; | |
5200 | return 0; | |
5201 | } | |
5202 | ||
9cd43611 | 5203 | static const struct kset_uevent_ops slab_uevent_ops = { |
81819f0f CL |
5204 | .filter = uevent_filter, |
5205 | }; | |
5206 | ||
27c3a314 | 5207 | static struct kset *slab_kset; |
81819f0f | 5208 | |
9a41707b VD |
5209 | static inline struct kset *cache_kset(struct kmem_cache *s) |
5210 | { | |
5211 | #ifdef CONFIG_MEMCG_KMEM | |
5212 | if (!is_root_cache(s)) | |
f7ce3190 | 5213 | return s->memcg_params.root_cache->memcg_kset; |
9a41707b VD |
5214 | #endif |
5215 | return slab_kset; | |
5216 | } | |
5217 | ||
81819f0f CL |
5218 | #define ID_STR_LENGTH 64 |
5219 | ||
5220 | /* Create a unique string id for a slab cache: | |
6446faa2 CL |
5221 | * |
5222 | * Format :[flags-]size | |
81819f0f CL |
5223 | */ |
5224 | static char *create_unique_id(struct kmem_cache *s) | |
5225 | { | |
5226 | char *name = kmalloc(ID_STR_LENGTH, GFP_KERNEL); | |
5227 | char *p = name; | |
5228 | ||
5229 | BUG_ON(!name); | |
5230 | ||
5231 | *p++ = ':'; | |
5232 | /* | |
5233 | * First flags affecting slabcache operations. We will only | |
5234 | * get here for aliasable slabs so we do not need to support | |
5235 | * too many flags. The flags here must cover all flags that | |
5236 | * are matched during merging to guarantee that the id is | |
5237 | * unique. | |
5238 | */ | |
5239 | if (s->flags & SLAB_CACHE_DMA) | |
5240 | *p++ = 'd'; | |
5241 | if (s->flags & SLAB_RECLAIM_ACCOUNT) | |
5242 | *p++ = 'a'; | |
5243 | if (s->flags & SLAB_DEBUG_FREE) | |
5244 | *p++ = 'F'; | |
5a896d9e VN |
5245 | if (!(s->flags & SLAB_NOTRACK)) |
5246 | *p++ = 't'; | |
81819f0f CL |
5247 | if (p != name + 1) |
5248 | *p++ = '-'; | |
5249 | p += sprintf(p, "%07d", s->size); | |
2633d7a0 | 5250 | |
81819f0f CL |
5251 | BUG_ON(p > name + ID_STR_LENGTH - 1); |
5252 | return name; | |
5253 | } | |
5254 | ||
5255 | static int sysfs_slab_add(struct kmem_cache *s) | |
5256 | { | |
5257 | int err; | |
5258 | const char *name; | |
45530c44 | 5259 | int unmergeable = slab_unmergeable(s); |
81819f0f | 5260 | |
81819f0f CL |
5261 | if (unmergeable) { |
5262 | /* | |
5263 | * Slabcache can never be merged so we can use the name proper. | |
5264 | * This is typically the case for debug situations. In that | |
5265 | * case we can catch duplicate names easily. | |
5266 | */ | |
27c3a314 | 5267 | sysfs_remove_link(&slab_kset->kobj, s->name); |
81819f0f CL |
5268 | name = s->name; |
5269 | } else { | |
5270 | /* | |
5271 | * Create a unique name for the slab as a target | |
5272 | * for the symlinks. | |
5273 | */ | |
5274 | name = create_unique_id(s); | |
5275 | } | |
5276 | ||
9a41707b | 5277 | s->kobj.kset = cache_kset(s); |
26e4f205 | 5278 | err = kobject_init_and_add(&s->kobj, &slab_ktype, NULL, "%s", name); |
54b6a731 | 5279 | if (err) |
80da026a | 5280 | goto out; |
81819f0f CL |
5281 | |
5282 | err = sysfs_create_group(&s->kobj, &slab_attr_group); | |
54b6a731 DJ |
5283 | if (err) |
5284 | goto out_del_kobj; | |
9a41707b VD |
5285 | |
5286 | #ifdef CONFIG_MEMCG_KMEM | |
5287 | if (is_root_cache(s)) { | |
5288 | s->memcg_kset = kset_create_and_add("cgroup", NULL, &s->kobj); | |
5289 | if (!s->memcg_kset) { | |
54b6a731 DJ |
5290 | err = -ENOMEM; |
5291 | goto out_del_kobj; | |
9a41707b VD |
5292 | } |
5293 | } | |
5294 | #endif | |
5295 | ||
81819f0f CL |
5296 | kobject_uevent(&s->kobj, KOBJ_ADD); |
5297 | if (!unmergeable) { | |
5298 | /* Setup first alias */ | |
5299 | sysfs_slab_alias(s, s->name); | |
81819f0f | 5300 | } |
54b6a731 DJ |
5301 | out: |
5302 | if (!unmergeable) | |
5303 | kfree(name); | |
5304 | return err; | |
5305 | out_del_kobj: | |
5306 | kobject_del(&s->kobj); | |
54b6a731 | 5307 | goto out; |
81819f0f CL |
5308 | } |
5309 | ||
41a21285 | 5310 | void sysfs_slab_remove(struct kmem_cache *s) |
81819f0f | 5311 | { |
97d06609 | 5312 | if (slab_state < FULL) |
2bce6485 CL |
5313 | /* |
5314 | * Sysfs has not been setup yet so no need to remove the | |
5315 | * cache from sysfs. | |
5316 | */ | |
5317 | return; | |
5318 | ||
9a41707b VD |
5319 | #ifdef CONFIG_MEMCG_KMEM |
5320 | kset_unregister(s->memcg_kset); | |
5321 | #endif | |
81819f0f CL |
5322 | kobject_uevent(&s->kobj, KOBJ_REMOVE); |
5323 | kobject_del(&s->kobj); | |
151c602f | 5324 | kobject_put(&s->kobj); |
81819f0f CL |
5325 | } |
5326 | ||
5327 | /* | |
5328 | * Need to buffer aliases during bootup until sysfs becomes | |
9f6c708e | 5329 | * available lest we lose that information. |
81819f0f CL |
5330 | */ |
5331 | struct saved_alias { | |
5332 | struct kmem_cache *s; | |
5333 | const char *name; | |
5334 | struct saved_alias *next; | |
5335 | }; | |
5336 | ||
5af328a5 | 5337 | static struct saved_alias *alias_list; |
81819f0f CL |
5338 | |
5339 | static int sysfs_slab_alias(struct kmem_cache *s, const char *name) | |
5340 | { | |
5341 | struct saved_alias *al; | |
5342 | ||
97d06609 | 5343 | if (slab_state == FULL) { |
81819f0f CL |
5344 | /* |
5345 | * If we have a leftover link then remove it. | |
5346 | */ | |
27c3a314 GKH |
5347 | sysfs_remove_link(&slab_kset->kobj, name); |
5348 | return sysfs_create_link(&slab_kset->kobj, &s->kobj, name); | |
81819f0f CL |
5349 | } |
5350 | ||
5351 | al = kmalloc(sizeof(struct saved_alias), GFP_KERNEL); | |
5352 | if (!al) | |
5353 | return -ENOMEM; | |
5354 | ||
5355 | al->s = s; | |
5356 | al->name = name; | |
5357 | al->next = alias_list; | |
5358 | alias_list = al; | |
5359 | return 0; | |
5360 | } | |
5361 | ||
5362 | static int __init slab_sysfs_init(void) | |
5363 | { | |
5b95a4ac | 5364 | struct kmem_cache *s; |
81819f0f CL |
5365 | int err; |
5366 | ||
18004c5d | 5367 | mutex_lock(&slab_mutex); |
2bce6485 | 5368 | |
0ff21e46 | 5369 | slab_kset = kset_create_and_add("slab", &slab_uevent_ops, kernel_kobj); |
27c3a314 | 5370 | if (!slab_kset) { |
18004c5d | 5371 | mutex_unlock(&slab_mutex); |
f9f58285 | 5372 | pr_err("Cannot register slab subsystem.\n"); |
81819f0f CL |
5373 | return -ENOSYS; |
5374 | } | |
5375 | ||
97d06609 | 5376 | slab_state = FULL; |
26a7bd03 | 5377 | |
5b95a4ac | 5378 | list_for_each_entry(s, &slab_caches, list) { |
26a7bd03 | 5379 | err = sysfs_slab_add(s); |
5d540fb7 | 5380 | if (err) |
f9f58285 FF |
5381 | pr_err("SLUB: Unable to add boot slab %s to sysfs\n", |
5382 | s->name); | |
26a7bd03 | 5383 | } |
81819f0f CL |
5384 | |
5385 | while (alias_list) { | |
5386 | struct saved_alias *al = alias_list; | |
5387 | ||
5388 | alias_list = alias_list->next; | |
5389 | err = sysfs_slab_alias(al->s, al->name); | |
5d540fb7 | 5390 | if (err) |
f9f58285 FF |
5391 | pr_err("SLUB: Unable to add boot slab alias %s to sysfs\n", |
5392 | al->name); | |
81819f0f CL |
5393 | kfree(al); |
5394 | } | |
5395 | ||
18004c5d | 5396 | mutex_unlock(&slab_mutex); |
81819f0f CL |
5397 | resiliency_test(); |
5398 | return 0; | |
5399 | } | |
5400 | ||
5401 | __initcall(slab_sysfs_init); | |
ab4d5ed5 | 5402 | #endif /* CONFIG_SYSFS */ |
57ed3eda PE |
5403 | |
5404 | /* | |
5405 | * The /proc/slabinfo ABI | |
5406 | */ | |
158a9624 | 5407 | #ifdef CONFIG_SLABINFO |
0d7561c6 | 5408 | void get_slabinfo(struct kmem_cache *s, struct slabinfo *sinfo) |
57ed3eda | 5409 | { |
57ed3eda | 5410 | unsigned long nr_slabs = 0; |
205ab99d CL |
5411 | unsigned long nr_objs = 0; |
5412 | unsigned long nr_free = 0; | |
57ed3eda | 5413 | int node; |
fa45dc25 | 5414 | struct kmem_cache_node *n; |
57ed3eda | 5415 | |
fa45dc25 | 5416 | for_each_kmem_cache_node(s, node, n) { |
c17fd13e WL |
5417 | nr_slabs += node_nr_slabs(n); |
5418 | nr_objs += node_nr_objs(n); | |
205ab99d | 5419 | nr_free += count_partial(n, count_free); |
57ed3eda PE |
5420 | } |
5421 | ||
0d7561c6 GC |
5422 | sinfo->active_objs = nr_objs - nr_free; |
5423 | sinfo->num_objs = nr_objs; | |
5424 | sinfo->active_slabs = nr_slabs; | |
5425 | sinfo->num_slabs = nr_slabs; | |
5426 | sinfo->objects_per_slab = oo_objects(s->oo); | |
5427 | sinfo->cache_order = oo_order(s->oo); | |
57ed3eda PE |
5428 | } |
5429 | ||
0d7561c6 | 5430 | void slabinfo_show_stats(struct seq_file *m, struct kmem_cache *s) |
7b3c3a50 | 5431 | { |
7b3c3a50 AD |
5432 | } |
5433 | ||
b7454ad3 GC |
5434 | ssize_t slabinfo_write(struct file *file, const char __user *buffer, |
5435 | size_t count, loff_t *ppos) | |
7b3c3a50 | 5436 | { |
b7454ad3 | 5437 | return -EIO; |
7b3c3a50 | 5438 | } |
158a9624 | 5439 | #endif /* CONFIG_SLABINFO */ |