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)) | |
377 | return 1; | |
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 CL |
386 | slab_unlock(page); |
387 | return 1; | |
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 | ||
399 | return 0; | |
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)) | |
413 | return 1; | |
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); |
b789ef51 CL |
427 | return 1; |
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 | ||
440 | return 0; | |
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 | */ | |
f0630fff CL |
462 | #ifdef CONFIG_SLUB_DEBUG_ON |
463 | static int slub_debug = DEBUG_DEFAULT_FLAGS; | |
464 | #else | |
41ecc55b | 465 | static int slub_debug; |
f0630fff | 466 | #endif |
41ecc55b CL |
467 | |
468 | static char *slub_debug_slabs; | |
fa5ec8a1 | 469 | static int disable_higher_order_debug; |
41ecc55b | 470 | |
a79316c6 AR |
471 | /* |
472 | * slub is about to manipulate internal object metadata. This memory lies | |
473 | * outside the range of the allocated object, so accessing it would normally | |
474 | * be reported by kasan as a bounds error. metadata_access_enable() is used | |
475 | * to tell kasan that these accesses are OK. | |
476 | */ | |
477 | static inline void metadata_access_enable(void) | |
478 | { | |
479 | kasan_disable_current(); | |
480 | } | |
481 | ||
482 | static inline void metadata_access_disable(void) | |
483 | { | |
484 | kasan_enable_current(); | |
485 | } | |
486 | ||
81819f0f CL |
487 | /* |
488 | * Object debugging | |
489 | */ | |
490 | static void print_section(char *text, u8 *addr, unsigned int length) | |
491 | { | |
a79316c6 | 492 | metadata_access_enable(); |
ffc79d28 SAS |
493 | print_hex_dump(KERN_ERR, text, DUMP_PREFIX_ADDRESS, 16, 1, addr, |
494 | length, 1); | |
a79316c6 | 495 | metadata_access_disable(); |
81819f0f CL |
496 | } |
497 | ||
81819f0f CL |
498 | static struct track *get_track(struct kmem_cache *s, void *object, |
499 | enum track_item alloc) | |
500 | { | |
501 | struct track *p; | |
502 | ||
503 | if (s->offset) | |
504 | p = object + s->offset + sizeof(void *); | |
505 | else | |
506 | p = object + s->inuse; | |
507 | ||
508 | return p + alloc; | |
509 | } | |
510 | ||
511 | static void set_track(struct kmem_cache *s, void *object, | |
ce71e27c | 512 | enum track_item alloc, unsigned long addr) |
81819f0f | 513 | { |
1a00df4a | 514 | struct track *p = get_track(s, object, alloc); |
81819f0f | 515 | |
81819f0f | 516 | if (addr) { |
d6543e39 BG |
517 | #ifdef CONFIG_STACKTRACE |
518 | struct stack_trace trace; | |
519 | int i; | |
520 | ||
521 | trace.nr_entries = 0; | |
522 | trace.max_entries = TRACK_ADDRS_COUNT; | |
523 | trace.entries = p->addrs; | |
524 | trace.skip = 3; | |
a79316c6 | 525 | metadata_access_enable(); |
d6543e39 | 526 | save_stack_trace(&trace); |
a79316c6 | 527 | metadata_access_disable(); |
d6543e39 BG |
528 | |
529 | /* See rant in lockdep.c */ | |
530 | if (trace.nr_entries != 0 && | |
531 | trace.entries[trace.nr_entries - 1] == ULONG_MAX) | |
532 | trace.nr_entries--; | |
533 | ||
534 | for (i = trace.nr_entries; i < TRACK_ADDRS_COUNT; i++) | |
535 | p->addrs[i] = 0; | |
536 | #endif | |
81819f0f CL |
537 | p->addr = addr; |
538 | p->cpu = smp_processor_id(); | |
88e4ccf2 | 539 | p->pid = current->pid; |
81819f0f CL |
540 | p->when = jiffies; |
541 | } else | |
542 | memset(p, 0, sizeof(struct track)); | |
543 | } | |
544 | ||
81819f0f CL |
545 | static void init_tracking(struct kmem_cache *s, void *object) |
546 | { | |
24922684 CL |
547 | if (!(s->flags & SLAB_STORE_USER)) |
548 | return; | |
549 | ||
ce71e27c EGM |
550 | set_track(s, object, TRACK_FREE, 0UL); |
551 | set_track(s, object, TRACK_ALLOC, 0UL); | |
81819f0f CL |
552 | } |
553 | ||
554 | static void print_track(const char *s, struct track *t) | |
555 | { | |
556 | if (!t->addr) | |
557 | return; | |
558 | ||
f9f58285 FF |
559 | pr_err("INFO: %s in %pS age=%lu cpu=%u pid=%d\n", |
560 | s, (void *)t->addr, jiffies - t->when, t->cpu, t->pid); | |
d6543e39 BG |
561 | #ifdef CONFIG_STACKTRACE |
562 | { | |
563 | int i; | |
564 | for (i = 0; i < TRACK_ADDRS_COUNT; i++) | |
565 | if (t->addrs[i]) | |
f9f58285 | 566 | pr_err("\t%pS\n", (void *)t->addrs[i]); |
d6543e39 BG |
567 | else |
568 | break; | |
569 | } | |
570 | #endif | |
24922684 CL |
571 | } |
572 | ||
573 | static void print_tracking(struct kmem_cache *s, void *object) | |
574 | { | |
575 | if (!(s->flags & SLAB_STORE_USER)) | |
576 | return; | |
577 | ||
578 | print_track("Allocated", get_track(s, object, TRACK_ALLOC)); | |
579 | print_track("Freed", get_track(s, object, TRACK_FREE)); | |
580 | } | |
581 | ||
582 | static void print_page_info(struct page *page) | |
583 | { | |
f9f58285 | 584 | pr_err("INFO: Slab 0x%p objects=%u used=%u fp=0x%p flags=0x%04lx\n", |
d0e0ac97 | 585 | page, page->objects, page->inuse, page->freelist, page->flags); |
24922684 CL |
586 | |
587 | } | |
588 | ||
589 | static void slab_bug(struct kmem_cache *s, char *fmt, ...) | |
590 | { | |
ecc42fbe | 591 | struct va_format vaf; |
24922684 | 592 | va_list args; |
24922684 CL |
593 | |
594 | va_start(args, fmt); | |
ecc42fbe FF |
595 | vaf.fmt = fmt; |
596 | vaf.va = &args; | |
f9f58285 | 597 | pr_err("=============================================================================\n"); |
ecc42fbe | 598 | pr_err("BUG %s (%s): %pV\n", s->name, print_tainted(), &vaf); |
f9f58285 | 599 | pr_err("-----------------------------------------------------------------------------\n\n"); |
645df230 | 600 | |
373d4d09 | 601 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); |
ecc42fbe | 602 | va_end(args); |
81819f0f CL |
603 | } |
604 | ||
24922684 CL |
605 | static void slab_fix(struct kmem_cache *s, char *fmt, ...) |
606 | { | |
ecc42fbe | 607 | struct va_format vaf; |
24922684 | 608 | va_list args; |
24922684 CL |
609 | |
610 | va_start(args, fmt); | |
ecc42fbe FF |
611 | vaf.fmt = fmt; |
612 | vaf.va = &args; | |
613 | pr_err("FIX %s: %pV\n", s->name, &vaf); | |
24922684 | 614 | va_end(args); |
24922684 CL |
615 | } |
616 | ||
617 | static void print_trailer(struct kmem_cache *s, struct page *page, u8 *p) | |
81819f0f CL |
618 | { |
619 | unsigned int off; /* Offset of last byte */ | |
a973e9dd | 620 | u8 *addr = page_address(page); |
24922684 CL |
621 | |
622 | print_tracking(s, p); | |
623 | ||
624 | print_page_info(page); | |
625 | ||
f9f58285 FF |
626 | pr_err("INFO: Object 0x%p @offset=%tu fp=0x%p\n\n", |
627 | p, p - addr, get_freepointer(s, p)); | |
24922684 CL |
628 | |
629 | if (p > addr + 16) | |
ffc79d28 | 630 | print_section("Bytes b4 ", p - 16, 16); |
81819f0f | 631 | |
3b0efdfa | 632 | print_section("Object ", p, min_t(unsigned long, s->object_size, |
ffc79d28 | 633 | PAGE_SIZE)); |
81819f0f | 634 | if (s->flags & SLAB_RED_ZONE) |
3b0efdfa CL |
635 | print_section("Redzone ", p + s->object_size, |
636 | s->inuse - s->object_size); | |
81819f0f | 637 | |
81819f0f CL |
638 | if (s->offset) |
639 | off = s->offset + sizeof(void *); | |
640 | else | |
641 | off = s->inuse; | |
642 | ||
24922684 | 643 | if (s->flags & SLAB_STORE_USER) |
81819f0f | 644 | off += 2 * sizeof(struct track); |
81819f0f CL |
645 | |
646 | if (off != s->size) | |
647 | /* Beginning of the filler is the free pointer */ | |
ffc79d28 | 648 | print_section("Padding ", p + off, s->size - off); |
24922684 CL |
649 | |
650 | dump_stack(); | |
81819f0f CL |
651 | } |
652 | ||
75c66def | 653 | void object_err(struct kmem_cache *s, struct page *page, |
81819f0f CL |
654 | u8 *object, char *reason) |
655 | { | |
3dc50637 | 656 | slab_bug(s, "%s", reason); |
24922684 | 657 | print_trailer(s, page, object); |
81819f0f CL |
658 | } |
659 | ||
d0e0ac97 CG |
660 | static void slab_err(struct kmem_cache *s, struct page *page, |
661 | const char *fmt, ...) | |
81819f0f CL |
662 | { |
663 | va_list args; | |
664 | char buf[100]; | |
665 | ||
24922684 CL |
666 | va_start(args, fmt); |
667 | vsnprintf(buf, sizeof(buf), fmt, args); | |
81819f0f | 668 | va_end(args); |
3dc50637 | 669 | slab_bug(s, "%s", buf); |
24922684 | 670 | print_page_info(page); |
81819f0f CL |
671 | dump_stack(); |
672 | } | |
673 | ||
f7cb1933 | 674 | static void init_object(struct kmem_cache *s, void *object, u8 val) |
81819f0f CL |
675 | { |
676 | u8 *p = object; | |
677 | ||
678 | if (s->flags & __OBJECT_POISON) { | |
3b0efdfa CL |
679 | memset(p, POISON_FREE, s->object_size - 1); |
680 | p[s->object_size - 1] = POISON_END; | |
81819f0f CL |
681 | } |
682 | ||
683 | if (s->flags & SLAB_RED_ZONE) | |
3b0efdfa | 684 | memset(p + s->object_size, val, s->inuse - s->object_size); |
81819f0f CL |
685 | } |
686 | ||
24922684 CL |
687 | static void restore_bytes(struct kmem_cache *s, char *message, u8 data, |
688 | void *from, void *to) | |
689 | { | |
690 | slab_fix(s, "Restoring 0x%p-0x%p=0x%x\n", from, to - 1, data); | |
691 | memset(from, data, to - from); | |
692 | } | |
693 | ||
694 | static int check_bytes_and_report(struct kmem_cache *s, struct page *page, | |
695 | u8 *object, char *what, | |
06428780 | 696 | u8 *start, unsigned int value, unsigned int bytes) |
24922684 CL |
697 | { |
698 | u8 *fault; | |
699 | u8 *end; | |
700 | ||
a79316c6 | 701 | metadata_access_enable(); |
79824820 | 702 | fault = memchr_inv(start, value, bytes); |
a79316c6 | 703 | metadata_access_disable(); |
24922684 CL |
704 | if (!fault) |
705 | return 1; | |
706 | ||
707 | end = start + bytes; | |
708 | while (end > fault && end[-1] == value) | |
709 | end--; | |
710 | ||
711 | slab_bug(s, "%s overwritten", what); | |
f9f58285 | 712 | pr_err("INFO: 0x%p-0x%p. First byte 0x%x instead of 0x%x\n", |
24922684 CL |
713 | fault, end - 1, fault[0], value); |
714 | print_trailer(s, page, object); | |
715 | ||
716 | restore_bytes(s, what, value, fault, end); | |
717 | return 0; | |
81819f0f CL |
718 | } |
719 | ||
81819f0f CL |
720 | /* |
721 | * Object layout: | |
722 | * | |
723 | * object address | |
724 | * Bytes of the object to be managed. | |
725 | * If the freepointer may overlay the object then the free | |
726 | * pointer is the first word of the object. | |
672bba3a | 727 | * |
81819f0f CL |
728 | * Poisoning uses 0x6b (POISON_FREE) and the last byte is |
729 | * 0xa5 (POISON_END) | |
730 | * | |
3b0efdfa | 731 | * object + s->object_size |
81819f0f | 732 | * Padding to reach word boundary. This is also used for Redzoning. |
672bba3a | 733 | * Padding is extended by another word if Redzoning is enabled and |
3b0efdfa | 734 | * object_size == inuse. |
672bba3a | 735 | * |
81819f0f CL |
736 | * We fill with 0xbb (RED_INACTIVE) for inactive objects and with |
737 | * 0xcc (RED_ACTIVE) for objects in use. | |
738 | * | |
739 | * object + s->inuse | |
672bba3a CL |
740 | * Meta data starts here. |
741 | * | |
81819f0f CL |
742 | * A. Free pointer (if we cannot overwrite object on free) |
743 | * B. Tracking data for SLAB_STORE_USER | |
672bba3a | 744 | * C. Padding to reach required alignment boundary or at mininum |
6446faa2 | 745 | * one word if debugging is on to be able to detect writes |
672bba3a CL |
746 | * before the word boundary. |
747 | * | |
748 | * Padding is done using 0x5a (POISON_INUSE) | |
81819f0f CL |
749 | * |
750 | * object + s->size | |
672bba3a | 751 | * Nothing is used beyond s->size. |
81819f0f | 752 | * |
3b0efdfa | 753 | * If slabcaches are merged then the object_size and inuse boundaries are mostly |
672bba3a | 754 | * ignored. And therefore no slab options that rely on these boundaries |
81819f0f CL |
755 | * may be used with merged slabcaches. |
756 | */ | |
757 | ||
81819f0f CL |
758 | static int check_pad_bytes(struct kmem_cache *s, struct page *page, u8 *p) |
759 | { | |
760 | unsigned long off = s->inuse; /* The end of info */ | |
761 | ||
762 | if (s->offset) | |
763 | /* Freepointer is placed after the object. */ | |
764 | off += sizeof(void *); | |
765 | ||
766 | if (s->flags & SLAB_STORE_USER) | |
767 | /* We also have user information there */ | |
768 | off += 2 * sizeof(struct track); | |
769 | ||
770 | if (s->size == off) | |
771 | return 1; | |
772 | ||
24922684 CL |
773 | return check_bytes_and_report(s, page, p, "Object padding", |
774 | p + off, POISON_INUSE, s->size - off); | |
81819f0f CL |
775 | } |
776 | ||
39b26464 | 777 | /* Check the pad bytes at the end of a slab page */ |
81819f0f CL |
778 | static int slab_pad_check(struct kmem_cache *s, struct page *page) |
779 | { | |
24922684 CL |
780 | u8 *start; |
781 | u8 *fault; | |
782 | u8 *end; | |
783 | int length; | |
784 | int remainder; | |
81819f0f CL |
785 | |
786 | if (!(s->flags & SLAB_POISON)) | |
787 | return 1; | |
788 | ||
a973e9dd | 789 | start = page_address(page); |
ab9a0f19 | 790 | length = (PAGE_SIZE << compound_order(page)) - s->reserved; |
39b26464 CL |
791 | end = start + length; |
792 | remainder = length % s->size; | |
81819f0f CL |
793 | if (!remainder) |
794 | return 1; | |
795 | ||
a79316c6 | 796 | metadata_access_enable(); |
79824820 | 797 | fault = memchr_inv(end - remainder, POISON_INUSE, remainder); |
a79316c6 | 798 | metadata_access_disable(); |
24922684 CL |
799 | if (!fault) |
800 | return 1; | |
801 | while (end > fault && end[-1] == POISON_INUSE) | |
802 | end--; | |
803 | ||
804 | slab_err(s, page, "Padding overwritten. 0x%p-0x%p", fault, end - 1); | |
ffc79d28 | 805 | print_section("Padding ", end - remainder, remainder); |
24922684 | 806 | |
8a3d271d | 807 | restore_bytes(s, "slab padding", POISON_INUSE, end - remainder, end); |
24922684 | 808 | return 0; |
81819f0f CL |
809 | } |
810 | ||
811 | static int check_object(struct kmem_cache *s, struct page *page, | |
f7cb1933 | 812 | void *object, u8 val) |
81819f0f CL |
813 | { |
814 | u8 *p = object; | |
3b0efdfa | 815 | u8 *endobject = object + s->object_size; |
81819f0f CL |
816 | |
817 | if (s->flags & SLAB_RED_ZONE) { | |
24922684 | 818 | if (!check_bytes_and_report(s, page, object, "Redzone", |
3b0efdfa | 819 | endobject, val, s->inuse - s->object_size)) |
81819f0f | 820 | return 0; |
81819f0f | 821 | } else { |
3b0efdfa | 822 | if ((s->flags & SLAB_POISON) && s->object_size < s->inuse) { |
3adbefee | 823 | check_bytes_and_report(s, page, p, "Alignment padding", |
d0e0ac97 CG |
824 | endobject, POISON_INUSE, |
825 | s->inuse - s->object_size); | |
3adbefee | 826 | } |
81819f0f CL |
827 | } |
828 | ||
829 | if (s->flags & SLAB_POISON) { | |
f7cb1933 | 830 | if (val != SLUB_RED_ACTIVE && (s->flags & __OBJECT_POISON) && |
24922684 | 831 | (!check_bytes_and_report(s, page, p, "Poison", p, |
3b0efdfa | 832 | POISON_FREE, s->object_size - 1) || |
24922684 | 833 | !check_bytes_and_report(s, page, p, "Poison", |
3b0efdfa | 834 | p + s->object_size - 1, POISON_END, 1))) |
81819f0f | 835 | return 0; |
81819f0f CL |
836 | /* |
837 | * check_pad_bytes cleans up on its own. | |
838 | */ | |
839 | check_pad_bytes(s, page, p); | |
840 | } | |
841 | ||
f7cb1933 | 842 | if (!s->offset && val == SLUB_RED_ACTIVE) |
81819f0f CL |
843 | /* |
844 | * Object and freepointer overlap. Cannot check | |
845 | * freepointer while object is allocated. | |
846 | */ | |
847 | return 1; | |
848 | ||
849 | /* Check free pointer validity */ | |
850 | if (!check_valid_pointer(s, page, get_freepointer(s, p))) { | |
851 | object_err(s, page, p, "Freepointer corrupt"); | |
852 | /* | |
9f6c708e | 853 | * No choice but to zap it and thus lose the remainder |
81819f0f | 854 | * of the free objects in this slab. May cause |
672bba3a | 855 | * another error because the object count is now wrong. |
81819f0f | 856 | */ |
a973e9dd | 857 | set_freepointer(s, p, NULL); |
81819f0f CL |
858 | return 0; |
859 | } | |
860 | return 1; | |
861 | } | |
862 | ||
863 | static int check_slab(struct kmem_cache *s, struct page *page) | |
864 | { | |
39b26464 CL |
865 | int maxobj; |
866 | ||
81819f0f CL |
867 | VM_BUG_ON(!irqs_disabled()); |
868 | ||
869 | if (!PageSlab(page)) { | |
24922684 | 870 | slab_err(s, page, "Not a valid slab page"); |
81819f0f CL |
871 | return 0; |
872 | } | |
39b26464 | 873 | |
ab9a0f19 | 874 | maxobj = order_objects(compound_order(page), s->size, s->reserved); |
39b26464 CL |
875 | if (page->objects > maxobj) { |
876 | slab_err(s, page, "objects %u > max %u", | |
f6edde9c | 877 | page->objects, maxobj); |
39b26464 CL |
878 | return 0; |
879 | } | |
880 | if (page->inuse > page->objects) { | |
24922684 | 881 | slab_err(s, page, "inuse %u > max %u", |
f6edde9c | 882 | page->inuse, page->objects); |
81819f0f CL |
883 | return 0; |
884 | } | |
885 | /* Slab_pad_check fixes things up after itself */ | |
886 | slab_pad_check(s, page); | |
887 | return 1; | |
888 | } | |
889 | ||
890 | /* | |
672bba3a CL |
891 | * Determine if a certain object on a page is on the freelist. Must hold the |
892 | * slab lock to guarantee that the chains are in a consistent state. | |
81819f0f CL |
893 | */ |
894 | static int on_freelist(struct kmem_cache *s, struct page *page, void *search) | |
895 | { | |
896 | int nr = 0; | |
881db7fb | 897 | void *fp; |
81819f0f | 898 | void *object = NULL; |
f6edde9c | 899 | int max_objects; |
81819f0f | 900 | |
881db7fb | 901 | fp = page->freelist; |
39b26464 | 902 | while (fp && nr <= page->objects) { |
81819f0f CL |
903 | if (fp == search) |
904 | return 1; | |
905 | if (!check_valid_pointer(s, page, fp)) { | |
906 | if (object) { | |
907 | object_err(s, page, object, | |
908 | "Freechain corrupt"); | |
a973e9dd | 909 | set_freepointer(s, object, NULL); |
81819f0f | 910 | } else { |
24922684 | 911 | slab_err(s, page, "Freepointer corrupt"); |
a973e9dd | 912 | page->freelist = NULL; |
39b26464 | 913 | page->inuse = page->objects; |
24922684 | 914 | slab_fix(s, "Freelist cleared"); |
81819f0f CL |
915 | return 0; |
916 | } | |
917 | break; | |
918 | } | |
919 | object = fp; | |
920 | fp = get_freepointer(s, object); | |
921 | nr++; | |
922 | } | |
923 | ||
ab9a0f19 | 924 | max_objects = order_objects(compound_order(page), s->size, s->reserved); |
210b5c06 CG |
925 | if (max_objects > MAX_OBJS_PER_PAGE) |
926 | max_objects = MAX_OBJS_PER_PAGE; | |
224a88be CL |
927 | |
928 | if (page->objects != max_objects) { | |
929 | slab_err(s, page, "Wrong number of objects. Found %d but " | |
930 | "should be %d", page->objects, max_objects); | |
931 | page->objects = max_objects; | |
932 | slab_fix(s, "Number of objects adjusted."); | |
933 | } | |
39b26464 | 934 | if (page->inuse != page->objects - nr) { |
70d71228 | 935 | slab_err(s, page, "Wrong object count. Counter is %d but " |
39b26464 CL |
936 | "counted were %d", page->inuse, page->objects - nr); |
937 | page->inuse = page->objects - nr; | |
24922684 | 938 | slab_fix(s, "Object count adjusted."); |
81819f0f CL |
939 | } |
940 | return search == NULL; | |
941 | } | |
942 | ||
0121c619 CL |
943 | static void trace(struct kmem_cache *s, struct page *page, void *object, |
944 | int alloc) | |
3ec09742 CL |
945 | { |
946 | if (s->flags & SLAB_TRACE) { | |
f9f58285 | 947 | pr_info("TRACE %s %s 0x%p inuse=%d fp=0x%p\n", |
3ec09742 CL |
948 | s->name, |
949 | alloc ? "alloc" : "free", | |
950 | object, page->inuse, | |
951 | page->freelist); | |
952 | ||
953 | if (!alloc) | |
d0e0ac97 CG |
954 | print_section("Object ", (void *)object, |
955 | s->object_size); | |
3ec09742 CL |
956 | |
957 | dump_stack(); | |
958 | } | |
959 | } | |
960 | ||
643b1138 | 961 | /* |
672bba3a | 962 | * Tracking of fully allocated slabs for debugging purposes. |
643b1138 | 963 | */ |
5cc6eee8 CL |
964 | static void add_full(struct kmem_cache *s, |
965 | struct kmem_cache_node *n, struct page *page) | |
643b1138 | 966 | { |
5cc6eee8 CL |
967 | if (!(s->flags & SLAB_STORE_USER)) |
968 | return; | |
969 | ||
255d0884 | 970 | lockdep_assert_held(&n->list_lock); |
643b1138 | 971 | list_add(&page->lru, &n->full); |
643b1138 CL |
972 | } |
973 | ||
c65c1877 | 974 | static void remove_full(struct kmem_cache *s, struct kmem_cache_node *n, struct page *page) |
643b1138 | 975 | { |
643b1138 CL |
976 | if (!(s->flags & SLAB_STORE_USER)) |
977 | return; | |
978 | ||
255d0884 | 979 | lockdep_assert_held(&n->list_lock); |
643b1138 | 980 | list_del(&page->lru); |
643b1138 CL |
981 | } |
982 | ||
0f389ec6 CL |
983 | /* Tracking of the number of slabs for debugging purposes */ |
984 | static inline unsigned long slabs_node(struct kmem_cache *s, int node) | |
985 | { | |
986 | struct kmem_cache_node *n = get_node(s, node); | |
987 | ||
988 | return atomic_long_read(&n->nr_slabs); | |
989 | } | |
990 | ||
26c02cf0 AB |
991 | static inline unsigned long node_nr_slabs(struct kmem_cache_node *n) |
992 | { | |
993 | return atomic_long_read(&n->nr_slabs); | |
994 | } | |
995 | ||
205ab99d | 996 | static inline void inc_slabs_node(struct kmem_cache *s, int node, int objects) |
0f389ec6 CL |
997 | { |
998 | struct kmem_cache_node *n = get_node(s, node); | |
999 | ||
1000 | /* | |
1001 | * May be called early in order to allocate a slab for the | |
1002 | * kmem_cache_node structure. Solve the chicken-egg | |
1003 | * dilemma by deferring the increment of the count during | |
1004 | * bootstrap (see early_kmem_cache_node_alloc). | |
1005 | */ | |
338b2642 | 1006 | if (likely(n)) { |
0f389ec6 | 1007 | atomic_long_inc(&n->nr_slabs); |
205ab99d CL |
1008 | atomic_long_add(objects, &n->total_objects); |
1009 | } | |
0f389ec6 | 1010 | } |
205ab99d | 1011 | static inline void dec_slabs_node(struct kmem_cache *s, int node, int objects) |
0f389ec6 CL |
1012 | { |
1013 | struct kmem_cache_node *n = get_node(s, node); | |
1014 | ||
1015 | atomic_long_dec(&n->nr_slabs); | |
205ab99d | 1016 | atomic_long_sub(objects, &n->total_objects); |
0f389ec6 CL |
1017 | } |
1018 | ||
1019 | /* Object debug checks for alloc/free paths */ | |
3ec09742 CL |
1020 | static void setup_object_debug(struct kmem_cache *s, struct page *page, |
1021 | void *object) | |
1022 | { | |
1023 | if (!(s->flags & (SLAB_STORE_USER|SLAB_RED_ZONE|__OBJECT_POISON))) | |
1024 | return; | |
1025 | ||
f7cb1933 | 1026 | init_object(s, object, SLUB_RED_INACTIVE); |
3ec09742 CL |
1027 | init_tracking(s, object); |
1028 | } | |
1029 | ||
d0e0ac97 CG |
1030 | static noinline int alloc_debug_processing(struct kmem_cache *s, |
1031 | struct page *page, | |
ce71e27c | 1032 | void *object, unsigned long addr) |
81819f0f CL |
1033 | { |
1034 | if (!check_slab(s, page)) | |
1035 | goto bad; | |
1036 | ||
81819f0f CL |
1037 | if (!check_valid_pointer(s, page, object)) { |
1038 | object_err(s, page, object, "Freelist Pointer check fails"); | |
70d71228 | 1039 | goto bad; |
81819f0f CL |
1040 | } |
1041 | ||
f7cb1933 | 1042 | if (!check_object(s, page, object, SLUB_RED_INACTIVE)) |
81819f0f | 1043 | goto bad; |
81819f0f | 1044 | |
3ec09742 CL |
1045 | /* Success perform special debug activities for allocs */ |
1046 | if (s->flags & SLAB_STORE_USER) | |
1047 | set_track(s, object, TRACK_ALLOC, addr); | |
1048 | trace(s, page, object, 1); | |
f7cb1933 | 1049 | init_object(s, object, SLUB_RED_ACTIVE); |
81819f0f | 1050 | return 1; |
3ec09742 | 1051 | |
81819f0f CL |
1052 | bad: |
1053 | if (PageSlab(page)) { | |
1054 | /* | |
1055 | * If this is a slab page then lets do the best we can | |
1056 | * to avoid issues in the future. Marking all objects | |
672bba3a | 1057 | * as used avoids touching the remaining objects. |
81819f0f | 1058 | */ |
24922684 | 1059 | slab_fix(s, "Marking all objects used"); |
39b26464 | 1060 | page->inuse = page->objects; |
a973e9dd | 1061 | page->freelist = NULL; |
81819f0f CL |
1062 | } |
1063 | return 0; | |
1064 | } | |
1065 | ||
19c7ff9e CL |
1066 | static noinline struct kmem_cache_node *free_debug_processing( |
1067 | struct kmem_cache *s, struct page *page, void *object, | |
1068 | unsigned long addr, unsigned long *flags) | |
81819f0f | 1069 | { |
19c7ff9e | 1070 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); |
5c2e4bbb | 1071 | |
19c7ff9e | 1072 | spin_lock_irqsave(&n->list_lock, *flags); |
881db7fb CL |
1073 | slab_lock(page); |
1074 | ||
81819f0f CL |
1075 | if (!check_slab(s, page)) |
1076 | goto fail; | |
1077 | ||
1078 | if (!check_valid_pointer(s, page, object)) { | |
70d71228 | 1079 | slab_err(s, page, "Invalid object pointer 0x%p", object); |
81819f0f CL |
1080 | goto fail; |
1081 | } | |
1082 | ||
1083 | if (on_freelist(s, page, object)) { | |
24922684 | 1084 | object_err(s, page, object, "Object already free"); |
81819f0f CL |
1085 | goto fail; |
1086 | } | |
1087 | ||
f7cb1933 | 1088 | if (!check_object(s, page, object, SLUB_RED_ACTIVE)) |
5c2e4bbb | 1089 | goto out; |
81819f0f | 1090 | |
1b4f59e3 | 1091 | if (unlikely(s != page->slab_cache)) { |
3adbefee | 1092 | if (!PageSlab(page)) { |
70d71228 CL |
1093 | slab_err(s, page, "Attempt to free object(0x%p) " |
1094 | "outside of slab", object); | |
1b4f59e3 | 1095 | } else if (!page->slab_cache) { |
f9f58285 FF |
1096 | pr_err("SLUB <none>: no slab for object 0x%p.\n", |
1097 | object); | |
70d71228 | 1098 | dump_stack(); |
06428780 | 1099 | } else |
24922684 CL |
1100 | object_err(s, page, object, |
1101 | "page slab pointer corrupt."); | |
81819f0f CL |
1102 | goto fail; |
1103 | } | |
3ec09742 | 1104 | |
3ec09742 CL |
1105 | if (s->flags & SLAB_STORE_USER) |
1106 | set_track(s, object, TRACK_FREE, addr); | |
1107 | trace(s, page, object, 0); | |
f7cb1933 | 1108 | init_object(s, object, SLUB_RED_INACTIVE); |
5c2e4bbb | 1109 | out: |
881db7fb | 1110 | slab_unlock(page); |
19c7ff9e CL |
1111 | /* |
1112 | * Keep node_lock to preserve integrity | |
1113 | * until the object is actually freed | |
1114 | */ | |
1115 | return n; | |
3ec09742 | 1116 | |
81819f0f | 1117 | fail: |
19c7ff9e CL |
1118 | slab_unlock(page); |
1119 | spin_unlock_irqrestore(&n->list_lock, *flags); | |
24922684 | 1120 | slab_fix(s, "Object at 0x%p not freed", object); |
19c7ff9e | 1121 | return NULL; |
81819f0f CL |
1122 | } |
1123 | ||
41ecc55b CL |
1124 | static int __init setup_slub_debug(char *str) |
1125 | { | |
f0630fff CL |
1126 | slub_debug = DEBUG_DEFAULT_FLAGS; |
1127 | if (*str++ != '=' || !*str) | |
1128 | /* | |
1129 | * No options specified. Switch on full debugging. | |
1130 | */ | |
1131 | goto out; | |
1132 | ||
1133 | if (*str == ',') | |
1134 | /* | |
1135 | * No options but restriction on slabs. This means full | |
1136 | * debugging for slabs matching a pattern. | |
1137 | */ | |
1138 | goto check_slabs; | |
1139 | ||
fa5ec8a1 DR |
1140 | if (tolower(*str) == 'o') { |
1141 | /* | |
1142 | * Avoid enabling debugging on caches if its minimum order | |
1143 | * would increase as a result. | |
1144 | */ | |
1145 | disable_higher_order_debug = 1; | |
1146 | goto out; | |
1147 | } | |
1148 | ||
f0630fff CL |
1149 | slub_debug = 0; |
1150 | if (*str == '-') | |
1151 | /* | |
1152 | * Switch off all debugging measures. | |
1153 | */ | |
1154 | goto out; | |
1155 | ||
1156 | /* | |
1157 | * Determine which debug features should be switched on | |
1158 | */ | |
06428780 | 1159 | for (; *str && *str != ','; str++) { |
f0630fff CL |
1160 | switch (tolower(*str)) { |
1161 | case 'f': | |
1162 | slub_debug |= SLAB_DEBUG_FREE; | |
1163 | break; | |
1164 | case 'z': | |
1165 | slub_debug |= SLAB_RED_ZONE; | |
1166 | break; | |
1167 | case 'p': | |
1168 | slub_debug |= SLAB_POISON; | |
1169 | break; | |
1170 | case 'u': | |
1171 | slub_debug |= SLAB_STORE_USER; | |
1172 | break; | |
1173 | case 't': | |
1174 | slub_debug |= SLAB_TRACE; | |
1175 | break; | |
4c13dd3b DM |
1176 | case 'a': |
1177 | slub_debug |= SLAB_FAILSLAB; | |
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); | |
1268 | might_sleep_if(flags & __GFP_WAIT); | |
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 | |
81819f0f CL |
1311 | /* |
1312 | * Slab allocation and freeing | |
1313 | */ | |
5dfb4175 VD |
1314 | static inline struct page *alloc_slab_page(struct kmem_cache *s, |
1315 | gfp_t flags, int node, struct kmem_cache_order_objects oo) | |
65c3376a | 1316 | { |
5dfb4175 | 1317 | struct page *page; |
65c3376a CL |
1318 | int order = oo_order(oo); |
1319 | ||
b1eeab67 VN |
1320 | flags |= __GFP_NOTRACK; |
1321 | ||
5dfb4175 VD |
1322 | if (memcg_charge_slab(s, flags, order)) |
1323 | return NULL; | |
1324 | ||
2154a336 | 1325 | if (node == NUMA_NO_NODE) |
5dfb4175 | 1326 | page = alloc_pages(flags, order); |
65c3376a | 1327 | else |
5dfb4175 VD |
1328 | page = alloc_pages_exact_node(node, flags, order); |
1329 | ||
1330 | if (!page) | |
1331 | memcg_uncharge_slab(s, order); | |
1332 | ||
1333 | return page; | |
65c3376a CL |
1334 | } |
1335 | ||
81819f0f CL |
1336 | static struct page *allocate_slab(struct kmem_cache *s, gfp_t flags, int node) |
1337 | { | |
06428780 | 1338 | struct page *page; |
834f3d11 | 1339 | struct kmem_cache_order_objects oo = s->oo; |
ba52270d | 1340 | gfp_t alloc_gfp; |
81819f0f | 1341 | |
7e0528da CL |
1342 | flags &= gfp_allowed_mask; |
1343 | ||
1344 | if (flags & __GFP_WAIT) | |
1345 | local_irq_enable(); | |
1346 | ||
b7a49f0d | 1347 | flags |= s->allocflags; |
e12ba74d | 1348 | |
ba52270d PE |
1349 | /* |
1350 | * Let the initial higher-order allocation fail under memory pressure | |
1351 | * so we fall-back to the minimum order allocation. | |
1352 | */ | |
1353 | alloc_gfp = (flags | __GFP_NOWARN | __GFP_NORETRY) & ~__GFP_NOFAIL; | |
1354 | ||
5dfb4175 | 1355 | page = alloc_slab_page(s, alloc_gfp, node, oo); |
65c3376a CL |
1356 | if (unlikely(!page)) { |
1357 | oo = s->min; | |
80c3a998 | 1358 | alloc_gfp = flags; |
65c3376a CL |
1359 | /* |
1360 | * Allocation may have failed due to fragmentation. | |
1361 | * Try a lower order alloc if possible | |
1362 | */ | |
5dfb4175 | 1363 | page = alloc_slab_page(s, alloc_gfp, node, oo); |
81819f0f | 1364 | |
7e0528da CL |
1365 | if (page) |
1366 | stat(s, ORDER_FALLBACK); | |
65c3376a | 1367 | } |
5a896d9e | 1368 | |
737b719e | 1369 | if (kmemcheck_enabled && page |
5086c389 | 1370 | && !(s->flags & (SLAB_NOTRACK | DEBUG_DEFAULT_FLAGS))) { |
b1eeab67 VN |
1371 | int pages = 1 << oo_order(oo); |
1372 | ||
80c3a998 | 1373 | kmemcheck_alloc_shadow(page, oo_order(oo), alloc_gfp, node); |
b1eeab67 VN |
1374 | |
1375 | /* | |
1376 | * Objects from caches that have a constructor don't get | |
1377 | * cleared when they're allocated, so we need to do it here. | |
1378 | */ | |
1379 | if (s->ctor) | |
1380 | kmemcheck_mark_uninitialized_pages(page, pages); | |
1381 | else | |
1382 | kmemcheck_mark_unallocated_pages(page, pages); | |
5a896d9e VN |
1383 | } |
1384 | ||
737b719e DR |
1385 | if (flags & __GFP_WAIT) |
1386 | local_irq_disable(); | |
1387 | if (!page) | |
1388 | return NULL; | |
1389 | ||
834f3d11 | 1390 | page->objects = oo_objects(oo); |
81819f0f CL |
1391 | mod_zone_page_state(page_zone(page), |
1392 | (s->flags & SLAB_RECLAIM_ACCOUNT) ? | |
1393 | NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE, | |
65c3376a | 1394 | 1 << oo_order(oo)); |
81819f0f CL |
1395 | |
1396 | return page; | |
1397 | } | |
1398 | ||
1399 | static void setup_object(struct kmem_cache *s, struct page *page, | |
1400 | void *object) | |
1401 | { | |
3ec09742 | 1402 | setup_object_debug(s, page, object); |
0316bec2 AR |
1403 | if (unlikely(s->ctor)) { |
1404 | kasan_unpoison_object_data(s, object); | |
51cc5068 | 1405 | s->ctor(object); |
0316bec2 AR |
1406 | kasan_poison_object_data(s, object); |
1407 | } | |
81819f0f CL |
1408 | } |
1409 | ||
1410 | static struct page *new_slab(struct kmem_cache *s, gfp_t flags, int node) | |
1411 | { | |
1412 | struct page *page; | |
81819f0f | 1413 | void *start; |
81819f0f | 1414 | void *p; |
1f458cbf | 1415 | int order; |
54266640 | 1416 | int idx; |
81819f0f | 1417 | |
c871ac4e AM |
1418 | if (unlikely(flags & GFP_SLAB_BUG_MASK)) { |
1419 | pr_emerg("gfp: %u\n", flags & GFP_SLAB_BUG_MASK); | |
1420 | BUG(); | |
1421 | } | |
81819f0f | 1422 | |
6cb06229 CL |
1423 | page = allocate_slab(s, |
1424 | flags & (GFP_RECLAIM_MASK | GFP_CONSTRAINT_MASK), node); | |
81819f0f CL |
1425 | if (!page) |
1426 | goto out; | |
1427 | ||
1f458cbf | 1428 | order = compound_order(page); |
205ab99d | 1429 | inc_slabs_node(s, page_to_nid(page), page->objects); |
1b4f59e3 | 1430 | page->slab_cache = s; |
c03f94cc | 1431 | __SetPageSlab(page); |
072bb0aa MG |
1432 | if (page->pfmemalloc) |
1433 | SetPageSlabPfmemalloc(page); | |
81819f0f CL |
1434 | |
1435 | start = page_address(page); | |
81819f0f CL |
1436 | |
1437 | if (unlikely(s->flags & SLAB_POISON)) | |
1f458cbf | 1438 | memset(start, POISON_INUSE, PAGE_SIZE << order); |
81819f0f | 1439 | |
0316bec2 AR |
1440 | kasan_poison_slab(page); |
1441 | ||
54266640 WY |
1442 | for_each_object_idx(p, idx, s, start, page->objects) { |
1443 | setup_object(s, page, p); | |
1444 | if (likely(idx < page->objects)) | |
1445 | set_freepointer(s, p, p + s->size); | |
1446 | else | |
1447 | set_freepointer(s, p, NULL); | |
81819f0f | 1448 | } |
81819f0f CL |
1449 | |
1450 | page->freelist = start; | |
e6e82ea1 | 1451 | page->inuse = page->objects; |
8cb0a506 | 1452 | page->frozen = 1; |
81819f0f | 1453 | out: |
81819f0f CL |
1454 | return page; |
1455 | } | |
1456 | ||
1457 | static void __free_slab(struct kmem_cache *s, struct page *page) | |
1458 | { | |
834f3d11 CL |
1459 | int order = compound_order(page); |
1460 | int pages = 1 << order; | |
81819f0f | 1461 | |
af537b0a | 1462 | if (kmem_cache_debug(s)) { |
81819f0f CL |
1463 | void *p; |
1464 | ||
1465 | slab_pad_check(s, page); | |
224a88be CL |
1466 | for_each_object(p, s, page_address(page), |
1467 | page->objects) | |
f7cb1933 | 1468 | check_object(s, page, p, SLUB_RED_INACTIVE); |
81819f0f CL |
1469 | } |
1470 | ||
b1eeab67 | 1471 | kmemcheck_free_shadow(page, compound_order(page)); |
5a896d9e | 1472 | |
81819f0f CL |
1473 | mod_zone_page_state(page_zone(page), |
1474 | (s->flags & SLAB_RECLAIM_ACCOUNT) ? | |
1475 | NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE, | |
06428780 | 1476 | -pages); |
81819f0f | 1477 | |
072bb0aa | 1478 | __ClearPageSlabPfmemalloc(page); |
49bd5221 | 1479 | __ClearPageSlab(page); |
1f458cbf | 1480 | |
22b751c3 | 1481 | page_mapcount_reset(page); |
1eb5ac64 NP |
1482 | if (current->reclaim_state) |
1483 | current->reclaim_state->reclaimed_slab += pages; | |
5dfb4175 VD |
1484 | __free_pages(page, order); |
1485 | memcg_uncharge_slab(s, order); | |
81819f0f CL |
1486 | } |
1487 | ||
da9a638c LJ |
1488 | #define need_reserve_slab_rcu \ |
1489 | (sizeof(((struct page *)NULL)->lru) < sizeof(struct rcu_head)) | |
1490 | ||
81819f0f CL |
1491 | static void rcu_free_slab(struct rcu_head *h) |
1492 | { | |
1493 | struct page *page; | |
1494 | ||
da9a638c LJ |
1495 | if (need_reserve_slab_rcu) |
1496 | page = virt_to_head_page(h); | |
1497 | else | |
1498 | page = container_of((struct list_head *)h, struct page, lru); | |
1499 | ||
1b4f59e3 | 1500 | __free_slab(page->slab_cache, page); |
81819f0f CL |
1501 | } |
1502 | ||
1503 | static void free_slab(struct kmem_cache *s, struct page *page) | |
1504 | { | |
1505 | if (unlikely(s->flags & SLAB_DESTROY_BY_RCU)) { | |
da9a638c LJ |
1506 | struct rcu_head *head; |
1507 | ||
1508 | if (need_reserve_slab_rcu) { | |
1509 | int order = compound_order(page); | |
1510 | int offset = (PAGE_SIZE << order) - s->reserved; | |
1511 | ||
1512 | VM_BUG_ON(s->reserved != sizeof(*head)); | |
1513 | head = page_address(page) + offset; | |
1514 | } else { | |
1515 | /* | |
1516 | * RCU free overloads the RCU head over the LRU | |
1517 | */ | |
1518 | head = (void *)&page->lru; | |
1519 | } | |
81819f0f CL |
1520 | |
1521 | call_rcu(head, rcu_free_slab); | |
1522 | } else | |
1523 | __free_slab(s, page); | |
1524 | } | |
1525 | ||
1526 | static void discard_slab(struct kmem_cache *s, struct page *page) | |
1527 | { | |
205ab99d | 1528 | dec_slabs_node(s, page_to_nid(page), page->objects); |
81819f0f CL |
1529 | free_slab(s, page); |
1530 | } | |
1531 | ||
1532 | /* | |
5cc6eee8 | 1533 | * Management of partially allocated slabs. |
81819f0f | 1534 | */ |
1e4dd946 SR |
1535 | static inline void |
1536 | __add_partial(struct kmem_cache_node *n, struct page *page, int tail) | |
81819f0f | 1537 | { |
e95eed57 | 1538 | n->nr_partial++; |
136333d1 | 1539 | if (tail == DEACTIVATE_TO_TAIL) |
7c2e132c CL |
1540 | list_add_tail(&page->lru, &n->partial); |
1541 | else | |
1542 | list_add(&page->lru, &n->partial); | |
81819f0f CL |
1543 | } |
1544 | ||
1e4dd946 SR |
1545 | static inline void add_partial(struct kmem_cache_node *n, |
1546 | struct page *page, int tail) | |
62e346a8 | 1547 | { |
c65c1877 | 1548 | lockdep_assert_held(&n->list_lock); |
1e4dd946 SR |
1549 | __add_partial(n, page, tail); |
1550 | } | |
c65c1877 | 1551 | |
1e4dd946 SR |
1552 | static inline void |
1553 | __remove_partial(struct kmem_cache_node *n, struct page *page) | |
1554 | { | |
62e346a8 CL |
1555 | list_del(&page->lru); |
1556 | n->nr_partial--; | |
1557 | } | |
1558 | ||
1e4dd946 SR |
1559 | static inline void remove_partial(struct kmem_cache_node *n, |
1560 | struct page *page) | |
1561 | { | |
1562 | lockdep_assert_held(&n->list_lock); | |
1563 | __remove_partial(n, page); | |
1564 | } | |
1565 | ||
81819f0f | 1566 | /* |
7ced3719 CL |
1567 | * Remove slab from the partial list, freeze it and |
1568 | * return the pointer to the freelist. | |
81819f0f | 1569 | * |
497b66f2 | 1570 | * Returns a list of objects or NULL if it fails. |
81819f0f | 1571 | */ |
497b66f2 | 1572 | static inline void *acquire_slab(struct kmem_cache *s, |
acd19fd1 | 1573 | struct kmem_cache_node *n, struct page *page, |
633b0764 | 1574 | int mode, int *objects) |
81819f0f | 1575 | { |
2cfb7455 CL |
1576 | void *freelist; |
1577 | unsigned long counters; | |
1578 | struct page new; | |
1579 | ||
c65c1877 PZ |
1580 | lockdep_assert_held(&n->list_lock); |
1581 | ||
2cfb7455 CL |
1582 | /* |
1583 | * Zap the freelist and set the frozen bit. | |
1584 | * The old freelist is the list of objects for the | |
1585 | * per cpu allocation list. | |
1586 | */ | |
7ced3719 CL |
1587 | freelist = page->freelist; |
1588 | counters = page->counters; | |
1589 | new.counters = counters; | |
633b0764 | 1590 | *objects = new.objects - new.inuse; |
23910c50 | 1591 | if (mode) { |
7ced3719 | 1592 | new.inuse = page->objects; |
23910c50 PE |
1593 | new.freelist = NULL; |
1594 | } else { | |
1595 | new.freelist = freelist; | |
1596 | } | |
2cfb7455 | 1597 | |
a0132ac0 | 1598 | VM_BUG_ON(new.frozen); |
7ced3719 | 1599 | new.frozen = 1; |
2cfb7455 | 1600 | |
7ced3719 | 1601 | if (!__cmpxchg_double_slab(s, page, |
2cfb7455 | 1602 | freelist, counters, |
02d7633f | 1603 | new.freelist, new.counters, |
7ced3719 | 1604 | "acquire_slab")) |
7ced3719 | 1605 | return NULL; |
2cfb7455 CL |
1606 | |
1607 | remove_partial(n, page); | |
7ced3719 | 1608 | WARN_ON(!freelist); |
49e22585 | 1609 | return freelist; |
81819f0f CL |
1610 | } |
1611 | ||
633b0764 | 1612 | static void put_cpu_partial(struct kmem_cache *s, struct page *page, int drain); |
8ba00bb6 | 1613 | static inline bool pfmemalloc_match(struct page *page, gfp_t gfpflags); |
49e22585 | 1614 | |
81819f0f | 1615 | /* |
672bba3a | 1616 | * Try to allocate a partial slab from a specific node. |
81819f0f | 1617 | */ |
8ba00bb6 JK |
1618 | static void *get_partial_node(struct kmem_cache *s, struct kmem_cache_node *n, |
1619 | struct kmem_cache_cpu *c, gfp_t flags) | |
81819f0f | 1620 | { |
49e22585 CL |
1621 | struct page *page, *page2; |
1622 | void *object = NULL; | |
633b0764 JK |
1623 | int available = 0; |
1624 | int objects; | |
81819f0f CL |
1625 | |
1626 | /* | |
1627 | * Racy check. If we mistakenly see no partial slabs then we | |
1628 | * just allocate an empty slab. If we mistakenly try to get a | |
672bba3a CL |
1629 | * partial slab and there is none available then get_partials() |
1630 | * will return NULL. | |
81819f0f CL |
1631 | */ |
1632 | if (!n || !n->nr_partial) | |
1633 | return NULL; | |
1634 | ||
1635 | spin_lock(&n->list_lock); | |
49e22585 | 1636 | list_for_each_entry_safe(page, page2, &n->partial, lru) { |
8ba00bb6 | 1637 | void *t; |
49e22585 | 1638 | |
8ba00bb6 JK |
1639 | if (!pfmemalloc_match(page, flags)) |
1640 | continue; | |
1641 | ||
633b0764 | 1642 | t = acquire_slab(s, n, page, object == NULL, &objects); |
49e22585 CL |
1643 | if (!t) |
1644 | break; | |
1645 | ||
633b0764 | 1646 | available += objects; |
12d79634 | 1647 | if (!object) { |
49e22585 | 1648 | c->page = page; |
49e22585 | 1649 | stat(s, ALLOC_FROM_PARTIAL); |
49e22585 | 1650 | object = t; |
49e22585 | 1651 | } else { |
633b0764 | 1652 | put_cpu_partial(s, page, 0); |
8028dcea | 1653 | stat(s, CPU_PARTIAL_NODE); |
49e22585 | 1654 | } |
345c905d JK |
1655 | if (!kmem_cache_has_cpu_partial(s) |
1656 | || available > s->cpu_partial / 2) | |
49e22585 CL |
1657 | break; |
1658 | ||
497b66f2 | 1659 | } |
81819f0f | 1660 | spin_unlock(&n->list_lock); |
497b66f2 | 1661 | return object; |
81819f0f CL |
1662 | } |
1663 | ||
1664 | /* | |
672bba3a | 1665 | * Get a page from somewhere. Search in increasing NUMA distances. |
81819f0f | 1666 | */ |
de3ec035 | 1667 | static void *get_any_partial(struct kmem_cache *s, gfp_t flags, |
acd19fd1 | 1668 | struct kmem_cache_cpu *c) |
81819f0f CL |
1669 | { |
1670 | #ifdef CONFIG_NUMA | |
1671 | struct zonelist *zonelist; | |
dd1a239f | 1672 | struct zoneref *z; |
54a6eb5c MG |
1673 | struct zone *zone; |
1674 | enum zone_type high_zoneidx = gfp_zone(flags); | |
497b66f2 | 1675 | void *object; |
cc9a6c87 | 1676 | unsigned int cpuset_mems_cookie; |
81819f0f CL |
1677 | |
1678 | /* | |
672bba3a CL |
1679 | * The defrag ratio allows a configuration of the tradeoffs between |
1680 | * inter node defragmentation and node local allocations. A lower | |
1681 | * defrag_ratio increases the tendency to do local allocations | |
1682 | * instead of attempting to obtain partial slabs from other nodes. | |
81819f0f | 1683 | * |
672bba3a CL |
1684 | * If the defrag_ratio is set to 0 then kmalloc() always |
1685 | * returns node local objects. If the ratio is higher then kmalloc() | |
1686 | * may return off node objects because partial slabs are obtained | |
1687 | * from other nodes and filled up. | |
81819f0f | 1688 | * |
6446faa2 | 1689 | * If /sys/kernel/slab/xx/defrag_ratio is set to 100 (which makes |
672bba3a CL |
1690 | * defrag_ratio = 1000) then every (well almost) allocation will |
1691 | * first attempt to defrag slab caches on other nodes. This means | |
1692 | * scanning over all nodes to look for partial slabs which may be | |
1693 | * expensive if we do it every time we are trying to find a slab | |
1694 | * with available objects. | |
81819f0f | 1695 | */ |
9824601e CL |
1696 | if (!s->remote_node_defrag_ratio || |
1697 | get_cycles() % 1024 > s->remote_node_defrag_ratio) | |
81819f0f CL |
1698 | return NULL; |
1699 | ||
cc9a6c87 | 1700 | do { |
d26914d1 | 1701 | cpuset_mems_cookie = read_mems_allowed_begin(); |
2a389610 | 1702 | zonelist = node_zonelist(mempolicy_slab_node(), flags); |
cc9a6c87 MG |
1703 | for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) { |
1704 | struct kmem_cache_node *n; | |
1705 | ||
1706 | n = get_node(s, zone_to_nid(zone)); | |
1707 | ||
dee2f8aa | 1708 | if (n && cpuset_zone_allowed(zone, flags) && |
cc9a6c87 | 1709 | n->nr_partial > s->min_partial) { |
8ba00bb6 | 1710 | object = get_partial_node(s, n, c, flags); |
cc9a6c87 MG |
1711 | if (object) { |
1712 | /* | |
d26914d1 MG |
1713 | * Don't check read_mems_allowed_retry() |
1714 | * here - if mems_allowed was updated in | |
1715 | * parallel, that was a harmless race | |
1716 | * between allocation and the cpuset | |
1717 | * update | |
cc9a6c87 | 1718 | */ |
cc9a6c87 MG |
1719 | return object; |
1720 | } | |
c0ff7453 | 1721 | } |
81819f0f | 1722 | } |
d26914d1 | 1723 | } while (read_mems_allowed_retry(cpuset_mems_cookie)); |
81819f0f CL |
1724 | #endif |
1725 | return NULL; | |
1726 | } | |
1727 | ||
1728 | /* | |
1729 | * Get a partial page, lock it and return it. | |
1730 | */ | |
497b66f2 | 1731 | static void *get_partial(struct kmem_cache *s, gfp_t flags, int node, |
acd19fd1 | 1732 | struct kmem_cache_cpu *c) |
81819f0f | 1733 | { |
497b66f2 | 1734 | void *object; |
a561ce00 JK |
1735 | int searchnode = node; |
1736 | ||
1737 | if (node == NUMA_NO_NODE) | |
1738 | searchnode = numa_mem_id(); | |
1739 | else if (!node_present_pages(node)) | |
1740 | searchnode = node_to_mem_node(node); | |
81819f0f | 1741 | |
8ba00bb6 | 1742 | object = get_partial_node(s, get_node(s, searchnode), c, flags); |
497b66f2 CL |
1743 | if (object || node != NUMA_NO_NODE) |
1744 | return object; | |
81819f0f | 1745 | |
acd19fd1 | 1746 | return get_any_partial(s, flags, c); |
81819f0f CL |
1747 | } |
1748 | ||
8a5ec0ba CL |
1749 | #ifdef CONFIG_PREEMPT |
1750 | /* | |
1751 | * Calculate the next globally unique transaction for disambiguiation | |
1752 | * during cmpxchg. The transactions start with the cpu number and are then | |
1753 | * incremented by CONFIG_NR_CPUS. | |
1754 | */ | |
1755 | #define TID_STEP roundup_pow_of_two(CONFIG_NR_CPUS) | |
1756 | #else | |
1757 | /* | |
1758 | * No preemption supported therefore also no need to check for | |
1759 | * different cpus. | |
1760 | */ | |
1761 | #define TID_STEP 1 | |
1762 | #endif | |
1763 | ||
1764 | static inline unsigned long next_tid(unsigned long tid) | |
1765 | { | |
1766 | return tid + TID_STEP; | |
1767 | } | |
1768 | ||
1769 | static inline unsigned int tid_to_cpu(unsigned long tid) | |
1770 | { | |
1771 | return tid % TID_STEP; | |
1772 | } | |
1773 | ||
1774 | static inline unsigned long tid_to_event(unsigned long tid) | |
1775 | { | |
1776 | return tid / TID_STEP; | |
1777 | } | |
1778 | ||
1779 | static inline unsigned int init_tid(int cpu) | |
1780 | { | |
1781 | return cpu; | |
1782 | } | |
1783 | ||
1784 | static inline void note_cmpxchg_failure(const char *n, | |
1785 | const struct kmem_cache *s, unsigned long tid) | |
1786 | { | |
1787 | #ifdef SLUB_DEBUG_CMPXCHG | |
1788 | unsigned long actual_tid = __this_cpu_read(s->cpu_slab->tid); | |
1789 | ||
f9f58285 | 1790 | pr_info("%s %s: cmpxchg redo ", n, s->name); |
8a5ec0ba CL |
1791 | |
1792 | #ifdef CONFIG_PREEMPT | |
1793 | if (tid_to_cpu(tid) != tid_to_cpu(actual_tid)) | |
f9f58285 | 1794 | pr_warn("due to cpu change %d -> %d\n", |
8a5ec0ba CL |
1795 | tid_to_cpu(tid), tid_to_cpu(actual_tid)); |
1796 | else | |
1797 | #endif | |
1798 | if (tid_to_event(tid) != tid_to_event(actual_tid)) | |
f9f58285 | 1799 | pr_warn("due to cpu running other code. Event %ld->%ld\n", |
8a5ec0ba CL |
1800 | tid_to_event(tid), tid_to_event(actual_tid)); |
1801 | else | |
f9f58285 | 1802 | pr_warn("for unknown reason: actual=%lx was=%lx target=%lx\n", |
8a5ec0ba CL |
1803 | actual_tid, tid, next_tid(tid)); |
1804 | #endif | |
4fdccdfb | 1805 | stat(s, CMPXCHG_DOUBLE_CPU_FAIL); |
8a5ec0ba CL |
1806 | } |
1807 | ||
788e1aad | 1808 | static void init_kmem_cache_cpus(struct kmem_cache *s) |
8a5ec0ba | 1809 | { |
8a5ec0ba CL |
1810 | int cpu; |
1811 | ||
1812 | for_each_possible_cpu(cpu) | |
1813 | per_cpu_ptr(s->cpu_slab, cpu)->tid = init_tid(cpu); | |
8a5ec0ba | 1814 | } |
2cfb7455 | 1815 | |
81819f0f CL |
1816 | /* |
1817 | * Remove the cpu slab | |
1818 | */ | |
d0e0ac97 CG |
1819 | static void deactivate_slab(struct kmem_cache *s, struct page *page, |
1820 | void *freelist) | |
81819f0f | 1821 | { |
2cfb7455 | 1822 | enum slab_modes { M_NONE, M_PARTIAL, M_FULL, M_FREE }; |
2cfb7455 CL |
1823 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); |
1824 | int lock = 0; | |
1825 | enum slab_modes l = M_NONE, m = M_NONE; | |
2cfb7455 | 1826 | void *nextfree; |
136333d1 | 1827 | int tail = DEACTIVATE_TO_HEAD; |
2cfb7455 CL |
1828 | struct page new; |
1829 | struct page old; | |
1830 | ||
1831 | if (page->freelist) { | |
84e554e6 | 1832 | stat(s, DEACTIVATE_REMOTE_FREES); |
136333d1 | 1833 | tail = DEACTIVATE_TO_TAIL; |
2cfb7455 CL |
1834 | } |
1835 | ||
894b8788 | 1836 | /* |
2cfb7455 CL |
1837 | * Stage one: Free all available per cpu objects back |
1838 | * to the page freelist while it is still frozen. Leave the | |
1839 | * last one. | |
1840 | * | |
1841 | * There is no need to take the list->lock because the page | |
1842 | * is still frozen. | |
1843 | */ | |
1844 | while (freelist && (nextfree = get_freepointer(s, freelist))) { | |
1845 | void *prior; | |
1846 | unsigned long counters; | |
1847 | ||
1848 | do { | |
1849 | prior = page->freelist; | |
1850 | counters = page->counters; | |
1851 | set_freepointer(s, freelist, prior); | |
1852 | new.counters = counters; | |
1853 | new.inuse--; | |
a0132ac0 | 1854 | VM_BUG_ON(!new.frozen); |
2cfb7455 | 1855 | |
1d07171c | 1856 | } while (!__cmpxchg_double_slab(s, page, |
2cfb7455 CL |
1857 | prior, counters, |
1858 | freelist, new.counters, | |
1859 | "drain percpu freelist")); | |
1860 | ||
1861 | freelist = nextfree; | |
1862 | } | |
1863 | ||
894b8788 | 1864 | /* |
2cfb7455 CL |
1865 | * Stage two: Ensure that the page is unfrozen while the |
1866 | * list presence reflects the actual number of objects | |
1867 | * during unfreeze. | |
1868 | * | |
1869 | * We setup the list membership and then perform a cmpxchg | |
1870 | * with the count. If there is a mismatch then the page | |
1871 | * is not unfrozen but the page is on the wrong list. | |
1872 | * | |
1873 | * Then we restart the process which may have to remove | |
1874 | * the page from the list that we just put it on again | |
1875 | * because the number of objects in the slab may have | |
1876 | * changed. | |
894b8788 | 1877 | */ |
2cfb7455 | 1878 | redo: |
894b8788 | 1879 | |
2cfb7455 CL |
1880 | old.freelist = page->freelist; |
1881 | old.counters = page->counters; | |
a0132ac0 | 1882 | VM_BUG_ON(!old.frozen); |
7c2e132c | 1883 | |
2cfb7455 CL |
1884 | /* Determine target state of the slab */ |
1885 | new.counters = old.counters; | |
1886 | if (freelist) { | |
1887 | new.inuse--; | |
1888 | set_freepointer(s, freelist, old.freelist); | |
1889 | new.freelist = freelist; | |
1890 | } else | |
1891 | new.freelist = old.freelist; | |
1892 | ||
1893 | new.frozen = 0; | |
1894 | ||
8a5b20ae | 1895 | if (!new.inuse && n->nr_partial >= s->min_partial) |
2cfb7455 CL |
1896 | m = M_FREE; |
1897 | else if (new.freelist) { | |
1898 | m = M_PARTIAL; | |
1899 | if (!lock) { | |
1900 | lock = 1; | |
1901 | /* | |
1902 | * Taking the spinlock removes the possiblity | |
1903 | * that acquire_slab() will see a slab page that | |
1904 | * is frozen | |
1905 | */ | |
1906 | spin_lock(&n->list_lock); | |
1907 | } | |
1908 | } else { | |
1909 | m = M_FULL; | |
1910 | if (kmem_cache_debug(s) && !lock) { | |
1911 | lock = 1; | |
1912 | /* | |
1913 | * This also ensures that the scanning of full | |
1914 | * slabs from diagnostic functions will not see | |
1915 | * any frozen slabs. | |
1916 | */ | |
1917 | spin_lock(&n->list_lock); | |
1918 | } | |
1919 | } | |
1920 | ||
1921 | if (l != m) { | |
1922 | ||
1923 | if (l == M_PARTIAL) | |
1924 | ||
1925 | remove_partial(n, page); | |
1926 | ||
1927 | else if (l == M_FULL) | |
894b8788 | 1928 | |
c65c1877 | 1929 | remove_full(s, n, page); |
2cfb7455 CL |
1930 | |
1931 | if (m == M_PARTIAL) { | |
1932 | ||
1933 | add_partial(n, page, tail); | |
136333d1 | 1934 | stat(s, tail); |
2cfb7455 CL |
1935 | |
1936 | } else if (m == M_FULL) { | |
894b8788 | 1937 | |
2cfb7455 CL |
1938 | stat(s, DEACTIVATE_FULL); |
1939 | add_full(s, n, page); | |
1940 | ||
1941 | } | |
1942 | } | |
1943 | ||
1944 | l = m; | |
1d07171c | 1945 | if (!__cmpxchg_double_slab(s, page, |
2cfb7455 CL |
1946 | old.freelist, old.counters, |
1947 | new.freelist, new.counters, | |
1948 | "unfreezing slab")) | |
1949 | goto redo; | |
1950 | ||
2cfb7455 CL |
1951 | if (lock) |
1952 | spin_unlock(&n->list_lock); | |
1953 | ||
1954 | if (m == M_FREE) { | |
1955 | stat(s, DEACTIVATE_EMPTY); | |
1956 | discard_slab(s, page); | |
1957 | stat(s, FREE_SLAB); | |
894b8788 | 1958 | } |
81819f0f CL |
1959 | } |
1960 | ||
d24ac77f JK |
1961 | /* |
1962 | * Unfreeze all the cpu partial slabs. | |
1963 | * | |
59a09917 CL |
1964 | * This function must be called with interrupts disabled |
1965 | * for the cpu using c (or some other guarantee must be there | |
1966 | * to guarantee no concurrent accesses). | |
d24ac77f | 1967 | */ |
59a09917 CL |
1968 | static void unfreeze_partials(struct kmem_cache *s, |
1969 | struct kmem_cache_cpu *c) | |
49e22585 | 1970 | { |
345c905d | 1971 | #ifdef CONFIG_SLUB_CPU_PARTIAL |
43d77867 | 1972 | struct kmem_cache_node *n = NULL, *n2 = NULL; |
9ada1934 | 1973 | struct page *page, *discard_page = NULL; |
49e22585 CL |
1974 | |
1975 | while ((page = c->partial)) { | |
49e22585 CL |
1976 | struct page new; |
1977 | struct page old; | |
1978 | ||
1979 | c->partial = page->next; | |
43d77867 JK |
1980 | |
1981 | n2 = get_node(s, page_to_nid(page)); | |
1982 | if (n != n2) { | |
1983 | if (n) | |
1984 | spin_unlock(&n->list_lock); | |
1985 | ||
1986 | n = n2; | |
1987 | spin_lock(&n->list_lock); | |
1988 | } | |
49e22585 CL |
1989 | |
1990 | do { | |
1991 | ||
1992 | old.freelist = page->freelist; | |
1993 | old.counters = page->counters; | |
a0132ac0 | 1994 | VM_BUG_ON(!old.frozen); |
49e22585 CL |
1995 | |
1996 | new.counters = old.counters; | |
1997 | new.freelist = old.freelist; | |
1998 | ||
1999 | new.frozen = 0; | |
2000 | ||
d24ac77f | 2001 | } while (!__cmpxchg_double_slab(s, page, |
49e22585 CL |
2002 | old.freelist, old.counters, |
2003 | new.freelist, new.counters, | |
2004 | "unfreezing slab")); | |
2005 | ||
8a5b20ae | 2006 | if (unlikely(!new.inuse && n->nr_partial >= s->min_partial)) { |
9ada1934 SL |
2007 | page->next = discard_page; |
2008 | discard_page = page; | |
43d77867 JK |
2009 | } else { |
2010 | add_partial(n, page, DEACTIVATE_TO_TAIL); | |
2011 | stat(s, FREE_ADD_PARTIAL); | |
49e22585 CL |
2012 | } |
2013 | } | |
2014 | ||
2015 | if (n) | |
2016 | spin_unlock(&n->list_lock); | |
9ada1934 SL |
2017 | |
2018 | while (discard_page) { | |
2019 | page = discard_page; | |
2020 | discard_page = discard_page->next; | |
2021 | ||
2022 | stat(s, DEACTIVATE_EMPTY); | |
2023 | discard_slab(s, page); | |
2024 | stat(s, FREE_SLAB); | |
2025 | } | |
345c905d | 2026 | #endif |
49e22585 CL |
2027 | } |
2028 | ||
2029 | /* | |
2030 | * Put a page that was just frozen (in __slab_free) into a partial page | |
2031 | * slot if available. This is done without interrupts disabled and without | |
2032 | * preemption disabled. The cmpxchg is racy and may put the partial page | |
2033 | * onto a random cpus partial slot. | |
2034 | * | |
2035 | * If we did not find a slot then simply move all the partials to the | |
2036 | * per node partial list. | |
2037 | */ | |
633b0764 | 2038 | static void put_cpu_partial(struct kmem_cache *s, struct page *page, int drain) |
49e22585 | 2039 | { |
345c905d | 2040 | #ifdef CONFIG_SLUB_CPU_PARTIAL |
49e22585 CL |
2041 | struct page *oldpage; |
2042 | int pages; | |
2043 | int pobjects; | |
2044 | ||
d6e0b7fa | 2045 | preempt_disable(); |
49e22585 CL |
2046 | do { |
2047 | pages = 0; | |
2048 | pobjects = 0; | |
2049 | oldpage = this_cpu_read(s->cpu_slab->partial); | |
2050 | ||
2051 | if (oldpage) { | |
2052 | pobjects = oldpage->pobjects; | |
2053 | pages = oldpage->pages; | |
2054 | if (drain && pobjects > s->cpu_partial) { | |
2055 | unsigned long flags; | |
2056 | /* | |
2057 | * partial array is full. Move the existing | |
2058 | * set to the per node partial list. | |
2059 | */ | |
2060 | local_irq_save(flags); | |
59a09917 | 2061 | unfreeze_partials(s, this_cpu_ptr(s->cpu_slab)); |
49e22585 | 2062 | local_irq_restore(flags); |
e24fc410 | 2063 | oldpage = NULL; |
49e22585 CL |
2064 | pobjects = 0; |
2065 | pages = 0; | |
8028dcea | 2066 | stat(s, CPU_PARTIAL_DRAIN); |
49e22585 CL |
2067 | } |
2068 | } | |
2069 | ||
2070 | pages++; | |
2071 | pobjects += page->objects - page->inuse; | |
2072 | ||
2073 | page->pages = pages; | |
2074 | page->pobjects = pobjects; | |
2075 | page->next = oldpage; | |
2076 | ||
d0e0ac97 CG |
2077 | } while (this_cpu_cmpxchg(s->cpu_slab->partial, oldpage, page) |
2078 | != oldpage); | |
d6e0b7fa VD |
2079 | if (unlikely(!s->cpu_partial)) { |
2080 | unsigned long flags; | |
2081 | ||
2082 | local_irq_save(flags); | |
2083 | unfreeze_partials(s, this_cpu_ptr(s->cpu_slab)); | |
2084 | local_irq_restore(flags); | |
2085 | } | |
2086 | preempt_enable(); | |
345c905d | 2087 | #endif |
49e22585 CL |
2088 | } |
2089 | ||
dfb4f096 | 2090 | static inline void flush_slab(struct kmem_cache *s, struct kmem_cache_cpu *c) |
81819f0f | 2091 | { |
84e554e6 | 2092 | stat(s, CPUSLAB_FLUSH); |
c17dda40 CL |
2093 | deactivate_slab(s, c->page, c->freelist); |
2094 | ||
2095 | c->tid = next_tid(c->tid); | |
2096 | c->page = NULL; | |
2097 | c->freelist = NULL; | |
81819f0f CL |
2098 | } |
2099 | ||
2100 | /* | |
2101 | * Flush cpu slab. | |
6446faa2 | 2102 | * |
81819f0f CL |
2103 | * Called from IPI handler with interrupts disabled. |
2104 | */ | |
0c710013 | 2105 | static inline void __flush_cpu_slab(struct kmem_cache *s, int cpu) |
81819f0f | 2106 | { |
9dfc6e68 | 2107 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu); |
81819f0f | 2108 | |
49e22585 CL |
2109 | if (likely(c)) { |
2110 | if (c->page) | |
2111 | flush_slab(s, c); | |
2112 | ||
59a09917 | 2113 | unfreeze_partials(s, c); |
49e22585 | 2114 | } |
81819f0f CL |
2115 | } |
2116 | ||
2117 | static void flush_cpu_slab(void *d) | |
2118 | { | |
2119 | struct kmem_cache *s = d; | |
81819f0f | 2120 | |
dfb4f096 | 2121 | __flush_cpu_slab(s, smp_processor_id()); |
81819f0f CL |
2122 | } |
2123 | ||
a8364d55 GBY |
2124 | static bool has_cpu_slab(int cpu, void *info) |
2125 | { | |
2126 | struct kmem_cache *s = info; | |
2127 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu); | |
2128 | ||
02e1a9cd | 2129 | return c->page || c->partial; |
a8364d55 GBY |
2130 | } |
2131 | ||
81819f0f CL |
2132 | static void flush_all(struct kmem_cache *s) |
2133 | { | |
a8364d55 | 2134 | on_each_cpu_cond(has_cpu_slab, flush_cpu_slab, s, 1, GFP_ATOMIC); |
81819f0f CL |
2135 | } |
2136 | ||
dfb4f096 CL |
2137 | /* |
2138 | * Check if the objects in a per cpu structure fit numa | |
2139 | * locality expectations. | |
2140 | */ | |
57d437d2 | 2141 | static inline int node_match(struct page *page, int node) |
dfb4f096 CL |
2142 | { |
2143 | #ifdef CONFIG_NUMA | |
4d7868e6 | 2144 | if (!page || (node != NUMA_NO_NODE && page_to_nid(page) != node)) |
dfb4f096 CL |
2145 | return 0; |
2146 | #endif | |
2147 | return 1; | |
2148 | } | |
2149 | ||
9a02d699 | 2150 | #ifdef CONFIG_SLUB_DEBUG |
781b2ba6 PE |
2151 | static int count_free(struct page *page) |
2152 | { | |
2153 | return page->objects - page->inuse; | |
2154 | } | |
2155 | ||
9a02d699 DR |
2156 | static inline unsigned long node_nr_objs(struct kmem_cache_node *n) |
2157 | { | |
2158 | return atomic_long_read(&n->total_objects); | |
2159 | } | |
2160 | #endif /* CONFIG_SLUB_DEBUG */ | |
2161 | ||
2162 | #if defined(CONFIG_SLUB_DEBUG) || defined(CONFIG_SYSFS) | |
781b2ba6 PE |
2163 | static unsigned long count_partial(struct kmem_cache_node *n, |
2164 | int (*get_count)(struct page *)) | |
2165 | { | |
2166 | unsigned long flags; | |
2167 | unsigned long x = 0; | |
2168 | struct page *page; | |
2169 | ||
2170 | spin_lock_irqsave(&n->list_lock, flags); | |
2171 | list_for_each_entry(page, &n->partial, lru) | |
2172 | x += get_count(page); | |
2173 | spin_unlock_irqrestore(&n->list_lock, flags); | |
2174 | return x; | |
2175 | } | |
9a02d699 | 2176 | #endif /* CONFIG_SLUB_DEBUG || CONFIG_SYSFS */ |
26c02cf0 | 2177 | |
781b2ba6 PE |
2178 | static noinline void |
2179 | slab_out_of_memory(struct kmem_cache *s, gfp_t gfpflags, int nid) | |
2180 | { | |
9a02d699 DR |
2181 | #ifdef CONFIG_SLUB_DEBUG |
2182 | static DEFINE_RATELIMIT_STATE(slub_oom_rs, DEFAULT_RATELIMIT_INTERVAL, | |
2183 | DEFAULT_RATELIMIT_BURST); | |
781b2ba6 | 2184 | int node; |
fa45dc25 | 2185 | struct kmem_cache_node *n; |
781b2ba6 | 2186 | |
9a02d699 DR |
2187 | if ((gfpflags & __GFP_NOWARN) || !__ratelimit(&slub_oom_rs)) |
2188 | return; | |
2189 | ||
f9f58285 | 2190 | pr_warn("SLUB: Unable to allocate memory on node %d (gfp=0x%x)\n", |
781b2ba6 | 2191 | nid, gfpflags); |
f9f58285 FF |
2192 | pr_warn(" cache: %s, object size: %d, buffer size: %d, default order: %d, min order: %d\n", |
2193 | s->name, s->object_size, s->size, oo_order(s->oo), | |
2194 | oo_order(s->min)); | |
781b2ba6 | 2195 | |
3b0efdfa | 2196 | if (oo_order(s->min) > get_order(s->object_size)) |
f9f58285 FF |
2197 | pr_warn(" %s debugging increased min order, use slub_debug=O to disable.\n", |
2198 | s->name); | |
fa5ec8a1 | 2199 | |
fa45dc25 | 2200 | for_each_kmem_cache_node(s, node, n) { |
781b2ba6 PE |
2201 | unsigned long nr_slabs; |
2202 | unsigned long nr_objs; | |
2203 | unsigned long nr_free; | |
2204 | ||
26c02cf0 AB |
2205 | nr_free = count_partial(n, count_free); |
2206 | nr_slabs = node_nr_slabs(n); | |
2207 | nr_objs = node_nr_objs(n); | |
781b2ba6 | 2208 | |
f9f58285 | 2209 | pr_warn(" node %d: slabs: %ld, objs: %ld, free: %ld\n", |
781b2ba6 PE |
2210 | node, nr_slabs, nr_objs, nr_free); |
2211 | } | |
9a02d699 | 2212 | #endif |
781b2ba6 PE |
2213 | } |
2214 | ||
497b66f2 CL |
2215 | static inline void *new_slab_objects(struct kmem_cache *s, gfp_t flags, |
2216 | int node, struct kmem_cache_cpu **pc) | |
2217 | { | |
6faa6833 | 2218 | void *freelist; |
188fd063 CL |
2219 | struct kmem_cache_cpu *c = *pc; |
2220 | struct page *page; | |
497b66f2 | 2221 | |
188fd063 | 2222 | freelist = get_partial(s, flags, node, c); |
497b66f2 | 2223 | |
188fd063 CL |
2224 | if (freelist) |
2225 | return freelist; | |
2226 | ||
2227 | page = new_slab(s, flags, node); | |
497b66f2 | 2228 | if (page) { |
7c8e0181 | 2229 | c = raw_cpu_ptr(s->cpu_slab); |
497b66f2 CL |
2230 | if (c->page) |
2231 | flush_slab(s, c); | |
2232 | ||
2233 | /* | |
2234 | * No other reference to the page yet so we can | |
2235 | * muck around with it freely without cmpxchg | |
2236 | */ | |
6faa6833 | 2237 | freelist = page->freelist; |
497b66f2 CL |
2238 | page->freelist = NULL; |
2239 | ||
2240 | stat(s, ALLOC_SLAB); | |
497b66f2 CL |
2241 | c->page = page; |
2242 | *pc = c; | |
2243 | } else | |
6faa6833 | 2244 | freelist = NULL; |
497b66f2 | 2245 | |
6faa6833 | 2246 | return freelist; |
497b66f2 CL |
2247 | } |
2248 | ||
072bb0aa MG |
2249 | static inline bool pfmemalloc_match(struct page *page, gfp_t gfpflags) |
2250 | { | |
2251 | if (unlikely(PageSlabPfmemalloc(page))) | |
2252 | return gfp_pfmemalloc_allowed(gfpflags); | |
2253 | ||
2254 | return true; | |
2255 | } | |
2256 | ||
213eeb9f | 2257 | /* |
d0e0ac97 CG |
2258 | * Check the page->freelist of a page and either transfer the freelist to the |
2259 | * per cpu freelist or deactivate the page. | |
213eeb9f CL |
2260 | * |
2261 | * The page is still frozen if the return value is not NULL. | |
2262 | * | |
2263 | * If this function returns NULL then the page has been unfrozen. | |
d24ac77f JK |
2264 | * |
2265 | * This function must be called with interrupt disabled. | |
213eeb9f CL |
2266 | */ |
2267 | static inline void *get_freelist(struct kmem_cache *s, struct page *page) | |
2268 | { | |
2269 | struct page new; | |
2270 | unsigned long counters; | |
2271 | void *freelist; | |
2272 | ||
2273 | do { | |
2274 | freelist = page->freelist; | |
2275 | counters = page->counters; | |
6faa6833 | 2276 | |
213eeb9f | 2277 | new.counters = counters; |
a0132ac0 | 2278 | VM_BUG_ON(!new.frozen); |
213eeb9f CL |
2279 | |
2280 | new.inuse = page->objects; | |
2281 | new.frozen = freelist != NULL; | |
2282 | ||
d24ac77f | 2283 | } while (!__cmpxchg_double_slab(s, page, |
213eeb9f CL |
2284 | freelist, counters, |
2285 | NULL, new.counters, | |
2286 | "get_freelist")); | |
2287 | ||
2288 | return freelist; | |
2289 | } | |
2290 | ||
81819f0f | 2291 | /* |
894b8788 CL |
2292 | * Slow path. The lockless freelist is empty or we need to perform |
2293 | * debugging duties. | |
2294 | * | |
894b8788 CL |
2295 | * Processing is still very fast if new objects have been freed to the |
2296 | * regular freelist. In that case we simply take over the regular freelist | |
2297 | * as the lockless freelist and zap the regular freelist. | |
81819f0f | 2298 | * |
894b8788 CL |
2299 | * If that is not working then we fall back to the partial lists. We take the |
2300 | * first element of the freelist as the object to allocate now and move the | |
2301 | * rest of the freelist to the lockless freelist. | |
81819f0f | 2302 | * |
894b8788 | 2303 | * And if we were unable to get a new slab from the partial slab lists then |
6446faa2 CL |
2304 | * we need to allocate a new slab. This is the slowest path since it involves |
2305 | * a call to the page allocator and the setup of a new slab. | |
81819f0f | 2306 | */ |
ce71e27c EGM |
2307 | static void *__slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node, |
2308 | unsigned long addr, struct kmem_cache_cpu *c) | |
81819f0f | 2309 | { |
6faa6833 | 2310 | void *freelist; |
f6e7def7 | 2311 | struct page *page; |
8a5ec0ba CL |
2312 | unsigned long flags; |
2313 | ||
2314 | local_irq_save(flags); | |
2315 | #ifdef CONFIG_PREEMPT | |
2316 | /* | |
2317 | * We may have been preempted and rescheduled on a different | |
2318 | * cpu before disabling interrupts. Need to reload cpu area | |
2319 | * pointer. | |
2320 | */ | |
2321 | c = this_cpu_ptr(s->cpu_slab); | |
8a5ec0ba | 2322 | #endif |
81819f0f | 2323 | |
f6e7def7 CL |
2324 | page = c->page; |
2325 | if (!page) | |
81819f0f | 2326 | goto new_slab; |
49e22585 | 2327 | redo: |
6faa6833 | 2328 | |
57d437d2 | 2329 | if (unlikely(!node_match(page, node))) { |
a561ce00 JK |
2330 | int searchnode = node; |
2331 | ||
2332 | if (node != NUMA_NO_NODE && !node_present_pages(node)) | |
2333 | searchnode = node_to_mem_node(node); | |
2334 | ||
2335 | if (unlikely(!node_match(page, searchnode))) { | |
2336 | stat(s, ALLOC_NODE_MISMATCH); | |
2337 | deactivate_slab(s, page, c->freelist); | |
2338 | c->page = NULL; | |
2339 | c->freelist = NULL; | |
2340 | goto new_slab; | |
2341 | } | |
fc59c053 | 2342 | } |
6446faa2 | 2343 | |
072bb0aa MG |
2344 | /* |
2345 | * By rights, we should be searching for a slab page that was | |
2346 | * PFMEMALLOC but right now, we are losing the pfmemalloc | |
2347 | * information when the page leaves the per-cpu allocator | |
2348 | */ | |
2349 | if (unlikely(!pfmemalloc_match(page, gfpflags))) { | |
2350 | deactivate_slab(s, page, c->freelist); | |
2351 | c->page = NULL; | |
2352 | c->freelist = NULL; | |
2353 | goto new_slab; | |
2354 | } | |
2355 | ||
73736e03 | 2356 | /* must check again c->freelist in case of cpu migration or IRQ */ |
6faa6833 CL |
2357 | freelist = c->freelist; |
2358 | if (freelist) | |
73736e03 | 2359 | goto load_freelist; |
03e404af | 2360 | |
f6e7def7 | 2361 | freelist = get_freelist(s, page); |
6446faa2 | 2362 | |
6faa6833 | 2363 | if (!freelist) { |
03e404af CL |
2364 | c->page = NULL; |
2365 | stat(s, DEACTIVATE_BYPASS); | |
fc59c053 | 2366 | goto new_slab; |
03e404af | 2367 | } |
6446faa2 | 2368 | |
84e554e6 | 2369 | stat(s, ALLOC_REFILL); |
6446faa2 | 2370 | |
894b8788 | 2371 | load_freelist: |
507effea CL |
2372 | /* |
2373 | * freelist is pointing to the list of objects to be used. | |
2374 | * page is pointing to the page from which the objects are obtained. | |
2375 | * That page must be frozen for per cpu allocations to work. | |
2376 | */ | |
a0132ac0 | 2377 | VM_BUG_ON(!c->page->frozen); |
6faa6833 | 2378 | c->freelist = get_freepointer(s, freelist); |
8a5ec0ba CL |
2379 | c->tid = next_tid(c->tid); |
2380 | local_irq_restore(flags); | |
6faa6833 | 2381 | return freelist; |
81819f0f | 2382 | |
81819f0f | 2383 | new_slab: |
2cfb7455 | 2384 | |
49e22585 | 2385 | if (c->partial) { |
f6e7def7 CL |
2386 | page = c->page = c->partial; |
2387 | c->partial = page->next; | |
49e22585 CL |
2388 | stat(s, CPU_PARTIAL_ALLOC); |
2389 | c->freelist = NULL; | |
2390 | goto redo; | |
81819f0f CL |
2391 | } |
2392 | ||
188fd063 | 2393 | freelist = new_slab_objects(s, gfpflags, node, &c); |
01ad8a7b | 2394 | |
f4697436 | 2395 | if (unlikely(!freelist)) { |
9a02d699 | 2396 | slab_out_of_memory(s, gfpflags, node); |
f4697436 CL |
2397 | local_irq_restore(flags); |
2398 | return NULL; | |
81819f0f | 2399 | } |
2cfb7455 | 2400 | |
f6e7def7 | 2401 | page = c->page; |
5091b74a | 2402 | if (likely(!kmem_cache_debug(s) && pfmemalloc_match(page, gfpflags))) |
4b6f0750 | 2403 | goto load_freelist; |
2cfb7455 | 2404 | |
497b66f2 | 2405 | /* Only entered in the debug case */ |
d0e0ac97 CG |
2406 | if (kmem_cache_debug(s) && |
2407 | !alloc_debug_processing(s, page, freelist, addr)) | |
497b66f2 | 2408 | goto new_slab; /* Slab failed checks. Next slab needed */ |
894b8788 | 2409 | |
f6e7def7 | 2410 | deactivate_slab(s, page, get_freepointer(s, freelist)); |
c17dda40 CL |
2411 | c->page = NULL; |
2412 | c->freelist = NULL; | |
a71ae47a | 2413 | local_irq_restore(flags); |
6faa6833 | 2414 | return freelist; |
894b8788 CL |
2415 | } |
2416 | ||
2417 | /* | |
2418 | * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc) | |
2419 | * have the fastpath folded into their functions. So no function call | |
2420 | * overhead for requests that can be satisfied on the fastpath. | |
2421 | * | |
2422 | * The fastpath works by first checking if the lockless freelist can be used. | |
2423 | * If not then __slab_alloc is called for slow processing. | |
2424 | * | |
2425 | * Otherwise we can simply pick the next object from the lockless free list. | |
2426 | */ | |
2b847c3c | 2427 | static __always_inline void *slab_alloc_node(struct kmem_cache *s, |
ce71e27c | 2428 | gfp_t gfpflags, int node, unsigned long addr) |
894b8788 | 2429 | { |
894b8788 | 2430 | void **object; |
dfb4f096 | 2431 | struct kmem_cache_cpu *c; |
57d437d2 | 2432 | struct page *page; |
8a5ec0ba | 2433 | unsigned long tid; |
1f84260c | 2434 | |
8135be5a VD |
2435 | s = slab_pre_alloc_hook(s, gfpflags); |
2436 | if (!s) | |
773ff60e | 2437 | return NULL; |
8a5ec0ba | 2438 | redo: |
8a5ec0ba CL |
2439 | /* |
2440 | * Must read kmem_cache cpu data via this cpu ptr. Preemption is | |
2441 | * enabled. We may switch back and forth between cpus while | |
2442 | * reading from one cpu area. That does not matter as long | |
2443 | * as we end up on the original cpu again when doing the cmpxchg. | |
7cccd80b | 2444 | * |
9aabf810 JK |
2445 | * We should guarantee that tid and kmem_cache are retrieved on |
2446 | * the same cpu. It could be different if CONFIG_PREEMPT so we need | |
2447 | * to check if it is matched or not. | |
8a5ec0ba | 2448 | */ |
9aabf810 JK |
2449 | do { |
2450 | tid = this_cpu_read(s->cpu_slab->tid); | |
2451 | c = raw_cpu_ptr(s->cpu_slab); | |
859b7a0e MR |
2452 | } while (IS_ENABLED(CONFIG_PREEMPT) && |
2453 | unlikely(tid != READ_ONCE(c->tid))); | |
9aabf810 JK |
2454 | |
2455 | /* | |
2456 | * Irqless object alloc/free algorithm used here depends on sequence | |
2457 | * of fetching cpu_slab's data. tid should be fetched before anything | |
2458 | * on c to guarantee that object and page associated with previous tid | |
2459 | * won't be used with current tid. If we fetch tid first, object and | |
2460 | * page could be one associated with next tid and our alloc/free | |
2461 | * request will be failed. In this case, we will retry. So, no problem. | |
2462 | */ | |
2463 | barrier(); | |
8a5ec0ba | 2464 | |
8a5ec0ba CL |
2465 | /* |
2466 | * The transaction ids are globally unique per cpu and per operation on | |
2467 | * a per cpu queue. Thus they can be guarantee that the cmpxchg_double | |
2468 | * occurs on the right processor and that there was no operation on the | |
2469 | * linked list in between. | |
2470 | */ | |
8a5ec0ba | 2471 | |
9dfc6e68 | 2472 | object = c->freelist; |
57d437d2 | 2473 | page = c->page; |
8eae1492 | 2474 | if (unlikely(!object || !node_match(page, node))) { |
dfb4f096 | 2475 | object = __slab_alloc(s, gfpflags, node, addr, c); |
8eae1492 DH |
2476 | stat(s, ALLOC_SLOWPATH); |
2477 | } else { | |
0ad9500e ED |
2478 | void *next_object = get_freepointer_safe(s, object); |
2479 | ||
8a5ec0ba | 2480 | /* |
25985edc | 2481 | * The cmpxchg will only match if there was no additional |
8a5ec0ba CL |
2482 | * operation and if we are on the right processor. |
2483 | * | |
d0e0ac97 CG |
2484 | * The cmpxchg does the following atomically (without lock |
2485 | * semantics!) | |
8a5ec0ba CL |
2486 | * 1. Relocate first pointer to the current per cpu area. |
2487 | * 2. Verify that tid and freelist have not been changed | |
2488 | * 3. If they were not changed replace tid and freelist | |
2489 | * | |
d0e0ac97 CG |
2490 | * Since this is without lock semantics the protection is only |
2491 | * against code executing on this cpu *not* from access by | |
2492 | * other cpus. | |
8a5ec0ba | 2493 | */ |
933393f5 | 2494 | if (unlikely(!this_cpu_cmpxchg_double( |
8a5ec0ba CL |
2495 | s->cpu_slab->freelist, s->cpu_slab->tid, |
2496 | object, tid, | |
0ad9500e | 2497 | next_object, next_tid(tid)))) { |
8a5ec0ba CL |
2498 | |
2499 | note_cmpxchg_failure("slab_alloc", s, tid); | |
2500 | goto redo; | |
2501 | } | |
0ad9500e | 2502 | prefetch_freepointer(s, next_object); |
84e554e6 | 2503 | stat(s, ALLOC_FASTPATH); |
894b8788 | 2504 | } |
8a5ec0ba | 2505 | |
74e2134f | 2506 | if (unlikely(gfpflags & __GFP_ZERO) && object) |
3b0efdfa | 2507 | memset(object, 0, s->object_size); |
d07dbea4 | 2508 | |
c016b0bd | 2509 | slab_post_alloc_hook(s, gfpflags, object); |
5a896d9e | 2510 | |
894b8788 | 2511 | return object; |
81819f0f CL |
2512 | } |
2513 | ||
2b847c3c EG |
2514 | static __always_inline void *slab_alloc(struct kmem_cache *s, |
2515 | gfp_t gfpflags, unsigned long addr) | |
2516 | { | |
2517 | return slab_alloc_node(s, gfpflags, NUMA_NO_NODE, addr); | |
2518 | } | |
2519 | ||
81819f0f CL |
2520 | void *kmem_cache_alloc(struct kmem_cache *s, gfp_t gfpflags) |
2521 | { | |
2b847c3c | 2522 | void *ret = slab_alloc(s, gfpflags, _RET_IP_); |
5b882be4 | 2523 | |
d0e0ac97 CG |
2524 | trace_kmem_cache_alloc(_RET_IP_, ret, s->object_size, |
2525 | s->size, gfpflags); | |
5b882be4 EGM |
2526 | |
2527 | return ret; | |
81819f0f CL |
2528 | } |
2529 | EXPORT_SYMBOL(kmem_cache_alloc); | |
2530 | ||
0f24f128 | 2531 | #ifdef CONFIG_TRACING |
4a92379b RK |
2532 | void *kmem_cache_alloc_trace(struct kmem_cache *s, gfp_t gfpflags, size_t size) |
2533 | { | |
2b847c3c | 2534 | void *ret = slab_alloc(s, gfpflags, _RET_IP_); |
4a92379b | 2535 | trace_kmalloc(_RET_IP_, ret, size, s->size, gfpflags); |
0316bec2 | 2536 | kasan_kmalloc(s, ret, size); |
4a92379b RK |
2537 | return ret; |
2538 | } | |
2539 | EXPORT_SYMBOL(kmem_cache_alloc_trace); | |
5b882be4 EGM |
2540 | #endif |
2541 | ||
81819f0f CL |
2542 | #ifdef CONFIG_NUMA |
2543 | void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t gfpflags, int node) | |
2544 | { | |
2b847c3c | 2545 | void *ret = slab_alloc_node(s, gfpflags, node, _RET_IP_); |
5b882be4 | 2546 | |
ca2b84cb | 2547 | trace_kmem_cache_alloc_node(_RET_IP_, ret, |
3b0efdfa | 2548 | s->object_size, s->size, gfpflags, node); |
5b882be4 EGM |
2549 | |
2550 | return ret; | |
81819f0f CL |
2551 | } |
2552 | EXPORT_SYMBOL(kmem_cache_alloc_node); | |
81819f0f | 2553 | |
0f24f128 | 2554 | #ifdef CONFIG_TRACING |
4a92379b | 2555 | void *kmem_cache_alloc_node_trace(struct kmem_cache *s, |
5b882be4 | 2556 | gfp_t gfpflags, |
4a92379b | 2557 | int node, size_t size) |
5b882be4 | 2558 | { |
2b847c3c | 2559 | void *ret = slab_alloc_node(s, gfpflags, node, _RET_IP_); |
4a92379b RK |
2560 | |
2561 | trace_kmalloc_node(_RET_IP_, ret, | |
2562 | size, s->size, gfpflags, node); | |
0316bec2 AR |
2563 | |
2564 | kasan_kmalloc(s, ret, size); | |
4a92379b | 2565 | return ret; |
5b882be4 | 2566 | } |
4a92379b | 2567 | EXPORT_SYMBOL(kmem_cache_alloc_node_trace); |
5b882be4 | 2568 | #endif |
5d1f57e4 | 2569 | #endif |
5b882be4 | 2570 | |
81819f0f | 2571 | /* |
94e4d712 | 2572 | * Slow path handling. This may still be called frequently since objects |
894b8788 | 2573 | * have a longer lifetime than the cpu slabs in most processing loads. |
81819f0f | 2574 | * |
894b8788 CL |
2575 | * So we still attempt to reduce cache line usage. Just take the slab |
2576 | * lock and free the item. If there is no additional partial page | |
2577 | * handling required then we can return immediately. | |
81819f0f | 2578 | */ |
894b8788 | 2579 | static void __slab_free(struct kmem_cache *s, struct page *page, |
ff12059e | 2580 | void *x, unsigned long addr) |
81819f0f CL |
2581 | { |
2582 | void *prior; | |
2583 | void **object = (void *)x; | |
2cfb7455 | 2584 | int was_frozen; |
2cfb7455 CL |
2585 | struct page new; |
2586 | unsigned long counters; | |
2587 | struct kmem_cache_node *n = NULL; | |
61728d1e | 2588 | unsigned long uninitialized_var(flags); |
81819f0f | 2589 | |
8a5ec0ba | 2590 | stat(s, FREE_SLOWPATH); |
81819f0f | 2591 | |
19c7ff9e CL |
2592 | if (kmem_cache_debug(s) && |
2593 | !(n = free_debug_processing(s, page, x, addr, &flags))) | |
80f08c19 | 2594 | return; |
6446faa2 | 2595 | |
2cfb7455 | 2596 | do { |
837d678d JK |
2597 | if (unlikely(n)) { |
2598 | spin_unlock_irqrestore(&n->list_lock, flags); | |
2599 | n = NULL; | |
2600 | } | |
2cfb7455 CL |
2601 | prior = page->freelist; |
2602 | counters = page->counters; | |
2603 | set_freepointer(s, object, prior); | |
2604 | new.counters = counters; | |
2605 | was_frozen = new.frozen; | |
2606 | new.inuse--; | |
837d678d | 2607 | if ((!new.inuse || !prior) && !was_frozen) { |
49e22585 | 2608 | |
c65c1877 | 2609 | if (kmem_cache_has_cpu_partial(s) && !prior) { |
49e22585 CL |
2610 | |
2611 | /* | |
d0e0ac97 CG |
2612 | * Slab was on no list before and will be |
2613 | * partially empty | |
2614 | * We can defer the list move and instead | |
2615 | * freeze it. | |
49e22585 CL |
2616 | */ |
2617 | new.frozen = 1; | |
2618 | ||
c65c1877 | 2619 | } else { /* Needs to be taken off a list */ |
49e22585 | 2620 | |
b455def2 | 2621 | n = get_node(s, page_to_nid(page)); |
49e22585 CL |
2622 | /* |
2623 | * Speculatively acquire the list_lock. | |
2624 | * If the cmpxchg does not succeed then we may | |
2625 | * drop the list_lock without any processing. | |
2626 | * | |
2627 | * Otherwise the list_lock will synchronize with | |
2628 | * other processors updating the list of slabs. | |
2629 | */ | |
2630 | spin_lock_irqsave(&n->list_lock, flags); | |
2631 | ||
2632 | } | |
2cfb7455 | 2633 | } |
81819f0f | 2634 | |
2cfb7455 CL |
2635 | } while (!cmpxchg_double_slab(s, page, |
2636 | prior, counters, | |
2637 | object, new.counters, | |
2638 | "__slab_free")); | |
81819f0f | 2639 | |
2cfb7455 | 2640 | if (likely(!n)) { |
49e22585 CL |
2641 | |
2642 | /* | |
2643 | * If we just froze the page then put it onto the | |
2644 | * per cpu partial list. | |
2645 | */ | |
8028dcea | 2646 | if (new.frozen && !was_frozen) { |
49e22585 | 2647 | put_cpu_partial(s, page, 1); |
8028dcea AS |
2648 | stat(s, CPU_PARTIAL_FREE); |
2649 | } | |
49e22585 | 2650 | /* |
2cfb7455 CL |
2651 | * The list lock was not taken therefore no list |
2652 | * activity can be necessary. | |
2653 | */ | |
b455def2 L |
2654 | if (was_frozen) |
2655 | stat(s, FREE_FROZEN); | |
2656 | return; | |
2657 | } | |
81819f0f | 2658 | |
8a5b20ae | 2659 | if (unlikely(!new.inuse && n->nr_partial >= s->min_partial)) |
837d678d JK |
2660 | goto slab_empty; |
2661 | ||
81819f0f | 2662 | /* |
837d678d JK |
2663 | * Objects left in the slab. If it was not on the partial list before |
2664 | * then add it. | |
81819f0f | 2665 | */ |
345c905d JK |
2666 | if (!kmem_cache_has_cpu_partial(s) && unlikely(!prior)) { |
2667 | if (kmem_cache_debug(s)) | |
c65c1877 | 2668 | remove_full(s, n, page); |
837d678d JK |
2669 | add_partial(n, page, DEACTIVATE_TO_TAIL); |
2670 | stat(s, FREE_ADD_PARTIAL); | |
8ff12cfc | 2671 | } |
80f08c19 | 2672 | spin_unlock_irqrestore(&n->list_lock, flags); |
81819f0f CL |
2673 | return; |
2674 | ||
2675 | slab_empty: | |
a973e9dd | 2676 | if (prior) { |
81819f0f | 2677 | /* |
6fbabb20 | 2678 | * Slab on the partial list. |
81819f0f | 2679 | */ |
5cc6eee8 | 2680 | remove_partial(n, page); |
84e554e6 | 2681 | stat(s, FREE_REMOVE_PARTIAL); |
c65c1877 | 2682 | } else { |
6fbabb20 | 2683 | /* Slab must be on the full list */ |
c65c1877 PZ |
2684 | remove_full(s, n, page); |
2685 | } | |
2cfb7455 | 2686 | |
80f08c19 | 2687 | spin_unlock_irqrestore(&n->list_lock, flags); |
84e554e6 | 2688 | stat(s, FREE_SLAB); |
81819f0f | 2689 | discard_slab(s, page); |
81819f0f CL |
2690 | } |
2691 | ||
894b8788 CL |
2692 | /* |
2693 | * Fastpath with forced inlining to produce a kfree and kmem_cache_free that | |
2694 | * can perform fastpath freeing without additional function calls. | |
2695 | * | |
2696 | * The fastpath is only possible if we are freeing to the current cpu slab | |
2697 | * of this processor. This typically the case if we have just allocated | |
2698 | * the item before. | |
2699 | * | |
2700 | * If fastpath is not possible then fall back to __slab_free where we deal | |
2701 | * with all sorts of special processing. | |
2702 | */ | |
06428780 | 2703 | static __always_inline void slab_free(struct kmem_cache *s, |
ce71e27c | 2704 | struct page *page, void *x, unsigned long addr) |
894b8788 CL |
2705 | { |
2706 | void **object = (void *)x; | |
dfb4f096 | 2707 | struct kmem_cache_cpu *c; |
8a5ec0ba | 2708 | unsigned long tid; |
1f84260c | 2709 | |
c016b0bd CL |
2710 | slab_free_hook(s, x); |
2711 | ||
8a5ec0ba CL |
2712 | redo: |
2713 | /* | |
2714 | * Determine the currently cpus per cpu slab. | |
2715 | * The cpu may change afterward. However that does not matter since | |
2716 | * data is retrieved via this pointer. If we are on the same cpu | |
2717 | * during the cmpxchg then the free will succedd. | |
2718 | */ | |
9aabf810 JK |
2719 | do { |
2720 | tid = this_cpu_read(s->cpu_slab->tid); | |
2721 | c = raw_cpu_ptr(s->cpu_slab); | |
859b7a0e MR |
2722 | } while (IS_ENABLED(CONFIG_PREEMPT) && |
2723 | unlikely(tid != READ_ONCE(c->tid))); | |
c016b0bd | 2724 | |
9aabf810 JK |
2725 | /* Same with comment on barrier() in slab_alloc_node() */ |
2726 | barrier(); | |
c016b0bd | 2727 | |
442b06bc | 2728 | if (likely(page == c->page)) { |
ff12059e | 2729 | set_freepointer(s, object, c->freelist); |
8a5ec0ba | 2730 | |
933393f5 | 2731 | if (unlikely(!this_cpu_cmpxchg_double( |
8a5ec0ba CL |
2732 | s->cpu_slab->freelist, s->cpu_slab->tid, |
2733 | c->freelist, tid, | |
2734 | object, next_tid(tid)))) { | |
2735 | ||
2736 | note_cmpxchg_failure("slab_free", s, tid); | |
2737 | goto redo; | |
2738 | } | |
84e554e6 | 2739 | stat(s, FREE_FASTPATH); |
894b8788 | 2740 | } else |
ff12059e | 2741 | __slab_free(s, page, x, addr); |
894b8788 | 2742 | |
894b8788 CL |
2743 | } |
2744 | ||
81819f0f CL |
2745 | void kmem_cache_free(struct kmem_cache *s, void *x) |
2746 | { | |
b9ce5ef4 GC |
2747 | s = cache_from_obj(s, x); |
2748 | if (!s) | |
79576102 | 2749 | return; |
b9ce5ef4 | 2750 | slab_free(s, virt_to_head_page(x), x, _RET_IP_); |
ca2b84cb | 2751 | trace_kmem_cache_free(_RET_IP_, x); |
81819f0f CL |
2752 | } |
2753 | EXPORT_SYMBOL(kmem_cache_free); | |
2754 | ||
81819f0f | 2755 | /* |
672bba3a CL |
2756 | * Object placement in a slab is made very easy because we always start at |
2757 | * offset 0. If we tune the size of the object to the alignment then we can | |
2758 | * get the required alignment by putting one properly sized object after | |
2759 | * another. | |
81819f0f CL |
2760 | * |
2761 | * Notice that the allocation order determines the sizes of the per cpu | |
2762 | * caches. Each processor has always one slab available for allocations. | |
2763 | * Increasing the allocation order reduces the number of times that slabs | |
672bba3a | 2764 | * must be moved on and off the partial lists and is therefore a factor in |
81819f0f | 2765 | * locking overhead. |
81819f0f CL |
2766 | */ |
2767 | ||
2768 | /* | |
2769 | * Mininum / Maximum order of slab pages. This influences locking overhead | |
2770 | * and slab fragmentation. A higher order reduces the number of partial slabs | |
2771 | * and increases the number of allocations possible without having to | |
2772 | * take the list_lock. | |
2773 | */ | |
2774 | static int slub_min_order; | |
114e9e89 | 2775 | static int slub_max_order = PAGE_ALLOC_COSTLY_ORDER; |
9b2cd506 | 2776 | static int slub_min_objects; |
81819f0f | 2777 | |
81819f0f CL |
2778 | /* |
2779 | * Calculate the order of allocation given an slab object size. | |
2780 | * | |
672bba3a CL |
2781 | * The order of allocation has significant impact on performance and other |
2782 | * system components. Generally order 0 allocations should be preferred since | |
2783 | * order 0 does not cause fragmentation in the page allocator. Larger objects | |
2784 | * be problematic to put into order 0 slabs because there may be too much | |
c124f5b5 | 2785 | * unused space left. We go to a higher order if more than 1/16th of the slab |
672bba3a CL |
2786 | * would be wasted. |
2787 | * | |
2788 | * In order to reach satisfactory performance we must ensure that a minimum | |
2789 | * number of objects is in one slab. Otherwise we may generate too much | |
2790 | * activity on the partial lists which requires taking the list_lock. This is | |
2791 | * less a concern for large slabs though which are rarely used. | |
81819f0f | 2792 | * |
672bba3a CL |
2793 | * slub_max_order specifies the order where we begin to stop considering the |
2794 | * number of objects in a slab as critical. If we reach slub_max_order then | |
2795 | * we try to keep the page order as low as possible. So we accept more waste | |
2796 | * of space in favor of a small page order. | |
81819f0f | 2797 | * |
672bba3a CL |
2798 | * Higher order allocations also allow the placement of more objects in a |
2799 | * slab and thereby reduce object handling overhead. If the user has | |
2800 | * requested a higher mininum order then we start with that one instead of | |
2801 | * the smallest order which will fit the object. | |
81819f0f | 2802 | */ |
5e6d444e | 2803 | static inline int slab_order(int size, int min_objects, |
ab9a0f19 | 2804 | int max_order, int fract_leftover, int reserved) |
81819f0f CL |
2805 | { |
2806 | int order; | |
2807 | int rem; | |
6300ea75 | 2808 | int min_order = slub_min_order; |
81819f0f | 2809 | |
ab9a0f19 | 2810 | if (order_objects(min_order, size, reserved) > MAX_OBJS_PER_PAGE) |
210b5c06 | 2811 | return get_order(size * MAX_OBJS_PER_PAGE) - 1; |
39b26464 | 2812 | |
6300ea75 | 2813 | for (order = max(min_order, |
5e6d444e CL |
2814 | fls(min_objects * size - 1) - PAGE_SHIFT); |
2815 | order <= max_order; order++) { | |
81819f0f | 2816 | |
5e6d444e | 2817 | unsigned long slab_size = PAGE_SIZE << order; |
81819f0f | 2818 | |
ab9a0f19 | 2819 | if (slab_size < min_objects * size + reserved) |
81819f0f CL |
2820 | continue; |
2821 | ||
ab9a0f19 | 2822 | rem = (slab_size - reserved) % size; |
81819f0f | 2823 | |
5e6d444e | 2824 | if (rem <= slab_size / fract_leftover) |
81819f0f CL |
2825 | break; |
2826 | ||
2827 | } | |
672bba3a | 2828 | |
81819f0f CL |
2829 | return order; |
2830 | } | |
2831 | ||
ab9a0f19 | 2832 | static inline int calculate_order(int size, int reserved) |
5e6d444e CL |
2833 | { |
2834 | int order; | |
2835 | int min_objects; | |
2836 | int fraction; | |
e8120ff1 | 2837 | int max_objects; |
5e6d444e CL |
2838 | |
2839 | /* | |
2840 | * Attempt to find best configuration for a slab. This | |
2841 | * works by first attempting to generate a layout with | |
2842 | * the best configuration and backing off gradually. | |
2843 | * | |
2844 | * First we reduce the acceptable waste in a slab. Then | |
2845 | * we reduce the minimum objects required in a slab. | |
2846 | */ | |
2847 | min_objects = slub_min_objects; | |
9b2cd506 CL |
2848 | if (!min_objects) |
2849 | min_objects = 4 * (fls(nr_cpu_ids) + 1); | |
ab9a0f19 | 2850 | max_objects = order_objects(slub_max_order, size, reserved); |
e8120ff1 ZY |
2851 | min_objects = min(min_objects, max_objects); |
2852 | ||
5e6d444e | 2853 | while (min_objects > 1) { |
c124f5b5 | 2854 | fraction = 16; |
5e6d444e CL |
2855 | while (fraction >= 4) { |
2856 | order = slab_order(size, min_objects, | |
ab9a0f19 | 2857 | slub_max_order, fraction, reserved); |
5e6d444e CL |
2858 | if (order <= slub_max_order) |
2859 | return order; | |
2860 | fraction /= 2; | |
2861 | } | |
5086c389 | 2862 | min_objects--; |
5e6d444e CL |
2863 | } |
2864 | ||
2865 | /* | |
2866 | * We were unable to place multiple objects in a slab. Now | |
2867 | * lets see if we can place a single object there. | |
2868 | */ | |
ab9a0f19 | 2869 | order = slab_order(size, 1, slub_max_order, 1, reserved); |
5e6d444e CL |
2870 | if (order <= slub_max_order) |
2871 | return order; | |
2872 | ||
2873 | /* | |
2874 | * Doh this slab cannot be placed using slub_max_order. | |
2875 | */ | |
ab9a0f19 | 2876 | order = slab_order(size, 1, MAX_ORDER, 1, reserved); |
818cf590 | 2877 | if (order < MAX_ORDER) |
5e6d444e CL |
2878 | return order; |
2879 | return -ENOSYS; | |
2880 | } | |
2881 | ||
5595cffc | 2882 | static void |
4053497d | 2883 | init_kmem_cache_node(struct kmem_cache_node *n) |
81819f0f CL |
2884 | { |
2885 | n->nr_partial = 0; | |
81819f0f CL |
2886 | spin_lock_init(&n->list_lock); |
2887 | INIT_LIST_HEAD(&n->partial); | |
8ab1372f | 2888 | #ifdef CONFIG_SLUB_DEBUG |
0f389ec6 | 2889 | atomic_long_set(&n->nr_slabs, 0); |
02b71b70 | 2890 | atomic_long_set(&n->total_objects, 0); |
643b1138 | 2891 | INIT_LIST_HEAD(&n->full); |
8ab1372f | 2892 | #endif |
81819f0f CL |
2893 | } |
2894 | ||
55136592 | 2895 | static inline int alloc_kmem_cache_cpus(struct kmem_cache *s) |
4c93c355 | 2896 | { |
6c182dc0 | 2897 | BUILD_BUG_ON(PERCPU_DYNAMIC_EARLY_SIZE < |
95a05b42 | 2898 | KMALLOC_SHIFT_HIGH * sizeof(struct kmem_cache_cpu)); |
4c93c355 | 2899 | |
8a5ec0ba | 2900 | /* |
d4d84fef CM |
2901 | * Must align to double word boundary for the double cmpxchg |
2902 | * instructions to work; see __pcpu_double_call_return_bool(). | |
8a5ec0ba | 2903 | */ |
d4d84fef CM |
2904 | s->cpu_slab = __alloc_percpu(sizeof(struct kmem_cache_cpu), |
2905 | 2 * sizeof(void *)); | |
8a5ec0ba CL |
2906 | |
2907 | if (!s->cpu_slab) | |
2908 | return 0; | |
2909 | ||
2910 | init_kmem_cache_cpus(s); | |
4c93c355 | 2911 | |
8a5ec0ba | 2912 | return 1; |
4c93c355 | 2913 | } |
4c93c355 | 2914 | |
51df1142 CL |
2915 | static struct kmem_cache *kmem_cache_node; |
2916 | ||
81819f0f CL |
2917 | /* |
2918 | * No kmalloc_node yet so do it by hand. We know that this is the first | |
2919 | * slab on the node for this slabcache. There are no concurrent accesses | |
2920 | * possible. | |
2921 | * | |
721ae22a ZYW |
2922 | * Note that this function only works on the kmem_cache_node |
2923 | * when allocating for the kmem_cache_node. This is used for bootstrapping | |
4c93c355 | 2924 | * memory on a fresh node that has no slab structures yet. |
81819f0f | 2925 | */ |
55136592 | 2926 | static void early_kmem_cache_node_alloc(int node) |
81819f0f CL |
2927 | { |
2928 | struct page *page; | |
2929 | struct kmem_cache_node *n; | |
2930 | ||
51df1142 | 2931 | BUG_ON(kmem_cache_node->size < sizeof(struct kmem_cache_node)); |
81819f0f | 2932 | |
51df1142 | 2933 | page = new_slab(kmem_cache_node, GFP_NOWAIT, node); |
81819f0f CL |
2934 | |
2935 | BUG_ON(!page); | |
a2f92ee7 | 2936 | if (page_to_nid(page) != node) { |
f9f58285 FF |
2937 | pr_err("SLUB: Unable to allocate memory from node %d\n", node); |
2938 | pr_err("SLUB: Allocating a useless per node structure in order to be able to continue\n"); | |
a2f92ee7 CL |
2939 | } |
2940 | ||
81819f0f CL |
2941 | n = page->freelist; |
2942 | BUG_ON(!n); | |
51df1142 | 2943 | page->freelist = get_freepointer(kmem_cache_node, n); |
e6e82ea1 | 2944 | page->inuse = 1; |
8cb0a506 | 2945 | page->frozen = 0; |
51df1142 | 2946 | kmem_cache_node->node[node] = n; |
8ab1372f | 2947 | #ifdef CONFIG_SLUB_DEBUG |
f7cb1933 | 2948 | init_object(kmem_cache_node, n, SLUB_RED_ACTIVE); |
51df1142 | 2949 | init_tracking(kmem_cache_node, n); |
8ab1372f | 2950 | #endif |
0316bec2 | 2951 | kasan_kmalloc(kmem_cache_node, n, sizeof(struct kmem_cache_node)); |
4053497d | 2952 | init_kmem_cache_node(n); |
51df1142 | 2953 | inc_slabs_node(kmem_cache_node, node, page->objects); |
6446faa2 | 2954 | |
67b6c900 | 2955 | /* |
1e4dd946 SR |
2956 | * No locks need to be taken here as it has just been |
2957 | * initialized and there is no concurrent access. | |
67b6c900 | 2958 | */ |
1e4dd946 | 2959 | __add_partial(n, page, DEACTIVATE_TO_HEAD); |
81819f0f CL |
2960 | } |
2961 | ||
2962 | static void free_kmem_cache_nodes(struct kmem_cache *s) | |
2963 | { | |
2964 | int node; | |
fa45dc25 | 2965 | struct kmem_cache_node *n; |
81819f0f | 2966 | |
fa45dc25 CL |
2967 | for_each_kmem_cache_node(s, node, n) { |
2968 | kmem_cache_free(kmem_cache_node, n); | |
81819f0f CL |
2969 | s->node[node] = NULL; |
2970 | } | |
2971 | } | |
2972 | ||
55136592 | 2973 | static int init_kmem_cache_nodes(struct kmem_cache *s) |
81819f0f CL |
2974 | { |
2975 | int node; | |
81819f0f | 2976 | |
f64dc58c | 2977 | for_each_node_state(node, N_NORMAL_MEMORY) { |
81819f0f CL |
2978 | struct kmem_cache_node *n; |
2979 | ||
73367bd8 | 2980 | if (slab_state == DOWN) { |
55136592 | 2981 | early_kmem_cache_node_alloc(node); |
73367bd8 AD |
2982 | continue; |
2983 | } | |
51df1142 | 2984 | n = kmem_cache_alloc_node(kmem_cache_node, |
55136592 | 2985 | GFP_KERNEL, node); |
81819f0f | 2986 | |
73367bd8 AD |
2987 | if (!n) { |
2988 | free_kmem_cache_nodes(s); | |
2989 | return 0; | |
81819f0f | 2990 | } |
73367bd8 | 2991 | |
81819f0f | 2992 | s->node[node] = n; |
4053497d | 2993 | init_kmem_cache_node(n); |
81819f0f CL |
2994 | } |
2995 | return 1; | |
2996 | } | |
81819f0f | 2997 | |
c0bdb232 | 2998 | static void set_min_partial(struct kmem_cache *s, unsigned long min) |
3b89d7d8 DR |
2999 | { |
3000 | if (min < MIN_PARTIAL) | |
3001 | min = MIN_PARTIAL; | |
3002 | else if (min > MAX_PARTIAL) | |
3003 | min = MAX_PARTIAL; | |
3004 | s->min_partial = min; | |
3005 | } | |
3006 | ||
81819f0f CL |
3007 | /* |
3008 | * calculate_sizes() determines the order and the distribution of data within | |
3009 | * a slab object. | |
3010 | */ | |
06b285dc | 3011 | static int calculate_sizes(struct kmem_cache *s, int forced_order) |
81819f0f CL |
3012 | { |
3013 | unsigned long flags = s->flags; | |
3b0efdfa | 3014 | unsigned long size = s->object_size; |
834f3d11 | 3015 | int order; |
81819f0f | 3016 | |
d8b42bf5 CL |
3017 | /* |
3018 | * Round up object size to the next word boundary. We can only | |
3019 | * place the free pointer at word boundaries and this determines | |
3020 | * the possible location of the free pointer. | |
3021 | */ | |
3022 | size = ALIGN(size, sizeof(void *)); | |
3023 | ||
3024 | #ifdef CONFIG_SLUB_DEBUG | |
81819f0f CL |
3025 | /* |
3026 | * Determine if we can poison the object itself. If the user of | |
3027 | * the slab may touch the object after free or before allocation | |
3028 | * then we should never poison the object itself. | |
3029 | */ | |
3030 | if ((flags & SLAB_POISON) && !(flags & SLAB_DESTROY_BY_RCU) && | |
c59def9f | 3031 | !s->ctor) |
81819f0f CL |
3032 | s->flags |= __OBJECT_POISON; |
3033 | else | |
3034 | s->flags &= ~__OBJECT_POISON; | |
3035 | ||
81819f0f CL |
3036 | |
3037 | /* | |
672bba3a | 3038 | * If we are Redzoning then check if there is some space between the |
81819f0f | 3039 | * end of the object and the free pointer. If not then add an |
672bba3a | 3040 | * additional word to have some bytes to store Redzone information. |
81819f0f | 3041 | */ |
3b0efdfa | 3042 | if ((flags & SLAB_RED_ZONE) && size == s->object_size) |
81819f0f | 3043 | size += sizeof(void *); |
41ecc55b | 3044 | #endif |
81819f0f CL |
3045 | |
3046 | /* | |
672bba3a CL |
3047 | * With that we have determined the number of bytes in actual use |
3048 | * by the object. This is the potential offset to the free pointer. | |
81819f0f CL |
3049 | */ |
3050 | s->inuse = size; | |
3051 | ||
3052 | if (((flags & (SLAB_DESTROY_BY_RCU | SLAB_POISON)) || | |
c59def9f | 3053 | s->ctor)) { |
81819f0f CL |
3054 | /* |
3055 | * Relocate free pointer after the object if it is not | |
3056 | * permitted to overwrite the first word of the object on | |
3057 | * kmem_cache_free. | |
3058 | * | |
3059 | * This is the case if we do RCU, have a constructor or | |
3060 | * destructor or are poisoning the objects. | |
3061 | */ | |
3062 | s->offset = size; | |
3063 | size += sizeof(void *); | |
3064 | } | |
3065 | ||
c12b3c62 | 3066 | #ifdef CONFIG_SLUB_DEBUG |
81819f0f CL |
3067 | if (flags & SLAB_STORE_USER) |
3068 | /* | |
3069 | * Need to store information about allocs and frees after | |
3070 | * the object. | |
3071 | */ | |
3072 | size += 2 * sizeof(struct track); | |
3073 | ||
be7b3fbc | 3074 | if (flags & SLAB_RED_ZONE) |
81819f0f CL |
3075 | /* |
3076 | * Add some empty padding so that we can catch | |
3077 | * overwrites from earlier objects rather than let | |
3078 | * tracking information or the free pointer be | |
0211a9c8 | 3079 | * corrupted if a user writes before the start |
81819f0f CL |
3080 | * of the object. |
3081 | */ | |
3082 | size += sizeof(void *); | |
41ecc55b | 3083 | #endif |
672bba3a | 3084 | |
81819f0f CL |
3085 | /* |
3086 | * SLUB stores one object immediately after another beginning from | |
3087 | * offset 0. In order to align the objects we have to simply size | |
3088 | * each object to conform to the alignment. | |
3089 | */ | |
45906855 | 3090 | size = ALIGN(size, s->align); |
81819f0f | 3091 | s->size = size; |
06b285dc CL |
3092 | if (forced_order >= 0) |
3093 | order = forced_order; | |
3094 | else | |
ab9a0f19 | 3095 | order = calculate_order(size, s->reserved); |
81819f0f | 3096 | |
834f3d11 | 3097 | if (order < 0) |
81819f0f CL |
3098 | return 0; |
3099 | ||
b7a49f0d | 3100 | s->allocflags = 0; |
834f3d11 | 3101 | if (order) |
b7a49f0d CL |
3102 | s->allocflags |= __GFP_COMP; |
3103 | ||
3104 | if (s->flags & SLAB_CACHE_DMA) | |
2c59dd65 | 3105 | s->allocflags |= GFP_DMA; |
b7a49f0d CL |
3106 | |
3107 | if (s->flags & SLAB_RECLAIM_ACCOUNT) | |
3108 | s->allocflags |= __GFP_RECLAIMABLE; | |
3109 | ||
81819f0f CL |
3110 | /* |
3111 | * Determine the number of objects per slab | |
3112 | */ | |
ab9a0f19 LJ |
3113 | s->oo = oo_make(order, size, s->reserved); |
3114 | s->min = oo_make(get_order(size), size, s->reserved); | |
205ab99d CL |
3115 | if (oo_objects(s->oo) > oo_objects(s->max)) |
3116 | s->max = s->oo; | |
81819f0f | 3117 | |
834f3d11 | 3118 | return !!oo_objects(s->oo); |
81819f0f CL |
3119 | } |
3120 | ||
8a13a4cc | 3121 | static int kmem_cache_open(struct kmem_cache *s, unsigned long flags) |
81819f0f | 3122 | { |
8a13a4cc | 3123 | s->flags = kmem_cache_flags(s->size, flags, s->name, s->ctor); |
ab9a0f19 | 3124 | s->reserved = 0; |
81819f0f | 3125 | |
da9a638c LJ |
3126 | if (need_reserve_slab_rcu && (s->flags & SLAB_DESTROY_BY_RCU)) |
3127 | s->reserved = sizeof(struct rcu_head); | |
81819f0f | 3128 | |
06b285dc | 3129 | if (!calculate_sizes(s, -1)) |
81819f0f | 3130 | goto error; |
3de47213 DR |
3131 | if (disable_higher_order_debug) { |
3132 | /* | |
3133 | * Disable debugging flags that store metadata if the min slab | |
3134 | * order increased. | |
3135 | */ | |
3b0efdfa | 3136 | if (get_order(s->size) > get_order(s->object_size)) { |
3de47213 DR |
3137 | s->flags &= ~DEBUG_METADATA_FLAGS; |
3138 | s->offset = 0; | |
3139 | if (!calculate_sizes(s, -1)) | |
3140 | goto error; | |
3141 | } | |
3142 | } | |
81819f0f | 3143 | |
2565409f HC |
3144 | #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \ |
3145 | defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE) | |
b789ef51 CL |
3146 | if (system_has_cmpxchg_double() && (s->flags & SLAB_DEBUG_FLAGS) == 0) |
3147 | /* Enable fast mode */ | |
3148 | s->flags |= __CMPXCHG_DOUBLE; | |
3149 | #endif | |
3150 | ||
3b89d7d8 DR |
3151 | /* |
3152 | * The larger the object size is, the more pages we want on the partial | |
3153 | * list to avoid pounding the page allocator excessively. | |
3154 | */ | |
49e22585 CL |
3155 | set_min_partial(s, ilog2(s->size) / 2); |
3156 | ||
3157 | /* | |
3158 | * cpu_partial determined the maximum number of objects kept in the | |
3159 | * per cpu partial lists of a processor. | |
3160 | * | |
3161 | * Per cpu partial lists mainly contain slabs that just have one | |
3162 | * object freed. If they are used for allocation then they can be | |
3163 | * filled up again with minimal effort. The slab will never hit the | |
3164 | * per node partial lists and therefore no locking will be required. | |
3165 | * | |
3166 | * This setting also determines | |
3167 | * | |
3168 | * A) The number of objects from per cpu partial slabs dumped to the | |
3169 | * per node list when we reach the limit. | |
9f264904 | 3170 | * B) The number of objects in cpu partial slabs to extract from the |
d0e0ac97 CG |
3171 | * per node list when we run out of per cpu objects. We only fetch |
3172 | * 50% to keep some capacity around for frees. | |
49e22585 | 3173 | */ |
345c905d | 3174 | if (!kmem_cache_has_cpu_partial(s)) |
8f1e33da CL |
3175 | s->cpu_partial = 0; |
3176 | else if (s->size >= PAGE_SIZE) | |
49e22585 CL |
3177 | s->cpu_partial = 2; |
3178 | else if (s->size >= 1024) | |
3179 | s->cpu_partial = 6; | |
3180 | else if (s->size >= 256) | |
3181 | s->cpu_partial = 13; | |
3182 | else | |
3183 | s->cpu_partial = 30; | |
3184 | ||
81819f0f | 3185 | #ifdef CONFIG_NUMA |
e2cb96b7 | 3186 | s->remote_node_defrag_ratio = 1000; |
81819f0f | 3187 | #endif |
55136592 | 3188 | if (!init_kmem_cache_nodes(s)) |
dfb4f096 | 3189 | goto error; |
81819f0f | 3190 | |
55136592 | 3191 | if (alloc_kmem_cache_cpus(s)) |
278b1bb1 | 3192 | return 0; |
ff12059e | 3193 | |
4c93c355 | 3194 | free_kmem_cache_nodes(s); |
81819f0f CL |
3195 | error: |
3196 | if (flags & SLAB_PANIC) | |
3197 | panic("Cannot create slab %s size=%lu realsize=%u " | |
3198 | "order=%u offset=%u flags=%lx\n", | |
d0e0ac97 CG |
3199 | s->name, (unsigned long)s->size, s->size, |
3200 | oo_order(s->oo), s->offset, flags); | |
278b1bb1 | 3201 | return -EINVAL; |
81819f0f | 3202 | } |
81819f0f | 3203 | |
33b12c38 CL |
3204 | static void list_slab_objects(struct kmem_cache *s, struct page *page, |
3205 | const char *text) | |
3206 | { | |
3207 | #ifdef CONFIG_SLUB_DEBUG | |
3208 | void *addr = page_address(page); | |
3209 | void *p; | |
a5dd5c11 NK |
3210 | unsigned long *map = kzalloc(BITS_TO_LONGS(page->objects) * |
3211 | sizeof(long), GFP_ATOMIC); | |
bbd7d57b ED |
3212 | if (!map) |
3213 | return; | |
945cf2b6 | 3214 | slab_err(s, page, text, s->name); |
33b12c38 | 3215 | slab_lock(page); |
33b12c38 | 3216 | |
5f80b13a | 3217 | get_map(s, page, map); |
33b12c38 CL |
3218 | for_each_object(p, s, addr, page->objects) { |
3219 | ||
3220 | if (!test_bit(slab_index(p, s, addr), map)) { | |
f9f58285 | 3221 | pr_err("INFO: Object 0x%p @offset=%tu\n", p, p - addr); |
33b12c38 CL |
3222 | print_tracking(s, p); |
3223 | } | |
3224 | } | |
3225 | slab_unlock(page); | |
bbd7d57b | 3226 | kfree(map); |
33b12c38 CL |
3227 | #endif |
3228 | } | |
3229 | ||
81819f0f | 3230 | /* |
599870b1 | 3231 | * Attempt to free all partial slabs on a node. |
69cb8e6b CL |
3232 | * This is called from kmem_cache_close(). We must be the last thread |
3233 | * using the cache and therefore we do not need to lock anymore. | |
81819f0f | 3234 | */ |
599870b1 | 3235 | static void free_partial(struct kmem_cache *s, struct kmem_cache_node *n) |
81819f0f | 3236 | { |
81819f0f CL |
3237 | struct page *page, *h; |
3238 | ||
33b12c38 | 3239 | list_for_each_entry_safe(page, h, &n->partial, lru) { |
81819f0f | 3240 | if (!page->inuse) { |
1e4dd946 | 3241 | __remove_partial(n, page); |
81819f0f | 3242 | discard_slab(s, page); |
33b12c38 CL |
3243 | } else { |
3244 | list_slab_objects(s, page, | |
945cf2b6 | 3245 | "Objects remaining in %s on kmem_cache_close()"); |
599870b1 | 3246 | } |
33b12c38 | 3247 | } |
81819f0f CL |
3248 | } |
3249 | ||
3250 | /* | |
672bba3a | 3251 | * Release all resources used by a slab cache. |
81819f0f | 3252 | */ |
0c710013 | 3253 | static inline int kmem_cache_close(struct kmem_cache *s) |
81819f0f CL |
3254 | { |
3255 | int node; | |
fa45dc25 | 3256 | struct kmem_cache_node *n; |
81819f0f CL |
3257 | |
3258 | flush_all(s); | |
81819f0f | 3259 | /* Attempt to free all objects */ |
fa45dc25 | 3260 | for_each_kmem_cache_node(s, node, n) { |
599870b1 CL |
3261 | free_partial(s, n); |
3262 | if (n->nr_partial || slabs_node(s, node)) | |
81819f0f CL |
3263 | return 1; |
3264 | } | |
945cf2b6 | 3265 | free_percpu(s->cpu_slab); |
81819f0f CL |
3266 | free_kmem_cache_nodes(s); |
3267 | return 0; | |
3268 | } | |
3269 | ||
945cf2b6 | 3270 | int __kmem_cache_shutdown(struct kmem_cache *s) |
81819f0f | 3271 | { |
41a21285 | 3272 | return kmem_cache_close(s); |
81819f0f | 3273 | } |
81819f0f CL |
3274 | |
3275 | /******************************************************************** | |
3276 | * Kmalloc subsystem | |
3277 | *******************************************************************/ | |
3278 | ||
81819f0f CL |
3279 | static int __init setup_slub_min_order(char *str) |
3280 | { | |
06428780 | 3281 | get_option(&str, &slub_min_order); |
81819f0f CL |
3282 | |
3283 | return 1; | |
3284 | } | |
3285 | ||
3286 | __setup("slub_min_order=", setup_slub_min_order); | |
3287 | ||
3288 | static int __init setup_slub_max_order(char *str) | |
3289 | { | |
06428780 | 3290 | get_option(&str, &slub_max_order); |
818cf590 | 3291 | slub_max_order = min(slub_max_order, MAX_ORDER - 1); |
81819f0f CL |
3292 | |
3293 | return 1; | |
3294 | } | |
3295 | ||
3296 | __setup("slub_max_order=", setup_slub_max_order); | |
3297 | ||
3298 | static int __init setup_slub_min_objects(char *str) | |
3299 | { | |
06428780 | 3300 | get_option(&str, &slub_min_objects); |
81819f0f CL |
3301 | |
3302 | return 1; | |
3303 | } | |
3304 | ||
3305 | __setup("slub_min_objects=", setup_slub_min_objects); | |
3306 | ||
81819f0f CL |
3307 | void *__kmalloc(size_t size, gfp_t flags) |
3308 | { | |
aadb4bc4 | 3309 | struct kmem_cache *s; |
5b882be4 | 3310 | void *ret; |
81819f0f | 3311 | |
95a05b42 | 3312 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) |
eada35ef | 3313 | return kmalloc_large(size, flags); |
aadb4bc4 | 3314 | |
2c59dd65 | 3315 | s = kmalloc_slab(size, flags); |
aadb4bc4 CL |
3316 | |
3317 | if (unlikely(ZERO_OR_NULL_PTR(s))) | |
6cb8f913 CL |
3318 | return s; |
3319 | ||
2b847c3c | 3320 | ret = slab_alloc(s, flags, _RET_IP_); |
5b882be4 | 3321 | |
ca2b84cb | 3322 | trace_kmalloc(_RET_IP_, ret, size, s->size, flags); |
5b882be4 | 3323 | |
0316bec2 AR |
3324 | kasan_kmalloc(s, ret, size); |
3325 | ||
5b882be4 | 3326 | return ret; |
81819f0f CL |
3327 | } |
3328 | EXPORT_SYMBOL(__kmalloc); | |
3329 | ||
5d1f57e4 | 3330 | #ifdef CONFIG_NUMA |
f619cfe1 CL |
3331 | static void *kmalloc_large_node(size_t size, gfp_t flags, int node) |
3332 | { | |
b1eeab67 | 3333 | struct page *page; |
e4f7c0b4 | 3334 | void *ptr = NULL; |
f619cfe1 | 3335 | |
52383431 VD |
3336 | flags |= __GFP_COMP | __GFP_NOTRACK; |
3337 | page = alloc_kmem_pages_node(node, flags, get_order(size)); | |
f619cfe1 | 3338 | if (page) |
e4f7c0b4 CM |
3339 | ptr = page_address(page); |
3340 | ||
d56791b3 | 3341 | kmalloc_large_node_hook(ptr, size, flags); |
e4f7c0b4 | 3342 | return ptr; |
f619cfe1 CL |
3343 | } |
3344 | ||
81819f0f CL |
3345 | void *__kmalloc_node(size_t size, gfp_t flags, int node) |
3346 | { | |
aadb4bc4 | 3347 | struct kmem_cache *s; |
5b882be4 | 3348 | void *ret; |
81819f0f | 3349 | |
95a05b42 | 3350 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) { |
5b882be4 EGM |
3351 | ret = kmalloc_large_node(size, flags, node); |
3352 | ||
ca2b84cb EGM |
3353 | trace_kmalloc_node(_RET_IP_, ret, |
3354 | size, PAGE_SIZE << get_order(size), | |
3355 | flags, node); | |
5b882be4 EGM |
3356 | |
3357 | return ret; | |
3358 | } | |
aadb4bc4 | 3359 | |
2c59dd65 | 3360 | s = kmalloc_slab(size, flags); |
aadb4bc4 CL |
3361 | |
3362 | if (unlikely(ZERO_OR_NULL_PTR(s))) | |
6cb8f913 CL |
3363 | return s; |
3364 | ||
2b847c3c | 3365 | ret = slab_alloc_node(s, flags, node, _RET_IP_); |
5b882be4 | 3366 | |
ca2b84cb | 3367 | trace_kmalloc_node(_RET_IP_, ret, size, s->size, flags, node); |
5b882be4 | 3368 | |
0316bec2 AR |
3369 | kasan_kmalloc(s, ret, size); |
3370 | ||
5b882be4 | 3371 | return ret; |
81819f0f CL |
3372 | } |
3373 | EXPORT_SYMBOL(__kmalloc_node); | |
3374 | #endif | |
3375 | ||
0316bec2 | 3376 | static size_t __ksize(const void *object) |
81819f0f | 3377 | { |
272c1d21 | 3378 | struct page *page; |
81819f0f | 3379 | |
ef8b4520 | 3380 | if (unlikely(object == ZERO_SIZE_PTR)) |
272c1d21 CL |
3381 | return 0; |
3382 | ||
294a80a8 | 3383 | page = virt_to_head_page(object); |
294a80a8 | 3384 | |
76994412 PE |
3385 | if (unlikely(!PageSlab(page))) { |
3386 | WARN_ON(!PageCompound(page)); | |
294a80a8 | 3387 | return PAGE_SIZE << compound_order(page); |
76994412 | 3388 | } |
81819f0f | 3389 | |
1b4f59e3 | 3390 | return slab_ksize(page->slab_cache); |
81819f0f | 3391 | } |
0316bec2 AR |
3392 | |
3393 | size_t ksize(const void *object) | |
3394 | { | |
3395 | size_t size = __ksize(object); | |
3396 | /* We assume that ksize callers could use whole allocated area, | |
3397 | so we need unpoison this area. */ | |
3398 | kasan_krealloc(object, size); | |
3399 | return size; | |
3400 | } | |
b1aabecd | 3401 | EXPORT_SYMBOL(ksize); |
81819f0f CL |
3402 | |
3403 | void kfree(const void *x) | |
3404 | { | |
81819f0f | 3405 | struct page *page; |
5bb983b0 | 3406 | void *object = (void *)x; |
81819f0f | 3407 | |
2121db74 PE |
3408 | trace_kfree(_RET_IP_, x); |
3409 | ||
2408c550 | 3410 | if (unlikely(ZERO_OR_NULL_PTR(x))) |
81819f0f CL |
3411 | return; |
3412 | ||
b49af68f | 3413 | page = virt_to_head_page(x); |
aadb4bc4 | 3414 | if (unlikely(!PageSlab(page))) { |
0937502a | 3415 | BUG_ON(!PageCompound(page)); |
d56791b3 | 3416 | kfree_hook(x); |
52383431 | 3417 | __free_kmem_pages(page, compound_order(page)); |
aadb4bc4 CL |
3418 | return; |
3419 | } | |
1b4f59e3 | 3420 | slab_free(page->slab_cache, page, object, _RET_IP_); |
81819f0f CL |
3421 | } |
3422 | EXPORT_SYMBOL(kfree); | |
3423 | ||
832f37f5 VD |
3424 | #define SHRINK_PROMOTE_MAX 32 |
3425 | ||
2086d26a | 3426 | /* |
832f37f5 VD |
3427 | * kmem_cache_shrink discards empty slabs and promotes the slabs filled |
3428 | * up most to the head of the partial lists. New allocations will then | |
3429 | * fill those up and thus they can be removed from the partial lists. | |
672bba3a CL |
3430 | * |
3431 | * The slabs with the least items are placed last. This results in them | |
3432 | * being allocated from last increasing the chance that the last objects | |
3433 | * are freed in them. | |
2086d26a | 3434 | */ |
d6e0b7fa | 3435 | int __kmem_cache_shrink(struct kmem_cache *s, bool deactivate) |
2086d26a CL |
3436 | { |
3437 | int node; | |
3438 | int i; | |
3439 | struct kmem_cache_node *n; | |
3440 | struct page *page; | |
3441 | struct page *t; | |
832f37f5 VD |
3442 | struct list_head discard; |
3443 | struct list_head promote[SHRINK_PROMOTE_MAX]; | |
2086d26a | 3444 | unsigned long flags; |
ce3712d7 | 3445 | int ret = 0; |
2086d26a | 3446 | |
d6e0b7fa VD |
3447 | if (deactivate) { |
3448 | /* | |
3449 | * Disable empty slabs caching. Used to avoid pinning offline | |
3450 | * memory cgroups by kmem pages that can be freed. | |
3451 | */ | |
3452 | s->cpu_partial = 0; | |
3453 | s->min_partial = 0; | |
3454 | ||
3455 | /* | |
3456 | * s->cpu_partial is checked locklessly (see put_cpu_partial), | |
3457 | * so we have to make sure the change is visible. | |
3458 | */ | |
3459 | kick_all_cpus_sync(); | |
3460 | } | |
3461 | ||
2086d26a | 3462 | flush_all(s); |
fa45dc25 | 3463 | for_each_kmem_cache_node(s, node, n) { |
832f37f5 VD |
3464 | INIT_LIST_HEAD(&discard); |
3465 | for (i = 0; i < SHRINK_PROMOTE_MAX; i++) | |
3466 | INIT_LIST_HEAD(promote + i); | |
2086d26a CL |
3467 | |
3468 | spin_lock_irqsave(&n->list_lock, flags); | |
3469 | ||
3470 | /* | |
832f37f5 | 3471 | * Build lists of slabs to discard or promote. |
2086d26a | 3472 | * |
672bba3a CL |
3473 | * Note that concurrent frees may occur while we hold the |
3474 | * list_lock. page->inuse here is the upper limit. | |
2086d26a CL |
3475 | */ |
3476 | list_for_each_entry_safe(page, t, &n->partial, lru) { | |
832f37f5 VD |
3477 | int free = page->objects - page->inuse; |
3478 | ||
3479 | /* Do not reread page->inuse */ | |
3480 | barrier(); | |
3481 | ||
3482 | /* We do not keep full slabs on the list */ | |
3483 | BUG_ON(free <= 0); | |
3484 | ||
3485 | if (free == page->objects) { | |
3486 | list_move(&page->lru, &discard); | |
69cb8e6b | 3487 | n->nr_partial--; |
832f37f5 VD |
3488 | } else if (free <= SHRINK_PROMOTE_MAX) |
3489 | list_move(&page->lru, promote + free - 1); | |
2086d26a CL |
3490 | } |
3491 | ||
2086d26a | 3492 | /* |
832f37f5 VD |
3493 | * Promote the slabs filled up most to the head of the |
3494 | * partial list. | |
2086d26a | 3495 | */ |
832f37f5 VD |
3496 | for (i = SHRINK_PROMOTE_MAX - 1; i >= 0; i--) |
3497 | list_splice(promote + i, &n->partial); | |
2086d26a | 3498 | |
2086d26a | 3499 | spin_unlock_irqrestore(&n->list_lock, flags); |
69cb8e6b CL |
3500 | |
3501 | /* Release empty slabs */ | |
832f37f5 | 3502 | list_for_each_entry_safe(page, t, &discard, lru) |
69cb8e6b | 3503 | discard_slab(s, page); |
ce3712d7 VD |
3504 | |
3505 | if (slabs_node(s, node)) | |
3506 | ret = 1; | |
2086d26a CL |
3507 | } |
3508 | ||
ce3712d7 | 3509 | return ret; |
2086d26a | 3510 | } |
2086d26a | 3511 | |
b9049e23 YG |
3512 | static int slab_mem_going_offline_callback(void *arg) |
3513 | { | |
3514 | struct kmem_cache *s; | |
3515 | ||
18004c5d | 3516 | mutex_lock(&slab_mutex); |
b9049e23 | 3517 | list_for_each_entry(s, &slab_caches, list) |
d6e0b7fa | 3518 | __kmem_cache_shrink(s, false); |
18004c5d | 3519 | mutex_unlock(&slab_mutex); |
b9049e23 YG |
3520 | |
3521 | return 0; | |
3522 | } | |
3523 | ||
3524 | static void slab_mem_offline_callback(void *arg) | |
3525 | { | |
3526 | struct kmem_cache_node *n; | |
3527 | struct kmem_cache *s; | |
3528 | struct memory_notify *marg = arg; | |
3529 | int offline_node; | |
3530 | ||
b9d5ab25 | 3531 | offline_node = marg->status_change_nid_normal; |
b9049e23 YG |
3532 | |
3533 | /* | |
3534 | * If the node still has available memory. we need kmem_cache_node | |
3535 | * for it yet. | |
3536 | */ | |
3537 | if (offline_node < 0) | |
3538 | return; | |
3539 | ||
18004c5d | 3540 | mutex_lock(&slab_mutex); |
b9049e23 YG |
3541 | list_for_each_entry(s, &slab_caches, list) { |
3542 | n = get_node(s, offline_node); | |
3543 | if (n) { | |
3544 | /* | |
3545 | * if n->nr_slabs > 0, slabs still exist on the node | |
3546 | * that is going down. We were unable to free them, | |
c9404c9c | 3547 | * and offline_pages() function shouldn't call this |
b9049e23 YG |
3548 | * callback. So, we must fail. |
3549 | */ | |
0f389ec6 | 3550 | BUG_ON(slabs_node(s, offline_node)); |
b9049e23 YG |
3551 | |
3552 | s->node[offline_node] = NULL; | |
8de66a0c | 3553 | kmem_cache_free(kmem_cache_node, n); |
b9049e23 YG |
3554 | } |
3555 | } | |
18004c5d | 3556 | mutex_unlock(&slab_mutex); |
b9049e23 YG |
3557 | } |
3558 | ||
3559 | static int slab_mem_going_online_callback(void *arg) | |
3560 | { | |
3561 | struct kmem_cache_node *n; | |
3562 | struct kmem_cache *s; | |
3563 | struct memory_notify *marg = arg; | |
b9d5ab25 | 3564 | int nid = marg->status_change_nid_normal; |
b9049e23 YG |
3565 | int ret = 0; |
3566 | ||
3567 | /* | |
3568 | * If the node's memory is already available, then kmem_cache_node is | |
3569 | * already created. Nothing to do. | |
3570 | */ | |
3571 | if (nid < 0) | |
3572 | return 0; | |
3573 | ||
3574 | /* | |
0121c619 | 3575 | * We are bringing a node online. No memory is available yet. We must |
b9049e23 YG |
3576 | * allocate a kmem_cache_node structure in order to bring the node |
3577 | * online. | |
3578 | */ | |
18004c5d | 3579 | mutex_lock(&slab_mutex); |
b9049e23 YG |
3580 | list_for_each_entry(s, &slab_caches, list) { |
3581 | /* | |
3582 | * XXX: kmem_cache_alloc_node will fallback to other nodes | |
3583 | * since memory is not yet available from the node that | |
3584 | * is brought up. | |
3585 | */ | |
8de66a0c | 3586 | n = kmem_cache_alloc(kmem_cache_node, GFP_KERNEL); |
b9049e23 YG |
3587 | if (!n) { |
3588 | ret = -ENOMEM; | |
3589 | goto out; | |
3590 | } | |
4053497d | 3591 | init_kmem_cache_node(n); |
b9049e23 YG |
3592 | s->node[nid] = n; |
3593 | } | |
3594 | out: | |
18004c5d | 3595 | mutex_unlock(&slab_mutex); |
b9049e23 YG |
3596 | return ret; |
3597 | } | |
3598 | ||
3599 | static int slab_memory_callback(struct notifier_block *self, | |
3600 | unsigned long action, void *arg) | |
3601 | { | |
3602 | int ret = 0; | |
3603 | ||
3604 | switch (action) { | |
3605 | case MEM_GOING_ONLINE: | |
3606 | ret = slab_mem_going_online_callback(arg); | |
3607 | break; | |
3608 | case MEM_GOING_OFFLINE: | |
3609 | ret = slab_mem_going_offline_callback(arg); | |
3610 | break; | |
3611 | case MEM_OFFLINE: | |
3612 | case MEM_CANCEL_ONLINE: | |
3613 | slab_mem_offline_callback(arg); | |
3614 | break; | |
3615 | case MEM_ONLINE: | |
3616 | case MEM_CANCEL_OFFLINE: | |
3617 | break; | |
3618 | } | |
dc19f9db KH |
3619 | if (ret) |
3620 | ret = notifier_from_errno(ret); | |
3621 | else | |
3622 | ret = NOTIFY_OK; | |
b9049e23 YG |
3623 | return ret; |
3624 | } | |
3625 | ||
3ac38faa AM |
3626 | static struct notifier_block slab_memory_callback_nb = { |
3627 | .notifier_call = slab_memory_callback, | |
3628 | .priority = SLAB_CALLBACK_PRI, | |
3629 | }; | |
b9049e23 | 3630 | |
81819f0f CL |
3631 | /******************************************************************** |
3632 | * Basic setup of slabs | |
3633 | *******************************************************************/ | |
3634 | ||
51df1142 CL |
3635 | /* |
3636 | * Used for early kmem_cache structures that were allocated using | |
dffb4d60 CL |
3637 | * the page allocator. Allocate them properly then fix up the pointers |
3638 | * that may be pointing to the wrong kmem_cache structure. | |
51df1142 CL |
3639 | */ |
3640 | ||
dffb4d60 | 3641 | static struct kmem_cache * __init bootstrap(struct kmem_cache *static_cache) |
51df1142 CL |
3642 | { |
3643 | int node; | |
dffb4d60 | 3644 | struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT); |
fa45dc25 | 3645 | struct kmem_cache_node *n; |
51df1142 | 3646 | |
dffb4d60 | 3647 | memcpy(s, static_cache, kmem_cache->object_size); |
51df1142 | 3648 | |
7d557b3c GC |
3649 | /* |
3650 | * This runs very early, and only the boot processor is supposed to be | |
3651 | * up. Even if it weren't true, IRQs are not up so we couldn't fire | |
3652 | * IPIs around. | |
3653 | */ | |
3654 | __flush_cpu_slab(s, smp_processor_id()); | |
fa45dc25 | 3655 | for_each_kmem_cache_node(s, node, n) { |
51df1142 CL |
3656 | struct page *p; |
3657 | ||
fa45dc25 CL |
3658 | list_for_each_entry(p, &n->partial, lru) |
3659 | p->slab_cache = s; | |
51df1142 | 3660 | |
607bf324 | 3661 | #ifdef CONFIG_SLUB_DEBUG |
fa45dc25 CL |
3662 | list_for_each_entry(p, &n->full, lru) |
3663 | p->slab_cache = s; | |
51df1142 | 3664 | #endif |
51df1142 | 3665 | } |
f7ce3190 | 3666 | slab_init_memcg_params(s); |
dffb4d60 CL |
3667 | list_add(&s->list, &slab_caches); |
3668 | return s; | |
51df1142 CL |
3669 | } |
3670 | ||
81819f0f CL |
3671 | void __init kmem_cache_init(void) |
3672 | { | |
dffb4d60 CL |
3673 | static __initdata struct kmem_cache boot_kmem_cache, |
3674 | boot_kmem_cache_node; | |
51df1142 | 3675 | |
fc8d8620 SG |
3676 | if (debug_guardpage_minorder()) |
3677 | slub_max_order = 0; | |
3678 | ||
dffb4d60 CL |
3679 | kmem_cache_node = &boot_kmem_cache_node; |
3680 | kmem_cache = &boot_kmem_cache; | |
51df1142 | 3681 | |
dffb4d60 CL |
3682 | create_boot_cache(kmem_cache_node, "kmem_cache_node", |
3683 | sizeof(struct kmem_cache_node), SLAB_HWCACHE_ALIGN); | |
b9049e23 | 3684 | |
3ac38faa | 3685 | register_hotmemory_notifier(&slab_memory_callback_nb); |
81819f0f CL |
3686 | |
3687 | /* Able to allocate the per node structures */ | |
3688 | slab_state = PARTIAL; | |
3689 | ||
dffb4d60 CL |
3690 | create_boot_cache(kmem_cache, "kmem_cache", |
3691 | offsetof(struct kmem_cache, node) + | |
3692 | nr_node_ids * sizeof(struct kmem_cache_node *), | |
3693 | SLAB_HWCACHE_ALIGN); | |
8a13a4cc | 3694 | |
dffb4d60 | 3695 | kmem_cache = bootstrap(&boot_kmem_cache); |
81819f0f | 3696 | |
51df1142 CL |
3697 | /* |
3698 | * Allocate kmem_cache_node properly from the kmem_cache slab. | |
3699 | * kmem_cache_node is separately allocated so no need to | |
3700 | * update any list pointers. | |
3701 | */ | |
dffb4d60 | 3702 | kmem_cache_node = bootstrap(&boot_kmem_cache_node); |
51df1142 CL |
3703 | |
3704 | /* Now we can use the kmem_cache to allocate kmalloc slabs */ | |
f97d5f63 | 3705 | create_kmalloc_caches(0); |
81819f0f CL |
3706 | |
3707 | #ifdef CONFIG_SMP | |
3708 | register_cpu_notifier(&slab_notifier); | |
9dfc6e68 | 3709 | #endif |
81819f0f | 3710 | |
f9f58285 | 3711 | pr_info("SLUB: HWalign=%d, Order=%d-%d, MinObjects=%d, CPUs=%d, Nodes=%d\n", |
f97d5f63 | 3712 | cache_line_size(), |
81819f0f CL |
3713 | slub_min_order, slub_max_order, slub_min_objects, |
3714 | nr_cpu_ids, nr_node_ids); | |
3715 | } | |
3716 | ||
7e85ee0c PE |
3717 | void __init kmem_cache_init_late(void) |
3718 | { | |
7e85ee0c PE |
3719 | } |
3720 | ||
2633d7a0 | 3721 | struct kmem_cache * |
a44cb944 VD |
3722 | __kmem_cache_alias(const char *name, size_t size, size_t align, |
3723 | unsigned long flags, void (*ctor)(void *)) | |
81819f0f | 3724 | { |
426589f5 | 3725 | struct kmem_cache *s, *c; |
81819f0f | 3726 | |
a44cb944 | 3727 | s = find_mergeable(size, align, flags, name, ctor); |
81819f0f CL |
3728 | if (s) { |
3729 | s->refcount++; | |
84d0ddd6 | 3730 | |
81819f0f CL |
3731 | /* |
3732 | * Adjust the object sizes so that we clear | |
3733 | * the complete object on kzalloc. | |
3734 | */ | |
3b0efdfa | 3735 | s->object_size = max(s->object_size, (int)size); |
81819f0f | 3736 | s->inuse = max_t(int, s->inuse, ALIGN(size, sizeof(void *))); |
6446faa2 | 3737 | |
426589f5 | 3738 | for_each_memcg_cache(c, s) { |
84d0ddd6 VD |
3739 | c->object_size = s->object_size; |
3740 | c->inuse = max_t(int, c->inuse, | |
3741 | ALIGN(size, sizeof(void *))); | |
3742 | } | |
3743 | ||
7b8f3b66 | 3744 | if (sysfs_slab_alias(s, name)) { |
7b8f3b66 | 3745 | s->refcount--; |
cbb79694 | 3746 | s = NULL; |
7b8f3b66 | 3747 | } |
a0e1d1be | 3748 | } |
6446faa2 | 3749 | |
cbb79694 CL |
3750 | return s; |
3751 | } | |
84c1cf62 | 3752 | |
8a13a4cc | 3753 | int __kmem_cache_create(struct kmem_cache *s, unsigned long flags) |
cbb79694 | 3754 | { |
aac3a166 PE |
3755 | int err; |
3756 | ||
3757 | err = kmem_cache_open(s, flags); | |
3758 | if (err) | |
3759 | return err; | |
20cea968 | 3760 | |
45530c44 CL |
3761 | /* Mutex is not taken during early boot */ |
3762 | if (slab_state <= UP) | |
3763 | return 0; | |
3764 | ||
107dab5c | 3765 | memcg_propagate_slab_attrs(s); |
aac3a166 | 3766 | err = sysfs_slab_add(s); |
aac3a166 PE |
3767 | if (err) |
3768 | kmem_cache_close(s); | |
20cea968 | 3769 | |
aac3a166 | 3770 | return err; |
81819f0f | 3771 | } |
81819f0f | 3772 | |
81819f0f | 3773 | #ifdef CONFIG_SMP |
81819f0f | 3774 | /* |
672bba3a CL |
3775 | * Use the cpu notifier to insure that the cpu slabs are flushed when |
3776 | * necessary. | |
81819f0f | 3777 | */ |
0db0628d | 3778 | static int slab_cpuup_callback(struct notifier_block *nfb, |
81819f0f CL |
3779 | unsigned long action, void *hcpu) |
3780 | { | |
3781 | long cpu = (long)hcpu; | |
5b95a4ac CL |
3782 | struct kmem_cache *s; |
3783 | unsigned long flags; | |
81819f0f CL |
3784 | |
3785 | switch (action) { | |
3786 | case CPU_UP_CANCELED: | |
8bb78442 | 3787 | case CPU_UP_CANCELED_FROZEN: |
81819f0f | 3788 | case CPU_DEAD: |
8bb78442 | 3789 | case CPU_DEAD_FROZEN: |
18004c5d | 3790 | mutex_lock(&slab_mutex); |
5b95a4ac CL |
3791 | list_for_each_entry(s, &slab_caches, list) { |
3792 | local_irq_save(flags); | |
3793 | __flush_cpu_slab(s, cpu); | |
3794 | local_irq_restore(flags); | |
3795 | } | |
18004c5d | 3796 | mutex_unlock(&slab_mutex); |
81819f0f CL |
3797 | break; |
3798 | default: | |
3799 | break; | |
3800 | } | |
3801 | return NOTIFY_OK; | |
3802 | } | |
3803 | ||
0db0628d | 3804 | static struct notifier_block slab_notifier = { |
3adbefee | 3805 | .notifier_call = slab_cpuup_callback |
06428780 | 3806 | }; |
81819f0f CL |
3807 | |
3808 | #endif | |
3809 | ||
ce71e27c | 3810 | void *__kmalloc_track_caller(size_t size, gfp_t gfpflags, unsigned long caller) |
81819f0f | 3811 | { |
aadb4bc4 | 3812 | struct kmem_cache *s; |
94b528d0 | 3813 | void *ret; |
aadb4bc4 | 3814 | |
95a05b42 | 3815 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) |
eada35ef PE |
3816 | return kmalloc_large(size, gfpflags); |
3817 | ||
2c59dd65 | 3818 | s = kmalloc_slab(size, gfpflags); |
81819f0f | 3819 | |
2408c550 | 3820 | if (unlikely(ZERO_OR_NULL_PTR(s))) |
6cb8f913 | 3821 | return s; |
81819f0f | 3822 | |
2b847c3c | 3823 | ret = slab_alloc(s, gfpflags, caller); |
94b528d0 | 3824 | |
25985edc | 3825 | /* Honor the call site pointer we received. */ |
ca2b84cb | 3826 | trace_kmalloc(caller, ret, size, s->size, gfpflags); |
94b528d0 EGM |
3827 | |
3828 | return ret; | |
81819f0f CL |
3829 | } |
3830 | ||
5d1f57e4 | 3831 | #ifdef CONFIG_NUMA |
81819f0f | 3832 | void *__kmalloc_node_track_caller(size_t size, gfp_t gfpflags, |
ce71e27c | 3833 | int node, unsigned long caller) |
81819f0f | 3834 | { |
aadb4bc4 | 3835 | struct kmem_cache *s; |
94b528d0 | 3836 | void *ret; |
aadb4bc4 | 3837 | |
95a05b42 | 3838 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) { |
d3e14aa3 XF |
3839 | ret = kmalloc_large_node(size, gfpflags, node); |
3840 | ||
3841 | trace_kmalloc_node(caller, ret, | |
3842 | size, PAGE_SIZE << get_order(size), | |
3843 | gfpflags, node); | |
3844 | ||
3845 | return ret; | |
3846 | } | |
eada35ef | 3847 | |
2c59dd65 | 3848 | s = kmalloc_slab(size, gfpflags); |
81819f0f | 3849 | |
2408c550 | 3850 | if (unlikely(ZERO_OR_NULL_PTR(s))) |
6cb8f913 | 3851 | return s; |
81819f0f | 3852 | |
2b847c3c | 3853 | ret = slab_alloc_node(s, gfpflags, node, caller); |
94b528d0 | 3854 | |
25985edc | 3855 | /* Honor the call site pointer we received. */ |
ca2b84cb | 3856 | trace_kmalloc_node(caller, ret, size, s->size, gfpflags, node); |
94b528d0 EGM |
3857 | |
3858 | return ret; | |
81819f0f | 3859 | } |
5d1f57e4 | 3860 | #endif |
81819f0f | 3861 | |
ab4d5ed5 | 3862 | #ifdef CONFIG_SYSFS |
205ab99d CL |
3863 | static int count_inuse(struct page *page) |
3864 | { | |
3865 | return page->inuse; | |
3866 | } | |
3867 | ||
3868 | static int count_total(struct page *page) | |
3869 | { | |
3870 | return page->objects; | |
3871 | } | |
ab4d5ed5 | 3872 | #endif |
205ab99d | 3873 | |
ab4d5ed5 | 3874 | #ifdef CONFIG_SLUB_DEBUG |
434e245d CL |
3875 | static int validate_slab(struct kmem_cache *s, struct page *page, |
3876 | unsigned long *map) | |
53e15af0 CL |
3877 | { |
3878 | void *p; | |
a973e9dd | 3879 | void *addr = page_address(page); |
53e15af0 CL |
3880 | |
3881 | if (!check_slab(s, page) || | |
3882 | !on_freelist(s, page, NULL)) | |
3883 | return 0; | |
3884 | ||
3885 | /* Now we know that a valid freelist exists */ | |
39b26464 | 3886 | bitmap_zero(map, page->objects); |
53e15af0 | 3887 | |
5f80b13a CL |
3888 | get_map(s, page, map); |
3889 | for_each_object(p, s, addr, page->objects) { | |
3890 | if (test_bit(slab_index(p, s, addr), map)) | |
3891 | if (!check_object(s, page, p, SLUB_RED_INACTIVE)) | |
3892 | return 0; | |
53e15af0 CL |
3893 | } |
3894 | ||
224a88be | 3895 | for_each_object(p, s, addr, page->objects) |
7656c72b | 3896 | if (!test_bit(slab_index(p, s, addr), map)) |
37d57443 | 3897 | if (!check_object(s, page, p, SLUB_RED_ACTIVE)) |
53e15af0 CL |
3898 | return 0; |
3899 | return 1; | |
3900 | } | |
3901 | ||
434e245d CL |
3902 | static void validate_slab_slab(struct kmem_cache *s, struct page *page, |
3903 | unsigned long *map) | |
53e15af0 | 3904 | { |
881db7fb CL |
3905 | slab_lock(page); |
3906 | validate_slab(s, page, map); | |
3907 | slab_unlock(page); | |
53e15af0 CL |
3908 | } |
3909 | ||
434e245d CL |
3910 | static int validate_slab_node(struct kmem_cache *s, |
3911 | struct kmem_cache_node *n, unsigned long *map) | |
53e15af0 CL |
3912 | { |
3913 | unsigned long count = 0; | |
3914 | struct page *page; | |
3915 | unsigned long flags; | |
3916 | ||
3917 | spin_lock_irqsave(&n->list_lock, flags); | |
3918 | ||
3919 | list_for_each_entry(page, &n->partial, lru) { | |
434e245d | 3920 | validate_slab_slab(s, page, map); |
53e15af0 CL |
3921 | count++; |
3922 | } | |
3923 | if (count != n->nr_partial) | |
f9f58285 FF |
3924 | pr_err("SLUB %s: %ld partial slabs counted but counter=%ld\n", |
3925 | s->name, count, n->nr_partial); | |
53e15af0 CL |
3926 | |
3927 | if (!(s->flags & SLAB_STORE_USER)) | |
3928 | goto out; | |
3929 | ||
3930 | list_for_each_entry(page, &n->full, lru) { | |
434e245d | 3931 | validate_slab_slab(s, page, map); |
53e15af0 CL |
3932 | count++; |
3933 | } | |
3934 | if (count != atomic_long_read(&n->nr_slabs)) | |
f9f58285 FF |
3935 | pr_err("SLUB: %s %ld slabs counted but counter=%ld\n", |
3936 | s->name, count, atomic_long_read(&n->nr_slabs)); | |
53e15af0 CL |
3937 | |
3938 | out: | |
3939 | spin_unlock_irqrestore(&n->list_lock, flags); | |
3940 | return count; | |
3941 | } | |
3942 | ||
434e245d | 3943 | static long validate_slab_cache(struct kmem_cache *s) |
53e15af0 CL |
3944 | { |
3945 | int node; | |
3946 | unsigned long count = 0; | |
205ab99d | 3947 | unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) * |
434e245d | 3948 | sizeof(unsigned long), GFP_KERNEL); |
fa45dc25 | 3949 | struct kmem_cache_node *n; |
434e245d CL |
3950 | |
3951 | if (!map) | |
3952 | return -ENOMEM; | |
53e15af0 CL |
3953 | |
3954 | flush_all(s); | |
fa45dc25 | 3955 | for_each_kmem_cache_node(s, node, n) |
434e245d | 3956 | count += validate_slab_node(s, n, map); |
434e245d | 3957 | kfree(map); |
53e15af0 CL |
3958 | return count; |
3959 | } | |
88a420e4 | 3960 | /* |
672bba3a | 3961 | * Generate lists of code addresses where slabcache objects are allocated |
88a420e4 CL |
3962 | * and freed. |
3963 | */ | |
3964 | ||
3965 | struct location { | |
3966 | unsigned long count; | |
ce71e27c | 3967 | unsigned long addr; |
45edfa58 CL |
3968 | long long sum_time; |
3969 | long min_time; | |
3970 | long max_time; | |
3971 | long min_pid; | |
3972 | long max_pid; | |
174596a0 | 3973 | DECLARE_BITMAP(cpus, NR_CPUS); |
45edfa58 | 3974 | nodemask_t nodes; |
88a420e4 CL |
3975 | }; |
3976 | ||
3977 | struct loc_track { | |
3978 | unsigned long max; | |
3979 | unsigned long count; | |
3980 | struct location *loc; | |
3981 | }; | |
3982 | ||
3983 | static void free_loc_track(struct loc_track *t) | |
3984 | { | |
3985 | if (t->max) | |
3986 | free_pages((unsigned long)t->loc, | |
3987 | get_order(sizeof(struct location) * t->max)); | |
3988 | } | |
3989 | ||
68dff6a9 | 3990 | static int alloc_loc_track(struct loc_track *t, unsigned long max, gfp_t flags) |
88a420e4 CL |
3991 | { |
3992 | struct location *l; | |
3993 | int order; | |
3994 | ||
88a420e4 CL |
3995 | order = get_order(sizeof(struct location) * max); |
3996 | ||
68dff6a9 | 3997 | l = (void *)__get_free_pages(flags, order); |
88a420e4 CL |
3998 | if (!l) |
3999 | return 0; | |
4000 | ||
4001 | if (t->count) { | |
4002 | memcpy(l, t->loc, sizeof(struct location) * t->count); | |
4003 | free_loc_track(t); | |
4004 | } | |
4005 | t->max = max; | |
4006 | t->loc = l; | |
4007 | return 1; | |
4008 | } | |
4009 | ||
4010 | static int add_location(struct loc_track *t, struct kmem_cache *s, | |
45edfa58 | 4011 | const struct track *track) |
88a420e4 CL |
4012 | { |
4013 | long start, end, pos; | |
4014 | struct location *l; | |
ce71e27c | 4015 | unsigned long caddr; |
45edfa58 | 4016 | unsigned long age = jiffies - track->when; |
88a420e4 CL |
4017 | |
4018 | start = -1; | |
4019 | end = t->count; | |
4020 | ||
4021 | for ( ; ; ) { | |
4022 | pos = start + (end - start + 1) / 2; | |
4023 | ||
4024 | /* | |
4025 | * There is nothing at "end". If we end up there | |
4026 | * we need to add something to before end. | |
4027 | */ | |
4028 | if (pos == end) | |
4029 | break; | |
4030 | ||
4031 | caddr = t->loc[pos].addr; | |
45edfa58 CL |
4032 | if (track->addr == caddr) { |
4033 | ||
4034 | l = &t->loc[pos]; | |
4035 | l->count++; | |
4036 | if (track->when) { | |
4037 | l->sum_time += age; | |
4038 | if (age < l->min_time) | |
4039 | l->min_time = age; | |
4040 | if (age > l->max_time) | |
4041 | l->max_time = age; | |
4042 | ||
4043 | if (track->pid < l->min_pid) | |
4044 | l->min_pid = track->pid; | |
4045 | if (track->pid > l->max_pid) | |
4046 | l->max_pid = track->pid; | |
4047 | ||
174596a0 RR |
4048 | cpumask_set_cpu(track->cpu, |
4049 | to_cpumask(l->cpus)); | |
45edfa58 CL |
4050 | } |
4051 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
88a420e4 CL |
4052 | return 1; |
4053 | } | |
4054 | ||
45edfa58 | 4055 | if (track->addr < caddr) |
88a420e4 CL |
4056 | end = pos; |
4057 | else | |
4058 | start = pos; | |
4059 | } | |
4060 | ||
4061 | /* | |
672bba3a | 4062 | * Not found. Insert new tracking element. |
88a420e4 | 4063 | */ |
68dff6a9 | 4064 | if (t->count >= t->max && !alloc_loc_track(t, 2 * t->max, GFP_ATOMIC)) |
88a420e4 CL |
4065 | return 0; |
4066 | ||
4067 | l = t->loc + pos; | |
4068 | if (pos < t->count) | |
4069 | memmove(l + 1, l, | |
4070 | (t->count - pos) * sizeof(struct location)); | |
4071 | t->count++; | |
4072 | l->count = 1; | |
45edfa58 CL |
4073 | l->addr = track->addr; |
4074 | l->sum_time = age; | |
4075 | l->min_time = age; | |
4076 | l->max_time = age; | |
4077 | l->min_pid = track->pid; | |
4078 | l->max_pid = track->pid; | |
174596a0 RR |
4079 | cpumask_clear(to_cpumask(l->cpus)); |
4080 | cpumask_set_cpu(track->cpu, to_cpumask(l->cpus)); | |
45edfa58 CL |
4081 | nodes_clear(l->nodes); |
4082 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
88a420e4 CL |
4083 | return 1; |
4084 | } | |
4085 | ||
4086 | static void process_slab(struct loc_track *t, struct kmem_cache *s, | |
bbd7d57b | 4087 | struct page *page, enum track_item alloc, |
a5dd5c11 | 4088 | unsigned long *map) |
88a420e4 | 4089 | { |
a973e9dd | 4090 | void *addr = page_address(page); |
88a420e4 CL |
4091 | void *p; |
4092 | ||
39b26464 | 4093 | bitmap_zero(map, page->objects); |
5f80b13a | 4094 | get_map(s, page, map); |
88a420e4 | 4095 | |
224a88be | 4096 | for_each_object(p, s, addr, page->objects) |
45edfa58 CL |
4097 | if (!test_bit(slab_index(p, s, addr), map)) |
4098 | add_location(t, s, get_track(s, p, alloc)); | |
88a420e4 CL |
4099 | } |
4100 | ||
4101 | static int list_locations(struct kmem_cache *s, char *buf, | |
4102 | enum track_item alloc) | |
4103 | { | |
e374d483 | 4104 | int len = 0; |
88a420e4 | 4105 | unsigned long i; |
68dff6a9 | 4106 | struct loc_track t = { 0, 0, NULL }; |
88a420e4 | 4107 | int node; |
bbd7d57b ED |
4108 | unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) * |
4109 | sizeof(unsigned long), GFP_KERNEL); | |
fa45dc25 | 4110 | struct kmem_cache_node *n; |
88a420e4 | 4111 | |
bbd7d57b ED |
4112 | if (!map || !alloc_loc_track(&t, PAGE_SIZE / sizeof(struct location), |
4113 | GFP_TEMPORARY)) { | |
4114 | kfree(map); | |
68dff6a9 | 4115 | return sprintf(buf, "Out of memory\n"); |
bbd7d57b | 4116 | } |
88a420e4 CL |
4117 | /* Push back cpu slabs */ |
4118 | flush_all(s); | |
4119 | ||
fa45dc25 | 4120 | for_each_kmem_cache_node(s, node, n) { |
88a420e4 CL |
4121 | unsigned long flags; |
4122 | struct page *page; | |
4123 | ||
9e86943b | 4124 | if (!atomic_long_read(&n->nr_slabs)) |
88a420e4 CL |
4125 | continue; |
4126 | ||
4127 | spin_lock_irqsave(&n->list_lock, flags); | |
4128 | list_for_each_entry(page, &n->partial, lru) | |
bbd7d57b | 4129 | process_slab(&t, s, page, alloc, map); |
88a420e4 | 4130 | list_for_each_entry(page, &n->full, lru) |
bbd7d57b | 4131 | process_slab(&t, s, page, alloc, map); |
88a420e4 CL |
4132 | spin_unlock_irqrestore(&n->list_lock, flags); |
4133 | } | |
4134 | ||
4135 | for (i = 0; i < t.count; i++) { | |
45edfa58 | 4136 | struct location *l = &t.loc[i]; |
88a420e4 | 4137 | |
9c246247 | 4138 | if (len > PAGE_SIZE - KSYM_SYMBOL_LEN - 100) |
88a420e4 | 4139 | break; |
e374d483 | 4140 | len += sprintf(buf + len, "%7ld ", l->count); |
45edfa58 CL |
4141 | |
4142 | if (l->addr) | |
62c70bce | 4143 | len += sprintf(buf + len, "%pS", (void *)l->addr); |
88a420e4 | 4144 | else |
e374d483 | 4145 | len += sprintf(buf + len, "<not-available>"); |
45edfa58 CL |
4146 | |
4147 | if (l->sum_time != l->min_time) { | |
e374d483 | 4148 | len += sprintf(buf + len, " age=%ld/%ld/%ld", |
f8bd2258 RZ |
4149 | l->min_time, |
4150 | (long)div_u64(l->sum_time, l->count), | |
4151 | l->max_time); | |
45edfa58 | 4152 | } else |
e374d483 | 4153 | len += sprintf(buf + len, " age=%ld", |
45edfa58 CL |
4154 | l->min_time); |
4155 | ||
4156 | if (l->min_pid != l->max_pid) | |
e374d483 | 4157 | len += sprintf(buf + len, " pid=%ld-%ld", |
45edfa58 CL |
4158 | l->min_pid, l->max_pid); |
4159 | else | |
e374d483 | 4160 | len += sprintf(buf + len, " pid=%ld", |
45edfa58 CL |
4161 | l->min_pid); |
4162 | ||
174596a0 RR |
4163 | if (num_online_cpus() > 1 && |
4164 | !cpumask_empty(to_cpumask(l->cpus)) && | |
5024c1d7 TH |
4165 | len < PAGE_SIZE - 60) |
4166 | len += scnprintf(buf + len, PAGE_SIZE - len - 50, | |
4167 | " cpus=%*pbl", | |
4168 | cpumask_pr_args(to_cpumask(l->cpus))); | |
45edfa58 | 4169 | |
62bc62a8 | 4170 | if (nr_online_nodes > 1 && !nodes_empty(l->nodes) && |
5024c1d7 TH |
4171 | len < PAGE_SIZE - 60) |
4172 | len += scnprintf(buf + len, PAGE_SIZE - len - 50, | |
4173 | " nodes=%*pbl", | |
4174 | nodemask_pr_args(&l->nodes)); | |
45edfa58 | 4175 | |
e374d483 | 4176 | len += sprintf(buf + len, "\n"); |
88a420e4 CL |
4177 | } |
4178 | ||
4179 | free_loc_track(&t); | |
bbd7d57b | 4180 | kfree(map); |
88a420e4 | 4181 | if (!t.count) |
e374d483 HH |
4182 | len += sprintf(buf, "No data\n"); |
4183 | return len; | |
88a420e4 | 4184 | } |
ab4d5ed5 | 4185 | #endif |
88a420e4 | 4186 | |
a5a84755 | 4187 | #ifdef SLUB_RESILIENCY_TEST |
c07b8183 | 4188 | static void __init resiliency_test(void) |
a5a84755 CL |
4189 | { |
4190 | u8 *p; | |
4191 | ||
95a05b42 | 4192 | BUILD_BUG_ON(KMALLOC_MIN_SIZE > 16 || KMALLOC_SHIFT_HIGH < 10); |
a5a84755 | 4193 | |
f9f58285 FF |
4194 | pr_err("SLUB resiliency testing\n"); |
4195 | pr_err("-----------------------\n"); | |
4196 | pr_err("A. Corruption after allocation\n"); | |
a5a84755 CL |
4197 | |
4198 | p = kzalloc(16, GFP_KERNEL); | |
4199 | p[16] = 0x12; | |
f9f58285 FF |
4200 | pr_err("\n1. kmalloc-16: Clobber Redzone/next pointer 0x12->0x%p\n\n", |
4201 | p + 16); | |
a5a84755 CL |
4202 | |
4203 | validate_slab_cache(kmalloc_caches[4]); | |
4204 | ||
4205 | /* Hmmm... The next two are dangerous */ | |
4206 | p = kzalloc(32, GFP_KERNEL); | |
4207 | p[32 + sizeof(void *)] = 0x34; | |
f9f58285 FF |
4208 | pr_err("\n2. kmalloc-32: Clobber next pointer/next slab 0x34 -> -0x%p\n", |
4209 | p); | |
4210 | pr_err("If allocated object is overwritten then not detectable\n\n"); | |
a5a84755 CL |
4211 | |
4212 | validate_slab_cache(kmalloc_caches[5]); | |
4213 | p = kzalloc(64, GFP_KERNEL); | |
4214 | p += 64 + (get_cycles() & 0xff) * sizeof(void *); | |
4215 | *p = 0x56; | |
f9f58285 FF |
4216 | pr_err("\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n", |
4217 | p); | |
4218 | pr_err("If allocated object is overwritten then not detectable\n\n"); | |
a5a84755 CL |
4219 | validate_slab_cache(kmalloc_caches[6]); |
4220 | ||
f9f58285 | 4221 | pr_err("\nB. Corruption after free\n"); |
a5a84755 CL |
4222 | p = kzalloc(128, GFP_KERNEL); |
4223 | kfree(p); | |
4224 | *p = 0x78; | |
f9f58285 | 4225 | pr_err("1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p); |
a5a84755 CL |
4226 | validate_slab_cache(kmalloc_caches[7]); |
4227 | ||
4228 | p = kzalloc(256, GFP_KERNEL); | |
4229 | kfree(p); | |
4230 | p[50] = 0x9a; | |
f9f58285 | 4231 | pr_err("\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n", p); |
a5a84755 CL |
4232 | validate_slab_cache(kmalloc_caches[8]); |
4233 | ||
4234 | p = kzalloc(512, GFP_KERNEL); | |
4235 | kfree(p); | |
4236 | p[512] = 0xab; | |
f9f58285 | 4237 | pr_err("\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p); |
a5a84755 CL |
4238 | validate_slab_cache(kmalloc_caches[9]); |
4239 | } | |
4240 | #else | |
4241 | #ifdef CONFIG_SYSFS | |
4242 | static void resiliency_test(void) {}; | |
4243 | #endif | |
4244 | #endif | |
4245 | ||
ab4d5ed5 | 4246 | #ifdef CONFIG_SYSFS |
81819f0f | 4247 | enum slab_stat_type { |
205ab99d CL |
4248 | SL_ALL, /* All slabs */ |
4249 | SL_PARTIAL, /* Only partially allocated slabs */ | |
4250 | SL_CPU, /* Only slabs used for cpu caches */ | |
4251 | SL_OBJECTS, /* Determine allocated objects not slabs */ | |
4252 | SL_TOTAL /* Determine object capacity not slabs */ | |
81819f0f CL |
4253 | }; |
4254 | ||
205ab99d | 4255 | #define SO_ALL (1 << SL_ALL) |
81819f0f CL |
4256 | #define SO_PARTIAL (1 << SL_PARTIAL) |
4257 | #define SO_CPU (1 << SL_CPU) | |
4258 | #define SO_OBJECTS (1 << SL_OBJECTS) | |
205ab99d | 4259 | #define SO_TOTAL (1 << SL_TOTAL) |
81819f0f | 4260 | |
62e5c4b4 CG |
4261 | static ssize_t show_slab_objects(struct kmem_cache *s, |
4262 | char *buf, unsigned long flags) | |
81819f0f CL |
4263 | { |
4264 | unsigned long total = 0; | |
81819f0f CL |
4265 | int node; |
4266 | int x; | |
4267 | unsigned long *nodes; | |
81819f0f | 4268 | |
e35e1a97 | 4269 | nodes = kzalloc(sizeof(unsigned long) * nr_node_ids, GFP_KERNEL); |
62e5c4b4 CG |
4270 | if (!nodes) |
4271 | return -ENOMEM; | |
81819f0f | 4272 | |
205ab99d CL |
4273 | if (flags & SO_CPU) { |
4274 | int cpu; | |
81819f0f | 4275 | |
205ab99d | 4276 | for_each_possible_cpu(cpu) { |
d0e0ac97 CG |
4277 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, |
4278 | cpu); | |
ec3ab083 | 4279 | int node; |
49e22585 | 4280 | struct page *page; |
dfb4f096 | 4281 | |
bc6697d8 | 4282 | page = ACCESS_ONCE(c->page); |
ec3ab083 CL |
4283 | if (!page) |
4284 | continue; | |
205ab99d | 4285 | |
ec3ab083 CL |
4286 | node = page_to_nid(page); |
4287 | if (flags & SO_TOTAL) | |
4288 | x = page->objects; | |
4289 | else if (flags & SO_OBJECTS) | |
4290 | x = page->inuse; | |
4291 | else | |
4292 | x = 1; | |
49e22585 | 4293 | |
ec3ab083 CL |
4294 | total += x; |
4295 | nodes[node] += x; | |
4296 | ||
4297 | page = ACCESS_ONCE(c->partial); | |
49e22585 | 4298 | if (page) { |
8afb1474 LZ |
4299 | node = page_to_nid(page); |
4300 | if (flags & SO_TOTAL) | |
4301 | WARN_ON_ONCE(1); | |
4302 | else if (flags & SO_OBJECTS) | |
4303 | WARN_ON_ONCE(1); | |
4304 | else | |
4305 | x = page->pages; | |
bc6697d8 ED |
4306 | total += x; |
4307 | nodes[node] += x; | |
49e22585 | 4308 | } |
81819f0f CL |
4309 | } |
4310 | } | |
4311 | ||
bfc8c901 | 4312 | get_online_mems(); |
ab4d5ed5 | 4313 | #ifdef CONFIG_SLUB_DEBUG |
205ab99d | 4314 | if (flags & SO_ALL) { |
fa45dc25 CL |
4315 | struct kmem_cache_node *n; |
4316 | ||
4317 | for_each_kmem_cache_node(s, node, n) { | |
205ab99d | 4318 | |
d0e0ac97 CG |
4319 | if (flags & SO_TOTAL) |
4320 | x = atomic_long_read(&n->total_objects); | |
4321 | else if (flags & SO_OBJECTS) | |
4322 | x = atomic_long_read(&n->total_objects) - | |
4323 | count_partial(n, count_free); | |
81819f0f | 4324 | else |
205ab99d | 4325 | x = atomic_long_read(&n->nr_slabs); |
81819f0f CL |
4326 | total += x; |
4327 | nodes[node] += x; | |
4328 | } | |
4329 | ||
ab4d5ed5 CL |
4330 | } else |
4331 | #endif | |
4332 | if (flags & SO_PARTIAL) { | |
fa45dc25 | 4333 | struct kmem_cache_node *n; |
81819f0f | 4334 | |
fa45dc25 | 4335 | for_each_kmem_cache_node(s, node, n) { |
205ab99d CL |
4336 | if (flags & SO_TOTAL) |
4337 | x = count_partial(n, count_total); | |
4338 | else if (flags & SO_OBJECTS) | |
4339 | x = count_partial(n, count_inuse); | |
81819f0f | 4340 | else |
205ab99d | 4341 | x = n->nr_partial; |
81819f0f CL |
4342 | total += x; |
4343 | nodes[node] += x; | |
4344 | } | |
4345 | } | |
81819f0f CL |
4346 | x = sprintf(buf, "%lu", total); |
4347 | #ifdef CONFIG_NUMA | |
fa45dc25 | 4348 | for (node = 0; node < nr_node_ids; node++) |
81819f0f CL |
4349 | if (nodes[node]) |
4350 | x += sprintf(buf + x, " N%d=%lu", | |
4351 | node, nodes[node]); | |
4352 | #endif | |
bfc8c901 | 4353 | put_online_mems(); |
81819f0f CL |
4354 | kfree(nodes); |
4355 | return x + sprintf(buf + x, "\n"); | |
4356 | } | |
4357 | ||
ab4d5ed5 | 4358 | #ifdef CONFIG_SLUB_DEBUG |
81819f0f CL |
4359 | static int any_slab_objects(struct kmem_cache *s) |
4360 | { | |
4361 | int node; | |
fa45dc25 | 4362 | struct kmem_cache_node *n; |
81819f0f | 4363 | |
fa45dc25 | 4364 | for_each_kmem_cache_node(s, node, n) |
4ea33e2d | 4365 | if (atomic_long_read(&n->total_objects)) |
81819f0f | 4366 | return 1; |
fa45dc25 | 4367 | |
81819f0f CL |
4368 | return 0; |
4369 | } | |
ab4d5ed5 | 4370 | #endif |
81819f0f CL |
4371 | |
4372 | #define to_slab_attr(n) container_of(n, struct slab_attribute, attr) | |
497888cf | 4373 | #define to_slab(n) container_of(n, struct kmem_cache, kobj) |
81819f0f CL |
4374 | |
4375 | struct slab_attribute { | |
4376 | struct attribute attr; | |
4377 | ssize_t (*show)(struct kmem_cache *s, char *buf); | |
4378 | ssize_t (*store)(struct kmem_cache *s, const char *x, size_t count); | |
4379 | }; | |
4380 | ||
4381 | #define SLAB_ATTR_RO(_name) \ | |
ab067e99 VK |
4382 | static struct slab_attribute _name##_attr = \ |
4383 | __ATTR(_name, 0400, _name##_show, NULL) | |
81819f0f CL |
4384 | |
4385 | #define SLAB_ATTR(_name) \ | |
4386 | static struct slab_attribute _name##_attr = \ | |
ab067e99 | 4387 | __ATTR(_name, 0600, _name##_show, _name##_store) |
81819f0f | 4388 | |
81819f0f CL |
4389 | static ssize_t slab_size_show(struct kmem_cache *s, char *buf) |
4390 | { | |
4391 | return sprintf(buf, "%d\n", s->size); | |
4392 | } | |
4393 | SLAB_ATTR_RO(slab_size); | |
4394 | ||
4395 | static ssize_t align_show(struct kmem_cache *s, char *buf) | |
4396 | { | |
4397 | return sprintf(buf, "%d\n", s->align); | |
4398 | } | |
4399 | SLAB_ATTR_RO(align); | |
4400 | ||
4401 | static ssize_t object_size_show(struct kmem_cache *s, char *buf) | |
4402 | { | |
3b0efdfa | 4403 | return sprintf(buf, "%d\n", s->object_size); |
81819f0f CL |
4404 | } |
4405 | SLAB_ATTR_RO(object_size); | |
4406 | ||
4407 | static ssize_t objs_per_slab_show(struct kmem_cache *s, char *buf) | |
4408 | { | |
834f3d11 | 4409 | return sprintf(buf, "%d\n", oo_objects(s->oo)); |
81819f0f CL |
4410 | } |
4411 | SLAB_ATTR_RO(objs_per_slab); | |
4412 | ||
06b285dc CL |
4413 | static ssize_t order_store(struct kmem_cache *s, |
4414 | const char *buf, size_t length) | |
4415 | { | |
0121c619 CL |
4416 | unsigned long order; |
4417 | int err; | |
4418 | ||
3dbb95f7 | 4419 | err = kstrtoul(buf, 10, &order); |
0121c619 CL |
4420 | if (err) |
4421 | return err; | |
06b285dc CL |
4422 | |
4423 | if (order > slub_max_order || order < slub_min_order) | |
4424 | return -EINVAL; | |
4425 | ||
4426 | calculate_sizes(s, order); | |
4427 | return length; | |
4428 | } | |
4429 | ||
81819f0f CL |
4430 | static ssize_t order_show(struct kmem_cache *s, char *buf) |
4431 | { | |
834f3d11 | 4432 | return sprintf(buf, "%d\n", oo_order(s->oo)); |
81819f0f | 4433 | } |
06b285dc | 4434 | SLAB_ATTR(order); |
81819f0f | 4435 | |
73d342b1 DR |
4436 | static ssize_t min_partial_show(struct kmem_cache *s, char *buf) |
4437 | { | |
4438 | return sprintf(buf, "%lu\n", s->min_partial); | |
4439 | } | |
4440 | ||
4441 | static ssize_t min_partial_store(struct kmem_cache *s, const char *buf, | |
4442 | size_t length) | |
4443 | { | |
4444 | unsigned long min; | |
4445 | int err; | |
4446 | ||
3dbb95f7 | 4447 | err = kstrtoul(buf, 10, &min); |
73d342b1 DR |
4448 | if (err) |
4449 | return err; | |
4450 | ||
c0bdb232 | 4451 | set_min_partial(s, min); |
73d342b1 DR |
4452 | return length; |
4453 | } | |
4454 | SLAB_ATTR(min_partial); | |
4455 | ||
49e22585 CL |
4456 | static ssize_t cpu_partial_show(struct kmem_cache *s, char *buf) |
4457 | { | |
4458 | return sprintf(buf, "%u\n", s->cpu_partial); | |
4459 | } | |
4460 | ||
4461 | static ssize_t cpu_partial_store(struct kmem_cache *s, const char *buf, | |
4462 | size_t length) | |
4463 | { | |
4464 | unsigned long objects; | |
4465 | int err; | |
4466 | ||
3dbb95f7 | 4467 | err = kstrtoul(buf, 10, &objects); |
49e22585 CL |
4468 | if (err) |
4469 | return err; | |
345c905d | 4470 | if (objects && !kmem_cache_has_cpu_partial(s)) |
74ee4ef1 | 4471 | return -EINVAL; |
49e22585 CL |
4472 | |
4473 | s->cpu_partial = objects; | |
4474 | flush_all(s); | |
4475 | return length; | |
4476 | } | |
4477 | SLAB_ATTR(cpu_partial); | |
4478 | ||
81819f0f CL |
4479 | static ssize_t ctor_show(struct kmem_cache *s, char *buf) |
4480 | { | |
62c70bce JP |
4481 | if (!s->ctor) |
4482 | return 0; | |
4483 | return sprintf(buf, "%pS\n", s->ctor); | |
81819f0f CL |
4484 | } |
4485 | SLAB_ATTR_RO(ctor); | |
4486 | ||
81819f0f CL |
4487 | static ssize_t aliases_show(struct kmem_cache *s, char *buf) |
4488 | { | |
4307c14f | 4489 | return sprintf(buf, "%d\n", s->refcount < 0 ? 0 : s->refcount - 1); |
81819f0f CL |
4490 | } |
4491 | SLAB_ATTR_RO(aliases); | |
4492 | ||
81819f0f CL |
4493 | static ssize_t partial_show(struct kmem_cache *s, char *buf) |
4494 | { | |
d9acf4b7 | 4495 | return show_slab_objects(s, buf, SO_PARTIAL); |
81819f0f CL |
4496 | } |
4497 | SLAB_ATTR_RO(partial); | |
4498 | ||
4499 | static ssize_t cpu_slabs_show(struct kmem_cache *s, char *buf) | |
4500 | { | |
d9acf4b7 | 4501 | return show_slab_objects(s, buf, SO_CPU); |
81819f0f CL |
4502 | } |
4503 | SLAB_ATTR_RO(cpu_slabs); | |
4504 | ||
4505 | static ssize_t objects_show(struct kmem_cache *s, char *buf) | |
4506 | { | |
205ab99d | 4507 | return show_slab_objects(s, buf, SO_ALL|SO_OBJECTS); |
81819f0f CL |
4508 | } |
4509 | SLAB_ATTR_RO(objects); | |
4510 | ||
205ab99d CL |
4511 | static ssize_t objects_partial_show(struct kmem_cache *s, char *buf) |
4512 | { | |
4513 | return show_slab_objects(s, buf, SO_PARTIAL|SO_OBJECTS); | |
4514 | } | |
4515 | SLAB_ATTR_RO(objects_partial); | |
4516 | ||
49e22585 CL |
4517 | static ssize_t slabs_cpu_partial_show(struct kmem_cache *s, char *buf) |
4518 | { | |
4519 | int objects = 0; | |
4520 | int pages = 0; | |
4521 | int cpu; | |
4522 | int len; | |
4523 | ||
4524 | for_each_online_cpu(cpu) { | |
4525 | struct page *page = per_cpu_ptr(s->cpu_slab, cpu)->partial; | |
4526 | ||
4527 | if (page) { | |
4528 | pages += page->pages; | |
4529 | objects += page->pobjects; | |
4530 | } | |
4531 | } | |
4532 | ||
4533 | len = sprintf(buf, "%d(%d)", objects, pages); | |
4534 | ||
4535 | #ifdef CONFIG_SMP | |
4536 | for_each_online_cpu(cpu) { | |
4537 | struct page *page = per_cpu_ptr(s->cpu_slab, cpu) ->partial; | |
4538 | ||
4539 | if (page && len < PAGE_SIZE - 20) | |
4540 | len += sprintf(buf + len, " C%d=%d(%d)", cpu, | |
4541 | page->pobjects, page->pages); | |
4542 | } | |
4543 | #endif | |
4544 | return len + sprintf(buf + len, "\n"); | |
4545 | } | |
4546 | SLAB_ATTR_RO(slabs_cpu_partial); | |
4547 | ||
a5a84755 CL |
4548 | static ssize_t reclaim_account_show(struct kmem_cache *s, char *buf) |
4549 | { | |
4550 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_RECLAIM_ACCOUNT)); | |
4551 | } | |
4552 | ||
4553 | static ssize_t reclaim_account_store(struct kmem_cache *s, | |
4554 | const char *buf, size_t length) | |
4555 | { | |
4556 | s->flags &= ~SLAB_RECLAIM_ACCOUNT; | |
4557 | if (buf[0] == '1') | |
4558 | s->flags |= SLAB_RECLAIM_ACCOUNT; | |
4559 | return length; | |
4560 | } | |
4561 | SLAB_ATTR(reclaim_account); | |
4562 | ||
4563 | static ssize_t hwcache_align_show(struct kmem_cache *s, char *buf) | |
4564 | { | |
4565 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_HWCACHE_ALIGN)); | |
4566 | } | |
4567 | SLAB_ATTR_RO(hwcache_align); | |
4568 | ||
4569 | #ifdef CONFIG_ZONE_DMA | |
4570 | static ssize_t cache_dma_show(struct kmem_cache *s, char *buf) | |
4571 | { | |
4572 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_CACHE_DMA)); | |
4573 | } | |
4574 | SLAB_ATTR_RO(cache_dma); | |
4575 | #endif | |
4576 | ||
4577 | static ssize_t destroy_by_rcu_show(struct kmem_cache *s, char *buf) | |
4578 | { | |
4579 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_DESTROY_BY_RCU)); | |
4580 | } | |
4581 | SLAB_ATTR_RO(destroy_by_rcu); | |
4582 | ||
ab9a0f19 LJ |
4583 | static ssize_t reserved_show(struct kmem_cache *s, char *buf) |
4584 | { | |
4585 | return sprintf(buf, "%d\n", s->reserved); | |
4586 | } | |
4587 | SLAB_ATTR_RO(reserved); | |
4588 | ||
ab4d5ed5 | 4589 | #ifdef CONFIG_SLUB_DEBUG |
a5a84755 CL |
4590 | static ssize_t slabs_show(struct kmem_cache *s, char *buf) |
4591 | { | |
4592 | return show_slab_objects(s, buf, SO_ALL); | |
4593 | } | |
4594 | SLAB_ATTR_RO(slabs); | |
4595 | ||
205ab99d CL |
4596 | static ssize_t total_objects_show(struct kmem_cache *s, char *buf) |
4597 | { | |
4598 | return show_slab_objects(s, buf, SO_ALL|SO_TOTAL); | |
4599 | } | |
4600 | SLAB_ATTR_RO(total_objects); | |
4601 | ||
81819f0f CL |
4602 | static ssize_t sanity_checks_show(struct kmem_cache *s, char *buf) |
4603 | { | |
4604 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_DEBUG_FREE)); | |
4605 | } | |
4606 | ||
4607 | static ssize_t sanity_checks_store(struct kmem_cache *s, | |
4608 | const char *buf, size_t length) | |
4609 | { | |
4610 | s->flags &= ~SLAB_DEBUG_FREE; | |
b789ef51 CL |
4611 | if (buf[0] == '1') { |
4612 | s->flags &= ~__CMPXCHG_DOUBLE; | |
81819f0f | 4613 | s->flags |= SLAB_DEBUG_FREE; |
b789ef51 | 4614 | } |
81819f0f CL |
4615 | return length; |
4616 | } | |
4617 | SLAB_ATTR(sanity_checks); | |
4618 | ||
4619 | static ssize_t trace_show(struct kmem_cache *s, char *buf) | |
4620 | { | |
4621 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_TRACE)); | |
4622 | } | |
4623 | ||
4624 | static ssize_t trace_store(struct kmem_cache *s, const char *buf, | |
4625 | size_t length) | |
4626 | { | |
c9e16131 CL |
4627 | /* |
4628 | * Tracing a merged cache is going to give confusing results | |
4629 | * as well as cause other issues like converting a mergeable | |
4630 | * cache into an umergeable one. | |
4631 | */ | |
4632 | if (s->refcount > 1) | |
4633 | return -EINVAL; | |
4634 | ||
81819f0f | 4635 | s->flags &= ~SLAB_TRACE; |
b789ef51 CL |
4636 | if (buf[0] == '1') { |
4637 | s->flags &= ~__CMPXCHG_DOUBLE; | |
81819f0f | 4638 | s->flags |= SLAB_TRACE; |
b789ef51 | 4639 | } |
81819f0f CL |
4640 | return length; |
4641 | } | |
4642 | SLAB_ATTR(trace); | |
4643 | ||
81819f0f CL |
4644 | static ssize_t red_zone_show(struct kmem_cache *s, char *buf) |
4645 | { | |
4646 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_RED_ZONE)); | |
4647 | } | |
4648 | ||
4649 | static ssize_t red_zone_store(struct kmem_cache *s, | |
4650 | const char *buf, size_t length) | |
4651 | { | |
4652 | if (any_slab_objects(s)) | |
4653 | return -EBUSY; | |
4654 | ||
4655 | s->flags &= ~SLAB_RED_ZONE; | |
b789ef51 CL |
4656 | if (buf[0] == '1') { |
4657 | s->flags &= ~__CMPXCHG_DOUBLE; | |
81819f0f | 4658 | s->flags |= SLAB_RED_ZONE; |
b789ef51 | 4659 | } |
06b285dc | 4660 | calculate_sizes(s, -1); |
81819f0f CL |
4661 | return length; |
4662 | } | |
4663 | SLAB_ATTR(red_zone); | |
4664 | ||
4665 | static ssize_t poison_show(struct kmem_cache *s, char *buf) | |
4666 | { | |
4667 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_POISON)); | |
4668 | } | |
4669 | ||
4670 | static ssize_t poison_store(struct kmem_cache *s, | |
4671 | const char *buf, size_t length) | |
4672 | { | |
4673 | if (any_slab_objects(s)) | |
4674 | return -EBUSY; | |
4675 | ||
4676 | s->flags &= ~SLAB_POISON; | |
b789ef51 CL |
4677 | if (buf[0] == '1') { |
4678 | s->flags &= ~__CMPXCHG_DOUBLE; | |
81819f0f | 4679 | s->flags |= SLAB_POISON; |
b789ef51 | 4680 | } |
06b285dc | 4681 | calculate_sizes(s, -1); |
81819f0f CL |
4682 | return length; |
4683 | } | |
4684 | SLAB_ATTR(poison); | |
4685 | ||
4686 | static ssize_t store_user_show(struct kmem_cache *s, char *buf) | |
4687 | { | |
4688 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_STORE_USER)); | |
4689 | } | |
4690 | ||
4691 | static ssize_t store_user_store(struct kmem_cache *s, | |
4692 | const char *buf, size_t length) | |
4693 | { | |
4694 | if (any_slab_objects(s)) | |
4695 | return -EBUSY; | |
4696 | ||
4697 | s->flags &= ~SLAB_STORE_USER; | |
b789ef51 CL |
4698 | if (buf[0] == '1') { |
4699 | s->flags &= ~__CMPXCHG_DOUBLE; | |
81819f0f | 4700 | s->flags |= SLAB_STORE_USER; |
b789ef51 | 4701 | } |
06b285dc | 4702 | calculate_sizes(s, -1); |
81819f0f CL |
4703 | return length; |
4704 | } | |
4705 | SLAB_ATTR(store_user); | |
4706 | ||
53e15af0 CL |
4707 | static ssize_t validate_show(struct kmem_cache *s, char *buf) |
4708 | { | |
4709 | return 0; | |
4710 | } | |
4711 | ||
4712 | static ssize_t validate_store(struct kmem_cache *s, | |
4713 | const char *buf, size_t length) | |
4714 | { | |
434e245d CL |
4715 | int ret = -EINVAL; |
4716 | ||
4717 | if (buf[0] == '1') { | |
4718 | ret = validate_slab_cache(s); | |
4719 | if (ret >= 0) | |
4720 | ret = length; | |
4721 | } | |
4722 | return ret; | |
53e15af0 CL |
4723 | } |
4724 | SLAB_ATTR(validate); | |
a5a84755 CL |
4725 | |
4726 | static ssize_t alloc_calls_show(struct kmem_cache *s, char *buf) | |
4727 | { | |
4728 | if (!(s->flags & SLAB_STORE_USER)) | |
4729 | return -ENOSYS; | |
4730 | return list_locations(s, buf, TRACK_ALLOC); | |
4731 | } | |
4732 | SLAB_ATTR_RO(alloc_calls); | |
4733 | ||
4734 | static ssize_t free_calls_show(struct kmem_cache *s, char *buf) | |
4735 | { | |
4736 | if (!(s->flags & SLAB_STORE_USER)) | |
4737 | return -ENOSYS; | |
4738 | return list_locations(s, buf, TRACK_FREE); | |
4739 | } | |
4740 | SLAB_ATTR_RO(free_calls); | |
4741 | #endif /* CONFIG_SLUB_DEBUG */ | |
4742 | ||
4743 | #ifdef CONFIG_FAILSLAB | |
4744 | static ssize_t failslab_show(struct kmem_cache *s, char *buf) | |
4745 | { | |
4746 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_FAILSLAB)); | |
4747 | } | |
4748 | ||
4749 | static ssize_t failslab_store(struct kmem_cache *s, const char *buf, | |
4750 | size_t length) | |
4751 | { | |
c9e16131 CL |
4752 | if (s->refcount > 1) |
4753 | return -EINVAL; | |
4754 | ||
a5a84755 CL |
4755 | s->flags &= ~SLAB_FAILSLAB; |
4756 | if (buf[0] == '1') | |
4757 | s->flags |= SLAB_FAILSLAB; | |
4758 | return length; | |
4759 | } | |
4760 | SLAB_ATTR(failslab); | |
ab4d5ed5 | 4761 | #endif |
53e15af0 | 4762 | |
2086d26a CL |
4763 | static ssize_t shrink_show(struct kmem_cache *s, char *buf) |
4764 | { | |
4765 | return 0; | |
4766 | } | |
4767 | ||
4768 | static ssize_t shrink_store(struct kmem_cache *s, | |
4769 | const char *buf, size_t length) | |
4770 | { | |
832f37f5 VD |
4771 | if (buf[0] == '1') |
4772 | kmem_cache_shrink(s); | |
4773 | else | |
2086d26a CL |
4774 | return -EINVAL; |
4775 | return length; | |
4776 | } | |
4777 | SLAB_ATTR(shrink); | |
4778 | ||
81819f0f | 4779 | #ifdef CONFIG_NUMA |
9824601e | 4780 | static ssize_t remote_node_defrag_ratio_show(struct kmem_cache *s, char *buf) |
81819f0f | 4781 | { |
9824601e | 4782 | return sprintf(buf, "%d\n", s->remote_node_defrag_ratio / 10); |
81819f0f CL |
4783 | } |
4784 | ||
9824601e | 4785 | static ssize_t remote_node_defrag_ratio_store(struct kmem_cache *s, |
81819f0f CL |
4786 | const char *buf, size_t length) |
4787 | { | |
0121c619 CL |
4788 | unsigned long ratio; |
4789 | int err; | |
4790 | ||
3dbb95f7 | 4791 | err = kstrtoul(buf, 10, &ratio); |
0121c619 CL |
4792 | if (err) |
4793 | return err; | |
4794 | ||
e2cb96b7 | 4795 | if (ratio <= 100) |
0121c619 | 4796 | s->remote_node_defrag_ratio = ratio * 10; |
81819f0f | 4797 | |
81819f0f CL |
4798 | return length; |
4799 | } | |
9824601e | 4800 | SLAB_ATTR(remote_node_defrag_ratio); |
81819f0f CL |
4801 | #endif |
4802 | ||
8ff12cfc | 4803 | #ifdef CONFIG_SLUB_STATS |
8ff12cfc CL |
4804 | static int show_stat(struct kmem_cache *s, char *buf, enum stat_item si) |
4805 | { | |
4806 | unsigned long sum = 0; | |
4807 | int cpu; | |
4808 | int len; | |
4809 | int *data = kmalloc(nr_cpu_ids * sizeof(int), GFP_KERNEL); | |
4810 | ||
4811 | if (!data) | |
4812 | return -ENOMEM; | |
4813 | ||
4814 | for_each_online_cpu(cpu) { | |
9dfc6e68 | 4815 | unsigned x = per_cpu_ptr(s->cpu_slab, cpu)->stat[si]; |
8ff12cfc CL |
4816 | |
4817 | data[cpu] = x; | |
4818 | sum += x; | |
4819 | } | |
4820 | ||
4821 | len = sprintf(buf, "%lu", sum); | |
4822 | ||
50ef37b9 | 4823 | #ifdef CONFIG_SMP |
8ff12cfc CL |
4824 | for_each_online_cpu(cpu) { |
4825 | if (data[cpu] && len < PAGE_SIZE - 20) | |
50ef37b9 | 4826 | len += sprintf(buf + len, " C%d=%u", cpu, data[cpu]); |
8ff12cfc | 4827 | } |
50ef37b9 | 4828 | #endif |
8ff12cfc CL |
4829 | kfree(data); |
4830 | return len + sprintf(buf + len, "\n"); | |
4831 | } | |
4832 | ||
78eb00cc DR |
4833 | static void clear_stat(struct kmem_cache *s, enum stat_item si) |
4834 | { | |
4835 | int cpu; | |
4836 | ||
4837 | for_each_online_cpu(cpu) | |
9dfc6e68 | 4838 | per_cpu_ptr(s->cpu_slab, cpu)->stat[si] = 0; |
78eb00cc DR |
4839 | } |
4840 | ||
8ff12cfc CL |
4841 | #define STAT_ATTR(si, text) \ |
4842 | static ssize_t text##_show(struct kmem_cache *s, char *buf) \ | |
4843 | { \ | |
4844 | return show_stat(s, buf, si); \ | |
4845 | } \ | |
78eb00cc DR |
4846 | static ssize_t text##_store(struct kmem_cache *s, \ |
4847 | const char *buf, size_t length) \ | |
4848 | { \ | |
4849 | if (buf[0] != '0') \ | |
4850 | return -EINVAL; \ | |
4851 | clear_stat(s, si); \ | |
4852 | return length; \ | |
4853 | } \ | |
4854 | SLAB_ATTR(text); \ | |
8ff12cfc CL |
4855 | |
4856 | STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath); | |
4857 | STAT_ATTR(ALLOC_SLOWPATH, alloc_slowpath); | |
4858 | STAT_ATTR(FREE_FASTPATH, free_fastpath); | |
4859 | STAT_ATTR(FREE_SLOWPATH, free_slowpath); | |
4860 | STAT_ATTR(FREE_FROZEN, free_frozen); | |
4861 | STAT_ATTR(FREE_ADD_PARTIAL, free_add_partial); | |
4862 | STAT_ATTR(FREE_REMOVE_PARTIAL, free_remove_partial); | |
4863 | STAT_ATTR(ALLOC_FROM_PARTIAL, alloc_from_partial); | |
4864 | STAT_ATTR(ALLOC_SLAB, alloc_slab); | |
4865 | STAT_ATTR(ALLOC_REFILL, alloc_refill); | |
e36a2652 | 4866 | STAT_ATTR(ALLOC_NODE_MISMATCH, alloc_node_mismatch); |
8ff12cfc CL |
4867 | STAT_ATTR(FREE_SLAB, free_slab); |
4868 | STAT_ATTR(CPUSLAB_FLUSH, cpuslab_flush); | |
4869 | STAT_ATTR(DEACTIVATE_FULL, deactivate_full); | |
4870 | STAT_ATTR(DEACTIVATE_EMPTY, deactivate_empty); | |
4871 | STAT_ATTR(DEACTIVATE_TO_HEAD, deactivate_to_head); | |
4872 | STAT_ATTR(DEACTIVATE_TO_TAIL, deactivate_to_tail); | |
4873 | STAT_ATTR(DEACTIVATE_REMOTE_FREES, deactivate_remote_frees); | |
03e404af | 4874 | STAT_ATTR(DEACTIVATE_BYPASS, deactivate_bypass); |
65c3376a | 4875 | STAT_ATTR(ORDER_FALLBACK, order_fallback); |
b789ef51 CL |
4876 | STAT_ATTR(CMPXCHG_DOUBLE_CPU_FAIL, cmpxchg_double_cpu_fail); |
4877 | STAT_ATTR(CMPXCHG_DOUBLE_FAIL, cmpxchg_double_fail); | |
49e22585 CL |
4878 | STAT_ATTR(CPU_PARTIAL_ALLOC, cpu_partial_alloc); |
4879 | STAT_ATTR(CPU_PARTIAL_FREE, cpu_partial_free); | |
8028dcea AS |
4880 | STAT_ATTR(CPU_PARTIAL_NODE, cpu_partial_node); |
4881 | STAT_ATTR(CPU_PARTIAL_DRAIN, cpu_partial_drain); | |
8ff12cfc CL |
4882 | #endif |
4883 | ||
06428780 | 4884 | static struct attribute *slab_attrs[] = { |
81819f0f CL |
4885 | &slab_size_attr.attr, |
4886 | &object_size_attr.attr, | |
4887 | &objs_per_slab_attr.attr, | |
4888 | &order_attr.attr, | |
73d342b1 | 4889 | &min_partial_attr.attr, |
49e22585 | 4890 | &cpu_partial_attr.attr, |
81819f0f | 4891 | &objects_attr.attr, |
205ab99d | 4892 | &objects_partial_attr.attr, |
81819f0f CL |
4893 | &partial_attr.attr, |
4894 | &cpu_slabs_attr.attr, | |
4895 | &ctor_attr.attr, | |
81819f0f CL |
4896 | &aliases_attr.attr, |
4897 | &align_attr.attr, | |
81819f0f CL |
4898 | &hwcache_align_attr.attr, |
4899 | &reclaim_account_attr.attr, | |
4900 | &destroy_by_rcu_attr.attr, | |
a5a84755 | 4901 | &shrink_attr.attr, |
ab9a0f19 | 4902 | &reserved_attr.attr, |
49e22585 | 4903 | &slabs_cpu_partial_attr.attr, |
ab4d5ed5 | 4904 | #ifdef CONFIG_SLUB_DEBUG |
a5a84755 CL |
4905 | &total_objects_attr.attr, |
4906 | &slabs_attr.attr, | |
4907 | &sanity_checks_attr.attr, | |
4908 | &trace_attr.attr, | |
81819f0f CL |
4909 | &red_zone_attr.attr, |
4910 | &poison_attr.attr, | |
4911 | &store_user_attr.attr, | |
53e15af0 | 4912 | &validate_attr.attr, |
88a420e4 CL |
4913 | &alloc_calls_attr.attr, |
4914 | &free_calls_attr.attr, | |
ab4d5ed5 | 4915 | #endif |
81819f0f CL |
4916 | #ifdef CONFIG_ZONE_DMA |
4917 | &cache_dma_attr.attr, | |
4918 | #endif | |
4919 | #ifdef CONFIG_NUMA | |
9824601e | 4920 | &remote_node_defrag_ratio_attr.attr, |
8ff12cfc CL |
4921 | #endif |
4922 | #ifdef CONFIG_SLUB_STATS | |
4923 | &alloc_fastpath_attr.attr, | |
4924 | &alloc_slowpath_attr.attr, | |
4925 | &free_fastpath_attr.attr, | |
4926 | &free_slowpath_attr.attr, | |
4927 | &free_frozen_attr.attr, | |
4928 | &free_add_partial_attr.attr, | |
4929 | &free_remove_partial_attr.attr, | |
4930 | &alloc_from_partial_attr.attr, | |
4931 | &alloc_slab_attr.attr, | |
4932 | &alloc_refill_attr.attr, | |
e36a2652 | 4933 | &alloc_node_mismatch_attr.attr, |
8ff12cfc CL |
4934 | &free_slab_attr.attr, |
4935 | &cpuslab_flush_attr.attr, | |
4936 | &deactivate_full_attr.attr, | |
4937 | &deactivate_empty_attr.attr, | |
4938 | &deactivate_to_head_attr.attr, | |
4939 | &deactivate_to_tail_attr.attr, | |
4940 | &deactivate_remote_frees_attr.attr, | |
03e404af | 4941 | &deactivate_bypass_attr.attr, |
65c3376a | 4942 | &order_fallback_attr.attr, |
b789ef51 CL |
4943 | &cmpxchg_double_fail_attr.attr, |
4944 | &cmpxchg_double_cpu_fail_attr.attr, | |
49e22585 CL |
4945 | &cpu_partial_alloc_attr.attr, |
4946 | &cpu_partial_free_attr.attr, | |
8028dcea AS |
4947 | &cpu_partial_node_attr.attr, |
4948 | &cpu_partial_drain_attr.attr, | |
81819f0f | 4949 | #endif |
4c13dd3b DM |
4950 | #ifdef CONFIG_FAILSLAB |
4951 | &failslab_attr.attr, | |
4952 | #endif | |
4953 | ||
81819f0f CL |
4954 | NULL |
4955 | }; | |
4956 | ||
4957 | static struct attribute_group slab_attr_group = { | |
4958 | .attrs = slab_attrs, | |
4959 | }; | |
4960 | ||
4961 | static ssize_t slab_attr_show(struct kobject *kobj, | |
4962 | struct attribute *attr, | |
4963 | char *buf) | |
4964 | { | |
4965 | struct slab_attribute *attribute; | |
4966 | struct kmem_cache *s; | |
4967 | int err; | |
4968 | ||
4969 | attribute = to_slab_attr(attr); | |
4970 | s = to_slab(kobj); | |
4971 | ||
4972 | if (!attribute->show) | |
4973 | return -EIO; | |
4974 | ||
4975 | err = attribute->show(s, buf); | |
4976 | ||
4977 | return err; | |
4978 | } | |
4979 | ||
4980 | static ssize_t slab_attr_store(struct kobject *kobj, | |
4981 | struct attribute *attr, | |
4982 | const char *buf, size_t len) | |
4983 | { | |
4984 | struct slab_attribute *attribute; | |
4985 | struct kmem_cache *s; | |
4986 | int err; | |
4987 | ||
4988 | attribute = to_slab_attr(attr); | |
4989 | s = to_slab(kobj); | |
4990 | ||
4991 | if (!attribute->store) | |
4992 | return -EIO; | |
4993 | ||
4994 | err = attribute->store(s, buf, len); | |
107dab5c GC |
4995 | #ifdef CONFIG_MEMCG_KMEM |
4996 | if (slab_state >= FULL && err >= 0 && is_root_cache(s)) { | |
426589f5 | 4997 | struct kmem_cache *c; |
81819f0f | 4998 | |
107dab5c GC |
4999 | mutex_lock(&slab_mutex); |
5000 | if (s->max_attr_size < len) | |
5001 | s->max_attr_size = len; | |
5002 | ||
ebe945c2 GC |
5003 | /* |
5004 | * This is a best effort propagation, so this function's return | |
5005 | * value will be determined by the parent cache only. This is | |
5006 | * basically because not all attributes will have a well | |
5007 | * defined semantics for rollbacks - most of the actions will | |
5008 | * have permanent effects. | |
5009 | * | |
5010 | * Returning the error value of any of the children that fail | |
5011 | * is not 100 % defined, in the sense that users seeing the | |
5012 | * error code won't be able to know anything about the state of | |
5013 | * the cache. | |
5014 | * | |
5015 | * Only returning the error code for the parent cache at least | |
5016 | * has well defined semantics. The cache being written to | |
5017 | * directly either failed or succeeded, in which case we loop | |
5018 | * through the descendants with best-effort propagation. | |
5019 | */ | |
426589f5 VD |
5020 | for_each_memcg_cache(c, s) |
5021 | attribute->store(c, buf, len); | |
107dab5c GC |
5022 | mutex_unlock(&slab_mutex); |
5023 | } | |
5024 | #endif | |
81819f0f CL |
5025 | return err; |
5026 | } | |
5027 | ||
107dab5c GC |
5028 | static void memcg_propagate_slab_attrs(struct kmem_cache *s) |
5029 | { | |
5030 | #ifdef CONFIG_MEMCG_KMEM | |
5031 | int i; | |
5032 | char *buffer = NULL; | |
93030d83 | 5033 | struct kmem_cache *root_cache; |
107dab5c | 5034 | |
93030d83 | 5035 | if (is_root_cache(s)) |
107dab5c GC |
5036 | return; |
5037 | ||
f7ce3190 | 5038 | root_cache = s->memcg_params.root_cache; |
93030d83 | 5039 | |
107dab5c GC |
5040 | /* |
5041 | * This mean this cache had no attribute written. Therefore, no point | |
5042 | * in copying default values around | |
5043 | */ | |
93030d83 | 5044 | if (!root_cache->max_attr_size) |
107dab5c GC |
5045 | return; |
5046 | ||
5047 | for (i = 0; i < ARRAY_SIZE(slab_attrs); i++) { | |
5048 | char mbuf[64]; | |
5049 | char *buf; | |
5050 | struct slab_attribute *attr = to_slab_attr(slab_attrs[i]); | |
5051 | ||
5052 | if (!attr || !attr->store || !attr->show) | |
5053 | continue; | |
5054 | ||
5055 | /* | |
5056 | * It is really bad that we have to allocate here, so we will | |
5057 | * do it only as a fallback. If we actually allocate, though, | |
5058 | * we can just use the allocated buffer until the end. | |
5059 | * | |
5060 | * Most of the slub attributes will tend to be very small in | |
5061 | * size, but sysfs allows buffers up to a page, so they can | |
5062 | * theoretically happen. | |
5063 | */ | |
5064 | if (buffer) | |
5065 | buf = buffer; | |
93030d83 | 5066 | else if (root_cache->max_attr_size < ARRAY_SIZE(mbuf)) |
107dab5c GC |
5067 | buf = mbuf; |
5068 | else { | |
5069 | buffer = (char *) get_zeroed_page(GFP_KERNEL); | |
5070 | if (WARN_ON(!buffer)) | |
5071 | continue; | |
5072 | buf = buffer; | |
5073 | } | |
5074 | ||
93030d83 | 5075 | attr->show(root_cache, buf); |
107dab5c GC |
5076 | attr->store(s, buf, strlen(buf)); |
5077 | } | |
5078 | ||
5079 | if (buffer) | |
5080 | free_page((unsigned long)buffer); | |
5081 | #endif | |
5082 | } | |
5083 | ||
41a21285 CL |
5084 | static void kmem_cache_release(struct kobject *k) |
5085 | { | |
5086 | slab_kmem_cache_release(to_slab(k)); | |
5087 | } | |
5088 | ||
52cf25d0 | 5089 | static const struct sysfs_ops slab_sysfs_ops = { |
81819f0f CL |
5090 | .show = slab_attr_show, |
5091 | .store = slab_attr_store, | |
5092 | }; | |
5093 | ||
5094 | static struct kobj_type slab_ktype = { | |
5095 | .sysfs_ops = &slab_sysfs_ops, | |
41a21285 | 5096 | .release = kmem_cache_release, |
81819f0f CL |
5097 | }; |
5098 | ||
5099 | static int uevent_filter(struct kset *kset, struct kobject *kobj) | |
5100 | { | |
5101 | struct kobj_type *ktype = get_ktype(kobj); | |
5102 | ||
5103 | if (ktype == &slab_ktype) | |
5104 | return 1; | |
5105 | return 0; | |
5106 | } | |
5107 | ||
9cd43611 | 5108 | static const struct kset_uevent_ops slab_uevent_ops = { |
81819f0f CL |
5109 | .filter = uevent_filter, |
5110 | }; | |
5111 | ||
27c3a314 | 5112 | static struct kset *slab_kset; |
81819f0f | 5113 | |
9a41707b VD |
5114 | static inline struct kset *cache_kset(struct kmem_cache *s) |
5115 | { | |
5116 | #ifdef CONFIG_MEMCG_KMEM | |
5117 | if (!is_root_cache(s)) | |
f7ce3190 | 5118 | return s->memcg_params.root_cache->memcg_kset; |
9a41707b VD |
5119 | #endif |
5120 | return slab_kset; | |
5121 | } | |
5122 | ||
81819f0f CL |
5123 | #define ID_STR_LENGTH 64 |
5124 | ||
5125 | /* Create a unique string id for a slab cache: | |
6446faa2 CL |
5126 | * |
5127 | * Format :[flags-]size | |
81819f0f CL |
5128 | */ |
5129 | static char *create_unique_id(struct kmem_cache *s) | |
5130 | { | |
5131 | char *name = kmalloc(ID_STR_LENGTH, GFP_KERNEL); | |
5132 | char *p = name; | |
5133 | ||
5134 | BUG_ON(!name); | |
5135 | ||
5136 | *p++ = ':'; | |
5137 | /* | |
5138 | * First flags affecting slabcache operations. We will only | |
5139 | * get here for aliasable slabs so we do not need to support | |
5140 | * too many flags. The flags here must cover all flags that | |
5141 | * are matched during merging to guarantee that the id is | |
5142 | * unique. | |
5143 | */ | |
5144 | if (s->flags & SLAB_CACHE_DMA) | |
5145 | *p++ = 'd'; | |
5146 | if (s->flags & SLAB_RECLAIM_ACCOUNT) | |
5147 | *p++ = 'a'; | |
5148 | if (s->flags & SLAB_DEBUG_FREE) | |
5149 | *p++ = 'F'; | |
5a896d9e VN |
5150 | if (!(s->flags & SLAB_NOTRACK)) |
5151 | *p++ = 't'; | |
81819f0f CL |
5152 | if (p != name + 1) |
5153 | *p++ = '-'; | |
5154 | p += sprintf(p, "%07d", s->size); | |
2633d7a0 | 5155 | |
81819f0f CL |
5156 | BUG_ON(p > name + ID_STR_LENGTH - 1); |
5157 | return name; | |
5158 | } | |
5159 | ||
5160 | static int sysfs_slab_add(struct kmem_cache *s) | |
5161 | { | |
5162 | int err; | |
5163 | const char *name; | |
45530c44 | 5164 | int unmergeable = slab_unmergeable(s); |
81819f0f | 5165 | |
81819f0f CL |
5166 | if (unmergeable) { |
5167 | /* | |
5168 | * Slabcache can never be merged so we can use the name proper. | |
5169 | * This is typically the case for debug situations. In that | |
5170 | * case we can catch duplicate names easily. | |
5171 | */ | |
27c3a314 | 5172 | sysfs_remove_link(&slab_kset->kobj, s->name); |
81819f0f CL |
5173 | name = s->name; |
5174 | } else { | |
5175 | /* | |
5176 | * Create a unique name for the slab as a target | |
5177 | * for the symlinks. | |
5178 | */ | |
5179 | name = create_unique_id(s); | |
5180 | } | |
5181 | ||
9a41707b | 5182 | s->kobj.kset = cache_kset(s); |
26e4f205 | 5183 | err = kobject_init_and_add(&s->kobj, &slab_ktype, NULL, "%s", name); |
54b6a731 DJ |
5184 | if (err) |
5185 | goto out_put_kobj; | |
81819f0f CL |
5186 | |
5187 | err = sysfs_create_group(&s->kobj, &slab_attr_group); | |
54b6a731 DJ |
5188 | if (err) |
5189 | goto out_del_kobj; | |
9a41707b VD |
5190 | |
5191 | #ifdef CONFIG_MEMCG_KMEM | |
5192 | if (is_root_cache(s)) { | |
5193 | s->memcg_kset = kset_create_and_add("cgroup", NULL, &s->kobj); | |
5194 | if (!s->memcg_kset) { | |
54b6a731 DJ |
5195 | err = -ENOMEM; |
5196 | goto out_del_kobj; | |
9a41707b VD |
5197 | } |
5198 | } | |
5199 | #endif | |
5200 | ||
81819f0f CL |
5201 | kobject_uevent(&s->kobj, KOBJ_ADD); |
5202 | if (!unmergeable) { | |
5203 | /* Setup first alias */ | |
5204 | sysfs_slab_alias(s, s->name); | |
81819f0f | 5205 | } |
54b6a731 DJ |
5206 | out: |
5207 | if (!unmergeable) | |
5208 | kfree(name); | |
5209 | return err; | |
5210 | out_del_kobj: | |
5211 | kobject_del(&s->kobj); | |
5212 | out_put_kobj: | |
5213 | kobject_put(&s->kobj); | |
5214 | goto out; | |
81819f0f CL |
5215 | } |
5216 | ||
41a21285 | 5217 | void sysfs_slab_remove(struct kmem_cache *s) |
81819f0f | 5218 | { |
97d06609 | 5219 | if (slab_state < FULL) |
2bce6485 CL |
5220 | /* |
5221 | * Sysfs has not been setup yet so no need to remove the | |
5222 | * cache from sysfs. | |
5223 | */ | |
5224 | return; | |
5225 | ||
9a41707b VD |
5226 | #ifdef CONFIG_MEMCG_KMEM |
5227 | kset_unregister(s->memcg_kset); | |
5228 | #endif | |
81819f0f CL |
5229 | kobject_uevent(&s->kobj, KOBJ_REMOVE); |
5230 | kobject_del(&s->kobj); | |
151c602f | 5231 | kobject_put(&s->kobj); |
81819f0f CL |
5232 | } |
5233 | ||
5234 | /* | |
5235 | * Need to buffer aliases during bootup until sysfs becomes | |
9f6c708e | 5236 | * available lest we lose that information. |
81819f0f CL |
5237 | */ |
5238 | struct saved_alias { | |
5239 | struct kmem_cache *s; | |
5240 | const char *name; | |
5241 | struct saved_alias *next; | |
5242 | }; | |
5243 | ||
5af328a5 | 5244 | static struct saved_alias *alias_list; |
81819f0f CL |
5245 | |
5246 | static int sysfs_slab_alias(struct kmem_cache *s, const char *name) | |
5247 | { | |
5248 | struct saved_alias *al; | |
5249 | ||
97d06609 | 5250 | if (slab_state == FULL) { |
81819f0f CL |
5251 | /* |
5252 | * If we have a leftover link then remove it. | |
5253 | */ | |
27c3a314 GKH |
5254 | sysfs_remove_link(&slab_kset->kobj, name); |
5255 | return sysfs_create_link(&slab_kset->kobj, &s->kobj, name); | |
81819f0f CL |
5256 | } |
5257 | ||
5258 | al = kmalloc(sizeof(struct saved_alias), GFP_KERNEL); | |
5259 | if (!al) | |
5260 | return -ENOMEM; | |
5261 | ||
5262 | al->s = s; | |
5263 | al->name = name; | |
5264 | al->next = alias_list; | |
5265 | alias_list = al; | |
5266 | return 0; | |
5267 | } | |
5268 | ||
5269 | static int __init slab_sysfs_init(void) | |
5270 | { | |
5b95a4ac | 5271 | struct kmem_cache *s; |
81819f0f CL |
5272 | int err; |
5273 | ||
18004c5d | 5274 | mutex_lock(&slab_mutex); |
2bce6485 | 5275 | |
0ff21e46 | 5276 | slab_kset = kset_create_and_add("slab", &slab_uevent_ops, kernel_kobj); |
27c3a314 | 5277 | if (!slab_kset) { |
18004c5d | 5278 | mutex_unlock(&slab_mutex); |
f9f58285 | 5279 | pr_err("Cannot register slab subsystem.\n"); |
81819f0f CL |
5280 | return -ENOSYS; |
5281 | } | |
5282 | ||
97d06609 | 5283 | slab_state = FULL; |
26a7bd03 | 5284 | |
5b95a4ac | 5285 | list_for_each_entry(s, &slab_caches, list) { |
26a7bd03 | 5286 | err = sysfs_slab_add(s); |
5d540fb7 | 5287 | if (err) |
f9f58285 FF |
5288 | pr_err("SLUB: Unable to add boot slab %s to sysfs\n", |
5289 | s->name); | |
26a7bd03 | 5290 | } |
81819f0f CL |
5291 | |
5292 | while (alias_list) { | |
5293 | struct saved_alias *al = alias_list; | |
5294 | ||
5295 | alias_list = alias_list->next; | |
5296 | err = sysfs_slab_alias(al->s, al->name); | |
5d540fb7 | 5297 | if (err) |
f9f58285 FF |
5298 | pr_err("SLUB: Unable to add boot slab alias %s to sysfs\n", |
5299 | al->name); | |
81819f0f CL |
5300 | kfree(al); |
5301 | } | |
5302 | ||
18004c5d | 5303 | mutex_unlock(&slab_mutex); |
81819f0f CL |
5304 | resiliency_test(); |
5305 | return 0; | |
5306 | } | |
5307 | ||
5308 | __initcall(slab_sysfs_init); | |
ab4d5ed5 | 5309 | #endif /* CONFIG_SYSFS */ |
57ed3eda PE |
5310 | |
5311 | /* | |
5312 | * The /proc/slabinfo ABI | |
5313 | */ | |
158a9624 | 5314 | #ifdef CONFIG_SLABINFO |
0d7561c6 | 5315 | void get_slabinfo(struct kmem_cache *s, struct slabinfo *sinfo) |
57ed3eda | 5316 | { |
57ed3eda | 5317 | unsigned long nr_slabs = 0; |
205ab99d CL |
5318 | unsigned long nr_objs = 0; |
5319 | unsigned long nr_free = 0; | |
57ed3eda | 5320 | int node; |
fa45dc25 | 5321 | struct kmem_cache_node *n; |
57ed3eda | 5322 | |
fa45dc25 | 5323 | for_each_kmem_cache_node(s, node, n) { |
c17fd13e WL |
5324 | nr_slabs += node_nr_slabs(n); |
5325 | nr_objs += node_nr_objs(n); | |
205ab99d | 5326 | nr_free += count_partial(n, count_free); |
57ed3eda PE |
5327 | } |
5328 | ||
0d7561c6 GC |
5329 | sinfo->active_objs = nr_objs - nr_free; |
5330 | sinfo->num_objs = nr_objs; | |
5331 | sinfo->active_slabs = nr_slabs; | |
5332 | sinfo->num_slabs = nr_slabs; | |
5333 | sinfo->objects_per_slab = oo_objects(s->oo); | |
5334 | sinfo->cache_order = oo_order(s->oo); | |
57ed3eda PE |
5335 | } |
5336 | ||
0d7561c6 | 5337 | void slabinfo_show_stats(struct seq_file *m, struct kmem_cache *s) |
7b3c3a50 | 5338 | { |
7b3c3a50 AD |
5339 | } |
5340 | ||
b7454ad3 GC |
5341 | ssize_t slabinfo_write(struct file *file, const char __user *buffer, |
5342 | size_t count, loff_t *ppos) | |
7b3c3a50 | 5343 | { |
b7454ad3 | 5344 | return -EIO; |
7b3c3a50 | 5345 | } |
158a9624 | 5346 | #endif /* CONFIG_SLABINFO */ |