Commit | Line | Data |
---|---|---|
81819f0f CL |
1 | /* |
2 | * SLUB: A slab allocator that limits cache line use instead of queuing | |
3 | * objects in per cpu and per node lists. | |
4 | * | |
881db7fb CL |
5 | * The allocator synchronizes using per slab locks or atomic operatios |
6 | * and only uses a centralized lock to manage a pool of partial slabs. | |
81819f0f | 7 | * |
cde53535 | 8 | * (C) 2007 SGI, Christoph Lameter |
881db7fb | 9 | * (C) 2011 Linux Foundation, Christoph Lameter |
81819f0f CL |
10 | */ |
11 | ||
12 | #include <linux/mm.h> | |
1eb5ac64 | 13 | #include <linux/swap.h> /* struct reclaim_state */ |
81819f0f CL |
14 | #include <linux/module.h> |
15 | #include <linux/bit_spinlock.h> | |
16 | #include <linux/interrupt.h> | |
17 | #include <linux/bitops.h> | |
18 | #include <linux/slab.h> | |
97d06609 | 19 | #include "slab.h" |
7b3c3a50 | 20 | #include <linux/proc_fs.h> |
3ac38faa | 21 | #include <linux/notifier.h> |
81819f0f | 22 | #include <linux/seq_file.h> |
5a896d9e | 23 | #include <linux/kmemcheck.h> |
81819f0f CL |
24 | #include <linux/cpu.h> |
25 | #include <linux/cpuset.h> | |
26 | #include <linux/mempolicy.h> | |
27 | #include <linux/ctype.h> | |
3ac7fe5a | 28 | #include <linux/debugobjects.h> |
81819f0f | 29 | #include <linux/kallsyms.h> |
b9049e23 | 30 | #include <linux/memory.h> |
f8bd2258 | 31 | #include <linux/math64.h> |
773ff60e | 32 | #include <linux/fault-inject.h> |
bfa71457 | 33 | #include <linux/stacktrace.h> |
4de900b4 | 34 | #include <linux/prefetch.h> |
2633d7a0 | 35 | #include <linux/memcontrol.h> |
81819f0f | 36 | |
4a92379b RK |
37 | #include <trace/events/kmem.h> |
38 | ||
072bb0aa MG |
39 | #include "internal.h" |
40 | ||
81819f0f CL |
41 | /* |
42 | * Lock order: | |
18004c5d | 43 | * 1. slab_mutex (Global Mutex) |
881db7fb CL |
44 | * 2. node->list_lock |
45 | * 3. slab_lock(page) (Only on some arches and for debugging) | |
81819f0f | 46 | * |
18004c5d | 47 | * slab_mutex |
881db7fb | 48 | * |
18004c5d | 49 | * The role of the slab_mutex is to protect the list of all the slabs |
881db7fb CL |
50 | * and to synchronize major metadata changes to slab cache structures. |
51 | * | |
52 | * The slab_lock is only used for debugging and on arches that do not | |
53 | * have the ability to do a cmpxchg_double. It only protects the second | |
54 | * double word in the page struct. Meaning | |
55 | * A. page->freelist -> List of object free in a page | |
56 | * B. page->counters -> Counters of objects | |
57 | * C. page->frozen -> frozen state | |
58 | * | |
59 | * If a slab is frozen then it is exempt from list management. It is not | |
60 | * on any list. The processor that froze the slab is the one who can | |
61 | * perform list operations on the page. Other processors may put objects | |
62 | * onto the freelist but the processor that froze the slab is the only | |
63 | * one that can retrieve the objects from the page's freelist. | |
81819f0f CL |
64 | * |
65 | * The list_lock protects the partial and full list on each node and | |
66 | * the partial slab counter. If taken then no new slabs may be added or | |
67 | * removed from the lists nor make the number of partial slabs be modified. | |
68 | * (Note that the total number of slabs is an atomic value that may be | |
69 | * modified without taking the list lock). | |
70 | * | |
71 | * The list_lock is a centralized lock and thus we avoid taking it as | |
72 | * much as possible. As long as SLUB does not have to handle partial | |
73 | * slabs, operations can continue without any centralized lock. F.e. | |
74 | * allocating a long series of objects that fill up slabs does not require | |
75 | * the list lock. | |
81819f0f CL |
76 | * Interrupts are disabled during allocation and deallocation in order to |
77 | * make the slab allocator safe to use in the context of an irq. In addition | |
78 | * interrupts are disabled to ensure that the processor does not change | |
79 | * while handling per_cpu slabs, due to kernel preemption. | |
80 | * | |
81 | * SLUB assigns one slab for allocation to each processor. | |
82 | * Allocations only occur from these slabs called cpu slabs. | |
83 | * | |
672bba3a CL |
84 | * Slabs with free elements are kept on a partial list and during regular |
85 | * operations no list for full slabs is used. If an object in a full slab is | |
81819f0f | 86 | * freed then the slab will show up again on the partial lists. |
672bba3a CL |
87 | * We track full slabs for debugging purposes though because otherwise we |
88 | * cannot scan all objects. | |
81819f0f CL |
89 | * |
90 | * Slabs are freed when they become empty. Teardown and setup is | |
91 | * minimal so we rely on the page allocators per cpu caches for | |
92 | * fast frees and allocs. | |
93 | * | |
94 | * Overloading of page flags that are otherwise used for LRU management. | |
95 | * | |
4b6f0750 CL |
96 | * PageActive The slab is frozen and exempt from list processing. |
97 | * This means that the slab is dedicated to a purpose | |
98 | * such as satisfying allocations for a specific | |
99 | * processor. Objects may be freed in the slab while | |
100 | * it is frozen but slab_free will then skip the usual | |
101 | * list operations. It is up to the processor holding | |
102 | * the slab to integrate the slab into the slab lists | |
103 | * when the slab is no longer needed. | |
104 | * | |
105 | * One use of this flag is to mark slabs that are | |
106 | * used for allocations. Then such a slab becomes a cpu | |
107 | * slab. The cpu slab may be equipped with an additional | |
dfb4f096 | 108 | * freelist that allows lockless access to |
894b8788 CL |
109 | * free objects in addition to the regular freelist |
110 | * that requires the slab lock. | |
81819f0f CL |
111 | * |
112 | * PageError Slab requires special handling due to debug | |
113 | * options set. This moves slab handling out of | |
894b8788 | 114 | * the fast path and disables lockless freelists. |
81819f0f CL |
115 | */ |
116 | ||
af537b0a CL |
117 | static inline int kmem_cache_debug(struct kmem_cache *s) |
118 | { | |
5577bd8a | 119 | #ifdef CONFIG_SLUB_DEBUG |
af537b0a | 120 | return unlikely(s->flags & SLAB_DEBUG_FLAGS); |
5577bd8a | 121 | #else |
af537b0a | 122 | return 0; |
5577bd8a | 123 | #endif |
af537b0a | 124 | } |
5577bd8a | 125 | |
345c905d JK |
126 | static inline bool kmem_cache_has_cpu_partial(struct kmem_cache *s) |
127 | { | |
128 | #ifdef CONFIG_SLUB_CPU_PARTIAL | |
129 | return !kmem_cache_debug(s); | |
130 | #else | |
131 | return false; | |
132 | #endif | |
133 | } | |
134 | ||
81819f0f CL |
135 | /* |
136 | * Issues still to be resolved: | |
137 | * | |
81819f0f CL |
138 | * - Support PAGE_ALLOC_DEBUG. Should be easy to do. |
139 | * | |
81819f0f CL |
140 | * - Variable sizing of the per node arrays |
141 | */ | |
142 | ||
143 | /* Enable to test recovery from slab corruption on boot */ | |
144 | #undef SLUB_RESILIENCY_TEST | |
145 | ||
b789ef51 CL |
146 | /* Enable to log cmpxchg failures */ |
147 | #undef SLUB_DEBUG_CMPXCHG | |
148 | ||
2086d26a CL |
149 | /* |
150 | * Mininum number of partial slabs. These will be left on the partial | |
151 | * lists even if they are empty. kmem_cache_shrink may reclaim them. | |
152 | */ | |
76be8950 | 153 | #define MIN_PARTIAL 5 |
e95eed57 | 154 | |
2086d26a CL |
155 | /* |
156 | * Maximum number of desirable partial slabs. | |
157 | * The existence of more partial slabs makes kmem_cache_shrink | |
721ae22a | 158 | * sort the partial list by the number of objects in use. |
2086d26a CL |
159 | */ |
160 | #define MAX_PARTIAL 10 | |
161 | ||
81819f0f CL |
162 | #define DEBUG_DEFAULT_FLAGS (SLAB_DEBUG_FREE | SLAB_RED_ZONE | \ |
163 | SLAB_POISON | SLAB_STORE_USER) | |
672bba3a | 164 | |
fa5ec8a1 | 165 | /* |
3de47213 DR |
166 | * Debugging flags that require metadata to be stored in the slab. These get |
167 | * disabled when slub_debug=O is used and a cache's min order increases with | |
168 | * metadata. | |
fa5ec8a1 | 169 | */ |
3de47213 | 170 | #define DEBUG_METADATA_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER) |
fa5ec8a1 | 171 | |
81819f0f CL |
172 | /* |
173 | * Set of flags that will prevent slab merging | |
174 | */ | |
175 | #define SLUB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \ | |
4c13dd3b DM |
176 | SLAB_TRACE | SLAB_DESTROY_BY_RCU | SLAB_NOLEAKTRACE | \ |
177 | SLAB_FAILSLAB) | |
81819f0f CL |
178 | |
179 | #define SLUB_MERGE_SAME (SLAB_DEBUG_FREE | SLAB_RECLAIM_ACCOUNT | \ | |
5a896d9e | 180 | SLAB_CACHE_DMA | SLAB_NOTRACK) |
81819f0f | 181 | |
210b5c06 CG |
182 | #define OO_SHIFT 16 |
183 | #define OO_MASK ((1 << OO_SHIFT) - 1) | |
50d5c41c | 184 | #define MAX_OBJS_PER_PAGE 32767 /* since page.objects is u15 */ |
210b5c06 | 185 | |
81819f0f | 186 | /* Internal SLUB flags */ |
f90ec390 | 187 | #define __OBJECT_POISON 0x80000000UL /* Poison object */ |
b789ef51 | 188 | #define __CMPXCHG_DOUBLE 0x40000000UL /* Use cmpxchg_double */ |
81819f0f | 189 | |
81819f0f CL |
190 | #ifdef CONFIG_SMP |
191 | static struct notifier_block slab_notifier; | |
192 | #endif | |
193 | ||
02cbc874 CL |
194 | /* |
195 | * Tracking user of a slab. | |
196 | */ | |
d6543e39 | 197 | #define TRACK_ADDRS_COUNT 16 |
02cbc874 | 198 | struct track { |
ce71e27c | 199 | unsigned long addr; /* Called from address */ |
d6543e39 BG |
200 | #ifdef CONFIG_STACKTRACE |
201 | unsigned long addrs[TRACK_ADDRS_COUNT]; /* Called from address */ | |
202 | #endif | |
02cbc874 CL |
203 | int cpu; /* Was running on cpu */ |
204 | int pid; /* Pid context */ | |
205 | unsigned long when; /* When did the operation occur */ | |
206 | }; | |
207 | ||
208 | enum track_item { TRACK_ALLOC, TRACK_FREE }; | |
209 | ||
ab4d5ed5 | 210 | #ifdef CONFIG_SYSFS |
81819f0f CL |
211 | static int sysfs_slab_add(struct kmem_cache *); |
212 | static int sysfs_slab_alias(struct kmem_cache *, const char *); | |
107dab5c | 213 | static void memcg_propagate_slab_attrs(struct kmem_cache *s); |
81819f0f | 214 | #else |
0c710013 CL |
215 | static inline int sysfs_slab_add(struct kmem_cache *s) { return 0; } |
216 | static inline int sysfs_slab_alias(struct kmem_cache *s, const char *p) | |
217 | { return 0; } | |
107dab5c | 218 | static inline void memcg_propagate_slab_attrs(struct kmem_cache *s) { } |
81819f0f CL |
219 | #endif |
220 | ||
4fdccdfb | 221 | static inline void stat(const struct kmem_cache *s, enum stat_item si) |
8ff12cfc CL |
222 | { |
223 | #ifdef CONFIG_SLUB_STATS | |
88da03a6 CL |
224 | /* |
225 | * The rmw is racy on a preemptible kernel but this is acceptable, so | |
226 | * avoid this_cpu_add()'s irq-disable overhead. | |
227 | */ | |
228 | raw_cpu_inc(s->cpu_slab->stat[si]); | |
8ff12cfc CL |
229 | #endif |
230 | } | |
231 | ||
81819f0f CL |
232 | /******************************************************************** |
233 | * Core slab cache functions | |
234 | *******************************************************************/ | |
235 | ||
6446faa2 | 236 | /* Verify that a pointer has an address that is valid within a slab page */ |
02cbc874 CL |
237 | static inline int check_valid_pointer(struct kmem_cache *s, |
238 | struct page *page, const void *object) | |
239 | { | |
240 | void *base; | |
241 | ||
a973e9dd | 242 | if (!object) |
02cbc874 CL |
243 | return 1; |
244 | ||
a973e9dd | 245 | base = page_address(page); |
39b26464 | 246 | if (object < base || object >= base + page->objects * s->size || |
02cbc874 CL |
247 | (object - base) % s->size) { |
248 | return 0; | |
249 | } | |
250 | ||
251 | return 1; | |
252 | } | |
253 | ||
7656c72b CL |
254 | static inline void *get_freepointer(struct kmem_cache *s, void *object) |
255 | { | |
256 | return *(void **)(object + s->offset); | |
257 | } | |
258 | ||
0ad9500e ED |
259 | static void prefetch_freepointer(const struct kmem_cache *s, void *object) |
260 | { | |
261 | prefetch(object + s->offset); | |
262 | } | |
263 | ||
1393d9a1 CL |
264 | static inline void *get_freepointer_safe(struct kmem_cache *s, void *object) |
265 | { | |
266 | void *p; | |
267 | ||
268 | #ifdef CONFIG_DEBUG_PAGEALLOC | |
269 | probe_kernel_read(&p, (void **)(object + s->offset), sizeof(p)); | |
270 | #else | |
271 | p = get_freepointer(s, object); | |
272 | #endif | |
273 | return p; | |
274 | } | |
275 | ||
7656c72b CL |
276 | static inline void set_freepointer(struct kmem_cache *s, void *object, void *fp) |
277 | { | |
278 | *(void **)(object + s->offset) = fp; | |
279 | } | |
280 | ||
281 | /* Loop over all objects in a slab */ | |
224a88be CL |
282 | #define for_each_object(__p, __s, __addr, __objects) \ |
283 | for (__p = (__addr); __p < (__addr) + (__objects) * (__s)->size;\ | |
7656c72b CL |
284 | __p += (__s)->size) |
285 | ||
7656c72b CL |
286 | /* Determine object index from a given position */ |
287 | static inline int slab_index(void *p, struct kmem_cache *s, void *addr) | |
288 | { | |
289 | return (p - addr) / s->size; | |
290 | } | |
291 | ||
d71f606f MK |
292 | static inline size_t slab_ksize(const struct kmem_cache *s) |
293 | { | |
294 | #ifdef CONFIG_SLUB_DEBUG | |
295 | /* | |
296 | * Debugging requires use of the padding between object | |
297 | * and whatever may come after it. | |
298 | */ | |
299 | if (s->flags & (SLAB_RED_ZONE | SLAB_POISON)) | |
3b0efdfa | 300 | return s->object_size; |
d71f606f MK |
301 | |
302 | #endif | |
303 | /* | |
304 | * If we have the need to store the freelist pointer | |
305 | * back there or track user information then we can | |
306 | * only use the space before that information. | |
307 | */ | |
308 | if (s->flags & (SLAB_DESTROY_BY_RCU | SLAB_STORE_USER)) | |
309 | return s->inuse; | |
310 | /* | |
311 | * Else we can use all the padding etc for the allocation | |
312 | */ | |
313 | return s->size; | |
314 | } | |
315 | ||
ab9a0f19 LJ |
316 | static inline int order_objects(int order, unsigned long size, int reserved) |
317 | { | |
318 | return ((PAGE_SIZE << order) - reserved) / size; | |
319 | } | |
320 | ||
834f3d11 | 321 | static inline struct kmem_cache_order_objects oo_make(int order, |
ab9a0f19 | 322 | unsigned long size, int reserved) |
834f3d11 CL |
323 | { |
324 | struct kmem_cache_order_objects x = { | |
ab9a0f19 | 325 | (order << OO_SHIFT) + order_objects(order, size, reserved) |
834f3d11 CL |
326 | }; |
327 | ||
328 | return x; | |
329 | } | |
330 | ||
331 | static inline int oo_order(struct kmem_cache_order_objects x) | |
332 | { | |
210b5c06 | 333 | return x.x >> OO_SHIFT; |
834f3d11 CL |
334 | } |
335 | ||
336 | static inline int oo_objects(struct kmem_cache_order_objects x) | |
337 | { | |
210b5c06 | 338 | return x.x & OO_MASK; |
834f3d11 CL |
339 | } |
340 | ||
881db7fb CL |
341 | /* |
342 | * Per slab locking using the pagelock | |
343 | */ | |
344 | static __always_inline void slab_lock(struct page *page) | |
345 | { | |
346 | bit_spin_lock(PG_locked, &page->flags); | |
347 | } | |
348 | ||
349 | static __always_inline void slab_unlock(struct page *page) | |
350 | { | |
351 | __bit_spin_unlock(PG_locked, &page->flags); | |
352 | } | |
353 | ||
a0320865 DH |
354 | static inline void set_page_slub_counters(struct page *page, unsigned long counters_new) |
355 | { | |
356 | struct page tmp; | |
357 | tmp.counters = counters_new; | |
358 | /* | |
359 | * page->counters can cover frozen/inuse/objects as well | |
360 | * as page->_count. If we assign to ->counters directly | |
361 | * we run the risk of losing updates to page->_count, so | |
362 | * be careful and only assign to the fields we need. | |
363 | */ | |
364 | page->frozen = tmp.frozen; | |
365 | page->inuse = tmp.inuse; | |
366 | page->objects = tmp.objects; | |
367 | } | |
368 | ||
1d07171c CL |
369 | /* Interrupts must be disabled (for the fallback code to work right) */ |
370 | static inline bool __cmpxchg_double_slab(struct kmem_cache *s, struct page *page, | |
371 | void *freelist_old, unsigned long counters_old, | |
372 | void *freelist_new, unsigned long counters_new, | |
373 | const char *n) | |
374 | { | |
375 | VM_BUG_ON(!irqs_disabled()); | |
2565409f HC |
376 | #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \ |
377 | defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE) | |
1d07171c | 378 | if (s->flags & __CMPXCHG_DOUBLE) { |
cdcd6298 | 379 | if (cmpxchg_double(&page->freelist, &page->counters, |
1d07171c CL |
380 | freelist_old, counters_old, |
381 | freelist_new, counters_new)) | |
382 | return 1; | |
383 | } else | |
384 | #endif | |
385 | { | |
386 | slab_lock(page); | |
d0e0ac97 CG |
387 | if (page->freelist == freelist_old && |
388 | page->counters == counters_old) { | |
1d07171c | 389 | page->freelist = freelist_new; |
a0320865 | 390 | set_page_slub_counters(page, counters_new); |
1d07171c CL |
391 | slab_unlock(page); |
392 | return 1; | |
393 | } | |
394 | slab_unlock(page); | |
395 | } | |
396 | ||
397 | cpu_relax(); | |
398 | stat(s, CMPXCHG_DOUBLE_FAIL); | |
399 | ||
400 | #ifdef SLUB_DEBUG_CMPXCHG | |
f9f58285 | 401 | pr_info("%s %s: cmpxchg double redo ", n, s->name); |
1d07171c CL |
402 | #endif |
403 | ||
404 | return 0; | |
405 | } | |
406 | ||
b789ef51 CL |
407 | static inline bool cmpxchg_double_slab(struct kmem_cache *s, struct page *page, |
408 | void *freelist_old, unsigned long counters_old, | |
409 | void *freelist_new, unsigned long counters_new, | |
410 | const char *n) | |
411 | { | |
2565409f HC |
412 | #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \ |
413 | defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE) | |
b789ef51 | 414 | if (s->flags & __CMPXCHG_DOUBLE) { |
cdcd6298 | 415 | if (cmpxchg_double(&page->freelist, &page->counters, |
b789ef51 CL |
416 | freelist_old, counters_old, |
417 | freelist_new, counters_new)) | |
418 | return 1; | |
419 | } else | |
420 | #endif | |
421 | { | |
1d07171c CL |
422 | unsigned long flags; |
423 | ||
424 | local_irq_save(flags); | |
881db7fb | 425 | slab_lock(page); |
d0e0ac97 CG |
426 | if (page->freelist == freelist_old && |
427 | page->counters == counters_old) { | |
b789ef51 | 428 | page->freelist = freelist_new; |
a0320865 | 429 | set_page_slub_counters(page, counters_new); |
881db7fb | 430 | slab_unlock(page); |
1d07171c | 431 | local_irq_restore(flags); |
b789ef51 CL |
432 | return 1; |
433 | } | |
881db7fb | 434 | slab_unlock(page); |
1d07171c | 435 | local_irq_restore(flags); |
b789ef51 CL |
436 | } |
437 | ||
438 | cpu_relax(); | |
439 | stat(s, CMPXCHG_DOUBLE_FAIL); | |
440 | ||
441 | #ifdef SLUB_DEBUG_CMPXCHG | |
f9f58285 | 442 | pr_info("%s %s: cmpxchg double redo ", n, s->name); |
b789ef51 CL |
443 | #endif |
444 | ||
445 | return 0; | |
446 | } | |
447 | ||
41ecc55b | 448 | #ifdef CONFIG_SLUB_DEBUG |
5f80b13a CL |
449 | /* |
450 | * Determine a map of object in use on a page. | |
451 | * | |
881db7fb | 452 | * Node listlock must be held to guarantee that the page does |
5f80b13a CL |
453 | * not vanish from under us. |
454 | */ | |
455 | static void get_map(struct kmem_cache *s, struct page *page, unsigned long *map) | |
456 | { | |
457 | void *p; | |
458 | void *addr = page_address(page); | |
459 | ||
460 | for (p = page->freelist; p; p = get_freepointer(s, p)) | |
461 | set_bit(slab_index(p, s, addr), map); | |
462 | } | |
463 | ||
41ecc55b CL |
464 | /* |
465 | * Debug settings: | |
466 | */ | |
f0630fff CL |
467 | #ifdef CONFIG_SLUB_DEBUG_ON |
468 | static int slub_debug = DEBUG_DEFAULT_FLAGS; | |
469 | #else | |
41ecc55b | 470 | static int slub_debug; |
f0630fff | 471 | #endif |
41ecc55b CL |
472 | |
473 | static char *slub_debug_slabs; | |
fa5ec8a1 | 474 | static int disable_higher_order_debug; |
41ecc55b | 475 | |
81819f0f CL |
476 | /* |
477 | * Object debugging | |
478 | */ | |
479 | static void print_section(char *text, u8 *addr, unsigned int length) | |
480 | { | |
ffc79d28 SAS |
481 | print_hex_dump(KERN_ERR, text, DUMP_PREFIX_ADDRESS, 16, 1, addr, |
482 | length, 1); | |
81819f0f CL |
483 | } |
484 | ||
81819f0f CL |
485 | static struct track *get_track(struct kmem_cache *s, void *object, |
486 | enum track_item alloc) | |
487 | { | |
488 | struct track *p; | |
489 | ||
490 | if (s->offset) | |
491 | p = object + s->offset + sizeof(void *); | |
492 | else | |
493 | p = object + s->inuse; | |
494 | ||
495 | return p + alloc; | |
496 | } | |
497 | ||
498 | static void set_track(struct kmem_cache *s, void *object, | |
ce71e27c | 499 | enum track_item alloc, unsigned long addr) |
81819f0f | 500 | { |
1a00df4a | 501 | struct track *p = get_track(s, object, alloc); |
81819f0f | 502 | |
81819f0f | 503 | if (addr) { |
d6543e39 BG |
504 | #ifdef CONFIG_STACKTRACE |
505 | struct stack_trace trace; | |
506 | int i; | |
507 | ||
508 | trace.nr_entries = 0; | |
509 | trace.max_entries = TRACK_ADDRS_COUNT; | |
510 | trace.entries = p->addrs; | |
511 | trace.skip = 3; | |
512 | save_stack_trace(&trace); | |
513 | ||
514 | /* See rant in lockdep.c */ | |
515 | if (trace.nr_entries != 0 && | |
516 | trace.entries[trace.nr_entries - 1] == ULONG_MAX) | |
517 | trace.nr_entries--; | |
518 | ||
519 | for (i = trace.nr_entries; i < TRACK_ADDRS_COUNT; i++) | |
520 | p->addrs[i] = 0; | |
521 | #endif | |
81819f0f CL |
522 | p->addr = addr; |
523 | p->cpu = smp_processor_id(); | |
88e4ccf2 | 524 | p->pid = current->pid; |
81819f0f CL |
525 | p->when = jiffies; |
526 | } else | |
527 | memset(p, 0, sizeof(struct track)); | |
528 | } | |
529 | ||
81819f0f CL |
530 | static void init_tracking(struct kmem_cache *s, void *object) |
531 | { | |
24922684 CL |
532 | if (!(s->flags & SLAB_STORE_USER)) |
533 | return; | |
534 | ||
ce71e27c EGM |
535 | set_track(s, object, TRACK_FREE, 0UL); |
536 | set_track(s, object, TRACK_ALLOC, 0UL); | |
81819f0f CL |
537 | } |
538 | ||
539 | static void print_track(const char *s, struct track *t) | |
540 | { | |
541 | if (!t->addr) | |
542 | return; | |
543 | ||
f9f58285 FF |
544 | pr_err("INFO: %s in %pS age=%lu cpu=%u pid=%d\n", |
545 | s, (void *)t->addr, jiffies - t->when, t->cpu, t->pid); | |
d6543e39 BG |
546 | #ifdef CONFIG_STACKTRACE |
547 | { | |
548 | int i; | |
549 | for (i = 0; i < TRACK_ADDRS_COUNT; i++) | |
550 | if (t->addrs[i]) | |
f9f58285 | 551 | pr_err("\t%pS\n", (void *)t->addrs[i]); |
d6543e39 BG |
552 | else |
553 | break; | |
554 | } | |
555 | #endif | |
24922684 CL |
556 | } |
557 | ||
558 | static void print_tracking(struct kmem_cache *s, void *object) | |
559 | { | |
560 | if (!(s->flags & SLAB_STORE_USER)) | |
561 | return; | |
562 | ||
563 | print_track("Allocated", get_track(s, object, TRACK_ALLOC)); | |
564 | print_track("Freed", get_track(s, object, TRACK_FREE)); | |
565 | } | |
566 | ||
567 | static void print_page_info(struct page *page) | |
568 | { | |
f9f58285 | 569 | pr_err("INFO: Slab 0x%p objects=%u used=%u fp=0x%p flags=0x%04lx\n", |
d0e0ac97 | 570 | page, page->objects, page->inuse, page->freelist, page->flags); |
24922684 CL |
571 | |
572 | } | |
573 | ||
574 | static void slab_bug(struct kmem_cache *s, char *fmt, ...) | |
575 | { | |
ecc42fbe | 576 | struct va_format vaf; |
24922684 | 577 | va_list args; |
24922684 CL |
578 | |
579 | va_start(args, fmt); | |
ecc42fbe FF |
580 | vaf.fmt = fmt; |
581 | vaf.va = &args; | |
f9f58285 | 582 | pr_err("=============================================================================\n"); |
ecc42fbe | 583 | pr_err("BUG %s (%s): %pV\n", s->name, print_tainted(), &vaf); |
f9f58285 | 584 | pr_err("-----------------------------------------------------------------------------\n\n"); |
645df230 | 585 | |
373d4d09 | 586 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); |
ecc42fbe | 587 | va_end(args); |
81819f0f CL |
588 | } |
589 | ||
24922684 CL |
590 | static void slab_fix(struct kmem_cache *s, char *fmt, ...) |
591 | { | |
ecc42fbe | 592 | struct va_format vaf; |
24922684 | 593 | va_list args; |
24922684 CL |
594 | |
595 | va_start(args, fmt); | |
ecc42fbe FF |
596 | vaf.fmt = fmt; |
597 | vaf.va = &args; | |
598 | pr_err("FIX %s: %pV\n", s->name, &vaf); | |
24922684 | 599 | va_end(args); |
24922684 CL |
600 | } |
601 | ||
602 | static void print_trailer(struct kmem_cache *s, struct page *page, u8 *p) | |
81819f0f CL |
603 | { |
604 | unsigned int off; /* Offset of last byte */ | |
a973e9dd | 605 | u8 *addr = page_address(page); |
24922684 CL |
606 | |
607 | print_tracking(s, p); | |
608 | ||
609 | print_page_info(page); | |
610 | ||
f9f58285 FF |
611 | pr_err("INFO: Object 0x%p @offset=%tu fp=0x%p\n\n", |
612 | p, p - addr, get_freepointer(s, p)); | |
24922684 CL |
613 | |
614 | if (p > addr + 16) | |
ffc79d28 | 615 | print_section("Bytes b4 ", p - 16, 16); |
81819f0f | 616 | |
3b0efdfa | 617 | print_section("Object ", p, min_t(unsigned long, s->object_size, |
ffc79d28 | 618 | PAGE_SIZE)); |
81819f0f | 619 | if (s->flags & SLAB_RED_ZONE) |
3b0efdfa CL |
620 | print_section("Redzone ", p + s->object_size, |
621 | s->inuse - s->object_size); | |
81819f0f | 622 | |
81819f0f CL |
623 | if (s->offset) |
624 | off = s->offset + sizeof(void *); | |
625 | else | |
626 | off = s->inuse; | |
627 | ||
24922684 | 628 | if (s->flags & SLAB_STORE_USER) |
81819f0f | 629 | off += 2 * sizeof(struct track); |
81819f0f CL |
630 | |
631 | if (off != s->size) | |
632 | /* Beginning of the filler is the free pointer */ | |
ffc79d28 | 633 | print_section("Padding ", p + off, s->size - off); |
24922684 CL |
634 | |
635 | dump_stack(); | |
81819f0f CL |
636 | } |
637 | ||
638 | static void object_err(struct kmem_cache *s, struct page *page, | |
639 | u8 *object, char *reason) | |
640 | { | |
3dc50637 | 641 | slab_bug(s, "%s", reason); |
24922684 | 642 | print_trailer(s, page, object); |
81819f0f CL |
643 | } |
644 | ||
d0e0ac97 CG |
645 | static void slab_err(struct kmem_cache *s, struct page *page, |
646 | const char *fmt, ...) | |
81819f0f CL |
647 | { |
648 | va_list args; | |
649 | char buf[100]; | |
650 | ||
24922684 CL |
651 | va_start(args, fmt); |
652 | vsnprintf(buf, sizeof(buf), fmt, args); | |
81819f0f | 653 | va_end(args); |
3dc50637 | 654 | slab_bug(s, "%s", buf); |
24922684 | 655 | print_page_info(page); |
81819f0f CL |
656 | dump_stack(); |
657 | } | |
658 | ||
f7cb1933 | 659 | static void init_object(struct kmem_cache *s, void *object, u8 val) |
81819f0f CL |
660 | { |
661 | u8 *p = object; | |
662 | ||
663 | if (s->flags & __OBJECT_POISON) { | |
3b0efdfa CL |
664 | memset(p, POISON_FREE, s->object_size - 1); |
665 | p[s->object_size - 1] = POISON_END; | |
81819f0f CL |
666 | } |
667 | ||
668 | if (s->flags & SLAB_RED_ZONE) | |
3b0efdfa | 669 | memset(p + s->object_size, val, s->inuse - s->object_size); |
81819f0f CL |
670 | } |
671 | ||
24922684 CL |
672 | static void restore_bytes(struct kmem_cache *s, char *message, u8 data, |
673 | void *from, void *to) | |
674 | { | |
675 | slab_fix(s, "Restoring 0x%p-0x%p=0x%x\n", from, to - 1, data); | |
676 | memset(from, data, to - from); | |
677 | } | |
678 | ||
679 | static int check_bytes_and_report(struct kmem_cache *s, struct page *page, | |
680 | u8 *object, char *what, | |
06428780 | 681 | u8 *start, unsigned int value, unsigned int bytes) |
24922684 CL |
682 | { |
683 | u8 *fault; | |
684 | u8 *end; | |
685 | ||
79824820 | 686 | fault = memchr_inv(start, value, bytes); |
24922684 CL |
687 | if (!fault) |
688 | return 1; | |
689 | ||
690 | end = start + bytes; | |
691 | while (end > fault && end[-1] == value) | |
692 | end--; | |
693 | ||
694 | slab_bug(s, "%s overwritten", what); | |
f9f58285 | 695 | pr_err("INFO: 0x%p-0x%p. First byte 0x%x instead of 0x%x\n", |
24922684 CL |
696 | fault, end - 1, fault[0], value); |
697 | print_trailer(s, page, object); | |
698 | ||
699 | restore_bytes(s, what, value, fault, end); | |
700 | return 0; | |
81819f0f CL |
701 | } |
702 | ||
81819f0f CL |
703 | /* |
704 | * Object layout: | |
705 | * | |
706 | * object address | |
707 | * Bytes of the object to be managed. | |
708 | * If the freepointer may overlay the object then the free | |
709 | * pointer is the first word of the object. | |
672bba3a | 710 | * |
81819f0f CL |
711 | * Poisoning uses 0x6b (POISON_FREE) and the last byte is |
712 | * 0xa5 (POISON_END) | |
713 | * | |
3b0efdfa | 714 | * object + s->object_size |
81819f0f | 715 | * Padding to reach word boundary. This is also used for Redzoning. |
672bba3a | 716 | * Padding is extended by another word if Redzoning is enabled and |
3b0efdfa | 717 | * object_size == inuse. |
672bba3a | 718 | * |
81819f0f CL |
719 | * We fill with 0xbb (RED_INACTIVE) for inactive objects and with |
720 | * 0xcc (RED_ACTIVE) for objects in use. | |
721 | * | |
722 | * object + s->inuse | |
672bba3a CL |
723 | * Meta data starts here. |
724 | * | |
81819f0f CL |
725 | * A. Free pointer (if we cannot overwrite object on free) |
726 | * B. Tracking data for SLAB_STORE_USER | |
672bba3a | 727 | * C. Padding to reach required alignment boundary or at mininum |
6446faa2 | 728 | * one word if debugging is on to be able to detect writes |
672bba3a CL |
729 | * before the word boundary. |
730 | * | |
731 | * Padding is done using 0x5a (POISON_INUSE) | |
81819f0f CL |
732 | * |
733 | * object + s->size | |
672bba3a | 734 | * Nothing is used beyond s->size. |
81819f0f | 735 | * |
3b0efdfa | 736 | * If slabcaches are merged then the object_size and inuse boundaries are mostly |
672bba3a | 737 | * ignored. And therefore no slab options that rely on these boundaries |
81819f0f CL |
738 | * may be used with merged slabcaches. |
739 | */ | |
740 | ||
81819f0f CL |
741 | static int check_pad_bytes(struct kmem_cache *s, struct page *page, u8 *p) |
742 | { | |
743 | unsigned long off = s->inuse; /* The end of info */ | |
744 | ||
745 | if (s->offset) | |
746 | /* Freepointer is placed after the object. */ | |
747 | off += sizeof(void *); | |
748 | ||
749 | if (s->flags & SLAB_STORE_USER) | |
750 | /* We also have user information there */ | |
751 | off += 2 * sizeof(struct track); | |
752 | ||
753 | if (s->size == off) | |
754 | return 1; | |
755 | ||
24922684 CL |
756 | return check_bytes_and_report(s, page, p, "Object padding", |
757 | p + off, POISON_INUSE, s->size - off); | |
81819f0f CL |
758 | } |
759 | ||
39b26464 | 760 | /* Check the pad bytes at the end of a slab page */ |
81819f0f CL |
761 | static int slab_pad_check(struct kmem_cache *s, struct page *page) |
762 | { | |
24922684 CL |
763 | u8 *start; |
764 | u8 *fault; | |
765 | u8 *end; | |
766 | int length; | |
767 | int remainder; | |
81819f0f CL |
768 | |
769 | if (!(s->flags & SLAB_POISON)) | |
770 | return 1; | |
771 | ||
a973e9dd | 772 | start = page_address(page); |
ab9a0f19 | 773 | length = (PAGE_SIZE << compound_order(page)) - s->reserved; |
39b26464 CL |
774 | end = start + length; |
775 | remainder = length % s->size; | |
81819f0f CL |
776 | if (!remainder) |
777 | return 1; | |
778 | ||
79824820 | 779 | fault = memchr_inv(end - remainder, POISON_INUSE, remainder); |
24922684 CL |
780 | if (!fault) |
781 | return 1; | |
782 | while (end > fault && end[-1] == POISON_INUSE) | |
783 | end--; | |
784 | ||
785 | slab_err(s, page, "Padding overwritten. 0x%p-0x%p", fault, end - 1); | |
ffc79d28 | 786 | print_section("Padding ", end - remainder, remainder); |
24922684 | 787 | |
8a3d271d | 788 | restore_bytes(s, "slab padding", POISON_INUSE, end - remainder, end); |
24922684 | 789 | return 0; |
81819f0f CL |
790 | } |
791 | ||
792 | static int check_object(struct kmem_cache *s, struct page *page, | |
f7cb1933 | 793 | void *object, u8 val) |
81819f0f CL |
794 | { |
795 | u8 *p = object; | |
3b0efdfa | 796 | u8 *endobject = object + s->object_size; |
81819f0f CL |
797 | |
798 | if (s->flags & SLAB_RED_ZONE) { | |
24922684 | 799 | if (!check_bytes_and_report(s, page, object, "Redzone", |
3b0efdfa | 800 | endobject, val, s->inuse - s->object_size)) |
81819f0f | 801 | return 0; |
81819f0f | 802 | } else { |
3b0efdfa | 803 | if ((s->flags & SLAB_POISON) && s->object_size < s->inuse) { |
3adbefee | 804 | check_bytes_and_report(s, page, p, "Alignment padding", |
d0e0ac97 CG |
805 | endobject, POISON_INUSE, |
806 | s->inuse - s->object_size); | |
3adbefee | 807 | } |
81819f0f CL |
808 | } |
809 | ||
810 | if (s->flags & SLAB_POISON) { | |
f7cb1933 | 811 | if (val != SLUB_RED_ACTIVE && (s->flags & __OBJECT_POISON) && |
24922684 | 812 | (!check_bytes_and_report(s, page, p, "Poison", p, |
3b0efdfa | 813 | POISON_FREE, s->object_size - 1) || |
24922684 | 814 | !check_bytes_and_report(s, page, p, "Poison", |
3b0efdfa | 815 | p + s->object_size - 1, POISON_END, 1))) |
81819f0f | 816 | return 0; |
81819f0f CL |
817 | /* |
818 | * check_pad_bytes cleans up on its own. | |
819 | */ | |
820 | check_pad_bytes(s, page, p); | |
821 | } | |
822 | ||
f7cb1933 | 823 | if (!s->offset && val == SLUB_RED_ACTIVE) |
81819f0f CL |
824 | /* |
825 | * Object and freepointer overlap. Cannot check | |
826 | * freepointer while object is allocated. | |
827 | */ | |
828 | return 1; | |
829 | ||
830 | /* Check free pointer validity */ | |
831 | if (!check_valid_pointer(s, page, get_freepointer(s, p))) { | |
832 | object_err(s, page, p, "Freepointer corrupt"); | |
833 | /* | |
9f6c708e | 834 | * No choice but to zap it and thus lose the remainder |
81819f0f | 835 | * of the free objects in this slab. May cause |
672bba3a | 836 | * another error because the object count is now wrong. |
81819f0f | 837 | */ |
a973e9dd | 838 | set_freepointer(s, p, NULL); |
81819f0f CL |
839 | return 0; |
840 | } | |
841 | return 1; | |
842 | } | |
843 | ||
844 | static int check_slab(struct kmem_cache *s, struct page *page) | |
845 | { | |
39b26464 CL |
846 | int maxobj; |
847 | ||
81819f0f CL |
848 | VM_BUG_ON(!irqs_disabled()); |
849 | ||
850 | if (!PageSlab(page)) { | |
24922684 | 851 | slab_err(s, page, "Not a valid slab page"); |
81819f0f CL |
852 | return 0; |
853 | } | |
39b26464 | 854 | |
ab9a0f19 | 855 | maxobj = order_objects(compound_order(page), s->size, s->reserved); |
39b26464 CL |
856 | if (page->objects > maxobj) { |
857 | slab_err(s, page, "objects %u > max %u", | |
858 | s->name, page->objects, maxobj); | |
859 | return 0; | |
860 | } | |
861 | if (page->inuse > page->objects) { | |
24922684 | 862 | slab_err(s, page, "inuse %u > max %u", |
39b26464 | 863 | s->name, page->inuse, page->objects); |
81819f0f CL |
864 | return 0; |
865 | } | |
866 | /* Slab_pad_check fixes things up after itself */ | |
867 | slab_pad_check(s, page); | |
868 | return 1; | |
869 | } | |
870 | ||
871 | /* | |
672bba3a CL |
872 | * Determine if a certain object on a page is on the freelist. Must hold the |
873 | * slab lock to guarantee that the chains are in a consistent state. | |
81819f0f CL |
874 | */ |
875 | static int on_freelist(struct kmem_cache *s, struct page *page, void *search) | |
876 | { | |
877 | int nr = 0; | |
881db7fb | 878 | void *fp; |
81819f0f | 879 | void *object = NULL; |
224a88be | 880 | unsigned long max_objects; |
81819f0f | 881 | |
881db7fb | 882 | fp = page->freelist; |
39b26464 | 883 | while (fp && nr <= page->objects) { |
81819f0f CL |
884 | if (fp == search) |
885 | return 1; | |
886 | if (!check_valid_pointer(s, page, fp)) { | |
887 | if (object) { | |
888 | object_err(s, page, object, | |
889 | "Freechain corrupt"); | |
a973e9dd | 890 | set_freepointer(s, object, NULL); |
81819f0f | 891 | } else { |
24922684 | 892 | slab_err(s, page, "Freepointer corrupt"); |
a973e9dd | 893 | page->freelist = NULL; |
39b26464 | 894 | page->inuse = page->objects; |
24922684 | 895 | slab_fix(s, "Freelist cleared"); |
81819f0f CL |
896 | return 0; |
897 | } | |
898 | break; | |
899 | } | |
900 | object = fp; | |
901 | fp = get_freepointer(s, object); | |
902 | nr++; | |
903 | } | |
904 | ||
ab9a0f19 | 905 | max_objects = order_objects(compound_order(page), s->size, s->reserved); |
210b5c06 CG |
906 | if (max_objects > MAX_OBJS_PER_PAGE) |
907 | max_objects = MAX_OBJS_PER_PAGE; | |
224a88be CL |
908 | |
909 | if (page->objects != max_objects) { | |
910 | slab_err(s, page, "Wrong number of objects. Found %d but " | |
911 | "should be %d", page->objects, max_objects); | |
912 | page->objects = max_objects; | |
913 | slab_fix(s, "Number of objects adjusted."); | |
914 | } | |
39b26464 | 915 | if (page->inuse != page->objects - nr) { |
70d71228 | 916 | slab_err(s, page, "Wrong object count. Counter is %d but " |
39b26464 CL |
917 | "counted were %d", page->inuse, page->objects - nr); |
918 | page->inuse = page->objects - nr; | |
24922684 | 919 | slab_fix(s, "Object count adjusted."); |
81819f0f CL |
920 | } |
921 | return search == NULL; | |
922 | } | |
923 | ||
0121c619 CL |
924 | static void trace(struct kmem_cache *s, struct page *page, void *object, |
925 | int alloc) | |
3ec09742 CL |
926 | { |
927 | if (s->flags & SLAB_TRACE) { | |
f9f58285 | 928 | pr_info("TRACE %s %s 0x%p inuse=%d fp=0x%p\n", |
3ec09742 CL |
929 | s->name, |
930 | alloc ? "alloc" : "free", | |
931 | object, page->inuse, | |
932 | page->freelist); | |
933 | ||
934 | if (!alloc) | |
d0e0ac97 CG |
935 | print_section("Object ", (void *)object, |
936 | s->object_size); | |
3ec09742 CL |
937 | |
938 | dump_stack(); | |
939 | } | |
940 | } | |
941 | ||
c016b0bd CL |
942 | /* |
943 | * Hooks for other subsystems that check memory allocations. In a typical | |
944 | * production configuration these hooks all should produce no code at all. | |
945 | */ | |
d56791b3 RB |
946 | static inline void kmalloc_large_node_hook(void *ptr, size_t size, gfp_t flags) |
947 | { | |
948 | kmemleak_alloc(ptr, size, 1, flags); | |
949 | } | |
950 | ||
951 | static inline void kfree_hook(const void *x) | |
952 | { | |
953 | kmemleak_free(x); | |
954 | } | |
955 | ||
c016b0bd CL |
956 | static inline int slab_pre_alloc_hook(struct kmem_cache *s, gfp_t flags) |
957 | { | |
c1d50836 | 958 | flags &= gfp_allowed_mask; |
c016b0bd CL |
959 | lockdep_trace_alloc(flags); |
960 | might_sleep_if(flags & __GFP_WAIT); | |
961 | ||
3b0efdfa | 962 | return should_failslab(s->object_size, flags, s->flags); |
c016b0bd CL |
963 | } |
964 | ||
d0e0ac97 CG |
965 | static inline void slab_post_alloc_hook(struct kmem_cache *s, |
966 | gfp_t flags, void *object) | |
c016b0bd | 967 | { |
c1d50836 | 968 | flags &= gfp_allowed_mask; |
b3d41885 | 969 | kmemcheck_slab_alloc(s, flags, object, slab_ksize(s)); |
3b0efdfa | 970 | kmemleak_alloc_recursive(object, s->object_size, 1, s->flags, flags); |
c016b0bd CL |
971 | } |
972 | ||
973 | static inline void slab_free_hook(struct kmem_cache *s, void *x) | |
974 | { | |
975 | kmemleak_free_recursive(x, s->flags); | |
c016b0bd | 976 | |
d3f661d6 | 977 | /* |
d1756174 | 978 | * Trouble is that we may no longer disable interrupts in the fast path |
d3f661d6 CL |
979 | * So in order to make the debug calls that expect irqs to be |
980 | * disabled we need to disable interrupts temporarily. | |
981 | */ | |
982 | #if defined(CONFIG_KMEMCHECK) || defined(CONFIG_LOCKDEP) | |
983 | { | |
984 | unsigned long flags; | |
985 | ||
986 | local_irq_save(flags); | |
3b0efdfa CL |
987 | kmemcheck_slab_free(s, x, s->object_size); |
988 | debug_check_no_locks_freed(x, s->object_size); | |
d3f661d6 CL |
989 | local_irq_restore(flags); |
990 | } | |
991 | #endif | |
f9b615de | 992 | if (!(s->flags & SLAB_DEBUG_OBJECTS)) |
3b0efdfa | 993 | debug_check_no_obj_freed(x, s->object_size); |
c016b0bd CL |
994 | } |
995 | ||
643b1138 | 996 | /* |
672bba3a | 997 | * Tracking of fully allocated slabs for debugging purposes. |
643b1138 | 998 | */ |
5cc6eee8 CL |
999 | static void add_full(struct kmem_cache *s, |
1000 | struct kmem_cache_node *n, struct page *page) | |
643b1138 | 1001 | { |
5cc6eee8 CL |
1002 | if (!(s->flags & SLAB_STORE_USER)) |
1003 | return; | |
1004 | ||
255d0884 | 1005 | lockdep_assert_held(&n->list_lock); |
643b1138 | 1006 | list_add(&page->lru, &n->full); |
643b1138 CL |
1007 | } |
1008 | ||
c65c1877 | 1009 | static void remove_full(struct kmem_cache *s, struct kmem_cache_node *n, struct page *page) |
643b1138 | 1010 | { |
643b1138 CL |
1011 | if (!(s->flags & SLAB_STORE_USER)) |
1012 | return; | |
1013 | ||
255d0884 | 1014 | lockdep_assert_held(&n->list_lock); |
643b1138 | 1015 | list_del(&page->lru); |
643b1138 CL |
1016 | } |
1017 | ||
0f389ec6 CL |
1018 | /* Tracking of the number of slabs for debugging purposes */ |
1019 | static inline unsigned long slabs_node(struct kmem_cache *s, int node) | |
1020 | { | |
1021 | struct kmem_cache_node *n = get_node(s, node); | |
1022 | ||
1023 | return atomic_long_read(&n->nr_slabs); | |
1024 | } | |
1025 | ||
26c02cf0 AB |
1026 | static inline unsigned long node_nr_slabs(struct kmem_cache_node *n) |
1027 | { | |
1028 | return atomic_long_read(&n->nr_slabs); | |
1029 | } | |
1030 | ||
205ab99d | 1031 | static inline void inc_slabs_node(struct kmem_cache *s, int node, int objects) |
0f389ec6 CL |
1032 | { |
1033 | struct kmem_cache_node *n = get_node(s, node); | |
1034 | ||
1035 | /* | |
1036 | * May be called early in order to allocate a slab for the | |
1037 | * kmem_cache_node structure. Solve the chicken-egg | |
1038 | * dilemma by deferring the increment of the count during | |
1039 | * bootstrap (see early_kmem_cache_node_alloc). | |
1040 | */ | |
338b2642 | 1041 | if (likely(n)) { |
0f389ec6 | 1042 | atomic_long_inc(&n->nr_slabs); |
205ab99d CL |
1043 | atomic_long_add(objects, &n->total_objects); |
1044 | } | |
0f389ec6 | 1045 | } |
205ab99d | 1046 | static inline void dec_slabs_node(struct kmem_cache *s, int node, int objects) |
0f389ec6 CL |
1047 | { |
1048 | struct kmem_cache_node *n = get_node(s, node); | |
1049 | ||
1050 | atomic_long_dec(&n->nr_slabs); | |
205ab99d | 1051 | atomic_long_sub(objects, &n->total_objects); |
0f389ec6 CL |
1052 | } |
1053 | ||
1054 | /* Object debug checks for alloc/free paths */ | |
3ec09742 CL |
1055 | static void setup_object_debug(struct kmem_cache *s, struct page *page, |
1056 | void *object) | |
1057 | { | |
1058 | if (!(s->flags & (SLAB_STORE_USER|SLAB_RED_ZONE|__OBJECT_POISON))) | |
1059 | return; | |
1060 | ||
f7cb1933 | 1061 | init_object(s, object, SLUB_RED_INACTIVE); |
3ec09742 CL |
1062 | init_tracking(s, object); |
1063 | } | |
1064 | ||
d0e0ac97 CG |
1065 | static noinline int alloc_debug_processing(struct kmem_cache *s, |
1066 | struct page *page, | |
ce71e27c | 1067 | void *object, unsigned long addr) |
81819f0f CL |
1068 | { |
1069 | if (!check_slab(s, page)) | |
1070 | goto bad; | |
1071 | ||
81819f0f CL |
1072 | if (!check_valid_pointer(s, page, object)) { |
1073 | object_err(s, page, object, "Freelist Pointer check fails"); | |
70d71228 | 1074 | goto bad; |
81819f0f CL |
1075 | } |
1076 | ||
f7cb1933 | 1077 | if (!check_object(s, page, object, SLUB_RED_INACTIVE)) |
81819f0f | 1078 | goto bad; |
81819f0f | 1079 | |
3ec09742 CL |
1080 | /* Success perform special debug activities for allocs */ |
1081 | if (s->flags & SLAB_STORE_USER) | |
1082 | set_track(s, object, TRACK_ALLOC, addr); | |
1083 | trace(s, page, object, 1); | |
f7cb1933 | 1084 | init_object(s, object, SLUB_RED_ACTIVE); |
81819f0f | 1085 | return 1; |
3ec09742 | 1086 | |
81819f0f CL |
1087 | bad: |
1088 | if (PageSlab(page)) { | |
1089 | /* | |
1090 | * If this is a slab page then lets do the best we can | |
1091 | * to avoid issues in the future. Marking all objects | |
672bba3a | 1092 | * as used avoids touching the remaining objects. |
81819f0f | 1093 | */ |
24922684 | 1094 | slab_fix(s, "Marking all objects used"); |
39b26464 | 1095 | page->inuse = page->objects; |
a973e9dd | 1096 | page->freelist = NULL; |
81819f0f CL |
1097 | } |
1098 | return 0; | |
1099 | } | |
1100 | ||
19c7ff9e CL |
1101 | static noinline struct kmem_cache_node *free_debug_processing( |
1102 | struct kmem_cache *s, struct page *page, void *object, | |
1103 | unsigned long addr, unsigned long *flags) | |
81819f0f | 1104 | { |
19c7ff9e | 1105 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); |
5c2e4bbb | 1106 | |
19c7ff9e | 1107 | spin_lock_irqsave(&n->list_lock, *flags); |
881db7fb CL |
1108 | slab_lock(page); |
1109 | ||
81819f0f CL |
1110 | if (!check_slab(s, page)) |
1111 | goto fail; | |
1112 | ||
1113 | if (!check_valid_pointer(s, page, object)) { | |
70d71228 | 1114 | slab_err(s, page, "Invalid object pointer 0x%p", object); |
81819f0f CL |
1115 | goto fail; |
1116 | } | |
1117 | ||
1118 | if (on_freelist(s, page, object)) { | |
24922684 | 1119 | object_err(s, page, object, "Object already free"); |
81819f0f CL |
1120 | goto fail; |
1121 | } | |
1122 | ||
f7cb1933 | 1123 | if (!check_object(s, page, object, SLUB_RED_ACTIVE)) |
5c2e4bbb | 1124 | goto out; |
81819f0f | 1125 | |
1b4f59e3 | 1126 | if (unlikely(s != page->slab_cache)) { |
3adbefee | 1127 | if (!PageSlab(page)) { |
70d71228 CL |
1128 | slab_err(s, page, "Attempt to free object(0x%p) " |
1129 | "outside of slab", object); | |
1b4f59e3 | 1130 | } else if (!page->slab_cache) { |
f9f58285 FF |
1131 | pr_err("SLUB <none>: no slab for object 0x%p.\n", |
1132 | object); | |
70d71228 | 1133 | dump_stack(); |
06428780 | 1134 | } else |
24922684 CL |
1135 | object_err(s, page, object, |
1136 | "page slab pointer corrupt."); | |
81819f0f CL |
1137 | goto fail; |
1138 | } | |
3ec09742 | 1139 | |
3ec09742 CL |
1140 | if (s->flags & SLAB_STORE_USER) |
1141 | set_track(s, object, TRACK_FREE, addr); | |
1142 | trace(s, page, object, 0); | |
f7cb1933 | 1143 | init_object(s, object, SLUB_RED_INACTIVE); |
5c2e4bbb | 1144 | out: |
881db7fb | 1145 | slab_unlock(page); |
19c7ff9e CL |
1146 | /* |
1147 | * Keep node_lock to preserve integrity | |
1148 | * until the object is actually freed | |
1149 | */ | |
1150 | return n; | |
3ec09742 | 1151 | |
81819f0f | 1152 | fail: |
19c7ff9e CL |
1153 | slab_unlock(page); |
1154 | spin_unlock_irqrestore(&n->list_lock, *flags); | |
24922684 | 1155 | slab_fix(s, "Object at 0x%p not freed", object); |
19c7ff9e | 1156 | return NULL; |
81819f0f CL |
1157 | } |
1158 | ||
41ecc55b CL |
1159 | static int __init setup_slub_debug(char *str) |
1160 | { | |
f0630fff CL |
1161 | slub_debug = DEBUG_DEFAULT_FLAGS; |
1162 | if (*str++ != '=' || !*str) | |
1163 | /* | |
1164 | * No options specified. Switch on full debugging. | |
1165 | */ | |
1166 | goto out; | |
1167 | ||
1168 | if (*str == ',') | |
1169 | /* | |
1170 | * No options but restriction on slabs. This means full | |
1171 | * debugging for slabs matching a pattern. | |
1172 | */ | |
1173 | goto check_slabs; | |
1174 | ||
fa5ec8a1 DR |
1175 | if (tolower(*str) == 'o') { |
1176 | /* | |
1177 | * Avoid enabling debugging on caches if its minimum order | |
1178 | * would increase as a result. | |
1179 | */ | |
1180 | disable_higher_order_debug = 1; | |
1181 | goto out; | |
1182 | } | |
1183 | ||
f0630fff CL |
1184 | slub_debug = 0; |
1185 | if (*str == '-') | |
1186 | /* | |
1187 | * Switch off all debugging measures. | |
1188 | */ | |
1189 | goto out; | |
1190 | ||
1191 | /* | |
1192 | * Determine which debug features should be switched on | |
1193 | */ | |
06428780 | 1194 | for (; *str && *str != ','; str++) { |
f0630fff CL |
1195 | switch (tolower(*str)) { |
1196 | case 'f': | |
1197 | slub_debug |= SLAB_DEBUG_FREE; | |
1198 | break; | |
1199 | case 'z': | |
1200 | slub_debug |= SLAB_RED_ZONE; | |
1201 | break; | |
1202 | case 'p': | |
1203 | slub_debug |= SLAB_POISON; | |
1204 | break; | |
1205 | case 'u': | |
1206 | slub_debug |= SLAB_STORE_USER; | |
1207 | break; | |
1208 | case 't': | |
1209 | slub_debug |= SLAB_TRACE; | |
1210 | break; | |
4c13dd3b DM |
1211 | case 'a': |
1212 | slub_debug |= SLAB_FAILSLAB; | |
1213 | break; | |
f0630fff | 1214 | default: |
f9f58285 FF |
1215 | pr_err("slub_debug option '%c' unknown. skipped\n", |
1216 | *str); | |
f0630fff | 1217 | } |
41ecc55b CL |
1218 | } |
1219 | ||
f0630fff | 1220 | check_slabs: |
41ecc55b CL |
1221 | if (*str == ',') |
1222 | slub_debug_slabs = str + 1; | |
f0630fff | 1223 | out: |
41ecc55b CL |
1224 | return 1; |
1225 | } | |
1226 | ||
1227 | __setup("slub_debug", setup_slub_debug); | |
1228 | ||
3b0efdfa | 1229 | static unsigned long kmem_cache_flags(unsigned long object_size, |
ba0268a8 | 1230 | unsigned long flags, const char *name, |
51cc5068 | 1231 | void (*ctor)(void *)) |
41ecc55b CL |
1232 | { |
1233 | /* | |
e153362a | 1234 | * Enable debugging if selected on the kernel commandline. |
41ecc55b | 1235 | */ |
c6f58d9b CL |
1236 | if (slub_debug && (!slub_debug_slabs || (name && |
1237 | !strncmp(slub_debug_slabs, name, strlen(slub_debug_slabs))))) | |
3de47213 | 1238 | flags |= slub_debug; |
ba0268a8 CL |
1239 | |
1240 | return flags; | |
41ecc55b CL |
1241 | } |
1242 | #else | |
3ec09742 CL |
1243 | static inline void setup_object_debug(struct kmem_cache *s, |
1244 | struct page *page, void *object) {} | |
41ecc55b | 1245 | |
3ec09742 | 1246 | static inline int alloc_debug_processing(struct kmem_cache *s, |
ce71e27c | 1247 | struct page *page, void *object, unsigned long addr) { return 0; } |
41ecc55b | 1248 | |
19c7ff9e CL |
1249 | static inline struct kmem_cache_node *free_debug_processing( |
1250 | struct kmem_cache *s, struct page *page, void *object, | |
1251 | unsigned long addr, unsigned long *flags) { return NULL; } | |
41ecc55b | 1252 | |
41ecc55b CL |
1253 | static inline int slab_pad_check(struct kmem_cache *s, struct page *page) |
1254 | { return 1; } | |
1255 | static inline int check_object(struct kmem_cache *s, struct page *page, | |
f7cb1933 | 1256 | void *object, u8 val) { return 1; } |
5cc6eee8 CL |
1257 | static inline void add_full(struct kmem_cache *s, struct kmem_cache_node *n, |
1258 | struct page *page) {} | |
c65c1877 PZ |
1259 | static inline void remove_full(struct kmem_cache *s, struct kmem_cache_node *n, |
1260 | struct page *page) {} | |
3b0efdfa | 1261 | static inline unsigned long kmem_cache_flags(unsigned long object_size, |
ba0268a8 | 1262 | unsigned long flags, const char *name, |
51cc5068 | 1263 | void (*ctor)(void *)) |
ba0268a8 CL |
1264 | { |
1265 | return flags; | |
1266 | } | |
41ecc55b | 1267 | #define slub_debug 0 |
0f389ec6 | 1268 | |
fdaa45e9 IM |
1269 | #define disable_higher_order_debug 0 |
1270 | ||
0f389ec6 CL |
1271 | static inline unsigned long slabs_node(struct kmem_cache *s, int node) |
1272 | { return 0; } | |
26c02cf0 AB |
1273 | static inline unsigned long node_nr_slabs(struct kmem_cache_node *n) |
1274 | { return 0; } | |
205ab99d CL |
1275 | static inline void inc_slabs_node(struct kmem_cache *s, int node, |
1276 | int objects) {} | |
1277 | static inline void dec_slabs_node(struct kmem_cache *s, int node, | |
1278 | int objects) {} | |
7d550c56 | 1279 | |
d56791b3 RB |
1280 | static inline void kmalloc_large_node_hook(void *ptr, size_t size, gfp_t flags) |
1281 | { | |
1282 | kmemleak_alloc(ptr, size, 1, flags); | |
1283 | } | |
1284 | ||
1285 | static inline void kfree_hook(const void *x) | |
1286 | { | |
1287 | kmemleak_free(x); | |
1288 | } | |
1289 | ||
7d550c56 CL |
1290 | static inline int slab_pre_alloc_hook(struct kmem_cache *s, gfp_t flags) |
1291 | { return 0; } | |
1292 | ||
1293 | static inline void slab_post_alloc_hook(struct kmem_cache *s, gfp_t flags, | |
d56791b3 RB |
1294 | void *object) |
1295 | { | |
1296 | kmemleak_alloc_recursive(object, s->object_size, 1, s->flags, | |
1297 | flags & gfp_allowed_mask); | |
1298 | } | |
7d550c56 | 1299 | |
d56791b3 RB |
1300 | static inline void slab_free_hook(struct kmem_cache *s, void *x) |
1301 | { | |
1302 | kmemleak_free_recursive(x, s->flags); | |
1303 | } | |
7d550c56 | 1304 | |
ab4d5ed5 | 1305 | #endif /* CONFIG_SLUB_DEBUG */ |
205ab99d | 1306 | |
81819f0f CL |
1307 | /* |
1308 | * Slab allocation and freeing | |
1309 | */ | |
5dfb4175 VD |
1310 | static inline struct page *alloc_slab_page(struct kmem_cache *s, |
1311 | gfp_t flags, int node, struct kmem_cache_order_objects oo) | |
65c3376a | 1312 | { |
5dfb4175 | 1313 | struct page *page; |
65c3376a CL |
1314 | int order = oo_order(oo); |
1315 | ||
b1eeab67 VN |
1316 | flags |= __GFP_NOTRACK; |
1317 | ||
5dfb4175 VD |
1318 | if (memcg_charge_slab(s, flags, order)) |
1319 | return NULL; | |
1320 | ||
2154a336 | 1321 | if (node == NUMA_NO_NODE) |
5dfb4175 | 1322 | page = alloc_pages(flags, order); |
65c3376a | 1323 | else |
5dfb4175 VD |
1324 | page = alloc_pages_exact_node(node, flags, order); |
1325 | ||
1326 | if (!page) | |
1327 | memcg_uncharge_slab(s, order); | |
1328 | ||
1329 | return page; | |
65c3376a CL |
1330 | } |
1331 | ||
81819f0f CL |
1332 | static struct page *allocate_slab(struct kmem_cache *s, gfp_t flags, int node) |
1333 | { | |
06428780 | 1334 | struct page *page; |
834f3d11 | 1335 | struct kmem_cache_order_objects oo = s->oo; |
ba52270d | 1336 | gfp_t alloc_gfp; |
81819f0f | 1337 | |
7e0528da CL |
1338 | flags &= gfp_allowed_mask; |
1339 | ||
1340 | if (flags & __GFP_WAIT) | |
1341 | local_irq_enable(); | |
1342 | ||
b7a49f0d | 1343 | flags |= s->allocflags; |
e12ba74d | 1344 | |
ba52270d PE |
1345 | /* |
1346 | * Let the initial higher-order allocation fail under memory pressure | |
1347 | * so we fall-back to the minimum order allocation. | |
1348 | */ | |
1349 | alloc_gfp = (flags | __GFP_NOWARN | __GFP_NORETRY) & ~__GFP_NOFAIL; | |
1350 | ||
5dfb4175 | 1351 | page = alloc_slab_page(s, alloc_gfp, node, oo); |
65c3376a CL |
1352 | if (unlikely(!page)) { |
1353 | oo = s->min; | |
80c3a998 | 1354 | alloc_gfp = flags; |
65c3376a CL |
1355 | /* |
1356 | * Allocation may have failed due to fragmentation. | |
1357 | * Try a lower order alloc if possible | |
1358 | */ | |
5dfb4175 | 1359 | page = alloc_slab_page(s, alloc_gfp, node, oo); |
81819f0f | 1360 | |
7e0528da CL |
1361 | if (page) |
1362 | stat(s, ORDER_FALLBACK); | |
65c3376a | 1363 | } |
5a896d9e | 1364 | |
737b719e | 1365 | if (kmemcheck_enabled && page |
5086c389 | 1366 | && !(s->flags & (SLAB_NOTRACK | DEBUG_DEFAULT_FLAGS))) { |
b1eeab67 VN |
1367 | int pages = 1 << oo_order(oo); |
1368 | ||
80c3a998 | 1369 | kmemcheck_alloc_shadow(page, oo_order(oo), alloc_gfp, node); |
b1eeab67 VN |
1370 | |
1371 | /* | |
1372 | * Objects from caches that have a constructor don't get | |
1373 | * cleared when they're allocated, so we need to do it here. | |
1374 | */ | |
1375 | if (s->ctor) | |
1376 | kmemcheck_mark_uninitialized_pages(page, pages); | |
1377 | else | |
1378 | kmemcheck_mark_unallocated_pages(page, pages); | |
5a896d9e VN |
1379 | } |
1380 | ||
737b719e DR |
1381 | if (flags & __GFP_WAIT) |
1382 | local_irq_disable(); | |
1383 | if (!page) | |
1384 | return NULL; | |
1385 | ||
834f3d11 | 1386 | page->objects = oo_objects(oo); |
81819f0f CL |
1387 | mod_zone_page_state(page_zone(page), |
1388 | (s->flags & SLAB_RECLAIM_ACCOUNT) ? | |
1389 | NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE, | |
65c3376a | 1390 | 1 << oo_order(oo)); |
81819f0f CL |
1391 | |
1392 | return page; | |
1393 | } | |
1394 | ||
1395 | static void setup_object(struct kmem_cache *s, struct page *page, | |
1396 | void *object) | |
1397 | { | |
3ec09742 | 1398 | setup_object_debug(s, page, object); |
4f104934 | 1399 | if (unlikely(s->ctor)) |
51cc5068 | 1400 | s->ctor(object); |
81819f0f CL |
1401 | } |
1402 | ||
1403 | static struct page *new_slab(struct kmem_cache *s, gfp_t flags, int node) | |
1404 | { | |
1405 | struct page *page; | |
81819f0f | 1406 | void *start; |
81819f0f CL |
1407 | void *last; |
1408 | void *p; | |
1f458cbf | 1409 | int order; |
81819f0f | 1410 | |
6cb06229 | 1411 | BUG_ON(flags & GFP_SLAB_BUG_MASK); |
81819f0f | 1412 | |
6cb06229 CL |
1413 | page = allocate_slab(s, |
1414 | flags & (GFP_RECLAIM_MASK | GFP_CONSTRAINT_MASK), node); | |
81819f0f CL |
1415 | if (!page) |
1416 | goto out; | |
1417 | ||
1f458cbf | 1418 | order = compound_order(page); |
205ab99d | 1419 | inc_slabs_node(s, page_to_nid(page), page->objects); |
1b4f59e3 | 1420 | page->slab_cache = s; |
c03f94cc | 1421 | __SetPageSlab(page); |
072bb0aa MG |
1422 | if (page->pfmemalloc) |
1423 | SetPageSlabPfmemalloc(page); | |
81819f0f CL |
1424 | |
1425 | start = page_address(page); | |
81819f0f CL |
1426 | |
1427 | if (unlikely(s->flags & SLAB_POISON)) | |
1f458cbf | 1428 | memset(start, POISON_INUSE, PAGE_SIZE << order); |
81819f0f CL |
1429 | |
1430 | last = start; | |
224a88be | 1431 | for_each_object(p, s, start, page->objects) { |
81819f0f CL |
1432 | setup_object(s, page, last); |
1433 | set_freepointer(s, last, p); | |
1434 | last = p; | |
1435 | } | |
1436 | setup_object(s, page, last); | |
a973e9dd | 1437 | set_freepointer(s, last, NULL); |
81819f0f CL |
1438 | |
1439 | page->freelist = start; | |
e6e82ea1 | 1440 | page->inuse = page->objects; |
8cb0a506 | 1441 | page->frozen = 1; |
81819f0f | 1442 | out: |
81819f0f CL |
1443 | return page; |
1444 | } | |
1445 | ||
1446 | static void __free_slab(struct kmem_cache *s, struct page *page) | |
1447 | { | |
834f3d11 CL |
1448 | int order = compound_order(page); |
1449 | int pages = 1 << order; | |
81819f0f | 1450 | |
af537b0a | 1451 | if (kmem_cache_debug(s)) { |
81819f0f CL |
1452 | void *p; |
1453 | ||
1454 | slab_pad_check(s, page); | |
224a88be CL |
1455 | for_each_object(p, s, page_address(page), |
1456 | page->objects) | |
f7cb1933 | 1457 | check_object(s, page, p, SLUB_RED_INACTIVE); |
81819f0f CL |
1458 | } |
1459 | ||
b1eeab67 | 1460 | kmemcheck_free_shadow(page, compound_order(page)); |
5a896d9e | 1461 | |
81819f0f CL |
1462 | mod_zone_page_state(page_zone(page), |
1463 | (s->flags & SLAB_RECLAIM_ACCOUNT) ? | |
1464 | NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE, | |
06428780 | 1465 | -pages); |
81819f0f | 1466 | |
072bb0aa | 1467 | __ClearPageSlabPfmemalloc(page); |
49bd5221 | 1468 | __ClearPageSlab(page); |
1f458cbf | 1469 | |
22b751c3 | 1470 | page_mapcount_reset(page); |
1eb5ac64 NP |
1471 | if (current->reclaim_state) |
1472 | current->reclaim_state->reclaimed_slab += pages; | |
5dfb4175 VD |
1473 | __free_pages(page, order); |
1474 | memcg_uncharge_slab(s, order); | |
81819f0f CL |
1475 | } |
1476 | ||
da9a638c LJ |
1477 | #define need_reserve_slab_rcu \ |
1478 | (sizeof(((struct page *)NULL)->lru) < sizeof(struct rcu_head)) | |
1479 | ||
81819f0f CL |
1480 | static void rcu_free_slab(struct rcu_head *h) |
1481 | { | |
1482 | struct page *page; | |
1483 | ||
da9a638c LJ |
1484 | if (need_reserve_slab_rcu) |
1485 | page = virt_to_head_page(h); | |
1486 | else | |
1487 | page = container_of((struct list_head *)h, struct page, lru); | |
1488 | ||
1b4f59e3 | 1489 | __free_slab(page->slab_cache, page); |
81819f0f CL |
1490 | } |
1491 | ||
1492 | static void free_slab(struct kmem_cache *s, struct page *page) | |
1493 | { | |
1494 | if (unlikely(s->flags & SLAB_DESTROY_BY_RCU)) { | |
da9a638c LJ |
1495 | struct rcu_head *head; |
1496 | ||
1497 | if (need_reserve_slab_rcu) { | |
1498 | int order = compound_order(page); | |
1499 | int offset = (PAGE_SIZE << order) - s->reserved; | |
1500 | ||
1501 | VM_BUG_ON(s->reserved != sizeof(*head)); | |
1502 | head = page_address(page) + offset; | |
1503 | } else { | |
1504 | /* | |
1505 | * RCU free overloads the RCU head over the LRU | |
1506 | */ | |
1507 | head = (void *)&page->lru; | |
1508 | } | |
81819f0f CL |
1509 | |
1510 | call_rcu(head, rcu_free_slab); | |
1511 | } else | |
1512 | __free_slab(s, page); | |
1513 | } | |
1514 | ||
1515 | static void discard_slab(struct kmem_cache *s, struct page *page) | |
1516 | { | |
205ab99d | 1517 | dec_slabs_node(s, page_to_nid(page), page->objects); |
81819f0f CL |
1518 | free_slab(s, page); |
1519 | } | |
1520 | ||
1521 | /* | |
5cc6eee8 | 1522 | * Management of partially allocated slabs. |
81819f0f | 1523 | */ |
1e4dd946 SR |
1524 | static inline void |
1525 | __add_partial(struct kmem_cache_node *n, struct page *page, int tail) | |
81819f0f | 1526 | { |
e95eed57 | 1527 | n->nr_partial++; |
136333d1 | 1528 | if (tail == DEACTIVATE_TO_TAIL) |
7c2e132c CL |
1529 | list_add_tail(&page->lru, &n->partial); |
1530 | else | |
1531 | list_add(&page->lru, &n->partial); | |
81819f0f CL |
1532 | } |
1533 | ||
1e4dd946 SR |
1534 | static inline void add_partial(struct kmem_cache_node *n, |
1535 | struct page *page, int tail) | |
62e346a8 | 1536 | { |
c65c1877 | 1537 | lockdep_assert_held(&n->list_lock); |
1e4dd946 SR |
1538 | __add_partial(n, page, tail); |
1539 | } | |
c65c1877 | 1540 | |
1e4dd946 SR |
1541 | static inline void |
1542 | __remove_partial(struct kmem_cache_node *n, struct page *page) | |
1543 | { | |
62e346a8 CL |
1544 | list_del(&page->lru); |
1545 | n->nr_partial--; | |
1546 | } | |
1547 | ||
1e4dd946 SR |
1548 | static inline void remove_partial(struct kmem_cache_node *n, |
1549 | struct page *page) | |
1550 | { | |
1551 | lockdep_assert_held(&n->list_lock); | |
1552 | __remove_partial(n, page); | |
1553 | } | |
1554 | ||
81819f0f | 1555 | /* |
7ced3719 CL |
1556 | * Remove slab from the partial list, freeze it and |
1557 | * return the pointer to the freelist. | |
81819f0f | 1558 | * |
497b66f2 | 1559 | * Returns a list of objects or NULL if it fails. |
81819f0f | 1560 | */ |
497b66f2 | 1561 | static inline void *acquire_slab(struct kmem_cache *s, |
acd19fd1 | 1562 | struct kmem_cache_node *n, struct page *page, |
633b0764 | 1563 | int mode, int *objects) |
81819f0f | 1564 | { |
2cfb7455 CL |
1565 | void *freelist; |
1566 | unsigned long counters; | |
1567 | struct page new; | |
1568 | ||
c65c1877 PZ |
1569 | lockdep_assert_held(&n->list_lock); |
1570 | ||
2cfb7455 CL |
1571 | /* |
1572 | * Zap the freelist and set the frozen bit. | |
1573 | * The old freelist is the list of objects for the | |
1574 | * per cpu allocation list. | |
1575 | */ | |
7ced3719 CL |
1576 | freelist = page->freelist; |
1577 | counters = page->counters; | |
1578 | new.counters = counters; | |
633b0764 | 1579 | *objects = new.objects - new.inuse; |
23910c50 | 1580 | if (mode) { |
7ced3719 | 1581 | new.inuse = page->objects; |
23910c50 PE |
1582 | new.freelist = NULL; |
1583 | } else { | |
1584 | new.freelist = freelist; | |
1585 | } | |
2cfb7455 | 1586 | |
a0132ac0 | 1587 | VM_BUG_ON(new.frozen); |
7ced3719 | 1588 | new.frozen = 1; |
2cfb7455 | 1589 | |
7ced3719 | 1590 | if (!__cmpxchg_double_slab(s, page, |
2cfb7455 | 1591 | freelist, counters, |
02d7633f | 1592 | new.freelist, new.counters, |
7ced3719 | 1593 | "acquire_slab")) |
7ced3719 | 1594 | return NULL; |
2cfb7455 CL |
1595 | |
1596 | remove_partial(n, page); | |
7ced3719 | 1597 | WARN_ON(!freelist); |
49e22585 | 1598 | return freelist; |
81819f0f CL |
1599 | } |
1600 | ||
633b0764 | 1601 | static void put_cpu_partial(struct kmem_cache *s, struct page *page, int drain); |
8ba00bb6 | 1602 | static inline bool pfmemalloc_match(struct page *page, gfp_t gfpflags); |
49e22585 | 1603 | |
81819f0f | 1604 | /* |
672bba3a | 1605 | * Try to allocate a partial slab from a specific node. |
81819f0f | 1606 | */ |
8ba00bb6 JK |
1607 | static void *get_partial_node(struct kmem_cache *s, struct kmem_cache_node *n, |
1608 | struct kmem_cache_cpu *c, gfp_t flags) | |
81819f0f | 1609 | { |
49e22585 CL |
1610 | struct page *page, *page2; |
1611 | void *object = NULL; | |
633b0764 JK |
1612 | int available = 0; |
1613 | int objects; | |
81819f0f CL |
1614 | |
1615 | /* | |
1616 | * Racy check. If we mistakenly see no partial slabs then we | |
1617 | * just allocate an empty slab. If we mistakenly try to get a | |
672bba3a CL |
1618 | * partial slab and there is none available then get_partials() |
1619 | * will return NULL. | |
81819f0f CL |
1620 | */ |
1621 | if (!n || !n->nr_partial) | |
1622 | return NULL; | |
1623 | ||
1624 | spin_lock(&n->list_lock); | |
49e22585 | 1625 | list_for_each_entry_safe(page, page2, &n->partial, lru) { |
8ba00bb6 | 1626 | void *t; |
49e22585 | 1627 | |
8ba00bb6 JK |
1628 | if (!pfmemalloc_match(page, flags)) |
1629 | continue; | |
1630 | ||
633b0764 | 1631 | t = acquire_slab(s, n, page, object == NULL, &objects); |
49e22585 CL |
1632 | if (!t) |
1633 | break; | |
1634 | ||
633b0764 | 1635 | available += objects; |
12d79634 | 1636 | if (!object) { |
49e22585 | 1637 | c->page = page; |
49e22585 | 1638 | stat(s, ALLOC_FROM_PARTIAL); |
49e22585 | 1639 | object = t; |
49e22585 | 1640 | } else { |
633b0764 | 1641 | put_cpu_partial(s, page, 0); |
8028dcea | 1642 | stat(s, CPU_PARTIAL_NODE); |
49e22585 | 1643 | } |
345c905d JK |
1644 | if (!kmem_cache_has_cpu_partial(s) |
1645 | || available > s->cpu_partial / 2) | |
49e22585 CL |
1646 | break; |
1647 | ||
497b66f2 | 1648 | } |
81819f0f | 1649 | spin_unlock(&n->list_lock); |
497b66f2 | 1650 | return object; |
81819f0f CL |
1651 | } |
1652 | ||
1653 | /* | |
672bba3a | 1654 | * Get a page from somewhere. Search in increasing NUMA distances. |
81819f0f | 1655 | */ |
de3ec035 | 1656 | static void *get_any_partial(struct kmem_cache *s, gfp_t flags, |
acd19fd1 | 1657 | struct kmem_cache_cpu *c) |
81819f0f CL |
1658 | { |
1659 | #ifdef CONFIG_NUMA | |
1660 | struct zonelist *zonelist; | |
dd1a239f | 1661 | struct zoneref *z; |
54a6eb5c MG |
1662 | struct zone *zone; |
1663 | enum zone_type high_zoneidx = gfp_zone(flags); | |
497b66f2 | 1664 | void *object; |
cc9a6c87 | 1665 | unsigned int cpuset_mems_cookie; |
81819f0f CL |
1666 | |
1667 | /* | |
672bba3a CL |
1668 | * The defrag ratio allows a configuration of the tradeoffs between |
1669 | * inter node defragmentation and node local allocations. A lower | |
1670 | * defrag_ratio increases the tendency to do local allocations | |
1671 | * instead of attempting to obtain partial slabs from other nodes. | |
81819f0f | 1672 | * |
672bba3a CL |
1673 | * If the defrag_ratio is set to 0 then kmalloc() always |
1674 | * returns node local objects. If the ratio is higher then kmalloc() | |
1675 | * may return off node objects because partial slabs are obtained | |
1676 | * from other nodes and filled up. | |
81819f0f | 1677 | * |
6446faa2 | 1678 | * If /sys/kernel/slab/xx/defrag_ratio is set to 100 (which makes |
672bba3a CL |
1679 | * defrag_ratio = 1000) then every (well almost) allocation will |
1680 | * first attempt to defrag slab caches on other nodes. This means | |
1681 | * scanning over all nodes to look for partial slabs which may be | |
1682 | * expensive if we do it every time we are trying to find a slab | |
1683 | * with available objects. | |
81819f0f | 1684 | */ |
9824601e CL |
1685 | if (!s->remote_node_defrag_ratio || |
1686 | get_cycles() % 1024 > s->remote_node_defrag_ratio) | |
81819f0f CL |
1687 | return NULL; |
1688 | ||
cc9a6c87 | 1689 | do { |
d26914d1 | 1690 | cpuset_mems_cookie = read_mems_allowed_begin(); |
2a389610 | 1691 | zonelist = node_zonelist(mempolicy_slab_node(), flags); |
cc9a6c87 MG |
1692 | for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) { |
1693 | struct kmem_cache_node *n; | |
1694 | ||
1695 | n = get_node(s, zone_to_nid(zone)); | |
1696 | ||
1697 | if (n && cpuset_zone_allowed_hardwall(zone, flags) && | |
1698 | n->nr_partial > s->min_partial) { | |
8ba00bb6 | 1699 | object = get_partial_node(s, n, c, flags); |
cc9a6c87 MG |
1700 | if (object) { |
1701 | /* | |
d26914d1 MG |
1702 | * Don't check read_mems_allowed_retry() |
1703 | * here - if mems_allowed was updated in | |
1704 | * parallel, that was a harmless race | |
1705 | * between allocation and the cpuset | |
1706 | * update | |
cc9a6c87 | 1707 | */ |
cc9a6c87 MG |
1708 | return object; |
1709 | } | |
c0ff7453 | 1710 | } |
81819f0f | 1711 | } |
d26914d1 | 1712 | } while (read_mems_allowed_retry(cpuset_mems_cookie)); |
81819f0f CL |
1713 | #endif |
1714 | return NULL; | |
1715 | } | |
1716 | ||
1717 | /* | |
1718 | * Get a partial page, lock it and return it. | |
1719 | */ | |
497b66f2 | 1720 | static void *get_partial(struct kmem_cache *s, gfp_t flags, int node, |
acd19fd1 | 1721 | struct kmem_cache_cpu *c) |
81819f0f | 1722 | { |
497b66f2 | 1723 | void *object; |
844e4d66 | 1724 | int searchnode = (node == NUMA_NO_NODE) ? numa_mem_id() : node; |
81819f0f | 1725 | |
8ba00bb6 | 1726 | object = get_partial_node(s, get_node(s, searchnode), c, flags); |
497b66f2 CL |
1727 | if (object || node != NUMA_NO_NODE) |
1728 | return object; | |
81819f0f | 1729 | |
acd19fd1 | 1730 | return get_any_partial(s, flags, c); |
81819f0f CL |
1731 | } |
1732 | ||
8a5ec0ba CL |
1733 | #ifdef CONFIG_PREEMPT |
1734 | /* | |
1735 | * Calculate the next globally unique transaction for disambiguiation | |
1736 | * during cmpxchg. The transactions start with the cpu number and are then | |
1737 | * incremented by CONFIG_NR_CPUS. | |
1738 | */ | |
1739 | #define TID_STEP roundup_pow_of_two(CONFIG_NR_CPUS) | |
1740 | #else | |
1741 | /* | |
1742 | * No preemption supported therefore also no need to check for | |
1743 | * different cpus. | |
1744 | */ | |
1745 | #define TID_STEP 1 | |
1746 | #endif | |
1747 | ||
1748 | static inline unsigned long next_tid(unsigned long tid) | |
1749 | { | |
1750 | return tid + TID_STEP; | |
1751 | } | |
1752 | ||
1753 | static inline unsigned int tid_to_cpu(unsigned long tid) | |
1754 | { | |
1755 | return tid % TID_STEP; | |
1756 | } | |
1757 | ||
1758 | static inline unsigned long tid_to_event(unsigned long tid) | |
1759 | { | |
1760 | return tid / TID_STEP; | |
1761 | } | |
1762 | ||
1763 | static inline unsigned int init_tid(int cpu) | |
1764 | { | |
1765 | return cpu; | |
1766 | } | |
1767 | ||
1768 | static inline void note_cmpxchg_failure(const char *n, | |
1769 | const struct kmem_cache *s, unsigned long tid) | |
1770 | { | |
1771 | #ifdef SLUB_DEBUG_CMPXCHG | |
1772 | unsigned long actual_tid = __this_cpu_read(s->cpu_slab->tid); | |
1773 | ||
f9f58285 | 1774 | pr_info("%s %s: cmpxchg redo ", n, s->name); |
8a5ec0ba CL |
1775 | |
1776 | #ifdef CONFIG_PREEMPT | |
1777 | if (tid_to_cpu(tid) != tid_to_cpu(actual_tid)) | |
f9f58285 | 1778 | pr_warn("due to cpu change %d -> %d\n", |
8a5ec0ba CL |
1779 | tid_to_cpu(tid), tid_to_cpu(actual_tid)); |
1780 | else | |
1781 | #endif | |
1782 | if (tid_to_event(tid) != tid_to_event(actual_tid)) | |
f9f58285 | 1783 | pr_warn("due to cpu running other code. Event %ld->%ld\n", |
8a5ec0ba CL |
1784 | tid_to_event(tid), tid_to_event(actual_tid)); |
1785 | else | |
f9f58285 | 1786 | pr_warn("for unknown reason: actual=%lx was=%lx target=%lx\n", |
8a5ec0ba CL |
1787 | actual_tid, tid, next_tid(tid)); |
1788 | #endif | |
4fdccdfb | 1789 | stat(s, CMPXCHG_DOUBLE_CPU_FAIL); |
8a5ec0ba CL |
1790 | } |
1791 | ||
788e1aad | 1792 | static void init_kmem_cache_cpus(struct kmem_cache *s) |
8a5ec0ba | 1793 | { |
8a5ec0ba CL |
1794 | int cpu; |
1795 | ||
1796 | for_each_possible_cpu(cpu) | |
1797 | per_cpu_ptr(s->cpu_slab, cpu)->tid = init_tid(cpu); | |
8a5ec0ba | 1798 | } |
2cfb7455 | 1799 | |
81819f0f CL |
1800 | /* |
1801 | * Remove the cpu slab | |
1802 | */ | |
d0e0ac97 CG |
1803 | static void deactivate_slab(struct kmem_cache *s, struct page *page, |
1804 | void *freelist) | |
81819f0f | 1805 | { |
2cfb7455 | 1806 | enum slab_modes { M_NONE, M_PARTIAL, M_FULL, M_FREE }; |
2cfb7455 CL |
1807 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); |
1808 | int lock = 0; | |
1809 | enum slab_modes l = M_NONE, m = M_NONE; | |
2cfb7455 | 1810 | void *nextfree; |
136333d1 | 1811 | int tail = DEACTIVATE_TO_HEAD; |
2cfb7455 CL |
1812 | struct page new; |
1813 | struct page old; | |
1814 | ||
1815 | if (page->freelist) { | |
84e554e6 | 1816 | stat(s, DEACTIVATE_REMOTE_FREES); |
136333d1 | 1817 | tail = DEACTIVATE_TO_TAIL; |
2cfb7455 CL |
1818 | } |
1819 | ||
894b8788 | 1820 | /* |
2cfb7455 CL |
1821 | * Stage one: Free all available per cpu objects back |
1822 | * to the page freelist while it is still frozen. Leave the | |
1823 | * last one. | |
1824 | * | |
1825 | * There is no need to take the list->lock because the page | |
1826 | * is still frozen. | |
1827 | */ | |
1828 | while (freelist && (nextfree = get_freepointer(s, freelist))) { | |
1829 | void *prior; | |
1830 | unsigned long counters; | |
1831 | ||
1832 | do { | |
1833 | prior = page->freelist; | |
1834 | counters = page->counters; | |
1835 | set_freepointer(s, freelist, prior); | |
1836 | new.counters = counters; | |
1837 | new.inuse--; | |
a0132ac0 | 1838 | VM_BUG_ON(!new.frozen); |
2cfb7455 | 1839 | |
1d07171c | 1840 | } while (!__cmpxchg_double_slab(s, page, |
2cfb7455 CL |
1841 | prior, counters, |
1842 | freelist, new.counters, | |
1843 | "drain percpu freelist")); | |
1844 | ||
1845 | freelist = nextfree; | |
1846 | } | |
1847 | ||
894b8788 | 1848 | /* |
2cfb7455 CL |
1849 | * Stage two: Ensure that the page is unfrozen while the |
1850 | * list presence reflects the actual number of objects | |
1851 | * during unfreeze. | |
1852 | * | |
1853 | * We setup the list membership and then perform a cmpxchg | |
1854 | * with the count. If there is a mismatch then the page | |
1855 | * is not unfrozen but the page is on the wrong list. | |
1856 | * | |
1857 | * Then we restart the process which may have to remove | |
1858 | * the page from the list that we just put it on again | |
1859 | * because the number of objects in the slab may have | |
1860 | * changed. | |
894b8788 | 1861 | */ |
2cfb7455 | 1862 | redo: |
894b8788 | 1863 | |
2cfb7455 CL |
1864 | old.freelist = page->freelist; |
1865 | old.counters = page->counters; | |
a0132ac0 | 1866 | VM_BUG_ON(!old.frozen); |
7c2e132c | 1867 | |
2cfb7455 CL |
1868 | /* Determine target state of the slab */ |
1869 | new.counters = old.counters; | |
1870 | if (freelist) { | |
1871 | new.inuse--; | |
1872 | set_freepointer(s, freelist, old.freelist); | |
1873 | new.freelist = freelist; | |
1874 | } else | |
1875 | new.freelist = old.freelist; | |
1876 | ||
1877 | new.frozen = 0; | |
1878 | ||
8a5b20ae | 1879 | if (!new.inuse && n->nr_partial >= s->min_partial) |
2cfb7455 CL |
1880 | m = M_FREE; |
1881 | else if (new.freelist) { | |
1882 | m = M_PARTIAL; | |
1883 | if (!lock) { | |
1884 | lock = 1; | |
1885 | /* | |
1886 | * Taking the spinlock removes the possiblity | |
1887 | * that acquire_slab() will see a slab page that | |
1888 | * is frozen | |
1889 | */ | |
1890 | spin_lock(&n->list_lock); | |
1891 | } | |
1892 | } else { | |
1893 | m = M_FULL; | |
1894 | if (kmem_cache_debug(s) && !lock) { | |
1895 | lock = 1; | |
1896 | /* | |
1897 | * This also ensures that the scanning of full | |
1898 | * slabs from diagnostic functions will not see | |
1899 | * any frozen slabs. | |
1900 | */ | |
1901 | spin_lock(&n->list_lock); | |
1902 | } | |
1903 | } | |
1904 | ||
1905 | if (l != m) { | |
1906 | ||
1907 | if (l == M_PARTIAL) | |
1908 | ||
1909 | remove_partial(n, page); | |
1910 | ||
1911 | else if (l == M_FULL) | |
894b8788 | 1912 | |
c65c1877 | 1913 | remove_full(s, n, page); |
2cfb7455 CL |
1914 | |
1915 | if (m == M_PARTIAL) { | |
1916 | ||
1917 | add_partial(n, page, tail); | |
136333d1 | 1918 | stat(s, tail); |
2cfb7455 CL |
1919 | |
1920 | } else if (m == M_FULL) { | |
894b8788 | 1921 | |
2cfb7455 CL |
1922 | stat(s, DEACTIVATE_FULL); |
1923 | add_full(s, n, page); | |
1924 | ||
1925 | } | |
1926 | } | |
1927 | ||
1928 | l = m; | |
1d07171c | 1929 | if (!__cmpxchg_double_slab(s, page, |
2cfb7455 CL |
1930 | old.freelist, old.counters, |
1931 | new.freelist, new.counters, | |
1932 | "unfreezing slab")) | |
1933 | goto redo; | |
1934 | ||
2cfb7455 CL |
1935 | if (lock) |
1936 | spin_unlock(&n->list_lock); | |
1937 | ||
1938 | if (m == M_FREE) { | |
1939 | stat(s, DEACTIVATE_EMPTY); | |
1940 | discard_slab(s, page); | |
1941 | stat(s, FREE_SLAB); | |
894b8788 | 1942 | } |
81819f0f CL |
1943 | } |
1944 | ||
d24ac77f JK |
1945 | /* |
1946 | * Unfreeze all the cpu partial slabs. | |
1947 | * | |
59a09917 CL |
1948 | * This function must be called with interrupts disabled |
1949 | * for the cpu using c (or some other guarantee must be there | |
1950 | * to guarantee no concurrent accesses). | |
d24ac77f | 1951 | */ |
59a09917 CL |
1952 | static void unfreeze_partials(struct kmem_cache *s, |
1953 | struct kmem_cache_cpu *c) | |
49e22585 | 1954 | { |
345c905d | 1955 | #ifdef CONFIG_SLUB_CPU_PARTIAL |
43d77867 | 1956 | struct kmem_cache_node *n = NULL, *n2 = NULL; |
9ada1934 | 1957 | struct page *page, *discard_page = NULL; |
49e22585 CL |
1958 | |
1959 | while ((page = c->partial)) { | |
49e22585 CL |
1960 | struct page new; |
1961 | struct page old; | |
1962 | ||
1963 | c->partial = page->next; | |
43d77867 JK |
1964 | |
1965 | n2 = get_node(s, page_to_nid(page)); | |
1966 | if (n != n2) { | |
1967 | if (n) | |
1968 | spin_unlock(&n->list_lock); | |
1969 | ||
1970 | n = n2; | |
1971 | spin_lock(&n->list_lock); | |
1972 | } | |
49e22585 CL |
1973 | |
1974 | do { | |
1975 | ||
1976 | old.freelist = page->freelist; | |
1977 | old.counters = page->counters; | |
a0132ac0 | 1978 | VM_BUG_ON(!old.frozen); |
49e22585 CL |
1979 | |
1980 | new.counters = old.counters; | |
1981 | new.freelist = old.freelist; | |
1982 | ||
1983 | new.frozen = 0; | |
1984 | ||
d24ac77f | 1985 | } while (!__cmpxchg_double_slab(s, page, |
49e22585 CL |
1986 | old.freelist, old.counters, |
1987 | new.freelist, new.counters, | |
1988 | "unfreezing slab")); | |
1989 | ||
8a5b20ae | 1990 | if (unlikely(!new.inuse && n->nr_partial >= s->min_partial)) { |
9ada1934 SL |
1991 | page->next = discard_page; |
1992 | discard_page = page; | |
43d77867 JK |
1993 | } else { |
1994 | add_partial(n, page, DEACTIVATE_TO_TAIL); | |
1995 | stat(s, FREE_ADD_PARTIAL); | |
49e22585 CL |
1996 | } |
1997 | } | |
1998 | ||
1999 | if (n) | |
2000 | spin_unlock(&n->list_lock); | |
9ada1934 SL |
2001 | |
2002 | while (discard_page) { | |
2003 | page = discard_page; | |
2004 | discard_page = discard_page->next; | |
2005 | ||
2006 | stat(s, DEACTIVATE_EMPTY); | |
2007 | discard_slab(s, page); | |
2008 | stat(s, FREE_SLAB); | |
2009 | } | |
345c905d | 2010 | #endif |
49e22585 CL |
2011 | } |
2012 | ||
2013 | /* | |
2014 | * Put a page that was just frozen (in __slab_free) into a partial page | |
2015 | * slot if available. This is done without interrupts disabled and without | |
2016 | * preemption disabled. The cmpxchg is racy and may put the partial page | |
2017 | * onto a random cpus partial slot. | |
2018 | * | |
2019 | * If we did not find a slot then simply move all the partials to the | |
2020 | * per node partial list. | |
2021 | */ | |
633b0764 | 2022 | static void put_cpu_partial(struct kmem_cache *s, struct page *page, int drain) |
49e22585 | 2023 | { |
345c905d | 2024 | #ifdef CONFIG_SLUB_CPU_PARTIAL |
49e22585 CL |
2025 | struct page *oldpage; |
2026 | int pages; | |
2027 | int pobjects; | |
2028 | ||
2029 | do { | |
2030 | pages = 0; | |
2031 | pobjects = 0; | |
2032 | oldpage = this_cpu_read(s->cpu_slab->partial); | |
2033 | ||
2034 | if (oldpage) { | |
2035 | pobjects = oldpage->pobjects; | |
2036 | pages = oldpage->pages; | |
2037 | if (drain && pobjects > s->cpu_partial) { | |
2038 | unsigned long flags; | |
2039 | /* | |
2040 | * partial array is full. Move the existing | |
2041 | * set to the per node partial list. | |
2042 | */ | |
2043 | local_irq_save(flags); | |
59a09917 | 2044 | unfreeze_partials(s, this_cpu_ptr(s->cpu_slab)); |
49e22585 | 2045 | local_irq_restore(flags); |
e24fc410 | 2046 | oldpage = NULL; |
49e22585 CL |
2047 | pobjects = 0; |
2048 | pages = 0; | |
8028dcea | 2049 | stat(s, CPU_PARTIAL_DRAIN); |
49e22585 CL |
2050 | } |
2051 | } | |
2052 | ||
2053 | pages++; | |
2054 | pobjects += page->objects - page->inuse; | |
2055 | ||
2056 | page->pages = pages; | |
2057 | page->pobjects = pobjects; | |
2058 | page->next = oldpage; | |
2059 | ||
d0e0ac97 CG |
2060 | } while (this_cpu_cmpxchg(s->cpu_slab->partial, oldpage, page) |
2061 | != oldpage); | |
345c905d | 2062 | #endif |
49e22585 CL |
2063 | } |
2064 | ||
dfb4f096 | 2065 | static inline void flush_slab(struct kmem_cache *s, struct kmem_cache_cpu *c) |
81819f0f | 2066 | { |
84e554e6 | 2067 | stat(s, CPUSLAB_FLUSH); |
c17dda40 CL |
2068 | deactivate_slab(s, c->page, c->freelist); |
2069 | ||
2070 | c->tid = next_tid(c->tid); | |
2071 | c->page = NULL; | |
2072 | c->freelist = NULL; | |
81819f0f CL |
2073 | } |
2074 | ||
2075 | /* | |
2076 | * Flush cpu slab. | |
6446faa2 | 2077 | * |
81819f0f CL |
2078 | * Called from IPI handler with interrupts disabled. |
2079 | */ | |
0c710013 | 2080 | static inline void __flush_cpu_slab(struct kmem_cache *s, int cpu) |
81819f0f | 2081 | { |
9dfc6e68 | 2082 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu); |
81819f0f | 2083 | |
49e22585 CL |
2084 | if (likely(c)) { |
2085 | if (c->page) | |
2086 | flush_slab(s, c); | |
2087 | ||
59a09917 | 2088 | unfreeze_partials(s, c); |
49e22585 | 2089 | } |
81819f0f CL |
2090 | } |
2091 | ||
2092 | static void flush_cpu_slab(void *d) | |
2093 | { | |
2094 | struct kmem_cache *s = d; | |
81819f0f | 2095 | |
dfb4f096 | 2096 | __flush_cpu_slab(s, smp_processor_id()); |
81819f0f CL |
2097 | } |
2098 | ||
a8364d55 GBY |
2099 | static bool has_cpu_slab(int cpu, void *info) |
2100 | { | |
2101 | struct kmem_cache *s = info; | |
2102 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu); | |
2103 | ||
02e1a9cd | 2104 | return c->page || c->partial; |
a8364d55 GBY |
2105 | } |
2106 | ||
81819f0f CL |
2107 | static void flush_all(struct kmem_cache *s) |
2108 | { | |
a8364d55 | 2109 | on_each_cpu_cond(has_cpu_slab, flush_cpu_slab, s, 1, GFP_ATOMIC); |
81819f0f CL |
2110 | } |
2111 | ||
dfb4f096 CL |
2112 | /* |
2113 | * Check if the objects in a per cpu structure fit numa | |
2114 | * locality expectations. | |
2115 | */ | |
57d437d2 | 2116 | static inline int node_match(struct page *page, int node) |
dfb4f096 CL |
2117 | { |
2118 | #ifdef CONFIG_NUMA | |
4d7868e6 | 2119 | if (!page || (node != NUMA_NO_NODE && page_to_nid(page) != node)) |
dfb4f096 CL |
2120 | return 0; |
2121 | #endif | |
2122 | return 1; | |
2123 | } | |
2124 | ||
9a02d699 | 2125 | #ifdef CONFIG_SLUB_DEBUG |
781b2ba6 PE |
2126 | static int count_free(struct page *page) |
2127 | { | |
2128 | return page->objects - page->inuse; | |
2129 | } | |
2130 | ||
9a02d699 DR |
2131 | static inline unsigned long node_nr_objs(struct kmem_cache_node *n) |
2132 | { | |
2133 | return atomic_long_read(&n->total_objects); | |
2134 | } | |
2135 | #endif /* CONFIG_SLUB_DEBUG */ | |
2136 | ||
2137 | #if defined(CONFIG_SLUB_DEBUG) || defined(CONFIG_SYSFS) | |
781b2ba6 PE |
2138 | static unsigned long count_partial(struct kmem_cache_node *n, |
2139 | int (*get_count)(struct page *)) | |
2140 | { | |
2141 | unsigned long flags; | |
2142 | unsigned long x = 0; | |
2143 | struct page *page; | |
2144 | ||
2145 | spin_lock_irqsave(&n->list_lock, flags); | |
2146 | list_for_each_entry(page, &n->partial, lru) | |
2147 | x += get_count(page); | |
2148 | spin_unlock_irqrestore(&n->list_lock, flags); | |
2149 | return x; | |
2150 | } | |
9a02d699 | 2151 | #endif /* CONFIG_SLUB_DEBUG || CONFIG_SYSFS */ |
26c02cf0 | 2152 | |
781b2ba6 PE |
2153 | static noinline void |
2154 | slab_out_of_memory(struct kmem_cache *s, gfp_t gfpflags, int nid) | |
2155 | { | |
9a02d699 DR |
2156 | #ifdef CONFIG_SLUB_DEBUG |
2157 | static DEFINE_RATELIMIT_STATE(slub_oom_rs, DEFAULT_RATELIMIT_INTERVAL, | |
2158 | DEFAULT_RATELIMIT_BURST); | |
781b2ba6 | 2159 | int node; |
fa45dc25 | 2160 | struct kmem_cache_node *n; |
781b2ba6 | 2161 | |
9a02d699 DR |
2162 | if ((gfpflags & __GFP_NOWARN) || !__ratelimit(&slub_oom_rs)) |
2163 | return; | |
2164 | ||
f9f58285 | 2165 | pr_warn("SLUB: Unable to allocate memory on node %d (gfp=0x%x)\n", |
781b2ba6 | 2166 | nid, gfpflags); |
f9f58285 FF |
2167 | pr_warn(" cache: %s, object size: %d, buffer size: %d, default order: %d, min order: %d\n", |
2168 | s->name, s->object_size, s->size, oo_order(s->oo), | |
2169 | oo_order(s->min)); | |
781b2ba6 | 2170 | |
3b0efdfa | 2171 | if (oo_order(s->min) > get_order(s->object_size)) |
f9f58285 FF |
2172 | pr_warn(" %s debugging increased min order, use slub_debug=O to disable.\n", |
2173 | s->name); | |
fa5ec8a1 | 2174 | |
fa45dc25 | 2175 | for_each_kmem_cache_node(s, node, n) { |
781b2ba6 PE |
2176 | unsigned long nr_slabs; |
2177 | unsigned long nr_objs; | |
2178 | unsigned long nr_free; | |
2179 | ||
26c02cf0 AB |
2180 | nr_free = count_partial(n, count_free); |
2181 | nr_slabs = node_nr_slabs(n); | |
2182 | nr_objs = node_nr_objs(n); | |
781b2ba6 | 2183 | |
f9f58285 | 2184 | pr_warn(" node %d: slabs: %ld, objs: %ld, free: %ld\n", |
781b2ba6 PE |
2185 | node, nr_slabs, nr_objs, nr_free); |
2186 | } | |
9a02d699 | 2187 | #endif |
781b2ba6 PE |
2188 | } |
2189 | ||
497b66f2 CL |
2190 | static inline void *new_slab_objects(struct kmem_cache *s, gfp_t flags, |
2191 | int node, struct kmem_cache_cpu **pc) | |
2192 | { | |
6faa6833 | 2193 | void *freelist; |
188fd063 CL |
2194 | struct kmem_cache_cpu *c = *pc; |
2195 | struct page *page; | |
497b66f2 | 2196 | |
188fd063 | 2197 | freelist = get_partial(s, flags, node, c); |
497b66f2 | 2198 | |
188fd063 CL |
2199 | if (freelist) |
2200 | return freelist; | |
2201 | ||
2202 | page = new_slab(s, flags, node); | |
497b66f2 | 2203 | if (page) { |
7c8e0181 | 2204 | c = raw_cpu_ptr(s->cpu_slab); |
497b66f2 CL |
2205 | if (c->page) |
2206 | flush_slab(s, c); | |
2207 | ||
2208 | /* | |
2209 | * No other reference to the page yet so we can | |
2210 | * muck around with it freely without cmpxchg | |
2211 | */ | |
6faa6833 | 2212 | freelist = page->freelist; |
497b66f2 CL |
2213 | page->freelist = NULL; |
2214 | ||
2215 | stat(s, ALLOC_SLAB); | |
497b66f2 CL |
2216 | c->page = page; |
2217 | *pc = c; | |
2218 | } else | |
6faa6833 | 2219 | freelist = NULL; |
497b66f2 | 2220 | |
6faa6833 | 2221 | return freelist; |
497b66f2 CL |
2222 | } |
2223 | ||
072bb0aa MG |
2224 | static inline bool pfmemalloc_match(struct page *page, gfp_t gfpflags) |
2225 | { | |
2226 | if (unlikely(PageSlabPfmemalloc(page))) | |
2227 | return gfp_pfmemalloc_allowed(gfpflags); | |
2228 | ||
2229 | return true; | |
2230 | } | |
2231 | ||
213eeb9f | 2232 | /* |
d0e0ac97 CG |
2233 | * Check the page->freelist of a page and either transfer the freelist to the |
2234 | * per cpu freelist or deactivate the page. | |
213eeb9f CL |
2235 | * |
2236 | * The page is still frozen if the return value is not NULL. | |
2237 | * | |
2238 | * If this function returns NULL then the page has been unfrozen. | |
d24ac77f JK |
2239 | * |
2240 | * This function must be called with interrupt disabled. | |
213eeb9f CL |
2241 | */ |
2242 | static inline void *get_freelist(struct kmem_cache *s, struct page *page) | |
2243 | { | |
2244 | struct page new; | |
2245 | unsigned long counters; | |
2246 | void *freelist; | |
2247 | ||
2248 | do { | |
2249 | freelist = page->freelist; | |
2250 | counters = page->counters; | |
6faa6833 | 2251 | |
213eeb9f | 2252 | new.counters = counters; |
a0132ac0 | 2253 | VM_BUG_ON(!new.frozen); |
213eeb9f CL |
2254 | |
2255 | new.inuse = page->objects; | |
2256 | new.frozen = freelist != NULL; | |
2257 | ||
d24ac77f | 2258 | } while (!__cmpxchg_double_slab(s, page, |
213eeb9f CL |
2259 | freelist, counters, |
2260 | NULL, new.counters, | |
2261 | "get_freelist")); | |
2262 | ||
2263 | return freelist; | |
2264 | } | |
2265 | ||
81819f0f | 2266 | /* |
894b8788 CL |
2267 | * Slow path. The lockless freelist is empty or we need to perform |
2268 | * debugging duties. | |
2269 | * | |
894b8788 CL |
2270 | * Processing is still very fast if new objects have been freed to the |
2271 | * regular freelist. In that case we simply take over the regular freelist | |
2272 | * as the lockless freelist and zap the regular freelist. | |
81819f0f | 2273 | * |
894b8788 CL |
2274 | * If that is not working then we fall back to the partial lists. We take the |
2275 | * first element of the freelist as the object to allocate now and move the | |
2276 | * rest of the freelist to the lockless freelist. | |
81819f0f | 2277 | * |
894b8788 | 2278 | * And if we were unable to get a new slab from the partial slab lists then |
6446faa2 CL |
2279 | * we need to allocate a new slab. This is the slowest path since it involves |
2280 | * a call to the page allocator and the setup of a new slab. | |
81819f0f | 2281 | */ |
ce71e27c EGM |
2282 | static void *__slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node, |
2283 | unsigned long addr, struct kmem_cache_cpu *c) | |
81819f0f | 2284 | { |
6faa6833 | 2285 | void *freelist; |
f6e7def7 | 2286 | struct page *page; |
8a5ec0ba CL |
2287 | unsigned long flags; |
2288 | ||
2289 | local_irq_save(flags); | |
2290 | #ifdef CONFIG_PREEMPT | |
2291 | /* | |
2292 | * We may have been preempted and rescheduled on a different | |
2293 | * cpu before disabling interrupts. Need to reload cpu area | |
2294 | * pointer. | |
2295 | */ | |
2296 | c = this_cpu_ptr(s->cpu_slab); | |
8a5ec0ba | 2297 | #endif |
81819f0f | 2298 | |
f6e7def7 CL |
2299 | page = c->page; |
2300 | if (!page) | |
81819f0f | 2301 | goto new_slab; |
49e22585 | 2302 | redo: |
6faa6833 | 2303 | |
57d437d2 | 2304 | if (unlikely(!node_match(page, node))) { |
e36a2652 | 2305 | stat(s, ALLOC_NODE_MISMATCH); |
f6e7def7 | 2306 | deactivate_slab(s, page, c->freelist); |
c17dda40 CL |
2307 | c->page = NULL; |
2308 | c->freelist = NULL; | |
fc59c053 CL |
2309 | goto new_slab; |
2310 | } | |
6446faa2 | 2311 | |
072bb0aa MG |
2312 | /* |
2313 | * By rights, we should be searching for a slab page that was | |
2314 | * PFMEMALLOC but right now, we are losing the pfmemalloc | |
2315 | * information when the page leaves the per-cpu allocator | |
2316 | */ | |
2317 | if (unlikely(!pfmemalloc_match(page, gfpflags))) { | |
2318 | deactivate_slab(s, page, c->freelist); | |
2319 | c->page = NULL; | |
2320 | c->freelist = NULL; | |
2321 | goto new_slab; | |
2322 | } | |
2323 | ||
73736e03 | 2324 | /* must check again c->freelist in case of cpu migration or IRQ */ |
6faa6833 CL |
2325 | freelist = c->freelist; |
2326 | if (freelist) | |
73736e03 | 2327 | goto load_freelist; |
03e404af | 2328 | |
f6e7def7 | 2329 | freelist = get_freelist(s, page); |
6446faa2 | 2330 | |
6faa6833 | 2331 | if (!freelist) { |
03e404af CL |
2332 | c->page = NULL; |
2333 | stat(s, DEACTIVATE_BYPASS); | |
fc59c053 | 2334 | goto new_slab; |
03e404af | 2335 | } |
6446faa2 | 2336 | |
84e554e6 | 2337 | stat(s, ALLOC_REFILL); |
6446faa2 | 2338 | |
894b8788 | 2339 | load_freelist: |
507effea CL |
2340 | /* |
2341 | * freelist is pointing to the list of objects to be used. | |
2342 | * page is pointing to the page from which the objects are obtained. | |
2343 | * That page must be frozen for per cpu allocations to work. | |
2344 | */ | |
a0132ac0 | 2345 | VM_BUG_ON(!c->page->frozen); |
6faa6833 | 2346 | c->freelist = get_freepointer(s, freelist); |
8a5ec0ba CL |
2347 | c->tid = next_tid(c->tid); |
2348 | local_irq_restore(flags); | |
6faa6833 | 2349 | return freelist; |
81819f0f | 2350 | |
81819f0f | 2351 | new_slab: |
2cfb7455 | 2352 | |
49e22585 | 2353 | if (c->partial) { |
f6e7def7 CL |
2354 | page = c->page = c->partial; |
2355 | c->partial = page->next; | |
49e22585 CL |
2356 | stat(s, CPU_PARTIAL_ALLOC); |
2357 | c->freelist = NULL; | |
2358 | goto redo; | |
81819f0f CL |
2359 | } |
2360 | ||
188fd063 | 2361 | freelist = new_slab_objects(s, gfpflags, node, &c); |
01ad8a7b | 2362 | |
f4697436 | 2363 | if (unlikely(!freelist)) { |
9a02d699 | 2364 | slab_out_of_memory(s, gfpflags, node); |
f4697436 CL |
2365 | local_irq_restore(flags); |
2366 | return NULL; | |
81819f0f | 2367 | } |
2cfb7455 | 2368 | |
f6e7def7 | 2369 | page = c->page; |
5091b74a | 2370 | if (likely(!kmem_cache_debug(s) && pfmemalloc_match(page, gfpflags))) |
4b6f0750 | 2371 | goto load_freelist; |
2cfb7455 | 2372 | |
497b66f2 | 2373 | /* Only entered in the debug case */ |
d0e0ac97 CG |
2374 | if (kmem_cache_debug(s) && |
2375 | !alloc_debug_processing(s, page, freelist, addr)) | |
497b66f2 | 2376 | goto new_slab; /* Slab failed checks. Next slab needed */ |
894b8788 | 2377 | |
f6e7def7 | 2378 | deactivate_slab(s, page, get_freepointer(s, freelist)); |
c17dda40 CL |
2379 | c->page = NULL; |
2380 | c->freelist = NULL; | |
a71ae47a | 2381 | local_irq_restore(flags); |
6faa6833 | 2382 | return freelist; |
894b8788 CL |
2383 | } |
2384 | ||
2385 | /* | |
2386 | * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc) | |
2387 | * have the fastpath folded into their functions. So no function call | |
2388 | * overhead for requests that can be satisfied on the fastpath. | |
2389 | * | |
2390 | * The fastpath works by first checking if the lockless freelist can be used. | |
2391 | * If not then __slab_alloc is called for slow processing. | |
2392 | * | |
2393 | * Otherwise we can simply pick the next object from the lockless free list. | |
2394 | */ | |
2b847c3c | 2395 | static __always_inline void *slab_alloc_node(struct kmem_cache *s, |
ce71e27c | 2396 | gfp_t gfpflags, int node, unsigned long addr) |
894b8788 | 2397 | { |
894b8788 | 2398 | void **object; |
dfb4f096 | 2399 | struct kmem_cache_cpu *c; |
57d437d2 | 2400 | struct page *page; |
8a5ec0ba | 2401 | unsigned long tid; |
1f84260c | 2402 | |
c016b0bd | 2403 | if (slab_pre_alloc_hook(s, gfpflags)) |
773ff60e | 2404 | return NULL; |
1f84260c | 2405 | |
d79923fa | 2406 | s = memcg_kmem_get_cache(s, gfpflags); |
8a5ec0ba | 2407 | redo: |
8a5ec0ba CL |
2408 | /* |
2409 | * Must read kmem_cache cpu data via this cpu ptr. Preemption is | |
2410 | * enabled. We may switch back and forth between cpus while | |
2411 | * reading from one cpu area. That does not matter as long | |
2412 | * as we end up on the original cpu again when doing the cmpxchg. | |
7cccd80b CL |
2413 | * |
2414 | * Preemption is disabled for the retrieval of the tid because that | |
2415 | * must occur from the current processor. We cannot allow rescheduling | |
2416 | * on a different processor between the determination of the pointer | |
2417 | * and the retrieval of the tid. | |
8a5ec0ba | 2418 | */ |
7cccd80b | 2419 | preempt_disable(); |
7c8e0181 | 2420 | c = this_cpu_ptr(s->cpu_slab); |
8a5ec0ba | 2421 | |
8a5ec0ba CL |
2422 | /* |
2423 | * The transaction ids are globally unique per cpu and per operation on | |
2424 | * a per cpu queue. Thus they can be guarantee that the cmpxchg_double | |
2425 | * occurs on the right processor and that there was no operation on the | |
2426 | * linked list in between. | |
2427 | */ | |
2428 | tid = c->tid; | |
7cccd80b | 2429 | preempt_enable(); |
8a5ec0ba | 2430 | |
9dfc6e68 | 2431 | object = c->freelist; |
57d437d2 | 2432 | page = c->page; |
8eae1492 | 2433 | if (unlikely(!object || !node_match(page, node))) { |
dfb4f096 | 2434 | object = __slab_alloc(s, gfpflags, node, addr, c); |
8eae1492 DH |
2435 | stat(s, ALLOC_SLOWPATH); |
2436 | } else { | |
0ad9500e ED |
2437 | void *next_object = get_freepointer_safe(s, object); |
2438 | ||
8a5ec0ba | 2439 | /* |
25985edc | 2440 | * The cmpxchg will only match if there was no additional |
8a5ec0ba CL |
2441 | * operation and if we are on the right processor. |
2442 | * | |
d0e0ac97 CG |
2443 | * The cmpxchg does the following atomically (without lock |
2444 | * semantics!) | |
8a5ec0ba CL |
2445 | * 1. Relocate first pointer to the current per cpu area. |
2446 | * 2. Verify that tid and freelist have not been changed | |
2447 | * 3. If they were not changed replace tid and freelist | |
2448 | * | |
d0e0ac97 CG |
2449 | * Since this is without lock semantics the protection is only |
2450 | * against code executing on this cpu *not* from access by | |
2451 | * other cpus. | |
8a5ec0ba | 2452 | */ |
933393f5 | 2453 | if (unlikely(!this_cpu_cmpxchg_double( |
8a5ec0ba CL |
2454 | s->cpu_slab->freelist, s->cpu_slab->tid, |
2455 | object, tid, | |
0ad9500e | 2456 | next_object, next_tid(tid)))) { |
8a5ec0ba CL |
2457 | |
2458 | note_cmpxchg_failure("slab_alloc", s, tid); | |
2459 | goto redo; | |
2460 | } | |
0ad9500e | 2461 | prefetch_freepointer(s, next_object); |
84e554e6 | 2462 | stat(s, ALLOC_FASTPATH); |
894b8788 | 2463 | } |
8a5ec0ba | 2464 | |
74e2134f | 2465 | if (unlikely(gfpflags & __GFP_ZERO) && object) |
3b0efdfa | 2466 | memset(object, 0, s->object_size); |
d07dbea4 | 2467 | |
c016b0bd | 2468 | slab_post_alloc_hook(s, gfpflags, object); |
5a896d9e | 2469 | |
894b8788 | 2470 | return object; |
81819f0f CL |
2471 | } |
2472 | ||
2b847c3c EG |
2473 | static __always_inline void *slab_alloc(struct kmem_cache *s, |
2474 | gfp_t gfpflags, unsigned long addr) | |
2475 | { | |
2476 | return slab_alloc_node(s, gfpflags, NUMA_NO_NODE, addr); | |
2477 | } | |
2478 | ||
81819f0f CL |
2479 | void *kmem_cache_alloc(struct kmem_cache *s, gfp_t gfpflags) |
2480 | { | |
2b847c3c | 2481 | void *ret = slab_alloc(s, gfpflags, _RET_IP_); |
5b882be4 | 2482 | |
d0e0ac97 CG |
2483 | trace_kmem_cache_alloc(_RET_IP_, ret, s->object_size, |
2484 | s->size, gfpflags); | |
5b882be4 EGM |
2485 | |
2486 | return ret; | |
81819f0f CL |
2487 | } |
2488 | EXPORT_SYMBOL(kmem_cache_alloc); | |
2489 | ||
0f24f128 | 2490 | #ifdef CONFIG_TRACING |
4a92379b RK |
2491 | void *kmem_cache_alloc_trace(struct kmem_cache *s, gfp_t gfpflags, size_t size) |
2492 | { | |
2b847c3c | 2493 | void *ret = slab_alloc(s, gfpflags, _RET_IP_); |
4a92379b RK |
2494 | trace_kmalloc(_RET_IP_, ret, size, s->size, gfpflags); |
2495 | return ret; | |
2496 | } | |
2497 | EXPORT_SYMBOL(kmem_cache_alloc_trace); | |
5b882be4 EGM |
2498 | #endif |
2499 | ||
81819f0f CL |
2500 | #ifdef CONFIG_NUMA |
2501 | void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t gfpflags, int node) | |
2502 | { | |
2b847c3c | 2503 | void *ret = slab_alloc_node(s, gfpflags, node, _RET_IP_); |
5b882be4 | 2504 | |
ca2b84cb | 2505 | trace_kmem_cache_alloc_node(_RET_IP_, ret, |
3b0efdfa | 2506 | s->object_size, s->size, gfpflags, node); |
5b882be4 EGM |
2507 | |
2508 | return ret; | |
81819f0f CL |
2509 | } |
2510 | EXPORT_SYMBOL(kmem_cache_alloc_node); | |
81819f0f | 2511 | |
0f24f128 | 2512 | #ifdef CONFIG_TRACING |
4a92379b | 2513 | void *kmem_cache_alloc_node_trace(struct kmem_cache *s, |
5b882be4 | 2514 | gfp_t gfpflags, |
4a92379b | 2515 | int node, size_t size) |
5b882be4 | 2516 | { |
2b847c3c | 2517 | void *ret = slab_alloc_node(s, gfpflags, node, _RET_IP_); |
4a92379b RK |
2518 | |
2519 | trace_kmalloc_node(_RET_IP_, ret, | |
2520 | size, s->size, gfpflags, node); | |
2521 | return ret; | |
5b882be4 | 2522 | } |
4a92379b | 2523 | EXPORT_SYMBOL(kmem_cache_alloc_node_trace); |
5b882be4 | 2524 | #endif |
5d1f57e4 | 2525 | #endif |
5b882be4 | 2526 | |
81819f0f | 2527 | /* |
894b8788 CL |
2528 | * Slow patch handling. This may still be called frequently since objects |
2529 | * have a longer lifetime than the cpu slabs in most processing loads. | |
81819f0f | 2530 | * |
894b8788 CL |
2531 | * So we still attempt to reduce cache line usage. Just take the slab |
2532 | * lock and free the item. If there is no additional partial page | |
2533 | * handling required then we can return immediately. | |
81819f0f | 2534 | */ |
894b8788 | 2535 | static void __slab_free(struct kmem_cache *s, struct page *page, |
ff12059e | 2536 | void *x, unsigned long addr) |
81819f0f CL |
2537 | { |
2538 | void *prior; | |
2539 | void **object = (void *)x; | |
2cfb7455 | 2540 | int was_frozen; |
2cfb7455 CL |
2541 | struct page new; |
2542 | unsigned long counters; | |
2543 | struct kmem_cache_node *n = NULL; | |
61728d1e | 2544 | unsigned long uninitialized_var(flags); |
81819f0f | 2545 | |
8a5ec0ba | 2546 | stat(s, FREE_SLOWPATH); |
81819f0f | 2547 | |
19c7ff9e CL |
2548 | if (kmem_cache_debug(s) && |
2549 | !(n = free_debug_processing(s, page, x, addr, &flags))) | |
80f08c19 | 2550 | return; |
6446faa2 | 2551 | |
2cfb7455 | 2552 | do { |
837d678d JK |
2553 | if (unlikely(n)) { |
2554 | spin_unlock_irqrestore(&n->list_lock, flags); | |
2555 | n = NULL; | |
2556 | } | |
2cfb7455 CL |
2557 | prior = page->freelist; |
2558 | counters = page->counters; | |
2559 | set_freepointer(s, object, prior); | |
2560 | new.counters = counters; | |
2561 | was_frozen = new.frozen; | |
2562 | new.inuse--; | |
837d678d | 2563 | if ((!new.inuse || !prior) && !was_frozen) { |
49e22585 | 2564 | |
c65c1877 | 2565 | if (kmem_cache_has_cpu_partial(s) && !prior) { |
49e22585 CL |
2566 | |
2567 | /* | |
d0e0ac97 CG |
2568 | * Slab was on no list before and will be |
2569 | * partially empty | |
2570 | * We can defer the list move and instead | |
2571 | * freeze it. | |
49e22585 CL |
2572 | */ |
2573 | new.frozen = 1; | |
2574 | ||
c65c1877 | 2575 | } else { /* Needs to be taken off a list */ |
49e22585 CL |
2576 | |
2577 | n = get_node(s, page_to_nid(page)); | |
2578 | /* | |
2579 | * Speculatively acquire the list_lock. | |
2580 | * If the cmpxchg does not succeed then we may | |
2581 | * drop the list_lock without any processing. | |
2582 | * | |
2583 | * Otherwise the list_lock will synchronize with | |
2584 | * other processors updating the list of slabs. | |
2585 | */ | |
2586 | spin_lock_irqsave(&n->list_lock, flags); | |
2587 | ||
2588 | } | |
2cfb7455 | 2589 | } |
81819f0f | 2590 | |
2cfb7455 CL |
2591 | } while (!cmpxchg_double_slab(s, page, |
2592 | prior, counters, | |
2593 | object, new.counters, | |
2594 | "__slab_free")); | |
81819f0f | 2595 | |
2cfb7455 | 2596 | if (likely(!n)) { |
49e22585 CL |
2597 | |
2598 | /* | |
2599 | * If we just froze the page then put it onto the | |
2600 | * per cpu partial list. | |
2601 | */ | |
8028dcea | 2602 | if (new.frozen && !was_frozen) { |
49e22585 | 2603 | put_cpu_partial(s, page, 1); |
8028dcea AS |
2604 | stat(s, CPU_PARTIAL_FREE); |
2605 | } | |
49e22585 | 2606 | /* |
2cfb7455 CL |
2607 | * The list lock was not taken therefore no list |
2608 | * activity can be necessary. | |
2609 | */ | |
2610 | if (was_frozen) | |
2611 | stat(s, FREE_FROZEN); | |
80f08c19 | 2612 | return; |
2cfb7455 | 2613 | } |
81819f0f | 2614 | |
8a5b20ae | 2615 | if (unlikely(!new.inuse && n->nr_partial >= s->min_partial)) |
837d678d JK |
2616 | goto slab_empty; |
2617 | ||
81819f0f | 2618 | /* |
837d678d JK |
2619 | * Objects left in the slab. If it was not on the partial list before |
2620 | * then add it. | |
81819f0f | 2621 | */ |
345c905d JK |
2622 | if (!kmem_cache_has_cpu_partial(s) && unlikely(!prior)) { |
2623 | if (kmem_cache_debug(s)) | |
c65c1877 | 2624 | remove_full(s, n, page); |
837d678d JK |
2625 | add_partial(n, page, DEACTIVATE_TO_TAIL); |
2626 | stat(s, FREE_ADD_PARTIAL); | |
8ff12cfc | 2627 | } |
80f08c19 | 2628 | spin_unlock_irqrestore(&n->list_lock, flags); |
81819f0f CL |
2629 | return; |
2630 | ||
2631 | slab_empty: | |
a973e9dd | 2632 | if (prior) { |
81819f0f | 2633 | /* |
6fbabb20 | 2634 | * Slab on the partial list. |
81819f0f | 2635 | */ |
5cc6eee8 | 2636 | remove_partial(n, page); |
84e554e6 | 2637 | stat(s, FREE_REMOVE_PARTIAL); |
c65c1877 | 2638 | } else { |
6fbabb20 | 2639 | /* Slab must be on the full list */ |
c65c1877 PZ |
2640 | remove_full(s, n, page); |
2641 | } | |
2cfb7455 | 2642 | |
80f08c19 | 2643 | spin_unlock_irqrestore(&n->list_lock, flags); |
84e554e6 | 2644 | stat(s, FREE_SLAB); |
81819f0f | 2645 | discard_slab(s, page); |
81819f0f CL |
2646 | } |
2647 | ||
894b8788 CL |
2648 | /* |
2649 | * Fastpath with forced inlining to produce a kfree and kmem_cache_free that | |
2650 | * can perform fastpath freeing without additional function calls. | |
2651 | * | |
2652 | * The fastpath is only possible if we are freeing to the current cpu slab | |
2653 | * of this processor. This typically the case if we have just allocated | |
2654 | * the item before. | |
2655 | * | |
2656 | * If fastpath is not possible then fall back to __slab_free where we deal | |
2657 | * with all sorts of special processing. | |
2658 | */ | |
06428780 | 2659 | static __always_inline void slab_free(struct kmem_cache *s, |
ce71e27c | 2660 | struct page *page, void *x, unsigned long addr) |
894b8788 CL |
2661 | { |
2662 | void **object = (void *)x; | |
dfb4f096 | 2663 | struct kmem_cache_cpu *c; |
8a5ec0ba | 2664 | unsigned long tid; |
1f84260c | 2665 | |
c016b0bd CL |
2666 | slab_free_hook(s, x); |
2667 | ||
8a5ec0ba CL |
2668 | redo: |
2669 | /* | |
2670 | * Determine the currently cpus per cpu slab. | |
2671 | * The cpu may change afterward. However that does not matter since | |
2672 | * data is retrieved via this pointer. If we are on the same cpu | |
2673 | * during the cmpxchg then the free will succedd. | |
2674 | */ | |
7cccd80b | 2675 | preempt_disable(); |
7c8e0181 | 2676 | c = this_cpu_ptr(s->cpu_slab); |
c016b0bd | 2677 | |
8a5ec0ba | 2678 | tid = c->tid; |
7cccd80b | 2679 | preempt_enable(); |
c016b0bd | 2680 | |
442b06bc | 2681 | if (likely(page == c->page)) { |
ff12059e | 2682 | set_freepointer(s, object, c->freelist); |
8a5ec0ba | 2683 | |
933393f5 | 2684 | if (unlikely(!this_cpu_cmpxchg_double( |
8a5ec0ba CL |
2685 | s->cpu_slab->freelist, s->cpu_slab->tid, |
2686 | c->freelist, tid, | |
2687 | object, next_tid(tid)))) { | |
2688 | ||
2689 | note_cmpxchg_failure("slab_free", s, tid); | |
2690 | goto redo; | |
2691 | } | |
84e554e6 | 2692 | stat(s, FREE_FASTPATH); |
894b8788 | 2693 | } else |
ff12059e | 2694 | __slab_free(s, page, x, addr); |
894b8788 | 2695 | |
894b8788 CL |
2696 | } |
2697 | ||
81819f0f CL |
2698 | void kmem_cache_free(struct kmem_cache *s, void *x) |
2699 | { | |
b9ce5ef4 GC |
2700 | s = cache_from_obj(s, x); |
2701 | if (!s) | |
79576102 | 2702 | return; |
b9ce5ef4 | 2703 | slab_free(s, virt_to_head_page(x), x, _RET_IP_); |
ca2b84cb | 2704 | trace_kmem_cache_free(_RET_IP_, x); |
81819f0f CL |
2705 | } |
2706 | EXPORT_SYMBOL(kmem_cache_free); | |
2707 | ||
81819f0f | 2708 | /* |
672bba3a CL |
2709 | * Object placement in a slab is made very easy because we always start at |
2710 | * offset 0. If we tune the size of the object to the alignment then we can | |
2711 | * get the required alignment by putting one properly sized object after | |
2712 | * another. | |
81819f0f CL |
2713 | * |
2714 | * Notice that the allocation order determines the sizes of the per cpu | |
2715 | * caches. Each processor has always one slab available for allocations. | |
2716 | * Increasing the allocation order reduces the number of times that slabs | |
672bba3a | 2717 | * must be moved on and off the partial lists and is therefore a factor in |
81819f0f | 2718 | * locking overhead. |
81819f0f CL |
2719 | */ |
2720 | ||
2721 | /* | |
2722 | * Mininum / Maximum order of slab pages. This influences locking overhead | |
2723 | * and slab fragmentation. A higher order reduces the number of partial slabs | |
2724 | * and increases the number of allocations possible without having to | |
2725 | * take the list_lock. | |
2726 | */ | |
2727 | static int slub_min_order; | |
114e9e89 | 2728 | static int slub_max_order = PAGE_ALLOC_COSTLY_ORDER; |
9b2cd506 | 2729 | static int slub_min_objects; |
81819f0f CL |
2730 | |
2731 | /* | |
2732 | * Merge control. If this is set then no merging of slab caches will occur. | |
672bba3a | 2733 | * (Could be removed. This was introduced to pacify the merge skeptics.) |
81819f0f CL |
2734 | */ |
2735 | static int slub_nomerge; | |
2736 | ||
81819f0f CL |
2737 | /* |
2738 | * Calculate the order of allocation given an slab object size. | |
2739 | * | |
672bba3a CL |
2740 | * The order of allocation has significant impact on performance and other |
2741 | * system components. Generally order 0 allocations should be preferred since | |
2742 | * order 0 does not cause fragmentation in the page allocator. Larger objects | |
2743 | * be problematic to put into order 0 slabs because there may be too much | |
c124f5b5 | 2744 | * unused space left. We go to a higher order if more than 1/16th of the slab |
672bba3a CL |
2745 | * would be wasted. |
2746 | * | |
2747 | * In order to reach satisfactory performance we must ensure that a minimum | |
2748 | * number of objects is in one slab. Otherwise we may generate too much | |
2749 | * activity on the partial lists which requires taking the list_lock. This is | |
2750 | * less a concern for large slabs though which are rarely used. | |
81819f0f | 2751 | * |
672bba3a CL |
2752 | * slub_max_order specifies the order where we begin to stop considering the |
2753 | * number of objects in a slab as critical. If we reach slub_max_order then | |
2754 | * we try to keep the page order as low as possible. So we accept more waste | |
2755 | * of space in favor of a small page order. | |
81819f0f | 2756 | * |
672bba3a CL |
2757 | * Higher order allocations also allow the placement of more objects in a |
2758 | * slab and thereby reduce object handling overhead. If the user has | |
2759 | * requested a higher mininum order then we start with that one instead of | |
2760 | * the smallest order which will fit the object. | |
81819f0f | 2761 | */ |
5e6d444e | 2762 | static inline int slab_order(int size, int min_objects, |
ab9a0f19 | 2763 | int max_order, int fract_leftover, int reserved) |
81819f0f CL |
2764 | { |
2765 | int order; | |
2766 | int rem; | |
6300ea75 | 2767 | int min_order = slub_min_order; |
81819f0f | 2768 | |
ab9a0f19 | 2769 | if (order_objects(min_order, size, reserved) > MAX_OBJS_PER_PAGE) |
210b5c06 | 2770 | return get_order(size * MAX_OBJS_PER_PAGE) - 1; |
39b26464 | 2771 | |
6300ea75 | 2772 | for (order = max(min_order, |
5e6d444e CL |
2773 | fls(min_objects * size - 1) - PAGE_SHIFT); |
2774 | order <= max_order; order++) { | |
81819f0f | 2775 | |
5e6d444e | 2776 | unsigned long slab_size = PAGE_SIZE << order; |
81819f0f | 2777 | |
ab9a0f19 | 2778 | if (slab_size < min_objects * size + reserved) |
81819f0f CL |
2779 | continue; |
2780 | ||
ab9a0f19 | 2781 | rem = (slab_size - reserved) % size; |
81819f0f | 2782 | |
5e6d444e | 2783 | if (rem <= slab_size / fract_leftover) |
81819f0f CL |
2784 | break; |
2785 | ||
2786 | } | |
672bba3a | 2787 | |
81819f0f CL |
2788 | return order; |
2789 | } | |
2790 | ||
ab9a0f19 | 2791 | static inline int calculate_order(int size, int reserved) |
5e6d444e CL |
2792 | { |
2793 | int order; | |
2794 | int min_objects; | |
2795 | int fraction; | |
e8120ff1 | 2796 | int max_objects; |
5e6d444e CL |
2797 | |
2798 | /* | |
2799 | * Attempt to find best configuration for a slab. This | |
2800 | * works by first attempting to generate a layout with | |
2801 | * the best configuration and backing off gradually. | |
2802 | * | |
2803 | * First we reduce the acceptable waste in a slab. Then | |
2804 | * we reduce the minimum objects required in a slab. | |
2805 | */ | |
2806 | min_objects = slub_min_objects; | |
9b2cd506 CL |
2807 | if (!min_objects) |
2808 | min_objects = 4 * (fls(nr_cpu_ids) + 1); | |
ab9a0f19 | 2809 | max_objects = order_objects(slub_max_order, size, reserved); |
e8120ff1 ZY |
2810 | min_objects = min(min_objects, max_objects); |
2811 | ||
5e6d444e | 2812 | while (min_objects > 1) { |
c124f5b5 | 2813 | fraction = 16; |
5e6d444e CL |
2814 | while (fraction >= 4) { |
2815 | order = slab_order(size, min_objects, | |
ab9a0f19 | 2816 | slub_max_order, fraction, reserved); |
5e6d444e CL |
2817 | if (order <= slub_max_order) |
2818 | return order; | |
2819 | fraction /= 2; | |
2820 | } | |
5086c389 | 2821 | min_objects--; |
5e6d444e CL |
2822 | } |
2823 | ||
2824 | /* | |
2825 | * We were unable to place multiple objects in a slab. Now | |
2826 | * lets see if we can place a single object there. | |
2827 | */ | |
ab9a0f19 | 2828 | order = slab_order(size, 1, slub_max_order, 1, reserved); |
5e6d444e CL |
2829 | if (order <= slub_max_order) |
2830 | return order; | |
2831 | ||
2832 | /* | |
2833 | * Doh this slab cannot be placed using slub_max_order. | |
2834 | */ | |
ab9a0f19 | 2835 | order = slab_order(size, 1, MAX_ORDER, 1, reserved); |
818cf590 | 2836 | if (order < MAX_ORDER) |
5e6d444e CL |
2837 | return order; |
2838 | return -ENOSYS; | |
2839 | } | |
2840 | ||
5595cffc | 2841 | static void |
4053497d | 2842 | init_kmem_cache_node(struct kmem_cache_node *n) |
81819f0f CL |
2843 | { |
2844 | n->nr_partial = 0; | |
81819f0f CL |
2845 | spin_lock_init(&n->list_lock); |
2846 | INIT_LIST_HEAD(&n->partial); | |
8ab1372f | 2847 | #ifdef CONFIG_SLUB_DEBUG |
0f389ec6 | 2848 | atomic_long_set(&n->nr_slabs, 0); |
02b71b70 | 2849 | atomic_long_set(&n->total_objects, 0); |
643b1138 | 2850 | INIT_LIST_HEAD(&n->full); |
8ab1372f | 2851 | #endif |
81819f0f CL |
2852 | } |
2853 | ||
55136592 | 2854 | static inline int alloc_kmem_cache_cpus(struct kmem_cache *s) |
4c93c355 | 2855 | { |
6c182dc0 | 2856 | BUILD_BUG_ON(PERCPU_DYNAMIC_EARLY_SIZE < |
95a05b42 | 2857 | KMALLOC_SHIFT_HIGH * sizeof(struct kmem_cache_cpu)); |
4c93c355 | 2858 | |
8a5ec0ba | 2859 | /* |
d4d84fef CM |
2860 | * Must align to double word boundary for the double cmpxchg |
2861 | * instructions to work; see __pcpu_double_call_return_bool(). | |
8a5ec0ba | 2862 | */ |
d4d84fef CM |
2863 | s->cpu_slab = __alloc_percpu(sizeof(struct kmem_cache_cpu), |
2864 | 2 * sizeof(void *)); | |
8a5ec0ba CL |
2865 | |
2866 | if (!s->cpu_slab) | |
2867 | return 0; | |
2868 | ||
2869 | init_kmem_cache_cpus(s); | |
4c93c355 | 2870 | |
8a5ec0ba | 2871 | return 1; |
4c93c355 | 2872 | } |
4c93c355 | 2873 | |
51df1142 CL |
2874 | static struct kmem_cache *kmem_cache_node; |
2875 | ||
81819f0f CL |
2876 | /* |
2877 | * No kmalloc_node yet so do it by hand. We know that this is the first | |
2878 | * slab on the node for this slabcache. There are no concurrent accesses | |
2879 | * possible. | |
2880 | * | |
721ae22a ZYW |
2881 | * Note that this function only works on the kmem_cache_node |
2882 | * when allocating for the kmem_cache_node. This is used for bootstrapping | |
4c93c355 | 2883 | * memory on a fresh node that has no slab structures yet. |
81819f0f | 2884 | */ |
55136592 | 2885 | static void early_kmem_cache_node_alloc(int node) |
81819f0f CL |
2886 | { |
2887 | struct page *page; | |
2888 | struct kmem_cache_node *n; | |
2889 | ||
51df1142 | 2890 | BUG_ON(kmem_cache_node->size < sizeof(struct kmem_cache_node)); |
81819f0f | 2891 | |
51df1142 | 2892 | page = new_slab(kmem_cache_node, GFP_NOWAIT, node); |
81819f0f CL |
2893 | |
2894 | BUG_ON(!page); | |
a2f92ee7 | 2895 | if (page_to_nid(page) != node) { |
f9f58285 FF |
2896 | pr_err("SLUB: Unable to allocate memory from node %d\n", node); |
2897 | pr_err("SLUB: Allocating a useless per node structure in order to be able to continue\n"); | |
a2f92ee7 CL |
2898 | } |
2899 | ||
81819f0f CL |
2900 | n = page->freelist; |
2901 | BUG_ON(!n); | |
51df1142 | 2902 | page->freelist = get_freepointer(kmem_cache_node, n); |
e6e82ea1 | 2903 | page->inuse = 1; |
8cb0a506 | 2904 | page->frozen = 0; |
51df1142 | 2905 | kmem_cache_node->node[node] = n; |
8ab1372f | 2906 | #ifdef CONFIG_SLUB_DEBUG |
f7cb1933 | 2907 | init_object(kmem_cache_node, n, SLUB_RED_ACTIVE); |
51df1142 | 2908 | init_tracking(kmem_cache_node, n); |
8ab1372f | 2909 | #endif |
4053497d | 2910 | init_kmem_cache_node(n); |
51df1142 | 2911 | inc_slabs_node(kmem_cache_node, node, page->objects); |
6446faa2 | 2912 | |
67b6c900 | 2913 | /* |
1e4dd946 SR |
2914 | * No locks need to be taken here as it has just been |
2915 | * initialized and there is no concurrent access. | |
67b6c900 | 2916 | */ |
1e4dd946 | 2917 | __add_partial(n, page, DEACTIVATE_TO_HEAD); |
81819f0f CL |
2918 | } |
2919 | ||
2920 | static void free_kmem_cache_nodes(struct kmem_cache *s) | |
2921 | { | |
2922 | int node; | |
fa45dc25 | 2923 | struct kmem_cache_node *n; |
81819f0f | 2924 | |
fa45dc25 CL |
2925 | for_each_kmem_cache_node(s, node, n) { |
2926 | kmem_cache_free(kmem_cache_node, n); | |
81819f0f CL |
2927 | s->node[node] = NULL; |
2928 | } | |
2929 | } | |
2930 | ||
55136592 | 2931 | static int init_kmem_cache_nodes(struct kmem_cache *s) |
81819f0f CL |
2932 | { |
2933 | int node; | |
81819f0f | 2934 | |
f64dc58c | 2935 | for_each_node_state(node, N_NORMAL_MEMORY) { |
81819f0f CL |
2936 | struct kmem_cache_node *n; |
2937 | ||
73367bd8 | 2938 | if (slab_state == DOWN) { |
55136592 | 2939 | early_kmem_cache_node_alloc(node); |
73367bd8 AD |
2940 | continue; |
2941 | } | |
51df1142 | 2942 | n = kmem_cache_alloc_node(kmem_cache_node, |
55136592 | 2943 | GFP_KERNEL, node); |
81819f0f | 2944 | |
73367bd8 AD |
2945 | if (!n) { |
2946 | free_kmem_cache_nodes(s); | |
2947 | return 0; | |
81819f0f | 2948 | } |
73367bd8 | 2949 | |
81819f0f | 2950 | s->node[node] = n; |
4053497d | 2951 | init_kmem_cache_node(n); |
81819f0f CL |
2952 | } |
2953 | return 1; | |
2954 | } | |
81819f0f | 2955 | |
c0bdb232 | 2956 | static void set_min_partial(struct kmem_cache *s, unsigned long min) |
3b89d7d8 DR |
2957 | { |
2958 | if (min < MIN_PARTIAL) | |
2959 | min = MIN_PARTIAL; | |
2960 | else if (min > MAX_PARTIAL) | |
2961 | min = MAX_PARTIAL; | |
2962 | s->min_partial = min; | |
2963 | } | |
2964 | ||
81819f0f CL |
2965 | /* |
2966 | * calculate_sizes() determines the order and the distribution of data within | |
2967 | * a slab object. | |
2968 | */ | |
06b285dc | 2969 | static int calculate_sizes(struct kmem_cache *s, int forced_order) |
81819f0f CL |
2970 | { |
2971 | unsigned long flags = s->flags; | |
3b0efdfa | 2972 | unsigned long size = s->object_size; |
834f3d11 | 2973 | int order; |
81819f0f | 2974 | |
d8b42bf5 CL |
2975 | /* |
2976 | * Round up object size to the next word boundary. We can only | |
2977 | * place the free pointer at word boundaries and this determines | |
2978 | * the possible location of the free pointer. | |
2979 | */ | |
2980 | size = ALIGN(size, sizeof(void *)); | |
2981 | ||
2982 | #ifdef CONFIG_SLUB_DEBUG | |
81819f0f CL |
2983 | /* |
2984 | * Determine if we can poison the object itself. If the user of | |
2985 | * the slab may touch the object after free or before allocation | |
2986 | * then we should never poison the object itself. | |
2987 | */ | |
2988 | if ((flags & SLAB_POISON) && !(flags & SLAB_DESTROY_BY_RCU) && | |
c59def9f | 2989 | !s->ctor) |
81819f0f CL |
2990 | s->flags |= __OBJECT_POISON; |
2991 | else | |
2992 | s->flags &= ~__OBJECT_POISON; | |
2993 | ||
81819f0f CL |
2994 | |
2995 | /* | |
672bba3a | 2996 | * If we are Redzoning then check if there is some space between the |
81819f0f | 2997 | * end of the object and the free pointer. If not then add an |
672bba3a | 2998 | * additional word to have some bytes to store Redzone information. |
81819f0f | 2999 | */ |
3b0efdfa | 3000 | if ((flags & SLAB_RED_ZONE) && size == s->object_size) |
81819f0f | 3001 | size += sizeof(void *); |
41ecc55b | 3002 | #endif |
81819f0f CL |
3003 | |
3004 | /* | |
672bba3a CL |
3005 | * With that we have determined the number of bytes in actual use |
3006 | * by the object. This is the potential offset to the free pointer. | |
81819f0f CL |
3007 | */ |
3008 | s->inuse = size; | |
3009 | ||
3010 | if (((flags & (SLAB_DESTROY_BY_RCU | SLAB_POISON)) || | |
c59def9f | 3011 | s->ctor)) { |
81819f0f CL |
3012 | /* |
3013 | * Relocate free pointer after the object if it is not | |
3014 | * permitted to overwrite the first word of the object on | |
3015 | * kmem_cache_free. | |
3016 | * | |
3017 | * This is the case if we do RCU, have a constructor or | |
3018 | * destructor or are poisoning the objects. | |
3019 | */ | |
3020 | s->offset = size; | |
3021 | size += sizeof(void *); | |
3022 | } | |
3023 | ||
c12b3c62 | 3024 | #ifdef CONFIG_SLUB_DEBUG |
81819f0f CL |
3025 | if (flags & SLAB_STORE_USER) |
3026 | /* | |
3027 | * Need to store information about allocs and frees after | |
3028 | * the object. | |
3029 | */ | |
3030 | size += 2 * sizeof(struct track); | |
3031 | ||
be7b3fbc | 3032 | if (flags & SLAB_RED_ZONE) |
81819f0f CL |
3033 | /* |
3034 | * Add some empty padding so that we can catch | |
3035 | * overwrites from earlier objects rather than let | |
3036 | * tracking information or the free pointer be | |
0211a9c8 | 3037 | * corrupted if a user writes before the start |
81819f0f CL |
3038 | * of the object. |
3039 | */ | |
3040 | size += sizeof(void *); | |
41ecc55b | 3041 | #endif |
672bba3a | 3042 | |
81819f0f CL |
3043 | /* |
3044 | * SLUB stores one object immediately after another beginning from | |
3045 | * offset 0. In order to align the objects we have to simply size | |
3046 | * each object to conform to the alignment. | |
3047 | */ | |
45906855 | 3048 | size = ALIGN(size, s->align); |
81819f0f | 3049 | s->size = size; |
06b285dc CL |
3050 | if (forced_order >= 0) |
3051 | order = forced_order; | |
3052 | else | |
ab9a0f19 | 3053 | order = calculate_order(size, s->reserved); |
81819f0f | 3054 | |
834f3d11 | 3055 | if (order < 0) |
81819f0f CL |
3056 | return 0; |
3057 | ||
b7a49f0d | 3058 | s->allocflags = 0; |
834f3d11 | 3059 | if (order) |
b7a49f0d CL |
3060 | s->allocflags |= __GFP_COMP; |
3061 | ||
3062 | if (s->flags & SLAB_CACHE_DMA) | |
2c59dd65 | 3063 | s->allocflags |= GFP_DMA; |
b7a49f0d CL |
3064 | |
3065 | if (s->flags & SLAB_RECLAIM_ACCOUNT) | |
3066 | s->allocflags |= __GFP_RECLAIMABLE; | |
3067 | ||
81819f0f CL |
3068 | /* |
3069 | * Determine the number of objects per slab | |
3070 | */ | |
ab9a0f19 LJ |
3071 | s->oo = oo_make(order, size, s->reserved); |
3072 | s->min = oo_make(get_order(size), size, s->reserved); | |
205ab99d CL |
3073 | if (oo_objects(s->oo) > oo_objects(s->max)) |
3074 | s->max = s->oo; | |
81819f0f | 3075 | |
834f3d11 | 3076 | return !!oo_objects(s->oo); |
81819f0f CL |
3077 | } |
3078 | ||
8a13a4cc | 3079 | static int kmem_cache_open(struct kmem_cache *s, unsigned long flags) |
81819f0f | 3080 | { |
8a13a4cc | 3081 | s->flags = kmem_cache_flags(s->size, flags, s->name, s->ctor); |
ab9a0f19 | 3082 | s->reserved = 0; |
81819f0f | 3083 | |
da9a638c LJ |
3084 | if (need_reserve_slab_rcu && (s->flags & SLAB_DESTROY_BY_RCU)) |
3085 | s->reserved = sizeof(struct rcu_head); | |
81819f0f | 3086 | |
06b285dc | 3087 | if (!calculate_sizes(s, -1)) |
81819f0f | 3088 | goto error; |
3de47213 DR |
3089 | if (disable_higher_order_debug) { |
3090 | /* | |
3091 | * Disable debugging flags that store metadata if the min slab | |
3092 | * order increased. | |
3093 | */ | |
3b0efdfa | 3094 | if (get_order(s->size) > get_order(s->object_size)) { |
3de47213 DR |
3095 | s->flags &= ~DEBUG_METADATA_FLAGS; |
3096 | s->offset = 0; | |
3097 | if (!calculate_sizes(s, -1)) | |
3098 | goto error; | |
3099 | } | |
3100 | } | |
81819f0f | 3101 | |
2565409f HC |
3102 | #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \ |
3103 | defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE) | |
b789ef51 CL |
3104 | if (system_has_cmpxchg_double() && (s->flags & SLAB_DEBUG_FLAGS) == 0) |
3105 | /* Enable fast mode */ | |
3106 | s->flags |= __CMPXCHG_DOUBLE; | |
3107 | #endif | |
3108 | ||
3b89d7d8 DR |
3109 | /* |
3110 | * The larger the object size is, the more pages we want on the partial | |
3111 | * list to avoid pounding the page allocator excessively. | |
3112 | */ | |
49e22585 CL |
3113 | set_min_partial(s, ilog2(s->size) / 2); |
3114 | ||
3115 | /* | |
3116 | * cpu_partial determined the maximum number of objects kept in the | |
3117 | * per cpu partial lists of a processor. | |
3118 | * | |
3119 | * Per cpu partial lists mainly contain slabs that just have one | |
3120 | * object freed. If they are used for allocation then they can be | |
3121 | * filled up again with minimal effort. The slab will never hit the | |
3122 | * per node partial lists and therefore no locking will be required. | |
3123 | * | |
3124 | * This setting also determines | |
3125 | * | |
3126 | * A) The number of objects from per cpu partial slabs dumped to the | |
3127 | * per node list when we reach the limit. | |
9f264904 | 3128 | * B) The number of objects in cpu partial slabs to extract from the |
d0e0ac97 CG |
3129 | * per node list when we run out of per cpu objects. We only fetch |
3130 | * 50% to keep some capacity around for frees. | |
49e22585 | 3131 | */ |
345c905d | 3132 | if (!kmem_cache_has_cpu_partial(s)) |
8f1e33da CL |
3133 | s->cpu_partial = 0; |
3134 | else if (s->size >= PAGE_SIZE) | |
49e22585 CL |
3135 | s->cpu_partial = 2; |
3136 | else if (s->size >= 1024) | |
3137 | s->cpu_partial = 6; | |
3138 | else if (s->size >= 256) | |
3139 | s->cpu_partial = 13; | |
3140 | else | |
3141 | s->cpu_partial = 30; | |
3142 | ||
81819f0f | 3143 | #ifdef CONFIG_NUMA |
e2cb96b7 | 3144 | s->remote_node_defrag_ratio = 1000; |
81819f0f | 3145 | #endif |
55136592 | 3146 | if (!init_kmem_cache_nodes(s)) |
dfb4f096 | 3147 | goto error; |
81819f0f | 3148 | |
55136592 | 3149 | if (alloc_kmem_cache_cpus(s)) |
278b1bb1 | 3150 | return 0; |
ff12059e | 3151 | |
4c93c355 | 3152 | free_kmem_cache_nodes(s); |
81819f0f CL |
3153 | error: |
3154 | if (flags & SLAB_PANIC) | |
3155 | panic("Cannot create slab %s size=%lu realsize=%u " | |
3156 | "order=%u offset=%u flags=%lx\n", | |
d0e0ac97 CG |
3157 | s->name, (unsigned long)s->size, s->size, |
3158 | oo_order(s->oo), s->offset, flags); | |
278b1bb1 | 3159 | return -EINVAL; |
81819f0f | 3160 | } |
81819f0f | 3161 | |
33b12c38 CL |
3162 | static void list_slab_objects(struct kmem_cache *s, struct page *page, |
3163 | const char *text) | |
3164 | { | |
3165 | #ifdef CONFIG_SLUB_DEBUG | |
3166 | void *addr = page_address(page); | |
3167 | void *p; | |
a5dd5c11 NK |
3168 | unsigned long *map = kzalloc(BITS_TO_LONGS(page->objects) * |
3169 | sizeof(long), GFP_ATOMIC); | |
bbd7d57b ED |
3170 | if (!map) |
3171 | return; | |
945cf2b6 | 3172 | slab_err(s, page, text, s->name); |
33b12c38 | 3173 | slab_lock(page); |
33b12c38 | 3174 | |
5f80b13a | 3175 | get_map(s, page, map); |
33b12c38 CL |
3176 | for_each_object(p, s, addr, page->objects) { |
3177 | ||
3178 | if (!test_bit(slab_index(p, s, addr), map)) { | |
f9f58285 | 3179 | pr_err("INFO: Object 0x%p @offset=%tu\n", p, p - addr); |
33b12c38 CL |
3180 | print_tracking(s, p); |
3181 | } | |
3182 | } | |
3183 | slab_unlock(page); | |
bbd7d57b | 3184 | kfree(map); |
33b12c38 CL |
3185 | #endif |
3186 | } | |
3187 | ||
81819f0f | 3188 | /* |
599870b1 | 3189 | * Attempt to free all partial slabs on a node. |
69cb8e6b CL |
3190 | * This is called from kmem_cache_close(). We must be the last thread |
3191 | * using the cache and therefore we do not need to lock anymore. | |
81819f0f | 3192 | */ |
599870b1 | 3193 | static void free_partial(struct kmem_cache *s, struct kmem_cache_node *n) |
81819f0f | 3194 | { |
81819f0f CL |
3195 | struct page *page, *h; |
3196 | ||
33b12c38 | 3197 | list_for_each_entry_safe(page, h, &n->partial, lru) { |
81819f0f | 3198 | if (!page->inuse) { |
1e4dd946 | 3199 | __remove_partial(n, page); |
81819f0f | 3200 | discard_slab(s, page); |
33b12c38 CL |
3201 | } else { |
3202 | list_slab_objects(s, page, | |
945cf2b6 | 3203 | "Objects remaining in %s on kmem_cache_close()"); |
599870b1 | 3204 | } |
33b12c38 | 3205 | } |
81819f0f CL |
3206 | } |
3207 | ||
3208 | /* | |
672bba3a | 3209 | * Release all resources used by a slab cache. |
81819f0f | 3210 | */ |
0c710013 | 3211 | static inline int kmem_cache_close(struct kmem_cache *s) |
81819f0f CL |
3212 | { |
3213 | int node; | |
fa45dc25 | 3214 | struct kmem_cache_node *n; |
81819f0f CL |
3215 | |
3216 | flush_all(s); | |
81819f0f | 3217 | /* Attempt to free all objects */ |
fa45dc25 | 3218 | for_each_kmem_cache_node(s, node, n) { |
599870b1 CL |
3219 | free_partial(s, n); |
3220 | if (n->nr_partial || slabs_node(s, node)) | |
81819f0f CL |
3221 | return 1; |
3222 | } | |
945cf2b6 | 3223 | free_percpu(s->cpu_slab); |
81819f0f CL |
3224 | free_kmem_cache_nodes(s); |
3225 | return 0; | |
3226 | } | |
3227 | ||
945cf2b6 | 3228 | int __kmem_cache_shutdown(struct kmem_cache *s) |
81819f0f | 3229 | { |
41a21285 | 3230 | return kmem_cache_close(s); |
81819f0f | 3231 | } |
81819f0f CL |
3232 | |
3233 | /******************************************************************** | |
3234 | * Kmalloc subsystem | |
3235 | *******************************************************************/ | |
3236 | ||
81819f0f CL |
3237 | static int __init setup_slub_min_order(char *str) |
3238 | { | |
06428780 | 3239 | get_option(&str, &slub_min_order); |
81819f0f CL |
3240 | |
3241 | return 1; | |
3242 | } | |
3243 | ||
3244 | __setup("slub_min_order=", setup_slub_min_order); | |
3245 | ||
3246 | static int __init setup_slub_max_order(char *str) | |
3247 | { | |
06428780 | 3248 | get_option(&str, &slub_max_order); |
818cf590 | 3249 | slub_max_order = min(slub_max_order, MAX_ORDER - 1); |
81819f0f CL |
3250 | |
3251 | return 1; | |
3252 | } | |
3253 | ||
3254 | __setup("slub_max_order=", setup_slub_max_order); | |
3255 | ||
3256 | static int __init setup_slub_min_objects(char *str) | |
3257 | { | |
06428780 | 3258 | get_option(&str, &slub_min_objects); |
81819f0f CL |
3259 | |
3260 | return 1; | |
3261 | } | |
3262 | ||
3263 | __setup("slub_min_objects=", setup_slub_min_objects); | |
3264 | ||
3265 | static int __init setup_slub_nomerge(char *str) | |
3266 | { | |
3267 | slub_nomerge = 1; | |
3268 | return 1; | |
3269 | } | |
3270 | ||
3271 | __setup("slub_nomerge", setup_slub_nomerge); | |
3272 | ||
81819f0f CL |
3273 | void *__kmalloc(size_t size, gfp_t flags) |
3274 | { | |
aadb4bc4 | 3275 | struct kmem_cache *s; |
5b882be4 | 3276 | void *ret; |
81819f0f | 3277 | |
95a05b42 | 3278 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) |
eada35ef | 3279 | return kmalloc_large(size, flags); |
aadb4bc4 | 3280 | |
2c59dd65 | 3281 | s = kmalloc_slab(size, flags); |
aadb4bc4 CL |
3282 | |
3283 | if (unlikely(ZERO_OR_NULL_PTR(s))) | |
6cb8f913 CL |
3284 | return s; |
3285 | ||
2b847c3c | 3286 | ret = slab_alloc(s, flags, _RET_IP_); |
5b882be4 | 3287 | |
ca2b84cb | 3288 | trace_kmalloc(_RET_IP_, ret, size, s->size, flags); |
5b882be4 EGM |
3289 | |
3290 | return ret; | |
81819f0f CL |
3291 | } |
3292 | EXPORT_SYMBOL(__kmalloc); | |
3293 | ||
5d1f57e4 | 3294 | #ifdef CONFIG_NUMA |
f619cfe1 CL |
3295 | static void *kmalloc_large_node(size_t size, gfp_t flags, int node) |
3296 | { | |
b1eeab67 | 3297 | struct page *page; |
e4f7c0b4 | 3298 | void *ptr = NULL; |
f619cfe1 | 3299 | |
52383431 VD |
3300 | flags |= __GFP_COMP | __GFP_NOTRACK; |
3301 | page = alloc_kmem_pages_node(node, flags, get_order(size)); | |
f619cfe1 | 3302 | if (page) |
e4f7c0b4 CM |
3303 | ptr = page_address(page); |
3304 | ||
d56791b3 | 3305 | kmalloc_large_node_hook(ptr, size, flags); |
e4f7c0b4 | 3306 | return ptr; |
f619cfe1 CL |
3307 | } |
3308 | ||
81819f0f CL |
3309 | void *__kmalloc_node(size_t size, gfp_t flags, int node) |
3310 | { | |
aadb4bc4 | 3311 | struct kmem_cache *s; |
5b882be4 | 3312 | void *ret; |
81819f0f | 3313 | |
95a05b42 | 3314 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) { |
5b882be4 EGM |
3315 | ret = kmalloc_large_node(size, flags, node); |
3316 | ||
ca2b84cb EGM |
3317 | trace_kmalloc_node(_RET_IP_, ret, |
3318 | size, PAGE_SIZE << get_order(size), | |
3319 | flags, node); | |
5b882be4 EGM |
3320 | |
3321 | return ret; | |
3322 | } | |
aadb4bc4 | 3323 | |
2c59dd65 | 3324 | s = kmalloc_slab(size, flags); |
aadb4bc4 CL |
3325 | |
3326 | if (unlikely(ZERO_OR_NULL_PTR(s))) | |
6cb8f913 CL |
3327 | return s; |
3328 | ||
2b847c3c | 3329 | ret = slab_alloc_node(s, flags, node, _RET_IP_); |
5b882be4 | 3330 | |
ca2b84cb | 3331 | trace_kmalloc_node(_RET_IP_, ret, size, s->size, flags, node); |
5b882be4 EGM |
3332 | |
3333 | return ret; | |
81819f0f CL |
3334 | } |
3335 | EXPORT_SYMBOL(__kmalloc_node); | |
3336 | #endif | |
3337 | ||
3338 | size_t ksize(const void *object) | |
3339 | { | |
272c1d21 | 3340 | struct page *page; |
81819f0f | 3341 | |
ef8b4520 | 3342 | if (unlikely(object == ZERO_SIZE_PTR)) |
272c1d21 CL |
3343 | return 0; |
3344 | ||
294a80a8 | 3345 | page = virt_to_head_page(object); |
294a80a8 | 3346 | |
76994412 PE |
3347 | if (unlikely(!PageSlab(page))) { |
3348 | WARN_ON(!PageCompound(page)); | |
294a80a8 | 3349 | return PAGE_SIZE << compound_order(page); |
76994412 | 3350 | } |
81819f0f | 3351 | |
1b4f59e3 | 3352 | return slab_ksize(page->slab_cache); |
81819f0f | 3353 | } |
b1aabecd | 3354 | EXPORT_SYMBOL(ksize); |
81819f0f CL |
3355 | |
3356 | void kfree(const void *x) | |
3357 | { | |
81819f0f | 3358 | struct page *page; |
5bb983b0 | 3359 | void *object = (void *)x; |
81819f0f | 3360 | |
2121db74 PE |
3361 | trace_kfree(_RET_IP_, x); |
3362 | ||
2408c550 | 3363 | if (unlikely(ZERO_OR_NULL_PTR(x))) |
81819f0f CL |
3364 | return; |
3365 | ||
b49af68f | 3366 | page = virt_to_head_page(x); |
aadb4bc4 | 3367 | if (unlikely(!PageSlab(page))) { |
0937502a | 3368 | BUG_ON(!PageCompound(page)); |
d56791b3 | 3369 | kfree_hook(x); |
52383431 | 3370 | __free_kmem_pages(page, compound_order(page)); |
aadb4bc4 CL |
3371 | return; |
3372 | } | |
1b4f59e3 | 3373 | slab_free(page->slab_cache, page, object, _RET_IP_); |
81819f0f CL |
3374 | } |
3375 | EXPORT_SYMBOL(kfree); | |
3376 | ||
2086d26a | 3377 | /* |
672bba3a CL |
3378 | * kmem_cache_shrink removes empty slabs from the partial lists and sorts |
3379 | * the remaining slabs by the number of items in use. The slabs with the | |
3380 | * most items in use come first. New allocations will then fill those up | |
3381 | * and thus they can be removed from the partial lists. | |
3382 | * | |
3383 | * The slabs with the least items are placed last. This results in them | |
3384 | * being allocated from last increasing the chance that the last objects | |
3385 | * are freed in them. | |
2086d26a | 3386 | */ |
03afc0e2 | 3387 | int __kmem_cache_shrink(struct kmem_cache *s) |
2086d26a CL |
3388 | { |
3389 | int node; | |
3390 | int i; | |
3391 | struct kmem_cache_node *n; | |
3392 | struct page *page; | |
3393 | struct page *t; | |
205ab99d | 3394 | int objects = oo_objects(s->max); |
2086d26a | 3395 | struct list_head *slabs_by_inuse = |
834f3d11 | 3396 | kmalloc(sizeof(struct list_head) * objects, GFP_KERNEL); |
2086d26a CL |
3397 | unsigned long flags; |
3398 | ||
3399 | if (!slabs_by_inuse) | |
3400 | return -ENOMEM; | |
3401 | ||
3402 | flush_all(s); | |
fa45dc25 | 3403 | for_each_kmem_cache_node(s, node, n) { |
2086d26a CL |
3404 | if (!n->nr_partial) |
3405 | continue; | |
3406 | ||
834f3d11 | 3407 | for (i = 0; i < objects; i++) |
2086d26a CL |
3408 | INIT_LIST_HEAD(slabs_by_inuse + i); |
3409 | ||
3410 | spin_lock_irqsave(&n->list_lock, flags); | |
3411 | ||
3412 | /* | |
672bba3a | 3413 | * Build lists indexed by the items in use in each slab. |
2086d26a | 3414 | * |
672bba3a CL |
3415 | * Note that concurrent frees may occur while we hold the |
3416 | * list_lock. page->inuse here is the upper limit. | |
2086d26a CL |
3417 | */ |
3418 | list_for_each_entry_safe(page, t, &n->partial, lru) { | |
69cb8e6b CL |
3419 | list_move(&page->lru, slabs_by_inuse + page->inuse); |
3420 | if (!page->inuse) | |
3421 | n->nr_partial--; | |
2086d26a CL |
3422 | } |
3423 | ||
2086d26a | 3424 | /* |
672bba3a CL |
3425 | * Rebuild the partial list with the slabs filled up most |
3426 | * first and the least used slabs at the end. | |
2086d26a | 3427 | */ |
69cb8e6b | 3428 | for (i = objects - 1; i > 0; i--) |
2086d26a CL |
3429 | list_splice(slabs_by_inuse + i, n->partial.prev); |
3430 | ||
2086d26a | 3431 | spin_unlock_irqrestore(&n->list_lock, flags); |
69cb8e6b CL |
3432 | |
3433 | /* Release empty slabs */ | |
3434 | list_for_each_entry_safe(page, t, slabs_by_inuse, lru) | |
3435 | discard_slab(s, page); | |
2086d26a CL |
3436 | } |
3437 | ||
3438 | kfree(slabs_by_inuse); | |
3439 | return 0; | |
3440 | } | |
2086d26a | 3441 | |
b9049e23 YG |
3442 | static int slab_mem_going_offline_callback(void *arg) |
3443 | { | |
3444 | struct kmem_cache *s; | |
3445 | ||
18004c5d | 3446 | mutex_lock(&slab_mutex); |
b9049e23 | 3447 | list_for_each_entry(s, &slab_caches, list) |
03afc0e2 | 3448 | __kmem_cache_shrink(s); |
18004c5d | 3449 | mutex_unlock(&slab_mutex); |
b9049e23 YG |
3450 | |
3451 | return 0; | |
3452 | } | |
3453 | ||
3454 | static void slab_mem_offline_callback(void *arg) | |
3455 | { | |
3456 | struct kmem_cache_node *n; | |
3457 | struct kmem_cache *s; | |
3458 | struct memory_notify *marg = arg; | |
3459 | int offline_node; | |
3460 | ||
b9d5ab25 | 3461 | offline_node = marg->status_change_nid_normal; |
b9049e23 YG |
3462 | |
3463 | /* | |
3464 | * If the node still has available memory. we need kmem_cache_node | |
3465 | * for it yet. | |
3466 | */ | |
3467 | if (offline_node < 0) | |
3468 | return; | |
3469 | ||
18004c5d | 3470 | mutex_lock(&slab_mutex); |
b9049e23 YG |
3471 | list_for_each_entry(s, &slab_caches, list) { |
3472 | n = get_node(s, offline_node); | |
3473 | if (n) { | |
3474 | /* | |
3475 | * if n->nr_slabs > 0, slabs still exist on the node | |
3476 | * that is going down. We were unable to free them, | |
c9404c9c | 3477 | * and offline_pages() function shouldn't call this |
b9049e23 YG |
3478 | * callback. So, we must fail. |
3479 | */ | |
0f389ec6 | 3480 | BUG_ON(slabs_node(s, offline_node)); |
b9049e23 YG |
3481 | |
3482 | s->node[offline_node] = NULL; | |
8de66a0c | 3483 | kmem_cache_free(kmem_cache_node, n); |
b9049e23 YG |
3484 | } |
3485 | } | |
18004c5d | 3486 | mutex_unlock(&slab_mutex); |
b9049e23 YG |
3487 | } |
3488 | ||
3489 | static int slab_mem_going_online_callback(void *arg) | |
3490 | { | |
3491 | struct kmem_cache_node *n; | |
3492 | struct kmem_cache *s; | |
3493 | struct memory_notify *marg = arg; | |
b9d5ab25 | 3494 | int nid = marg->status_change_nid_normal; |
b9049e23 YG |
3495 | int ret = 0; |
3496 | ||
3497 | /* | |
3498 | * If the node's memory is already available, then kmem_cache_node is | |
3499 | * already created. Nothing to do. | |
3500 | */ | |
3501 | if (nid < 0) | |
3502 | return 0; | |
3503 | ||
3504 | /* | |
0121c619 | 3505 | * We are bringing a node online. No memory is available yet. We must |
b9049e23 YG |
3506 | * allocate a kmem_cache_node structure in order to bring the node |
3507 | * online. | |
3508 | */ | |
18004c5d | 3509 | mutex_lock(&slab_mutex); |
b9049e23 YG |
3510 | list_for_each_entry(s, &slab_caches, list) { |
3511 | /* | |
3512 | * XXX: kmem_cache_alloc_node will fallback to other nodes | |
3513 | * since memory is not yet available from the node that | |
3514 | * is brought up. | |
3515 | */ | |
8de66a0c | 3516 | n = kmem_cache_alloc(kmem_cache_node, GFP_KERNEL); |
b9049e23 YG |
3517 | if (!n) { |
3518 | ret = -ENOMEM; | |
3519 | goto out; | |
3520 | } | |
4053497d | 3521 | init_kmem_cache_node(n); |
b9049e23 YG |
3522 | s->node[nid] = n; |
3523 | } | |
3524 | out: | |
18004c5d | 3525 | mutex_unlock(&slab_mutex); |
b9049e23 YG |
3526 | return ret; |
3527 | } | |
3528 | ||
3529 | static int slab_memory_callback(struct notifier_block *self, | |
3530 | unsigned long action, void *arg) | |
3531 | { | |
3532 | int ret = 0; | |
3533 | ||
3534 | switch (action) { | |
3535 | case MEM_GOING_ONLINE: | |
3536 | ret = slab_mem_going_online_callback(arg); | |
3537 | break; | |
3538 | case MEM_GOING_OFFLINE: | |
3539 | ret = slab_mem_going_offline_callback(arg); | |
3540 | break; | |
3541 | case MEM_OFFLINE: | |
3542 | case MEM_CANCEL_ONLINE: | |
3543 | slab_mem_offline_callback(arg); | |
3544 | break; | |
3545 | case MEM_ONLINE: | |
3546 | case MEM_CANCEL_OFFLINE: | |
3547 | break; | |
3548 | } | |
dc19f9db KH |
3549 | if (ret) |
3550 | ret = notifier_from_errno(ret); | |
3551 | else | |
3552 | ret = NOTIFY_OK; | |
b9049e23 YG |
3553 | return ret; |
3554 | } | |
3555 | ||
3ac38faa AM |
3556 | static struct notifier_block slab_memory_callback_nb = { |
3557 | .notifier_call = slab_memory_callback, | |
3558 | .priority = SLAB_CALLBACK_PRI, | |
3559 | }; | |
b9049e23 | 3560 | |
81819f0f CL |
3561 | /******************************************************************** |
3562 | * Basic setup of slabs | |
3563 | *******************************************************************/ | |
3564 | ||
51df1142 CL |
3565 | /* |
3566 | * Used for early kmem_cache structures that were allocated using | |
dffb4d60 CL |
3567 | * the page allocator. Allocate them properly then fix up the pointers |
3568 | * that may be pointing to the wrong kmem_cache structure. | |
51df1142 CL |
3569 | */ |
3570 | ||
dffb4d60 | 3571 | static struct kmem_cache * __init bootstrap(struct kmem_cache *static_cache) |
51df1142 CL |
3572 | { |
3573 | int node; | |
dffb4d60 | 3574 | struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT); |
fa45dc25 | 3575 | struct kmem_cache_node *n; |
51df1142 | 3576 | |
dffb4d60 | 3577 | memcpy(s, static_cache, kmem_cache->object_size); |
51df1142 | 3578 | |
7d557b3c GC |
3579 | /* |
3580 | * This runs very early, and only the boot processor is supposed to be | |
3581 | * up. Even if it weren't true, IRQs are not up so we couldn't fire | |
3582 | * IPIs around. | |
3583 | */ | |
3584 | __flush_cpu_slab(s, smp_processor_id()); | |
fa45dc25 | 3585 | for_each_kmem_cache_node(s, node, n) { |
51df1142 CL |
3586 | struct page *p; |
3587 | ||
fa45dc25 CL |
3588 | list_for_each_entry(p, &n->partial, lru) |
3589 | p->slab_cache = s; | |
51df1142 | 3590 | |
607bf324 | 3591 | #ifdef CONFIG_SLUB_DEBUG |
fa45dc25 CL |
3592 | list_for_each_entry(p, &n->full, lru) |
3593 | p->slab_cache = s; | |
51df1142 | 3594 | #endif |
51df1142 | 3595 | } |
dffb4d60 CL |
3596 | list_add(&s->list, &slab_caches); |
3597 | return s; | |
51df1142 CL |
3598 | } |
3599 | ||
81819f0f CL |
3600 | void __init kmem_cache_init(void) |
3601 | { | |
dffb4d60 CL |
3602 | static __initdata struct kmem_cache boot_kmem_cache, |
3603 | boot_kmem_cache_node; | |
51df1142 | 3604 | |
fc8d8620 SG |
3605 | if (debug_guardpage_minorder()) |
3606 | slub_max_order = 0; | |
3607 | ||
dffb4d60 CL |
3608 | kmem_cache_node = &boot_kmem_cache_node; |
3609 | kmem_cache = &boot_kmem_cache; | |
51df1142 | 3610 | |
dffb4d60 CL |
3611 | create_boot_cache(kmem_cache_node, "kmem_cache_node", |
3612 | sizeof(struct kmem_cache_node), SLAB_HWCACHE_ALIGN); | |
b9049e23 | 3613 | |
3ac38faa | 3614 | register_hotmemory_notifier(&slab_memory_callback_nb); |
81819f0f CL |
3615 | |
3616 | /* Able to allocate the per node structures */ | |
3617 | slab_state = PARTIAL; | |
3618 | ||
dffb4d60 CL |
3619 | create_boot_cache(kmem_cache, "kmem_cache", |
3620 | offsetof(struct kmem_cache, node) + | |
3621 | nr_node_ids * sizeof(struct kmem_cache_node *), | |
3622 | SLAB_HWCACHE_ALIGN); | |
8a13a4cc | 3623 | |
dffb4d60 | 3624 | kmem_cache = bootstrap(&boot_kmem_cache); |
81819f0f | 3625 | |
51df1142 CL |
3626 | /* |
3627 | * Allocate kmem_cache_node properly from the kmem_cache slab. | |
3628 | * kmem_cache_node is separately allocated so no need to | |
3629 | * update any list pointers. | |
3630 | */ | |
dffb4d60 | 3631 | kmem_cache_node = bootstrap(&boot_kmem_cache_node); |
51df1142 CL |
3632 | |
3633 | /* Now we can use the kmem_cache to allocate kmalloc slabs */ | |
f97d5f63 | 3634 | create_kmalloc_caches(0); |
81819f0f CL |
3635 | |
3636 | #ifdef CONFIG_SMP | |
3637 | register_cpu_notifier(&slab_notifier); | |
9dfc6e68 | 3638 | #endif |
81819f0f | 3639 | |
f9f58285 | 3640 | pr_info("SLUB: HWalign=%d, Order=%d-%d, MinObjects=%d, CPUs=%d, Nodes=%d\n", |
f97d5f63 | 3641 | cache_line_size(), |
81819f0f CL |
3642 | slub_min_order, slub_max_order, slub_min_objects, |
3643 | nr_cpu_ids, nr_node_ids); | |
3644 | } | |
3645 | ||
7e85ee0c PE |
3646 | void __init kmem_cache_init_late(void) |
3647 | { | |
7e85ee0c PE |
3648 | } |
3649 | ||
81819f0f CL |
3650 | /* |
3651 | * Find a mergeable slab cache | |
3652 | */ | |
3653 | static int slab_unmergeable(struct kmem_cache *s) | |
3654 | { | |
3655 | if (slub_nomerge || (s->flags & SLUB_NEVER_MERGE)) | |
3656 | return 1; | |
3657 | ||
a44cb944 VD |
3658 | if (!is_root_cache(s)) |
3659 | return 1; | |
3660 | ||
c59def9f | 3661 | if (s->ctor) |
81819f0f CL |
3662 | return 1; |
3663 | ||
8ffa6875 CL |
3664 | /* |
3665 | * We may have set a slab to be unmergeable during bootstrap. | |
3666 | */ | |
3667 | if (s->refcount < 0) | |
3668 | return 1; | |
3669 | ||
81819f0f CL |
3670 | return 0; |
3671 | } | |
3672 | ||
a44cb944 VD |
3673 | static struct kmem_cache *find_mergeable(size_t size, size_t align, |
3674 | unsigned long flags, const char *name, void (*ctor)(void *)) | |
81819f0f | 3675 | { |
5b95a4ac | 3676 | struct kmem_cache *s; |
81819f0f CL |
3677 | |
3678 | if (slub_nomerge || (flags & SLUB_NEVER_MERGE)) | |
3679 | return NULL; | |
3680 | ||
c59def9f | 3681 | if (ctor) |
81819f0f CL |
3682 | return NULL; |
3683 | ||
3684 | size = ALIGN(size, sizeof(void *)); | |
3685 | align = calculate_alignment(flags, align, size); | |
3686 | size = ALIGN(size, align); | |
ba0268a8 | 3687 | flags = kmem_cache_flags(size, flags, name, NULL); |
81819f0f | 3688 | |
5b95a4ac | 3689 | list_for_each_entry(s, &slab_caches, list) { |
81819f0f CL |
3690 | if (slab_unmergeable(s)) |
3691 | continue; | |
3692 | ||
3693 | if (size > s->size) | |
3694 | continue; | |
3695 | ||
ba0268a8 | 3696 | if ((flags & SLUB_MERGE_SAME) != (s->flags & SLUB_MERGE_SAME)) |
a44cb944 | 3697 | continue; |
81819f0f CL |
3698 | /* |
3699 | * Check if alignment is compatible. | |
3700 | * Courtesy of Adrian Drzewiecki | |
3701 | */ | |
06428780 | 3702 | if ((s->size & ~(align - 1)) != s->size) |
81819f0f CL |
3703 | continue; |
3704 | ||
3705 | if (s->size - size >= sizeof(void *)) | |
3706 | continue; | |
3707 | ||
3708 | return s; | |
3709 | } | |
3710 | return NULL; | |
3711 | } | |
3712 | ||
2633d7a0 | 3713 | struct kmem_cache * |
a44cb944 VD |
3714 | __kmem_cache_alias(const char *name, size_t size, size_t align, |
3715 | unsigned long flags, void (*ctor)(void *)) | |
81819f0f CL |
3716 | { |
3717 | struct kmem_cache *s; | |
3718 | ||
a44cb944 | 3719 | s = find_mergeable(size, align, flags, name, ctor); |
81819f0f | 3720 | if (s) { |
84d0ddd6 VD |
3721 | int i; |
3722 | struct kmem_cache *c; | |
3723 | ||
81819f0f | 3724 | s->refcount++; |
84d0ddd6 | 3725 | |
81819f0f CL |
3726 | /* |
3727 | * Adjust the object sizes so that we clear | |
3728 | * the complete object on kzalloc. | |
3729 | */ | |
3b0efdfa | 3730 | s->object_size = max(s->object_size, (int)size); |
81819f0f | 3731 | s->inuse = max_t(int, s->inuse, ALIGN(size, sizeof(void *))); |
6446faa2 | 3732 | |
84d0ddd6 VD |
3733 | for_each_memcg_cache_index(i) { |
3734 | c = cache_from_memcg_idx(s, i); | |
3735 | if (!c) | |
3736 | continue; | |
3737 | c->object_size = s->object_size; | |
3738 | c->inuse = max_t(int, c->inuse, | |
3739 | ALIGN(size, sizeof(void *))); | |
3740 | } | |
3741 | ||
7b8f3b66 | 3742 | if (sysfs_slab_alias(s, name)) { |
7b8f3b66 | 3743 | s->refcount--; |
cbb79694 | 3744 | s = NULL; |
7b8f3b66 | 3745 | } |
a0e1d1be | 3746 | } |
6446faa2 | 3747 | |
cbb79694 CL |
3748 | return s; |
3749 | } | |
84c1cf62 | 3750 | |
8a13a4cc | 3751 | int __kmem_cache_create(struct kmem_cache *s, unsigned long flags) |
cbb79694 | 3752 | { |
aac3a166 PE |
3753 | int err; |
3754 | ||
3755 | err = kmem_cache_open(s, flags); | |
3756 | if (err) | |
3757 | return err; | |
20cea968 | 3758 | |
45530c44 CL |
3759 | /* Mutex is not taken during early boot */ |
3760 | if (slab_state <= UP) | |
3761 | return 0; | |
3762 | ||
107dab5c | 3763 | memcg_propagate_slab_attrs(s); |
aac3a166 | 3764 | err = sysfs_slab_add(s); |
aac3a166 PE |
3765 | if (err) |
3766 | kmem_cache_close(s); | |
20cea968 | 3767 | |
aac3a166 | 3768 | return err; |
81819f0f | 3769 | } |
81819f0f | 3770 | |
81819f0f | 3771 | #ifdef CONFIG_SMP |
81819f0f | 3772 | /* |
672bba3a CL |
3773 | * Use the cpu notifier to insure that the cpu slabs are flushed when |
3774 | * necessary. | |
81819f0f | 3775 | */ |
0db0628d | 3776 | static int slab_cpuup_callback(struct notifier_block *nfb, |
81819f0f CL |
3777 | unsigned long action, void *hcpu) |
3778 | { | |
3779 | long cpu = (long)hcpu; | |
5b95a4ac CL |
3780 | struct kmem_cache *s; |
3781 | unsigned long flags; | |
81819f0f CL |
3782 | |
3783 | switch (action) { | |
3784 | case CPU_UP_CANCELED: | |
8bb78442 | 3785 | case CPU_UP_CANCELED_FROZEN: |
81819f0f | 3786 | case CPU_DEAD: |
8bb78442 | 3787 | case CPU_DEAD_FROZEN: |
18004c5d | 3788 | mutex_lock(&slab_mutex); |
5b95a4ac CL |
3789 | list_for_each_entry(s, &slab_caches, list) { |
3790 | local_irq_save(flags); | |
3791 | __flush_cpu_slab(s, cpu); | |
3792 | local_irq_restore(flags); | |
3793 | } | |
18004c5d | 3794 | mutex_unlock(&slab_mutex); |
81819f0f CL |
3795 | break; |
3796 | default: | |
3797 | break; | |
3798 | } | |
3799 | return NOTIFY_OK; | |
3800 | } | |
3801 | ||
0db0628d | 3802 | static struct notifier_block slab_notifier = { |
3adbefee | 3803 | .notifier_call = slab_cpuup_callback |
06428780 | 3804 | }; |
81819f0f CL |
3805 | |
3806 | #endif | |
3807 | ||
ce71e27c | 3808 | void *__kmalloc_track_caller(size_t size, gfp_t gfpflags, unsigned long caller) |
81819f0f | 3809 | { |
aadb4bc4 | 3810 | struct kmem_cache *s; |
94b528d0 | 3811 | void *ret; |
aadb4bc4 | 3812 | |
95a05b42 | 3813 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) |
eada35ef PE |
3814 | return kmalloc_large(size, gfpflags); |
3815 | ||
2c59dd65 | 3816 | s = kmalloc_slab(size, gfpflags); |
81819f0f | 3817 | |
2408c550 | 3818 | if (unlikely(ZERO_OR_NULL_PTR(s))) |
6cb8f913 | 3819 | return s; |
81819f0f | 3820 | |
2b847c3c | 3821 | ret = slab_alloc(s, gfpflags, caller); |
94b528d0 | 3822 | |
25985edc | 3823 | /* Honor the call site pointer we received. */ |
ca2b84cb | 3824 | trace_kmalloc(caller, ret, size, s->size, gfpflags); |
94b528d0 EGM |
3825 | |
3826 | return ret; | |
81819f0f CL |
3827 | } |
3828 | ||
5d1f57e4 | 3829 | #ifdef CONFIG_NUMA |
81819f0f | 3830 | void *__kmalloc_node_track_caller(size_t size, gfp_t gfpflags, |
ce71e27c | 3831 | int node, unsigned long caller) |
81819f0f | 3832 | { |
aadb4bc4 | 3833 | struct kmem_cache *s; |
94b528d0 | 3834 | void *ret; |
aadb4bc4 | 3835 | |
95a05b42 | 3836 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) { |
d3e14aa3 XF |
3837 | ret = kmalloc_large_node(size, gfpflags, node); |
3838 | ||
3839 | trace_kmalloc_node(caller, ret, | |
3840 | size, PAGE_SIZE << get_order(size), | |
3841 | gfpflags, node); | |
3842 | ||
3843 | return ret; | |
3844 | } | |
eada35ef | 3845 | |
2c59dd65 | 3846 | s = kmalloc_slab(size, gfpflags); |
81819f0f | 3847 | |
2408c550 | 3848 | if (unlikely(ZERO_OR_NULL_PTR(s))) |
6cb8f913 | 3849 | return s; |
81819f0f | 3850 | |
2b847c3c | 3851 | ret = slab_alloc_node(s, gfpflags, node, caller); |
94b528d0 | 3852 | |
25985edc | 3853 | /* Honor the call site pointer we received. */ |
ca2b84cb | 3854 | trace_kmalloc_node(caller, ret, size, s->size, gfpflags, node); |
94b528d0 EGM |
3855 | |
3856 | return ret; | |
81819f0f | 3857 | } |
5d1f57e4 | 3858 | #endif |
81819f0f | 3859 | |
ab4d5ed5 | 3860 | #ifdef CONFIG_SYSFS |
205ab99d CL |
3861 | static int count_inuse(struct page *page) |
3862 | { | |
3863 | return page->inuse; | |
3864 | } | |
3865 | ||
3866 | static int count_total(struct page *page) | |
3867 | { | |
3868 | return page->objects; | |
3869 | } | |
ab4d5ed5 | 3870 | #endif |
205ab99d | 3871 | |
ab4d5ed5 | 3872 | #ifdef CONFIG_SLUB_DEBUG |
434e245d CL |
3873 | static int validate_slab(struct kmem_cache *s, struct page *page, |
3874 | unsigned long *map) | |
53e15af0 CL |
3875 | { |
3876 | void *p; | |
a973e9dd | 3877 | void *addr = page_address(page); |
53e15af0 CL |
3878 | |
3879 | if (!check_slab(s, page) || | |
3880 | !on_freelist(s, page, NULL)) | |
3881 | return 0; | |
3882 | ||
3883 | /* Now we know that a valid freelist exists */ | |
39b26464 | 3884 | bitmap_zero(map, page->objects); |
53e15af0 | 3885 | |
5f80b13a CL |
3886 | get_map(s, page, map); |
3887 | for_each_object(p, s, addr, page->objects) { | |
3888 | if (test_bit(slab_index(p, s, addr), map)) | |
3889 | if (!check_object(s, page, p, SLUB_RED_INACTIVE)) | |
3890 | return 0; | |
53e15af0 CL |
3891 | } |
3892 | ||
224a88be | 3893 | for_each_object(p, s, addr, page->objects) |
7656c72b | 3894 | if (!test_bit(slab_index(p, s, addr), map)) |
37d57443 | 3895 | if (!check_object(s, page, p, SLUB_RED_ACTIVE)) |
53e15af0 CL |
3896 | return 0; |
3897 | return 1; | |
3898 | } | |
3899 | ||
434e245d CL |
3900 | static void validate_slab_slab(struct kmem_cache *s, struct page *page, |
3901 | unsigned long *map) | |
53e15af0 | 3902 | { |
881db7fb CL |
3903 | slab_lock(page); |
3904 | validate_slab(s, page, map); | |
3905 | slab_unlock(page); | |
53e15af0 CL |
3906 | } |
3907 | ||
434e245d CL |
3908 | static int validate_slab_node(struct kmem_cache *s, |
3909 | struct kmem_cache_node *n, unsigned long *map) | |
53e15af0 CL |
3910 | { |
3911 | unsigned long count = 0; | |
3912 | struct page *page; | |
3913 | unsigned long flags; | |
3914 | ||
3915 | spin_lock_irqsave(&n->list_lock, flags); | |
3916 | ||
3917 | list_for_each_entry(page, &n->partial, lru) { | |
434e245d | 3918 | validate_slab_slab(s, page, map); |
53e15af0 CL |
3919 | count++; |
3920 | } | |
3921 | if (count != n->nr_partial) | |
f9f58285 FF |
3922 | pr_err("SLUB %s: %ld partial slabs counted but counter=%ld\n", |
3923 | s->name, count, n->nr_partial); | |
53e15af0 CL |
3924 | |
3925 | if (!(s->flags & SLAB_STORE_USER)) | |
3926 | goto out; | |
3927 | ||
3928 | list_for_each_entry(page, &n->full, lru) { | |
434e245d | 3929 | validate_slab_slab(s, page, map); |
53e15af0 CL |
3930 | count++; |
3931 | } | |
3932 | if (count != atomic_long_read(&n->nr_slabs)) | |
f9f58285 FF |
3933 | pr_err("SLUB: %s %ld slabs counted but counter=%ld\n", |
3934 | s->name, count, atomic_long_read(&n->nr_slabs)); | |
53e15af0 CL |
3935 | |
3936 | out: | |
3937 | spin_unlock_irqrestore(&n->list_lock, flags); | |
3938 | return count; | |
3939 | } | |
3940 | ||
434e245d | 3941 | static long validate_slab_cache(struct kmem_cache *s) |
53e15af0 CL |
3942 | { |
3943 | int node; | |
3944 | unsigned long count = 0; | |
205ab99d | 3945 | unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) * |
434e245d | 3946 | sizeof(unsigned long), GFP_KERNEL); |
fa45dc25 | 3947 | struct kmem_cache_node *n; |
434e245d CL |
3948 | |
3949 | if (!map) | |
3950 | return -ENOMEM; | |
53e15af0 CL |
3951 | |
3952 | flush_all(s); | |
fa45dc25 | 3953 | for_each_kmem_cache_node(s, node, n) |
434e245d | 3954 | count += validate_slab_node(s, n, map); |
434e245d | 3955 | kfree(map); |
53e15af0 CL |
3956 | return count; |
3957 | } | |
88a420e4 | 3958 | /* |
672bba3a | 3959 | * Generate lists of code addresses where slabcache objects are allocated |
88a420e4 CL |
3960 | * and freed. |
3961 | */ | |
3962 | ||
3963 | struct location { | |
3964 | unsigned long count; | |
ce71e27c | 3965 | unsigned long addr; |
45edfa58 CL |
3966 | long long sum_time; |
3967 | long min_time; | |
3968 | long max_time; | |
3969 | long min_pid; | |
3970 | long max_pid; | |
174596a0 | 3971 | DECLARE_BITMAP(cpus, NR_CPUS); |
45edfa58 | 3972 | nodemask_t nodes; |
88a420e4 CL |
3973 | }; |
3974 | ||
3975 | struct loc_track { | |
3976 | unsigned long max; | |
3977 | unsigned long count; | |
3978 | struct location *loc; | |
3979 | }; | |
3980 | ||
3981 | static void free_loc_track(struct loc_track *t) | |
3982 | { | |
3983 | if (t->max) | |
3984 | free_pages((unsigned long)t->loc, | |
3985 | get_order(sizeof(struct location) * t->max)); | |
3986 | } | |
3987 | ||
68dff6a9 | 3988 | static int alloc_loc_track(struct loc_track *t, unsigned long max, gfp_t flags) |
88a420e4 CL |
3989 | { |
3990 | struct location *l; | |
3991 | int order; | |
3992 | ||
88a420e4 CL |
3993 | order = get_order(sizeof(struct location) * max); |
3994 | ||
68dff6a9 | 3995 | l = (void *)__get_free_pages(flags, order); |
88a420e4 CL |
3996 | if (!l) |
3997 | return 0; | |
3998 | ||
3999 | if (t->count) { | |
4000 | memcpy(l, t->loc, sizeof(struct location) * t->count); | |
4001 | free_loc_track(t); | |
4002 | } | |
4003 | t->max = max; | |
4004 | t->loc = l; | |
4005 | return 1; | |
4006 | } | |
4007 | ||
4008 | static int add_location(struct loc_track *t, struct kmem_cache *s, | |
45edfa58 | 4009 | const struct track *track) |
88a420e4 CL |
4010 | { |
4011 | long start, end, pos; | |
4012 | struct location *l; | |
ce71e27c | 4013 | unsigned long caddr; |
45edfa58 | 4014 | unsigned long age = jiffies - track->when; |
88a420e4 CL |
4015 | |
4016 | start = -1; | |
4017 | end = t->count; | |
4018 | ||
4019 | for ( ; ; ) { | |
4020 | pos = start + (end - start + 1) / 2; | |
4021 | ||
4022 | /* | |
4023 | * There is nothing at "end". If we end up there | |
4024 | * we need to add something to before end. | |
4025 | */ | |
4026 | if (pos == end) | |
4027 | break; | |
4028 | ||
4029 | caddr = t->loc[pos].addr; | |
45edfa58 CL |
4030 | if (track->addr == caddr) { |
4031 | ||
4032 | l = &t->loc[pos]; | |
4033 | l->count++; | |
4034 | if (track->when) { | |
4035 | l->sum_time += age; | |
4036 | if (age < l->min_time) | |
4037 | l->min_time = age; | |
4038 | if (age > l->max_time) | |
4039 | l->max_time = age; | |
4040 | ||
4041 | if (track->pid < l->min_pid) | |
4042 | l->min_pid = track->pid; | |
4043 | if (track->pid > l->max_pid) | |
4044 | l->max_pid = track->pid; | |
4045 | ||
174596a0 RR |
4046 | cpumask_set_cpu(track->cpu, |
4047 | to_cpumask(l->cpus)); | |
45edfa58 CL |
4048 | } |
4049 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
88a420e4 CL |
4050 | return 1; |
4051 | } | |
4052 | ||
45edfa58 | 4053 | if (track->addr < caddr) |
88a420e4 CL |
4054 | end = pos; |
4055 | else | |
4056 | start = pos; | |
4057 | } | |
4058 | ||
4059 | /* | |
672bba3a | 4060 | * Not found. Insert new tracking element. |
88a420e4 | 4061 | */ |
68dff6a9 | 4062 | if (t->count >= t->max && !alloc_loc_track(t, 2 * t->max, GFP_ATOMIC)) |
88a420e4 CL |
4063 | return 0; |
4064 | ||
4065 | l = t->loc + pos; | |
4066 | if (pos < t->count) | |
4067 | memmove(l + 1, l, | |
4068 | (t->count - pos) * sizeof(struct location)); | |
4069 | t->count++; | |
4070 | l->count = 1; | |
45edfa58 CL |
4071 | l->addr = track->addr; |
4072 | l->sum_time = age; | |
4073 | l->min_time = age; | |
4074 | l->max_time = age; | |
4075 | l->min_pid = track->pid; | |
4076 | l->max_pid = track->pid; | |
174596a0 RR |
4077 | cpumask_clear(to_cpumask(l->cpus)); |
4078 | cpumask_set_cpu(track->cpu, to_cpumask(l->cpus)); | |
45edfa58 CL |
4079 | nodes_clear(l->nodes); |
4080 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
88a420e4 CL |
4081 | return 1; |
4082 | } | |
4083 | ||
4084 | static void process_slab(struct loc_track *t, struct kmem_cache *s, | |
bbd7d57b | 4085 | struct page *page, enum track_item alloc, |
a5dd5c11 | 4086 | unsigned long *map) |
88a420e4 | 4087 | { |
a973e9dd | 4088 | void *addr = page_address(page); |
88a420e4 CL |
4089 | void *p; |
4090 | ||
39b26464 | 4091 | bitmap_zero(map, page->objects); |
5f80b13a | 4092 | get_map(s, page, map); |
88a420e4 | 4093 | |
224a88be | 4094 | for_each_object(p, s, addr, page->objects) |
45edfa58 CL |
4095 | if (!test_bit(slab_index(p, s, addr), map)) |
4096 | add_location(t, s, get_track(s, p, alloc)); | |
88a420e4 CL |
4097 | } |
4098 | ||
4099 | static int list_locations(struct kmem_cache *s, char *buf, | |
4100 | enum track_item alloc) | |
4101 | { | |
e374d483 | 4102 | int len = 0; |
88a420e4 | 4103 | unsigned long i; |
68dff6a9 | 4104 | struct loc_track t = { 0, 0, NULL }; |
88a420e4 | 4105 | int node; |
bbd7d57b ED |
4106 | unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) * |
4107 | sizeof(unsigned long), GFP_KERNEL); | |
fa45dc25 | 4108 | struct kmem_cache_node *n; |
88a420e4 | 4109 | |
bbd7d57b ED |
4110 | if (!map || !alloc_loc_track(&t, PAGE_SIZE / sizeof(struct location), |
4111 | GFP_TEMPORARY)) { | |
4112 | kfree(map); | |
68dff6a9 | 4113 | return sprintf(buf, "Out of memory\n"); |
bbd7d57b | 4114 | } |
88a420e4 CL |
4115 | /* Push back cpu slabs */ |
4116 | flush_all(s); | |
4117 | ||
fa45dc25 | 4118 | for_each_kmem_cache_node(s, node, n) { |
88a420e4 CL |
4119 | unsigned long flags; |
4120 | struct page *page; | |
4121 | ||
9e86943b | 4122 | if (!atomic_long_read(&n->nr_slabs)) |
88a420e4 CL |
4123 | continue; |
4124 | ||
4125 | spin_lock_irqsave(&n->list_lock, flags); | |
4126 | list_for_each_entry(page, &n->partial, lru) | |
bbd7d57b | 4127 | process_slab(&t, s, page, alloc, map); |
88a420e4 | 4128 | list_for_each_entry(page, &n->full, lru) |
bbd7d57b | 4129 | process_slab(&t, s, page, alloc, map); |
88a420e4 CL |
4130 | spin_unlock_irqrestore(&n->list_lock, flags); |
4131 | } | |
4132 | ||
4133 | for (i = 0; i < t.count; i++) { | |
45edfa58 | 4134 | struct location *l = &t.loc[i]; |
88a420e4 | 4135 | |
9c246247 | 4136 | if (len > PAGE_SIZE - KSYM_SYMBOL_LEN - 100) |
88a420e4 | 4137 | break; |
e374d483 | 4138 | len += sprintf(buf + len, "%7ld ", l->count); |
45edfa58 CL |
4139 | |
4140 | if (l->addr) | |
62c70bce | 4141 | len += sprintf(buf + len, "%pS", (void *)l->addr); |
88a420e4 | 4142 | else |
e374d483 | 4143 | len += sprintf(buf + len, "<not-available>"); |
45edfa58 CL |
4144 | |
4145 | if (l->sum_time != l->min_time) { | |
e374d483 | 4146 | len += sprintf(buf + len, " age=%ld/%ld/%ld", |
f8bd2258 RZ |
4147 | l->min_time, |
4148 | (long)div_u64(l->sum_time, l->count), | |
4149 | l->max_time); | |
45edfa58 | 4150 | } else |
e374d483 | 4151 | len += sprintf(buf + len, " age=%ld", |
45edfa58 CL |
4152 | l->min_time); |
4153 | ||
4154 | if (l->min_pid != l->max_pid) | |
e374d483 | 4155 | len += sprintf(buf + len, " pid=%ld-%ld", |
45edfa58 CL |
4156 | l->min_pid, l->max_pid); |
4157 | else | |
e374d483 | 4158 | len += sprintf(buf + len, " pid=%ld", |
45edfa58 CL |
4159 | l->min_pid); |
4160 | ||
174596a0 RR |
4161 | if (num_online_cpus() > 1 && |
4162 | !cpumask_empty(to_cpumask(l->cpus)) && | |
e374d483 HH |
4163 | len < PAGE_SIZE - 60) { |
4164 | len += sprintf(buf + len, " cpus="); | |
d0e0ac97 CG |
4165 | len += cpulist_scnprintf(buf + len, |
4166 | PAGE_SIZE - len - 50, | |
174596a0 | 4167 | to_cpumask(l->cpus)); |
45edfa58 CL |
4168 | } |
4169 | ||
62bc62a8 | 4170 | if (nr_online_nodes > 1 && !nodes_empty(l->nodes) && |
e374d483 HH |
4171 | len < PAGE_SIZE - 60) { |
4172 | len += sprintf(buf + len, " nodes="); | |
d0e0ac97 CG |
4173 | len += nodelist_scnprintf(buf + len, |
4174 | PAGE_SIZE - len - 50, | |
4175 | l->nodes); | |
45edfa58 CL |
4176 | } |
4177 | ||
e374d483 | 4178 | len += sprintf(buf + len, "\n"); |
88a420e4 CL |
4179 | } |
4180 | ||
4181 | free_loc_track(&t); | |
bbd7d57b | 4182 | kfree(map); |
88a420e4 | 4183 | if (!t.count) |
e374d483 HH |
4184 | len += sprintf(buf, "No data\n"); |
4185 | return len; | |
88a420e4 | 4186 | } |
ab4d5ed5 | 4187 | #endif |
88a420e4 | 4188 | |
a5a84755 CL |
4189 | #ifdef SLUB_RESILIENCY_TEST |
4190 | static void resiliency_test(void) | |
4191 | { | |
4192 | u8 *p; | |
4193 | ||
95a05b42 | 4194 | BUILD_BUG_ON(KMALLOC_MIN_SIZE > 16 || KMALLOC_SHIFT_HIGH < 10); |
a5a84755 | 4195 | |
f9f58285 FF |
4196 | pr_err("SLUB resiliency testing\n"); |
4197 | pr_err("-----------------------\n"); | |
4198 | pr_err("A. Corruption after allocation\n"); | |
a5a84755 CL |
4199 | |
4200 | p = kzalloc(16, GFP_KERNEL); | |
4201 | p[16] = 0x12; | |
f9f58285 FF |
4202 | pr_err("\n1. kmalloc-16: Clobber Redzone/next pointer 0x12->0x%p\n\n", |
4203 | p + 16); | |
a5a84755 CL |
4204 | |
4205 | validate_slab_cache(kmalloc_caches[4]); | |
4206 | ||
4207 | /* Hmmm... The next two are dangerous */ | |
4208 | p = kzalloc(32, GFP_KERNEL); | |
4209 | p[32 + sizeof(void *)] = 0x34; | |
f9f58285 FF |
4210 | pr_err("\n2. kmalloc-32: Clobber next pointer/next slab 0x34 -> -0x%p\n", |
4211 | p); | |
4212 | pr_err("If allocated object is overwritten then not detectable\n\n"); | |
a5a84755 CL |
4213 | |
4214 | validate_slab_cache(kmalloc_caches[5]); | |
4215 | p = kzalloc(64, GFP_KERNEL); | |
4216 | p += 64 + (get_cycles() & 0xff) * sizeof(void *); | |
4217 | *p = 0x56; | |
f9f58285 FF |
4218 | pr_err("\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n", |
4219 | p); | |
4220 | pr_err("If allocated object is overwritten then not detectable\n\n"); | |
a5a84755 CL |
4221 | validate_slab_cache(kmalloc_caches[6]); |
4222 | ||
f9f58285 | 4223 | pr_err("\nB. Corruption after free\n"); |
a5a84755 CL |
4224 | p = kzalloc(128, GFP_KERNEL); |
4225 | kfree(p); | |
4226 | *p = 0x78; | |
f9f58285 | 4227 | pr_err("1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p); |
a5a84755 CL |
4228 | validate_slab_cache(kmalloc_caches[7]); |
4229 | ||
4230 | p = kzalloc(256, GFP_KERNEL); | |
4231 | kfree(p); | |
4232 | p[50] = 0x9a; | |
f9f58285 | 4233 | pr_err("\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n", p); |
a5a84755 CL |
4234 | validate_slab_cache(kmalloc_caches[8]); |
4235 | ||
4236 | p = kzalloc(512, GFP_KERNEL); | |
4237 | kfree(p); | |
4238 | p[512] = 0xab; | |
f9f58285 | 4239 | pr_err("\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p); |
a5a84755 CL |
4240 | validate_slab_cache(kmalloc_caches[9]); |
4241 | } | |
4242 | #else | |
4243 | #ifdef CONFIG_SYSFS | |
4244 | static void resiliency_test(void) {}; | |
4245 | #endif | |
4246 | #endif | |
4247 | ||
ab4d5ed5 | 4248 | #ifdef CONFIG_SYSFS |
81819f0f | 4249 | enum slab_stat_type { |
205ab99d CL |
4250 | SL_ALL, /* All slabs */ |
4251 | SL_PARTIAL, /* Only partially allocated slabs */ | |
4252 | SL_CPU, /* Only slabs used for cpu caches */ | |
4253 | SL_OBJECTS, /* Determine allocated objects not slabs */ | |
4254 | SL_TOTAL /* Determine object capacity not slabs */ | |
81819f0f CL |
4255 | }; |
4256 | ||
205ab99d | 4257 | #define SO_ALL (1 << SL_ALL) |
81819f0f CL |
4258 | #define SO_PARTIAL (1 << SL_PARTIAL) |
4259 | #define SO_CPU (1 << SL_CPU) | |
4260 | #define SO_OBJECTS (1 << SL_OBJECTS) | |
205ab99d | 4261 | #define SO_TOTAL (1 << SL_TOTAL) |
81819f0f | 4262 | |
62e5c4b4 CG |
4263 | static ssize_t show_slab_objects(struct kmem_cache *s, |
4264 | char *buf, unsigned long flags) | |
81819f0f CL |
4265 | { |
4266 | unsigned long total = 0; | |
81819f0f CL |
4267 | int node; |
4268 | int x; | |
4269 | unsigned long *nodes; | |
81819f0f | 4270 | |
e35e1a97 | 4271 | nodes = kzalloc(sizeof(unsigned long) * nr_node_ids, GFP_KERNEL); |
62e5c4b4 CG |
4272 | if (!nodes) |
4273 | return -ENOMEM; | |
81819f0f | 4274 | |
205ab99d CL |
4275 | if (flags & SO_CPU) { |
4276 | int cpu; | |
81819f0f | 4277 | |
205ab99d | 4278 | for_each_possible_cpu(cpu) { |
d0e0ac97 CG |
4279 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, |
4280 | cpu); | |
ec3ab083 | 4281 | int node; |
49e22585 | 4282 | struct page *page; |
dfb4f096 | 4283 | |
bc6697d8 | 4284 | page = ACCESS_ONCE(c->page); |
ec3ab083 CL |
4285 | if (!page) |
4286 | continue; | |
205ab99d | 4287 | |
ec3ab083 CL |
4288 | node = page_to_nid(page); |
4289 | if (flags & SO_TOTAL) | |
4290 | x = page->objects; | |
4291 | else if (flags & SO_OBJECTS) | |
4292 | x = page->inuse; | |
4293 | else | |
4294 | x = 1; | |
49e22585 | 4295 | |
ec3ab083 CL |
4296 | total += x; |
4297 | nodes[node] += x; | |
4298 | ||
4299 | page = ACCESS_ONCE(c->partial); | |
49e22585 | 4300 | if (page) { |
8afb1474 LZ |
4301 | node = page_to_nid(page); |
4302 | if (flags & SO_TOTAL) | |
4303 | WARN_ON_ONCE(1); | |
4304 | else if (flags & SO_OBJECTS) | |
4305 | WARN_ON_ONCE(1); | |
4306 | else | |
4307 | x = page->pages; | |
bc6697d8 ED |
4308 | total += x; |
4309 | nodes[node] += x; | |
49e22585 | 4310 | } |
81819f0f CL |
4311 | } |
4312 | } | |
4313 | ||
bfc8c901 | 4314 | get_online_mems(); |
ab4d5ed5 | 4315 | #ifdef CONFIG_SLUB_DEBUG |
205ab99d | 4316 | if (flags & SO_ALL) { |
fa45dc25 CL |
4317 | struct kmem_cache_node *n; |
4318 | ||
4319 | for_each_kmem_cache_node(s, node, n) { | |
205ab99d | 4320 | |
d0e0ac97 CG |
4321 | if (flags & SO_TOTAL) |
4322 | x = atomic_long_read(&n->total_objects); | |
4323 | else if (flags & SO_OBJECTS) | |
4324 | x = atomic_long_read(&n->total_objects) - | |
4325 | count_partial(n, count_free); | |
81819f0f | 4326 | else |
205ab99d | 4327 | x = atomic_long_read(&n->nr_slabs); |
81819f0f CL |
4328 | total += x; |
4329 | nodes[node] += x; | |
4330 | } | |
4331 | ||
ab4d5ed5 CL |
4332 | } else |
4333 | #endif | |
4334 | if (flags & SO_PARTIAL) { | |
fa45dc25 | 4335 | struct kmem_cache_node *n; |
81819f0f | 4336 | |
fa45dc25 | 4337 | for_each_kmem_cache_node(s, node, n) { |
205ab99d CL |
4338 | if (flags & SO_TOTAL) |
4339 | x = count_partial(n, count_total); | |
4340 | else if (flags & SO_OBJECTS) | |
4341 | x = count_partial(n, count_inuse); | |
81819f0f | 4342 | else |
205ab99d | 4343 | x = n->nr_partial; |
81819f0f CL |
4344 | total += x; |
4345 | nodes[node] += x; | |
4346 | } | |
4347 | } | |
81819f0f CL |
4348 | x = sprintf(buf, "%lu", total); |
4349 | #ifdef CONFIG_NUMA | |
fa45dc25 | 4350 | for (node = 0; node < nr_node_ids; node++) |
81819f0f CL |
4351 | if (nodes[node]) |
4352 | x += sprintf(buf + x, " N%d=%lu", | |
4353 | node, nodes[node]); | |
4354 | #endif | |
bfc8c901 | 4355 | put_online_mems(); |
81819f0f CL |
4356 | kfree(nodes); |
4357 | return x + sprintf(buf + x, "\n"); | |
4358 | } | |
4359 | ||
ab4d5ed5 | 4360 | #ifdef CONFIG_SLUB_DEBUG |
81819f0f CL |
4361 | static int any_slab_objects(struct kmem_cache *s) |
4362 | { | |
4363 | int node; | |
fa45dc25 | 4364 | struct kmem_cache_node *n; |
81819f0f | 4365 | |
fa45dc25 | 4366 | for_each_kmem_cache_node(s, node, n) |
4ea33e2d | 4367 | if (atomic_long_read(&n->total_objects)) |
81819f0f | 4368 | return 1; |
fa45dc25 | 4369 | |
81819f0f CL |
4370 | return 0; |
4371 | } | |
ab4d5ed5 | 4372 | #endif |
81819f0f CL |
4373 | |
4374 | #define to_slab_attr(n) container_of(n, struct slab_attribute, attr) | |
497888cf | 4375 | #define to_slab(n) container_of(n, struct kmem_cache, kobj) |
81819f0f CL |
4376 | |
4377 | struct slab_attribute { | |
4378 | struct attribute attr; | |
4379 | ssize_t (*show)(struct kmem_cache *s, char *buf); | |
4380 | ssize_t (*store)(struct kmem_cache *s, const char *x, size_t count); | |
4381 | }; | |
4382 | ||
4383 | #define SLAB_ATTR_RO(_name) \ | |
ab067e99 VK |
4384 | static struct slab_attribute _name##_attr = \ |
4385 | __ATTR(_name, 0400, _name##_show, NULL) | |
81819f0f CL |
4386 | |
4387 | #define SLAB_ATTR(_name) \ | |
4388 | static struct slab_attribute _name##_attr = \ | |
ab067e99 | 4389 | __ATTR(_name, 0600, _name##_show, _name##_store) |
81819f0f | 4390 | |
81819f0f CL |
4391 | static ssize_t slab_size_show(struct kmem_cache *s, char *buf) |
4392 | { | |
4393 | return sprintf(buf, "%d\n", s->size); | |
4394 | } | |
4395 | SLAB_ATTR_RO(slab_size); | |
4396 | ||
4397 | static ssize_t align_show(struct kmem_cache *s, char *buf) | |
4398 | { | |
4399 | return sprintf(buf, "%d\n", s->align); | |
4400 | } | |
4401 | SLAB_ATTR_RO(align); | |
4402 | ||
4403 | static ssize_t object_size_show(struct kmem_cache *s, char *buf) | |
4404 | { | |
3b0efdfa | 4405 | return sprintf(buf, "%d\n", s->object_size); |
81819f0f CL |
4406 | } |
4407 | SLAB_ATTR_RO(object_size); | |
4408 | ||
4409 | static ssize_t objs_per_slab_show(struct kmem_cache *s, char *buf) | |
4410 | { | |
834f3d11 | 4411 | return sprintf(buf, "%d\n", oo_objects(s->oo)); |
81819f0f CL |
4412 | } |
4413 | SLAB_ATTR_RO(objs_per_slab); | |
4414 | ||
06b285dc CL |
4415 | static ssize_t order_store(struct kmem_cache *s, |
4416 | const char *buf, size_t length) | |
4417 | { | |
0121c619 CL |
4418 | unsigned long order; |
4419 | int err; | |
4420 | ||
3dbb95f7 | 4421 | err = kstrtoul(buf, 10, &order); |
0121c619 CL |
4422 | if (err) |
4423 | return err; | |
06b285dc CL |
4424 | |
4425 | if (order > slub_max_order || order < slub_min_order) | |
4426 | return -EINVAL; | |
4427 | ||
4428 | calculate_sizes(s, order); | |
4429 | return length; | |
4430 | } | |
4431 | ||
81819f0f CL |
4432 | static ssize_t order_show(struct kmem_cache *s, char *buf) |
4433 | { | |
834f3d11 | 4434 | return sprintf(buf, "%d\n", oo_order(s->oo)); |
81819f0f | 4435 | } |
06b285dc | 4436 | SLAB_ATTR(order); |
81819f0f | 4437 | |
73d342b1 DR |
4438 | static ssize_t min_partial_show(struct kmem_cache *s, char *buf) |
4439 | { | |
4440 | return sprintf(buf, "%lu\n", s->min_partial); | |
4441 | } | |
4442 | ||
4443 | static ssize_t min_partial_store(struct kmem_cache *s, const char *buf, | |
4444 | size_t length) | |
4445 | { | |
4446 | unsigned long min; | |
4447 | int err; | |
4448 | ||
3dbb95f7 | 4449 | err = kstrtoul(buf, 10, &min); |
73d342b1 DR |
4450 | if (err) |
4451 | return err; | |
4452 | ||
c0bdb232 | 4453 | set_min_partial(s, min); |
73d342b1 DR |
4454 | return length; |
4455 | } | |
4456 | SLAB_ATTR(min_partial); | |
4457 | ||
49e22585 CL |
4458 | static ssize_t cpu_partial_show(struct kmem_cache *s, char *buf) |
4459 | { | |
4460 | return sprintf(buf, "%u\n", s->cpu_partial); | |
4461 | } | |
4462 | ||
4463 | static ssize_t cpu_partial_store(struct kmem_cache *s, const char *buf, | |
4464 | size_t length) | |
4465 | { | |
4466 | unsigned long objects; | |
4467 | int err; | |
4468 | ||
3dbb95f7 | 4469 | err = kstrtoul(buf, 10, &objects); |
49e22585 CL |
4470 | if (err) |
4471 | return err; | |
345c905d | 4472 | if (objects && !kmem_cache_has_cpu_partial(s)) |
74ee4ef1 | 4473 | return -EINVAL; |
49e22585 CL |
4474 | |
4475 | s->cpu_partial = objects; | |
4476 | flush_all(s); | |
4477 | return length; | |
4478 | } | |
4479 | SLAB_ATTR(cpu_partial); | |
4480 | ||
81819f0f CL |
4481 | static ssize_t ctor_show(struct kmem_cache *s, char *buf) |
4482 | { | |
62c70bce JP |
4483 | if (!s->ctor) |
4484 | return 0; | |
4485 | return sprintf(buf, "%pS\n", s->ctor); | |
81819f0f CL |
4486 | } |
4487 | SLAB_ATTR_RO(ctor); | |
4488 | ||
81819f0f CL |
4489 | static ssize_t aliases_show(struct kmem_cache *s, char *buf) |
4490 | { | |
4491 | return sprintf(buf, "%d\n", s->refcount - 1); | |
4492 | } | |
4493 | SLAB_ATTR_RO(aliases); | |
4494 | ||
81819f0f CL |
4495 | static ssize_t partial_show(struct kmem_cache *s, char *buf) |
4496 | { | |
d9acf4b7 | 4497 | return show_slab_objects(s, buf, SO_PARTIAL); |
81819f0f CL |
4498 | } |
4499 | SLAB_ATTR_RO(partial); | |
4500 | ||
4501 | static ssize_t cpu_slabs_show(struct kmem_cache *s, char *buf) | |
4502 | { | |
d9acf4b7 | 4503 | return show_slab_objects(s, buf, SO_CPU); |
81819f0f CL |
4504 | } |
4505 | SLAB_ATTR_RO(cpu_slabs); | |
4506 | ||
4507 | static ssize_t objects_show(struct kmem_cache *s, char *buf) | |
4508 | { | |
205ab99d | 4509 | return show_slab_objects(s, buf, SO_ALL|SO_OBJECTS); |
81819f0f CL |
4510 | } |
4511 | SLAB_ATTR_RO(objects); | |
4512 | ||
205ab99d CL |
4513 | static ssize_t objects_partial_show(struct kmem_cache *s, char *buf) |
4514 | { | |
4515 | return show_slab_objects(s, buf, SO_PARTIAL|SO_OBJECTS); | |
4516 | } | |
4517 | SLAB_ATTR_RO(objects_partial); | |
4518 | ||
49e22585 CL |
4519 | static ssize_t slabs_cpu_partial_show(struct kmem_cache *s, char *buf) |
4520 | { | |
4521 | int objects = 0; | |
4522 | int pages = 0; | |
4523 | int cpu; | |
4524 | int len; | |
4525 | ||
4526 | for_each_online_cpu(cpu) { | |
4527 | struct page *page = per_cpu_ptr(s->cpu_slab, cpu)->partial; | |
4528 | ||
4529 | if (page) { | |
4530 | pages += page->pages; | |
4531 | objects += page->pobjects; | |
4532 | } | |
4533 | } | |
4534 | ||
4535 | len = sprintf(buf, "%d(%d)", objects, pages); | |
4536 | ||
4537 | #ifdef CONFIG_SMP | |
4538 | for_each_online_cpu(cpu) { | |
4539 | struct page *page = per_cpu_ptr(s->cpu_slab, cpu) ->partial; | |
4540 | ||
4541 | if (page && len < PAGE_SIZE - 20) | |
4542 | len += sprintf(buf + len, " C%d=%d(%d)", cpu, | |
4543 | page->pobjects, page->pages); | |
4544 | } | |
4545 | #endif | |
4546 | return len + sprintf(buf + len, "\n"); | |
4547 | } | |
4548 | SLAB_ATTR_RO(slabs_cpu_partial); | |
4549 | ||
a5a84755 CL |
4550 | static ssize_t reclaim_account_show(struct kmem_cache *s, char *buf) |
4551 | { | |
4552 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_RECLAIM_ACCOUNT)); | |
4553 | } | |
4554 | ||
4555 | static ssize_t reclaim_account_store(struct kmem_cache *s, | |
4556 | const char *buf, size_t length) | |
4557 | { | |
4558 | s->flags &= ~SLAB_RECLAIM_ACCOUNT; | |
4559 | if (buf[0] == '1') | |
4560 | s->flags |= SLAB_RECLAIM_ACCOUNT; | |
4561 | return length; | |
4562 | } | |
4563 | SLAB_ATTR(reclaim_account); | |
4564 | ||
4565 | static ssize_t hwcache_align_show(struct kmem_cache *s, char *buf) | |
4566 | { | |
4567 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_HWCACHE_ALIGN)); | |
4568 | } | |
4569 | SLAB_ATTR_RO(hwcache_align); | |
4570 | ||
4571 | #ifdef CONFIG_ZONE_DMA | |
4572 | static ssize_t cache_dma_show(struct kmem_cache *s, char *buf) | |
4573 | { | |
4574 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_CACHE_DMA)); | |
4575 | } | |
4576 | SLAB_ATTR_RO(cache_dma); | |
4577 | #endif | |
4578 | ||
4579 | static ssize_t destroy_by_rcu_show(struct kmem_cache *s, char *buf) | |
4580 | { | |
4581 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_DESTROY_BY_RCU)); | |
4582 | } | |
4583 | SLAB_ATTR_RO(destroy_by_rcu); | |
4584 | ||
ab9a0f19 LJ |
4585 | static ssize_t reserved_show(struct kmem_cache *s, char *buf) |
4586 | { | |
4587 | return sprintf(buf, "%d\n", s->reserved); | |
4588 | } | |
4589 | SLAB_ATTR_RO(reserved); | |
4590 | ||
ab4d5ed5 | 4591 | #ifdef CONFIG_SLUB_DEBUG |
a5a84755 CL |
4592 | static ssize_t slabs_show(struct kmem_cache *s, char *buf) |
4593 | { | |
4594 | return show_slab_objects(s, buf, SO_ALL); | |
4595 | } | |
4596 | SLAB_ATTR_RO(slabs); | |
4597 | ||
205ab99d CL |
4598 | static ssize_t total_objects_show(struct kmem_cache *s, char *buf) |
4599 | { | |
4600 | return show_slab_objects(s, buf, SO_ALL|SO_TOTAL); | |
4601 | } | |
4602 | SLAB_ATTR_RO(total_objects); | |
4603 | ||
81819f0f CL |
4604 | static ssize_t sanity_checks_show(struct kmem_cache *s, char *buf) |
4605 | { | |
4606 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_DEBUG_FREE)); | |
4607 | } | |
4608 | ||
4609 | static ssize_t sanity_checks_store(struct kmem_cache *s, | |
4610 | const char *buf, size_t length) | |
4611 | { | |
4612 | s->flags &= ~SLAB_DEBUG_FREE; | |
b789ef51 CL |
4613 | if (buf[0] == '1') { |
4614 | s->flags &= ~__CMPXCHG_DOUBLE; | |
81819f0f | 4615 | s->flags |= SLAB_DEBUG_FREE; |
b789ef51 | 4616 | } |
81819f0f CL |
4617 | return length; |
4618 | } | |
4619 | SLAB_ATTR(sanity_checks); | |
4620 | ||
4621 | static ssize_t trace_show(struct kmem_cache *s, char *buf) | |
4622 | { | |
4623 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_TRACE)); | |
4624 | } | |
4625 | ||
4626 | static ssize_t trace_store(struct kmem_cache *s, const char *buf, | |
4627 | size_t length) | |
4628 | { | |
4629 | s->flags &= ~SLAB_TRACE; | |
b789ef51 CL |
4630 | if (buf[0] == '1') { |
4631 | s->flags &= ~__CMPXCHG_DOUBLE; | |
81819f0f | 4632 | s->flags |= SLAB_TRACE; |
b789ef51 | 4633 | } |
81819f0f CL |
4634 | return length; |
4635 | } | |
4636 | SLAB_ATTR(trace); | |
4637 | ||
81819f0f CL |
4638 | static ssize_t red_zone_show(struct kmem_cache *s, char *buf) |
4639 | { | |
4640 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_RED_ZONE)); | |
4641 | } | |
4642 | ||
4643 | static ssize_t red_zone_store(struct kmem_cache *s, | |
4644 | const char *buf, size_t length) | |
4645 | { | |
4646 | if (any_slab_objects(s)) | |
4647 | return -EBUSY; | |
4648 | ||
4649 | s->flags &= ~SLAB_RED_ZONE; | |
b789ef51 CL |
4650 | if (buf[0] == '1') { |
4651 | s->flags &= ~__CMPXCHG_DOUBLE; | |
81819f0f | 4652 | s->flags |= SLAB_RED_ZONE; |
b789ef51 | 4653 | } |
06b285dc | 4654 | calculate_sizes(s, -1); |
81819f0f CL |
4655 | return length; |
4656 | } | |
4657 | SLAB_ATTR(red_zone); | |
4658 | ||
4659 | static ssize_t poison_show(struct kmem_cache *s, char *buf) | |
4660 | { | |
4661 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_POISON)); | |
4662 | } | |
4663 | ||
4664 | static ssize_t poison_store(struct kmem_cache *s, | |
4665 | const char *buf, size_t length) | |
4666 | { | |
4667 | if (any_slab_objects(s)) | |
4668 | return -EBUSY; | |
4669 | ||
4670 | s->flags &= ~SLAB_POISON; | |
b789ef51 CL |
4671 | if (buf[0] == '1') { |
4672 | s->flags &= ~__CMPXCHG_DOUBLE; | |
81819f0f | 4673 | s->flags |= SLAB_POISON; |
b789ef51 | 4674 | } |
06b285dc | 4675 | calculate_sizes(s, -1); |
81819f0f CL |
4676 | return length; |
4677 | } | |
4678 | SLAB_ATTR(poison); | |
4679 | ||
4680 | static ssize_t store_user_show(struct kmem_cache *s, char *buf) | |
4681 | { | |
4682 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_STORE_USER)); | |
4683 | } | |
4684 | ||
4685 | static ssize_t store_user_store(struct kmem_cache *s, | |
4686 | const char *buf, size_t length) | |
4687 | { | |
4688 | if (any_slab_objects(s)) | |
4689 | return -EBUSY; | |
4690 | ||
4691 | s->flags &= ~SLAB_STORE_USER; | |
b789ef51 CL |
4692 | if (buf[0] == '1') { |
4693 | s->flags &= ~__CMPXCHG_DOUBLE; | |
81819f0f | 4694 | s->flags |= SLAB_STORE_USER; |
b789ef51 | 4695 | } |
06b285dc | 4696 | calculate_sizes(s, -1); |
81819f0f CL |
4697 | return length; |
4698 | } | |
4699 | SLAB_ATTR(store_user); | |
4700 | ||
53e15af0 CL |
4701 | static ssize_t validate_show(struct kmem_cache *s, char *buf) |
4702 | { | |
4703 | return 0; | |
4704 | } | |
4705 | ||
4706 | static ssize_t validate_store(struct kmem_cache *s, | |
4707 | const char *buf, size_t length) | |
4708 | { | |
434e245d CL |
4709 | int ret = -EINVAL; |
4710 | ||
4711 | if (buf[0] == '1') { | |
4712 | ret = validate_slab_cache(s); | |
4713 | if (ret >= 0) | |
4714 | ret = length; | |
4715 | } | |
4716 | return ret; | |
53e15af0 CL |
4717 | } |
4718 | SLAB_ATTR(validate); | |
a5a84755 CL |
4719 | |
4720 | static ssize_t alloc_calls_show(struct kmem_cache *s, char *buf) | |
4721 | { | |
4722 | if (!(s->flags & SLAB_STORE_USER)) | |
4723 | return -ENOSYS; | |
4724 | return list_locations(s, buf, TRACK_ALLOC); | |
4725 | } | |
4726 | SLAB_ATTR_RO(alloc_calls); | |
4727 | ||
4728 | static ssize_t free_calls_show(struct kmem_cache *s, char *buf) | |
4729 | { | |
4730 | if (!(s->flags & SLAB_STORE_USER)) | |
4731 | return -ENOSYS; | |
4732 | return list_locations(s, buf, TRACK_FREE); | |
4733 | } | |
4734 | SLAB_ATTR_RO(free_calls); | |
4735 | #endif /* CONFIG_SLUB_DEBUG */ | |
4736 | ||
4737 | #ifdef CONFIG_FAILSLAB | |
4738 | static ssize_t failslab_show(struct kmem_cache *s, char *buf) | |
4739 | { | |
4740 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_FAILSLAB)); | |
4741 | } | |
4742 | ||
4743 | static ssize_t failslab_store(struct kmem_cache *s, const char *buf, | |
4744 | size_t length) | |
4745 | { | |
4746 | s->flags &= ~SLAB_FAILSLAB; | |
4747 | if (buf[0] == '1') | |
4748 | s->flags |= SLAB_FAILSLAB; | |
4749 | return length; | |
4750 | } | |
4751 | SLAB_ATTR(failslab); | |
ab4d5ed5 | 4752 | #endif |
53e15af0 | 4753 | |
2086d26a CL |
4754 | static ssize_t shrink_show(struct kmem_cache *s, char *buf) |
4755 | { | |
4756 | return 0; | |
4757 | } | |
4758 | ||
4759 | static ssize_t shrink_store(struct kmem_cache *s, | |
4760 | const char *buf, size_t length) | |
4761 | { | |
4762 | if (buf[0] == '1') { | |
4763 | int rc = kmem_cache_shrink(s); | |
4764 | ||
4765 | if (rc) | |
4766 | return rc; | |
4767 | } else | |
4768 | return -EINVAL; | |
4769 | return length; | |
4770 | } | |
4771 | SLAB_ATTR(shrink); | |
4772 | ||
81819f0f | 4773 | #ifdef CONFIG_NUMA |
9824601e | 4774 | static ssize_t remote_node_defrag_ratio_show(struct kmem_cache *s, char *buf) |
81819f0f | 4775 | { |
9824601e | 4776 | return sprintf(buf, "%d\n", s->remote_node_defrag_ratio / 10); |
81819f0f CL |
4777 | } |
4778 | ||
9824601e | 4779 | static ssize_t remote_node_defrag_ratio_store(struct kmem_cache *s, |
81819f0f CL |
4780 | const char *buf, size_t length) |
4781 | { | |
0121c619 CL |
4782 | unsigned long ratio; |
4783 | int err; | |
4784 | ||
3dbb95f7 | 4785 | err = kstrtoul(buf, 10, &ratio); |
0121c619 CL |
4786 | if (err) |
4787 | return err; | |
4788 | ||
e2cb96b7 | 4789 | if (ratio <= 100) |
0121c619 | 4790 | s->remote_node_defrag_ratio = ratio * 10; |
81819f0f | 4791 | |
81819f0f CL |
4792 | return length; |
4793 | } | |
9824601e | 4794 | SLAB_ATTR(remote_node_defrag_ratio); |
81819f0f CL |
4795 | #endif |
4796 | ||
8ff12cfc | 4797 | #ifdef CONFIG_SLUB_STATS |
8ff12cfc CL |
4798 | static int show_stat(struct kmem_cache *s, char *buf, enum stat_item si) |
4799 | { | |
4800 | unsigned long sum = 0; | |
4801 | int cpu; | |
4802 | int len; | |
4803 | int *data = kmalloc(nr_cpu_ids * sizeof(int), GFP_KERNEL); | |
4804 | ||
4805 | if (!data) | |
4806 | return -ENOMEM; | |
4807 | ||
4808 | for_each_online_cpu(cpu) { | |
9dfc6e68 | 4809 | unsigned x = per_cpu_ptr(s->cpu_slab, cpu)->stat[si]; |
8ff12cfc CL |
4810 | |
4811 | data[cpu] = x; | |
4812 | sum += x; | |
4813 | } | |
4814 | ||
4815 | len = sprintf(buf, "%lu", sum); | |
4816 | ||
50ef37b9 | 4817 | #ifdef CONFIG_SMP |
8ff12cfc CL |
4818 | for_each_online_cpu(cpu) { |
4819 | if (data[cpu] && len < PAGE_SIZE - 20) | |
50ef37b9 | 4820 | len += sprintf(buf + len, " C%d=%u", cpu, data[cpu]); |
8ff12cfc | 4821 | } |
50ef37b9 | 4822 | #endif |
8ff12cfc CL |
4823 | kfree(data); |
4824 | return len + sprintf(buf + len, "\n"); | |
4825 | } | |
4826 | ||
78eb00cc DR |
4827 | static void clear_stat(struct kmem_cache *s, enum stat_item si) |
4828 | { | |
4829 | int cpu; | |
4830 | ||
4831 | for_each_online_cpu(cpu) | |
9dfc6e68 | 4832 | per_cpu_ptr(s->cpu_slab, cpu)->stat[si] = 0; |
78eb00cc DR |
4833 | } |
4834 | ||
8ff12cfc CL |
4835 | #define STAT_ATTR(si, text) \ |
4836 | static ssize_t text##_show(struct kmem_cache *s, char *buf) \ | |
4837 | { \ | |
4838 | return show_stat(s, buf, si); \ | |
4839 | } \ | |
78eb00cc DR |
4840 | static ssize_t text##_store(struct kmem_cache *s, \ |
4841 | const char *buf, size_t length) \ | |
4842 | { \ | |
4843 | if (buf[0] != '0') \ | |
4844 | return -EINVAL; \ | |
4845 | clear_stat(s, si); \ | |
4846 | return length; \ | |
4847 | } \ | |
4848 | SLAB_ATTR(text); \ | |
8ff12cfc CL |
4849 | |
4850 | STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath); | |
4851 | STAT_ATTR(ALLOC_SLOWPATH, alloc_slowpath); | |
4852 | STAT_ATTR(FREE_FASTPATH, free_fastpath); | |
4853 | STAT_ATTR(FREE_SLOWPATH, free_slowpath); | |
4854 | STAT_ATTR(FREE_FROZEN, free_frozen); | |
4855 | STAT_ATTR(FREE_ADD_PARTIAL, free_add_partial); | |
4856 | STAT_ATTR(FREE_REMOVE_PARTIAL, free_remove_partial); | |
4857 | STAT_ATTR(ALLOC_FROM_PARTIAL, alloc_from_partial); | |
4858 | STAT_ATTR(ALLOC_SLAB, alloc_slab); | |
4859 | STAT_ATTR(ALLOC_REFILL, alloc_refill); | |
e36a2652 | 4860 | STAT_ATTR(ALLOC_NODE_MISMATCH, alloc_node_mismatch); |
8ff12cfc CL |
4861 | STAT_ATTR(FREE_SLAB, free_slab); |
4862 | STAT_ATTR(CPUSLAB_FLUSH, cpuslab_flush); | |
4863 | STAT_ATTR(DEACTIVATE_FULL, deactivate_full); | |
4864 | STAT_ATTR(DEACTIVATE_EMPTY, deactivate_empty); | |
4865 | STAT_ATTR(DEACTIVATE_TO_HEAD, deactivate_to_head); | |
4866 | STAT_ATTR(DEACTIVATE_TO_TAIL, deactivate_to_tail); | |
4867 | STAT_ATTR(DEACTIVATE_REMOTE_FREES, deactivate_remote_frees); | |
03e404af | 4868 | STAT_ATTR(DEACTIVATE_BYPASS, deactivate_bypass); |
65c3376a | 4869 | STAT_ATTR(ORDER_FALLBACK, order_fallback); |
b789ef51 CL |
4870 | STAT_ATTR(CMPXCHG_DOUBLE_CPU_FAIL, cmpxchg_double_cpu_fail); |
4871 | STAT_ATTR(CMPXCHG_DOUBLE_FAIL, cmpxchg_double_fail); | |
49e22585 CL |
4872 | STAT_ATTR(CPU_PARTIAL_ALLOC, cpu_partial_alloc); |
4873 | STAT_ATTR(CPU_PARTIAL_FREE, cpu_partial_free); | |
8028dcea AS |
4874 | STAT_ATTR(CPU_PARTIAL_NODE, cpu_partial_node); |
4875 | STAT_ATTR(CPU_PARTIAL_DRAIN, cpu_partial_drain); | |
8ff12cfc CL |
4876 | #endif |
4877 | ||
06428780 | 4878 | static struct attribute *slab_attrs[] = { |
81819f0f CL |
4879 | &slab_size_attr.attr, |
4880 | &object_size_attr.attr, | |
4881 | &objs_per_slab_attr.attr, | |
4882 | &order_attr.attr, | |
73d342b1 | 4883 | &min_partial_attr.attr, |
49e22585 | 4884 | &cpu_partial_attr.attr, |
81819f0f | 4885 | &objects_attr.attr, |
205ab99d | 4886 | &objects_partial_attr.attr, |
81819f0f CL |
4887 | &partial_attr.attr, |
4888 | &cpu_slabs_attr.attr, | |
4889 | &ctor_attr.attr, | |
81819f0f CL |
4890 | &aliases_attr.attr, |
4891 | &align_attr.attr, | |
81819f0f CL |
4892 | &hwcache_align_attr.attr, |
4893 | &reclaim_account_attr.attr, | |
4894 | &destroy_by_rcu_attr.attr, | |
a5a84755 | 4895 | &shrink_attr.attr, |
ab9a0f19 | 4896 | &reserved_attr.attr, |
49e22585 | 4897 | &slabs_cpu_partial_attr.attr, |
ab4d5ed5 | 4898 | #ifdef CONFIG_SLUB_DEBUG |
a5a84755 CL |
4899 | &total_objects_attr.attr, |
4900 | &slabs_attr.attr, | |
4901 | &sanity_checks_attr.attr, | |
4902 | &trace_attr.attr, | |
81819f0f CL |
4903 | &red_zone_attr.attr, |
4904 | &poison_attr.attr, | |
4905 | &store_user_attr.attr, | |
53e15af0 | 4906 | &validate_attr.attr, |
88a420e4 CL |
4907 | &alloc_calls_attr.attr, |
4908 | &free_calls_attr.attr, | |
ab4d5ed5 | 4909 | #endif |
81819f0f CL |
4910 | #ifdef CONFIG_ZONE_DMA |
4911 | &cache_dma_attr.attr, | |
4912 | #endif | |
4913 | #ifdef CONFIG_NUMA | |
9824601e | 4914 | &remote_node_defrag_ratio_attr.attr, |
8ff12cfc CL |
4915 | #endif |
4916 | #ifdef CONFIG_SLUB_STATS | |
4917 | &alloc_fastpath_attr.attr, | |
4918 | &alloc_slowpath_attr.attr, | |
4919 | &free_fastpath_attr.attr, | |
4920 | &free_slowpath_attr.attr, | |
4921 | &free_frozen_attr.attr, | |
4922 | &free_add_partial_attr.attr, | |
4923 | &free_remove_partial_attr.attr, | |
4924 | &alloc_from_partial_attr.attr, | |
4925 | &alloc_slab_attr.attr, | |
4926 | &alloc_refill_attr.attr, | |
e36a2652 | 4927 | &alloc_node_mismatch_attr.attr, |
8ff12cfc CL |
4928 | &free_slab_attr.attr, |
4929 | &cpuslab_flush_attr.attr, | |
4930 | &deactivate_full_attr.attr, | |
4931 | &deactivate_empty_attr.attr, | |
4932 | &deactivate_to_head_attr.attr, | |
4933 | &deactivate_to_tail_attr.attr, | |
4934 | &deactivate_remote_frees_attr.attr, | |
03e404af | 4935 | &deactivate_bypass_attr.attr, |
65c3376a | 4936 | &order_fallback_attr.attr, |
b789ef51 CL |
4937 | &cmpxchg_double_fail_attr.attr, |
4938 | &cmpxchg_double_cpu_fail_attr.attr, | |
49e22585 CL |
4939 | &cpu_partial_alloc_attr.attr, |
4940 | &cpu_partial_free_attr.attr, | |
8028dcea AS |
4941 | &cpu_partial_node_attr.attr, |
4942 | &cpu_partial_drain_attr.attr, | |
81819f0f | 4943 | #endif |
4c13dd3b DM |
4944 | #ifdef CONFIG_FAILSLAB |
4945 | &failslab_attr.attr, | |
4946 | #endif | |
4947 | ||
81819f0f CL |
4948 | NULL |
4949 | }; | |
4950 | ||
4951 | static struct attribute_group slab_attr_group = { | |
4952 | .attrs = slab_attrs, | |
4953 | }; | |
4954 | ||
4955 | static ssize_t slab_attr_show(struct kobject *kobj, | |
4956 | struct attribute *attr, | |
4957 | char *buf) | |
4958 | { | |
4959 | struct slab_attribute *attribute; | |
4960 | struct kmem_cache *s; | |
4961 | int err; | |
4962 | ||
4963 | attribute = to_slab_attr(attr); | |
4964 | s = to_slab(kobj); | |
4965 | ||
4966 | if (!attribute->show) | |
4967 | return -EIO; | |
4968 | ||
4969 | err = attribute->show(s, buf); | |
4970 | ||
4971 | return err; | |
4972 | } | |
4973 | ||
4974 | static ssize_t slab_attr_store(struct kobject *kobj, | |
4975 | struct attribute *attr, | |
4976 | const char *buf, size_t len) | |
4977 | { | |
4978 | struct slab_attribute *attribute; | |
4979 | struct kmem_cache *s; | |
4980 | int err; | |
4981 | ||
4982 | attribute = to_slab_attr(attr); | |
4983 | s = to_slab(kobj); | |
4984 | ||
4985 | if (!attribute->store) | |
4986 | return -EIO; | |
4987 | ||
4988 | err = attribute->store(s, buf, len); | |
107dab5c GC |
4989 | #ifdef CONFIG_MEMCG_KMEM |
4990 | if (slab_state >= FULL && err >= 0 && is_root_cache(s)) { | |
4991 | int i; | |
81819f0f | 4992 | |
107dab5c GC |
4993 | mutex_lock(&slab_mutex); |
4994 | if (s->max_attr_size < len) | |
4995 | s->max_attr_size = len; | |
4996 | ||
ebe945c2 GC |
4997 | /* |
4998 | * This is a best effort propagation, so this function's return | |
4999 | * value will be determined by the parent cache only. This is | |
5000 | * basically because not all attributes will have a well | |
5001 | * defined semantics for rollbacks - most of the actions will | |
5002 | * have permanent effects. | |
5003 | * | |
5004 | * Returning the error value of any of the children that fail | |
5005 | * is not 100 % defined, in the sense that users seeing the | |
5006 | * error code won't be able to know anything about the state of | |
5007 | * the cache. | |
5008 | * | |
5009 | * Only returning the error code for the parent cache at least | |
5010 | * has well defined semantics. The cache being written to | |
5011 | * directly either failed or succeeded, in which case we loop | |
5012 | * through the descendants with best-effort propagation. | |
5013 | */ | |
107dab5c | 5014 | for_each_memcg_cache_index(i) { |
2ade4de8 | 5015 | struct kmem_cache *c = cache_from_memcg_idx(s, i); |
107dab5c GC |
5016 | if (c) |
5017 | attribute->store(c, buf, len); | |
5018 | } | |
5019 | mutex_unlock(&slab_mutex); | |
5020 | } | |
5021 | #endif | |
81819f0f CL |
5022 | return err; |
5023 | } | |
5024 | ||
107dab5c GC |
5025 | static void memcg_propagate_slab_attrs(struct kmem_cache *s) |
5026 | { | |
5027 | #ifdef CONFIG_MEMCG_KMEM | |
5028 | int i; | |
5029 | char *buffer = NULL; | |
93030d83 | 5030 | struct kmem_cache *root_cache; |
107dab5c | 5031 | |
93030d83 | 5032 | if (is_root_cache(s)) |
107dab5c GC |
5033 | return; |
5034 | ||
93030d83 VD |
5035 | root_cache = s->memcg_params->root_cache; |
5036 | ||
107dab5c GC |
5037 | /* |
5038 | * This mean this cache had no attribute written. Therefore, no point | |
5039 | * in copying default values around | |
5040 | */ | |
93030d83 | 5041 | if (!root_cache->max_attr_size) |
107dab5c GC |
5042 | return; |
5043 | ||
5044 | for (i = 0; i < ARRAY_SIZE(slab_attrs); i++) { | |
5045 | char mbuf[64]; | |
5046 | char *buf; | |
5047 | struct slab_attribute *attr = to_slab_attr(slab_attrs[i]); | |
5048 | ||
5049 | if (!attr || !attr->store || !attr->show) | |
5050 | continue; | |
5051 | ||
5052 | /* | |
5053 | * It is really bad that we have to allocate here, so we will | |
5054 | * do it only as a fallback. If we actually allocate, though, | |
5055 | * we can just use the allocated buffer until the end. | |
5056 | * | |
5057 | * Most of the slub attributes will tend to be very small in | |
5058 | * size, but sysfs allows buffers up to a page, so they can | |
5059 | * theoretically happen. | |
5060 | */ | |
5061 | if (buffer) | |
5062 | buf = buffer; | |
93030d83 | 5063 | else if (root_cache->max_attr_size < ARRAY_SIZE(mbuf)) |
107dab5c GC |
5064 | buf = mbuf; |
5065 | else { | |
5066 | buffer = (char *) get_zeroed_page(GFP_KERNEL); | |
5067 | if (WARN_ON(!buffer)) | |
5068 | continue; | |
5069 | buf = buffer; | |
5070 | } | |
5071 | ||
93030d83 | 5072 | attr->show(root_cache, buf); |
107dab5c GC |
5073 | attr->store(s, buf, strlen(buf)); |
5074 | } | |
5075 | ||
5076 | if (buffer) | |
5077 | free_page((unsigned long)buffer); | |
5078 | #endif | |
5079 | } | |
5080 | ||
41a21285 CL |
5081 | static void kmem_cache_release(struct kobject *k) |
5082 | { | |
5083 | slab_kmem_cache_release(to_slab(k)); | |
5084 | } | |
5085 | ||
52cf25d0 | 5086 | static const struct sysfs_ops slab_sysfs_ops = { |
81819f0f CL |
5087 | .show = slab_attr_show, |
5088 | .store = slab_attr_store, | |
5089 | }; | |
5090 | ||
5091 | static struct kobj_type slab_ktype = { | |
5092 | .sysfs_ops = &slab_sysfs_ops, | |
41a21285 | 5093 | .release = kmem_cache_release, |
81819f0f CL |
5094 | }; |
5095 | ||
5096 | static int uevent_filter(struct kset *kset, struct kobject *kobj) | |
5097 | { | |
5098 | struct kobj_type *ktype = get_ktype(kobj); | |
5099 | ||
5100 | if (ktype == &slab_ktype) | |
5101 | return 1; | |
5102 | return 0; | |
5103 | } | |
5104 | ||
9cd43611 | 5105 | static const struct kset_uevent_ops slab_uevent_ops = { |
81819f0f CL |
5106 | .filter = uevent_filter, |
5107 | }; | |
5108 | ||
27c3a314 | 5109 | static struct kset *slab_kset; |
81819f0f | 5110 | |
9a41707b VD |
5111 | static inline struct kset *cache_kset(struct kmem_cache *s) |
5112 | { | |
5113 | #ifdef CONFIG_MEMCG_KMEM | |
5114 | if (!is_root_cache(s)) | |
5115 | return s->memcg_params->root_cache->memcg_kset; | |
5116 | #endif | |
5117 | return slab_kset; | |
5118 | } | |
5119 | ||
81819f0f CL |
5120 | #define ID_STR_LENGTH 64 |
5121 | ||
5122 | /* Create a unique string id for a slab cache: | |
6446faa2 CL |
5123 | * |
5124 | * Format :[flags-]size | |
81819f0f CL |
5125 | */ |
5126 | static char *create_unique_id(struct kmem_cache *s) | |
5127 | { | |
5128 | char *name = kmalloc(ID_STR_LENGTH, GFP_KERNEL); | |
5129 | char *p = name; | |
5130 | ||
5131 | BUG_ON(!name); | |
5132 | ||
5133 | *p++ = ':'; | |
5134 | /* | |
5135 | * First flags affecting slabcache operations. We will only | |
5136 | * get here for aliasable slabs so we do not need to support | |
5137 | * too many flags. The flags here must cover all flags that | |
5138 | * are matched during merging to guarantee that the id is | |
5139 | * unique. | |
5140 | */ | |
5141 | if (s->flags & SLAB_CACHE_DMA) | |
5142 | *p++ = 'd'; | |
5143 | if (s->flags & SLAB_RECLAIM_ACCOUNT) | |
5144 | *p++ = 'a'; | |
5145 | if (s->flags & SLAB_DEBUG_FREE) | |
5146 | *p++ = 'F'; | |
5a896d9e VN |
5147 | if (!(s->flags & SLAB_NOTRACK)) |
5148 | *p++ = 't'; | |
81819f0f CL |
5149 | if (p != name + 1) |
5150 | *p++ = '-'; | |
5151 | p += sprintf(p, "%07d", s->size); | |
2633d7a0 GC |
5152 | |
5153 | #ifdef CONFIG_MEMCG_KMEM | |
5154 | if (!is_root_cache(s)) | |
d0e0ac97 CG |
5155 | p += sprintf(p, "-%08d", |
5156 | memcg_cache_id(s->memcg_params->memcg)); | |
2633d7a0 GC |
5157 | #endif |
5158 | ||
81819f0f CL |
5159 | BUG_ON(p > name + ID_STR_LENGTH - 1); |
5160 | return name; | |
5161 | } | |
5162 | ||
5163 | static int sysfs_slab_add(struct kmem_cache *s) | |
5164 | { | |
5165 | int err; | |
5166 | const char *name; | |
45530c44 | 5167 | int unmergeable = slab_unmergeable(s); |
81819f0f | 5168 | |
81819f0f CL |
5169 | if (unmergeable) { |
5170 | /* | |
5171 | * Slabcache can never be merged so we can use the name proper. | |
5172 | * This is typically the case for debug situations. In that | |
5173 | * case we can catch duplicate names easily. | |
5174 | */ | |
27c3a314 | 5175 | sysfs_remove_link(&slab_kset->kobj, s->name); |
81819f0f CL |
5176 | name = s->name; |
5177 | } else { | |
5178 | /* | |
5179 | * Create a unique name for the slab as a target | |
5180 | * for the symlinks. | |
5181 | */ | |
5182 | name = create_unique_id(s); | |
5183 | } | |
5184 | ||
9a41707b | 5185 | s->kobj.kset = cache_kset(s); |
26e4f205 | 5186 | err = kobject_init_and_add(&s->kobj, &slab_ktype, NULL, "%s", name); |
54b6a731 DJ |
5187 | if (err) |
5188 | goto out_put_kobj; | |
81819f0f CL |
5189 | |
5190 | err = sysfs_create_group(&s->kobj, &slab_attr_group); | |
54b6a731 DJ |
5191 | if (err) |
5192 | goto out_del_kobj; | |
9a41707b VD |
5193 | |
5194 | #ifdef CONFIG_MEMCG_KMEM | |
5195 | if (is_root_cache(s)) { | |
5196 | s->memcg_kset = kset_create_and_add("cgroup", NULL, &s->kobj); | |
5197 | if (!s->memcg_kset) { | |
54b6a731 DJ |
5198 | err = -ENOMEM; |
5199 | goto out_del_kobj; | |
9a41707b VD |
5200 | } |
5201 | } | |
5202 | #endif | |
5203 | ||
81819f0f CL |
5204 | kobject_uevent(&s->kobj, KOBJ_ADD); |
5205 | if (!unmergeable) { | |
5206 | /* Setup first alias */ | |
5207 | sysfs_slab_alias(s, s->name); | |
81819f0f | 5208 | } |
54b6a731 DJ |
5209 | out: |
5210 | if (!unmergeable) | |
5211 | kfree(name); | |
5212 | return err; | |
5213 | out_del_kobj: | |
5214 | kobject_del(&s->kobj); | |
5215 | out_put_kobj: | |
5216 | kobject_put(&s->kobj); | |
5217 | goto out; | |
81819f0f CL |
5218 | } |
5219 | ||
41a21285 | 5220 | void sysfs_slab_remove(struct kmem_cache *s) |
81819f0f | 5221 | { |
97d06609 | 5222 | if (slab_state < FULL) |
2bce6485 CL |
5223 | /* |
5224 | * Sysfs has not been setup yet so no need to remove the | |
5225 | * cache from sysfs. | |
5226 | */ | |
5227 | return; | |
5228 | ||
9a41707b VD |
5229 | #ifdef CONFIG_MEMCG_KMEM |
5230 | kset_unregister(s->memcg_kset); | |
5231 | #endif | |
81819f0f CL |
5232 | kobject_uevent(&s->kobj, KOBJ_REMOVE); |
5233 | kobject_del(&s->kobj); | |
151c602f | 5234 | kobject_put(&s->kobj); |
81819f0f CL |
5235 | } |
5236 | ||
5237 | /* | |
5238 | * Need to buffer aliases during bootup until sysfs becomes | |
9f6c708e | 5239 | * available lest we lose that information. |
81819f0f CL |
5240 | */ |
5241 | struct saved_alias { | |
5242 | struct kmem_cache *s; | |
5243 | const char *name; | |
5244 | struct saved_alias *next; | |
5245 | }; | |
5246 | ||
5af328a5 | 5247 | static struct saved_alias *alias_list; |
81819f0f CL |
5248 | |
5249 | static int sysfs_slab_alias(struct kmem_cache *s, const char *name) | |
5250 | { | |
5251 | struct saved_alias *al; | |
5252 | ||
97d06609 | 5253 | if (slab_state == FULL) { |
81819f0f CL |
5254 | /* |
5255 | * If we have a leftover link then remove it. | |
5256 | */ | |
27c3a314 GKH |
5257 | sysfs_remove_link(&slab_kset->kobj, name); |
5258 | return sysfs_create_link(&slab_kset->kobj, &s->kobj, name); | |
81819f0f CL |
5259 | } |
5260 | ||
5261 | al = kmalloc(sizeof(struct saved_alias), GFP_KERNEL); | |
5262 | if (!al) | |
5263 | return -ENOMEM; | |
5264 | ||
5265 | al->s = s; | |
5266 | al->name = name; | |
5267 | al->next = alias_list; | |
5268 | alias_list = al; | |
5269 | return 0; | |
5270 | } | |
5271 | ||
5272 | static int __init slab_sysfs_init(void) | |
5273 | { | |
5b95a4ac | 5274 | struct kmem_cache *s; |
81819f0f CL |
5275 | int err; |
5276 | ||
18004c5d | 5277 | mutex_lock(&slab_mutex); |
2bce6485 | 5278 | |
0ff21e46 | 5279 | slab_kset = kset_create_and_add("slab", &slab_uevent_ops, kernel_kobj); |
27c3a314 | 5280 | if (!slab_kset) { |
18004c5d | 5281 | mutex_unlock(&slab_mutex); |
f9f58285 | 5282 | pr_err("Cannot register slab subsystem.\n"); |
81819f0f CL |
5283 | return -ENOSYS; |
5284 | } | |
5285 | ||
97d06609 | 5286 | slab_state = FULL; |
26a7bd03 | 5287 | |
5b95a4ac | 5288 | list_for_each_entry(s, &slab_caches, list) { |
26a7bd03 | 5289 | err = sysfs_slab_add(s); |
5d540fb7 | 5290 | if (err) |
f9f58285 FF |
5291 | pr_err("SLUB: Unable to add boot slab %s to sysfs\n", |
5292 | s->name); | |
26a7bd03 | 5293 | } |
81819f0f CL |
5294 | |
5295 | while (alias_list) { | |
5296 | struct saved_alias *al = alias_list; | |
5297 | ||
5298 | alias_list = alias_list->next; | |
5299 | err = sysfs_slab_alias(al->s, al->name); | |
5d540fb7 | 5300 | if (err) |
f9f58285 FF |
5301 | pr_err("SLUB: Unable to add boot slab alias %s to sysfs\n", |
5302 | al->name); | |
81819f0f CL |
5303 | kfree(al); |
5304 | } | |
5305 | ||
18004c5d | 5306 | mutex_unlock(&slab_mutex); |
81819f0f CL |
5307 | resiliency_test(); |
5308 | return 0; | |
5309 | } | |
5310 | ||
5311 | __initcall(slab_sysfs_init); | |
ab4d5ed5 | 5312 | #endif /* CONFIG_SYSFS */ |
57ed3eda PE |
5313 | |
5314 | /* | |
5315 | * The /proc/slabinfo ABI | |
5316 | */ | |
158a9624 | 5317 | #ifdef CONFIG_SLABINFO |
0d7561c6 | 5318 | void get_slabinfo(struct kmem_cache *s, struct slabinfo *sinfo) |
57ed3eda | 5319 | { |
57ed3eda | 5320 | unsigned long nr_slabs = 0; |
205ab99d CL |
5321 | unsigned long nr_objs = 0; |
5322 | unsigned long nr_free = 0; | |
57ed3eda | 5323 | int node; |
fa45dc25 | 5324 | struct kmem_cache_node *n; |
57ed3eda | 5325 | |
fa45dc25 | 5326 | for_each_kmem_cache_node(s, node, n) { |
c17fd13e WL |
5327 | nr_slabs += node_nr_slabs(n); |
5328 | nr_objs += node_nr_objs(n); | |
205ab99d | 5329 | nr_free += count_partial(n, count_free); |
57ed3eda PE |
5330 | } |
5331 | ||
0d7561c6 GC |
5332 | sinfo->active_objs = nr_objs - nr_free; |
5333 | sinfo->num_objs = nr_objs; | |
5334 | sinfo->active_slabs = nr_slabs; | |
5335 | sinfo->num_slabs = nr_slabs; | |
5336 | sinfo->objects_per_slab = oo_objects(s->oo); | |
5337 | sinfo->cache_order = oo_order(s->oo); | |
57ed3eda PE |
5338 | } |
5339 | ||
0d7561c6 | 5340 | void slabinfo_show_stats(struct seq_file *m, struct kmem_cache *s) |
7b3c3a50 | 5341 | { |
7b3c3a50 AD |
5342 | } |
5343 | ||
b7454ad3 GC |
5344 | ssize_t slabinfo_write(struct file *file, const char __user *buffer, |
5345 | size_t count, loff_t *ppos) | |
7b3c3a50 | 5346 | { |
b7454ad3 | 5347 | return -EIO; |
7b3c3a50 | 5348 | } |
158a9624 | 5349 | #endif /* CONFIG_SLABINFO */ |