]>
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 | ||
54266640 WY |
286 | #define for_each_object_idx(__p, __idx, __s, __addr, __objects) \ |
287 | for (__p = (__addr), __idx = 1; __idx <= __objects;\ | |
288 | __p += (__s)->size, __idx++) | |
289 | ||
7656c72b CL |
290 | /* Determine object index from a given position */ |
291 | static inline int slab_index(void *p, struct kmem_cache *s, void *addr) | |
292 | { | |
293 | return (p - addr) / s->size; | |
294 | } | |
295 | ||
d71f606f MK |
296 | static inline size_t slab_ksize(const struct kmem_cache *s) |
297 | { | |
298 | #ifdef CONFIG_SLUB_DEBUG | |
299 | /* | |
300 | * Debugging requires use of the padding between object | |
301 | * and whatever may come after it. | |
302 | */ | |
303 | if (s->flags & (SLAB_RED_ZONE | SLAB_POISON)) | |
3b0efdfa | 304 | return s->object_size; |
d71f606f MK |
305 | |
306 | #endif | |
307 | /* | |
308 | * If we have the need to store the freelist pointer | |
309 | * back there or track user information then we can | |
310 | * only use the space before that information. | |
311 | */ | |
312 | if (s->flags & (SLAB_DESTROY_BY_RCU | SLAB_STORE_USER)) | |
313 | return s->inuse; | |
314 | /* | |
315 | * Else we can use all the padding etc for the allocation | |
316 | */ | |
317 | return s->size; | |
318 | } | |
319 | ||
ab9a0f19 LJ |
320 | static inline int order_objects(int order, unsigned long size, int reserved) |
321 | { | |
322 | return ((PAGE_SIZE << order) - reserved) / size; | |
323 | } | |
324 | ||
834f3d11 | 325 | static inline struct kmem_cache_order_objects oo_make(int order, |
ab9a0f19 | 326 | unsigned long size, int reserved) |
834f3d11 CL |
327 | { |
328 | struct kmem_cache_order_objects x = { | |
ab9a0f19 | 329 | (order << OO_SHIFT) + order_objects(order, size, reserved) |
834f3d11 CL |
330 | }; |
331 | ||
332 | return x; | |
333 | } | |
334 | ||
335 | static inline int oo_order(struct kmem_cache_order_objects x) | |
336 | { | |
210b5c06 | 337 | return x.x >> OO_SHIFT; |
834f3d11 CL |
338 | } |
339 | ||
340 | static inline int oo_objects(struct kmem_cache_order_objects x) | |
341 | { | |
210b5c06 | 342 | return x.x & OO_MASK; |
834f3d11 CL |
343 | } |
344 | ||
881db7fb CL |
345 | /* |
346 | * Per slab locking using the pagelock | |
347 | */ | |
348 | static __always_inline void slab_lock(struct page *page) | |
349 | { | |
350 | bit_spin_lock(PG_locked, &page->flags); | |
351 | } | |
352 | ||
353 | static __always_inline void slab_unlock(struct page *page) | |
354 | { | |
355 | __bit_spin_unlock(PG_locked, &page->flags); | |
356 | } | |
357 | ||
a0320865 DH |
358 | static inline void set_page_slub_counters(struct page *page, unsigned long counters_new) |
359 | { | |
360 | struct page tmp; | |
361 | tmp.counters = counters_new; | |
362 | /* | |
363 | * page->counters can cover frozen/inuse/objects as well | |
364 | * as page->_count. If we assign to ->counters directly | |
365 | * we run the risk of losing updates to page->_count, so | |
366 | * be careful and only assign to the fields we need. | |
367 | */ | |
368 | page->frozen = tmp.frozen; | |
369 | page->inuse = tmp.inuse; | |
370 | page->objects = tmp.objects; | |
371 | } | |
372 | ||
1d07171c CL |
373 | /* Interrupts must be disabled (for the fallback code to work right) */ |
374 | static inline bool __cmpxchg_double_slab(struct kmem_cache *s, struct page *page, | |
375 | void *freelist_old, unsigned long counters_old, | |
376 | void *freelist_new, unsigned long counters_new, | |
377 | const char *n) | |
378 | { | |
379 | VM_BUG_ON(!irqs_disabled()); | |
2565409f HC |
380 | #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \ |
381 | defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE) | |
1d07171c | 382 | if (s->flags & __CMPXCHG_DOUBLE) { |
cdcd6298 | 383 | if (cmpxchg_double(&page->freelist, &page->counters, |
0aa9a13d DC |
384 | freelist_old, counters_old, |
385 | freelist_new, counters_new)) | |
386 | return 1; | |
1d07171c CL |
387 | } else |
388 | #endif | |
389 | { | |
390 | slab_lock(page); | |
d0e0ac97 CG |
391 | if (page->freelist == freelist_old && |
392 | page->counters == counters_old) { | |
1d07171c | 393 | page->freelist = freelist_new; |
a0320865 | 394 | set_page_slub_counters(page, counters_new); |
1d07171c CL |
395 | slab_unlock(page); |
396 | return 1; | |
397 | } | |
398 | slab_unlock(page); | |
399 | } | |
400 | ||
401 | cpu_relax(); | |
402 | stat(s, CMPXCHG_DOUBLE_FAIL); | |
403 | ||
404 | #ifdef SLUB_DEBUG_CMPXCHG | |
f9f58285 | 405 | pr_info("%s %s: cmpxchg double redo ", n, s->name); |
1d07171c CL |
406 | #endif |
407 | ||
408 | return 0; | |
409 | } | |
410 | ||
b789ef51 CL |
411 | static inline bool cmpxchg_double_slab(struct kmem_cache *s, struct page *page, |
412 | void *freelist_old, unsigned long counters_old, | |
413 | void *freelist_new, unsigned long counters_new, | |
414 | const char *n) | |
415 | { | |
2565409f HC |
416 | #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \ |
417 | defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE) | |
b789ef51 | 418 | if (s->flags & __CMPXCHG_DOUBLE) { |
cdcd6298 | 419 | if (cmpxchg_double(&page->freelist, &page->counters, |
0aa9a13d DC |
420 | freelist_old, counters_old, |
421 | freelist_new, counters_new)) | |
422 | return 1; | |
b789ef51 CL |
423 | } else |
424 | #endif | |
425 | { | |
1d07171c CL |
426 | unsigned long flags; |
427 | ||
428 | local_irq_save(flags); | |
881db7fb | 429 | slab_lock(page); |
d0e0ac97 CG |
430 | if (page->freelist == freelist_old && |
431 | page->counters == counters_old) { | |
b789ef51 | 432 | page->freelist = freelist_new; |
a0320865 | 433 | set_page_slub_counters(page, counters_new); |
881db7fb | 434 | slab_unlock(page); |
1d07171c | 435 | local_irq_restore(flags); |
b789ef51 CL |
436 | return 1; |
437 | } | |
881db7fb | 438 | slab_unlock(page); |
1d07171c | 439 | local_irq_restore(flags); |
b789ef51 CL |
440 | } |
441 | ||
442 | cpu_relax(); | |
443 | stat(s, CMPXCHG_DOUBLE_FAIL); | |
444 | ||
445 | #ifdef SLUB_DEBUG_CMPXCHG | |
f9f58285 | 446 | pr_info("%s %s: cmpxchg double redo ", n, s->name); |
b789ef51 CL |
447 | #endif |
448 | ||
449 | return 0; | |
450 | } | |
451 | ||
41ecc55b | 452 | #ifdef CONFIG_SLUB_DEBUG |
5f80b13a CL |
453 | /* |
454 | * Determine a map of object in use on a page. | |
455 | * | |
881db7fb | 456 | * Node listlock must be held to guarantee that the page does |
5f80b13a CL |
457 | * not vanish from under us. |
458 | */ | |
459 | static void get_map(struct kmem_cache *s, struct page *page, unsigned long *map) | |
460 | { | |
461 | void *p; | |
462 | void *addr = page_address(page); | |
463 | ||
464 | for (p = page->freelist; p; p = get_freepointer(s, p)) | |
465 | set_bit(slab_index(p, s, addr), map); | |
466 | } | |
467 | ||
41ecc55b CL |
468 | /* |
469 | * Debug settings: | |
470 | */ | |
f0630fff CL |
471 | #ifdef CONFIG_SLUB_DEBUG_ON |
472 | static int slub_debug = DEBUG_DEFAULT_FLAGS; | |
473 | #else | |
41ecc55b | 474 | static int slub_debug; |
f0630fff | 475 | #endif |
41ecc55b CL |
476 | |
477 | static char *slub_debug_slabs; | |
fa5ec8a1 | 478 | static int disable_higher_order_debug; |
41ecc55b | 479 | |
81819f0f CL |
480 | /* |
481 | * Object debugging | |
482 | */ | |
483 | static void print_section(char *text, u8 *addr, unsigned int length) | |
484 | { | |
ffc79d28 SAS |
485 | print_hex_dump(KERN_ERR, text, DUMP_PREFIX_ADDRESS, 16, 1, addr, |
486 | length, 1); | |
81819f0f CL |
487 | } |
488 | ||
81819f0f CL |
489 | static struct track *get_track(struct kmem_cache *s, void *object, |
490 | enum track_item alloc) | |
491 | { | |
492 | struct track *p; | |
493 | ||
494 | if (s->offset) | |
495 | p = object + s->offset + sizeof(void *); | |
496 | else | |
497 | p = object + s->inuse; | |
498 | ||
499 | return p + alloc; | |
500 | } | |
501 | ||
502 | static void set_track(struct kmem_cache *s, void *object, | |
ce71e27c | 503 | enum track_item alloc, unsigned long addr) |
81819f0f | 504 | { |
1a00df4a | 505 | struct track *p = get_track(s, object, alloc); |
81819f0f | 506 | |
81819f0f | 507 | if (addr) { |
d6543e39 BG |
508 | #ifdef CONFIG_STACKTRACE |
509 | struct stack_trace trace; | |
510 | int i; | |
511 | ||
512 | trace.nr_entries = 0; | |
513 | trace.max_entries = TRACK_ADDRS_COUNT; | |
514 | trace.entries = p->addrs; | |
515 | trace.skip = 3; | |
516 | save_stack_trace(&trace); | |
517 | ||
518 | /* See rant in lockdep.c */ | |
519 | if (trace.nr_entries != 0 && | |
520 | trace.entries[trace.nr_entries - 1] == ULONG_MAX) | |
521 | trace.nr_entries--; | |
522 | ||
523 | for (i = trace.nr_entries; i < TRACK_ADDRS_COUNT; i++) | |
524 | p->addrs[i] = 0; | |
525 | #endif | |
81819f0f CL |
526 | p->addr = addr; |
527 | p->cpu = smp_processor_id(); | |
88e4ccf2 | 528 | p->pid = current->pid; |
81819f0f CL |
529 | p->when = jiffies; |
530 | } else | |
531 | memset(p, 0, sizeof(struct track)); | |
532 | } | |
533 | ||
81819f0f CL |
534 | static void init_tracking(struct kmem_cache *s, void *object) |
535 | { | |
24922684 CL |
536 | if (!(s->flags & SLAB_STORE_USER)) |
537 | return; | |
538 | ||
ce71e27c EGM |
539 | set_track(s, object, TRACK_FREE, 0UL); |
540 | set_track(s, object, TRACK_ALLOC, 0UL); | |
81819f0f CL |
541 | } |
542 | ||
543 | static void print_track(const char *s, struct track *t) | |
544 | { | |
545 | if (!t->addr) | |
546 | return; | |
547 | ||
f9f58285 FF |
548 | pr_err("INFO: %s in %pS age=%lu cpu=%u pid=%d\n", |
549 | s, (void *)t->addr, jiffies - t->when, t->cpu, t->pid); | |
d6543e39 BG |
550 | #ifdef CONFIG_STACKTRACE |
551 | { | |
552 | int i; | |
553 | for (i = 0; i < TRACK_ADDRS_COUNT; i++) | |
554 | if (t->addrs[i]) | |
f9f58285 | 555 | pr_err("\t%pS\n", (void *)t->addrs[i]); |
d6543e39 BG |
556 | else |
557 | break; | |
558 | } | |
559 | #endif | |
24922684 CL |
560 | } |
561 | ||
562 | static void print_tracking(struct kmem_cache *s, void *object) | |
563 | { | |
564 | if (!(s->flags & SLAB_STORE_USER)) | |
565 | return; | |
566 | ||
567 | print_track("Allocated", get_track(s, object, TRACK_ALLOC)); | |
568 | print_track("Freed", get_track(s, object, TRACK_FREE)); | |
569 | } | |
570 | ||
571 | static void print_page_info(struct page *page) | |
572 | { | |
f9f58285 | 573 | pr_err("INFO: Slab 0x%p objects=%u used=%u fp=0x%p flags=0x%04lx\n", |
d0e0ac97 | 574 | page, page->objects, page->inuse, page->freelist, page->flags); |
24922684 CL |
575 | |
576 | } | |
577 | ||
578 | static void slab_bug(struct kmem_cache *s, char *fmt, ...) | |
579 | { | |
ecc42fbe | 580 | struct va_format vaf; |
24922684 | 581 | va_list args; |
24922684 CL |
582 | |
583 | va_start(args, fmt); | |
ecc42fbe FF |
584 | vaf.fmt = fmt; |
585 | vaf.va = &args; | |
f9f58285 | 586 | pr_err("=============================================================================\n"); |
ecc42fbe | 587 | pr_err("BUG %s (%s): %pV\n", s->name, print_tainted(), &vaf); |
f9f58285 | 588 | pr_err("-----------------------------------------------------------------------------\n\n"); |
645df230 | 589 | |
373d4d09 | 590 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); |
ecc42fbe | 591 | va_end(args); |
81819f0f CL |
592 | } |
593 | ||
24922684 CL |
594 | static void slab_fix(struct kmem_cache *s, char *fmt, ...) |
595 | { | |
ecc42fbe | 596 | struct va_format vaf; |
24922684 | 597 | va_list args; |
24922684 CL |
598 | |
599 | va_start(args, fmt); | |
ecc42fbe FF |
600 | vaf.fmt = fmt; |
601 | vaf.va = &args; | |
602 | pr_err("FIX %s: %pV\n", s->name, &vaf); | |
24922684 | 603 | va_end(args); |
24922684 CL |
604 | } |
605 | ||
606 | static void print_trailer(struct kmem_cache *s, struct page *page, u8 *p) | |
81819f0f CL |
607 | { |
608 | unsigned int off; /* Offset of last byte */ | |
a973e9dd | 609 | u8 *addr = page_address(page); |
24922684 CL |
610 | |
611 | print_tracking(s, p); | |
612 | ||
613 | print_page_info(page); | |
614 | ||
f9f58285 FF |
615 | pr_err("INFO: Object 0x%p @offset=%tu fp=0x%p\n\n", |
616 | p, p - addr, get_freepointer(s, p)); | |
24922684 CL |
617 | |
618 | if (p > addr + 16) | |
ffc79d28 | 619 | print_section("Bytes b4 ", p - 16, 16); |
81819f0f | 620 | |
3b0efdfa | 621 | print_section("Object ", p, min_t(unsigned long, s->object_size, |
ffc79d28 | 622 | PAGE_SIZE)); |
81819f0f | 623 | if (s->flags & SLAB_RED_ZONE) |
3b0efdfa CL |
624 | print_section("Redzone ", p + s->object_size, |
625 | s->inuse - s->object_size); | |
81819f0f | 626 | |
81819f0f CL |
627 | if (s->offset) |
628 | off = s->offset + sizeof(void *); | |
629 | else | |
630 | off = s->inuse; | |
631 | ||
24922684 | 632 | if (s->flags & SLAB_STORE_USER) |
81819f0f | 633 | off += 2 * sizeof(struct track); |
81819f0f CL |
634 | |
635 | if (off != s->size) | |
636 | /* Beginning of the filler is the free pointer */ | |
ffc79d28 | 637 | print_section("Padding ", p + off, s->size - off); |
24922684 CL |
638 | |
639 | dump_stack(); | |
81819f0f CL |
640 | } |
641 | ||
642 | static void object_err(struct kmem_cache *s, struct page *page, | |
643 | u8 *object, char *reason) | |
644 | { | |
3dc50637 | 645 | slab_bug(s, "%s", reason); |
24922684 | 646 | print_trailer(s, page, object); |
81819f0f CL |
647 | } |
648 | ||
d0e0ac97 CG |
649 | static void slab_err(struct kmem_cache *s, struct page *page, |
650 | const char *fmt, ...) | |
81819f0f CL |
651 | { |
652 | va_list args; | |
653 | char buf[100]; | |
654 | ||
24922684 CL |
655 | va_start(args, fmt); |
656 | vsnprintf(buf, sizeof(buf), fmt, args); | |
81819f0f | 657 | va_end(args); |
3dc50637 | 658 | slab_bug(s, "%s", buf); |
24922684 | 659 | print_page_info(page); |
81819f0f CL |
660 | dump_stack(); |
661 | } | |
662 | ||
f7cb1933 | 663 | static void init_object(struct kmem_cache *s, void *object, u8 val) |
81819f0f CL |
664 | { |
665 | u8 *p = object; | |
666 | ||
667 | if (s->flags & __OBJECT_POISON) { | |
3b0efdfa CL |
668 | memset(p, POISON_FREE, s->object_size - 1); |
669 | p[s->object_size - 1] = POISON_END; | |
81819f0f CL |
670 | } |
671 | ||
672 | if (s->flags & SLAB_RED_ZONE) | |
3b0efdfa | 673 | memset(p + s->object_size, val, s->inuse - s->object_size); |
81819f0f CL |
674 | } |
675 | ||
24922684 CL |
676 | static void restore_bytes(struct kmem_cache *s, char *message, u8 data, |
677 | void *from, void *to) | |
678 | { | |
679 | slab_fix(s, "Restoring 0x%p-0x%p=0x%x\n", from, to - 1, data); | |
680 | memset(from, data, to - from); | |
681 | } | |
682 | ||
683 | static int check_bytes_and_report(struct kmem_cache *s, struct page *page, | |
684 | u8 *object, char *what, | |
06428780 | 685 | u8 *start, unsigned int value, unsigned int bytes) |
24922684 CL |
686 | { |
687 | u8 *fault; | |
688 | u8 *end; | |
689 | ||
79824820 | 690 | fault = memchr_inv(start, value, bytes); |
24922684 CL |
691 | if (!fault) |
692 | return 1; | |
693 | ||
694 | end = start + bytes; | |
695 | while (end > fault && end[-1] == value) | |
696 | end--; | |
697 | ||
698 | slab_bug(s, "%s overwritten", what); | |
f9f58285 | 699 | pr_err("INFO: 0x%p-0x%p. First byte 0x%x instead of 0x%x\n", |
24922684 CL |
700 | fault, end - 1, fault[0], value); |
701 | print_trailer(s, page, object); | |
702 | ||
703 | restore_bytes(s, what, value, fault, end); | |
704 | return 0; | |
81819f0f CL |
705 | } |
706 | ||
81819f0f CL |
707 | /* |
708 | * Object layout: | |
709 | * | |
710 | * object address | |
711 | * Bytes of the object to be managed. | |
712 | * If the freepointer may overlay the object then the free | |
713 | * pointer is the first word of the object. | |
672bba3a | 714 | * |
81819f0f CL |
715 | * Poisoning uses 0x6b (POISON_FREE) and the last byte is |
716 | * 0xa5 (POISON_END) | |
717 | * | |
3b0efdfa | 718 | * object + s->object_size |
81819f0f | 719 | * Padding to reach word boundary. This is also used for Redzoning. |
672bba3a | 720 | * Padding is extended by another word if Redzoning is enabled and |
3b0efdfa | 721 | * object_size == inuse. |
672bba3a | 722 | * |
81819f0f CL |
723 | * We fill with 0xbb (RED_INACTIVE) for inactive objects and with |
724 | * 0xcc (RED_ACTIVE) for objects in use. | |
725 | * | |
726 | * object + s->inuse | |
672bba3a CL |
727 | * Meta data starts here. |
728 | * | |
81819f0f CL |
729 | * A. Free pointer (if we cannot overwrite object on free) |
730 | * B. Tracking data for SLAB_STORE_USER | |
672bba3a | 731 | * C. Padding to reach required alignment boundary or at mininum |
6446faa2 | 732 | * one word if debugging is on to be able to detect writes |
672bba3a CL |
733 | * before the word boundary. |
734 | * | |
735 | * Padding is done using 0x5a (POISON_INUSE) | |
81819f0f CL |
736 | * |
737 | * object + s->size | |
672bba3a | 738 | * Nothing is used beyond s->size. |
81819f0f | 739 | * |
3b0efdfa | 740 | * If slabcaches are merged then the object_size and inuse boundaries are mostly |
672bba3a | 741 | * ignored. And therefore no slab options that rely on these boundaries |
81819f0f CL |
742 | * may be used with merged slabcaches. |
743 | */ | |
744 | ||
81819f0f CL |
745 | static int check_pad_bytes(struct kmem_cache *s, struct page *page, u8 *p) |
746 | { | |
747 | unsigned long off = s->inuse; /* The end of info */ | |
748 | ||
749 | if (s->offset) | |
750 | /* Freepointer is placed after the object. */ | |
751 | off += sizeof(void *); | |
752 | ||
753 | if (s->flags & SLAB_STORE_USER) | |
754 | /* We also have user information there */ | |
755 | off += 2 * sizeof(struct track); | |
756 | ||
757 | if (s->size == off) | |
758 | return 1; | |
759 | ||
24922684 CL |
760 | return check_bytes_and_report(s, page, p, "Object padding", |
761 | p + off, POISON_INUSE, s->size - off); | |
81819f0f CL |
762 | } |
763 | ||
39b26464 | 764 | /* Check the pad bytes at the end of a slab page */ |
81819f0f CL |
765 | static int slab_pad_check(struct kmem_cache *s, struct page *page) |
766 | { | |
24922684 CL |
767 | u8 *start; |
768 | u8 *fault; | |
769 | u8 *end; | |
770 | int length; | |
771 | int remainder; | |
81819f0f CL |
772 | |
773 | if (!(s->flags & SLAB_POISON)) | |
774 | return 1; | |
775 | ||
a973e9dd | 776 | start = page_address(page); |
ab9a0f19 | 777 | length = (PAGE_SIZE << compound_order(page)) - s->reserved; |
39b26464 CL |
778 | end = start + length; |
779 | remainder = length % s->size; | |
81819f0f CL |
780 | if (!remainder) |
781 | return 1; | |
782 | ||
79824820 | 783 | fault = memchr_inv(end - remainder, POISON_INUSE, remainder); |
24922684 CL |
784 | if (!fault) |
785 | return 1; | |
786 | while (end > fault && end[-1] == POISON_INUSE) | |
787 | end--; | |
788 | ||
789 | slab_err(s, page, "Padding overwritten. 0x%p-0x%p", fault, end - 1); | |
ffc79d28 | 790 | print_section("Padding ", end - remainder, remainder); |
24922684 | 791 | |
8a3d271d | 792 | restore_bytes(s, "slab padding", POISON_INUSE, end - remainder, end); |
24922684 | 793 | return 0; |
81819f0f CL |
794 | } |
795 | ||
796 | static int check_object(struct kmem_cache *s, struct page *page, | |
f7cb1933 | 797 | void *object, u8 val) |
81819f0f CL |
798 | { |
799 | u8 *p = object; | |
3b0efdfa | 800 | u8 *endobject = object + s->object_size; |
81819f0f CL |
801 | |
802 | if (s->flags & SLAB_RED_ZONE) { | |
24922684 | 803 | if (!check_bytes_and_report(s, page, object, "Redzone", |
3b0efdfa | 804 | endobject, val, s->inuse - s->object_size)) |
81819f0f | 805 | return 0; |
81819f0f | 806 | } else { |
3b0efdfa | 807 | if ((s->flags & SLAB_POISON) && s->object_size < s->inuse) { |
3adbefee | 808 | check_bytes_and_report(s, page, p, "Alignment padding", |
d0e0ac97 CG |
809 | endobject, POISON_INUSE, |
810 | s->inuse - s->object_size); | |
3adbefee | 811 | } |
81819f0f CL |
812 | } |
813 | ||
814 | if (s->flags & SLAB_POISON) { | |
f7cb1933 | 815 | if (val != SLUB_RED_ACTIVE && (s->flags & __OBJECT_POISON) && |
24922684 | 816 | (!check_bytes_and_report(s, page, p, "Poison", p, |
3b0efdfa | 817 | POISON_FREE, s->object_size - 1) || |
24922684 | 818 | !check_bytes_and_report(s, page, p, "Poison", |
3b0efdfa | 819 | p + s->object_size - 1, POISON_END, 1))) |
81819f0f | 820 | return 0; |
81819f0f CL |
821 | /* |
822 | * check_pad_bytes cleans up on its own. | |
823 | */ | |
824 | check_pad_bytes(s, page, p); | |
825 | } | |
826 | ||
f7cb1933 | 827 | if (!s->offset && val == SLUB_RED_ACTIVE) |
81819f0f CL |
828 | /* |
829 | * Object and freepointer overlap. Cannot check | |
830 | * freepointer while object is allocated. | |
831 | */ | |
832 | return 1; | |
833 | ||
834 | /* Check free pointer validity */ | |
835 | if (!check_valid_pointer(s, page, get_freepointer(s, p))) { | |
836 | object_err(s, page, p, "Freepointer corrupt"); | |
837 | /* | |
9f6c708e | 838 | * No choice but to zap it and thus lose the remainder |
81819f0f | 839 | * of the free objects in this slab. May cause |
672bba3a | 840 | * another error because the object count is now wrong. |
81819f0f | 841 | */ |
a973e9dd | 842 | set_freepointer(s, p, NULL); |
81819f0f CL |
843 | return 0; |
844 | } | |
845 | return 1; | |
846 | } | |
847 | ||
848 | static int check_slab(struct kmem_cache *s, struct page *page) | |
849 | { | |
39b26464 CL |
850 | int maxobj; |
851 | ||
81819f0f CL |
852 | VM_BUG_ON(!irqs_disabled()); |
853 | ||
854 | if (!PageSlab(page)) { | |
24922684 | 855 | slab_err(s, page, "Not a valid slab page"); |
81819f0f CL |
856 | return 0; |
857 | } | |
39b26464 | 858 | |
ab9a0f19 | 859 | maxobj = order_objects(compound_order(page), s->size, s->reserved); |
39b26464 CL |
860 | if (page->objects > maxobj) { |
861 | slab_err(s, page, "objects %u > max %u", | |
862 | s->name, page->objects, maxobj); | |
863 | return 0; | |
864 | } | |
865 | if (page->inuse > page->objects) { | |
24922684 | 866 | slab_err(s, page, "inuse %u > max %u", |
39b26464 | 867 | s->name, page->inuse, page->objects); |
81819f0f CL |
868 | return 0; |
869 | } | |
870 | /* Slab_pad_check fixes things up after itself */ | |
871 | slab_pad_check(s, page); | |
872 | return 1; | |
873 | } | |
874 | ||
875 | /* | |
672bba3a CL |
876 | * Determine if a certain object on a page is on the freelist. Must hold the |
877 | * slab lock to guarantee that the chains are in a consistent state. | |
81819f0f CL |
878 | */ |
879 | static int on_freelist(struct kmem_cache *s, struct page *page, void *search) | |
880 | { | |
881 | int nr = 0; | |
881db7fb | 882 | void *fp; |
81819f0f | 883 | void *object = NULL; |
224a88be | 884 | unsigned long max_objects; |
81819f0f | 885 | |
881db7fb | 886 | fp = page->freelist; |
39b26464 | 887 | while (fp && nr <= page->objects) { |
81819f0f CL |
888 | if (fp == search) |
889 | return 1; | |
890 | if (!check_valid_pointer(s, page, fp)) { | |
891 | if (object) { | |
892 | object_err(s, page, object, | |
893 | "Freechain corrupt"); | |
a973e9dd | 894 | set_freepointer(s, object, NULL); |
81819f0f | 895 | } else { |
24922684 | 896 | slab_err(s, page, "Freepointer corrupt"); |
a973e9dd | 897 | page->freelist = NULL; |
39b26464 | 898 | page->inuse = page->objects; |
24922684 | 899 | slab_fix(s, "Freelist cleared"); |
81819f0f CL |
900 | return 0; |
901 | } | |
902 | break; | |
903 | } | |
904 | object = fp; | |
905 | fp = get_freepointer(s, object); | |
906 | nr++; | |
907 | } | |
908 | ||
ab9a0f19 | 909 | max_objects = order_objects(compound_order(page), s->size, s->reserved); |
210b5c06 CG |
910 | if (max_objects > MAX_OBJS_PER_PAGE) |
911 | max_objects = MAX_OBJS_PER_PAGE; | |
224a88be CL |
912 | |
913 | if (page->objects != max_objects) { | |
914 | slab_err(s, page, "Wrong number of objects. Found %d but " | |
915 | "should be %d", page->objects, max_objects); | |
916 | page->objects = max_objects; | |
917 | slab_fix(s, "Number of objects adjusted."); | |
918 | } | |
39b26464 | 919 | if (page->inuse != page->objects - nr) { |
70d71228 | 920 | slab_err(s, page, "Wrong object count. Counter is %d but " |
39b26464 CL |
921 | "counted were %d", page->inuse, page->objects - nr); |
922 | page->inuse = page->objects - nr; | |
24922684 | 923 | slab_fix(s, "Object count adjusted."); |
81819f0f CL |
924 | } |
925 | return search == NULL; | |
926 | } | |
927 | ||
0121c619 CL |
928 | static void trace(struct kmem_cache *s, struct page *page, void *object, |
929 | int alloc) | |
3ec09742 CL |
930 | { |
931 | if (s->flags & SLAB_TRACE) { | |
f9f58285 | 932 | pr_info("TRACE %s %s 0x%p inuse=%d fp=0x%p\n", |
3ec09742 CL |
933 | s->name, |
934 | alloc ? "alloc" : "free", | |
935 | object, page->inuse, | |
936 | page->freelist); | |
937 | ||
938 | if (!alloc) | |
d0e0ac97 CG |
939 | print_section("Object ", (void *)object, |
940 | s->object_size); | |
3ec09742 CL |
941 | |
942 | dump_stack(); | |
943 | } | |
944 | } | |
945 | ||
643b1138 | 946 | /* |
672bba3a | 947 | * Tracking of fully allocated slabs for debugging purposes. |
643b1138 | 948 | */ |
5cc6eee8 CL |
949 | static void add_full(struct kmem_cache *s, |
950 | struct kmem_cache_node *n, struct page *page) | |
643b1138 | 951 | { |
5cc6eee8 CL |
952 | if (!(s->flags & SLAB_STORE_USER)) |
953 | return; | |
954 | ||
255d0884 | 955 | lockdep_assert_held(&n->list_lock); |
643b1138 | 956 | list_add(&page->lru, &n->full); |
643b1138 CL |
957 | } |
958 | ||
c65c1877 | 959 | static void remove_full(struct kmem_cache *s, struct kmem_cache_node *n, struct page *page) |
643b1138 | 960 | { |
643b1138 CL |
961 | if (!(s->flags & SLAB_STORE_USER)) |
962 | return; | |
963 | ||
255d0884 | 964 | lockdep_assert_held(&n->list_lock); |
643b1138 | 965 | list_del(&page->lru); |
643b1138 CL |
966 | } |
967 | ||
0f389ec6 CL |
968 | /* Tracking of the number of slabs for debugging purposes */ |
969 | static inline unsigned long slabs_node(struct kmem_cache *s, int node) | |
970 | { | |
971 | struct kmem_cache_node *n = get_node(s, node); | |
972 | ||
973 | return atomic_long_read(&n->nr_slabs); | |
974 | } | |
975 | ||
26c02cf0 AB |
976 | static inline unsigned long node_nr_slabs(struct kmem_cache_node *n) |
977 | { | |
978 | return atomic_long_read(&n->nr_slabs); | |
979 | } | |
980 | ||
205ab99d | 981 | static inline void inc_slabs_node(struct kmem_cache *s, int node, int objects) |
0f389ec6 CL |
982 | { |
983 | struct kmem_cache_node *n = get_node(s, node); | |
984 | ||
985 | /* | |
986 | * May be called early in order to allocate a slab for the | |
987 | * kmem_cache_node structure. Solve the chicken-egg | |
988 | * dilemma by deferring the increment of the count during | |
989 | * bootstrap (see early_kmem_cache_node_alloc). | |
990 | */ | |
338b2642 | 991 | if (likely(n)) { |
0f389ec6 | 992 | atomic_long_inc(&n->nr_slabs); |
205ab99d CL |
993 | atomic_long_add(objects, &n->total_objects); |
994 | } | |
0f389ec6 | 995 | } |
205ab99d | 996 | static inline void dec_slabs_node(struct kmem_cache *s, int node, int objects) |
0f389ec6 CL |
997 | { |
998 | struct kmem_cache_node *n = get_node(s, node); | |
999 | ||
1000 | atomic_long_dec(&n->nr_slabs); | |
205ab99d | 1001 | atomic_long_sub(objects, &n->total_objects); |
0f389ec6 CL |
1002 | } |
1003 | ||
1004 | /* Object debug checks for alloc/free paths */ | |
3ec09742 CL |
1005 | static void setup_object_debug(struct kmem_cache *s, struct page *page, |
1006 | void *object) | |
1007 | { | |
1008 | if (!(s->flags & (SLAB_STORE_USER|SLAB_RED_ZONE|__OBJECT_POISON))) | |
1009 | return; | |
1010 | ||
f7cb1933 | 1011 | init_object(s, object, SLUB_RED_INACTIVE); |
3ec09742 CL |
1012 | init_tracking(s, object); |
1013 | } | |
1014 | ||
d0e0ac97 CG |
1015 | static noinline int alloc_debug_processing(struct kmem_cache *s, |
1016 | struct page *page, | |
ce71e27c | 1017 | void *object, unsigned long addr) |
81819f0f CL |
1018 | { |
1019 | if (!check_slab(s, page)) | |
1020 | goto bad; | |
1021 | ||
81819f0f CL |
1022 | if (!check_valid_pointer(s, page, object)) { |
1023 | object_err(s, page, object, "Freelist Pointer check fails"); | |
70d71228 | 1024 | goto bad; |
81819f0f CL |
1025 | } |
1026 | ||
f7cb1933 | 1027 | if (!check_object(s, page, object, SLUB_RED_INACTIVE)) |
81819f0f | 1028 | goto bad; |
81819f0f | 1029 | |
3ec09742 CL |
1030 | /* Success perform special debug activities for allocs */ |
1031 | if (s->flags & SLAB_STORE_USER) | |
1032 | set_track(s, object, TRACK_ALLOC, addr); | |
1033 | trace(s, page, object, 1); | |
f7cb1933 | 1034 | init_object(s, object, SLUB_RED_ACTIVE); |
81819f0f | 1035 | return 1; |
3ec09742 | 1036 | |
81819f0f CL |
1037 | bad: |
1038 | if (PageSlab(page)) { | |
1039 | /* | |
1040 | * If this is a slab page then lets do the best we can | |
1041 | * to avoid issues in the future. Marking all objects | |
672bba3a | 1042 | * as used avoids touching the remaining objects. |
81819f0f | 1043 | */ |
24922684 | 1044 | slab_fix(s, "Marking all objects used"); |
39b26464 | 1045 | page->inuse = page->objects; |
a973e9dd | 1046 | page->freelist = NULL; |
81819f0f CL |
1047 | } |
1048 | return 0; | |
1049 | } | |
1050 | ||
19c7ff9e CL |
1051 | static noinline struct kmem_cache_node *free_debug_processing( |
1052 | struct kmem_cache *s, struct page *page, void *object, | |
1053 | unsigned long addr, unsigned long *flags) | |
81819f0f | 1054 | { |
19c7ff9e | 1055 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); |
5c2e4bbb | 1056 | |
19c7ff9e | 1057 | spin_lock_irqsave(&n->list_lock, *flags); |
881db7fb CL |
1058 | slab_lock(page); |
1059 | ||
81819f0f CL |
1060 | if (!check_slab(s, page)) |
1061 | goto fail; | |
1062 | ||
1063 | if (!check_valid_pointer(s, page, object)) { | |
70d71228 | 1064 | slab_err(s, page, "Invalid object pointer 0x%p", object); |
81819f0f CL |
1065 | goto fail; |
1066 | } | |
1067 | ||
1068 | if (on_freelist(s, page, object)) { | |
24922684 | 1069 | object_err(s, page, object, "Object already free"); |
81819f0f CL |
1070 | goto fail; |
1071 | } | |
1072 | ||
f7cb1933 | 1073 | if (!check_object(s, page, object, SLUB_RED_ACTIVE)) |
5c2e4bbb | 1074 | goto out; |
81819f0f | 1075 | |
1b4f59e3 | 1076 | if (unlikely(s != page->slab_cache)) { |
3adbefee | 1077 | if (!PageSlab(page)) { |
70d71228 CL |
1078 | slab_err(s, page, "Attempt to free object(0x%p) " |
1079 | "outside of slab", object); | |
1b4f59e3 | 1080 | } else if (!page->slab_cache) { |
f9f58285 FF |
1081 | pr_err("SLUB <none>: no slab for object 0x%p.\n", |
1082 | object); | |
70d71228 | 1083 | dump_stack(); |
06428780 | 1084 | } else |
24922684 CL |
1085 | object_err(s, page, object, |
1086 | "page slab pointer corrupt."); | |
81819f0f CL |
1087 | goto fail; |
1088 | } | |
3ec09742 | 1089 | |
3ec09742 CL |
1090 | if (s->flags & SLAB_STORE_USER) |
1091 | set_track(s, object, TRACK_FREE, addr); | |
1092 | trace(s, page, object, 0); | |
f7cb1933 | 1093 | init_object(s, object, SLUB_RED_INACTIVE); |
5c2e4bbb | 1094 | out: |
881db7fb | 1095 | slab_unlock(page); |
19c7ff9e CL |
1096 | /* |
1097 | * Keep node_lock to preserve integrity | |
1098 | * until the object is actually freed | |
1099 | */ | |
1100 | return n; | |
3ec09742 | 1101 | |
81819f0f | 1102 | fail: |
19c7ff9e CL |
1103 | slab_unlock(page); |
1104 | spin_unlock_irqrestore(&n->list_lock, *flags); | |
24922684 | 1105 | slab_fix(s, "Object at 0x%p not freed", object); |
19c7ff9e | 1106 | return NULL; |
81819f0f CL |
1107 | } |
1108 | ||
41ecc55b CL |
1109 | static int __init setup_slub_debug(char *str) |
1110 | { | |
f0630fff CL |
1111 | slub_debug = DEBUG_DEFAULT_FLAGS; |
1112 | if (*str++ != '=' || !*str) | |
1113 | /* | |
1114 | * No options specified. Switch on full debugging. | |
1115 | */ | |
1116 | goto out; | |
1117 | ||
1118 | if (*str == ',') | |
1119 | /* | |
1120 | * No options but restriction on slabs. This means full | |
1121 | * debugging for slabs matching a pattern. | |
1122 | */ | |
1123 | goto check_slabs; | |
1124 | ||
fa5ec8a1 DR |
1125 | if (tolower(*str) == 'o') { |
1126 | /* | |
1127 | * Avoid enabling debugging on caches if its minimum order | |
1128 | * would increase as a result. | |
1129 | */ | |
1130 | disable_higher_order_debug = 1; | |
1131 | goto out; | |
1132 | } | |
1133 | ||
f0630fff CL |
1134 | slub_debug = 0; |
1135 | if (*str == '-') | |
1136 | /* | |
1137 | * Switch off all debugging measures. | |
1138 | */ | |
1139 | goto out; | |
1140 | ||
1141 | /* | |
1142 | * Determine which debug features should be switched on | |
1143 | */ | |
06428780 | 1144 | for (; *str && *str != ','; str++) { |
f0630fff CL |
1145 | switch (tolower(*str)) { |
1146 | case 'f': | |
1147 | slub_debug |= SLAB_DEBUG_FREE; | |
1148 | break; | |
1149 | case 'z': | |
1150 | slub_debug |= SLAB_RED_ZONE; | |
1151 | break; | |
1152 | case 'p': | |
1153 | slub_debug |= SLAB_POISON; | |
1154 | break; | |
1155 | case 'u': | |
1156 | slub_debug |= SLAB_STORE_USER; | |
1157 | break; | |
1158 | case 't': | |
1159 | slub_debug |= SLAB_TRACE; | |
1160 | break; | |
4c13dd3b DM |
1161 | case 'a': |
1162 | slub_debug |= SLAB_FAILSLAB; | |
1163 | break; | |
f0630fff | 1164 | default: |
f9f58285 FF |
1165 | pr_err("slub_debug option '%c' unknown. skipped\n", |
1166 | *str); | |
f0630fff | 1167 | } |
41ecc55b CL |
1168 | } |
1169 | ||
f0630fff | 1170 | check_slabs: |
41ecc55b CL |
1171 | if (*str == ',') |
1172 | slub_debug_slabs = str + 1; | |
f0630fff | 1173 | out: |
41ecc55b CL |
1174 | return 1; |
1175 | } | |
1176 | ||
1177 | __setup("slub_debug", setup_slub_debug); | |
1178 | ||
3b0efdfa | 1179 | static unsigned long kmem_cache_flags(unsigned long object_size, |
ba0268a8 | 1180 | unsigned long flags, const char *name, |
51cc5068 | 1181 | void (*ctor)(void *)) |
41ecc55b CL |
1182 | { |
1183 | /* | |
e153362a | 1184 | * Enable debugging if selected on the kernel commandline. |
41ecc55b | 1185 | */ |
c6f58d9b CL |
1186 | if (slub_debug && (!slub_debug_slabs || (name && |
1187 | !strncmp(slub_debug_slabs, name, strlen(slub_debug_slabs))))) | |
3de47213 | 1188 | flags |= slub_debug; |
ba0268a8 CL |
1189 | |
1190 | return flags; | |
41ecc55b CL |
1191 | } |
1192 | #else | |
3ec09742 CL |
1193 | static inline void setup_object_debug(struct kmem_cache *s, |
1194 | struct page *page, void *object) {} | |
41ecc55b | 1195 | |
3ec09742 | 1196 | static inline int alloc_debug_processing(struct kmem_cache *s, |
ce71e27c | 1197 | struct page *page, void *object, unsigned long addr) { return 0; } |
41ecc55b | 1198 | |
19c7ff9e CL |
1199 | static inline struct kmem_cache_node *free_debug_processing( |
1200 | struct kmem_cache *s, struct page *page, void *object, | |
1201 | unsigned long addr, unsigned long *flags) { return NULL; } | |
41ecc55b | 1202 | |
41ecc55b CL |
1203 | static inline int slab_pad_check(struct kmem_cache *s, struct page *page) |
1204 | { return 1; } | |
1205 | static inline int check_object(struct kmem_cache *s, struct page *page, | |
f7cb1933 | 1206 | void *object, u8 val) { return 1; } |
5cc6eee8 CL |
1207 | static inline void add_full(struct kmem_cache *s, struct kmem_cache_node *n, |
1208 | struct page *page) {} | |
c65c1877 PZ |
1209 | static inline void remove_full(struct kmem_cache *s, struct kmem_cache_node *n, |
1210 | struct page *page) {} | |
3b0efdfa | 1211 | static inline unsigned long kmem_cache_flags(unsigned long object_size, |
ba0268a8 | 1212 | unsigned long flags, const char *name, |
51cc5068 | 1213 | void (*ctor)(void *)) |
ba0268a8 CL |
1214 | { |
1215 | return flags; | |
1216 | } | |
41ecc55b | 1217 | #define slub_debug 0 |
0f389ec6 | 1218 | |
fdaa45e9 IM |
1219 | #define disable_higher_order_debug 0 |
1220 | ||
0f389ec6 CL |
1221 | static inline unsigned long slabs_node(struct kmem_cache *s, int node) |
1222 | { return 0; } | |
26c02cf0 AB |
1223 | static inline unsigned long node_nr_slabs(struct kmem_cache_node *n) |
1224 | { return 0; } | |
205ab99d CL |
1225 | static inline void inc_slabs_node(struct kmem_cache *s, int node, |
1226 | int objects) {} | |
1227 | static inline void dec_slabs_node(struct kmem_cache *s, int node, | |
1228 | int objects) {} | |
7d550c56 | 1229 | |
02e72cc6 AR |
1230 | #endif /* CONFIG_SLUB_DEBUG */ |
1231 | ||
1232 | /* | |
1233 | * Hooks for other subsystems that check memory allocations. In a typical | |
1234 | * production configuration these hooks all should produce no code at all. | |
1235 | */ | |
d56791b3 RB |
1236 | static inline void kmalloc_large_node_hook(void *ptr, size_t size, gfp_t flags) |
1237 | { | |
1238 | kmemleak_alloc(ptr, size, 1, flags); | |
1239 | } | |
1240 | ||
1241 | static inline void kfree_hook(const void *x) | |
1242 | { | |
1243 | kmemleak_free(x); | |
1244 | } | |
1245 | ||
7d550c56 | 1246 | static inline int slab_pre_alloc_hook(struct kmem_cache *s, gfp_t flags) |
02e72cc6 AR |
1247 | { |
1248 | flags &= gfp_allowed_mask; | |
1249 | lockdep_trace_alloc(flags); | |
1250 | might_sleep_if(flags & __GFP_WAIT); | |
7d550c56 | 1251 | |
02e72cc6 AR |
1252 | return should_failslab(s->object_size, flags, s->flags); |
1253 | } | |
1254 | ||
1255 | static inline void slab_post_alloc_hook(struct kmem_cache *s, | |
1256 | gfp_t flags, void *object) | |
d56791b3 | 1257 | { |
02e72cc6 AR |
1258 | flags &= gfp_allowed_mask; |
1259 | kmemcheck_slab_alloc(s, flags, object, slab_ksize(s)); | |
1260 | kmemleak_alloc_recursive(object, s->object_size, 1, s->flags, flags); | |
d56791b3 | 1261 | } |
7d550c56 | 1262 | |
d56791b3 RB |
1263 | static inline void slab_free_hook(struct kmem_cache *s, void *x) |
1264 | { | |
1265 | kmemleak_free_recursive(x, s->flags); | |
7d550c56 | 1266 | |
02e72cc6 AR |
1267 | /* |
1268 | * Trouble is that we may no longer disable interrupts in the fast path | |
1269 | * So in order to make the debug calls that expect irqs to be | |
1270 | * disabled we need to disable interrupts temporarily. | |
1271 | */ | |
1272 | #if defined(CONFIG_KMEMCHECK) || defined(CONFIG_LOCKDEP) | |
1273 | { | |
1274 | unsigned long flags; | |
1275 | ||
1276 | local_irq_save(flags); | |
1277 | kmemcheck_slab_free(s, x, s->object_size); | |
1278 | debug_check_no_locks_freed(x, s->object_size); | |
1279 | local_irq_restore(flags); | |
1280 | } | |
1281 | #endif | |
1282 | if (!(s->flags & SLAB_DEBUG_OBJECTS)) | |
1283 | debug_check_no_obj_freed(x, s->object_size); | |
1284 | } | |
205ab99d | 1285 | |
81819f0f CL |
1286 | /* |
1287 | * Slab allocation and freeing | |
1288 | */ | |
5dfb4175 VD |
1289 | static inline struct page *alloc_slab_page(struct kmem_cache *s, |
1290 | gfp_t flags, int node, struct kmem_cache_order_objects oo) | |
65c3376a | 1291 | { |
5dfb4175 | 1292 | struct page *page; |
65c3376a CL |
1293 | int order = oo_order(oo); |
1294 | ||
b1eeab67 VN |
1295 | flags |= __GFP_NOTRACK; |
1296 | ||
5dfb4175 VD |
1297 | if (memcg_charge_slab(s, flags, order)) |
1298 | return NULL; | |
1299 | ||
2154a336 | 1300 | if (node == NUMA_NO_NODE) |
5dfb4175 | 1301 | page = alloc_pages(flags, order); |
65c3376a | 1302 | else |
5dfb4175 VD |
1303 | page = alloc_pages_exact_node(node, flags, order); |
1304 | ||
1305 | if (!page) | |
1306 | memcg_uncharge_slab(s, order); | |
1307 | ||
1308 | return page; | |
65c3376a CL |
1309 | } |
1310 | ||
81819f0f CL |
1311 | static struct page *allocate_slab(struct kmem_cache *s, gfp_t flags, int node) |
1312 | { | |
06428780 | 1313 | struct page *page; |
834f3d11 | 1314 | struct kmem_cache_order_objects oo = s->oo; |
ba52270d | 1315 | gfp_t alloc_gfp; |
81819f0f | 1316 | |
7e0528da CL |
1317 | flags &= gfp_allowed_mask; |
1318 | ||
1319 | if (flags & __GFP_WAIT) | |
1320 | local_irq_enable(); | |
1321 | ||
b7a49f0d | 1322 | flags |= s->allocflags; |
e12ba74d | 1323 | |
ba52270d PE |
1324 | /* |
1325 | * Let the initial higher-order allocation fail under memory pressure | |
1326 | * so we fall-back to the minimum order allocation. | |
1327 | */ | |
1328 | alloc_gfp = (flags | __GFP_NOWARN | __GFP_NORETRY) & ~__GFP_NOFAIL; | |
1329 | ||
5dfb4175 | 1330 | page = alloc_slab_page(s, alloc_gfp, node, oo); |
65c3376a CL |
1331 | if (unlikely(!page)) { |
1332 | oo = s->min; | |
80c3a998 | 1333 | alloc_gfp = flags; |
65c3376a CL |
1334 | /* |
1335 | * Allocation may have failed due to fragmentation. | |
1336 | * Try a lower order alloc if possible | |
1337 | */ | |
5dfb4175 | 1338 | page = alloc_slab_page(s, alloc_gfp, node, oo); |
81819f0f | 1339 | |
7e0528da CL |
1340 | if (page) |
1341 | stat(s, ORDER_FALLBACK); | |
65c3376a | 1342 | } |
5a896d9e | 1343 | |
737b719e | 1344 | if (kmemcheck_enabled && page |
5086c389 | 1345 | && !(s->flags & (SLAB_NOTRACK | DEBUG_DEFAULT_FLAGS))) { |
b1eeab67 VN |
1346 | int pages = 1 << oo_order(oo); |
1347 | ||
80c3a998 | 1348 | kmemcheck_alloc_shadow(page, oo_order(oo), alloc_gfp, node); |
b1eeab67 VN |
1349 | |
1350 | /* | |
1351 | * Objects from caches that have a constructor don't get | |
1352 | * cleared when they're allocated, so we need to do it here. | |
1353 | */ | |
1354 | if (s->ctor) | |
1355 | kmemcheck_mark_uninitialized_pages(page, pages); | |
1356 | else | |
1357 | kmemcheck_mark_unallocated_pages(page, pages); | |
5a896d9e VN |
1358 | } |
1359 | ||
737b719e DR |
1360 | if (flags & __GFP_WAIT) |
1361 | local_irq_disable(); | |
1362 | if (!page) | |
1363 | return NULL; | |
1364 | ||
834f3d11 | 1365 | page->objects = oo_objects(oo); |
81819f0f CL |
1366 | mod_zone_page_state(page_zone(page), |
1367 | (s->flags & SLAB_RECLAIM_ACCOUNT) ? | |
1368 | NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE, | |
65c3376a | 1369 | 1 << oo_order(oo)); |
81819f0f CL |
1370 | |
1371 | return page; | |
1372 | } | |
1373 | ||
1374 | static void setup_object(struct kmem_cache *s, struct page *page, | |
1375 | void *object) | |
1376 | { | |
3ec09742 | 1377 | setup_object_debug(s, page, object); |
4f104934 | 1378 | if (unlikely(s->ctor)) |
51cc5068 | 1379 | s->ctor(object); |
81819f0f CL |
1380 | } |
1381 | ||
1382 | static struct page *new_slab(struct kmem_cache *s, gfp_t flags, int node) | |
1383 | { | |
1384 | struct page *page; | |
81819f0f | 1385 | void *start; |
81819f0f | 1386 | void *p; |
1f458cbf | 1387 | int order; |
54266640 | 1388 | int idx; |
81819f0f | 1389 | |
6cb06229 | 1390 | BUG_ON(flags & GFP_SLAB_BUG_MASK); |
81819f0f | 1391 | |
6cb06229 CL |
1392 | page = allocate_slab(s, |
1393 | flags & (GFP_RECLAIM_MASK | GFP_CONSTRAINT_MASK), node); | |
81819f0f CL |
1394 | if (!page) |
1395 | goto out; | |
1396 | ||
1f458cbf | 1397 | order = compound_order(page); |
205ab99d | 1398 | inc_slabs_node(s, page_to_nid(page), page->objects); |
1b4f59e3 | 1399 | page->slab_cache = s; |
c03f94cc | 1400 | __SetPageSlab(page); |
072bb0aa MG |
1401 | if (page->pfmemalloc) |
1402 | SetPageSlabPfmemalloc(page); | |
81819f0f CL |
1403 | |
1404 | start = page_address(page); | |
81819f0f CL |
1405 | |
1406 | if (unlikely(s->flags & SLAB_POISON)) | |
1f458cbf | 1407 | memset(start, POISON_INUSE, PAGE_SIZE << order); |
81819f0f | 1408 | |
54266640 WY |
1409 | for_each_object_idx(p, idx, s, start, page->objects) { |
1410 | setup_object(s, page, p); | |
1411 | if (likely(idx < page->objects)) | |
1412 | set_freepointer(s, p, p + s->size); | |
1413 | else | |
1414 | set_freepointer(s, p, NULL); | |
81819f0f | 1415 | } |
81819f0f CL |
1416 | |
1417 | page->freelist = start; | |
e6e82ea1 | 1418 | page->inuse = page->objects; |
8cb0a506 | 1419 | page->frozen = 1; |
81819f0f | 1420 | out: |
81819f0f CL |
1421 | return page; |
1422 | } | |
1423 | ||
1424 | static void __free_slab(struct kmem_cache *s, struct page *page) | |
1425 | { | |
834f3d11 CL |
1426 | int order = compound_order(page); |
1427 | int pages = 1 << order; | |
81819f0f | 1428 | |
af537b0a | 1429 | if (kmem_cache_debug(s)) { |
81819f0f CL |
1430 | void *p; |
1431 | ||
1432 | slab_pad_check(s, page); | |
224a88be CL |
1433 | for_each_object(p, s, page_address(page), |
1434 | page->objects) | |
f7cb1933 | 1435 | check_object(s, page, p, SLUB_RED_INACTIVE); |
81819f0f CL |
1436 | } |
1437 | ||
b1eeab67 | 1438 | kmemcheck_free_shadow(page, compound_order(page)); |
5a896d9e | 1439 | |
81819f0f CL |
1440 | mod_zone_page_state(page_zone(page), |
1441 | (s->flags & SLAB_RECLAIM_ACCOUNT) ? | |
1442 | NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE, | |
06428780 | 1443 | -pages); |
81819f0f | 1444 | |
072bb0aa | 1445 | __ClearPageSlabPfmemalloc(page); |
49bd5221 | 1446 | __ClearPageSlab(page); |
1f458cbf | 1447 | |
22b751c3 | 1448 | page_mapcount_reset(page); |
1eb5ac64 NP |
1449 | if (current->reclaim_state) |
1450 | current->reclaim_state->reclaimed_slab += pages; | |
5dfb4175 VD |
1451 | __free_pages(page, order); |
1452 | memcg_uncharge_slab(s, order); | |
81819f0f CL |
1453 | } |
1454 | ||
da9a638c LJ |
1455 | #define need_reserve_slab_rcu \ |
1456 | (sizeof(((struct page *)NULL)->lru) < sizeof(struct rcu_head)) | |
1457 | ||
81819f0f CL |
1458 | static void rcu_free_slab(struct rcu_head *h) |
1459 | { | |
1460 | struct page *page; | |
1461 | ||
da9a638c LJ |
1462 | if (need_reserve_slab_rcu) |
1463 | page = virt_to_head_page(h); | |
1464 | else | |
1465 | page = container_of((struct list_head *)h, struct page, lru); | |
1466 | ||
1b4f59e3 | 1467 | __free_slab(page->slab_cache, page); |
81819f0f CL |
1468 | } |
1469 | ||
1470 | static void free_slab(struct kmem_cache *s, struct page *page) | |
1471 | { | |
1472 | if (unlikely(s->flags & SLAB_DESTROY_BY_RCU)) { | |
da9a638c LJ |
1473 | struct rcu_head *head; |
1474 | ||
1475 | if (need_reserve_slab_rcu) { | |
1476 | int order = compound_order(page); | |
1477 | int offset = (PAGE_SIZE << order) - s->reserved; | |
1478 | ||
1479 | VM_BUG_ON(s->reserved != sizeof(*head)); | |
1480 | head = page_address(page) + offset; | |
1481 | } else { | |
1482 | /* | |
1483 | * RCU free overloads the RCU head over the LRU | |
1484 | */ | |
1485 | head = (void *)&page->lru; | |
1486 | } | |
81819f0f CL |
1487 | |
1488 | call_rcu(head, rcu_free_slab); | |
1489 | } else | |
1490 | __free_slab(s, page); | |
1491 | } | |
1492 | ||
1493 | static void discard_slab(struct kmem_cache *s, struct page *page) | |
1494 | { | |
205ab99d | 1495 | dec_slabs_node(s, page_to_nid(page), page->objects); |
81819f0f CL |
1496 | free_slab(s, page); |
1497 | } | |
1498 | ||
1499 | /* | |
5cc6eee8 | 1500 | * Management of partially allocated slabs. |
81819f0f | 1501 | */ |
1e4dd946 SR |
1502 | static inline void |
1503 | __add_partial(struct kmem_cache_node *n, struct page *page, int tail) | |
81819f0f | 1504 | { |
e95eed57 | 1505 | n->nr_partial++; |
136333d1 | 1506 | if (tail == DEACTIVATE_TO_TAIL) |
7c2e132c CL |
1507 | list_add_tail(&page->lru, &n->partial); |
1508 | else | |
1509 | list_add(&page->lru, &n->partial); | |
81819f0f CL |
1510 | } |
1511 | ||
1e4dd946 SR |
1512 | static inline void add_partial(struct kmem_cache_node *n, |
1513 | struct page *page, int tail) | |
62e346a8 | 1514 | { |
c65c1877 | 1515 | lockdep_assert_held(&n->list_lock); |
1e4dd946 SR |
1516 | __add_partial(n, page, tail); |
1517 | } | |
c65c1877 | 1518 | |
1e4dd946 SR |
1519 | static inline void |
1520 | __remove_partial(struct kmem_cache_node *n, struct page *page) | |
1521 | { | |
62e346a8 CL |
1522 | list_del(&page->lru); |
1523 | n->nr_partial--; | |
1524 | } | |
1525 | ||
1e4dd946 SR |
1526 | static inline void remove_partial(struct kmem_cache_node *n, |
1527 | struct page *page) | |
1528 | { | |
1529 | lockdep_assert_held(&n->list_lock); | |
1530 | __remove_partial(n, page); | |
1531 | } | |
1532 | ||
81819f0f | 1533 | /* |
7ced3719 CL |
1534 | * Remove slab from the partial list, freeze it and |
1535 | * return the pointer to the freelist. | |
81819f0f | 1536 | * |
497b66f2 | 1537 | * Returns a list of objects or NULL if it fails. |
81819f0f | 1538 | */ |
497b66f2 | 1539 | static inline void *acquire_slab(struct kmem_cache *s, |
acd19fd1 | 1540 | struct kmem_cache_node *n, struct page *page, |
633b0764 | 1541 | int mode, int *objects) |
81819f0f | 1542 | { |
2cfb7455 CL |
1543 | void *freelist; |
1544 | unsigned long counters; | |
1545 | struct page new; | |
1546 | ||
c65c1877 PZ |
1547 | lockdep_assert_held(&n->list_lock); |
1548 | ||
2cfb7455 CL |
1549 | /* |
1550 | * Zap the freelist and set the frozen bit. | |
1551 | * The old freelist is the list of objects for the | |
1552 | * per cpu allocation list. | |
1553 | */ | |
7ced3719 CL |
1554 | freelist = page->freelist; |
1555 | counters = page->counters; | |
1556 | new.counters = counters; | |
633b0764 | 1557 | *objects = new.objects - new.inuse; |
23910c50 | 1558 | if (mode) { |
7ced3719 | 1559 | new.inuse = page->objects; |
23910c50 PE |
1560 | new.freelist = NULL; |
1561 | } else { | |
1562 | new.freelist = freelist; | |
1563 | } | |
2cfb7455 | 1564 | |
a0132ac0 | 1565 | VM_BUG_ON(new.frozen); |
7ced3719 | 1566 | new.frozen = 1; |
2cfb7455 | 1567 | |
7ced3719 | 1568 | if (!__cmpxchg_double_slab(s, page, |
2cfb7455 | 1569 | freelist, counters, |
02d7633f | 1570 | new.freelist, new.counters, |
7ced3719 | 1571 | "acquire_slab")) |
7ced3719 | 1572 | return NULL; |
2cfb7455 CL |
1573 | |
1574 | remove_partial(n, page); | |
7ced3719 | 1575 | WARN_ON(!freelist); |
49e22585 | 1576 | return freelist; |
81819f0f CL |
1577 | } |
1578 | ||
633b0764 | 1579 | static void put_cpu_partial(struct kmem_cache *s, struct page *page, int drain); |
8ba00bb6 | 1580 | static inline bool pfmemalloc_match(struct page *page, gfp_t gfpflags); |
49e22585 | 1581 | |
81819f0f | 1582 | /* |
672bba3a | 1583 | * Try to allocate a partial slab from a specific node. |
81819f0f | 1584 | */ |
8ba00bb6 JK |
1585 | static void *get_partial_node(struct kmem_cache *s, struct kmem_cache_node *n, |
1586 | struct kmem_cache_cpu *c, gfp_t flags) | |
81819f0f | 1587 | { |
49e22585 CL |
1588 | struct page *page, *page2; |
1589 | void *object = NULL; | |
633b0764 JK |
1590 | int available = 0; |
1591 | int objects; | |
81819f0f CL |
1592 | |
1593 | /* | |
1594 | * Racy check. If we mistakenly see no partial slabs then we | |
1595 | * just allocate an empty slab. If we mistakenly try to get a | |
672bba3a CL |
1596 | * partial slab and there is none available then get_partials() |
1597 | * will return NULL. | |
81819f0f CL |
1598 | */ |
1599 | if (!n || !n->nr_partial) | |
1600 | return NULL; | |
1601 | ||
1602 | spin_lock(&n->list_lock); | |
49e22585 | 1603 | list_for_each_entry_safe(page, page2, &n->partial, lru) { |
8ba00bb6 | 1604 | void *t; |
49e22585 | 1605 | |
8ba00bb6 JK |
1606 | if (!pfmemalloc_match(page, flags)) |
1607 | continue; | |
1608 | ||
633b0764 | 1609 | t = acquire_slab(s, n, page, object == NULL, &objects); |
49e22585 CL |
1610 | if (!t) |
1611 | break; | |
1612 | ||
633b0764 | 1613 | available += objects; |
12d79634 | 1614 | if (!object) { |
49e22585 | 1615 | c->page = page; |
49e22585 | 1616 | stat(s, ALLOC_FROM_PARTIAL); |
49e22585 | 1617 | object = t; |
49e22585 | 1618 | } else { |
633b0764 | 1619 | put_cpu_partial(s, page, 0); |
8028dcea | 1620 | stat(s, CPU_PARTIAL_NODE); |
49e22585 | 1621 | } |
345c905d JK |
1622 | if (!kmem_cache_has_cpu_partial(s) |
1623 | || available > s->cpu_partial / 2) | |
49e22585 CL |
1624 | break; |
1625 | ||
497b66f2 | 1626 | } |
81819f0f | 1627 | spin_unlock(&n->list_lock); |
497b66f2 | 1628 | return object; |
81819f0f CL |
1629 | } |
1630 | ||
1631 | /* | |
672bba3a | 1632 | * Get a page from somewhere. Search in increasing NUMA distances. |
81819f0f | 1633 | */ |
de3ec035 | 1634 | static void *get_any_partial(struct kmem_cache *s, gfp_t flags, |
acd19fd1 | 1635 | struct kmem_cache_cpu *c) |
81819f0f CL |
1636 | { |
1637 | #ifdef CONFIG_NUMA | |
1638 | struct zonelist *zonelist; | |
dd1a239f | 1639 | struct zoneref *z; |
54a6eb5c MG |
1640 | struct zone *zone; |
1641 | enum zone_type high_zoneidx = gfp_zone(flags); | |
497b66f2 | 1642 | void *object; |
cc9a6c87 | 1643 | unsigned int cpuset_mems_cookie; |
81819f0f CL |
1644 | |
1645 | /* | |
672bba3a CL |
1646 | * The defrag ratio allows a configuration of the tradeoffs between |
1647 | * inter node defragmentation and node local allocations. A lower | |
1648 | * defrag_ratio increases the tendency to do local allocations | |
1649 | * instead of attempting to obtain partial slabs from other nodes. | |
81819f0f | 1650 | * |
672bba3a CL |
1651 | * If the defrag_ratio is set to 0 then kmalloc() always |
1652 | * returns node local objects. If the ratio is higher then kmalloc() | |
1653 | * may return off node objects because partial slabs are obtained | |
1654 | * from other nodes and filled up. | |
81819f0f | 1655 | * |
6446faa2 | 1656 | * If /sys/kernel/slab/xx/defrag_ratio is set to 100 (which makes |
672bba3a CL |
1657 | * defrag_ratio = 1000) then every (well almost) allocation will |
1658 | * first attempt to defrag slab caches on other nodes. This means | |
1659 | * scanning over all nodes to look for partial slabs which may be | |
1660 | * expensive if we do it every time we are trying to find a slab | |
1661 | * with available objects. | |
81819f0f | 1662 | */ |
9824601e CL |
1663 | if (!s->remote_node_defrag_ratio || |
1664 | get_cycles() % 1024 > s->remote_node_defrag_ratio) | |
81819f0f CL |
1665 | return NULL; |
1666 | ||
cc9a6c87 | 1667 | do { |
d26914d1 | 1668 | cpuset_mems_cookie = read_mems_allowed_begin(); |
2a389610 | 1669 | zonelist = node_zonelist(mempolicy_slab_node(), flags); |
cc9a6c87 MG |
1670 | for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) { |
1671 | struct kmem_cache_node *n; | |
1672 | ||
1673 | n = get_node(s, zone_to_nid(zone)); | |
1674 | ||
1675 | if (n && cpuset_zone_allowed_hardwall(zone, flags) && | |
1676 | n->nr_partial > s->min_partial) { | |
8ba00bb6 | 1677 | object = get_partial_node(s, n, c, flags); |
cc9a6c87 MG |
1678 | if (object) { |
1679 | /* | |
d26914d1 MG |
1680 | * Don't check read_mems_allowed_retry() |
1681 | * here - if mems_allowed was updated in | |
1682 | * parallel, that was a harmless race | |
1683 | * between allocation and the cpuset | |
1684 | * update | |
cc9a6c87 | 1685 | */ |
cc9a6c87 MG |
1686 | return object; |
1687 | } | |
c0ff7453 | 1688 | } |
81819f0f | 1689 | } |
d26914d1 | 1690 | } while (read_mems_allowed_retry(cpuset_mems_cookie)); |
81819f0f CL |
1691 | #endif |
1692 | return NULL; | |
1693 | } | |
1694 | ||
1695 | /* | |
1696 | * Get a partial page, lock it and return it. | |
1697 | */ | |
497b66f2 | 1698 | static void *get_partial(struct kmem_cache *s, gfp_t flags, int node, |
acd19fd1 | 1699 | struct kmem_cache_cpu *c) |
81819f0f | 1700 | { |
497b66f2 | 1701 | void *object; |
a561ce00 JK |
1702 | int searchnode = node; |
1703 | ||
1704 | if (node == NUMA_NO_NODE) | |
1705 | searchnode = numa_mem_id(); | |
1706 | else if (!node_present_pages(node)) | |
1707 | searchnode = node_to_mem_node(node); | |
81819f0f | 1708 | |
8ba00bb6 | 1709 | object = get_partial_node(s, get_node(s, searchnode), c, flags); |
497b66f2 CL |
1710 | if (object || node != NUMA_NO_NODE) |
1711 | return object; | |
81819f0f | 1712 | |
acd19fd1 | 1713 | return get_any_partial(s, flags, c); |
81819f0f CL |
1714 | } |
1715 | ||
8a5ec0ba CL |
1716 | #ifdef CONFIG_PREEMPT |
1717 | /* | |
1718 | * Calculate the next globally unique transaction for disambiguiation | |
1719 | * during cmpxchg. The transactions start with the cpu number and are then | |
1720 | * incremented by CONFIG_NR_CPUS. | |
1721 | */ | |
1722 | #define TID_STEP roundup_pow_of_two(CONFIG_NR_CPUS) | |
1723 | #else | |
1724 | /* | |
1725 | * No preemption supported therefore also no need to check for | |
1726 | * different cpus. | |
1727 | */ | |
1728 | #define TID_STEP 1 | |
1729 | #endif | |
1730 | ||
1731 | static inline unsigned long next_tid(unsigned long tid) | |
1732 | { | |
1733 | return tid + TID_STEP; | |
1734 | } | |
1735 | ||
1736 | static inline unsigned int tid_to_cpu(unsigned long tid) | |
1737 | { | |
1738 | return tid % TID_STEP; | |
1739 | } | |
1740 | ||
1741 | static inline unsigned long tid_to_event(unsigned long tid) | |
1742 | { | |
1743 | return tid / TID_STEP; | |
1744 | } | |
1745 | ||
1746 | static inline unsigned int init_tid(int cpu) | |
1747 | { | |
1748 | return cpu; | |
1749 | } | |
1750 | ||
1751 | static inline void note_cmpxchg_failure(const char *n, | |
1752 | const struct kmem_cache *s, unsigned long tid) | |
1753 | { | |
1754 | #ifdef SLUB_DEBUG_CMPXCHG | |
1755 | unsigned long actual_tid = __this_cpu_read(s->cpu_slab->tid); | |
1756 | ||
f9f58285 | 1757 | pr_info("%s %s: cmpxchg redo ", n, s->name); |
8a5ec0ba CL |
1758 | |
1759 | #ifdef CONFIG_PREEMPT | |
1760 | if (tid_to_cpu(tid) != tid_to_cpu(actual_tid)) | |
f9f58285 | 1761 | pr_warn("due to cpu change %d -> %d\n", |
8a5ec0ba CL |
1762 | tid_to_cpu(tid), tid_to_cpu(actual_tid)); |
1763 | else | |
1764 | #endif | |
1765 | if (tid_to_event(tid) != tid_to_event(actual_tid)) | |
f9f58285 | 1766 | pr_warn("due to cpu running other code. Event %ld->%ld\n", |
8a5ec0ba CL |
1767 | tid_to_event(tid), tid_to_event(actual_tid)); |
1768 | else | |
f9f58285 | 1769 | pr_warn("for unknown reason: actual=%lx was=%lx target=%lx\n", |
8a5ec0ba CL |
1770 | actual_tid, tid, next_tid(tid)); |
1771 | #endif | |
4fdccdfb | 1772 | stat(s, CMPXCHG_DOUBLE_CPU_FAIL); |
8a5ec0ba CL |
1773 | } |
1774 | ||
788e1aad | 1775 | static void init_kmem_cache_cpus(struct kmem_cache *s) |
8a5ec0ba | 1776 | { |
8a5ec0ba CL |
1777 | int cpu; |
1778 | ||
1779 | for_each_possible_cpu(cpu) | |
1780 | per_cpu_ptr(s->cpu_slab, cpu)->tid = init_tid(cpu); | |
8a5ec0ba | 1781 | } |
2cfb7455 | 1782 | |
81819f0f CL |
1783 | /* |
1784 | * Remove the cpu slab | |
1785 | */ | |
d0e0ac97 CG |
1786 | static void deactivate_slab(struct kmem_cache *s, struct page *page, |
1787 | void *freelist) | |
81819f0f | 1788 | { |
2cfb7455 | 1789 | enum slab_modes { M_NONE, M_PARTIAL, M_FULL, M_FREE }; |
2cfb7455 CL |
1790 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); |
1791 | int lock = 0; | |
1792 | enum slab_modes l = M_NONE, m = M_NONE; | |
2cfb7455 | 1793 | void *nextfree; |
136333d1 | 1794 | int tail = DEACTIVATE_TO_HEAD; |
2cfb7455 CL |
1795 | struct page new; |
1796 | struct page old; | |
1797 | ||
1798 | if (page->freelist) { | |
84e554e6 | 1799 | stat(s, DEACTIVATE_REMOTE_FREES); |
136333d1 | 1800 | tail = DEACTIVATE_TO_TAIL; |
2cfb7455 CL |
1801 | } |
1802 | ||
894b8788 | 1803 | /* |
2cfb7455 CL |
1804 | * Stage one: Free all available per cpu objects back |
1805 | * to the page freelist while it is still frozen. Leave the | |
1806 | * last one. | |
1807 | * | |
1808 | * There is no need to take the list->lock because the page | |
1809 | * is still frozen. | |
1810 | */ | |
1811 | while (freelist && (nextfree = get_freepointer(s, freelist))) { | |
1812 | void *prior; | |
1813 | unsigned long counters; | |
1814 | ||
1815 | do { | |
1816 | prior = page->freelist; | |
1817 | counters = page->counters; | |
1818 | set_freepointer(s, freelist, prior); | |
1819 | new.counters = counters; | |
1820 | new.inuse--; | |
a0132ac0 | 1821 | VM_BUG_ON(!new.frozen); |
2cfb7455 | 1822 | |
1d07171c | 1823 | } while (!__cmpxchg_double_slab(s, page, |
2cfb7455 CL |
1824 | prior, counters, |
1825 | freelist, new.counters, | |
1826 | "drain percpu freelist")); | |
1827 | ||
1828 | freelist = nextfree; | |
1829 | } | |
1830 | ||
894b8788 | 1831 | /* |
2cfb7455 CL |
1832 | * Stage two: Ensure that the page is unfrozen while the |
1833 | * list presence reflects the actual number of objects | |
1834 | * during unfreeze. | |
1835 | * | |
1836 | * We setup the list membership and then perform a cmpxchg | |
1837 | * with the count. If there is a mismatch then the page | |
1838 | * is not unfrozen but the page is on the wrong list. | |
1839 | * | |
1840 | * Then we restart the process which may have to remove | |
1841 | * the page from the list that we just put it on again | |
1842 | * because the number of objects in the slab may have | |
1843 | * changed. | |
894b8788 | 1844 | */ |
2cfb7455 | 1845 | redo: |
894b8788 | 1846 | |
2cfb7455 CL |
1847 | old.freelist = page->freelist; |
1848 | old.counters = page->counters; | |
a0132ac0 | 1849 | VM_BUG_ON(!old.frozen); |
7c2e132c | 1850 | |
2cfb7455 CL |
1851 | /* Determine target state of the slab */ |
1852 | new.counters = old.counters; | |
1853 | if (freelist) { | |
1854 | new.inuse--; | |
1855 | set_freepointer(s, freelist, old.freelist); | |
1856 | new.freelist = freelist; | |
1857 | } else | |
1858 | new.freelist = old.freelist; | |
1859 | ||
1860 | new.frozen = 0; | |
1861 | ||
8a5b20ae | 1862 | if (!new.inuse && n->nr_partial >= s->min_partial) |
2cfb7455 CL |
1863 | m = M_FREE; |
1864 | else if (new.freelist) { | |
1865 | m = M_PARTIAL; | |
1866 | if (!lock) { | |
1867 | lock = 1; | |
1868 | /* | |
1869 | * Taking the spinlock removes the possiblity | |
1870 | * that acquire_slab() will see a slab page that | |
1871 | * is frozen | |
1872 | */ | |
1873 | spin_lock(&n->list_lock); | |
1874 | } | |
1875 | } else { | |
1876 | m = M_FULL; | |
1877 | if (kmem_cache_debug(s) && !lock) { | |
1878 | lock = 1; | |
1879 | /* | |
1880 | * This also ensures that the scanning of full | |
1881 | * slabs from diagnostic functions will not see | |
1882 | * any frozen slabs. | |
1883 | */ | |
1884 | spin_lock(&n->list_lock); | |
1885 | } | |
1886 | } | |
1887 | ||
1888 | if (l != m) { | |
1889 | ||
1890 | if (l == M_PARTIAL) | |
1891 | ||
1892 | remove_partial(n, page); | |
1893 | ||
1894 | else if (l == M_FULL) | |
894b8788 | 1895 | |
c65c1877 | 1896 | remove_full(s, n, page); |
2cfb7455 CL |
1897 | |
1898 | if (m == M_PARTIAL) { | |
1899 | ||
1900 | add_partial(n, page, tail); | |
136333d1 | 1901 | stat(s, tail); |
2cfb7455 CL |
1902 | |
1903 | } else if (m == M_FULL) { | |
894b8788 | 1904 | |
2cfb7455 CL |
1905 | stat(s, DEACTIVATE_FULL); |
1906 | add_full(s, n, page); | |
1907 | ||
1908 | } | |
1909 | } | |
1910 | ||
1911 | l = m; | |
1d07171c | 1912 | if (!__cmpxchg_double_slab(s, page, |
2cfb7455 CL |
1913 | old.freelist, old.counters, |
1914 | new.freelist, new.counters, | |
1915 | "unfreezing slab")) | |
1916 | goto redo; | |
1917 | ||
2cfb7455 CL |
1918 | if (lock) |
1919 | spin_unlock(&n->list_lock); | |
1920 | ||
1921 | if (m == M_FREE) { | |
1922 | stat(s, DEACTIVATE_EMPTY); | |
1923 | discard_slab(s, page); | |
1924 | stat(s, FREE_SLAB); | |
894b8788 | 1925 | } |
81819f0f CL |
1926 | } |
1927 | ||
d24ac77f JK |
1928 | /* |
1929 | * Unfreeze all the cpu partial slabs. | |
1930 | * | |
59a09917 CL |
1931 | * This function must be called with interrupts disabled |
1932 | * for the cpu using c (or some other guarantee must be there | |
1933 | * to guarantee no concurrent accesses). | |
d24ac77f | 1934 | */ |
59a09917 CL |
1935 | static void unfreeze_partials(struct kmem_cache *s, |
1936 | struct kmem_cache_cpu *c) | |
49e22585 | 1937 | { |
345c905d | 1938 | #ifdef CONFIG_SLUB_CPU_PARTIAL |
43d77867 | 1939 | struct kmem_cache_node *n = NULL, *n2 = NULL; |
9ada1934 | 1940 | struct page *page, *discard_page = NULL; |
49e22585 CL |
1941 | |
1942 | while ((page = c->partial)) { | |
49e22585 CL |
1943 | struct page new; |
1944 | struct page old; | |
1945 | ||
1946 | c->partial = page->next; | |
43d77867 JK |
1947 | |
1948 | n2 = get_node(s, page_to_nid(page)); | |
1949 | if (n != n2) { | |
1950 | if (n) | |
1951 | spin_unlock(&n->list_lock); | |
1952 | ||
1953 | n = n2; | |
1954 | spin_lock(&n->list_lock); | |
1955 | } | |
49e22585 CL |
1956 | |
1957 | do { | |
1958 | ||
1959 | old.freelist = page->freelist; | |
1960 | old.counters = page->counters; | |
a0132ac0 | 1961 | VM_BUG_ON(!old.frozen); |
49e22585 CL |
1962 | |
1963 | new.counters = old.counters; | |
1964 | new.freelist = old.freelist; | |
1965 | ||
1966 | new.frozen = 0; | |
1967 | ||
d24ac77f | 1968 | } while (!__cmpxchg_double_slab(s, page, |
49e22585 CL |
1969 | old.freelist, old.counters, |
1970 | new.freelist, new.counters, | |
1971 | "unfreezing slab")); | |
1972 | ||
8a5b20ae | 1973 | if (unlikely(!new.inuse && n->nr_partial >= s->min_partial)) { |
9ada1934 SL |
1974 | page->next = discard_page; |
1975 | discard_page = page; | |
43d77867 JK |
1976 | } else { |
1977 | add_partial(n, page, DEACTIVATE_TO_TAIL); | |
1978 | stat(s, FREE_ADD_PARTIAL); | |
49e22585 CL |
1979 | } |
1980 | } | |
1981 | ||
1982 | if (n) | |
1983 | spin_unlock(&n->list_lock); | |
9ada1934 SL |
1984 | |
1985 | while (discard_page) { | |
1986 | page = discard_page; | |
1987 | discard_page = discard_page->next; | |
1988 | ||
1989 | stat(s, DEACTIVATE_EMPTY); | |
1990 | discard_slab(s, page); | |
1991 | stat(s, FREE_SLAB); | |
1992 | } | |
345c905d | 1993 | #endif |
49e22585 CL |
1994 | } |
1995 | ||
1996 | /* | |
1997 | * Put a page that was just frozen (in __slab_free) into a partial page | |
1998 | * slot if available. This is done without interrupts disabled and without | |
1999 | * preemption disabled. The cmpxchg is racy and may put the partial page | |
2000 | * onto a random cpus partial slot. | |
2001 | * | |
2002 | * If we did not find a slot then simply move all the partials to the | |
2003 | * per node partial list. | |
2004 | */ | |
633b0764 | 2005 | static void put_cpu_partial(struct kmem_cache *s, struct page *page, int drain) |
49e22585 | 2006 | { |
345c905d | 2007 | #ifdef CONFIG_SLUB_CPU_PARTIAL |
49e22585 CL |
2008 | struct page *oldpage; |
2009 | int pages; | |
2010 | int pobjects; | |
2011 | ||
2012 | do { | |
2013 | pages = 0; | |
2014 | pobjects = 0; | |
2015 | oldpage = this_cpu_read(s->cpu_slab->partial); | |
2016 | ||
2017 | if (oldpage) { | |
2018 | pobjects = oldpage->pobjects; | |
2019 | pages = oldpage->pages; | |
2020 | if (drain && pobjects > s->cpu_partial) { | |
2021 | unsigned long flags; | |
2022 | /* | |
2023 | * partial array is full. Move the existing | |
2024 | * set to the per node partial list. | |
2025 | */ | |
2026 | local_irq_save(flags); | |
59a09917 | 2027 | unfreeze_partials(s, this_cpu_ptr(s->cpu_slab)); |
49e22585 | 2028 | local_irq_restore(flags); |
e24fc410 | 2029 | oldpage = NULL; |
49e22585 CL |
2030 | pobjects = 0; |
2031 | pages = 0; | |
8028dcea | 2032 | stat(s, CPU_PARTIAL_DRAIN); |
49e22585 CL |
2033 | } |
2034 | } | |
2035 | ||
2036 | pages++; | |
2037 | pobjects += page->objects - page->inuse; | |
2038 | ||
2039 | page->pages = pages; | |
2040 | page->pobjects = pobjects; | |
2041 | page->next = oldpage; | |
2042 | ||
d0e0ac97 CG |
2043 | } while (this_cpu_cmpxchg(s->cpu_slab->partial, oldpage, page) |
2044 | != oldpage); | |
345c905d | 2045 | #endif |
49e22585 CL |
2046 | } |
2047 | ||
dfb4f096 | 2048 | static inline void flush_slab(struct kmem_cache *s, struct kmem_cache_cpu *c) |
81819f0f | 2049 | { |
84e554e6 | 2050 | stat(s, CPUSLAB_FLUSH); |
c17dda40 CL |
2051 | deactivate_slab(s, c->page, c->freelist); |
2052 | ||
2053 | c->tid = next_tid(c->tid); | |
2054 | c->page = NULL; | |
2055 | c->freelist = NULL; | |
81819f0f CL |
2056 | } |
2057 | ||
2058 | /* | |
2059 | * Flush cpu slab. | |
6446faa2 | 2060 | * |
81819f0f CL |
2061 | * Called from IPI handler with interrupts disabled. |
2062 | */ | |
0c710013 | 2063 | static inline void __flush_cpu_slab(struct kmem_cache *s, int cpu) |
81819f0f | 2064 | { |
9dfc6e68 | 2065 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu); |
81819f0f | 2066 | |
49e22585 CL |
2067 | if (likely(c)) { |
2068 | if (c->page) | |
2069 | flush_slab(s, c); | |
2070 | ||
59a09917 | 2071 | unfreeze_partials(s, c); |
49e22585 | 2072 | } |
81819f0f CL |
2073 | } |
2074 | ||
2075 | static void flush_cpu_slab(void *d) | |
2076 | { | |
2077 | struct kmem_cache *s = d; | |
81819f0f | 2078 | |
dfb4f096 | 2079 | __flush_cpu_slab(s, smp_processor_id()); |
81819f0f CL |
2080 | } |
2081 | ||
a8364d55 GBY |
2082 | static bool has_cpu_slab(int cpu, void *info) |
2083 | { | |
2084 | struct kmem_cache *s = info; | |
2085 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu); | |
2086 | ||
02e1a9cd | 2087 | return c->page || c->partial; |
a8364d55 GBY |
2088 | } |
2089 | ||
81819f0f CL |
2090 | static void flush_all(struct kmem_cache *s) |
2091 | { | |
a8364d55 | 2092 | on_each_cpu_cond(has_cpu_slab, flush_cpu_slab, s, 1, GFP_ATOMIC); |
81819f0f CL |
2093 | } |
2094 | ||
dfb4f096 CL |
2095 | /* |
2096 | * Check if the objects in a per cpu structure fit numa | |
2097 | * locality expectations. | |
2098 | */ | |
57d437d2 | 2099 | static inline int node_match(struct page *page, int node) |
dfb4f096 CL |
2100 | { |
2101 | #ifdef CONFIG_NUMA | |
4d7868e6 | 2102 | if (!page || (node != NUMA_NO_NODE && page_to_nid(page) != node)) |
dfb4f096 CL |
2103 | return 0; |
2104 | #endif | |
2105 | return 1; | |
2106 | } | |
2107 | ||
9a02d699 | 2108 | #ifdef CONFIG_SLUB_DEBUG |
781b2ba6 PE |
2109 | static int count_free(struct page *page) |
2110 | { | |
2111 | return page->objects - page->inuse; | |
2112 | } | |
2113 | ||
9a02d699 DR |
2114 | static inline unsigned long node_nr_objs(struct kmem_cache_node *n) |
2115 | { | |
2116 | return atomic_long_read(&n->total_objects); | |
2117 | } | |
2118 | #endif /* CONFIG_SLUB_DEBUG */ | |
2119 | ||
2120 | #if defined(CONFIG_SLUB_DEBUG) || defined(CONFIG_SYSFS) | |
781b2ba6 PE |
2121 | static unsigned long count_partial(struct kmem_cache_node *n, |
2122 | int (*get_count)(struct page *)) | |
2123 | { | |
2124 | unsigned long flags; | |
2125 | unsigned long x = 0; | |
2126 | struct page *page; | |
2127 | ||
2128 | spin_lock_irqsave(&n->list_lock, flags); | |
2129 | list_for_each_entry(page, &n->partial, lru) | |
2130 | x += get_count(page); | |
2131 | spin_unlock_irqrestore(&n->list_lock, flags); | |
2132 | return x; | |
2133 | } | |
9a02d699 | 2134 | #endif /* CONFIG_SLUB_DEBUG || CONFIG_SYSFS */ |
26c02cf0 | 2135 | |
781b2ba6 PE |
2136 | static noinline void |
2137 | slab_out_of_memory(struct kmem_cache *s, gfp_t gfpflags, int nid) | |
2138 | { | |
9a02d699 DR |
2139 | #ifdef CONFIG_SLUB_DEBUG |
2140 | static DEFINE_RATELIMIT_STATE(slub_oom_rs, DEFAULT_RATELIMIT_INTERVAL, | |
2141 | DEFAULT_RATELIMIT_BURST); | |
781b2ba6 | 2142 | int node; |
fa45dc25 | 2143 | struct kmem_cache_node *n; |
781b2ba6 | 2144 | |
9a02d699 DR |
2145 | if ((gfpflags & __GFP_NOWARN) || !__ratelimit(&slub_oom_rs)) |
2146 | return; | |
2147 | ||
f9f58285 | 2148 | pr_warn("SLUB: Unable to allocate memory on node %d (gfp=0x%x)\n", |
781b2ba6 | 2149 | nid, gfpflags); |
f9f58285 FF |
2150 | pr_warn(" cache: %s, object size: %d, buffer size: %d, default order: %d, min order: %d\n", |
2151 | s->name, s->object_size, s->size, oo_order(s->oo), | |
2152 | oo_order(s->min)); | |
781b2ba6 | 2153 | |
3b0efdfa | 2154 | if (oo_order(s->min) > get_order(s->object_size)) |
f9f58285 FF |
2155 | pr_warn(" %s debugging increased min order, use slub_debug=O to disable.\n", |
2156 | s->name); | |
fa5ec8a1 | 2157 | |
fa45dc25 | 2158 | for_each_kmem_cache_node(s, node, n) { |
781b2ba6 PE |
2159 | unsigned long nr_slabs; |
2160 | unsigned long nr_objs; | |
2161 | unsigned long nr_free; | |
2162 | ||
26c02cf0 AB |
2163 | nr_free = count_partial(n, count_free); |
2164 | nr_slabs = node_nr_slabs(n); | |
2165 | nr_objs = node_nr_objs(n); | |
781b2ba6 | 2166 | |
f9f58285 | 2167 | pr_warn(" node %d: slabs: %ld, objs: %ld, free: %ld\n", |
781b2ba6 PE |
2168 | node, nr_slabs, nr_objs, nr_free); |
2169 | } | |
9a02d699 | 2170 | #endif |
781b2ba6 PE |
2171 | } |
2172 | ||
497b66f2 CL |
2173 | static inline void *new_slab_objects(struct kmem_cache *s, gfp_t flags, |
2174 | int node, struct kmem_cache_cpu **pc) | |
2175 | { | |
6faa6833 | 2176 | void *freelist; |
188fd063 CL |
2177 | struct kmem_cache_cpu *c = *pc; |
2178 | struct page *page; | |
497b66f2 | 2179 | |
188fd063 | 2180 | freelist = get_partial(s, flags, node, c); |
497b66f2 | 2181 | |
188fd063 CL |
2182 | if (freelist) |
2183 | return freelist; | |
2184 | ||
2185 | page = new_slab(s, flags, node); | |
497b66f2 | 2186 | if (page) { |
7c8e0181 | 2187 | c = raw_cpu_ptr(s->cpu_slab); |
497b66f2 CL |
2188 | if (c->page) |
2189 | flush_slab(s, c); | |
2190 | ||
2191 | /* | |
2192 | * No other reference to the page yet so we can | |
2193 | * muck around with it freely without cmpxchg | |
2194 | */ | |
6faa6833 | 2195 | freelist = page->freelist; |
497b66f2 CL |
2196 | page->freelist = NULL; |
2197 | ||
2198 | stat(s, ALLOC_SLAB); | |
497b66f2 CL |
2199 | c->page = page; |
2200 | *pc = c; | |
2201 | } else | |
6faa6833 | 2202 | freelist = NULL; |
497b66f2 | 2203 | |
6faa6833 | 2204 | return freelist; |
497b66f2 CL |
2205 | } |
2206 | ||
072bb0aa MG |
2207 | static inline bool pfmemalloc_match(struct page *page, gfp_t gfpflags) |
2208 | { | |
2209 | if (unlikely(PageSlabPfmemalloc(page))) | |
2210 | return gfp_pfmemalloc_allowed(gfpflags); | |
2211 | ||
2212 | return true; | |
2213 | } | |
2214 | ||
213eeb9f | 2215 | /* |
d0e0ac97 CG |
2216 | * Check the page->freelist of a page and either transfer the freelist to the |
2217 | * per cpu freelist or deactivate the page. | |
213eeb9f CL |
2218 | * |
2219 | * The page is still frozen if the return value is not NULL. | |
2220 | * | |
2221 | * If this function returns NULL then the page has been unfrozen. | |
d24ac77f JK |
2222 | * |
2223 | * This function must be called with interrupt disabled. | |
213eeb9f CL |
2224 | */ |
2225 | static inline void *get_freelist(struct kmem_cache *s, struct page *page) | |
2226 | { | |
2227 | struct page new; | |
2228 | unsigned long counters; | |
2229 | void *freelist; | |
2230 | ||
2231 | do { | |
2232 | freelist = page->freelist; | |
2233 | counters = page->counters; | |
6faa6833 | 2234 | |
213eeb9f | 2235 | new.counters = counters; |
a0132ac0 | 2236 | VM_BUG_ON(!new.frozen); |
213eeb9f CL |
2237 | |
2238 | new.inuse = page->objects; | |
2239 | new.frozen = freelist != NULL; | |
2240 | ||
d24ac77f | 2241 | } while (!__cmpxchg_double_slab(s, page, |
213eeb9f CL |
2242 | freelist, counters, |
2243 | NULL, new.counters, | |
2244 | "get_freelist")); | |
2245 | ||
2246 | return freelist; | |
2247 | } | |
2248 | ||
81819f0f | 2249 | /* |
894b8788 CL |
2250 | * Slow path. The lockless freelist is empty or we need to perform |
2251 | * debugging duties. | |
2252 | * | |
894b8788 CL |
2253 | * Processing is still very fast if new objects have been freed to the |
2254 | * regular freelist. In that case we simply take over the regular freelist | |
2255 | * as the lockless freelist and zap the regular freelist. | |
81819f0f | 2256 | * |
894b8788 CL |
2257 | * If that is not working then we fall back to the partial lists. We take the |
2258 | * first element of the freelist as the object to allocate now and move the | |
2259 | * rest of the freelist to the lockless freelist. | |
81819f0f | 2260 | * |
894b8788 | 2261 | * And if we were unable to get a new slab from the partial slab lists then |
6446faa2 CL |
2262 | * we need to allocate a new slab. This is the slowest path since it involves |
2263 | * a call to the page allocator and the setup of a new slab. | |
81819f0f | 2264 | */ |
ce71e27c EGM |
2265 | static void *__slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node, |
2266 | unsigned long addr, struct kmem_cache_cpu *c) | |
81819f0f | 2267 | { |
6faa6833 | 2268 | void *freelist; |
f6e7def7 | 2269 | struct page *page; |
8a5ec0ba CL |
2270 | unsigned long flags; |
2271 | ||
2272 | local_irq_save(flags); | |
2273 | #ifdef CONFIG_PREEMPT | |
2274 | /* | |
2275 | * We may have been preempted and rescheduled on a different | |
2276 | * cpu before disabling interrupts. Need to reload cpu area | |
2277 | * pointer. | |
2278 | */ | |
2279 | c = this_cpu_ptr(s->cpu_slab); | |
8a5ec0ba | 2280 | #endif |
81819f0f | 2281 | |
f6e7def7 CL |
2282 | page = c->page; |
2283 | if (!page) | |
81819f0f | 2284 | goto new_slab; |
49e22585 | 2285 | redo: |
6faa6833 | 2286 | |
57d437d2 | 2287 | if (unlikely(!node_match(page, node))) { |
a561ce00 JK |
2288 | int searchnode = node; |
2289 | ||
2290 | if (node != NUMA_NO_NODE && !node_present_pages(node)) | |
2291 | searchnode = node_to_mem_node(node); | |
2292 | ||
2293 | if (unlikely(!node_match(page, searchnode))) { | |
2294 | stat(s, ALLOC_NODE_MISMATCH); | |
2295 | deactivate_slab(s, page, c->freelist); | |
2296 | c->page = NULL; | |
2297 | c->freelist = NULL; | |
2298 | goto new_slab; | |
2299 | } | |
fc59c053 | 2300 | } |
6446faa2 | 2301 | |
072bb0aa MG |
2302 | /* |
2303 | * By rights, we should be searching for a slab page that was | |
2304 | * PFMEMALLOC but right now, we are losing the pfmemalloc | |
2305 | * information when the page leaves the per-cpu allocator | |
2306 | */ | |
2307 | if (unlikely(!pfmemalloc_match(page, gfpflags))) { | |
2308 | deactivate_slab(s, page, c->freelist); | |
2309 | c->page = NULL; | |
2310 | c->freelist = NULL; | |
2311 | goto new_slab; | |
2312 | } | |
2313 | ||
73736e03 | 2314 | /* must check again c->freelist in case of cpu migration or IRQ */ |
6faa6833 CL |
2315 | freelist = c->freelist; |
2316 | if (freelist) | |
73736e03 | 2317 | goto load_freelist; |
03e404af | 2318 | |
f6e7def7 | 2319 | freelist = get_freelist(s, page); |
6446faa2 | 2320 | |
6faa6833 | 2321 | if (!freelist) { |
03e404af CL |
2322 | c->page = NULL; |
2323 | stat(s, DEACTIVATE_BYPASS); | |
fc59c053 | 2324 | goto new_slab; |
03e404af | 2325 | } |
6446faa2 | 2326 | |
84e554e6 | 2327 | stat(s, ALLOC_REFILL); |
6446faa2 | 2328 | |
894b8788 | 2329 | load_freelist: |
507effea CL |
2330 | /* |
2331 | * freelist is pointing to the list of objects to be used. | |
2332 | * page is pointing to the page from which the objects are obtained. | |
2333 | * That page must be frozen for per cpu allocations to work. | |
2334 | */ | |
a0132ac0 | 2335 | VM_BUG_ON(!c->page->frozen); |
6faa6833 | 2336 | c->freelist = get_freepointer(s, freelist); |
8a5ec0ba CL |
2337 | c->tid = next_tid(c->tid); |
2338 | local_irq_restore(flags); | |
6faa6833 | 2339 | return freelist; |
81819f0f | 2340 | |
81819f0f | 2341 | new_slab: |
2cfb7455 | 2342 | |
49e22585 | 2343 | if (c->partial) { |
f6e7def7 CL |
2344 | page = c->page = c->partial; |
2345 | c->partial = page->next; | |
49e22585 CL |
2346 | stat(s, CPU_PARTIAL_ALLOC); |
2347 | c->freelist = NULL; | |
2348 | goto redo; | |
81819f0f CL |
2349 | } |
2350 | ||
188fd063 | 2351 | freelist = new_slab_objects(s, gfpflags, node, &c); |
01ad8a7b | 2352 | |
f4697436 | 2353 | if (unlikely(!freelist)) { |
9a02d699 | 2354 | slab_out_of_memory(s, gfpflags, node); |
f4697436 CL |
2355 | local_irq_restore(flags); |
2356 | return NULL; | |
81819f0f | 2357 | } |
2cfb7455 | 2358 | |
f6e7def7 | 2359 | page = c->page; |
5091b74a | 2360 | if (likely(!kmem_cache_debug(s) && pfmemalloc_match(page, gfpflags))) |
4b6f0750 | 2361 | goto load_freelist; |
2cfb7455 | 2362 | |
497b66f2 | 2363 | /* Only entered in the debug case */ |
d0e0ac97 CG |
2364 | if (kmem_cache_debug(s) && |
2365 | !alloc_debug_processing(s, page, freelist, addr)) | |
497b66f2 | 2366 | goto new_slab; /* Slab failed checks. Next slab needed */ |
894b8788 | 2367 | |
f6e7def7 | 2368 | deactivate_slab(s, page, get_freepointer(s, freelist)); |
c17dda40 CL |
2369 | c->page = NULL; |
2370 | c->freelist = NULL; | |
a71ae47a | 2371 | local_irq_restore(flags); |
6faa6833 | 2372 | return freelist; |
894b8788 CL |
2373 | } |
2374 | ||
2375 | /* | |
2376 | * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc) | |
2377 | * have the fastpath folded into their functions. So no function call | |
2378 | * overhead for requests that can be satisfied on the fastpath. | |
2379 | * | |
2380 | * The fastpath works by first checking if the lockless freelist can be used. | |
2381 | * If not then __slab_alloc is called for slow processing. | |
2382 | * | |
2383 | * Otherwise we can simply pick the next object from the lockless free list. | |
2384 | */ | |
2b847c3c | 2385 | static __always_inline void *slab_alloc_node(struct kmem_cache *s, |
ce71e27c | 2386 | gfp_t gfpflags, int node, unsigned long addr) |
894b8788 | 2387 | { |
894b8788 | 2388 | void **object; |
dfb4f096 | 2389 | struct kmem_cache_cpu *c; |
57d437d2 | 2390 | struct page *page; |
8a5ec0ba | 2391 | unsigned long tid; |
1f84260c | 2392 | |
c016b0bd | 2393 | if (slab_pre_alloc_hook(s, gfpflags)) |
773ff60e | 2394 | return NULL; |
1f84260c | 2395 | |
d79923fa | 2396 | s = memcg_kmem_get_cache(s, gfpflags); |
8a5ec0ba | 2397 | redo: |
8a5ec0ba CL |
2398 | /* |
2399 | * Must read kmem_cache cpu data via this cpu ptr. Preemption is | |
2400 | * enabled. We may switch back and forth between cpus while | |
2401 | * reading from one cpu area. That does not matter as long | |
2402 | * as we end up on the original cpu again when doing the cmpxchg. | |
7cccd80b CL |
2403 | * |
2404 | * Preemption is disabled for the retrieval of the tid because that | |
2405 | * must occur from the current processor. We cannot allow rescheduling | |
2406 | * on a different processor between the determination of the pointer | |
2407 | * and the retrieval of the tid. | |
8a5ec0ba | 2408 | */ |
7cccd80b | 2409 | preempt_disable(); |
7c8e0181 | 2410 | c = this_cpu_ptr(s->cpu_slab); |
8a5ec0ba | 2411 | |
8a5ec0ba CL |
2412 | /* |
2413 | * The transaction ids are globally unique per cpu and per operation on | |
2414 | * a per cpu queue. Thus they can be guarantee that the cmpxchg_double | |
2415 | * occurs on the right processor and that there was no operation on the | |
2416 | * linked list in between. | |
2417 | */ | |
2418 | tid = c->tid; | |
7cccd80b | 2419 | preempt_enable(); |
8a5ec0ba | 2420 | |
9dfc6e68 | 2421 | object = c->freelist; |
57d437d2 | 2422 | page = c->page; |
8eae1492 | 2423 | if (unlikely(!object || !node_match(page, node))) { |
dfb4f096 | 2424 | object = __slab_alloc(s, gfpflags, node, addr, c); |
8eae1492 DH |
2425 | stat(s, ALLOC_SLOWPATH); |
2426 | } else { | |
0ad9500e ED |
2427 | void *next_object = get_freepointer_safe(s, object); |
2428 | ||
8a5ec0ba | 2429 | /* |
25985edc | 2430 | * The cmpxchg will only match if there was no additional |
8a5ec0ba CL |
2431 | * operation and if we are on the right processor. |
2432 | * | |
d0e0ac97 CG |
2433 | * The cmpxchg does the following atomically (without lock |
2434 | * semantics!) | |
8a5ec0ba CL |
2435 | * 1. Relocate first pointer to the current per cpu area. |
2436 | * 2. Verify that tid and freelist have not been changed | |
2437 | * 3. If they were not changed replace tid and freelist | |
2438 | * | |
d0e0ac97 CG |
2439 | * Since this is without lock semantics the protection is only |
2440 | * against code executing on this cpu *not* from access by | |
2441 | * other cpus. | |
8a5ec0ba | 2442 | */ |
933393f5 | 2443 | if (unlikely(!this_cpu_cmpxchg_double( |
8a5ec0ba CL |
2444 | s->cpu_slab->freelist, s->cpu_slab->tid, |
2445 | object, tid, | |
0ad9500e | 2446 | next_object, next_tid(tid)))) { |
8a5ec0ba CL |
2447 | |
2448 | note_cmpxchg_failure("slab_alloc", s, tid); | |
2449 | goto redo; | |
2450 | } | |
0ad9500e | 2451 | prefetch_freepointer(s, next_object); |
84e554e6 | 2452 | stat(s, ALLOC_FASTPATH); |
894b8788 | 2453 | } |
8a5ec0ba | 2454 | |
74e2134f | 2455 | if (unlikely(gfpflags & __GFP_ZERO) && object) |
3b0efdfa | 2456 | memset(object, 0, s->object_size); |
d07dbea4 | 2457 | |
c016b0bd | 2458 | slab_post_alloc_hook(s, gfpflags, object); |
5a896d9e | 2459 | |
894b8788 | 2460 | return object; |
81819f0f CL |
2461 | } |
2462 | ||
2b847c3c EG |
2463 | static __always_inline void *slab_alloc(struct kmem_cache *s, |
2464 | gfp_t gfpflags, unsigned long addr) | |
2465 | { | |
2466 | return slab_alloc_node(s, gfpflags, NUMA_NO_NODE, addr); | |
2467 | } | |
2468 | ||
81819f0f CL |
2469 | void *kmem_cache_alloc(struct kmem_cache *s, gfp_t gfpflags) |
2470 | { | |
2b847c3c | 2471 | void *ret = slab_alloc(s, gfpflags, _RET_IP_); |
5b882be4 | 2472 | |
d0e0ac97 CG |
2473 | trace_kmem_cache_alloc(_RET_IP_, ret, s->object_size, |
2474 | s->size, gfpflags); | |
5b882be4 EGM |
2475 | |
2476 | return ret; | |
81819f0f CL |
2477 | } |
2478 | EXPORT_SYMBOL(kmem_cache_alloc); | |
2479 | ||
0f24f128 | 2480 | #ifdef CONFIG_TRACING |
4a92379b RK |
2481 | void *kmem_cache_alloc_trace(struct kmem_cache *s, gfp_t gfpflags, size_t size) |
2482 | { | |
2b847c3c | 2483 | void *ret = slab_alloc(s, gfpflags, _RET_IP_); |
4a92379b RK |
2484 | trace_kmalloc(_RET_IP_, ret, size, s->size, gfpflags); |
2485 | return ret; | |
2486 | } | |
2487 | EXPORT_SYMBOL(kmem_cache_alloc_trace); | |
5b882be4 EGM |
2488 | #endif |
2489 | ||
81819f0f CL |
2490 | #ifdef CONFIG_NUMA |
2491 | void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t gfpflags, int node) | |
2492 | { | |
2b847c3c | 2493 | void *ret = slab_alloc_node(s, gfpflags, node, _RET_IP_); |
5b882be4 | 2494 | |
ca2b84cb | 2495 | trace_kmem_cache_alloc_node(_RET_IP_, ret, |
3b0efdfa | 2496 | s->object_size, s->size, gfpflags, node); |
5b882be4 EGM |
2497 | |
2498 | return ret; | |
81819f0f CL |
2499 | } |
2500 | EXPORT_SYMBOL(kmem_cache_alloc_node); | |
81819f0f | 2501 | |
0f24f128 | 2502 | #ifdef CONFIG_TRACING |
4a92379b | 2503 | void *kmem_cache_alloc_node_trace(struct kmem_cache *s, |
5b882be4 | 2504 | gfp_t gfpflags, |
4a92379b | 2505 | int node, size_t size) |
5b882be4 | 2506 | { |
2b847c3c | 2507 | void *ret = slab_alloc_node(s, gfpflags, node, _RET_IP_); |
4a92379b RK |
2508 | |
2509 | trace_kmalloc_node(_RET_IP_, ret, | |
2510 | size, s->size, gfpflags, node); | |
2511 | return ret; | |
5b882be4 | 2512 | } |
4a92379b | 2513 | EXPORT_SYMBOL(kmem_cache_alloc_node_trace); |
5b882be4 | 2514 | #endif |
5d1f57e4 | 2515 | #endif |
5b882be4 | 2516 | |
81819f0f | 2517 | /* |
894b8788 CL |
2518 | * Slow patch handling. This may still be called frequently since objects |
2519 | * have a longer lifetime than the cpu slabs in most processing loads. | |
81819f0f | 2520 | * |
894b8788 CL |
2521 | * So we still attempt to reduce cache line usage. Just take the slab |
2522 | * lock and free the item. If there is no additional partial page | |
2523 | * handling required then we can return immediately. | |
81819f0f | 2524 | */ |
894b8788 | 2525 | static void __slab_free(struct kmem_cache *s, struct page *page, |
ff12059e | 2526 | void *x, unsigned long addr) |
81819f0f CL |
2527 | { |
2528 | void *prior; | |
2529 | void **object = (void *)x; | |
2cfb7455 | 2530 | int was_frozen; |
2cfb7455 CL |
2531 | struct page new; |
2532 | unsigned long counters; | |
2533 | struct kmem_cache_node *n = NULL; | |
61728d1e | 2534 | unsigned long uninitialized_var(flags); |
81819f0f | 2535 | |
8a5ec0ba | 2536 | stat(s, FREE_SLOWPATH); |
81819f0f | 2537 | |
19c7ff9e CL |
2538 | if (kmem_cache_debug(s) && |
2539 | !(n = free_debug_processing(s, page, x, addr, &flags))) | |
80f08c19 | 2540 | return; |
6446faa2 | 2541 | |
2cfb7455 | 2542 | do { |
837d678d JK |
2543 | if (unlikely(n)) { |
2544 | spin_unlock_irqrestore(&n->list_lock, flags); | |
2545 | n = NULL; | |
2546 | } | |
2cfb7455 CL |
2547 | prior = page->freelist; |
2548 | counters = page->counters; | |
2549 | set_freepointer(s, object, prior); | |
2550 | new.counters = counters; | |
2551 | was_frozen = new.frozen; | |
2552 | new.inuse--; | |
837d678d | 2553 | if ((!new.inuse || !prior) && !was_frozen) { |
49e22585 | 2554 | |
c65c1877 | 2555 | if (kmem_cache_has_cpu_partial(s) && !prior) { |
49e22585 CL |
2556 | |
2557 | /* | |
d0e0ac97 CG |
2558 | * Slab was on no list before and will be |
2559 | * partially empty | |
2560 | * We can defer the list move and instead | |
2561 | * freeze it. | |
49e22585 CL |
2562 | */ |
2563 | new.frozen = 1; | |
2564 | ||
c65c1877 | 2565 | } else { /* Needs to be taken off a list */ |
49e22585 CL |
2566 | |
2567 | n = get_node(s, page_to_nid(page)); | |
2568 | /* | |
2569 | * Speculatively acquire the list_lock. | |
2570 | * If the cmpxchg does not succeed then we may | |
2571 | * drop the list_lock without any processing. | |
2572 | * | |
2573 | * Otherwise the list_lock will synchronize with | |
2574 | * other processors updating the list of slabs. | |
2575 | */ | |
2576 | spin_lock_irqsave(&n->list_lock, flags); | |
2577 | ||
2578 | } | |
2cfb7455 | 2579 | } |
81819f0f | 2580 | |
2cfb7455 CL |
2581 | } while (!cmpxchg_double_slab(s, page, |
2582 | prior, counters, | |
2583 | object, new.counters, | |
2584 | "__slab_free")); | |
81819f0f | 2585 | |
2cfb7455 | 2586 | if (likely(!n)) { |
49e22585 CL |
2587 | |
2588 | /* | |
2589 | * If we just froze the page then put it onto the | |
2590 | * per cpu partial list. | |
2591 | */ | |
8028dcea | 2592 | if (new.frozen && !was_frozen) { |
49e22585 | 2593 | put_cpu_partial(s, page, 1); |
8028dcea AS |
2594 | stat(s, CPU_PARTIAL_FREE); |
2595 | } | |
49e22585 | 2596 | /* |
2cfb7455 CL |
2597 | * The list lock was not taken therefore no list |
2598 | * activity can be necessary. | |
2599 | */ | |
2600 | if (was_frozen) | |
2601 | stat(s, FREE_FROZEN); | |
80f08c19 | 2602 | return; |
2cfb7455 | 2603 | } |
81819f0f | 2604 | |
8a5b20ae | 2605 | if (unlikely(!new.inuse && n->nr_partial >= s->min_partial)) |
837d678d JK |
2606 | goto slab_empty; |
2607 | ||
81819f0f | 2608 | /* |
837d678d JK |
2609 | * Objects left in the slab. If it was not on the partial list before |
2610 | * then add it. | |
81819f0f | 2611 | */ |
345c905d JK |
2612 | if (!kmem_cache_has_cpu_partial(s) && unlikely(!prior)) { |
2613 | if (kmem_cache_debug(s)) | |
c65c1877 | 2614 | remove_full(s, n, page); |
837d678d JK |
2615 | add_partial(n, page, DEACTIVATE_TO_TAIL); |
2616 | stat(s, FREE_ADD_PARTIAL); | |
8ff12cfc | 2617 | } |
80f08c19 | 2618 | spin_unlock_irqrestore(&n->list_lock, flags); |
81819f0f CL |
2619 | return; |
2620 | ||
2621 | slab_empty: | |
a973e9dd | 2622 | if (prior) { |
81819f0f | 2623 | /* |
6fbabb20 | 2624 | * Slab on the partial list. |
81819f0f | 2625 | */ |
5cc6eee8 | 2626 | remove_partial(n, page); |
84e554e6 | 2627 | stat(s, FREE_REMOVE_PARTIAL); |
c65c1877 | 2628 | } else { |
6fbabb20 | 2629 | /* Slab must be on the full list */ |
c65c1877 PZ |
2630 | remove_full(s, n, page); |
2631 | } | |
2cfb7455 | 2632 | |
80f08c19 | 2633 | spin_unlock_irqrestore(&n->list_lock, flags); |
84e554e6 | 2634 | stat(s, FREE_SLAB); |
81819f0f | 2635 | discard_slab(s, page); |
81819f0f CL |
2636 | } |
2637 | ||
894b8788 CL |
2638 | /* |
2639 | * Fastpath with forced inlining to produce a kfree and kmem_cache_free that | |
2640 | * can perform fastpath freeing without additional function calls. | |
2641 | * | |
2642 | * The fastpath is only possible if we are freeing to the current cpu slab | |
2643 | * of this processor. This typically the case if we have just allocated | |
2644 | * the item before. | |
2645 | * | |
2646 | * If fastpath is not possible then fall back to __slab_free where we deal | |
2647 | * with all sorts of special processing. | |
2648 | */ | |
06428780 | 2649 | static __always_inline void slab_free(struct kmem_cache *s, |
ce71e27c | 2650 | struct page *page, void *x, unsigned long addr) |
894b8788 CL |
2651 | { |
2652 | void **object = (void *)x; | |
dfb4f096 | 2653 | struct kmem_cache_cpu *c; |
8a5ec0ba | 2654 | unsigned long tid; |
1f84260c | 2655 | |
c016b0bd CL |
2656 | slab_free_hook(s, x); |
2657 | ||
8a5ec0ba CL |
2658 | redo: |
2659 | /* | |
2660 | * Determine the currently cpus per cpu slab. | |
2661 | * The cpu may change afterward. However that does not matter since | |
2662 | * data is retrieved via this pointer. If we are on the same cpu | |
2663 | * during the cmpxchg then the free will succedd. | |
2664 | */ | |
7cccd80b | 2665 | preempt_disable(); |
7c8e0181 | 2666 | c = this_cpu_ptr(s->cpu_slab); |
c016b0bd | 2667 | |
8a5ec0ba | 2668 | tid = c->tid; |
7cccd80b | 2669 | preempt_enable(); |
c016b0bd | 2670 | |
442b06bc | 2671 | if (likely(page == c->page)) { |
ff12059e | 2672 | set_freepointer(s, object, c->freelist); |
8a5ec0ba | 2673 | |
933393f5 | 2674 | if (unlikely(!this_cpu_cmpxchg_double( |
8a5ec0ba CL |
2675 | s->cpu_slab->freelist, s->cpu_slab->tid, |
2676 | c->freelist, tid, | |
2677 | object, next_tid(tid)))) { | |
2678 | ||
2679 | note_cmpxchg_failure("slab_free", s, tid); | |
2680 | goto redo; | |
2681 | } | |
84e554e6 | 2682 | stat(s, FREE_FASTPATH); |
894b8788 | 2683 | } else |
ff12059e | 2684 | __slab_free(s, page, x, addr); |
894b8788 | 2685 | |
894b8788 CL |
2686 | } |
2687 | ||
81819f0f CL |
2688 | void kmem_cache_free(struct kmem_cache *s, void *x) |
2689 | { | |
b9ce5ef4 GC |
2690 | s = cache_from_obj(s, x); |
2691 | if (!s) | |
79576102 | 2692 | return; |
b9ce5ef4 | 2693 | slab_free(s, virt_to_head_page(x), x, _RET_IP_); |
ca2b84cb | 2694 | trace_kmem_cache_free(_RET_IP_, x); |
81819f0f CL |
2695 | } |
2696 | EXPORT_SYMBOL(kmem_cache_free); | |
2697 | ||
81819f0f | 2698 | /* |
672bba3a CL |
2699 | * Object placement in a slab is made very easy because we always start at |
2700 | * offset 0. If we tune the size of the object to the alignment then we can | |
2701 | * get the required alignment by putting one properly sized object after | |
2702 | * another. | |
81819f0f CL |
2703 | * |
2704 | * Notice that the allocation order determines the sizes of the per cpu | |
2705 | * caches. Each processor has always one slab available for allocations. | |
2706 | * Increasing the allocation order reduces the number of times that slabs | |
672bba3a | 2707 | * must be moved on and off the partial lists and is therefore a factor in |
81819f0f | 2708 | * locking overhead. |
81819f0f CL |
2709 | */ |
2710 | ||
2711 | /* | |
2712 | * Mininum / Maximum order of slab pages. This influences locking overhead | |
2713 | * and slab fragmentation. A higher order reduces the number of partial slabs | |
2714 | * and increases the number of allocations possible without having to | |
2715 | * take the list_lock. | |
2716 | */ | |
2717 | static int slub_min_order; | |
114e9e89 | 2718 | static int slub_max_order = PAGE_ALLOC_COSTLY_ORDER; |
9b2cd506 | 2719 | static int slub_min_objects; |
81819f0f CL |
2720 | |
2721 | /* | |
2722 | * Merge control. If this is set then no merging of slab caches will occur. | |
672bba3a | 2723 | * (Could be removed. This was introduced to pacify the merge skeptics.) |
81819f0f CL |
2724 | */ |
2725 | static int slub_nomerge; | |
2726 | ||
81819f0f CL |
2727 | /* |
2728 | * Calculate the order of allocation given an slab object size. | |
2729 | * | |
672bba3a CL |
2730 | * The order of allocation has significant impact on performance and other |
2731 | * system components. Generally order 0 allocations should be preferred since | |
2732 | * order 0 does not cause fragmentation in the page allocator. Larger objects | |
2733 | * be problematic to put into order 0 slabs because there may be too much | |
c124f5b5 | 2734 | * unused space left. We go to a higher order if more than 1/16th of the slab |
672bba3a CL |
2735 | * would be wasted. |
2736 | * | |
2737 | * In order to reach satisfactory performance we must ensure that a minimum | |
2738 | * number of objects is in one slab. Otherwise we may generate too much | |
2739 | * activity on the partial lists which requires taking the list_lock. This is | |
2740 | * less a concern for large slabs though which are rarely used. | |
81819f0f | 2741 | * |
672bba3a CL |
2742 | * slub_max_order specifies the order where we begin to stop considering the |
2743 | * number of objects in a slab as critical. If we reach slub_max_order then | |
2744 | * we try to keep the page order as low as possible. So we accept more waste | |
2745 | * of space in favor of a small page order. | |
81819f0f | 2746 | * |
672bba3a CL |
2747 | * Higher order allocations also allow the placement of more objects in a |
2748 | * slab and thereby reduce object handling overhead. If the user has | |
2749 | * requested a higher mininum order then we start with that one instead of | |
2750 | * the smallest order which will fit the object. | |
81819f0f | 2751 | */ |
5e6d444e | 2752 | static inline int slab_order(int size, int min_objects, |
ab9a0f19 | 2753 | int max_order, int fract_leftover, int reserved) |
81819f0f CL |
2754 | { |
2755 | int order; | |
2756 | int rem; | |
6300ea75 | 2757 | int min_order = slub_min_order; |
81819f0f | 2758 | |
ab9a0f19 | 2759 | if (order_objects(min_order, size, reserved) > MAX_OBJS_PER_PAGE) |
210b5c06 | 2760 | return get_order(size * MAX_OBJS_PER_PAGE) - 1; |
39b26464 | 2761 | |
6300ea75 | 2762 | for (order = max(min_order, |
5e6d444e CL |
2763 | fls(min_objects * size - 1) - PAGE_SHIFT); |
2764 | order <= max_order; order++) { | |
81819f0f | 2765 | |
5e6d444e | 2766 | unsigned long slab_size = PAGE_SIZE << order; |
81819f0f | 2767 | |
ab9a0f19 | 2768 | if (slab_size < min_objects * size + reserved) |
81819f0f CL |
2769 | continue; |
2770 | ||
ab9a0f19 | 2771 | rem = (slab_size - reserved) % size; |
81819f0f | 2772 | |
5e6d444e | 2773 | if (rem <= slab_size / fract_leftover) |
81819f0f CL |
2774 | break; |
2775 | ||
2776 | } | |
672bba3a | 2777 | |
81819f0f CL |
2778 | return order; |
2779 | } | |
2780 | ||
ab9a0f19 | 2781 | static inline int calculate_order(int size, int reserved) |
5e6d444e CL |
2782 | { |
2783 | int order; | |
2784 | int min_objects; | |
2785 | int fraction; | |
e8120ff1 | 2786 | int max_objects; |
5e6d444e CL |
2787 | |
2788 | /* | |
2789 | * Attempt to find best configuration for a slab. This | |
2790 | * works by first attempting to generate a layout with | |
2791 | * the best configuration and backing off gradually. | |
2792 | * | |
2793 | * First we reduce the acceptable waste in a slab. Then | |
2794 | * we reduce the minimum objects required in a slab. | |
2795 | */ | |
2796 | min_objects = slub_min_objects; | |
9b2cd506 CL |
2797 | if (!min_objects) |
2798 | min_objects = 4 * (fls(nr_cpu_ids) + 1); | |
ab9a0f19 | 2799 | max_objects = order_objects(slub_max_order, size, reserved); |
e8120ff1 ZY |
2800 | min_objects = min(min_objects, max_objects); |
2801 | ||
5e6d444e | 2802 | while (min_objects > 1) { |
c124f5b5 | 2803 | fraction = 16; |
5e6d444e CL |
2804 | while (fraction >= 4) { |
2805 | order = slab_order(size, min_objects, | |
ab9a0f19 | 2806 | slub_max_order, fraction, reserved); |
5e6d444e CL |
2807 | if (order <= slub_max_order) |
2808 | return order; | |
2809 | fraction /= 2; | |
2810 | } | |
5086c389 | 2811 | min_objects--; |
5e6d444e CL |
2812 | } |
2813 | ||
2814 | /* | |
2815 | * We were unable to place multiple objects in a slab. Now | |
2816 | * lets see if we can place a single object there. | |
2817 | */ | |
ab9a0f19 | 2818 | order = slab_order(size, 1, slub_max_order, 1, reserved); |
5e6d444e CL |
2819 | if (order <= slub_max_order) |
2820 | return order; | |
2821 | ||
2822 | /* | |
2823 | * Doh this slab cannot be placed using slub_max_order. | |
2824 | */ | |
ab9a0f19 | 2825 | order = slab_order(size, 1, MAX_ORDER, 1, reserved); |
818cf590 | 2826 | if (order < MAX_ORDER) |
5e6d444e CL |
2827 | return order; |
2828 | return -ENOSYS; | |
2829 | } | |
2830 | ||
5595cffc | 2831 | static void |
4053497d | 2832 | init_kmem_cache_node(struct kmem_cache_node *n) |
81819f0f CL |
2833 | { |
2834 | n->nr_partial = 0; | |
81819f0f CL |
2835 | spin_lock_init(&n->list_lock); |
2836 | INIT_LIST_HEAD(&n->partial); | |
8ab1372f | 2837 | #ifdef CONFIG_SLUB_DEBUG |
0f389ec6 | 2838 | atomic_long_set(&n->nr_slabs, 0); |
02b71b70 | 2839 | atomic_long_set(&n->total_objects, 0); |
643b1138 | 2840 | INIT_LIST_HEAD(&n->full); |
8ab1372f | 2841 | #endif |
81819f0f CL |
2842 | } |
2843 | ||
55136592 | 2844 | static inline int alloc_kmem_cache_cpus(struct kmem_cache *s) |
4c93c355 | 2845 | { |
6c182dc0 | 2846 | BUILD_BUG_ON(PERCPU_DYNAMIC_EARLY_SIZE < |
95a05b42 | 2847 | KMALLOC_SHIFT_HIGH * sizeof(struct kmem_cache_cpu)); |
4c93c355 | 2848 | |
8a5ec0ba | 2849 | /* |
d4d84fef CM |
2850 | * Must align to double word boundary for the double cmpxchg |
2851 | * instructions to work; see __pcpu_double_call_return_bool(). | |
8a5ec0ba | 2852 | */ |
d4d84fef CM |
2853 | s->cpu_slab = __alloc_percpu(sizeof(struct kmem_cache_cpu), |
2854 | 2 * sizeof(void *)); | |
8a5ec0ba CL |
2855 | |
2856 | if (!s->cpu_slab) | |
2857 | return 0; | |
2858 | ||
2859 | init_kmem_cache_cpus(s); | |
4c93c355 | 2860 | |
8a5ec0ba | 2861 | return 1; |
4c93c355 | 2862 | } |
4c93c355 | 2863 | |
51df1142 CL |
2864 | static struct kmem_cache *kmem_cache_node; |
2865 | ||
81819f0f CL |
2866 | /* |
2867 | * No kmalloc_node yet so do it by hand. We know that this is the first | |
2868 | * slab on the node for this slabcache. There are no concurrent accesses | |
2869 | * possible. | |
2870 | * | |
721ae22a ZYW |
2871 | * Note that this function only works on the kmem_cache_node |
2872 | * when allocating for the kmem_cache_node. This is used for bootstrapping | |
4c93c355 | 2873 | * memory on a fresh node that has no slab structures yet. |
81819f0f | 2874 | */ |
55136592 | 2875 | static void early_kmem_cache_node_alloc(int node) |
81819f0f CL |
2876 | { |
2877 | struct page *page; | |
2878 | struct kmem_cache_node *n; | |
2879 | ||
51df1142 | 2880 | BUG_ON(kmem_cache_node->size < sizeof(struct kmem_cache_node)); |
81819f0f | 2881 | |
51df1142 | 2882 | page = new_slab(kmem_cache_node, GFP_NOWAIT, node); |
81819f0f CL |
2883 | |
2884 | BUG_ON(!page); | |
a2f92ee7 | 2885 | if (page_to_nid(page) != node) { |
f9f58285 FF |
2886 | pr_err("SLUB: Unable to allocate memory from node %d\n", node); |
2887 | pr_err("SLUB: Allocating a useless per node structure in order to be able to continue\n"); | |
a2f92ee7 CL |
2888 | } |
2889 | ||
81819f0f CL |
2890 | n = page->freelist; |
2891 | BUG_ON(!n); | |
51df1142 | 2892 | page->freelist = get_freepointer(kmem_cache_node, n); |
e6e82ea1 | 2893 | page->inuse = 1; |
8cb0a506 | 2894 | page->frozen = 0; |
51df1142 | 2895 | kmem_cache_node->node[node] = n; |
8ab1372f | 2896 | #ifdef CONFIG_SLUB_DEBUG |
f7cb1933 | 2897 | init_object(kmem_cache_node, n, SLUB_RED_ACTIVE); |
51df1142 | 2898 | init_tracking(kmem_cache_node, n); |
8ab1372f | 2899 | #endif |
4053497d | 2900 | init_kmem_cache_node(n); |
51df1142 | 2901 | inc_slabs_node(kmem_cache_node, node, page->objects); |
6446faa2 | 2902 | |
67b6c900 | 2903 | /* |
1e4dd946 SR |
2904 | * No locks need to be taken here as it has just been |
2905 | * initialized and there is no concurrent access. | |
67b6c900 | 2906 | */ |
1e4dd946 | 2907 | __add_partial(n, page, DEACTIVATE_TO_HEAD); |
81819f0f CL |
2908 | } |
2909 | ||
2910 | static void free_kmem_cache_nodes(struct kmem_cache *s) | |
2911 | { | |
2912 | int node; | |
fa45dc25 | 2913 | struct kmem_cache_node *n; |
81819f0f | 2914 | |
fa45dc25 CL |
2915 | for_each_kmem_cache_node(s, node, n) { |
2916 | kmem_cache_free(kmem_cache_node, n); | |
81819f0f CL |
2917 | s->node[node] = NULL; |
2918 | } | |
2919 | } | |
2920 | ||
55136592 | 2921 | static int init_kmem_cache_nodes(struct kmem_cache *s) |
81819f0f CL |
2922 | { |
2923 | int node; | |
81819f0f | 2924 | |
f64dc58c | 2925 | for_each_node_state(node, N_NORMAL_MEMORY) { |
81819f0f CL |
2926 | struct kmem_cache_node *n; |
2927 | ||
73367bd8 | 2928 | if (slab_state == DOWN) { |
55136592 | 2929 | early_kmem_cache_node_alloc(node); |
73367bd8 AD |
2930 | continue; |
2931 | } | |
51df1142 | 2932 | n = kmem_cache_alloc_node(kmem_cache_node, |
55136592 | 2933 | GFP_KERNEL, node); |
81819f0f | 2934 | |
73367bd8 AD |
2935 | if (!n) { |
2936 | free_kmem_cache_nodes(s); | |
2937 | return 0; | |
81819f0f | 2938 | } |
73367bd8 | 2939 | |
81819f0f | 2940 | s->node[node] = n; |
4053497d | 2941 | init_kmem_cache_node(n); |
81819f0f CL |
2942 | } |
2943 | return 1; | |
2944 | } | |
81819f0f | 2945 | |
c0bdb232 | 2946 | static void set_min_partial(struct kmem_cache *s, unsigned long min) |
3b89d7d8 DR |
2947 | { |
2948 | if (min < MIN_PARTIAL) | |
2949 | min = MIN_PARTIAL; | |
2950 | else if (min > MAX_PARTIAL) | |
2951 | min = MAX_PARTIAL; | |
2952 | s->min_partial = min; | |
2953 | } | |
2954 | ||
81819f0f CL |
2955 | /* |
2956 | * calculate_sizes() determines the order and the distribution of data within | |
2957 | * a slab object. | |
2958 | */ | |
06b285dc | 2959 | static int calculate_sizes(struct kmem_cache *s, int forced_order) |
81819f0f CL |
2960 | { |
2961 | unsigned long flags = s->flags; | |
3b0efdfa | 2962 | unsigned long size = s->object_size; |
834f3d11 | 2963 | int order; |
81819f0f | 2964 | |
d8b42bf5 CL |
2965 | /* |
2966 | * Round up object size to the next word boundary. We can only | |
2967 | * place the free pointer at word boundaries and this determines | |
2968 | * the possible location of the free pointer. | |
2969 | */ | |
2970 | size = ALIGN(size, sizeof(void *)); | |
2971 | ||
2972 | #ifdef CONFIG_SLUB_DEBUG | |
81819f0f CL |
2973 | /* |
2974 | * Determine if we can poison the object itself. If the user of | |
2975 | * the slab may touch the object after free or before allocation | |
2976 | * then we should never poison the object itself. | |
2977 | */ | |
2978 | if ((flags & SLAB_POISON) && !(flags & SLAB_DESTROY_BY_RCU) && | |
c59def9f | 2979 | !s->ctor) |
81819f0f CL |
2980 | s->flags |= __OBJECT_POISON; |
2981 | else | |
2982 | s->flags &= ~__OBJECT_POISON; | |
2983 | ||
81819f0f CL |
2984 | |
2985 | /* | |
672bba3a | 2986 | * If we are Redzoning then check if there is some space between the |
81819f0f | 2987 | * end of the object and the free pointer. If not then add an |
672bba3a | 2988 | * additional word to have some bytes to store Redzone information. |
81819f0f | 2989 | */ |
3b0efdfa | 2990 | if ((flags & SLAB_RED_ZONE) && size == s->object_size) |
81819f0f | 2991 | size += sizeof(void *); |
41ecc55b | 2992 | #endif |
81819f0f CL |
2993 | |
2994 | /* | |
672bba3a CL |
2995 | * With that we have determined the number of bytes in actual use |
2996 | * by the object. This is the potential offset to the free pointer. | |
81819f0f CL |
2997 | */ |
2998 | s->inuse = size; | |
2999 | ||
3000 | if (((flags & (SLAB_DESTROY_BY_RCU | SLAB_POISON)) || | |
c59def9f | 3001 | s->ctor)) { |
81819f0f CL |
3002 | /* |
3003 | * Relocate free pointer after the object if it is not | |
3004 | * permitted to overwrite the first word of the object on | |
3005 | * kmem_cache_free. | |
3006 | * | |
3007 | * This is the case if we do RCU, have a constructor or | |
3008 | * destructor or are poisoning the objects. | |
3009 | */ | |
3010 | s->offset = size; | |
3011 | size += sizeof(void *); | |
3012 | } | |
3013 | ||
c12b3c62 | 3014 | #ifdef CONFIG_SLUB_DEBUG |
81819f0f CL |
3015 | if (flags & SLAB_STORE_USER) |
3016 | /* | |
3017 | * Need to store information about allocs and frees after | |
3018 | * the object. | |
3019 | */ | |
3020 | size += 2 * sizeof(struct track); | |
3021 | ||
be7b3fbc | 3022 | if (flags & SLAB_RED_ZONE) |
81819f0f CL |
3023 | /* |
3024 | * Add some empty padding so that we can catch | |
3025 | * overwrites from earlier objects rather than let | |
3026 | * tracking information or the free pointer be | |
0211a9c8 | 3027 | * corrupted if a user writes before the start |
81819f0f CL |
3028 | * of the object. |
3029 | */ | |
3030 | size += sizeof(void *); | |
41ecc55b | 3031 | #endif |
672bba3a | 3032 | |
81819f0f CL |
3033 | /* |
3034 | * SLUB stores one object immediately after another beginning from | |
3035 | * offset 0. In order to align the objects we have to simply size | |
3036 | * each object to conform to the alignment. | |
3037 | */ | |
45906855 | 3038 | size = ALIGN(size, s->align); |
81819f0f | 3039 | s->size = size; |
06b285dc CL |
3040 | if (forced_order >= 0) |
3041 | order = forced_order; | |
3042 | else | |
ab9a0f19 | 3043 | order = calculate_order(size, s->reserved); |
81819f0f | 3044 | |
834f3d11 | 3045 | if (order < 0) |
81819f0f CL |
3046 | return 0; |
3047 | ||
b7a49f0d | 3048 | s->allocflags = 0; |
834f3d11 | 3049 | if (order) |
b7a49f0d CL |
3050 | s->allocflags |= __GFP_COMP; |
3051 | ||
3052 | if (s->flags & SLAB_CACHE_DMA) | |
2c59dd65 | 3053 | s->allocflags |= GFP_DMA; |
b7a49f0d CL |
3054 | |
3055 | if (s->flags & SLAB_RECLAIM_ACCOUNT) | |
3056 | s->allocflags |= __GFP_RECLAIMABLE; | |
3057 | ||
81819f0f CL |
3058 | /* |
3059 | * Determine the number of objects per slab | |
3060 | */ | |
ab9a0f19 LJ |
3061 | s->oo = oo_make(order, size, s->reserved); |
3062 | s->min = oo_make(get_order(size), size, s->reserved); | |
205ab99d CL |
3063 | if (oo_objects(s->oo) > oo_objects(s->max)) |
3064 | s->max = s->oo; | |
81819f0f | 3065 | |
834f3d11 | 3066 | return !!oo_objects(s->oo); |
81819f0f CL |
3067 | } |
3068 | ||
8a13a4cc | 3069 | static int kmem_cache_open(struct kmem_cache *s, unsigned long flags) |
81819f0f | 3070 | { |
8a13a4cc | 3071 | s->flags = kmem_cache_flags(s->size, flags, s->name, s->ctor); |
ab9a0f19 | 3072 | s->reserved = 0; |
81819f0f | 3073 | |
da9a638c LJ |
3074 | if (need_reserve_slab_rcu && (s->flags & SLAB_DESTROY_BY_RCU)) |
3075 | s->reserved = sizeof(struct rcu_head); | |
81819f0f | 3076 | |
06b285dc | 3077 | if (!calculate_sizes(s, -1)) |
81819f0f | 3078 | goto error; |
3de47213 DR |
3079 | if (disable_higher_order_debug) { |
3080 | /* | |
3081 | * Disable debugging flags that store metadata if the min slab | |
3082 | * order increased. | |
3083 | */ | |
3b0efdfa | 3084 | if (get_order(s->size) > get_order(s->object_size)) { |
3de47213 DR |
3085 | s->flags &= ~DEBUG_METADATA_FLAGS; |
3086 | s->offset = 0; | |
3087 | if (!calculate_sizes(s, -1)) | |
3088 | goto error; | |
3089 | } | |
3090 | } | |
81819f0f | 3091 | |
2565409f HC |
3092 | #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \ |
3093 | defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE) | |
b789ef51 CL |
3094 | if (system_has_cmpxchg_double() && (s->flags & SLAB_DEBUG_FLAGS) == 0) |
3095 | /* Enable fast mode */ | |
3096 | s->flags |= __CMPXCHG_DOUBLE; | |
3097 | #endif | |
3098 | ||
3b89d7d8 DR |
3099 | /* |
3100 | * The larger the object size is, the more pages we want on the partial | |
3101 | * list to avoid pounding the page allocator excessively. | |
3102 | */ | |
49e22585 CL |
3103 | set_min_partial(s, ilog2(s->size) / 2); |
3104 | ||
3105 | /* | |
3106 | * cpu_partial determined the maximum number of objects kept in the | |
3107 | * per cpu partial lists of a processor. | |
3108 | * | |
3109 | * Per cpu partial lists mainly contain slabs that just have one | |
3110 | * object freed. If they are used for allocation then they can be | |
3111 | * filled up again with minimal effort. The slab will never hit the | |
3112 | * per node partial lists and therefore no locking will be required. | |
3113 | * | |
3114 | * This setting also determines | |
3115 | * | |
3116 | * A) The number of objects from per cpu partial slabs dumped to the | |
3117 | * per node list when we reach the limit. | |
9f264904 | 3118 | * B) The number of objects in cpu partial slabs to extract from the |
d0e0ac97 CG |
3119 | * per node list when we run out of per cpu objects. We only fetch |
3120 | * 50% to keep some capacity around for frees. | |
49e22585 | 3121 | */ |
345c905d | 3122 | if (!kmem_cache_has_cpu_partial(s)) |
8f1e33da CL |
3123 | s->cpu_partial = 0; |
3124 | else if (s->size >= PAGE_SIZE) | |
49e22585 CL |
3125 | s->cpu_partial = 2; |
3126 | else if (s->size >= 1024) | |
3127 | s->cpu_partial = 6; | |
3128 | else if (s->size >= 256) | |
3129 | s->cpu_partial = 13; | |
3130 | else | |
3131 | s->cpu_partial = 30; | |
3132 | ||
81819f0f | 3133 | #ifdef CONFIG_NUMA |
e2cb96b7 | 3134 | s->remote_node_defrag_ratio = 1000; |
81819f0f | 3135 | #endif |
55136592 | 3136 | if (!init_kmem_cache_nodes(s)) |
dfb4f096 | 3137 | goto error; |
81819f0f | 3138 | |
55136592 | 3139 | if (alloc_kmem_cache_cpus(s)) |
278b1bb1 | 3140 | return 0; |
ff12059e | 3141 | |
4c93c355 | 3142 | free_kmem_cache_nodes(s); |
81819f0f CL |
3143 | error: |
3144 | if (flags & SLAB_PANIC) | |
3145 | panic("Cannot create slab %s size=%lu realsize=%u " | |
3146 | "order=%u offset=%u flags=%lx\n", | |
d0e0ac97 CG |
3147 | s->name, (unsigned long)s->size, s->size, |
3148 | oo_order(s->oo), s->offset, flags); | |
278b1bb1 | 3149 | return -EINVAL; |
81819f0f | 3150 | } |
81819f0f | 3151 | |
33b12c38 CL |
3152 | static void list_slab_objects(struct kmem_cache *s, struct page *page, |
3153 | const char *text) | |
3154 | { | |
3155 | #ifdef CONFIG_SLUB_DEBUG | |
3156 | void *addr = page_address(page); | |
3157 | void *p; | |
a5dd5c11 NK |
3158 | unsigned long *map = kzalloc(BITS_TO_LONGS(page->objects) * |
3159 | sizeof(long), GFP_ATOMIC); | |
bbd7d57b ED |
3160 | if (!map) |
3161 | return; | |
945cf2b6 | 3162 | slab_err(s, page, text, s->name); |
33b12c38 | 3163 | slab_lock(page); |
33b12c38 | 3164 | |
5f80b13a | 3165 | get_map(s, page, map); |
33b12c38 CL |
3166 | for_each_object(p, s, addr, page->objects) { |
3167 | ||
3168 | if (!test_bit(slab_index(p, s, addr), map)) { | |
f9f58285 | 3169 | pr_err("INFO: Object 0x%p @offset=%tu\n", p, p - addr); |
33b12c38 CL |
3170 | print_tracking(s, p); |
3171 | } | |
3172 | } | |
3173 | slab_unlock(page); | |
bbd7d57b | 3174 | kfree(map); |
33b12c38 CL |
3175 | #endif |
3176 | } | |
3177 | ||
81819f0f | 3178 | /* |
599870b1 | 3179 | * Attempt to free all partial slabs on a node. |
69cb8e6b CL |
3180 | * This is called from kmem_cache_close(). We must be the last thread |
3181 | * using the cache and therefore we do not need to lock anymore. | |
81819f0f | 3182 | */ |
599870b1 | 3183 | static void free_partial(struct kmem_cache *s, struct kmem_cache_node *n) |
81819f0f | 3184 | { |
81819f0f CL |
3185 | struct page *page, *h; |
3186 | ||
33b12c38 | 3187 | list_for_each_entry_safe(page, h, &n->partial, lru) { |
81819f0f | 3188 | if (!page->inuse) { |
1e4dd946 | 3189 | __remove_partial(n, page); |
81819f0f | 3190 | discard_slab(s, page); |
33b12c38 CL |
3191 | } else { |
3192 | list_slab_objects(s, page, | |
945cf2b6 | 3193 | "Objects remaining in %s on kmem_cache_close()"); |
599870b1 | 3194 | } |
33b12c38 | 3195 | } |
81819f0f CL |
3196 | } |
3197 | ||
3198 | /* | |
672bba3a | 3199 | * Release all resources used by a slab cache. |
81819f0f | 3200 | */ |
0c710013 | 3201 | static inline int kmem_cache_close(struct kmem_cache *s) |
81819f0f CL |
3202 | { |
3203 | int node; | |
fa45dc25 | 3204 | struct kmem_cache_node *n; |
81819f0f CL |
3205 | |
3206 | flush_all(s); | |
81819f0f | 3207 | /* Attempt to free all objects */ |
fa45dc25 | 3208 | for_each_kmem_cache_node(s, node, n) { |
599870b1 CL |
3209 | free_partial(s, n); |
3210 | if (n->nr_partial || slabs_node(s, node)) | |
81819f0f CL |
3211 | return 1; |
3212 | } | |
945cf2b6 | 3213 | free_percpu(s->cpu_slab); |
81819f0f CL |
3214 | free_kmem_cache_nodes(s); |
3215 | return 0; | |
3216 | } | |
3217 | ||
945cf2b6 | 3218 | int __kmem_cache_shutdown(struct kmem_cache *s) |
81819f0f | 3219 | { |
41a21285 | 3220 | return kmem_cache_close(s); |
81819f0f | 3221 | } |
81819f0f CL |
3222 | |
3223 | /******************************************************************** | |
3224 | * Kmalloc subsystem | |
3225 | *******************************************************************/ | |
3226 | ||
81819f0f CL |
3227 | static int __init setup_slub_min_order(char *str) |
3228 | { | |
06428780 | 3229 | get_option(&str, &slub_min_order); |
81819f0f CL |
3230 | |
3231 | return 1; | |
3232 | } | |
3233 | ||
3234 | __setup("slub_min_order=", setup_slub_min_order); | |
3235 | ||
3236 | static int __init setup_slub_max_order(char *str) | |
3237 | { | |
06428780 | 3238 | get_option(&str, &slub_max_order); |
818cf590 | 3239 | slub_max_order = min(slub_max_order, MAX_ORDER - 1); |
81819f0f CL |
3240 | |
3241 | return 1; | |
3242 | } | |
3243 | ||
3244 | __setup("slub_max_order=", setup_slub_max_order); | |
3245 | ||
3246 | static int __init setup_slub_min_objects(char *str) | |
3247 | { | |
06428780 | 3248 | get_option(&str, &slub_min_objects); |
81819f0f CL |
3249 | |
3250 | return 1; | |
3251 | } | |
3252 | ||
3253 | __setup("slub_min_objects=", setup_slub_min_objects); | |
3254 | ||
3255 | static int __init setup_slub_nomerge(char *str) | |
3256 | { | |
3257 | slub_nomerge = 1; | |
3258 | return 1; | |
3259 | } | |
3260 | ||
3261 | __setup("slub_nomerge", setup_slub_nomerge); | |
3262 | ||
81819f0f CL |
3263 | void *__kmalloc(size_t size, gfp_t flags) |
3264 | { | |
aadb4bc4 | 3265 | struct kmem_cache *s; |
5b882be4 | 3266 | void *ret; |
81819f0f | 3267 | |
95a05b42 | 3268 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) |
eada35ef | 3269 | return kmalloc_large(size, flags); |
aadb4bc4 | 3270 | |
2c59dd65 | 3271 | s = kmalloc_slab(size, flags); |
aadb4bc4 CL |
3272 | |
3273 | if (unlikely(ZERO_OR_NULL_PTR(s))) | |
6cb8f913 CL |
3274 | return s; |
3275 | ||
2b847c3c | 3276 | ret = slab_alloc(s, flags, _RET_IP_); |
5b882be4 | 3277 | |
ca2b84cb | 3278 | trace_kmalloc(_RET_IP_, ret, size, s->size, flags); |
5b882be4 EGM |
3279 | |
3280 | return ret; | |
81819f0f CL |
3281 | } |
3282 | EXPORT_SYMBOL(__kmalloc); | |
3283 | ||
5d1f57e4 | 3284 | #ifdef CONFIG_NUMA |
f619cfe1 CL |
3285 | static void *kmalloc_large_node(size_t size, gfp_t flags, int node) |
3286 | { | |
b1eeab67 | 3287 | struct page *page; |
e4f7c0b4 | 3288 | void *ptr = NULL; |
f619cfe1 | 3289 | |
52383431 VD |
3290 | flags |= __GFP_COMP | __GFP_NOTRACK; |
3291 | page = alloc_kmem_pages_node(node, flags, get_order(size)); | |
f619cfe1 | 3292 | if (page) |
e4f7c0b4 CM |
3293 | ptr = page_address(page); |
3294 | ||
d56791b3 | 3295 | kmalloc_large_node_hook(ptr, size, flags); |
e4f7c0b4 | 3296 | return ptr; |
f619cfe1 CL |
3297 | } |
3298 | ||
81819f0f CL |
3299 | void *__kmalloc_node(size_t size, gfp_t flags, int node) |
3300 | { | |
aadb4bc4 | 3301 | struct kmem_cache *s; |
5b882be4 | 3302 | void *ret; |
81819f0f | 3303 | |
95a05b42 | 3304 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) { |
5b882be4 EGM |
3305 | ret = kmalloc_large_node(size, flags, node); |
3306 | ||
ca2b84cb EGM |
3307 | trace_kmalloc_node(_RET_IP_, ret, |
3308 | size, PAGE_SIZE << get_order(size), | |
3309 | flags, node); | |
5b882be4 EGM |
3310 | |
3311 | return ret; | |
3312 | } | |
aadb4bc4 | 3313 | |
2c59dd65 | 3314 | s = kmalloc_slab(size, flags); |
aadb4bc4 CL |
3315 | |
3316 | if (unlikely(ZERO_OR_NULL_PTR(s))) | |
6cb8f913 CL |
3317 | return s; |
3318 | ||
2b847c3c | 3319 | ret = slab_alloc_node(s, flags, node, _RET_IP_); |
5b882be4 | 3320 | |
ca2b84cb | 3321 | trace_kmalloc_node(_RET_IP_, ret, size, s->size, flags, node); |
5b882be4 EGM |
3322 | |
3323 | return ret; | |
81819f0f CL |
3324 | } |
3325 | EXPORT_SYMBOL(__kmalloc_node); | |
3326 | #endif | |
3327 | ||
3328 | size_t ksize(const void *object) | |
3329 | { | |
272c1d21 | 3330 | struct page *page; |
81819f0f | 3331 | |
ef8b4520 | 3332 | if (unlikely(object == ZERO_SIZE_PTR)) |
272c1d21 CL |
3333 | return 0; |
3334 | ||
294a80a8 | 3335 | page = virt_to_head_page(object); |
294a80a8 | 3336 | |
76994412 PE |
3337 | if (unlikely(!PageSlab(page))) { |
3338 | WARN_ON(!PageCompound(page)); | |
294a80a8 | 3339 | return PAGE_SIZE << compound_order(page); |
76994412 | 3340 | } |
81819f0f | 3341 | |
1b4f59e3 | 3342 | return slab_ksize(page->slab_cache); |
81819f0f | 3343 | } |
b1aabecd | 3344 | EXPORT_SYMBOL(ksize); |
81819f0f CL |
3345 | |
3346 | void kfree(const void *x) | |
3347 | { | |
81819f0f | 3348 | struct page *page; |
5bb983b0 | 3349 | void *object = (void *)x; |
81819f0f | 3350 | |
2121db74 PE |
3351 | trace_kfree(_RET_IP_, x); |
3352 | ||
2408c550 | 3353 | if (unlikely(ZERO_OR_NULL_PTR(x))) |
81819f0f CL |
3354 | return; |
3355 | ||
b49af68f | 3356 | page = virt_to_head_page(x); |
aadb4bc4 | 3357 | if (unlikely(!PageSlab(page))) { |
0937502a | 3358 | BUG_ON(!PageCompound(page)); |
d56791b3 | 3359 | kfree_hook(x); |
52383431 | 3360 | __free_kmem_pages(page, compound_order(page)); |
aadb4bc4 CL |
3361 | return; |
3362 | } | |
1b4f59e3 | 3363 | slab_free(page->slab_cache, page, object, _RET_IP_); |
81819f0f CL |
3364 | } |
3365 | EXPORT_SYMBOL(kfree); | |
3366 | ||
2086d26a | 3367 | /* |
672bba3a CL |
3368 | * kmem_cache_shrink removes empty slabs from the partial lists and sorts |
3369 | * the remaining slabs by the number of items in use. The slabs with the | |
3370 | * most items in use come first. New allocations will then fill those up | |
3371 | * and thus they can be removed from the partial lists. | |
3372 | * | |
3373 | * The slabs with the least items are placed last. This results in them | |
3374 | * being allocated from last increasing the chance that the last objects | |
3375 | * are freed in them. | |
2086d26a | 3376 | */ |
03afc0e2 | 3377 | int __kmem_cache_shrink(struct kmem_cache *s) |
2086d26a CL |
3378 | { |
3379 | int node; | |
3380 | int i; | |
3381 | struct kmem_cache_node *n; | |
3382 | struct page *page; | |
3383 | struct page *t; | |
205ab99d | 3384 | int objects = oo_objects(s->max); |
2086d26a | 3385 | struct list_head *slabs_by_inuse = |
834f3d11 | 3386 | kmalloc(sizeof(struct list_head) * objects, GFP_KERNEL); |
2086d26a CL |
3387 | unsigned long flags; |
3388 | ||
3389 | if (!slabs_by_inuse) | |
3390 | return -ENOMEM; | |
3391 | ||
3392 | flush_all(s); | |
fa45dc25 | 3393 | for_each_kmem_cache_node(s, node, n) { |
2086d26a CL |
3394 | if (!n->nr_partial) |
3395 | continue; | |
3396 | ||
834f3d11 | 3397 | for (i = 0; i < objects; i++) |
2086d26a CL |
3398 | INIT_LIST_HEAD(slabs_by_inuse + i); |
3399 | ||
3400 | spin_lock_irqsave(&n->list_lock, flags); | |
3401 | ||
3402 | /* | |
672bba3a | 3403 | * Build lists indexed by the items in use in each slab. |
2086d26a | 3404 | * |
672bba3a CL |
3405 | * Note that concurrent frees may occur while we hold the |
3406 | * list_lock. page->inuse here is the upper limit. | |
2086d26a CL |
3407 | */ |
3408 | list_for_each_entry_safe(page, t, &n->partial, lru) { | |
69cb8e6b CL |
3409 | list_move(&page->lru, slabs_by_inuse + page->inuse); |
3410 | if (!page->inuse) | |
3411 | n->nr_partial--; | |
2086d26a CL |
3412 | } |
3413 | ||
2086d26a | 3414 | /* |
672bba3a CL |
3415 | * Rebuild the partial list with the slabs filled up most |
3416 | * first and the least used slabs at the end. | |
2086d26a | 3417 | */ |
69cb8e6b | 3418 | for (i = objects - 1; i > 0; i--) |
2086d26a CL |
3419 | list_splice(slabs_by_inuse + i, n->partial.prev); |
3420 | ||
2086d26a | 3421 | spin_unlock_irqrestore(&n->list_lock, flags); |
69cb8e6b CL |
3422 | |
3423 | /* Release empty slabs */ | |
3424 | list_for_each_entry_safe(page, t, slabs_by_inuse, lru) | |
3425 | discard_slab(s, page); | |
2086d26a CL |
3426 | } |
3427 | ||
3428 | kfree(slabs_by_inuse); | |
3429 | return 0; | |
3430 | } | |
2086d26a | 3431 | |
b9049e23 YG |
3432 | static int slab_mem_going_offline_callback(void *arg) |
3433 | { | |
3434 | struct kmem_cache *s; | |
3435 | ||
18004c5d | 3436 | mutex_lock(&slab_mutex); |
b9049e23 | 3437 | list_for_each_entry(s, &slab_caches, list) |
03afc0e2 | 3438 | __kmem_cache_shrink(s); |
18004c5d | 3439 | mutex_unlock(&slab_mutex); |
b9049e23 YG |
3440 | |
3441 | return 0; | |
3442 | } | |
3443 | ||
3444 | static void slab_mem_offline_callback(void *arg) | |
3445 | { | |
3446 | struct kmem_cache_node *n; | |
3447 | struct kmem_cache *s; | |
3448 | struct memory_notify *marg = arg; | |
3449 | int offline_node; | |
3450 | ||
b9d5ab25 | 3451 | offline_node = marg->status_change_nid_normal; |
b9049e23 YG |
3452 | |
3453 | /* | |
3454 | * If the node still has available memory. we need kmem_cache_node | |
3455 | * for it yet. | |
3456 | */ | |
3457 | if (offline_node < 0) | |
3458 | return; | |
3459 | ||
18004c5d | 3460 | mutex_lock(&slab_mutex); |
b9049e23 YG |
3461 | list_for_each_entry(s, &slab_caches, list) { |
3462 | n = get_node(s, offline_node); | |
3463 | if (n) { | |
3464 | /* | |
3465 | * if n->nr_slabs > 0, slabs still exist on the node | |
3466 | * that is going down. We were unable to free them, | |
c9404c9c | 3467 | * and offline_pages() function shouldn't call this |
b9049e23 YG |
3468 | * callback. So, we must fail. |
3469 | */ | |
0f389ec6 | 3470 | BUG_ON(slabs_node(s, offline_node)); |
b9049e23 YG |
3471 | |
3472 | s->node[offline_node] = NULL; | |
8de66a0c | 3473 | kmem_cache_free(kmem_cache_node, n); |
b9049e23 YG |
3474 | } |
3475 | } | |
18004c5d | 3476 | mutex_unlock(&slab_mutex); |
b9049e23 YG |
3477 | } |
3478 | ||
3479 | static int slab_mem_going_online_callback(void *arg) | |
3480 | { | |
3481 | struct kmem_cache_node *n; | |
3482 | struct kmem_cache *s; | |
3483 | struct memory_notify *marg = arg; | |
b9d5ab25 | 3484 | int nid = marg->status_change_nid_normal; |
b9049e23 YG |
3485 | int ret = 0; |
3486 | ||
3487 | /* | |
3488 | * If the node's memory is already available, then kmem_cache_node is | |
3489 | * already created. Nothing to do. | |
3490 | */ | |
3491 | if (nid < 0) | |
3492 | return 0; | |
3493 | ||
3494 | /* | |
0121c619 | 3495 | * We are bringing a node online. No memory is available yet. We must |
b9049e23 YG |
3496 | * allocate a kmem_cache_node structure in order to bring the node |
3497 | * online. | |
3498 | */ | |
18004c5d | 3499 | mutex_lock(&slab_mutex); |
b9049e23 YG |
3500 | list_for_each_entry(s, &slab_caches, list) { |
3501 | /* | |
3502 | * XXX: kmem_cache_alloc_node will fallback to other nodes | |
3503 | * since memory is not yet available from the node that | |
3504 | * is brought up. | |
3505 | */ | |
8de66a0c | 3506 | n = kmem_cache_alloc(kmem_cache_node, GFP_KERNEL); |
b9049e23 YG |
3507 | if (!n) { |
3508 | ret = -ENOMEM; | |
3509 | goto out; | |
3510 | } | |
4053497d | 3511 | init_kmem_cache_node(n); |
b9049e23 YG |
3512 | s->node[nid] = n; |
3513 | } | |
3514 | out: | |
18004c5d | 3515 | mutex_unlock(&slab_mutex); |
b9049e23 YG |
3516 | return ret; |
3517 | } | |
3518 | ||
3519 | static int slab_memory_callback(struct notifier_block *self, | |
3520 | unsigned long action, void *arg) | |
3521 | { | |
3522 | int ret = 0; | |
3523 | ||
3524 | switch (action) { | |
3525 | case MEM_GOING_ONLINE: | |
3526 | ret = slab_mem_going_online_callback(arg); | |
3527 | break; | |
3528 | case MEM_GOING_OFFLINE: | |
3529 | ret = slab_mem_going_offline_callback(arg); | |
3530 | break; | |
3531 | case MEM_OFFLINE: | |
3532 | case MEM_CANCEL_ONLINE: | |
3533 | slab_mem_offline_callback(arg); | |
3534 | break; | |
3535 | case MEM_ONLINE: | |
3536 | case MEM_CANCEL_OFFLINE: | |
3537 | break; | |
3538 | } | |
dc19f9db KH |
3539 | if (ret) |
3540 | ret = notifier_from_errno(ret); | |
3541 | else | |
3542 | ret = NOTIFY_OK; | |
b9049e23 YG |
3543 | return ret; |
3544 | } | |
3545 | ||
3ac38faa AM |
3546 | static struct notifier_block slab_memory_callback_nb = { |
3547 | .notifier_call = slab_memory_callback, | |
3548 | .priority = SLAB_CALLBACK_PRI, | |
3549 | }; | |
b9049e23 | 3550 | |
81819f0f CL |
3551 | /******************************************************************** |
3552 | * Basic setup of slabs | |
3553 | *******************************************************************/ | |
3554 | ||
51df1142 CL |
3555 | /* |
3556 | * Used for early kmem_cache structures that were allocated using | |
dffb4d60 CL |
3557 | * the page allocator. Allocate them properly then fix up the pointers |
3558 | * that may be pointing to the wrong kmem_cache structure. | |
51df1142 CL |
3559 | */ |
3560 | ||
dffb4d60 | 3561 | static struct kmem_cache * __init bootstrap(struct kmem_cache *static_cache) |
51df1142 CL |
3562 | { |
3563 | int node; | |
dffb4d60 | 3564 | struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT); |
fa45dc25 | 3565 | struct kmem_cache_node *n; |
51df1142 | 3566 | |
dffb4d60 | 3567 | memcpy(s, static_cache, kmem_cache->object_size); |
51df1142 | 3568 | |
7d557b3c GC |
3569 | /* |
3570 | * This runs very early, and only the boot processor is supposed to be | |
3571 | * up. Even if it weren't true, IRQs are not up so we couldn't fire | |
3572 | * IPIs around. | |
3573 | */ | |
3574 | __flush_cpu_slab(s, smp_processor_id()); | |
fa45dc25 | 3575 | for_each_kmem_cache_node(s, node, n) { |
51df1142 CL |
3576 | struct page *p; |
3577 | ||
fa45dc25 CL |
3578 | list_for_each_entry(p, &n->partial, lru) |
3579 | p->slab_cache = s; | |
51df1142 | 3580 | |
607bf324 | 3581 | #ifdef CONFIG_SLUB_DEBUG |
fa45dc25 CL |
3582 | list_for_each_entry(p, &n->full, lru) |
3583 | p->slab_cache = s; | |
51df1142 | 3584 | #endif |
51df1142 | 3585 | } |
dffb4d60 CL |
3586 | list_add(&s->list, &slab_caches); |
3587 | return s; | |
51df1142 CL |
3588 | } |
3589 | ||
81819f0f CL |
3590 | void __init kmem_cache_init(void) |
3591 | { | |
dffb4d60 CL |
3592 | static __initdata struct kmem_cache boot_kmem_cache, |
3593 | boot_kmem_cache_node; | |
51df1142 | 3594 | |
fc8d8620 SG |
3595 | if (debug_guardpage_minorder()) |
3596 | slub_max_order = 0; | |
3597 | ||
dffb4d60 CL |
3598 | kmem_cache_node = &boot_kmem_cache_node; |
3599 | kmem_cache = &boot_kmem_cache; | |
51df1142 | 3600 | |
dffb4d60 CL |
3601 | create_boot_cache(kmem_cache_node, "kmem_cache_node", |
3602 | sizeof(struct kmem_cache_node), SLAB_HWCACHE_ALIGN); | |
b9049e23 | 3603 | |
3ac38faa | 3604 | register_hotmemory_notifier(&slab_memory_callback_nb); |
81819f0f CL |
3605 | |
3606 | /* Able to allocate the per node structures */ | |
3607 | slab_state = PARTIAL; | |
3608 | ||
dffb4d60 CL |
3609 | create_boot_cache(kmem_cache, "kmem_cache", |
3610 | offsetof(struct kmem_cache, node) + | |
3611 | nr_node_ids * sizeof(struct kmem_cache_node *), | |
3612 | SLAB_HWCACHE_ALIGN); | |
8a13a4cc | 3613 | |
dffb4d60 | 3614 | kmem_cache = bootstrap(&boot_kmem_cache); |
81819f0f | 3615 | |
51df1142 CL |
3616 | /* |
3617 | * Allocate kmem_cache_node properly from the kmem_cache slab. | |
3618 | * kmem_cache_node is separately allocated so no need to | |
3619 | * update any list pointers. | |
3620 | */ | |
dffb4d60 | 3621 | kmem_cache_node = bootstrap(&boot_kmem_cache_node); |
51df1142 CL |
3622 | |
3623 | /* Now we can use the kmem_cache to allocate kmalloc slabs */ | |
f97d5f63 | 3624 | create_kmalloc_caches(0); |
81819f0f CL |
3625 | |
3626 | #ifdef CONFIG_SMP | |
3627 | register_cpu_notifier(&slab_notifier); | |
9dfc6e68 | 3628 | #endif |
81819f0f | 3629 | |
f9f58285 | 3630 | pr_info("SLUB: HWalign=%d, Order=%d-%d, MinObjects=%d, CPUs=%d, Nodes=%d\n", |
f97d5f63 | 3631 | cache_line_size(), |
81819f0f CL |
3632 | slub_min_order, slub_max_order, slub_min_objects, |
3633 | nr_cpu_ids, nr_node_ids); | |
3634 | } | |
3635 | ||
7e85ee0c PE |
3636 | void __init kmem_cache_init_late(void) |
3637 | { | |
7e85ee0c PE |
3638 | } |
3639 | ||
81819f0f CL |
3640 | /* |
3641 | * Find a mergeable slab cache | |
3642 | */ | |
3643 | static int slab_unmergeable(struct kmem_cache *s) | |
3644 | { | |
3645 | if (slub_nomerge || (s->flags & SLUB_NEVER_MERGE)) | |
3646 | return 1; | |
3647 | ||
a44cb944 VD |
3648 | if (!is_root_cache(s)) |
3649 | return 1; | |
3650 | ||
c59def9f | 3651 | if (s->ctor) |
81819f0f CL |
3652 | return 1; |
3653 | ||
8ffa6875 CL |
3654 | /* |
3655 | * We may have set a slab to be unmergeable during bootstrap. | |
3656 | */ | |
3657 | if (s->refcount < 0) | |
3658 | return 1; | |
3659 | ||
81819f0f CL |
3660 | return 0; |
3661 | } | |
3662 | ||
a44cb944 VD |
3663 | static struct kmem_cache *find_mergeable(size_t size, size_t align, |
3664 | unsigned long flags, const char *name, void (*ctor)(void *)) | |
81819f0f | 3665 | { |
5b95a4ac | 3666 | struct kmem_cache *s; |
81819f0f CL |
3667 | |
3668 | if (slub_nomerge || (flags & SLUB_NEVER_MERGE)) | |
3669 | return NULL; | |
3670 | ||
c59def9f | 3671 | if (ctor) |
81819f0f CL |
3672 | return NULL; |
3673 | ||
3674 | size = ALIGN(size, sizeof(void *)); | |
3675 | align = calculate_alignment(flags, align, size); | |
3676 | size = ALIGN(size, align); | |
ba0268a8 | 3677 | flags = kmem_cache_flags(size, flags, name, NULL); |
81819f0f | 3678 | |
5b95a4ac | 3679 | list_for_each_entry(s, &slab_caches, list) { |
81819f0f CL |
3680 | if (slab_unmergeable(s)) |
3681 | continue; | |
3682 | ||
3683 | if (size > s->size) | |
3684 | continue; | |
3685 | ||
ba0268a8 | 3686 | if ((flags & SLUB_MERGE_SAME) != (s->flags & SLUB_MERGE_SAME)) |
a44cb944 | 3687 | continue; |
81819f0f CL |
3688 | /* |
3689 | * Check if alignment is compatible. | |
3690 | * Courtesy of Adrian Drzewiecki | |
3691 | */ | |
06428780 | 3692 | if ((s->size & ~(align - 1)) != s->size) |
81819f0f CL |
3693 | continue; |
3694 | ||
3695 | if (s->size - size >= sizeof(void *)) | |
3696 | continue; | |
3697 | ||
3698 | return s; | |
3699 | } | |
3700 | return NULL; | |
3701 | } | |
3702 | ||
2633d7a0 | 3703 | struct kmem_cache * |
a44cb944 VD |
3704 | __kmem_cache_alias(const char *name, size_t size, size_t align, |
3705 | unsigned long flags, void (*ctor)(void *)) | |
81819f0f CL |
3706 | { |
3707 | struct kmem_cache *s; | |
3708 | ||
a44cb944 | 3709 | s = find_mergeable(size, align, flags, name, ctor); |
81819f0f | 3710 | if (s) { |
84d0ddd6 VD |
3711 | int i; |
3712 | struct kmem_cache *c; | |
3713 | ||
81819f0f | 3714 | s->refcount++; |
84d0ddd6 | 3715 | |
81819f0f CL |
3716 | /* |
3717 | * Adjust the object sizes so that we clear | |
3718 | * the complete object on kzalloc. | |
3719 | */ | |
3b0efdfa | 3720 | s->object_size = max(s->object_size, (int)size); |
81819f0f | 3721 | s->inuse = max_t(int, s->inuse, ALIGN(size, sizeof(void *))); |
6446faa2 | 3722 | |
84d0ddd6 VD |
3723 | for_each_memcg_cache_index(i) { |
3724 | c = cache_from_memcg_idx(s, i); | |
3725 | if (!c) | |
3726 | continue; | |
3727 | c->object_size = s->object_size; | |
3728 | c->inuse = max_t(int, c->inuse, | |
3729 | ALIGN(size, sizeof(void *))); | |
3730 | } | |
3731 | ||
7b8f3b66 | 3732 | if (sysfs_slab_alias(s, name)) { |
7b8f3b66 | 3733 | s->refcount--; |
cbb79694 | 3734 | s = NULL; |
7b8f3b66 | 3735 | } |
a0e1d1be | 3736 | } |
6446faa2 | 3737 | |
cbb79694 CL |
3738 | return s; |
3739 | } | |
84c1cf62 | 3740 | |
8a13a4cc | 3741 | int __kmem_cache_create(struct kmem_cache *s, unsigned long flags) |
cbb79694 | 3742 | { |
aac3a166 PE |
3743 | int err; |
3744 | ||
3745 | err = kmem_cache_open(s, flags); | |
3746 | if (err) | |
3747 | return err; | |
20cea968 | 3748 | |
45530c44 CL |
3749 | /* Mutex is not taken during early boot */ |
3750 | if (slab_state <= UP) | |
3751 | return 0; | |
3752 | ||
107dab5c | 3753 | memcg_propagate_slab_attrs(s); |
aac3a166 | 3754 | err = sysfs_slab_add(s); |
aac3a166 PE |
3755 | if (err) |
3756 | kmem_cache_close(s); | |
20cea968 | 3757 | |
aac3a166 | 3758 | return err; |
81819f0f | 3759 | } |
81819f0f | 3760 | |
81819f0f | 3761 | #ifdef CONFIG_SMP |
81819f0f | 3762 | /* |
672bba3a CL |
3763 | * Use the cpu notifier to insure that the cpu slabs are flushed when |
3764 | * necessary. | |
81819f0f | 3765 | */ |
0db0628d | 3766 | static int slab_cpuup_callback(struct notifier_block *nfb, |
81819f0f CL |
3767 | unsigned long action, void *hcpu) |
3768 | { | |
3769 | long cpu = (long)hcpu; | |
5b95a4ac CL |
3770 | struct kmem_cache *s; |
3771 | unsigned long flags; | |
81819f0f CL |
3772 | |
3773 | switch (action) { | |
3774 | case CPU_UP_CANCELED: | |
8bb78442 | 3775 | case CPU_UP_CANCELED_FROZEN: |
81819f0f | 3776 | case CPU_DEAD: |
8bb78442 | 3777 | case CPU_DEAD_FROZEN: |
18004c5d | 3778 | mutex_lock(&slab_mutex); |
5b95a4ac CL |
3779 | list_for_each_entry(s, &slab_caches, list) { |
3780 | local_irq_save(flags); | |
3781 | __flush_cpu_slab(s, cpu); | |
3782 | local_irq_restore(flags); | |
3783 | } | |
18004c5d | 3784 | mutex_unlock(&slab_mutex); |
81819f0f CL |
3785 | break; |
3786 | default: | |
3787 | break; | |
3788 | } | |
3789 | return NOTIFY_OK; | |
3790 | } | |
3791 | ||
0db0628d | 3792 | static struct notifier_block slab_notifier = { |
3adbefee | 3793 | .notifier_call = slab_cpuup_callback |
06428780 | 3794 | }; |
81819f0f CL |
3795 | |
3796 | #endif | |
3797 | ||
ce71e27c | 3798 | void *__kmalloc_track_caller(size_t size, gfp_t gfpflags, unsigned long caller) |
81819f0f | 3799 | { |
aadb4bc4 | 3800 | struct kmem_cache *s; |
94b528d0 | 3801 | void *ret; |
aadb4bc4 | 3802 | |
95a05b42 | 3803 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) |
eada35ef PE |
3804 | return kmalloc_large(size, gfpflags); |
3805 | ||
2c59dd65 | 3806 | s = kmalloc_slab(size, gfpflags); |
81819f0f | 3807 | |
2408c550 | 3808 | if (unlikely(ZERO_OR_NULL_PTR(s))) |
6cb8f913 | 3809 | return s; |
81819f0f | 3810 | |
2b847c3c | 3811 | ret = slab_alloc(s, gfpflags, caller); |
94b528d0 | 3812 | |
25985edc | 3813 | /* Honor the call site pointer we received. */ |
ca2b84cb | 3814 | trace_kmalloc(caller, ret, size, s->size, gfpflags); |
94b528d0 EGM |
3815 | |
3816 | return ret; | |
81819f0f CL |
3817 | } |
3818 | ||
5d1f57e4 | 3819 | #ifdef CONFIG_NUMA |
81819f0f | 3820 | void *__kmalloc_node_track_caller(size_t size, gfp_t gfpflags, |
ce71e27c | 3821 | int node, unsigned long caller) |
81819f0f | 3822 | { |
aadb4bc4 | 3823 | struct kmem_cache *s; |
94b528d0 | 3824 | void *ret; |
aadb4bc4 | 3825 | |
95a05b42 | 3826 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) { |
d3e14aa3 XF |
3827 | ret = kmalloc_large_node(size, gfpflags, node); |
3828 | ||
3829 | trace_kmalloc_node(caller, ret, | |
3830 | size, PAGE_SIZE << get_order(size), | |
3831 | gfpflags, node); | |
3832 | ||
3833 | return ret; | |
3834 | } | |
eada35ef | 3835 | |
2c59dd65 | 3836 | s = kmalloc_slab(size, gfpflags); |
81819f0f | 3837 | |
2408c550 | 3838 | if (unlikely(ZERO_OR_NULL_PTR(s))) |
6cb8f913 | 3839 | return s; |
81819f0f | 3840 | |
2b847c3c | 3841 | ret = slab_alloc_node(s, gfpflags, node, caller); |
94b528d0 | 3842 | |
25985edc | 3843 | /* Honor the call site pointer we received. */ |
ca2b84cb | 3844 | trace_kmalloc_node(caller, ret, size, s->size, gfpflags, node); |
94b528d0 EGM |
3845 | |
3846 | return ret; | |
81819f0f | 3847 | } |
5d1f57e4 | 3848 | #endif |
81819f0f | 3849 | |
ab4d5ed5 | 3850 | #ifdef CONFIG_SYSFS |
205ab99d CL |
3851 | static int count_inuse(struct page *page) |
3852 | { | |
3853 | return page->inuse; | |
3854 | } | |
3855 | ||
3856 | static int count_total(struct page *page) | |
3857 | { | |
3858 | return page->objects; | |
3859 | } | |
ab4d5ed5 | 3860 | #endif |
205ab99d | 3861 | |
ab4d5ed5 | 3862 | #ifdef CONFIG_SLUB_DEBUG |
434e245d CL |
3863 | static int validate_slab(struct kmem_cache *s, struct page *page, |
3864 | unsigned long *map) | |
53e15af0 CL |
3865 | { |
3866 | void *p; | |
a973e9dd | 3867 | void *addr = page_address(page); |
53e15af0 CL |
3868 | |
3869 | if (!check_slab(s, page) || | |
3870 | !on_freelist(s, page, NULL)) | |
3871 | return 0; | |
3872 | ||
3873 | /* Now we know that a valid freelist exists */ | |
39b26464 | 3874 | bitmap_zero(map, page->objects); |
53e15af0 | 3875 | |
5f80b13a CL |
3876 | get_map(s, page, map); |
3877 | for_each_object(p, s, addr, page->objects) { | |
3878 | if (test_bit(slab_index(p, s, addr), map)) | |
3879 | if (!check_object(s, page, p, SLUB_RED_INACTIVE)) | |
3880 | return 0; | |
53e15af0 CL |
3881 | } |
3882 | ||
224a88be | 3883 | for_each_object(p, s, addr, page->objects) |
7656c72b | 3884 | if (!test_bit(slab_index(p, s, addr), map)) |
37d57443 | 3885 | if (!check_object(s, page, p, SLUB_RED_ACTIVE)) |
53e15af0 CL |
3886 | return 0; |
3887 | return 1; | |
3888 | } | |
3889 | ||
434e245d CL |
3890 | static void validate_slab_slab(struct kmem_cache *s, struct page *page, |
3891 | unsigned long *map) | |
53e15af0 | 3892 | { |
881db7fb CL |
3893 | slab_lock(page); |
3894 | validate_slab(s, page, map); | |
3895 | slab_unlock(page); | |
53e15af0 CL |
3896 | } |
3897 | ||
434e245d CL |
3898 | static int validate_slab_node(struct kmem_cache *s, |
3899 | struct kmem_cache_node *n, unsigned long *map) | |
53e15af0 CL |
3900 | { |
3901 | unsigned long count = 0; | |
3902 | struct page *page; | |
3903 | unsigned long flags; | |
3904 | ||
3905 | spin_lock_irqsave(&n->list_lock, flags); | |
3906 | ||
3907 | list_for_each_entry(page, &n->partial, lru) { | |
434e245d | 3908 | validate_slab_slab(s, page, map); |
53e15af0 CL |
3909 | count++; |
3910 | } | |
3911 | if (count != n->nr_partial) | |
f9f58285 FF |
3912 | pr_err("SLUB %s: %ld partial slabs counted but counter=%ld\n", |
3913 | s->name, count, n->nr_partial); | |
53e15af0 CL |
3914 | |
3915 | if (!(s->flags & SLAB_STORE_USER)) | |
3916 | goto out; | |
3917 | ||
3918 | list_for_each_entry(page, &n->full, lru) { | |
434e245d | 3919 | validate_slab_slab(s, page, map); |
53e15af0 CL |
3920 | count++; |
3921 | } | |
3922 | if (count != atomic_long_read(&n->nr_slabs)) | |
f9f58285 FF |
3923 | pr_err("SLUB: %s %ld slabs counted but counter=%ld\n", |
3924 | s->name, count, atomic_long_read(&n->nr_slabs)); | |
53e15af0 CL |
3925 | |
3926 | out: | |
3927 | spin_unlock_irqrestore(&n->list_lock, flags); | |
3928 | return count; | |
3929 | } | |
3930 | ||
434e245d | 3931 | static long validate_slab_cache(struct kmem_cache *s) |
53e15af0 CL |
3932 | { |
3933 | int node; | |
3934 | unsigned long count = 0; | |
205ab99d | 3935 | unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) * |
434e245d | 3936 | sizeof(unsigned long), GFP_KERNEL); |
fa45dc25 | 3937 | struct kmem_cache_node *n; |
434e245d CL |
3938 | |
3939 | if (!map) | |
3940 | return -ENOMEM; | |
53e15af0 CL |
3941 | |
3942 | flush_all(s); | |
fa45dc25 | 3943 | for_each_kmem_cache_node(s, node, n) |
434e245d | 3944 | count += validate_slab_node(s, n, map); |
434e245d | 3945 | kfree(map); |
53e15af0 CL |
3946 | return count; |
3947 | } | |
88a420e4 | 3948 | /* |
672bba3a | 3949 | * Generate lists of code addresses where slabcache objects are allocated |
88a420e4 CL |
3950 | * and freed. |
3951 | */ | |
3952 | ||
3953 | struct location { | |
3954 | unsigned long count; | |
ce71e27c | 3955 | unsigned long addr; |
45edfa58 CL |
3956 | long long sum_time; |
3957 | long min_time; | |
3958 | long max_time; | |
3959 | long min_pid; | |
3960 | long max_pid; | |
174596a0 | 3961 | DECLARE_BITMAP(cpus, NR_CPUS); |
45edfa58 | 3962 | nodemask_t nodes; |
88a420e4 CL |
3963 | }; |
3964 | ||
3965 | struct loc_track { | |
3966 | unsigned long max; | |
3967 | unsigned long count; | |
3968 | struct location *loc; | |
3969 | }; | |
3970 | ||
3971 | static void free_loc_track(struct loc_track *t) | |
3972 | { | |
3973 | if (t->max) | |
3974 | free_pages((unsigned long)t->loc, | |
3975 | get_order(sizeof(struct location) * t->max)); | |
3976 | } | |
3977 | ||
68dff6a9 | 3978 | static int alloc_loc_track(struct loc_track *t, unsigned long max, gfp_t flags) |
88a420e4 CL |
3979 | { |
3980 | struct location *l; | |
3981 | int order; | |
3982 | ||
88a420e4 CL |
3983 | order = get_order(sizeof(struct location) * max); |
3984 | ||
68dff6a9 | 3985 | l = (void *)__get_free_pages(flags, order); |
88a420e4 CL |
3986 | if (!l) |
3987 | return 0; | |
3988 | ||
3989 | if (t->count) { | |
3990 | memcpy(l, t->loc, sizeof(struct location) * t->count); | |
3991 | free_loc_track(t); | |
3992 | } | |
3993 | t->max = max; | |
3994 | t->loc = l; | |
3995 | return 1; | |
3996 | } | |
3997 | ||
3998 | static int add_location(struct loc_track *t, struct kmem_cache *s, | |
45edfa58 | 3999 | const struct track *track) |
88a420e4 CL |
4000 | { |
4001 | long start, end, pos; | |
4002 | struct location *l; | |
ce71e27c | 4003 | unsigned long caddr; |
45edfa58 | 4004 | unsigned long age = jiffies - track->when; |
88a420e4 CL |
4005 | |
4006 | start = -1; | |
4007 | end = t->count; | |
4008 | ||
4009 | for ( ; ; ) { | |
4010 | pos = start + (end - start + 1) / 2; | |
4011 | ||
4012 | /* | |
4013 | * There is nothing at "end". If we end up there | |
4014 | * we need to add something to before end. | |
4015 | */ | |
4016 | if (pos == end) | |
4017 | break; | |
4018 | ||
4019 | caddr = t->loc[pos].addr; | |
45edfa58 CL |
4020 | if (track->addr == caddr) { |
4021 | ||
4022 | l = &t->loc[pos]; | |
4023 | l->count++; | |
4024 | if (track->when) { | |
4025 | l->sum_time += age; | |
4026 | if (age < l->min_time) | |
4027 | l->min_time = age; | |
4028 | if (age > l->max_time) | |
4029 | l->max_time = age; | |
4030 | ||
4031 | if (track->pid < l->min_pid) | |
4032 | l->min_pid = track->pid; | |
4033 | if (track->pid > l->max_pid) | |
4034 | l->max_pid = track->pid; | |
4035 | ||
174596a0 RR |
4036 | cpumask_set_cpu(track->cpu, |
4037 | to_cpumask(l->cpus)); | |
45edfa58 CL |
4038 | } |
4039 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
88a420e4 CL |
4040 | return 1; |
4041 | } | |
4042 | ||
45edfa58 | 4043 | if (track->addr < caddr) |
88a420e4 CL |
4044 | end = pos; |
4045 | else | |
4046 | start = pos; | |
4047 | } | |
4048 | ||
4049 | /* | |
672bba3a | 4050 | * Not found. Insert new tracking element. |
88a420e4 | 4051 | */ |
68dff6a9 | 4052 | if (t->count >= t->max && !alloc_loc_track(t, 2 * t->max, GFP_ATOMIC)) |
88a420e4 CL |
4053 | return 0; |
4054 | ||
4055 | l = t->loc + pos; | |
4056 | if (pos < t->count) | |
4057 | memmove(l + 1, l, | |
4058 | (t->count - pos) * sizeof(struct location)); | |
4059 | t->count++; | |
4060 | l->count = 1; | |
45edfa58 CL |
4061 | l->addr = track->addr; |
4062 | l->sum_time = age; | |
4063 | l->min_time = age; | |
4064 | l->max_time = age; | |
4065 | l->min_pid = track->pid; | |
4066 | l->max_pid = track->pid; | |
174596a0 RR |
4067 | cpumask_clear(to_cpumask(l->cpus)); |
4068 | cpumask_set_cpu(track->cpu, to_cpumask(l->cpus)); | |
45edfa58 CL |
4069 | nodes_clear(l->nodes); |
4070 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
88a420e4 CL |
4071 | return 1; |
4072 | } | |
4073 | ||
4074 | static void process_slab(struct loc_track *t, struct kmem_cache *s, | |
bbd7d57b | 4075 | struct page *page, enum track_item alloc, |
a5dd5c11 | 4076 | unsigned long *map) |
88a420e4 | 4077 | { |
a973e9dd | 4078 | void *addr = page_address(page); |
88a420e4 CL |
4079 | void *p; |
4080 | ||
39b26464 | 4081 | bitmap_zero(map, page->objects); |
5f80b13a | 4082 | get_map(s, page, map); |
88a420e4 | 4083 | |
224a88be | 4084 | for_each_object(p, s, addr, page->objects) |
45edfa58 CL |
4085 | if (!test_bit(slab_index(p, s, addr), map)) |
4086 | add_location(t, s, get_track(s, p, alloc)); | |
88a420e4 CL |
4087 | } |
4088 | ||
4089 | static int list_locations(struct kmem_cache *s, char *buf, | |
4090 | enum track_item alloc) | |
4091 | { | |
e374d483 | 4092 | int len = 0; |
88a420e4 | 4093 | unsigned long i; |
68dff6a9 | 4094 | struct loc_track t = { 0, 0, NULL }; |
88a420e4 | 4095 | int node; |
bbd7d57b ED |
4096 | unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) * |
4097 | sizeof(unsigned long), GFP_KERNEL); | |
fa45dc25 | 4098 | struct kmem_cache_node *n; |
88a420e4 | 4099 | |
bbd7d57b ED |
4100 | if (!map || !alloc_loc_track(&t, PAGE_SIZE / sizeof(struct location), |
4101 | GFP_TEMPORARY)) { | |
4102 | kfree(map); | |
68dff6a9 | 4103 | return sprintf(buf, "Out of memory\n"); |
bbd7d57b | 4104 | } |
88a420e4 CL |
4105 | /* Push back cpu slabs */ |
4106 | flush_all(s); | |
4107 | ||
fa45dc25 | 4108 | for_each_kmem_cache_node(s, node, n) { |
88a420e4 CL |
4109 | unsigned long flags; |
4110 | struct page *page; | |
4111 | ||
9e86943b | 4112 | if (!atomic_long_read(&n->nr_slabs)) |
88a420e4 CL |
4113 | continue; |
4114 | ||
4115 | spin_lock_irqsave(&n->list_lock, flags); | |
4116 | list_for_each_entry(page, &n->partial, lru) | |
bbd7d57b | 4117 | process_slab(&t, s, page, alloc, map); |
88a420e4 | 4118 | list_for_each_entry(page, &n->full, lru) |
bbd7d57b | 4119 | process_slab(&t, s, page, alloc, map); |
88a420e4 CL |
4120 | spin_unlock_irqrestore(&n->list_lock, flags); |
4121 | } | |
4122 | ||
4123 | for (i = 0; i < t.count; i++) { | |
45edfa58 | 4124 | struct location *l = &t.loc[i]; |
88a420e4 | 4125 | |
9c246247 | 4126 | if (len > PAGE_SIZE - KSYM_SYMBOL_LEN - 100) |
88a420e4 | 4127 | break; |
e374d483 | 4128 | len += sprintf(buf + len, "%7ld ", l->count); |
45edfa58 CL |
4129 | |
4130 | if (l->addr) | |
62c70bce | 4131 | len += sprintf(buf + len, "%pS", (void *)l->addr); |
88a420e4 | 4132 | else |
e374d483 | 4133 | len += sprintf(buf + len, "<not-available>"); |
45edfa58 CL |
4134 | |
4135 | if (l->sum_time != l->min_time) { | |
e374d483 | 4136 | len += sprintf(buf + len, " age=%ld/%ld/%ld", |
f8bd2258 RZ |
4137 | l->min_time, |
4138 | (long)div_u64(l->sum_time, l->count), | |
4139 | l->max_time); | |
45edfa58 | 4140 | } else |
e374d483 | 4141 | len += sprintf(buf + len, " age=%ld", |
45edfa58 CL |
4142 | l->min_time); |
4143 | ||
4144 | if (l->min_pid != l->max_pid) | |
e374d483 | 4145 | len += sprintf(buf + len, " pid=%ld-%ld", |
45edfa58 CL |
4146 | l->min_pid, l->max_pid); |
4147 | else | |
e374d483 | 4148 | len += sprintf(buf + len, " pid=%ld", |
45edfa58 CL |
4149 | l->min_pid); |
4150 | ||
174596a0 RR |
4151 | if (num_online_cpus() > 1 && |
4152 | !cpumask_empty(to_cpumask(l->cpus)) && | |
e374d483 HH |
4153 | len < PAGE_SIZE - 60) { |
4154 | len += sprintf(buf + len, " cpus="); | |
d0e0ac97 CG |
4155 | len += cpulist_scnprintf(buf + len, |
4156 | PAGE_SIZE - len - 50, | |
174596a0 | 4157 | to_cpumask(l->cpus)); |
45edfa58 CL |
4158 | } |
4159 | ||
62bc62a8 | 4160 | if (nr_online_nodes > 1 && !nodes_empty(l->nodes) && |
e374d483 HH |
4161 | len < PAGE_SIZE - 60) { |
4162 | len += sprintf(buf + len, " nodes="); | |
d0e0ac97 CG |
4163 | len += nodelist_scnprintf(buf + len, |
4164 | PAGE_SIZE - len - 50, | |
4165 | l->nodes); | |
45edfa58 CL |
4166 | } |
4167 | ||
e374d483 | 4168 | len += sprintf(buf + len, "\n"); |
88a420e4 CL |
4169 | } |
4170 | ||
4171 | free_loc_track(&t); | |
bbd7d57b | 4172 | kfree(map); |
88a420e4 | 4173 | if (!t.count) |
e374d483 HH |
4174 | len += sprintf(buf, "No data\n"); |
4175 | return len; | |
88a420e4 | 4176 | } |
ab4d5ed5 | 4177 | #endif |
88a420e4 | 4178 | |
a5a84755 | 4179 | #ifdef SLUB_RESILIENCY_TEST |
c07b8183 | 4180 | static void __init resiliency_test(void) |
a5a84755 CL |
4181 | { |
4182 | u8 *p; | |
4183 | ||
95a05b42 | 4184 | BUILD_BUG_ON(KMALLOC_MIN_SIZE > 16 || KMALLOC_SHIFT_HIGH < 10); |
a5a84755 | 4185 | |
f9f58285 FF |
4186 | pr_err("SLUB resiliency testing\n"); |
4187 | pr_err("-----------------------\n"); | |
4188 | pr_err("A. Corruption after allocation\n"); | |
a5a84755 CL |
4189 | |
4190 | p = kzalloc(16, GFP_KERNEL); | |
4191 | p[16] = 0x12; | |
f9f58285 FF |
4192 | pr_err("\n1. kmalloc-16: Clobber Redzone/next pointer 0x12->0x%p\n\n", |
4193 | p + 16); | |
a5a84755 CL |
4194 | |
4195 | validate_slab_cache(kmalloc_caches[4]); | |
4196 | ||
4197 | /* Hmmm... The next two are dangerous */ | |
4198 | p = kzalloc(32, GFP_KERNEL); | |
4199 | p[32 + sizeof(void *)] = 0x34; | |
f9f58285 FF |
4200 | pr_err("\n2. kmalloc-32: Clobber next pointer/next slab 0x34 -> -0x%p\n", |
4201 | p); | |
4202 | pr_err("If allocated object is overwritten then not detectable\n\n"); | |
a5a84755 CL |
4203 | |
4204 | validate_slab_cache(kmalloc_caches[5]); | |
4205 | p = kzalloc(64, GFP_KERNEL); | |
4206 | p += 64 + (get_cycles() & 0xff) * sizeof(void *); | |
4207 | *p = 0x56; | |
f9f58285 FF |
4208 | pr_err("\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n", |
4209 | p); | |
4210 | pr_err("If allocated object is overwritten then not detectable\n\n"); | |
a5a84755 CL |
4211 | validate_slab_cache(kmalloc_caches[6]); |
4212 | ||
f9f58285 | 4213 | pr_err("\nB. Corruption after free\n"); |
a5a84755 CL |
4214 | p = kzalloc(128, GFP_KERNEL); |
4215 | kfree(p); | |
4216 | *p = 0x78; | |
f9f58285 | 4217 | pr_err("1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p); |
a5a84755 CL |
4218 | validate_slab_cache(kmalloc_caches[7]); |
4219 | ||
4220 | p = kzalloc(256, GFP_KERNEL); | |
4221 | kfree(p); | |
4222 | p[50] = 0x9a; | |
f9f58285 | 4223 | pr_err("\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n", p); |
a5a84755 CL |
4224 | validate_slab_cache(kmalloc_caches[8]); |
4225 | ||
4226 | p = kzalloc(512, GFP_KERNEL); | |
4227 | kfree(p); | |
4228 | p[512] = 0xab; | |
f9f58285 | 4229 | pr_err("\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p); |
a5a84755 CL |
4230 | validate_slab_cache(kmalloc_caches[9]); |
4231 | } | |
4232 | #else | |
4233 | #ifdef CONFIG_SYSFS | |
4234 | static void resiliency_test(void) {}; | |
4235 | #endif | |
4236 | #endif | |
4237 | ||
ab4d5ed5 | 4238 | #ifdef CONFIG_SYSFS |
81819f0f | 4239 | enum slab_stat_type { |
205ab99d CL |
4240 | SL_ALL, /* All slabs */ |
4241 | SL_PARTIAL, /* Only partially allocated slabs */ | |
4242 | SL_CPU, /* Only slabs used for cpu caches */ | |
4243 | SL_OBJECTS, /* Determine allocated objects not slabs */ | |
4244 | SL_TOTAL /* Determine object capacity not slabs */ | |
81819f0f CL |
4245 | }; |
4246 | ||
205ab99d | 4247 | #define SO_ALL (1 << SL_ALL) |
81819f0f CL |
4248 | #define SO_PARTIAL (1 << SL_PARTIAL) |
4249 | #define SO_CPU (1 << SL_CPU) | |
4250 | #define SO_OBJECTS (1 << SL_OBJECTS) | |
205ab99d | 4251 | #define SO_TOTAL (1 << SL_TOTAL) |
81819f0f | 4252 | |
62e5c4b4 CG |
4253 | static ssize_t show_slab_objects(struct kmem_cache *s, |
4254 | char *buf, unsigned long flags) | |
81819f0f CL |
4255 | { |
4256 | unsigned long total = 0; | |
81819f0f CL |
4257 | int node; |
4258 | int x; | |
4259 | unsigned long *nodes; | |
81819f0f | 4260 | |
e35e1a97 | 4261 | nodes = kzalloc(sizeof(unsigned long) * nr_node_ids, GFP_KERNEL); |
62e5c4b4 CG |
4262 | if (!nodes) |
4263 | return -ENOMEM; | |
81819f0f | 4264 | |
205ab99d CL |
4265 | if (flags & SO_CPU) { |
4266 | int cpu; | |
81819f0f | 4267 | |
205ab99d | 4268 | for_each_possible_cpu(cpu) { |
d0e0ac97 CG |
4269 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, |
4270 | cpu); | |
ec3ab083 | 4271 | int node; |
49e22585 | 4272 | struct page *page; |
dfb4f096 | 4273 | |
bc6697d8 | 4274 | page = ACCESS_ONCE(c->page); |
ec3ab083 CL |
4275 | if (!page) |
4276 | continue; | |
205ab99d | 4277 | |
ec3ab083 CL |
4278 | node = page_to_nid(page); |
4279 | if (flags & SO_TOTAL) | |
4280 | x = page->objects; | |
4281 | else if (flags & SO_OBJECTS) | |
4282 | x = page->inuse; | |
4283 | else | |
4284 | x = 1; | |
49e22585 | 4285 | |
ec3ab083 CL |
4286 | total += x; |
4287 | nodes[node] += x; | |
4288 | ||
4289 | page = ACCESS_ONCE(c->partial); | |
49e22585 | 4290 | if (page) { |
8afb1474 LZ |
4291 | node = page_to_nid(page); |
4292 | if (flags & SO_TOTAL) | |
4293 | WARN_ON_ONCE(1); | |
4294 | else if (flags & SO_OBJECTS) | |
4295 | WARN_ON_ONCE(1); | |
4296 | else | |
4297 | x = page->pages; | |
bc6697d8 ED |
4298 | total += x; |
4299 | nodes[node] += x; | |
49e22585 | 4300 | } |
81819f0f CL |
4301 | } |
4302 | } | |
4303 | ||
bfc8c901 | 4304 | get_online_mems(); |
ab4d5ed5 | 4305 | #ifdef CONFIG_SLUB_DEBUG |
205ab99d | 4306 | if (flags & SO_ALL) { |
fa45dc25 CL |
4307 | struct kmem_cache_node *n; |
4308 | ||
4309 | for_each_kmem_cache_node(s, node, n) { | |
205ab99d | 4310 | |
d0e0ac97 CG |
4311 | if (flags & SO_TOTAL) |
4312 | x = atomic_long_read(&n->total_objects); | |
4313 | else if (flags & SO_OBJECTS) | |
4314 | x = atomic_long_read(&n->total_objects) - | |
4315 | count_partial(n, count_free); | |
81819f0f | 4316 | else |
205ab99d | 4317 | x = atomic_long_read(&n->nr_slabs); |
81819f0f CL |
4318 | total += x; |
4319 | nodes[node] += x; | |
4320 | } | |
4321 | ||
ab4d5ed5 CL |
4322 | } else |
4323 | #endif | |
4324 | if (flags & SO_PARTIAL) { | |
fa45dc25 | 4325 | struct kmem_cache_node *n; |
81819f0f | 4326 | |
fa45dc25 | 4327 | for_each_kmem_cache_node(s, node, n) { |
205ab99d CL |
4328 | if (flags & SO_TOTAL) |
4329 | x = count_partial(n, count_total); | |
4330 | else if (flags & SO_OBJECTS) | |
4331 | x = count_partial(n, count_inuse); | |
81819f0f | 4332 | else |
205ab99d | 4333 | x = n->nr_partial; |
81819f0f CL |
4334 | total += x; |
4335 | nodes[node] += x; | |
4336 | } | |
4337 | } | |
81819f0f CL |
4338 | x = sprintf(buf, "%lu", total); |
4339 | #ifdef CONFIG_NUMA | |
fa45dc25 | 4340 | for (node = 0; node < nr_node_ids; node++) |
81819f0f CL |
4341 | if (nodes[node]) |
4342 | x += sprintf(buf + x, " N%d=%lu", | |
4343 | node, nodes[node]); | |
4344 | #endif | |
bfc8c901 | 4345 | put_online_mems(); |
81819f0f CL |
4346 | kfree(nodes); |
4347 | return x + sprintf(buf + x, "\n"); | |
4348 | } | |
4349 | ||
ab4d5ed5 | 4350 | #ifdef CONFIG_SLUB_DEBUG |
81819f0f CL |
4351 | static int any_slab_objects(struct kmem_cache *s) |
4352 | { | |
4353 | int node; | |
fa45dc25 | 4354 | struct kmem_cache_node *n; |
81819f0f | 4355 | |
fa45dc25 | 4356 | for_each_kmem_cache_node(s, node, n) |
4ea33e2d | 4357 | if (atomic_long_read(&n->total_objects)) |
81819f0f | 4358 | return 1; |
fa45dc25 | 4359 | |
81819f0f CL |
4360 | return 0; |
4361 | } | |
ab4d5ed5 | 4362 | #endif |
81819f0f CL |
4363 | |
4364 | #define to_slab_attr(n) container_of(n, struct slab_attribute, attr) | |
497888cf | 4365 | #define to_slab(n) container_of(n, struct kmem_cache, kobj) |
81819f0f CL |
4366 | |
4367 | struct slab_attribute { | |
4368 | struct attribute attr; | |
4369 | ssize_t (*show)(struct kmem_cache *s, char *buf); | |
4370 | ssize_t (*store)(struct kmem_cache *s, const char *x, size_t count); | |
4371 | }; | |
4372 | ||
4373 | #define SLAB_ATTR_RO(_name) \ | |
ab067e99 VK |
4374 | static struct slab_attribute _name##_attr = \ |
4375 | __ATTR(_name, 0400, _name##_show, NULL) | |
81819f0f CL |
4376 | |
4377 | #define SLAB_ATTR(_name) \ | |
4378 | static struct slab_attribute _name##_attr = \ | |
ab067e99 | 4379 | __ATTR(_name, 0600, _name##_show, _name##_store) |
81819f0f | 4380 | |
81819f0f CL |
4381 | static ssize_t slab_size_show(struct kmem_cache *s, char *buf) |
4382 | { | |
4383 | return sprintf(buf, "%d\n", s->size); | |
4384 | } | |
4385 | SLAB_ATTR_RO(slab_size); | |
4386 | ||
4387 | static ssize_t align_show(struct kmem_cache *s, char *buf) | |
4388 | { | |
4389 | return sprintf(buf, "%d\n", s->align); | |
4390 | } | |
4391 | SLAB_ATTR_RO(align); | |
4392 | ||
4393 | static ssize_t object_size_show(struct kmem_cache *s, char *buf) | |
4394 | { | |
3b0efdfa | 4395 | return sprintf(buf, "%d\n", s->object_size); |
81819f0f CL |
4396 | } |
4397 | SLAB_ATTR_RO(object_size); | |
4398 | ||
4399 | static ssize_t objs_per_slab_show(struct kmem_cache *s, char *buf) | |
4400 | { | |
834f3d11 | 4401 | return sprintf(buf, "%d\n", oo_objects(s->oo)); |
81819f0f CL |
4402 | } |
4403 | SLAB_ATTR_RO(objs_per_slab); | |
4404 | ||
06b285dc CL |
4405 | static ssize_t order_store(struct kmem_cache *s, |
4406 | const char *buf, size_t length) | |
4407 | { | |
0121c619 CL |
4408 | unsigned long order; |
4409 | int err; | |
4410 | ||
3dbb95f7 | 4411 | err = kstrtoul(buf, 10, &order); |
0121c619 CL |
4412 | if (err) |
4413 | return err; | |
06b285dc CL |
4414 | |
4415 | if (order > slub_max_order || order < slub_min_order) | |
4416 | return -EINVAL; | |
4417 | ||
4418 | calculate_sizes(s, order); | |
4419 | return length; | |
4420 | } | |
4421 | ||
81819f0f CL |
4422 | static ssize_t order_show(struct kmem_cache *s, char *buf) |
4423 | { | |
834f3d11 | 4424 | return sprintf(buf, "%d\n", oo_order(s->oo)); |
81819f0f | 4425 | } |
06b285dc | 4426 | SLAB_ATTR(order); |
81819f0f | 4427 | |
73d342b1 DR |
4428 | static ssize_t min_partial_show(struct kmem_cache *s, char *buf) |
4429 | { | |
4430 | return sprintf(buf, "%lu\n", s->min_partial); | |
4431 | } | |
4432 | ||
4433 | static ssize_t min_partial_store(struct kmem_cache *s, const char *buf, | |
4434 | size_t length) | |
4435 | { | |
4436 | unsigned long min; | |
4437 | int err; | |
4438 | ||
3dbb95f7 | 4439 | err = kstrtoul(buf, 10, &min); |
73d342b1 DR |
4440 | if (err) |
4441 | return err; | |
4442 | ||
c0bdb232 | 4443 | set_min_partial(s, min); |
73d342b1 DR |
4444 | return length; |
4445 | } | |
4446 | SLAB_ATTR(min_partial); | |
4447 | ||
49e22585 CL |
4448 | static ssize_t cpu_partial_show(struct kmem_cache *s, char *buf) |
4449 | { | |
4450 | return sprintf(buf, "%u\n", s->cpu_partial); | |
4451 | } | |
4452 | ||
4453 | static ssize_t cpu_partial_store(struct kmem_cache *s, const char *buf, | |
4454 | size_t length) | |
4455 | { | |
4456 | unsigned long objects; | |
4457 | int err; | |
4458 | ||
3dbb95f7 | 4459 | err = kstrtoul(buf, 10, &objects); |
49e22585 CL |
4460 | if (err) |
4461 | return err; | |
345c905d | 4462 | if (objects && !kmem_cache_has_cpu_partial(s)) |
74ee4ef1 | 4463 | return -EINVAL; |
49e22585 CL |
4464 | |
4465 | s->cpu_partial = objects; | |
4466 | flush_all(s); | |
4467 | return length; | |
4468 | } | |
4469 | SLAB_ATTR(cpu_partial); | |
4470 | ||
81819f0f CL |
4471 | static ssize_t ctor_show(struct kmem_cache *s, char *buf) |
4472 | { | |
62c70bce JP |
4473 | if (!s->ctor) |
4474 | return 0; | |
4475 | return sprintf(buf, "%pS\n", s->ctor); | |
81819f0f CL |
4476 | } |
4477 | SLAB_ATTR_RO(ctor); | |
4478 | ||
81819f0f CL |
4479 | static ssize_t aliases_show(struct kmem_cache *s, char *buf) |
4480 | { | |
4307c14f | 4481 | return sprintf(buf, "%d\n", s->refcount < 0 ? 0 : s->refcount - 1); |
81819f0f CL |
4482 | } |
4483 | SLAB_ATTR_RO(aliases); | |
4484 | ||
81819f0f CL |
4485 | static ssize_t partial_show(struct kmem_cache *s, char *buf) |
4486 | { | |
d9acf4b7 | 4487 | return show_slab_objects(s, buf, SO_PARTIAL); |
81819f0f CL |
4488 | } |
4489 | SLAB_ATTR_RO(partial); | |
4490 | ||
4491 | static ssize_t cpu_slabs_show(struct kmem_cache *s, char *buf) | |
4492 | { | |
d9acf4b7 | 4493 | return show_slab_objects(s, buf, SO_CPU); |
81819f0f CL |
4494 | } |
4495 | SLAB_ATTR_RO(cpu_slabs); | |
4496 | ||
4497 | static ssize_t objects_show(struct kmem_cache *s, char *buf) | |
4498 | { | |
205ab99d | 4499 | return show_slab_objects(s, buf, SO_ALL|SO_OBJECTS); |
81819f0f CL |
4500 | } |
4501 | SLAB_ATTR_RO(objects); | |
4502 | ||
205ab99d CL |
4503 | static ssize_t objects_partial_show(struct kmem_cache *s, char *buf) |
4504 | { | |
4505 | return show_slab_objects(s, buf, SO_PARTIAL|SO_OBJECTS); | |
4506 | } | |
4507 | SLAB_ATTR_RO(objects_partial); | |
4508 | ||
49e22585 CL |
4509 | static ssize_t slabs_cpu_partial_show(struct kmem_cache *s, char *buf) |
4510 | { | |
4511 | int objects = 0; | |
4512 | int pages = 0; | |
4513 | int cpu; | |
4514 | int len; | |
4515 | ||
4516 | for_each_online_cpu(cpu) { | |
4517 | struct page *page = per_cpu_ptr(s->cpu_slab, cpu)->partial; | |
4518 | ||
4519 | if (page) { | |
4520 | pages += page->pages; | |
4521 | objects += page->pobjects; | |
4522 | } | |
4523 | } | |
4524 | ||
4525 | len = sprintf(buf, "%d(%d)", objects, pages); | |
4526 | ||
4527 | #ifdef CONFIG_SMP | |
4528 | for_each_online_cpu(cpu) { | |
4529 | struct page *page = per_cpu_ptr(s->cpu_slab, cpu) ->partial; | |
4530 | ||
4531 | if (page && len < PAGE_SIZE - 20) | |
4532 | len += sprintf(buf + len, " C%d=%d(%d)", cpu, | |
4533 | page->pobjects, page->pages); | |
4534 | } | |
4535 | #endif | |
4536 | return len + sprintf(buf + len, "\n"); | |
4537 | } | |
4538 | SLAB_ATTR_RO(slabs_cpu_partial); | |
4539 | ||
a5a84755 CL |
4540 | static ssize_t reclaim_account_show(struct kmem_cache *s, char *buf) |
4541 | { | |
4542 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_RECLAIM_ACCOUNT)); | |
4543 | } | |
4544 | ||
4545 | static ssize_t reclaim_account_store(struct kmem_cache *s, | |
4546 | const char *buf, size_t length) | |
4547 | { | |
4548 | s->flags &= ~SLAB_RECLAIM_ACCOUNT; | |
4549 | if (buf[0] == '1') | |
4550 | s->flags |= SLAB_RECLAIM_ACCOUNT; | |
4551 | return length; | |
4552 | } | |
4553 | SLAB_ATTR(reclaim_account); | |
4554 | ||
4555 | static ssize_t hwcache_align_show(struct kmem_cache *s, char *buf) | |
4556 | { | |
4557 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_HWCACHE_ALIGN)); | |
4558 | } | |
4559 | SLAB_ATTR_RO(hwcache_align); | |
4560 | ||
4561 | #ifdef CONFIG_ZONE_DMA | |
4562 | static ssize_t cache_dma_show(struct kmem_cache *s, char *buf) | |
4563 | { | |
4564 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_CACHE_DMA)); | |
4565 | } | |
4566 | SLAB_ATTR_RO(cache_dma); | |
4567 | #endif | |
4568 | ||
4569 | static ssize_t destroy_by_rcu_show(struct kmem_cache *s, char *buf) | |
4570 | { | |
4571 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_DESTROY_BY_RCU)); | |
4572 | } | |
4573 | SLAB_ATTR_RO(destroy_by_rcu); | |
4574 | ||
ab9a0f19 LJ |
4575 | static ssize_t reserved_show(struct kmem_cache *s, char *buf) |
4576 | { | |
4577 | return sprintf(buf, "%d\n", s->reserved); | |
4578 | } | |
4579 | SLAB_ATTR_RO(reserved); | |
4580 | ||
ab4d5ed5 | 4581 | #ifdef CONFIG_SLUB_DEBUG |
a5a84755 CL |
4582 | static ssize_t slabs_show(struct kmem_cache *s, char *buf) |
4583 | { | |
4584 | return show_slab_objects(s, buf, SO_ALL); | |
4585 | } | |
4586 | SLAB_ATTR_RO(slabs); | |
4587 | ||
205ab99d CL |
4588 | static ssize_t total_objects_show(struct kmem_cache *s, char *buf) |
4589 | { | |
4590 | return show_slab_objects(s, buf, SO_ALL|SO_TOTAL); | |
4591 | } | |
4592 | SLAB_ATTR_RO(total_objects); | |
4593 | ||
81819f0f CL |
4594 | static ssize_t sanity_checks_show(struct kmem_cache *s, char *buf) |
4595 | { | |
4596 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_DEBUG_FREE)); | |
4597 | } | |
4598 | ||
4599 | static ssize_t sanity_checks_store(struct kmem_cache *s, | |
4600 | const char *buf, size_t length) | |
4601 | { | |
4602 | s->flags &= ~SLAB_DEBUG_FREE; | |
b789ef51 CL |
4603 | if (buf[0] == '1') { |
4604 | s->flags &= ~__CMPXCHG_DOUBLE; | |
81819f0f | 4605 | s->flags |= SLAB_DEBUG_FREE; |
b789ef51 | 4606 | } |
81819f0f CL |
4607 | return length; |
4608 | } | |
4609 | SLAB_ATTR(sanity_checks); | |
4610 | ||
4611 | static ssize_t trace_show(struct kmem_cache *s, char *buf) | |
4612 | { | |
4613 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_TRACE)); | |
4614 | } | |
4615 | ||
4616 | static ssize_t trace_store(struct kmem_cache *s, const char *buf, | |
4617 | size_t length) | |
4618 | { | |
c9e16131 CL |
4619 | /* |
4620 | * Tracing a merged cache is going to give confusing results | |
4621 | * as well as cause other issues like converting a mergeable | |
4622 | * cache into an umergeable one. | |
4623 | */ | |
4624 | if (s->refcount > 1) | |
4625 | return -EINVAL; | |
4626 | ||
81819f0f | 4627 | s->flags &= ~SLAB_TRACE; |
b789ef51 CL |
4628 | if (buf[0] == '1') { |
4629 | s->flags &= ~__CMPXCHG_DOUBLE; | |
81819f0f | 4630 | s->flags |= SLAB_TRACE; |
b789ef51 | 4631 | } |
81819f0f CL |
4632 | return length; |
4633 | } | |
4634 | SLAB_ATTR(trace); | |
4635 | ||
81819f0f CL |
4636 | static ssize_t red_zone_show(struct kmem_cache *s, char *buf) |
4637 | { | |
4638 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_RED_ZONE)); | |
4639 | } | |
4640 | ||
4641 | static ssize_t red_zone_store(struct kmem_cache *s, | |
4642 | const char *buf, size_t length) | |
4643 | { | |
4644 | if (any_slab_objects(s)) | |
4645 | return -EBUSY; | |
4646 | ||
4647 | s->flags &= ~SLAB_RED_ZONE; | |
b789ef51 CL |
4648 | if (buf[0] == '1') { |
4649 | s->flags &= ~__CMPXCHG_DOUBLE; | |
81819f0f | 4650 | s->flags |= SLAB_RED_ZONE; |
b789ef51 | 4651 | } |
06b285dc | 4652 | calculate_sizes(s, -1); |
81819f0f CL |
4653 | return length; |
4654 | } | |
4655 | SLAB_ATTR(red_zone); | |
4656 | ||
4657 | static ssize_t poison_show(struct kmem_cache *s, char *buf) | |
4658 | { | |
4659 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_POISON)); | |
4660 | } | |
4661 | ||
4662 | static ssize_t poison_store(struct kmem_cache *s, | |
4663 | const char *buf, size_t length) | |
4664 | { | |
4665 | if (any_slab_objects(s)) | |
4666 | return -EBUSY; | |
4667 | ||
4668 | s->flags &= ~SLAB_POISON; | |
b789ef51 CL |
4669 | if (buf[0] == '1') { |
4670 | s->flags &= ~__CMPXCHG_DOUBLE; | |
81819f0f | 4671 | s->flags |= SLAB_POISON; |
b789ef51 | 4672 | } |
06b285dc | 4673 | calculate_sizes(s, -1); |
81819f0f CL |
4674 | return length; |
4675 | } | |
4676 | SLAB_ATTR(poison); | |
4677 | ||
4678 | static ssize_t store_user_show(struct kmem_cache *s, char *buf) | |
4679 | { | |
4680 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_STORE_USER)); | |
4681 | } | |
4682 | ||
4683 | static ssize_t store_user_store(struct kmem_cache *s, | |
4684 | const char *buf, size_t length) | |
4685 | { | |
4686 | if (any_slab_objects(s)) | |
4687 | return -EBUSY; | |
4688 | ||
4689 | s->flags &= ~SLAB_STORE_USER; | |
b789ef51 CL |
4690 | if (buf[0] == '1') { |
4691 | s->flags &= ~__CMPXCHG_DOUBLE; | |
81819f0f | 4692 | s->flags |= SLAB_STORE_USER; |
b789ef51 | 4693 | } |
06b285dc | 4694 | calculate_sizes(s, -1); |
81819f0f CL |
4695 | return length; |
4696 | } | |
4697 | SLAB_ATTR(store_user); | |
4698 | ||
53e15af0 CL |
4699 | static ssize_t validate_show(struct kmem_cache *s, char *buf) |
4700 | { | |
4701 | return 0; | |
4702 | } | |
4703 | ||
4704 | static ssize_t validate_store(struct kmem_cache *s, | |
4705 | const char *buf, size_t length) | |
4706 | { | |
434e245d CL |
4707 | int ret = -EINVAL; |
4708 | ||
4709 | if (buf[0] == '1') { | |
4710 | ret = validate_slab_cache(s); | |
4711 | if (ret >= 0) | |
4712 | ret = length; | |
4713 | } | |
4714 | return ret; | |
53e15af0 CL |
4715 | } |
4716 | SLAB_ATTR(validate); | |
a5a84755 CL |
4717 | |
4718 | static ssize_t alloc_calls_show(struct kmem_cache *s, char *buf) | |
4719 | { | |
4720 | if (!(s->flags & SLAB_STORE_USER)) | |
4721 | return -ENOSYS; | |
4722 | return list_locations(s, buf, TRACK_ALLOC); | |
4723 | } | |
4724 | SLAB_ATTR_RO(alloc_calls); | |
4725 | ||
4726 | static ssize_t free_calls_show(struct kmem_cache *s, char *buf) | |
4727 | { | |
4728 | if (!(s->flags & SLAB_STORE_USER)) | |
4729 | return -ENOSYS; | |
4730 | return list_locations(s, buf, TRACK_FREE); | |
4731 | } | |
4732 | SLAB_ATTR_RO(free_calls); | |
4733 | #endif /* CONFIG_SLUB_DEBUG */ | |
4734 | ||
4735 | #ifdef CONFIG_FAILSLAB | |
4736 | static ssize_t failslab_show(struct kmem_cache *s, char *buf) | |
4737 | { | |
4738 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_FAILSLAB)); | |
4739 | } | |
4740 | ||
4741 | static ssize_t failslab_store(struct kmem_cache *s, const char *buf, | |
4742 | size_t length) | |
4743 | { | |
c9e16131 CL |
4744 | if (s->refcount > 1) |
4745 | return -EINVAL; | |
4746 | ||
a5a84755 CL |
4747 | s->flags &= ~SLAB_FAILSLAB; |
4748 | if (buf[0] == '1') | |
4749 | s->flags |= SLAB_FAILSLAB; | |
4750 | return length; | |
4751 | } | |
4752 | SLAB_ATTR(failslab); | |
ab4d5ed5 | 4753 | #endif |
53e15af0 | 4754 | |
2086d26a CL |
4755 | static ssize_t shrink_show(struct kmem_cache *s, char *buf) |
4756 | { | |
4757 | return 0; | |
4758 | } | |
4759 | ||
4760 | static ssize_t shrink_store(struct kmem_cache *s, | |
4761 | const char *buf, size_t length) | |
4762 | { | |
4763 | if (buf[0] == '1') { | |
4764 | int rc = kmem_cache_shrink(s); | |
4765 | ||
4766 | if (rc) | |
4767 | return rc; | |
4768 | } else | |
4769 | return -EINVAL; | |
4770 | return length; | |
4771 | } | |
4772 | SLAB_ATTR(shrink); | |
4773 | ||
81819f0f | 4774 | #ifdef CONFIG_NUMA |
9824601e | 4775 | static ssize_t remote_node_defrag_ratio_show(struct kmem_cache *s, char *buf) |
81819f0f | 4776 | { |
9824601e | 4777 | return sprintf(buf, "%d\n", s->remote_node_defrag_ratio / 10); |
81819f0f CL |
4778 | } |
4779 | ||
9824601e | 4780 | static ssize_t remote_node_defrag_ratio_store(struct kmem_cache *s, |
81819f0f CL |
4781 | const char *buf, size_t length) |
4782 | { | |
0121c619 CL |
4783 | unsigned long ratio; |
4784 | int err; | |
4785 | ||
3dbb95f7 | 4786 | err = kstrtoul(buf, 10, &ratio); |
0121c619 CL |
4787 | if (err) |
4788 | return err; | |
4789 | ||
e2cb96b7 | 4790 | if (ratio <= 100) |
0121c619 | 4791 | s->remote_node_defrag_ratio = ratio * 10; |
81819f0f | 4792 | |
81819f0f CL |
4793 | return length; |
4794 | } | |
9824601e | 4795 | SLAB_ATTR(remote_node_defrag_ratio); |
81819f0f CL |
4796 | #endif |
4797 | ||
8ff12cfc | 4798 | #ifdef CONFIG_SLUB_STATS |
8ff12cfc CL |
4799 | static int show_stat(struct kmem_cache *s, char *buf, enum stat_item si) |
4800 | { | |
4801 | unsigned long sum = 0; | |
4802 | int cpu; | |
4803 | int len; | |
4804 | int *data = kmalloc(nr_cpu_ids * sizeof(int), GFP_KERNEL); | |
4805 | ||
4806 | if (!data) | |
4807 | return -ENOMEM; | |
4808 | ||
4809 | for_each_online_cpu(cpu) { | |
9dfc6e68 | 4810 | unsigned x = per_cpu_ptr(s->cpu_slab, cpu)->stat[si]; |
8ff12cfc CL |
4811 | |
4812 | data[cpu] = x; | |
4813 | sum += x; | |
4814 | } | |
4815 | ||
4816 | len = sprintf(buf, "%lu", sum); | |
4817 | ||
50ef37b9 | 4818 | #ifdef CONFIG_SMP |
8ff12cfc CL |
4819 | for_each_online_cpu(cpu) { |
4820 | if (data[cpu] && len < PAGE_SIZE - 20) | |
50ef37b9 | 4821 | len += sprintf(buf + len, " C%d=%u", cpu, data[cpu]); |
8ff12cfc | 4822 | } |
50ef37b9 | 4823 | #endif |
8ff12cfc CL |
4824 | kfree(data); |
4825 | return len + sprintf(buf + len, "\n"); | |
4826 | } | |
4827 | ||
78eb00cc DR |
4828 | static void clear_stat(struct kmem_cache *s, enum stat_item si) |
4829 | { | |
4830 | int cpu; | |
4831 | ||
4832 | for_each_online_cpu(cpu) | |
9dfc6e68 | 4833 | per_cpu_ptr(s->cpu_slab, cpu)->stat[si] = 0; |
78eb00cc DR |
4834 | } |
4835 | ||
8ff12cfc CL |
4836 | #define STAT_ATTR(si, text) \ |
4837 | static ssize_t text##_show(struct kmem_cache *s, char *buf) \ | |
4838 | { \ | |
4839 | return show_stat(s, buf, si); \ | |
4840 | } \ | |
78eb00cc DR |
4841 | static ssize_t text##_store(struct kmem_cache *s, \ |
4842 | const char *buf, size_t length) \ | |
4843 | { \ | |
4844 | if (buf[0] != '0') \ | |
4845 | return -EINVAL; \ | |
4846 | clear_stat(s, si); \ | |
4847 | return length; \ | |
4848 | } \ | |
4849 | SLAB_ATTR(text); \ | |
8ff12cfc CL |
4850 | |
4851 | STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath); | |
4852 | STAT_ATTR(ALLOC_SLOWPATH, alloc_slowpath); | |
4853 | STAT_ATTR(FREE_FASTPATH, free_fastpath); | |
4854 | STAT_ATTR(FREE_SLOWPATH, free_slowpath); | |
4855 | STAT_ATTR(FREE_FROZEN, free_frozen); | |
4856 | STAT_ATTR(FREE_ADD_PARTIAL, free_add_partial); | |
4857 | STAT_ATTR(FREE_REMOVE_PARTIAL, free_remove_partial); | |
4858 | STAT_ATTR(ALLOC_FROM_PARTIAL, alloc_from_partial); | |
4859 | STAT_ATTR(ALLOC_SLAB, alloc_slab); | |
4860 | STAT_ATTR(ALLOC_REFILL, alloc_refill); | |
e36a2652 | 4861 | STAT_ATTR(ALLOC_NODE_MISMATCH, alloc_node_mismatch); |
8ff12cfc CL |
4862 | STAT_ATTR(FREE_SLAB, free_slab); |
4863 | STAT_ATTR(CPUSLAB_FLUSH, cpuslab_flush); | |
4864 | STAT_ATTR(DEACTIVATE_FULL, deactivate_full); | |
4865 | STAT_ATTR(DEACTIVATE_EMPTY, deactivate_empty); | |
4866 | STAT_ATTR(DEACTIVATE_TO_HEAD, deactivate_to_head); | |
4867 | STAT_ATTR(DEACTIVATE_TO_TAIL, deactivate_to_tail); | |
4868 | STAT_ATTR(DEACTIVATE_REMOTE_FREES, deactivate_remote_frees); | |
03e404af | 4869 | STAT_ATTR(DEACTIVATE_BYPASS, deactivate_bypass); |
65c3376a | 4870 | STAT_ATTR(ORDER_FALLBACK, order_fallback); |
b789ef51 CL |
4871 | STAT_ATTR(CMPXCHG_DOUBLE_CPU_FAIL, cmpxchg_double_cpu_fail); |
4872 | STAT_ATTR(CMPXCHG_DOUBLE_FAIL, cmpxchg_double_fail); | |
49e22585 CL |
4873 | STAT_ATTR(CPU_PARTIAL_ALLOC, cpu_partial_alloc); |
4874 | STAT_ATTR(CPU_PARTIAL_FREE, cpu_partial_free); | |
8028dcea AS |
4875 | STAT_ATTR(CPU_PARTIAL_NODE, cpu_partial_node); |
4876 | STAT_ATTR(CPU_PARTIAL_DRAIN, cpu_partial_drain); | |
8ff12cfc CL |
4877 | #endif |
4878 | ||
06428780 | 4879 | static struct attribute *slab_attrs[] = { |
81819f0f CL |
4880 | &slab_size_attr.attr, |
4881 | &object_size_attr.attr, | |
4882 | &objs_per_slab_attr.attr, | |
4883 | &order_attr.attr, | |
73d342b1 | 4884 | &min_partial_attr.attr, |
49e22585 | 4885 | &cpu_partial_attr.attr, |
81819f0f | 4886 | &objects_attr.attr, |
205ab99d | 4887 | &objects_partial_attr.attr, |
81819f0f CL |
4888 | &partial_attr.attr, |
4889 | &cpu_slabs_attr.attr, | |
4890 | &ctor_attr.attr, | |
81819f0f CL |
4891 | &aliases_attr.attr, |
4892 | &align_attr.attr, | |
81819f0f CL |
4893 | &hwcache_align_attr.attr, |
4894 | &reclaim_account_attr.attr, | |
4895 | &destroy_by_rcu_attr.attr, | |
a5a84755 | 4896 | &shrink_attr.attr, |
ab9a0f19 | 4897 | &reserved_attr.attr, |
49e22585 | 4898 | &slabs_cpu_partial_attr.attr, |
ab4d5ed5 | 4899 | #ifdef CONFIG_SLUB_DEBUG |
a5a84755 CL |
4900 | &total_objects_attr.attr, |
4901 | &slabs_attr.attr, | |
4902 | &sanity_checks_attr.attr, | |
4903 | &trace_attr.attr, | |
81819f0f CL |
4904 | &red_zone_attr.attr, |
4905 | &poison_attr.attr, | |
4906 | &store_user_attr.attr, | |
53e15af0 | 4907 | &validate_attr.attr, |
88a420e4 CL |
4908 | &alloc_calls_attr.attr, |
4909 | &free_calls_attr.attr, | |
ab4d5ed5 | 4910 | #endif |
81819f0f CL |
4911 | #ifdef CONFIG_ZONE_DMA |
4912 | &cache_dma_attr.attr, | |
4913 | #endif | |
4914 | #ifdef CONFIG_NUMA | |
9824601e | 4915 | &remote_node_defrag_ratio_attr.attr, |
8ff12cfc CL |
4916 | #endif |
4917 | #ifdef CONFIG_SLUB_STATS | |
4918 | &alloc_fastpath_attr.attr, | |
4919 | &alloc_slowpath_attr.attr, | |
4920 | &free_fastpath_attr.attr, | |
4921 | &free_slowpath_attr.attr, | |
4922 | &free_frozen_attr.attr, | |
4923 | &free_add_partial_attr.attr, | |
4924 | &free_remove_partial_attr.attr, | |
4925 | &alloc_from_partial_attr.attr, | |
4926 | &alloc_slab_attr.attr, | |
4927 | &alloc_refill_attr.attr, | |
e36a2652 | 4928 | &alloc_node_mismatch_attr.attr, |
8ff12cfc CL |
4929 | &free_slab_attr.attr, |
4930 | &cpuslab_flush_attr.attr, | |
4931 | &deactivate_full_attr.attr, | |
4932 | &deactivate_empty_attr.attr, | |
4933 | &deactivate_to_head_attr.attr, | |
4934 | &deactivate_to_tail_attr.attr, | |
4935 | &deactivate_remote_frees_attr.attr, | |
03e404af | 4936 | &deactivate_bypass_attr.attr, |
65c3376a | 4937 | &order_fallback_attr.attr, |
b789ef51 CL |
4938 | &cmpxchg_double_fail_attr.attr, |
4939 | &cmpxchg_double_cpu_fail_attr.attr, | |
49e22585 CL |
4940 | &cpu_partial_alloc_attr.attr, |
4941 | &cpu_partial_free_attr.attr, | |
8028dcea AS |
4942 | &cpu_partial_node_attr.attr, |
4943 | &cpu_partial_drain_attr.attr, | |
81819f0f | 4944 | #endif |
4c13dd3b DM |
4945 | #ifdef CONFIG_FAILSLAB |
4946 | &failslab_attr.attr, | |
4947 | #endif | |
4948 | ||
81819f0f CL |
4949 | NULL |
4950 | }; | |
4951 | ||
4952 | static struct attribute_group slab_attr_group = { | |
4953 | .attrs = slab_attrs, | |
4954 | }; | |
4955 | ||
4956 | static ssize_t slab_attr_show(struct kobject *kobj, | |
4957 | struct attribute *attr, | |
4958 | char *buf) | |
4959 | { | |
4960 | struct slab_attribute *attribute; | |
4961 | struct kmem_cache *s; | |
4962 | int err; | |
4963 | ||
4964 | attribute = to_slab_attr(attr); | |
4965 | s = to_slab(kobj); | |
4966 | ||
4967 | if (!attribute->show) | |
4968 | return -EIO; | |
4969 | ||
4970 | err = attribute->show(s, buf); | |
4971 | ||
4972 | return err; | |
4973 | } | |
4974 | ||
4975 | static ssize_t slab_attr_store(struct kobject *kobj, | |
4976 | struct attribute *attr, | |
4977 | const char *buf, size_t len) | |
4978 | { | |
4979 | struct slab_attribute *attribute; | |
4980 | struct kmem_cache *s; | |
4981 | int err; | |
4982 | ||
4983 | attribute = to_slab_attr(attr); | |
4984 | s = to_slab(kobj); | |
4985 | ||
4986 | if (!attribute->store) | |
4987 | return -EIO; | |
4988 | ||
4989 | err = attribute->store(s, buf, len); | |
107dab5c GC |
4990 | #ifdef CONFIG_MEMCG_KMEM |
4991 | if (slab_state >= FULL && err >= 0 && is_root_cache(s)) { | |
4992 | int i; | |
81819f0f | 4993 | |
107dab5c GC |
4994 | mutex_lock(&slab_mutex); |
4995 | if (s->max_attr_size < len) | |
4996 | s->max_attr_size = len; | |
4997 | ||
ebe945c2 GC |
4998 | /* |
4999 | * This is a best effort propagation, so this function's return | |
5000 | * value will be determined by the parent cache only. This is | |
5001 | * basically because not all attributes will have a well | |
5002 | * defined semantics for rollbacks - most of the actions will | |
5003 | * have permanent effects. | |
5004 | * | |
5005 | * Returning the error value of any of the children that fail | |
5006 | * is not 100 % defined, in the sense that users seeing the | |
5007 | * error code won't be able to know anything about the state of | |
5008 | * the cache. | |
5009 | * | |
5010 | * Only returning the error code for the parent cache at least | |
5011 | * has well defined semantics. The cache being written to | |
5012 | * directly either failed or succeeded, in which case we loop | |
5013 | * through the descendants with best-effort propagation. | |
5014 | */ | |
107dab5c | 5015 | for_each_memcg_cache_index(i) { |
2ade4de8 | 5016 | struct kmem_cache *c = cache_from_memcg_idx(s, i); |
107dab5c GC |
5017 | if (c) |
5018 | attribute->store(c, buf, len); | |
5019 | } | |
5020 | mutex_unlock(&slab_mutex); | |
5021 | } | |
5022 | #endif | |
81819f0f CL |
5023 | return err; |
5024 | } | |
5025 | ||
107dab5c GC |
5026 | static void memcg_propagate_slab_attrs(struct kmem_cache *s) |
5027 | { | |
5028 | #ifdef CONFIG_MEMCG_KMEM | |
5029 | int i; | |
5030 | char *buffer = NULL; | |
93030d83 | 5031 | struct kmem_cache *root_cache; |
107dab5c | 5032 | |
93030d83 | 5033 | if (is_root_cache(s)) |
107dab5c GC |
5034 | return; |
5035 | ||
93030d83 VD |
5036 | root_cache = s->memcg_params->root_cache; |
5037 | ||
107dab5c GC |
5038 | /* |
5039 | * This mean this cache had no attribute written. Therefore, no point | |
5040 | * in copying default values around | |
5041 | */ | |
93030d83 | 5042 | if (!root_cache->max_attr_size) |
107dab5c GC |
5043 | return; |
5044 | ||
5045 | for (i = 0; i < ARRAY_SIZE(slab_attrs); i++) { | |
5046 | char mbuf[64]; | |
5047 | char *buf; | |
5048 | struct slab_attribute *attr = to_slab_attr(slab_attrs[i]); | |
5049 | ||
5050 | if (!attr || !attr->store || !attr->show) | |
5051 | continue; | |
5052 | ||
5053 | /* | |
5054 | * It is really bad that we have to allocate here, so we will | |
5055 | * do it only as a fallback. If we actually allocate, though, | |
5056 | * we can just use the allocated buffer until the end. | |
5057 | * | |
5058 | * Most of the slub attributes will tend to be very small in | |
5059 | * size, but sysfs allows buffers up to a page, so they can | |
5060 | * theoretically happen. | |
5061 | */ | |
5062 | if (buffer) | |
5063 | buf = buffer; | |
93030d83 | 5064 | else if (root_cache->max_attr_size < ARRAY_SIZE(mbuf)) |
107dab5c GC |
5065 | buf = mbuf; |
5066 | else { | |
5067 | buffer = (char *) get_zeroed_page(GFP_KERNEL); | |
5068 | if (WARN_ON(!buffer)) | |
5069 | continue; | |
5070 | buf = buffer; | |
5071 | } | |
5072 | ||
93030d83 | 5073 | attr->show(root_cache, buf); |
107dab5c GC |
5074 | attr->store(s, buf, strlen(buf)); |
5075 | } | |
5076 | ||
5077 | if (buffer) | |
5078 | free_page((unsigned long)buffer); | |
5079 | #endif | |
5080 | } | |
5081 | ||
41a21285 CL |
5082 | static void kmem_cache_release(struct kobject *k) |
5083 | { | |
5084 | slab_kmem_cache_release(to_slab(k)); | |
5085 | } | |
5086 | ||
52cf25d0 | 5087 | static const struct sysfs_ops slab_sysfs_ops = { |
81819f0f CL |
5088 | .show = slab_attr_show, |
5089 | .store = slab_attr_store, | |
5090 | }; | |
5091 | ||
5092 | static struct kobj_type slab_ktype = { | |
5093 | .sysfs_ops = &slab_sysfs_ops, | |
41a21285 | 5094 | .release = kmem_cache_release, |
81819f0f CL |
5095 | }; |
5096 | ||
5097 | static int uevent_filter(struct kset *kset, struct kobject *kobj) | |
5098 | { | |
5099 | struct kobj_type *ktype = get_ktype(kobj); | |
5100 | ||
5101 | if (ktype == &slab_ktype) | |
5102 | return 1; | |
5103 | return 0; | |
5104 | } | |
5105 | ||
9cd43611 | 5106 | static const struct kset_uevent_ops slab_uevent_ops = { |
81819f0f CL |
5107 | .filter = uevent_filter, |
5108 | }; | |
5109 | ||
27c3a314 | 5110 | static struct kset *slab_kset; |
81819f0f | 5111 | |
9a41707b VD |
5112 | static inline struct kset *cache_kset(struct kmem_cache *s) |
5113 | { | |
5114 | #ifdef CONFIG_MEMCG_KMEM | |
5115 | if (!is_root_cache(s)) | |
5116 | return s->memcg_params->root_cache->memcg_kset; | |
5117 | #endif | |
5118 | return slab_kset; | |
5119 | } | |
5120 | ||
81819f0f CL |
5121 | #define ID_STR_LENGTH 64 |
5122 | ||
5123 | /* Create a unique string id for a slab cache: | |
6446faa2 CL |
5124 | * |
5125 | * Format :[flags-]size | |
81819f0f CL |
5126 | */ |
5127 | static char *create_unique_id(struct kmem_cache *s) | |
5128 | { | |
5129 | char *name = kmalloc(ID_STR_LENGTH, GFP_KERNEL); | |
5130 | char *p = name; | |
5131 | ||
5132 | BUG_ON(!name); | |
5133 | ||
5134 | *p++ = ':'; | |
5135 | /* | |
5136 | * First flags affecting slabcache operations. We will only | |
5137 | * get here for aliasable slabs so we do not need to support | |
5138 | * too many flags. The flags here must cover all flags that | |
5139 | * are matched during merging to guarantee that the id is | |
5140 | * unique. | |
5141 | */ | |
5142 | if (s->flags & SLAB_CACHE_DMA) | |
5143 | *p++ = 'd'; | |
5144 | if (s->flags & SLAB_RECLAIM_ACCOUNT) | |
5145 | *p++ = 'a'; | |
5146 | if (s->flags & SLAB_DEBUG_FREE) | |
5147 | *p++ = 'F'; | |
5a896d9e VN |
5148 | if (!(s->flags & SLAB_NOTRACK)) |
5149 | *p++ = 't'; | |
81819f0f CL |
5150 | if (p != name + 1) |
5151 | *p++ = '-'; | |
5152 | p += sprintf(p, "%07d", s->size); | |
2633d7a0 | 5153 | |
81819f0f CL |
5154 | BUG_ON(p > name + ID_STR_LENGTH - 1); |
5155 | return name; | |
5156 | } | |
5157 | ||
5158 | static int sysfs_slab_add(struct kmem_cache *s) | |
5159 | { | |
5160 | int err; | |
5161 | const char *name; | |
45530c44 | 5162 | int unmergeable = slab_unmergeable(s); |
81819f0f | 5163 | |
81819f0f CL |
5164 | if (unmergeable) { |
5165 | /* | |
5166 | * Slabcache can never be merged so we can use the name proper. | |
5167 | * This is typically the case for debug situations. In that | |
5168 | * case we can catch duplicate names easily. | |
5169 | */ | |
27c3a314 | 5170 | sysfs_remove_link(&slab_kset->kobj, s->name); |
81819f0f CL |
5171 | name = s->name; |
5172 | } else { | |
5173 | /* | |
5174 | * Create a unique name for the slab as a target | |
5175 | * for the symlinks. | |
5176 | */ | |
5177 | name = create_unique_id(s); | |
5178 | } | |
5179 | ||
9a41707b | 5180 | s->kobj.kset = cache_kset(s); |
26e4f205 | 5181 | err = kobject_init_and_add(&s->kobj, &slab_ktype, NULL, "%s", name); |
54b6a731 DJ |
5182 | if (err) |
5183 | goto out_put_kobj; | |
81819f0f CL |
5184 | |
5185 | err = sysfs_create_group(&s->kobj, &slab_attr_group); | |
54b6a731 DJ |
5186 | if (err) |
5187 | goto out_del_kobj; | |
9a41707b VD |
5188 | |
5189 | #ifdef CONFIG_MEMCG_KMEM | |
5190 | if (is_root_cache(s)) { | |
5191 | s->memcg_kset = kset_create_and_add("cgroup", NULL, &s->kobj); | |
5192 | if (!s->memcg_kset) { | |
54b6a731 DJ |
5193 | err = -ENOMEM; |
5194 | goto out_del_kobj; | |
9a41707b VD |
5195 | } |
5196 | } | |
5197 | #endif | |
5198 | ||
81819f0f CL |
5199 | kobject_uevent(&s->kobj, KOBJ_ADD); |
5200 | if (!unmergeable) { | |
5201 | /* Setup first alias */ | |
5202 | sysfs_slab_alias(s, s->name); | |
81819f0f | 5203 | } |
54b6a731 DJ |
5204 | out: |
5205 | if (!unmergeable) | |
5206 | kfree(name); | |
5207 | return err; | |
5208 | out_del_kobj: | |
5209 | kobject_del(&s->kobj); | |
5210 | out_put_kobj: | |
5211 | kobject_put(&s->kobj); | |
5212 | goto out; | |
81819f0f CL |
5213 | } |
5214 | ||
41a21285 | 5215 | void sysfs_slab_remove(struct kmem_cache *s) |
81819f0f | 5216 | { |
97d06609 | 5217 | if (slab_state < FULL) |
2bce6485 CL |
5218 | /* |
5219 | * Sysfs has not been setup yet so no need to remove the | |
5220 | * cache from sysfs. | |
5221 | */ | |
5222 | return; | |
5223 | ||
9a41707b VD |
5224 | #ifdef CONFIG_MEMCG_KMEM |
5225 | kset_unregister(s->memcg_kset); | |
5226 | #endif | |
81819f0f CL |
5227 | kobject_uevent(&s->kobj, KOBJ_REMOVE); |
5228 | kobject_del(&s->kobj); | |
151c602f | 5229 | kobject_put(&s->kobj); |
81819f0f CL |
5230 | } |
5231 | ||
5232 | /* | |
5233 | * Need to buffer aliases during bootup until sysfs becomes | |
9f6c708e | 5234 | * available lest we lose that information. |
81819f0f CL |
5235 | */ |
5236 | struct saved_alias { | |
5237 | struct kmem_cache *s; | |
5238 | const char *name; | |
5239 | struct saved_alias *next; | |
5240 | }; | |
5241 | ||
5af328a5 | 5242 | static struct saved_alias *alias_list; |
81819f0f CL |
5243 | |
5244 | static int sysfs_slab_alias(struct kmem_cache *s, const char *name) | |
5245 | { | |
5246 | struct saved_alias *al; | |
5247 | ||
97d06609 | 5248 | if (slab_state == FULL) { |
81819f0f CL |
5249 | /* |
5250 | * If we have a leftover link then remove it. | |
5251 | */ | |
27c3a314 GKH |
5252 | sysfs_remove_link(&slab_kset->kobj, name); |
5253 | return sysfs_create_link(&slab_kset->kobj, &s->kobj, name); | |
81819f0f CL |
5254 | } |
5255 | ||
5256 | al = kmalloc(sizeof(struct saved_alias), GFP_KERNEL); | |
5257 | if (!al) | |
5258 | return -ENOMEM; | |
5259 | ||
5260 | al->s = s; | |
5261 | al->name = name; | |
5262 | al->next = alias_list; | |
5263 | alias_list = al; | |
5264 | return 0; | |
5265 | } | |
5266 | ||
5267 | static int __init slab_sysfs_init(void) | |
5268 | { | |
5b95a4ac | 5269 | struct kmem_cache *s; |
81819f0f CL |
5270 | int err; |
5271 | ||
18004c5d | 5272 | mutex_lock(&slab_mutex); |
2bce6485 | 5273 | |
0ff21e46 | 5274 | slab_kset = kset_create_and_add("slab", &slab_uevent_ops, kernel_kobj); |
27c3a314 | 5275 | if (!slab_kset) { |
18004c5d | 5276 | mutex_unlock(&slab_mutex); |
f9f58285 | 5277 | pr_err("Cannot register slab subsystem.\n"); |
81819f0f CL |
5278 | return -ENOSYS; |
5279 | } | |
5280 | ||
97d06609 | 5281 | slab_state = FULL; |
26a7bd03 | 5282 | |
5b95a4ac | 5283 | list_for_each_entry(s, &slab_caches, list) { |
26a7bd03 | 5284 | err = sysfs_slab_add(s); |
5d540fb7 | 5285 | if (err) |
f9f58285 FF |
5286 | pr_err("SLUB: Unable to add boot slab %s to sysfs\n", |
5287 | s->name); | |
26a7bd03 | 5288 | } |
81819f0f CL |
5289 | |
5290 | while (alias_list) { | |
5291 | struct saved_alias *al = alias_list; | |
5292 | ||
5293 | alias_list = alias_list->next; | |
5294 | err = sysfs_slab_alias(al->s, al->name); | |
5d540fb7 | 5295 | if (err) |
f9f58285 FF |
5296 | pr_err("SLUB: Unable to add boot slab alias %s to sysfs\n", |
5297 | al->name); | |
81819f0f CL |
5298 | kfree(al); |
5299 | } | |
5300 | ||
18004c5d | 5301 | mutex_unlock(&slab_mutex); |
81819f0f CL |
5302 | resiliency_test(); |
5303 | return 0; | |
5304 | } | |
5305 | ||
5306 | __initcall(slab_sysfs_init); | |
ab4d5ed5 | 5307 | #endif /* CONFIG_SYSFS */ |
57ed3eda PE |
5308 | |
5309 | /* | |
5310 | * The /proc/slabinfo ABI | |
5311 | */ | |
158a9624 | 5312 | #ifdef CONFIG_SLABINFO |
0d7561c6 | 5313 | void get_slabinfo(struct kmem_cache *s, struct slabinfo *sinfo) |
57ed3eda | 5314 | { |
57ed3eda | 5315 | unsigned long nr_slabs = 0; |
205ab99d CL |
5316 | unsigned long nr_objs = 0; |
5317 | unsigned long nr_free = 0; | |
57ed3eda | 5318 | int node; |
fa45dc25 | 5319 | struct kmem_cache_node *n; |
57ed3eda | 5320 | |
fa45dc25 | 5321 | for_each_kmem_cache_node(s, node, n) { |
c17fd13e WL |
5322 | nr_slabs += node_nr_slabs(n); |
5323 | nr_objs += node_nr_objs(n); | |
205ab99d | 5324 | nr_free += count_partial(n, count_free); |
57ed3eda PE |
5325 | } |
5326 | ||
0d7561c6 GC |
5327 | sinfo->active_objs = nr_objs - nr_free; |
5328 | sinfo->num_objs = nr_objs; | |
5329 | sinfo->active_slabs = nr_slabs; | |
5330 | sinfo->num_slabs = nr_slabs; | |
5331 | sinfo->objects_per_slab = oo_objects(s->oo); | |
5332 | sinfo->cache_order = oo_order(s->oo); | |
57ed3eda PE |
5333 | } |
5334 | ||
0d7561c6 | 5335 | void slabinfo_show_stats(struct seq_file *m, struct kmem_cache *s) |
7b3c3a50 | 5336 | { |
7b3c3a50 AD |
5337 | } |
5338 | ||
b7454ad3 GC |
5339 | ssize_t slabinfo_write(struct file *file, const char __user *buffer, |
5340 | size_t count, loff_t *ppos) | |
7b3c3a50 | 5341 | { |
b7454ad3 | 5342 | return -EIO; |
7b3c3a50 | 5343 | } |
158a9624 | 5344 | #endif /* CONFIG_SLABINFO */ |