]>
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 *); | |
213 | static void sysfs_slab_remove(struct kmem_cache *); | |
107dab5c | 214 | static void memcg_propagate_slab_attrs(struct kmem_cache *s); |
81819f0f | 215 | #else |
0c710013 CL |
216 | static inline int sysfs_slab_add(struct kmem_cache *s) { return 0; } |
217 | static inline int sysfs_slab_alias(struct kmem_cache *s, const char *p) | |
218 | { return 0; } | |
db265eca | 219 | static inline void sysfs_slab_remove(struct kmem_cache *s) { } |
8ff12cfc | 220 | |
107dab5c | 221 | static inline void memcg_propagate_slab_attrs(struct kmem_cache *s) { } |
81819f0f CL |
222 | #endif |
223 | ||
4fdccdfb | 224 | static inline void stat(const struct kmem_cache *s, enum stat_item si) |
8ff12cfc CL |
225 | { |
226 | #ifdef CONFIG_SLUB_STATS | |
84e554e6 | 227 | __this_cpu_inc(s->cpu_slab->stat[si]); |
8ff12cfc CL |
228 | #endif |
229 | } | |
230 | ||
81819f0f CL |
231 | /******************************************************************** |
232 | * Core slab cache functions | |
233 | *******************************************************************/ | |
234 | ||
81819f0f CL |
235 | static inline struct kmem_cache_node *get_node(struct kmem_cache *s, int node) |
236 | { | |
81819f0f | 237 | return s->node[node]; |
81819f0f CL |
238 | } |
239 | ||
6446faa2 | 240 | /* Verify that a pointer has an address that is valid within a slab page */ |
02cbc874 CL |
241 | static inline int check_valid_pointer(struct kmem_cache *s, |
242 | struct page *page, const void *object) | |
243 | { | |
244 | void *base; | |
245 | ||
a973e9dd | 246 | if (!object) |
02cbc874 CL |
247 | return 1; |
248 | ||
a973e9dd | 249 | base = page_address(page); |
39b26464 | 250 | if (object < base || object >= base + page->objects * s->size || |
02cbc874 CL |
251 | (object - base) % s->size) { |
252 | return 0; | |
253 | } | |
254 | ||
255 | return 1; | |
256 | } | |
257 | ||
7656c72b CL |
258 | static inline void *get_freepointer(struct kmem_cache *s, void *object) |
259 | { | |
260 | return *(void **)(object + s->offset); | |
261 | } | |
262 | ||
0ad9500e ED |
263 | static void prefetch_freepointer(const struct kmem_cache *s, void *object) |
264 | { | |
265 | prefetch(object + s->offset); | |
266 | } | |
267 | ||
1393d9a1 CL |
268 | static inline void *get_freepointer_safe(struct kmem_cache *s, void *object) |
269 | { | |
270 | void *p; | |
271 | ||
272 | #ifdef CONFIG_DEBUG_PAGEALLOC | |
273 | probe_kernel_read(&p, (void **)(object + s->offset), sizeof(p)); | |
274 | #else | |
275 | p = get_freepointer(s, object); | |
276 | #endif | |
277 | return p; | |
278 | } | |
279 | ||
7656c72b CL |
280 | static inline void set_freepointer(struct kmem_cache *s, void *object, void *fp) |
281 | { | |
282 | *(void **)(object + s->offset) = fp; | |
283 | } | |
284 | ||
285 | /* Loop over all objects in a slab */ | |
224a88be CL |
286 | #define for_each_object(__p, __s, __addr, __objects) \ |
287 | for (__p = (__addr); __p < (__addr) + (__objects) * (__s)->size;\ | |
7656c72b CL |
288 | __p += (__s)->size) |
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, |
1d07171c CL |
384 | freelist_old, counters_old, |
385 | freelist_new, counters_new)) | |
386 | return 1; | |
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 | |
405 | printk(KERN_INFO "%s %s: cmpxchg double redo ", n, s->name); | |
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, |
b789ef51 CL |
420 | freelist_old, counters_old, |
421 | freelist_new, counters_new)) | |
422 | return 1; | |
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 | |
446 | printk(KERN_INFO "%s %s: cmpxchg double redo ", n, s->name); | |
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 | ||
7daf705f | 548 | printk(KERN_ERR "INFO: %s in %pS age=%lu cpu=%u pid=%d\n", |
ce71e27c | 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]) | |
555 | printk(KERN_ERR "\t%pS\n", (void *)t->addrs[i]); | |
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 | { | |
d0e0ac97 CG |
573 | printk(KERN_ERR |
574 | "INFO: Slab 0x%p objects=%u used=%u fp=0x%p flags=0x%04lx\n", | |
575 | page, page->objects, page->inuse, page->freelist, page->flags); | |
24922684 CL |
576 | |
577 | } | |
578 | ||
579 | static void slab_bug(struct kmem_cache *s, char *fmt, ...) | |
580 | { | |
581 | va_list args; | |
582 | char buf[100]; | |
583 | ||
584 | va_start(args, fmt); | |
585 | vsnprintf(buf, sizeof(buf), fmt, args); | |
586 | va_end(args); | |
587 | printk(KERN_ERR "========================================" | |
588 | "=====================================\n"); | |
265d47e7 | 589 | printk(KERN_ERR "BUG %s (%s): %s\n", s->name, print_tainted(), buf); |
24922684 CL |
590 | printk(KERN_ERR "----------------------------------------" |
591 | "-------------------------------------\n\n"); | |
645df230 | 592 | |
373d4d09 | 593 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); |
81819f0f CL |
594 | } |
595 | ||
24922684 CL |
596 | static void slab_fix(struct kmem_cache *s, char *fmt, ...) |
597 | { | |
598 | va_list args; | |
599 | char buf[100]; | |
600 | ||
601 | va_start(args, fmt); | |
602 | vsnprintf(buf, sizeof(buf), fmt, args); | |
603 | va_end(args); | |
604 | printk(KERN_ERR "FIX %s: %s\n", s->name, buf); | |
605 | } | |
606 | ||
607 | static void print_trailer(struct kmem_cache *s, struct page *page, u8 *p) | |
81819f0f CL |
608 | { |
609 | unsigned int off; /* Offset of last byte */ | |
a973e9dd | 610 | u8 *addr = page_address(page); |
24922684 CL |
611 | |
612 | print_tracking(s, p); | |
613 | ||
614 | print_page_info(page); | |
615 | ||
616 | printk(KERN_ERR "INFO: Object 0x%p @offset=%tu fp=0x%p\n\n", | |
617 | p, p - addr, get_freepointer(s, p)); | |
618 | ||
619 | if (p > addr + 16) | |
ffc79d28 | 620 | print_section("Bytes b4 ", p - 16, 16); |
81819f0f | 621 | |
3b0efdfa | 622 | print_section("Object ", p, min_t(unsigned long, s->object_size, |
ffc79d28 | 623 | PAGE_SIZE)); |
81819f0f | 624 | if (s->flags & SLAB_RED_ZONE) |
3b0efdfa CL |
625 | print_section("Redzone ", p + s->object_size, |
626 | s->inuse - s->object_size); | |
81819f0f | 627 | |
81819f0f CL |
628 | if (s->offset) |
629 | off = s->offset + sizeof(void *); | |
630 | else | |
631 | off = s->inuse; | |
632 | ||
24922684 | 633 | if (s->flags & SLAB_STORE_USER) |
81819f0f | 634 | off += 2 * sizeof(struct track); |
81819f0f CL |
635 | |
636 | if (off != s->size) | |
637 | /* Beginning of the filler is the free pointer */ | |
ffc79d28 | 638 | print_section("Padding ", p + off, s->size - off); |
24922684 CL |
639 | |
640 | dump_stack(); | |
81819f0f CL |
641 | } |
642 | ||
643 | static void object_err(struct kmem_cache *s, struct page *page, | |
644 | u8 *object, char *reason) | |
645 | { | |
3dc50637 | 646 | slab_bug(s, "%s", reason); |
24922684 | 647 | print_trailer(s, page, object); |
81819f0f CL |
648 | } |
649 | ||
d0e0ac97 CG |
650 | static void slab_err(struct kmem_cache *s, struct page *page, |
651 | const char *fmt, ...) | |
81819f0f CL |
652 | { |
653 | va_list args; | |
654 | char buf[100]; | |
655 | ||
24922684 CL |
656 | va_start(args, fmt); |
657 | vsnprintf(buf, sizeof(buf), fmt, args); | |
81819f0f | 658 | va_end(args); |
3dc50637 | 659 | slab_bug(s, "%s", buf); |
24922684 | 660 | print_page_info(page); |
81819f0f CL |
661 | dump_stack(); |
662 | } | |
663 | ||
f7cb1933 | 664 | static void init_object(struct kmem_cache *s, void *object, u8 val) |
81819f0f CL |
665 | { |
666 | u8 *p = object; | |
667 | ||
668 | if (s->flags & __OBJECT_POISON) { | |
3b0efdfa CL |
669 | memset(p, POISON_FREE, s->object_size - 1); |
670 | p[s->object_size - 1] = POISON_END; | |
81819f0f CL |
671 | } |
672 | ||
673 | if (s->flags & SLAB_RED_ZONE) | |
3b0efdfa | 674 | memset(p + s->object_size, val, s->inuse - s->object_size); |
81819f0f CL |
675 | } |
676 | ||
24922684 CL |
677 | static void restore_bytes(struct kmem_cache *s, char *message, u8 data, |
678 | void *from, void *to) | |
679 | { | |
680 | slab_fix(s, "Restoring 0x%p-0x%p=0x%x\n", from, to - 1, data); | |
681 | memset(from, data, to - from); | |
682 | } | |
683 | ||
684 | static int check_bytes_and_report(struct kmem_cache *s, struct page *page, | |
685 | u8 *object, char *what, | |
06428780 | 686 | u8 *start, unsigned int value, unsigned int bytes) |
24922684 CL |
687 | { |
688 | u8 *fault; | |
689 | u8 *end; | |
690 | ||
79824820 | 691 | fault = memchr_inv(start, value, bytes); |
24922684 CL |
692 | if (!fault) |
693 | return 1; | |
694 | ||
695 | end = start + bytes; | |
696 | while (end > fault && end[-1] == value) | |
697 | end--; | |
698 | ||
699 | slab_bug(s, "%s overwritten", what); | |
700 | printk(KERN_ERR "INFO: 0x%p-0x%p. First byte 0x%x instead of 0x%x\n", | |
701 | fault, end - 1, fault[0], value); | |
702 | print_trailer(s, page, object); | |
703 | ||
704 | restore_bytes(s, what, value, fault, end); | |
705 | return 0; | |
81819f0f CL |
706 | } |
707 | ||
81819f0f CL |
708 | /* |
709 | * Object layout: | |
710 | * | |
711 | * object address | |
712 | * Bytes of the object to be managed. | |
713 | * If the freepointer may overlay the object then the free | |
714 | * pointer is the first word of the object. | |
672bba3a | 715 | * |
81819f0f CL |
716 | * Poisoning uses 0x6b (POISON_FREE) and the last byte is |
717 | * 0xa5 (POISON_END) | |
718 | * | |
3b0efdfa | 719 | * object + s->object_size |
81819f0f | 720 | * Padding to reach word boundary. This is also used for Redzoning. |
672bba3a | 721 | * Padding is extended by another word if Redzoning is enabled and |
3b0efdfa | 722 | * object_size == inuse. |
672bba3a | 723 | * |
81819f0f CL |
724 | * We fill with 0xbb (RED_INACTIVE) for inactive objects and with |
725 | * 0xcc (RED_ACTIVE) for objects in use. | |
726 | * | |
727 | * object + s->inuse | |
672bba3a CL |
728 | * Meta data starts here. |
729 | * | |
81819f0f CL |
730 | * A. Free pointer (if we cannot overwrite object on free) |
731 | * B. Tracking data for SLAB_STORE_USER | |
672bba3a | 732 | * C. Padding to reach required alignment boundary or at mininum |
6446faa2 | 733 | * one word if debugging is on to be able to detect writes |
672bba3a CL |
734 | * before the word boundary. |
735 | * | |
736 | * Padding is done using 0x5a (POISON_INUSE) | |
81819f0f CL |
737 | * |
738 | * object + s->size | |
672bba3a | 739 | * Nothing is used beyond s->size. |
81819f0f | 740 | * |
3b0efdfa | 741 | * If slabcaches are merged then the object_size and inuse boundaries are mostly |
672bba3a | 742 | * ignored. And therefore no slab options that rely on these boundaries |
81819f0f CL |
743 | * may be used with merged slabcaches. |
744 | */ | |
745 | ||
81819f0f CL |
746 | static int check_pad_bytes(struct kmem_cache *s, struct page *page, u8 *p) |
747 | { | |
748 | unsigned long off = s->inuse; /* The end of info */ | |
749 | ||
750 | if (s->offset) | |
751 | /* Freepointer is placed after the object. */ | |
752 | off += sizeof(void *); | |
753 | ||
754 | if (s->flags & SLAB_STORE_USER) | |
755 | /* We also have user information there */ | |
756 | off += 2 * sizeof(struct track); | |
757 | ||
758 | if (s->size == off) | |
759 | return 1; | |
760 | ||
24922684 CL |
761 | return check_bytes_and_report(s, page, p, "Object padding", |
762 | p + off, POISON_INUSE, s->size - off); | |
81819f0f CL |
763 | } |
764 | ||
39b26464 | 765 | /* Check the pad bytes at the end of a slab page */ |
81819f0f CL |
766 | static int slab_pad_check(struct kmem_cache *s, struct page *page) |
767 | { | |
24922684 CL |
768 | u8 *start; |
769 | u8 *fault; | |
770 | u8 *end; | |
771 | int length; | |
772 | int remainder; | |
81819f0f CL |
773 | |
774 | if (!(s->flags & SLAB_POISON)) | |
775 | return 1; | |
776 | ||
a973e9dd | 777 | start = page_address(page); |
ab9a0f19 | 778 | length = (PAGE_SIZE << compound_order(page)) - s->reserved; |
39b26464 CL |
779 | end = start + length; |
780 | remainder = length % s->size; | |
81819f0f CL |
781 | if (!remainder) |
782 | return 1; | |
783 | ||
79824820 | 784 | fault = memchr_inv(end - remainder, POISON_INUSE, remainder); |
24922684 CL |
785 | if (!fault) |
786 | return 1; | |
787 | while (end > fault && end[-1] == POISON_INUSE) | |
788 | end--; | |
789 | ||
790 | slab_err(s, page, "Padding overwritten. 0x%p-0x%p", fault, end - 1); | |
ffc79d28 | 791 | print_section("Padding ", end - remainder, remainder); |
24922684 | 792 | |
8a3d271d | 793 | restore_bytes(s, "slab padding", POISON_INUSE, end - remainder, end); |
24922684 | 794 | return 0; |
81819f0f CL |
795 | } |
796 | ||
797 | static int check_object(struct kmem_cache *s, struct page *page, | |
f7cb1933 | 798 | void *object, u8 val) |
81819f0f CL |
799 | { |
800 | u8 *p = object; | |
3b0efdfa | 801 | u8 *endobject = object + s->object_size; |
81819f0f CL |
802 | |
803 | if (s->flags & SLAB_RED_ZONE) { | |
24922684 | 804 | if (!check_bytes_and_report(s, page, object, "Redzone", |
3b0efdfa | 805 | endobject, val, s->inuse - s->object_size)) |
81819f0f | 806 | return 0; |
81819f0f | 807 | } else { |
3b0efdfa | 808 | if ((s->flags & SLAB_POISON) && s->object_size < s->inuse) { |
3adbefee | 809 | check_bytes_and_report(s, page, p, "Alignment padding", |
d0e0ac97 CG |
810 | endobject, POISON_INUSE, |
811 | s->inuse - s->object_size); | |
3adbefee | 812 | } |
81819f0f CL |
813 | } |
814 | ||
815 | if (s->flags & SLAB_POISON) { | |
f7cb1933 | 816 | if (val != SLUB_RED_ACTIVE && (s->flags & __OBJECT_POISON) && |
24922684 | 817 | (!check_bytes_and_report(s, page, p, "Poison", p, |
3b0efdfa | 818 | POISON_FREE, s->object_size - 1) || |
24922684 | 819 | !check_bytes_and_report(s, page, p, "Poison", |
3b0efdfa | 820 | p + s->object_size - 1, POISON_END, 1))) |
81819f0f | 821 | return 0; |
81819f0f CL |
822 | /* |
823 | * check_pad_bytes cleans up on its own. | |
824 | */ | |
825 | check_pad_bytes(s, page, p); | |
826 | } | |
827 | ||
f7cb1933 | 828 | if (!s->offset && val == SLUB_RED_ACTIVE) |
81819f0f CL |
829 | /* |
830 | * Object and freepointer overlap. Cannot check | |
831 | * freepointer while object is allocated. | |
832 | */ | |
833 | return 1; | |
834 | ||
835 | /* Check free pointer validity */ | |
836 | if (!check_valid_pointer(s, page, get_freepointer(s, p))) { | |
837 | object_err(s, page, p, "Freepointer corrupt"); | |
838 | /* | |
9f6c708e | 839 | * No choice but to zap it and thus lose the remainder |
81819f0f | 840 | * of the free objects in this slab. May cause |
672bba3a | 841 | * another error because the object count is now wrong. |
81819f0f | 842 | */ |
a973e9dd | 843 | set_freepointer(s, p, NULL); |
81819f0f CL |
844 | return 0; |
845 | } | |
846 | return 1; | |
847 | } | |
848 | ||
849 | static int check_slab(struct kmem_cache *s, struct page *page) | |
850 | { | |
39b26464 CL |
851 | int maxobj; |
852 | ||
81819f0f CL |
853 | VM_BUG_ON(!irqs_disabled()); |
854 | ||
855 | if (!PageSlab(page)) { | |
24922684 | 856 | slab_err(s, page, "Not a valid slab page"); |
81819f0f CL |
857 | return 0; |
858 | } | |
39b26464 | 859 | |
ab9a0f19 | 860 | maxobj = order_objects(compound_order(page), s->size, s->reserved); |
39b26464 CL |
861 | if (page->objects > maxobj) { |
862 | slab_err(s, page, "objects %u > max %u", | |
863 | s->name, page->objects, maxobj); | |
864 | return 0; | |
865 | } | |
866 | if (page->inuse > page->objects) { | |
24922684 | 867 | slab_err(s, page, "inuse %u > max %u", |
39b26464 | 868 | s->name, page->inuse, page->objects); |
81819f0f CL |
869 | return 0; |
870 | } | |
871 | /* Slab_pad_check fixes things up after itself */ | |
872 | slab_pad_check(s, page); | |
873 | return 1; | |
874 | } | |
875 | ||
876 | /* | |
672bba3a CL |
877 | * Determine if a certain object on a page is on the freelist. Must hold the |
878 | * slab lock to guarantee that the chains are in a consistent state. | |
81819f0f CL |
879 | */ |
880 | static int on_freelist(struct kmem_cache *s, struct page *page, void *search) | |
881 | { | |
882 | int nr = 0; | |
881db7fb | 883 | void *fp; |
81819f0f | 884 | void *object = NULL; |
224a88be | 885 | unsigned long max_objects; |
81819f0f | 886 | |
881db7fb | 887 | fp = page->freelist; |
39b26464 | 888 | while (fp && nr <= page->objects) { |
81819f0f CL |
889 | if (fp == search) |
890 | return 1; | |
891 | if (!check_valid_pointer(s, page, fp)) { | |
892 | if (object) { | |
893 | object_err(s, page, object, | |
894 | "Freechain corrupt"); | |
a973e9dd | 895 | set_freepointer(s, object, NULL); |
81819f0f | 896 | } else { |
24922684 | 897 | slab_err(s, page, "Freepointer corrupt"); |
a973e9dd | 898 | page->freelist = NULL; |
39b26464 | 899 | page->inuse = page->objects; |
24922684 | 900 | slab_fix(s, "Freelist cleared"); |
81819f0f CL |
901 | return 0; |
902 | } | |
903 | break; | |
904 | } | |
905 | object = fp; | |
906 | fp = get_freepointer(s, object); | |
907 | nr++; | |
908 | } | |
909 | ||
ab9a0f19 | 910 | max_objects = order_objects(compound_order(page), s->size, s->reserved); |
210b5c06 CG |
911 | if (max_objects > MAX_OBJS_PER_PAGE) |
912 | max_objects = MAX_OBJS_PER_PAGE; | |
224a88be CL |
913 | |
914 | if (page->objects != max_objects) { | |
915 | slab_err(s, page, "Wrong number of objects. Found %d but " | |
916 | "should be %d", page->objects, max_objects); | |
917 | page->objects = max_objects; | |
918 | slab_fix(s, "Number of objects adjusted."); | |
919 | } | |
39b26464 | 920 | if (page->inuse != page->objects - nr) { |
70d71228 | 921 | slab_err(s, page, "Wrong object count. Counter is %d but " |
39b26464 CL |
922 | "counted were %d", page->inuse, page->objects - nr); |
923 | page->inuse = page->objects - nr; | |
24922684 | 924 | slab_fix(s, "Object count adjusted."); |
81819f0f CL |
925 | } |
926 | return search == NULL; | |
927 | } | |
928 | ||
0121c619 CL |
929 | static void trace(struct kmem_cache *s, struct page *page, void *object, |
930 | int alloc) | |
3ec09742 CL |
931 | { |
932 | if (s->flags & SLAB_TRACE) { | |
933 | printk(KERN_INFO "TRACE %s %s 0x%p inuse=%d fp=0x%p\n", | |
934 | s->name, | |
935 | alloc ? "alloc" : "free", | |
936 | object, page->inuse, | |
937 | page->freelist); | |
938 | ||
939 | if (!alloc) | |
d0e0ac97 CG |
940 | print_section("Object ", (void *)object, |
941 | s->object_size); | |
3ec09742 CL |
942 | |
943 | dump_stack(); | |
944 | } | |
945 | } | |
946 | ||
c016b0bd CL |
947 | /* |
948 | * Hooks for other subsystems that check memory allocations. In a typical | |
949 | * production configuration these hooks all should produce no code at all. | |
950 | */ | |
d56791b3 RB |
951 | static inline void kmalloc_large_node_hook(void *ptr, size_t size, gfp_t flags) |
952 | { | |
953 | kmemleak_alloc(ptr, size, 1, flags); | |
954 | } | |
955 | ||
956 | static inline void kfree_hook(const void *x) | |
957 | { | |
958 | kmemleak_free(x); | |
959 | } | |
960 | ||
c016b0bd CL |
961 | static inline int slab_pre_alloc_hook(struct kmem_cache *s, gfp_t flags) |
962 | { | |
c1d50836 | 963 | flags &= gfp_allowed_mask; |
c016b0bd CL |
964 | lockdep_trace_alloc(flags); |
965 | might_sleep_if(flags & __GFP_WAIT); | |
966 | ||
3b0efdfa | 967 | return should_failslab(s->object_size, flags, s->flags); |
c016b0bd CL |
968 | } |
969 | ||
d0e0ac97 CG |
970 | static inline void slab_post_alloc_hook(struct kmem_cache *s, |
971 | gfp_t flags, void *object) | |
c016b0bd | 972 | { |
c1d50836 | 973 | flags &= gfp_allowed_mask; |
b3d41885 | 974 | kmemcheck_slab_alloc(s, flags, object, slab_ksize(s)); |
3b0efdfa | 975 | kmemleak_alloc_recursive(object, s->object_size, 1, s->flags, flags); |
c016b0bd CL |
976 | } |
977 | ||
978 | static inline void slab_free_hook(struct kmem_cache *s, void *x) | |
979 | { | |
980 | kmemleak_free_recursive(x, s->flags); | |
c016b0bd | 981 | |
d3f661d6 | 982 | /* |
d1756174 | 983 | * Trouble is that we may no longer disable interrupts in the fast path |
d3f661d6 CL |
984 | * So in order to make the debug calls that expect irqs to be |
985 | * disabled we need to disable interrupts temporarily. | |
986 | */ | |
987 | #if defined(CONFIG_KMEMCHECK) || defined(CONFIG_LOCKDEP) | |
988 | { | |
989 | unsigned long flags; | |
990 | ||
991 | local_irq_save(flags); | |
3b0efdfa CL |
992 | kmemcheck_slab_free(s, x, s->object_size); |
993 | debug_check_no_locks_freed(x, s->object_size); | |
d3f661d6 CL |
994 | local_irq_restore(flags); |
995 | } | |
996 | #endif | |
f9b615de | 997 | if (!(s->flags & SLAB_DEBUG_OBJECTS)) |
3b0efdfa | 998 | debug_check_no_obj_freed(x, s->object_size); |
c016b0bd CL |
999 | } |
1000 | ||
643b1138 | 1001 | /* |
672bba3a | 1002 | * Tracking of fully allocated slabs for debugging purposes. |
643b1138 | 1003 | */ |
5cc6eee8 CL |
1004 | static void add_full(struct kmem_cache *s, |
1005 | struct kmem_cache_node *n, struct page *page) | |
643b1138 | 1006 | { |
5cc6eee8 CL |
1007 | if (!(s->flags & SLAB_STORE_USER)) |
1008 | return; | |
1009 | ||
255d0884 | 1010 | lockdep_assert_held(&n->list_lock); |
643b1138 | 1011 | list_add(&page->lru, &n->full); |
643b1138 CL |
1012 | } |
1013 | ||
c65c1877 | 1014 | static void remove_full(struct kmem_cache *s, struct kmem_cache_node *n, struct page *page) |
643b1138 | 1015 | { |
643b1138 CL |
1016 | if (!(s->flags & SLAB_STORE_USER)) |
1017 | return; | |
1018 | ||
255d0884 | 1019 | lockdep_assert_held(&n->list_lock); |
643b1138 | 1020 | list_del(&page->lru); |
643b1138 CL |
1021 | } |
1022 | ||
0f389ec6 CL |
1023 | /* Tracking of the number of slabs for debugging purposes */ |
1024 | static inline unsigned long slabs_node(struct kmem_cache *s, int node) | |
1025 | { | |
1026 | struct kmem_cache_node *n = get_node(s, node); | |
1027 | ||
1028 | return atomic_long_read(&n->nr_slabs); | |
1029 | } | |
1030 | ||
26c02cf0 AB |
1031 | static inline unsigned long node_nr_slabs(struct kmem_cache_node *n) |
1032 | { | |
1033 | return atomic_long_read(&n->nr_slabs); | |
1034 | } | |
1035 | ||
205ab99d | 1036 | static inline void inc_slabs_node(struct kmem_cache *s, int node, int objects) |
0f389ec6 CL |
1037 | { |
1038 | struct kmem_cache_node *n = get_node(s, node); | |
1039 | ||
1040 | /* | |
1041 | * May be called early in order to allocate a slab for the | |
1042 | * kmem_cache_node structure. Solve the chicken-egg | |
1043 | * dilemma by deferring the increment of the count during | |
1044 | * bootstrap (see early_kmem_cache_node_alloc). | |
1045 | */ | |
338b2642 | 1046 | if (likely(n)) { |
0f389ec6 | 1047 | atomic_long_inc(&n->nr_slabs); |
205ab99d CL |
1048 | atomic_long_add(objects, &n->total_objects); |
1049 | } | |
0f389ec6 | 1050 | } |
205ab99d | 1051 | static inline void dec_slabs_node(struct kmem_cache *s, int node, int objects) |
0f389ec6 CL |
1052 | { |
1053 | struct kmem_cache_node *n = get_node(s, node); | |
1054 | ||
1055 | atomic_long_dec(&n->nr_slabs); | |
205ab99d | 1056 | atomic_long_sub(objects, &n->total_objects); |
0f389ec6 CL |
1057 | } |
1058 | ||
1059 | /* Object debug checks for alloc/free paths */ | |
3ec09742 CL |
1060 | static void setup_object_debug(struct kmem_cache *s, struct page *page, |
1061 | void *object) | |
1062 | { | |
1063 | if (!(s->flags & (SLAB_STORE_USER|SLAB_RED_ZONE|__OBJECT_POISON))) | |
1064 | return; | |
1065 | ||
f7cb1933 | 1066 | init_object(s, object, SLUB_RED_INACTIVE); |
3ec09742 CL |
1067 | init_tracking(s, object); |
1068 | } | |
1069 | ||
d0e0ac97 CG |
1070 | static noinline int alloc_debug_processing(struct kmem_cache *s, |
1071 | struct page *page, | |
ce71e27c | 1072 | void *object, unsigned long addr) |
81819f0f CL |
1073 | { |
1074 | if (!check_slab(s, page)) | |
1075 | goto bad; | |
1076 | ||
81819f0f CL |
1077 | if (!check_valid_pointer(s, page, object)) { |
1078 | object_err(s, page, object, "Freelist Pointer check fails"); | |
70d71228 | 1079 | goto bad; |
81819f0f CL |
1080 | } |
1081 | ||
f7cb1933 | 1082 | if (!check_object(s, page, object, SLUB_RED_INACTIVE)) |
81819f0f | 1083 | goto bad; |
81819f0f | 1084 | |
3ec09742 CL |
1085 | /* Success perform special debug activities for allocs */ |
1086 | if (s->flags & SLAB_STORE_USER) | |
1087 | set_track(s, object, TRACK_ALLOC, addr); | |
1088 | trace(s, page, object, 1); | |
f7cb1933 | 1089 | init_object(s, object, SLUB_RED_ACTIVE); |
81819f0f | 1090 | return 1; |
3ec09742 | 1091 | |
81819f0f CL |
1092 | bad: |
1093 | if (PageSlab(page)) { | |
1094 | /* | |
1095 | * If this is a slab page then lets do the best we can | |
1096 | * to avoid issues in the future. Marking all objects | |
672bba3a | 1097 | * as used avoids touching the remaining objects. |
81819f0f | 1098 | */ |
24922684 | 1099 | slab_fix(s, "Marking all objects used"); |
39b26464 | 1100 | page->inuse = page->objects; |
a973e9dd | 1101 | page->freelist = NULL; |
81819f0f CL |
1102 | } |
1103 | return 0; | |
1104 | } | |
1105 | ||
19c7ff9e CL |
1106 | static noinline struct kmem_cache_node *free_debug_processing( |
1107 | struct kmem_cache *s, struct page *page, void *object, | |
1108 | unsigned long addr, unsigned long *flags) | |
81819f0f | 1109 | { |
19c7ff9e | 1110 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); |
5c2e4bbb | 1111 | |
19c7ff9e | 1112 | spin_lock_irqsave(&n->list_lock, *flags); |
881db7fb CL |
1113 | slab_lock(page); |
1114 | ||
81819f0f CL |
1115 | if (!check_slab(s, page)) |
1116 | goto fail; | |
1117 | ||
1118 | if (!check_valid_pointer(s, page, object)) { | |
70d71228 | 1119 | slab_err(s, page, "Invalid object pointer 0x%p", object); |
81819f0f CL |
1120 | goto fail; |
1121 | } | |
1122 | ||
1123 | if (on_freelist(s, page, object)) { | |
24922684 | 1124 | object_err(s, page, object, "Object already free"); |
81819f0f CL |
1125 | goto fail; |
1126 | } | |
1127 | ||
f7cb1933 | 1128 | if (!check_object(s, page, object, SLUB_RED_ACTIVE)) |
5c2e4bbb | 1129 | goto out; |
81819f0f | 1130 | |
1b4f59e3 | 1131 | if (unlikely(s != page->slab_cache)) { |
3adbefee | 1132 | if (!PageSlab(page)) { |
70d71228 CL |
1133 | slab_err(s, page, "Attempt to free object(0x%p) " |
1134 | "outside of slab", object); | |
1b4f59e3 | 1135 | } else if (!page->slab_cache) { |
81819f0f | 1136 | printk(KERN_ERR |
70d71228 | 1137 | "SLUB <none>: no slab for object 0x%p.\n", |
81819f0f | 1138 | object); |
70d71228 | 1139 | dump_stack(); |
06428780 | 1140 | } else |
24922684 CL |
1141 | object_err(s, page, object, |
1142 | "page slab pointer corrupt."); | |
81819f0f CL |
1143 | goto fail; |
1144 | } | |
3ec09742 | 1145 | |
3ec09742 CL |
1146 | if (s->flags & SLAB_STORE_USER) |
1147 | set_track(s, object, TRACK_FREE, addr); | |
1148 | trace(s, page, object, 0); | |
f7cb1933 | 1149 | init_object(s, object, SLUB_RED_INACTIVE); |
5c2e4bbb | 1150 | out: |
881db7fb | 1151 | slab_unlock(page); |
19c7ff9e CL |
1152 | /* |
1153 | * Keep node_lock to preserve integrity | |
1154 | * until the object is actually freed | |
1155 | */ | |
1156 | return n; | |
3ec09742 | 1157 | |
81819f0f | 1158 | fail: |
19c7ff9e CL |
1159 | slab_unlock(page); |
1160 | spin_unlock_irqrestore(&n->list_lock, *flags); | |
24922684 | 1161 | slab_fix(s, "Object at 0x%p not freed", object); |
19c7ff9e | 1162 | return NULL; |
81819f0f CL |
1163 | } |
1164 | ||
41ecc55b CL |
1165 | static int __init setup_slub_debug(char *str) |
1166 | { | |
f0630fff CL |
1167 | slub_debug = DEBUG_DEFAULT_FLAGS; |
1168 | if (*str++ != '=' || !*str) | |
1169 | /* | |
1170 | * No options specified. Switch on full debugging. | |
1171 | */ | |
1172 | goto out; | |
1173 | ||
1174 | if (*str == ',') | |
1175 | /* | |
1176 | * No options but restriction on slabs. This means full | |
1177 | * debugging for slabs matching a pattern. | |
1178 | */ | |
1179 | goto check_slabs; | |
1180 | ||
fa5ec8a1 DR |
1181 | if (tolower(*str) == 'o') { |
1182 | /* | |
1183 | * Avoid enabling debugging on caches if its minimum order | |
1184 | * would increase as a result. | |
1185 | */ | |
1186 | disable_higher_order_debug = 1; | |
1187 | goto out; | |
1188 | } | |
1189 | ||
f0630fff CL |
1190 | slub_debug = 0; |
1191 | if (*str == '-') | |
1192 | /* | |
1193 | * Switch off all debugging measures. | |
1194 | */ | |
1195 | goto out; | |
1196 | ||
1197 | /* | |
1198 | * Determine which debug features should be switched on | |
1199 | */ | |
06428780 | 1200 | for (; *str && *str != ','; str++) { |
f0630fff CL |
1201 | switch (tolower(*str)) { |
1202 | case 'f': | |
1203 | slub_debug |= SLAB_DEBUG_FREE; | |
1204 | break; | |
1205 | case 'z': | |
1206 | slub_debug |= SLAB_RED_ZONE; | |
1207 | break; | |
1208 | case 'p': | |
1209 | slub_debug |= SLAB_POISON; | |
1210 | break; | |
1211 | case 'u': | |
1212 | slub_debug |= SLAB_STORE_USER; | |
1213 | break; | |
1214 | case 't': | |
1215 | slub_debug |= SLAB_TRACE; | |
1216 | break; | |
4c13dd3b DM |
1217 | case 'a': |
1218 | slub_debug |= SLAB_FAILSLAB; | |
1219 | break; | |
f0630fff CL |
1220 | default: |
1221 | printk(KERN_ERR "slub_debug option '%c' " | |
06428780 | 1222 | "unknown. skipped\n", *str); |
f0630fff | 1223 | } |
41ecc55b CL |
1224 | } |
1225 | ||
f0630fff | 1226 | check_slabs: |
41ecc55b CL |
1227 | if (*str == ',') |
1228 | slub_debug_slabs = str + 1; | |
f0630fff | 1229 | out: |
41ecc55b CL |
1230 | return 1; |
1231 | } | |
1232 | ||
1233 | __setup("slub_debug", setup_slub_debug); | |
1234 | ||
3b0efdfa | 1235 | static unsigned long kmem_cache_flags(unsigned long object_size, |
ba0268a8 | 1236 | unsigned long flags, const char *name, |
51cc5068 | 1237 | void (*ctor)(void *)) |
41ecc55b CL |
1238 | { |
1239 | /* | |
e153362a | 1240 | * Enable debugging if selected on the kernel commandline. |
41ecc55b | 1241 | */ |
c6f58d9b CL |
1242 | if (slub_debug && (!slub_debug_slabs || (name && |
1243 | !strncmp(slub_debug_slabs, name, strlen(slub_debug_slabs))))) | |
3de47213 | 1244 | flags |= slub_debug; |
ba0268a8 CL |
1245 | |
1246 | return flags; | |
41ecc55b CL |
1247 | } |
1248 | #else | |
3ec09742 CL |
1249 | static inline void setup_object_debug(struct kmem_cache *s, |
1250 | struct page *page, void *object) {} | |
41ecc55b | 1251 | |
3ec09742 | 1252 | static inline int alloc_debug_processing(struct kmem_cache *s, |
ce71e27c | 1253 | struct page *page, void *object, unsigned long addr) { return 0; } |
41ecc55b | 1254 | |
19c7ff9e CL |
1255 | static inline struct kmem_cache_node *free_debug_processing( |
1256 | struct kmem_cache *s, struct page *page, void *object, | |
1257 | unsigned long addr, unsigned long *flags) { return NULL; } | |
41ecc55b | 1258 | |
41ecc55b CL |
1259 | static inline int slab_pad_check(struct kmem_cache *s, struct page *page) |
1260 | { return 1; } | |
1261 | static inline int check_object(struct kmem_cache *s, struct page *page, | |
f7cb1933 | 1262 | void *object, u8 val) { return 1; } |
5cc6eee8 CL |
1263 | static inline void add_full(struct kmem_cache *s, struct kmem_cache_node *n, |
1264 | struct page *page) {} | |
c65c1877 PZ |
1265 | static inline void remove_full(struct kmem_cache *s, struct kmem_cache_node *n, |
1266 | struct page *page) {} | |
3b0efdfa | 1267 | static inline unsigned long kmem_cache_flags(unsigned long object_size, |
ba0268a8 | 1268 | unsigned long flags, const char *name, |
51cc5068 | 1269 | void (*ctor)(void *)) |
ba0268a8 CL |
1270 | { |
1271 | return flags; | |
1272 | } | |
41ecc55b | 1273 | #define slub_debug 0 |
0f389ec6 | 1274 | |
fdaa45e9 IM |
1275 | #define disable_higher_order_debug 0 |
1276 | ||
0f389ec6 CL |
1277 | static inline unsigned long slabs_node(struct kmem_cache *s, int node) |
1278 | { return 0; } | |
26c02cf0 AB |
1279 | static inline unsigned long node_nr_slabs(struct kmem_cache_node *n) |
1280 | { return 0; } | |
205ab99d CL |
1281 | static inline void inc_slabs_node(struct kmem_cache *s, int node, |
1282 | int objects) {} | |
1283 | static inline void dec_slabs_node(struct kmem_cache *s, int node, | |
1284 | int objects) {} | |
7d550c56 | 1285 | |
d56791b3 RB |
1286 | static inline void kmalloc_large_node_hook(void *ptr, size_t size, gfp_t flags) |
1287 | { | |
1288 | kmemleak_alloc(ptr, size, 1, flags); | |
1289 | } | |
1290 | ||
1291 | static inline void kfree_hook(const void *x) | |
1292 | { | |
1293 | kmemleak_free(x); | |
1294 | } | |
1295 | ||
7d550c56 CL |
1296 | static inline int slab_pre_alloc_hook(struct kmem_cache *s, gfp_t flags) |
1297 | { return 0; } | |
1298 | ||
1299 | static inline void slab_post_alloc_hook(struct kmem_cache *s, gfp_t flags, | |
d56791b3 RB |
1300 | void *object) |
1301 | { | |
1302 | kmemleak_alloc_recursive(object, s->object_size, 1, s->flags, | |
1303 | flags & gfp_allowed_mask); | |
1304 | } | |
7d550c56 | 1305 | |
d56791b3 RB |
1306 | static inline void slab_free_hook(struct kmem_cache *s, void *x) |
1307 | { | |
1308 | kmemleak_free_recursive(x, s->flags); | |
1309 | } | |
7d550c56 | 1310 | |
ab4d5ed5 | 1311 | #endif /* CONFIG_SLUB_DEBUG */ |
205ab99d | 1312 | |
81819f0f CL |
1313 | /* |
1314 | * Slab allocation and freeing | |
1315 | */ | |
65c3376a CL |
1316 | static inline struct page *alloc_slab_page(gfp_t flags, int node, |
1317 | struct kmem_cache_order_objects oo) | |
1318 | { | |
1319 | int order = oo_order(oo); | |
1320 | ||
b1eeab67 VN |
1321 | flags |= __GFP_NOTRACK; |
1322 | ||
2154a336 | 1323 | if (node == NUMA_NO_NODE) |
65c3376a CL |
1324 | return alloc_pages(flags, order); |
1325 | else | |
6b65aaf3 | 1326 | return alloc_pages_exact_node(node, flags, order); |
65c3376a CL |
1327 | } |
1328 | ||
81819f0f CL |
1329 | static struct page *allocate_slab(struct kmem_cache *s, gfp_t flags, int node) |
1330 | { | |
06428780 | 1331 | struct page *page; |
834f3d11 | 1332 | struct kmem_cache_order_objects oo = s->oo; |
ba52270d | 1333 | gfp_t alloc_gfp; |
81819f0f | 1334 | |
7e0528da CL |
1335 | flags &= gfp_allowed_mask; |
1336 | ||
1337 | if (flags & __GFP_WAIT) | |
1338 | local_irq_enable(); | |
1339 | ||
b7a49f0d | 1340 | flags |= s->allocflags; |
e12ba74d | 1341 | |
ba52270d PE |
1342 | /* |
1343 | * Let the initial higher-order allocation fail under memory pressure | |
1344 | * so we fall-back to the minimum order allocation. | |
1345 | */ | |
1346 | alloc_gfp = (flags | __GFP_NOWARN | __GFP_NORETRY) & ~__GFP_NOFAIL; | |
1347 | ||
1348 | page = alloc_slab_page(alloc_gfp, node, oo); | |
65c3376a CL |
1349 | if (unlikely(!page)) { |
1350 | oo = s->min; | |
1351 | /* | |
1352 | * Allocation may have failed due to fragmentation. | |
1353 | * Try a lower order alloc if possible | |
1354 | */ | |
1355 | page = alloc_slab_page(flags, node, oo); | |
81819f0f | 1356 | |
7e0528da CL |
1357 | if (page) |
1358 | stat(s, ORDER_FALLBACK); | |
65c3376a | 1359 | } |
5a896d9e | 1360 | |
737b719e | 1361 | if (kmemcheck_enabled && page |
5086c389 | 1362 | && !(s->flags & (SLAB_NOTRACK | DEBUG_DEFAULT_FLAGS))) { |
b1eeab67 VN |
1363 | int pages = 1 << oo_order(oo); |
1364 | ||
1365 | kmemcheck_alloc_shadow(page, oo_order(oo), flags, node); | |
1366 | ||
1367 | /* | |
1368 | * Objects from caches that have a constructor don't get | |
1369 | * cleared when they're allocated, so we need to do it here. | |
1370 | */ | |
1371 | if (s->ctor) | |
1372 | kmemcheck_mark_uninitialized_pages(page, pages); | |
1373 | else | |
1374 | kmemcheck_mark_unallocated_pages(page, pages); | |
5a896d9e VN |
1375 | } |
1376 | ||
737b719e DR |
1377 | if (flags & __GFP_WAIT) |
1378 | local_irq_disable(); | |
1379 | if (!page) | |
1380 | return NULL; | |
1381 | ||
834f3d11 | 1382 | page->objects = oo_objects(oo); |
81819f0f CL |
1383 | mod_zone_page_state(page_zone(page), |
1384 | (s->flags & SLAB_RECLAIM_ACCOUNT) ? | |
1385 | NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE, | |
65c3376a | 1386 | 1 << oo_order(oo)); |
81819f0f CL |
1387 | |
1388 | return page; | |
1389 | } | |
1390 | ||
1391 | static void setup_object(struct kmem_cache *s, struct page *page, | |
1392 | void *object) | |
1393 | { | |
3ec09742 | 1394 | setup_object_debug(s, page, object); |
4f104934 | 1395 | if (unlikely(s->ctor)) |
51cc5068 | 1396 | s->ctor(object); |
81819f0f CL |
1397 | } |
1398 | ||
1399 | static struct page *new_slab(struct kmem_cache *s, gfp_t flags, int node) | |
1400 | { | |
1401 | struct page *page; | |
81819f0f | 1402 | void *start; |
81819f0f CL |
1403 | void *last; |
1404 | void *p; | |
1f458cbf | 1405 | int order; |
81819f0f | 1406 | |
6cb06229 | 1407 | BUG_ON(flags & GFP_SLAB_BUG_MASK); |
81819f0f | 1408 | |
6cb06229 CL |
1409 | page = allocate_slab(s, |
1410 | flags & (GFP_RECLAIM_MASK | GFP_CONSTRAINT_MASK), node); | |
81819f0f CL |
1411 | if (!page) |
1412 | goto out; | |
1413 | ||
1f458cbf | 1414 | order = compound_order(page); |
205ab99d | 1415 | inc_slabs_node(s, page_to_nid(page), page->objects); |
1f458cbf | 1416 | memcg_bind_pages(s, order); |
1b4f59e3 | 1417 | page->slab_cache = s; |
c03f94cc | 1418 | __SetPageSlab(page); |
072bb0aa MG |
1419 | if (page->pfmemalloc) |
1420 | SetPageSlabPfmemalloc(page); | |
81819f0f CL |
1421 | |
1422 | start = page_address(page); | |
81819f0f CL |
1423 | |
1424 | if (unlikely(s->flags & SLAB_POISON)) | |
1f458cbf | 1425 | memset(start, POISON_INUSE, PAGE_SIZE << order); |
81819f0f CL |
1426 | |
1427 | last = start; | |
224a88be | 1428 | for_each_object(p, s, start, page->objects) { |
81819f0f CL |
1429 | setup_object(s, page, last); |
1430 | set_freepointer(s, last, p); | |
1431 | last = p; | |
1432 | } | |
1433 | setup_object(s, page, last); | |
a973e9dd | 1434 | set_freepointer(s, last, NULL); |
81819f0f CL |
1435 | |
1436 | page->freelist = start; | |
e6e82ea1 | 1437 | page->inuse = page->objects; |
8cb0a506 | 1438 | page->frozen = 1; |
81819f0f | 1439 | out: |
81819f0f CL |
1440 | return page; |
1441 | } | |
1442 | ||
1443 | static void __free_slab(struct kmem_cache *s, struct page *page) | |
1444 | { | |
834f3d11 CL |
1445 | int order = compound_order(page); |
1446 | int pages = 1 << order; | |
81819f0f | 1447 | |
af537b0a | 1448 | if (kmem_cache_debug(s)) { |
81819f0f CL |
1449 | void *p; |
1450 | ||
1451 | slab_pad_check(s, page); | |
224a88be CL |
1452 | for_each_object(p, s, page_address(page), |
1453 | page->objects) | |
f7cb1933 | 1454 | check_object(s, page, p, SLUB_RED_INACTIVE); |
81819f0f CL |
1455 | } |
1456 | ||
b1eeab67 | 1457 | kmemcheck_free_shadow(page, compound_order(page)); |
5a896d9e | 1458 | |
81819f0f CL |
1459 | mod_zone_page_state(page_zone(page), |
1460 | (s->flags & SLAB_RECLAIM_ACCOUNT) ? | |
1461 | NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE, | |
06428780 | 1462 | -pages); |
81819f0f | 1463 | |
072bb0aa | 1464 | __ClearPageSlabPfmemalloc(page); |
49bd5221 | 1465 | __ClearPageSlab(page); |
1f458cbf GC |
1466 | |
1467 | memcg_release_pages(s, order); | |
22b751c3 | 1468 | page_mapcount_reset(page); |
1eb5ac64 NP |
1469 | if (current->reclaim_state) |
1470 | current->reclaim_state->reclaimed_slab += pages; | |
d79923fa | 1471 | __free_memcg_kmem_pages(page, order); |
81819f0f CL |
1472 | } |
1473 | ||
da9a638c LJ |
1474 | #define need_reserve_slab_rcu \ |
1475 | (sizeof(((struct page *)NULL)->lru) < sizeof(struct rcu_head)) | |
1476 | ||
81819f0f CL |
1477 | static void rcu_free_slab(struct rcu_head *h) |
1478 | { | |
1479 | struct page *page; | |
1480 | ||
da9a638c LJ |
1481 | if (need_reserve_slab_rcu) |
1482 | page = virt_to_head_page(h); | |
1483 | else | |
1484 | page = container_of((struct list_head *)h, struct page, lru); | |
1485 | ||
1b4f59e3 | 1486 | __free_slab(page->slab_cache, page); |
81819f0f CL |
1487 | } |
1488 | ||
1489 | static void free_slab(struct kmem_cache *s, struct page *page) | |
1490 | { | |
1491 | if (unlikely(s->flags & SLAB_DESTROY_BY_RCU)) { | |
da9a638c LJ |
1492 | struct rcu_head *head; |
1493 | ||
1494 | if (need_reserve_slab_rcu) { | |
1495 | int order = compound_order(page); | |
1496 | int offset = (PAGE_SIZE << order) - s->reserved; | |
1497 | ||
1498 | VM_BUG_ON(s->reserved != sizeof(*head)); | |
1499 | head = page_address(page) + offset; | |
1500 | } else { | |
1501 | /* | |
1502 | * RCU free overloads the RCU head over the LRU | |
1503 | */ | |
1504 | head = (void *)&page->lru; | |
1505 | } | |
81819f0f CL |
1506 | |
1507 | call_rcu(head, rcu_free_slab); | |
1508 | } else | |
1509 | __free_slab(s, page); | |
1510 | } | |
1511 | ||
1512 | static void discard_slab(struct kmem_cache *s, struct page *page) | |
1513 | { | |
205ab99d | 1514 | dec_slabs_node(s, page_to_nid(page), page->objects); |
81819f0f CL |
1515 | free_slab(s, page); |
1516 | } | |
1517 | ||
1518 | /* | |
5cc6eee8 | 1519 | * Management of partially allocated slabs. |
81819f0f | 1520 | */ |
1e4dd946 SR |
1521 | static inline void |
1522 | __add_partial(struct kmem_cache_node *n, struct page *page, int tail) | |
81819f0f | 1523 | { |
e95eed57 | 1524 | n->nr_partial++; |
136333d1 | 1525 | if (tail == DEACTIVATE_TO_TAIL) |
7c2e132c CL |
1526 | list_add_tail(&page->lru, &n->partial); |
1527 | else | |
1528 | list_add(&page->lru, &n->partial); | |
81819f0f CL |
1529 | } |
1530 | ||
1e4dd946 SR |
1531 | static inline void add_partial(struct kmem_cache_node *n, |
1532 | struct page *page, int tail) | |
62e346a8 | 1533 | { |
c65c1877 | 1534 | lockdep_assert_held(&n->list_lock); |
1e4dd946 SR |
1535 | __add_partial(n, page, tail); |
1536 | } | |
c65c1877 | 1537 | |
1e4dd946 SR |
1538 | static inline void |
1539 | __remove_partial(struct kmem_cache_node *n, struct page *page) | |
1540 | { | |
62e346a8 CL |
1541 | list_del(&page->lru); |
1542 | n->nr_partial--; | |
1543 | } | |
1544 | ||
1e4dd946 SR |
1545 | static inline void remove_partial(struct kmem_cache_node *n, |
1546 | struct page *page) | |
1547 | { | |
1548 | lockdep_assert_held(&n->list_lock); | |
1549 | __remove_partial(n, page); | |
1550 | } | |
1551 | ||
81819f0f | 1552 | /* |
7ced3719 CL |
1553 | * Remove slab from the partial list, freeze it and |
1554 | * return the pointer to the freelist. | |
81819f0f | 1555 | * |
497b66f2 | 1556 | * Returns a list of objects or NULL if it fails. |
81819f0f | 1557 | */ |
497b66f2 | 1558 | static inline void *acquire_slab(struct kmem_cache *s, |
acd19fd1 | 1559 | struct kmem_cache_node *n, struct page *page, |
633b0764 | 1560 | int mode, int *objects) |
81819f0f | 1561 | { |
2cfb7455 CL |
1562 | void *freelist; |
1563 | unsigned long counters; | |
1564 | struct page new; | |
1565 | ||
c65c1877 PZ |
1566 | lockdep_assert_held(&n->list_lock); |
1567 | ||
2cfb7455 CL |
1568 | /* |
1569 | * Zap the freelist and set the frozen bit. | |
1570 | * The old freelist is the list of objects for the | |
1571 | * per cpu allocation list. | |
1572 | */ | |
7ced3719 CL |
1573 | freelist = page->freelist; |
1574 | counters = page->counters; | |
1575 | new.counters = counters; | |
633b0764 | 1576 | *objects = new.objects - new.inuse; |
23910c50 | 1577 | if (mode) { |
7ced3719 | 1578 | new.inuse = page->objects; |
23910c50 PE |
1579 | new.freelist = NULL; |
1580 | } else { | |
1581 | new.freelist = freelist; | |
1582 | } | |
2cfb7455 | 1583 | |
a0132ac0 | 1584 | VM_BUG_ON(new.frozen); |
7ced3719 | 1585 | new.frozen = 1; |
2cfb7455 | 1586 | |
7ced3719 | 1587 | if (!__cmpxchg_double_slab(s, page, |
2cfb7455 | 1588 | freelist, counters, |
02d7633f | 1589 | new.freelist, new.counters, |
7ced3719 | 1590 | "acquire_slab")) |
7ced3719 | 1591 | return NULL; |
2cfb7455 CL |
1592 | |
1593 | remove_partial(n, page); | |
7ced3719 | 1594 | WARN_ON(!freelist); |
49e22585 | 1595 | return freelist; |
81819f0f CL |
1596 | } |
1597 | ||
633b0764 | 1598 | static void put_cpu_partial(struct kmem_cache *s, struct page *page, int drain); |
8ba00bb6 | 1599 | static inline bool pfmemalloc_match(struct page *page, gfp_t gfpflags); |
49e22585 | 1600 | |
81819f0f | 1601 | /* |
672bba3a | 1602 | * Try to allocate a partial slab from a specific node. |
81819f0f | 1603 | */ |
8ba00bb6 JK |
1604 | static void *get_partial_node(struct kmem_cache *s, struct kmem_cache_node *n, |
1605 | struct kmem_cache_cpu *c, gfp_t flags) | |
81819f0f | 1606 | { |
49e22585 CL |
1607 | struct page *page, *page2; |
1608 | void *object = NULL; | |
633b0764 JK |
1609 | int available = 0; |
1610 | int objects; | |
81819f0f CL |
1611 | |
1612 | /* | |
1613 | * Racy check. If we mistakenly see no partial slabs then we | |
1614 | * just allocate an empty slab. If we mistakenly try to get a | |
672bba3a CL |
1615 | * partial slab and there is none available then get_partials() |
1616 | * will return NULL. | |
81819f0f CL |
1617 | */ |
1618 | if (!n || !n->nr_partial) | |
1619 | return NULL; | |
1620 | ||
1621 | spin_lock(&n->list_lock); | |
49e22585 | 1622 | list_for_each_entry_safe(page, page2, &n->partial, lru) { |
8ba00bb6 | 1623 | void *t; |
49e22585 | 1624 | |
8ba00bb6 JK |
1625 | if (!pfmemalloc_match(page, flags)) |
1626 | continue; | |
1627 | ||
633b0764 | 1628 | t = acquire_slab(s, n, page, object == NULL, &objects); |
49e22585 CL |
1629 | if (!t) |
1630 | break; | |
1631 | ||
633b0764 | 1632 | available += objects; |
12d79634 | 1633 | if (!object) { |
49e22585 | 1634 | c->page = page; |
49e22585 | 1635 | stat(s, ALLOC_FROM_PARTIAL); |
49e22585 | 1636 | object = t; |
49e22585 | 1637 | } else { |
633b0764 | 1638 | put_cpu_partial(s, page, 0); |
8028dcea | 1639 | stat(s, CPU_PARTIAL_NODE); |
49e22585 | 1640 | } |
345c905d JK |
1641 | if (!kmem_cache_has_cpu_partial(s) |
1642 | || available > s->cpu_partial / 2) | |
49e22585 CL |
1643 | break; |
1644 | ||
497b66f2 | 1645 | } |
81819f0f | 1646 | spin_unlock(&n->list_lock); |
497b66f2 | 1647 | return object; |
81819f0f CL |
1648 | } |
1649 | ||
1650 | /* | |
672bba3a | 1651 | * Get a page from somewhere. Search in increasing NUMA distances. |
81819f0f | 1652 | */ |
de3ec035 | 1653 | static void *get_any_partial(struct kmem_cache *s, gfp_t flags, |
acd19fd1 | 1654 | struct kmem_cache_cpu *c) |
81819f0f CL |
1655 | { |
1656 | #ifdef CONFIG_NUMA | |
1657 | struct zonelist *zonelist; | |
dd1a239f | 1658 | struct zoneref *z; |
54a6eb5c MG |
1659 | struct zone *zone; |
1660 | enum zone_type high_zoneidx = gfp_zone(flags); | |
497b66f2 | 1661 | void *object; |
cc9a6c87 | 1662 | unsigned int cpuset_mems_cookie; |
81819f0f CL |
1663 | |
1664 | /* | |
672bba3a CL |
1665 | * The defrag ratio allows a configuration of the tradeoffs between |
1666 | * inter node defragmentation and node local allocations. A lower | |
1667 | * defrag_ratio increases the tendency to do local allocations | |
1668 | * instead of attempting to obtain partial slabs from other nodes. | |
81819f0f | 1669 | * |
672bba3a CL |
1670 | * If the defrag_ratio is set to 0 then kmalloc() always |
1671 | * returns node local objects. If the ratio is higher then kmalloc() | |
1672 | * may return off node objects because partial slabs are obtained | |
1673 | * from other nodes and filled up. | |
81819f0f | 1674 | * |
6446faa2 | 1675 | * If /sys/kernel/slab/xx/defrag_ratio is set to 100 (which makes |
672bba3a CL |
1676 | * defrag_ratio = 1000) then every (well almost) allocation will |
1677 | * first attempt to defrag slab caches on other nodes. This means | |
1678 | * scanning over all nodes to look for partial slabs which may be | |
1679 | * expensive if we do it every time we are trying to find a slab | |
1680 | * with available objects. | |
81819f0f | 1681 | */ |
9824601e CL |
1682 | if (!s->remote_node_defrag_ratio || |
1683 | get_cycles() % 1024 > s->remote_node_defrag_ratio) | |
81819f0f CL |
1684 | return NULL; |
1685 | ||
cc9a6c87 | 1686 | do { |
d26914d1 | 1687 | cpuset_mems_cookie = read_mems_allowed_begin(); |
2a389610 | 1688 | zonelist = node_zonelist(mempolicy_slab_node(), flags); |
cc9a6c87 MG |
1689 | for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) { |
1690 | struct kmem_cache_node *n; | |
1691 | ||
1692 | n = get_node(s, zone_to_nid(zone)); | |
1693 | ||
1694 | if (n && cpuset_zone_allowed_hardwall(zone, flags) && | |
1695 | n->nr_partial > s->min_partial) { | |
8ba00bb6 | 1696 | object = get_partial_node(s, n, c, flags); |
cc9a6c87 MG |
1697 | if (object) { |
1698 | /* | |
d26914d1 MG |
1699 | * Don't check read_mems_allowed_retry() |
1700 | * here - if mems_allowed was updated in | |
1701 | * parallel, that was a harmless race | |
1702 | * between allocation and the cpuset | |
1703 | * update | |
cc9a6c87 | 1704 | */ |
cc9a6c87 MG |
1705 | return object; |
1706 | } | |
c0ff7453 | 1707 | } |
81819f0f | 1708 | } |
d26914d1 | 1709 | } while (read_mems_allowed_retry(cpuset_mems_cookie)); |
81819f0f CL |
1710 | #endif |
1711 | return NULL; | |
1712 | } | |
1713 | ||
1714 | /* | |
1715 | * Get a partial page, lock it and return it. | |
1716 | */ | |
497b66f2 | 1717 | static void *get_partial(struct kmem_cache *s, gfp_t flags, int node, |
acd19fd1 | 1718 | struct kmem_cache_cpu *c) |
81819f0f | 1719 | { |
497b66f2 | 1720 | void *object; |
2154a336 | 1721 | int searchnode = (node == NUMA_NO_NODE) ? numa_node_id() : node; |
81819f0f | 1722 | |
8ba00bb6 | 1723 | object = get_partial_node(s, get_node(s, searchnode), c, flags); |
497b66f2 CL |
1724 | if (object || node != NUMA_NO_NODE) |
1725 | return object; | |
81819f0f | 1726 | |
acd19fd1 | 1727 | return get_any_partial(s, flags, c); |
81819f0f CL |
1728 | } |
1729 | ||
8a5ec0ba CL |
1730 | #ifdef CONFIG_PREEMPT |
1731 | /* | |
1732 | * Calculate the next globally unique transaction for disambiguiation | |
1733 | * during cmpxchg. The transactions start with the cpu number and are then | |
1734 | * incremented by CONFIG_NR_CPUS. | |
1735 | */ | |
1736 | #define TID_STEP roundup_pow_of_two(CONFIG_NR_CPUS) | |
1737 | #else | |
1738 | /* | |
1739 | * No preemption supported therefore also no need to check for | |
1740 | * different cpus. | |
1741 | */ | |
1742 | #define TID_STEP 1 | |
1743 | #endif | |
1744 | ||
1745 | static inline unsigned long next_tid(unsigned long tid) | |
1746 | { | |
1747 | return tid + TID_STEP; | |
1748 | } | |
1749 | ||
1750 | static inline unsigned int tid_to_cpu(unsigned long tid) | |
1751 | { | |
1752 | return tid % TID_STEP; | |
1753 | } | |
1754 | ||
1755 | static inline unsigned long tid_to_event(unsigned long tid) | |
1756 | { | |
1757 | return tid / TID_STEP; | |
1758 | } | |
1759 | ||
1760 | static inline unsigned int init_tid(int cpu) | |
1761 | { | |
1762 | return cpu; | |
1763 | } | |
1764 | ||
1765 | static inline void note_cmpxchg_failure(const char *n, | |
1766 | const struct kmem_cache *s, unsigned long tid) | |
1767 | { | |
1768 | #ifdef SLUB_DEBUG_CMPXCHG | |
1769 | unsigned long actual_tid = __this_cpu_read(s->cpu_slab->tid); | |
1770 | ||
1771 | printk(KERN_INFO "%s %s: cmpxchg redo ", n, s->name); | |
1772 | ||
1773 | #ifdef CONFIG_PREEMPT | |
1774 | if (tid_to_cpu(tid) != tid_to_cpu(actual_tid)) | |
1775 | printk("due to cpu change %d -> %d\n", | |
1776 | tid_to_cpu(tid), tid_to_cpu(actual_tid)); | |
1777 | else | |
1778 | #endif | |
1779 | if (tid_to_event(tid) != tid_to_event(actual_tid)) | |
1780 | printk("due to cpu running other code. Event %ld->%ld\n", | |
1781 | tid_to_event(tid), tid_to_event(actual_tid)); | |
1782 | else | |
1783 | printk("for unknown reason: actual=%lx was=%lx target=%lx\n", | |
1784 | actual_tid, tid, next_tid(tid)); | |
1785 | #endif | |
4fdccdfb | 1786 | stat(s, CMPXCHG_DOUBLE_CPU_FAIL); |
8a5ec0ba CL |
1787 | } |
1788 | ||
788e1aad | 1789 | static void init_kmem_cache_cpus(struct kmem_cache *s) |
8a5ec0ba | 1790 | { |
8a5ec0ba CL |
1791 | int cpu; |
1792 | ||
1793 | for_each_possible_cpu(cpu) | |
1794 | per_cpu_ptr(s->cpu_slab, cpu)->tid = init_tid(cpu); | |
8a5ec0ba | 1795 | } |
2cfb7455 | 1796 | |
81819f0f CL |
1797 | /* |
1798 | * Remove the cpu slab | |
1799 | */ | |
d0e0ac97 CG |
1800 | static void deactivate_slab(struct kmem_cache *s, struct page *page, |
1801 | void *freelist) | |
81819f0f | 1802 | { |
2cfb7455 | 1803 | enum slab_modes { M_NONE, M_PARTIAL, M_FULL, M_FREE }; |
2cfb7455 CL |
1804 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); |
1805 | int lock = 0; | |
1806 | enum slab_modes l = M_NONE, m = M_NONE; | |
2cfb7455 | 1807 | void *nextfree; |
136333d1 | 1808 | int tail = DEACTIVATE_TO_HEAD; |
2cfb7455 CL |
1809 | struct page new; |
1810 | struct page old; | |
1811 | ||
1812 | if (page->freelist) { | |
84e554e6 | 1813 | stat(s, DEACTIVATE_REMOTE_FREES); |
136333d1 | 1814 | tail = DEACTIVATE_TO_TAIL; |
2cfb7455 CL |
1815 | } |
1816 | ||
894b8788 | 1817 | /* |
2cfb7455 CL |
1818 | * Stage one: Free all available per cpu objects back |
1819 | * to the page freelist while it is still frozen. Leave the | |
1820 | * last one. | |
1821 | * | |
1822 | * There is no need to take the list->lock because the page | |
1823 | * is still frozen. | |
1824 | */ | |
1825 | while (freelist && (nextfree = get_freepointer(s, freelist))) { | |
1826 | void *prior; | |
1827 | unsigned long counters; | |
1828 | ||
1829 | do { | |
1830 | prior = page->freelist; | |
1831 | counters = page->counters; | |
1832 | set_freepointer(s, freelist, prior); | |
1833 | new.counters = counters; | |
1834 | new.inuse--; | |
a0132ac0 | 1835 | VM_BUG_ON(!new.frozen); |
2cfb7455 | 1836 | |
1d07171c | 1837 | } while (!__cmpxchg_double_slab(s, page, |
2cfb7455 CL |
1838 | prior, counters, |
1839 | freelist, new.counters, | |
1840 | "drain percpu freelist")); | |
1841 | ||
1842 | freelist = nextfree; | |
1843 | } | |
1844 | ||
894b8788 | 1845 | /* |
2cfb7455 CL |
1846 | * Stage two: Ensure that the page is unfrozen while the |
1847 | * list presence reflects the actual number of objects | |
1848 | * during unfreeze. | |
1849 | * | |
1850 | * We setup the list membership and then perform a cmpxchg | |
1851 | * with the count. If there is a mismatch then the page | |
1852 | * is not unfrozen but the page is on the wrong list. | |
1853 | * | |
1854 | * Then we restart the process which may have to remove | |
1855 | * the page from the list that we just put it on again | |
1856 | * because the number of objects in the slab may have | |
1857 | * changed. | |
894b8788 | 1858 | */ |
2cfb7455 | 1859 | redo: |
894b8788 | 1860 | |
2cfb7455 CL |
1861 | old.freelist = page->freelist; |
1862 | old.counters = page->counters; | |
a0132ac0 | 1863 | VM_BUG_ON(!old.frozen); |
7c2e132c | 1864 | |
2cfb7455 CL |
1865 | /* Determine target state of the slab */ |
1866 | new.counters = old.counters; | |
1867 | if (freelist) { | |
1868 | new.inuse--; | |
1869 | set_freepointer(s, freelist, old.freelist); | |
1870 | new.freelist = freelist; | |
1871 | } else | |
1872 | new.freelist = old.freelist; | |
1873 | ||
1874 | new.frozen = 0; | |
1875 | ||
81107188 | 1876 | if (!new.inuse && n->nr_partial > s->min_partial) |
2cfb7455 CL |
1877 | m = M_FREE; |
1878 | else if (new.freelist) { | |
1879 | m = M_PARTIAL; | |
1880 | if (!lock) { | |
1881 | lock = 1; | |
1882 | /* | |
1883 | * Taking the spinlock removes the possiblity | |
1884 | * that acquire_slab() will see a slab page that | |
1885 | * is frozen | |
1886 | */ | |
1887 | spin_lock(&n->list_lock); | |
1888 | } | |
1889 | } else { | |
1890 | m = M_FULL; | |
1891 | if (kmem_cache_debug(s) && !lock) { | |
1892 | lock = 1; | |
1893 | /* | |
1894 | * This also ensures that the scanning of full | |
1895 | * slabs from diagnostic functions will not see | |
1896 | * any frozen slabs. | |
1897 | */ | |
1898 | spin_lock(&n->list_lock); | |
1899 | } | |
1900 | } | |
1901 | ||
1902 | if (l != m) { | |
1903 | ||
1904 | if (l == M_PARTIAL) | |
1905 | ||
1906 | remove_partial(n, page); | |
1907 | ||
1908 | else if (l == M_FULL) | |
894b8788 | 1909 | |
c65c1877 | 1910 | remove_full(s, n, page); |
2cfb7455 CL |
1911 | |
1912 | if (m == M_PARTIAL) { | |
1913 | ||
1914 | add_partial(n, page, tail); | |
136333d1 | 1915 | stat(s, tail); |
2cfb7455 CL |
1916 | |
1917 | } else if (m == M_FULL) { | |
894b8788 | 1918 | |
2cfb7455 CL |
1919 | stat(s, DEACTIVATE_FULL); |
1920 | add_full(s, n, page); | |
1921 | ||
1922 | } | |
1923 | } | |
1924 | ||
1925 | l = m; | |
1d07171c | 1926 | if (!__cmpxchg_double_slab(s, page, |
2cfb7455 CL |
1927 | old.freelist, old.counters, |
1928 | new.freelist, new.counters, | |
1929 | "unfreezing slab")) | |
1930 | goto redo; | |
1931 | ||
2cfb7455 CL |
1932 | if (lock) |
1933 | spin_unlock(&n->list_lock); | |
1934 | ||
1935 | if (m == M_FREE) { | |
1936 | stat(s, DEACTIVATE_EMPTY); | |
1937 | discard_slab(s, page); | |
1938 | stat(s, FREE_SLAB); | |
894b8788 | 1939 | } |
81819f0f CL |
1940 | } |
1941 | ||
d24ac77f JK |
1942 | /* |
1943 | * Unfreeze all the cpu partial slabs. | |
1944 | * | |
59a09917 CL |
1945 | * This function must be called with interrupts disabled |
1946 | * for the cpu using c (or some other guarantee must be there | |
1947 | * to guarantee no concurrent accesses). | |
d24ac77f | 1948 | */ |
59a09917 CL |
1949 | static void unfreeze_partials(struct kmem_cache *s, |
1950 | struct kmem_cache_cpu *c) | |
49e22585 | 1951 | { |
345c905d | 1952 | #ifdef CONFIG_SLUB_CPU_PARTIAL |
43d77867 | 1953 | struct kmem_cache_node *n = NULL, *n2 = NULL; |
9ada1934 | 1954 | struct page *page, *discard_page = NULL; |
49e22585 CL |
1955 | |
1956 | while ((page = c->partial)) { | |
49e22585 CL |
1957 | struct page new; |
1958 | struct page old; | |
1959 | ||
1960 | c->partial = page->next; | |
43d77867 JK |
1961 | |
1962 | n2 = get_node(s, page_to_nid(page)); | |
1963 | if (n != n2) { | |
1964 | if (n) | |
1965 | spin_unlock(&n->list_lock); | |
1966 | ||
1967 | n = n2; | |
1968 | spin_lock(&n->list_lock); | |
1969 | } | |
49e22585 CL |
1970 | |
1971 | do { | |
1972 | ||
1973 | old.freelist = page->freelist; | |
1974 | old.counters = page->counters; | |
a0132ac0 | 1975 | VM_BUG_ON(!old.frozen); |
49e22585 CL |
1976 | |
1977 | new.counters = old.counters; | |
1978 | new.freelist = old.freelist; | |
1979 | ||
1980 | new.frozen = 0; | |
1981 | ||
d24ac77f | 1982 | } while (!__cmpxchg_double_slab(s, page, |
49e22585 CL |
1983 | old.freelist, old.counters, |
1984 | new.freelist, new.counters, | |
1985 | "unfreezing slab")); | |
1986 | ||
43d77867 | 1987 | if (unlikely(!new.inuse && n->nr_partial > s->min_partial)) { |
9ada1934 SL |
1988 | page->next = discard_page; |
1989 | discard_page = page; | |
43d77867 JK |
1990 | } else { |
1991 | add_partial(n, page, DEACTIVATE_TO_TAIL); | |
1992 | stat(s, FREE_ADD_PARTIAL); | |
49e22585 CL |
1993 | } |
1994 | } | |
1995 | ||
1996 | if (n) | |
1997 | spin_unlock(&n->list_lock); | |
9ada1934 SL |
1998 | |
1999 | while (discard_page) { | |
2000 | page = discard_page; | |
2001 | discard_page = discard_page->next; | |
2002 | ||
2003 | stat(s, DEACTIVATE_EMPTY); | |
2004 | discard_slab(s, page); | |
2005 | stat(s, FREE_SLAB); | |
2006 | } | |
345c905d | 2007 | #endif |
49e22585 CL |
2008 | } |
2009 | ||
2010 | /* | |
2011 | * Put a page that was just frozen (in __slab_free) into a partial page | |
2012 | * slot if available. This is done without interrupts disabled and without | |
2013 | * preemption disabled. The cmpxchg is racy and may put the partial page | |
2014 | * onto a random cpus partial slot. | |
2015 | * | |
2016 | * If we did not find a slot then simply move all the partials to the | |
2017 | * per node partial list. | |
2018 | */ | |
633b0764 | 2019 | static void put_cpu_partial(struct kmem_cache *s, struct page *page, int drain) |
49e22585 | 2020 | { |
345c905d | 2021 | #ifdef CONFIG_SLUB_CPU_PARTIAL |
49e22585 CL |
2022 | struct page *oldpage; |
2023 | int pages; | |
2024 | int pobjects; | |
2025 | ||
2026 | do { | |
2027 | pages = 0; | |
2028 | pobjects = 0; | |
2029 | oldpage = this_cpu_read(s->cpu_slab->partial); | |
2030 | ||
2031 | if (oldpage) { | |
2032 | pobjects = oldpage->pobjects; | |
2033 | pages = oldpage->pages; | |
2034 | if (drain && pobjects > s->cpu_partial) { | |
2035 | unsigned long flags; | |
2036 | /* | |
2037 | * partial array is full. Move the existing | |
2038 | * set to the per node partial list. | |
2039 | */ | |
2040 | local_irq_save(flags); | |
59a09917 | 2041 | unfreeze_partials(s, this_cpu_ptr(s->cpu_slab)); |
49e22585 | 2042 | local_irq_restore(flags); |
e24fc410 | 2043 | oldpage = NULL; |
49e22585 CL |
2044 | pobjects = 0; |
2045 | pages = 0; | |
8028dcea | 2046 | stat(s, CPU_PARTIAL_DRAIN); |
49e22585 CL |
2047 | } |
2048 | } | |
2049 | ||
2050 | pages++; | |
2051 | pobjects += page->objects - page->inuse; | |
2052 | ||
2053 | page->pages = pages; | |
2054 | page->pobjects = pobjects; | |
2055 | page->next = oldpage; | |
2056 | ||
d0e0ac97 CG |
2057 | } while (this_cpu_cmpxchg(s->cpu_slab->partial, oldpage, page) |
2058 | != oldpage); | |
345c905d | 2059 | #endif |
49e22585 CL |
2060 | } |
2061 | ||
dfb4f096 | 2062 | static inline void flush_slab(struct kmem_cache *s, struct kmem_cache_cpu *c) |
81819f0f | 2063 | { |
84e554e6 | 2064 | stat(s, CPUSLAB_FLUSH); |
c17dda40 CL |
2065 | deactivate_slab(s, c->page, c->freelist); |
2066 | ||
2067 | c->tid = next_tid(c->tid); | |
2068 | c->page = NULL; | |
2069 | c->freelist = NULL; | |
81819f0f CL |
2070 | } |
2071 | ||
2072 | /* | |
2073 | * Flush cpu slab. | |
6446faa2 | 2074 | * |
81819f0f CL |
2075 | * Called from IPI handler with interrupts disabled. |
2076 | */ | |
0c710013 | 2077 | static inline void __flush_cpu_slab(struct kmem_cache *s, int cpu) |
81819f0f | 2078 | { |
9dfc6e68 | 2079 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu); |
81819f0f | 2080 | |
49e22585 CL |
2081 | if (likely(c)) { |
2082 | if (c->page) | |
2083 | flush_slab(s, c); | |
2084 | ||
59a09917 | 2085 | unfreeze_partials(s, c); |
49e22585 | 2086 | } |
81819f0f CL |
2087 | } |
2088 | ||
2089 | static void flush_cpu_slab(void *d) | |
2090 | { | |
2091 | struct kmem_cache *s = d; | |
81819f0f | 2092 | |
dfb4f096 | 2093 | __flush_cpu_slab(s, smp_processor_id()); |
81819f0f CL |
2094 | } |
2095 | ||
a8364d55 GBY |
2096 | static bool has_cpu_slab(int cpu, void *info) |
2097 | { | |
2098 | struct kmem_cache *s = info; | |
2099 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu); | |
2100 | ||
02e1a9cd | 2101 | return c->page || c->partial; |
a8364d55 GBY |
2102 | } |
2103 | ||
81819f0f CL |
2104 | static void flush_all(struct kmem_cache *s) |
2105 | { | |
a8364d55 | 2106 | on_each_cpu_cond(has_cpu_slab, flush_cpu_slab, s, 1, GFP_ATOMIC); |
81819f0f CL |
2107 | } |
2108 | ||
dfb4f096 CL |
2109 | /* |
2110 | * Check if the objects in a per cpu structure fit numa | |
2111 | * locality expectations. | |
2112 | */ | |
57d437d2 | 2113 | static inline int node_match(struct page *page, int node) |
dfb4f096 CL |
2114 | { |
2115 | #ifdef CONFIG_NUMA | |
4d7868e6 | 2116 | if (!page || (node != NUMA_NO_NODE && page_to_nid(page) != node)) |
dfb4f096 CL |
2117 | return 0; |
2118 | #endif | |
2119 | return 1; | |
2120 | } | |
2121 | ||
781b2ba6 PE |
2122 | static int count_free(struct page *page) |
2123 | { | |
2124 | return page->objects - page->inuse; | |
2125 | } | |
2126 | ||
2127 | static unsigned long count_partial(struct kmem_cache_node *n, | |
2128 | int (*get_count)(struct page *)) | |
2129 | { | |
2130 | unsigned long flags; | |
2131 | unsigned long x = 0; | |
2132 | struct page *page; | |
2133 | ||
2134 | spin_lock_irqsave(&n->list_lock, flags); | |
2135 | list_for_each_entry(page, &n->partial, lru) | |
2136 | x += get_count(page); | |
2137 | spin_unlock_irqrestore(&n->list_lock, flags); | |
2138 | return x; | |
2139 | } | |
2140 | ||
26c02cf0 AB |
2141 | static inline unsigned long node_nr_objs(struct kmem_cache_node *n) |
2142 | { | |
2143 | #ifdef CONFIG_SLUB_DEBUG | |
2144 | return atomic_long_read(&n->total_objects); | |
2145 | #else | |
2146 | return 0; | |
2147 | #endif | |
2148 | } | |
2149 | ||
781b2ba6 PE |
2150 | static noinline void |
2151 | slab_out_of_memory(struct kmem_cache *s, gfp_t gfpflags, int nid) | |
2152 | { | |
2153 | int node; | |
2154 | ||
2155 | printk(KERN_WARNING | |
2156 | "SLUB: Unable to allocate memory on node %d (gfp=0x%x)\n", | |
2157 | nid, gfpflags); | |
2158 | printk(KERN_WARNING " cache: %s, object size: %d, buffer size: %d, " | |
3b0efdfa | 2159 | "default order: %d, min order: %d\n", s->name, s->object_size, |
781b2ba6 PE |
2160 | s->size, oo_order(s->oo), oo_order(s->min)); |
2161 | ||
3b0efdfa | 2162 | if (oo_order(s->min) > get_order(s->object_size)) |
fa5ec8a1 DR |
2163 | printk(KERN_WARNING " %s debugging increased min order, use " |
2164 | "slub_debug=O to disable.\n", s->name); | |
2165 | ||
781b2ba6 PE |
2166 | for_each_online_node(node) { |
2167 | struct kmem_cache_node *n = get_node(s, node); | |
2168 | unsigned long nr_slabs; | |
2169 | unsigned long nr_objs; | |
2170 | unsigned long nr_free; | |
2171 | ||
2172 | if (!n) | |
2173 | continue; | |
2174 | ||
26c02cf0 AB |
2175 | nr_free = count_partial(n, count_free); |
2176 | nr_slabs = node_nr_slabs(n); | |
2177 | nr_objs = node_nr_objs(n); | |
781b2ba6 PE |
2178 | |
2179 | printk(KERN_WARNING | |
2180 | " node %d: slabs: %ld, objs: %ld, free: %ld\n", | |
2181 | node, nr_slabs, nr_objs, nr_free); | |
2182 | } | |
2183 | } | |
2184 | ||
497b66f2 CL |
2185 | static inline void *new_slab_objects(struct kmem_cache *s, gfp_t flags, |
2186 | int node, struct kmem_cache_cpu **pc) | |
2187 | { | |
6faa6833 | 2188 | void *freelist; |
188fd063 CL |
2189 | struct kmem_cache_cpu *c = *pc; |
2190 | struct page *page; | |
497b66f2 | 2191 | |
188fd063 | 2192 | freelist = get_partial(s, flags, node, c); |
497b66f2 | 2193 | |
188fd063 CL |
2194 | if (freelist) |
2195 | return freelist; | |
2196 | ||
2197 | page = new_slab(s, flags, node); | |
497b66f2 CL |
2198 | if (page) { |
2199 | c = __this_cpu_ptr(s->cpu_slab); | |
2200 | if (c->page) | |
2201 | flush_slab(s, c); | |
2202 | ||
2203 | /* | |
2204 | * No other reference to the page yet so we can | |
2205 | * muck around with it freely without cmpxchg | |
2206 | */ | |
6faa6833 | 2207 | freelist = page->freelist; |
497b66f2 CL |
2208 | page->freelist = NULL; |
2209 | ||
2210 | stat(s, ALLOC_SLAB); | |
497b66f2 CL |
2211 | c->page = page; |
2212 | *pc = c; | |
2213 | } else | |
6faa6833 | 2214 | freelist = NULL; |
497b66f2 | 2215 | |
6faa6833 | 2216 | return freelist; |
497b66f2 CL |
2217 | } |
2218 | ||
072bb0aa MG |
2219 | static inline bool pfmemalloc_match(struct page *page, gfp_t gfpflags) |
2220 | { | |
2221 | if (unlikely(PageSlabPfmemalloc(page))) | |
2222 | return gfp_pfmemalloc_allowed(gfpflags); | |
2223 | ||
2224 | return true; | |
2225 | } | |
2226 | ||
213eeb9f | 2227 | /* |
d0e0ac97 CG |
2228 | * Check the page->freelist of a page and either transfer the freelist to the |
2229 | * per cpu freelist or deactivate the page. | |
213eeb9f CL |
2230 | * |
2231 | * The page is still frozen if the return value is not NULL. | |
2232 | * | |
2233 | * If this function returns NULL then the page has been unfrozen. | |
d24ac77f JK |
2234 | * |
2235 | * This function must be called with interrupt disabled. | |
213eeb9f CL |
2236 | */ |
2237 | static inline void *get_freelist(struct kmem_cache *s, struct page *page) | |
2238 | { | |
2239 | struct page new; | |
2240 | unsigned long counters; | |
2241 | void *freelist; | |
2242 | ||
2243 | do { | |
2244 | freelist = page->freelist; | |
2245 | counters = page->counters; | |
6faa6833 | 2246 | |
213eeb9f | 2247 | new.counters = counters; |
a0132ac0 | 2248 | VM_BUG_ON(!new.frozen); |
213eeb9f CL |
2249 | |
2250 | new.inuse = page->objects; | |
2251 | new.frozen = freelist != NULL; | |
2252 | ||
d24ac77f | 2253 | } while (!__cmpxchg_double_slab(s, page, |
213eeb9f CL |
2254 | freelist, counters, |
2255 | NULL, new.counters, | |
2256 | "get_freelist")); | |
2257 | ||
2258 | return freelist; | |
2259 | } | |
2260 | ||
81819f0f | 2261 | /* |
894b8788 CL |
2262 | * Slow path. The lockless freelist is empty or we need to perform |
2263 | * debugging duties. | |
2264 | * | |
894b8788 CL |
2265 | * Processing is still very fast if new objects have been freed to the |
2266 | * regular freelist. In that case we simply take over the regular freelist | |
2267 | * as the lockless freelist and zap the regular freelist. | |
81819f0f | 2268 | * |
894b8788 CL |
2269 | * If that is not working then we fall back to the partial lists. We take the |
2270 | * first element of the freelist as the object to allocate now and move the | |
2271 | * rest of the freelist to the lockless freelist. | |
81819f0f | 2272 | * |
894b8788 | 2273 | * And if we were unable to get a new slab from the partial slab lists then |
6446faa2 CL |
2274 | * we need to allocate a new slab. This is the slowest path since it involves |
2275 | * a call to the page allocator and the setup of a new slab. | |
81819f0f | 2276 | */ |
ce71e27c EGM |
2277 | static void *__slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node, |
2278 | unsigned long addr, struct kmem_cache_cpu *c) | |
81819f0f | 2279 | { |
6faa6833 | 2280 | void *freelist; |
f6e7def7 | 2281 | struct page *page; |
8a5ec0ba CL |
2282 | unsigned long flags; |
2283 | ||
2284 | local_irq_save(flags); | |
2285 | #ifdef CONFIG_PREEMPT | |
2286 | /* | |
2287 | * We may have been preempted and rescheduled on a different | |
2288 | * cpu before disabling interrupts. Need to reload cpu area | |
2289 | * pointer. | |
2290 | */ | |
2291 | c = this_cpu_ptr(s->cpu_slab); | |
8a5ec0ba | 2292 | #endif |
81819f0f | 2293 | |
f6e7def7 CL |
2294 | page = c->page; |
2295 | if (!page) | |
81819f0f | 2296 | goto new_slab; |
49e22585 | 2297 | redo: |
6faa6833 | 2298 | |
57d437d2 | 2299 | if (unlikely(!node_match(page, node))) { |
e36a2652 | 2300 | stat(s, ALLOC_NODE_MISMATCH); |
f6e7def7 | 2301 | deactivate_slab(s, page, c->freelist); |
c17dda40 CL |
2302 | c->page = NULL; |
2303 | c->freelist = NULL; | |
fc59c053 CL |
2304 | goto new_slab; |
2305 | } | |
6446faa2 | 2306 | |
072bb0aa MG |
2307 | /* |
2308 | * By rights, we should be searching for a slab page that was | |
2309 | * PFMEMALLOC but right now, we are losing the pfmemalloc | |
2310 | * information when the page leaves the per-cpu allocator | |
2311 | */ | |
2312 | if (unlikely(!pfmemalloc_match(page, gfpflags))) { | |
2313 | deactivate_slab(s, page, c->freelist); | |
2314 | c->page = NULL; | |
2315 | c->freelist = NULL; | |
2316 | goto new_slab; | |
2317 | } | |
2318 | ||
73736e03 | 2319 | /* must check again c->freelist in case of cpu migration or IRQ */ |
6faa6833 CL |
2320 | freelist = c->freelist; |
2321 | if (freelist) | |
73736e03 | 2322 | goto load_freelist; |
03e404af | 2323 | |
2cfb7455 | 2324 | stat(s, ALLOC_SLOWPATH); |
03e404af | 2325 | |
f6e7def7 | 2326 | freelist = get_freelist(s, page); |
6446faa2 | 2327 | |
6faa6833 | 2328 | if (!freelist) { |
03e404af CL |
2329 | c->page = NULL; |
2330 | stat(s, DEACTIVATE_BYPASS); | |
fc59c053 | 2331 | goto new_slab; |
03e404af | 2332 | } |
6446faa2 | 2333 | |
84e554e6 | 2334 | stat(s, ALLOC_REFILL); |
6446faa2 | 2335 | |
894b8788 | 2336 | load_freelist: |
507effea CL |
2337 | /* |
2338 | * freelist is pointing to the list of objects to be used. | |
2339 | * page is pointing to the page from which the objects are obtained. | |
2340 | * That page must be frozen for per cpu allocations to work. | |
2341 | */ | |
a0132ac0 | 2342 | VM_BUG_ON(!c->page->frozen); |
6faa6833 | 2343 | c->freelist = get_freepointer(s, freelist); |
8a5ec0ba CL |
2344 | c->tid = next_tid(c->tid); |
2345 | local_irq_restore(flags); | |
6faa6833 | 2346 | return freelist; |
81819f0f | 2347 | |
81819f0f | 2348 | new_slab: |
2cfb7455 | 2349 | |
49e22585 | 2350 | if (c->partial) { |
f6e7def7 CL |
2351 | page = c->page = c->partial; |
2352 | c->partial = page->next; | |
49e22585 CL |
2353 | stat(s, CPU_PARTIAL_ALLOC); |
2354 | c->freelist = NULL; | |
2355 | goto redo; | |
81819f0f CL |
2356 | } |
2357 | ||
188fd063 | 2358 | freelist = new_slab_objects(s, gfpflags, node, &c); |
01ad8a7b | 2359 | |
f4697436 CL |
2360 | if (unlikely(!freelist)) { |
2361 | if (!(gfpflags & __GFP_NOWARN) && printk_ratelimit()) | |
2362 | slab_out_of_memory(s, gfpflags, node); | |
2cfb7455 | 2363 | |
f4697436 CL |
2364 | local_irq_restore(flags); |
2365 | return NULL; | |
81819f0f | 2366 | } |
2cfb7455 | 2367 | |
f6e7def7 | 2368 | page = c->page; |
5091b74a | 2369 | if (likely(!kmem_cache_debug(s) && pfmemalloc_match(page, gfpflags))) |
4b6f0750 | 2370 | goto load_freelist; |
2cfb7455 | 2371 | |
497b66f2 | 2372 | /* Only entered in the debug case */ |
d0e0ac97 CG |
2373 | if (kmem_cache_debug(s) && |
2374 | !alloc_debug_processing(s, page, freelist, addr)) | |
497b66f2 | 2375 | goto new_slab; /* Slab failed checks. Next slab needed */ |
894b8788 | 2376 | |
f6e7def7 | 2377 | deactivate_slab(s, page, get_freepointer(s, freelist)); |
c17dda40 CL |
2378 | c->page = NULL; |
2379 | c->freelist = NULL; | |
a71ae47a | 2380 | local_irq_restore(flags); |
6faa6833 | 2381 | return freelist; |
894b8788 CL |
2382 | } |
2383 | ||
2384 | /* | |
2385 | * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc) | |
2386 | * have the fastpath folded into their functions. So no function call | |
2387 | * overhead for requests that can be satisfied on the fastpath. | |
2388 | * | |
2389 | * The fastpath works by first checking if the lockless freelist can be used. | |
2390 | * If not then __slab_alloc is called for slow processing. | |
2391 | * | |
2392 | * Otherwise we can simply pick the next object from the lockless free list. | |
2393 | */ | |
2b847c3c | 2394 | static __always_inline void *slab_alloc_node(struct kmem_cache *s, |
ce71e27c | 2395 | gfp_t gfpflags, int node, unsigned long addr) |
894b8788 | 2396 | { |
894b8788 | 2397 | void **object; |
dfb4f096 | 2398 | struct kmem_cache_cpu *c; |
57d437d2 | 2399 | struct page *page; |
8a5ec0ba | 2400 | unsigned long tid; |
1f84260c | 2401 | |
c016b0bd | 2402 | if (slab_pre_alloc_hook(s, gfpflags)) |
773ff60e | 2403 | return NULL; |
1f84260c | 2404 | |
d79923fa | 2405 | s = memcg_kmem_get_cache(s, gfpflags); |
8a5ec0ba | 2406 | redo: |
8a5ec0ba CL |
2407 | /* |
2408 | * Must read kmem_cache cpu data via this cpu ptr. Preemption is | |
2409 | * enabled. We may switch back and forth between cpus while | |
2410 | * reading from one cpu area. That does not matter as long | |
2411 | * as we end up on the original cpu again when doing the cmpxchg. | |
7cccd80b CL |
2412 | * |
2413 | * Preemption is disabled for the retrieval of the tid because that | |
2414 | * must occur from the current processor. We cannot allow rescheduling | |
2415 | * on a different processor between the determination of the pointer | |
2416 | * and the retrieval of the tid. | |
8a5ec0ba | 2417 | */ |
7cccd80b | 2418 | preempt_disable(); |
9dfc6e68 | 2419 | c = __this_cpu_ptr(s->cpu_slab); |
8a5ec0ba | 2420 | |
8a5ec0ba CL |
2421 | /* |
2422 | * The transaction ids are globally unique per cpu and per operation on | |
2423 | * a per cpu queue. Thus they can be guarantee that the cmpxchg_double | |
2424 | * occurs on the right processor and that there was no operation on the | |
2425 | * linked list in between. | |
2426 | */ | |
2427 | tid = c->tid; | |
7cccd80b | 2428 | preempt_enable(); |
8a5ec0ba | 2429 | |
9dfc6e68 | 2430 | object = c->freelist; |
57d437d2 | 2431 | page = c->page; |
ac6434e6 | 2432 | if (unlikely(!object || !node_match(page, node))) |
dfb4f096 | 2433 | object = __slab_alloc(s, gfpflags, node, addr, c); |
894b8788 CL |
2434 | |
2435 | else { | |
0ad9500e ED |
2436 | void *next_object = get_freepointer_safe(s, object); |
2437 | ||
8a5ec0ba | 2438 | /* |
25985edc | 2439 | * The cmpxchg will only match if there was no additional |
8a5ec0ba CL |
2440 | * operation and if we are on the right processor. |
2441 | * | |
d0e0ac97 CG |
2442 | * The cmpxchg does the following atomically (without lock |
2443 | * semantics!) | |
8a5ec0ba CL |
2444 | * 1. Relocate first pointer to the current per cpu area. |
2445 | * 2. Verify that tid and freelist have not been changed | |
2446 | * 3. If they were not changed replace tid and freelist | |
2447 | * | |
d0e0ac97 CG |
2448 | * Since this is without lock semantics the protection is only |
2449 | * against code executing on this cpu *not* from access by | |
2450 | * other cpus. | |
8a5ec0ba | 2451 | */ |
933393f5 | 2452 | if (unlikely(!this_cpu_cmpxchg_double( |
8a5ec0ba CL |
2453 | s->cpu_slab->freelist, s->cpu_slab->tid, |
2454 | object, tid, | |
0ad9500e | 2455 | next_object, next_tid(tid)))) { |
8a5ec0ba CL |
2456 | |
2457 | note_cmpxchg_failure("slab_alloc", s, tid); | |
2458 | goto redo; | |
2459 | } | |
0ad9500e | 2460 | prefetch_freepointer(s, next_object); |
84e554e6 | 2461 | stat(s, ALLOC_FASTPATH); |
894b8788 | 2462 | } |
8a5ec0ba | 2463 | |
74e2134f | 2464 | if (unlikely(gfpflags & __GFP_ZERO) && object) |
3b0efdfa | 2465 | memset(object, 0, s->object_size); |
d07dbea4 | 2466 | |
c016b0bd | 2467 | slab_post_alloc_hook(s, gfpflags, object); |
5a896d9e | 2468 | |
894b8788 | 2469 | return object; |
81819f0f CL |
2470 | } |
2471 | ||
2b847c3c EG |
2472 | static __always_inline void *slab_alloc(struct kmem_cache *s, |
2473 | gfp_t gfpflags, unsigned long addr) | |
2474 | { | |
2475 | return slab_alloc_node(s, gfpflags, NUMA_NO_NODE, addr); | |
2476 | } | |
2477 | ||
81819f0f CL |
2478 | void *kmem_cache_alloc(struct kmem_cache *s, gfp_t gfpflags) |
2479 | { | |
2b847c3c | 2480 | void *ret = slab_alloc(s, gfpflags, _RET_IP_); |
5b882be4 | 2481 | |
d0e0ac97 CG |
2482 | trace_kmem_cache_alloc(_RET_IP_, ret, s->object_size, |
2483 | s->size, gfpflags); | |
5b882be4 EGM |
2484 | |
2485 | return ret; | |
81819f0f CL |
2486 | } |
2487 | EXPORT_SYMBOL(kmem_cache_alloc); | |
2488 | ||
0f24f128 | 2489 | #ifdef CONFIG_TRACING |
4a92379b RK |
2490 | void *kmem_cache_alloc_trace(struct kmem_cache *s, gfp_t gfpflags, size_t size) |
2491 | { | |
2b847c3c | 2492 | void *ret = slab_alloc(s, gfpflags, _RET_IP_); |
4a92379b RK |
2493 | trace_kmalloc(_RET_IP_, ret, size, s->size, gfpflags); |
2494 | return ret; | |
2495 | } | |
2496 | EXPORT_SYMBOL(kmem_cache_alloc_trace); | |
5b882be4 EGM |
2497 | #endif |
2498 | ||
81819f0f CL |
2499 | #ifdef CONFIG_NUMA |
2500 | void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t gfpflags, int node) | |
2501 | { | |
2b847c3c | 2502 | void *ret = slab_alloc_node(s, gfpflags, node, _RET_IP_); |
5b882be4 | 2503 | |
ca2b84cb | 2504 | trace_kmem_cache_alloc_node(_RET_IP_, ret, |
3b0efdfa | 2505 | s->object_size, s->size, gfpflags, node); |
5b882be4 EGM |
2506 | |
2507 | return ret; | |
81819f0f CL |
2508 | } |
2509 | EXPORT_SYMBOL(kmem_cache_alloc_node); | |
81819f0f | 2510 | |
0f24f128 | 2511 | #ifdef CONFIG_TRACING |
4a92379b | 2512 | void *kmem_cache_alloc_node_trace(struct kmem_cache *s, |
5b882be4 | 2513 | gfp_t gfpflags, |
4a92379b | 2514 | int node, size_t size) |
5b882be4 | 2515 | { |
2b847c3c | 2516 | void *ret = slab_alloc_node(s, gfpflags, node, _RET_IP_); |
4a92379b RK |
2517 | |
2518 | trace_kmalloc_node(_RET_IP_, ret, | |
2519 | size, s->size, gfpflags, node); | |
2520 | return ret; | |
5b882be4 | 2521 | } |
4a92379b | 2522 | EXPORT_SYMBOL(kmem_cache_alloc_node_trace); |
5b882be4 | 2523 | #endif |
5d1f57e4 | 2524 | #endif |
5b882be4 | 2525 | |
81819f0f | 2526 | /* |
894b8788 CL |
2527 | * Slow patch handling. This may still be called frequently since objects |
2528 | * have a longer lifetime than the cpu slabs in most processing loads. | |
81819f0f | 2529 | * |
894b8788 CL |
2530 | * So we still attempt to reduce cache line usage. Just take the slab |
2531 | * lock and free the item. If there is no additional partial page | |
2532 | * handling required then we can return immediately. | |
81819f0f | 2533 | */ |
894b8788 | 2534 | static void __slab_free(struct kmem_cache *s, struct page *page, |
ff12059e | 2535 | void *x, unsigned long addr) |
81819f0f CL |
2536 | { |
2537 | void *prior; | |
2538 | void **object = (void *)x; | |
2cfb7455 | 2539 | int was_frozen; |
2cfb7455 CL |
2540 | struct page new; |
2541 | unsigned long counters; | |
2542 | struct kmem_cache_node *n = NULL; | |
61728d1e | 2543 | unsigned long uninitialized_var(flags); |
81819f0f | 2544 | |
8a5ec0ba | 2545 | stat(s, FREE_SLOWPATH); |
81819f0f | 2546 | |
19c7ff9e CL |
2547 | if (kmem_cache_debug(s) && |
2548 | !(n = free_debug_processing(s, page, x, addr, &flags))) | |
80f08c19 | 2549 | return; |
6446faa2 | 2550 | |
2cfb7455 | 2551 | do { |
837d678d JK |
2552 | if (unlikely(n)) { |
2553 | spin_unlock_irqrestore(&n->list_lock, flags); | |
2554 | n = NULL; | |
2555 | } | |
2cfb7455 CL |
2556 | prior = page->freelist; |
2557 | counters = page->counters; | |
2558 | set_freepointer(s, object, prior); | |
2559 | new.counters = counters; | |
2560 | was_frozen = new.frozen; | |
2561 | new.inuse--; | |
837d678d | 2562 | if ((!new.inuse || !prior) && !was_frozen) { |
49e22585 | 2563 | |
c65c1877 | 2564 | if (kmem_cache_has_cpu_partial(s) && !prior) { |
49e22585 CL |
2565 | |
2566 | /* | |
d0e0ac97 CG |
2567 | * Slab was on no list before and will be |
2568 | * partially empty | |
2569 | * We can defer the list move and instead | |
2570 | * freeze it. | |
49e22585 CL |
2571 | */ |
2572 | new.frozen = 1; | |
2573 | ||
c65c1877 | 2574 | } else { /* Needs to be taken off a list */ |
49e22585 CL |
2575 | |
2576 | n = get_node(s, page_to_nid(page)); | |
2577 | /* | |
2578 | * Speculatively acquire the list_lock. | |
2579 | * If the cmpxchg does not succeed then we may | |
2580 | * drop the list_lock without any processing. | |
2581 | * | |
2582 | * Otherwise the list_lock will synchronize with | |
2583 | * other processors updating the list of slabs. | |
2584 | */ | |
2585 | spin_lock_irqsave(&n->list_lock, flags); | |
2586 | ||
2587 | } | |
2cfb7455 | 2588 | } |
81819f0f | 2589 | |
2cfb7455 CL |
2590 | } while (!cmpxchg_double_slab(s, page, |
2591 | prior, counters, | |
2592 | object, new.counters, | |
2593 | "__slab_free")); | |
81819f0f | 2594 | |
2cfb7455 | 2595 | if (likely(!n)) { |
49e22585 CL |
2596 | |
2597 | /* | |
2598 | * If we just froze the page then put it onto the | |
2599 | * per cpu partial list. | |
2600 | */ | |
8028dcea | 2601 | if (new.frozen && !was_frozen) { |
49e22585 | 2602 | put_cpu_partial(s, page, 1); |
8028dcea AS |
2603 | stat(s, CPU_PARTIAL_FREE); |
2604 | } | |
49e22585 | 2605 | /* |
2cfb7455 CL |
2606 | * The list lock was not taken therefore no list |
2607 | * activity can be necessary. | |
2608 | */ | |
2609 | if (was_frozen) | |
2610 | stat(s, FREE_FROZEN); | |
80f08c19 | 2611 | return; |
2cfb7455 | 2612 | } |
81819f0f | 2613 | |
837d678d JK |
2614 | if (unlikely(!new.inuse && n->nr_partial > s->min_partial)) |
2615 | goto slab_empty; | |
2616 | ||
81819f0f | 2617 | /* |
837d678d JK |
2618 | * Objects left in the slab. If it was not on the partial list before |
2619 | * then add it. | |
81819f0f | 2620 | */ |
345c905d JK |
2621 | if (!kmem_cache_has_cpu_partial(s) && unlikely(!prior)) { |
2622 | if (kmem_cache_debug(s)) | |
c65c1877 | 2623 | remove_full(s, n, page); |
837d678d JK |
2624 | add_partial(n, page, DEACTIVATE_TO_TAIL); |
2625 | stat(s, FREE_ADD_PARTIAL); | |
8ff12cfc | 2626 | } |
80f08c19 | 2627 | spin_unlock_irqrestore(&n->list_lock, flags); |
81819f0f CL |
2628 | return; |
2629 | ||
2630 | slab_empty: | |
a973e9dd | 2631 | if (prior) { |
81819f0f | 2632 | /* |
6fbabb20 | 2633 | * Slab on the partial list. |
81819f0f | 2634 | */ |
5cc6eee8 | 2635 | remove_partial(n, page); |
84e554e6 | 2636 | stat(s, FREE_REMOVE_PARTIAL); |
c65c1877 | 2637 | } else { |
6fbabb20 | 2638 | /* Slab must be on the full list */ |
c65c1877 PZ |
2639 | remove_full(s, n, page); |
2640 | } | |
2cfb7455 | 2641 | |
80f08c19 | 2642 | spin_unlock_irqrestore(&n->list_lock, flags); |
84e554e6 | 2643 | stat(s, FREE_SLAB); |
81819f0f | 2644 | discard_slab(s, page); |
81819f0f CL |
2645 | } |
2646 | ||
894b8788 CL |
2647 | /* |
2648 | * Fastpath with forced inlining to produce a kfree and kmem_cache_free that | |
2649 | * can perform fastpath freeing without additional function calls. | |
2650 | * | |
2651 | * The fastpath is only possible if we are freeing to the current cpu slab | |
2652 | * of this processor. This typically the case if we have just allocated | |
2653 | * the item before. | |
2654 | * | |
2655 | * If fastpath is not possible then fall back to __slab_free where we deal | |
2656 | * with all sorts of special processing. | |
2657 | */ | |
06428780 | 2658 | static __always_inline void slab_free(struct kmem_cache *s, |
ce71e27c | 2659 | struct page *page, void *x, unsigned long addr) |
894b8788 CL |
2660 | { |
2661 | void **object = (void *)x; | |
dfb4f096 | 2662 | struct kmem_cache_cpu *c; |
8a5ec0ba | 2663 | unsigned long tid; |
1f84260c | 2664 | |
c016b0bd CL |
2665 | slab_free_hook(s, x); |
2666 | ||
8a5ec0ba CL |
2667 | redo: |
2668 | /* | |
2669 | * Determine the currently cpus per cpu slab. | |
2670 | * The cpu may change afterward. However that does not matter since | |
2671 | * data is retrieved via this pointer. If we are on the same cpu | |
2672 | * during the cmpxchg then the free will succedd. | |
2673 | */ | |
7cccd80b | 2674 | preempt_disable(); |
9dfc6e68 | 2675 | c = __this_cpu_ptr(s->cpu_slab); |
c016b0bd | 2676 | |
8a5ec0ba | 2677 | tid = c->tid; |
7cccd80b | 2678 | preempt_enable(); |
c016b0bd | 2679 | |
442b06bc | 2680 | if (likely(page == c->page)) { |
ff12059e | 2681 | set_freepointer(s, object, c->freelist); |
8a5ec0ba | 2682 | |
933393f5 | 2683 | if (unlikely(!this_cpu_cmpxchg_double( |
8a5ec0ba CL |
2684 | s->cpu_slab->freelist, s->cpu_slab->tid, |
2685 | c->freelist, tid, | |
2686 | object, next_tid(tid)))) { | |
2687 | ||
2688 | note_cmpxchg_failure("slab_free", s, tid); | |
2689 | goto redo; | |
2690 | } | |
84e554e6 | 2691 | stat(s, FREE_FASTPATH); |
894b8788 | 2692 | } else |
ff12059e | 2693 | __slab_free(s, page, x, addr); |
894b8788 | 2694 | |
894b8788 CL |
2695 | } |
2696 | ||
81819f0f CL |
2697 | void kmem_cache_free(struct kmem_cache *s, void *x) |
2698 | { | |
b9ce5ef4 GC |
2699 | s = cache_from_obj(s, x); |
2700 | if (!s) | |
79576102 | 2701 | return; |
b9ce5ef4 | 2702 | slab_free(s, virt_to_head_page(x), x, _RET_IP_); |
ca2b84cb | 2703 | trace_kmem_cache_free(_RET_IP_, x); |
81819f0f CL |
2704 | } |
2705 | EXPORT_SYMBOL(kmem_cache_free); | |
2706 | ||
81819f0f | 2707 | /* |
672bba3a CL |
2708 | * Object placement in a slab is made very easy because we always start at |
2709 | * offset 0. If we tune the size of the object to the alignment then we can | |
2710 | * get the required alignment by putting one properly sized object after | |
2711 | * another. | |
81819f0f CL |
2712 | * |
2713 | * Notice that the allocation order determines the sizes of the per cpu | |
2714 | * caches. Each processor has always one slab available for allocations. | |
2715 | * Increasing the allocation order reduces the number of times that slabs | |
672bba3a | 2716 | * must be moved on and off the partial lists and is therefore a factor in |
81819f0f | 2717 | * locking overhead. |
81819f0f CL |
2718 | */ |
2719 | ||
2720 | /* | |
2721 | * Mininum / Maximum order of slab pages. This influences locking overhead | |
2722 | * and slab fragmentation. A higher order reduces the number of partial slabs | |
2723 | * and increases the number of allocations possible without having to | |
2724 | * take the list_lock. | |
2725 | */ | |
2726 | static int slub_min_order; | |
114e9e89 | 2727 | static int slub_max_order = PAGE_ALLOC_COSTLY_ORDER; |
9b2cd506 | 2728 | static int slub_min_objects; |
81819f0f CL |
2729 | |
2730 | /* | |
2731 | * Merge control. If this is set then no merging of slab caches will occur. | |
672bba3a | 2732 | * (Could be removed. This was introduced to pacify the merge skeptics.) |
81819f0f CL |
2733 | */ |
2734 | static int slub_nomerge; | |
2735 | ||
81819f0f CL |
2736 | /* |
2737 | * Calculate the order of allocation given an slab object size. | |
2738 | * | |
672bba3a CL |
2739 | * The order of allocation has significant impact on performance and other |
2740 | * system components. Generally order 0 allocations should be preferred since | |
2741 | * order 0 does not cause fragmentation in the page allocator. Larger objects | |
2742 | * be problematic to put into order 0 slabs because there may be too much | |
c124f5b5 | 2743 | * unused space left. We go to a higher order if more than 1/16th of the slab |
672bba3a CL |
2744 | * would be wasted. |
2745 | * | |
2746 | * In order to reach satisfactory performance we must ensure that a minimum | |
2747 | * number of objects is in one slab. Otherwise we may generate too much | |
2748 | * activity on the partial lists which requires taking the list_lock. This is | |
2749 | * less a concern for large slabs though which are rarely used. | |
81819f0f | 2750 | * |
672bba3a CL |
2751 | * slub_max_order specifies the order where we begin to stop considering the |
2752 | * number of objects in a slab as critical. If we reach slub_max_order then | |
2753 | * we try to keep the page order as low as possible. So we accept more waste | |
2754 | * of space in favor of a small page order. | |
81819f0f | 2755 | * |
672bba3a CL |
2756 | * Higher order allocations also allow the placement of more objects in a |
2757 | * slab and thereby reduce object handling overhead. If the user has | |
2758 | * requested a higher mininum order then we start with that one instead of | |
2759 | * the smallest order which will fit the object. | |
81819f0f | 2760 | */ |
5e6d444e | 2761 | static inline int slab_order(int size, int min_objects, |
ab9a0f19 | 2762 | int max_order, int fract_leftover, int reserved) |
81819f0f CL |
2763 | { |
2764 | int order; | |
2765 | int rem; | |
6300ea75 | 2766 | int min_order = slub_min_order; |
81819f0f | 2767 | |
ab9a0f19 | 2768 | if (order_objects(min_order, size, reserved) > MAX_OBJS_PER_PAGE) |
210b5c06 | 2769 | return get_order(size * MAX_OBJS_PER_PAGE) - 1; |
39b26464 | 2770 | |
6300ea75 | 2771 | for (order = max(min_order, |
5e6d444e CL |
2772 | fls(min_objects * size - 1) - PAGE_SHIFT); |
2773 | order <= max_order; order++) { | |
81819f0f | 2774 | |
5e6d444e | 2775 | unsigned long slab_size = PAGE_SIZE << order; |
81819f0f | 2776 | |
ab9a0f19 | 2777 | if (slab_size < min_objects * size + reserved) |
81819f0f CL |
2778 | continue; |
2779 | ||
ab9a0f19 | 2780 | rem = (slab_size - reserved) % size; |
81819f0f | 2781 | |
5e6d444e | 2782 | if (rem <= slab_size / fract_leftover) |
81819f0f CL |
2783 | break; |
2784 | ||
2785 | } | |
672bba3a | 2786 | |
81819f0f CL |
2787 | return order; |
2788 | } | |
2789 | ||
ab9a0f19 | 2790 | static inline int calculate_order(int size, int reserved) |
5e6d444e CL |
2791 | { |
2792 | int order; | |
2793 | int min_objects; | |
2794 | int fraction; | |
e8120ff1 | 2795 | int max_objects; |
5e6d444e CL |
2796 | |
2797 | /* | |
2798 | * Attempt to find best configuration for a slab. This | |
2799 | * works by first attempting to generate a layout with | |
2800 | * the best configuration and backing off gradually. | |
2801 | * | |
2802 | * First we reduce the acceptable waste in a slab. Then | |
2803 | * we reduce the minimum objects required in a slab. | |
2804 | */ | |
2805 | min_objects = slub_min_objects; | |
9b2cd506 CL |
2806 | if (!min_objects) |
2807 | min_objects = 4 * (fls(nr_cpu_ids) + 1); | |
ab9a0f19 | 2808 | max_objects = order_objects(slub_max_order, size, reserved); |
e8120ff1 ZY |
2809 | min_objects = min(min_objects, max_objects); |
2810 | ||
5e6d444e | 2811 | while (min_objects > 1) { |
c124f5b5 | 2812 | fraction = 16; |
5e6d444e CL |
2813 | while (fraction >= 4) { |
2814 | order = slab_order(size, min_objects, | |
ab9a0f19 | 2815 | slub_max_order, fraction, reserved); |
5e6d444e CL |
2816 | if (order <= slub_max_order) |
2817 | return order; | |
2818 | fraction /= 2; | |
2819 | } | |
5086c389 | 2820 | min_objects--; |
5e6d444e CL |
2821 | } |
2822 | ||
2823 | /* | |
2824 | * We were unable to place multiple objects in a slab. Now | |
2825 | * lets see if we can place a single object there. | |
2826 | */ | |
ab9a0f19 | 2827 | order = slab_order(size, 1, slub_max_order, 1, reserved); |
5e6d444e CL |
2828 | if (order <= slub_max_order) |
2829 | return order; | |
2830 | ||
2831 | /* | |
2832 | * Doh this slab cannot be placed using slub_max_order. | |
2833 | */ | |
ab9a0f19 | 2834 | order = slab_order(size, 1, MAX_ORDER, 1, reserved); |
818cf590 | 2835 | if (order < MAX_ORDER) |
5e6d444e CL |
2836 | return order; |
2837 | return -ENOSYS; | |
2838 | } | |
2839 | ||
5595cffc | 2840 | static void |
4053497d | 2841 | init_kmem_cache_node(struct kmem_cache_node *n) |
81819f0f CL |
2842 | { |
2843 | n->nr_partial = 0; | |
81819f0f CL |
2844 | spin_lock_init(&n->list_lock); |
2845 | INIT_LIST_HEAD(&n->partial); | |
8ab1372f | 2846 | #ifdef CONFIG_SLUB_DEBUG |
0f389ec6 | 2847 | atomic_long_set(&n->nr_slabs, 0); |
02b71b70 | 2848 | atomic_long_set(&n->total_objects, 0); |
643b1138 | 2849 | INIT_LIST_HEAD(&n->full); |
8ab1372f | 2850 | #endif |
81819f0f CL |
2851 | } |
2852 | ||
55136592 | 2853 | static inline int alloc_kmem_cache_cpus(struct kmem_cache *s) |
4c93c355 | 2854 | { |
6c182dc0 | 2855 | BUILD_BUG_ON(PERCPU_DYNAMIC_EARLY_SIZE < |
95a05b42 | 2856 | KMALLOC_SHIFT_HIGH * sizeof(struct kmem_cache_cpu)); |
4c93c355 | 2857 | |
8a5ec0ba | 2858 | /* |
d4d84fef CM |
2859 | * Must align to double word boundary for the double cmpxchg |
2860 | * instructions to work; see __pcpu_double_call_return_bool(). | |
8a5ec0ba | 2861 | */ |
d4d84fef CM |
2862 | s->cpu_slab = __alloc_percpu(sizeof(struct kmem_cache_cpu), |
2863 | 2 * sizeof(void *)); | |
8a5ec0ba CL |
2864 | |
2865 | if (!s->cpu_slab) | |
2866 | return 0; | |
2867 | ||
2868 | init_kmem_cache_cpus(s); | |
4c93c355 | 2869 | |
8a5ec0ba | 2870 | return 1; |
4c93c355 | 2871 | } |
4c93c355 | 2872 | |
51df1142 CL |
2873 | static struct kmem_cache *kmem_cache_node; |
2874 | ||
81819f0f CL |
2875 | /* |
2876 | * No kmalloc_node yet so do it by hand. We know that this is the first | |
2877 | * slab on the node for this slabcache. There are no concurrent accesses | |
2878 | * possible. | |
2879 | * | |
721ae22a ZYW |
2880 | * Note that this function only works on the kmem_cache_node |
2881 | * when allocating for the kmem_cache_node. This is used for bootstrapping | |
4c93c355 | 2882 | * memory on a fresh node that has no slab structures yet. |
81819f0f | 2883 | */ |
55136592 | 2884 | static void early_kmem_cache_node_alloc(int node) |
81819f0f CL |
2885 | { |
2886 | struct page *page; | |
2887 | struct kmem_cache_node *n; | |
2888 | ||
51df1142 | 2889 | BUG_ON(kmem_cache_node->size < sizeof(struct kmem_cache_node)); |
81819f0f | 2890 | |
51df1142 | 2891 | page = new_slab(kmem_cache_node, GFP_NOWAIT, node); |
81819f0f CL |
2892 | |
2893 | BUG_ON(!page); | |
a2f92ee7 CL |
2894 | if (page_to_nid(page) != node) { |
2895 | printk(KERN_ERR "SLUB: Unable to allocate memory from " | |
2896 | "node %d\n", node); | |
2897 | printk(KERN_ERR "SLUB: Allocating a useless per node structure " | |
2898 | "in order to be able to continue\n"); | |
2899 | } | |
2900 | ||
81819f0f CL |
2901 | n = page->freelist; |
2902 | BUG_ON(!n); | |
51df1142 | 2903 | page->freelist = get_freepointer(kmem_cache_node, n); |
e6e82ea1 | 2904 | page->inuse = 1; |
8cb0a506 | 2905 | page->frozen = 0; |
51df1142 | 2906 | kmem_cache_node->node[node] = n; |
8ab1372f | 2907 | #ifdef CONFIG_SLUB_DEBUG |
f7cb1933 | 2908 | init_object(kmem_cache_node, n, SLUB_RED_ACTIVE); |
51df1142 | 2909 | init_tracking(kmem_cache_node, n); |
8ab1372f | 2910 | #endif |
4053497d | 2911 | init_kmem_cache_node(n); |
51df1142 | 2912 | inc_slabs_node(kmem_cache_node, node, page->objects); |
6446faa2 | 2913 | |
67b6c900 | 2914 | /* |
1e4dd946 SR |
2915 | * No locks need to be taken here as it has just been |
2916 | * initialized and there is no concurrent access. | |
67b6c900 | 2917 | */ |
1e4dd946 | 2918 | __add_partial(n, page, DEACTIVATE_TO_HEAD); |
81819f0f CL |
2919 | } |
2920 | ||
2921 | static void free_kmem_cache_nodes(struct kmem_cache *s) | |
2922 | { | |
2923 | int node; | |
2924 | ||
f64dc58c | 2925 | for_each_node_state(node, N_NORMAL_MEMORY) { |
81819f0f | 2926 | struct kmem_cache_node *n = s->node[node]; |
51df1142 | 2927 | |
73367bd8 | 2928 | if (n) |
51df1142 CL |
2929 | kmem_cache_free(kmem_cache_node, n); |
2930 | ||
81819f0f CL |
2931 | s->node[node] = NULL; |
2932 | } | |
2933 | } | |
2934 | ||
55136592 | 2935 | static int init_kmem_cache_nodes(struct kmem_cache *s) |
81819f0f CL |
2936 | { |
2937 | int node; | |
81819f0f | 2938 | |
f64dc58c | 2939 | for_each_node_state(node, N_NORMAL_MEMORY) { |
81819f0f CL |
2940 | struct kmem_cache_node *n; |
2941 | ||
73367bd8 | 2942 | if (slab_state == DOWN) { |
55136592 | 2943 | early_kmem_cache_node_alloc(node); |
73367bd8 AD |
2944 | continue; |
2945 | } | |
51df1142 | 2946 | n = kmem_cache_alloc_node(kmem_cache_node, |
55136592 | 2947 | GFP_KERNEL, node); |
81819f0f | 2948 | |
73367bd8 AD |
2949 | if (!n) { |
2950 | free_kmem_cache_nodes(s); | |
2951 | return 0; | |
81819f0f | 2952 | } |
73367bd8 | 2953 | |
81819f0f | 2954 | s->node[node] = n; |
4053497d | 2955 | init_kmem_cache_node(n); |
81819f0f CL |
2956 | } |
2957 | return 1; | |
2958 | } | |
81819f0f | 2959 | |
c0bdb232 | 2960 | static void set_min_partial(struct kmem_cache *s, unsigned long min) |
3b89d7d8 DR |
2961 | { |
2962 | if (min < MIN_PARTIAL) | |
2963 | min = MIN_PARTIAL; | |
2964 | else if (min > MAX_PARTIAL) | |
2965 | min = MAX_PARTIAL; | |
2966 | s->min_partial = min; | |
2967 | } | |
2968 | ||
81819f0f CL |
2969 | /* |
2970 | * calculate_sizes() determines the order and the distribution of data within | |
2971 | * a slab object. | |
2972 | */ | |
06b285dc | 2973 | static int calculate_sizes(struct kmem_cache *s, int forced_order) |
81819f0f CL |
2974 | { |
2975 | unsigned long flags = s->flags; | |
3b0efdfa | 2976 | unsigned long size = s->object_size; |
834f3d11 | 2977 | int order; |
81819f0f | 2978 | |
d8b42bf5 CL |
2979 | /* |
2980 | * Round up object size to the next word boundary. We can only | |
2981 | * place the free pointer at word boundaries and this determines | |
2982 | * the possible location of the free pointer. | |
2983 | */ | |
2984 | size = ALIGN(size, sizeof(void *)); | |
2985 | ||
2986 | #ifdef CONFIG_SLUB_DEBUG | |
81819f0f CL |
2987 | /* |
2988 | * Determine if we can poison the object itself. If the user of | |
2989 | * the slab may touch the object after free or before allocation | |
2990 | * then we should never poison the object itself. | |
2991 | */ | |
2992 | if ((flags & SLAB_POISON) && !(flags & SLAB_DESTROY_BY_RCU) && | |
c59def9f | 2993 | !s->ctor) |
81819f0f CL |
2994 | s->flags |= __OBJECT_POISON; |
2995 | else | |
2996 | s->flags &= ~__OBJECT_POISON; | |
2997 | ||
81819f0f CL |
2998 | |
2999 | /* | |
672bba3a | 3000 | * If we are Redzoning then check if there is some space between the |
81819f0f | 3001 | * end of the object and the free pointer. If not then add an |
672bba3a | 3002 | * additional word to have some bytes to store Redzone information. |
81819f0f | 3003 | */ |
3b0efdfa | 3004 | if ((flags & SLAB_RED_ZONE) && size == s->object_size) |
81819f0f | 3005 | size += sizeof(void *); |
41ecc55b | 3006 | #endif |
81819f0f CL |
3007 | |
3008 | /* | |
672bba3a CL |
3009 | * With that we have determined the number of bytes in actual use |
3010 | * by the object. This is the potential offset to the free pointer. | |
81819f0f CL |
3011 | */ |
3012 | s->inuse = size; | |
3013 | ||
3014 | if (((flags & (SLAB_DESTROY_BY_RCU | SLAB_POISON)) || | |
c59def9f | 3015 | s->ctor)) { |
81819f0f CL |
3016 | /* |
3017 | * Relocate free pointer after the object if it is not | |
3018 | * permitted to overwrite the first word of the object on | |
3019 | * kmem_cache_free. | |
3020 | * | |
3021 | * This is the case if we do RCU, have a constructor or | |
3022 | * destructor or are poisoning the objects. | |
3023 | */ | |
3024 | s->offset = size; | |
3025 | size += sizeof(void *); | |
3026 | } | |
3027 | ||
c12b3c62 | 3028 | #ifdef CONFIG_SLUB_DEBUG |
81819f0f CL |
3029 | if (flags & SLAB_STORE_USER) |
3030 | /* | |
3031 | * Need to store information about allocs and frees after | |
3032 | * the object. | |
3033 | */ | |
3034 | size += 2 * sizeof(struct track); | |
3035 | ||
be7b3fbc | 3036 | if (flags & SLAB_RED_ZONE) |
81819f0f CL |
3037 | /* |
3038 | * Add some empty padding so that we can catch | |
3039 | * overwrites from earlier objects rather than let | |
3040 | * tracking information or the free pointer be | |
0211a9c8 | 3041 | * corrupted if a user writes before the start |
81819f0f CL |
3042 | * of the object. |
3043 | */ | |
3044 | size += sizeof(void *); | |
41ecc55b | 3045 | #endif |
672bba3a | 3046 | |
81819f0f CL |
3047 | /* |
3048 | * SLUB stores one object immediately after another beginning from | |
3049 | * offset 0. In order to align the objects we have to simply size | |
3050 | * each object to conform to the alignment. | |
3051 | */ | |
45906855 | 3052 | size = ALIGN(size, s->align); |
81819f0f | 3053 | s->size = size; |
06b285dc CL |
3054 | if (forced_order >= 0) |
3055 | order = forced_order; | |
3056 | else | |
ab9a0f19 | 3057 | order = calculate_order(size, s->reserved); |
81819f0f | 3058 | |
834f3d11 | 3059 | if (order < 0) |
81819f0f CL |
3060 | return 0; |
3061 | ||
b7a49f0d | 3062 | s->allocflags = 0; |
834f3d11 | 3063 | if (order) |
b7a49f0d CL |
3064 | s->allocflags |= __GFP_COMP; |
3065 | ||
3066 | if (s->flags & SLAB_CACHE_DMA) | |
2c59dd65 | 3067 | s->allocflags |= GFP_DMA; |
b7a49f0d CL |
3068 | |
3069 | if (s->flags & SLAB_RECLAIM_ACCOUNT) | |
3070 | s->allocflags |= __GFP_RECLAIMABLE; | |
3071 | ||
81819f0f CL |
3072 | /* |
3073 | * Determine the number of objects per slab | |
3074 | */ | |
ab9a0f19 LJ |
3075 | s->oo = oo_make(order, size, s->reserved); |
3076 | s->min = oo_make(get_order(size), size, s->reserved); | |
205ab99d CL |
3077 | if (oo_objects(s->oo) > oo_objects(s->max)) |
3078 | s->max = s->oo; | |
81819f0f | 3079 | |
834f3d11 | 3080 | return !!oo_objects(s->oo); |
81819f0f CL |
3081 | } |
3082 | ||
8a13a4cc | 3083 | static int kmem_cache_open(struct kmem_cache *s, unsigned long flags) |
81819f0f | 3084 | { |
8a13a4cc | 3085 | s->flags = kmem_cache_flags(s->size, flags, s->name, s->ctor); |
ab9a0f19 | 3086 | s->reserved = 0; |
81819f0f | 3087 | |
da9a638c LJ |
3088 | if (need_reserve_slab_rcu && (s->flags & SLAB_DESTROY_BY_RCU)) |
3089 | s->reserved = sizeof(struct rcu_head); | |
81819f0f | 3090 | |
06b285dc | 3091 | if (!calculate_sizes(s, -1)) |
81819f0f | 3092 | goto error; |
3de47213 DR |
3093 | if (disable_higher_order_debug) { |
3094 | /* | |
3095 | * Disable debugging flags that store metadata if the min slab | |
3096 | * order increased. | |
3097 | */ | |
3b0efdfa | 3098 | if (get_order(s->size) > get_order(s->object_size)) { |
3de47213 DR |
3099 | s->flags &= ~DEBUG_METADATA_FLAGS; |
3100 | s->offset = 0; | |
3101 | if (!calculate_sizes(s, -1)) | |
3102 | goto error; | |
3103 | } | |
3104 | } | |
81819f0f | 3105 | |
2565409f HC |
3106 | #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \ |
3107 | defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE) | |
b789ef51 CL |
3108 | if (system_has_cmpxchg_double() && (s->flags & SLAB_DEBUG_FLAGS) == 0) |
3109 | /* Enable fast mode */ | |
3110 | s->flags |= __CMPXCHG_DOUBLE; | |
3111 | #endif | |
3112 | ||
3b89d7d8 DR |
3113 | /* |
3114 | * The larger the object size is, the more pages we want on the partial | |
3115 | * list to avoid pounding the page allocator excessively. | |
3116 | */ | |
49e22585 CL |
3117 | set_min_partial(s, ilog2(s->size) / 2); |
3118 | ||
3119 | /* | |
3120 | * cpu_partial determined the maximum number of objects kept in the | |
3121 | * per cpu partial lists of a processor. | |
3122 | * | |
3123 | * Per cpu partial lists mainly contain slabs that just have one | |
3124 | * object freed. If they are used for allocation then they can be | |
3125 | * filled up again with minimal effort. The slab will never hit the | |
3126 | * per node partial lists and therefore no locking will be required. | |
3127 | * | |
3128 | * This setting also determines | |
3129 | * | |
3130 | * A) The number of objects from per cpu partial slabs dumped to the | |
3131 | * per node list when we reach the limit. | |
9f264904 | 3132 | * B) The number of objects in cpu partial slabs to extract from the |
d0e0ac97 CG |
3133 | * per node list when we run out of per cpu objects. We only fetch |
3134 | * 50% to keep some capacity around for frees. | |
49e22585 | 3135 | */ |
345c905d | 3136 | if (!kmem_cache_has_cpu_partial(s)) |
8f1e33da CL |
3137 | s->cpu_partial = 0; |
3138 | else if (s->size >= PAGE_SIZE) | |
49e22585 CL |
3139 | s->cpu_partial = 2; |
3140 | else if (s->size >= 1024) | |
3141 | s->cpu_partial = 6; | |
3142 | else if (s->size >= 256) | |
3143 | s->cpu_partial = 13; | |
3144 | else | |
3145 | s->cpu_partial = 30; | |
3146 | ||
81819f0f | 3147 | #ifdef CONFIG_NUMA |
e2cb96b7 | 3148 | s->remote_node_defrag_ratio = 1000; |
81819f0f | 3149 | #endif |
55136592 | 3150 | if (!init_kmem_cache_nodes(s)) |
dfb4f096 | 3151 | goto error; |
81819f0f | 3152 | |
55136592 | 3153 | if (alloc_kmem_cache_cpus(s)) |
278b1bb1 | 3154 | return 0; |
ff12059e | 3155 | |
4c93c355 | 3156 | free_kmem_cache_nodes(s); |
81819f0f CL |
3157 | error: |
3158 | if (flags & SLAB_PANIC) | |
3159 | panic("Cannot create slab %s size=%lu realsize=%u " | |
3160 | "order=%u offset=%u flags=%lx\n", | |
d0e0ac97 CG |
3161 | s->name, (unsigned long)s->size, s->size, |
3162 | oo_order(s->oo), s->offset, flags); | |
278b1bb1 | 3163 | return -EINVAL; |
81819f0f | 3164 | } |
81819f0f | 3165 | |
33b12c38 CL |
3166 | static void list_slab_objects(struct kmem_cache *s, struct page *page, |
3167 | const char *text) | |
3168 | { | |
3169 | #ifdef CONFIG_SLUB_DEBUG | |
3170 | void *addr = page_address(page); | |
3171 | void *p; | |
a5dd5c11 NK |
3172 | unsigned long *map = kzalloc(BITS_TO_LONGS(page->objects) * |
3173 | sizeof(long), GFP_ATOMIC); | |
bbd7d57b ED |
3174 | if (!map) |
3175 | return; | |
945cf2b6 | 3176 | slab_err(s, page, text, s->name); |
33b12c38 | 3177 | slab_lock(page); |
33b12c38 | 3178 | |
5f80b13a | 3179 | get_map(s, page, map); |
33b12c38 CL |
3180 | for_each_object(p, s, addr, page->objects) { |
3181 | ||
3182 | if (!test_bit(slab_index(p, s, addr), map)) { | |
3183 | printk(KERN_ERR "INFO: Object 0x%p @offset=%tu\n", | |
3184 | p, p - addr); | |
3185 | print_tracking(s, p); | |
3186 | } | |
3187 | } | |
3188 | slab_unlock(page); | |
bbd7d57b | 3189 | kfree(map); |
33b12c38 CL |
3190 | #endif |
3191 | } | |
3192 | ||
81819f0f | 3193 | /* |
599870b1 | 3194 | * Attempt to free all partial slabs on a node. |
69cb8e6b CL |
3195 | * This is called from kmem_cache_close(). We must be the last thread |
3196 | * using the cache and therefore we do not need to lock anymore. | |
81819f0f | 3197 | */ |
599870b1 | 3198 | static void free_partial(struct kmem_cache *s, struct kmem_cache_node *n) |
81819f0f | 3199 | { |
81819f0f CL |
3200 | struct page *page, *h; |
3201 | ||
33b12c38 | 3202 | list_for_each_entry_safe(page, h, &n->partial, lru) { |
81819f0f | 3203 | if (!page->inuse) { |
1e4dd946 | 3204 | __remove_partial(n, page); |
81819f0f | 3205 | discard_slab(s, page); |
33b12c38 CL |
3206 | } else { |
3207 | list_slab_objects(s, page, | |
945cf2b6 | 3208 | "Objects remaining in %s on kmem_cache_close()"); |
599870b1 | 3209 | } |
33b12c38 | 3210 | } |
81819f0f CL |
3211 | } |
3212 | ||
3213 | /* | |
672bba3a | 3214 | * Release all resources used by a slab cache. |
81819f0f | 3215 | */ |
0c710013 | 3216 | static inline int kmem_cache_close(struct kmem_cache *s) |
81819f0f CL |
3217 | { |
3218 | int node; | |
3219 | ||
3220 | flush_all(s); | |
81819f0f | 3221 | /* Attempt to free all objects */ |
f64dc58c | 3222 | for_each_node_state(node, N_NORMAL_MEMORY) { |
81819f0f CL |
3223 | struct kmem_cache_node *n = get_node(s, node); |
3224 | ||
599870b1 CL |
3225 | free_partial(s, n); |
3226 | if (n->nr_partial || slabs_node(s, node)) | |
81819f0f CL |
3227 | return 1; |
3228 | } | |
945cf2b6 | 3229 | free_percpu(s->cpu_slab); |
81819f0f CL |
3230 | free_kmem_cache_nodes(s); |
3231 | return 0; | |
3232 | } | |
3233 | ||
945cf2b6 | 3234 | int __kmem_cache_shutdown(struct kmem_cache *s) |
81819f0f | 3235 | { |
12c3667f | 3236 | int rc = kmem_cache_close(s); |
945cf2b6 | 3237 | |
5413dfba GC |
3238 | if (!rc) { |
3239 | /* | |
421af243 VD |
3240 | * Since slab_attr_store may take the slab_mutex, we should |
3241 | * release the lock while removing the sysfs entry in order to | |
3242 | * avoid a deadlock. Because this is pretty much the last | |
5413dfba GC |
3243 | * operation we do and the lock will be released shortly after |
3244 | * that in slab_common.c, we could just move sysfs_slab_remove | |
3245 | * to a later point in common code. We should do that when we | |
3246 | * have a common sysfs framework for all allocators. | |
3247 | */ | |
3248 | mutex_unlock(&slab_mutex); | |
81819f0f | 3249 | sysfs_slab_remove(s); |
5413dfba GC |
3250 | mutex_lock(&slab_mutex); |
3251 | } | |
12c3667f CL |
3252 | |
3253 | return rc; | |
81819f0f | 3254 | } |
81819f0f CL |
3255 | |
3256 | /******************************************************************** | |
3257 | * Kmalloc subsystem | |
3258 | *******************************************************************/ | |
3259 | ||
81819f0f CL |
3260 | static int __init setup_slub_min_order(char *str) |
3261 | { | |
06428780 | 3262 | get_option(&str, &slub_min_order); |
81819f0f CL |
3263 | |
3264 | return 1; | |
3265 | } | |
3266 | ||
3267 | __setup("slub_min_order=", setup_slub_min_order); | |
3268 | ||
3269 | static int __init setup_slub_max_order(char *str) | |
3270 | { | |
06428780 | 3271 | get_option(&str, &slub_max_order); |
818cf590 | 3272 | slub_max_order = min(slub_max_order, MAX_ORDER - 1); |
81819f0f CL |
3273 | |
3274 | return 1; | |
3275 | } | |
3276 | ||
3277 | __setup("slub_max_order=", setup_slub_max_order); | |
3278 | ||
3279 | static int __init setup_slub_min_objects(char *str) | |
3280 | { | |
06428780 | 3281 | get_option(&str, &slub_min_objects); |
81819f0f CL |
3282 | |
3283 | return 1; | |
3284 | } | |
3285 | ||
3286 | __setup("slub_min_objects=", setup_slub_min_objects); | |
3287 | ||
3288 | static int __init setup_slub_nomerge(char *str) | |
3289 | { | |
3290 | slub_nomerge = 1; | |
3291 | return 1; | |
3292 | } | |
3293 | ||
3294 | __setup("slub_nomerge", setup_slub_nomerge); | |
3295 | ||
81819f0f CL |
3296 | void *__kmalloc(size_t size, gfp_t flags) |
3297 | { | |
aadb4bc4 | 3298 | struct kmem_cache *s; |
5b882be4 | 3299 | void *ret; |
81819f0f | 3300 | |
95a05b42 | 3301 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) |
eada35ef | 3302 | return kmalloc_large(size, flags); |
aadb4bc4 | 3303 | |
2c59dd65 | 3304 | s = kmalloc_slab(size, flags); |
aadb4bc4 CL |
3305 | |
3306 | if (unlikely(ZERO_OR_NULL_PTR(s))) | |
6cb8f913 CL |
3307 | return s; |
3308 | ||
2b847c3c | 3309 | ret = slab_alloc(s, flags, _RET_IP_); |
5b882be4 | 3310 | |
ca2b84cb | 3311 | trace_kmalloc(_RET_IP_, ret, size, s->size, flags); |
5b882be4 EGM |
3312 | |
3313 | return ret; | |
81819f0f CL |
3314 | } |
3315 | EXPORT_SYMBOL(__kmalloc); | |
3316 | ||
5d1f57e4 | 3317 | #ifdef CONFIG_NUMA |
f619cfe1 CL |
3318 | static void *kmalloc_large_node(size_t size, gfp_t flags, int node) |
3319 | { | |
b1eeab67 | 3320 | struct page *page; |
e4f7c0b4 | 3321 | void *ptr = NULL; |
f619cfe1 | 3322 | |
d79923fa | 3323 | flags |= __GFP_COMP | __GFP_NOTRACK | __GFP_KMEMCG; |
b1eeab67 | 3324 | page = alloc_pages_node(node, flags, get_order(size)); |
f619cfe1 | 3325 | if (page) |
e4f7c0b4 CM |
3326 | ptr = page_address(page); |
3327 | ||
d56791b3 | 3328 | kmalloc_large_node_hook(ptr, size, flags); |
e4f7c0b4 | 3329 | return ptr; |
f619cfe1 CL |
3330 | } |
3331 | ||
81819f0f CL |
3332 | void *__kmalloc_node(size_t size, gfp_t flags, int node) |
3333 | { | |
aadb4bc4 | 3334 | struct kmem_cache *s; |
5b882be4 | 3335 | void *ret; |
81819f0f | 3336 | |
95a05b42 | 3337 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) { |
5b882be4 EGM |
3338 | ret = kmalloc_large_node(size, flags, node); |
3339 | ||
ca2b84cb EGM |
3340 | trace_kmalloc_node(_RET_IP_, ret, |
3341 | size, PAGE_SIZE << get_order(size), | |
3342 | flags, node); | |
5b882be4 EGM |
3343 | |
3344 | return ret; | |
3345 | } | |
aadb4bc4 | 3346 | |
2c59dd65 | 3347 | s = kmalloc_slab(size, flags); |
aadb4bc4 CL |
3348 | |
3349 | if (unlikely(ZERO_OR_NULL_PTR(s))) | |
6cb8f913 CL |
3350 | return s; |
3351 | ||
2b847c3c | 3352 | ret = slab_alloc_node(s, flags, node, _RET_IP_); |
5b882be4 | 3353 | |
ca2b84cb | 3354 | trace_kmalloc_node(_RET_IP_, ret, size, s->size, flags, node); |
5b882be4 EGM |
3355 | |
3356 | return ret; | |
81819f0f CL |
3357 | } |
3358 | EXPORT_SYMBOL(__kmalloc_node); | |
3359 | #endif | |
3360 | ||
3361 | size_t ksize(const void *object) | |
3362 | { | |
272c1d21 | 3363 | struct page *page; |
81819f0f | 3364 | |
ef8b4520 | 3365 | if (unlikely(object == ZERO_SIZE_PTR)) |
272c1d21 CL |
3366 | return 0; |
3367 | ||
294a80a8 | 3368 | page = virt_to_head_page(object); |
294a80a8 | 3369 | |
76994412 PE |
3370 | if (unlikely(!PageSlab(page))) { |
3371 | WARN_ON(!PageCompound(page)); | |
294a80a8 | 3372 | return PAGE_SIZE << compound_order(page); |
76994412 | 3373 | } |
81819f0f | 3374 | |
1b4f59e3 | 3375 | return slab_ksize(page->slab_cache); |
81819f0f | 3376 | } |
b1aabecd | 3377 | EXPORT_SYMBOL(ksize); |
81819f0f CL |
3378 | |
3379 | void kfree(const void *x) | |
3380 | { | |
81819f0f | 3381 | struct page *page; |
5bb983b0 | 3382 | void *object = (void *)x; |
81819f0f | 3383 | |
2121db74 PE |
3384 | trace_kfree(_RET_IP_, x); |
3385 | ||
2408c550 | 3386 | if (unlikely(ZERO_OR_NULL_PTR(x))) |
81819f0f CL |
3387 | return; |
3388 | ||
b49af68f | 3389 | page = virt_to_head_page(x); |
aadb4bc4 | 3390 | if (unlikely(!PageSlab(page))) { |
0937502a | 3391 | BUG_ON(!PageCompound(page)); |
d56791b3 | 3392 | kfree_hook(x); |
d79923fa | 3393 | __free_memcg_kmem_pages(page, compound_order(page)); |
aadb4bc4 CL |
3394 | return; |
3395 | } | |
1b4f59e3 | 3396 | slab_free(page->slab_cache, page, object, _RET_IP_); |
81819f0f CL |
3397 | } |
3398 | EXPORT_SYMBOL(kfree); | |
3399 | ||
2086d26a | 3400 | /* |
672bba3a CL |
3401 | * kmem_cache_shrink removes empty slabs from the partial lists and sorts |
3402 | * the remaining slabs by the number of items in use. The slabs with the | |
3403 | * most items in use come first. New allocations will then fill those up | |
3404 | * and thus they can be removed from the partial lists. | |
3405 | * | |
3406 | * The slabs with the least items are placed last. This results in them | |
3407 | * being allocated from last increasing the chance that the last objects | |
3408 | * are freed in them. | |
2086d26a CL |
3409 | */ |
3410 | int kmem_cache_shrink(struct kmem_cache *s) | |
3411 | { | |
3412 | int node; | |
3413 | int i; | |
3414 | struct kmem_cache_node *n; | |
3415 | struct page *page; | |
3416 | struct page *t; | |
205ab99d | 3417 | int objects = oo_objects(s->max); |
2086d26a | 3418 | struct list_head *slabs_by_inuse = |
834f3d11 | 3419 | kmalloc(sizeof(struct list_head) * objects, GFP_KERNEL); |
2086d26a CL |
3420 | unsigned long flags; |
3421 | ||
3422 | if (!slabs_by_inuse) | |
3423 | return -ENOMEM; | |
3424 | ||
3425 | flush_all(s); | |
f64dc58c | 3426 | for_each_node_state(node, N_NORMAL_MEMORY) { |
2086d26a CL |
3427 | n = get_node(s, node); |
3428 | ||
3429 | if (!n->nr_partial) | |
3430 | continue; | |
3431 | ||
834f3d11 | 3432 | for (i = 0; i < objects; i++) |
2086d26a CL |
3433 | INIT_LIST_HEAD(slabs_by_inuse + i); |
3434 | ||
3435 | spin_lock_irqsave(&n->list_lock, flags); | |
3436 | ||
3437 | /* | |
672bba3a | 3438 | * Build lists indexed by the items in use in each slab. |
2086d26a | 3439 | * |
672bba3a CL |
3440 | * Note that concurrent frees may occur while we hold the |
3441 | * list_lock. page->inuse here is the upper limit. | |
2086d26a CL |
3442 | */ |
3443 | list_for_each_entry_safe(page, t, &n->partial, lru) { | |
69cb8e6b CL |
3444 | list_move(&page->lru, slabs_by_inuse + page->inuse); |
3445 | if (!page->inuse) | |
3446 | n->nr_partial--; | |
2086d26a CL |
3447 | } |
3448 | ||
2086d26a | 3449 | /* |
672bba3a CL |
3450 | * Rebuild the partial list with the slabs filled up most |
3451 | * first and the least used slabs at the end. | |
2086d26a | 3452 | */ |
69cb8e6b | 3453 | for (i = objects - 1; i > 0; i--) |
2086d26a CL |
3454 | list_splice(slabs_by_inuse + i, n->partial.prev); |
3455 | ||
2086d26a | 3456 | spin_unlock_irqrestore(&n->list_lock, flags); |
69cb8e6b CL |
3457 | |
3458 | /* Release empty slabs */ | |
3459 | list_for_each_entry_safe(page, t, slabs_by_inuse, lru) | |
3460 | discard_slab(s, page); | |
2086d26a CL |
3461 | } |
3462 | ||
3463 | kfree(slabs_by_inuse); | |
3464 | return 0; | |
3465 | } | |
3466 | EXPORT_SYMBOL(kmem_cache_shrink); | |
3467 | ||
b9049e23 YG |
3468 | static int slab_mem_going_offline_callback(void *arg) |
3469 | { | |
3470 | struct kmem_cache *s; | |
3471 | ||
18004c5d | 3472 | mutex_lock(&slab_mutex); |
b9049e23 YG |
3473 | list_for_each_entry(s, &slab_caches, list) |
3474 | kmem_cache_shrink(s); | |
18004c5d | 3475 | mutex_unlock(&slab_mutex); |
b9049e23 YG |
3476 | |
3477 | return 0; | |
3478 | } | |
3479 | ||
3480 | static void slab_mem_offline_callback(void *arg) | |
3481 | { | |
3482 | struct kmem_cache_node *n; | |
3483 | struct kmem_cache *s; | |
3484 | struct memory_notify *marg = arg; | |
3485 | int offline_node; | |
3486 | ||
b9d5ab25 | 3487 | offline_node = marg->status_change_nid_normal; |
b9049e23 YG |
3488 | |
3489 | /* | |
3490 | * If the node still has available memory. we need kmem_cache_node | |
3491 | * for it yet. | |
3492 | */ | |
3493 | if (offline_node < 0) | |
3494 | return; | |
3495 | ||
18004c5d | 3496 | mutex_lock(&slab_mutex); |
b9049e23 YG |
3497 | list_for_each_entry(s, &slab_caches, list) { |
3498 | n = get_node(s, offline_node); | |
3499 | if (n) { | |
3500 | /* | |
3501 | * if n->nr_slabs > 0, slabs still exist on the node | |
3502 | * that is going down. We were unable to free them, | |
c9404c9c | 3503 | * and offline_pages() function shouldn't call this |
b9049e23 YG |
3504 | * callback. So, we must fail. |
3505 | */ | |
0f389ec6 | 3506 | BUG_ON(slabs_node(s, offline_node)); |
b9049e23 YG |
3507 | |
3508 | s->node[offline_node] = NULL; | |
8de66a0c | 3509 | kmem_cache_free(kmem_cache_node, n); |
b9049e23 YG |
3510 | } |
3511 | } | |
18004c5d | 3512 | mutex_unlock(&slab_mutex); |
b9049e23 YG |
3513 | } |
3514 | ||
3515 | static int slab_mem_going_online_callback(void *arg) | |
3516 | { | |
3517 | struct kmem_cache_node *n; | |
3518 | struct kmem_cache *s; | |
3519 | struct memory_notify *marg = arg; | |
b9d5ab25 | 3520 | int nid = marg->status_change_nid_normal; |
b9049e23 YG |
3521 | int ret = 0; |
3522 | ||
3523 | /* | |
3524 | * If the node's memory is already available, then kmem_cache_node is | |
3525 | * already created. Nothing to do. | |
3526 | */ | |
3527 | if (nid < 0) | |
3528 | return 0; | |
3529 | ||
3530 | /* | |
0121c619 | 3531 | * We are bringing a node online. No memory is available yet. We must |
b9049e23 YG |
3532 | * allocate a kmem_cache_node structure in order to bring the node |
3533 | * online. | |
3534 | */ | |
18004c5d | 3535 | mutex_lock(&slab_mutex); |
b9049e23 YG |
3536 | list_for_each_entry(s, &slab_caches, list) { |
3537 | /* | |
3538 | * XXX: kmem_cache_alloc_node will fallback to other nodes | |
3539 | * since memory is not yet available from the node that | |
3540 | * is brought up. | |
3541 | */ | |
8de66a0c | 3542 | n = kmem_cache_alloc(kmem_cache_node, GFP_KERNEL); |
b9049e23 YG |
3543 | if (!n) { |
3544 | ret = -ENOMEM; | |
3545 | goto out; | |
3546 | } | |
4053497d | 3547 | init_kmem_cache_node(n); |
b9049e23 YG |
3548 | s->node[nid] = n; |
3549 | } | |
3550 | out: | |
18004c5d | 3551 | mutex_unlock(&slab_mutex); |
b9049e23 YG |
3552 | return ret; |
3553 | } | |
3554 | ||
3555 | static int slab_memory_callback(struct notifier_block *self, | |
3556 | unsigned long action, void *arg) | |
3557 | { | |
3558 | int ret = 0; | |
3559 | ||
3560 | switch (action) { | |
3561 | case MEM_GOING_ONLINE: | |
3562 | ret = slab_mem_going_online_callback(arg); | |
3563 | break; | |
3564 | case MEM_GOING_OFFLINE: | |
3565 | ret = slab_mem_going_offline_callback(arg); | |
3566 | break; | |
3567 | case MEM_OFFLINE: | |
3568 | case MEM_CANCEL_ONLINE: | |
3569 | slab_mem_offline_callback(arg); | |
3570 | break; | |
3571 | case MEM_ONLINE: | |
3572 | case MEM_CANCEL_OFFLINE: | |
3573 | break; | |
3574 | } | |
dc19f9db KH |
3575 | if (ret) |
3576 | ret = notifier_from_errno(ret); | |
3577 | else | |
3578 | ret = NOTIFY_OK; | |
b9049e23 YG |
3579 | return ret; |
3580 | } | |
3581 | ||
3ac38faa AM |
3582 | static struct notifier_block slab_memory_callback_nb = { |
3583 | .notifier_call = slab_memory_callback, | |
3584 | .priority = SLAB_CALLBACK_PRI, | |
3585 | }; | |
b9049e23 | 3586 | |
81819f0f CL |
3587 | /******************************************************************** |
3588 | * Basic setup of slabs | |
3589 | *******************************************************************/ | |
3590 | ||
51df1142 CL |
3591 | /* |
3592 | * Used for early kmem_cache structures that were allocated using | |
dffb4d60 CL |
3593 | * the page allocator. Allocate them properly then fix up the pointers |
3594 | * that may be pointing to the wrong kmem_cache structure. | |
51df1142 CL |
3595 | */ |
3596 | ||
dffb4d60 | 3597 | static struct kmem_cache * __init bootstrap(struct kmem_cache *static_cache) |
51df1142 CL |
3598 | { |
3599 | int node; | |
dffb4d60 | 3600 | struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT); |
51df1142 | 3601 | |
dffb4d60 | 3602 | memcpy(s, static_cache, kmem_cache->object_size); |
51df1142 | 3603 | |
7d557b3c GC |
3604 | /* |
3605 | * This runs very early, and only the boot processor is supposed to be | |
3606 | * up. Even if it weren't true, IRQs are not up so we couldn't fire | |
3607 | * IPIs around. | |
3608 | */ | |
3609 | __flush_cpu_slab(s, smp_processor_id()); | |
51df1142 CL |
3610 | for_each_node_state(node, N_NORMAL_MEMORY) { |
3611 | struct kmem_cache_node *n = get_node(s, node); | |
3612 | struct page *p; | |
3613 | ||
3614 | if (n) { | |
3615 | list_for_each_entry(p, &n->partial, lru) | |
1b4f59e3 | 3616 | p->slab_cache = s; |
51df1142 | 3617 | |
607bf324 | 3618 | #ifdef CONFIG_SLUB_DEBUG |
51df1142 | 3619 | list_for_each_entry(p, &n->full, lru) |
1b4f59e3 | 3620 | p->slab_cache = s; |
51df1142 CL |
3621 | #endif |
3622 | } | |
3623 | } | |
dffb4d60 CL |
3624 | list_add(&s->list, &slab_caches); |
3625 | return s; | |
51df1142 CL |
3626 | } |
3627 | ||
81819f0f CL |
3628 | void __init kmem_cache_init(void) |
3629 | { | |
dffb4d60 CL |
3630 | static __initdata struct kmem_cache boot_kmem_cache, |
3631 | boot_kmem_cache_node; | |
51df1142 | 3632 | |
fc8d8620 SG |
3633 | if (debug_guardpage_minorder()) |
3634 | slub_max_order = 0; | |
3635 | ||
dffb4d60 CL |
3636 | kmem_cache_node = &boot_kmem_cache_node; |
3637 | kmem_cache = &boot_kmem_cache; | |
51df1142 | 3638 | |
dffb4d60 CL |
3639 | create_boot_cache(kmem_cache_node, "kmem_cache_node", |
3640 | sizeof(struct kmem_cache_node), SLAB_HWCACHE_ALIGN); | |
b9049e23 | 3641 | |
3ac38faa | 3642 | register_hotmemory_notifier(&slab_memory_callback_nb); |
81819f0f CL |
3643 | |
3644 | /* Able to allocate the per node structures */ | |
3645 | slab_state = PARTIAL; | |
3646 | ||
dffb4d60 CL |
3647 | create_boot_cache(kmem_cache, "kmem_cache", |
3648 | offsetof(struct kmem_cache, node) + | |
3649 | nr_node_ids * sizeof(struct kmem_cache_node *), | |
3650 | SLAB_HWCACHE_ALIGN); | |
8a13a4cc | 3651 | |
dffb4d60 | 3652 | kmem_cache = bootstrap(&boot_kmem_cache); |
81819f0f | 3653 | |
51df1142 CL |
3654 | /* |
3655 | * Allocate kmem_cache_node properly from the kmem_cache slab. | |
3656 | * kmem_cache_node is separately allocated so no need to | |
3657 | * update any list pointers. | |
3658 | */ | |
dffb4d60 | 3659 | kmem_cache_node = bootstrap(&boot_kmem_cache_node); |
51df1142 CL |
3660 | |
3661 | /* Now we can use the kmem_cache to allocate kmalloc slabs */ | |
f97d5f63 | 3662 | create_kmalloc_caches(0); |
81819f0f CL |
3663 | |
3664 | #ifdef CONFIG_SMP | |
3665 | register_cpu_notifier(&slab_notifier); | |
9dfc6e68 | 3666 | #endif |
81819f0f | 3667 | |
3adbefee | 3668 | printk(KERN_INFO |
f97d5f63 | 3669 | "SLUB: HWalign=%d, Order=%d-%d, MinObjects=%d," |
4b356be0 | 3670 | " CPUs=%d, Nodes=%d\n", |
f97d5f63 | 3671 | cache_line_size(), |
81819f0f CL |
3672 | slub_min_order, slub_max_order, slub_min_objects, |
3673 | nr_cpu_ids, nr_node_ids); | |
3674 | } | |
3675 | ||
7e85ee0c PE |
3676 | void __init kmem_cache_init_late(void) |
3677 | { | |
7e85ee0c PE |
3678 | } |
3679 | ||
81819f0f CL |
3680 | /* |
3681 | * Find a mergeable slab cache | |
3682 | */ | |
3683 | static int slab_unmergeable(struct kmem_cache *s) | |
3684 | { | |
3685 | if (slub_nomerge || (s->flags & SLUB_NEVER_MERGE)) | |
3686 | return 1; | |
3687 | ||
a44cb944 VD |
3688 | if (!is_root_cache(s)) |
3689 | return 1; | |
3690 | ||
c59def9f | 3691 | if (s->ctor) |
81819f0f CL |
3692 | return 1; |
3693 | ||
8ffa6875 CL |
3694 | /* |
3695 | * We may have set a slab to be unmergeable during bootstrap. | |
3696 | */ | |
3697 | if (s->refcount < 0) | |
3698 | return 1; | |
3699 | ||
81819f0f CL |
3700 | return 0; |
3701 | } | |
3702 | ||
a44cb944 VD |
3703 | static struct kmem_cache *find_mergeable(size_t size, size_t align, |
3704 | unsigned long flags, const char *name, void (*ctor)(void *)) | |
81819f0f | 3705 | { |
5b95a4ac | 3706 | struct kmem_cache *s; |
81819f0f CL |
3707 | |
3708 | if (slub_nomerge || (flags & SLUB_NEVER_MERGE)) | |
3709 | return NULL; | |
3710 | ||
c59def9f | 3711 | if (ctor) |
81819f0f CL |
3712 | return NULL; |
3713 | ||
3714 | size = ALIGN(size, sizeof(void *)); | |
3715 | align = calculate_alignment(flags, align, size); | |
3716 | size = ALIGN(size, align); | |
ba0268a8 | 3717 | flags = kmem_cache_flags(size, flags, name, NULL); |
81819f0f | 3718 | |
5b95a4ac | 3719 | list_for_each_entry(s, &slab_caches, list) { |
81819f0f CL |
3720 | if (slab_unmergeable(s)) |
3721 | continue; | |
3722 | ||
3723 | if (size > s->size) | |
3724 | continue; | |
3725 | ||
ba0268a8 | 3726 | if ((flags & SLUB_MERGE_SAME) != (s->flags & SLUB_MERGE_SAME)) |
a44cb944 | 3727 | continue; |
81819f0f CL |
3728 | /* |
3729 | * Check if alignment is compatible. | |
3730 | * Courtesy of Adrian Drzewiecki | |
3731 | */ | |
06428780 | 3732 | if ((s->size & ~(align - 1)) != s->size) |
81819f0f CL |
3733 | continue; |
3734 | ||
3735 | if (s->size - size >= sizeof(void *)) | |
3736 | continue; | |
3737 | ||
3738 | return s; | |
3739 | } | |
3740 | return NULL; | |
3741 | } | |
3742 | ||
2633d7a0 | 3743 | struct kmem_cache * |
a44cb944 VD |
3744 | __kmem_cache_alias(const char *name, size_t size, size_t align, |
3745 | unsigned long flags, void (*ctor)(void *)) | |
81819f0f CL |
3746 | { |
3747 | struct kmem_cache *s; | |
3748 | ||
a44cb944 | 3749 | s = find_mergeable(size, align, flags, name, ctor); |
81819f0f | 3750 | if (s) { |
84d0ddd6 VD |
3751 | int i; |
3752 | struct kmem_cache *c; | |
3753 | ||
81819f0f | 3754 | s->refcount++; |
84d0ddd6 | 3755 | |
81819f0f CL |
3756 | /* |
3757 | * Adjust the object sizes so that we clear | |
3758 | * the complete object on kzalloc. | |
3759 | */ | |
3b0efdfa | 3760 | s->object_size = max(s->object_size, (int)size); |
81819f0f | 3761 | s->inuse = max_t(int, s->inuse, ALIGN(size, sizeof(void *))); |
6446faa2 | 3762 | |
84d0ddd6 VD |
3763 | for_each_memcg_cache_index(i) { |
3764 | c = cache_from_memcg_idx(s, i); | |
3765 | if (!c) | |
3766 | continue; | |
3767 | c->object_size = s->object_size; | |
3768 | c->inuse = max_t(int, c->inuse, | |
3769 | ALIGN(size, sizeof(void *))); | |
3770 | } | |
3771 | ||
7b8f3b66 | 3772 | if (sysfs_slab_alias(s, name)) { |
7b8f3b66 | 3773 | s->refcount--; |
cbb79694 | 3774 | s = NULL; |
7b8f3b66 | 3775 | } |
a0e1d1be | 3776 | } |
6446faa2 | 3777 | |
cbb79694 CL |
3778 | return s; |
3779 | } | |
84c1cf62 | 3780 | |
8a13a4cc | 3781 | int __kmem_cache_create(struct kmem_cache *s, unsigned long flags) |
cbb79694 | 3782 | { |
aac3a166 PE |
3783 | int err; |
3784 | ||
3785 | err = kmem_cache_open(s, flags); | |
3786 | if (err) | |
3787 | return err; | |
20cea968 | 3788 | |
45530c44 CL |
3789 | /* Mutex is not taken during early boot */ |
3790 | if (slab_state <= UP) | |
3791 | return 0; | |
3792 | ||
107dab5c | 3793 | memcg_propagate_slab_attrs(s); |
aac3a166 | 3794 | err = sysfs_slab_add(s); |
aac3a166 PE |
3795 | if (err) |
3796 | kmem_cache_close(s); | |
20cea968 | 3797 | |
aac3a166 | 3798 | return err; |
81819f0f | 3799 | } |
81819f0f | 3800 | |
81819f0f | 3801 | #ifdef CONFIG_SMP |
81819f0f | 3802 | /* |
672bba3a CL |
3803 | * Use the cpu notifier to insure that the cpu slabs are flushed when |
3804 | * necessary. | |
81819f0f | 3805 | */ |
0db0628d | 3806 | static int slab_cpuup_callback(struct notifier_block *nfb, |
81819f0f CL |
3807 | unsigned long action, void *hcpu) |
3808 | { | |
3809 | long cpu = (long)hcpu; | |
5b95a4ac CL |
3810 | struct kmem_cache *s; |
3811 | unsigned long flags; | |
81819f0f CL |
3812 | |
3813 | switch (action) { | |
3814 | case CPU_UP_CANCELED: | |
8bb78442 | 3815 | case CPU_UP_CANCELED_FROZEN: |
81819f0f | 3816 | case CPU_DEAD: |
8bb78442 | 3817 | case CPU_DEAD_FROZEN: |
18004c5d | 3818 | mutex_lock(&slab_mutex); |
5b95a4ac CL |
3819 | list_for_each_entry(s, &slab_caches, list) { |
3820 | local_irq_save(flags); | |
3821 | __flush_cpu_slab(s, cpu); | |
3822 | local_irq_restore(flags); | |
3823 | } | |
18004c5d | 3824 | mutex_unlock(&slab_mutex); |
81819f0f CL |
3825 | break; |
3826 | default: | |
3827 | break; | |
3828 | } | |
3829 | return NOTIFY_OK; | |
3830 | } | |
3831 | ||
0db0628d | 3832 | static struct notifier_block slab_notifier = { |
3adbefee | 3833 | .notifier_call = slab_cpuup_callback |
06428780 | 3834 | }; |
81819f0f CL |
3835 | |
3836 | #endif | |
3837 | ||
ce71e27c | 3838 | void *__kmalloc_track_caller(size_t size, gfp_t gfpflags, unsigned long caller) |
81819f0f | 3839 | { |
aadb4bc4 | 3840 | struct kmem_cache *s; |
94b528d0 | 3841 | void *ret; |
aadb4bc4 | 3842 | |
95a05b42 | 3843 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) |
eada35ef PE |
3844 | return kmalloc_large(size, gfpflags); |
3845 | ||
2c59dd65 | 3846 | s = kmalloc_slab(size, gfpflags); |
81819f0f | 3847 | |
2408c550 | 3848 | if (unlikely(ZERO_OR_NULL_PTR(s))) |
6cb8f913 | 3849 | return s; |
81819f0f | 3850 | |
2b847c3c | 3851 | ret = slab_alloc(s, gfpflags, caller); |
94b528d0 | 3852 | |
25985edc | 3853 | /* Honor the call site pointer we received. */ |
ca2b84cb | 3854 | trace_kmalloc(caller, ret, size, s->size, gfpflags); |
94b528d0 EGM |
3855 | |
3856 | return ret; | |
81819f0f CL |
3857 | } |
3858 | ||
5d1f57e4 | 3859 | #ifdef CONFIG_NUMA |
81819f0f | 3860 | void *__kmalloc_node_track_caller(size_t size, gfp_t gfpflags, |
ce71e27c | 3861 | int node, unsigned long caller) |
81819f0f | 3862 | { |
aadb4bc4 | 3863 | struct kmem_cache *s; |
94b528d0 | 3864 | void *ret; |
aadb4bc4 | 3865 | |
95a05b42 | 3866 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) { |
d3e14aa3 XF |
3867 | ret = kmalloc_large_node(size, gfpflags, node); |
3868 | ||
3869 | trace_kmalloc_node(caller, ret, | |
3870 | size, PAGE_SIZE << get_order(size), | |
3871 | gfpflags, node); | |
3872 | ||
3873 | return ret; | |
3874 | } | |
eada35ef | 3875 | |
2c59dd65 | 3876 | s = kmalloc_slab(size, gfpflags); |
81819f0f | 3877 | |
2408c550 | 3878 | if (unlikely(ZERO_OR_NULL_PTR(s))) |
6cb8f913 | 3879 | return s; |
81819f0f | 3880 | |
2b847c3c | 3881 | ret = slab_alloc_node(s, gfpflags, node, caller); |
94b528d0 | 3882 | |
25985edc | 3883 | /* Honor the call site pointer we received. */ |
ca2b84cb | 3884 | trace_kmalloc_node(caller, ret, size, s->size, gfpflags, node); |
94b528d0 EGM |
3885 | |
3886 | return ret; | |
81819f0f | 3887 | } |
5d1f57e4 | 3888 | #endif |
81819f0f | 3889 | |
ab4d5ed5 | 3890 | #ifdef CONFIG_SYSFS |
205ab99d CL |
3891 | static int count_inuse(struct page *page) |
3892 | { | |
3893 | return page->inuse; | |
3894 | } | |
3895 | ||
3896 | static int count_total(struct page *page) | |
3897 | { | |
3898 | return page->objects; | |
3899 | } | |
ab4d5ed5 | 3900 | #endif |
205ab99d | 3901 | |
ab4d5ed5 | 3902 | #ifdef CONFIG_SLUB_DEBUG |
434e245d CL |
3903 | static int validate_slab(struct kmem_cache *s, struct page *page, |
3904 | unsigned long *map) | |
53e15af0 CL |
3905 | { |
3906 | void *p; | |
a973e9dd | 3907 | void *addr = page_address(page); |
53e15af0 CL |
3908 | |
3909 | if (!check_slab(s, page) || | |
3910 | !on_freelist(s, page, NULL)) | |
3911 | return 0; | |
3912 | ||
3913 | /* Now we know that a valid freelist exists */ | |
39b26464 | 3914 | bitmap_zero(map, page->objects); |
53e15af0 | 3915 | |
5f80b13a CL |
3916 | get_map(s, page, map); |
3917 | for_each_object(p, s, addr, page->objects) { | |
3918 | if (test_bit(slab_index(p, s, addr), map)) | |
3919 | if (!check_object(s, page, p, SLUB_RED_INACTIVE)) | |
3920 | return 0; | |
53e15af0 CL |
3921 | } |
3922 | ||
224a88be | 3923 | for_each_object(p, s, addr, page->objects) |
7656c72b | 3924 | if (!test_bit(slab_index(p, s, addr), map)) |
37d57443 | 3925 | if (!check_object(s, page, p, SLUB_RED_ACTIVE)) |
53e15af0 CL |
3926 | return 0; |
3927 | return 1; | |
3928 | } | |
3929 | ||
434e245d CL |
3930 | static void validate_slab_slab(struct kmem_cache *s, struct page *page, |
3931 | unsigned long *map) | |
53e15af0 | 3932 | { |
881db7fb CL |
3933 | slab_lock(page); |
3934 | validate_slab(s, page, map); | |
3935 | slab_unlock(page); | |
53e15af0 CL |
3936 | } |
3937 | ||
434e245d CL |
3938 | static int validate_slab_node(struct kmem_cache *s, |
3939 | struct kmem_cache_node *n, unsigned long *map) | |
53e15af0 CL |
3940 | { |
3941 | unsigned long count = 0; | |
3942 | struct page *page; | |
3943 | unsigned long flags; | |
3944 | ||
3945 | spin_lock_irqsave(&n->list_lock, flags); | |
3946 | ||
3947 | list_for_each_entry(page, &n->partial, lru) { | |
434e245d | 3948 | validate_slab_slab(s, page, map); |
53e15af0 CL |
3949 | count++; |
3950 | } | |
3951 | if (count != n->nr_partial) | |
3952 | printk(KERN_ERR "SLUB %s: %ld partial slabs counted but " | |
3953 | "counter=%ld\n", s->name, count, n->nr_partial); | |
3954 | ||
3955 | if (!(s->flags & SLAB_STORE_USER)) | |
3956 | goto out; | |
3957 | ||
3958 | list_for_each_entry(page, &n->full, lru) { | |
434e245d | 3959 | validate_slab_slab(s, page, map); |
53e15af0 CL |
3960 | count++; |
3961 | } | |
3962 | if (count != atomic_long_read(&n->nr_slabs)) | |
3963 | printk(KERN_ERR "SLUB: %s %ld slabs counted but " | |
3964 | "counter=%ld\n", s->name, count, | |
3965 | atomic_long_read(&n->nr_slabs)); | |
3966 | ||
3967 | out: | |
3968 | spin_unlock_irqrestore(&n->list_lock, flags); | |
3969 | return count; | |
3970 | } | |
3971 | ||
434e245d | 3972 | static long validate_slab_cache(struct kmem_cache *s) |
53e15af0 CL |
3973 | { |
3974 | int node; | |
3975 | unsigned long count = 0; | |
205ab99d | 3976 | unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) * |
434e245d CL |
3977 | sizeof(unsigned long), GFP_KERNEL); |
3978 | ||
3979 | if (!map) | |
3980 | return -ENOMEM; | |
53e15af0 CL |
3981 | |
3982 | flush_all(s); | |
f64dc58c | 3983 | for_each_node_state(node, N_NORMAL_MEMORY) { |
53e15af0 CL |
3984 | struct kmem_cache_node *n = get_node(s, node); |
3985 | ||
434e245d | 3986 | count += validate_slab_node(s, n, map); |
53e15af0 | 3987 | } |
434e245d | 3988 | kfree(map); |
53e15af0 CL |
3989 | return count; |
3990 | } | |
88a420e4 | 3991 | /* |
672bba3a | 3992 | * Generate lists of code addresses where slabcache objects are allocated |
88a420e4 CL |
3993 | * and freed. |
3994 | */ | |
3995 | ||
3996 | struct location { | |
3997 | unsigned long count; | |
ce71e27c | 3998 | unsigned long addr; |
45edfa58 CL |
3999 | long long sum_time; |
4000 | long min_time; | |
4001 | long max_time; | |
4002 | long min_pid; | |
4003 | long max_pid; | |
174596a0 | 4004 | DECLARE_BITMAP(cpus, NR_CPUS); |
45edfa58 | 4005 | nodemask_t nodes; |
88a420e4 CL |
4006 | }; |
4007 | ||
4008 | struct loc_track { | |
4009 | unsigned long max; | |
4010 | unsigned long count; | |
4011 | struct location *loc; | |
4012 | }; | |
4013 | ||
4014 | static void free_loc_track(struct loc_track *t) | |
4015 | { | |
4016 | if (t->max) | |
4017 | free_pages((unsigned long)t->loc, | |
4018 | get_order(sizeof(struct location) * t->max)); | |
4019 | } | |
4020 | ||
68dff6a9 | 4021 | static int alloc_loc_track(struct loc_track *t, unsigned long max, gfp_t flags) |
88a420e4 CL |
4022 | { |
4023 | struct location *l; | |
4024 | int order; | |
4025 | ||
88a420e4 CL |
4026 | order = get_order(sizeof(struct location) * max); |
4027 | ||
68dff6a9 | 4028 | l = (void *)__get_free_pages(flags, order); |
88a420e4 CL |
4029 | if (!l) |
4030 | return 0; | |
4031 | ||
4032 | if (t->count) { | |
4033 | memcpy(l, t->loc, sizeof(struct location) * t->count); | |
4034 | free_loc_track(t); | |
4035 | } | |
4036 | t->max = max; | |
4037 | t->loc = l; | |
4038 | return 1; | |
4039 | } | |
4040 | ||
4041 | static int add_location(struct loc_track *t, struct kmem_cache *s, | |
45edfa58 | 4042 | const struct track *track) |
88a420e4 CL |
4043 | { |
4044 | long start, end, pos; | |
4045 | struct location *l; | |
ce71e27c | 4046 | unsigned long caddr; |
45edfa58 | 4047 | unsigned long age = jiffies - track->when; |
88a420e4 CL |
4048 | |
4049 | start = -1; | |
4050 | end = t->count; | |
4051 | ||
4052 | for ( ; ; ) { | |
4053 | pos = start + (end - start + 1) / 2; | |
4054 | ||
4055 | /* | |
4056 | * There is nothing at "end". If we end up there | |
4057 | * we need to add something to before end. | |
4058 | */ | |
4059 | if (pos == end) | |
4060 | break; | |
4061 | ||
4062 | caddr = t->loc[pos].addr; | |
45edfa58 CL |
4063 | if (track->addr == caddr) { |
4064 | ||
4065 | l = &t->loc[pos]; | |
4066 | l->count++; | |
4067 | if (track->when) { | |
4068 | l->sum_time += age; | |
4069 | if (age < l->min_time) | |
4070 | l->min_time = age; | |
4071 | if (age > l->max_time) | |
4072 | l->max_time = age; | |
4073 | ||
4074 | if (track->pid < l->min_pid) | |
4075 | l->min_pid = track->pid; | |
4076 | if (track->pid > l->max_pid) | |
4077 | l->max_pid = track->pid; | |
4078 | ||
174596a0 RR |
4079 | cpumask_set_cpu(track->cpu, |
4080 | to_cpumask(l->cpus)); | |
45edfa58 CL |
4081 | } |
4082 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
88a420e4 CL |
4083 | return 1; |
4084 | } | |
4085 | ||
45edfa58 | 4086 | if (track->addr < caddr) |
88a420e4 CL |
4087 | end = pos; |
4088 | else | |
4089 | start = pos; | |
4090 | } | |
4091 | ||
4092 | /* | |
672bba3a | 4093 | * Not found. Insert new tracking element. |
88a420e4 | 4094 | */ |
68dff6a9 | 4095 | if (t->count >= t->max && !alloc_loc_track(t, 2 * t->max, GFP_ATOMIC)) |
88a420e4 CL |
4096 | return 0; |
4097 | ||
4098 | l = t->loc + pos; | |
4099 | if (pos < t->count) | |
4100 | memmove(l + 1, l, | |
4101 | (t->count - pos) * sizeof(struct location)); | |
4102 | t->count++; | |
4103 | l->count = 1; | |
45edfa58 CL |
4104 | l->addr = track->addr; |
4105 | l->sum_time = age; | |
4106 | l->min_time = age; | |
4107 | l->max_time = age; | |
4108 | l->min_pid = track->pid; | |
4109 | l->max_pid = track->pid; | |
174596a0 RR |
4110 | cpumask_clear(to_cpumask(l->cpus)); |
4111 | cpumask_set_cpu(track->cpu, to_cpumask(l->cpus)); | |
45edfa58 CL |
4112 | nodes_clear(l->nodes); |
4113 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
88a420e4 CL |
4114 | return 1; |
4115 | } | |
4116 | ||
4117 | static void process_slab(struct loc_track *t, struct kmem_cache *s, | |
bbd7d57b | 4118 | struct page *page, enum track_item alloc, |
a5dd5c11 | 4119 | unsigned long *map) |
88a420e4 | 4120 | { |
a973e9dd | 4121 | void *addr = page_address(page); |
88a420e4 CL |
4122 | void *p; |
4123 | ||
39b26464 | 4124 | bitmap_zero(map, page->objects); |
5f80b13a | 4125 | get_map(s, page, map); |
88a420e4 | 4126 | |
224a88be | 4127 | for_each_object(p, s, addr, page->objects) |
45edfa58 CL |
4128 | if (!test_bit(slab_index(p, s, addr), map)) |
4129 | add_location(t, s, get_track(s, p, alloc)); | |
88a420e4 CL |
4130 | } |
4131 | ||
4132 | static int list_locations(struct kmem_cache *s, char *buf, | |
4133 | enum track_item alloc) | |
4134 | { | |
e374d483 | 4135 | int len = 0; |
88a420e4 | 4136 | unsigned long i; |
68dff6a9 | 4137 | struct loc_track t = { 0, 0, NULL }; |
88a420e4 | 4138 | int node; |
bbd7d57b ED |
4139 | unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) * |
4140 | sizeof(unsigned long), GFP_KERNEL); | |
88a420e4 | 4141 | |
bbd7d57b ED |
4142 | if (!map || !alloc_loc_track(&t, PAGE_SIZE / sizeof(struct location), |
4143 | GFP_TEMPORARY)) { | |
4144 | kfree(map); | |
68dff6a9 | 4145 | return sprintf(buf, "Out of memory\n"); |
bbd7d57b | 4146 | } |
88a420e4 CL |
4147 | /* Push back cpu slabs */ |
4148 | flush_all(s); | |
4149 | ||
f64dc58c | 4150 | for_each_node_state(node, N_NORMAL_MEMORY) { |
88a420e4 CL |
4151 | struct kmem_cache_node *n = get_node(s, node); |
4152 | unsigned long flags; | |
4153 | struct page *page; | |
4154 | ||
9e86943b | 4155 | if (!atomic_long_read(&n->nr_slabs)) |
88a420e4 CL |
4156 | continue; |
4157 | ||
4158 | spin_lock_irqsave(&n->list_lock, flags); | |
4159 | list_for_each_entry(page, &n->partial, lru) | |
bbd7d57b | 4160 | process_slab(&t, s, page, alloc, map); |
88a420e4 | 4161 | list_for_each_entry(page, &n->full, lru) |
bbd7d57b | 4162 | process_slab(&t, s, page, alloc, map); |
88a420e4 CL |
4163 | spin_unlock_irqrestore(&n->list_lock, flags); |
4164 | } | |
4165 | ||
4166 | for (i = 0; i < t.count; i++) { | |
45edfa58 | 4167 | struct location *l = &t.loc[i]; |
88a420e4 | 4168 | |
9c246247 | 4169 | if (len > PAGE_SIZE - KSYM_SYMBOL_LEN - 100) |
88a420e4 | 4170 | break; |
e374d483 | 4171 | len += sprintf(buf + len, "%7ld ", l->count); |
45edfa58 CL |
4172 | |
4173 | if (l->addr) | |
62c70bce | 4174 | len += sprintf(buf + len, "%pS", (void *)l->addr); |
88a420e4 | 4175 | else |
e374d483 | 4176 | len += sprintf(buf + len, "<not-available>"); |
45edfa58 CL |
4177 | |
4178 | if (l->sum_time != l->min_time) { | |
e374d483 | 4179 | len += sprintf(buf + len, " age=%ld/%ld/%ld", |
f8bd2258 RZ |
4180 | l->min_time, |
4181 | (long)div_u64(l->sum_time, l->count), | |
4182 | l->max_time); | |
45edfa58 | 4183 | } else |
e374d483 | 4184 | len += sprintf(buf + len, " age=%ld", |
45edfa58 CL |
4185 | l->min_time); |
4186 | ||
4187 | if (l->min_pid != l->max_pid) | |
e374d483 | 4188 | len += sprintf(buf + len, " pid=%ld-%ld", |
45edfa58 CL |
4189 | l->min_pid, l->max_pid); |
4190 | else | |
e374d483 | 4191 | len += sprintf(buf + len, " pid=%ld", |
45edfa58 CL |
4192 | l->min_pid); |
4193 | ||
174596a0 RR |
4194 | if (num_online_cpus() > 1 && |
4195 | !cpumask_empty(to_cpumask(l->cpus)) && | |
e374d483 HH |
4196 | len < PAGE_SIZE - 60) { |
4197 | len += sprintf(buf + len, " cpus="); | |
d0e0ac97 CG |
4198 | len += cpulist_scnprintf(buf + len, |
4199 | PAGE_SIZE - len - 50, | |
174596a0 | 4200 | to_cpumask(l->cpus)); |
45edfa58 CL |
4201 | } |
4202 | ||
62bc62a8 | 4203 | if (nr_online_nodes > 1 && !nodes_empty(l->nodes) && |
e374d483 HH |
4204 | len < PAGE_SIZE - 60) { |
4205 | len += sprintf(buf + len, " nodes="); | |
d0e0ac97 CG |
4206 | len += nodelist_scnprintf(buf + len, |
4207 | PAGE_SIZE - len - 50, | |
4208 | l->nodes); | |
45edfa58 CL |
4209 | } |
4210 | ||
e374d483 | 4211 | len += sprintf(buf + len, "\n"); |
88a420e4 CL |
4212 | } |
4213 | ||
4214 | free_loc_track(&t); | |
bbd7d57b | 4215 | kfree(map); |
88a420e4 | 4216 | if (!t.count) |
e374d483 HH |
4217 | len += sprintf(buf, "No data\n"); |
4218 | return len; | |
88a420e4 | 4219 | } |
ab4d5ed5 | 4220 | #endif |
88a420e4 | 4221 | |
a5a84755 CL |
4222 | #ifdef SLUB_RESILIENCY_TEST |
4223 | static void resiliency_test(void) | |
4224 | { | |
4225 | u8 *p; | |
4226 | ||
95a05b42 | 4227 | BUILD_BUG_ON(KMALLOC_MIN_SIZE > 16 || KMALLOC_SHIFT_HIGH < 10); |
a5a84755 CL |
4228 | |
4229 | printk(KERN_ERR "SLUB resiliency testing\n"); | |
4230 | printk(KERN_ERR "-----------------------\n"); | |
4231 | printk(KERN_ERR "A. Corruption after allocation\n"); | |
4232 | ||
4233 | p = kzalloc(16, GFP_KERNEL); | |
4234 | p[16] = 0x12; | |
4235 | printk(KERN_ERR "\n1. kmalloc-16: Clobber Redzone/next pointer" | |
4236 | " 0x12->0x%p\n\n", p + 16); | |
4237 | ||
4238 | validate_slab_cache(kmalloc_caches[4]); | |
4239 | ||
4240 | /* Hmmm... The next two are dangerous */ | |
4241 | p = kzalloc(32, GFP_KERNEL); | |
4242 | p[32 + sizeof(void *)] = 0x34; | |
4243 | printk(KERN_ERR "\n2. kmalloc-32: Clobber next pointer/next slab" | |
4244 | " 0x34 -> -0x%p\n", p); | |
4245 | printk(KERN_ERR | |
4246 | "If allocated object is overwritten then not detectable\n\n"); | |
4247 | ||
4248 | validate_slab_cache(kmalloc_caches[5]); | |
4249 | p = kzalloc(64, GFP_KERNEL); | |
4250 | p += 64 + (get_cycles() & 0xff) * sizeof(void *); | |
4251 | *p = 0x56; | |
4252 | printk(KERN_ERR "\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n", | |
4253 | p); | |
4254 | printk(KERN_ERR | |
4255 | "If allocated object is overwritten then not detectable\n\n"); | |
4256 | validate_slab_cache(kmalloc_caches[6]); | |
4257 | ||
4258 | printk(KERN_ERR "\nB. Corruption after free\n"); | |
4259 | p = kzalloc(128, GFP_KERNEL); | |
4260 | kfree(p); | |
4261 | *p = 0x78; | |
4262 | printk(KERN_ERR "1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p); | |
4263 | validate_slab_cache(kmalloc_caches[7]); | |
4264 | ||
4265 | p = kzalloc(256, GFP_KERNEL); | |
4266 | kfree(p); | |
4267 | p[50] = 0x9a; | |
4268 | printk(KERN_ERR "\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n", | |
4269 | p); | |
4270 | validate_slab_cache(kmalloc_caches[8]); | |
4271 | ||
4272 | p = kzalloc(512, GFP_KERNEL); | |
4273 | kfree(p); | |
4274 | p[512] = 0xab; | |
4275 | printk(KERN_ERR "\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p); | |
4276 | validate_slab_cache(kmalloc_caches[9]); | |
4277 | } | |
4278 | #else | |
4279 | #ifdef CONFIG_SYSFS | |
4280 | static void resiliency_test(void) {}; | |
4281 | #endif | |
4282 | #endif | |
4283 | ||
ab4d5ed5 | 4284 | #ifdef CONFIG_SYSFS |
81819f0f | 4285 | enum slab_stat_type { |
205ab99d CL |
4286 | SL_ALL, /* All slabs */ |
4287 | SL_PARTIAL, /* Only partially allocated slabs */ | |
4288 | SL_CPU, /* Only slabs used for cpu caches */ | |
4289 | SL_OBJECTS, /* Determine allocated objects not slabs */ | |
4290 | SL_TOTAL /* Determine object capacity not slabs */ | |
81819f0f CL |
4291 | }; |
4292 | ||
205ab99d | 4293 | #define SO_ALL (1 << SL_ALL) |
81819f0f CL |
4294 | #define SO_PARTIAL (1 << SL_PARTIAL) |
4295 | #define SO_CPU (1 << SL_CPU) | |
4296 | #define SO_OBJECTS (1 << SL_OBJECTS) | |
205ab99d | 4297 | #define SO_TOTAL (1 << SL_TOTAL) |
81819f0f | 4298 | |
62e5c4b4 CG |
4299 | static ssize_t show_slab_objects(struct kmem_cache *s, |
4300 | char *buf, unsigned long flags) | |
81819f0f CL |
4301 | { |
4302 | unsigned long total = 0; | |
81819f0f CL |
4303 | int node; |
4304 | int x; | |
4305 | unsigned long *nodes; | |
81819f0f | 4306 | |
e35e1a97 | 4307 | nodes = kzalloc(sizeof(unsigned long) * nr_node_ids, GFP_KERNEL); |
62e5c4b4 CG |
4308 | if (!nodes) |
4309 | return -ENOMEM; | |
81819f0f | 4310 | |
205ab99d CL |
4311 | if (flags & SO_CPU) { |
4312 | int cpu; | |
81819f0f | 4313 | |
205ab99d | 4314 | for_each_possible_cpu(cpu) { |
d0e0ac97 CG |
4315 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, |
4316 | cpu); | |
ec3ab083 | 4317 | int node; |
49e22585 | 4318 | struct page *page; |
dfb4f096 | 4319 | |
bc6697d8 | 4320 | page = ACCESS_ONCE(c->page); |
ec3ab083 CL |
4321 | if (!page) |
4322 | continue; | |
205ab99d | 4323 | |
ec3ab083 CL |
4324 | node = page_to_nid(page); |
4325 | if (flags & SO_TOTAL) | |
4326 | x = page->objects; | |
4327 | else if (flags & SO_OBJECTS) | |
4328 | x = page->inuse; | |
4329 | else | |
4330 | x = 1; | |
49e22585 | 4331 | |
ec3ab083 CL |
4332 | total += x; |
4333 | nodes[node] += x; | |
4334 | ||
4335 | page = ACCESS_ONCE(c->partial); | |
49e22585 | 4336 | if (page) { |
8afb1474 LZ |
4337 | node = page_to_nid(page); |
4338 | if (flags & SO_TOTAL) | |
4339 | WARN_ON_ONCE(1); | |
4340 | else if (flags & SO_OBJECTS) | |
4341 | WARN_ON_ONCE(1); | |
4342 | else | |
4343 | x = page->pages; | |
bc6697d8 ED |
4344 | total += x; |
4345 | nodes[node] += x; | |
49e22585 | 4346 | } |
81819f0f CL |
4347 | } |
4348 | } | |
4349 | ||
04d94879 | 4350 | lock_memory_hotplug(); |
ab4d5ed5 | 4351 | #ifdef CONFIG_SLUB_DEBUG |
205ab99d CL |
4352 | if (flags & SO_ALL) { |
4353 | for_each_node_state(node, N_NORMAL_MEMORY) { | |
4354 | struct kmem_cache_node *n = get_node(s, node); | |
4355 | ||
d0e0ac97 CG |
4356 | if (flags & SO_TOTAL) |
4357 | x = atomic_long_read(&n->total_objects); | |
4358 | else if (flags & SO_OBJECTS) | |
4359 | x = atomic_long_read(&n->total_objects) - | |
4360 | count_partial(n, count_free); | |
81819f0f | 4361 | else |
205ab99d | 4362 | x = atomic_long_read(&n->nr_slabs); |
81819f0f CL |
4363 | total += x; |
4364 | nodes[node] += x; | |
4365 | } | |
4366 | ||
ab4d5ed5 CL |
4367 | } else |
4368 | #endif | |
4369 | if (flags & SO_PARTIAL) { | |
205ab99d CL |
4370 | for_each_node_state(node, N_NORMAL_MEMORY) { |
4371 | struct kmem_cache_node *n = get_node(s, node); | |
81819f0f | 4372 | |
205ab99d CL |
4373 | if (flags & SO_TOTAL) |
4374 | x = count_partial(n, count_total); | |
4375 | else if (flags & SO_OBJECTS) | |
4376 | x = count_partial(n, count_inuse); | |
81819f0f | 4377 | else |
205ab99d | 4378 | x = n->nr_partial; |
81819f0f CL |
4379 | total += x; |
4380 | nodes[node] += x; | |
4381 | } | |
4382 | } | |
81819f0f CL |
4383 | x = sprintf(buf, "%lu", total); |
4384 | #ifdef CONFIG_NUMA | |
f64dc58c | 4385 | for_each_node_state(node, N_NORMAL_MEMORY) |
81819f0f CL |
4386 | if (nodes[node]) |
4387 | x += sprintf(buf + x, " N%d=%lu", | |
4388 | node, nodes[node]); | |
4389 | #endif | |
04d94879 | 4390 | unlock_memory_hotplug(); |
81819f0f CL |
4391 | kfree(nodes); |
4392 | return x + sprintf(buf + x, "\n"); | |
4393 | } | |
4394 | ||
ab4d5ed5 | 4395 | #ifdef CONFIG_SLUB_DEBUG |
81819f0f CL |
4396 | static int any_slab_objects(struct kmem_cache *s) |
4397 | { | |
4398 | int node; | |
81819f0f | 4399 | |
dfb4f096 | 4400 | for_each_online_node(node) { |
81819f0f CL |
4401 | struct kmem_cache_node *n = get_node(s, node); |
4402 | ||
dfb4f096 CL |
4403 | if (!n) |
4404 | continue; | |
4405 | ||
4ea33e2d | 4406 | if (atomic_long_read(&n->total_objects)) |
81819f0f CL |
4407 | return 1; |
4408 | } | |
4409 | return 0; | |
4410 | } | |
ab4d5ed5 | 4411 | #endif |
81819f0f CL |
4412 | |
4413 | #define to_slab_attr(n) container_of(n, struct slab_attribute, attr) | |
497888cf | 4414 | #define to_slab(n) container_of(n, struct kmem_cache, kobj) |
81819f0f CL |
4415 | |
4416 | struct slab_attribute { | |
4417 | struct attribute attr; | |
4418 | ssize_t (*show)(struct kmem_cache *s, char *buf); | |
4419 | ssize_t (*store)(struct kmem_cache *s, const char *x, size_t count); | |
4420 | }; | |
4421 | ||
4422 | #define SLAB_ATTR_RO(_name) \ | |
ab067e99 VK |
4423 | static struct slab_attribute _name##_attr = \ |
4424 | __ATTR(_name, 0400, _name##_show, NULL) | |
81819f0f CL |
4425 | |
4426 | #define SLAB_ATTR(_name) \ | |
4427 | static struct slab_attribute _name##_attr = \ | |
ab067e99 | 4428 | __ATTR(_name, 0600, _name##_show, _name##_store) |
81819f0f | 4429 | |
81819f0f CL |
4430 | static ssize_t slab_size_show(struct kmem_cache *s, char *buf) |
4431 | { | |
4432 | return sprintf(buf, "%d\n", s->size); | |
4433 | } | |
4434 | SLAB_ATTR_RO(slab_size); | |
4435 | ||
4436 | static ssize_t align_show(struct kmem_cache *s, char *buf) | |
4437 | { | |
4438 | return sprintf(buf, "%d\n", s->align); | |
4439 | } | |
4440 | SLAB_ATTR_RO(align); | |
4441 | ||
4442 | static ssize_t object_size_show(struct kmem_cache *s, char *buf) | |
4443 | { | |
3b0efdfa | 4444 | return sprintf(buf, "%d\n", s->object_size); |
81819f0f CL |
4445 | } |
4446 | SLAB_ATTR_RO(object_size); | |
4447 | ||
4448 | static ssize_t objs_per_slab_show(struct kmem_cache *s, char *buf) | |
4449 | { | |
834f3d11 | 4450 | return sprintf(buf, "%d\n", oo_objects(s->oo)); |
81819f0f CL |
4451 | } |
4452 | SLAB_ATTR_RO(objs_per_slab); | |
4453 | ||
06b285dc CL |
4454 | static ssize_t order_store(struct kmem_cache *s, |
4455 | const char *buf, size_t length) | |
4456 | { | |
0121c619 CL |
4457 | unsigned long order; |
4458 | int err; | |
4459 | ||
3dbb95f7 | 4460 | err = kstrtoul(buf, 10, &order); |
0121c619 CL |
4461 | if (err) |
4462 | return err; | |
06b285dc CL |
4463 | |
4464 | if (order > slub_max_order || order < slub_min_order) | |
4465 | return -EINVAL; | |
4466 | ||
4467 | calculate_sizes(s, order); | |
4468 | return length; | |
4469 | } | |
4470 | ||
81819f0f CL |
4471 | static ssize_t order_show(struct kmem_cache *s, char *buf) |
4472 | { | |
834f3d11 | 4473 | return sprintf(buf, "%d\n", oo_order(s->oo)); |
81819f0f | 4474 | } |
06b285dc | 4475 | SLAB_ATTR(order); |
81819f0f | 4476 | |
73d342b1 DR |
4477 | static ssize_t min_partial_show(struct kmem_cache *s, char *buf) |
4478 | { | |
4479 | return sprintf(buf, "%lu\n", s->min_partial); | |
4480 | } | |
4481 | ||
4482 | static ssize_t min_partial_store(struct kmem_cache *s, const char *buf, | |
4483 | size_t length) | |
4484 | { | |
4485 | unsigned long min; | |
4486 | int err; | |
4487 | ||
3dbb95f7 | 4488 | err = kstrtoul(buf, 10, &min); |
73d342b1 DR |
4489 | if (err) |
4490 | return err; | |
4491 | ||
c0bdb232 | 4492 | set_min_partial(s, min); |
73d342b1 DR |
4493 | return length; |
4494 | } | |
4495 | SLAB_ATTR(min_partial); | |
4496 | ||
49e22585 CL |
4497 | static ssize_t cpu_partial_show(struct kmem_cache *s, char *buf) |
4498 | { | |
4499 | return sprintf(buf, "%u\n", s->cpu_partial); | |
4500 | } | |
4501 | ||
4502 | static ssize_t cpu_partial_store(struct kmem_cache *s, const char *buf, | |
4503 | size_t length) | |
4504 | { | |
4505 | unsigned long objects; | |
4506 | int err; | |
4507 | ||
3dbb95f7 | 4508 | err = kstrtoul(buf, 10, &objects); |
49e22585 CL |
4509 | if (err) |
4510 | return err; | |
345c905d | 4511 | if (objects && !kmem_cache_has_cpu_partial(s)) |
74ee4ef1 | 4512 | return -EINVAL; |
49e22585 CL |
4513 | |
4514 | s->cpu_partial = objects; | |
4515 | flush_all(s); | |
4516 | return length; | |
4517 | } | |
4518 | SLAB_ATTR(cpu_partial); | |
4519 | ||
81819f0f CL |
4520 | static ssize_t ctor_show(struct kmem_cache *s, char *buf) |
4521 | { | |
62c70bce JP |
4522 | if (!s->ctor) |
4523 | return 0; | |
4524 | return sprintf(buf, "%pS\n", s->ctor); | |
81819f0f CL |
4525 | } |
4526 | SLAB_ATTR_RO(ctor); | |
4527 | ||
81819f0f CL |
4528 | static ssize_t aliases_show(struct kmem_cache *s, char *buf) |
4529 | { | |
4530 | return sprintf(buf, "%d\n", s->refcount - 1); | |
4531 | } | |
4532 | SLAB_ATTR_RO(aliases); | |
4533 | ||
81819f0f CL |
4534 | static ssize_t partial_show(struct kmem_cache *s, char *buf) |
4535 | { | |
d9acf4b7 | 4536 | return show_slab_objects(s, buf, SO_PARTIAL); |
81819f0f CL |
4537 | } |
4538 | SLAB_ATTR_RO(partial); | |
4539 | ||
4540 | static ssize_t cpu_slabs_show(struct kmem_cache *s, char *buf) | |
4541 | { | |
d9acf4b7 | 4542 | return show_slab_objects(s, buf, SO_CPU); |
81819f0f CL |
4543 | } |
4544 | SLAB_ATTR_RO(cpu_slabs); | |
4545 | ||
4546 | static ssize_t objects_show(struct kmem_cache *s, char *buf) | |
4547 | { | |
205ab99d | 4548 | return show_slab_objects(s, buf, SO_ALL|SO_OBJECTS); |
81819f0f CL |
4549 | } |
4550 | SLAB_ATTR_RO(objects); | |
4551 | ||
205ab99d CL |
4552 | static ssize_t objects_partial_show(struct kmem_cache *s, char *buf) |
4553 | { | |
4554 | return show_slab_objects(s, buf, SO_PARTIAL|SO_OBJECTS); | |
4555 | } | |
4556 | SLAB_ATTR_RO(objects_partial); | |
4557 | ||
49e22585 CL |
4558 | static ssize_t slabs_cpu_partial_show(struct kmem_cache *s, char *buf) |
4559 | { | |
4560 | int objects = 0; | |
4561 | int pages = 0; | |
4562 | int cpu; | |
4563 | int len; | |
4564 | ||
4565 | for_each_online_cpu(cpu) { | |
4566 | struct page *page = per_cpu_ptr(s->cpu_slab, cpu)->partial; | |
4567 | ||
4568 | if (page) { | |
4569 | pages += page->pages; | |
4570 | objects += page->pobjects; | |
4571 | } | |
4572 | } | |
4573 | ||
4574 | len = sprintf(buf, "%d(%d)", objects, pages); | |
4575 | ||
4576 | #ifdef CONFIG_SMP | |
4577 | for_each_online_cpu(cpu) { | |
4578 | struct page *page = per_cpu_ptr(s->cpu_slab, cpu) ->partial; | |
4579 | ||
4580 | if (page && len < PAGE_SIZE - 20) | |
4581 | len += sprintf(buf + len, " C%d=%d(%d)", cpu, | |
4582 | page->pobjects, page->pages); | |
4583 | } | |
4584 | #endif | |
4585 | return len + sprintf(buf + len, "\n"); | |
4586 | } | |
4587 | SLAB_ATTR_RO(slabs_cpu_partial); | |
4588 | ||
a5a84755 CL |
4589 | static ssize_t reclaim_account_show(struct kmem_cache *s, char *buf) |
4590 | { | |
4591 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_RECLAIM_ACCOUNT)); | |
4592 | } | |
4593 | ||
4594 | static ssize_t reclaim_account_store(struct kmem_cache *s, | |
4595 | const char *buf, size_t length) | |
4596 | { | |
4597 | s->flags &= ~SLAB_RECLAIM_ACCOUNT; | |
4598 | if (buf[0] == '1') | |
4599 | s->flags |= SLAB_RECLAIM_ACCOUNT; | |
4600 | return length; | |
4601 | } | |
4602 | SLAB_ATTR(reclaim_account); | |
4603 | ||
4604 | static ssize_t hwcache_align_show(struct kmem_cache *s, char *buf) | |
4605 | { | |
4606 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_HWCACHE_ALIGN)); | |
4607 | } | |
4608 | SLAB_ATTR_RO(hwcache_align); | |
4609 | ||
4610 | #ifdef CONFIG_ZONE_DMA | |
4611 | static ssize_t cache_dma_show(struct kmem_cache *s, char *buf) | |
4612 | { | |
4613 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_CACHE_DMA)); | |
4614 | } | |
4615 | SLAB_ATTR_RO(cache_dma); | |
4616 | #endif | |
4617 | ||
4618 | static ssize_t destroy_by_rcu_show(struct kmem_cache *s, char *buf) | |
4619 | { | |
4620 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_DESTROY_BY_RCU)); | |
4621 | } | |
4622 | SLAB_ATTR_RO(destroy_by_rcu); | |
4623 | ||
ab9a0f19 LJ |
4624 | static ssize_t reserved_show(struct kmem_cache *s, char *buf) |
4625 | { | |
4626 | return sprintf(buf, "%d\n", s->reserved); | |
4627 | } | |
4628 | SLAB_ATTR_RO(reserved); | |
4629 | ||
ab4d5ed5 | 4630 | #ifdef CONFIG_SLUB_DEBUG |
a5a84755 CL |
4631 | static ssize_t slabs_show(struct kmem_cache *s, char *buf) |
4632 | { | |
4633 | return show_slab_objects(s, buf, SO_ALL); | |
4634 | } | |
4635 | SLAB_ATTR_RO(slabs); | |
4636 | ||
205ab99d CL |
4637 | static ssize_t total_objects_show(struct kmem_cache *s, char *buf) |
4638 | { | |
4639 | return show_slab_objects(s, buf, SO_ALL|SO_TOTAL); | |
4640 | } | |
4641 | SLAB_ATTR_RO(total_objects); | |
4642 | ||
81819f0f CL |
4643 | static ssize_t sanity_checks_show(struct kmem_cache *s, char *buf) |
4644 | { | |
4645 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_DEBUG_FREE)); | |
4646 | } | |
4647 | ||
4648 | static ssize_t sanity_checks_store(struct kmem_cache *s, | |
4649 | const char *buf, size_t length) | |
4650 | { | |
4651 | s->flags &= ~SLAB_DEBUG_FREE; | |
b789ef51 CL |
4652 | if (buf[0] == '1') { |
4653 | s->flags &= ~__CMPXCHG_DOUBLE; | |
81819f0f | 4654 | s->flags |= SLAB_DEBUG_FREE; |
b789ef51 | 4655 | } |
81819f0f CL |
4656 | return length; |
4657 | } | |
4658 | SLAB_ATTR(sanity_checks); | |
4659 | ||
4660 | static ssize_t trace_show(struct kmem_cache *s, char *buf) | |
4661 | { | |
4662 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_TRACE)); | |
4663 | } | |
4664 | ||
4665 | static ssize_t trace_store(struct kmem_cache *s, const char *buf, | |
4666 | size_t length) | |
4667 | { | |
4668 | s->flags &= ~SLAB_TRACE; | |
b789ef51 CL |
4669 | if (buf[0] == '1') { |
4670 | s->flags &= ~__CMPXCHG_DOUBLE; | |
81819f0f | 4671 | s->flags |= SLAB_TRACE; |
b789ef51 | 4672 | } |
81819f0f CL |
4673 | return length; |
4674 | } | |
4675 | SLAB_ATTR(trace); | |
4676 | ||
81819f0f CL |
4677 | static ssize_t red_zone_show(struct kmem_cache *s, char *buf) |
4678 | { | |
4679 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_RED_ZONE)); | |
4680 | } | |
4681 | ||
4682 | static ssize_t red_zone_store(struct kmem_cache *s, | |
4683 | const char *buf, size_t length) | |
4684 | { | |
4685 | if (any_slab_objects(s)) | |
4686 | return -EBUSY; | |
4687 | ||
4688 | s->flags &= ~SLAB_RED_ZONE; | |
b789ef51 CL |
4689 | if (buf[0] == '1') { |
4690 | s->flags &= ~__CMPXCHG_DOUBLE; | |
81819f0f | 4691 | s->flags |= SLAB_RED_ZONE; |
b789ef51 | 4692 | } |
06b285dc | 4693 | calculate_sizes(s, -1); |
81819f0f CL |
4694 | return length; |
4695 | } | |
4696 | SLAB_ATTR(red_zone); | |
4697 | ||
4698 | static ssize_t poison_show(struct kmem_cache *s, char *buf) | |
4699 | { | |
4700 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_POISON)); | |
4701 | } | |
4702 | ||
4703 | static ssize_t poison_store(struct kmem_cache *s, | |
4704 | const char *buf, size_t length) | |
4705 | { | |
4706 | if (any_slab_objects(s)) | |
4707 | return -EBUSY; | |
4708 | ||
4709 | s->flags &= ~SLAB_POISON; | |
b789ef51 CL |
4710 | if (buf[0] == '1') { |
4711 | s->flags &= ~__CMPXCHG_DOUBLE; | |
81819f0f | 4712 | s->flags |= SLAB_POISON; |
b789ef51 | 4713 | } |
06b285dc | 4714 | calculate_sizes(s, -1); |
81819f0f CL |
4715 | return length; |
4716 | } | |
4717 | SLAB_ATTR(poison); | |
4718 | ||
4719 | static ssize_t store_user_show(struct kmem_cache *s, char *buf) | |
4720 | { | |
4721 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_STORE_USER)); | |
4722 | } | |
4723 | ||
4724 | static ssize_t store_user_store(struct kmem_cache *s, | |
4725 | const char *buf, size_t length) | |
4726 | { | |
4727 | if (any_slab_objects(s)) | |
4728 | return -EBUSY; | |
4729 | ||
4730 | s->flags &= ~SLAB_STORE_USER; | |
b789ef51 CL |
4731 | if (buf[0] == '1') { |
4732 | s->flags &= ~__CMPXCHG_DOUBLE; | |
81819f0f | 4733 | s->flags |= SLAB_STORE_USER; |
b789ef51 | 4734 | } |
06b285dc | 4735 | calculate_sizes(s, -1); |
81819f0f CL |
4736 | return length; |
4737 | } | |
4738 | SLAB_ATTR(store_user); | |
4739 | ||
53e15af0 CL |
4740 | static ssize_t validate_show(struct kmem_cache *s, char *buf) |
4741 | { | |
4742 | return 0; | |
4743 | } | |
4744 | ||
4745 | static ssize_t validate_store(struct kmem_cache *s, | |
4746 | const char *buf, size_t length) | |
4747 | { | |
434e245d CL |
4748 | int ret = -EINVAL; |
4749 | ||
4750 | if (buf[0] == '1') { | |
4751 | ret = validate_slab_cache(s); | |
4752 | if (ret >= 0) | |
4753 | ret = length; | |
4754 | } | |
4755 | return ret; | |
53e15af0 CL |
4756 | } |
4757 | SLAB_ATTR(validate); | |
a5a84755 CL |
4758 | |
4759 | static ssize_t alloc_calls_show(struct kmem_cache *s, char *buf) | |
4760 | { | |
4761 | if (!(s->flags & SLAB_STORE_USER)) | |
4762 | return -ENOSYS; | |
4763 | return list_locations(s, buf, TRACK_ALLOC); | |
4764 | } | |
4765 | SLAB_ATTR_RO(alloc_calls); | |
4766 | ||
4767 | static ssize_t free_calls_show(struct kmem_cache *s, char *buf) | |
4768 | { | |
4769 | if (!(s->flags & SLAB_STORE_USER)) | |
4770 | return -ENOSYS; | |
4771 | return list_locations(s, buf, TRACK_FREE); | |
4772 | } | |
4773 | SLAB_ATTR_RO(free_calls); | |
4774 | #endif /* CONFIG_SLUB_DEBUG */ | |
4775 | ||
4776 | #ifdef CONFIG_FAILSLAB | |
4777 | static ssize_t failslab_show(struct kmem_cache *s, char *buf) | |
4778 | { | |
4779 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_FAILSLAB)); | |
4780 | } | |
4781 | ||
4782 | static ssize_t failslab_store(struct kmem_cache *s, const char *buf, | |
4783 | size_t length) | |
4784 | { | |
4785 | s->flags &= ~SLAB_FAILSLAB; | |
4786 | if (buf[0] == '1') | |
4787 | s->flags |= SLAB_FAILSLAB; | |
4788 | return length; | |
4789 | } | |
4790 | SLAB_ATTR(failslab); | |
ab4d5ed5 | 4791 | #endif |
53e15af0 | 4792 | |
2086d26a CL |
4793 | static ssize_t shrink_show(struct kmem_cache *s, char *buf) |
4794 | { | |
4795 | return 0; | |
4796 | } | |
4797 | ||
4798 | static ssize_t shrink_store(struct kmem_cache *s, | |
4799 | const char *buf, size_t length) | |
4800 | { | |
4801 | if (buf[0] == '1') { | |
4802 | int rc = kmem_cache_shrink(s); | |
4803 | ||
4804 | if (rc) | |
4805 | return rc; | |
4806 | } else | |
4807 | return -EINVAL; | |
4808 | return length; | |
4809 | } | |
4810 | SLAB_ATTR(shrink); | |
4811 | ||
81819f0f | 4812 | #ifdef CONFIG_NUMA |
9824601e | 4813 | static ssize_t remote_node_defrag_ratio_show(struct kmem_cache *s, char *buf) |
81819f0f | 4814 | { |
9824601e | 4815 | return sprintf(buf, "%d\n", s->remote_node_defrag_ratio / 10); |
81819f0f CL |
4816 | } |
4817 | ||
9824601e | 4818 | static ssize_t remote_node_defrag_ratio_store(struct kmem_cache *s, |
81819f0f CL |
4819 | const char *buf, size_t length) |
4820 | { | |
0121c619 CL |
4821 | unsigned long ratio; |
4822 | int err; | |
4823 | ||
3dbb95f7 | 4824 | err = kstrtoul(buf, 10, &ratio); |
0121c619 CL |
4825 | if (err) |
4826 | return err; | |
4827 | ||
e2cb96b7 | 4828 | if (ratio <= 100) |
0121c619 | 4829 | s->remote_node_defrag_ratio = ratio * 10; |
81819f0f | 4830 | |
81819f0f CL |
4831 | return length; |
4832 | } | |
9824601e | 4833 | SLAB_ATTR(remote_node_defrag_ratio); |
81819f0f CL |
4834 | #endif |
4835 | ||
8ff12cfc | 4836 | #ifdef CONFIG_SLUB_STATS |
8ff12cfc CL |
4837 | static int show_stat(struct kmem_cache *s, char *buf, enum stat_item si) |
4838 | { | |
4839 | unsigned long sum = 0; | |
4840 | int cpu; | |
4841 | int len; | |
4842 | int *data = kmalloc(nr_cpu_ids * sizeof(int), GFP_KERNEL); | |
4843 | ||
4844 | if (!data) | |
4845 | return -ENOMEM; | |
4846 | ||
4847 | for_each_online_cpu(cpu) { | |
9dfc6e68 | 4848 | unsigned x = per_cpu_ptr(s->cpu_slab, cpu)->stat[si]; |
8ff12cfc CL |
4849 | |
4850 | data[cpu] = x; | |
4851 | sum += x; | |
4852 | } | |
4853 | ||
4854 | len = sprintf(buf, "%lu", sum); | |
4855 | ||
50ef37b9 | 4856 | #ifdef CONFIG_SMP |
8ff12cfc CL |
4857 | for_each_online_cpu(cpu) { |
4858 | if (data[cpu] && len < PAGE_SIZE - 20) | |
50ef37b9 | 4859 | len += sprintf(buf + len, " C%d=%u", cpu, data[cpu]); |
8ff12cfc | 4860 | } |
50ef37b9 | 4861 | #endif |
8ff12cfc CL |
4862 | kfree(data); |
4863 | return len + sprintf(buf + len, "\n"); | |
4864 | } | |
4865 | ||
78eb00cc DR |
4866 | static void clear_stat(struct kmem_cache *s, enum stat_item si) |
4867 | { | |
4868 | int cpu; | |
4869 | ||
4870 | for_each_online_cpu(cpu) | |
9dfc6e68 | 4871 | per_cpu_ptr(s->cpu_slab, cpu)->stat[si] = 0; |
78eb00cc DR |
4872 | } |
4873 | ||
8ff12cfc CL |
4874 | #define STAT_ATTR(si, text) \ |
4875 | static ssize_t text##_show(struct kmem_cache *s, char *buf) \ | |
4876 | { \ | |
4877 | return show_stat(s, buf, si); \ | |
4878 | } \ | |
78eb00cc DR |
4879 | static ssize_t text##_store(struct kmem_cache *s, \ |
4880 | const char *buf, size_t length) \ | |
4881 | { \ | |
4882 | if (buf[0] != '0') \ | |
4883 | return -EINVAL; \ | |
4884 | clear_stat(s, si); \ | |
4885 | return length; \ | |
4886 | } \ | |
4887 | SLAB_ATTR(text); \ | |
8ff12cfc CL |
4888 | |
4889 | STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath); | |
4890 | STAT_ATTR(ALLOC_SLOWPATH, alloc_slowpath); | |
4891 | STAT_ATTR(FREE_FASTPATH, free_fastpath); | |
4892 | STAT_ATTR(FREE_SLOWPATH, free_slowpath); | |
4893 | STAT_ATTR(FREE_FROZEN, free_frozen); | |
4894 | STAT_ATTR(FREE_ADD_PARTIAL, free_add_partial); | |
4895 | STAT_ATTR(FREE_REMOVE_PARTIAL, free_remove_partial); | |
4896 | STAT_ATTR(ALLOC_FROM_PARTIAL, alloc_from_partial); | |
4897 | STAT_ATTR(ALLOC_SLAB, alloc_slab); | |
4898 | STAT_ATTR(ALLOC_REFILL, alloc_refill); | |
e36a2652 | 4899 | STAT_ATTR(ALLOC_NODE_MISMATCH, alloc_node_mismatch); |
8ff12cfc CL |
4900 | STAT_ATTR(FREE_SLAB, free_slab); |
4901 | STAT_ATTR(CPUSLAB_FLUSH, cpuslab_flush); | |
4902 | STAT_ATTR(DEACTIVATE_FULL, deactivate_full); | |
4903 | STAT_ATTR(DEACTIVATE_EMPTY, deactivate_empty); | |
4904 | STAT_ATTR(DEACTIVATE_TO_HEAD, deactivate_to_head); | |
4905 | STAT_ATTR(DEACTIVATE_TO_TAIL, deactivate_to_tail); | |
4906 | STAT_ATTR(DEACTIVATE_REMOTE_FREES, deactivate_remote_frees); | |
03e404af | 4907 | STAT_ATTR(DEACTIVATE_BYPASS, deactivate_bypass); |
65c3376a | 4908 | STAT_ATTR(ORDER_FALLBACK, order_fallback); |
b789ef51 CL |
4909 | STAT_ATTR(CMPXCHG_DOUBLE_CPU_FAIL, cmpxchg_double_cpu_fail); |
4910 | STAT_ATTR(CMPXCHG_DOUBLE_FAIL, cmpxchg_double_fail); | |
49e22585 CL |
4911 | STAT_ATTR(CPU_PARTIAL_ALLOC, cpu_partial_alloc); |
4912 | STAT_ATTR(CPU_PARTIAL_FREE, cpu_partial_free); | |
8028dcea AS |
4913 | STAT_ATTR(CPU_PARTIAL_NODE, cpu_partial_node); |
4914 | STAT_ATTR(CPU_PARTIAL_DRAIN, cpu_partial_drain); | |
8ff12cfc CL |
4915 | #endif |
4916 | ||
06428780 | 4917 | static struct attribute *slab_attrs[] = { |
81819f0f CL |
4918 | &slab_size_attr.attr, |
4919 | &object_size_attr.attr, | |
4920 | &objs_per_slab_attr.attr, | |
4921 | &order_attr.attr, | |
73d342b1 | 4922 | &min_partial_attr.attr, |
49e22585 | 4923 | &cpu_partial_attr.attr, |
81819f0f | 4924 | &objects_attr.attr, |
205ab99d | 4925 | &objects_partial_attr.attr, |
81819f0f CL |
4926 | &partial_attr.attr, |
4927 | &cpu_slabs_attr.attr, | |
4928 | &ctor_attr.attr, | |
81819f0f CL |
4929 | &aliases_attr.attr, |
4930 | &align_attr.attr, | |
81819f0f CL |
4931 | &hwcache_align_attr.attr, |
4932 | &reclaim_account_attr.attr, | |
4933 | &destroy_by_rcu_attr.attr, | |
a5a84755 | 4934 | &shrink_attr.attr, |
ab9a0f19 | 4935 | &reserved_attr.attr, |
49e22585 | 4936 | &slabs_cpu_partial_attr.attr, |
ab4d5ed5 | 4937 | #ifdef CONFIG_SLUB_DEBUG |
a5a84755 CL |
4938 | &total_objects_attr.attr, |
4939 | &slabs_attr.attr, | |
4940 | &sanity_checks_attr.attr, | |
4941 | &trace_attr.attr, | |
81819f0f CL |
4942 | &red_zone_attr.attr, |
4943 | &poison_attr.attr, | |
4944 | &store_user_attr.attr, | |
53e15af0 | 4945 | &validate_attr.attr, |
88a420e4 CL |
4946 | &alloc_calls_attr.attr, |
4947 | &free_calls_attr.attr, | |
ab4d5ed5 | 4948 | #endif |
81819f0f CL |
4949 | #ifdef CONFIG_ZONE_DMA |
4950 | &cache_dma_attr.attr, | |
4951 | #endif | |
4952 | #ifdef CONFIG_NUMA | |
9824601e | 4953 | &remote_node_defrag_ratio_attr.attr, |
8ff12cfc CL |
4954 | #endif |
4955 | #ifdef CONFIG_SLUB_STATS | |
4956 | &alloc_fastpath_attr.attr, | |
4957 | &alloc_slowpath_attr.attr, | |
4958 | &free_fastpath_attr.attr, | |
4959 | &free_slowpath_attr.attr, | |
4960 | &free_frozen_attr.attr, | |
4961 | &free_add_partial_attr.attr, | |
4962 | &free_remove_partial_attr.attr, | |
4963 | &alloc_from_partial_attr.attr, | |
4964 | &alloc_slab_attr.attr, | |
4965 | &alloc_refill_attr.attr, | |
e36a2652 | 4966 | &alloc_node_mismatch_attr.attr, |
8ff12cfc CL |
4967 | &free_slab_attr.attr, |
4968 | &cpuslab_flush_attr.attr, | |
4969 | &deactivate_full_attr.attr, | |
4970 | &deactivate_empty_attr.attr, | |
4971 | &deactivate_to_head_attr.attr, | |
4972 | &deactivate_to_tail_attr.attr, | |
4973 | &deactivate_remote_frees_attr.attr, | |
03e404af | 4974 | &deactivate_bypass_attr.attr, |
65c3376a | 4975 | &order_fallback_attr.attr, |
b789ef51 CL |
4976 | &cmpxchg_double_fail_attr.attr, |
4977 | &cmpxchg_double_cpu_fail_attr.attr, | |
49e22585 CL |
4978 | &cpu_partial_alloc_attr.attr, |
4979 | &cpu_partial_free_attr.attr, | |
8028dcea AS |
4980 | &cpu_partial_node_attr.attr, |
4981 | &cpu_partial_drain_attr.attr, | |
81819f0f | 4982 | #endif |
4c13dd3b DM |
4983 | #ifdef CONFIG_FAILSLAB |
4984 | &failslab_attr.attr, | |
4985 | #endif | |
4986 | ||
81819f0f CL |
4987 | NULL |
4988 | }; | |
4989 | ||
4990 | static struct attribute_group slab_attr_group = { | |
4991 | .attrs = slab_attrs, | |
4992 | }; | |
4993 | ||
4994 | static ssize_t slab_attr_show(struct kobject *kobj, | |
4995 | struct attribute *attr, | |
4996 | char *buf) | |
4997 | { | |
4998 | struct slab_attribute *attribute; | |
4999 | struct kmem_cache *s; | |
5000 | int err; | |
5001 | ||
5002 | attribute = to_slab_attr(attr); | |
5003 | s = to_slab(kobj); | |
5004 | ||
5005 | if (!attribute->show) | |
5006 | return -EIO; | |
5007 | ||
5008 | err = attribute->show(s, buf); | |
5009 | ||
5010 | return err; | |
5011 | } | |
5012 | ||
5013 | static ssize_t slab_attr_store(struct kobject *kobj, | |
5014 | struct attribute *attr, | |
5015 | const char *buf, size_t len) | |
5016 | { | |
5017 | struct slab_attribute *attribute; | |
5018 | struct kmem_cache *s; | |
5019 | int err; | |
5020 | ||
5021 | attribute = to_slab_attr(attr); | |
5022 | s = to_slab(kobj); | |
5023 | ||
5024 | if (!attribute->store) | |
5025 | return -EIO; | |
5026 | ||
5027 | err = attribute->store(s, buf, len); | |
107dab5c GC |
5028 | #ifdef CONFIG_MEMCG_KMEM |
5029 | if (slab_state >= FULL && err >= 0 && is_root_cache(s)) { | |
5030 | int i; | |
81819f0f | 5031 | |
107dab5c GC |
5032 | mutex_lock(&slab_mutex); |
5033 | if (s->max_attr_size < len) | |
5034 | s->max_attr_size = len; | |
5035 | ||
ebe945c2 GC |
5036 | /* |
5037 | * This is a best effort propagation, so this function's return | |
5038 | * value will be determined by the parent cache only. This is | |
5039 | * basically because not all attributes will have a well | |
5040 | * defined semantics for rollbacks - most of the actions will | |
5041 | * have permanent effects. | |
5042 | * | |
5043 | * Returning the error value of any of the children that fail | |
5044 | * is not 100 % defined, in the sense that users seeing the | |
5045 | * error code won't be able to know anything about the state of | |
5046 | * the cache. | |
5047 | * | |
5048 | * Only returning the error code for the parent cache at least | |
5049 | * has well defined semantics. The cache being written to | |
5050 | * directly either failed or succeeded, in which case we loop | |
5051 | * through the descendants with best-effort propagation. | |
5052 | */ | |
107dab5c | 5053 | for_each_memcg_cache_index(i) { |
2ade4de8 | 5054 | struct kmem_cache *c = cache_from_memcg_idx(s, i); |
107dab5c GC |
5055 | if (c) |
5056 | attribute->store(c, buf, len); | |
5057 | } | |
5058 | mutex_unlock(&slab_mutex); | |
5059 | } | |
5060 | #endif | |
81819f0f CL |
5061 | return err; |
5062 | } | |
5063 | ||
107dab5c GC |
5064 | static void memcg_propagate_slab_attrs(struct kmem_cache *s) |
5065 | { | |
5066 | #ifdef CONFIG_MEMCG_KMEM | |
5067 | int i; | |
5068 | char *buffer = NULL; | |
5069 | ||
5070 | if (!is_root_cache(s)) | |
5071 | return; | |
5072 | ||
5073 | /* | |
5074 | * This mean this cache had no attribute written. Therefore, no point | |
5075 | * in copying default values around | |
5076 | */ | |
5077 | if (!s->max_attr_size) | |
5078 | return; | |
5079 | ||
5080 | for (i = 0; i < ARRAY_SIZE(slab_attrs); i++) { | |
5081 | char mbuf[64]; | |
5082 | char *buf; | |
5083 | struct slab_attribute *attr = to_slab_attr(slab_attrs[i]); | |
5084 | ||
5085 | if (!attr || !attr->store || !attr->show) | |
5086 | continue; | |
5087 | ||
5088 | /* | |
5089 | * It is really bad that we have to allocate here, so we will | |
5090 | * do it only as a fallback. If we actually allocate, though, | |
5091 | * we can just use the allocated buffer until the end. | |
5092 | * | |
5093 | * Most of the slub attributes will tend to be very small in | |
5094 | * size, but sysfs allows buffers up to a page, so they can | |
5095 | * theoretically happen. | |
5096 | */ | |
5097 | if (buffer) | |
5098 | buf = buffer; | |
5099 | else if (s->max_attr_size < ARRAY_SIZE(mbuf)) | |
5100 | buf = mbuf; | |
5101 | else { | |
5102 | buffer = (char *) get_zeroed_page(GFP_KERNEL); | |
5103 | if (WARN_ON(!buffer)) | |
5104 | continue; | |
5105 | buf = buffer; | |
5106 | } | |
5107 | ||
5108 | attr->show(s->memcg_params->root_cache, buf); | |
5109 | attr->store(s, buf, strlen(buf)); | |
5110 | } | |
5111 | ||
5112 | if (buffer) | |
5113 | free_page((unsigned long)buffer); | |
5114 | #endif | |
5115 | } | |
5116 | ||
52cf25d0 | 5117 | static const struct sysfs_ops slab_sysfs_ops = { |
81819f0f CL |
5118 | .show = slab_attr_show, |
5119 | .store = slab_attr_store, | |
5120 | }; | |
5121 | ||
5122 | static struct kobj_type slab_ktype = { | |
5123 | .sysfs_ops = &slab_sysfs_ops, | |
5124 | }; | |
5125 | ||
5126 | static int uevent_filter(struct kset *kset, struct kobject *kobj) | |
5127 | { | |
5128 | struct kobj_type *ktype = get_ktype(kobj); | |
5129 | ||
5130 | if (ktype == &slab_ktype) | |
5131 | return 1; | |
5132 | return 0; | |
5133 | } | |
5134 | ||
9cd43611 | 5135 | static const struct kset_uevent_ops slab_uevent_ops = { |
81819f0f CL |
5136 | .filter = uevent_filter, |
5137 | }; | |
5138 | ||
27c3a314 | 5139 | static struct kset *slab_kset; |
81819f0f | 5140 | |
9a41707b VD |
5141 | static inline struct kset *cache_kset(struct kmem_cache *s) |
5142 | { | |
5143 | #ifdef CONFIG_MEMCG_KMEM | |
5144 | if (!is_root_cache(s)) | |
5145 | return s->memcg_params->root_cache->memcg_kset; | |
5146 | #endif | |
5147 | return slab_kset; | |
5148 | } | |
5149 | ||
81819f0f CL |
5150 | #define ID_STR_LENGTH 64 |
5151 | ||
5152 | /* Create a unique string id for a slab cache: | |
6446faa2 CL |
5153 | * |
5154 | * Format :[flags-]size | |
81819f0f CL |
5155 | */ |
5156 | static char *create_unique_id(struct kmem_cache *s) | |
5157 | { | |
5158 | char *name = kmalloc(ID_STR_LENGTH, GFP_KERNEL); | |
5159 | char *p = name; | |
5160 | ||
5161 | BUG_ON(!name); | |
5162 | ||
5163 | *p++ = ':'; | |
5164 | /* | |
5165 | * First flags affecting slabcache operations. We will only | |
5166 | * get here for aliasable slabs so we do not need to support | |
5167 | * too many flags. The flags here must cover all flags that | |
5168 | * are matched during merging to guarantee that the id is | |
5169 | * unique. | |
5170 | */ | |
5171 | if (s->flags & SLAB_CACHE_DMA) | |
5172 | *p++ = 'd'; | |
5173 | if (s->flags & SLAB_RECLAIM_ACCOUNT) | |
5174 | *p++ = 'a'; | |
5175 | if (s->flags & SLAB_DEBUG_FREE) | |
5176 | *p++ = 'F'; | |
5a896d9e VN |
5177 | if (!(s->flags & SLAB_NOTRACK)) |
5178 | *p++ = 't'; | |
81819f0f CL |
5179 | if (p != name + 1) |
5180 | *p++ = '-'; | |
5181 | p += sprintf(p, "%07d", s->size); | |
2633d7a0 GC |
5182 | |
5183 | #ifdef CONFIG_MEMCG_KMEM | |
5184 | if (!is_root_cache(s)) | |
d0e0ac97 CG |
5185 | p += sprintf(p, "-%08d", |
5186 | memcg_cache_id(s->memcg_params->memcg)); | |
2633d7a0 GC |
5187 | #endif |
5188 | ||
81819f0f CL |
5189 | BUG_ON(p > name + ID_STR_LENGTH - 1); |
5190 | return name; | |
5191 | } | |
5192 | ||
5193 | static int sysfs_slab_add(struct kmem_cache *s) | |
5194 | { | |
5195 | int err; | |
5196 | const char *name; | |
45530c44 | 5197 | int unmergeable = slab_unmergeable(s); |
81819f0f | 5198 | |
81819f0f CL |
5199 | if (unmergeable) { |
5200 | /* | |
5201 | * Slabcache can never be merged so we can use the name proper. | |
5202 | * This is typically the case for debug situations. In that | |
5203 | * case we can catch duplicate names easily. | |
5204 | */ | |
27c3a314 | 5205 | sysfs_remove_link(&slab_kset->kobj, s->name); |
81819f0f CL |
5206 | name = s->name; |
5207 | } else { | |
5208 | /* | |
5209 | * Create a unique name for the slab as a target | |
5210 | * for the symlinks. | |
5211 | */ | |
5212 | name = create_unique_id(s); | |
5213 | } | |
5214 | ||
9a41707b | 5215 | s->kobj.kset = cache_kset(s); |
26e4f205 | 5216 | err = kobject_init_and_add(&s->kobj, &slab_ktype, NULL, "%s", name); |
1eada11c GKH |
5217 | if (err) { |
5218 | kobject_put(&s->kobj); | |
81819f0f | 5219 | return err; |
1eada11c | 5220 | } |
81819f0f CL |
5221 | |
5222 | err = sysfs_create_group(&s->kobj, &slab_attr_group); | |
5788d8ad XF |
5223 | if (err) { |
5224 | kobject_del(&s->kobj); | |
5225 | kobject_put(&s->kobj); | |
81819f0f | 5226 | return err; |
5788d8ad | 5227 | } |
9a41707b VD |
5228 | |
5229 | #ifdef CONFIG_MEMCG_KMEM | |
5230 | if (is_root_cache(s)) { | |
5231 | s->memcg_kset = kset_create_and_add("cgroup", NULL, &s->kobj); | |
5232 | if (!s->memcg_kset) { | |
5233 | kobject_del(&s->kobj); | |
5234 | kobject_put(&s->kobj); | |
5235 | return -ENOMEM; | |
5236 | } | |
5237 | } | |
5238 | #endif | |
5239 | ||
81819f0f CL |
5240 | kobject_uevent(&s->kobj, KOBJ_ADD); |
5241 | if (!unmergeable) { | |
5242 | /* Setup first alias */ | |
5243 | sysfs_slab_alias(s, s->name); | |
5244 | kfree(name); | |
5245 | } | |
5246 | return 0; | |
5247 | } | |
5248 | ||
5249 | static void sysfs_slab_remove(struct kmem_cache *s) | |
5250 | { | |
97d06609 | 5251 | if (slab_state < FULL) |
2bce6485 CL |
5252 | /* |
5253 | * Sysfs has not been setup yet so no need to remove the | |
5254 | * cache from sysfs. | |
5255 | */ | |
5256 | return; | |
5257 | ||
9a41707b VD |
5258 | #ifdef CONFIG_MEMCG_KMEM |
5259 | kset_unregister(s->memcg_kset); | |
5260 | #endif | |
81819f0f CL |
5261 | kobject_uevent(&s->kobj, KOBJ_REMOVE); |
5262 | kobject_del(&s->kobj); | |
151c602f | 5263 | kobject_put(&s->kobj); |
81819f0f CL |
5264 | } |
5265 | ||
5266 | /* | |
5267 | * Need to buffer aliases during bootup until sysfs becomes | |
9f6c708e | 5268 | * available lest we lose that information. |
81819f0f CL |
5269 | */ |
5270 | struct saved_alias { | |
5271 | struct kmem_cache *s; | |
5272 | const char *name; | |
5273 | struct saved_alias *next; | |
5274 | }; | |
5275 | ||
5af328a5 | 5276 | static struct saved_alias *alias_list; |
81819f0f CL |
5277 | |
5278 | static int sysfs_slab_alias(struct kmem_cache *s, const char *name) | |
5279 | { | |
5280 | struct saved_alias *al; | |
5281 | ||
97d06609 | 5282 | if (slab_state == FULL) { |
81819f0f CL |
5283 | /* |
5284 | * If we have a leftover link then remove it. | |
5285 | */ | |
27c3a314 GKH |
5286 | sysfs_remove_link(&slab_kset->kobj, name); |
5287 | return sysfs_create_link(&slab_kset->kobj, &s->kobj, name); | |
81819f0f CL |
5288 | } |
5289 | ||
5290 | al = kmalloc(sizeof(struct saved_alias), GFP_KERNEL); | |
5291 | if (!al) | |
5292 | return -ENOMEM; | |
5293 | ||
5294 | al->s = s; | |
5295 | al->name = name; | |
5296 | al->next = alias_list; | |
5297 | alias_list = al; | |
5298 | return 0; | |
5299 | } | |
5300 | ||
5301 | static int __init slab_sysfs_init(void) | |
5302 | { | |
5b95a4ac | 5303 | struct kmem_cache *s; |
81819f0f CL |
5304 | int err; |
5305 | ||
18004c5d | 5306 | mutex_lock(&slab_mutex); |
2bce6485 | 5307 | |
0ff21e46 | 5308 | slab_kset = kset_create_and_add("slab", &slab_uevent_ops, kernel_kobj); |
27c3a314 | 5309 | if (!slab_kset) { |
18004c5d | 5310 | mutex_unlock(&slab_mutex); |
81819f0f CL |
5311 | printk(KERN_ERR "Cannot register slab subsystem.\n"); |
5312 | return -ENOSYS; | |
5313 | } | |
5314 | ||
97d06609 | 5315 | slab_state = FULL; |
26a7bd03 | 5316 | |
5b95a4ac | 5317 | list_for_each_entry(s, &slab_caches, list) { |
26a7bd03 | 5318 | err = sysfs_slab_add(s); |
5d540fb7 CL |
5319 | if (err) |
5320 | printk(KERN_ERR "SLUB: Unable to add boot slab %s" | |
5321 | " to sysfs\n", s->name); | |
26a7bd03 | 5322 | } |
81819f0f CL |
5323 | |
5324 | while (alias_list) { | |
5325 | struct saved_alias *al = alias_list; | |
5326 | ||
5327 | alias_list = alias_list->next; | |
5328 | err = sysfs_slab_alias(al->s, al->name); | |
5d540fb7 CL |
5329 | if (err) |
5330 | printk(KERN_ERR "SLUB: Unable to add boot slab alias" | |
068ce415 | 5331 | " %s to sysfs\n", al->name); |
81819f0f CL |
5332 | kfree(al); |
5333 | } | |
5334 | ||
18004c5d | 5335 | mutex_unlock(&slab_mutex); |
81819f0f CL |
5336 | resiliency_test(); |
5337 | return 0; | |
5338 | } | |
5339 | ||
5340 | __initcall(slab_sysfs_init); | |
ab4d5ed5 | 5341 | #endif /* CONFIG_SYSFS */ |
57ed3eda PE |
5342 | |
5343 | /* | |
5344 | * The /proc/slabinfo ABI | |
5345 | */ | |
158a9624 | 5346 | #ifdef CONFIG_SLABINFO |
0d7561c6 | 5347 | void get_slabinfo(struct kmem_cache *s, struct slabinfo *sinfo) |
57ed3eda | 5348 | { |
57ed3eda | 5349 | unsigned long nr_slabs = 0; |
205ab99d CL |
5350 | unsigned long nr_objs = 0; |
5351 | unsigned long nr_free = 0; | |
57ed3eda PE |
5352 | int node; |
5353 | ||
57ed3eda PE |
5354 | for_each_online_node(node) { |
5355 | struct kmem_cache_node *n = get_node(s, node); | |
5356 | ||
5357 | if (!n) | |
5358 | continue; | |
5359 | ||
c17fd13e WL |
5360 | nr_slabs += node_nr_slabs(n); |
5361 | nr_objs += node_nr_objs(n); | |
205ab99d | 5362 | nr_free += count_partial(n, count_free); |
57ed3eda PE |
5363 | } |
5364 | ||
0d7561c6 GC |
5365 | sinfo->active_objs = nr_objs - nr_free; |
5366 | sinfo->num_objs = nr_objs; | |
5367 | sinfo->active_slabs = nr_slabs; | |
5368 | sinfo->num_slabs = nr_slabs; | |
5369 | sinfo->objects_per_slab = oo_objects(s->oo); | |
5370 | sinfo->cache_order = oo_order(s->oo); | |
57ed3eda PE |
5371 | } |
5372 | ||
0d7561c6 | 5373 | void slabinfo_show_stats(struct seq_file *m, struct kmem_cache *s) |
7b3c3a50 | 5374 | { |
7b3c3a50 AD |
5375 | } |
5376 | ||
b7454ad3 GC |
5377 | ssize_t slabinfo_write(struct file *file, const char __user *buffer, |
5378 | size_t count, loff_t *ppos) | |
7b3c3a50 | 5379 | { |
b7454ad3 | 5380 | return -EIO; |
7b3c3a50 | 5381 | } |
158a9624 | 5382 | #endif /* CONFIG_SLABINFO */ |