Commit | Line | Data |
---|---|---|
81819f0f CL |
1 | /* |
2 | * SLUB: A slab allocator that limits cache line use instead of queuing | |
3 | * objects in per cpu and per node lists. | |
4 | * | |
881db7fb CL |
5 | * The allocator synchronizes using per slab locks or atomic operatios |
6 | * and only uses a centralized lock to manage a pool of partial slabs. | |
81819f0f | 7 | * |
cde53535 | 8 | * (C) 2007 SGI, Christoph Lameter |
881db7fb | 9 | * (C) 2011 Linux Foundation, Christoph Lameter |
81819f0f CL |
10 | */ |
11 | ||
12 | #include <linux/mm.h> | |
1eb5ac64 | 13 | #include <linux/swap.h> /* struct reclaim_state */ |
81819f0f CL |
14 | #include <linux/module.h> |
15 | #include <linux/bit_spinlock.h> | |
16 | #include <linux/interrupt.h> | |
17 | #include <linux/bitops.h> | |
18 | #include <linux/slab.h> | |
97d06609 | 19 | #include "slab.h" |
7b3c3a50 | 20 | #include <linux/proc_fs.h> |
3ac38faa | 21 | #include <linux/notifier.h> |
81819f0f | 22 | #include <linux/seq_file.h> |
a79316c6 | 23 | #include <linux/kasan.h> |
5a896d9e | 24 | #include <linux/kmemcheck.h> |
81819f0f CL |
25 | #include <linux/cpu.h> |
26 | #include <linux/cpuset.h> | |
27 | #include <linux/mempolicy.h> | |
28 | #include <linux/ctype.h> | |
3ac7fe5a | 29 | #include <linux/debugobjects.h> |
81819f0f | 30 | #include <linux/kallsyms.h> |
b9049e23 | 31 | #include <linux/memory.h> |
f8bd2258 | 32 | #include <linux/math64.h> |
773ff60e | 33 | #include <linux/fault-inject.h> |
bfa71457 | 34 | #include <linux/stacktrace.h> |
4de900b4 | 35 | #include <linux/prefetch.h> |
2633d7a0 | 36 | #include <linux/memcontrol.h> |
81819f0f | 37 | |
4a92379b RK |
38 | #include <trace/events/kmem.h> |
39 | ||
072bb0aa MG |
40 | #include "internal.h" |
41 | ||
81819f0f CL |
42 | /* |
43 | * Lock order: | |
18004c5d | 44 | * 1. slab_mutex (Global Mutex) |
881db7fb CL |
45 | * 2. node->list_lock |
46 | * 3. slab_lock(page) (Only on some arches and for debugging) | |
81819f0f | 47 | * |
18004c5d | 48 | * slab_mutex |
881db7fb | 49 | * |
18004c5d | 50 | * The role of the slab_mutex is to protect the list of all the slabs |
881db7fb CL |
51 | * and to synchronize major metadata changes to slab cache structures. |
52 | * | |
53 | * The slab_lock is only used for debugging and on arches that do not | |
54 | * have the ability to do a cmpxchg_double. It only protects the second | |
55 | * double word in the page struct. Meaning | |
56 | * A. page->freelist -> List of object free in a page | |
57 | * B. page->counters -> Counters of objects | |
58 | * C. page->frozen -> frozen state | |
59 | * | |
60 | * If a slab is frozen then it is exempt from list management. It is not | |
61 | * on any list. The processor that froze the slab is the one who can | |
62 | * perform list operations on the page. Other processors may put objects | |
63 | * onto the freelist but the processor that froze the slab is the only | |
64 | * one that can retrieve the objects from the page's freelist. | |
81819f0f CL |
65 | * |
66 | * The list_lock protects the partial and full list on each node and | |
67 | * the partial slab counter. If taken then no new slabs may be added or | |
68 | * removed from the lists nor make the number of partial slabs be modified. | |
69 | * (Note that the total number of slabs is an atomic value that may be | |
70 | * modified without taking the list lock). | |
71 | * | |
72 | * The list_lock is a centralized lock and thus we avoid taking it as | |
73 | * much as possible. As long as SLUB does not have to handle partial | |
74 | * slabs, operations can continue without any centralized lock. F.e. | |
75 | * allocating a long series of objects that fill up slabs does not require | |
76 | * the list lock. | |
81819f0f CL |
77 | * Interrupts are disabled during allocation and deallocation in order to |
78 | * make the slab allocator safe to use in the context of an irq. In addition | |
79 | * interrupts are disabled to ensure that the processor does not change | |
80 | * while handling per_cpu slabs, due to kernel preemption. | |
81 | * | |
82 | * SLUB assigns one slab for allocation to each processor. | |
83 | * Allocations only occur from these slabs called cpu slabs. | |
84 | * | |
672bba3a CL |
85 | * Slabs with free elements are kept on a partial list and during regular |
86 | * operations no list for full slabs is used. If an object in a full slab is | |
81819f0f | 87 | * freed then the slab will show up again on the partial lists. |
672bba3a CL |
88 | * We track full slabs for debugging purposes though because otherwise we |
89 | * cannot scan all objects. | |
81819f0f CL |
90 | * |
91 | * Slabs are freed when they become empty. Teardown and setup is | |
92 | * minimal so we rely on the page allocators per cpu caches for | |
93 | * fast frees and allocs. | |
94 | * | |
95 | * Overloading of page flags that are otherwise used for LRU management. | |
96 | * | |
4b6f0750 CL |
97 | * PageActive The slab is frozen and exempt from list processing. |
98 | * This means that the slab is dedicated to a purpose | |
99 | * such as satisfying allocations for a specific | |
100 | * processor. Objects may be freed in the slab while | |
101 | * it is frozen but slab_free will then skip the usual | |
102 | * list operations. It is up to the processor holding | |
103 | * the slab to integrate the slab into the slab lists | |
104 | * when the slab is no longer needed. | |
105 | * | |
106 | * One use of this flag is to mark slabs that are | |
107 | * used for allocations. Then such a slab becomes a cpu | |
108 | * slab. The cpu slab may be equipped with an additional | |
dfb4f096 | 109 | * freelist that allows lockless access to |
894b8788 CL |
110 | * free objects in addition to the regular freelist |
111 | * that requires the slab lock. | |
81819f0f CL |
112 | * |
113 | * PageError Slab requires special handling due to debug | |
114 | * options set. This moves slab handling out of | |
894b8788 | 115 | * the fast path and disables lockless freelists. |
81819f0f CL |
116 | */ |
117 | ||
af537b0a CL |
118 | static inline int kmem_cache_debug(struct kmem_cache *s) |
119 | { | |
5577bd8a | 120 | #ifdef CONFIG_SLUB_DEBUG |
af537b0a | 121 | return unlikely(s->flags & SLAB_DEBUG_FLAGS); |
5577bd8a | 122 | #else |
af537b0a | 123 | return 0; |
5577bd8a | 124 | #endif |
af537b0a | 125 | } |
5577bd8a | 126 | |
345c905d JK |
127 | static inline bool kmem_cache_has_cpu_partial(struct kmem_cache *s) |
128 | { | |
129 | #ifdef CONFIG_SLUB_CPU_PARTIAL | |
130 | return !kmem_cache_debug(s); | |
131 | #else | |
132 | return false; | |
133 | #endif | |
134 | } | |
135 | ||
81819f0f CL |
136 | /* |
137 | * Issues still to be resolved: | |
138 | * | |
81819f0f CL |
139 | * - Support PAGE_ALLOC_DEBUG. Should be easy to do. |
140 | * | |
81819f0f CL |
141 | * - Variable sizing of the per node arrays |
142 | */ | |
143 | ||
144 | /* Enable to test recovery from slab corruption on boot */ | |
145 | #undef SLUB_RESILIENCY_TEST | |
146 | ||
b789ef51 CL |
147 | /* Enable to log cmpxchg failures */ |
148 | #undef SLUB_DEBUG_CMPXCHG | |
149 | ||
2086d26a CL |
150 | /* |
151 | * Mininum number of partial slabs. These will be left on the partial | |
152 | * lists even if they are empty. kmem_cache_shrink may reclaim them. | |
153 | */ | |
76be8950 | 154 | #define MIN_PARTIAL 5 |
e95eed57 | 155 | |
2086d26a CL |
156 | /* |
157 | * Maximum number of desirable partial slabs. | |
158 | * The existence of more partial slabs makes kmem_cache_shrink | |
721ae22a | 159 | * sort the partial list by the number of objects in use. |
2086d26a CL |
160 | */ |
161 | #define MAX_PARTIAL 10 | |
162 | ||
81819f0f CL |
163 | #define DEBUG_DEFAULT_FLAGS (SLAB_DEBUG_FREE | SLAB_RED_ZONE | \ |
164 | SLAB_POISON | SLAB_STORE_USER) | |
672bba3a | 165 | |
fa5ec8a1 | 166 | /* |
3de47213 DR |
167 | * Debugging flags that require metadata to be stored in the slab. These get |
168 | * disabled when slub_debug=O is used and a cache's min order increases with | |
169 | * metadata. | |
fa5ec8a1 | 170 | */ |
3de47213 | 171 | #define DEBUG_METADATA_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER) |
fa5ec8a1 | 172 | |
210b5c06 CG |
173 | #define OO_SHIFT 16 |
174 | #define OO_MASK ((1 << OO_SHIFT) - 1) | |
50d5c41c | 175 | #define MAX_OBJS_PER_PAGE 32767 /* since page.objects is u15 */ |
210b5c06 | 176 | |
81819f0f | 177 | /* Internal SLUB flags */ |
f90ec390 | 178 | #define __OBJECT_POISON 0x80000000UL /* Poison object */ |
b789ef51 | 179 | #define __CMPXCHG_DOUBLE 0x40000000UL /* Use cmpxchg_double */ |
81819f0f | 180 | |
81819f0f CL |
181 | #ifdef CONFIG_SMP |
182 | static struct notifier_block slab_notifier; | |
183 | #endif | |
184 | ||
02cbc874 CL |
185 | /* |
186 | * Tracking user of a slab. | |
187 | */ | |
d6543e39 | 188 | #define TRACK_ADDRS_COUNT 16 |
02cbc874 | 189 | struct track { |
ce71e27c | 190 | unsigned long addr; /* Called from address */ |
d6543e39 BG |
191 | #ifdef CONFIG_STACKTRACE |
192 | unsigned long addrs[TRACK_ADDRS_COUNT]; /* Called from address */ | |
193 | #endif | |
02cbc874 CL |
194 | int cpu; /* Was running on cpu */ |
195 | int pid; /* Pid context */ | |
196 | unsigned long when; /* When did the operation occur */ | |
197 | }; | |
198 | ||
199 | enum track_item { TRACK_ALLOC, TRACK_FREE }; | |
200 | ||
ab4d5ed5 | 201 | #ifdef CONFIG_SYSFS |
81819f0f CL |
202 | static int sysfs_slab_add(struct kmem_cache *); |
203 | static int sysfs_slab_alias(struct kmem_cache *, const char *); | |
107dab5c | 204 | static void memcg_propagate_slab_attrs(struct kmem_cache *s); |
81819f0f | 205 | #else |
0c710013 CL |
206 | static inline int sysfs_slab_add(struct kmem_cache *s) { return 0; } |
207 | static inline int sysfs_slab_alias(struct kmem_cache *s, const char *p) | |
208 | { return 0; } | |
107dab5c | 209 | static inline void memcg_propagate_slab_attrs(struct kmem_cache *s) { } |
81819f0f CL |
210 | #endif |
211 | ||
4fdccdfb | 212 | static inline void stat(const struct kmem_cache *s, enum stat_item si) |
8ff12cfc CL |
213 | { |
214 | #ifdef CONFIG_SLUB_STATS | |
88da03a6 CL |
215 | /* |
216 | * The rmw is racy on a preemptible kernel but this is acceptable, so | |
217 | * avoid this_cpu_add()'s irq-disable overhead. | |
218 | */ | |
219 | raw_cpu_inc(s->cpu_slab->stat[si]); | |
8ff12cfc CL |
220 | #endif |
221 | } | |
222 | ||
81819f0f CL |
223 | /******************************************************************** |
224 | * Core slab cache functions | |
225 | *******************************************************************/ | |
226 | ||
6446faa2 | 227 | /* Verify that a pointer has an address that is valid within a slab page */ |
02cbc874 CL |
228 | static inline int check_valid_pointer(struct kmem_cache *s, |
229 | struct page *page, const void *object) | |
230 | { | |
231 | void *base; | |
232 | ||
a973e9dd | 233 | if (!object) |
02cbc874 CL |
234 | return 1; |
235 | ||
a973e9dd | 236 | base = page_address(page); |
39b26464 | 237 | if (object < base || object >= base + page->objects * s->size || |
02cbc874 CL |
238 | (object - base) % s->size) { |
239 | return 0; | |
240 | } | |
241 | ||
242 | return 1; | |
243 | } | |
244 | ||
7656c72b CL |
245 | static inline void *get_freepointer(struct kmem_cache *s, void *object) |
246 | { | |
247 | return *(void **)(object + s->offset); | |
248 | } | |
249 | ||
0ad9500e ED |
250 | static void prefetch_freepointer(const struct kmem_cache *s, void *object) |
251 | { | |
252 | prefetch(object + s->offset); | |
253 | } | |
254 | ||
1393d9a1 CL |
255 | static inline void *get_freepointer_safe(struct kmem_cache *s, void *object) |
256 | { | |
257 | void *p; | |
258 | ||
259 | #ifdef CONFIG_DEBUG_PAGEALLOC | |
260 | probe_kernel_read(&p, (void **)(object + s->offset), sizeof(p)); | |
261 | #else | |
262 | p = get_freepointer(s, object); | |
263 | #endif | |
264 | return p; | |
265 | } | |
266 | ||
7656c72b CL |
267 | static inline void set_freepointer(struct kmem_cache *s, void *object, void *fp) |
268 | { | |
269 | *(void **)(object + s->offset) = fp; | |
270 | } | |
271 | ||
272 | /* Loop over all objects in a slab */ | |
224a88be CL |
273 | #define for_each_object(__p, __s, __addr, __objects) \ |
274 | for (__p = (__addr); __p < (__addr) + (__objects) * (__s)->size;\ | |
7656c72b CL |
275 | __p += (__s)->size) |
276 | ||
54266640 WY |
277 | #define for_each_object_idx(__p, __idx, __s, __addr, __objects) \ |
278 | for (__p = (__addr), __idx = 1; __idx <= __objects;\ | |
279 | __p += (__s)->size, __idx++) | |
280 | ||
7656c72b CL |
281 | /* Determine object index from a given position */ |
282 | static inline int slab_index(void *p, struct kmem_cache *s, void *addr) | |
283 | { | |
284 | return (p - addr) / s->size; | |
285 | } | |
286 | ||
d71f606f MK |
287 | static inline size_t slab_ksize(const struct kmem_cache *s) |
288 | { | |
289 | #ifdef CONFIG_SLUB_DEBUG | |
290 | /* | |
291 | * Debugging requires use of the padding between object | |
292 | * and whatever may come after it. | |
293 | */ | |
294 | if (s->flags & (SLAB_RED_ZONE | SLAB_POISON)) | |
3b0efdfa | 295 | return s->object_size; |
d71f606f MK |
296 | |
297 | #endif | |
298 | /* | |
299 | * If we have the need to store the freelist pointer | |
300 | * back there or track user information then we can | |
301 | * only use the space before that information. | |
302 | */ | |
303 | if (s->flags & (SLAB_DESTROY_BY_RCU | SLAB_STORE_USER)) | |
304 | return s->inuse; | |
305 | /* | |
306 | * Else we can use all the padding etc for the allocation | |
307 | */ | |
308 | return s->size; | |
309 | } | |
310 | ||
ab9a0f19 LJ |
311 | static inline int order_objects(int order, unsigned long size, int reserved) |
312 | { | |
313 | return ((PAGE_SIZE << order) - reserved) / size; | |
314 | } | |
315 | ||
834f3d11 | 316 | static inline struct kmem_cache_order_objects oo_make(int order, |
ab9a0f19 | 317 | unsigned long size, int reserved) |
834f3d11 CL |
318 | { |
319 | struct kmem_cache_order_objects x = { | |
ab9a0f19 | 320 | (order << OO_SHIFT) + order_objects(order, size, reserved) |
834f3d11 CL |
321 | }; |
322 | ||
323 | return x; | |
324 | } | |
325 | ||
326 | static inline int oo_order(struct kmem_cache_order_objects x) | |
327 | { | |
210b5c06 | 328 | return x.x >> OO_SHIFT; |
834f3d11 CL |
329 | } |
330 | ||
331 | static inline int oo_objects(struct kmem_cache_order_objects x) | |
332 | { | |
210b5c06 | 333 | return x.x & OO_MASK; |
834f3d11 CL |
334 | } |
335 | ||
881db7fb CL |
336 | /* |
337 | * Per slab locking using the pagelock | |
338 | */ | |
339 | static __always_inline void slab_lock(struct page *page) | |
340 | { | |
48c935ad | 341 | VM_BUG_ON_PAGE(PageTail(page), page); |
881db7fb CL |
342 | bit_spin_lock(PG_locked, &page->flags); |
343 | } | |
344 | ||
345 | static __always_inline void slab_unlock(struct page *page) | |
346 | { | |
48c935ad | 347 | VM_BUG_ON_PAGE(PageTail(page), page); |
881db7fb CL |
348 | __bit_spin_unlock(PG_locked, &page->flags); |
349 | } | |
350 | ||
a0320865 DH |
351 | static inline void set_page_slub_counters(struct page *page, unsigned long counters_new) |
352 | { | |
353 | struct page tmp; | |
354 | tmp.counters = counters_new; | |
355 | /* | |
356 | * page->counters can cover frozen/inuse/objects as well | |
357 | * as page->_count. If we assign to ->counters directly | |
358 | * we run the risk of losing updates to page->_count, so | |
359 | * be careful and only assign to the fields we need. | |
360 | */ | |
361 | page->frozen = tmp.frozen; | |
362 | page->inuse = tmp.inuse; | |
363 | page->objects = tmp.objects; | |
364 | } | |
365 | ||
1d07171c CL |
366 | /* Interrupts must be disabled (for the fallback code to work right) */ |
367 | static inline bool __cmpxchg_double_slab(struct kmem_cache *s, struct page *page, | |
368 | void *freelist_old, unsigned long counters_old, | |
369 | void *freelist_new, unsigned long counters_new, | |
370 | const char *n) | |
371 | { | |
372 | VM_BUG_ON(!irqs_disabled()); | |
2565409f HC |
373 | #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \ |
374 | defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE) | |
1d07171c | 375 | if (s->flags & __CMPXCHG_DOUBLE) { |
cdcd6298 | 376 | if (cmpxchg_double(&page->freelist, &page->counters, |
0aa9a13d DC |
377 | freelist_old, counters_old, |
378 | freelist_new, counters_new)) | |
6f6528a1 | 379 | return true; |
1d07171c CL |
380 | } else |
381 | #endif | |
382 | { | |
383 | slab_lock(page); | |
d0e0ac97 CG |
384 | if (page->freelist == freelist_old && |
385 | page->counters == counters_old) { | |
1d07171c | 386 | page->freelist = freelist_new; |
a0320865 | 387 | set_page_slub_counters(page, counters_new); |
1d07171c | 388 | slab_unlock(page); |
6f6528a1 | 389 | return true; |
1d07171c CL |
390 | } |
391 | slab_unlock(page); | |
392 | } | |
393 | ||
394 | cpu_relax(); | |
395 | stat(s, CMPXCHG_DOUBLE_FAIL); | |
396 | ||
397 | #ifdef SLUB_DEBUG_CMPXCHG | |
f9f58285 | 398 | pr_info("%s %s: cmpxchg double redo ", n, s->name); |
1d07171c CL |
399 | #endif |
400 | ||
6f6528a1 | 401 | return false; |
1d07171c CL |
402 | } |
403 | ||
b789ef51 CL |
404 | static inline bool cmpxchg_double_slab(struct kmem_cache *s, struct page *page, |
405 | void *freelist_old, unsigned long counters_old, | |
406 | void *freelist_new, unsigned long counters_new, | |
407 | const char *n) | |
408 | { | |
2565409f HC |
409 | #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \ |
410 | defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE) | |
b789ef51 | 411 | if (s->flags & __CMPXCHG_DOUBLE) { |
cdcd6298 | 412 | if (cmpxchg_double(&page->freelist, &page->counters, |
0aa9a13d DC |
413 | freelist_old, counters_old, |
414 | freelist_new, counters_new)) | |
6f6528a1 | 415 | return true; |
b789ef51 CL |
416 | } else |
417 | #endif | |
418 | { | |
1d07171c CL |
419 | unsigned long flags; |
420 | ||
421 | local_irq_save(flags); | |
881db7fb | 422 | slab_lock(page); |
d0e0ac97 CG |
423 | if (page->freelist == freelist_old && |
424 | page->counters == counters_old) { | |
b789ef51 | 425 | page->freelist = freelist_new; |
a0320865 | 426 | set_page_slub_counters(page, counters_new); |
881db7fb | 427 | slab_unlock(page); |
1d07171c | 428 | local_irq_restore(flags); |
6f6528a1 | 429 | return true; |
b789ef51 | 430 | } |
881db7fb | 431 | slab_unlock(page); |
1d07171c | 432 | local_irq_restore(flags); |
b789ef51 CL |
433 | } |
434 | ||
435 | cpu_relax(); | |
436 | stat(s, CMPXCHG_DOUBLE_FAIL); | |
437 | ||
438 | #ifdef SLUB_DEBUG_CMPXCHG | |
f9f58285 | 439 | pr_info("%s %s: cmpxchg double redo ", n, s->name); |
b789ef51 CL |
440 | #endif |
441 | ||
6f6528a1 | 442 | return false; |
b789ef51 CL |
443 | } |
444 | ||
41ecc55b | 445 | #ifdef CONFIG_SLUB_DEBUG |
5f80b13a CL |
446 | /* |
447 | * Determine a map of object in use on a page. | |
448 | * | |
881db7fb | 449 | * Node listlock must be held to guarantee that the page does |
5f80b13a CL |
450 | * not vanish from under us. |
451 | */ | |
452 | static void get_map(struct kmem_cache *s, struct page *page, unsigned long *map) | |
453 | { | |
454 | void *p; | |
455 | void *addr = page_address(page); | |
456 | ||
457 | for (p = page->freelist; p; p = get_freepointer(s, p)) | |
458 | set_bit(slab_index(p, s, addr), map); | |
459 | } | |
460 | ||
41ecc55b CL |
461 | /* |
462 | * Debug settings: | |
463 | */ | |
89d3c87e | 464 | #if defined(CONFIG_SLUB_DEBUG_ON) |
f0630fff | 465 | static int slub_debug = DEBUG_DEFAULT_FLAGS; |
89d3c87e AR |
466 | #elif defined(CONFIG_KASAN) |
467 | static int slub_debug = SLAB_STORE_USER; | |
f0630fff | 468 | #else |
41ecc55b | 469 | static int slub_debug; |
f0630fff | 470 | #endif |
41ecc55b CL |
471 | |
472 | static char *slub_debug_slabs; | |
fa5ec8a1 | 473 | static int disable_higher_order_debug; |
41ecc55b | 474 | |
a79316c6 AR |
475 | /* |
476 | * slub is about to manipulate internal object metadata. This memory lies | |
477 | * outside the range of the allocated object, so accessing it would normally | |
478 | * be reported by kasan as a bounds error. metadata_access_enable() is used | |
479 | * to tell kasan that these accesses are OK. | |
480 | */ | |
481 | static inline void metadata_access_enable(void) | |
482 | { | |
483 | kasan_disable_current(); | |
484 | } | |
485 | ||
486 | static inline void metadata_access_disable(void) | |
487 | { | |
488 | kasan_enable_current(); | |
489 | } | |
490 | ||
81819f0f CL |
491 | /* |
492 | * Object debugging | |
493 | */ | |
494 | static void print_section(char *text, u8 *addr, unsigned int length) | |
495 | { | |
a79316c6 | 496 | metadata_access_enable(); |
ffc79d28 SAS |
497 | print_hex_dump(KERN_ERR, text, DUMP_PREFIX_ADDRESS, 16, 1, addr, |
498 | length, 1); | |
a79316c6 | 499 | metadata_access_disable(); |
81819f0f CL |
500 | } |
501 | ||
81819f0f CL |
502 | static struct track *get_track(struct kmem_cache *s, void *object, |
503 | enum track_item alloc) | |
504 | { | |
505 | struct track *p; | |
506 | ||
507 | if (s->offset) | |
508 | p = object + s->offset + sizeof(void *); | |
509 | else | |
510 | p = object + s->inuse; | |
511 | ||
512 | return p + alloc; | |
513 | } | |
514 | ||
515 | static void set_track(struct kmem_cache *s, void *object, | |
ce71e27c | 516 | enum track_item alloc, unsigned long addr) |
81819f0f | 517 | { |
1a00df4a | 518 | struct track *p = get_track(s, object, alloc); |
81819f0f | 519 | |
81819f0f | 520 | if (addr) { |
d6543e39 BG |
521 | #ifdef CONFIG_STACKTRACE |
522 | struct stack_trace trace; | |
523 | int i; | |
524 | ||
525 | trace.nr_entries = 0; | |
526 | trace.max_entries = TRACK_ADDRS_COUNT; | |
527 | trace.entries = p->addrs; | |
528 | trace.skip = 3; | |
a79316c6 | 529 | metadata_access_enable(); |
d6543e39 | 530 | save_stack_trace(&trace); |
a79316c6 | 531 | metadata_access_disable(); |
d6543e39 BG |
532 | |
533 | /* See rant in lockdep.c */ | |
534 | if (trace.nr_entries != 0 && | |
535 | trace.entries[trace.nr_entries - 1] == ULONG_MAX) | |
536 | trace.nr_entries--; | |
537 | ||
538 | for (i = trace.nr_entries; i < TRACK_ADDRS_COUNT; i++) | |
539 | p->addrs[i] = 0; | |
540 | #endif | |
81819f0f CL |
541 | p->addr = addr; |
542 | p->cpu = smp_processor_id(); | |
88e4ccf2 | 543 | p->pid = current->pid; |
81819f0f CL |
544 | p->when = jiffies; |
545 | } else | |
546 | memset(p, 0, sizeof(struct track)); | |
547 | } | |
548 | ||
81819f0f CL |
549 | static void init_tracking(struct kmem_cache *s, void *object) |
550 | { | |
24922684 CL |
551 | if (!(s->flags & SLAB_STORE_USER)) |
552 | return; | |
553 | ||
ce71e27c EGM |
554 | set_track(s, object, TRACK_FREE, 0UL); |
555 | set_track(s, object, TRACK_ALLOC, 0UL); | |
81819f0f CL |
556 | } |
557 | ||
558 | static void print_track(const char *s, struct track *t) | |
559 | { | |
560 | if (!t->addr) | |
561 | return; | |
562 | ||
f9f58285 FF |
563 | pr_err("INFO: %s in %pS age=%lu cpu=%u pid=%d\n", |
564 | s, (void *)t->addr, jiffies - t->when, t->cpu, t->pid); | |
d6543e39 BG |
565 | #ifdef CONFIG_STACKTRACE |
566 | { | |
567 | int i; | |
568 | for (i = 0; i < TRACK_ADDRS_COUNT; i++) | |
569 | if (t->addrs[i]) | |
f9f58285 | 570 | pr_err("\t%pS\n", (void *)t->addrs[i]); |
d6543e39 BG |
571 | else |
572 | break; | |
573 | } | |
574 | #endif | |
24922684 CL |
575 | } |
576 | ||
577 | static void print_tracking(struct kmem_cache *s, void *object) | |
578 | { | |
579 | if (!(s->flags & SLAB_STORE_USER)) | |
580 | return; | |
581 | ||
582 | print_track("Allocated", get_track(s, object, TRACK_ALLOC)); | |
583 | print_track("Freed", get_track(s, object, TRACK_FREE)); | |
584 | } | |
585 | ||
586 | static void print_page_info(struct page *page) | |
587 | { | |
f9f58285 | 588 | pr_err("INFO: Slab 0x%p objects=%u used=%u fp=0x%p flags=0x%04lx\n", |
d0e0ac97 | 589 | page, page->objects, page->inuse, page->freelist, page->flags); |
24922684 CL |
590 | |
591 | } | |
592 | ||
593 | static void slab_bug(struct kmem_cache *s, char *fmt, ...) | |
594 | { | |
ecc42fbe | 595 | struct va_format vaf; |
24922684 | 596 | va_list args; |
24922684 CL |
597 | |
598 | va_start(args, fmt); | |
ecc42fbe FF |
599 | vaf.fmt = fmt; |
600 | vaf.va = &args; | |
f9f58285 | 601 | pr_err("=============================================================================\n"); |
ecc42fbe | 602 | pr_err("BUG %s (%s): %pV\n", s->name, print_tainted(), &vaf); |
f9f58285 | 603 | pr_err("-----------------------------------------------------------------------------\n\n"); |
645df230 | 604 | |
373d4d09 | 605 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); |
ecc42fbe | 606 | va_end(args); |
81819f0f CL |
607 | } |
608 | ||
24922684 CL |
609 | static void slab_fix(struct kmem_cache *s, char *fmt, ...) |
610 | { | |
ecc42fbe | 611 | struct va_format vaf; |
24922684 | 612 | va_list args; |
24922684 CL |
613 | |
614 | va_start(args, fmt); | |
ecc42fbe FF |
615 | vaf.fmt = fmt; |
616 | vaf.va = &args; | |
617 | pr_err("FIX %s: %pV\n", s->name, &vaf); | |
24922684 | 618 | va_end(args); |
24922684 CL |
619 | } |
620 | ||
621 | static void print_trailer(struct kmem_cache *s, struct page *page, u8 *p) | |
81819f0f CL |
622 | { |
623 | unsigned int off; /* Offset of last byte */ | |
a973e9dd | 624 | u8 *addr = page_address(page); |
24922684 CL |
625 | |
626 | print_tracking(s, p); | |
627 | ||
628 | print_page_info(page); | |
629 | ||
f9f58285 FF |
630 | pr_err("INFO: Object 0x%p @offset=%tu fp=0x%p\n\n", |
631 | p, p - addr, get_freepointer(s, p)); | |
24922684 CL |
632 | |
633 | if (p > addr + 16) | |
ffc79d28 | 634 | print_section("Bytes b4 ", p - 16, 16); |
81819f0f | 635 | |
3b0efdfa | 636 | print_section("Object ", p, min_t(unsigned long, s->object_size, |
ffc79d28 | 637 | PAGE_SIZE)); |
81819f0f | 638 | if (s->flags & SLAB_RED_ZONE) |
3b0efdfa CL |
639 | print_section("Redzone ", p + s->object_size, |
640 | s->inuse - s->object_size); | |
81819f0f | 641 | |
81819f0f CL |
642 | if (s->offset) |
643 | off = s->offset + sizeof(void *); | |
644 | else | |
645 | off = s->inuse; | |
646 | ||
24922684 | 647 | if (s->flags & SLAB_STORE_USER) |
81819f0f | 648 | off += 2 * sizeof(struct track); |
81819f0f CL |
649 | |
650 | if (off != s->size) | |
651 | /* Beginning of the filler is the free pointer */ | |
ffc79d28 | 652 | print_section("Padding ", p + off, s->size - off); |
24922684 CL |
653 | |
654 | dump_stack(); | |
81819f0f CL |
655 | } |
656 | ||
75c66def | 657 | void object_err(struct kmem_cache *s, struct page *page, |
81819f0f CL |
658 | u8 *object, char *reason) |
659 | { | |
3dc50637 | 660 | slab_bug(s, "%s", reason); |
24922684 | 661 | print_trailer(s, page, object); |
81819f0f CL |
662 | } |
663 | ||
d0e0ac97 CG |
664 | static void slab_err(struct kmem_cache *s, struct page *page, |
665 | const char *fmt, ...) | |
81819f0f CL |
666 | { |
667 | va_list args; | |
668 | char buf[100]; | |
669 | ||
24922684 CL |
670 | va_start(args, fmt); |
671 | vsnprintf(buf, sizeof(buf), fmt, args); | |
81819f0f | 672 | va_end(args); |
3dc50637 | 673 | slab_bug(s, "%s", buf); |
24922684 | 674 | print_page_info(page); |
81819f0f CL |
675 | dump_stack(); |
676 | } | |
677 | ||
f7cb1933 | 678 | static void init_object(struct kmem_cache *s, void *object, u8 val) |
81819f0f CL |
679 | { |
680 | u8 *p = object; | |
681 | ||
682 | if (s->flags & __OBJECT_POISON) { | |
3b0efdfa CL |
683 | memset(p, POISON_FREE, s->object_size - 1); |
684 | p[s->object_size - 1] = POISON_END; | |
81819f0f CL |
685 | } |
686 | ||
687 | if (s->flags & SLAB_RED_ZONE) | |
3b0efdfa | 688 | memset(p + s->object_size, val, s->inuse - s->object_size); |
81819f0f CL |
689 | } |
690 | ||
24922684 CL |
691 | static void restore_bytes(struct kmem_cache *s, char *message, u8 data, |
692 | void *from, void *to) | |
693 | { | |
694 | slab_fix(s, "Restoring 0x%p-0x%p=0x%x\n", from, to - 1, data); | |
695 | memset(from, data, to - from); | |
696 | } | |
697 | ||
698 | static int check_bytes_and_report(struct kmem_cache *s, struct page *page, | |
699 | u8 *object, char *what, | |
06428780 | 700 | u8 *start, unsigned int value, unsigned int bytes) |
24922684 CL |
701 | { |
702 | u8 *fault; | |
703 | u8 *end; | |
704 | ||
a79316c6 | 705 | metadata_access_enable(); |
79824820 | 706 | fault = memchr_inv(start, value, bytes); |
a79316c6 | 707 | metadata_access_disable(); |
24922684 CL |
708 | if (!fault) |
709 | return 1; | |
710 | ||
711 | end = start + bytes; | |
712 | while (end > fault && end[-1] == value) | |
713 | end--; | |
714 | ||
715 | slab_bug(s, "%s overwritten", what); | |
f9f58285 | 716 | pr_err("INFO: 0x%p-0x%p. First byte 0x%x instead of 0x%x\n", |
24922684 CL |
717 | fault, end - 1, fault[0], value); |
718 | print_trailer(s, page, object); | |
719 | ||
720 | restore_bytes(s, what, value, fault, end); | |
721 | return 0; | |
81819f0f CL |
722 | } |
723 | ||
81819f0f CL |
724 | /* |
725 | * Object layout: | |
726 | * | |
727 | * object address | |
728 | * Bytes of the object to be managed. | |
729 | * If the freepointer may overlay the object then the free | |
730 | * pointer is the first word of the object. | |
672bba3a | 731 | * |
81819f0f CL |
732 | * Poisoning uses 0x6b (POISON_FREE) and the last byte is |
733 | * 0xa5 (POISON_END) | |
734 | * | |
3b0efdfa | 735 | * object + s->object_size |
81819f0f | 736 | * Padding to reach word boundary. This is also used for Redzoning. |
672bba3a | 737 | * Padding is extended by another word if Redzoning is enabled and |
3b0efdfa | 738 | * object_size == inuse. |
672bba3a | 739 | * |
81819f0f CL |
740 | * We fill with 0xbb (RED_INACTIVE) for inactive objects and with |
741 | * 0xcc (RED_ACTIVE) for objects in use. | |
742 | * | |
743 | * object + s->inuse | |
672bba3a CL |
744 | * Meta data starts here. |
745 | * | |
81819f0f CL |
746 | * A. Free pointer (if we cannot overwrite object on free) |
747 | * B. Tracking data for SLAB_STORE_USER | |
672bba3a | 748 | * C. Padding to reach required alignment boundary or at mininum |
6446faa2 | 749 | * one word if debugging is on to be able to detect writes |
672bba3a CL |
750 | * before the word boundary. |
751 | * | |
752 | * Padding is done using 0x5a (POISON_INUSE) | |
81819f0f CL |
753 | * |
754 | * object + s->size | |
672bba3a | 755 | * Nothing is used beyond s->size. |
81819f0f | 756 | * |
3b0efdfa | 757 | * If slabcaches are merged then the object_size and inuse boundaries are mostly |
672bba3a | 758 | * ignored. And therefore no slab options that rely on these boundaries |
81819f0f CL |
759 | * may be used with merged slabcaches. |
760 | */ | |
761 | ||
81819f0f CL |
762 | static int check_pad_bytes(struct kmem_cache *s, struct page *page, u8 *p) |
763 | { | |
764 | unsigned long off = s->inuse; /* The end of info */ | |
765 | ||
766 | if (s->offset) | |
767 | /* Freepointer is placed after the object. */ | |
768 | off += sizeof(void *); | |
769 | ||
770 | if (s->flags & SLAB_STORE_USER) | |
771 | /* We also have user information there */ | |
772 | off += 2 * sizeof(struct track); | |
773 | ||
774 | if (s->size == off) | |
775 | return 1; | |
776 | ||
24922684 CL |
777 | return check_bytes_and_report(s, page, p, "Object padding", |
778 | p + off, POISON_INUSE, s->size - off); | |
81819f0f CL |
779 | } |
780 | ||
39b26464 | 781 | /* Check the pad bytes at the end of a slab page */ |
81819f0f CL |
782 | static int slab_pad_check(struct kmem_cache *s, struct page *page) |
783 | { | |
24922684 CL |
784 | u8 *start; |
785 | u8 *fault; | |
786 | u8 *end; | |
787 | int length; | |
788 | int remainder; | |
81819f0f CL |
789 | |
790 | if (!(s->flags & SLAB_POISON)) | |
791 | return 1; | |
792 | ||
a973e9dd | 793 | start = page_address(page); |
ab9a0f19 | 794 | length = (PAGE_SIZE << compound_order(page)) - s->reserved; |
39b26464 CL |
795 | end = start + length; |
796 | remainder = length % s->size; | |
81819f0f CL |
797 | if (!remainder) |
798 | return 1; | |
799 | ||
a79316c6 | 800 | metadata_access_enable(); |
79824820 | 801 | fault = memchr_inv(end - remainder, POISON_INUSE, remainder); |
a79316c6 | 802 | metadata_access_disable(); |
24922684 CL |
803 | if (!fault) |
804 | return 1; | |
805 | while (end > fault && end[-1] == POISON_INUSE) | |
806 | end--; | |
807 | ||
808 | slab_err(s, page, "Padding overwritten. 0x%p-0x%p", fault, end - 1); | |
ffc79d28 | 809 | print_section("Padding ", end - remainder, remainder); |
24922684 | 810 | |
8a3d271d | 811 | restore_bytes(s, "slab padding", POISON_INUSE, end - remainder, end); |
24922684 | 812 | return 0; |
81819f0f CL |
813 | } |
814 | ||
815 | static int check_object(struct kmem_cache *s, struct page *page, | |
f7cb1933 | 816 | void *object, u8 val) |
81819f0f CL |
817 | { |
818 | u8 *p = object; | |
3b0efdfa | 819 | u8 *endobject = object + s->object_size; |
81819f0f CL |
820 | |
821 | if (s->flags & SLAB_RED_ZONE) { | |
24922684 | 822 | if (!check_bytes_and_report(s, page, object, "Redzone", |
3b0efdfa | 823 | endobject, val, s->inuse - s->object_size)) |
81819f0f | 824 | return 0; |
81819f0f | 825 | } else { |
3b0efdfa | 826 | if ((s->flags & SLAB_POISON) && s->object_size < s->inuse) { |
3adbefee | 827 | check_bytes_and_report(s, page, p, "Alignment padding", |
d0e0ac97 CG |
828 | endobject, POISON_INUSE, |
829 | s->inuse - s->object_size); | |
3adbefee | 830 | } |
81819f0f CL |
831 | } |
832 | ||
833 | if (s->flags & SLAB_POISON) { | |
f7cb1933 | 834 | if (val != SLUB_RED_ACTIVE && (s->flags & __OBJECT_POISON) && |
24922684 | 835 | (!check_bytes_and_report(s, page, p, "Poison", p, |
3b0efdfa | 836 | POISON_FREE, s->object_size - 1) || |
24922684 | 837 | !check_bytes_and_report(s, page, p, "Poison", |
3b0efdfa | 838 | p + s->object_size - 1, POISON_END, 1))) |
81819f0f | 839 | return 0; |
81819f0f CL |
840 | /* |
841 | * check_pad_bytes cleans up on its own. | |
842 | */ | |
843 | check_pad_bytes(s, page, p); | |
844 | } | |
845 | ||
f7cb1933 | 846 | if (!s->offset && val == SLUB_RED_ACTIVE) |
81819f0f CL |
847 | /* |
848 | * Object and freepointer overlap. Cannot check | |
849 | * freepointer while object is allocated. | |
850 | */ | |
851 | return 1; | |
852 | ||
853 | /* Check free pointer validity */ | |
854 | if (!check_valid_pointer(s, page, get_freepointer(s, p))) { | |
855 | object_err(s, page, p, "Freepointer corrupt"); | |
856 | /* | |
9f6c708e | 857 | * No choice but to zap it and thus lose the remainder |
81819f0f | 858 | * of the free objects in this slab. May cause |
672bba3a | 859 | * another error because the object count is now wrong. |
81819f0f | 860 | */ |
a973e9dd | 861 | set_freepointer(s, p, NULL); |
81819f0f CL |
862 | return 0; |
863 | } | |
864 | return 1; | |
865 | } | |
866 | ||
867 | static int check_slab(struct kmem_cache *s, struct page *page) | |
868 | { | |
39b26464 CL |
869 | int maxobj; |
870 | ||
81819f0f CL |
871 | VM_BUG_ON(!irqs_disabled()); |
872 | ||
873 | if (!PageSlab(page)) { | |
24922684 | 874 | slab_err(s, page, "Not a valid slab page"); |
81819f0f CL |
875 | return 0; |
876 | } | |
39b26464 | 877 | |
ab9a0f19 | 878 | maxobj = order_objects(compound_order(page), s->size, s->reserved); |
39b26464 CL |
879 | if (page->objects > maxobj) { |
880 | slab_err(s, page, "objects %u > max %u", | |
f6edde9c | 881 | page->objects, maxobj); |
39b26464 CL |
882 | return 0; |
883 | } | |
884 | if (page->inuse > page->objects) { | |
24922684 | 885 | slab_err(s, page, "inuse %u > max %u", |
f6edde9c | 886 | page->inuse, page->objects); |
81819f0f CL |
887 | return 0; |
888 | } | |
889 | /* Slab_pad_check fixes things up after itself */ | |
890 | slab_pad_check(s, page); | |
891 | return 1; | |
892 | } | |
893 | ||
894 | /* | |
672bba3a CL |
895 | * Determine if a certain object on a page is on the freelist. Must hold the |
896 | * slab lock to guarantee that the chains are in a consistent state. | |
81819f0f CL |
897 | */ |
898 | static int on_freelist(struct kmem_cache *s, struct page *page, void *search) | |
899 | { | |
900 | int nr = 0; | |
881db7fb | 901 | void *fp; |
81819f0f | 902 | void *object = NULL; |
f6edde9c | 903 | int max_objects; |
81819f0f | 904 | |
881db7fb | 905 | fp = page->freelist; |
39b26464 | 906 | while (fp && nr <= page->objects) { |
81819f0f CL |
907 | if (fp == search) |
908 | return 1; | |
909 | if (!check_valid_pointer(s, page, fp)) { | |
910 | if (object) { | |
911 | object_err(s, page, object, | |
912 | "Freechain corrupt"); | |
a973e9dd | 913 | set_freepointer(s, object, NULL); |
81819f0f | 914 | } else { |
24922684 | 915 | slab_err(s, page, "Freepointer corrupt"); |
a973e9dd | 916 | page->freelist = NULL; |
39b26464 | 917 | page->inuse = page->objects; |
24922684 | 918 | slab_fix(s, "Freelist cleared"); |
81819f0f CL |
919 | return 0; |
920 | } | |
921 | break; | |
922 | } | |
923 | object = fp; | |
924 | fp = get_freepointer(s, object); | |
925 | nr++; | |
926 | } | |
927 | ||
ab9a0f19 | 928 | max_objects = order_objects(compound_order(page), s->size, s->reserved); |
210b5c06 CG |
929 | if (max_objects > MAX_OBJS_PER_PAGE) |
930 | max_objects = MAX_OBJS_PER_PAGE; | |
224a88be CL |
931 | |
932 | if (page->objects != max_objects) { | |
933 | slab_err(s, page, "Wrong number of objects. Found %d but " | |
934 | "should be %d", page->objects, max_objects); | |
935 | page->objects = max_objects; | |
936 | slab_fix(s, "Number of objects adjusted."); | |
937 | } | |
39b26464 | 938 | if (page->inuse != page->objects - nr) { |
70d71228 | 939 | slab_err(s, page, "Wrong object count. Counter is %d but " |
39b26464 CL |
940 | "counted were %d", page->inuse, page->objects - nr); |
941 | page->inuse = page->objects - nr; | |
24922684 | 942 | slab_fix(s, "Object count adjusted."); |
81819f0f CL |
943 | } |
944 | return search == NULL; | |
945 | } | |
946 | ||
0121c619 CL |
947 | static void trace(struct kmem_cache *s, struct page *page, void *object, |
948 | int alloc) | |
3ec09742 CL |
949 | { |
950 | if (s->flags & SLAB_TRACE) { | |
f9f58285 | 951 | pr_info("TRACE %s %s 0x%p inuse=%d fp=0x%p\n", |
3ec09742 CL |
952 | s->name, |
953 | alloc ? "alloc" : "free", | |
954 | object, page->inuse, | |
955 | page->freelist); | |
956 | ||
957 | if (!alloc) | |
d0e0ac97 CG |
958 | print_section("Object ", (void *)object, |
959 | s->object_size); | |
3ec09742 CL |
960 | |
961 | dump_stack(); | |
962 | } | |
963 | } | |
964 | ||
643b1138 | 965 | /* |
672bba3a | 966 | * Tracking of fully allocated slabs for debugging purposes. |
643b1138 | 967 | */ |
5cc6eee8 CL |
968 | static void add_full(struct kmem_cache *s, |
969 | struct kmem_cache_node *n, struct page *page) | |
643b1138 | 970 | { |
5cc6eee8 CL |
971 | if (!(s->flags & SLAB_STORE_USER)) |
972 | return; | |
973 | ||
255d0884 | 974 | lockdep_assert_held(&n->list_lock); |
643b1138 | 975 | list_add(&page->lru, &n->full); |
643b1138 CL |
976 | } |
977 | ||
c65c1877 | 978 | static void remove_full(struct kmem_cache *s, struct kmem_cache_node *n, struct page *page) |
643b1138 | 979 | { |
643b1138 CL |
980 | if (!(s->flags & SLAB_STORE_USER)) |
981 | return; | |
982 | ||
255d0884 | 983 | lockdep_assert_held(&n->list_lock); |
643b1138 | 984 | list_del(&page->lru); |
643b1138 CL |
985 | } |
986 | ||
0f389ec6 CL |
987 | /* Tracking of the number of slabs for debugging purposes */ |
988 | static inline unsigned long slabs_node(struct kmem_cache *s, int node) | |
989 | { | |
990 | struct kmem_cache_node *n = get_node(s, node); | |
991 | ||
992 | return atomic_long_read(&n->nr_slabs); | |
993 | } | |
994 | ||
26c02cf0 AB |
995 | static inline unsigned long node_nr_slabs(struct kmem_cache_node *n) |
996 | { | |
997 | return atomic_long_read(&n->nr_slabs); | |
998 | } | |
999 | ||
205ab99d | 1000 | static inline void inc_slabs_node(struct kmem_cache *s, int node, int objects) |
0f389ec6 CL |
1001 | { |
1002 | struct kmem_cache_node *n = get_node(s, node); | |
1003 | ||
1004 | /* | |
1005 | * May be called early in order to allocate a slab for the | |
1006 | * kmem_cache_node structure. Solve the chicken-egg | |
1007 | * dilemma by deferring the increment of the count during | |
1008 | * bootstrap (see early_kmem_cache_node_alloc). | |
1009 | */ | |
338b2642 | 1010 | if (likely(n)) { |
0f389ec6 | 1011 | atomic_long_inc(&n->nr_slabs); |
205ab99d CL |
1012 | atomic_long_add(objects, &n->total_objects); |
1013 | } | |
0f389ec6 | 1014 | } |
205ab99d | 1015 | static inline void dec_slabs_node(struct kmem_cache *s, int node, int objects) |
0f389ec6 CL |
1016 | { |
1017 | struct kmem_cache_node *n = get_node(s, node); | |
1018 | ||
1019 | atomic_long_dec(&n->nr_slabs); | |
205ab99d | 1020 | atomic_long_sub(objects, &n->total_objects); |
0f389ec6 CL |
1021 | } |
1022 | ||
1023 | /* Object debug checks for alloc/free paths */ | |
3ec09742 CL |
1024 | static void setup_object_debug(struct kmem_cache *s, struct page *page, |
1025 | void *object) | |
1026 | { | |
1027 | if (!(s->flags & (SLAB_STORE_USER|SLAB_RED_ZONE|__OBJECT_POISON))) | |
1028 | return; | |
1029 | ||
f7cb1933 | 1030 | init_object(s, object, SLUB_RED_INACTIVE); |
3ec09742 CL |
1031 | init_tracking(s, object); |
1032 | } | |
1033 | ||
d0e0ac97 CG |
1034 | static noinline int alloc_debug_processing(struct kmem_cache *s, |
1035 | struct page *page, | |
ce71e27c | 1036 | void *object, unsigned long addr) |
81819f0f CL |
1037 | { |
1038 | if (!check_slab(s, page)) | |
1039 | goto bad; | |
1040 | ||
81819f0f CL |
1041 | if (!check_valid_pointer(s, page, object)) { |
1042 | object_err(s, page, object, "Freelist Pointer check fails"); | |
70d71228 | 1043 | goto bad; |
81819f0f CL |
1044 | } |
1045 | ||
f7cb1933 | 1046 | if (!check_object(s, page, object, SLUB_RED_INACTIVE)) |
81819f0f | 1047 | goto bad; |
81819f0f | 1048 | |
3ec09742 CL |
1049 | /* Success perform special debug activities for allocs */ |
1050 | if (s->flags & SLAB_STORE_USER) | |
1051 | set_track(s, object, TRACK_ALLOC, addr); | |
1052 | trace(s, page, object, 1); | |
f7cb1933 | 1053 | init_object(s, object, SLUB_RED_ACTIVE); |
81819f0f | 1054 | return 1; |
3ec09742 | 1055 | |
81819f0f CL |
1056 | bad: |
1057 | if (PageSlab(page)) { | |
1058 | /* | |
1059 | * If this is a slab page then lets do the best we can | |
1060 | * to avoid issues in the future. Marking all objects | |
672bba3a | 1061 | * as used avoids touching the remaining objects. |
81819f0f | 1062 | */ |
24922684 | 1063 | slab_fix(s, "Marking all objects used"); |
39b26464 | 1064 | page->inuse = page->objects; |
a973e9dd | 1065 | page->freelist = NULL; |
81819f0f CL |
1066 | } |
1067 | return 0; | |
1068 | } | |
1069 | ||
81084651 | 1070 | /* Supports checking bulk free of a constructed freelist */ |
19c7ff9e | 1071 | static noinline struct kmem_cache_node *free_debug_processing( |
81084651 JDB |
1072 | struct kmem_cache *s, struct page *page, |
1073 | void *head, void *tail, int bulk_cnt, | |
19c7ff9e | 1074 | unsigned long addr, unsigned long *flags) |
81819f0f | 1075 | { |
19c7ff9e | 1076 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); |
81084651 JDB |
1077 | void *object = head; |
1078 | int cnt = 0; | |
5c2e4bbb | 1079 | |
19c7ff9e | 1080 | spin_lock_irqsave(&n->list_lock, *flags); |
881db7fb CL |
1081 | slab_lock(page); |
1082 | ||
81819f0f CL |
1083 | if (!check_slab(s, page)) |
1084 | goto fail; | |
1085 | ||
81084651 JDB |
1086 | next_object: |
1087 | cnt++; | |
1088 | ||
81819f0f | 1089 | if (!check_valid_pointer(s, page, object)) { |
70d71228 | 1090 | slab_err(s, page, "Invalid object pointer 0x%p", object); |
81819f0f CL |
1091 | goto fail; |
1092 | } | |
1093 | ||
1094 | if (on_freelist(s, page, object)) { | |
24922684 | 1095 | object_err(s, page, object, "Object already free"); |
81819f0f CL |
1096 | goto fail; |
1097 | } | |
1098 | ||
f7cb1933 | 1099 | if (!check_object(s, page, object, SLUB_RED_ACTIVE)) |
5c2e4bbb | 1100 | goto out; |
81819f0f | 1101 | |
1b4f59e3 | 1102 | if (unlikely(s != page->slab_cache)) { |
3adbefee | 1103 | if (!PageSlab(page)) { |
70d71228 CL |
1104 | slab_err(s, page, "Attempt to free object(0x%p) " |
1105 | "outside of slab", object); | |
1b4f59e3 | 1106 | } else if (!page->slab_cache) { |
f9f58285 FF |
1107 | pr_err("SLUB <none>: no slab for object 0x%p.\n", |
1108 | object); | |
70d71228 | 1109 | dump_stack(); |
06428780 | 1110 | } else |
24922684 CL |
1111 | object_err(s, page, object, |
1112 | "page slab pointer corrupt."); | |
81819f0f CL |
1113 | goto fail; |
1114 | } | |
3ec09742 | 1115 | |
3ec09742 CL |
1116 | if (s->flags & SLAB_STORE_USER) |
1117 | set_track(s, object, TRACK_FREE, addr); | |
1118 | trace(s, page, object, 0); | |
81084651 | 1119 | /* Freepointer not overwritten by init_object(), SLAB_POISON moved it */ |
f7cb1933 | 1120 | init_object(s, object, SLUB_RED_INACTIVE); |
81084651 JDB |
1121 | |
1122 | /* Reached end of constructed freelist yet? */ | |
1123 | if (object != tail) { | |
1124 | object = get_freepointer(s, object); | |
1125 | goto next_object; | |
1126 | } | |
5c2e4bbb | 1127 | out: |
81084651 JDB |
1128 | if (cnt != bulk_cnt) |
1129 | slab_err(s, page, "Bulk freelist count(%d) invalid(%d)\n", | |
1130 | bulk_cnt, cnt); | |
1131 | ||
881db7fb | 1132 | slab_unlock(page); |
19c7ff9e CL |
1133 | /* |
1134 | * Keep node_lock to preserve integrity | |
1135 | * until the object is actually freed | |
1136 | */ | |
1137 | return n; | |
3ec09742 | 1138 | |
81819f0f | 1139 | fail: |
19c7ff9e CL |
1140 | slab_unlock(page); |
1141 | spin_unlock_irqrestore(&n->list_lock, *flags); | |
24922684 | 1142 | slab_fix(s, "Object at 0x%p not freed", object); |
19c7ff9e | 1143 | return NULL; |
81819f0f CL |
1144 | } |
1145 | ||
41ecc55b CL |
1146 | static int __init setup_slub_debug(char *str) |
1147 | { | |
f0630fff CL |
1148 | slub_debug = DEBUG_DEFAULT_FLAGS; |
1149 | if (*str++ != '=' || !*str) | |
1150 | /* | |
1151 | * No options specified. Switch on full debugging. | |
1152 | */ | |
1153 | goto out; | |
1154 | ||
1155 | if (*str == ',') | |
1156 | /* | |
1157 | * No options but restriction on slabs. This means full | |
1158 | * debugging for slabs matching a pattern. | |
1159 | */ | |
1160 | goto check_slabs; | |
1161 | ||
1162 | slub_debug = 0; | |
1163 | if (*str == '-') | |
1164 | /* | |
1165 | * Switch off all debugging measures. | |
1166 | */ | |
1167 | goto out; | |
1168 | ||
1169 | /* | |
1170 | * Determine which debug features should be switched on | |
1171 | */ | |
06428780 | 1172 | for (; *str && *str != ','; str++) { |
f0630fff CL |
1173 | switch (tolower(*str)) { |
1174 | case 'f': | |
1175 | slub_debug |= SLAB_DEBUG_FREE; | |
1176 | break; | |
1177 | case 'z': | |
1178 | slub_debug |= SLAB_RED_ZONE; | |
1179 | break; | |
1180 | case 'p': | |
1181 | slub_debug |= SLAB_POISON; | |
1182 | break; | |
1183 | case 'u': | |
1184 | slub_debug |= SLAB_STORE_USER; | |
1185 | break; | |
1186 | case 't': | |
1187 | slub_debug |= SLAB_TRACE; | |
1188 | break; | |
4c13dd3b DM |
1189 | case 'a': |
1190 | slub_debug |= SLAB_FAILSLAB; | |
1191 | break; | |
08303a73 CA |
1192 | case 'o': |
1193 | /* | |
1194 | * Avoid enabling debugging on caches if its minimum | |
1195 | * order would increase as a result. | |
1196 | */ | |
1197 | disable_higher_order_debug = 1; | |
1198 | break; | |
f0630fff | 1199 | default: |
f9f58285 FF |
1200 | pr_err("slub_debug option '%c' unknown. skipped\n", |
1201 | *str); | |
f0630fff | 1202 | } |
41ecc55b CL |
1203 | } |
1204 | ||
f0630fff | 1205 | check_slabs: |
41ecc55b CL |
1206 | if (*str == ',') |
1207 | slub_debug_slabs = str + 1; | |
f0630fff | 1208 | out: |
41ecc55b CL |
1209 | return 1; |
1210 | } | |
1211 | ||
1212 | __setup("slub_debug", setup_slub_debug); | |
1213 | ||
423c929c | 1214 | unsigned long kmem_cache_flags(unsigned long object_size, |
ba0268a8 | 1215 | unsigned long flags, const char *name, |
51cc5068 | 1216 | void (*ctor)(void *)) |
41ecc55b CL |
1217 | { |
1218 | /* | |
e153362a | 1219 | * Enable debugging if selected on the kernel commandline. |
41ecc55b | 1220 | */ |
c6f58d9b CL |
1221 | if (slub_debug && (!slub_debug_slabs || (name && |
1222 | !strncmp(slub_debug_slabs, name, strlen(slub_debug_slabs))))) | |
3de47213 | 1223 | flags |= slub_debug; |
ba0268a8 CL |
1224 | |
1225 | return flags; | |
41ecc55b | 1226 | } |
b4a64718 | 1227 | #else /* !CONFIG_SLUB_DEBUG */ |
3ec09742 CL |
1228 | static inline void setup_object_debug(struct kmem_cache *s, |
1229 | struct page *page, void *object) {} | |
41ecc55b | 1230 | |
3ec09742 | 1231 | static inline int alloc_debug_processing(struct kmem_cache *s, |
ce71e27c | 1232 | struct page *page, void *object, unsigned long addr) { return 0; } |
41ecc55b | 1233 | |
19c7ff9e | 1234 | static inline struct kmem_cache_node *free_debug_processing( |
81084651 JDB |
1235 | struct kmem_cache *s, struct page *page, |
1236 | void *head, void *tail, int bulk_cnt, | |
19c7ff9e | 1237 | unsigned long addr, unsigned long *flags) { return NULL; } |
41ecc55b | 1238 | |
41ecc55b CL |
1239 | static inline int slab_pad_check(struct kmem_cache *s, struct page *page) |
1240 | { return 1; } | |
1241 | static inline int check_object(struct kmem_cache *s, struct page *page, | |
f7cb1933 | 1242 | void *object, u8 val) { return 1; } |
5cc6eee8 CL |
1243 | static inline void add_full(struct kmem_cache *s, struct kmem_cache_node *n, |
1244 | struct page *page) {} | |
c65c1877 PZ |
1245 | static inline void remove_full(struct kmem_cache *s, struct kmem_cache_node *n, |
1246 | struct page *page) {} | |
423c929c | 1247 | unsigned long kmem_cache_flags(unsigned long object_size, |
ba0268a8 | 1248 | unsigned long flags, const char *name, |
51cc5068 | 1249 | void (*ctor)(void *)) |
ba0268a8 CL |
1250 | { |
1251 | return flags; | |
1252 | } | |
41ecc55b | 1253 | #define slub_debug 0 |
0f389ec6 | 1254 | |
fdaa45e9 IM |
1255 | #define disable_higher_order_debug 0 |
1256 | ||
0f389ec6 CL |
1257 | static inline unsigned long slabs_node(struct kmem_cache *s, int node) |
1258 | { return 0; } | |
26c02cf0 AB |
1259 | static inline unsigned long node_nr_slabs(struct kmem_cache_node *n) |
1260 | { return 0; } | |
205ab99d CL |
1261 | static inline void inc_slabs_node(struct kmem_cache *s, int node, |
1262 | int objects) {} | |
1263 | static inline void dec_slabs_node(struct kmem_cache *s, int node, | |
1264 | int objects) {} | |
7d550c56 | 1265 | |
02e72cc6 AR |
1266 | #endif /* CONFIG_SLUB_DEBUG */ |
1267 | ||
1268 | /* | |
1269 | * Hooks for other subsystems that check memory allocations. In a typical | |
1270 | * production configuration these hooks all should produce no code at all. | |
1271 | */ | |
d56791b3 RB |
1272 | static inline void kmalloc_large_node_hook(void *ptr, size_t size, gfp_t flags) |
1273 | { | |
1274 | kmemleak_alloc(ptr, size, 1, flags); | |
0316bec2 | 1275 | kasan_kmalloc_large(ptr, size); |
d56791b3 RB |
1276 | } |
1277 | ||
1278 | static inline void kfree_hook(const void *x) | |
1279 | { | |
1280 | kmemleak_free(x); | |
0316bec2 | 1281 | kasan_kfree_large(x); |
d56791b3 RB |
1282 | } |
1283 | ||
8135be5a VD |
1284 | static inline struct kmem_cache *slab_pre_alloc_hook(struct kmem_cache *s, |
1285 | gfp_t flags) | |
02e72cc6 AR |
1286 | { |
1287 | flags &= gfp_allowed_mask; | |
1288 | lockdep_trace_alloc(flags); | |
d0164adc | 1289 | might_sleep_if(gfpflags_allow_blocking(flags)); |
7d550c56 | 1290 | |
8135be5a VD |
1291 | if (should_failslab(s->object_size, flags, s->flags)) |
1292 | return NULL; | |
1293 | ||
1294 | return memcg_kmem_get_cache(s, flags); | |
02e72cc6 AR |
1295 | } |
1296 | ||
03ec0ed5 JDB |
1297 | static inline void slab_post_alloc_hook(struct kmem_cache *s, gfp_t flags, |
1298 | size_t size, void **p) | |
d56791b3 | 1299 | { |
03ec0ed5 JDB |
1300 | size_t i; |
1301 | ||
02e72cc6 | 1302 | flags &= gfp_allowed_mask; |
03ec0ed5 JDB |
1303 | for (i = 0; i < size; i++) { |
1304 | void *object = p[i]; | |
1305 | ||
1306 | kmemcheck_slab_alloc(s, flags, object, slab_ksize(s)); | |
1307 | kmemleak_alloc_recursive(object, s->object_size, 1, | |
1308 | s->flags, flags); | |
1309 | kasan_slab_alloc(s, object); | |
1310 | } | |
8135be5a | 1311 | memcg_kmem_put_cache(s); |
d56791b3 | 1312 | } |
7d550c56 | 1313 | |
d56791b3 RB |
1314 | static inline void slab_free_hook(struct kmem_cache *s, void *x) |
1315 | { | |
1316 | kmemleak_free_recursive(x, s->flags); | |
7d550c56 | 1317 | |
02e72cc6 AR |
1318 | /* |
1319 | * Trouble is that we may no longer disable interrupts in the fast path | |
1320 | * So in order to make the debug calls that expect irqs to be | |
1321 | * disabled we need to disable interrupts temporarily. | |
1322 | */ | |
1323 | #if defined(CONFIG_KMEMCHECK) || defined(CONFIG_LOCKDEP) | |
1324 | { | |
1325 | unsigned long flags; | |
1326 | ||
1327 | local_irq_save(flags); | |
1328 | kmemcheck_slab_free(s, x, s->object_size); | |
1329 | debug_check_no_locks_freed(x, s->object_size); | |
1330 | local_irq_restore(flags); | |
1331 | } | |
1332 | #endif | |
1333 | if (!(s->flags & SLAB_DEBUG_OBJECTS)) | |
1334 | debug_check_no_obj_freed(x, s->object_size); | |
0316bec2 AR |
1335 | |
1336 | kasan_slab_free(s, x); | |
02e72cc6 | 1337 | } |
205ab99d | 1338 | |
81084651 JDB |
1339 | static inline void slab_free_freelist_hook(struct kmem_cache *s, |
1340 | void *head, void *tail) | |
1341 | { | |
1342 | /* | |
1343 | * Compiler cannot detect this function can be removed if slab_free_hook() | |
1344 | * evaluates to nothing. Thus, catch all relevant config debug options here. | |
1345 | */ | |
1346 | #if defined(CONFIG_KMEMCHECK) || \ | |
1347 | defined(CONFIG_LOCKDEP) || \ | |
1348 | defined(CONFIG_DEBUG_KMEMLEAK) || \ | |
1349 | defined(CONFIG_DEBUG_OBJECTS_FREE) || \ | |
1350 | defined(CONFIG_KASAN) | |
1351 | ||
1352 | void *object = head; | |
1353 | void *tail_obj = tail ? : head; | |
1354 | ||
1355 | do { | |
1356 | slab_free_hook(s, object); | |
1357 | } while ((object != tail_obj) && | |
1358 | (object = get_freepointer(s, object))); | |
1359 | #endif | |
1360 | } | |
1361 | ||
588f8ba9 TG |
1362 | static void setup_object(struct kmem_cache *s, struct page *page, |
1363 | void *object) | |
1364 | { | |
1365 | setup_object_debug(s, page, object); | |
1366 | if (unlikely(s->ctor)) { | |
1367 | kasan_unpoison_object_data(s, object); | |
1368 | s->ctor(object); | |
1369 | kasan_poison_object_data(s, object); | |
1370 | } | |
1371 | } | |
1372 | ||
81819f0f CL |
1373 | /* |
1374 | * Slab allocation and freeing | |
1375 | */ | |
5dfb4175 VD |
1376 | static inline struct page *alloc_slab_page(struct kmem_cache *s, |
1377 | gfp_t flags, int node, struct kmem_cache_order_objects oo) | |
65c3376a | 1378 | { |
5dfb4175 | 1379 | struct page *page; |
65c3376a CL |
1380 | int order = oo_order(oo); |
1381 | ||
b1eeab67 VN |
1382 | flags |= __GFP_NOTRACK; |
1383 | ||
2154a336 | 1384 | if (node == NUMA_NO_NODE) |
5dfb4175 | 1385 | page = alloc_pages(flags, order); |
65c3376a | 1386 | else |
96db800f | 1387 | page = __alloc_pages_node(node, flags, order); |
5dfb4175 | 1388 | |
f3ccb2c4 VD |
1389 | if (page && memcg_charge_slab(page, flags, order, s)) { |
1390 | __free_pages(page, order); | |
1391 | page = NULL; | |
1392 | } | |
5dfb4175 VD |
1393 | |
1394 | return page; | |
65c3376a CL |
1395 | } |
1396 | ||
81819f0f CL |
1397 | static struct page *allocate_slab(struct kmem_cache *s, gfp_t flags, int node) |
1398 | { | |
06428780 | 1399 | struct page *page; |
834f3d11 | 1400 | struct kmem_cache_order_objects oo = s->oo; |
ba52270d | 1401 | gfp_t alloc_gfp; |
588f8ba9 TG |
1402 | void *start, *p; |
1403 | int idx, order; | |
81819f0f | 1404 | |
7e0528da CL |
1405 | flags &= gfp_allowed_mask; |
1406 | ||
d0164adc | 1407 | if (gfpflags_allow_blocking(flags)) |
7e0528da CL |
1408 | local_irq_enable(); |
1409 | ||
b7a49f0d | 1410 | flags |= s->allocflags; |
e12ba74d | 1411 | |
ba52270d PE |
1412 | /* |
1413 | * Let the initial higher-order allocation fail under memory pressure | |
1414 | * so we fall-back to the minimum order allocation. | |
1415 | */ | |
1416 | alloc_gfp = (flags | __GFP_NOWARN | __GFP_NORETRY) & ~__GFP_NOFAIL; | |
d0164adc MG |
1417 | if ((alloc_gfp & __GFP_DIRECT_RECLAIM) && oo_order(oo) > oo_order(s->min)) |
1418 | alloc_gfp = (alloc_gfp | __GFP_NOMEMALLOC) & ~__GFP_DIRECT_RECLAIM; | |
ba52270d | 1419 | |
5dfb4175 | 1420 | page = alloc_slab_page(s, alloc_gfp, node, oo); |
65c3376a CL |
1421 | if (unlikely(!page)) { |
1422 | oo = s->min; | |
80c3a998 | 1423 | alloc_gfp = flags; |
65c3376a CL |
1424 | /* |
1425 | * Allocation may have failed due to fragmentation. | |
1426 | * Try a lower order alloc if possible | |
1427 | */ | |
5dfb4175 | 1428 | page = alloc_slab_page(s, alloc_gfp, node, oo); |
588f8ba9 TG |
1429 | if (unlikely(!page)) |
1430 | goto out; | |
1431 | stat(s, ORDER_FALLBACK); | |
65c3376a | 1432 | } |
5a896d9e | 1433 | |
588f8ba9 TG |
1434 | if (kmemcheck_enabled && |
1435 | !(s->flags & (SLAB_NOTRACK | DEBUG_DEFAULT_FLAGS))) { | |
b1eeab67 VN |
1436 | int pages = 1 << oo_order(oo); |
1437 | ||
80c3a998 | 1438 | kmemcheck_alloc_shadow(page, oo_order(oo), alloc_gfp, node); |
b1eeab67 VN |
1439 | |
1440 | /* | |
1441 | * Objects from caches that have a constructor don't get | |
1442 | * cleared when they're allocated, so we need to do it here. | |
1443 | */ | |
1444 | if (s->ctor) | |
1445 | kmemcheck_mark_uninitialized_pages(page, pages); | |
1446 | else | |
1447 | kmemcheck_mark_unallocated_pages(page, pages); | |
5a896d9e VN |
1448 | } |
1449 | ||
834f3d11 | 1450 | page->objects = oo_objects(oo); |
81819f0f | 1451 | |
1f458cbf | 1452 | order = compound_order(page); |
1b4f59e3 | 1453 | page->slab_cache = s; |
c03f94cc | 1454 | __SetPageSlab(page); |
2f064f34 | 1455 | if (page_is_pfmemalloc(page)) |
072bb0aa | 1456 | SetPageSlabPfmemalloc(page); |
81819f0f CL |
1457 | |
1458 | start = page_address(page); | |
81819f0f CL |
1459 | |
1460 | if (unlikely(s->flags & SLAB_POISON)) | |
1f458cbf | 1461 | memset(start, POISON_INUSE, PAGE_SIZE << order); |
81819f0f | 1462 | |
0316bec2 AR |
1463 | kasan_poison_slab(page); |
1464 | ||
54266640 WY |
1465 | for_each_object_idx(p, idx, s, start, page->objects) { |
1466 | setup_object(s, page, p); | |
1467 | if (likely(idx < page->objects)) | |
1468 | set_freepointer(s, p, p + s->size); | |
1469 | else | |
1470 | set_freepointer(s, p, NULL); | |
81819f0f | 1471 | } |
81819f0f CL |
1472 | |
1473 | page->freelist = start; | |
e6e82ea1 | 1474 | page->inuse = page->objects; |
8cb0a506 | 1475 | page->frozen = 1; |
588f8ba9 | 1476 | |
81819f0f | 1477 | out: |
d0164adc | 1478 | if (gfpflags_allow_blocking(flags)) |
588f8ba9 TG |
1479 | local_irq_disable(); |
1480 | if (!page) | |
1481 | return NULL; | |
1482 | ||
1483 | mod_zone_page_state(page_zone(page), | |
1484 | (s->flags & SLAB_RECLAIM_ACCOUNT) ? | |
1485 | NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE, | |
1486 | 1 << oo_order(oo)); | |
1487 | ||
1488 | inc_slabs_node(s, page_to_nid(page), page->objects); | |
1489 | ||
81819f0f CL |
1490 | return page; |
1491 | } | |
1492 | ||
588f8ba9 TG |
1493 | static struct page *new_slab(struct kmem_cache *s, gfp_t flags, int node) |
1494 | { | |
1495 | if (unlikely(flags & GFP_SLAB_BUG_MASK)) { | |
1496 | pr_emerg("gfp: %u\n", flags & GFP_SLAB_BUG_MASK); | |
1497 | BUG(); | |
1498 | } | |
1499 | ||
1500 | return allocate_slab(s, | |
1501 | flags & (GFP_RECLAIM_MASK | GFP_CONSTRAINT_MASK), node); | |
1502 | } | |
1503 | ||
81819f0f CL |
1504 | static void __free_slab(struct kmem_cache *s, struct page *page) |
1505 | { | |
834f3d11 CL |
1506 | int order = compound_order(page); |
1507 | int pages = 1 << order; | |
81819f0f | 1508 | |
af537b0a | 1509 | if (kmem_cache_debug(s)) { |
81819f0f CL |
1510 | void *p; |
1511 | ||
1512 | slab_pad_check(s, page); | |
224a88be CL |
1513 | for_each_object(p, s, page_address(page), |
1514 | page->objects) | |
f7cb1933 | 1515 | check_object(s, page, p, SLUB_RED_INACTIVE); |
81819f0f CL |
1516 | } |
1517 | ||
b1eeab67 | 1518 | kmemcheck_free_shadow(page, compound_order(page)); |
5a896d9e | 1519 | |
81819f0f CL |
1520 | mod_zone_page_state(page_zone(page), |
1521 | (s->flags & SLAB_RECLAIM_ACCOUNT) ? | |
1522 | NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE, | |
06428780 | 1523 | -pages); |
81819f0f | 1524 | |
072bb0aa | 1525 | __ClearPageSlabPfmemalloc(page); |
49bd5221 | 1526 | __ClearPageSlab(page); |
1f458cbf | 1527 | |
22b751c3 | 1528 | page_mapcount_reset(page); |
1eb5ac64 NP |
1529 | if (current->reclaim_state) |
1530 | current->reclaim_state->reclaimed_slab += pages; | |
f3ccb2c4 | 1531 | __free_kmem_pages(page, order); |
81819f0f CL |
1532 | } |
1533 | ||
da9a638c LJ |
1534 | #define need_reserve_slab_rcu \ |
1535 | (sizeof(((struct page *)NULL)->lru) < sizeof(struct rcu_head)) | |
1536 | ||
81819f0f CL |
1537 | static void rcu_free_slab(struct rcu_head *h) |
1538 | { | |
1539 | struct page *page; | |
1540 | ||
da9a638c LJ |
1541 | if (need_reserve_slab_rcu) |
1542 | page = virt_to_head_page(h); | |
1543 | else | |
1544 | page = container_of((struct list_head *)h, struct page, lru); | |
1545 | ||
1b4f59e3 | 1546 | __free_slab(page->slab_cache, page); |
81819f0f CL |
1547 | } |
1548 | ||
1549 | static void free_slab(struct kmem_cache *s, struct page *page) | |
1550 | { | |
1551 | if (unlikely(s->flags & SLAB_DESTROY_BY_RCU)) { | |
da9a638c LJ |
1552 | struct rcu_head *head; |
1553 | ||
1554 | if (need_reserve_slab_rcu) { | |
1555 | int order = compound_order(page); | |
1556 | int offset = (PAGE_SIZE << order) - s->reserved; | |
1557 | ||
1558 | VM_BUG_ON(s->reserved != sizeof(*head)); | |
1559 | head = page_address(page) + offset; | |
1560 | } else { | |
bc4f610d | 1561 | head = &page->rcu_head; |
da9a638c | 1562 | } |
81819f0f CL |
1563 | |
1564 | call_rcu(head, rcu_free_slab); | |
1565 | } else | |
1566 | __free_slab(s, page); | |
1567 | } | |
1568 | ||
1569 | static void discard_slab(struct kmem_cache *s, struct page *page) | |
1570 | { | |
205ab99d | 1571 | dec_slabs_node(s, page_to_nid(page), page->objects); |
81819f0f CL |
1572 | free_slab(s, page); |
1573 | } | |
1574 | ||
1575 | /* | |
5cc6eee8 | 1576 | * Management of partially allocated slabs. |
81819f0f | 1577 | */ |
1e4dd946 SR |
1578 | static inline void |
1579 | __add_partial(struct kmem_cache_node *n, struct page *page, int tail) | |
81819f0f | 1580 | { |
e95eed57 | 1581 | n->nr_partial++; |
136333d1 | 1582 | if (tail == DEACTIVATE_TO_TAIL) |
7c2e132c CL |
1583 | list_add_tail(&page->lru, &n->partial); |
1584 | else | |
1585 | list_add(&page->lru, &n->partial); | |
81819f0f CL |
1586 | } |
1587 | ||
1e4dd946 SR |
1588 | static inline void add_partial(struct kmem_cache_node *n, |
1589 | struct page *page, int tail) | |
62e346a8 | 1590 | { |
c65c1877 | 1591 | lockdep_assert_held(&n->list_lock); |
1e4dd946 SR |
1592 | __add_partial(n, page, tail); |
1593 | } | |
c65c1877 | 1594 | |
1e4dd946 SR |
1595 | static inline void remove_partial(struct kmem_cache_node *n, |
1596 | struct page *page) | |
1597 | { | |
1598 | lockdep_assert_held(&n->list_lock); | |
52b4b950 DS |
1599 | list_del(&page->lru); |
1600 | n->nr_partial--; | |
1e4dd946 SR |
1601 | } |
1602 | ||
81819f0f | 1603 | /* |
7ced3719 CL |
1604 | * Remove slab from the partial list, freeze it and |
1605 | * return the pointer to the freelist. | |
81819f0f | 1606 | * |
497b66f2 | 1607 | * Returns a list of objects or NULL if it fails. |
81819f0f | 1608 | */ |
497b66f2 | 1609 | static inline void *acquire_slab(struct kmem_cache *s, |
acd19fd1 | 1610 | struct kmem_cache_node *n, struct page *page, |
633b0764 | 1611 | int mode, int *objects) |
81819f0f | 1612 | { |
2cfb7455 CL |
1613 | void *freelist; |
1614 | unsigned long counters; | |
1615 | struct page new; | |
1616 | ||
c65c1877 PZ |
1617 | lockdep_assert_held(&n->list_lock); |
1618 | ||
2cfb7455 CL |
1619 | /* |
1620 | * Zap the freelist and set the frozen bit. | |
1621 | * The old freelist is the list of objects for the | |
1622 | * per cpu allocation list. | |
1623 | */ | |
7ced3719 CL |
1624 | freelist = page->freelist; |
1625 | counters = page->counters; | |
1626 | new.counters = counters; | |
633b0764 | 1627 | *objects = new.objects - new.inuse; |
23910c50 | 1628 | if (mode) { |
7ced3719 | 1629 | new.inuse = page->objects; |
23910c50 PE |
1630 | new.freelist = NULL; |
1631 | } else { | |
1632 | new.freelist = freelist; | |
1633 | } | |
2cfb7455 | 1634 | |
a0132ac0 | 1635 | VM_BUG_ON(new.frozen); |
7ced3719 | 1636 | new.frozen = 1; |
2cfb7455 | 1637 | |
7ced3719 | 1638 | if (!__cmpxchg_double_slab(s, page, |
2cfb7455 | 1639 | freelist, counters, |
02d7633f | 1640 | new.freelist, new.counters, |
7ced3719 | 1641 | "acquire_slab")) |
7ced3719 | 1642 | return NULL; |
2cfb7455 CL |
1643 | |
1644 | remove_partial(n, page); | |
7ced3719 | 1645 | WARN_ON(!freelist); |
49e22585 | 1646 | return freelist; |
81819f0f CL |
1647 | } |
1648 | ||
633b0764 | 1649 | static void put_cpu_partial(struct kmem_cache *s, struct page *page, int drain); |
8ba00bb6 | 1650 | static inline bool pfmemalloc_match(struct page *page, gfp_t gfpflags); |
49e22585 | 1651 | |
81819f0f | 1652 | /* |
672bba3a | 1653 | * Try to allocate a partial slab from a specific node. |
81819f0f | 1654 | */ |
8ba00bb6 JK |
1655 | static void *get_partial_node(struct kmem_cache *s, struct kmem_cache_node *n, |
1656 | struct kmem_cache_cpu *c, gfp_t flags) | |
81819f0f | 1657 | { |
49e22585 CL |
1658 | struct page *page, *page2; |
1659 | void *object = NULL; | |
633b0764 JK |
1660 | int available = 0; |
1661 | int objects; | |
81819f0f CL |
1662 | |
1663 | /* | |
1664 | * Racy check. If we mistakenly see no partial slabs then we | |
1665 | * just allocate an empty slab. If we mistakenly try to get a | |
672bba3a CL |
1666 | * partial slab and there is none available then get_partials() |
1667 | * will return NULL. | |
81819f0f CL |
1668 | */ |
1669 | if (!n || !n->nr_partial) | |
1670 | return NULL; | |
1671 | ||
1672 | spin_lock(&n->list_lock); | |
49e22585 | 1673 | list_for_each_entry_safe(page, page2, &n->partial, lru) { |
8ba00bb6 | 1674 | void *t; |
49e22585 | 1675 | |
8ba00bb6 JK |
1676 | if (!pfmemalloc_match(page, flags)) |
1677 | continue; | |
1678 | ||
633b0764 | 1679 | t = acquire_slab(s, n, page, object == NULL, &objects); |
49e22585 CL |
1680 | if (!t) |
1681 | break; | |
1682 | ||
633b0764 | 1683 | available += objects; |
12d79634 | 1684 | if (!object) { |
49e22585 | 1685 | c->page = page; |
49e22585 | 1686 | stat(s, ALLOC_FROM_PARTIAL); |
49e22585 | 1687 | object = t; |
49e22585 | 1688 | } else { |
633b0764 | 1689 | put_cpu_partial(s, page, 0); |
8028dcea | 1690 | stat(s, CPU_PARTIAL_NODE); |
49e22585 | 1691 | } |
345c905d JK |
1692 | if (!kmem_cache_has_cpu_partial(s) |
1693 | || available > s->cpu_partial / 2) | |
49e22585 CL |
1694 | break; |
1695 | ||
497b66f2 | 1696 | } |
81819f0f | 1697 | spin_unlock(&n->list_lock); |
497b66f2 | 1698 | return object; |
81819f0f CL |
1699 | } |
1700 | ||
1701 | /* | |
672bba3a | 1702 | * Get a page from somewhere. Search in increasing NUMA distances. |
81819f0f | 1703 | */ |
de3ec035 | 1704 | static void *get_any_partial(struct kmem_cache *s, gfp_t flags, |
acd19fd1 | 1705 | struct kmem_cache_cpu *c) |
81819f0f CL |
1706 | { |
1707 | #ifdef CONFIG_NUMA | |
1708 | struct zonelist *zonelist; | |
dd1a239f | 1709 | struct zoneref *z; |
54a6eb5c MG |
1710 | struct zone *zone; |
1711 | enum zone_type high_zoneidx = gfp_zone(flags); | |
497b66f2 | 1712 | void *object; |
cc9a6c87 | 1713 | unsigned int cpuset_mems_cookie; |
81819f0f CL |
1714 | |
1715 | /* | |
672bba3a CL |
1716 | * The defrag ratio allows a configuration of the tradeoffs between |
1717 | * inter node defragmentation and node local allocations. A lower | |
1718 | * defrag_ratio increases the tendency to do local allocations | |
1719 | * instead of attempting to obtain partial slabs from other nodes. | |
81819f0f | 1720 | * |
672bba3a CL |
1721 | * If the defrag_ratio is set to 0 then kmalloc() always |
1722 | * returns node local objects. If the ratio is higher then kmalloc() | |
1723 | * may return off node objects because partial slabs are obtained | |
1724 | * from other nodes and filled up. | |
81819f0f | 1725 | * |
6446faa2 | 1726 | * If /sys/kernel/slab/xx/defrag_ratio is set to 100 (which makes |
672bba3a CL |
1727 | * defrag_ratio = 1000) then every (well almost) allocation will |
1728 | * first attempt to defrag slab caches on other nodes. This means | |
1729 | * scanning over all nodes to look for partial slabs which may be | |
1730 | * expensive if we do it every time we are trying to find a slab | |
1731 | * with available objects. | |
81819f0f | 1732 | */ |
9824601e CL |
1733 | if (!s->remote_node_defrag_ratio || |
1734 | get_cycles() % 1024 > s->remote_node_defrag_ratio) | |
81819f0f CL |
1735 | return NULL; |
1736 | ||
cc9a6c87 | 1737 | do { |
d26914d1 | 1738 | cpuset_mems_cookie = read_mems_allowed_begin(); |
2a389610 | 1739 | zonelist = node_zonelist(mempolicy_slab_node(), flags); |
cc9a6c87 MG |
1740 | for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) { |
1741 | struct kmem_cache_node *n; | |
1742 | ||
1743 | n = get_node(s, zone_to_nid(zone)); | |
1744 | ||
dee2f8aa | 1745 | if (n && cpuset_zone_allowed(zone, flags) && |
cc9a6c87 | 1746 | n->nr_partial > s->min_partial) { |
8ba00bb6 | 1747 | object = get_partial_node(s, n, c, flags); |
cc9a6c87 MG |
1748 | if (object) { |
1749 | /* | |
d26914d1 MG |
1750 | * Don't check read_mems_allowed_retry() |
1751 | * here - if mems_allowed was updated in | |
1752 | * parallel, that was a harmless race | |
1753 | * between allocation and the cpuset | |
1754 | * update | |
cc9a6c87 | 1755 | */ |
cc9a6c87 MG |
1756 | return object; |
1757 | } | |
c0ff7453 | 1758 | } |
81819f0f | 1759 | } |
d26914d1 | 1760 | } while (read_mems_allowed_retry(cpuset_mems_cookie)); |
81819f0f CL |
1761 | #endif |
1762 | return NULL; | |
1763 | } | |
1764 | ||
1765 | /* | |
1766 | * Get a partial page, lock it and return it. | |
1767 | */ | |
497b66f2 | 1768 | static void *get_partial(struct kmem_cache *s, gfp_t flags, int node, |
acd19fd1 | 1769 | struct kmem_cache_cpu *c) |
81819f0f | 1770 | { |
497b66f2 | 1771 | void *object; |
a561ce00 JK |
1772 | int searchnode = node; |
1773 | ||
1774 | if (node == NUMA_NO_NODE) | |
1775 | searchnode = numa_mem_id(); | |
1776 | else if (!node_present_pages(node)) | |
1777 | searchnode = node_to_mem_node(node); | |
81819f0f | 1778 | |
8ba00bb6 | 1779 | object = get_partial_node(s, get_node(s, searchnode), c, flags); |
497b66f2 CL |
1780 | if (object || node != NUMA_NO_NODE) |
1781 | return object; | |
81819f0f | 1782 | |
acd19fd1 | 1783 | return get_any_partial(s, flags, c); |
81819f0f CL |
1784 | } |
1785 | ||
8a5ec0ba CL |
1786 | #ifdef CONFIG_PREEMPT |
1787 | /* | |
1788 | * Calculate the next globally unique transaction for disambiguiation | |
1789 | * during cmpxchg. The transactions start with the cpu number and are then | |
1790 | * incremented by CONFIG_NR_CPUS. | |
1791 | */ | |
1792 | #define TID_STEP roundup_pow_of_two(CONFIG_NR_CPUS) | |
1793 | #else | |
1794 | /* | |
1795 | * No preemption supported therefore also no need to check for | |
1796 | * different cpus. | |
1797 | */ | |
1798 | #define TID_STEP 1 | |
1799 | #endif | |
1800 | ||
1801 | static inline unsigned long next_tid(unsigned long tid) | |
1802 | { | |
1803 | return tid + TID_STEP; | |
1804 | } | |
1805 | ||
1806 | static inline unsigned int tid_to_cpu(unsigned long tid) | |
1807 | { | |
1808 | return tid % TID_STEP; | |
1809 | } | |
1810 | ||
1811 | static inline unsigned long tid_to_event(unsigned long tid) | |
1812 | { | |
1813 | return tid / TID_STEP; | |
1814 | } | |
1815 | ||
1816 | static inline unsigned int init_tid(int cpu) | |
1817 | { | |
1818 | return cpu; | |
1819 | } | |
1820 | ||
1821 | static inline void note_cmpxchg_failure(const char *n, | |
1822 | const struct kmem_cache *s, unsigned long tid) | |
1823 | { | |
1824 | #ifdef SLUB_DEBUG_CMPXCHG | |
1825 | unsigned long actual_tid = __this_cpu_read(s->cpu_slab->tid); | |
1826 | ||
f9f58285 | 1827 | pr_info("%s %s: cmpxchg redo ", n, s->name); |
8a5ec0ba CL |
1828 | |
1829 | #ifdef CONFIG_PREEMPT | |
1830 | if (tid_to_cpu(tid) != tid_to_cpu(actual_tid)) | |
f9f58285 | 1831 | pr_warn("due to cpu change %d -> %d\n", |
8a5ec0ba CL |
1832 | tid_to_cpu(tid), tid_to_cpu(actual_tid)); |
1833 | else | |
1834 | #endif | |
1835 | if (tid_to_event(tid) != tid_to_event(actual_tid)) | |
f9f58285 | 1836 | pr_warn("due to cpu running other code. Event %ld->%ld\n", |
8a5ec0ba CL |
1837 | tid_to_event(tid), tid_to_event(actual_tid)); |
1838 | else | |
f9f58285 | 1839 | pr_warn("for unknown reason: actual=%lx was=%lx target=%lx\n", |
8a5ec0ba CL |
1840 | actual_tid, tid, next_tid(tid)); |
1841 | #endif | |
4fdccdfb | 1842 | stat(s, CMPXCHG_DOUBLE_CPU_FAIL); |
8a5ec0ba CL |
1843 | } |
1844 | ||
788e1aad | 1845 | static void init_kmem_cache_cpus(struct kmem_cache *s) |
8a5ec0ba | 1846 | { |
8a5ec0ba CL |
1847 | int cpu; |
1848 | ||
1849 | for_each_possible_cpu(cpu) | |
1850 | per_cpu_ptr(s->cpu_slab, cpu)->tid = init_tid(cpu); | |
8a5ec0ba | 1851 | } |
2cfb7455 | 1852 | |
81819f0f CL |
1853 | /* |
1854 | * Remove the cpu slab | |
1855 | */ | |
d0e0ac97 CG |
1856 | static void deactivate_slab(struct kmem_cache *s, struct page *page, |
1857 | void *freelist) | |
81819f0f | 1858 | { |
2cfb7455 | 1859 | enum slab_modes { M_NONE, M_PARTIAL, M_FULL, M_FREE }; |
2cfb7455 CL |
1860 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); |
1861 | int lock = 0; | |
1862 | enum slab_modes l = M_NONE, m = M_NONE; | |
2cfb7455 | 1863 | void *nextfree; |
136333d1 | 1864 | int tail = DEACTIVATE_TO_HEAD; |
2cfb7455 CL |
1865 | struct page new; |
1866 | struct page old; | |
1867 | ||
1868 | if (page->freelist) { | |
84e554e6 | 1869 | stat(s, DEACTIVATE_REMOTE_FREES); |
136333d1 | 1870 | tail = DEACTIVATE_TO_TAIL; |
2cfb7455 CL |
1871 | } |
1872 | ||
894b8788 | 1873 | /* |
2cfb7455 CL |
1874 | * Stage one: Free all available per cpu objects back |
1875 | * to the page freelist while it is still frozen. Leave the | |
1876 | * last one. | |
1877 | * | |
1878 | * There is no need to take the list->lock because the page | |
1879 | * is still frozen. | |
1880 | */ | |
1881 | while (freelist && (nextfree = get_freepointer(s, freelist))) { | |
1882 | void *prior; | |
1883 | unsigned long counters; | |
1884 | ||
1885 | do { | |
1886 | prior = page->freelist; | |
1887 | counters = page->counters; | |
1888 | set_freepointer(s, freelist, prior); | |
1889 | new.counters = counters; | |
1890 | new.inuse--; | |
a0132ac0 | 1891 | VM_BUG_ON(!new.frozen); |
2cfb7455 | 1892 | |
1d07171c | 1893 | } while (!__cmpxchg_double_slab(s, page, |
2cfb7455 CL |
1894 | prior, counters, |
1895 | freelist, new.counters, | |
1896 | "drain percpu freelist")); | |
1897 | ||
1898 | freelist = nextfree; | |
1899 | } | |
1900 | ||
894b8788 | 1901 | /* |
2cfb7455 CL |
1902 | * Stage two: Ensure that the page is unfrozen while the |
1903 | * list presence reflects the actual number of objects | |
1904 | * during unfreeze. | |
1905 | * | |
1906 | * We setup the list membership and then perform a cmpxchg | |
1907 | * with the count. If there is a mismatch then the page | |
1908 | * is not unfrozen but the page is on the wrong list. | |
1909 | * | |
1910 | * Then we restart the process which may have to remove | |
1911 | * the page from the list that we just put it on again | |
1912 | * because the number of objects in the slab may have | |
1913 | * changed. | |
894b8788 | 1914 | */ |
2cfb7455 | 1915 | redo: |
894b8788 | 1916 | |
2cfb7455 CL |
1917 | old.freelist = page->freelist; |
1918 | old.counters = page->counters; | |
a0132ac0 | 1919 | VM_BUG_ON(!old.frozen); |
7c2e132c | 1920 | |
2cfb7455 CL |
1921 | /* Determine target state of the slab */ |
1922 | new.counters = old.counters; | |
1923 | if (freelist) { | |
1924 | new.inuse--; | |
1925 | set_freepointer(s, freelist, old.freelist); | |
1926 | new.freelist = freelist; | |
1927 | } else | |
1928 | new.freelist = old.freelist; | |
1929 | ||
1930 | new.frozen = 0; | |
1931 | ||
8a5b20ae | 1932 | if (!new.inuse && n->nr_partial >= s->min_partial) |
2cfb7455 CL |
1933 | m = M_FREE; |
1934 | else if (new.freelist) { | |
1935 | m = M_PARTIAL; | |
1936 | if (!lock) { | |
1937 | lock = 1; | |
1938 | /* | |
1939 | * Taking the spinlock removes the possiblity | |
1940 | * that acquire_slab() will see a slab page that | |
1941 | * is frozen | |
1942 | */ | |
1943 | spin_lock(&n->list_lock); | |
1944 | } | |
1945 | } else { | |
1946 | m = M_FULL; | |
1947 | if (kmem_cache_debug(s) && !lock) { | |
1948 | lock = 1; | |
1949 | /* | |
1950 | * This also ensures that the scanning of full | |
1951 | * slabs from diagnostic functions will not see | |
1952 | * any frozen slabs. | |
1953 | */ | |
1954 | spin_lock(&n->list_lock); | |
1955 | } | |
1956 | } | |
1957 | ||
1958 | if (l != m) { | |
1959 | ||
1960 | if (l == M_PARTIAL) | |
1961 | ||
1962 | remove_partial(n, page); | |
1963 | ||
1964 | else if (l == M_FULL) | |
894b8788 | 1965 | |
c65c1877 | 1966 | remove_full(s, n, page); |
2cfb7455 CL |
1967 | |
1968 | if (m == M_PARTIAL) { | |
1969 | ||
1970 | add_partial(n, page, tail); | |
136333d1 | 1971 | stat(s, tail); |
2cfb7455 CL |
1972 | |
1973 | } else if (m == M_FULL) { | |
894b8788 | 1974 | |
2cfb7455 CL |
1975 | stat(s, DEACTIVATE_FULL); |
1976 | add_full(s, n, page); | |
1977 | ||
1978 | } | |
1979 | } | |
1980 | ||
1981 | l = m; | |
1d07171c | 1982 | if (!__cmpxchg_double_slab(s, page, |
2cfb7455 CL |
1983 | old.freelist, old.counters, |
1984 | new.freelist, new.counters, | |
1985 | "unfreezing slab")) | |
1986 | goto redo; | |
1987 | ||
2cfb7455 CL |
1988 | if (lock) |
1989 | spin_unlock(&n->list_lock); | |
1990 | ||
1991 | if (m == M_FREE) { | |
1992 | stat(s, DEACTIVATE_EMPTY); | |
1993 | discard_slab(s, page); | |
1994 | stat(s, FREE_SLAB); | |
894b8788 | 1995 | } |
81819f0f CL |
1996 | } |
1997 | ||
d24ac77f JK |
1998 | /* |
1999 | * Unfreeze all the cpu partial slabs. | |
2000 | * | |
59a09917 CL |
2001 | * This function must be called with interrupts disabled |
2002 | * for the cpu using c (or some other guarantee must be there | |
2003 | * to guarantee no concurrent accesses). | |
d24ac77f | 2004 | */ |
59a09917 CL |
2005 | static void unfreeze_partials(struct kmem_cache *s, |
2006 | struct kmem_cache_cpu *c) | |
49e22585 | 2007 | { |
345c905d | 2008 | #ifdef CONFIG_SLUB_CPU_PARTIAL |
43d77867 | 2009 | struct kmem_cache_node *n = NULL, *n2 = NULL; |
9ada1934 | 2010 | struct page *page, *discard_page = NULL; |
49e22585 CL |
2011 | |
2012 | while ((page = c->partial)) { | |
49e22585 CL |
2013 | struct page new; |
2014 | struct page old; | |
2015 | ||
2016 | c->partial = page->next; | |
43d77867 JK |
2017 | |
2018 | n2 = get_node(s, page_to_nid(page)); | |
2019 | if (n != n2) { | |
2020 | if (n) | |
2021 | spin_unlock(&n->list_lock); | |
2022 | ||
2023 | n = n2; | |
2024 | spin_lock(&n->list_lock); | |
2025 | } | |
49e22585 CL |
2026 | |
2027 | do { | |
2028 | ||
2029 | old.freelist = page->freelist; | |
2030 | old.counters = page->counters; | |
a0132ac0 | 2031 | VM_BUG_ON(!old.frozen); |
49e22585 CL |
2032 | |
2033 | new.counters = old.counters; | |
2034 | new.freelist = old.freelist; | |
2035 | ||
2036 | new.frozen = 0; | |
2037 | ||
d24ac77f | 2038 | } while (!__cmpxchg_double_slab(s, page, |
49e22585 CL |
2039 | old.freelist, old.counters, |
2040 | new.freelist, new.counters, | |
2041 | "unfreezing slab")); | |
2042 | ||
8a5b20ae | 2043 | if (unlikely(!new.inuse && n->nr_partial >= s->min_partial)) { |
9ada1934 SL |
2044 | page->next = discard_page; |
2045 | discard_page = page; | |
43d77867 JK |
2046 | } else { |
2047 | add_partial(n, page, DEACTIVATE_TO_TAIL); | |
2048 | stat(s, FREE_ADD_PARTIAL); | |
49e22585 CL |
2049 | } |
2050 | } | |
2051 | ||
2052 | if (n) | |
2053 | spin_unlock(&n->list_lock); | |
9ada1934 SL |
2054 | |
2055 | while (discard_page) { | |
2056 | page = discard_page; | |
2057 | discard_page = discard_page->next; | |
2058 | ||
2059 | stat(s, DEACTIVATE_EMPTY); | |
2060 | discard_slab(s, page); | |
2061 | stat(s, FREE_SLAB); | |
2062 | } | |
345c905d | 2063 | #endif |
49e22585 CL |
2064 | } |
2065 | ||
2066 | /* | |
2067 | * Put a page that was just frozen (in __slab_free) into a partial page | |
2068 | * slot if available. This is done without interrupts disabled and without | |
2069 | * preemption disabled. The cmpxchg is racy and may put the partial page | |
2070 | * onto a random cpus partial slot. | |
2071 | * | |
2072 | * If we did not find a slot then simply move all the partials to the | |
2073 | * per node partial list. | |
2074 | */ | |
633b0764 | 2075 | static void put_cpu_partial(struct kmem_cache *s, struct page *page, int drain) |
49e22585 | 2076 | { |
345c905d | 2077 | #ifdef CONFIG_SLUB_CPU_PARTIAL |
49e22585 CL |
2078 | struct page *oldpage; |
2079 | int pages; | |
2080 | int pobjects; | |
2081 | ||
d6e0b7fa | 2082 | preempt_disable(); |
49e22585 CL |
2083 | do { |
2084 | pages = 0; | |
2085 | pobjects = 0; | |
2086 | oldpage = this_cpu_read(s->cpu_slab->partial); | |
2087 | ||
2088 | if (oldpage) { | |
2089 | pobjects = oldpage->pobjects; | |
2090 | pages = oldpage->pages; | |
2091 | if (drain && pobjects > s->cpu_partial) { | |
2092 | unsigned long flags; | |
2093 | /* | |
2094 | * partial array is full. Move the existing | |
2095 | * set to the per node partial list. | |
2096 | */ | |
2097 | local_irq_save(flags); | |
59a09917 | 2098 | unfreeze_partials(s, this_cpu_ptr(s->cpu_slab)); |
49e22585 | 2099 | local_irq_restore(flags); |
e24fc410 | 2100 | oldpage = NULL; |
49e22585 CL |
2101 | pobjects = 0; |
2102 | pages = 0; | |
8028dcea | 2103 | stat(s, CPU_PARTIAL_DRAIN); |
49e22585 CL |
2104 | } |
2105 | } | |
2106 | ||
2107 | pages++; | |
2108 | pobjects += page->objects - page->inuse; | |
2109 | ||
2110 | page->pages = pages; | |
2111 | page->pobjects = pobjects; | |
2112 | page->next = oldpage; | |
2113 | ||
d0e0ac97 CG |
2114 | } while (this_cpu_cmpxchg(s->cpu_slab->partial, oldpage, page) |
2115 | != oldpage); | |
d6e0b7fa VD |
2116 | if (unlikely(!s->cpu_partial)) { |
2117 | unsigned long flags; | |
2118 | ||
2119 | local_irq_save(flags); | |
2120 | unfreeze_partials(s, this_cpu_ptr(s->cpu_slab)); | |
2121 | local_irq_restore(flags); | |
2122 | } | |
2123 | preempt_enable(); | |
345c905d | 2124 | #endif |
49e22585 CL |
2125 | } |
2126 | ||
dfb4f096 | 2127 | static inline void flush_slab(struct kmem_cache *s, struct kmem_cache_cpu *c) |
81819f0f | 2128 | { |
84e554e6 | 2129 | stat(s, CPUSLAB_FLUSH); |
c17dda40 CL |
2130 | deactivate_slab(s, c->page, c->freelist); |
2131 | ||
2132 | c->tid = next_tid(c->tid); | |
2133 | c->page = NULL; | |
2134 | c->freelist = NULL; | |
81819f0f CL |
2135 | } |
2136 | ||
2137 | /* | |
2138 | * Flush cpu slab. | |
6446faa2 | 2139 | * |
81819f0f CL |
2140 | * Called from IPI handler with interrupts disabled. |
2141 | */ | |
0c710013 | 2142 | static inline void __flush_cpu_slab(struct kmem_cache *s, int cpu) |
81819f0f | 2143 | { |
9dfc6e68 | 2144 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu); |
81819f0f | 2145 | |
49e22585 CL |
2146 | if (likely(c)) { |
2147 | if (c->page) | |
2148 | flush_slab(s, c); | |
2149 | ||
59a09917 | 2150 | unfreeze_partials(s, c); |
49e22585 | 2151 | } |
81819f0f CL |
2152 | } |
2153 | ||
2154 | static void flush_cpu_slab(void *d) | |
2155 | { | |
2156 | struct kmem_cache *s = d; | |
81819f0f | 2157 | |
dfb4f096 | 2158 | __flush_cpu_slab(s, smp_processor_id()); |
81819f0f CL |
2159 | } |
2160 | ||
a8364d55 GBY |
2161 | static bool has_cpu_slab(int cpu, void *info) |
2162 | { | |
2163 | struct kmem_cache *s = info; | |
2164 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu); | |
2165 | ||
02e1a9cd | 2166 | return c->page || c->partial; |
a8364d55 GBY |
2167 | } |
2168 | ||
81819f0f CL |
2169 | static void flush_all(struct kmem_cache *s) |
2170 | { | |
a8364d55 | 2171 | on_each_cpu_cond(has_cpu_slab, flush_cpu_slab, s, 1, GFP_ATOMIC); |
81819f0f CL |
2172 | } |
2173 | ||
dfb4f096 CL |
2174 | /* |
2175 | * Check if the objects in a per cpu structure fit numa | |
2176 | * locality expectations. | |
2177 | */ | |
57d437d2 | 2178 | static inline int node_match(struct page *page, int node) |
dfb4f096 CL |
2179 | { |
2180 | #ifdef CONFIG_NUMA | |
4d7868e6 | 2181 | if (!page || (node != NUMA_NO_NODE && page_to_nid(page) != node)) |
dfb4f096 CL |
2182 | return 0; |
2183 | #endif | |
2184 | return 1; | |
2185 | } | |
2186 | ||
9a02d699 | 2187 | #ifdef CONFIG_SLUB_DEBUG |
781b2ba6 PE |
2188 | static int count_free(struct page *page) |
2189 | { | |
2190 | return page->objects - page->inuse; | |
2191 | } | |
2192 | ||
9a02d699 DR |
2193 | static inline unsigned long node_nr_objs(struct kmem_cache_node *n) |
2194 | { | |
2195 | return atomic_long_read(&n->total_objects); | |
2196 | } | |
2197 | #endif /* CONFIG_SLUB_DEBUG */ | |
2198 | ||
2199 | #if defined(CONFIG_SLUB_DEBUG) || defined(CONFIG_SYSFS) | |
781b2ba6 PE |
2200 | static unsigned long count_partial(struct kmem_cache_node *n, |
2201 | int (*get_count)(struct page *)) | |
2202 | { | |
2203 | unsigned long flags; | |
2204 | unsigned long x = 0; | |
2205 | struct page *page; | |
2206 | ||
2207 | spin_lock_irqsave(&n->list_lock, flags); | |
2208 | list_for_each_entry(page, &n->partial, lru) | |
2209 | x += get_count(page); | |
2210 | spin_unlock_irqrestore(&n->list_lock, flags); | |
2211 | return x; | |
2212 | } | |
9a02d699 | 2213 | #endif /* CONFIG_SLUB_DEBUG || CONFIG_SYSFS */ |
26c02cf0 | 2214 | |
781b2ba6 PE |
2215 | static noinline void |
2216 | slab_out_of_memory(struct kmem_cache *s, gfp_t gfpflags, int nid) | |
2217 | { | |
9a02d699 DR |
2218 | #ifdef CONFIG_SLUB_DEBUG |
2219 | static DEFINE_RATELIMIT_STATE(slub_oom_rs, DEFAULT_RATELIMIT_INTERVAL, | |
2220 | DEFAULT_RATELIMIT_BURST); | |
781b2ba6 | 2221 | int node; |
fa45dc25 | 2222 | struct kmem_cache_node *n; |
781b2ba6 | 2223 | |
9a02d699 DR |
2224 | if ((gfpflags & __GFP_NOWARN) || !__ratelimit(&slub_oom_rs)) |
2225 | return; | |
2226 | ||
f9f58285 | 2227 | pr_warn("SLUB: Unable to allocate memory on node %d (gfp=0x%x)\n", |
781b2ba6 | 2228 | nid, gfpflags); |
f9f58285 FF |
2229 | pr_warn(" cache: %s, object size: %d, buffer size: %d, default order: %d, min order: %d\n", |
2230 | s->name, s->object_size, s->size, oo_order(s->oo), | |
2231 | oo_order(s->min)); | |
781b2ba6 | 2232 | |
3b0efdfa | 2233 | if (oo_order(s->min) > get_order(s->object_size)) |
f9f58285 FF |
2234 | pr_warn(" %s debugging increased min order, use slub_debug=O to disable.\n", |
2235 | s->name); | |
fa5ec8a1 | 2236 | |
fa45dc25 | 2237 | for_each_kmem_cache_node(s, node, n) { |
781b2ba6 PE |
2238 | unsigned long nr_slabs; |
2239 | unsigned long nr_objs; | |
2240 | unsigned long nr_free; | |
2241 | ||
26c02cf0 AB |
2242 | nr_free = count_partial(n, count_free); |
2243 | nr_slabs = node_nr_slabs(n); | |
2244 | nr_objs = node_nr_objs(n); | |
781b2ba6 | 2245 | |
f9f58285 | 2246 | pr_warn(" node %d: slabs: %ld, objs: %ld, free: %ld\n", |
781b2ba6 PE |
2247 | node, nr_slabs, nr_objs, nr_free); |
2248 | } | |
9a02d699 | 2249 | #endif |
781b2ba6 PE |
2250 | } |
2251 | ||
497b66f2 CL |
2252 | static inline void *new_slab_objects(struct kmem_cache *s, gfp_t flags, |
2253 | int node, struct kmem_cache_cpu **pc) | |
2254 | { | |
6faa6833 | 2255 | void *freelist; |
188fd063 CL |
2256 | struct kmem_cache_cpu *c = *pc; |
2257 | struct page *page; | |
497b66f2 | 2258 | |
188fd063 | 2259 | freelist = get_partial(s, flags, node, c); |
497b66f2 | 2260 | |
188fd063 CL |
2261 | if (freelist) |
2262 | return freelist; | |
2263 | ||
2264 | page = new_slab(s, flags, node); | |
497b66f2 | 2265 | if (page) { |
7c8e0181 | 2266 | c = raw_cpu_ptr(s->cpu_slab); |
497b66f2 CL |
2267 | if (c->page) |
2268 | flush_slab(s, c); | |
2269 | ||
2270 | /* | |
2271 | * No other reference to the page yet so we can | |
2272 | * muck around with it freely without cmpxchg | |
2273 | */ | |
6faa6833 | 2274 | freelist = page->freelist; |
497b66f2 CL |
2275 | page->freelist = NULL; |
2276 | ||
2277 | stat(s, ALLOC_SLAB); | |
497b66f2 CL |
2278 | c->page = page; |
2279 | *pc = c; | |
2280 | } else | |
6faa6833 | 2281 | freelist = NULL; |
497b66f2 | 2282 | |
6faa6833 | 2283 | return freelist; |
497b66f2 CL |
2284 | } |
2285 | ||
072bb0aa MG |
2286 | static inline bool pfmemalloc_match(struct page *page, gfp_t gfpflags) |
2287 | { | |
2288 | if (unlikely(PageSlabPfmemalloc(page))) | |
2289 | return gfp_pfmemalloc_allowed(gfpflags); | |
2290 | ||
2291 | return true; | |
2292 | } | |
2293 | ||
213eeb9f | 2294 | /* |
d0e0ac97 CG |
2295 | * Check the page->freelist of a page and either transfer the freelist to the |
2296 | * per cpu freelist or deactivate the page. | |
213eeb9f CL |
2297 | * |
2298 | * The page is still frozen if the return value is not NULL. | |
2299 | * | |
2300 | * If this function returns NULL then the page has been unfrozen. | |
d24ac77f JK |
2301 | * |
2302 | * This function must be called with interrupt disabled. | |
213eeb9f CL |
2303 | */ |
2304 | static inline void *get_freelist(struct kmem_cache *s, struct page *page) | |
2305 | { | |
2306 | struct page new; | |
2307 | unsigned long counters; | |
2308 | void *freelist; | |
2309 | ||
2310 | do { | |
2311 | freelist = page->freelist; | |
2312 | counters = page->counters; | |
6faa6833 | 2313 | |
213eeb9f | 2314 | new.counters = counters; |
a0132ac0 | 2315 | VM_BUG_ON(!new.frozen); |
213eeb9f CL |
2316 | |
2317 | new.inuse = page->objects; | |
2318 | new.frozen = freelist != NULL; | |
2319 | ||
d24ac77f | 2320 | } while (!__cmpxchg_double_slab(s, page, |
213eeb9f CL |
2321 | freelist, counters, |
2322 | NULL, new.counters, | |
2323 | "get_freelist")); | |
2324 | ||
2325 | return freelist; | |
2326 | } | |
2327 | ||
81819f0f | 2328 | /* |
894b8788 CL |
2329 | * Slow path. The lockless freelist is empty or we need to perform |
2330 | * debugging duties. | |
2331 | * | |
894b8788 CL |
2332 | * Processing is still very fast if new objects have been freed to the |
2333 | * regular freelist. In that case we simply take over the regular freelist | |
2334 | * as the lockless freelist and zap the regular freelist. | |
81819f0f | 2335 | * |
894b8788 CL |
2336 | * If that is not working then we fall back to the partial lists. We take the |
2337 | * first element of the freelist as the object to allocate now and move the | |
2338 | * rest of the freelist to the lockless freelist. | |
81819f0f | 2339 | * |
894b8788 | 2340 | * And if we were unable to get a new slab from the partial slab lists then |
6446faa2 CL |
2341 | * we need to allocate a new slab. This is the slowest path since it involves |
2342 | * a call to the page allocator and the setup of a new slab. | |
a380a3c7 CL |
2343 | * |
2344 | * Version of __slab_alloc to use when we know that interrupts are | |
2345 | * already disabled (which is the case for bulk allocation). | |
81819f0f | 2346 | */ |
a380a3c7 | 2347 | static void *___slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node, |
ce71e27c | 2348 | unsigned long addr, struct kmem_cache_cpu *c) |
81819f0f | 2349 | { |
6faa6833 | 2350 | void *freelist; |
f6e7def7 | 2351 | struct page *page; |
81819f0f | 2352 | |
f6e7def7 CL |
2353 | page = c->page; |
2354 | if (!page) | |
81819f0f | 2355 | goto new_slab; |
49e22585 | 2356 | redo: |
6faa6833 | 2357 | |
57d437d2 | 2358 | if (unlikely(!node_match(page, node))) { |
a561ce00 JK |
2359 | int searchnode = node; |
2360 | ||
2361 | if (node != NUMA_NO_NODE && !node_present_pages(node)) | |
2362 | searchnode = node_to_mem_node(node); | |
2363 | ||
2364 | if (unlikely(!node_match(page, searchnode))) { | |
2365 | stat(s, ALLOC_NODE_MISMATCH); | |
2366 | deactivate_slab(s, page, c->freelist); | |
2367 | c->page = NULL; | |
2368 | c->freelist = NULL; | |
2369 | goto new_slab; | |
2370 | } | |
fc59c053 | 2371 | } |
6446faa2 | 2372 | |
072bb0aa MG |
2373 | /* |
2374 | * By rights, we should be searching for a slab page that was | |
2375 | * PFMEMALLOC but right now, we are losing the pfmemalloc | |
2376 | * information when the page leaves the per-cpu allocator | |
2377 | */ | |
2378 | if (unlikely(!pfmemalloc_match(page, gfpflags))) { | |
2379 | deactivate_slab(s, page, c->freelist); | |
2380 | c->page = NULL; | |
2381 | c->freelist = NULL; | |
2382 | goto new_slab; | |
2383 | } | |
2384 | ||
73736e03 | 2385 | /* must check again c->freelist in case of cpu migration or IRQ */ |
6faa6833 CL |
2386 | freelist = c->freelist; |
2387 | if (freelist) | |
73736e03 | 2388 | goto load_freelist; |
03e404af | 2389 | |
f6e7def7 | 2390 | freelist = get_freelist(s, page); |
6446faa2 | 2391 | |
6faa6833 | 2392 | if (!freelist) { |
03e404af CL |
2393 | c->page = NULL; |
2394 | stat(s, DEACTIVATE_BYPASS); | |
fc59c053 | 2395 | goto new_slab; |
03e404af | 2396 | } |
6446faa2 | 2397 | |
84e554e6 | 2398 | stat(s, ALLOC_REFILL); |
6446faa2 | 2399 | |
894b8788 | 2400 | load_freelist: |
507effea CL |
2401 | /* |
2402 | * freelist is pointing to the list of objects to be used. | |
2403 | * page is pointing to the page from which the objects are obtained. | |
2404 | * That page must be frozen for per cpu allocations to work. | |
2405 | */ | |
a0132ac0 | 2406 | VM_BUG_ON(!c->page->frozen); |
6faa6833 | 2407 | c->freelist = get_freepointer(s, freelist); |
8a5ec0ba | 2408 | c->tid = next_tid(c->tid); |
6faa6833 | 2409 | return freelist; |
81819f0f | 2410 | |
81819f0f | 2411 | new_slab: |
2cfb7455 | 2412 | |
49e22585 | 2413 | if (c->partial) { |
f6e7def7 CL |
2414 | page = c->page = c->partial; |
2415 | c->partial = page->next; | |
49e22585 CL |
2416 | stat(s, CPU_PARTIAL_ALLOC); |
2417 | c->freelist = NULL; | |
2418 | goto redo; | |
81819f0f CL |
2419 | } |
2420 | ||
188fd063 | 2421 | freelist = new_slab_objects(s, gfpflags, node, &c); |
01ad8a7b | 2422 | |
f4697436 | 2423 | if (unlikely(!freelist)) { |
9a02d699 | 2424 | slab_out_of_memory(s, gfpflags, node); |
f4697436 | 2425 | return NULL; |
81819f0f | 2426 | } |
2cfb7455 | 2427 | |
f6e7def7 | 2428 | page = c->page; |
5091b74a | 2429 | if (likely(!kmem_cache_debug(s) && pfmemalloc_match(page, gfpflags))) |
4b6f0750 | 2430 | goto load_freelist; |
2cfb7455 | 2431 | |
497b66f2 | 2432 | /* Only entered in the debug case */ |
d0e0ac97 CG |
2433 | if (kmem_cache_debug(s) && |
2434 | !alloc_debug_processing(s, page, freelist, addr)) | |
497b66f2 | 2435 | goto new_slab; /* Slab failed checks. Next slab needed */ |
894b8788 | 2436 | |
f6e7def7 | 2437 | deactivate_slab(s, page, get_freepointer(s, freelist)); |
c17dda40 CL |
2438 | c->page = NULL; |
2439 | c->freelist = NULL; | |
6faa6833 | 2440 | return freelist; |
894b8788 CL |
2441 | } |
2442 | ||
a380a3c7 CL |
2443 | /* |
2444 | * Another one that disabled interrupt and compensates for possible | |
2445 | * cpu changes by refetching the per cpu area pointer. | |
2446 | */ | |
2447 | static void *__slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node, | |
2448 | unsigned long addr, struct kmem_cache_cpu *c) | |
2449 | { | |
2450 | void *p; | |
2451 | unsigned long flags; | |
2452 | ||
2453 | local_irq_save(flags); | |
2454 | #ifdef CONFIG_PREEMPT | |
2455 | /* | |
2456 | * We may have been preempted and rescheduled on a different | |
2457 | * cpu before disabling interrupts. Need to reload cpu area | |
2458 | * pointer. | |
2459 | */ | |
2460 | c = this_cpu_ptr(s->cpu_slab); | |
2461 | #endif | |
2462 | ||
2463 | p = ___slab_alloc(s, gfpflags, node, addr, c); | |
2464 | local_irq_restore(flags); | |
2465 | return p; | |
2466 | } | |
2467 | ||
894b8788 CL |
2468 | /* |
2469 | * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc) | |
2470 | * have the fastpath folded into their functions. So no function call | |
2471 | * overhead for requests that can be satisfied on the fastpath. | |
2472 | * | |
2473 | * The fastpath works by first checking if the lockless freelist can be used. | |
2474 | * If not then __slab_alloc is called for slow processing. | |
2475 | * | |
2476 | * Otherwise we can simply pick the next object from the lockless free list. | |
2477 | */ | |
2b847c3c | 2478 | static __always_inline void *slab_alloc_node(struct kmem_cache *s, |
ce71e27c | 2479 | gfp_t gfpflags, int node, unsigned long addr) |
894b8788 | 2480 | { |
03ec0ed5 | 2481 | void *object; |
dfb4f096 | 2482 | struct kmem_cache_cpu *c; |
57d437d2 | 2483 | struct page *page; |
8a5ec0ba | 2484 | unsigned long tid; |
1f84260c | 2485 | |
8135be5a VD |
2486 | s = slab_pre_alloc_hook(s, gfpflags); |
2487 | if (!s) | |
773ff60e | 2488 | return NULL; |
8a5ec0ba | 2489 | redo: |
8a5ec0ba CL |
2490 | /* |
2491 | * Must read kmem_cache cpu data via this cpu ptr. Preemption is | |
2492 | * enabled. We may switch back and forth between cpus while | |
2493 | * reading from one cpu area. That does not matter as long | |
2494 | * as we end up on the original cpu again when doing the cmpxchg. | |
7cccd80b | 2495 | * |
9aabf810 JK |
2496 | * We should guarantee that tid and kmem_cache are retrieved on |
2497 | * the same cpu. It could be different if CONFIG_PREEMPT so we need | |
2498 | * to check if it is matched or not. | |
8a5ec0ba | 2499 | */ |
9aabf810 JK |
2500 | do { |
2501 | tid = this_cpu_read(s->cpu_slab->tid); | |
2502 | c = raw_cpu_ptr(s->cpu_slab); | |
859b7a0e MR |
2503 | } while (IS_ENABLED(CONFIG_PREEMPT) && |
2504 | unlikely(tid != READ_ONCE(c->tid))); | |
9aabf810 JK |
2505 | |
2506 | /* | |
2507 | * Irqless object alloc/free algorithm used here depends on sequence | |
2508 | * of fetching cpu_slab's data. tid should be fetched before anything | |
2509 | * on c to guarantee that object and page associated with previous tid | |
2510 | * won't be used with current tid. If we fetch tid first, object and | |
2511 | * page could be one associated with next tid and our alloc/free | |
2512 | * request will be failed. In this case, we will retry. So, no problem. | |
2513 | */ | |
2514 | barrier(); | |
8a5ec0ba | 2515 | |
8a5ec0ba CL |
2516 | /* |
2517 | * The transaction ids are globally unique per cpu and per operation on | |
2518 | * a per cpu queue. Thus they can be guarantee that the cmpxchg_double | |
2519 | * occurs on the right processor and that there was no operation on the | |
2520 | * linked list in between. | |
2521 | */ | |
8a5ec0ba | 2522 | |
9dfc6e68 | 2523 | object = c->freelist; |
57d437d2 | 2524 | page = c->page; |
8eae1492 | 2525 | if (unlikely(!object || !node_match(page, node))) { |
dfb4f096 | 2526 | object = __slab_alloc(s, gfpflags, node, addr, c); |
8eae1492 DH |
2527 | stat(s, ALLOC_SLOWPATH); |
2528 | } else { | |
0ad9500e ED |
2529 | void *next_object = get_freepointer_safe(s, object); |
2530 | ||
8a5ec0ba | 2531 | /* |
25985edc | 2532 | * The cmpxchg will only match if there was no additional |
8a5ec0ba CL |
2533 | * operation and if we are on the right processor. |
2534 | * | |
d0e0ac97 CG |
2535 | * The cmpxchg does the following atomically (without lock |
2536 | * semantics!) | |
8a5ec0ba CL |
2537 | * 1. Relocate first pointer to the current per cpu area. |
2538 | * 2. Verify that tid and freelist have not been changed | |
2539 | * 3. If they were not changed replace tid and freelist | |
2540 | * | |
d0e0ac97 CG |
2541 | * Since this is without lock semantics the protection is only |
2542 | * against code executing on this cpu *not* from access by | |
2543 | * other cpus. | |
8a5ec0ba | 2544 | */ |
933393f5 | 2545 | if (unlikely(!this_cpu_cmpxchg_double( |
8a5ec0ba CL |
2546 | s->cpu_slab->freelist, s->cpu_slab->tid, |
2547 | object, tid, | |
0ad9500e | 2548 | next_object, next_tid(tid)))) { |
8a5ec0ba CL |
2549 | |
2550 | note_cmpxchg_failure("slab_alloc", s, tid); | |
2551 | goto redo; | |
2552 | } | |
0ad9500e | 2553 | prefetch_freepointer(s, next_object); |
84e554e6 | 2554 | stat(s, ALLOC_FASTPATH); |
894b8788 | 2555 | } |
8a5ec0ba | 2556 | |
74e2134f | 2557 | if (unlikely(gfpflags & __GFP_ZERO) && object) |
3b0efdfa | 2558 | memset(object, 0, s->object_size); |
d07dbea4 | 2559 | |
03ec0ed5 | 2560 | slab_post_alloc_hook(s, gfpflags, 1, &object); |
5a896d9e | 2561 | |
894b8788 | 2562 | return object; |
81819f0f CL |
2563 | } |
2564 | ||
2b847c3c EG |
2565 | static __always_inline void *slab_alloc(struct kmem_cache *s, |
2566 | gfp_t gfpflags, unsigned long addr) | |
2567 | { | |
2568 | return slab_alloc_node(s, gfpflags, NUMA_NO_NODE, addr); | |
2569 | } | |
2570 | ||
81819f0f CL |
2571 | void *kmem_cache_alloc(struct kmem_cache *s, gfp_t gfpflags) |
2572 | { | |
2b847c3c | 2573 | void *ret = slab_alloc(s, gfpflags, _RET_IP_); |
5b882be4 | 2574 | |
d0e0ac97 CG |
2575 | trace_kmem_cache_alloc(_RET_IP_, ret, s->object_size, |
2576 | s->size, gfpflags); | |
5b882be4 EGM |
2577 | |
2578 | return ret; | |
81819f0f CL |
2579 | } |
2580 | EXPORT_SYMBOL(kmem_cache_alloc); | |
2581 | ||
0f24f128 | 2582 | #ifdef CONFIG_TRACING |
4a92379b RK |
2583 | void *kmem_cache_alloc_trace(struct kmem_cache *s, gfp_t gfpflags, size_t size) |
2584 | { | |
2b847c3c | 2585 | void *ret = slab_alloc(s, gfpflags, _RET_IP_); |
4a92379b | 2586 | trace_kmalloc(_RET_IP_, ret, size, s->size, gfpflags); |
0316bec2 | 2587 | kasan_kmalloc(s, ret, size); |
4a92379b RK |
2588 | return ret; |
2589 | } | |
2590 | EXPORT_SYMBOL(kmem_cache_alloc_trace); | |
5b882be4 EGM |
2591 | #endif |
2592 | ||
81819f0f CL |
2593 | #ifdef CONFIG_NUMA |
2594 | void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t gfpflags, int node) | |
2595 | { | |
2b847c3c | 2596 | void *ret = slab_alloc_node(s, gfpflags, node, _RET_IP_); |
5b882be4 | 2597 | |
ca2b84cb | 2598 | trace_kmem_cache_alloc_node(_RET_IP_, ret, |
3b0efdfa | 2599 | s->object_size, s->size, gfpflags, node); |
5b882be4 EGM |
2600 | |
2601 | return ret; | |
81819f0f CL |
2602 | } |
2603 | EXPORT_SYMBOL(kmem_cache_alloc_node); | |
81819f0f | 2604 | |
0f24f128 | 2605 | #ifdef CONFIG_TRACING |
4a92379b | 2606 | void *kmem_cache_alloc_node_trace(struct kmem_cache *s, |
5b882be4 | 2607 | gfp_t gfpflags, |
4a92379b | 2608 | int node, size_t size) |
5b882be4 | 2609 | { |
2b847c3c | 2610 | void *ret = slab_alloc_node(s, gfpflags, node, _RET_IP_); |
4a92379b RK |
2611 | |
2612 | trace_kmalloc_node(_RET_IP_, ret, | |
2613 | size, s->size, gfpflags, node); | |
0316bec2 AR |
2614 | |
2615 | kasan_kmalloc(s, ret, size); | |
4a92379b | 2616 | return ret; |
5b882be4 | 2617 | } |
4a92379b | 2618 | EXPORT_SYMBOL(kmem_cache_alloc_node_trace); |
5b882be4 | 2619 | #endif |
5d1f57e4 | 2620 | #endif |
5b882be4 | 2621 | |
81819f0f | 2622 | /* |
94e4d712 | 2623 | * Slow path handling. This may still be called frequently since objects |
894b8788 | 2624 | * have a longer lifetime than the cpu slabs in most processing loads. |
81819f0f | 2625 | * |
894b8788 CL |
2626 | * So we still attempt to reduce cache line usage. Just take the slab |
2627 | * lock and free the item. If there is no additional partial page | |
2628 | * handling required then we can return immediately. | |
81819f0f | 2629 | */ |
894b8788 | 2630 | static void __slab_free(struct kmem_cache *s, struct page *page, |
81084651 JDB |
2631 | void *head, void *tail, int cnt, |
2632 | unsigned long addr) | |
2633 | ||
81819f0f CL |
2634 | { |
2635 | void *prior; | |
2cfb7455 | 2636 | int was_frozen; |
2cfb7455 CL |
2637 | struct page new; |
2638 | unsigned long counters; | |
2639 | struct kmem_cache_node *n = NULL; | |
61728d1e | 2640 | unsigned long uninitialized_var(flags); |
81819f0f | 2641 | |
8a5ec0ba | 2642 | stat(s, FREE_SLOWPATH); |
81819f0f | 2643 | |
19c7ff9e | 2644 | if (kmem_cache_debug(s) && |
81084651 JDB |
2645 | !(n = free_debug_processing(s, page, head, tail, cnt, |
2646 | addr, &flags))) | |
80f08c19 | 2647 | return; |
6446faa2 | 2648 | |
2cfb7455 | 2649 | do { |
837d678d JK |
2650 | if (unlikely(n)) { |
2651 | spin_unlock_irqrestore(&n->list_lock, flags); | |
2652 | n = NULL; | |
2653 | } | |
2cfb7455 CL |
2654 | prior = page->freelist; |
2655 | counters = page->counters; | |
81084651 | 2656 | set_freepointer(s, tail, prior); |
2cfb7455 CL |
2657 | new.counters = counters; |
2658 | was_frozen = new.frozen; | |
81084651 | 2659 | new.inuse -= cnt; |
837d678d | 2660 | if ((!new.inuse || !prior) && !was_frozen) { |
49e22585 | 2661 | |
c65c1877 | 2662 | if (kmem_cache_has_cpu_partial(s) && !prior) { |
49e22585 CL |
2663 | |
2664 | /* | |
d0e0ac97 CG |
2665 | * Slab was on no list before and will be |
2666 | * partially empty | |
2667 | * We can defer the list move and instead | |
2668 | * freeze it. | |
49e22585 CL |
2669 | */ |
2670 | new.frozen = 1; | |
2671 | ||
c65c1877 | 2672 | } else { /* Needs to be taken off a list */ |
49e22585 | 2673 | |
b455def2 | 2674 | n = get_node(s, page_to_nid(page)); |
49e22585 CL |
2675 | /* |
2676 | * Speculatively acquire the list_lock. | |
2677 | * If the cmpxchg does not succeed then we may | |
2678 | * drop the list_lock without any processing. | |
2679 | * | |
2680 | * Otherwise the list_lock will synchronize with | |
2681 | * other processors updating the list of slabs. | |
2682 | */ | |
2683 | spin_lock_irqsave(&n->list_lock, flags); | |
2684 | ||
2685 | } | |
2cfb7455 | 2686 | } |
81819f0f | 2687 | |
2cfb7455 CL |
2688 | } while (!cmpxchg_double_slab(s, page, |
2689 | prior, counters, | |
81084651 | 2690 | head, new.counters, |
2cfb7455 | 2691 | "__slab_free")); |
81819f0f | 2692 | |
2cfb7455 | 2693 | if (likely(!n)) { |
49e22585 CL |
2694 | |
2695 | /* | |
2696 | * If we just froze the page then put it onto the | |
2697 | * per cpu partial list. | |
2698 | */ | |
8028dcea | 2699 | if (new.frozen && !was_frozen) { |
49e22585 | 2700 | put_cpu_partial(s, page, 1); |
8028dcea AS |
2701 | stat(s, CPU_PARTIAL_FREE); |
2702 | } | |
49e22585 | 2703 | /* |
2cfb7455 CL |
2704 | * The list lock was not taken therefore no list |
2705 | * activity can be necessary. | |
2706 | */ | |
b455def2 L |
2707 | if (was_frozen) |
2708 | stat(s, FREE_FROZEN); | |
2709 | return; | |
2710 | } | |
81819f0f | 2711 | |
8a5b20ae | 2712 | if (unlikely(!new.inuse && n->nr_partial >= s->min_partial)) |
837d678d JK |
2713 | goto slab_empty; |
2714 | ||
81819f0f | 2715 | /* |
837d678d JK |
2716 | * Objects left in the slab. If it was not on the partial list before |
2717 | * then add it. | |
81819f0f | 2718 | */ |
345c905d JK |
2719 | if (!kmem_cache_has_cpu_partial(s) && unlikely(!prior)) { |
2720 | if (kmem_cache_debug(s)) | |
c65c1877 | 2721 | remove_full(s, n, page); |
837d678d JK |
2722 | add_partial(n, page, DEACTIVATE_TO_TAIL); |
2723 | stat(s, FREE_ADD_PARTIAL); | |
8ff12cfc | 2724 | } |
80f08c19 | 2725 | spin_unlock_irqrestore(&n->list_lock, flags); |
81819f0f CL |
2726 | return; |
2727 | ||
2728 | slab_empty: | |
a973e9dd | 2729 | if (prior) { |
81819f0f | 2730 | /* |
6fbabb20 | 2731 | * Slab on the partial list. |
81819f0f | 2732 | */ |
5cc6eee8 | 2733 | remove_partial(n, page); |
84e554e6 | 2734 | stat(s, FREE_REMOVE_PARTIAL); |
c65c1877 | 2735 | } else { |
6fbabb20 | 2736 | /* Slab must be on the full list */ |
c65c1877 PZ |
2737 | remove_full(s, n, page); |
2738 | } | |
2cfb7455 | 2739 | |
80f08c19 | 2740 | spin_unlock_irqrestore(&n->list_lock, flags); |
84e554e6 | 2741 | stat(s, FREE_SLAB); |
81819f0f | 2742 | discard_slab(s, page); |
81819f0f CL |
2743 | } |
2744 | ||
894b8788 CL |
2745 | /* |
2746 | * Fastpath with forced inlining to produce a kfree and kmem_cache_free that | |
2747 | * can perform fastpath freeing without additional function calls. | |
2748 | * | |
2749 | * The fastpath is only possible if we are freeing to the current cpu slab | |
2750 | * of this processor. This typically the case if we have just allocated | |
2751 | * the item before. | |
2752 | * | |
2753 | * If fastpath is not possible then fall back to __slab_free where we deal | |
2754 | * with all sorts of special processing. | |
81084651 JDB |
2755 | * |
2756 | * Bulk free of a freelist with several objects (all pointing to the | |
2757 | * same page) possible by specifying head and tail ptr, plus objects | |
2758 | * count (cnt). Bulk free indicated by tail pointer being set. | |
894b8788 | 2759 | */ |
81084651 JDB |
2760 | static __always_inline void slab_free(struct kmem_cache *s, struct page *page, |
2761 | void *head, void *tail, int cnt, | |
2762 | unsigned long addr) | |
894b8788 | 2763 | { |
81084651 | 2764 | void *tail_obj = tail ? : head; |
dfb4f096 | 2765 | struct kmem_cache_cpu *c; |
8a5ec0ba | 2766 | unsigned long tid; |
1f84260c | 2767 | |
81084651 | 2768 | slab_free_freelist_hook(s, head, tail); |
c016b0bd | 2769 | |
8a5ec0ba CL |
2770 | redo: |
2771 | /* | |
2772 | * Determine the currently cpus per cpu slab. | |
2773 | * The cpu may change afterward. However that does not matter since | |
2774 | * data is retrieved via this pointer. If we are on the same cpu | |
2ae44005 | 2775 | * during the cmpxchg then the free will succeed. |
8a5ec0ba | 2776 | */ |
9aabf810 JK |
2777 | do { |
2778 | tid = this_cpu_read(s->cpu_slab->tid); | |
2779 | c = raw_cpu_ptr(s->cpu_slab); | |
859b7a0e MR |
2780 | } while (IS_ENABLED(CONFIG_PREEMPT) && |
2781 | unlikely(tid != READ_ONCE(c->tid))); | |
c016b0bd | 2782 | |
9aabf810 JK |
2783 | /* Same with comment on barrier() in slab_alloc_node() */ |
2784 | barrier(); | |
c016b0bd | 2785 | |
442b06bc | 2786 | if (likely(page == c->page)) { |
81084651 | 2787 | set_freepointer(s, tail_obj, c->freelist); |
8a5ec0ba | 2788 | |
933393f5 | 2789 | if (unlikely(!this_cpu_cmpxchg_double( |
8a5ec0ba CL |
2790 | s->cpu_slab->freelist, s->cpu_slab->tid, |
2791 | c->freelist, tid, | |
81084651 | 2792 | head, next_tid(tid)))) { |
8a5ec0ba CL |
2793 | |
2794 | note_cmpxchg_failure("slab_free", s, tid); | |
2795 | goto redo; | |
2796 | } | |
84e554e6 | 2797 | stat(s, FREE_FASTPATH); |
894b8788 | 2798 | } else |
81084651 | 2799 | __slab_free(s, page, head, tail_obj, cnt, addr); |
894b8788 | 2800 | |
894b8788 CL |
2801 | } |
2802 | ||
81819f0f CL |
2803 | void kmem_cache_free(struct kmem_cache *s, void *x) |
2804 | { | |
b9ce5ef4 GC |
2805 | s = cache_from_obj(s, x); |
2806 | if (!s) | |
79576102 | 2807 | return; |
81084651 | 2808 | slab_free(s, virt_to_head_page(x), x, NULL, 1, _RET_IP_); |
ca2b84cb | 2809 | trace_kmem_cache_free(_RET_IP_, x); |
81819f0f CL |
2810 | } |
2811 | EXPORT_SYMBOL(kmem_cache_free); | |
2812 | ||
d0ecd894 | 2813 | struct detached_freelist { |
fbd02630 | 2814 | struct page *page; |
d0ecd894 JDB |
2815 | void *tail; |
2816 | void *freelist; | |
2817 | int cnt; | |
2818 | }; | |
fbd02630 | 2819 | |
d0ecd894 JDB |
2820 | /* |
2821 | * This function progressively scans the array with free objects (with | |
2822 | * a limited look ahead) and extract objects belonging to the same | |
2823 | * page. It builds a detached freelist directly within the given | |
2824 | * page/objects. This can happen without any need for | |
2825 | * synchronization, because the objects are owned by running process. | |
2826 | * The freelist is build up as a single linked list in the objects. | |
2827 | * The idea is, that this detached freelist can then be bulk | |
2828 | * transferred to the real freelist(s), but only requiring a single | |
2829 | * synchronization primitive. Look ahead in the array is limited due | |
2830 | * to performance reasons. | |
2831 | */ | |
2832 | static int build_detached_freelist(struct kmem_cache *s, size_t size, | |
2833 | void **p, struct detached_freelist *df) | |
2834 | { | |
2835 | size_t first_skipped_index = 0; | |
2836 | int lookahead = 3; | |
2837 | void *object; | |
fbd02630 | 2838 | |
d0ecd894 JDB |
2839 | /* Always re-init detached_freelist */ |
2840 | df->page = NULL; | |
fbd02630 | 2841 | |
d0ecd894 JDB |
2842 | do { |
2843 | object = p[--size]; | |
2844 | } while (!object && size); | |
3eed034d | 2845 | |
d0ecd894 JDB |
2846 | if (!object) |
2847 | return 0; | |
fbd02630 | 2848 | |
d0ecd894 JDB |
2849 | /* Start new detached freelist */ |
2850 | set_freepointer(s, object, NULL); | |
2851 | df->page = virt_to_head_page(object); | |
2852 | df->tail = object; | |
2853 | df->freelist = object; | |
2854 | p[size] = NULL; /* mark object processed */ | |
2855 | df->cnt = 1; | |
2856 | ||
2857 | while (size) { | |
2858 | object = p[--size]; | |
2859 | if (!object) | |
2860 | continue; /* Skip processed objects */ | |
2861 | ||
2862 | /* df->page is always set at this point */ | |
2863 | if (df->page == virt_to_head_page(object)) { | |
2864 | /* Opportunity build freelist */ | |
2865 | set_freepointer(s, object, df->freelist); | |
2866 | df->freelist = object; | |
2867 | df->cnt++; | |
2868 | p[size] = NULL; /* mark object processed */ | |
2869 | ||
2870 | continue; | |
fbd02630 | 2871 | } |
d0ecd894 JDB |
2872 | |
2873 | /* Limit look ahead search */ | |
2874 | if (!--lookahead) | |
2875 | break; | |
2876 | ||
2877 | if (!first_skipped_index) | |
2878 | first_skipped_index = size + 1; | |
fbd02630 | 2879 | } |
d0ecd894 JDB |
2880 | |
2881 | return first_skipped_index; | |
2882 | } | |
2883 | ||
2884 | ||
2885 | /* Note that interrupts must be enabled when calling this function. */ | |
03374518 | 2886 | void kmem_cache_free_bulk(struct kmem_cache *orig_s, size_t size, void **p) |
d0ecd894 JDB |
2887 | { |
2888 | if (WARN_ON(!size)) | |
2889 | return; | |
2890 | ||
2891 | do { | |
2892 | struct detached_freelist df; | |
03374518 JDB |
2893 | struct kmem_cache *s; |
2894 | ||
2895 | /* Support for memcg */ | |
2896 | s = cache_from_obj(orig_s, p[size - 1]); | |
d0ecd894 JDB |
2897 | |
2898 | size = build_detached_freelist(s, size, p, &df); | |
2899 | if (unlikely(!df.page)) | |
2900 | continue; | |
2901 | ||
2902 | slab_free(s, df.page, df.freelist, df.tail, df.cnt, _RET_IP_); | |
2903 | } while (likely(size)); | |
484748f0 CL |
2904 | } |
2905 | EXPORT_SYMBOL(kmem_cache_free_bulk); | |
2906 | ||
994eb764 | 2907 | /* Note that interrupts must be enabled when calling this function. */ |
865762a8 JDB |
2908 | int kmem_cache_alloc_bulk(struct kmem_cache *s, gfp_t flags, size_t size, |
2909 | void **p) | |
484748f0 | 2910 | { |
994eb764 JDB |
2911 | struct kmem_cache_cpu *c; |
2912 | int i; | |
2913 | ||
03ec0ed5 JDB |
2914 | /* memcg and kmem_cache debug support */ |
2915 | s = slab_pre_alloc_hook(s, flags); | |
2916 | if (unlikely(!s)) | |
2917 | return false; | |
994eb764 JDB |
2918 | /* |
2919 | * Drain objects in the per cpu slab, while disabling local | |
2920 | * IRQs, which protects against PREEMPT and interrupts | |
2921 | * handlers invoking normal fastpath. | |
2922 | */ | |
2923 | local_irq_disable(); | |
2924 | c = this_cpu_ptr(s->cpu_slab); | |
2925 | ||
2926 | for (i = 0; i < size; i++) { | |
2927 | void *object = c->freelist; | |
2928 | ||
ebe909e0 | 2929 | if (unlikely(!object)) { |
ebe909e0 JDB |
2930 | /* |
2931 | * Invoking slow path likely have side-effect | |
2932 | * of re-populating per CPU c->freelist | |
2933 | */ | |
87098373 | 2934 | p[i] = ___slab_alloc(s, flags, NUMA_NO_NODE, |
ebe909e0 | 2935 | _RET_IP_, c); |
87098373 CL |
2936 | if (unlikely(!p[i])) |
2937 | goto error; | |
2938 | ||
ebe909e0 JDB |
2939 | c = this_cpu_ptr(s->cpu_slab); |
2940 | continue; /* goto for-loop */ | |
2941 | } | |
994eb764 JDB |
2942 | c->freelist = get_freepointer(s, object); |
2943 | p[i] = object; | |
2944 | } | |
2945 | c->tid = next_tid(c->tid); | |
2946 | local_irq_enable(); | |
2947 | ||
2948 | /* Clear memory outside IRQ disabled fastpath loop */ | |
2949 | if (unlikely(flags & __GFP_ZERO)) { | |
2950 | int j; | |
2951 | ||
2952 | for (j = 0; j < i; j++) | |
2953 | memset(p[j], 0, s->object_size); | |
2954 | } | |
2955 | ||
03ec0ed5 JDB |
2956 | /* memcg and kmem_cache debug support */ |
2957 | slab_post_alloc_hook(s, flags, size, p); | |
865762a8 | 2958 | return i; |
87098373 | 2959 | error: |
87098373 | 2960 | local_irq_enable(); |
03ec0ed5 JDB |
2961 | slab_post_alloc_hook(s, flags, i, p); |
2962 | __kmem_cache_free_bulk(s, i, p); | |
865762a8 | 2963 | return 0; |
484748f0 CL |
2964 | } |
2965 | EXPORT_SYMBOL(kmem_cache_alloc_bulk); | |
2966 | ||
2967 | ||
81819f0f | 2968 | /* |
672bba3a CL |
2969 | * Object placement in a slab is made very easy because we always start at |
2970 | * offset 0. If we tune the size of the object to the alignment then we can | |
2971 | * get the required alignment by putting one properly sized object after | |
2972 | * another. | |
81819f0f CL |
2973 | * |
2974 | * Notice that the allocation order determines the sizes of the per cpu | |
2975 | * caches. Each processor has always one slab available for allocations. | |
2976 | * Increasing the allocation order reduces the number of times that slabs | |
672bba3a | 2977 | * must be moved on and off the partial lists and is therefore a factor in |
81819f0f | 2978 | * locking overhead. |
81819f0f CL |
2979 | */ |
2980 | ||
2981 | /* | |
2982 | * Mininum / Maximum order of slab pages. This influences locking overhead | |
2983 | * and slab fragmentation. A higher order reduces the number of partial slabs | |
2984 | * and increases the number of allocations possible without having to | |
2985 | * take the list_lock. | |
2986 | */ | |
2987 | static int slub_min_order; | |
114e9e89 | 2988 | static int slub_max_order = PAGE_ALLOC_COSTLY_ORDER; |
9b2cd506 | 2989 | static int slub_min_objects; |
81819f0f | 2990 | |
81819f0f CL |
2991 | /* |
2992 | * Calculate the order of allocation given an slab object size. | |
2993 | * | |
672bba3a CL |
2994 | * The order of allocation has significant impact on performance and other |
2995 | * system components. Generally order 0 allocations should be preferred since | |
2996 | * order 0 does not cause fragmentation in the page allocator. Larger objects | |
2997 | * be problematic to put into order 0 slabs because there may be too much | |
c124f5b5 | 2998 | * unused space left. We go to a higher order if more than 1/16th of the slab |
672bba3a CL |
2999 | * would be wasted. |
3000 | * | |
3001 | * In order to reach satisfactory performance we must ensure that a minimum | |
3002 | * number of objects is in one slab. Otherwise we may generate too much | |
3003 | * activity on the partial lists which requires taking the list_lock. This is | |
3004 | * less a concern for large slabs though which are rarely used. | |
81819f0f | 3005 | * |
672bba3a CL |
3006 | * slub_max_order specifies the order where we begin to stop considering the |
3007 | * number of objects in a slab as critical. If we reach slub_max_order then | |
3008 | * we try to keep the page order as low as possible. So we accept more waste | |
3009 | * of space in favor of a small page order. | |
81819f0f | 3010 | * |
672bba3a CL |
3011 | * Higher order allocations also allow the placement of more objects in a |
3012 | * slab and thereby reduce object handling overhead. If the user has | |
3013 | * requested a higher mininum order then we start with that one instead of | |
3014 | * the smallest order which will fit the object. | |
81819f0f | 3015 | */ |
5e6d444e | 3016 | static inline int slab_order(int size, int min_objects, |
ab9a0f19 | 3017 | int max_order, int fract_leftover, int reserved) |
81819f0f CL |
3018 | { |
3019 | int order; | |
3020 | int rem; | |
6300ea75 | 3021 | int min_order = slub_min_order; |
81819f0f | 3022 | |
ab9a0f19 | 3023 | if (order_objects(min_order, size, reserved) > MAX_OBJS_PER_PAGE) |
210b5c06 | 3024 | return get_order(size * MAX_OBJS_PER_PAGE) - 1; |
39b26464 | 3025 | |
9f835703 | 3026 | for (order = max(min_order, get_order(min_objects * size + reserved)); |
5e6d444e | 3027 | order <= max_order; order++) { |
81819f0f | 3028 | |
5e6d444e | 3029 | unsigned long slab_size = PAGE_SIZE << order; |
81819f0f | 3030 | |
ab9a0f19 | 3031 | rem = (slab_size - reserved) % size; |
81819f0f | 3032 | |
5e6d444e | 3033 | if (rem <= slab_size / fract_leftover) |
81819f0f | 3034 | break; |
81819f0f | 3035 | } |
672bba3a | 3036 | |
81819f0f CL |
3037 | return order; |
3038 | } | |
3039 | ||
ab9a0f19 | 3040 | static inline int calculate_order(int size, int reserved) |
5e6d444e CL |
3041 | { |
3042 | int order; | |
3043 | int min_objects; | |
3044 | int fraction; | |
e8120ff1 | 3045 | int max_objects; |
5e6d444e CL |
3046 | |
3047 | /* | |
3048 | * Attempt to find best configuration for a slab. This | |
3049 | * works by first attempting to generate a layout with | |
3050 | * the best configuration and backing off gradually. | |
3051 | * | |
422ff4d7 | 3052 | * First we increase the acceptable waste in a slab. Then |
5e6d444e CL |
3053 | * we reduce the minimum objects required in a slab. |
3054 | */ | |
3055 | min_objects = slub_min_objects; | |
9b2cd506 CL |
3056 | if (!min_objects) |
3057 | min_objects = 4 * (fls(nr_cpu_ids) + 1); | |
ab9a0f19 | 3058 | max_objects = order_objects(slub_max_order, size, reserved); |
e8120ff1 ZY |
3059 | min_objects = min(min_objects, max_objects); |
3060 | ||
5e6d444e | 3061 | while (min_objects > 1) { |
c124f5b5 | 3062 | fraction = 16; |
5e6d444e CL |
3063 | while (fraction >= 4) { |
3064 | order = slab_order(size, min_objects, | |
ab9a0f19 | 3065 | slub_max_order, fraction, reserved); |
5e6d444e CL |
3066 | if (order <= slub_max_order) |
3067 | return order; | |
3068 | fraction /= 2; | |
3069 | } | |
5086c389 | 3070 | min_objects--; |
5e6d444e CL |
3071 | } |
3072 | ||
3073 | /* | |
3074 | * We were unable to place multiple objects in a slab. Now | |
3075 | * lets see if we can place a single object there. | |
3076 | */ | |
ab9a0f19 | 3077 | order = slab_order(size, 1, slub_max_order, 1, reserved); |
5e6d444e CL |
3078 | if (order <= slub_max_order) |
3079 | return order; | |
3080 | ||
3081 | /* | |
3082 | * Doh this slab cannot be placed using slub_max_order. | |
3083 | */ | |
ab9a0f19 | 3084 | order = slab_order(size, 1, MAX_ORDER, 1, reserved); |
818cf590 | 3085 | if (order < MAX_ORDER) |
5e6d444e CL |
3086 | return order; |
3087 | return -ENOSYS; | |
3088 | } | |
3089 | ||
5595cffc | 3090 | static void |
4053497d | 3091 | init_kmem_cache_node(struct kmem_cache_node *n) |
81819f0f CL |
3092 | { |
3093 | n->nr_partial = 0; | |
81819f0f CL |
3094 | spin_lock_init(&n->list_lock); |
3095 | INIT_LIST_HEAD(&n->partial); | |
8ab1372f | 3096 | #ifdef CONFIG_SLUB_DEBUG |
0f389ec6 | 3097 | atomic_long_set(&n->nr_slabs, 0); |
02b71b70 | 3098 | atomic_long_set(&n->total_objects, 0); |
643b1138 | 3099 | INIT_LIST_HEAD(&n->full); |
8ab1372f | 3100 | #endif |
81819f0f CL |
3101 | } |
3102 | ||
55136592 | 3103 | static inline int alloc_kmem_cache_cpus(struct kmem_cache *s) |
4c93c355 | 3104 | { |
6c182dc0 | 3105 | BUILD_BUG_ON(PERCPU_DYNAMIC_EARLY_SIZE < |
95a05b42 | 3106 | KMALLOC_SHIFT_HIGH * sizeof(struct kmem_cache_cpu)); |
4c93c355 | 3107 | |
8a5ec0ba | 3108 | /* |
d4d84fef CM |
3109 | * Must align to double word boundary for the double cmpxchg |
3110 | * instructions to work; see __pcpu_double_call_return_bool(). | |
8a5ec0ba | 3111 | */ |
d4d84fef CM |
3112 | s->cpu_slab = __alloc_percpu(sizeof(struct kmem_cache_cpu), |
3113 | 2 * sizeof(void *)); | |
8a5ec0ba CL |
3114 | |
3115 | if (!s->cpu_slab) | |
3116 | return 0; | |
3117 | ||
3118 | init_kmem_cache_cpus(s); | |
4c93c355 | 3119 | |
8a5ec0ba | 3120 | return 1; |
4c93c355 | 3121 | } |
4c93c355 | 3122 | |
51df1142 CL |
3123 | static struct kmem_cache *kmem_cache_node; |
3124 | ||
81819f0f CL |
3125 | /* |
3126 | * No kmalloc_node yet so do it by hand. We know that this is the first | |
3127 | * slab on the node for this slabcache. There are no concurrent accesses | |
3128 | * possible. | |
3129 | * | |
721ae22a ZYW |
3130 | * Note that this function only works on the kmem_cache_node |
3131 | * when allocating for the kmem_cache_node. This is used for bootstrapping | |
4c93c355 | 3132 | * memory on a fresh node that has no slab structures yet. |
81819f0f | 3133 | */ |
55136592 | 3134 | static void early_kmem_cache_node_alloc(int node) |
81819f0f CL |
3135 | { |
3136 | struct page *page; | |
3137 | struct kmem_cache_node *n; | |
3138 | ||
51df1142 | 3139 | BUG_ON(kmem_cache_node->size < sizeof(struct kmem_cache_node)); |
81819f0f | 3140 | |
51df1142 | 3141 | page = new_slab(kmem_cache_node, GFP_NOWAIT, node); |
81819f0f CL |
3142 | |
3143 | BUG_ON(!page); | |
a2f92ee7 | 3144 | if (page_to_nid(page) != node) { |
f9f58285 FF |
3145 | pr_err("SLUB: Unable to allocate memory from node %d\n", node); |
3146 | pr_err("SLUB: Allocating a useless per node structure in order to be able to continue\n"); | |
a2f92ee7 CL |
3147 | } |
3148 | ||
81819f0f CL |
3149 | n = page->freelist; |
3150 | BUG_ON(!n); | |
51df1142 | 3151 | page->freelist = get_freepointer(kmem_cache_node, n); |
e6e82ea1 | 3152 | page->inuse = 1; |
8cb0a506 | 3153 | page->frozen = 0; |
51df1142 | 3154 | kmem_cache_node->node[node] = n; |
8ab1372f | 3155 | #ifdef CONFIG_SLUB_DEBUG |
f7cb1933 | 3156 | init_object(kmem_cache_node, n, SLUB_RED_ACTIVE); |
51df1142 | 3157 | init_tracking(kmem_cache_node, n); |
8ab1372f | 3158 | #endif |
0316bec2 | 3159 | kasan_kmalloc(kmem_cache_node, n, sizeof(struct kmem_cache_node)); |
4053497d | 3160 | init_kmem_cache_node(n); |
51df1142 | 3161 | inc_slabs_node(kmem_cache_node, node, page->objects); |
6446faa2 | 3162 | |
67b6c900 | 3163 | /* |
1e4dd946 SR |
3164 | * No locks need to be taken here as it has just been |
3165 | * initialized and there is no concurrent access. | |
67b6c900 | 3166 | */ |
1e4dd946 | 3167 | __add_partial(n, page, DEACTIVATE_TO_HEAD); |
81819f0f CL |
3168 | } |
3169 | ||
3170 | static void free_kmem_cache_nodes(struct kmem_cache *s) | |
3171 | { | |
3172 | int node; | |
fa45dc25 | 3173 | struct kmem_cache_node *n; |
81819f0f | 3174 | |
fa45dc25 CL |
3175 | for_each_kmem_cache_node(s, node, n) { |
3176 | kmem_cache_free(kmem_cache_node, n); | |
81819f0f CL |
3177 | s->node[node] = NULL; |
3178 | } | |
3179 | } | |
3180 | ||
52b4b950 DS |
3181 | void __kmem_cache_release(struct kmem_cache *s) |
3182 | { | |
3183 | free_percpu(s->cpu_slab); | |
3184 | free_kmem_cache_nodes(s); | |
3185 | } | |
3186 | ||
55136592 | 3187 | static int init_kmem_cache_nodes(struct kmem_cache *s) |
81819f0f CL |
3188 | { |
3189 | int node; | |
81819f0f | 3190 | |
f64dc58c | 3191 | for_each_node_state(node, N_NORMAL_MEMORY) { |
81819f0f CL |
3192 | struct kmem_cache_node *n; |
3193 | ||
73367bd8 | 3194 | if (slab_state == DOWN) { |
55136592 | 3195 | early_kmem_cache_node_alloc(node); |
73367bd8 AD |
3196 | continue; |
3197 | } | |
51df1142 | 3198 | n = kmem_cache_alloc_node(kmem_cache_node, |
55136592 | 3199 | GFP_KERNEL, node); |
81819f0f | 3200 | |
73367bd8 AD |
3201 | if (!n) { |
3202 | free_kmem_cache_nodes(s); | |
3203 | return 0; | |
81819f0f | 3204 | } |
73367bd8 | 3205 | |
81819f0f | 3206 | s->node[node] = n; |
4053497d | 3207 | init_kmem_cache_node(n); |
81819f0f CL |
3208 | } |
3209 | return 1; | |
3210 | } | |
81819f0f | 3211 | |
c0bdb232 | 3212 | static void set_min_partial(struct kmem_cache *s, unsigned long min) |
3b89d7d8 DR |
3213 | { |
3214 | if (min < MIN_PARTIAL) | |
3215 | min = MIN_PARTIAL; | |
3216 | else if (min > MAX_PARTIAL) | |
3217 | min = MAX_PARTIAL; | |
3218 | s->min_partial = min; | |
3219 | } | |
3220 | ||
81819f0f CL |
3221 | /* |
3222 | * calculate_sizes() determines the order and the distribution of data within | |
3223 | * a slab object. | |
3224 | */ | |
06b285dc | 3225 | static int calculate_sizes(struct kmem_cache *s, int forced_order) |
81819f0f CL |
3226 | { |
3227 | unsigned long flags = s->flags; | |
3b0efdfa | 3228 | unsigned long size = s->object_size; |
834f3d11 | 3229 | int order; |
81819f0f | 3230 | |
d8b42bf5 CL |
3231 | /* |
3232 | * Round up object size to the next word boundary. We can only | |
3233 | * place the free pointer at word boundaries and this determines | |
3234 | * the possible location of the free pointer. | |
3235 | */ | |
3236 | size = ALIGN(size, sizeof(void *)); | |
3237 | ||
3238 | #ifdef CONFIG_SLUB_DEBUG | |
81819f0f CL |
3239 | /* |
3240 | * Determine if we can poison the object itself. If the user of | |
3241 | * the slab may touch the object after free or before allocation | |
3242 | * then we should never poison the object itself. | |
3243 | */ | |
3244 | if ((flags & SLAB_POISON) && !(flags & SLAB_DESTROY_BY_RCU) && | |
c59def9f | 3245 | !s->ctor) |
81819f0f CL |
3246 | s->flags |= __OBJECT_POISON; |
3247 | else | |
3248 | s->flags &= ~__OBJECT_POISON; | |
3249 | ||
81819f0f CL |
3250 | |
3251 | /* | |
672bba3a | 3252 | * If we are Redzoning then check if there is some space between the |
81819f0f | 3253 | * end of the object and the free pointer. If not then add an |
672bba3a | 3254 | * additional word to have some bytes to store Redzone information. |
81819f0f | 3255 | */ |
3b0efdfa | 3256 | if ((flags & SLAB_RED_ZONE) && size == s->object_size) |
81819f0f | 3257 | size += sizeof(void *); |
41ecc55b | 3258 | #endif |
81819f0f CL |
3259 | |
3260 | /* | |
672bba3a CL |
3261 | * With that we have determined the number of bytes in actual use |
3262 | * by the object. This is the potential offset to the free pointer. | |
81819f0f CL |
3263 | */ |
3264 | s->inuse = size; | |
3265 | ||
3266 | if (((flags & (SLAB_DESTROY_BY_RCU | SLAB_POISON)) || | |
c59def9f | 3267 | s->ctor)) { |
81819f0f CL |
3268 | /* |
3269 | * Relocate free pointer after the object if it is not | |
3270 | * permitted to overwrite the first word of the object on | |
3271 | * kmem_cache_free. | |
3272 | * | |
3273 | * This is the case if we do RCU, have a constructor or | |
3274 | * destructor or are poisoning the objects. | |
3275 | */ | |
3276 | s->offset = size; | |
3277 | size += sizeof(void *); | |
3278 | } | |
3279 | ||
c12b3c62 | 3280 | #ifdef CONFIG_SLUB_DEBUG |
81819f0f CL |
3281 | if (flags & SLAB_STORE_USER) |
3282 | /* | |
3283 | * Need to store information about allocs and frees after | |
3284 | * the object. | |
3285 | */ | |
3286 | size += 2 * sizeof(struct track); | |
3287 | ||
be7b3fbc | 3288 | if (flags & SLAB_RED_ZONE) |
81819f0f CL |
3289 | /* |
3290 | * Add some empty padding so that we can catch | |
3291 | * overwrites from earlier objects rather than let | |
3292 | * tracking information or the free pointer be | |
0211a9c8 | 3293 | * corrupted if a user writes before the start |
81819f0f CL |
3294 | * of the object. |
3295 | */ | |
3296 | size += sizeof(void *); | |
41ecc55b | 3297 | #endif |
672bba3a | 3298 | |
81819f0f CL |
3299 | /* |
3300 | * SLUB stores one object immediately after another beginning from | |
3301 | * offset 0. In order to align the objects we have to simply size | |
3302 | * each object to conform to the alignment. | |
3303 | */ | |
45906855 | 3304 | size = ALIGN(size, s->align); |
81819f0f | 3305 | s->size = size; |
06b285dc CL |
3306 | if (forced_order >= 0) |
3307 | order = forced_order; | |
3308 | else | |
ab9a0f19 | 3309 | order = calculate_order(size, s->reserved); |
81819f0f | 3310 | |
834f3d11 | 3311 | if (order < 0) |
81819f0f CL |
3312 | return 0; |
3313 | ||
b7a49f0d | 3314 | s->allocflags = 0; |
834f3d11 | 3315 | if (order) |
b7a49f0d CL |
3316 | s->allocflags |= __GFP_COMP; |
3317 | ||
3318 | if (s->flags & SLAB_CACHE_DMA) | |
2c59dd65 | 3319 | s->allocflags |= GFP_DMA; |
b7a49f0d CL |
3320 | |
3321 | if (s->flags & SLAB_RECLAIM_ACCOUNT) | |
3322 | s->allocflags |= __GFP_RECLAIMABLE; | |
3323 | ||
81819f0f CL |
3324 | /* |
3325 | * Determine the number of objects per slab | |
3326 | */ | |
ab9a0f19 LJ |
3327 | s->oo = oo_make(order, size, s->reserved); |
3328 | s->min = oo_make(get_order(size), size, s->reserved); | |
205ab99d CL |
3329 | if (oo_objects(s->oo) > oo_objects(s->max)) |
3330 | s->max = s->oo; | |
81819f0f | 3331 | |
834f3d11 | 3332 | return !!oo_objects(s->oo); |
81819f0f CL |
3333 | } |
3334 | ||
8a13a4cc | 3335 | static int kmem_cache_open(struct kmem_cache *s, unsigned long flags) |
81819f0f | 3336 | { |
8a13a4cc | 3337 | s->flags = kmem_cache_flags(s->size, flags, s->name, s->ctor); |
ab9a0f19 | 3338 | s->reserved = 0; |
81819f0f | 3339 | |
da9a638c LJ |
3340 | if (need_reserve_slab_rcu && (s->flags & SLAB_DESTROY_BY_RCU)) |
3341 | s->reserved = sizeof(struct rcu_head); | |
81819f0f | 3342 | |
06b285dc | 3343 | if (!calculate_sizes(s, -1)) |
81819f0f | 3344 | goto error; |
3de47213 DR |
3345 | if (disable_higher_order_debug) { |
3346 | /* | |
3347 | * Disable debugging flags that store metadata if the min slab | |
3348 | * order increased. | |
3349 | */ | |
3b0efdfa | 3350 | if (get_order(s->size) > get_order(s->object_size)) { |
3de47213 DR |
3351 | s->flags &= ~DEBUG_METADATA_FLAGS; |
3352 | s->offset = 0; | |
3353 | if (!calculate_sizes(s, -1)) | |
3354 | goto error; | |
3355 | } | |
3356 | } | |
81819f0f | 3357 | |
2565409f HC |
3358 | #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \ |
3359 | defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE) | |
b789ef51 CL |
3360 | if (system_has_cmpxchg_double() && (s->flags & SLAB_DEBUG_FLAGS) == 0) |
3361 | /* Enable fast mode */ | |
3362 | s->flags |= __CMPXCHG_DOUBLE; | |
3363 | #endif | |
3364 | ||
3b89d7d8 DR |
3365 | /* |
3366 | * The larger the object size is, the more pages we want on the partial | |
3367 | * list to avoid pounding the page allocator excessively. | |
3368 | */ | |
49e22585 CL |
3369 | set_min_partial(s, ilog2(s->size) / 2); |
3370 | ||
3371 | /* | |
3372 | * cpu_partial determined the maximum number of objects kept in the | |
3373 | * per cpu partial lists of a processor. | |
3374 | * | |
3375 | * Per cpu partial lists mainly contain slabs that just have one | |
3376 | * object freed. If they are used for allocation then they can be | |
3377 | * filled up again with minimal effort. The slab will never hit the | |
3378 | * per node partial lists and therefore no locking will be required. | |
3379 | * | |
3380 | * This setting also determines | |
3381 | * | |
3382 | * A) The number of objects from per cpu partial slabs dumped to the | |
3383 | * per node list when we reach the limit. | |
9f264904 | 3384 | * B) The number of objects in cpu partial slabs to extract from the |
d0e0ac97 CG |
3385 | * per node list when we run out of per cpu objects. We only fetch |
3386 | * 50% to keep some capacity around for frees. | |
49e22585 | 3387 | */ |
345c905d | 3388 | if (!kmem_cache_has_cpu_partial(s)) |
8f1e33da CL |
3389 | s->cpu_partial = 0; |
3390 | else if (s->size >= PAGE_SIZE) | |
49e22585 CL |
3391 | s->cpu_partial = 2; |
3392 | else if (s->size >= 1024) | |
3393 | s->cpu_partial = 6; | |
3394 | else if (s->size >= 256) | |
3395 | s->cpu_partial = 13; | |
3396 | else | |
3397 | s->cpu_partial = 30; | |
3398 | ||
81819f0f | 3399 | #ifdef CONFIG_NUMA |
e2cb96b7 | 3400 | s->remote_node_defrag_ratio = 1000; |
81819f0f | 3401 | #endif |
55136592 | 3402 | if (!init_kmem_cache_nodes(s)) |
dfb4f096 | 3403 | goto error; |
81819f0f | 3404 | |
55136592 | 3405 | if (alloc_kmem_cache_cpus(s)) |
278b1bb1 | 3406 | return 0; |
ff12059e | 3407 | |
4c93c355 | 3408 | free_kmem_cache_nodes(s); |
81819f0f CL |
3409 | error: |
3410 | if (flags & SLAB_PANIC) | |
3411 | panic("Cannot create slab %s size=%lu realsize=%u " | |
3412 | "order=%u offset=%u flags=%lx\n", | |
d0e0ac97 CG |
3413 | s->name, (unsigned long)s->size, s->size, |
3414 | oo_order(s->oo), s->offset, flags); | |
278b1bb1 | 3415 | return -EINVAL; |
81819f0f | 3416 | } |
81819f0f | 3417 | |
33b12c38 CL |
3418 | static void list_slab_objects(struct kmem_cache *s, struct page *page, |
3419 | const char *text) | |
3420 | { | |
3421 | #ifdef CONFIG_SLUB_DEBUG | |
3422 | void *addr = page_address(page); | |
3423 | void *p; | |
a5dd5c11 NK |
3424 | unsigned long *map = kzalloc(BITS_TO_LONGS(page->objects) * |
3425 | sizeof(long), GFP_ATOMIC); | |
bbd7d57b ED |
3426 | if (!map) |
3427 | return; | |
945cf2b6 | 3428 | slab_err(s, page, text, s->name); |
33b12c38 | 3429 | slab_lock(page); |
33b12c38 | 3430 | |
5f80b13a | 3431 | get_map(s, page, map); |
33b12c38 CL |
3432 | for_each_object(p, s, addr, page->objects) { |
3433 | ||
3434 | if (!test_bit(slab_index(p, s, addr), map)) { | |
f9f58285 | 3435 | pr_err("INFO: Object 0x%p @offset=%tu\n", p, p - addr); |
33b12c38 CL |
3436 | print_tracking(s, p); |
3437 | } | |
3438 | } | |
3439 | slab_unlock(page); | |
bbd7d57b | 3440 | kfree(map); |
33b12c38 CL |
3441 | #endif |
3442 | } | |
3443 | ||
81819f0f | 3444 | /* |
599870b1 | 3445 | * Attempt to free all partial slabs on a node. |
52b4b950 DS |
3446 | * This is called from __kmem_cache_shutdown(). We must take list_lock |
3447 | * because sysfs file might still access partial list after the shutdowning. | |
81819f0f | 3448 | */ |
599870b1 | 3449 | static void free_partial(struct kmem_cache *s, struct kmem_cache_node *n) |
81819f0f | 3450 | { |
81819f0f CL |
3451 | struct page *page, *h; |
3452 | ||
52b4b950 DS |
3453 | BUG_ON(irqs_disabled()); |
3454 | spin_lock_irq(&n->list_lock); | |
33b12c38 | 3455 | list_for_each_entry_safe(page, h, &n->partial, lru) { |
81819f0f | 3456 | if (!page->inuse) { |
52b4b950 | 3457 | remove_partial(n, page); |
81819f0f | 3458 | discard_slab(s, page); |
33b12c38 CL |
3459 | } else { |
3460 | list_slab_objects(s, page, | |
52b4b950 | 3461 | "Objects remaining in %s on __kmem_cache_shutdown()"); |
599870b1 | 3462 | } |
33b12c38 | 3463 | } |
52b4b950 | 3464 | spin_unlock_irq(&n->list_lock); |
81819f0f CL |
3465 | } |
3466 | ||
3467 | /* | |
672bba3a | 3468 | * Release all resources used by a slab cache. |
81819f0f | 3469 | */ |
52b4b950 | 3470 | int __kmem_cache_shutdown(struct kmem_cache *s) |
81819f0f CL |
3471 | { |
3472 | int node; | |
fa45dc25 | 3473 | struct kmem_cache_node *n; |
81819f0f CL |
3474 | |
3475 | flush_all(s); | |
81819f0f | 3476 | /* Attempt to free all objects */ |
fa45dc25 | 3477 | for_each_kmem_cache_node(s, node, n) { |
599870b1 CL |
3478 | free_partial(s, n); |
3479 | if (n->nr_partial || slabs_node(s, node)) | |
81819f0f CL |
3480 | return 1; |
3481 | } | |
81819f0f CL |
3482 | return 0; |
3483 | } | |
3484 | ||
81819f0f CL |
3485 | /******************************************************************** |
3486 | * Kmalloc subsystem | |
3487 | *******************************************************************/ | |
3488 | ||
81819f0f CL |
3489 | static int __init setup_slub_min_order(char *str) |
3490 | { | |
06428780 | 3491 | get_option(&str, &slub_min_order); |
81819f0f CL |
3492 | |
3493 | return 1; | |
3494 | } | |
3495 | ||
3496 | __setup("slub_min_order=", setup_slub_min_order); | |
3497 | ||
3498 | static int __init setup_slub_max_order(char *str) | |
3499 | { | |
06428780 | 3500 | get_option(&str, &slub_max_order); |
818cf590 | 3501 | slub_max_order = min(slub_max_order, MAX_ORDER - 1); |
81819f0f CL |
3502 | |
3503 | return 1; | |
3504 | } | |
3505 | ||
3506 | __setup("slub_max_order=", setup_slub_max_order); | |
3507 | ||
3508 | static int __init setup_slub_min_objects(char *str) | |
3509 | { | |
06428780 | 3510 | get_option(&str, &slub_min_objects); |
81819f0f CL |
3511 | |
3512 | return 1; | |
3513 | } | |
3514 | ||
3515 | __setup("slub_min_objects=", setup_slub_min_objects); | |
3516 | ||
81819f0f CL |
3517 | void *__kmalloc(size_t size, gfp_t flags) |
3518 | { | |
aadb4bc4 | 3519 | struct kmem_cache *s; |
5b882be4 | 3520 | void *ret; |
81819f0f | 3521 | |
95a05b42 | 3522 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) |
eada35ef | 3523 | return kmalloc_large(size, flags); |
aadb4bc4 | 3524 | |
2c59dd65 | 3525 | s = kmalloc_slab(size, flags); |
aadb4bc4 CL |
3526 | |
3527 | if (unlikely(ZERO_OR_NULL_PTR(s))) | |
6cb8f913 CL |
3528 | return s; |
3529 | ||
2b847c3c | 3530 | ret = slab_alloc(s, flags, _RET_IP_); |
5b882be4 | 3531 | |
ca2b84cb | 3532 | trace_kmalloc(_RET_IP_, ret, size, s->size, flags); |
5b882be4 | 3533 | |
0316bec2 AR |
3534 | kasan_kmalloc(s, ret, size); |
3535 | ||
5b882be4 | 3536 | return ret; |
81819f0f CL |
3537 | } |
3538 | EXPORT_SYMBOL(__kmalloc); | |
3539 | ||
5d1f57e4 | 3540 | #ifdef CONFIG_NUMA |
f619cfe1 CL |
3541 | static void *kmalloc_large_node(size_t size, gfp_t flags, int node) |
3542 | { | |
b1eeab67 | 3543 | struct page *page; |
e4f7c0b4 | 3544 | void *ptr = NULL; |
f619cfe1 | 3545 | |
52383431 VD |
3546 | flags |= __GFP_COMP | __GFP_NOTRACK; |
3547 | page = alloc_kmem_pages_node(node, flags, get_order(size)); | |
f619cfe1 | 3548 | if (page) |
e4f7c0b4 CM |
3549 | ptr = page_address(page); |
3550 | ||
d56791b3 | 3551 | kmalloc_large_node_hook(ptr, size, flags); |
e4f7c0b4 | 3552 | return ptr; |
f619cfe1 CL |
3553 | } |
3554 | ||
81819f0f CL |
3555 | void *__kmalloc_node(size_t size, gfp_t flags, int node) |
3556 | { | |
aadb4bc4 | 3557 | struct kmem_cache *s; |
5b882be4 | 3558 | void *ret; |
81819f0f | 3559 | |
95a05b42 | 3560 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) { |
5b882be4 EGM |
3561 | ret = kmalloc_large_node(size, flags, node); |
3562 | ||
ca2b84cb EGM |
3563 | trace_kmalloc_node(_RET_IP_, ret, |
3564 | size, PAGE_SIZE << get_order(size), | |
3565 | flags, node); | |
5b882be4 EGM |
3566 | |
3567 | return ret; | |
3568 | } | |
aadb4bc4 | 3569 | |
2c59dd65 | 3570 | s = kmalloc_slab(size, flags); |
aadb4bc4 CL |
3571 | |
3572 | if (unlikely(ZERO_OR_NULL_PTR(s))) | |
6cb8f913 CL |
3573 | return s; |
3574 | ||
2b847c3c | 3575 | ret = slab_alloc_node(s, flags, node, _RET_IP_); |
5b882be4 | 3576 | |
ca2b84cb | 3577 | trace_kmalloc_node(_RET_IP_, ret, size, s->size, flags, node); |
5b882be4 | 3578 | |
0316bec2 AR |
3579 | kasan_kmalloc(s, ret, size); |
3580 | ||
5b882be4 | 3581 | return ret; |
81819f0f CL |
3582 | } |
3583 | EXPORT_SYMBOL(__kmalloc_node); | |
3584 | #endif | |
3585 | ||
0316bec2 | 3586 | static size_t __ksize(const void *object) |
81819f0f | 3587 | { |
272c1d21 | 3588 | struct page *page; |
81819f0f | 3589 | |
ef8b4520 | 3590 | if (unlikely(object == ZERO_SIZE_PTR)) |
272c1d21 CL |
3591 | return 0; |
3592 | ||
294a80a8 | 3593 | page = virt_to_head_page(object); |
294a80a8 | 3594 | |
76994412 PE |
3595 | if (unlikely(!PageSlab(page))) { |
3596 | WARN_ON(!PageCompound(page)); | |
294a80a8 | 3597 | return PAGE_SIZE << compound_order(page); |
76994412 | 3598 | } |
81819f0f | 3599 | |
1b4f59e3 | 3600 | return slab_ksize(page->slab_cache); |
81819f0f | 3601 | } |
0316bec2 AR |
3602 | |
3603 | size_t ksize(const void *object) | |
3604 | { | |
3605 | size_t size = __ksize(object); | |
3606 | /* We assume that ksize callers could use whole allocated area, | |
3607 | so we need unpoison this area. */ | |
3608 | kasan_krealloc(object, size); | |
3609 | return size; | |
3610 | } | |
b1aabecd | 3611 | EXPORT_SYMBOL(ksize); |
81819f0f CL |
3612 | |
3613 | void kfree(const void *x) | |
3614 | { | |
81819f0f | 3615 | struct page *page; |
5bb983b0 | 3616 | void *object = (void *)x; |
81819f0f | 3617 | |
2121db74 PE |
3618 | trace_kfree(_RET_IP_, x); |
3619 | ||
2408c550 | 3620 | if (unlikely(ZERO_OR_NULL_PTR(x))) |
81819f0f CL |
3621 | return; |
3622 | ||
b49af68f | 3623 | page = virt_to_head_page(x); |
aadb4bc4 | 3624 | if (unlikely(!PageSlab(page))) { |
0937502a | 3625 | BUG_ON(!PageCompound(page)); |
d56791b3 | 3626 | kfree_hook(x); |
52383431 | 3627 | __free_kmem_pages(page, compound_order(page)); |
aadb4bc4 CL |
3628 | return; |
3629 | } | |
81084651 | 3630 | slab_free(page->slab_cache, page, object, NULL, 1, _RET_IP_); |
81819f0f CL |
3631 | } |
3632 | EXPORT_SYMBOL(kfree); | |
3633 | ||
832f37f5 VD |
3634 | #define SHRINK_PROMOTE_MAX 32 |
3635 | ||
2086d26a | 3636 | /* |
832f37f5 VD |
3637 | * kmem_cache_shrink discards empty slabs and promotes the slabs filled |
3638 | * up most to the head of the partial lists. New allocations will then | |
3639 | * fill those up and thus they can be removed from the partial lists. | |
672bba3a CL |
3640 | * |
3641 | * The slabs with the least items are placed last. This results in them | |
3642 | * being allocated from last increasing the chance that the last objects | |
3643 | * are freed in them. | |
2086d26a | 3644 | */ |
d6e0b7fa | 3645 | int __kmem_cache_shrink(struct kmem_cache *s, bool deactivate) |
2086d26a CL |
3646 | { |
3647 | int node; | |
3648 | int i; | |
3649 | struct kmem_cache_node *n; | |
3650 | struct page *page; | |
3651 | struct page *t; | |
832f37f5 VD |
3652 | struct list_head discard; |
3653 | struct list_head promote[SHRINK_PROMOTE_MAX]; | |
2086d26a | 3654 | unsigned long flags; |
ce3712d7 | 3655 | int ret = 0; |
2086d26a | 3656 | |
d6e0b7fa VD |
3657 | if (deactivate) { |
3658 | /* | |
3659 | * Disable empty slabs caching. Used to avoid pinning offline | |
3660 | * memory cgroups by kmem pages that can be freed. | |
3661 | */ | |
3662 | s->cpu_partial = 0; | |
3663 | s->min_partial = 0; | |
3664 | ||
3665 | /* | |
3666 | * s->cpu_partial is checked locklessly (see put_cpu_partial), | |
3667 | * so we have to make sure the change is visible. | |
3668 | */ | |
3669 | kick_all_cpus_sync(); | |
3670 | } | |
3671 | ||
2086d26a | 3672 | flush_all(s); |
fa45dc25 | 3673 | for_each_kmem_cache_node(s, node, n) { |
832f37f5 VD |
3674 | INIT_LIST_HEAD(&discard); |
3675 | for (i = 0; i < SHRINK_PROMOTE_MAX; i++) | |
3676 | INIT_LIST_HEAD(promote + i); | |
2086d26a CL |
3677 | |
3678 | spin_lock_irqsave(&n->list_lock, flags); | |
3679 | ||
3680 | /* | |
832f37f5 | 3681 | * Build lists of slabs to discard or promote. |
2086d26a | 3682 | * |
672bba3a CL |
3683 | * Note that concurrent frees may occur while we hold the |
3684 | * list_lock. page->inuse here is the upper limit. | |
2086d26a CL |
3685 | */ |
3686 | list_for_each_entry_safe(page, t, &n->partial, lru) { | |
832f37f5 VD |
3687 | int free = page->objects - page->inuse; |
3688 | ||
3689 | /* Do not reread page->inuse */ | |
3690 | barrier(); | |
3691 | ||
3692 | /* We do not keep full slabs on the list */ | |
3693 | BUG_ON(free <= 0); | |
3694 | ||
3695 | if (free == page->objects) { | |
3696 | list_move(&page->lru, &discard); | |
69cb8e6b | 3697 | n->nr_partial--; |
832f37f5 VD |
3698 | } else if (free <= SHRINK_PROMOTE_MAX) |
3699 | list_move(&page->lru, promote + free - 1); | |
2086d26a CL |
3700 | } |
3701 | ||
2086d26a | 3702 | /* |
832f37f5 VD |
3703 | * Promote the slabs filled up most to the head of the |
3704 | * partial list. | |
2086d26a | 3705 | */ |
832f37f5 VD |
3706 | for (i = SHRINK_PROMOTE_MAX - 1; i >= 0; i--) |
3707 | list_splice(promote + i, &n->partial); | |
2086d26a | 3708 | |
2086d26a | 3709 | spin_unlock_irqrestore(&n->list_lock, flags); |
69cb8e6b CL |
3710 | |
3711 | /* Release empty slabs */ | |
832f37f5 | 3712 | list_for_each_entry_safe(page, t, &discard, lru) |
69cb8e6b | 3713 | discard_slab(s, page); |
ce3712d7 VD |
3714 | |
3715 | if (slabs_node(s, node)) | |
3716 | ret = 1; | |
2086d26a CL |
3717 | } |
3718 | ||
ce3712d7 | 3719 | return ret; |
2086d26a | 3720 | } |
2086d26a | 3721 | |
b9049e23 YG |
3722 | static int slab_mem_going_offline_callback(void *arg) |
3723 | { | |
3724 | struct kmem_cache *s; | |
3725 | ||
18004c5d | 3726 | mutex_lock(&slab_mutex); |
b9049e23 | 3727 | list_for_each_entry(s, &slab_caches, list) |
d6e0b7fa | 3728 | __kmem_cache_shrink(s, false); |
18004c5d | 3729 | mutex_unlock(&slab_mutex); |
b9049e23 YG |
3730 | |
3731 | return 0; | |
3732 | } | |
3733 | ||
3734 | static void slab_mem_offline_callback(void *arg) | |
3735 | { | |
3736 | struct kmem_cache_node *n; | |
3737 | struct kmem_cache *s; | |
3738 | struct memory_notify *marg = arg; | |
3739 | int offline_node; | |
3740 | ||
b9d5ab25 | 3741 | offline_node = marg->status_change_nid_normal; |
b9049e23 YG |
3742 | |
3743 | /* | |
3744 | * If the node still has available memory. we need kmem_cache_node | |
3745 | * for it yet. | |
3746 | */ | |
3747 | if (offline_node < 0) | |
3748 | return; | |
3749 | ||
18004c5d | 3750 | mutex_lock(&slab_mutex); |
b9049e23 YG |
3751 | list_for_each_entry(s, &slab_caches, list) { |
3752 | n = get_node(s, offline_node); | |
3753 | if (n) { | |
3754 | /* | |
3755 | * if n->nr_slabs > 0, slabs still exist on the node | |
3756 | * that is going down. We were unable to free them, | |
c9404c9c | 3757 | * and offline_pages() function shouldn't call this |
b9049e23 YG |
3758 | * callback. So, we must fail. |
3759 | */ | |
0f389ec6 | 3760 | BUG_ON(slabs_node(s, offline_node)); |
b9049e23 YG |
3761 | |
3762 | s->node[offline_node] = NULL; | |
8de66a0c | 3763 | kmem_cache_free(kmem_cache_node, n); |
b9049e23 YG |
3764 | } |
3765 | } | |
18004c5d | 3766 | mutex_unlock(&slab_mutex); |
b9049e23 YG |
3767 | } |
3768 | ||
3769 | static int slab_mem_going_online_callback(void *arg) | |
3770 | { | |
3771 | struct kmem_cache_node *n; | |
3772 | struct kmem_cache *s; | |
3773 | struct memory_notify *marg = arg; | |
b9d5ab25 | 3774 | int nid = marg->status_change_nid_normal; |
b9049e23 YG |
3775 | int ret = 0; |
3776 | ||
3777 | /* | |
3778 | * If the node's memory is already available, then kmem_cache_node is | |
3779 | * already created. Nothing to do. | |
3780 | */ | |
3781 | if (nid < 0) | |
3782 | return 0; | |
3783 | ||
3784 | /* | |
0121c619 | 3785 | * We are bringing a node online. No memory is available yet. We must |
b9049e23 YG |
3786 | * allocate a kmem_cache_node structure in order to bring the node |
3787 | * online. | |
3788 | */ | |
18004c5d | 3789 | mutex_lock(&slab_mutex); |
b9049e23 YG |
3790 | list_for_each_entry(s, &slab_caches, list) { |
3791 | /* | |
3792 | * XXX: kmem_cache_alloc_node will fallback to other nodes | |
3793 | * since memory is not yet available from the node that | |
3794 | * is brought up. | |
3795 | */ | |
8de66a0c | 3796 | n = kmem_cache_alloc(kmem_cache_node, GFP_KERNEL); |
b9049e23 YG |
3797 | if (!n) { |
3798 | ret = -ENOMEM; | |
3799 | goto out; | |
3800 | } | |
4053497d | 3801 | init_kmem_cache_node(n); |
b9049e23 YG |
3802 | s->node[nid] = n; |
3803 | } | |
3804 | out: | |
18004c5d | 3805 | mutex_unlock(&slab_mutex); |
b9049e23 YG |
3806 | return ret; |
3807 | } | |
3808 | ||
3809 | static int slab_memory_callback(struct notifier_block *self, | |
3810 | unsigned long action, void *arg) | |
3811 | { | |
3812 | int ret = 0; | |
3813 | ||
3814 | switch (action) { | |
3815 | case MEM_GOING_ONLINE: | |
3816 | ret = slab_mem_going_online_callback(arg); | |
3817 | break; | |
3818 | case MEM_GOING_OFFLINE: | |
3819 | ret = slab_mem_going_offline_callback(arg); | |
3820 | break; | |
3821 | case MEM_OFFLINE: | |
3822 | case MEM_CANCEL_ONLINE: | |
3823 | slab_mem_offline_callback(arg); | |
3824 | break; | |
3825 | case MEM_ONLINE: | |
3826 | case MEM_CANCEL_OFFLINE: | |
3827 | break; | |
3828 | } | |
dc19f9db KH |
3829 | if (ret) |
3830 | ret = notifier_from_errno(ret); | |
3831 | else | |
3832 | ret = NOTIFY_OK; | |
b9049e23 YG |
3833 | return ret; |
3834 | } | |
3835 | ||
3ac38faa AM |
3836 | static struct notifier_block slab_memory_callback_nb = { |
3837 | .notifier_call = slab_memory_callback, | |
3838 | .priority = SLAB_CALLBACK_PRI, | |
3839 | }; | |
b9049e23 | 3840 | |
81819f0f CL |
3841 | /******************************************************************** |
3842 | * Basic setup of slabs | |
3843 | *******************************************************************/ | |
3844 | ||
51df1142 CL |
3845 | /* |
3846 | * Used for early kmem_cache structures that were allocated using | |
dffb4d60 CL |
3847 | * the page allocator. Allocate them properly then fix up the pointers |
3848 | * that may be pointing to the wrong kmem_cache structure. | |
51df1142 CL |
3849 | */ |
3850 | ||
dffb4d60 | 3851 | static struct kmem_cache * __init bootstrap(struct kmem_cache *static_cache) |
51df1142 CL |
3852 | { |
3853 | int node; | |
dffb4d60 | 3854 | struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT); |
fa45dc25 | 3855 | struct kmem_cache_node *n; |
51df1142 | 3856 | |
dffb4d60 | 3857 | memcpy(s, static_cache, kmem_cache->object_size); |
51df1142 | 3858 | |
7d557b3c GC |
3859 | /* |
3860 | * This runs very early, and only the boot processor is supposed to be | |
3861 | * up. Even if it weren't true, IRQs are not up so we couldn't fire | |
3862 | * IPIs around. | |
3863 | */ | |
3864 | __flush_cpu_slab(s, smp_processor_id()); | |
fa45dc25 | 3865 | for_each_kmem_cache_node(s, node, n) { |
51df1142 CL |
3866 | struct page *p; |
3867 | ||
fa45dc25 CL |
3868 | list_for_each_entry(p, &n->partial, lru) |
3869 | p->slab_cache = s; | |
51df1142 | 3870 | |
607bf324 | 3871 | #ifdef CONFIG_SLUB_DEBUG |
fa45dc25 CL |
3872 | list_for_each_entry(p, &n->full, lru) |
3873 | p->slab_cache = s; | |
51df1142 | 3874 | #endif |
51df1142 | 3875 | } |
f7ce3190 | 3876 | slab_init_memcg_params(s); |
dffb4d60 CL |
3877 | list_add(&s->list, &slab_caches); |
3878 | return s; | |
51df1142 CL |
3879 | } |
3880 | ||
81819f0f CL |
3881 | void __init kmem_cache_init(void) |
3882 | { | |
dffb4d60 CL |
3883 | static __initdata struct kmem_cache boot_kmem_cache, |
3884 | boot_kmem_cache_node; | |
51df1142 | 3885 | |
fc8d8620 SG |
3886 | if (debug_guardpage_minorder()) |
3887 | slub_max_order = 0; | |
3888 | ||
dffb4d60 CL |
3889 | kmem_cache_node = &boot_kmem_cache_node; |
3890 | kmem_cache = &boot_kmem_cache; | |
51df1142 | 3891 | |
dffb4d60 CL |
3892 | create_boot_cache(kmem_cache_node, "kmem_cache_node", |
3893 | sizeof(struct kmem_cache_node), SLAB_HWCACHE_ALIGN); | |
b9049e23 | 3894 | |
3ac38faa | 3895 | register_hotmemory_notifier(&slab_memory_callback_nb); |
81819f0f CL |
3896 | |
3897 | /* Able to allocate the per node structures */ | |
3898 | slab_state = PARTIAL; | |
3899 | ||
dffb4d60 CL |
3900 | create_boot_cache(kmem_cache, "kmem_cache", |
3901 | offsetof(struct kmem_cache, node) + | |
3902 | nr_node_ids * sizeof(struct kmem_cache_node *), | |
3903 | SLAB_HWCACHE_ALIGN); | |
8a13a4cc | 3904 | |
dffb4d60 | 3905 | kmem_cache = bootstrap(&boot_kmem_cache); |
81819f0f | 3906 | |
51df1142 CL |
3907 | /* |
3908 | * Allocate kmem_cache_node properly from the kmem_cache slab. | |
3909 | * kmem_cache_node is separately allocated so no need to | |
3910 | * update any list pointers. | |
3911 | */ | |
dffb4d60 | 3912 | kmem_cache_node = bootstrap(&boot_kmem_cache_node); |
51df1142 CL |
3913 | |
3914 | /* Now we can use the kmem_cache to allocate kmalloc slabs */ | |
34cc6990 | 3915 | setup_kmalloc_cache_index_table(); |
f97d5f63 | 3916 | create_kmalloc_caches(0); |
81819f0f CL |
3917 | |
3918 | #ifdef CONFIG_SMP | |
3919 | register_cpu_notifier(&slab_notifier); | |
9dfc6e68 | 3920 | #endif |
81819f0f | 3921 | |
f9f58285 | 3922 | pr_info("SLUB: HWalign=%d, Order=%d-%d, MinObjects=%d, CPUs=%d, Nodes=%d\n", |
f97d5f63 | 3923 | cache_line_size(), |
81819f0f CL |
3924 | slub_min_order, slub_max_order, slub_min_objects, |
3925 | nr_cpu_ids, nr_node_ids); | |
3926 | } | |
3927 | ||
7e85ee0c PE |
3928 | void __init kmem_cache_init_late(void) |
3929 | { | |
7e85ee0c PE |
3930 | } |
3931 | ||
2633d7a0 | 3932 | struct kmem_cache * |
a44cb944 VD |
3933 | __kmem_cache_alias(const char *name, size_t size, size_t align, |
3934 | unsigned long flags, void (*ctor)(void *)) | |
81819f0f | 3935 | { |
426589f5 | 3936 | struct kmem_cache *s, *c; |
81819f0f | 3937 | |
a44cb944 | 3938 | s = find_mergeable(size, align, flags, name, ctor); |
81819f0f CL |
3939 | if (s) { |
3940 | s->refcount++; | |
84d0ddd6 | 3941 | |
81819f0f CL |
3942 | /* |
3943 | * Adjust the object sizes so that we clear | |
3944 | * the complete object on kzalloc. | |
3945 | */ | |
3b0efdfa | 3946 | s->object_size = max(s->object_size, (int)size); |
81819f0f | 3947 | s->inuse = max_t(int, s->inuse, ALIGN(size, sizeof(void *))); |
6446faa2 | 3948 | |
426589f5 | 3949 | for_each_memcg_cache(c, s) { |
84d0ddd6 VD |
3950 | c->object_size = s->object_size; |
3951 | c->inuse = max_t(int, c->inuse, | |
3952 | ALIGN(size, sizeof(void *))); | |
3953 | } | |
3954 | ||
7b8f3b66 | 3955 | if (sysfs_slab_alias(s, name)) { |
7b8f3b66 | 3956 | s->refcount--; |
cbb79694 | 3957 | s = NULL; |
7b8f3b66 | 3958 | } |
a0e1d1be | 3959 | } |
6446faa2 | 3960 | |
cbb79694 CL |
3961 | return s; |
3962 | } | |
84c1cf62 | 3963 | |
8a13a4cc | 3964 | int __kmem_cache_create(struct kmem_cache *s, unsigned long flags) |
cbb79694 | 3965 | { |
aac3a166 PE |
3966 | int err; |
3967 | ||
3968 | err = kmem_cache_open(s, flags); | |
3969 | if (err) | |
3970 | return err; | |
20cea968 | 3971 | |
45530c44 CL |
3972 | /* Mutex is not taken during early boot */ |
3973 | if (slab_state <= UP) | |
3974 | return 0; | |
3975 | ||
107dab5c | 3976 | memcg_propagate_slab_attrs(s); |
aac3a166 | 3977 | err = sysfs_slab_add(s); |
aac3a166 | 3978 | if (err) |
52b4b950 | 3979 | __kmem_cache_release(s); |
20cea968 | 3980 | |
aac3a166 | 3981 | return err; |
81819f0f | 3982 | } |
81819f0f | 3983 | |
81819f0f | 3984 | #ifdef CONFIG_SMP |
81819f0f | 3985 | /* |
672bba3a CL |
3986 | * Use the cpu notifier to insure that the cpu slabs are flushed when |
3987 | * necessary. | |
81819f0f | 3988 | */ |
0db0628d | 3989 | static int slab_cpuup_callback(struct notifier_block *nfb, |
81819f0f CL |
3990 | unsigned long action, void *hcpu) |
3991 | { | |
3992 | long cpu = (long)hcpu; | |
5b95a4ac CL |
3993 | struct kmem_cache *s; |
3994 | unsigned long flags; | |
81819f0f CL |
3995 | |
3996 | switch (action) { | |
3997 | case CPU_UP_CANCELED: | |
8bb78442 | 3998 | case CPU_UP_CANCELED_FROZEN: |
81819f0f | 3999 | case CPU_DEAD: |
8bb78442 | 4000 | case CPU_DEAD_FROZEN: |
18004c5d | 4001 | mutex_lock(&slab_mutex); |
5b95a4ac CL |
4002 | list_for_each_entry(s, &slab_caches, list) { |
4003 | local_irq_save(flags); | |
4004 | __flush_cpu_slab(s, cpu); | |
4005 | local_irq_restore(flags); | |
4006 | } | |
18004c5d | 4007 | mutex_unlock(&slab_mutex); |
81819f0f CL |
4008 | break; |
4009 | default: | |
4010 | break; | |
4011 | } | |
4012 | return NOTIFY_OK; | |
4013 | } | |
4014 | ||
0db0628d | 4015 | static struct notifier_block slab_notifier = { |
3adbefee | 4016 | .notifier_call = slab_cpuup_callback |
06428780 | 4017 | }; |
81819f0f CL |
4018 | |
4019 | #endif | |
4020 | ||
ce71e27c | 4021 | void *__kmalloc_track_caller(size_t size, gfp_t gfpflags, unsigned long caller) |
81819f0f | 4022 | { |
aadb4bc4 | 4023 | struct kmem_cache *s; |
94b528d0 | 4024 | void *ret; |
aadb4bc4 | 4025 | |
95a05b42 | 4026 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) |
eada35ef PE |
4027 | return kmalloc_large(size, gfpflags); |
4028 | ||
2c59dd65 | 4029 | s = kmalloc_slab(size, gfpflags); |
81819f0f | 4030 | |
2408c550 | 4031 | if (unlikely(ZERO_OR_NULL_PTR(s))) |
6cb8f913 | 4032 | return s; |
81819f0f | 4033 | |
2b847c3c | 4034 | ret = slab_alloc(s, gfpflags, caller); |
94b528d0 | 4035 | |
25985edc | 4036 | /* Honor the call site pointer we received. */ |
ca2b84cb | 4037 | trace_kmalloc(caller, ret, size, s->size, gfpflags); |
94b528d0 EGM |
4038 | |
4039 | return ret; | |
81819f0f CL |
4040 | } |
4041 | ||
5d1f57e4 | 4042 | #ifdef CONFIG_NUMA |
81819f0f | 4043 | void *__kmalloc_node_track_caller(size_t size, gfp_t gfpflags, |
ce71e27c | 4044 | int node, unsigned long caller) |
81819f0f | 4045 | { |
aadb4bc4 | 4046 | struct kmem_cache *s; |
94b528d0 | 4047 | void *ret; |
aadb4bc4 | 4048 | |
95a05b42 | 4049 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) { |
d3e14aa3 XF |
4050 | ret = kmalloc_large_node(size, gfpflags, node); |
4051 | ||
4052 | trace_kmalloc_node(caller, ret, | |
4053 | size, PAGE_SIZE << get_order(size), | |
4054 | gfpflags, node); | |
4055 | ||
4056 | return ret; | |
4057 | } | |
eada35ef | 4058 | |
2c59dd65 | 4059 | s = kmalloc_slab(size, gfpflags); |
81819f0f | 4060 | |
2408c550 | 4061 | if (unlikely(ZERO_OR_NULL_PTR(s))) |
6cb8f913 | 4062 | return s; |
81819f0f | 4063 | |
2b847c3c | 4064 | ret = slab_alloc_node(s, gfpflags, node, caller); |
94b528d0 | 4065 | |
25985edc | 4066 | /* Honor the call site pointer we received. */ |
ca2b84cb | 4067 | trace_kmalloc_node(caller, ret, size, s->size, gfpflags, node); |
94b528d0 EGM |
4068 | |
4069 | return ret; | |
81819f0f | 4070 | } |
5d1f57e4 | 4071 | #endif |
81819f0f | 4072 | |
ab4d5ed5 | 4073 | #ifdef CONFIG_SYSFS |
205ab99d CL |
4074 | static int count_inuse(struct page *page) |
4075 | { | |
4076 | return page->inuse; | |
4077 | } | |
4078 | ||
4079 | static int count_total(struct page *page) | |
4080 | { | |
4081 | return page->objects; | |
4082 | } | |
ab4d5ed5 | 4083 | #endif |
205ab99d | 4084 | |
ab4d5ed5 | 4085 | #ifdef CONFIG_SLUB_DEBUG |
434e245d CL |
4086 | static int validate_slab(struct kmem_cache *s, struct page *page, |
4087 | unsigned long *map) | |
53e15af0 CL |
4088 | { |
4089 | void *p; | |
a973e9dd | 4090 | void *addr = page_address(page); |
53e15af0 CL |
4091 | |
4092 | if (!check_slab(s, page) || | |
4093 | !on_freelist(s, page, NULL)) | |
4094 | return 0; | |
4095 | ||
4096 | /* Now we know that a valid freelist exists */ | |
39b26464 | 4097 | bitmap_zero(map, page->objects); |
53e15af0 | 4098 | |
5f80b13a CL |
4099 | get_map(s, page, map); |
4100 | for_each_object(p, s, addr, page->objects) { | |
4101 | if (test_bit(slab_index(p, s, addr), map)) | |
4102 | if (!check_object(s, page, p, SLUB_RED_INACTIVE)) | |
4103 | return 0; | |
53e15af0 CL |
4104 | } |
4105 | ||
224a88be | 4106 | for_each_object(p, s, addr, page->objects) |
7656c72b | 4107 | if (!test_bit(slab_index(p, s, addr), map)) |
37d57443 | 4108 | if (!check_object(s, page, p, SLUB_RED_ACTIVE)) |
53e15af0 CL |
4109 | return 0; |
4110 | return 1; | |
4111 | } | |
4112 | ||
434e245d CL |
4113 | static void validate_slab_slab(struct kmem_cache *s, struct page *page, |
4114 | unsigned long *map) | |
53e15af0 | 4115 | { |
881db7fb CL |
4116 | slab_lock(page); |
4117 | validate_slab(s, page, map); | |
4118 | slab_unlock(page); | |
53e15af0 CL |
4119 | } |
4120 | ||
434e245d CL |
4121 | static int validate_slab_node(struct kmem_cache *s, |
4122 | struct kmem_cache_node *n, unsigned long *map) | |
53e15af0 CL |
4123 | { |
4124 | unsigned long count = 0; | |
4125 | struct page *page; | |
4126 | unsigned long flags; | |
4127 | ||
4128 | spin_lock_irqsave(&n->list_lock, flags); | |
4129 | ||
4130 | list_for_each_entry(page, &n->partial, lru) { | |
434e245d | 4131 | validate_slab_slab(s, page, map); |
53e15af0 CL |
4132 | count++; |
4133 | } | |
4134 | if (count != n->nr_partial) | |
f9f58285 FF |
4135 | pr_err("SLUB %s: %ld partial slabs counted but counter=%ld\n", |
4136 | s->name, count, n->nr_partial); | |
53e15af0 CL |
4137 | |
4138 | if (!(s->flags & SLAB_STORE_USER)) | |
4139 | goto out; | |
4140 | ||
4141 | list_for_each_entry(page, &n->full, lru) { | |
434e245d | 4142 | validate_slab_slab(s, page, map); |
53e15af0 CL |
4143 | count++; |
4144 | } | |
4145 | if (count != atomic_long_read(&n->nr_slabs)) | |
f9f58285 FF |
4146 | pr_err("SLUB: %s %ld slabs counted but counter=%ld\n", |
4147 | s->name, count, atomic_long_read(&n->nr_slabs)); | |
53e15af0 CL |
4148 | |
4149 | out: | |
4150 | spin_unlock_irqrestore(&n->list_lock, flags); | |
4151 | return count; | |
4152 | } | |
4153 | ||
434e245d | 4154 | static long validate_slab_cache(struct kmem_cache *s) |
53e15af0 CL |
4155 | { |
4156 | int node; | |
4157 | unsigned long count = 0; | |
205ab99d | 4158 | unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) * |
434e245d | 4159 | sizeof(unsigned long), GFP_KERNEL); |
fa45dc25 | 4160 | struct kmem_cache_node *n; |
434e245d CL |
4161 | |
4162 | if (!map) | |
4163 | return -ENOMEM; | |
53e15af0 CL |
4164 | |
4165 | flush_all(s); | |
fa45dc25 | 4166 | for_each_kmem_cache_node(s, node, n) |
434e245d | 4167 | count += validate_slab_node(s, n, map); |
434e245d | 4168 | kfree(map); |
53e15af0 CL |
4169 | return count; |
4170 | } | |
88a420e4 | 4171 | /* |
672bba3a | 4172 | * Generate lists of code addresses where slabcache objects are allocated |
88a420e4 CL |
4173 | * and freed. |
4174 | */ | |
4175 | ||
4176 | struct location { | |
4177 | unsigned long count; | |
ce71e27c | 4178 | unsigned long addr; |
45edfa58 CL |
4179 | long long sum_time; |
4180 | long min_time; | |
4181 | long max_time; | |
4182 | long min_pid; | |
4183 | long max_pid; | |
174596a0 | 4184 | DECLARE_BITMAP(cpus, NR_CPUS); |
45edfa58 | 4185 | nodemask_t nodes; |
88a420e4 CL |
4186 | }; |
4187 | ||
4188 | struct loc_track { | |
4189 | unsigned long max; | |
4190 | unsigned long count; | |
4191 | struct location *loc; | |
4192 | }; | |
4193 | ||
4194 | static void free_loc_track(struct loc_track *t) | |
4195 | { | |
4196 | if (t->max) | |
4197 | free_pages((unsigned long)t->loc, | |
4198 | get_order(sizeof(struct location) * t->max)); | |
4199 | } | |
4200 | ||
68dff6a9 | 4201 | static int alloc_loc_track(struct loc_track *t, unsigned long max, gfp_t flags) |
88a420e4 CL |
4202 | { |
4203 | struct location *l; | |
4204 | int order; | |
4205 | ||
88a420e4 CL |
4206 | order = get_order(sizeof(struct location) * max); |
4207 | ||
68dff6a9 | 4208 | l = (void *)__get_free_pages(flags, order); |
88a420e4 CL |
4209 | if (!l) |
4210 | return 0; | |
4211 | ||
4212 | if (t->count) { | |
4213 | memcpy(l, t->loc, sizeof(struct location) * t->count); | |
4214 | free_loc_track(t); | |
4215 | } | |
4216 | t->max = max; | |
4217 | t->loc = l; | |
4218 | return 1; | |
4219 | } | |
4220 | ||
4221 | static int add_location(struct loc_track *t, struct kmem_cache *s, | |
45edfa58 | 4222 | const struct track *track) |
88a420e4 CL |
4223 | { |
4224 | long start, end, pos; | |
4225 | struct location *l; | |
ce71e27c | 4226 | unsigned long caddr; |
45edfa58 | 4227 | unsigned long age = jiffies - track->when; |
88a420e4 CL |
4228 | |
4229 | start = -1; | |
4230 | end = t->count; | |
4231 | ||
4232 | for ( ; ; ) { | |
4233 | pos = start + (end - start + 1) / 2; | |
4234 | ||
4235 | /* | |
4236 | * There is nothing at "end". If we end up there | |
4237 | * we need to add something to before end. | |
4238 | */ | |
4239 | if (pos == end) | |
4240 | break; | |
4241 | ||
4242 | caddr = t->loc[pos].addr; | |
45edfa58 CL |
4243 | if (track->addr == caddr) { |
4244 | ||
4245 | l = &t->loc[pos]; | |
4246 | l->count++; | |
4247 | if (track->when) { | |
4248 | l->sum_time += age; | |
4249 | if (age < l->min_time) | |
4250 | l->min_time = age; | |
4251 | if (age > l->max_time) | |
4252 | l->max_time = age; | |
4253 | ||
4254 | if (track->pid < l->min_pid) | |
4255 | l->min_pid = track->pid; | |
4256 | if (track->pid > l->max_pid) | |
4257 | l->max_pid = track->pid; | |
4258 | ||
174596a0 RR |
4259 | cpumask_set_cpu(track->cpu, |
4260 | to_cpumask(l->cpus)); | |
45edfa58 CL |
4261 | } |
4262 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
88a420e4 CL |
4263 | return 1; |
4264 | } | |
4265 | ||
45edfa58 | 4266 | if (track->addr < caddr) |
88a420e4 CL |
4267 | end = pos; |
4268 | else | |
4269 | start = pos; | |
4270 | } | |
4271 | ||
4272 | /* | |
672bba3a | 4273 | * Not found. Insert new tracking element. |
88a420e4 | 4274 | */ |
68dff6a9 | 4275 | if (t->count >= t->max && !alloc_loc_track(t, 2 * t->max, GFP_ATOMIC)) |
88a420e4 CL |
4276 | return 0; |
4277 | ||
4278 | l = t->loc + pos; | |
4279 | if (pos < t->count) | |
4280 | memmove(l + 1, l, | |
4281 | (t->count - pos) * sizeof(struct location)); | |
4282 | t->count++; | |
4283 | l->count = 1; | |
45edfa58 CL |
4284 | l->addr = track->addr; |
4285 | l->sum_time = age; | |
4286 | l->min_time = age; | |
4287 | l->max_time = age; | |
4288 | l->min_pid = track->pid; | |
4289 | l->max_pid = track->pid; | |
174596a0 RR |
4290 | cpumask_clear(to_cpumask(l->cpus)); |
4291 | cpumask_set_cpu(track->cpu, to_cpumask(l->cpus)); | |
45edfa58 CL |
4292 | nodes_clear(l->nodes); |
4293 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
88a420e4 CL |
4294 | return 1; |
4295 | } | |
4296 | ||
4297 | static void process_slab(struct loc_track *t, struct kmem_cache *s, | |
bbd7d57b | 4298 | struct page *page, enum track_item alloc, |
a5dd5c11 | 4299 | unsigned long *map) |
88a420e4 | 4300 | { |
a973e9dd | 4301 | void *addr = page_address(page); |
88a420e4 CL |
4302 | void *p; |
4303 | ||
39b26464 | 4304 | bitmap_zero(map, page->objects); |
5f80b13a | 4305 | get_map(s, page, map); |
88a420e4 | 4306 | |
224a88be | 4307 | for_each_object(p, s, addr, page->objects) |
45edfa58 CL |
4308 | if (!test_bit(slab_index(p, s, addr), map)) |
4309 | add_location(t, s, get_track(s, p, alloc)); | |
88a420e4 CL |
4310 | } |
4311 | ||
4312 | static int list_locations(struct kmem_cache *s, char *buf, | |
4313 | enum track_item alloc) | |
4314 | { | |
e374d483 | 4315 | int len = 0; |
88a420e4 | 4316 | unsigned long i; |
68dff6a9 | 4317 | struct loc_track t = { 0, 0, NULL }; |
88a420e4 | 4318 | int node; |
bbd7d57b ED |
4319 | unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) * |
4320 | sizeof(unsigned long), GFP_KERNEL); | |
fa45dc25 | 4321 | struct kmem_cache_node *n; |
88a420e4 | 4322 | |
bbd7d57b ED |
4323 | if (!map || !alloc_loc_track(&t, PAGE_SIZE / sizeof(struct location), |
4324 | GFP_TEMPORARY)) { | |
4325 | kfree(map); | |
68dff6a9 | 4326 | return sprintf(buf, "Out of memory\n"); |
bbd7d57b | 4327 | } |
88a420e4 CL |
4328 | /* Push back cpu slabs */ |
4329 | flush_all(s); | |
4330 | ||
fa45dc25 | 4331 | for_each_kmem_cache_node(s, node, n) { |
88a420e4 CL |
4332 | unsigned long flags; |
4333 | struct page *page; | |
4334 | ||
9e86943b | 4335 | if (!atomic_long_read(&n->nr_slabs)) |
88a420e4 CL |
4336 | continue; |
4337 | ||
4338 | spin_lock_irqsave(&n->list_lock, flags); | |
4339 | list_for_each_entry(page, &n->partial, lru) | |
bbd7d57b | 4340 | process_slab(&t, s, page, alloc, map); |
88a420e4 | 4341 | list_for_each_entry(page, &n->full, lru) |
bbd7d57b | 4342 | process_slab(&t, s, page, alloc, map); |
88a420e4 CL |
4343 | spin_unlock_irqrestore(&n->list_lock, flags); |
4344 | } | |
4345 | ||
4346 | for (i = 0; i < t.count; i++) { | |
45edfa58 | 4347 | struct location *l = &t.loc[i]; |
88a420e4 | 4348 | |
9c246247 | 4349 | if (len > PAGE_SIZE - KSYM_SYMBOL_LEN - 100) |
88a420e4 | 4350 | break; |
e374d483 | 4351 | len += sprintf(buf + len, "%7ld ", l->count); |
45edfa58 CL |
4352 | |
4353 | if (l->addr) | |
62c70bce | 4354 | len += sprintf(buf + len, "%pS", (void *)l->addr); |
88a420e4 | 4355 | else |
e374d483 | 4356 | len += sprintf(buf + len, "<not-available>"); |
45edfa58 CL |
4357 | |
4358 | if (l->sum_time != l->min_time) { | |
e374d483 | 4359 | len += sprintf(buf + len, " age=%ld/%ld/%ld", |
f8bd2258 RZ |
4360 | l->min_time, |
4361 | (long)div_u64(l->sum_time, l->count), | |
4362 | l->max_time); | |
45edfa58 | 4363 | } else |
e374d483 | 4364 | len += sprintf(buf + len, " age=%ld", |
45edfa58 CL |
4365 | l->min_time); |
4366 | ||
4367 | if (l->min_pid != l->max_pid) | |
e374d483 | 4368 | len += sprintf(buf + len, " pid=%ld-%ld", |
45edfa58 CL |
4369 | l->min_pid, l->max_pid); |
4370 | else | |
e374d483 | 4371 | len += sprintf(buf + len, " pid=%ld", |
45edfa58 CL |
4372 | l->min_pid); |
4373 | ||
174596a0 RR |
4374 | if (num_online_cpus() > 1 && |
4375 | !cpumask_empty(to_cpumask(l->cpus)) && | |
5024c1d7 TH |
4376 | len < PAGE_SIZE - 60) |
4377 | len += scnprintf(buf + len, PAGE_SIZE - len - 50, | |
4378 | " cpus=%*pbl", | |
4379 | cpumask_pr_args(to_cpumask(l->cpus))); | |
45edfa58 | 4380 | |
62bc62a8 | 4381 | if (nr_online_nodes > 1 && !nodes_empty(l->nodes) && |
5024c1d7 TH |
4382 | len < PAGE_SIZE - 60) |
4383 | len += scnprintf(buf + len, PAGE_SIZE - len - 50, | |
4384 | " nodes=%*pbl", | |
4385 | nodemask_pr_args(&l->nodes)); | |
45edfa58 | 4386 | |
e374d483 | 4387 | len += sprintf(buf + len, "\n"); |
88a420e4 CL |
4388 | } |
4389 | ||
4390 | free_loc_track(&t); | |
bbd7d57b | 4391 | kfree(map); |
88a420e4 | 4392 | if (!t.count) |
e374d483 HH |
4393 | len += sprintf(buf, "No data\n"); |
4394 | return len; | |
88a420e4 | 4395 | } |
ab4d5ed5 | 4396 | #endif |
88a420e4 | 4397 | |
a5a84755 | 4398 | #ifdef SLUB_RESILIENCY_TEST |
c07b8183 | 4399 | static void __init resiliency_test(void) |
a5a84755 CL |
4400 | { |
4401 | u8 *p; | |
4402 | ||
95a05b42 | 4403 | BUILD_BUG_ON(KMALLOC_MIN_SIZE > 16 || KMALLOC_SHIFT_HIGH < 10); |
a5a84755 | 4404 | |
f9f58285 FF |
4405 | pr_err("SLUB resiliency testing\n"); |
4406 | pr_err("-----------------------\n"); | |
4407 | pr_err("A. Corruption after allocation\n"); | |
a5a84755 CL |
4408 | |
4409 | p = kzalloc(16, GFP_KERNEL); | |
4410 | p[16] = 0x12; | |
f9f58285 FF |
4411 | pr_err("\n1. kmalloc-16: Clobber Redzone/next pointer 0x12->0x%p\n\n", |
4412 | p + 16); | |
a5a84755 CL |
4413 | |
4414 | validate_slab_cache(kmalloc_caches[4]); | |
4415 | ||
4416 | /* Hmmm... The next two are dangerous */ | |
4417 | p = kzalloc(32, GFP_KERNEL); | |
4418 | p[32 + sizeof(void *)] = 0x34; | |
f9f58285 FF |
4419 | pr_err("\n2. kmalloc-32: Clobber next pointer/next slab 0x34 -> -0x%p\n", |
4420 | p); | |
4421 | pr_err("If allocated object is overwritten then not detectable\n\n"); | |
a5a84755 CL |
4422 | |
4423 | validate_slab_cache(kmalloc_caches[5]); | |
4424 | p = kzalloc(64, GFP_KERNEL); | |
4425 | p += 64 + (get_cycles() & 0xff) * sizeof(void *); | |
4426 | *p = 0x56; | |
f9f58285 FF |
4427 | pr_err("\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n", |
4428 | p); | |
4429 | pr_err("If allocated object is overwritten then not detectable\n\n"); | |
a5a84755 CL |
4430 | validate_slab_cache(kmalloc_caches[6]); |
4431 | ||
f9f58285 | 4432 | pr_err("\nB. Corruption after free\n"); |
a5a84755 CL |
4433 | p = kzalloc(128, GFP_KERNEL); |
4434 | kfree(p); | |
4435 | *p = 0x78; | |
f9f58285 | 4436 | pr_err("1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p); |
a5a84755 CL |
4437 | validate_slab_cache(kmalloc_caches[7]); |
4438 | ||
4439 | p = kzalloc(256, GFP_KERNEL); | |
4440 | kfree(p); | |
4441 | p[50] = 0x9a; | |
f9f58285 | 4442 | pr_err("\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n", p); |
a5a84755 CL |
4443 | validate_slab_cache(kmalloc_caches[8]); |
4444 | ||
4445 | p = kzalloc(512, GFP_KERNEL); | |
4446 | kfree(p); | |
4447 | p[512] = 0xab; | |
f9f58285 | 4448 | pr_err("\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p); |
a5a84755 CL |
4449 | validate_slab_cache(kmalloc_caches[9]); |
4450 | } | |
4451 | #else | |
4452 | #ifdef CONFIG_SYSFS | |
4453 | static void resiliency_test(void) {}; | |
4454 | #endif | |
4455 | #endif | |
4456 | ||
ab4d5ed5 | 4457 | #ifdef CONFIG_SYSFS |
81819f0f | 4458 | enum slab_stat_type { |
205ab99d CL |
4459 | SL_ALL, /* All slabs */ |
4460 | SL_PARTIAL, /* Only partially allocated slabs */ | |
4461 | SL_CPU, /* Only slabs used for cpu caches */ | |
4462 | SL_OBJECTS, /* Determine allocated objects not slabs */ | |
4463 | SL_TOTAL /* Determine object capacity not slabs */ | |
81819f0f CL |
4464 | }; |
4465 | ||
205ab99d | 4466 | #define SO_ALL (1 << SL_ALL) |
81819f0f CL |
4467 | #define SO_PARTIAL (1 << SL_PARTIAL) |
4468 | #define SO_CPU (1 << SL_CPU) | |
4469 | #define SO_OBJECTS (1 << SL_OBJECTS) | |
205ab99d | 4470 | #define SO_TOTAL (1 << SL_TOTAL) |
81819f0f | 4471 | |
62e5c4b4 CG |
4472 | static ssize_t show_slab_objects(struct kmem_cache *s, |
4473 | char *buf, unsigned long flags) | |
81819f0f CL |
4474 | { |
4475 | unsigned long total = 0; | |
81819f0f CL |
4476 | int node; |
4477 | int x; | |
4478 | unsigned long *nodes; | |
81819f0f | 4479 | |
e35e1a97 | 4480 | nodes = kzalloc(sizeof(unsigned long) * nr_node_ids, GFP_KERNEL); |
62e5c4b4 CG |
4481 | if (!nodes) |
4482 | return -ENOMEM; | |
81819f0f | 4483 | |
205ab99d CL |
4484 | if (flags & SO_CPU) { |
4485 | int cpu; | |
81819f0f | 4486 | |
205ab99d | 4487 | for_each_possible_cpu(cpu) { |
d0e0ac97 CG |
4488 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, |
4489 | cpu); | |
ec3ab083 | 4490 | int node; |
49e22585 | 4491 | struct page *page; |
dfb4f096 | 4492 | |
4db0c3c2 | 4493 | page = READ_ONCE(c->page); |
ec3ab083 CL |
4494 | if (!page) |
4495 | continue; | |
205ab99d | 4496 | |
ec3ab083 CL |
4497 | node = page_to_nid(page); |
4498 | if (flags & SO_TOTAL) | |
4499 | x = page->objects; | |
4500 | else if (flags & SO_OBJECTS) | |
4501 | x = page->inuse; | |
4502 | else | |
4503 | x = 1; | |
49e22585 | 4504 | |
ec3ab083 CL |
4505 | total += x; |
4506 | nodes[node] += x; | |
4507 | ||
4db0c3c2 | 4508 | page = READ_ONCE(c->partial); |
49e22585 | 4509 | if (page) { |
8afb1474 LZ |
4510 | node = page_to_nid(page); |
4511 | if (flags & SO_TOTAL) | |
4512 | WARN_ON_ONCE(1); | |
4513 | else if (flags & SO_OBJECTS) | |
4514 | WARN_ON_ONCE(1); | |
4515 | else | |
4516 | x = page->pages; | |
bc6697d8 ED |
4517 | total += x; |
4518 | nodes[node] += x; | |
49e22585 | 4519 | } |
81819f0f CL |
4520 | } |
4521 | } | |
4522 | ||
bfc8c901 | 4523 | get_online_mems(); |
ab4d5ed5 | 4524 | #ifdef CONFIG_SLUB_DEBUG |
205ab99d | 4525 | if (flags & SO_ALL) { |
fa45dc25 CL |
4526 | struct kmem_cache_node *n; |
4527 | ||
4528 | for_each_kmem_cache_node(s, node, n) { | |
205ab99d | 4529 | |
d0e0ac97 CG |
4530 | if (flags & SO_TOTAL) |
4531 | x = atomic_long_read(&n->total_objects); | |
4532 | else if (flags & SO_OBJECTS) | |
4533 | x = atomic_long_read(&n->total_objects) - | |
4534 | count_partial(n, count_free); | |
81819f0f | 4535 | else |
205ab99d | 4536 | x = atomic_long_read(&n->nr_slabs); |
81819f0f CL |
4537 | total += x; |
4538 | nodes[node] += x; | |
4539 | } | |
4540 | ||
ab4d5ed5 CL |
4541 | } else |
4542 | #endif | |
4543 | if (flags & SO_PARTIAL) { | |
fa45dc25 | 4544 | struct kmem_cache_node *n; |
81819f0f | 4545 | |
fa45dc25 | 4546 | for_each_kmem_cache_node(s, node, n) { |
205ab99d CL |
4547 | if (flags & SO_TOTAL) |
4548 | x = count_partial(n, count_total); | |
4549 | else if (flags & SO_OBJECTS) | |
4550 | x = count_partial(n, count_inuse); | |
81819f0f | 4551 | else |
205ab99d | 4552 | x = n->nr_partial; |
81819f0f CL |
4553 | total += x; |
4554 | nodes[node] += x; | |
4555 | } | |
4556 | } | |
81819f0f CL |
4557 | x = sprintf(buf, "%lu", total); |
4558 | #ifdef CONFIG_NUMA | |
fa45dc25 | 4559 | for (node = 0; node < nr_node_ids; node++) |
81819f0f CL |
4560 | if (nodes[node]) |
4561 | x += sprintf(buf + x, " N%d=%lu", | |
4562 | node, nodes[node]); | |
4563 | #endif | |
bfc8c901 | 4564 | put_online_mems(); |
81819f0f CL |
4565 | kfree(nodes); |
4566 | return x + sprintf(buf + x, "\n"); | |
4567 | } | |
4568 | ||
ab4d5ed5 | 4569 | #ifdef CONFIG_SLUB_DEBUG |
81819f0f CL |
4570 | static int any_slab_objects(struct kmem_cache *s) |
4571 | { | |
4572 | int node; | |
fa45dc25 | 4573 | struct kmem_cache_node *n; |
81819f0f | 4574 | |
fa45dc25 | 4575 | for_each_kmem_cache_node(s, node, n) |
4ea33e2d | 4576 | if (atomic_long_read(&n->total_objects)) |
81819f0f | 4577 | return 1; |
fa45dc25 | 4578 | |
81819f0f CL |
4579 | return 0; |
4580 | } | |
ab4d5ed5 | 4581 | #endif |
81819f0f CL |
4582 | |
4583 | #define to_slab_attr(n) container_of(n, struct slab_attribute, attr) | |
497888cf | 4584 | #define to_slab(n) container_of(n, struct kmem_cache, kobj) |
81819f0f CL |
4585 | |
4586 | struct slab_attribute { | |
4587 | struct attribute attr; | |
4588 | ssize_t (*show)(struct kmem_cache *s, char *buf); | |
4589 | ssize_t (*store)(struct kmem_cache *s, const char *x, size_t count); | |
4590 | }; | |
4591 | ||
4592 | #define SLAB_ATTR_RO(_name) \ | |
ab067e99 VK |
4593 | static struct slab_attribute _name##_attr = \ |
4594 | __ATTR(_name, 0400, _name##_show, NULL) | |
81819f0f CL |
4595 | |
4596 | #define SLAB_ATTR(_name) \ | |
4597 | static struct slab_attribute _name##_attr = \ | |
ab067e99 | 4598 | __ATTR(_name, 0600, _name##_show, _name##_store) |
81819f0f | 4599 | |
81819f0f CL |
4600 | static ssize_t slab_size_show(struct kmem_cache *s, char *buf) |
4601 | { | |
4602 | return sprintf(buf, "%d\n", s->size); | |
4603 | } | |
4604 | SLAB_ATTR_RO(slab_size); | |
4605 | ||
4606 | static ssize_t align_show(struct kmem_cache *s, char *buf) | |
4607 | { | |
4608 | return sprintf(buf, "%d\n", s->align); | |
4609 | } | |
4610 | SLAB_ATTR_RO(align); | |
4611 | ||
4612 | static ssize_t object_size_show(struct kmem_cache *s, char *buf) | |
4613 | { | |
3b0efdfa | 4614 | return sprintf(buf, "%d\n", s->object_size); |
81819f0f CL |
4615 | } |
4616 | SLAB_ATTR_RO(object_size); | |
4617 | ||
4618 | static ssize_t objs_per_slab_show(struct kmem_cache *s, char *buf) | |
4619 | { | |
834f3d11 | 4620 | return sprintf(buf, "%d\n", oo_objects(s->oo)); |
81819f0f CL |
4621 | } |
4622 | SLAB_ATTR_RO(objs_per_slab); | |
4623 | ||
06b285dc CL |
4624 | static ssize_t order_store(struct kmem_cache *s, |
4625 | const char *buf, size_t length) | |
4626 | { | |
0121c619 CL |
4627 | unsigned long order; |
4628 | int err; | |
4629 | ||
3dbb95f7 | 4630 | err = kstrtoul(buf, 10, &order); |
0121c619 CL |
4631 | if (err) |
4632 | return err; | |
06b285dc CL |
4633 | |
4634 | if (order > slub_max_order || order < slub_min_order) | |
4635 | return -EINVAL; | |
4636 | ||
4637 | calculate_sizes(s, order); | |
4638 | return length; | |
4639 | } | |
4640 | ||
81819f0f CL |
4641 | static ssize_t order_show(struct kmem_cache *s, char *buf) |
4642 | { | |
834f3d11 | 4643 | return sprintf(buf, "%d\n", oo_order(s->oo)); |
81819f0f | 4644 | } |
06b285dc | 4645 | SLAB_ATTR(order); |
81819f0f | 4646 | |
73d342b1 DR |
4647 | static ssize_t min_partial_show(struct kmem_cache *s, char *buf) |
4648 | { | |
4649 | return sprintf(buf, "%lu\n", s->min_partial); | |
4650 | } | |
4651 | ||
4652 | static ssize_t min_partial_store(struct kmem_cache *s, const char *buf, | |
4653 | size_t length) | |
4654 | { | |
4655 | unsigned long min; | |
4656 | int err; | |
4657 | ||
3dbb95f7 | 4658 | err = kstrtoul(buf, 10, &min); |
73d342b1 DR |
4659 | if (err) |
4660 | return err; | |
4661 | ||
c0bdb232 | 4662 | set_min_partial(s, min); |
73d342b1 DR |
4663 | return length; |
4664 | } | |
4665 | SLAB_ATTR(min_partial); | |
4666 | ||
49e22585 CL |
4667 | static ssize_t cpu_partial_show(struct kmem_cache *s, char *buf) |
4668 | { | |
4669 | return sprintf(buf, "%u\n", s->cpu_partial); | |
4670 | } | |
4671 | ||
4672 | static ssize_t cpu_partial_store(struct kmem_cache *s, const char *buf, | |
4673 | size_t length) | |
4674 | { | |
4675 | unsigned long objects; | |
4676 | int err; | |
4677 | ||
3dbb95f7 | 4678 | err = kstrtoul(buf, 10, &objects); |
49e22585 CL |
4679 | if (err) |
4680 | return err; | |
345c905d | 4681 | if (objects && !kmem_cache_has_cpu_partial(s)) |
74ee4ef1 | 4682 | return -EINVAL; |
49e22585 CL |
4683 | |
4684 | s->cpu_partial = objects; | |
4685 | flush_all(s); | |
4686 | return length; | |
4687 | } | |
4688 | SLAB_ATTR(cpu_partial); | |
4689 | ||
81819f0f CL |
4690 | static ssize_t ctor_show(struct kmem_cache *s, char *buf) |
4691 | { | |
62c70bce JP |
4692 | if (!s->ctor) |
4693 | return 0; | |
4694 | return sprintf(buf, "%pS\n", s->ctor); | |
81819f0f CL |
4695 | } |
4696 | SLAB_ATTR_RO(ctor); | |
4697 | ||
81819f0f CL |
4698 | static ssize_t aliases_show(struct kmem_cache *s, char *buf) |
4699 | { | |
4307c14f | 4700 | return sprintf(buf, "%d\n", s->refcount < 0 ? 0 : s->refcount - 1); |
81819f0f CL |
4701 | } |
4702 | SLAB_ATTR_RO(aliases); | |
4703 | ||
81819f0f CL |
4704 | static ssize_t partial_show(struct kmem_cache *s, char *buf) |
4705 | { | |
d9acf4b7 | 4706 | return show_slab_objects(s, buf, SO_PARTIAL); |
81819f0f CL |
4707 | } |
4708 | SLAB_ATTR_RO(partial); | |
4709 | ||
4710 | static ssize_t cpu_slabs_show(struct kmem_cache *s, char *buf) | |
4711 | { | |
d9acf4b7 | 4712 | return show_slab_objects(s, buf, SO_CPU); |
81819f0f CL |
4713 | } |
4714 | SLAB_ATTR_RO(cpu_slabs); | |
4715 | ||
4716 | static ssize_t objects_show(struct kmem_cache *s, char *buf) | |
4717 | { | |
205ab99d | 4718 | return show_slab_objects(s, buf, SO_ALL|SO_OBJECTS); |
81819f0f CL |
4719 | } |
4720 | SLAB_ATTR_RO(objects); | |
4721 | ||
205ab99d CL |
4722 | static ssize_t objects_partial_show(struct kmem_cache *s, char *buf) |
4723 | { | |
4724 | return show_slab_objects(s, buf, SO_PARTIAL|SO_OBJECTS); | |
4725 | } | |
4726 | SLAB_ATTR_RO(objects_partial); | |
4727 | ||
49e22585 CL |
4728 | static ssize_t slabs_cpu_partial_show(struct kmem_cache *s, char *buf) |
4729 | { | |
4730 | int objects = 0; | |
4731 | int pages = 0; | |
4732 | int cpu; | |
4733 | int len; | |
4734 | ||
4735 | for_each_online_cpu(cpu) { | |
4736 | struct page *page = per_cpu_ptr(s->cpu_slab, cpu)->partial; | |
4737 | ||
4738 | if (page) { | |
4739 | pages += page->pages; | |
4740 | objects += page->pobjects; | |
4741 | } | |
4742 | } | |
4743 | ||
4744 | len = sprintf(buf, "%d(%d)", objects, pages); | |
4745 | ||
4746 | #ifdef CONFIG_SMP | |
4747 | for_each_online_cpu(cpu) { | |
4748 | struct page *page = per_cpu_ptr(s->cpu_slab, cpu) ->partial; | |
4749 | ||
4750 | if (page && len < PAGE_SIZE - 20) | |
4751 | len += sprintf(buf + len, " C%d=%d(%d)", cpu, | |
4752 | page->pobjects, page->pages); | |
4753 | } | |
4754 | #endif | |
4755 | return len + sprintf(buf + len, "\n"); | |
4756 | } | |
4757 | SLAB_ATTR_RO(slabs_cpu_partial); | |
4758 | ||
a5a84755 CL |
4759 | static ssize_t reclaim_account_show(struct kmem_cache *s, char *buf) |
4760 | { | |
4761 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_RECLAIM_ACCOUNT)); | |
4762 | } | |
4763 | ||
4764 | static ssize_t reclaim_account_store(struct kmem_cache *s, | |
4765 | const char *buf, size_t length) | |
4766 | { | |
4767 | s->flags &= ~SLAB_RECLAIM_ACCOUNT; | |
4768 | if (buf[0] == '1') | |
4769 | s->flags |= SLAB_RECLAIM_ACCOUNT; | |
4770 | return length; | |
4771 | } | |
4772 | SLAB_ATTR(reclaim_account); | |
4773 | ||
4774 | static ssize_t hwcache_align_show(struct kmem_cache *s, char *buf) | |
4775 | { | |
4776 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_HWCACHE_ALIGN)); | |
4777 | } | |
4778 | SLAB_ATTR_RO(hwcache_align); | |
4779 | ||
4780 | #ifdef CONFIG_ZONE_DMA | |
4781 | static ssize_t cache_dma_show(struct kmem_cache *s, char *buf) | |
4782 | { | |
4783 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_CACHE_DMA)); | |
4784 | } | |
4785 | SLAB_ATTR_RO(cache_dma); | |
4786 | #endif | |
4787 | ||
4788 | static ssize_t destroy_by_rcu_show(struct kmem_cache *s, char *buf) | |
4789 | { | |
4790 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_DESTROY_BY_RCU)); | |
4791 | } | |
4792 | SLAB_ATTR_RO(destroy_by_rcu); | |
4793 | ||
ab9a0f19 LJ |
4794 | static ssize_t reserved_show(struct kmem_cache *s, char *buf) |
4795 | { | |
4796 | return sprintf(buf, "%d\n", s->reserved); | |
4797 | } | |
4798 | SLAB_ATTR_RO(reserved); | |
4799 | ||
ab4d5ed5 | 4800 | #ifdef CONFIG_SLUB_DEBUG |
a5a84755 CL |
4801 | static ssize_t slabs_show(struct kmem_cache *s, char *buf) |
4802 | { | |
4803 | return show_slab_objects(s, buf, SO_ALL); | |
4804 | } | |
4805 | SLAB_ATTR_RO(slabs); | |
4806 | ||
205ab99d CL |
4807 | static ssize_t total_objects_show(struct kmem_cache *s, char *buf) |
4808 | { | |
4809 | return show_slab_objects(s, buf, SO_ALL|SO_TOTAL); | |
4810 | } | |
4811 | SLAB_ATTR_RO(total_objects); | |
4812 | ||
81819f0f CL |
4813 | static ssize_t sanity_checks_show(struct kmem_cache *s, char *buf) |
4814 | { | |
4815 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_DEBUG_FREE)); | |
4816 | } | |
4817 | ||
4818 | static ssize_t sanity_checks_store(struct kmem_cache *s, | |
4819 | const char *buf, size_t length) | |
4820 | { | |
4821 | s->flags &= ~SLAB_DEBUG_FREE; | |
b789ef51 CL |
4822 | if (buf[0] == '1') { |
4823 | s->flags &= ~__CMPXCHG_DOUBLE; | |
81819f0f | 4824 | s->flags |= SLAB_DEBUG_FREE; |
b789ef51 | 4825 | } |
81819f0f CL |
4826 | return length; |
4827 | } | |
4828 | SLAB_ATTR(sanity_checks); | |
4829 | ||
4830 | static ssize_t trace_show(struct kmem_cache *s, char *buf) | |
4831 | { | |
4832 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_TRACE)); | |
4833 | } | |
4834 | ||
4835 | static ssize_t trace_store(struct kmem_cache *s, const char *buf, | |
4836 | size_t length) | |
4837 | { | |
c9e16131 CL |
4838 | /* |
4839 | * Tracing a merged cache is going to give confusing results | |
4840 | * as well as cause other issues like converting a mergeable | |
4841 | * cache into an umergeable one. | |
4842 | */ | |
4843 | if (s->refcount > 1) | |
4844 | return -EINVAL; | |
4845 | ||
81819f0f | 4846 | s->flags &= ~SLAB_TRACE; |
b789ef51 CL |
4847 | if (buf[0] == '1') { |
4848 | s->flags &= ~__CMPXCHG_DOUBLE; | |
81819f0f | 4849 | s->flags |= SLAB_TRACE; |
b789ef51 | 4850 | } |
81819f0f CL |
4851 | return length; |
4852 | } | |
4853 | SLAB_ATTR(trace); | |
4854 | ||
81819f0f CL |
4855 | static ssize_t red_zone_show(struct kmem_cache *s, char *buf) |
4856 | { | |
4857 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_RED_ZONE)); | |
4858 | } | |
4859 | ||
4860 | static ssize_t red_zone_store(struct kmem_cache *s, | |
4861 | const char *buf, size_t length) | |
4862 | { | |
4863 | if (any_slab_objects(s)) | |
4864 | return -EBUSY; | |
4865 | ||
4866 | s->flags &= ~SLAB_RED_ZONE; | |
b789ef51 CL |
4867 | if (buf[0] == '1') { |
4868 | s->flags &= ~__CMPXCHG_DOUBLE; | |
81819f0f | 4869 | s->flags |= SLAB_RED_ZONE; |
b789ef51 | 4870 | } |
06b285dc | 4871 | calculate_sizes(s, -1); |
81819f0f CL |
4872 | return length; |
4873 | } | |
4874 | SLAB_ATTR(red_zone); | |
4875 | ||
4876 | static ssize_t poison_show(struct kmem_cache *s, char *buf) | |
4877 | { | |
4878 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_POISON)); | |
4879 | } | |
4880 | ||
4881 | static ssize_t poison_store(struct kmem_cache *s, | |
4882 | const char *buf, size_t length) | |
4883 | { | |
4884 | if (any_slab_objects(s)) | |
4885 | return -EBUSY; | |
4886 | ||
4887 | s->flags &= ~SLAB_POISON; | |
b789ef51 CL |
4888 | if (buf[0] == '1') { |
4889 | s->flags &= ~__CMPXCHG_DOUBLE; | |
81819f0f | 4890 | s->flags |= SLAB_POISON; |
b789ef51 | 4891 | } |
06b285dc | 4892 | calculate_sizes(s, -1); |
81819f0f CL |
4893 | return length; |
4894 | } | |
4895 | SLAB_ATTR(poison); | |
4896 | ||
4897 | static ssize_t store_user_show(struct kmem_cache *s, char *buf) | |
4898 | { | |
4899 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_STORE_USER)); | |
4900 | } | |
4901 | ||
4902 | static ssize_t store_user_store(struct kmem_cache *s, | |
4903 | const char *buf, size_t length) | |
4904 | { | |
4905 | if (any_slab_objects(s)) | |
4906 | return -EBUSY; | |
4907 | ||
4908 | s->flags &= ~SLAB_STORE_USER; | |
b789ef51 CL |
4909 | if (buf[0] == '1') { |
4910 | s->flags &= ~__CMPXCHG_DOUBLE; | |
81819f0f | 4911 | s->flags |= SLAB_STORE_USER; |
b789ef51 | 4912 | } |
06b285dc | 4913 | calculate_sizes(s, -1); |
81819f0f CL |
4914 | return length; |
4915 | } | |
4916 | SLAB_ATTR(store_user); | |
4917 | ||
53e15af0 CL |
4918 | static ssize_t validate_show(struct kmem_cache *s, char *buf) |
4919 | { | |
4920 | return 0; | |
4921 | } | |
4922 | ||
4923 | static ssize_t validate_store(struct kmem_cache *s, | |
4924 | const char *buf, size_t length) | |
4925 | { | |
434e245d CL |
4926 | int ret = -EINVAL; |
4927 | ||
4928 | if (buf[0] == '1') { | |
4929 | ret = validate_slab_cache(s); | |
4930 | if (ret >= 0) | |
4931 | ret = length; | |
4932 | } | |
4933 | return ret; | |
53e15af0 CL |
4934 | } |
4935 | SLAB_ATTR(validate); | |
a5a84755 CL |
4936 | |
4937 | static ssize_t alloc_calls_show(struct kmem_cache *s, char *buf) | |
4938 | { | |
4939 | if (!(s->flags & SLAB_STORE_USER)) | |
4940 | return -ENOSYS; | |
4941 | return list_locations(s, buf, TRACK_ALLOC); | |
4942 | } | |
4943 | SLAB_ATTR_RO(alloc_calls); | |
4944 | ||
4945 | static ssize_t free_calls_show(struct kmem_cache *s, char *buf) | |
4946 | { | |
4947 | if (!(s->flags & SLAB_STORE_USER)) | |
4948 | return -ENOSYS; | |
4949 | return list_locations(s, buf, TRACK_FREE); | |
4950 | } | |
4951 | SLAB_ATTR_RO(free_calls); | |
4952 | #endif /* CONFIG_SLUB_DEBUG */ | |
4953 | ||
4954 | #ifdef CONFIG_FAILSLAB | |
4955 | static ssize_t failslab_show(struct kmem_cache *s, char *buf) | |
4956 | { | |
4957 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_FAILSLAB)); | |
4958 | } | |
4959 | ||
4960 | static ssize_t failslab_store(struct kmem_cache *s, const char *buf, | |
4961 | size_t length) | |
4962 | { | |
c9e16131 CL |
4963 | if (s->refcount > 1) |
4964 | return -EINVAL; | |
4965 | ||
a5a84755 CL |
4966 | s->flags &= ~SLAB_FAILSLAB; |
4967 | if (buf[0] == '1') | |
4968 | s->flags |= SLAB_FAILSLAB; | |
4969 | return length; | |
4970 | } | |
4971 | SLAB_ATTR(failslab); | |
ab4d5ed5 | 4972 | #endif |
53e15af0 | 4973 | |
2086d26a CL |
4974 | static ssize_t shrink_show(struct kmem_cache *s, char *buf) |
4975 | { | |
4976 | return 0; | |
4977 | } | |
4978 | ||
4979 | static ssize_t shrink_store(struct kmem_cache *s, | |
4980 | const char *buf, size_t length) | |
4981 | { | |
832f37f5 VD |
4982 | if (buf[0] == '1') |
4983 | kmem_cache_shrink(s); | |
4984 | else | |
2086d26a CL |
4985 | return -EINVAL; |
4986 | return length; | |
4987 | } | |
4988 | SLAB_ATTR(shrink); | |
4989 | ||
81819f0f | 4990 | #ifdef CONFIG_NUMA |
9824601e | 4991 | static ssize_t remote_node_defrag_ratio_show(struct kmem_cache *s, char *buf) |
81819f0f | 4992 | { |
9824601e | 4993 | return sprintf(buf, "%d\n", s->remote_node_defrag_ratio / 10); |
81819f0f CL |
4994 | } |
4995 | ||
9824601e | 4996 | static ssize_t remote_node_defrag_ratio_store(struct kmem_cache *s, |
81819f0f CL |
4997 | const char *buf, size_t length) |
4998 | { | |
0121c619 CL |
4999 | unsigned long ratio; |
5000 | int err; | |
5001 | ||
3dbb95f7 | 5002 | err = kstrtoul(buf, 10, &ratio); |
0121c619 CL |
5003 | if (err) |
5004 | return err; | |
5005 | ||
e2cb96b7 | 5006 | if (ratio <= 100) |
0121c619 | 5007 | s->remote_node_defrag_ratio = ratio * 10; |
81819f0f | 5008 | |
81819f0f CL |
5009 | return length; |
5010 | } | |
9824601e | 5011 | SLAB_ATTR(remote_node_defrag_ratio); |
81819f0f CL |
5012 | #endif |
5013 | ||
8ff12cfc | 5014 | #ifdef CONFIG_SLUB_STATS |
8ff12cfc CL |
5015 | static int show_stat(struct kmem_cache *s, char *buf, enum stat_item si) |
5016 | { | |
5017 | unsigned long sum = 0; | |
5018 | int cpu; | |
5019 | int len; | |
5020 | int *data = kmalloc(nr_cpu_ids * sizeof(int), GFP_KERNEL); | |
5021 | ||
5022 | if (!data) | |
5023 | return -ENOMEM; | |
5024 | ||
5025 | for_each_online_cpu(cpu) { | |
9dfc6e68 | 5026 | unsigned x = per_cpu_ptr(s->cpu_slab, cpu)->stat[si]; |
8ff12cfc CL |
5027 | |
5028 | data[cpu] = x; | |
5029 | sum += x; | |
5030 | } | |
5031 | ||
5032 | len = sprintf(buf, "%lu", sum); | |
5033 | ||
50ef37b9 | 5034 | #ifdef CONFIG_SMP |
8ff12cfc CL |
5035 | for_each_online_cpu(cpu) { |
5036 | if (data[cpu] && len < PAGE_SIZE - 20) | |
50ef37b9 | 5037 | len += sprintf(buf + len, " C%d=%u", cpu, data[cpu]); |
8ff12cfc | 5038 | } |
50ef37b9 | 5039 | #endif |
8ff12cfc CL |
5040 | kfree(data); |
5041 | return len + sprintf(buf + len, "\n"); | |
5042 | } | |
5043 | ||
78eb00cc DR |
5044 | static void clear_stat(struct kmem_cache *s, enum stat_item si) |
5045 | { | |
5046 | int cpu; | |
5047 | ||
5048 | for_each_online_cpu(cpu) | |
9dfc6e68 | 5049 | per_cpu_ptr(s->cpu_slab, cpu)->stat[si] = 0; |
78eb00cc DR |
5050 | } |
5051 | ||
8ff12cfc CL |
5052 | #define STAT_ATTR(si, text) \ |
5053 | static ssize_t text##_show(struct kmem_cache *s, char *buf) \ | |
5054 | { \ | |
5055 | return show_stat(s, buf, si); \ | |
5056 | } \ | |
78eb00cc DR |
5057 | static ssize_t text##_store(struct kmem_cache *s, \ |
5058 | const char *buf, size_t length) \ | |
5059 | { \ | |
5060 | if (buf[0] != '0') \ | |
5061 | return -EINVAL; \ | |
5062 | clear_stat(s, si); \ | |
5063 | return length; \ | |
5064 | } \ | |
5065 | SLAB_ATTR(text); \ | |
8ff12cfc CL |
5066 | |
5067 | STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath); | |
5068 | STAT_ATTR(ALLOC_SLOWPATH, alloc_slowpath); | |
5069 | STAT_ATTR(FREE_FASTPATH, free_fastpath); | |
5070 | STAT_ATTR(FREE_SLOWPATH, free_slowpath); | |
5071 | STAT_ATTR(FREE_FROZEN, free_frozen); | |
5072 | STAT_ATTR(FREE_ADD_PARTIAL, free_add_partial); | |
5073 | STAT_ATTR(FREE_REMOVE_PARTIAL, free_remove_partial); | |
5074 | STAT_ATTR(ALLOC_FROM_PARTIAL, alloc_from_partial); | |
5075 | STAT_ATTR(ALLOC_SLAB, alloc_slab); | |
5076 | STAT_ATTR(ALLOC_REFILL, alloc_refill); | |
e36a2652 | 5077 | STAT_ATTR(ALLOC_NODE_MISMATCH, alloc_node_mismatch); |
8ff12cfc CL |
5078 | STAT_ATTR(FREE_SLAB, free_slab); |
5079 | STAT_ATTR(CPUSLAB_FLUSH, cpuslab_flush); | |
5080 | STAT_ATTR(DEACTIVATE_FULL, deactivate_full); | |
5081 | STAT_ATTR(DEACTIVATE_EMPTY, deactivate_empty); | |
5082 | STAT_ATTR(DEACTIVATE_TO_HEAD, deactivate_to_head); | |
5083 | STAT_ATTR(DEACTIVATE_TO_TAIL, deactivate_to_tail); | |
5084 | STAT_ATTR(DEACTIVATE_REMOTE_FREES, deactivate_remote_frees); | |
03e404af | 5085 | STAT_ATTR(DEACTIVATE_BYPASS, deactivate_bypass); |
65c3376a | 5086 | STAT_ATTR(ORDER_FALLBACK, order_fallback); |
b789ef51 CL |
5087 | STAT_ATTR(CMPXCHG_DOUBLE_CPU_FAIL, cmpxchg_double_cpu_fail); |
5088 | STAT_ATTR(CMPXCHG_DOUBLE_FAIL, cmpxchg_double_fail); | |
49e22585 CL |
5089 | STAT_ATTR(CPU_PARTIAL_ALLOC, cpu_partial_alloc); |
5090 | STAT_ATTR(CPU_PARTIAL_FREE, cpu_partial_free); | |
8028dcea AS |
5091 | STAT_ATTR(CPU_PARTIAL_NODE, cpu_partial_node); |
5092 | STAT_ATTR(CPU_PARTIAL_DRAIN, cpu_partial_drain); | |
8ff12cfc CL |
5093 | #endif |
5094 | ||
06428780 | 5095 | static struct attribute *slab_attrs[] = { |
81819f0f CL |
5096 | &slab_size_attr.attr, |
5097 | &object_size_attr.attr, | |
5098 | &objs_per_slab_attr.attr, | |
5099 | &order_attr.attr, | |
73d342b1 | 5100 | &min_partial_attr.attr, |
49e22585 | 5101 | &cpu_partial_attr.attr, |
81819f0f | 5102 | &objects_attr.attr, |
205ab99d | 5103 | &objects_partial_attr.attr, |
81819f0f CL |
5104 | &partial_attr.attr, |
5105 | &cpu_slabs_attr.attr, | |
5106 | &ctor_attr.attr, | |
81819f0f CL |
5107 | &aliases_attr.attr, |
5108 | &align_attr.attr, | |
81819f0f CL |
5109 | &hwcache_align_attr.attr, |
5110 | &reclaim_account_attr.attr, | |
5111 | &destroy_by_rcu_attr.attr, | |
a5a84755 | 5112 | &shrink_attr.attr, |
ab9a0f19 | 5113 | &reserved_attr.attr, |
49e22585 | 5114 | &slabs_cpu_partial_attr.attr, |
ab4d5ed5 | 5115 | #ifdef CONFIG_SLUB_DEBUG |
a5a84755 CL |
5116 | &total_objects_attr.attr, |
5117 | &slabs_attr.attr, | |
5118 | &sanity_checks_attr.attr, | |
5119 | &trace_attr.attr, | |
81819f0f CL |
5120 | &red_zone_attr.attr, |
5121 | &poison_attr.attr, | |
5122 | &store_user_attr.attr, | |
53e15af0 | 5123 | &validate_attr.attr, |
88a420e4 CL |
5124 | &alloc_calls_attr.attr, |
5125 | &free_calls_attr.attr, | |
ab4d5ed5 | 5126 | #endif |
81819f0f CL |
5127 | #ifdef CONFIG_ZONE_DMA |
5128 | &cache_dma_attr.attr, | |
5129 | #endif | |
5130 | #ifdef CONFIG_NUMA | |
9824601e | 5131 | &remote_node_defrag_ratio_attr.attr, |
8ff12cfc CL |
5132 | #endif |
5133 | #ifdef CONFIG_SLUB_STATS | |
5134 | &alloc_fastpath_attr.attr, | |
5135 | &alloc_slowpath_attr.attr, | |
5136 | &free_fastpath_attr.attr, | |
5137 | &free_slowpath_attr.attr, | |
5138 | &free_frozen_attr.attr, | |
5139 | &free_add_partial_attr.attr, | |
5140 | &free_remove_partial_attr.attr, | |
5141 | &alloc_from_partial_attr.attr, | |
5142 | &alloc_slab_attr.attr, | |
5143 | &alloc_refill_attr.attr, | |
e36a2652 | 5144 | &alloc_node_mismatch_attr.attr, |
8ff12cfc CL |
5145 | &free_slab_attr.attr, |
5146 | &cpuslab_flush_attr.attr, | |
5147 | &deactivate_full_attr.attr, | |
5148 | &deactivate_empty_attr.attr, | |
5149 | &deactivate_to_head_attr.attr, | |
5150 | &deactivate_to_tail_attr.attr, | |
5151 | &deactivate_remote_frees_attr.attr, | |
03e404af | 5152 | &deactivate_bypass_attr.attr, |
65c3376a | 5153 | &order_fallback_attr.attr, |
b789ef51 CL |
5154 | &cmpxchg_double_fail_attr.attr, |
5155 | &cmpxchg_double_cpu_fail_attr.attr, | |
49e22585 CL |
5156 | &cpu_partial_alloc_attr.attr, |
5157 | &cpu_partial_free_attr.attr, | |
8028dcea AS |
5158 | &cpu_partial_node_attr.attr, |
5159 | &cpu_partial_drain_attr.attr, | |
81819f0f | 5160 | #endif |
4c13dd3b DM |
5161 | #ifdef CONFIG_FAILSLAB |
5162 | &failslab_attr.attr, | |
5163 | #endif | |
5164 | ||
81819f0f CL |
5165 | NULL |
5166 | }; | |
5167 | ||
5168 | static struct attribute_group slab_attr_group = { | |
5169 | .attrs = slab_attrs, | |
5170 | }; | |
5171 | ||
5172 | static ssize_t slab_attr_show(struct kobject *kobj, | |
5173 | struct attribute *attr, | |
5174 | char *buf) | |
5175 | { | |
5176 | struct slab_attribute *attribute; | |
5177 | struct kmem_cache *s; | |
5178 | int err; | |
5179 | ||
5180 | attribute = to_slab_attr(attr); | |
5181 | s = to_slab(kobj); | |
5182 | ||
5183 | if (!attribute->show) | |
5184 | return -EIO; | |
5185 | ||
5186 | err = attribute->show(s, buf); | |
5187 | ||
5188 | return err; | |
5189 | } | |
5190 | ||
5191 | static ssize_t slab_attr_store(struct kobject *kobj, | |
5192 | struct attribute *attr, | |
5193 | const char *buf, size_t len) | |
5194 | { | |
5195 | struct slab_attribute *attribute; | |
5196 | struct kmem_cache *s; | |
5197 | int err; | |
5198 | ||
5199 | attribute = to_slab_attr(attr); | |
5200 | s = to_slab(kobj); | |
5201 | ||
5202 | if (!attribute->store) | |
5203 | return -EIO; | |
5204 | ||
5205 | err = attribute->store(s, buf, len); | |
127424c8 | 5206 | #ifdef CONFIG_MEMCG |
107dab5c | 5207 | if (slab_state >= FULL && err >= 0 && is_root_cache(s)) { |
426589f5 | 5208 | struct kmem_cache *c; |
81819f0f | 5209 | |
107dab5c GC |
5210 | mutex_lock(&slab_mutex); |
5211 | if (s->max_attr_size < len) | |
5212 | s->max_attr_size = len; | |
5213 | ||
ebe945c2 GC |
5214 | /* |
5215 | * This is a best effort propagation, so this function's return | |
5216 | * value will be determined by the parent cache only. This is | |
5217 | * basically because not all attributes will have a well | |
5218 | * defined semantics for rollbacks - most of the actions will | |
5219 | * have permanent effects. | |
5220 | * | |
5221 | * Returning the error value of any of the children that fail | |
5222 | * is not 100 % defined, in the sense that users seeing the | |
5223 | * error code won't be able to know anything about the state of | |
5224 | * the cache. | |
5225 | * | |
5226 | * Only returning the error code for the parent cache at least | |
5227 | * has well defined semantics. The cache being written to | |
5228 | * directly either failed or succeeded, in which case we loop | |
5229 | * through the descendants with best-effort propagation. | |
5230 | */ | |
426589f5 VD |
5231 | for_each_memcg_cache(c, s) |
5232 | attribute->store(c, buf, len); | |
107dab5c GC |
5233 | mutex_unlock(&slab_mutex); |
5234 | } | |
5235 | #endif | |
81819f0f CL |
5236 | return err; |
5237 | } | |
5238 | ||
107dab5c GC |
5239 | static void memcg_propagate_slab_attrs(struct kmem_cache *s) |
5240 | { | |
127424c8 | 5241 | #ifdef CONFIG_MEMCG |
107dab5c GC |
5242 | int i; |
5243 | char *buffer = NULL; | |
93030d83 | 5244 | struct kmem_cache *root_cache; |
107dab5c | 5245 | |
93030d83 | 5246 | if (is_root_cache(s)) |
107dab5c GC |
5247 | return; |
5248 | ||
f7ce3190 | 5249 | root_cache = s->memcg_params.root_cache; |
93030d83 | 5250 | |
107dab5c GC |
5251 | /* |
5252 | * This mean this cache had no attribute written. Therefore, no point | |
5253 | * in copying default values around | |
5254 | */ | |
93030d83 | 5255 | if (!root_cache->max_attr_size) |
107dab5c GC |
5256 | return; |
5257 | ||
5258 | for (i = 0; i < ARRAY_SIZE(slab_attrs); i++) { | |
5259 | char mbuf[64]; | |
5260 | char *buf; | |
5261 | struct slab_attribute *attr = to_slab_attr(slab_attrs[i]); | |
5262 | ||
5263 | if (!attr || !attr->store || !attr->show) | |
5264 | continue; | |
5265 | ||
5266 | /* | |
5267 | * It is really bad that we have to allocate here, so we will | |
5268 | * do it only as a fallback. If we actually allocate, though, | |
5269 | * we can just use the allocated buffer until the end. | |
5270 | * | |
5271 | * Most of the slub attributes will tend to be very small in | |
5272 | * size, but sysfs allows buffers up to a page, so they can | |
5273 | * theoretically happen. | |
5274 | */ | |
5275 | if (buffer) | |
5276 | buf = buffer; | |
93030d83 | 5277 | else if (root_cache->max_attr_size < ARRAY_SIZE(mbuf)) |
107dab5c GC |
5278 | buf = mbuf; |
5279 | else { | |
5280 | buffer = (char *) get_zeroed_page(GFP_KERNEL); | |
5281 | if (WARN_ON(!buffer)) | |
5282 | continue; | |
5283 | buf = buffer; | |
5284 | } | |
5285 | ||
93030d83 | 5286 | attr->show(root_cache, buf); |
107dab5c GC |
5287 | attr->store(s, buf, strlen(buf)); |
5288 | } | |
5289 | ||
5290 | if (buffer) | |
5291 | free_page((unsigned long)buffer); | |
5292 | #endif | |
5293 | } | |
5294 | ||
41a21285 CL |
5295 | static void kmem_cache_release(struct kobject *k) |
5296 | { | |
5297 | slab_kmem_cache_release(to_slab(k)); | |
5298 | } | |
5299 | ||
52cf25d0 | 5300 | static const struct sysfs_ops slab_sysfs_ops = { |
81819f0f CL |
5301 | .show = slab_attr_show, |
5302 | .store = slab_attr_store, | |
5303 | }; | |
5304 | ||
5305 | static struct kobj_type slab_ktype = { | |
5306 | .sysfs_ops = &slab_sysfs_ops, | |
41a21285 | 5307 | .release = kmem_cache_release, |
81819f0f CL |
5308 | }; |
5309 | ||
5310 | static int uevent_filter(struct kset *kset, struct kobject *kobj) | |
5311 | { | |
5312 | struct kobj_type *ktype = get_ktype(kobj); | |
5313 | ||
5314 | if (ktype == &slab_ktype) | |
5315 | return 1; | |
5316 | return 0; | |
5317 | } | |
5318 | ||
9cd43611 | 5319 | static const struct kset_uevent_ops slab_uevent_ops = { |
81819f0f CL |
5320 | .filter = uevent_filter, |
5321 | }; | |
5322 | ||
27c3a314 | 5323 | static struct kset *slab_kset; |
81819f0f | 5324 | |
9a41707b VD |
5325 | static inline struct kset *cache_kset(struct kmem_cache *s) |
5326 | { | |
127424c8 | 5327 | #ifdef CONFIG_MEMCG |
9a41707b | 5328 | if (!is_root_cache(s)) |
f7ce3190 | 5329 | return s->memcg_params.root_cache->memcg_kset; |
9a41707b VD |
5330 | #endif |
5331 | return slab_kset; | |
5332 | } | |
5333 | ||
81819f0f CL |
5334 | #define ID_STR_LENGTH 64 |
5335 | ||
5336 | /* Create a unique string id for a slab cache: | |
6446faa2 CL |
5337 | * |
5338 | * Format :[flags-]size | |
81819f0f CL |
5339 | */ |
5340 | static char *create_unique_id(struct kmem_cache *s) | |
5341 | { | |
5342 | char *name = kmalloc(ID_STR_LENGTH, GFP_KERNEL); | |
5343 | char *p = name; | |
5344 | ||
5345 | BUG_ON(!name); | |
5346 | ||
5347 | *p++ = ':'; | |
5348 | /* | |
5349 | * First flags affecting slabcache operations. We will only | |
5350 | * get here for aliasable slabs so we do not need to support | |
5351 | * too many flags. The flags here must cover all flags that | |
5352 | * are matched during merging to guarantee that the id is | |
5353 | * unique. | |
5354 | */ | |
5355 | if (s->flags & SLAB_CACHE_DMA) | |
5356 | *p++ = 'd'; | |
5357 | if (s->flags & SLAB_RECLAIM_ACCOUNT) | |
5358 | *p++ = 'a'; | |
5359 | if (s->flags & SLAB_DEBUG_FREE) | |
5360 | *p++ = 'F'; | |
5a896d9e VN |
5361 | if (!(s->flags & SLAB_NOTRACK)) |
5362 | *p++ = 't'; | |
230e9fc2 VD |
5363 | if (s->flags & SLAB_ACCOUNT) |
5364 | *p++ = 'A'; | |
81819f0f CL |
5365 | if (p != name + 1) |
5366 | *p++ = '-'; | |
5367 | p += sprintf(p, "%07d", s->size); | |
2633d7a0 | 5368 | |
81819f0f CL |
5369 | BUG_ON(p > name + ID_STR_LENGTH - 1); |
5370 | return name; | |
5371 | } | |
5372 | ||
5373 | static int sysfs_slab_add(struct kmem_cache *s) | |
5374 | { | |
5375 | int err; | |
5376 | const char *name; | |
45530c44 | 5377 | int unmergeable = slab_unmergeable(s); |
81819f0f | 5378 | |
81819f0f CL |
5379 | if (unmergeable) { |
5380 | /* | |
5381 | * Slabcache can never be merged so we can use the name proper. | |
5382 | * This is typically the case for debug situations. In that | |
5383 | * case we can catch duplicate names easily. | |
5384 | */ | |
27c3a314 | 5385 | sysfs_remove_link(&slab_kset->kobj, s->name); |
81819f0f CL |
5386 | name = s->name; |
5387 | } else { | |
5388 | /* | |
5389 | * Create a unique name for the slab as a target | |
5390 | * for the symlinks. | |
5391 | */ | |
5392 | name = create_unique_id(s); | |
5393 | } | |
5394 | ||
9a41707b | 5395 | s->kobj.kset = cache_kset(s); |
26e4f205 | 5396 | err = kobject_init_and_add(&s->kobj, &slab_ktype, NULL, "%s", name); |
54b6a731 | 5397 | if (err) |
80da026a | 5398 | goto out; |
81819f0f CL |
5399 | |
5400 | err = sysfs_create_group(&s->kobj, &slab_attr_group); | |
54b6a731 DJ |
5401 | if (err) |
5402 | goto out_del_kobj; | |
9a41707b | 5403 | |
127424c8 | 5404 | #ifdef CONFIG_MEMCG |
9a41707b VD |
5405 | if (is_root_cache(s)) { |
5406 | s->memcg_kset = kset_create_and_add("cgroup", NULL, &s->kobj); | |
5407 | if (!s->memcg_kset) { | |
54b6a731 DJ |
5408 | err = -ENOMEM; |
5409 | goto out_del_kobj; | |
9a41707b VD |
5410 | } |
5411 | } | |
5412 | #endif | |
5413 | ||
81819f0f CL |
5414 | kobject_uevent(&s->kobj, KOBJ_ADD); |
5415 | if (!unmergeable) { | |
5416 | /* Setup first alias */ | |
5417 | sysfs_slab_alias(s, s->name); | |
81819f0f | 5418 | } |
54b6a731 DJ |
5419 | out: |
5420 | if (!unmergeable) | |
5421 | kfree(name); | |
5422 | return err; | |
5423 | out_del_kobj: | |
5424 | kobject_del(&s->kobj); | |
54b6a731 | 5425 | goto out; |
81819f0f CL |
5426 | } |
5427 | ||
41a21285 | 5428 | void sysfs_slab_remove(struct kmem_cache *s) |
81819f0f | 5429 | { |
97d06609 | 5430 | if (slab_state < FULL) |
2bce6485 CL |
5431 | /* |
5432 | * Sysfs has not been setup yet so no need to remove the | |
5433 | * cache from sysfs. | |
5434 | */ | |
5435 | return; | |
5436 | ||
127424c8 | 5437 | #ifdef CONFIG_MEMCG |
9a41707b VD |
5438 | kset_unregister(s->memcg_kset); |
5439 | #endif | |
81819f0f CL |
5440 | kobject_uevent(&s->kobj, KOBJ_REMOVE); |
5441 | kobject_del(&s->kobj); | |
151c602f | 5442 | kobject_put(&s->kobj); |
81819f0f CL |
5443 | } |
5444 | ||
5445 | /* | |
5446 | * Need to buffer aliases during bootup until sysfs becomes | |
9f6c708e | 5447 | * available lest we lose that information. |
81819f0f CL |
5448 | */ |
5449 | struct saved_alias { | |
5450 | struct kmem_cache *s; | |
5451 | const char *name; | |
5452 | struct saved_alias *next; | |
5453 | }; | |
5454 | ||
5af328a5 | 5455 | static struct saved_alias *alias_list; |
81819f0f CL |
5456 | |
5457 | static int sysfs_slab_alias(struct kmem_cache *s, const char *name) | |
5458 | { | |
5459 | struct saved_alias *al; | |
5460 | ||
97d06609 | 5461 | if (slab_state == FULL) { |
81819f0f CL |
5462 | /* |
5463 | * If we have a leftover link then remove it. | |
5464 | */ | |
27c3a314 GKH |
5465 | sysfs_remove_link(&slab_kset->kobj, name); |
5466 | return sysfs_create_link(&slab_kset->kobj, &s->kobj, name); | |
81819f0f CL |
5467 | } |
5468 | ||
5469 | al = kmalloc(sizeof(struct saved_alias), GFP_KERNEL); | |
5470 | if (!al) | |
5471 | return -ENOMEM; | |
5472 | ||
5473 | al->s = s; | |
5474 | al->name = name; | |
5475 | al->next = alias_list; | |
5476 | alias_list = al; | |
5477 | return 0; | |
5478 | } | |
5479 | ||
5480 | static int __init slab_sysfs_init(void) | |
5481 | { | |
5b95a4ac | 5482 | struct kmem_cache *s; |
81819f0f CL |
5483 | int err; |
5484 | ||
18004c5d | 5485 | mutex_lock(&slab_mutex); |
2bce6485 | 5486 | |
0ff21e46 | 5487 | slab_kset = kset_create_and_add("slab", &slab_uevent_ops, kernel_kobj); |
27c3a314 | 5488 | if (!slab_kset) { |
18004c5d | 5489 | mutex_unlock(&slab_mutex); |
f9f58285 | 5490 | pr_err("Cannot register slab subsystem.\n"); |
81819f0f CL |
5491 | return -ENOSYS; |
5492 | } | |
5493 | ||
97d06609 | 5494 | slab_state = FULL; |
26a7bd03 | 5495 | |
5b95a4ac | 5496 | list_for_each_entry(s, &slab_caches, list) { |
26a7bd03 | 5497 | err = sysfs_slab_add(s); |
5d540fb7 | 5498 | if (err) |
f9f58285 FF |
5499 | pr_err("SLUB: Unable to add boot slab %s to sysfs\n", |
5500 | s->name); | |
26a7bd03 | 5501 | } |
81819f0f CL |
5502 | |
5503 | while (alias_list) { | |
5504 | struct saved_alias *al = alias_list; | |
5505 | ||
5506 | alias_list = alias_list->next; | |
5507 | err = sysfs_slab_alias(al->s, al->name); | |
5d540fb7 | 5508 | if (err) |
f9f58285 FF |
5509 | pr_err("SLUB: Unable to add boot slab alias %s to sysfs\n", |
5510 | al->name); | |
81819f0f CL |
5511 | kfree(al); |
5512 | } | |
5513 | ||
18004c5d | 5514 | mutex_unlock(&slab_mutex); |
81819f0f CL |
5515 | resiliency_test(); |
5516 | return 0; | |
5517 | } | |
5518 | ||
5519 | __initcall(slab_sysfs_init); | |
ab4d5ed5 | 5520 | #endif /* CONFIG_SYSFS */ |
57ed3eda PE |
5521 | |
5522 | /* | |
5523 | * The /proc/slabinfo ABI | |
5524 | */ | |
158a9624 | 5525 | #ifdef CONFIG_SLABINFO |
0d7561c6 | 5526 | void get_slabinfo(struct kmem_cache *s, struct slabinfo *sinfo) |
57ed3eda | 5527 | { |
57ed3eda | 5528 | unsigned long nr_slabs = 0; |
205ab99d CL |
5529 | unsigned long nr_objs = 0; |
5530 | unsigned long nr_free = 0; | |
57ed3eda | 5531 | int node; |
fa45dc25 | 5532 | struct kmem_cache_node *n; |
57ed3eda | 5533 | |
fa45dc25 | 5534 | for_each_kmem_cache_node(s, node, n) { |
c17fd13e WL |
5535 | nr_slabs += node_nr_slabs(n); |
5536 | nr_objs += node_nr_objs(n); | |
205ab99d | 5537 | nr_free += count_partial(n, count_free); |
57ed3eda PE |
5538 | } |
5539 | ||
0d7561c6 GC |
5540 | sinfo->active_objs = nr_objs - nr_free; |
5541 | sinfo->num_objs = nr_objs; | |
5542 | sinfo->active_slabs = nr_slabs; | |
5543 | sinfo->num_slabs = nr_slabs; | |
5544 | sinfo->objects_per_slab = oo_objects(s->oo); | |
5545 | sinfo->cache_order = oo_order(s->oo); | |
57ed3eda PE |
5546 | } |
5547 | ||
0d7561c6 | 5548 | void slabinfo_show_stats(struct seq_file *m, struct kmem_cache *s) |
7b3c3a50 | 5549 | { |
7b3c3a50 AD |
5550 | } |
5551 | ||
b7454ad3 GC |
5552 | ssize_t slabinfo_write(struct file *file, const char __user *buffer, |
5553 | size_t count, loff_t *ppos) | |
7b3c3a50 | 5554 | { |
b7454ad3 | 5555 | return -EIO; |
7b3c3a50 | 5556 | } |
158a9624 | 5557 | #endif /* CONFIG_SLABINFO */ |