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 |
1596 | __remove_partial(struct kmem_cache_node *n, struct page *page) | |
1597 | { | |
62e346a8 CL |
1598 | list_del(&page->lru); |
1599 | n->nr_partial--; | |
1600 | } | |
1601 | ||
1e4dd946 SR |
1602 | static inline void remove_partial(struct kmem_cache_node *n, |
1603 | struct page *page) | |
1604 | { | |
1605 | lockdep_assert_held(&n->list_lock); | |
1606 | __remove_partial(n, page); | |
1607 | } | |
1608 | ||
81819f0f | 1609 | /* |
7ced3719 CL |
1610 | * Remove slab from the partial list, freeze it and |
1611 | * return the pointer to the freelist. | |
81819f0f | 1612 | * |
497b66f2 | 1613 | * Returns a list of objects or NULL if it fails. |
81819f0f | 1614 | */ |
497b66f2 | 1615 | static inline void *acquire_slab(struct kmem_cache *s, |
acd19fd1 | 1616 | struct kmem_cache_node *n, struct page *page, |
633b0764 | 1617 | int mode, int *objects) |
81819f0f | 1618 | { |
2cfb7455 CL |
1619 | void *freelist; |
1620 | unsigned long counters; | |
1621 | struct page new; | |
1622 | ||
c65c1877 PZ |
1623 | lockdep_assert_held(&n->list_lock); |
1624 | ||
2cfb7455 CL |
1625 | /* |
1626 | * Zap the freelist and set the frozen bit. | |
1627 | * The old freelist is the list of objects for the | |
1628 | * per cpu allocation list. | |
1629 | */ | |
7ced3719 CL |
1630 | freelist = page->freelist; |
1631 | counters = page->counters; | |
1632 | new.counters = counters; | |
633b0764 | 1633 | *objects = new.objects - new.inuse; |
23910c50 | 1634 | if (mode) { |
7ced3719 | 1635 | new.inuse = page->objects; |
23910c50 PE |
1636 | new.freelist = NULL; |
1637 | } else { | |
1638 | new.freelist = freelist; | |
1639 | } | |
2cfb7455 | 1640 | |
a0132ac0 | 1641 | VM_BUG_ON(new.frozen); |
7ced3719 | 1642 | new.frozen = 1; |
2cfb7455 | 1643 | |
7ced3719 | 1644 | if (!__cmpxchg_double_slab(s, page, |
2cfb7455 | 1645 | freelist, counters, |
02d7633f | 1646 | new.freelist, new.counters, |
7ced3719 | 1647 | "acquire_slab")) |
7ced3719 | 1648 | return NULL; |
2cfb7455 CL |
1649 | |
1650 | remove_partial(n, page); | |
7ced3719 | 1651 | WARN_ON(!freelist); |
49e22585 | 1652 | return freelist; |
81819f0f CL |
1653 | } |
1654 | ||
633b0764 | 1655 | static void put_cpu_partial(struct kmem_cache *s, struct page *page, int drain); |
8ba00bb6 | 1656 | static inline bool pfmemalloc_match(struct page *page, gfp_t gfpflags); |
49e22585 | 1657 | |
81819f0f | 1658 | /* |
672bba3a | 1659 | * Try to allocate a partial slab from a specific node. |
81819f0f | 1660 | */ |
8ba00bb6 JK |
1661 | static void *get_partial_node(struct kmem_cache *s, struct kmem_cache_node *n, |
1662 | struct kmem_cache_cpu *c, gfp_t flags) | |
81819f0f | 1663 | { |
49e22585 CL |
1664 | struct page *page, *page2; |
1665 | void *object = NULL; | |
633b0764 JK |
1666 | int available = 0; |
1667 | int objects; | |
81819f0f CL |
1668 | |
1669 | /* | |
1670 | * Racy check. If we mistakenly see no partial slabs then we | |
1671 | * just allocate an empty slab. If we mistakenly try to get a | |
672bba3a CL |
1672 | * partial slab and there is none available then get_partials() |
1673 | * will return NULL. | |
81819f0f CL |
1674 | */ |
1675 | if (!n || !n->nr_partial) | |
1676 | return NULL; | |
1677 | ||
1678 | spin_lock(&n->list_lock); | |
49e22585 | 1679 | list_for_each_entry_safe(page, page2, &n->partial, lru) { |
8ba00bb6 | 1680 | void *t; |
49e22585 | 1681 | |
8ba00bb6 JK |
1682 | if (!pfmemalloc_match(page, flags)) |
1683 | continue; | |
1684 | ||
633b0764 | 1685 | t = acquire_slab(s, n, page, object == NULL, &objects); |
49e22585 CL |
1686 | if (!t) |
1687 | break; | |
1688 | ||
633b0764 | 1689 | available += objects; |
12d79634 | 1690 | if (!object) { |
49e22585 | 1691 | c->page = page; |
49e22585 | 1692 | stat(s, ALLOC_FROM_PARTIAL); |
49e22585 | 1693 | object = t; |
49e22585 | 1694 | } else { |
633b0764 | 1695 | put_cpu_partial(s, page, 0); |
8028dcea | 1696 | stat(s, CPU_PARTIAL_NODE); |
49e22585 | 1697 | } |
345c905d JK |
1698 | if (!kmem_cache_has_cpu_partial(s) |
1699 | || available > s->cpu_partial / 2) | |
49e22585 CL |
1700 | break; |
1701 | ||
497b66f2 | 1702 | } |
81819f0f | 1703 | spin_unlock(&n->list_lock); |
497b66f2 | 1704 | return object; |
81819f0f CL |
1705 | } |
1706 | ||
1707 | /* | |
672bba3a | 1708 | * Get a page from somewhere. Search in increasing NUMA distances. |
81819f0f | 1709 | */ |
de3ec035 | 1710 | static void *get_any_partial(struct kmem_cache *s, gfp_t flags, |
acd19fd1 | 1711 | struct kmem_cache_cpu *c) |
81819f0f CL |
1712 | { |
1713 | #ifdef CONFIG_NUMA | |
1714 | struct zonelist *zonelist; | |
dd1a239f | 1715 | struct zoneref *z; |
54a6eb5c MG |
1716 | struct zone *zone; |
1717 | enum zone_type high_zoneidx = gfp_zone(flags); | |
497b66f2 | 1718 | void *object; |
cc9a6c87 | 1719 | unsigned int cpuset_mems_cookie; |
81819f0f CL |
1720 | |
1721 | /* | |
672bba3a CL |
1722 | * The defrag ratio allows a configuration of the tradeoffs between |
1723 | * inter node defragmentation and node local allocations. A lower | |
1724 | * defrag_ratio increases the tendency to do local allocations | |
1725 | * instead of attempting to obtain partial slabs from other nodes. | |
81819f0f | 1726 | * |
672bba3a CL |
1727 | * If the defrag_ratio is set to 0 then kmalloc() always |
1728 | * returns node local objects. If the ratio is higher then kmalloc() | |
1729 | * may return off node objects because partial slabs are obtained | |
1730 | * from other nodes and filled up. | |
81819f0f | 1731 | * |
6446faa2 | 1732 | * If /sys/kernel/slab/xx/defrag_ratio is set to 100 (which makes |
672bba3a CL |
1733 | * defrag_ratio = 1000) then every (well almost) allocation will |
1734 | * first attempt to defrag slab caches on other nodes. This means | |
1735 | * scanning over all nodes to look for partial slabs which may be | |
1736 | * expensive if we do it every time we are trying to find a slab | |
1737 | * with available objects. | |
81819f0f | 1738 | */ |
9824601e CL |
1739 | if (!s->remote_node_defrag_ratio || |
1740 | get_cycles() % 1024 > s->remote_node_defrag_ratio) | |
81819f0f CL |
1741 | return NULL; |
1742 | ||
cc9a6c87 | 1743 | do { |
d26914d1 | 1744 | cpuset_mems_cookie = read_mems_allowed_begin(); |
2a389610 | 1745 | zonelist = node_zonelist(mempolicy_slab_node(), flags); |
cc9a6c87 MG |
1746 | for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) { |
1747 | struct kmem_cache_node *n; | |
1748 | ||
1749 | n = get_node(s, zone_to_nid(zone)); | |
1750 | ||
dee2f8aa | 1751 | if (n && cpuset_zone_allowed(zone, flags) && |
cc9a6c87 | 1752 | n->nr_partial > s->min_partial) { |
8ba00bb6 | 1753 | object = get_partial_node(s, n, c, flags); |
cc9a6c87 MG |
1754 | if (object) { |
1755 | /* | |
d26914d1 MG |
1756 | * Don't check read_mems_allowed_retry() |
1757 | * here - if mems_allowed was updated in | |
1758 | * parallel, that was a harmless race | |
1759 | * between allocation and the cpuset | |
1760 | * update | |
cc9a6c87 | 1761 | */ |
cc9a6c87 MG |
1762 | return object; |
1763 | } | |
c0ff7453 | 1764 | } |
81819f0f | 1765 | } |
d26914d1 | 1766 | } while (read_mems_allowed_retry(cpuset_mems_cookie)); |
81819f0f CL |
1767 | #endif |
1768 | return NULL; | |
1769 | } | |
1770 | ||
1771 | /* | |
1772 | * Get a partial page, lock it and return it. | |
1773 | */ | |
497b66f2 | 1774 | static void *get_partial(struct kmem_cache *s, gfp_t flags, int node, |
acd19fd1 | 1775 | struct kmem_cache_cpu *c) |
81819f0f | 1776 | { |
497b66f2 | 1777 | void *object; |
a561ce00 JK |
1778 | int searchnode = node; |
1779 | ||
1780 | if (node == NUMA_NO_NODE) | |
1781 | searchnode = numa_mem_id(); | |
1782 | else if (!node_present_pages(node)) | |
1783 | searchnode = node_to_mem_node(node); | |
81819f0f | 1784 | |
8ba00bb6 | 1785 | object = get_partial_node(s, get_node(s, searchnode), c, flags); |
497b66f2 CL |
1786 | if (object || node != NUMA_NO_NODE) |
1787 | return object; | |
81819f0f | 1788 | |
acd19fd1 | 1789 | return get_any_partial(s, flags, c); |
81819f0f CL |
1790 | } |
1791 | ||
8a5ec0ba CL |
1792 | #ifdef CONFIG_PREEMPT |
1793 | /* | |
1794 | * Calculate the next globally unique transaction for disambiguiation | |
1795 | * during cmpxchg. The transactions start with the cpu number and are then | |
1796 | * incremented by CONFIG_NR_CPUS. | |
1797 | */ | |
1798 | #define TID_STEP roundup_pow_of_two(CONFIG_NR_CPUS) | |
1799 | #else | |
1800 | /* | |
1801 | * No preemption supported therefore also no need to check for | |
1802 | * different cpus. | |
1803 | */ | |
1804 | #define TID_STEP 1 | |
1805 | #endif | |
1806 | ||
1807 | static inline unsigned long next_tid(unsigned long tid) | |
1808 | { | |
1809 | return tid + TID_STEP; | |
1810 | } | |
1811 | ||
1812 | static inline unsigned int tid_to_cpu(unsigned long tid) | |
1813 | { | |
1814 | return tid % TID_STEP; | |
1815 | } | |
1816 | ||
1817 | static inline unsigned long tid_to_event(unsigned long tid) | |
1818 | { | |
1819 | return tid / TID_STEP; | |
1820 | } | |
1821 | ||
1822 | static inline unsigned int init_tid(int cpu) | |
1823 | { | |
1824 | return cpu; | |
1825 | } | |
1826 | ||
1827 | static inline void note_cmpxchg_failure(const char *n, | |
1828 | const struct kmem_cache *s, unsigned long tid) | |
1829 | { | |
1830 | #ifdef SLUB_DEBUG_CMPXCHG | |
1831 | unsigned long actual_tid = __this_cpu_read(s->cpu_slab->tid); | |
1832 | ||
f9f58285 | 1833 | pr_info("%s %s: cmpxchg redo ", n, s->name); |
8a5ec0ba CL |
1834 | |
1835 | #ifdef CONFIG_PREEMPT | |
1836 | if (tid_to_cpu(tid) != tid_to_cpu(actual_tid)) | |
f9f58285 | 1837 | pr_warn("due to cpu change %d -> %d\n", |
8a5ec0ba CL |
1838 | tid_to_cpu(tid), tid_to_cpu(actual_tid)); |
1839 | else | |
1840 | #endif | |
1841 | if (tid_to_event(tid) != tid_to_event(actual_tid)) | |
f9f58285 | 1842 | pr_warn("due to cpu running other code. Event %ld->%ld\n", |
8a5ec0ba CL |
1843 | tid_to_event(tid), tid_to_event(actual_tid)); |
1844 | else | |
f9f58285 | 1845 | pr_warn("for unknown reason: actual=%lx was=%lx target=%lx\n", |
8a5ec0ba CL |
1846 | actual_tid, tid, next_tid(tid)); |
1847 | #endif | |
4fdccdfb | 1848 | stat(s, CMPXCHG_DOUBLE_CPU_FAIL); |
8a5ec0ba CL |
1849 | } |
1850 | ||
788e1aad | 1851 | static void init_kmem_cache_cpus(struct kmem_cache *s) |
8a5ec0ba | 1852 | { |
8a5ec0ba CL |
1853 | int cpu; |
1854 | ||
1855 | for_each_possible_cpu(cpu) | |
1856 | per_cpu_ptr(s->cpu_slab, cpu)->tid = init_tid(cpu); | |
8a5ec0ba | 1857 | } |
2cfb7455 | 1858 | |
81819f0f CL |
1859 | /* |
1860 | * Remove the cpu slab | |
1861 | */ | |
d0e0ac97 CG |
1862 | static void deactivate_slab(struct kmem_cache *s, struct page *page, |
1863 | void *freelist) | |
81819f0f | 1864 | { |
2cfb7455 | 1865 | enum slab_modes { M_NONE, M_PARTIAL, M_FULL, M_FREE }; |
2cfb7455 CL |
1866 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); |
1867 | int lock = 0; | |
1868 | enum slab_modes l = M_NONE, m = M_NONE; | |
2cfb7455 | 1869 | void *nextfree; |
136333d1 | 1870 | int tail = DEACTIVATE_TO_HEAD; |
2cfb7455 CL |
1871 | struct page new; |
1872 | struct page old; | |
1873 | ||
1874 | if (page->freelist) { | |
84e554e6 | 1875 | stat(s, DEACTIVATE_REMOTE_FREES); |
136333d1 | 1876 | tail = DEACTIVATE_TO_TAIL; |
2cfb7455 CL |
1877 | } |
1878 | ||
894b8788 | 1879 | /* |
2cfb7455 CL |
1880 | * Stage one: Free all available per cpu objects back |
1881 | * to the page freelist while it is still frozen. Leave the | |
1882 | * last one. | |
1883 | * | |
1884 | * There is no need to take the list->lock because the page | |
1885 | * is still frozen. | |
1886 | */ | |
1887 | while (freelist && (nextfree = get_freepointer(s, freelist))) { | |
1888 | void *prior; | |
1889 | unsigned long counters; | |
1890 | ||
1891 | do { | |
1892 | prior = page->freelist; | |
1893 | counters = page->counters; | |
1894 | set_freepointer(s, freelist, prior); | |
1895 | new.counters = counters; | |
1896 | new.inuse--; | |
a0132ac0 | 1897 | VM_BUG_ON(!new.frozen); |
2cfb7455 | 1898 | |
1d07171c | 1899 | } while (!__cmpxchg_double_slab(s, page, |
2cfb7455 CL |
1900 | prior, counters, |
1901 | freelist, new.counters, | |
1902 | "drain percpu freelist")); | |
1903 | ||
1904 | freelist = nextfree; | |
1905 | } | |
1906 | ||
894b8788 | 1907 | /* |
2cfb7455 CL |
1908 | * Stage two: Ensure that the page is unfrozen while the |
1909 | * list presence reflects the actual number of objects | |
1910 | * during unfreeze. | |
1911 | * | |
1912 | * We setup the list membership and then perform a cmpxchg | |
1913 | * with the count. If there is a mismatch then the page | |
1914 | * is not unfrozen but the page is on the wrong list. | |
1915 | * | |
1916 | * Then we restart the process which may have to remove | |
1917 | * the page from the list that we just put it on again | |
1918 | * because the number of objects in the slab may have | |
1919 | * changed. | |
894b8788 | 1920 | */ |
2cfb7455 | 1921 | redo: |
894b8788 | 1922 | |
2cfb7455 CL |
1923 | old.freelist = page->freelist; |
1924 | old.counters = page->counters; | |
a0132ac0 | 1925 | VM_BUG_ON(!old.frozen); |
7c2e132c | 1926 | |
2cfb7455 CL |
1927 | /* Determine target state of the slab */ |
1928 | new.counters = old.counters; | |
1929 | if (freelist) { | |
1930 | new.inuse--; | |
1931 | set_freepointer(s, freelist, old.freelist); | |
1932 | new.freelist = freelist; | |
1933 | } else | |
1934 | new.freelist = old.freelist; | |
1935 | ||
1936 | new.frozen = 0; | |
1937 | ||
8a5b20ae | 1938 | if (!new.inuse && n->nr_partial >= s->min_partial) |
2cfb7455 CL |
1939 | m = M_FREE; |
1940 | else if (new.freelist) { | |
1941 | m = M_PARTIAL; | |
1942 | if (!lock) { | |
1943 | lock = 1; | |
1944 | /* | |
1945 | * Taking the spinlock removes the possiblity | |
1946 | * that acquire_slab() will see a slab page that | |
1947 | * is frozen | |
1948 | */ | |
1949 | spin_lock(&n->list_lock); | |
1950 | } | |
1951 | } else { | |
1952 | m = M_FULL; | |
1953 | if (kmem_cache_debug(s) && !lock) { | |
1954 | lock = 1; | |
1955 | /* | |
1956 | * This also ensures that the scanning of full | |
1957 | * slabs from diagnostic functions will not see | |
1958 | * any frozen slabs. | |
1959 | */ | |
1960 | spin_lock(&n->list_lock); | |
1961 | } | |
1962 | } | |
1963 | ||
1964 | if (l != m) { | |
1965 | ||
1966 | if (l == M_PARTIAL) | |
1967 | ||
1968 | remove_partial(n, page); | |
1969 | ||
1970 | else if (l == M_FULL) | |
894b8788 | 1971 | |
c65c1877 | 1972 | remove_full(s, n, page); |
2cfb7455 CL |
1973 | |
1974 | if (m == M_PARTIAL) { | |
1975 | ||
1976 | add_partial(n, page, tail); | |
136333d1 | 1977 | stat(s, tail); |
2cfb7455 CL |
1978 | |
1979 | } else if (m == M_FULL) { | |
894b8788 | 1980 | |
2cfb7455 CL |
1981 | stat(s, DEACTIVATE_FULL); |
1982 | add_full(s, n, page); | |
1983 | ||
1984 | } | |
1985 | } | |
1986 | ||
1987 | l = m; | |
1d07171c | 1988 | if (!__cmpxchg_double_slab(s, page, |
2cfb7455 CL |
1989 | old.freelist, old.counters, |
1990 | new.freelist, new.counters, | |
1991 | "unfreezing slab")) | |
1992 | goto redo; | |
1993 | ||
2cfb7455 CL |
1994 | if (lock) |
1995 | spin_unlock(&n->list_lock); | |
1996 | ||
1997 | if (m == M_FREE) { | |
1998 | stat(s, DEACTIVATE_EMPTY); | |
1999 | discard_slab(s, page); | |
2000 | stat(s, FREE_SLAB); | |
894b8788 | 2001 | } |
81819f0f CL |
2002 | } |
2003 | ||
d24ac77f JK |
2004 | /* |
2005 | * Unfreeze all the cpu partial slabs. | |
2006 | * | |
59a09917 CL |
2007 | * This function must be called with interrupts disabled |
2008 | * for the cpu using c (or some other guarantee must be there | |
2009 | * to guarantee no concurrent accesses). | |
d24ac77f | 2010 | */ |
59a09917 CL |
2011 | static void unfreeze_partials(struct kmem_cache *s, |
2012 | struct kmem_cache_cpu *c) | |
49e22585 | 2013 | { |
345c905d | 2014 | #ifdef CONFIG_SLUB_CPU_PARTIAL |
43d77867 | 2015 | struct kmem_cache_node *n = NULL, *n2 = NULL; |
9ada1934 | 2016 | struct page *page, *discard_page = NULL; |
49e22585 CL |
2017 | |
2018 | while ((page = c->partial)) { | |
49e22585 CL |
2019 | struct page new; |
2020 | struct page old; | |
2021 | ||
2022 | c->partial = page->next; | |
43d77867 JK |
2023 | |
2024 | n2 = get_node(s, page_to_nid(page)); | |
2025 | if (n != n2) { | |
2026 | if (n) | |
2027 | spin_unlock(&n->list_lock); | |
2028 | ||
2029 | n = n2; | |
2030 | spin_lock(&n->list_lock); | |
2031 | } | |
49e22585 CL |
2032 | |
2033 | do { | |
2034 | ||
2035 | old.freelist = page->freelist; | |
2036 | old.counters = page->counters; | |
a0132ac0 | 2037 | VM_BUG_ON(!old.frozen); |
49e22585 CL |
2038 | |
2039 | new.counters = old.counters; | |
2040 | new.freelist = old.freelist; | |
2041 | ||
2042 | new.frozen = 0; | |
2043 | ||
d24ac77f | 2044 | } while (!__cmpxchg_double_slab(s, page, |
49e22585 CL |
2045 | old.freelist, old.counters, |
2046 | new.freelist, new.counters, | |
2047 | "unfreezing slab")); | |
2048 | ||
8a5b20ae | 2049 | if (unlikely(!new.inuse && n->nr_partial >= s->min_partial)) { |
9ada1934 SL |
2050 | page->next = discard_page; |
2051 | discard_page = page; | |
43d77867 JK |
2052 | } else { |
2053 | add_partial(n, page, DEACTIVATE_TO_TAIL); | |
2054 | stat(s, FREE_ADD_PARTIAL); | |
49e22585 CL |
2055 | } |
2056 | } | |
2057 | ||
2058 | if (n) | |
2059 | spin_unlock(&n->list_lock); | |
9ada1934 SL |
2060 | |
2061 | while (discard_page) { | |
2062 | page = discard_page; | |
2063 | discard_page = discard_page->next; | |
2064 | ||
2065 | stat(s, DEACTIVATE_EMPTY); | |
2066 | discard_slab(s, page); | |
2067 | stat(s, FREE_SLAB); | |
2068 | } | |
345c905d | 2069 | #endif |
49e22585 CL |
2070 | } |
2071 | ||
2072 | /* | |
2073 | * Put a page that was just frozen (in __slab_free) into a partial page | |
2074 | * slot if available. This is done without interrupts disabled and without | |
2075 | * preemption disabled. The cmpxchg is racy and may put the partial page | |
2076 | * onto a random cpus partial slot. | |
2077 | * | |
2078 | * If we did not find a slot then simply move all the partials to the | |
2079 | * per node partial list. | |
2080 | */ | |
633b0764 | 2081 | static void put_cpu_partial(struct kmem_cache *s, struct page *page, int drain) |
49e22585 | 2082 | { |
345c905d | 2083 | #ifdef CONFIG_SLUB_CPU_PARTIAL |
49e22585 CL |
2084 | struct page *oldpage; |
2085 | int pages; | |
2086 | int pobjects; | |
2087 | ||
d6e0b7fa | 2088 | preempt_disable(); |
49e22585 CL |
2089 | do { |
2090 | pages = 0; | |
2091 | pobjects = 0; | |
2092 | oldpage = this_cpu_read(s->cpu_slab->partial); | |
2093 | ||
2094 | if (oldpage) { | |
2095 | pobjects = oldpage->pobjects; | |
2096 | pages = oldpage->pages; | |
2097 | if (drain && pobjects > s->cpu_partial) { | |
2098 | unsigned long flags; | |
2099 | /* | |
2100 | * partial array is full. Move the existing | |
2101 | * set to the per node partial list. | |
2102 | */ | |
2103 | local_irq_save(flags); | |
59a09917 | 2104 | unfreeze_partials(s, this_cpu_ptr(s->cpu_slab)); |
49e22585 | 2105 | local_irq_restore(flags); |
e24fc410 | 2106 | oldpage = NULL; |
49e22585 CL |
2107 | pobjects = 0; |
2108 | pages = 0; | |
8028dcea | 2109 | stat(s, CPU_PARTIAL_DRAIN); |
49e22585 CL |
2110 | } |
2111 | } | |
2112 | ||
2113 | pages++; | |
2114 | pobjects += page->objects - page->inuse; | |
2115 | ||
2116 | page->pages = pages; | |
2117 | page->pobjects = pobjects; | |
2118 | page->next = oldpage; | |
2119 | ||
d0e0ac97 CG |
2120 | } while (this_cpu_cmpxchg(s->cpu_slab->partial, oldpage, page) |
2121 | != oldpage); | |
d6e0b7fa VD |
2122 | if (unlikely(!s->cpu_partial)) { |
2123 | unsigned long flags; | |
2124 | ||
2125 | local_irq_save(flags); | |
2126 | unfreeze_partials(s, this_cpu_ptr(s->cpu_slab)); | |
2127 | local_irq_restore(flags); | |
2128 | } | |
2129 | preempt_enable(); | |
345c905d | 2130 | #endif |
49e22585 CL |
2131 | } |
2132 | ||
dfb4f096 | 2133 | static inline void flush_slab(struct kmem_cache *s, struct kmem_cache_cpu *c) |
81819f0f | 2134 | { |
84e554e6 | 2135 | stat(s, CPUSLAB_FLUSH); |
c17dda40 CL |
2136 | deactivate_slab(s, c->page, c->freelist); |
2137 | ||
2138 | c->tid = next_tid(c->tid); | |
2139 | c->page = NULL; | |
2140 | c->freelist = NULL; | |
81819f0f CL |
2141 | } |
2142 | ||
2143 | /* | |
2144 | * Flush cpu slab. | |
6446faa2 | 2145 | * |
81819f0f CL |
2146 | * Called from IPI handler with interrupts disabled. |
2147 | */ | |
0c710013 | 2148 | static inline void __flush_cpu_slab(struct kmem_cache *s, int cpu) |
81819f0f | 2149 | { |
9dfc6e68 | 2150 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu); |
81819f0f | 2151 | |
49e22585 CL |
2152 | if (likely(c)) { |
2153 | if (c->page) | |
2154 | flush_slab(s, c); | |
2155 | ||
59a09917 | 2156 | unfreeze_partials(s, c); |
49e22585 | 2157 | } |
81819f0f CL |
2158 | } |
2159 | ||
2160 | static void flush_cpu_slab(void *d) | |
2161 | { | |
2162 | struct kmem_cache *s = d; | |
81819f0f | 2163 | |
dfb4f096 | 2164 | __flush_cpu_slab(s, smp_processor_id()); |
81819f0f CL |
2165 | } |
2166 | ||
a8364d55 GBY |
2167 | static bool has_cpu_slab(int cpu, void *info) |
2168 | { | |
2169 | struct kmem_cache *s = info; | |
2170 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu); | |
2171 | ||
02e1a9cd | 2172 | return c->page || c->partial; |
a8364d55 GBY |
2173 | } |
2174 | ||
81819f0f CL |
2175 | static void flush_all(struct kmem_cache *s) |
2176 | { | |
a8364d55 | 2177 | on_each_cpu_cond(has_cpu_slab, flush_cpu_slab, s, 1, GFP_ATOMIC); |
81819f0f CL |
2178 | } |
2179 | ||
dfb4f096 CL |
2180 | /* |
2181 | * Check if the objects in a per cpu structure fit numa | |
2182 | * locality expectations. | |
2183 | */ | |
57d437d2 | 2184 | static inline int node_match(struct page *page, int node) |
dfb4f096 CL |
2185 | { |
2186 | #ifdef CONFIG_NUMA | |
4d7868e6 | 2187 | if (!page || (node != NUMA_NO_NODE && page_to_nid(page) != node)) |
dfb4f096 CL |
2188 | return 0; |
2189 | #endif | |
2190 | return 1; | |
2191 | } | |
2192 | ||
9a02d699 | 2193 | #ifdef CONFIG_SLUB_DEBUG |
781b2ba6 PE |
2194 | static int count_free(struct page *page) |
2195 | { | |
2196 | return page->objects - page->inuse; | |
2197 | } | |
2198 | ||
9a02d699 DR |
2199 | static inline unsigned long node_nr_objs(struct kmem_cache_node *n) |
2200 | { | |
2201 | return atomic_long_read(&n->total_objects); | |
2202 | } | |
2203 | #endif /* CONFIG_SLUB_DEBUG */ | |
2204 | ||
2205 | #if defined(CONFIG_SLUB_DEBUG) || defined(CONFIG_SYSFS) | |
781b2ba6 PE |
2206 | static unsigned long count_partial(struct kmem_cache_node *n, |
2207 | int (*get_count)(struct page *)) | |
2208 | { | |
2209 | unsigned long flags; | |
2210 | unsigned long x = 0; | |
2211 | struct page *page; | |
2212 | ||
2213 | spin_lock_irqsave(&n->list_lock, flags); | |
2214 | list_for_each_entry(page, &n->partial, lru) | |
2215 | x += get_count(page); | |
2216 | spin_unlock_irqrestore(&n->list_lock, flags); | |
2217 | return x; | |
2218 | } | |
9a02d699 | 2219 | #endif /* CONFIG_SLUB_DEBUG || CONFIG_SYSFS */ |
26c02cf0 | 2220 | |
781b2ba6 PE |
2221 | static noinline void |
2222 | slab_out_of_memory(struct kmem_cache *s, gfp_t gfpflags, int nid) | |
2223 | { | |
9a02d699 DR |
2224 | #ifdef CONFIG_SLUB_DEBUG |
2225 | static DEFINE_RATELIMIT_STATE(slub_oom_rs, DEFAULT_RATELIMIT_INTERVAL, | |
2226 | DEFAULT_RATELIMIT_BURST); | |
781b2ba6 | 2227 | int node; |
fa45dc25 | 2228 | struct kmem_cache_node *n; |
781b2ba6 | 2229 | |
9a02d699 DR |
2230 | if ((gfpflags & __GFP_NOWARN) || !__ratelimit(&slub_oom_rs)) |
2231 | return; | |
2232 | ||
f9f58285 | 2233 | pr_warn("SLUB: Unable to allocate memory on node %d (gfp=0x%x)\n", |
781b2ba6 | 2234 | nid, gfpflags); |
f9f58285 FF |
2235 | pr_warn(" cache: %s, object size: %d, buffer size: %d, default order: %d, min order: %d\n", |
2236 | s->name, s->object_size, s->size, oo_order(s->oo), | |
2237 | oo_order(s->min)); | |
781b2ba6 | 2238 | |
3b0efdfa | 2239 | if (oo_order(s->min) > get_order(s->object_size)) |
f9f58285 FF |
2240 | pr_warn(" %s debugging increased min order, use slub_debug=O to disable.\n", |
2241 | s->name); | |
fa5ec8a1 | 2242 | |
fa45dc25 | 2243 | for_each_kmem_cache_node(s, node, n) { |
781b2ba6 PE |
2244 | unsigned long nr_slabs; |
2245 | unsigned long nr_objs; | |
2246 | unsigned long nr_free; | |
2247 | ||
26c02cf0 AB |
2248 | nr_free = count_partial(n, count_free); |
2249 | nr_slabs = node_nr_slabs(n); | |
2250 | nr_objs = node_nr_objs(n); | |
781b2ba6 | 2251 | |
f9f58285 | 2252 | pr_warn(" node %d: slabs: %ld, objs: %ld, free: %ld\n", |
781b2ba6 PE |
2253 | node, nr_slabs, nr_objs, nr_free); |
2254 | } | |
9a02d699 | 2255 | #endif |
781b2ba6 PE |
2256 | } |
2257 | ||
497b66f2 CL |
2258 | static inline void *new_slab_objects(struct kmem_cache *s, gfp_t flags, |
2259 | int node, struct kmem_cache_cpu **pc) | |
2260 | { | |
6faa6833 | 2261 | void *freelist; |
188fd063 CL |
2262 | struct kmem_cache_cpu *c = *pc; |
2263 | struct page *page; | |
497b66f2 | 2264 | |
188fd063 | 2265 | freelist = get_partial(s, flags, node, c); |
497b66f2 | 2266 | |
188fd063 CL |
2267 | if (freelist) |
2268 | return freelist; | |
2269 | ||
2270 | page = new_slab(s, flags, node); | |
497b66f2 | 2271 | if (page) { |
7c8e0181 | 2272 | c = raw_cpu_ptr(s->cpu_slab); |
497b66f2 CL |
2273 | if (c->page) |
2274 | flush_slab(s, c); | |
2275 | ||
2276 | /* | |
2277 | * No other reference to the page yet so we can | |
2278 | * muck around with it freely without cmpxchg | |
2279 | */ | |
6faa6833 | 2280 | freelist = page->freelist; |
497b66f2 CL |
2281 | page->freelist = NULL; |
2282 | ||
2283 | stat(s, ALLOC_SLAB); | |
497b66f2 CL |
2284 | c->page = page; |
2285 | *pc = c; | |
2286 | } else | |
6faa6833 | 2287 | freelist = NULL; |
497b66f2 | 2288 | |
6faa6833 | 2289 | return freelist; |
497b66f2 CL |
2290 | } |
2291 | ||
072bb0aa MG |
2292 | static inline bool pfmemalloc_match(struct page *page, gfp_t gfpflags) |
2293 | { | |
2294 | if (unlikely(PageSlabPfmemalloc(page))) | |
2295 | return gfp_pfmemalloc_allowed(gfpflags); | |
2296 | ||
2297 | return true; | |
2298 | } | |
2299 | ||
213eeb9f | 2300 | /* |
d0e0ac97 CG |
2301 | * Check the page->freelist of a page and either transfer the freelist to the |
2302 | * per cpu freelist or deactivate the page. | |
213eeb9f CL |
2303 | * |
2304 | * The page is still frozen if the return value is not NULL. | |
2305 | * | |
2306 | * If this function returns NULL then the page has been unfrozen. | |
d24ac77f JK |
2307 | * |
2308 | * This function must be called with interrupt disabled. | |
213eeb9f CL |
2309 | */ |
2310 | static inline void *get_freelist(struct kmem_cache *s, struct page *page) | |
2311 | { | |
2312 | struct page new; | |
2313 | unsigned long counters; | |
2314 | void *freelist; | |
2315 | ||
2316 | do { | |
2317 | freelist = page->freelist; | |
2318 | counters = page->counters; | |
6faa6833 | 2319 | |
213eeb9f | 2320 | new.counters = counters; |
a0132ac0 | 2321 | VM_BUG_ON(!new.frozen); |
213eeb9f CL |
2322 | |
2323 | new.inuse = page->objects; | |
2324 | new.frozen = freelist != NULL; | |
2325 | ||
d24ac77f | 2326 | } while (!__cmpxchg_double_slab(s, page, |
213eeb9f CL |
2327 | freelist, counters, |
2328 | NULL, new.counters, | |
2329 | "get_freelist")); | |
2330 | ||
2331 | return freelist; | |
2332 | } | |
2333 | ||
81819f0f | 2334 | /* |
894b8788 CL |
2335 | * Slow path. The lockless freelist is empty or we need to perform |
2336 | * debugging duties. | |
2337 | * | |
894b8788 CL |
2338 | * Processing is still very fast if new objects have been freed to the |
2339 | * regular freelist. In that case we simply take over the regular freelist | |
2340 | * as the lockless freelist and zap the regular freelist. | |
81819f0f | 2341 | * |
894b8788 CL |
2342 | * If that is not working then we fall back to the partial lists. We take the |
2343 | * first element of the freelist as the object to allocate now and move the | |
2344 | * rest of the freelist to the lockless freelist. | |
81819f0f | 2345 | * |
894b8788 | 2346 | * And if we were unable to get a new slab from the partial slab lists then |
6446faa2 CL |
2347 | * we need to allocate a new slab. This is the slowest path since it involves |
2348 | * a call to the page allocator and the setup of a new slab. | |
a380a3c7 CL |
2349 | * |
2350 | * Version of __slab_alloc to use when we know that interrupts are | |
2351 | * already disabled (which is the case for bulk allocation). | |
81819f0f | 2352 | */ |
a380a3c7 | 2353 | static void *___slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node, |
ce71e27c | 2354 | unsigned long addr, struct kmem_cache_cpu *c) |
81819f0f | 2355 | { |
6faa6833 | 2356 | void *freelist; |
f6e7def7 | 2357 | struct page *page; |
81819f0f | 2358 | |
f6e7def7 CL |
2359 | page = c->page; |
2360 | if (!page) | |
81819f0f | 2361 | goto new_slab; |
49e22585 | 2362 | redo: |
6faa6833 | 2363 | |
57d437d2 | 2364 | if (unlikely(!node_match(page, node))) { |
a561ce00 JK |
2365 | int searchnode = node; |
2366 | ||
2367 | if (node != NUMA_NO_NODE && !node_present_pages(node)) | |
2368 | searchnode = node_to_mem_node(node); | |
2369 | ||
2370 | if (unlikely(!node_match(page, searchnode))) { | |
2371 | stat(s, ALLOC_NODE_MISMATCH); | |
2372 | deactivate_slab(s, page, c->freelist); | |
2373 | c->page = NULL; | |
2374 | c->freelist = NULL; | |
2375 | goto new_slab; | |
2376 | } | |
fc59c053 | 2377 | } |
6446faa2 | 2378 | |
072bb0aa MG |
2379 | /* |
2380 | * By rights, we should be searching for a slab page that was | |
2381 | * PFMEMALLOC but right now, we are losing the pfmemalloc | |
2382 | * information when the page leaves the per-cpu allocator | |
2383 | */ | |
2384 | if (unlikely(!pfmemalloc_match(page, gfpflags))) { | |
2385 | deactivate_slab(s, page, c->freelist); | |
2386 | c->page = NULL; | |
2387 | c->freelist = NULL; | |
2388 | goto new_slab; | |
2389 | } | |
2390 | ||
73736e03 | 2391 | /* must check again c->freelist in case of cpu migration or IRQ */ |
6faa6833 CL |
2392 | freelist = c->freelist; |
2393 | if (freelist) | |
73736e03 | 2394 | goto load_freelist; |
03e404af | 2395 | |
f6e7def7 | 2396 | freelist = get_freelist(s, page); |
6446faa2 | 2397 | |
6faa6833 | 2398 | if (!freelist) { |
03e404af CL |
2399 | c->page = NULL; |
2400 | stat(s, DEACTIVATE_BYPASS); | |
fc59c053 | 2401 | goto new_slab; |
03e404af | 2402 | } |
6446faa2 | 2403 | |
84e554e6 | 2404 | stat(s, ALLOC_REFILL); |
6446faa2 | 2405 | |
894b8788 | 2406 | load_freelist: |
507effea CL |
2407 | /* |
2408 | * freelist is pointing to the list of objects to be used. | |
2409 | * page is pointing to the page from which the objects are obtained. | |
2410 | * That page must be frozen for per cpu allocations to work. | |
2411 | */ | |
a0132ac0 | 2412 | VM_BUG_ON(!c->page->frozen); |
6faa6833 | 2413 | c->freelist = get_freepointer(s, freelist); |
8a5ec0ba | 2414 | c->tid = next_tid(c->tid); |
6faa6833 | 2415 | return freelist; |
81819f0f | 2416 | |
81819f0f | 2417 | new_slab: |
2cfb7455 | 2418 | |
49e22585 | 2419 | if (c->partial) { |
f6e7def7 CL |
2420 | page = c->page = c->partial; |
2421 | c->partial = page->next; | |
49e22585 CL |
2422 | stat(s, CPU_PARTIAL_ALLOC); |
2423 | c->freelist = NULL; | |
2424 | goto redo; | |
81819f0f CL |
2425 | } |
2426 | ||
188fd063 | 2427 | freelist = new_slab_objects(s, gfpflags, node, &c); |
01ad8a7b | 2428 | |
f4697436 | 2429 | if (unlikely(!freelist)) { |
9a02d699 | 2430 | slab_out_of_memory(s, gfpflags, node); |
f4697436 | 2431 | return NULL; |
81819f0f | 2432 | } |
2cfb7455 | 2433 | |
f6e7def7 | 2434 | page = c->page; |
5091b74a | 2435 | if (likely(!kmem_cache_debug(s) && pfmemalloc_match(page, gfpflags))) |
4b6f0750 | 2436 | goto load_freelist; |
2cfb7455 | 2437 | |
497b66f2 | 2438 | /* Only entered in the debug case */ |
d0e0ac97 CG |
2439 | if (kmem_cache_debug(s) && |
2440 | !alloc_debug_processing(s, page, freelist, addr)) | |
497b66f2 | 2441 | goto new_slab; /* Slab failed checks. Next slab needed */ |
894b8788 | 2442 | |
f6e7def7 | 2443 | deactivate_slab(s, page, get_freepointer(s, freelist)); |
c17dda40 CL |
2444 | c->page = NULL; |
2445 | c->freelist = NULL; | |
6faa6833 | 2446 | return freelist; |
894b8788 CL |
2447 | } |
2448 | ||
a380a3c7 CL |
2449 | /* |
2450 | * Another one that disabled interrupt and compensates for possible | |
2451 | * cpu changes by refetching the per cpu area pointer. | |
2452 | */ | |
2453 | static void *__slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node, | |
2454 | unsigned long addr, struct kmem_cache_cpu *c) | |
2455 | { | |
2456 | void *p; | |
2457 | unsigned long flags; | |
2458 | ||
2459 | local_irq_save(flags); | |
2460 | #ifdef CONFIG_PREEMPT | |
2461 | /* | |
2462 | * We may have been preempted and rescheduled on a different | |
2463 | * cpu before disabling interrupts. Need to reload cpu area | |
2464 | * pointer. | |
2465 | */ | |
2466 | c = this_cpu_ptr(s->cpu_slab); | |
2467 | #endif | |
2468 | ||
2469 | p = ___slab_alloc(s, gfpflags, node, addr, c); | |
2470 | local_irq_restore(flags); | |
2471 | return p; | |
2472 | } | |
2473 | ||
894b8788 CL |
2474 | /* |
2475 | * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc) | |
2476 | * have the fastpath folded into their functions. So no function call | |
2477 | * overhead for requests that can be satisfied on the fastpath. | |
2478 | * | |
2479 | * The fastpath works by first checking if the lockless freelist can be used. | |
2480 | * If not then __slab_alloc is called for slow processing. | |
2481 | * | |
2482 | * Otherwise we can simply pick the next object from the lockless free list. | |
2483 | */ | |
2b847c3c | 2484 | static __always_inline void *slab_alloc_node(struct kmem_cache *s, |
ce71e27c | 2485 | gfp_t gfpflags, int node, unsigned long addr) |
894b8788 | 2486 | { |
03ec0ed5 | 2487 | void *object; |
dfb4f096 | 2488 | struct kmem_cache_cpu *c; |
57d437d2 | 2489 | struct page *page; |
8a5ec0ba | 2490 | unsigned long tid; |
1f84260c | 2491 | |
8135be5a VD |
2492 | s = slab_pre_alloc_hook(s, gfpflags); |
2493 | if (!s) | |
773ff60e | 2494 | return NULL; |
8a5ec0ba | 2495 | redo: |
8a5ec0ba CL |
2496 | /* |
2497 | * Must read kmem_cache cpu data via this cpu ptr. Preemption is | |
2498 | * enabled. We may switch back and forth between cpus while | |
2499 | * reading from one cpu area. That does not matter as long | |
2500 | * as we end up on the original cpu again when doing the cmpxchg. | |
7cccd80b | 2501 | * |
9aabf810 JK |
2502 | * We should guarantee that tid and kmem_cache are retrieved on |
2503 | * the same cpu. It could be different if CONFIG_PREEMPT so we need | |
2504 | * to check if it is matched or not. | |
8a5ec0ba | 2505 | */ |
9aabf810 JK |
2506 | do { |
2507 | tid = this_cpu_read(s->cpu_slab->tid); | |
2508 | c = raw_cpu_ptr(s->cpu_slab); | |
859b7a0e MR |
2509 | } while (IS_ENABLED(CONFIG_PREEMPT) && |
2510 | unlikely(tid != READ_ONCE(c->tid))); | |
9aabf810 JK |
2511 | |
2512 | /* | |
2513 | * Irqless object alloc/free algorithm used here depends on sequence | |
2514 | * of fetching cpu_slab's data. tid should be fetched before anything | |
2515 | * on c to guarantee that object and page associated with previous tid | |
2516 | * won't be used with current tid. If we fetch tid first, object and | |
2517 | * page could be one associated with next tid and our alloc/free | |
2518 | * request will be failed. In this case, we will retry. So, no problem. | |
2519 | */ | |
2520 | barrier(); | |
8a5ec0ba | 2521 | |
8a5ec0ba CL |
2522 | /* |
2523 | * The transaction ids are globally unique per cpu and per operation on | |
2524 | * a per cpu queue. Thus they can be guarantee that the cmpxchg_double | |
2525 | * occurs on the right processor and that there was no operation on the | |
2526 | * linked list in between. | |
2527 | */ | |
8a5ec0ba | 2528 | |
9dfc6e68 | 2529 | object = c->freelist; |
57d437d2 | 2530 | page = c->page; |
8eae1492 | 2531 | if (unlikely(!object || !node_match(page, node))) { |
dfb4f096 | 2532 | object = __slab_alloc(s, gfpflags, node, addr, c); |
8eae1492 DH |
2533 | stat(s, ALLOC_SLOWPATH); |
2534 | } else { | |
0ad9500e ED |
2535 | void *next_object = get_freepointer_safe(s, object); |
2536 | ||
8a5ec0ba | 2537 | /* |
25985edc | 2538 | * The cmpxchg will only match if there was no additional |
8a5ec0ba CL |
2539 | * operation and if we are on the right processor. |
2540 | * | |
d0e0ac97 CG |
2541 | * The cmpxchg does the following atomically (without lock |
2542 | * semantics!) | |
8a5ec0ba CL |
2543 | * 1. Relocate first pointer to the current per cpu area. |
2544 | * 2. Verify that tid and freelist have not been changed | |
2545 | * 3. If they were not changed replace tid and freelist | |
2546 | * | |
d0e0ac97 CG |
2547 | * Since this is without lock semantics the protection is only |
2548 | * against code executing on this cpu *not* from access by | |
2549 | * other cpus. | |
8a5ec0ba | 2550 | */ |
933393f5 | 2551 | if (unlikely(!this_cpu_cmpxchg_double( |
8a5ec0ba CL |
2552 | s->cpu_slab->freelist, s->cpu_slab->tid, |
2553 | object, tid, | |
0ad9500e | 2554 | next_object, next_tid(tid)))) { |
8a5ec0ba CL |
2555 | |
2556 | note_cmpxchg_failure("slab_alloc", s, tid); | |
2557 | goto redo; | |
2558 | } | |
0ad9500e | 2559 | prefetch_freepointer(s, next_object); |
84e554e6 | 2560 | stat(s, ALLOC_FASTPATH); |
894b8788 | 2561 | } |
8a5ec0ba | 2562 | |
74e2134f | 2563 | if (unlikely(gfpflags & __GFP_ZERO) && object) |
3b0efdfa | 2564 | memset(object, 0, s->object_size); |
d07dbea4 | 2565 | |
03ec0ed5 | 2566 | slab_post_alloc_hook(s, gfpflags, 1, &object); |
5a896d9e | 2567 | |
894b8788 | 2568 | return object; |
81819f0f CL |
2569 | } |
2570 | ||
2b847c3c EG |
2571 | static __always_inline void *slab_alloc(struct kmem_cache *s, |
2572 | gfp_t gfpflags, unsigned long addr) | |
2573 | { | |
2574 | return slab_alloc_node(s, gfpflags, NUMA_NO_NODE, addr); | |
2575 | } | |
2576 | ||
81819f0f CL |
2577 | void *kmem_cache_alloc(struct kmem_cache *s, gfp_t gfpflags) |
2578 | { | |
2b847c3c | 2579 | void *ret = slab_alloc(s, gfpflags, _RET_IP_); |
5b882be4 | 2580 | |
d0e0ac97 CG |
2581 | trace_kmem_cache_alloc(_RET_IP_, ret, s->object_size, |
2582 | s->size, gfpflags); | |
5b882be4 EGM |
2583 | |
2584 | return ret; | |
81819f0f CL |
2585 | } |
2586 | EXPORT_SYMBOL(kmem_cache_alloc); | |
2587 | ||
0f24f128 | 2588 | #ifdef CONFIG_TRACING |
4a92379b RK |
2589 | void *kmem_cache_alloc_trace(struct kmem_cache *s, gfp_t gfpflags, size_t size) |
2590 | { | |
2b847c3c | 2591 | void *ret = slab_alloc(s, gfpflags, _RET_IP_); |
4a92379b | 2592 | trace_kmalloc(_RET_IP_, ret, size, s->size, gfpflags); |
0316bec2 | 2593 | kasan_kmalloc(s, ret, size); |
4a92379b RK |
2594 | return ret; |
2595 | } | |
2596 | EXPORT_SYMBOL(kmem_cache_alloc_trace); | |
5b882be4 EGM |
2597 | #endif |
2598 | ||
81819f0f CL |
2599 | #ifdef CONFIG_NUMA |
2600 | void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t gfpflags, int node) | |
2601 | { | |
2b847c3c | 2602 | void *ret = slab_alloc_node(s, gfpflags, node, _RET_IP_); |
5b882be4 | 2603 | |
ca2b84cb | 2604 | trace_kmem_cache_alloc_node(_RET_IP_, ret, |
3b0efdfa | 2605 | s->object_size, s->size, gfpflags, node); |
5b882be4 EGM |
2606 | |
2607 | return ret; | |
81819f0f CL |
2608 | } |
2609 | EXPORT_SYMBOL(kmem_cache_alloc_node); | |
81819f0f | 2610 | |
0f24f128 | 2611 | #ifdef CONFIG_TRACING |
4a92379b | 2612 | void *kmem_cache_alloc_node_trace(struct kmem_cache *s, |
5b882be4 | 2613 | gfp_t gfpflags, |
4a92379b | 2614 | int node, size_t size) |
5b882be4 | 2615 | { |
2b847c3c | 2616 | void *ret = slab_alloc_node(s, gfpflags, node, _RET_IP_); |
4a92379b RK |
2617 | |
2618 | trace_kmalloc_node(_RET_IP_, ret, | |
2619 | size, s->size, gfpflags, node); | |
0316bec2 AR |
2620 | |
2621 | kasan_kmalloc(s, ret, size); | |
4a92379b | 2622 | return ret; |
5b882be4 | 2623 | } |
4a92379b | 2624 | EXPORT_SYMBOL(kmem_cache_alloc_node_trace); |
5b882be4 | 2625 | #endif |
5d1f57e4 | 2626 | #endif |
5b882be4 | 2627 | |
81819f0f | 2628 | /* |
94e4d712 | 2629 | * Slow path handling. This may still be called frequently since objects |
894b8788 | 2630 | * have a longer lifetime than the cpu slabs in most processing loads. |
81819f0f | 2631 | * |
894b8788 CL |
2632 | * So we still attempt to reduce cache line usage. Just take the slab |
2633 | * lock and free the item. If there is no additional partial page | |
2634 | * handling required then we can return immediately. | |
81819f0f | 2635 | */ |
894b8788 | 2636 | static void __slab_free(struct kmem_cache *s, struct page *page, |
81084651 JDB |
2637 | void *head, void *tail, int cnt, |
2638 | unsigned long addr) | |
2639 | ||
81819f0f CL |
2640 | { |
2641 | void *prior; | |
2cfb7455 | 2642 | int was_frozen; |
2cfb7455 CL |
2643 | struct page new; |
2644 | unsigned long counters; | |
2645 | struct kmem_cache_node *n = NULL; | |
61728d1e | 2646 | unsigned long uninitialized_var(flags); |
81819f0f | 2647 | |
8a5ec0ba | 2648 | stat(s, FREE_SLOWPATH); |
81819f0f | 2649 | |
19c7ff9e | 2650 | if (kmem_cache_debug(s) && |
81084651 JDB |
2651 | !(n = free_debug_processing(s, page, head, tail, cnt, |
2652 | addr, &flags))) | |
80f08c19 | 2653 | return; |
6446faa2 | 2654 | |
2cfb7455 | 2655 | do { |
837d678d JK |
2656 | if (unlikely(n)) { |
2657 | spin_unlock_irqrestore(&n->list_lock, flags); | |
2658 | n = NULL; | |
2659 | } | |
2cfb7455 CL |
2660 | prior = page->freelist; |
2661 | counters = page->counters; | |
81084651 | 2662 | set_freepointer(s, tail, prior); |
2cfb7455 CL |
2663 | new.counters = counters; |
2664 | was_frozen = new.frozen; | |
81084651 | 2665 | new.inuse -= cnt; |
837d678d | 2666 | if ((!new.inuse || !prior) && !was_frozen) { |
49e22585 | 2667 | |
c65c1877 | 2668 | if (kmem_cache_has_cpu_partial(s) && !prior) { |
49e22585 CL |
2669 | |
2670 | /* | |
d0e0ac97 CG |
2671 | * Slab was on no list before and will be |
2672 | * partially empty | |
2673 | * We can defer the list move and instead | |
2674 | * freeze it. | |
49e22585 CL |
2675 | */ |
2676 | new.frozen = 1; | |
2677 | ||
c65c1877 | 2678 | } else { /* Needs to be taken off a list */ |
49e22585 | 2679 | |
b455def2 | 2680 | n = get_node(s, page_to_nid(page)); |
49e22585 CL |
2681 | /* |
2682 | * Speculatively acquire the list_lock. | |
2683 | * If the cmpxchg does not succeed then we may | |
2684 | * drop the list_lock without any processing. | |
2685 | * | |
2686 | * Otherwise the list_lock will synchronize with | |
2687 | * other processors updating the list of slabs. | |
2688 | */ | |
2689 | spin_lock_irqsave(&n->list_lock, flags); | |
2690 | ||
2691 | } | |
2cfb7455 | 2692 | } |
81819f0f | 2693 | |
2cfb7455 CL |
2694 | } while (!cmpxchg_double_slab(s, page, |
2695 | prior, counters, | |
81084651 | 2696 | head, new.counters, |
2cfb7455 | 2697 | "__slab_free")); |
81819f0f | 2698 | |
2cfb7455 | 2699 | if (likely(!n)) { |
49e22585 CL |
2700 | |
2701 | /* | |
2702 | * If we just froze the page then put it onto the | |
2703 | * per cpu partial list. | |
2704 | */ | |
8028dcea | 2705 | if (new.frozen && !was_frozen) { |
49e22585 | 2706 | put_cpu_partial(s, page, 1); |
8028dcea AS |
2707 | stat(s, CPU_PARTIAL_FREE); |
2708 | } | |
49e22585 | 2709 | /* |
2cfb7455 CL |
2710 | * The list lock was not taken therefore no list |
2711 | * activity can be necessary. | |
2712 | */ | |
b455def2 L |
2713 | if (was_frozen) |
2714 | stat(s, FREE_FROZEN); | |
2715 | return; | |
2716 | } | |
81819f0f | 2717 | |
8a5b20ae | 2718 | if (unlikely(!new.inuse && n->nr_partial >= s->min_partial)) |
837d678d JK |
2719 | goto slab_empty; |
2720 | ||
81819f0f | 2721 | /* |
837d678d JK |
2722 | * Objects left in the slab. If it was not on the partial list before |
2723 | * then add it. | |
81819f0f | 2724 | */ |
345c905d JK |
2725 | if (!kmem_cache_has_cpu_partial(s) && unlikely(!prior)) { |
2726 | if (kmem_cache_debug(s)) | |
c65c1877 | 2727 | remove_full(s, n, page); |
837d678d JK |
2728 | add_partial(n, page, DEACTIVATE_TO_TAIL); |
2729 | stat(s, FREE_ADD_PARTIAL); | |
8ff12cfc | 2730 | } |
80f08c19 | 2731 | spin_unlock_irqrestore(&n->list_lock, flags); |
81819f0f CL |
2732 | return; |
2733 | ||
2734 | slab_empty: | |
a973e9dd | 2735 | if (prior) { |
81819f0f | 2736 | /* |
6fbabb20 | 2737 | * Slab on the partial list. |
81819f0f | 2738 | */ |
5cc6eee8 | 2739 | remove_partial(n, page); |
84e554e6 | 2740 | stat(s, FREE_REMOVE_PARTIAL); |
c65c1877 | 2741 | } else { |
6fbabb20 | 2742 | /* Slab must be on the full list */ |
c65c1877 PZ |
2743 | remove_full(s, n, page); |
2744 | } | |
2cfb7455 | 2745 | |
80f08c19 | 2746 | spin_unlock_irqrestore(&n->list_lock, flags); |
84e554e6 | 2747 | stat(s, FREE_SLAB); |
81819f0f | 2748 | discard_slab(s, page); |
81819f0f CL |
2749 | } |
2750 | ||
894b8788 CL |
2751 | /* |
2752 | * Fastpath with forced inlining to produce a kfree and kmem_cache_free that | |
2753 | * can perform fastpath freeing without additional function calls. | |
2754 | * | |
2755 | * The fastpath is only possible if we are freeing to the current cpu slab | |
2756 | * of this processor. This typically the case if we have just allocated | |
2757 | * the item before. | |
2758 | * | |
2759 | * If fastpath is not possible then fall back to __slab_free where we deal | |
2760 | * with all sorts of special processing. | |
81084651 JDB |
2761 | * |
2762 | * Bulk free of a freelist with several objects (all pointing to the | |
2763 | * same page) possible by specifying head and tail ptr, plus objects | |
2764 | * count (cnt). Bulk free indicated by tail pointer being set. | |
894b8788 | 2765 | */ |
81084651 JDB |
2766 | static __always_inline void slab_free(struct kmem_cache *s, struct page *page, |
2767 | void *head, void *tail, int cnt, | |
2768 | unsigned long addr) | |
894b8788 | 2769 | { |
81084651 | 2770 | void *tail_obj = tail ? : head; |
dfb4f096 | 2771 | struct kmem_cache_cpu *c; |
8a5ec0ba | 2772 | unsigned long tid; |
1f84260c | 2773 | |
81084651 | 2774 | slab_free_freelist_hook(s, head, tail); |
c016b0bd | 2775 | |
8a5ec0ba CL |
2776 | redo: |
2777 | /* | |
2778 | * Determine the currently cpus per cpu slab. | |
2779 | * The cpu may change afterward. However that does not matter since | |
2780 | * data is retrieved via this pointer. If we are on the same cpu | |
2ae44005 | 2781 | * during the cmpxchg then the free will succeed. |
8a5ec0ba | 2782 | */ |
9aabf810 JK |
2783 | do { |
2784 | tid = this_cpu_read(s->cpu_slab->tid); | |
2785 | c = raw_cpu_ptr(s->cpu_slab); | |
859b7a0e MR |
2786 | } while (IS_ENABLED(CONFIG_PREEMPT) && |
2787 | unlikely(tid != READ_ONCE(c->tid))); | |
c016b0bd | 2788 | |
9aabf810 JK |
2789 | /* Same with comment on barrier() in slab_alloc_node() */ |
2790 | barrier(); | |
c016b0bd | 2791 | |
442b06bc | 2792 | if (likely(page == c->page)) { |
81084651 | 2793 | set_freepointer(s, tail_obj, c->freelist); |
8a5ec0ba | 2794 | |
933393f5 | 2795 | if (unlikely(!this_cpu_cmpxchg_double( |
8a5ec0ba CL |
2796 | s->cpu_slab->freelist, s->cpu_slab->tid, |
2797 | c->freelist, tid, | |
81084651 | 2798 | head, next_tid(tid)))) { |
8a5ec0ba CL |
2799 | |
2800 | note_cmpxchg_failure("slab_free", s, tid); | |
2801 | goto redo; | |
2802 | } | |
84e554e6 | 2803 | stat(s, FREE_FASTPATH); |
894b8788 | 2804 | } else |
81084651 | 2805 | __slab_free(s, page, head, tail_obj, cnt, addr); |
894b8788 | 2806 | |
894b8788 CL |
2807 | } |
2808 | ||
81819f0f CL |
2809 | void kmem_cache_free(struct kmem_cache *s, void *x) |
2810 | { | |
b9ce5ef4 GC |
2811 | s = cache_from_obj(s, x); |
2812 | if (!s) | |
79576102 | 2813 | return; |
81084651 | 2814 | slab_free(s, virt_to_head_page(x), x, NULL, 1, _RET_IP_); |
ca2b84cb | 2815 | trace_kmem_cache_free(_RET_IP_, x); |
81819f0f CL |
2816 | } |
2817 | EXPORT_SYMBOL(kmem_cache_free); | |
2818 | ||
d0ecd894 | 2819 | struct detached_freelist { |
fbd02630 | 2820 | struct page *page; |
d0ecd894 JDB |
2821 | void *tail; |
2822 | void *freelist; | |
2823 | int cnt; | |
2824 | }; | |
fbd02630 | 2825 | |
d0ecd894 JDB |
2826 | /* |
2827 | * This function progressively scans the array with free objects (with | |
2828 | * a limited look ahead) and extract objects belonging to the same | |
2829 | * page. It builds a detached freelist directly within the given | |
2830 | * page/objects. This can happen without any need for | |
2831 | * synchronization, because the objects are owned by running process. | |
2832 | * The freelist is build up as a single linked list in the objects. | |
2833 | * The idea is, that this detached freelist can then be bulk | |
2834 | * transferred to the real freelist(s), but only requiring a single | |
2835 | * synchronization primitive. Look ahead in the array is limited due | |
2836 | * to performance reasons. | |
2837 | */ | |
2838 | static int build_detached_freelist(struct kmem_cache *s, size_t size, | |
2839 | void **p, struct detached_freelist *df) | |
2840 | { | |
2841 | size_t first_skipped_index = 0; | |
2842 | int lookahead = 3; | |
2843 | void *object; | |
fbd02630 | 2844 | |
d0ecd894 JDB |
2845 | /* Always re-init detached_freelist */ |
2846 | df->page = NULL; | |
fbd02630 | 2847 | |
d0ecd894 JDB |
2848 | do { |
2849 | object = p[--size]; | |
2850 | } while (!object && size); | |
3eed034d | 2851 | |
d0ecd894 JDB |
2852 | if (!object) |
2853 | return 0; | |
fbd02630 | 2854 | |
d0ecd894 JDB |
2855 | /* Start new detached freelist */ |
2856 | set_freepointer(s, object, NULL); | |
2857 | df->page = virt_to_head_page(object); | |
2858 | df->tail = object; | |
2859 | df->freelist = object; | |
2860 | p[size] = NULL; /* mark object processed */ | |
2861 | df->cnt = 1; | |
2862 | ||
2863 | while (size) { | |
2864 | object = p[--size]; | |
2865 | if (!object) | |
2866 | continue; /* Skip processed objects */ | |
2867 | ||
2868 | /* df->page is always set at this point */ | |
2869 | if (df->page == virt_to_head_page(object)) { | |
2870 | /* Opportunity build freelist */ | |
2871 | set_freepointer(s, object, df->freelist); | |
2872 | df->freelist = object; | |
2873 | df->cnt++; | |
2874 | p[size] = NULL; /* mark object processed */ | |
2875 | ||
2876 | continue; | |
fbd02630 | 2877 | } |
d0ecd894 JDB |
2878 | |
2879 | /* Limit look ahead search */ | |
2880 | if (!--lookahead) | |
2881 | break; | |
2882 | ||
2883 | if (!first_skipped_index) | |
2884 | first_skipped_index = size + 1; | |
fbd02630 | 2885 | } |
d0ecd894 JDB |
2886 | |
2887 | return first_skipped_index; | |
2888 | } | |
2889 | ||
2890 | ||
2891 | /* Note that interrupts must be enabled when calling this function. */ | |
03374518 | 2892 | void kmem_cache_free_bulk(struct kmem_cache *orig_s, size_t size, void **p) |
d0ecd894 JDB |
2893 | { |
2894 | if (WARN_ON(!size)) | |
2895 | return; | |
2896 | ||
2897 | do { | |
2898 | struct detached_freelist df; | |
03374518 JDB |
2899 | struct kmem_cache *s; |
2900 | ||
2901 | /* Support for memcg */ | |
2902 | s = cache_from_obj(orig_s, p[size - 1]); | |
d0ecd894 JDB |
2903 | |
2904 | size = build_detached_freelist(s, size, p, &df); | |
2905 | if (unlikely(!df.page)) | |
2906 | continue; | |
2907 | ||
2908 | slab_free(s, df.page, df.freelist, df.tail, df.cnt, _RET_IP_); | |
2909 | } while (likely(size)); | |
484748f0 CL |
2910 | } |
2911 | EXPORT_SYMBOL(kmem_cache_free_bulk); | |
2912 | ||
994eb764 | 2913 | /* Note that interrupts must be enabled when calling this function. */ |
865762a8 JDB |
2914 | int kmem_cache_alloc_bulk(struct kmem_cache *s, gfp_t flags, size_t size, |
2915 | void **p) | |
484748f0 | 2916 | { |
994eb764 JDB |
2917 | struct kmem_cache_cpu *c; |
2918 | int i; | |
2919 | ||
03ec0ed5 JDB |
2920 | /* memcg and kmem_cache debug support */ |
2921 | s = slab_pre_alloc_hook(s, flags); | |
2922 | if (unlikely(!s)) | |
2923 | return false; | |
994eb764 JDB |
2924 | /* |
2925 | * Drain objects in the per cpu slab, while disabling local | |
2926 | * IRQs, which protects against PREEMPT and interrupts | |
2927 | * handlers invoking normal fastpath. | |
2928 | */ | |
2929 | local_irq_disable(); | |
2930 | c = this_cpu_ptr(s->cpu_slab); | |
2931 | ||
2932 | for (i = 0; i < size; i++) { | |
2933 | void *object = c->freelist; | |
2934 | ||
ebe909e0 | 2935 | if (unlikely(!object)) { |
ebe909e0 JDB |
2936 | /* |
2937 | * Invoking slow path likely have side-effect | |
2938 | * of re-populating per CPU c->freelist | |
2939 | */ | |
87098373 | 2940 | p[i] = ___slab_alloc(s, flags, NUMA_NO_NODE, |
ebe909e0 | 2941 | _RET_IP_, c); |
87098373 CL |
2942 | if (unlikely(!p[i])) |
2943 | goto error; | |
2944 | ||
ebe909e0 JDB |
2945 | c = this_cpu_ptr(s->cpu_slab); |
2946 | continue; /* goto for-loop */ | |
2947 | } | |
994eb764 JDB |
2948 | c->freelist = get_freepointer(s, object); |
2949 | p[i] = object; | |
2950 | } | |
2951 | c->tid = next_tid(c->tid); | |
2952 | local_irq_enable(); | |
2953 | ||
2954 | /* Clear memory outside IRQ disabled fastpath loop */ | |
2955 | if (unlikely(flags & __GFP_ZERO)) { | |
2956 | int j; | |
2957 | ||
2958 | for (j = 0; j < i; j++) | |
2959 | memset(p[j], 0, s->object_size); | |
2960 | } | |
2961 | ||
03ec0ed5 JDB |
2962 | /* memcg and kmem_cache debug support */ |
2963 | slab_post_alloc_hook(s, flags, size, p); | |
865762a8 | 2964 | return i; |
87098373 | 2965 | error: |
87098373 | 2966 | local_irq_enable(); |
03ec0ed5 JDB |
2967 | slab_post_alloc_hook(s, flags, i, p); |
2968 | __kmem_cache_free_bulk(s, i, p); | |
865762a8 | 2969 | return 0; |
484748f0 CL |
2970 | } |
2971 | EXPORT_SYMBOL(kmem_cache_alloc_bulk); | |
2972 | ||
2973 | ||
81819f0f | 2974 | /* |
672bba3a CL |
2975 | * Object placement in a slab is made very easy because we always start at |
2976 | * offset 0. If we tune the size of the object to the alignment then we can | |
2977 | * get the required alignment by putting one properly sized object after | |
2978 | * another. | |
81819f0f CL |
2979 | * |
2980 | * Notice that the allocation order determines the sizes of the per cpu | |
2981 | * caches. Each processor has always one slab available for allocations. | |
2982 | * Increasing the allocation order reduces the number of times that slabs | |
672bba3a | 2983 | * must be moved on and off the partial lists and is therefore a factor in |
81819f0f | 2984 | * locking overhead. |
81819f0f CL |
2985 | */ |
2986 | ||
2987 | /* | |
2988 | * Mininum / Maximum order of slab pages. This influences locking overhead | |
2989 | * and slab fragmentation. A higher order reduces the number of partial slabs | |
2990 | * and increases the number of allocations possible without having to | |
2991 | * take the list_lock. | |
2992 | */ | |
2993 | static int slub_min_order; | |
114e9e89 | 2994 | static int slub_max_order = PAGE_ALLOC_COSTLY_ORDER; |
9b2cd506 | 2995 | static int slub_min_objects; |
81819f0f | 2996 | |
81819f0f CL |
2997 | /* |
2998 | * Calculate the order of allocation given an slab object size. | |
2999 | * | |
672bba3a CL |
3000 | * The order of allocation has significant impact on performance and other |
3001 | * system components. Generally order 0 allocations should be preferred since | |
3002 | * order 0 does not cause fragmentation in the page allocator. Larger objects | |
3003 | * be problematic to put into order 0 slabs because there may be too much | |
c124f5b5 | 3004 | * unused space left. We go to a higher order if more than 1/16th of the slab |
672bba3a CL |
3005 | * would be wasted. |
3006 | * | |
3007 | * In order to reach satisfactory performance we must ensure that a minimum | |
3008 | * number of objects is in one slab. Otherwise we may generate too much | |
3009 | * activity on the partial lists which requires taking the list_lock. This is | |
3010 | * less a concern for large slabs though which are rarely used. | |
81819f0f | 3011 | * |
672bba3a CL |
3012 | * slub_max_order specifies the order where we begin to stop considering the |
3013 | * number of objects in a slab as critical. If we reach slub_max_order then | |
3014 | * we try to keep the page order as low as possible. So we accept more waste | |
3015 | * of space in favor of a small page order. | |
81819f0f | 3016 | * |
672bba3a CL |
3017 | * Higher order allocations also allow the placement of more objects in a |
3018 | * slab and thereby reduce object handling overhead. If the user has | |
3019 | * requested a higher mininum order then we start with that one instead of | |
3020 | * the smallest order which will fit the object. | |
81819f0f | 3021 | */ |
5e6d444e | 3022 | static inline int slab_order(int size, int min_objects, |
ab9a0f19 | 3023 | int max_order, int fract_leftover, int reserved) |
81819f0f CL |
3024 | { |
3025 | int order; | |
3026 | int rem; | |
6300ea75 | 3027 | int min_order = slub_min_order; |
81819f0f | 3028 | |
ab9a0f19 | 3029 | if (order_objects(min_order, size, reserved) > MAX_OBJS_PER_PAGE) |
210b5c06 | 3030 | return get_order(size * MAX_OBJS_PER_PAGE) - 1; |
39b26464 | 3031 | |
9f835703 | 3032 | for (order = max(min_order, get_order(min_objects * size + reserved)); |
5e6d444e | 3033 | order <= max_order; order++) { |
81819f0f | 3034 | |
5e6d444e | 3035 | unsigned long slab_size = PAGE_SIZE << order; |
81819f0f | 3036 | |
ab9a0f19 | 3037 | rem = (slab_size - reserved) % size; |
81819f0f | 3038 | |
5e6d444e | 3039 | if (rem <= slab_size / fract_leftover) |
81819f0f | 3040 | break; |
81819f0f | 3041 | } |
672bba3a | 3042 | |
81819f0f CL |
3043 | return order; |
3044 | } | |
3045 | ||
ab9a0f19 | 3046 | static inline int calculate_order(int size, int reserved) |
5e6d444e CL |
3047 | { |
3048 | int order; | |
3049 | int min_objects; | |
3050 | int fraction; | |
e8120ff1 | 3051 | int max_objects; |
5e6d444e CL |
3052 | |
3053 | /* | |
3054 | * Attempt to find best configuration for a slab. This | |
3055 | * works by first attempting to generate a layout with | |
3056 | * the best configuration and backing off gradually. | |
3057 | * | |
422ff4d7 | 3058 | * First we increase the acceptable waste in a slab. Then |
5e6d444e CL |
3059 | * we reduce the minimum objects required in a slab. |
3060 | */ | |
3061 | min_objects = slub_min_objects; | |
9b2cd506 CL |
3062 | if (!min_objects) |
3063 | min_objects = 4 * (fls(nr_cpu_ids) + 1); | |
ab9a0f19 | 3064 | max_objects = order_objects(slub_max_order, size, reserved); |
e8120ff1 ZY |
3065 | min_objects = min(min_objects, max_objects); |
3066 | ||
5e6d444e | 3067 | while (min_objects > 1) { |
c124f5b5 | 3068 | fraction = 16; |
5e6d444e CL |
3069 | while (fraction >= 4) { |
3070 | order = slab_order(size, min_objects, | |
ab9a0f19 | 3071 | slub_max_order, fraction, reserved); |
5e6d444e CL |
3072 | if (order <= slub_max_order) |
3073 | return order; | |
3074 | fraction /= 2; | |
3075 | } | |
5086c389 | 3076 | min_objects--; |
5e6d444e CL |
3077 | } |
3078 | ||
3079 | /* | |
3080 | * We were unable to place multiple objects in a slab. Now | |
3081 | * lets see if we can place a single object there. | |
3082 | */ | |
ab9a0f19 | 3083 | order = slab_order(size, 1, slub_max_order, 1, reserved); |
5e6d444e CL |
3084 | if (order <= slub_max_order) |
3085 | return order; | |
3086 | ||
3087 | /* | |
3088 | * Doh this slab cannot be placed using slub_max_order. | |
3089 | */ | |
ab9a0f19 | 3090 | order = slab_order(size, 1, MAX_ORDER, 1, reserved); |
818cf590 | 3091 | if (order < MAX_ORDER) |
5e6d444e CL |
3092 | return order; |
3093 | return -ENOSYS; | |
3094 | } | |
3095 | ||
5595cffc | 3096 | static void |
4053497d | 3097 | init_kmem_cache_node(struct kmem_cache_node *n) |
81819f0f CL |
3098 | { |
3099 | n->nr_partial = 0; | |
81819f0f CL |
3100 | spin_lock_init(&n->list_lock); |
3101 | INIT_LIST_HEAD(&n->partial); | |
8ab1372f | 3102 | #ifdef CONFIG_SLUB_DEBUG |
0f389ec6 | 3103 | atomic_long_set(&n->nr_slabs, 0); |
02b71b70 | 3104 | atomic_long_set(&n->total_objects, 0); |
643b1138 | 3105 | INIT_LIST_HEAD(&n->full); |
8ab1372f | 3106 | #endif |
81819f0f CL |
3107 | } |
3108 | ||
55136592 | 3109 | static inline int alloc_kmem_cache_cpus(struct kmem_cache *s) |
4c93c355 | 3110 | { |
6c182dc0 | 3111 | BUILD_BUG_ON(PERCPU_DYNAMIC_EARLY_SIZE < |
95a05b42 | 3112 | KMALLOC_SHIFT_HIGH * sizeof(struct kmem_cache_cpu)); |
4c93c355 | 3113 | |
8a5ec0ba | 3114 | /* |
d4d84fef CM |
3115 | * Must align to double word boundary for the double cmpxchg |
3116 | * instructions to work; see __pcpu_double_call_return_bool(). | |
8a5ec0ba | 3117 | */ |
d4d84fef CM |
3118 | s->cpu_slab = __alloc_percpu(sizeof(struct kmem_cache_cpu), |
3119 | 2 * sizeof(void *)); | |
8a5ec0ba CL |
3120 | |
3121 | if (!s->cpu_slab) | |
3122 | return 0; | |
3123 | ||
3124 | init_kmem_cache_cpus(s); | |
4c93c355 | 3125 | |
8a5ec0ba | 3126 | return 1; |
4c93c355 | 3127 | } |
4c93c355 | 3128 | |
51df1142 CL |
3129 | static struct kmem_cache *kmem_cache_node; |
3130 | ||
81819f0f CL |
3131 | /* |
3132 | * No kmalloc_node yet so do it by hand. We know that this is the first | |
3133 | * slab on the node for this slabcache. There are no concurrent accesses | |
3134 | * possible. | |
3135 | * | |
721ae22a ZYW |
3136 | * Note that this function only works on the kmem_cache_node |
3137 | * when allocating for the kmem_cache_node. This is used for bootstrapping | |
4c93c355 | 3138 | * memory on a fresh node that has no slab structures yet. |
81819f0f | 3139 | */ |
55136592 | 3140 | static void early_kmem_cache_node_alloc(int node) |
81819f0f CL |
3141 | { |
3142 | struct page *page; | |
3143 | struct kmem_cache_node *n; | |
3144 | ||
51df1142 | 3145 | BUG_ON(kmem_cache_node->size < sizeof(struct kmem_cache_node)); |
81819f0f | 3146 | |
51df1142 | 3147 | page = new_slab(kmem_cache_node, GFP_NOWAIT, node); |
81819f0f CL |
3148 | |
3149 | BUG_ON(!page); | |
a2f92ee7 | 3150 | if (page_to_nid(page) != node) { |
f9f58285 FF |
3151 | pr_err("SLUB: Unable to allocate memory from node %d\n", node); |
3152 | pr_err("SLUB: Allocating a useless per node structure in order to be able to continue\n"); | |
a2f92ee7 CL |
3153 | } |
3154 | ||
81819f0f CL |
3155 | n = page->freelist; |
3156 | BUG_ON(!n); | |
51df1142 | 3157 | page->freelist = get_freepointer(kmem_cache_node, n); |
e6e82ea1 | 3158 | page->inuse = 1; |
8cb0a506 | 3159 | page->frozen = 0; |
51df1142 | 3160 | kmem_cache_node->node[node] = n; |
8ab1372f | 3161 | #ifdef CONFIG_SLUB_DEBUG |
f7cb1933 | 3162 | init_object(kmem_cache_node, n, SLUB_RED_ACTIVE); |
51df1142 | 3163 | init_tracking(kmem_cache_node, n); |
8ab1372f | 3164 | #endif |
0316bec2 | 3165 | kasan_kmalloc(kmem_cache_node, n, sizeof(struct kmem_cache_node)); |
4053497d | 3166 | init_kmem_cache_node(n); |
51df1142 | 3167 | inc_slabs_node(kmem_cache_node, node, page->objects); |
6446faa2 | 3168 | |
67b6c900 | 3169 | /* |
1e4dd946 SR |
3170 | * No locks need to be taken here as it has just been |
3171 | * initialized and there is no concurrent access. | |
67b6c900 | 3172 | */ |
1e4dd946 | 3173 | __add_partial(n, page, DEACTIVATE_TO_HEAD); |
81819f0f CL |
3174 | } |
3175 | ||
3176 | static void free_kmem_cache_nodes(struct kmem_cache *s) | |
3177 | { | |
3178 | int node; | |
fa45dc25 | 3179 | struct kmem_cache_node *n; |
81819f0f | 3180 | |
fa45dc25 CL |
3181 | for_each_kmem_cache_node(s, node, n) { |
3182 | kmem_cache_free(kmem_cache_node, n); | |
81819f0f CL |
3183 | s->node[node] = NULL; |
3184 | } | |
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. |
69cb8e6b CL |
3446 | * This is called from kmem_cache_close(). We must be the last thread |
3447 | * using the cache and therefore we do not need to lock anymore. | |
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 | ||
33b12c38 | 3453 | list_for_each_entry_safe(page, h, &n->partial, lru) { |
81819f0f | 3454 | if (!page->inuse) { |
1e4dd946 | 3455 | __remove_partial(n, page); |
81819f0f | 3456 | discard_slab(s, page); |
33b12c38 CL |
3457 | } else { |
3458 | list_slab_objects(s, page, | |
945cf2b6 | 3459 | "Objects remaining in %s on kmem_cache_close()"); |
599870b1 | 3460 | } |
33b12c38 | 3461 | } |
81819f0f CL |
3462 | } |
3463 | ||
3464 | /* | |
672bba3a | 3465 | * Release all resources used by a slab cache. |
81819f0f | 3466 | */ |
0c710013 | 3467 | static inline int kmem_cache_close(struct kmem_cache *s) |
81819f0f CL |
3468 | { |
3469 | int node; | |
fa45dc25 | 3470 | struct kmem_cache_node *n; |
81819f0f CL |
3471 | |
3472 | flush_all(s); | |
81819f0f | 3473 | /* Attempt to free all objects */ |
fa45dc25 | 3474 | for_each_kmem_cache_node(s, node, n) { |
599870b1 CL |
3475 | free_partial(s, n); |
3476 | if (n->nr_partial || slabs_node(s, node)) | |
81819f0f CL |
3477 | return 1; |
3478 | } | |
945cf2b6 | 3479 | free_percpu(s->cpu_slab); |
81819f0f CL |
3480 | free_kmem_cache_nodes(s); |
3481 | return 0; | |
3482 | } | |
3483 | ||
945cf2b6 | 3484 | int __kmem_cache_shutdown(struct kmem_cache *s) |
81819f0f | 3485 | { |
41a21285 | 3486 | return kmem_cache_close(s); |
81819f0f | 3487 | } |
81819f0f CL |
3488 | |
3489 | /******************************************************************** | |
3490 | * Kmalloc subsystem | |
3491 | *******************************************************************/ | |
3492 | ||
81819f0f CL |
3493 | static int __init setup_slub_min_order(char *str) |
3494 | { | |
06428780 | 3495 | get_option(&str, &slub_min_order); |
81819f0f CL |
3496 | |
3497 | return 1; | |
3498 | } | |
3499 | ||
3500 | __setup("slub_min_order=", setup_slub_min_order); | |
3501 | ||
3502 | static int __init setup_slub_max_order(char *str) | |
3503 | { | |
06428780 | 3504 | get_option(&str, &slub_max_order); |
818cf590 | 3505 | slub_max_order = min(slub_max_order, MAX_ORDER - 1); |
81819f0f CL |
3506 | |
3507 | return 1; | |
3508 | } | |
3509 | ||
3510 | __setup("slub_max_order=", setup_slub_max_order); | |
3511 | ||
3512 | static int __init setup_slub_min_objects(char *str) | |
3513 | { | |
06428780 | 3514 | get_option(&str, &slub_min_objects); |
81819f0f CL |
3515 | |
3516 | return 1; | |
3517 | } | |
3518 | ||
3519 | __setup("slub_min_objects=", setup_slub_min_objects); | |
3520 | ||
81819f0f CL |
3521 | void *__kmalloc(size_t size, gfp_t flags) |
3522 | { | |
aadb4bc4 | 3523 | struct kmem_cache *s; |
5b882be4 | 3524 | void *ret; |
81819f0f | 3525 | |
95a05b42 | 3526 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) |
eada35ef | 3527 | return kmalloc_large(size, flags); |
aadb4bc4 | 3528 | |
2c59dd65 | 3529 | s = kmalloc_slab(size, flags); |
aadb4bc4 CL |
3530 | |
3531 | if (unlikely(ZERO_OR_NULL_PTR(s))) | |
6cb8f913 CL |
3532 | return s; |
3533 | ||
2b847c3c | 3534 | ret = slab_alloc(s, flags, _RET_IP_); |
5b882be4 | 3535 | |
ca2b84cb | 3536 | trace_kmalloc(_RET_IP_, ret, size, s->size, flags); |
5b882be4 | 3537 | |
0316bec2 AR |
3538 | kasan_kmalloc(s, ret, size); |
3539 | ||
5b882be4 | 3540 | return ret; |
81819f0f CL |
3541 | } |
3542 | EXPORT_SYMBOL(__kmalloc); | |
3543 | ||
5d1f57e4 | 3544 | #ifdef CONFIG_NUMA |
f619cfe1 CL |
3545 | static void *kmalloc_large_node(size_t size, gfp_t flags, int node) |
3546 | { | |
b1eeab67 | 3547 | struct page *page; |
e4f7c0b4 | 3548 | void *ptr = NULL; |
f619cfe1 | 3549 | |
52383431 VD |
3550 | flags |= __GFP_COMP | __GFP_NOTRACK; |
3551 | page = alloc_kmem_pages_node(node, flags, get_order(size)); | |
f619cfe1 | 3552 | if (page) |
e4f7c0b4 CM |
3553 | ptr = page_address(page); |
3554 | ||
d56791b3 | 3555 | kmalloc_large_node_hook(ptr, size, flags); |
e4f7c0b4 | 3556 | return ptr; |
f619cfe1 CL |
3557 | } |
3558 | ||
81819f0f CL |
3559 | void *__kmalloc_node(size_t size, gfp_t flags, int node) |
3560 | { | |
aadb4bc4 | 3561 | struct kmem_cache *s; |
5b882be4 | 3562 | void *ret; |
81819f0f | 3563 | |
95a05b42 | 3564 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) { |
5b882be4 EGM |
3565 | ret = kmalloc_large_node(size, flags, node); |
3566 | ||
ca2b84cb EGM |
3567 | trace_kmalloc_node(_RET_IP_, ret, |
3568 | size, PAGE_SIZE << get_order(size), | |
3569 | flags, node); | |
5b882be4 EGM |
3570 | |
3571 | return ret; | |
3572 | } | |
aadb4bc4 | 3573 | |
2c59dd65 | 3574 | s = kmalloc_slab(size, flags); |
aadb4bc4 CL |
3575 | |
3576 | if (unlikely(ZERO_OR_NULL_PTR(s))) | |
6cb8f913 CL |
3577 | return s; |
3578 | ||
2b847c3c | 3579 | ret = slab_alloc_node(s, flags, node, _RET_IP_); |
5b882be4 | 3580 | |
ca2b84cb | 3581 | trace_kmalloc_node(_RET_IP_, ret, size, s->size, flags, node); |
5b882be4 | 3582 | |
0316bec2 AR |
3583 | kasan_kmalloc(s, ret, size); |
3584 | ||
5b882be4 | 3585 | return ret; |
81819f0f CL |
3586 | } |
3587 | EXPORT_SYMBOL(__kmalloc_node); | |
3588 | #endif | |
3589 | ||
0316bec2 | 3590 | static size_t __ksize(const void *object) |
81819f0f | 3591 | { |
272c1d21 | 3592 | struct page *page; |
81819f0f | 3593 | |
ef8b4520 | 3594 | if (unlikely(object == ZERO_SIZE_PTR)) |
272c1d21 CL |
3595 | return 0; |
3596 | ||
294a80a8 | 3597 | page = virt_to_head_page(object); |
294a80a8 | 3598 | |
76994412 PE |
3599 | if (unlikely(!PageSlab(page))) { |
3600 | WARN_ON(!PageCompound(page)); | |
294a80a8 | 3601 | return PAGE_SIZE << compound_order(page); |
76994412 | 3602 | } |
81819f0f | 3603 | |
1b4f59e3 | 3604 | return slab_ksize(page->slab_cache); |
81819f0f | 3605 | } |
0316bec2 AR |
3606 | |
3607 | size_t ksize(const void *object) | |
3608 | { | |
3609 | size_t size = __ksize(object); | |
3610 | /* We assume that ksize callers could use whole allocated area, | |
3611 | so we need unpoison this area. */ | |
3612 | kasan_krealloc(object, size); | |
3613 | return size; | |
3614 | } | |
b1aabecd | 3615 | EXPORT_SYMBOL(ksize); |
81819f0f CL |
3616 | |
3617 | void kfree(const void *x) | |
3618 | { | |
81819f0f | 3619 | struct page *page; |
5bb983b0 | 3620 | void *object = (void *)x; |
81819f0f | 3621 | |
2121db74 PE |
3622 | trace_kfree(_RET_IP_, x); |
3623 | ||
2408c550 | 3624 | if (unlikely(ZERO_OR_NULL_PTR(x))) |
81819f0f CL |
3625 | return; |
3626 | ||
b49af68f | 3627 | page = virt_to_head_page(x); |
aadb4bc4 | 3628 | if (unlikely(!PageSlab(page))) { |
0937502a | 3629 | BUG_ON(!PageCompound(page)); |
d56791b3 | 3630 | kfree_hook(x); |
52383431 | 3631 | __free_kmem_pages(page, compound_order(page)); |
aadb4bc4 CL |
3632 | return; |
3633 | } | |
81084651 | 3634 | slab_free(page->slab_cache, page, object, NULL, 1, _RET_IP_); |
81819f0f CL |
3635 | } |
3636 | EXPORT_SYMBOL(kfree); | |
3637 | ||
832f37f5 VD |
3638 | #define SHRINK_PROMOTE_MAX 32 |
3639 | ||
2086d26a | 3640 | /* |
832f37f5 VD |
3641 | * kmem_cache_shrink discards empty slabs and promotes the slabs filled |
3642 | * up most to the head of the partial lists. New allocations will then | |
3643 | * fill those up and thus they can be removed from the partial lists. | |
672bba3a CL |
3644 | * |
3645 | * The slabs with the least items are placed last. This results in them | |
3646 | * being allocated from last increasing the chance that the last objects | |
3647 | * are freed in them. | |
2086d26a | 3648 | */ |
d6e0b7fa | 3649 | int __kmem_cache_shrink(struct kmem_cache *s, bool deactivate) |
2086d26a CL |
3650 | { |
3651 | int node; | |
3652 | int i; | |
3653 | struct kmem_cache_node *n; | |
3654 | struct page *page; | |
3655 | struct page *t; | |
832f37f5 VD |
3656 | struct list_head discard; |
3657 | struct list_head promote[SHRINK_PROMOTE_MAX]; | |
2086d26a | 3658 | unsigned long flags; |
ce3712d7 | 3659 | int ret = 0; |
2086d26a | 3660 | |
d6e0b7fa VD |
3661 | if (deactivate) { |
3662 | /* | |
3663 | * Disable empty slabs caching. Used to avoid pinning offline | |
3664 | * memory cgroups by kmem pages that can be freed. | |
3665 | */ | |
3666 | s->cpu_partial = 0; | |
3667 | s->min_partial = 0; | |
3668 | ||
3669 | /* | |
3670 | * s->cpu_partial is checked locklessly (see put_cpu_partial), | |
3671 | * so we have to make sure the change is visible. | |
3672 | */ | |
3673 | kick_all_cpus_sync(); | |
3674 | } | |
3675 | ||
2086d26a | 3676 | flush_all(s); |
fa45dc25 | 3677 | for_each_kmem_cache_node(s, node, n) { |
832f37f5 VD |
3678 | INIT_LIST_HEAD(&discard); |
3679 | for (i = 0; i < SHRINK_PROMOTE_MAX; i++) | |
3680 | INIT_LIST_HEAD(promote + i); | |
2086d26a CL |
3681 | |
3682 | spin_lock_irqsave(&n->list_lock, flags); | |
3683 | ||
3684 | /* | |
832f37f5 | 3685 | * Build lists of slabs to discard or promote. |
2086d26a | 3686 | * |
672bba3a CL |
3687 | * Note that concurrent frees may occur while we hold the |
3688 | * list_lock. page->inuse here is the upper limit. | |
2086d26a CL |
3689 | */ |
3690 | list_for_each_entry_safe(page, t, &n->partial, lru) { | |
832f37f5 VD |
3691 | int free = page->objects - page->inuse; |
3692 | ||
3693 | /* Do not reread page->inuse */ | |
3694 | barrier(); | |
3695 | ||
3696 | /* We do not keep full slabs on the list */ | |
3697 | BUG_ON(free <= 0); | |
3698 | ||
3699 | if (free == page->objects) { | |
3700 | list_move(&page->lru, &discard); | |
69cb8e6b | 3701 | n->nr_partial--; |
832f37f5 VD |
3702 | } else if (free <= SHRINK_PROMOTE_MAX) |
3703 | list_move(&page->lru, promote + free - 1); | |
2086d26a CL |
3704 | } |
3705 | ||
2086d26a | 3706 | /* |
832f37f5 VD |
3707 | * Promote the slabs filled up most to the head of the |
3708 | * partial list. | |
2086d26a | 3709 | */ |
832f37f5 VD |
3710 | for (i = SHRINK_PROMOTE_MAX - 1; i >= 0; i--) |
3711 | list_splice(promote + i, &n->partial); | |
2086d26a | 3712 | |
2086d26a | 3713 | spin_unlock_irqrestore(&n->list_lock, flags); |
69cb8e6b CL |
3714 | |
3715 | /* Release empty slabs */ | |
832f37f5 | 3716 | list_for_each_entry_safe(page, t, &discard, lru) |
69cb8e6b | 3717 | discard_slab(s, page); |
ce3712d7 VD |
3718 | |
3719 | if (slabs_node(s, node)) | |
3720 | ret = 1; | |
2086d26a CL |
3721 | } |
3722 | ||
ce3712d7 | 3723 | return ret; |
2086d26a | 3724 | } |
2086d26a | 3725 | |
b9049e23 YG |
3726 | static int slab_mem_going_offline_callback(void *arg) |
3727 | { | |
3728 | struct kmem_cache *s; | |
3729 | ||
18004c5d | 3730 | mutex_lock(&slab_mutex); |
b9049e23 | 3731 | list_for_each_entry(s, &slab_caches, list) |
d6e0b7fa | 3732 | __kmem_cache_shrink(s, false); |
18004c5d | 3733 | mutex_unlock(&slab_mutex); |
b9049e23 YG |
3734 | |
3735 | return 0; | |
3736 | } | |
3737 | ||
3738 | static void slab_mem_offline_callback(void *arg) | |
3739 | { | |
3740 | struct kmem_cache_node *n; | |
3741 | struct kmem_cache *s; | |
3742 | struct memory_notify *marg = arg; | |
3743 | int offline_node; | |
3744 | ||
b9d5ab25 | 3745 | offline_node = marg->status_change_nid_normal; |
b9049e23 YG |
3746 | |
3747 | /* | |
3748 | * If the node still has available memory. we need kmem_cache_node | |
3749 | * for it yet. | |
3750 | */ | |
3751 | if (offline_node < 0) | |
3752 | return; | |
3753 | ||
18004c5d | 3754 | mutex_lock(&slab_mutex); |
b9049e23 YG |
3755 | list_for_each_entry(s, &slab_caches, list) { |
3756 | n = get_node(s, offline_node); | |
3757 | if (n) { | |
3758 | /* | |
3759 | * if n->nr_slabs > 0, slabs still exist on the node | |
3760 | * that is going down. We were unable to free them, | |
c9404c9c | 3761 | * and offline_pages() function shouldn't call this |
b9049e23 YG |
3762 | * callback. So, we must fail. |
3763 | */ | |
0f389ec6 | 3764 | BUG_ON(slabs_node(s, offline_node)); |
b9049e23 YG |
3765 | |
3766 | s->node[offline_node] = NULL; | |
8de66a0c | 3767 | kmem_cache_free(kmem_cache_node, n); |
b9049e23 YG |
3768 | } |
3769 | } | |
18004c5d | 3770 | mutex_unlock(&slab_mutex); |
b9049e23 YG |
3771 | } |
3772 | ||
3773 | static int slab_mem_going_online_callback(void *arg) | |
3774 | { | |
3775 | struct kmem_cache_node *n; | |
3776 | struct kmem_cache *s; | |
3777 | struct memory_notify *marg = arg; | |
b9d5ab25 | 3778 | int nid = marg->status_change_nid_normal; |
b9049e23 YG |
3779 | int ret = 0; |
3780 | ||
3781 | /* | |
3782 | * If the node's memory is already available, then kmem_cache_node is | |
3783 | * already created. Nothing to do. | |
3784 | */ | |
3785 | if (nid < 0) | |
3786 | return 0; | |
3787 | ||
3788 | /* | |
0121c619 | 3789 | * We are bringing a node online. No memory is available yet. We must |
b9049e23 YG |
3790 | * allocate a kmem_cache_node structure in order to bring the node |
3791 | * online. | |
3792 | */ | |
18004c5d | 3793 | mutex_lock(&slab_mutex); |
b9049e23 YG |
3794 | list_for_each_entry(s, &slab_caches, list) { |
3795 | /* | |
3796 | * XXX: kmem_cache_alloc_node will fallback to other nodes | |
3797 | * since memory is not yet available from the node that | |
3798 | * is brought up. | |
3799 | */ | |
8de66a0c | 3800 | n = kmem_cache_alloc(kmem_cache_node, GFP_KERNEL); |
b9049e23 YG |
3801 | if (!n) { |
3802 | ret = -ENOMEM; | |
3803 | goto out; | |
3804 | } | |
4053497d | 3805 | init_kmem_cache_node(n); |
b9049e23 YG |
3806 | s->node[nid] = n; |
3807 | } | |
3808 | out: | |
18004c5d | 3809 | mutex_unlock(&slab_mutex); |
b9049e23 YG |
3810 | return ret; |
3811 | } | |
3812 | ||
3813 | static int slab_memory_callback(struct notifier_block *self, | |
3814 | unsigned long action, void *arg) | |
3815 | { | |
3816 | int ret = 0; | |
3817 | ||
3818 | switch (action) { | |
3819 | case MEM_GOING_ONLINE: | |
3820 | ret = slab_mem_going_online_callback(arg); | |
3821 | break; | |
3822 | case MEM_GOING_OFFLINE: | |
3823 | ret = slab_mem_going_offline_callback(arg); | |
3824 | break; | |
3825 | case MEM_OFFLINE: | |
3826 | case MEM_CANCEL_ONLINE: | |
3827 | slab_mem_offline_callback(arg); | |
3828 | break; | |
3829 | case MEM_ONLINE: | |
3830 | case MEM_CANCEL_OFFLINE: | |
3831 | break; | |
3832 | } | |
dc19f9db KH |
3833 | if (ret) |
3834 | ret = notifier_from_errno(ret); | |
3835 | else | |
3836 | ret = NOTIFY_OK; | |
b9049e23 YG |
3837 | return ret; |
3838 | } | |
3839 | ||
3ac38faa AM |
3840 | static struct notifier_block slab_memory_callback_nb = { |
3841 | .notifier_call = slab_memory_callback, | |
3842 | .priority = SLAB_CALLBACK_PRI, | |
3843 | }; | |
b9049e23 | 3844 | |
81819f0f CL |
3845 | /******************************************************************** |
3846 | * Basic setup of slabs | |
3847 | *******************************************************************/ | |
3848 | ||
51df1142 CL |
3849 | /* |
3850 | * Used for early kmem_cache structures that were allocated using | |
dffb4d60 CL |
3851 | * the page allocator. Allocate them properly then fix up the pointers |
3852 | * that may be pointing to the wrong kmem_cache structure. | |
51df1142 CL |
3853 | */ |
3854 | ||
dffb4d60 | 3855 | static struct kmem_cache * __init bootstrap(struct kmem_cache *static_cache) |
51df1142 CL |
3856 | { |
3857 | int node; | |
dffb4d60 | 3858 | struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT); |
fa45dc25 | 3859 | struct kmem_cache_node *n; |
51df1142 | 3860 | |
dffb4d60 | 3861 | memcpy(s, static_cache, kmem_cache->object_size); |
51df1142 | 3862 | |
7d557b3c GC |
3863 | /* |
3864 | * This runs very early, and only the boot processor is supposed to be | |
3865 | * up. Even if it weren't true, IRQs are not up so we couldn't fire | |
3866 | * IPIs around. | |
3867 | */ | |
3868 | __flush_cpu_slab(s, smp_processor_id()); | |
fa45dc25 | 3869 | for_each_kmem_cache_node(s, node, n) { |
51df1142 CL |
3870 | struct page *p; |
3871 | ||
fa45dc25 CL |
3872 | list_for_each_entry(p, &n->partial, lru) |
3873 | p->slab_cache = s; | |
51df1142 | 3874 | |
607bf324 | 3875 | #ifdef CONFIG_SLUB_DEBUG |
fa45dc25 CL |
3876 | list_for_each_entry(p, &n->full, lru) |
3877 | p->slab_cache = s; | |
51df1142 | 3878 | #endif |
51df1142 | 3879 | } |
f7ce3190 | 3880 | slab_init_memcg_params(s); |
dffb4d60 CL |
3881 | list_add(&s->list, &slab_caches); |
3882 | return s; | |
51df1142 CL |
3883 | } |
3884 | ||
81819f0f CL |
3885 | void __init kmem_cache_init(void) |
3886 | { | |
dffb4d60 CL |
3887 | static __initdata struct kmem_cache boot_kmem_cache, |
3888 | boot_kmem_cache_node; | |
51df1142 | 3889 | |
fc8d8620 SG |
3890 | if (debug_guardpage_minorder()) |
3891 | slub_max_order = 0; | |
3892 | ||
dffb4d60 CL |
3893 | kmem_cache_node = &boot_kmem_cache_node; |
3894 | kmem_cache = &boot_kmem_cache; | |
51df1142 | 3895 | |
dffb4d60 CL |
3896 | create_boot_cache(kmem_cache_node, "kmem_cache_node", |
3897 | sizeof(struct kmem_cache_node), SLAB_HWCACHE_ALIGN); | |
b9049e23 | 3898 | |
3ac38faa | 3899 | register_hotmemory_notifier(&slab_memory_callback_nb); |
81819f0f CL |
3900 | |
3901 | /* Able to allocate the per node structures */ | |
3902 | slab_state = PARTIAL; | |
3903 | ||
dffb4d60 CL |
3904 | create_boot_cache(kmem_cache, "kmem_cache", |
3905 | offsetof(struct kmem_cache, node) + | |
3906 | nr_node_ids * sizeof(struct kmem_cache_node *), | |
3907 | SLAB_HWCACHE_ALIGN); | |
8a13a4cc | 3908 | |
dffb4d60 | 3909 | kmem_cache = bootstrap(&boot_kmem_cache); |
81819f0f | 3910 | |
51df1142 CL |
3911 | /* |
3912 | * Allocate kmem_cache_node properly from the kmem_cache slab. | |
3913 | * kmem_cache_node is separately allocated so no need to | |
3914 | * update any list pointers. | |
3915 | */ | |
dffb4d60 | 3916 | kmem_cache_node = bootstrap(&boot_kmem_cache_node); |
51df1142 CL |
3917 | |
3918 | /* Now we can use the kmem_cache to allocate kmalloc slabs */ | |
34cc6990 | 3919 | setup_kmalloc_cache_index_table(); |
f97d5f63 | 3920 | create_kmalloc_caches(0); |
81819f0f CL |
3921 | |
3922 | #ifdef CONFIG_SMP | |
3923 | register_cpu_notifier(&slab_notifier); | |
9dfc6e68 | 3924 | #endif |
81819f0f | 3925 | |
f9f58285 | 3926 | pr_info("SLUB: HWalign=%d, Order=%d-%d, MinObjects=%d, CPUs=%d, Nodes=%d\n", |
f97d5f63 | 3927 | cache_line_size(), |
81819f0f CL |
3928 | slub_min_order, slub_max_order, slub_min_objects, |
3929 | nr_cpu_ids, nr_node_ids); | |
3930 | } | |
3931 | ||
7e85ee0c PE |
3932 | void __init kmem_cache_init_late(void) |
3933 | { | |
7e85ee0c PE |
3934 | } |
3935 | ||
2633d7a0 | 3936 | struct kmem_cache * |
a44cb944 VD |
3937 | __kmem_cache_alias(const char *name, size_t size, size_t align, |
3938 | unsigned long flags, void (*ctor)(void *)) | |
81819f0f | 3939 | { |
426589f5 | 3940 | struct kmem_cache *s, *c; |
81819f0f | 3941 | |
a44cb944 | 3942 | s = find_mergeable(size, align, flags, name, ctor); |
81819f0f CL |
3943 | if (s) { |
3944 | s->refcount++; | |
84d0ddd6 | 3945 | |
81819f0f CL |
3946 | /* |
3947 | * Adjust the object sizes so that we clear | |
3948 | * the complete object on kzalloc. | |
3949 | */ | |
3b0efdfa | 3950 | s->object_size = max(s->object_size, (int)size); |
81819f0f | 3951 | s->inuse = max_t(int, s->inuse, ALIGN(size, sizeof(void *))); |
6446faa2 | 3952 | |
426589f5 | 3953 | for_each_memcg_cache(c, s) { |
84d0ddd6 VD |
3954 | c->object_size = s->object_size; |
3955 | c->inuse = max_t(int, c->inuse, | |
3956 | ALIGN(size, sizeof(void *))); | |
3957 | } | |
3958 | ||
7b8f3b66 | 3959 | if (sysfs_slab_alias(s, name)) { |
7b8f3b66 | 3960 | s->refcount--; |
cbb79694 | 3961 | s = NULL; |
7b8f3b66 | 3962 | } |
a0e1d1be | 3963 | } |
6446faa2 | 3964 | |
cbb79694 CL |
3965 | return s; |
3966 | } | |
84c1cf62 | 3967 | |
8a13a4cc | 3968 | int __kmem_cache_create(struct kmem_cache *s, unsigned long flags) |
cbb79694 | 3969 | { |
aac3a166 PE |
3970 | int err; |
3971 | ||
3972 | err = kmem_cache_open(s, flags); | |
3973 | if (err) | |
3974 | return err; | |
20cea968 | 3975 | |
45530c44 CL |
3976 | /* Mutex is not taken during early boot */ |
3977 | if (slab_state <= UP) | |
3978 | return 0; | |
3979 | ||
107dab5c | 3980 | memcg_propagate_slab_attrs(s); |
aac3a166 | 3981 | err = sysfs_slab_add(s); |
aac3a166 PE |
3982 | if (err) |
3983 | kmem_cache_close(s); | |
20cea968 | 3984 | |
aac3a166 | 3985 | return err; |
81819f0f | 3986 | } |
81819f0f | 3987 | |
81819f0f | 3988 | #ifdef CONFIG_SMP |
81819f0f | 3989 | /* |
672bba3a CL |
3990 | * Use the cpu notifier to insure that the cpu slabs are flushed when |
3991 | * necessary. | |
81819f0f | 3992 | */ |
0db0628d | 3993 | static int slab_cpuup_callback(struct notifier_block *nfb, |
81819f0f CL |
3994 | unsigned long action, void *hcpu) |
3995 | { | |
3996 | long cpu = (long)hcpu; | |
5b95a4ac CL |
3997 | struct kmem_cache *s; |
3998 | unsigned long flags; | |
81819f0f CL |
3999 | |
4000 | switch (action) { | |
4001 | case CPU_UP_CANCELED: | |
8bb78442 | 4002 | case CPU_UP_CANCELED_FROZEN: |
81819f0f | 4003 | case CPU_DEAD: |
8bb78442 | 4004 | case CPU_DEAD_FROZEN: |
18004c5d | 4005 | mutex_lock(&slab_mutex); |
5b95a4ac CL |
4006 | list_for_each_entry(s, &slab_caches, list) { |
4007 | local_irq_save(flags); | |
4008 | __flush_cpu_slab(s, cpu); | |
4009 | local_irq_restore(flags); | |
4010 | } | |
18004c5d | 4011 | mutex_unlock(&slab_mutex); |
81819f0f CL |
4012 | break; |
4013 | default: | |
4014 | break; | |
4015 | } | |
4016 | return NOTIFY_OK; | |
4017 | } | |
4018 | ||
0db0628d | 4019 | static struct notifier_block slab_notifier = { |
3adbefee | 4020 | .notifier_call = slab_cpuup_callback |
06428780 | 4021 | }; |
81819f0f CL |
4022 | |
4023 | #endif | |
4024 | ||
ce71e27c | 4025 | void *__kmalloc_track_caller(size_t size, gfp_t gfpflags, unsigned long caller) |
81819f0f | 4026 | { |
aadb4bc4 | 4027 | struct kmem_cache *s; |
94b528d0 | 4028 | void *ret; |
aadb4bc4 | 4029 | |
95a05b42 | 4030 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) |
eada35ef PE |
4031 | return kmalloc_large(size, gfpflags); |
4032 | ||
2c59dd65 | 4033 | s = kmalloc_slab(size, gfpflags); |
81819f0f | 4034 | |
2408c550 | 4035 | if (unlikely(ZERO_OR_NULL_PTR(s))) |
6cb8f913 | 4036 | return s; |
81819f0f | 4037 | |
2b847c3c | 4038 | ret = slab_alloc(s, gfpflags, caller); |
94b528d0 | 4039 | |
25985edc | 4040 | /* Honor the call site pointer we received. */ |
ca2b84cb | 4041 | trace_kmalloc(caller, ret, size, s->size, gfpflags); |
94b528d0 EGM |
4042 | |
4043 | return ret; | |
81819f0f CL |
4044 | } |
4045 | ||
5d1f57e4 | 4046 | #ifdef CONFIG_NUMA |
81819f0f | 4047 | void *__kmalloc_node_track_caller(size_t size, gfp_t gfpflags, |
ce71e27c | 4048 | int node, unsigned long caller) |
81819f0f | 4049 | { |
aadb4bc4 | 4050 | struct kmem_cache *s; |
94b528d0 | 4051 | void *ret; |
aadb4bc4 | 4052 | |
95a05b42 | 4053 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) { |
d3e14aa3 XF |
4054 | ret = kmalloc_large_node(size, gfpflags, node); |
4055 | ||
4056 | trace_kmalloc_node(caller, ret, | |
4057 | size, PAGE_SIZE << get_order(size), | |
4058 | gfpflags, node); | |
4059 | ||
4060 | return ret; | |
4061 | } | |
eada35ef | 4062 | |
2c59dd65 | 4063 | s = kmalloc_slab(size, gfpflags); |
81819f0f | 4064 | |
2408c550 | 4065 | if (unlikely(ZERO_OR_NULL_PTR(s))) |
6cb8f913 | 4066 | return s; |
81819f0f | 4067 | |
2b847c3c | 4068 | ret = slab_alloc_node(s, gfpflags, node, caller); |
94b528d0 | 4069 | |
25985edc | 4070 | /* Honor the call site pointer we received. */ |
ca2b84cb | 4071 | trace_kmalloc_node(caller, ret, size, s->size, gfpflags, node); |
94b528d0 EGM |
4072 | |
4073 | return ret; | |
81819f0f | 4074 | } |
5d1f57e4 | 4075 | #endif |
81819f0f | 4076 | |
ab4d5ed5 | 4077 | #ifdef CONFIG_SYSFS |
205ab99d CL |
4078 | static int count_inuse(struct page *page) |
4079 | { | |
4080 | return page->inuse; | |
4081 | } | |
4082 | ||
4083 | static int count_total(struct page *page) | |
4084 | { | |
4085 | return page->objects; | |
4086 | } | |
ab4d5ed5 | 4087 | #endif |
205ab99d | 4088 | |
ab4d5ed5 | 4089 | #ifdef CONFIG_SLUB_DEBUG |
434e245d CL |
4090 | static int validate_slab(struct kmem_cache *s, struct page *page, |
4091 | unsigned long *map) | |
53e15af0 CL |
4092 | { |
4093 | void *p; | |
a973e9dd | 4094 | void *addr = page_address(page); |
53e15af0 CL |
4095 | |
4096 | if (!check_slab(s, page) || | |
4097 | !on_freelist(s, page, NULL)) | |
4098 | return 0; | |
4099 | ||
4100 | /* Now we know that a valid freelist exists */ | |
39b26464 | 4101 | bitmap_zero(map, page->objects); |
53e15af0 | 4102 | |
5f80b13a CL |
4103 | get_map(s, page, map); |
4104 | for_each_object(p, s, addr, page->objects) { | |
4105 | if (test_bit(slab_index(p, s, addr), map)) | |
4106 | if (!check_object(s, page, p, SLUB_RED_INACTIVE)) | |
4107 | return 0; | |
53e15af0 CL |
4108 | } |
4109 | ||
224a88be | 4110 | for_each_object(p, s, addr, page->objects) |
7656c72b | 4111 | if (!test_bit(slab_index(p, s, addr), map)) |
37d57443 | 4112 | if (!check_object(s, page, p, SLUB_RED_ACTIVE)) |
53e15af0 CL |
4113 | return 0; |
4114 | return 1; | |
4115 | } | |
4116 | ||
434e245d CL |
4117 | static void validate_slab_slab(struct kmem_cache *s, struct page *page, |
4118 | unsigned long *map) | |
53e15af0 | 4119 | { |
881db7fb CL |
4120 | slab_lock(page); |
4121 | validate_slab(s, page, map); | |
4122 | slab_unlock(page); | |
53e15af0 CL |
4123 | } |
4124 | ||
434e245d CL |
4125 | static int validate_slab_node(struct kmem_cache *s, |
4126 | struct kmem_cache_node *n, unsigned long *map) | |
53e15af0 CL |
4127 | { |
4128 | unsigned long count = 0; | |
4129 | struct page *page; | |
4130 | unsigned long flags; | |
4131 | ||
4132 | spin_lock_irqsave(&n->list_lock, flags); | |
4133 | ||
4134 | list_for_each_entry(page, &n->partial, lru) { | |
434e245d | 4135 | validate_slab_slab(s, page, map); |
53e15af0 CL |
4136 | count++; |
4137 | } | |
4138 | if (count != n->nr_partial) | |
f9f58285 FF |
4139 | pr_err("SLUB %s: %ld partial slabs counted but counter=%ld\n", |
4140 | s->name, count, n->nr_partial); | |
53e15af0 CL |
4141 | |
4142 | if (!(s->flags & SLAB_STORE_USER)) | |
4143 | goto out; | |
4144 | ||
4145 | list_for_each_entry(page, &n->full, lru) { | |
434e245d | 4146 | validate_slab_slab(s, page, map); |
53e15af0 CL |
4147 | count++; |
4148 | } | |
4149 | if (count != atomic_long_read(&n->nr_slabs)) | |
f9f58285 FF |
4150 | pr_err("SLUB: %s %ld slabs counted but counter=%ld\n", |
4151 | s->name, count, atomic_long_read(&n->nr_slabs)); | |
53e15af0 CL |
4152 | |
4153 | out: | |
4154 | spin_unlock_irqrestore(&n->list_lock, flags); | |
4155 | return count; | |
4156 | } | |
4157 | ||
434e245d | 4158 | static long validate_slab_cache(struct kmem_cache *s) |
53e15af0 CL |
4159 | { |
4160 | int node; | |
4161 | unsigned long count = 0; | |
205ab99d | 4162 | unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) * |
434e245d | 4163 | sizeof(unsigned long), GFP_KERNEL); |
fa45dc25 | 4164 | struct kmem_cache_node *n; |
434e245d CL |
4165 | |
4166 | if (!map) | |
4167 | return -ENOMEM; | |
53e15af0 CL |
4168 | |
4169 | flush_all(s); | |
fa45dc25 | 4170 | for_each_kmem_cache_node(s, node, n) |
434e245d | 4171 | count += validate_slab_node(s, n, map); |
434e245d | 4172 | kfree(map); |
53e15af0 CL |
4173 | return count; |
4174 | } | |
88a420e4 | 4175 | /* |
672bba3a | 4176 | * Generate lists of code addresses where slabcache objects are allocated |
88a420e4 CL |
4177 | * and freed. |
4178 | */ | |
4179 | ||
4180 | struct location { | |
4181 | unsigned long count; | |
ce71e27c | 4182 | unsigned long addr; |
45edfa58 CL |
4183 | long long sum_time; |
4184 | long min_time; | |
4185 | long max_time; | |
4186 | long min_pid; | |
4187 | long max_pid; | |
174596a0 | 4188 | DECLARE_BITMAP(cpus, NR_CPUS); |
45edfa58 | 4189 | nodemask_t nodes; |
88a420e4 CL |
4190 | }; |
4191 | ||
4192 | struct loc_track { | |
4193 | unsigned long max; | |
4194 | unsigned long count; | |
4195 | struct location *loc; | |
4196 | }; | |
4197 | ||
4198 | static void free_loc_track(struct loc_track *t) | |
4199 | { | |
4200 | if (t->max) | |
4201 | free_pages((unsigned long)t->loc, | |
4202 | get_order(sizeof(struct location) * t->max)); | |
4203 | } | |
4204 | ||
68dff6a9 | 4205 | static int alloc_loc_track(struct loc_track *t, unsigned long max, gfp_t flags) |
88a420e4 CL |
4206 | { |
4207 | struct location *l; | |
4208 | int order; | |
4209 | ||
88a420e4 CL |
4210 | order = get_order(sizeof(struct location) * max); |
4211 | ||
68dff6a9 | 4212 | l = (void *)__get_free_pages(flags, order); |
88a420e4 CL |
4213 | if (!l) |
4214 | return 0; | |
4215 | ||
4216 | if (t->count) { | |
4217 | memcpy(l, t->loc, sizeof(struct location) * t->count); | |
4218 | free_loc_track(t); | |
4219 | } | |
4220 | t->max = max; | |
4221 | t->loc = l; | |
4222 | return 1; | |
4223 | } | |
4224 | ||
4225 | static int add_location(struct loc_track *t, struct kmem_cache *s, | |
45edfa58 | 4226 | const struct track *track) |
88a420e4 CL |
4227 | { |
4228 | long start, end, pos; | |
4229 | struct location *l; | |
ce71e27c | 4230 | unsigned long caddr; |
45edfa58 | 4231 | unsigned long age = jiffies - track->when; |
88a420e4 CL |
4232 | |
4233 | start = -1; | |
4234 | end = t->count; | |
4235 | ||
4236 | for ( ; ; ) { | |
4237 | pos = start + (end - start + 1) / 2; | |
4238 | ||
4239 | /* | |
4240 | * There is nothing at "end". If we end up there | |
4241 | * we need to add something to before end. | |
4242 | */ | |
4243 | if (pos == end) | |
4244 | break; | |
4245 | ||
4246 | caddr = t->loc[pos].addr; | |
45edfa58 CL |
4247 | if (track->addr == caddr) { |
4248 | ||
4249 | l = &t->loc[pos]; | |
4250 | l->count++; | |
4251 | if (track->when) { | |
4252 | l->sum_time += age; | |
4253 | if (age < l->min_time) | |
4254 | l->min_time = age; | |
4255 | if (age > l->max_time) | |
4256 | l->max_time = age; | |
4257 | ||
4258 | if (track->pid < l->min_pid) | |
4259 | l->min_pid = track->pid; | |
4260 | if (track->pid > l->max_pid) | |
4261 | l->max_pid = track->pid; | |
4262 | ||
174596a0 RR |
4263 | cpumask_set_cpu(track->cpu, |
4264 | to_cpumask(l->cpus)); | |
45edfa58 CL |
4265 | } |
4266 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
88a420e4 CL |
4267 | return 1; |
4268 | } | |
4269 | ||
45edfa58 | 4270 | if (track->addr < caddr) |
88a420e4 CL |
4271 | end = pos; |
4272 | else | |
4273 | start = pos; | |
4274 | } | |
4275 | ||
4276 | /* | |
672bba3a | 4277 | * Not found. Insert new tracking element. |
88a420e4 | 4278 | */ |
68dff6a9 | 4279 | if (t->count >= t->max && !alloc_loc_track(t, 2 * t->max, GFP_ATOMIC)) |
88a420e4 CL |
4280 | return 0; |
4281 | ||
4282 | l = t->loc + pos; | |
4283 | if (pos < t->count) | |
4284 | memmove(l + 1, l, | |
4285 | (t->count - pos) * sizeof(struct location)); | |
4286 | t->count++; | |
4287 | l->count = 1; | |
45edfa58 CL |
4288 | l->addr = track->addr; |
4289 | l->sum_time = age; | |
4290 | l->min_time = age; | |
4291 | l->max_time = age; | |
4292 | l->min_pid = track->pid; | |
4293 | l->max_pid = track->pid; | |
174596a0 RR |
4294 | cpumask_clear(to_cpumask(l->cpus)); |
4295 | cpumask_set_cpu(track->cpu, to_cpumask(l->cpus)); | |
45edfa58 CL |
4296 | nodes_clear(l->nodes); |
4297 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
88a420e4 CL |
4298 | return 1; |
4299 | } | |
4300 | ||
4301 | static void process_slab(struct loc_track *t, struct kmem_cache *s, | |
bbd7d57b | 4302 | struct page *page, enum track_item alloc, |
a5dd5c11 | 4303 | unsigned long *map) |
88a420e4 | 4304 | { |
a973e9dd | 4305 | void *addr = page_address(page); |
88a420e4 CL |
4306 | void *p; |
4307 | ||
39b26464 | 4308 | bitmap_zero(map, page->objects); |
5f80b13a | 4309 | get_map(s, page, map); |
88a420e4 | 4310 | |
224a88be | 4311 | for_each_object(p, s, addr, page->objects) |
45edfa58 CL |
4312 | if (!test_bit(slab_index(p, s, addr), map)) |
4313 | add_location(t, s, get_track(s, p, alloc)); | |
88a420e4 CL |
4314 | } |
4315 | ||
4316 | static int list_locations(struct kmem_cache *s, char *buf, | |
4317 | enum track_item alloc) | |
4318 | { | |
e374d483 | 4319 | int len = 0; |
88a420e4 | 4320 | unsigned long i; |
68dff6a9 | 4321 | struct loc_track t = { 0, 0, NULL }; |
88a420e4 | 4322 | int node; |
bbd7d57b ED |
4323 | unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) * |
4324 | sizeof(unsigned long), GFP_KERNEL); | |
fa45dc25 | 4325 | struct kmem_cache_node *n; |
88a420e4 | 4326 | |
bbd7d57b ED |
4327 | if (!map || !alloc_loc_track(&t, PAGE_SIZE / sizeof(struct location), |
4328 | GFP_TEMPORARY)) { | |
4329 | kfree(map); | |
68dff6a9 | 4330 | return sprintf(buf, "Out of memory\n"); |
bbd7d57b | 4331 | } |
88a420e4 CL |
4332 | /* Push back cpu slabs */ |
4333 | flush_all(s); | |
4334 | ||
fa45dc25 | 4335 | for_each_kmem_cache_node(s, node, n) { |
88a420e4 CL |
4336 | unsigned long flags; |
4337 | struct page *page; | |
4338 | ||
9e86943b | 4339 | if (!atomic_long_read(&n->nr_slabs)) |
88a420e4 CL |
4340 | continue; |
4341 | ||
4342 | spin_lock_irqsave(&n->list_lock, flags); | |
4343 | list_for_each_entry(page, &n->partial, lru) | |
bbd7d57b | 4344 | process_slab(&t, s, page, alloc, map); |
88a420e4 | 4345 | list_for_each_entry(page, &n->full, lru) |
bbd7d57b | 4346 | process_slab(&t, s, page, alloc, map); |
88a420e4 CL |
4347 | spin_unlock_irqrestore(&n->list_lock, flags); |
4348 | } | |
4349 | ||
4350 | for (i = 0; i < t.count; i++) { | |
45edfa58 | 4351 | struct location *l = &t.loc[i]; |
88a420e4 | 4352 | |
9c246247 | 4353 | if (len > PAGE_SIZE - KSYM_SYMBOL_LEN - 100) |
88a420e4 | 4354 | break; |
e374d483 | 4355 | len += sprintf(buf + len, "%7ld ", l->count); |
45edfa58 CL |
4356 | |
4357 | if (l->addr) | |
62c70bce | 4358 | len += sprintf(buf + len, "%pS", (void *)l->addr); |
88a420e4 | 4359 | else |
e374d483 | 4360 | len += sprintf(buf + len, "<not-available>"); |
45edfa58 CL |
4361 | |
4362 | if (l->sum_time != l->min_time) { | |
e374d483 | 4363 | len += sprintf(buf + len, " age=%ld/%ld/%ld", |
f8bd2258 RZ |
4364 | l->min_time, |
4365 | (long)div_u64(l->sum_time, l->count), | |
4366 | l->max_time); | |
45edfa58 | 4367 | } else |
e374d483 | 4368 | len += sprintf(buf + len, " age=%ld", |
45edfa58 CL |
4369 | l->min_time); |
4370 | ||
4371 | if (l->min_pid != l->max_pid) | |
e374d483 | 4372 | len += sprintf(buf + len, " pid=%ld-%ld", |
45edfa58 CL |
4373 | l->min_pid, l->max_pid); |
4374 | else | |
e374d483 | 4375 | len += sprintf(buf + len, " pid=%ld", |
45edfa58 CL |
4376 | l->min_pid); |
4377 | ||
174596a0 RR |
4378 | if (num_online_cpus() > 1 && |
4379 | !cpumask_empty(to_cpumask(l->cpus)) && | |
5024c1d7 TH |
4380 | len < PAGE_SIZE - 60) |
4381 | len += scnprintf(buf + len, PAGE_SIZE - len - 50, | |
4382 | " cpus=%*pbl", | |
4383 | cpumask_pr_args(to_cpumask(l->cpus))); | |
45edfa58 | 4384 | |
62bc62a8 | 4385 | if (nr_online_nodes > 1 && !nodes_empty(l->nodes) && |
5024c1d7 TH |
4386 | len < PAGE_SIZE - 60) |
4387 | len += scnprintf(buf + len, PAGE_SIZE - len - 50, | |
4388 | " nodes=%*pbl", | |
4389 | nodemask_pr_args(&l->nodes)); | |
45edfa58 | 4390 | |
e374d483 | 4391 | len += sprintf(buf + len, "\n"); |
88a420e4 CL |
4392 | } |
4393 | ||
4394 | free_loc_track(&t); | |
bbd7d57b | 4395 | kfree(map); |
88a420e4 | 4396 | if (!t.count) |
e374d483 HH |
4397 | len += sprintf(buf, "No data\n"); |
4398 | return len; | |
88a420e4 | 4399 | } |
ab4d5ed5 | 4400 | #endif |
88a420e4 | 4401 | |
a5a84755 | 4402 | #ifdef SLUB_RESILIENCY_TEST |
c07b8183 | 4403 | static void __init resiliency_test(void) |
a5a84755 CL |
4404 | { |
4405 | u8 *p; | |
4406 | ||
95a05b42 | 4407 | BUILD_BUG_ON(KMALLOC_MIN_SIZE > 16 || KMALLOC_SHIFT_HIGH < 10); |
a5a84755 | 4408 | |
f9f58285 FF |
4409 | pr_err("SLUB resiliency testing\n"); |
4410 | pr_err("-----------------------\n"); | |
4411 | pr_err("A. Corruption after allocation\n"); | |
a5a84755 CL |
4412 | |
4413 | p = kzalloc(16, GFP_KERNEL); | |
4414 | p[16] = 0x12; | |
f9f58285 FF |
4415 | pr_err("\n1. kmalloc-16: Clobber Redzone/next pointer 0x12->0x%p\n\n", |
4416 | p + 16); | |
a5a84755 CL |
4417 | |
4418 | validate_slab_cache(kmalloc_caches[4]); | |
4419 | ||
4420 | /* Hmmm... The next two are dangerous */ | |
4421 | p = kzalloc(32, GFP_KERNEL); | |
4422 | p[32 + sizeof(void *)] = 0x34; | |
f9f58285 FF |
4423 | pr_err("\n2. kmalloc-32: Clobber next pointer/next slab 0x34 -> -0x%p\n", |
4424 | p); | |
4425 | pr_err("If allocated object is overwritten then not detectable\n\n"); | |
a5a84755 CL |
4426 | |
4427 | validate_slab_cache(kmalloc_caches[5]); | |
4428 | p = kzalloc(64, GFP_KERNEL); | |
4429 | p += 64 + (get_cycles() & 0xff) * sizeof(void *); | |
4430 | *p = 0x56; | |
f9f58285 FF |
4431 | pr_err("\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n", |
4432 | p); | |
4433 | pr_err("If allocated object is overwritten then not detectable\n\n"); | |
a5a84755 CL |
4434 | validate_slab_cache(kmalloc_caches[6]); |
4435 | ||
f9f58285 | 4436 | pr_err("\nB. Corruption after free\n"); |
a5a84755 CL |
4437 | p = kzalloc(128, GFP_KERNEL); |
4438 | kfree(p); | |
4439 | *p = 0x78; | |
f9f58285 | 4440 | pr_err("1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p); |
a5a84755 CL |
4441 | validate_slab_cache(kmalloc_caches[7]); |
4442 | ||
4443 | p = kzalloc(256, GFP_KERNEL); | |
4444 | kfree(p); | |
4445 | p[50] = 0x9a; | |
f9f58285 | 4446 | pr_err("\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n", p); |
a5a84755 CL |
4447 | validate_slab_cache(kmalloc_caches[8]); |
4448 | ||
4449 | p = kzalloc(512, GFP_KERNEL); | |
4450 | kfree(p); | |
4451 | p[512] = 0xab; | |
f9f58285 | 4452 | pr_err("\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p); |
a5a84755 CL |
4453 | validate_slab_cache(kmalloc_caches[9]); |
4454 | } | |
4455 | #else | |
4456 | #ifdef CONFIG_SYSFS | |
4457 | static void resiliency_test(void) {}; | |
4458 | #endif | |
4459 | #endif | |
4460 | ||
ab4d5ed5 | 4461 | #ifdef CONFIG_SYSFS |
81819f0f | 4462 | enum slab_stat_type { |
205ab99d CL |
4463 | SL_ALL, /* All slabs */ |
4464 | SL_PARTIAL, /* Only partially allocated slabs */ | |
4465 | SL_CPU, /* Only slabs used for cpu caches */ | |
4466 | SL_OBJECTS, /* Determine allocated objects not slabs */ | |
4467 | SL_TOTAL /* Determine object capacity not slabs */ | |
81819f0f CL |
4468 | }; |
4469 | ||
205ab99d | 4470 | #define SO_ALL (1 << SL_ALL) |
81819f0f CL |
4471 | #define SO_PARTIAL (1 << SL_PARTIAL) |
4472 | #define SO_CPU (1 << SL_CPU) | |
4473 | #define SO_OBJECTS (1 << SL_OBJECTS) | |
205ab99d | 4474 | #define SO_TOTAL (1 << SL_TOTAL) |
81819f0f | 4475 | |
62e5c4b4 CG |
4476 | static ssize_t show_slab_objects(struct kmem_cache *s, |
4477 | char *buf, unsigned long flags) | |
81819f0f CL |
4478 | { |
4479 | unsigned long total = 0; | |
81819f0f CL |
4480 | int node; |
4481 | int x; | |
4482 | unsigned long *nodes; | |
81819f0f | 4483 | |
e35e1a97 | 4484 | nodes = kzalloc(sizeof(unsigned long) * nr_node_ids, GFP_KERNEL); |
62e5c4b4 CG |
4485 | if (!nodes) |
4486 | return -ENOMEM; | |
81819f0f | 4487 | |
205ab99d CL |
4488 | if (flags & SO_CPU) { |
4489 | int cpu; | |
81819f0f | 4490 | |
205ab99d | 4491 | for_each_possible_cpu(cpu) { |
d0e0ac97 CG |
4492 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, |
4493 | cpu); | |
ec3ab083 | 4494 | int node; |
49e22585 | 4495 | struct page *page; |
dfb4f096 | 4496 | |
4db0c3c2 | 4497 | page = READ_ONCE(c->page); |
ec3ab083 CL |
4498 | if (!page) |
4499 | continue; | |
205ab99d | 4500 | |
ec3ab083 CL |
4501 | node = page_to_nid(page); |
4502 | if (flags & SO_TOTAL) | |
4503 | x = page->objects; | |
4504 | else if (flags & SO_OBJECTS) | |
4505 | x = page->inuse; | |
4506 | else | |
4507 | x = 1; | |
49e22585 | 4508 | |
ec3ab083 CL |
4509 | total += x; |
4510 | nodes[node] += x; | |
4511 | ||
4db0c3c2 | 4512 | page = READ_ONCE(c->partial); |
49e22585 | 4513 | if (page) { |
8afb1474 LZ |
4514 | node = page_to_nid(page); |
4515 | if (flags & SO_TOTAL) | |
4516 | WARN_ON_ONCE(1); | |
4517 | else if (flags & SO_OBJECTS) | |
4518 | WARN_ON_ONCE(1); | |
4519 | else | |
4520 | x = page->pages; | |
bc6697d8 ED |
4521 | total += x; |
4522 | nodes[node] += x; | |
49e22585 | 4523 | } |
81819f0f CL |
4524 | } |
4525 | } | |
4526 | ||
bfc8c901 | 4527 | get_online_mems(); |
ab4d5ed5 | 4528 | #ifdef CONFIG_SLUB_DEBUG |
205ab99d | 4529 | if (flags & SO_ALL) { |
fa45dc25 CL |
4530 | struct kmem_cache_node *n; |
4531 | ||
4532 | for_each_kmem_cache_node(s, node, n) { | |
205ab99d | 4533 | |
d0e0ac97 CG |
4534 | if (flags & SO_TOTAL) |
4535 | x = atomic_long_read(&n->total_objects); | |
4536 | else if (flags & SO_OBJECTS) | |
4537 | x = atomic_long_read(&n->total_objects) - | |
4538 | count_partial(n, count_free); | |
81819f0f | 4539 | else |
205ab99d | 4540 | x = atomic_long_read(&n->nr_slabs); |
81819f0f CL |
4541 | total += x; |
4542 | nodes[node] += x; | |
4543 | } | |
4544 | ||
ab4d5ed5 CL |
4545 | } else |
4546 | #endif | |
4547 | if (flags & SO_PARTIAL) { | |
fa45dc25 | 4548 | struct kmem_cache_node *n; |
81819f0f | 4549 | |
fa45dc25 | 4550 | for_each_kmem_cache_node(s, node, n) { |
205ab99d CL |
4551 | if (flags & SO_TOTAL) |
4552 | x = count_partial(n, count_total); | |
4553 | else if (flags & SO_OBJECTS) | |
4554 | x = count_partial(n, count_inuse); | |
81819f0f | 4555 | else |
205ab99d | 4556 | x = n->nr_partial; |
81819f0f CL |
4557 | total += x; |
4558 | nodes[node] += x; | |
4559 | } | |
4560 | } | |
81819f0f CL |
4561 | x = sprintf(buf, "%lu", total); |
4562 | #ifdef CONFIG_NUMA | |
fa45dc25 | 4563 | for (node = 0; node < nr_node_ids; node++) |
81819f0f CL |
4564 | if (nodes[node]) |
4565 | x += sprintf(buf + x, " N%d=%lu", | |
4566 | node, nodes[node]); | |
4567 | #endif | |
bfc8c901 | 4568 | put_online_mems(); |
81819f0f CL |
4569 | kfree(nodes); |
4570 | return x + sprintf(buf + x, "\n"); | |
4571 | } | |
4572 | ||
ab4d5ed5 | 4573 | #ifdef CONFIG_SLUB_DEBUG |
81819f0f CL |
4574 | static int any_slab_objects(struct kmem_cache *s) |
4575 | { | |
4576 | int node; | |
fa45dc25 | 4577 | struct kmem_cache_node *n; |
81819f0f | 4578 | |
fa45dc25 | 4579 | for_each_kmem_cache_node(s, node, n) |
4ea33e2d | 4580 | if (atomic_long_read(&n->total_objects)) |
81819f0f | 4581 | return 1; |
fa45dc25 | 4582 | |
81819f0f CL |
4583 | return 0; |
4584 | } | |
ab4d5ed5 | 4585 | #endif |
81819f0f CL |
4586 | |
4587 | #define to_slab_attr(n) container_of(n, struct slab_attribute, attr) | |
497888cf | 4588 | #define to_slab(n) container_of(n, struct kmem_cache, kobj) |
81819f0f CL |
4589 | |
4590 | struct slab_attribute { | |
4591 | struct attribute attr; | |
4592 | ssize_t (*show)(struct kmem_cache *s, char *buf); | |
4593 | ssize_t (*store)(struct kmem_cache *s, const char *x, size_t count); | |
4594 | }; | |
4595 | ||
4596 | #define SLAB_ATTR_RO(_name) \ | |
ab067e99 VK |
4597 | static struct slab_attribute _name##_attr = \ |
4598 | __ATTR(_name, 0400, _name##_show, NULL) | |
81819f0f CL |
4599 | |
4600 | #define SLAB_ATTR(_name) \ | |
4601 | static struct slab_attribute _name##_attr = \ | |
ab067e99 | 4602 | __ATTR(_name, 0600, _name##_show, _name##_store) |
81819f0f | 4603 | |
81819f0f CL |
4604 | static ssize_t slab_size_show(struct kmem_cache *s, char *buf) |
4605 | { | |
4606 | return sprintf(buf, "%d\n", s->size); | |
4607 | } | |
4608 | SLAB_ATTR_RO(slab_size); | |
4609 | ||
4610 | static ssize_t align_show(struct kmem_cache *s, char *buf) | |
4611 | { | |
4612 | return sprintf(buf, "%d\n", s->align); | |
4613 | } | |
4614 | SLAB_ATTR_RO(align); | |
4615 | ||
4616 | static ssize_t object_size_show(struct kmem_cache *s, char *buf) | |
4617 | { | |
3b0efdfa | 4618 | return sprintf(buf, "%d\n", s->object_size); |
81819f0f CL |
4619 | } |
4620 | SLAB_ATTR_RO(object_size); | |
4621 | ||
4622 | static ssize_t objs_per_slab_show(struct kmem_cache *s, char *buf) | |
4623 | { | |
834f3d11 | 4624 | return sprintf(buf, "%d\n", oo_objects(s->oo)); |
81819f0f CL |
4625 | } |
4626 | SLAB_ATTR_RO(objs_per_slab); | |
4627 | ||
06b285dc CL |
4628 | static ssize_t order_store(struct kmem_cache *s, |
4629 | const char *buf, size_t length) | |
4630 | { | |
0121c619 CL |
4631 | unsigned long order; |
4632 | int err; | |
4633 | ||
3dbb95f7 | 4634 | err = kstrtoul(buf, 10, &order); |
0121c619 CL |
4635 | if (err) |
4636 | return err; | |
06b285dc CL |
4637 | |
4638 | if (order > slub_max_order || order < slub_min_order) | |
4639 | return -EINVAL; | |
4640 | ||
4641 | calculate_sizes(s, order); | |
4642 | return length; | |
4643 | } | |
4644 | ||
81819f0f CL |
4645 | static ssize_t order_show(struct kmem_cache *s, char *buf) |
4646 | { | |
834f3d11 | 4647 | return sprintf(buf, "%d\n", oo_order(s->oo)); |
81819f0f | 4648 | } |
06b285dc | 4649 | SLAB_ATTR(order); |
81819f0f | 4650 | |
73d342b1 DR |
4651 | static ssize_t min_partial_show(struct kmem_cache *s, char *buf) |
4652 | { | |
4653 | return sprintf(buf, "%lu\n", s->min_partial); | |
4654 | } | |
4655 | ||
4656 | static ssize_t min_partial_store(struct kmem_cache *s, const char *buf, | |
4657 | size_t length) | |
4658 | { | |
4659 | unsigned long min; | |
4660 | int err; | |
4661 | ||
3dbb95f7 | 4662 | err = kstrtoul(buf, 10, &min); |
73d342b1 DR |
4663 | if (err) |
4664 | return err; | |
4665 | ||
c0bdb232 | 4666 | set_min_partial(s, min); |
73d342b1 DR |
4667 | return length; |
4668 | } | |
4669 | SLAB_ATTR(min_partial); | |
4670 | ||
49e22585 CL |
4671 | static ssize_t cpu_partial_show(struct kmem_cache *s, char *buf) |
4672 | { | |
4673 | return sprintf(buf, "%u\n", s->cpu_partial); | |
4674 | } | |
4675 | ||
4676 | static ssize_t cpu_partial_store(struct kmem_cache *s, const char *buf, | |
4677 | size_t length) | |
4678 | { | |
4679 | unsigned long objects; | |
4680 | int err; | |
4681 | ||
3dbb95f7 | 4682 | err = kstrtoul(buf, 10, &objects); |
49e22585 CL |
4683 | if (err) |
4684 | return err; | |
345c905d | 4685 | if (objects && !kmem_cache_has_cpu_partial(s)) |
74ee4ef1 | 4686 | return -EINVAL; |
49e22585 CL |
4687 | |
4688 | s->cpu_partial = objects; | |
4689 | flush_all(s); | |
4690 | return length; | |
4691 | } | |
4692 | SLAB_ATTR(cpu_partial); | |
4693 | ||
81819f0f CL |
4694 | static ssize_t ctor_show(struct kmem_cache *s, char *buf) |
4695 | { | |
62c70bce JP |
4696 | if (!s->ctor) |
4697 | return 0; | |
4698 | return sprintf(buf, "%pS\n", s->ctor); | |
81819f0f CL |
4699 | } |
4700 | SLAB_ATTR_RO(ctor); | |
4701 | ||
81819f0f CL |
4702 | static ssize_t aliases_show(struct kmem_cache *s, char *buf) |
4703 | { | |
4307c14f | 4704 | return sprintf(buf, "%d\n", s->refcount < 0 ? 0 : s->refcount - 1); |
81819f0f CL |
4705 | } |
4706 | SLAB_ATTR_RO(aliases); | |
4707 | ||
81819f0f CL |
4708 | static ssize_t partial_show(struct kmem_cache *s, char *buf) |
4709 | { | |
d9acf4b7 | 4710 | return show_slab_objects(s, buf, SO_PARTIAL); |
81819f0f CL |
4711 | } |
4712 | SLAB_ATTR_RO(partial); | |
4713 | ||
4714 | static ssize_t cpu_slabs_show(struct kmem_cache *s, char *buf) | |
4715 | { | |
d9acf4b7 | 4716 | return show_slab_objects(s, buf, SO_CPU); |
81819f0f CL |
4717 | } |
4718 | SLAB_ATTR_RO(cpu_slabs); | |
4719 | ||
4720 | static ssize_t objects_show(struct kmem_cache *s, char *buf) | |
4721 | { | |
205ab99d | 4722 | return show_slab_objects(s, buf, SO_ALL|SO_OBJECTS); |
81819f0f CL |
4723 | } |
4724 | SLAB_ATTR_RO(objects); | |
4725 | ||
205ab99d CL |
4726 | static ssize_t objects_partial_show(struct kmem_cache *s, char *buf) |
4727 | { | |
4728 | return show_slab_objects(s, buf, SO_PARTIAL|SO_OBJECTS); | |
4729 | } | |
4730 | SLAB_ATTR_RO(objects_partial); | |
4731 | ||
49e22585 CL |
4732 | static ssize_t slabs_cpu_partial_show(struct kmem_cache *s, char *buf) |
4733 | { | |
4734 | int objects = 0; | |
4735 | int pages = 0; | |
4736 | int cpu; | |
4737 | int len; | |
4738 | ||
4739 | for_each_online_cpu(cpu) { | |
4740 | struct page *page = per_cpu_ptr(s->cpu_slab, cpu)->partial; | |
4741 | ||
4742 | if (page) { | |
4743 | pages += page->pages; | |
4744 | objects += page->pobjects; | |
4745 | } | |
4746 | } | |
4747 | ||
4748 | len = sprintf(buf, "%d(%d)", objects, pages); | |
4749 | ||
4750 | #ifdef CONFIG_SMP | |
4751 | for_each_online_cpu(cpu) { | |
4752 | struct page *page = per_cpu_ptr(s->cpu_slab, cpu) ->partial; | |
4753 | ||
4754 | if (page && len < PAGE_SIZE - 20) | |
4755 | len += sprintf(buf + len, " C%d=%d(%d)", cpu, | |
4756 | page->pobjects, page->pages); | |
4757 | } | |
4758 | #endif | |
4759 | return len + sprintf(buf + len, "\n"); | |
4760 | } | |
4761 | SLAB_ATTR_RO(slabs_cpu_partial); | |
4762 | ||
a5a84755 CL |
4763 | static ssize_t reclaim_account_show(struct kmem_cache *s, char *buf) |
4764 | { | |
4765 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_RECLAIM_ACCOUNT)); | |
4766 | } | |
4767 | ||
4768 | static ssize_t reclaim_account_store(struct kmem_cache *s, | |
4769 | const char *buf, size_t length) | |
4770 | { | |
4771 | s->flags &= ~SLAB_RECLAIM_ACCOUNT; | |
4772 | if (buf[0] == '1') | |
4773 | s->flags |= SLAB_RECLAIM_ACCOUNT; | |
4774 | return length; | |
4775 | } | |
4776 | SLAB_ATTR(reclaim_account); | |
4777 | ||
4778 | static ssize_t hwcache_align_show(struct kmem_cache *s, char *buf) | |
4779 | { | |
4780 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_HWCACHE_ALIGN)); | |
4781 | } | |
4782 | SLAB_ATTR_RO(hwcache_align); | |
4783 | ||
4784 | #ifdef CONFIG_ZONE_DMA | |
4785 | static ssize_t cache_dma_show(struct kmem_cache *s, char *buf) | |
4786 | { | |
4787 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_CACHE_DMA)); | |
4788 | } | |
4789 | SLAB_ATTR_RO(cache_dma); | |
4790 | #endif | |
4791 | ||
4792 | static ssize_t destroy_by_rcu_show(struct kmem_cache *s, char *buf) | |
4793 | { | |
4794 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_DESTROY_BY_RCU)); | |
4795 | } | |
4796 | SLAB_ATTR_RO(destroy_by_rcu); | |
4797 | ||
ab9a0f19 LJ |
4798 | static ssize_t reserved_show(struct kmem_cache *s, char *buf) |
4799 | { | |
4800 | return sprintf(buf, "%d\n", s->reserved); | |
4801 | } | |
4802 | SLAB_ATTR_RO(reserved); | |
4803 | ||
ab4d5ed5 | 4804 | #ifdef CONFIG_SLUB_DEBUG |
a5a84755 CL |
4805 | static ssize_t slabs_show(struct kmem_cache *s, char *buf) |
4806 | { | |
4807 | return show_slab_objects(s, buf, SO_ALL); | |
4808 | } | |
4809 | SLAB_ATTR_RO(slabs); | |
4810 | ||
205ab99d CL |
4811 | static ssize_t total_objects_show(struct kmem_cache *s, char *buf) |
4812 | { | |
4813 | return show_slab_objects(s, buf, SO_ALL|SO_TOTAL); | |
4814 | } | |
4815 | SLAB_ATTR_RO(total_objects); | |
4816 | ||
81819f0f CL |
4817 | static ssize_t sanity_checks_show(struct kmem_cache *s, char *buf) |
4818 | { | |
4819 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_DEBUG_FREE)); | |
4820 | } | |
4821 | ||
4822 | static ssize_t sanity_checks_store(struct kmem_cache *s, | |
4823 | const char *buf, size_t length) | |
4824 | { | |
4825 | s->flags &= ~SLAB_DEBUG_FREE; | |
b789ef51 CL |
4826 | if (buf[0] == '1') { |
4827 | s->flags &= ~__CMPXCHG_DOUBLE; | |
81819f0f | 4828 | s->flags |= SLAB_DEBUG_FREE; |
b789ef51 | 4829 | } |
81819f0f CL |
4830 | return length; |
4831 | } | |
4832 | SLAB_ATTR(sanity_checks); | |
4833 | ||
4834 | static ssize_t trace_show(struct kmem_cache *s, char *buf) | |
4835 | { | |
4836 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_TRACE)); | |
4837 | } | |
4838 | ||
4839 | static ssize_t trace_store(struct kmem_cache *s, const char *buf, | |
4840 | size_t length) | |
4841 | { | |
c9e16131 CL |
4842 | /* |
4843 | * Tracing a merged cache is going to give confusing results | |
4844 | * as well as cause other issues like converting a mergeable | |
4845 | * cache into an umergeable one. | |
4846 | */ | |
4847 | if (s->refcount > 1) | |
4848 | return -EINVAL; | |
4849 | ||
81819f0f | 4850 | s->flags &= ~SLAB_TRACE; |
b789ef51 CL |
4851 | if (buf[0] == '1') { |
4852 | s->flags &= ~__CMPXCHG_DOUBLE; | |
81819f0f | 4853 | s->flags |= SLAB_TRACE; |
b789ef51 | 4854 | } |
81819f0f CL |
4855 | return length; |
4856 | } | |
4857 | SLAB_ATTR(trace); | |
4858 | ||
81819f0f CL |
4859 | static ssize_t red_zone_show(struct kmem_cache *s, char *buf) |
4860 | { | |
4861 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_RED_ZONE)); | |
4862 | } | |
4863 | ||
4864 | static ssize_t red_zone_store(struct kmem_cache *s, | |
4865 | const char *buf, size_t length) | |
4866 | { | |
4867 | if (any_slab_objects(s)) | |
4868 | return -EBUSY; | |
4869 | ||
4870 | s->flags &= ~SLAB_RED_ZONE; | |
b789ef51 CL |
4871 | if (buf[0] == '1') { |
4872 | s->flags &= ~__CMPXCHG_DOUBLE; | |
81819f0f | 4873 | s->flags |= SLAB_RED_ZONE; |
b789ef51 | 4874 | } |
06b285dc | 4875 | calculate_sizes(s, -1); |
81819f0f CL |
4876 | return length; |
4877 | } | |
4878 | SLAB_ATTR(red_zone); | |
4879 | ||
4880 | static ssize_t poison_show(struct kmem_cache *s, char *buf) | |
4881 | { | |
4882 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_POISON)); | |
4883 | } | |
4884 | ||
4885 | static ssize_t poison_store(struct kmem_cache *s, | |
4886 | const char *buf, size_t length) | |
4887 | { | |
4888 | if (any_slab_objects(s)) | |
4889 | return -EBUSY; | |
4890 | ||
4891 | s->flags &= ~SLAB_POISON; | |
b789ef51 CL |
4892 | if (buf[0] == '1') { |
4893 | s->flags &= ~__CMPXCHG_DOUBLE; | |
81819f0f | 4894 | s->flags |= SLAB_POISON; |
b789ef51 | 4895 | } |
06b285dc | 4896 | calculate_sizes(s, -1); |
81819f0f CL |
4897 | return length; |
4898 | } | |
4899 | SLAB_ATTR(poison); | |
4900 | ||
4901 | static ssize_t store_user_show(struct kmem_cache *s, char *buf) | |
4902 | { | |
4903 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_STORE_USER)); | |
4904 | } | |
4905 | ||
4906 | static ssize_t store_user_store(struct kmem_cache *s, | |
4907 | const char *buf, size_t length) | |
4908 | { | |
4909 | if (any_slab_objects(s)) | |
4910 | return -EBUSY; | |
4911 | ||
4912 | s->flags &= ~SLAB_STORE_USER; | |
b789ef51 CL |
4913 | if (buf[0] == '1') { |
4914 | s->flags &= ~__CMPXCHG_DOUBLE; | |
81819f0f | 4915 | s->flags |= SLAB_STORE_USER; |
b789ef51 | 4916 | } |
06b285dc | 4917 | calculate_sizes(s, -1); |
81819f0f CL |
4918 | return length; |
4919 | } | |
4920 | SLAB_ATTR(store_user); | |
4921 | ||
53e15af0 CL |
4922 | static ssize_t validate_show(struct kmem_cache *s, char *buf) |
4923 | { | |
4924 | return 0; | |
4925 | } | |
4926 | ||
4927 | static ssize_t validate_store(struct kmem_cache *s, | |
4928 | const char *buf, size_t length) | |
4929 | { | |
434e245d CL |
4930 | int ret = -EINVAL; |
4931 | ||
4932 | if (buf[0] == '1') { | |
4933 | ret = validate_slab_cache(s); | |
4934 | if (ret >= 0) | |
4935 | ret = length; | |
4936 | } | |
4937 | return ret; | |
53e15af0 CL |
4938 | } |
4939 | SLAB_ATTR(validate); | |
a5a84755 CL |
4940 | |
4941 | static ssize_t alloc_calls_show(struct kmem_cache *s, char *buf) | |
4942 | { | |
4943 | if (!(s->flags & SLAB_STORE_USER)) | |
4944 | return -ENOSYS; | |
4945 | return list_locations(s, buf, TRACK_ALLOC); | |
4946 | } | |
4947 | SLAB_ATTR_RO(alloc_calls); | |
4948 | ||
4949 | static ssize_t free_calls_show(struct kmem_cache *s, char *buf) | |
4950 | { | |
4951 | if (!(s->flags & SLAB_STORE_USER)) | |
4952 | return -ENOSYS; | |
4953 | return list_locations(s, buf, TRACK_FREE); | |
4954 | } | |
4955 | SLAB_ATTR_RO(free_calls); | |
4956 | #endif /* CONFIG_SLUB_DEBUG */ | |
4957 | ||
4958 | #ifdef CONFIG_FAILSLAB | |
4959 | static ssize_t failslab_show(struct kmem_cache *s, char *buf) | |
4960 | { | |
4961 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_FAILSLAB)); | |
4962 | } | |
4963 | ||
4964 | static ssize_t failslab_store(struct kmem_cache *s, const char *buf, | |
4965 | size_t length) | |
4966 | { | |
c9e16131 CL |
4967 | if (s->refcount > 1) |
4968 | return -EINVAL; | |
4969 | ||
a5a84755 CL |
4970 | s->flags &= ~SLAB_FAILSLAB; |
4971 | if (buf[0] == '1') | |
4972 | s->flags |= SLAB_FAILSLAB; | |
4973 | return length; | |
4974 | } | |
4975 | SLAB_ATTR(failslab); | |
ab4d5ed5 | 4976 | #endif |
53e15af0 | 4977 | |
2086d26a CL |
4978 | static ssize_t shrink_show(struct kmem_cache *s, char *buf) |
4979 | { | |
4980 | return 0; | |
4981 | } | |
4982 | ||
4983 | static ssize_t shrink_store(struct kmem_cache *s, | |
4984 | const char *buf, size_t length) | |
4985 | { | |
832f37f5 VD |
4986 | if (buf[0] == '1') |
4987 | kmem_cache_shrink(s); | |
4988 | else | |
2086d26a CL |
4989 | return -EINVAL; |
4990 | return length; | |
4991 | } | |
4992 | SLAB_ATTR(shrink); | |
4993 | ||
81819f0f | 4994 | #ifdef CONFIG_NUMA |
9824601e | 4995 | static ssize_t remote_node_defrag_ratio_show(struct kmem_cache *s, char *buf) |
81819f0f | 4996 | { |
9824601e | 4997 | return sprintf(buf, "%d\n", s->remote_node_defrag_ratio / 10); |
81819f0f CL |
4998 | } |
4999 | ||
9824601e | 5000 | static ssize_t remote_node_defrag_ratio_store(struct kmem_cache *s, |
81819f0f CL |
5001 | const char *buf, size_t length) |
5002 | { | |
0121c619 CL |
5003 | unsigned long ratio; |
5004 | int err; | |
5005 | ||
3dbb95f7 | 5006 | err = kstrtoul(buf, 10, &ratio); |
0121c619 CL |
5007 | if (err) |
5008 | return err; | |
5009 | ||
e2cb96b7 | 5010 | if (ratio <= 100) |
0121c619 | 5011 | s->remote_node_defrag_ratio = ratio * 10; |
81819f0f | 5012 | |
81819f0f CL |
5013 | return length; |
5014 | } | |
9824601e | 5015 | SLAB_ATTR(remote_node_defrag_ratio); |
81819f0f CL |
5016 | #endif |
5017 | ||
8ff12cfc | 5018 | #ifdef CONFIG_SLUB_STATS |
8ff12cfc CL |
5019 | static int show_stat(struct kmem_cache *s, char *buf, enum stat_item si) |
5020 | { | |
5021 | unsigned long sum = 0; | |
5022 | int cpu; | |
5023 | int len; | |
5024 | int *data = kmalloc(nr_cpu_ids * sizeof(int), GFP_KERNEL); | |
5025 | ||
5026 | if (!data) | |
5027 | return -ENOMEM; | |
5028 | ||
5029 | for_each_online_cpu(cpu) { | |
9dfc6e68 | 5030 | unsigned x = per_cpu_ptr(s->cpu_slab, cpu)->stat[si]; |
8ff12cfc CL |
5031 | |
5032 | data[cpu] = x; | |
5033 | sum += x; | |
5034 | } | |
5035 | ||
5036 | len = sprintf(buf, "%lu", sum); | |
5037 | ||
50ef37b9 | 5038 | #ifdef CONFIG_SMP |
8ff12cfc CL |
5039 | for_each_online_cpu(cpu) { |
5040 | if (data[cpu] && len < PAGE_SIZE - 20) | |
50ef37b9 | 5041 | len += sprintf(buf + len, " C%d=%u", cpu, data[cpu]); |
8ff12cfc | 5042 | } |
50ef37b9 | 5043 | #endif |
8ff12cfc CL |
5044 | kfree(data); |
5045 | return len + sprintf(buf + len, "\n"); | |
5046 | } | |
5047 | ||
78eb00cc DR |
5048 | static void clear_stat(struct kmem_cache *s, enum stat_item si) |
5049 | { | |
5050 | int cpu; | |
5051 | ||
5052 | for_each_online_cpu(cpu) | |
9dfc6e68 | 5053 | per_cpu_ptr(s->cpu_slab, cpu)->stat[si] = 0; |
78eb00cc DR |
5054 | } |
5055 | ||
8ff12cfc CL |
5056 | #define STAT_ATTR(si, text) \ |
5057 | static ssize_t text##_show(struct kmem_cache *s, char *buf) \ | |
5058 | { \ | |
5059 | return show_stat(s, buf, si); \ | |
5060 | } \ | |
78eb00cc DR |
5061 | static ssize_t text##_store(struct kmem_cache *s, \ |
5062 | const char *buf, size_t length) \ | |
5063 | { \ | |
5064 | if (buf[0] != '0') \ | |
5065 | return -EINVAL; \ | |
5066 | clear_stat(s, si); \ | |
5067 | return length; \ | |
5068 | } \ | |
5069 | SLAB_ATTR(text); \ | |
8ff12cfc CL |
5070 | |
5071 | STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath); | |
5072 | STAT_ATTR(ALLOC_SLOWPATH, alloc_slowpath); | |
5073 | STAT_ATTR(FREE_FASTPATH, free_fastpath); | |
5074 | STAT_ATTR(FREE_SLOWPATH, free_slowpath); | |
5075 | STAT_ATTR(FREE_FROZEN, free_frozen); | |
5076 | STAT_ATTR(FREE_ADD_PARTIAL, free_add_partial); | |
5077 | STAT_ATTR(FREE_REMOVE_PARTIAL, free_remove_partial); | |
5078 | STAT_ATTR(ALLOC_FROM_PARTIAL, alloc_from_partial); | |
5079 | STAT_ATTR(ALLOC_SLAB, alloc_slab); | |
5080 | STAT_ATTR(ALLOC_REFILL, alloc_refill); | |
e36a2652 | 5081 | STAT_ATTR(ALLOC_NODE_MISMATCH, alloc_node_mismatch); |
8ff12cfc CL |
5082 | STAT_ATTR(FREE_SLAB, free_slab); |
5083 | STAT_ATTR(CPUSLAB_FLUSH, cpuslab_flush); | |
5084 | STAT_ATTR(DEACTIVATE_FULL, deactivate_full); | |
5085 | STAT_ATTR(DEACTIVATE_EMPTY, deactivate_empty); | |
5086 | STAT_ATTR(DEACTIVATE_TO_HEAD, deactivate_to_head); | |
5087 | STAT_ATTR(DEACTIVATE_TO_TAIL, deactivate_to_tail); | |
5088 | STAT_ATTR(DEACTIVATE_REMOTE_FREES, deactivate_remote_frees); | |
03e404af | 5089 | STAT_ATTR(DEACTIVATE_BYPASS, deactivate_bypass); |
65c3376a | 5090 | STAT_ATTR(ORDER_FALLBACK, order_fallback); |
b789ef51 CL |
5091 | STAT_ATTR(CMPXCHG_DOUBLE_CPU_FAIL, cmpxchg_double_cpu_fail); |
5092 | STAT_ATTR(CMPXCHG_DOUBLE_FAIL, cmpxchg_double_fail); | |
49e22585 CL |
5093 | STAT_ATTR(CPU_PARTIAL_ALLOC, cpu_partial_alloc); |
5094 | STAT_ATTR(CPU_PARTIAL_FREE, cpu_partial_free); | |
8028dcea AS |
5095 | STAT_ATTR(CPU_PARTIAL_NODE, cpu_partial_node); |
5096 | STAT_ATTR(CPU_PARTIAL_DRAIN, cpu_partial_drain); | |
8ff12cfc CL |
5097 | #endif |
5098 | ||
06428780 | 5099 | static struct attribute *slab_attrs[] = { |
81819f0f CL |
5100 | &slab_size_attr.attr, |
5101 | &object_size_attr.attr, | |
5102 | &objs_per_slab_attr.attr, | |
5103 | &order_attr.attr, | |
73d342b1 | 5104 | &min_partial_attr.attr, |
49e22585 | 5105 | &cpu_partial_attr.attr, |
81819f0f | 5106 | &objects_attr.attr, |
205ab99d | 5107 | &objects_partial_attr.attr, |
81819f0f CL |
5108 | &partial_attr.attr, |
5109 | &cpu_slabs_attr.attr, | |
5110 | &ctor_attr.attr, | |
81819f0f CL |
5111 | &aliases_attr.attr, |
5112 | &align_attr.attr, | |
81819f0f CL |
5113 | &hwcache_align_attr.attr, |
5114 | &reclaim_account_attr.attr, | |
5115 | &destroy_by_rcu_attr.attr, | |
a5a84755 | 5116 | &shrink_attr.attr, |
ab9a0f19 | 5117 | &reserved_attr.attr, |
49e22585 | 5118 | &slabs_cpu_partial_attr.attr, |
ab4d5ed5 | 5119 | #ifdef CONFIG_SLUB_DEBUG |
a5a84755 CL |
5120 | &total_objects_attr.attr, |
5121 | &slabs_attr.attr, | |
5122 | &sanity_checks_attr.attr, | |
5123 | &trace_attr.attr, | |
81819f0f CL |
5124 | &red_zone_attr.attr, |
5125 | &poison_attr.attr, | |
5126 | &store_user_attr.attr, | |
53e15af0 | 5127 | &validate_attr.attr, |
88a420e4 CL |
5128 | &alloc_calls_attr.attr, |
5129 | &free_calls_attr.attr, | |
ab4d5ed5 | 5130 | #endif |
81819f0f CL |
5131 | #ifdef CONFIG_ZONE_DMA |
5132 | &cache_dma_attr.attr, | |
5133 | #endif | |
5134 | #ifdef CONFIG_NUMA | |
9824601e | 5135 | &remote_node_defrag_ratio_attr.attr, |
8ff12cfc CL |
5136 | #endif |
5137 | #ifdef CONFIG_SLUB_STATS | |
5138 | &alloc_fastpath_attr.attr, | |
5139 | &alloc_slowpath_attr.attr, | |
5140 | &free_fastpath_attr.attr, | |
5141 | &free_slowpath_attr.attr, | |
5142 | &free_frozen_attr.attr, | |
5143 | &free_add_partial_attr.attr, | |
5144 | &free_remove_partial_attr.attr, | |
5145 | &alloc_from_partial_attr.attr, | |
5146 | &alloc_slab_attr.attr, | |
5147 | &alloc_refill_attr.attr, | |
e36a2652 | 5148 | &alloc_node_mismatch_attr.attr, |
8ff12cfc CL |
5149 | &free_slab_attr.attr, |
5150 | &cpuslab_flush_attr.attr, | |
5151 | &deactivate_full_attr.attr, | |
5152 | &deactivate_empty_attr.attr, | |
5153 | &deactivate_to_head_attr.attr, | |
5154 | &deactivate_to_tail_attr.attr, | |
5155 | &deactivate_remote_frees_attr.attr, | |
03e404af | 5156 | &deactivate_bypass_attr.attr, |
65c3376a | 5157 | &order_fallback_attr.attr, |
b789ef51 CL |
5158 | &cmpxchg_double_fail_attr.attr, |
5159 | &cmpxchg_double_cpu_fail_attr.attr, | |
49e22585 CL |
5160 | &cpu_partial_alloc_attr.attr, |
5161 | &cpu_partial_free_attr.attr, | |
8028dcea AS |
5162 | &cpu_partial_node_attr.attr, |
5163 | &cpu_partial_drain_attr.attr, | |
81819f0f | 5164 | #endif |
4c13dd3b DM |
5165 | #ifdef CONFIG_FAILSLAB |
5166 | &failslab_attr.attr, | |
5167 | #endif | |
5168 | ||
81819f0f CL |
5169 | NULL |
5170 | }; | |
5171 | ||
5172 | static struct attribute_group slab_attr_group = { | |
5173 | .attrs = slab_attrs, | |
5174 | }; | |
5175 | ||
5176 | static ssize_t slab_attr_show(struct kobject *kobj, | |
5177 | struct attribute *attr, | |
5178 | char *buf) | |
5179 | { | |
5180 | struct slab_attribute *attribute; | |
5181 | struct kmem_cache *s; | |
5182 | int err; | |
5183 | ||
5184 | attribute = to_slab_attr(attr); | |
5185 | s = to_slab(kobj); | |
5186 | ||
5187 | if (!attribute->show) | |
5188 | return -EIO; | |
5189 | ||
5190 | err = attribute->show(s, buf); | |
5191 | ||
5192 | return err; | |
5193 | } | |
5194 | ||
5195 | static ssize_t slab_attr_store(struct kobject *kobj, | |
5196 | struct attribute *attr, | |
5197 | const char *buf, size_t len) | |
5198 | { | |
5199 | struct slab_attribute *attribute; | |
5200 | struct kmem_cache *s; | |
5201 | int err; | |
5202 | ||
5203 | attribute = to_slab_attr(attr); | |
5204 | s = to_slab(kobj); | |
5205 | ||
5206 | if (!attribute->store) | |
5207 | return -EIO; | |
5208 | ||
5209 | err = attribute->store(s, buf, len); | |
127424c8 | 5210 | #ifdef CONFIG_MEMCG |
107dab5c | 5211 | if (slab_state >= FULL && err >= 0 && is_root_cache(s)) { |
426589f5 | 5212 | struct kmem_cache *c; |
81819f0f | 5213 | |
107dab5c GC |
5214 | mutex_lock(&slab_mutex); |
5215 | if (s->max_attr_size < len) | |
5216 | s->max_attr_size = len; | |
5217 | ||
ebe945c2 GC |
5218 | /* |
5219 | * This is a best effort propagation, so this function's return | |
5220 | * value will be determined by the parent cache only. This is | |
5221 | * basically because not all attributes will have a well | |
5222 | * defined semantics for rollbacks - most of the actions will | |
5223 | * have permanent effects. | |
5224 | * | |
5225 | * Returning the error value of any of the children that fail | |
5226 | * is not 100 % defined, in the sense that users seeing the | |
5227 | * error code won't be able to know anything about the state of | |
5228 | * the cache. | |
5229 | * | |
5230 | * Only returning the error code for the parent cache at least | |
5231 | * has well defined semantics. The cache being written to | |
5232 | * directly either failed or succeeded, in which case we loop | |
5233 | * through the descendants with best-effort propagation. | |
5234 | */ | |
426589f5 VD |
5235 | for_each_memcg_cache(c, s) |
5236 | attribute->store(c, buf, len); | |
107dab5c GC |
5237 | mutex_unlock(&slab_mutex); |
5238 | } | |
5239 | #endif | |
81819f0f CL |
5240 | return err; |
5241 | } | |
5242 | ||
107dab5c GC |
5243 | static void memcg_propagate_slab_attrs(struct kmem_cache *s) |
5244 | { | |
127424c8 | 5245 | #ifdef CONFIG_MEMCG |
107dab5c GC |
5246 | int i; |
5247 | char *buffer = NULL; | |
93030d83 | 5248 | struct kmem_cache *root_cache; |
107dab5c | 5249 | |
93030d83 | 5250 | if (is_root_cache(s)) |
107dab5c GC |
5251 | return; |
5252 | ||
f7ce3190 | 5253 | root_cache = s->memcg_params.root_cache; |
93030d83 | 5254 | |
107dab5c GC |
5255 | /* |
5256 | * This mean this cache had no attribute written. Therefore, no point | |
5257 | * in copying default values around | |
5258 | */ | |
93030d83 | 5259 | if (!root_cache->max_attr_size) |
107dab5c GC |
5260 | return; |
5261 | ||
5262 | for (i = 0; i < ARRAY_SIZE(slab_attrs); i++) { | |
5263 | char mbuf[64]; | |
5264 | char *buf; | |
5265 | struct slab_attribute *attr = to_slab_attr(slab_attrs[i]); | |
5266 | ||
5267 | if (!attr || !attr->store || !attr->show) | |
5268 | continue; | |
5269 | ||
5270 | /* | |
5271 | * It is really bad that we have to allocate here, so we will | |
5272 | * do it only as a fallback. If we actually allocate, though, | |
5273 | * we can just use the allocated buffer until the end. | |
5274 | * | |
5275 | * Most of the slub attributes will tend to be very small in | |
5276 | * size, but sysfs allows buffers up to a page, so they can | |
5277 | * theoretically happen. | |
5278 | */ | |
5279 | if (buffer) | |
5280 | buf = buffer; | |
93030d83 | 5281 | else if (root_cache->max_attr_size < ARRAY_SIZE(mbuf)) |
107dab5c GC |
5282 | buf = mbuf; |
5283 | else { | |
5284 | buffer = (char *) get_zeroed_page(GFP_KERNEL); | |
5285 | if (WARN_ON(!buffer)) | |
5286 | continue; | |
5287 | buf = buffer; | |
5288 | } | |
5289 | ||
93030d83 | 5290 | attr->show(root_cache, buf); |
107dab5c GC |
5291 | attr->store(s, buf, strlen(buf)); |
5292 | } | |
5293 | ||
5294 | if (buffer) | |
5295 | free_page((unsigned long)buffer); | |
5296 | #endif | |
5297 | } | |
5298 | ||
41a21285 CL |
5299 | static void kmem_cache_release(struct kobject *k) |
5300 | { | |
5301 | slab_kmem_cache_release(to_slab(k)); | |
5302 | } | |
5303 | ||
52cf25d0 | 5304 | static const struct sysfs_ops slab_sysfs_ops = { |
81819f0f CL |
5305 | .show = slab_attr_show, |
5306 | .store = slab_attr_store, | |
5307 | }; | |
5308 | ||
5309 | static struct kobj_type slab_ktype = { | |
5310 | .sysfs_ops = &slab_sysfs_ops, | |
41a21285 | 5311 | .release = kmem_cache_release, |
81819f0f CL |
5312 | }; |
5313 | ||
5314 | static int uevent_filter(struct kset *kset, struct kobject *kobj) | |
5315 | { | |
5316 | struct kobj_type *ktype = get_ktype(kobj); | |
5317 | ||
5318 | if (ktype == &slab_ktype) | |
5319 | return 1; | |
5320 | return 0; | |
5321 | } | |
5322 | ||
9cd43611 | 5323 | static const struct kset_uevent_ops slab_uevent_ops = { |
81819f0f CL |
5324 | .filter = uevent_filter, |
5325 | }; | |
5326 | ||
27c3a314 | 5327 | static struct kset *slab_kset; |
81819f0f | 5328 | |
9a41707b VD |
5329 | static inline struct kset *cache_kset(struct kmem_cache *s) |
5330 | { | |
127424c8 | 5331 | #ifdef CONFIG_MEMCG |
9a41707b | 5332 | if (!is_root_cache(s)) |
f7ce3190 | 5333 | return s->memcg_params.root_cache->memcg_kset; |
9a41707b VD |
5334 | #endif |
5335 | return slab_kset; | |
5336 | } | |
5337 | ||
81819f0f CL |
5338 | #define ID_STR_LENGTH 64 |
5339 | ||
5340 | /* Create a unique string id for a slab cache: | |
6446faa2 CL |
5341 | * |
5342 | * Format :[flags-]size | |
81819f0f CL |
5343 | */ |
5344 | static char *create_unique_id(struct kmem_cache *s) | |
5345 | { | |
5346 | char *name = kmalloc(ID_STR_LENGTH, GFP_KERNEL); | |
5347 | char *p = name; | |
5348 | ||
5349 | BUG_ON(!name); | |
5350 | ||
5351 | *p++ = ':'; | |
5352 | /* | |
5353 | * First flags affecting slabcache operations. We will only | |
5354 | * get here for aliasable slabs so we do not need to support | |
5355 | * too many flags. The flags here must cover all flags that | |
5356 | * are matched during merging to guarantee that the id is | |
5357 | * unique. | |
5358 | */ | |
5359 | if (s->flags & SLAB_CACHE_DMA) | |
5360 | *p++ = 'd'; | |
5361 | if (s->flags & SLAB_RECLAIM_ACCOUNT) | |
5362 | *p++ = 'a'; | |
5363 | if (s->flags & SLAB_DEBUG_FREE) | |
5364 | *p++ = 'F'; | |
5a896d9e VN |
5365 | if (!(s->flags & SLAB_NOTRACK)) |
5366 | *p++ = 't'; | |
230e9fc2 VD |
5367 | if (s->flags & SLAB_ACCOUNT) |
5368 | *p++ = 'A'; | |
81819f0f CL |
5369 | if (p != name + 1) |
5370 | *p++ = '-'; | |
5371 | p += sprintf(p, "%07d", s->size); | |
2633d7a0 | 5372 | |
81819f0f CL |
5373 | BUG_ON(p > name + ID_STR_LENGTH - 1); |
5374 | return name; | |
5375 | } | |
5376 | ||
5377 | static int sysfs_slab_add(struct kmem_cache *s) | |
5378 | { | |
5379 | int err; | |
5380 | const char *name; | |
45530c44 | 5381 | int unmergeable = slab_unmergeable(s); |
81819f0f | 5382 | |
81819f0f CL |
5383 | if (unmergeable) { |
5384 | /* | |
5385 | * Slabcache can never be merged so we can use the name proper. | |
5386 | * This is typically the case for debug situations. In that | |
5387 | * case we can catch duplicate names easily. | |
5388 | */ | |
27c3a314 | 5389 | sysfs_remove_link(&slab_kset->kobj, s->name); |
81819f0f CL |
5390 | name = s->name; |
5391 | } else { | |
5392 | /* | |
5393 | * Create a unique name for the slab as a target | |
5394 | * for the symlinks. | |
5395 | */ | |
5396 | name = create_unique_id(s); | |
5397 | } | |
5398 | ||
9a41707b | 5399 | s->kobj.kset = cache_kset(s); |
26e4f205 | 5400 | err = kobject_init_and_add(&s->kobj, &slab_ktype, NULL, "%s", name); |
54b6a731 | 5401 | if (err) |
80da026a | 5402 | goto out; |
81819f0f CL |
5403 | |
5404 | err = sysfs_create_group(&s->kobj, &slab_attr_group); | |
54b6a731 DJ |
5405 | if (err) |
5406 | goto out_del_kobj; | |
9a41707b | 5407 | |
127424c8 | 5408 | #ifdef CONFIG_MEMCG |
9a41707b VD |
5409 | if (is_root_cache(s)) { |
5410 | s->memcg_kset = kset_create_and_add("cgroup", NULL, &s->kobj); | |
5411 | if (!s->memcg_kset) { | |
54b6a731 DJ |
5412 | err = -ENOMEM; |
5413 | goto out_del_kobj; | |
9a41707b VD |
5414 | } |
5415 | } | |
5416 | #endif | |
5417 | ||
81819f0f CL |
5418 | kobject_uevent(&s->kobj, KOBJ_ADD); |
5419 | if (!unmergeable) { | |
5420 | /* Setup first alias */ | |
5421 | sysfs_slab_alias(s, s->name); | |
81819f0f | 5422 | } |
54b6a731 DJ |
5423 | out: |
5424 | if (!unmergeable) | |
5425 | kfree(name); | |
5426 | return err; | |
5427 | out_del_kobj: | |
5428 | kobject_del(&s->kobj); | |
54b6a731 | 5429 | goto out; |
81819f0f CL |
5430 | } |
5431 | ||
41a21285 | 5432 | void sysfs_slab_remove(struct kmem_cache *s) |
81819f0f | 5433 | { |
97d06609 | 5434 | if (slab_state < FULL) |
2bce6485 CL |
5435 | /* |
5436 | * Sysfs has not been setup yet so no need to remove the | |
5437 | * cache from sysfs. | |
5438 | */ | |
5439 | return; | |
5440 | ||
127424c8 | 5441 | #ifdef CONFIG_MEMCG |
9a41707b VD |
5442 | kset_unregister(s->memcg_kset); |
5443 | #endif | |
81819f0f CL |
5444 | kobject_uevent(&s->kobj, KOBJ_REMOVE); |
5445 | kobject_del(&s->kobj); | |
151c602f | 5446 | kobject_put(&s->kobj); |
81819f0f CL |
5447 | } |
5448 | ||
5449 | /* | |
5450 | * Need to buffer aliases during bootup until sysfs becomes | |
9f6c708e | 5451 | * available lest we lose that information. |
81819f0f CL |
5452 | */ |
5453 | struct saved_alias { | |
5454 | struct kmem_cache *s; | |
5455 | const char *name; | |
5456 | struct saved_alias *next; | |
5457 | }; | |
5458 | ||
5af328a5 | 5459 | static struct saved_alias *alias_list; |
81819f0f CL |
5460 | |
5461 | static int sysfs_slab_alias(struct kmem_cache *s, const char *name) | |
5462 | { | |
5463 | struct saved_alias *al; | |
5464 | ||
97d06609 | 5465 | if (slab_state == FULL) { |
81819f0f CL |
5466 | /* |
5467 | * If we have a leftover link then remove it. | |
5468 | */ | |
27c3a314 GKH |
5469 | sysfs_remove_link(&slab_kset->kobj, name); |
5470 | return sysfs_create_link(&slab_kset->kobj, &s->kobj, name); | |
81819f0f CL |
5471 | } |
5472 | ||
5473 | al = kmalloc(sizeof(struct saved_alias), GFP_KERNEL); | |
5474 | if (!al) | |
5475 | return -ENOMEM; | |
5476 | ||
5477 | al->s = s; | |
5478 | al->name = name; | |
5479 | al->next = alias_list; | |
5480 | alias_list = al; | |
5481 | return 0; | |
5482 | } | |
5483 | ||
5484 | static int __init slab_sysfs_init(void) | |
5485 | { | |
5b95a4ac | 5486 | struct kmem_cache *s; |
81819f0f CL |
5487 | int err; |
5488 | ||
18004c5d | 5489 | mutex_lock(&slab_mutex); |
2bce6485 | 5490 | |
0ff21e46 | 5491 | slab_kset = kset_create_and_add("slab", &slab_uevent_ops, kernel_kobj); |
27c3a314 | 5492 | if (!slab_kset) { |
18004c5d | 5493 | mutex_unlock(&slab_mutex); |
f9f58285 | 5494 | pr_err("Cannot register slab subsystem.\n"); |
81819f0f CL |
5495 | return -ENOSYS; |
5496 | } | |
5497 | ||
97d06609 | 5498 | slab_state = FULL; |
26a7bd03 | 5499 | |
5b95a4ac | 5500 | list_for_each_entry(s, &slab_caches, list) { |
26a7bd03 | 5501 | err = sysfs_slab_add(s); |
5d540fb7 | 5502 | if (err) |
f9f58285 FF |
5503 | pr_err("SLUB: Unable to add boot slab %s to sysfs\n", |
5504 | s->name); | |
26a7bd03 | 5505 | } |
81819f0f CL |
5506 | |
5507 | while (alias_list) { | |
5508 | struct saved_alias *al = alias_list; | |
5509 | ||
5510 | alias_list = alias_list->next; | |
5511 | err = sysfs_slab_alias(al->s, al->name); | |
5d540fb7 | 5512 | if (err) |
f9f58285 FF |
5513 | pr_err("SLUB: Unable to add boot slab alias %s to sysfs\n", |
5514 | al->name); | |
81819f0f CL |
5515 | kfree(al); |
5516 | } | |
5517 | ||
18004c5d | 5518 | mutex_unlock(&slab_mutex); |
81819f0f CL |
5519 | resiliency_test(); |
5520 | return 0; | |
5521 | } | |
5522 | ||
5523 | __initcall(slab_sysfs_init); | |
ab4d5ed5 | 5524 | #endif /* CONFIG_SYSFS */ |
57ed3eda PE |
5525 | |
5526 | /* | |
5527 | * The /proc/slabinfo ABI | |
5528 | */ | |
158a9624 | 5529 | #ifdef CONFIG_SLABINFO |
0d7561c6 | 5530 | void get_slabinfo(struct kmem_cache *s, struct slabinfo *sinfo) |
57ed3eda | 5531 | { |
57ed3eda | 5532 | unsigned long nr_slabs = 0; |
205ab99d CL |
5533 | unsigned long nr_objs = 0; |
5534 | unsigned long nr_free = 0; | |
57ed3eda | 5535 | int node; |
fa45dc25 | 5536 | struct kmem_cache_node *n; |
57ed3eda | 5537 | |
fa45dc25 | 5538 | for_each_kmem_cache_node(s, node, n) { |
c17fd13e WL |
5539 | nr_slabs += node_nr_slabs(n); |
5540 | nr_objs += node_nr_objs(n); | |
205ab99d | 5541 | nr_free += count_partial(n, count_free); |
57ed3eda PE |
5542 | } |
5543 | ||
0d7561c6 GC |
5544 | sinfo->active_objs = nr_objs - nr_free; |
5545 | sinfo->num_objs = nr_objs; | |
5546 | sinfo->active_slabs = nr_slabs; | |
5547 | sinfo->num_slabs = nr_slabs; | |
5548 | sinfo->objects_per_slab = oo_objects(s->oo); | |
5549 | sinfo->cache_order = oo_order(s->oo); | |
57ed3eda PE |
5550 | } |
5551 | ||
0d7561c6 | 5552 | void slabinfo_show_stats(struct seq_file *m, struct kmem_cache *s) |
7b3c3a50 | 5553 | { |
7b3c3a50 AD |
5554 | } |
5555 | ||
b7454ad3 GC |
5556 | ssize_t slabinfo_write(struct file *file, const char __user *buffer, |
5557 | size_t count, loff_t *ppos) | |
7b3c3a50 | 5558 | { |
b7454ad3 | 5559 | return -EIO; |
7b3c3a50 | 5560 | } |
158a9624 | 5561 | #endif /* CONFIG_SLABINFO */ |