]>
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 | ||
ab9a0f19 LJ |
287 | static inline int order_objects(int order, unsigned long size, int reserved) |
288 | { | |
289 | return ((PAGE_SIZE << order) - reserved) / size; | |
290 | } | |
291 | ||
834f3d11 | 292 | static inline struct kmem_cache_order_objects oo_make(int order, |
ab9a0f19 | 293 | unsigned long size, int reserved) |
834f3d11 CL |
294 | { |
295 | struct kmem_cache_order_objects x = { | |
ab9a0f19 | 296 | (order << OO_SHIFT) + order_objects(order, size, reserved) |
834f3d11 CL |
297 | }; |
298 | ||
299 | return x; | |
300 | } | |
301 | ||
302 | static inline int oo_order(struct kmem_cache_order_objects x) | |
303 | { | |
210b5c06 | 304 | return x.x >> OO_SHIFT; |
834f3d11 CL |
305 | } |
306 | ||
307 | static inline int oo_objects(struct kmem_cache_order_objects x) | |
308 | { | |
210b5c06 | 309 | return x.x & OO_MASK; |
834f3d11 CL |
310 | } |
311 | ||
881db7fb CL |
312 | /* |
313 | * Per slab locking using the pagelock | |
314 | */ | |
315 | static __always_inline void slab_lock(struct page *page) | |
316 | { | |
48c935ad | 317 | VM_BUG_ON_PAGE(PageTail(page), page); |
881db7fb CL |
318 | bit_spin_lock(PG_locked, &page->flags); |
319 | } | |
320 | ||
321 | static __always_inline void slab_unlock(struct page *page) | |
322 | { | |
48c935ad | 323 | VM_BUG_ON_PAGE(PageTail(page), page); |
881db7fb CL |
324 | __bit_spin_unlock(PG_locked, &page->flags); |
325 | } | |
326 | ||
a0320865 DH |
327 | static inline void set_page_slub_counters(struct page *page, unsigned long counters_new) |
328 | { | |
329 | struct page tmp; | |
330 | tmp.counters = counters_new; | |
331 | /* | |
332 | * page->counters can cover frozen/inuse/objects as well | |
333 | * as page->_count. If we assign to ->counters directly | |
334 | * we run the risk of losing updates to page->_count, so | |
335 | * be careful and only assign to the fields we need. | |
336 | */ | |
337 | page->frozen = tmp.frozen; | |
338 | page->inuse = tmp.inuse; | |
339 | page->objects = tmp.objects; | |
340 | } | |
341 | ||
1d07171c CL |
342 | /* Interrupts must be disabled (for the fallback code to work right) */ |
343 | static inline bool __cmpxchg_double_slab(struct kmem_cache *s, struct page *page, | |
344 | void *freelist_old, unsigned long counters_old, | |
345 | void *freelist_new, unsigned long counters_new, | |
346 | const char *n) | |
347 | { | |
348 | VM_BUG_ON(!irqs_disabled()); | |
2565409f HC |
349 | #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \ |
350 | defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE) | |
1d07171c | 351 | if (s->flags & __CMPXCHG_DOUBLE) { |
cdcd6298 | 352 | if (cmpxchg_double(&page->freelist, &page->counters, |
0aa9a13d DC |
353 | freelist_old, counters_old, |
354 | freelist_new, counters_new)) | |
6f6528a1 | 355 | return true; |
1d07171c CL |
356 | } else |
357 | #endif | |
358 | { | |
359 | slab_lock(page); | |
d0e0ac97 CG |
360 | if (page->freelist == freelist_old && |
361 | page->counters == counters_old) { | |
1d07171c | 362 | page->freelist = freelist_new; |
a0320865 | 363 | set_page_slub_counters(page, counters_new); |
1d07171c | 364 | slab_unlock(page); |
6f6528a1 | 365 | return true; |
1d07171c CL |
366 | } |
367 | slab_unlock(page); | |
368 | } | |
369 | ||
370 | cpu_relax(); | |
371 | stat(s, CMPXCHG_DOUBLE_FAIL); | |
372 | ||
373 | #ifdef SLUB_DEBUG_CMPXCHG | |
f9f58285 | 374 | pr_info("%s %s: cmpxchg double redo ", n, s->name); |
1d07171c CL |
375 | #endif |
376 | ||
6f6528a1 | 377 | return false; |
1d07171c CL |
378 | } |
379 | ||
b789ef51 CL |
380 | static inline bool cmpxchg_double_slab(struct kmem_cache *s, struct page *page, |
381 | void *freelist_old, unsigned long counters_old, | |
382 | void *freelist_new, unsigned long counters_new, | |
383 | const char *n) | |
384 | { | |
2565409f HC |
385 | #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \ |
386 | defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE) | |
b789ef51 | 387 | if (s->flags & __CMPXCHG_DOUBLE) { |
cdcd6298 | 388 | if (cmpxchg_double(&page->freelist, &page->counters, |
0aa9a13d DC |
389 | freelist_old, counters_old, |
390 | freelist_new, counters_new)) | |
6f6528a1 | 391 | return true; |
b789ef51 CL |
392 | } else |
393 | #endif | |
394 | { | |
1d07171c CL |
395 | unsigned long flags; |
396 | ||
397 | local_irq_save(flags); | |
881db7fb | 398 | slab_lock(page); |
d0e0ac97 CG |
399 | if (page->freelist == freelist_old && |
400 | page->counters == counters_old) { | |
b789ef51 | 401 | page->freelist = freelist_new; |
a0320865 | 402 | set_page_slub_counters(page, counters_new); |
881db7fb | 403 | slab_unlock(page); |
1d07171c | 404 | local_irq_restore(flags); |
6f6528a1 | 405 | return true; |
b789ef51 | 406 | } |
881db7fb | 407 | slab_unlock(page); |
1d07171c | 408 | local_irq_restore(flags); |
b789ef51 CL |
409 | } |
410 | ||
411 | cpu_relax(); | |
412 | stat(s, CMPXCHG_DOUBLE_FAIL); | |
413 | ||
414 | #ifdef SLUB_DEBUG_CMPXCHG | |
f9f58285 | 415 | pr_info("%s %s: cmpxchg double redo ", n, s->name); |
b789ef51 CL |
416 | #endif |
417 | ||
6f6528a1 | 418 | return false; |
b789ef51 CL |
419 | } |
420 | ||
41ecc55b | 421 | #ifdef CONFIG_SLUB_DEBUG |
5f80b13a CL |
422 | /* |
423 | * Determine a map of object in use on a page. | |
424 | * | |
881db7fb | 425 | * Node listlock must be held to guarantee that the page does |
5f80b13a CL |
426 | * not vanish from under us. |
427 | */ | |
428 | static void get_map(struct kmem_cache *s, struct page *page, unsigned long *map) | |
429 | { | |
430 | void *p; | |
431 | void *addr = page_address(page); | |
432 | ||
433 | for (p = page->freelist; p; p = get_freepointer(s, p)) | |
434 | set_bit(slab_index(p, s, addr), map); | |
435 | } | |
436 | ||
41ecc55b CL |
437 | /* |
438 | * Debug settings: | |
439 | */ | |
89d3c87e | 440 | #if defined(CONFIG_SLUB_DEBUG_ON) |
f0630fff | 441 | static int slub_debug = DEBUG_DEFAULT_FLAGS; |
89d3c87e AR |
442 | #elif defined(CONFIG_KASAN) |
443 | static int slub_debug = SLAB_STORE_USER; | |
f0630fff | 444 | #else |
41ecc55b | 445 | static int slub_debug; |
f0630fff | 446 | #endif |
41ecc55b CL |
447 | |
448 | static char *slub_debug_slabs; | |
fa5ec8a1 | 449 | static int disable_higher_order_debug; |
41ecc55b | 450 | |
a79316c6 AR |
451 | /* |
452 | * slub is about to manipulate internal object metadata. This memory lies | |
453 | * outside the range of the allocated object, so accessing it would normally | |
454 | * be reported by kasan as a bounds error. metadata_access_enable() is used | |
455 | * to tell kasan that these accesses are OK. | |
456 | */ | |
457 | static inline void metadata_access_enable(void) | |
458 | { | |
459 | kasan_disable_current(); | |
460 | } | |
461 | ||
462 | static inline void metadata_access_disable(void) | |
463 | { | |
464 | kasan_enable_current(); | |
465 | } | |
466 | ||
81819f0f CL |
467 | /* |
468 | * Object debugging | |
469 | */ | |
470 | static void print_section(char *text, u8 *addr, unsigned int length) | |
471 | { | |
a79316c6 | 472 | metadata_access_enable(); |
ffc79d28 SAS |
473 | print_hex_dump(KERN_ERR, text, DUMP_PREFIX_ADDRESS, 16, 1, addr, |
474 | length, 1); | |
a79316c6 | 475 | metadata_access_disable(); |
81819f0f CL |
476 | } |
477 | ||
81819f0f CL |
478 | static struct track *get_track(struct kmem_cache *s, void *object, |
479 | enum track_item alloc) | |
480 | { | |
481 | struct track *p; | |
482 | ||
483 | if (s->offset) | |
484 | p = object + s->offset + sizeof(void *); | |
485 | else | |
486 | p = object + s->inuse; | |
487 | ||
488 | return p + alloc; | |
489 | } | |
490 | ||
491 | static void set_track(struct kmem_cache *s, void *object, | |
ce71e27c | 492 | enum track_item alloc, unsigned long addr) |
81819f0f | 493 | { |
1a00df4a | 494 | struct track *p = get_track(s, object, alloc); |
81819f0f | 495 | |
81819f0f | 496 | if (addr) { |
d6543e39 BG |
497 | #ifdef CONFIG_STACKTRACE |
498 | struct stack_trace trace; | |
499 | int i; | |
500 | ||
501 | trace.nr_entries = 0; | |
502 | trace.max_entries = TRACK_ADDRS_COUNT; | |
503 | trace.entries = p->addrs; | |
504 | trace.skip = 3; | |
a79316c6 | 505 | metadata_access_enable(); |
d6543e39 | 506 | save_stack_trace(&trace); |
a79316c6 | 507 | metadata_access_disable(); |
d6543e39 BG |
508 | |
509 | /* See rant in lockdep.c */ | |
510 | if (trace.nr_entries != 0 && | |
511 | trace.entries[trace.nr_entries - 1] == ULONG_MAX) | |
512 | trace.nr_entries--; | |
513 | ||
514 | for (i = trace.nr_entries; i < TRACK_ADDRS_COUNT; i++) | |
515 | p->addrs[i] = 0; | |
516 | #endif | |
81819f0f CL |
517 | p->addr = addr; |
518 | p->cpu = smp_processor_id(); | |
88e4ccf2 | 519 | p->pid = current->pid; |
81819f0f CL |
520 | p->when = jiffies; |
521 | } else | |
522 | memset(p, 0, sizeof(struct track)); | |
523 | } | |
524 | ||
81819f0f CL |
525 | static void init_tracking(struct kmem_cache *s, void *object) |
526 | { | |
24922684 CL |
527 | if (!(s->flags & SLAB_STORE_USER)) |
528 | return; | |
529 | ||
ce71e27c EGM |
530 | set_track(s, object, TRACK_FREE, 0UL); |
531 | set_track(s, object, TRACK_ALLOC, 0UL); | |
81819f0f CL |
532 | } |
533 | ||
534 | static void print_track(const char *s, struct track *t) | |
535 | { | |
536 | if (!t->addr) | |
537 | return; | |
538 | ||
f9f58285 FF |
539 | pr_err("INFO: %s in %pS age=%lu cpu=%u pid=%d\n", |
540 | s, (void *)t->addr, jiffies - t->when, t->cpu, t->pid); | |
d6543e39 BG |
541 | #ifdef CONFIG_STACKTRACE |
542 | { | |
543 | int i; | |
544 | for (i = 0; i < TRACK_ADDRS_COUNT; i++) | |
545 | if (t->addrs[i]) | |
f9f58285 | 546 | pr_err("\t%pS\n", (void *)t->addrs[i]); |
d6543e39 BG |
547 | else |
548 | break; | |
549 | } | |
550 | #endif | |
24922684 CL |
551 | } |
552 | ||
553 | static void print_tracking(struct kmem_cache *s, void *object) | |
554 | { | |
555 | if (!(s->flags & SLAB_STORE_USER)) | |
556 | return; | |
557 | ||
558 | print_track("Allocated", get_track(s, object, TRACK_ALLOC)); | |
559 | print_track("Freed", get_track(s, object, TRACK_FREE)); | |
560 | } | |
561 | ||
562 | static void print_page_info(struct page *page) | |
563 | { | |
f9f58285 | 564 | pr_err("INFO: Slab 0x%p objects=%u used=%u fp=0x%p flags=0x%04lx\n", |
d0e0ac97 | 565 | page, page->objects, page->inuse, page->freelist, page->flags); |
24922684 CL |
566 | |
567 | } | |
568 | ||
569 | static void slab_bug(struct kmem_cache *s, char *fmt, ...) | |
570 | { | |
ecc42fbe | 571 | struct va_format vaf; |
24922684 | 572 | va_list args; |
24922684 CL |
573 | |
574 | va_start(args, fmt); | |
ecc42fbe FF |
575 | vaf.fmt = fmt; |
576 | vaf.va = &args; | |
f9f58285 | 577 | pr_err("=============================================================================\n"); |
ecc42fbe | 578 | pr_err("BUG %s (%s): %pV\n", s->name, print_tainted(), &vaf); |
f9f58285 | 579 | pr_err("-----------------------------------------------------------------------------\n\n"); |
645df230 | 580 | |
373d4d09 | 581 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); |
ecc42fbe | 582 | va_end(args); |
81819f0f CL |
583 | } |
584 | ||
24922684 CL |
585 | static void slab_fix(struct kmem_cache *s, char *fmt, ...) |
586 | { | |
ecc42fbe | 587 | struct va_format vaf; |
24922684 | 588 | va_list args; |
24922684 CL |
589 | |
590 | va_start(args, fmt); | |
ecc42fbe FF |
591 | vaf.fmt = fmt; |
592 | vaf.va = &args; | |
593 | pr_err("FIX %s: %pV\n", s->name, &vaf); | |
24922684 | 594 | va_end(args); |
24922684 CL |
595 | } |
596 | ||
597 | static void print_trailer(struct kmem_cache *s, struct page *page, u8 *p) | |
81819f0f CL |
598 | { |
599 | unsigned int off; /* Offset of last byte */ | |
a973e9dd | 600 | u8 *addr = page_address(page); |
24922684 CL |
601 | |
602 | print_tracking(s, p); | |
603 | ||
604 | print_page_info(page); | |
605 | ||
f9f58285 FF |
606 | pr_err("INFO: Object 0x%p @offset=%tu fp=0x%p\n\n", |
607 | p, p - addr, get_freepointer(s, p)); | |
24922684 CL |
608 | |
609 | if (p > addr + 16) | |
ffc79d28 | 610 | print_section("Bytes b4 ", p - 16, 16); |
81819f0f | 611 | |
3b0efdfa | 612 | print_section("Object ", p, min_t(unsigned long, s->object_size, |
ffc79d28 | 613 | PAGE_SIZE)); |
81819f0f | 614 | if (s->flags & SLAB_RED_ZONE) |
3b0efdfa CL |
615 | print_section("Redzone ", p + s->object_size, |
616 | s->inuse - s->object_size); | |
81819f0f | 617 | |
81819f0f CL |
618 | if (s->offset) |
619 | off = s->offset + sizeof(void *); | |
620 | else | |
621 | off = s->inuse; | |
622 | ||
24922684 | 623 | if (s->flags & SLAB_STORE_USER) |
81819f0f | 624 | off += 2 * sizeof(struct track); |
81819f0f CL |
625 | |
626 | if (off != s->size) | |
627 | /* Beginning of the filler is the free pointer */ | |
ffc79d28 | 628 | print_section("Padding ", p + off, s->size - off); |
24922684 CL |
629 | |
630 | dump_stack(); | |
81819f0f CL |
631 | } |
632 | ||
75c66def | 633 | void object_err(struct kmem_cache *s, struct page *page, |
81819f0f CL |
634 | u8 *object, char *reason) |
635 | { | |
3dc50637 | 636 | slab_bug(s, "%s", reason); |
24922684 | 637 | print_trailer(s, page, object); |
81819f0f CL |
638 | } |
639 | ||
d0e0ac97 CG |
640 | static void slab_err(struct kmem_cache *s, struct page *page, |
641 | const char *fmt, ...) | |
81819f0f CL |
642 | { |
643 | va_list args; | |
644 | char buf[100]; | |
645 | ||
24922684 CL |
646 | va_start(args, fmt); |
647 | vsnprintf(buf, sizeof(buf), fmt, args); | |
81819f0f | 648 | va_end(args); |
3dc50637 | 649 | slab_bug(s, "%s", buf); |
24922684 | 650 | print_page_info(page); |
81819f0f CL |
651 | dump_stack(); |
652 | } | |
653 | ||
f7cb1933 | 654 | static void init_object(struct kmem_cache *s, void *object, u8 val) |
81819f0f CL |
655 | { |
656 | u8 *p = object; | |
657 | ||
658 | if (s->flags & __OBJECT_POISON) { | |
3b0efdfa CL |
659 | memset(p, POISON_FREE, s->object_size - 1); |
660 | p[s->object_size - 1] = POISON_END; | |
81819f0f CL |
661 | } |
662 | ||
663 | if (s->flags & SLAB_RED_ZONE) | |
3b0efdfa | 664 | memset(p + s->object_size, val, s->inuse - s->object_size); |
81819f0f CL |
665 | } |
666 | ||
24922684 CL |
667 | static void restore_bytes(struct kmem_cache *s, char *message, u8 data, |
668 | void *from, void *to) | |
669 | { | |
670 | slab_fix(s, "Restoring 0x%p-0x%p=0x%x\n", from, to - 1, data); | |
671 | memset(from, data, to - from); | |
672 | } | |
673 | ||
674 | static int check_bytes_and_report(struct kmem_cache *s, struct page *page, | |
675 | u8 *object, char *what, | |
06428780 | 676 | u8 *start, unsigned int value, unsigned int bytes) |
24922684 CL |
677 | { |
678 | u8 *fault; | |
679 | u8 *end; | |
680 | ||
a79316c6 | 681 | metadata_access_enable(); |
79824820 | 682 | fault = memchr_inv(start, value, bytes); |
a79316c6 | 683 | metadata_access_disable(); |
24922684 CL |
684 | if (!fault) |
685 | return 1; | |
686 | ||
687 | end = start + bytes; | |
688 | while (end > fault && end[-1] == value) | |
689 | end--; | |
690 | ||
691 | slab_bug(s, "%s overwritten", what); | |
f9f58285 | 692 | pr_err("INFO: 0x%p-0x%p. First byte 0x%x instead of 0x%x\n", |
24922684 CL |
693 | fault, end - 1, fault[0], value); |
694 | print_trailer(s, page, object); | |
695 | ||
696 | restore_bytes(s, what, value, fault, end); | |
697 | return 0; | |
81819f0f CL |
698 | } |
699 | ||
81819f0f CL |
700 | /* |
701 | * Object layout: | |
702 | * | |
703 | * object address | |
704 | * Bytes of the object to be managed. | |
705 | * If the freepointer may overlay the object then the free | |
706 | * pointer is the first word of the object. | |
672bba3a | 707 | * |
81819f0f CL |
708 | * Poisoning uses 0x6b (POISON_FREE) and the last byte is |
709 | * 0xa5 (POISON_END) | |
710 | * | |
3b0efdfa | 711 | * object + s->object_size |
81819f0f | 712 | * Padding to reach word boundary. This is also used for Redzoning. |
672bba3a | 713 | * Padding is extended by another word if Redzoning is enabled and |
3b0efdfa | 714 | * object_size == inuse. |
672bba3a | 715 | * |
81819f0f CL |
716 | * We fill with 0xbb (RED_INACTIVE) for inactive objects and with |
717 | * 0xcc (RED_ACTIVE) for objects in use. | |
718 | * | |
719 | * object + s->inuse | |
672bba3a CL |
720 | * Meta data starts here. |
721 | * | |
81819f0f CL |
722 | * A. Free pointer (if we cannot overwrite object on free) |
723 | * B. Tracking data for SLAB_STORE_USER | |
672bba3a | 724 | * C. Padding to reach required alignment boundary or at mininum |
6446faa2 | 725 | * one word if debugging is on to be able to detect writes |
672bba3a CL |
726 | * before the word boundary. |
727 | * | |
728 | * Padding is done using 0x5a (POISON_INUSE) | |
81819f0f CL |
729 | * |
730 | * object + s->size | |
672bba3a | 731 | * Nothing is used beyond s->size. |
81819f0f | 732 | * |
3b0efdfa | 733 | * If slabcaches are merged then the object_size and inuse boundaries are mostly |
672bba3a | 734 | * ignored. And therefore no slab options that rely on these boundaries |
81819f0f CL |
735 | * may be used with merged slabcaches. |
736 | */ | |
737 | ||
81819f0f CL |
738 | static int check_pad_bytes(struct kmem_cache *s, struct page *page, u8 *p) |
739 | { | |
740 | unsigned long off = s->inuse; /* The end of info */ | |
741 | ||
742 | if (s->offset) | |
743 | /* Freepointer is placed after the object. */ | |
744 | off += sizeof(void *); | |
745 | ||
746 | if (s->flags & SLAB_STORE_USER) | |
747 | /* We also have user information there */ | |
748 | off += 2 * sizeof(struct track); | |
749 | ||
750 | if (s->size == off) | |
751 | return 1; | |
752 | ||
24922684 CL |
753 | return check_bytes_and_report(s, page, p, "Object padding", |
754 | p + off, POISON_INUSE, s->size - off); | |
81819f0f CL |
755 | } |
756 | ||
39b26464 | 757 | /* Check the pad bytes at the end of a slab page */ |
81819f0f CL |
758 | static int slab_pad_check(struct kmem_cache *s, struct page *page) |
759 | { | |
24922684 CL |
760 | u8 *start; |
761 | u8 *fault; | |
762 | u8 *end; | |
763 | int length; | |
764 | int remainder; | |
81819f0f CL |
765 | |
766 | if (!(s->flags & SLAB_POISON)) | |
767 | return 1; | |
768 | ||
a973e9dd | 769 | start = page_address(page); |
ab9a0f19 | 770 | length = (PAGE_SIZE << compound_order(page)) - s->reserved; |
39b26464 CL |
771 | end = start + length; |
772 | remainder = length % s->size; | |
81819f0f CL |
773 | if (!remainder) |
774 | return 1; | |
775 | ||
a79316c6 | 776 | metadata_access_enable(); |
79824820 | 777 | fault = memchr_inv(end - remainder, POISON_INUSE, remainder); |
a79316c6 | 778 | metadata_access_disable(); |
24922684 CL |
779 | if (!fault) |
780 | return 1; | |
781 | while (end > fault && end[-1] == POISON_INUSE) | |
782 | end--; | |
783 | ||
784 | slab_err(s, page, "Padding overwritten. 0x%p-0x%p", fault, end - 1); | |
ffc79d28 | 785 | print_section("Padding ", end - remainder, remainder); |
24922684 | 786 | |
8a3d271d | 787 | restore_bytes(s, "slab padding", POISON_INUSE, end - remainder, end); |
24922684 | 788 | return 0; |
81819f0f CL |
789 | } |
790 | ||
791 | static int check_object(struct kmem_cache *s, struct page *page, | |
f7cb1933 | 792 | void *object, u8 val) |
81819f0f CL |
793 | { |
794 | u8 *p = object; | |
3b0efdfa | 795 | u8 *endobject = object + s->object_size; |
81819f0f CL |
796 | |
797 | if (s->flags & SLAB_RED_ZONE) { | |
24922684 | 798 | if (!check_bytes_and_report(s, page, object, "Redzone", |
3b0efdfa | 799 | endobject, val, s->inuse - s->object_size)) |
81819f0f | 800 | return 0; |
81819f0f | 801 | } else { |
3b0efdfa | 802 | if ((s->flags & SLAB_POISON) && s->object_size < s->inuse) { |
3adbefee | 803 | check_bytes_and_report(s, page, p, "Alignment padding", |
d0e0ac97 CG |
804 | endobject, POISON_INUSE, |
805 | s->inuse - s->object_size); | |
3adbefee | 806 | } |
81819f0f CL |
807 | } |
808 | ||
809 | if (s->flags & SLAB_POISON) { | |
f7cb1933 | 810 | if (val != SLUB_RED_ACTIVE && (s->flags & __OBJECT_POISON) && |
24922684 | 811 | (!check_bytes_and_report(s, page, p, "Poison", p, |
3b0efdfa | 812 | POISON_FREE, s->object_size - 1) || |
24922684 | 813 | !check_bytes_and_report(s, page, p, "Poison", |
3b0efdfa | 814 | p + s->object_size - 1, POISON_END, 1))) |
81819f0f | 815 | return 0; |
81819f0f CL |
816 | /* |
817 | * check_pad_bytes cleans up on its own. | |
818 | */ | |
819 | check_pad_bytes(s, page, p); | |
820 | } | |
821 | ||
f7cb1933 | 822 | if (!s->offset && val == SLUB_RED_ACTIVE) |
81819f0f CL |
823 | /* |
824 | * Object and freepointer overlap. Cannot check | |
825 | * freepointer while object is allocated. | |
826 | */ | |
827 | return 1; | |
828 | ||
829 | /* Check free pointer validity */ | |
830 | if (!check_valid_pointer(s, page, get_freepointer(s, p))) { | |
831 | object_err(s, page, p, "Freepointer corrupt"); | |
832 | /* | |
9f6c708e | 833 | * No choice but to zap it and thus lose the remainder |
81819f0f | 834 | * of the free objects in this slab. May cause |
672bba3a | 835 | * another error because the object count is now wrong. |
81819f0f | 836 | */ |
a973e9dd | 837 | set_freepointer(s, p, NULL); |
81819f0f CL |
838 | return 0; |
839 | } | |
840 | return 1; | |
841 | } | |
842 | ||
843 | static int check_slab(struct kmem_cache *s, struct page *page) | |
844 | { | |
39b26464 CL |
845 | int maxobj; |
846 | ||
81819f0f CL |
847 | VM_BUG_ON(!irqs_disabled()); |
848 | ||
849 | if (!PageSlab(page)) { | |
24922684 | 850 | slab_err(s, page, "Not a valid slab page"); |
81819f0f CL |
851 | return 0; |
852 | } | |
39b26464 | 853 | |
ab9a0f19 | 854 | maxobj = order_objects(compound_order(page), s->size, s->reserved); |
39b26464 CL |
855 | if (page->objects > maxobj) { |
856 | slab_err(s, page, "objects %u > max %u", | |
f6edde9c | 857 | page->objects, maxobj); |
39b26464 CL |
858 | return 0; |
859 | } | |
860 | if (page->inuse > page->objects) { | |
24922684 | 861 | slab_err(s, page, "inuse %u > max %u", |
f6edde9c | 862 | page->inuse, page->objects); |
81819f0f CL |
863 | return 0; |
864 | } | |
865 | /* Slab_pad_check fixes things up after itself */ | |
866 | slab_pad_check(s, page); | |
867 | return 1; | |
868 | } | |
869 | ||
870 | /* | |
672bba3a CL |
871 | * Determine if a certain object on a page is on the freelist. Must hold the |
872 | * slab lock to guarantee that the chains are in a consistent state. | |
81819f0f CL |
873 | */ |
874 | static int on_freelist(struct kmem_cache *s, struct page *page, void *search) | |
875 | { | |
876 | int nr = 0; | |
881db7fb | 877 | void *fp; |
81819f0f | 878 | void *object = NULL; |
f6edde9c | 879 | int max_objects; |
81819f0f | 880 | |
881db7fb | 881 | fp = page->freelist; |
39b26464 | 882 | while (fp && nr <= page->objects) { |
81819f0f CL |
883 | if (fp == search) |
884 | return 1; | |
885 | if (!check_valid_pointer(s, page, fp)) { | |
886 | if (object) { | |
887 | object_err(s, page, object, | |
888 | "Freechain corrupt"); | |
a973e9dd | 889 | set_freepointer(s, object, NULL); |
81819f0f | 890 | } else { |
24922684 | 891 | slab_err(s, page, "Freepointer corrupt"); |
a973e9dd | 892 | page->freelist = NULL; |
39b26464 | 893 | page->inuse = page->objects; |
24922684 | 894 | slab_fix(s, "Freelist cleared"); |
81819f0f CL |
895 | return 0; |
896 | } | |
897 | break; | |
898 | } | |
899 | object = fp; | |
900 | fp = get_freepointer(s, object); | |
901 | nr++; | |
902 | } | |
903 | ||
ab9a0f19 | 904 | max_objects = order_objects(compound_order(page), s->size, s->reserved); |
210b5c06 CG |
905 | if (max_objects > MAX_OBJS_PER_PAGE) |
906 | max_objects = MAX_OBJS_PER_PAGE; | |
224a88be CL |
907 | |
908 | if (page->objects != max_objects) { | |
909 | slab_err(s, page, "Wrong number of objects. Found %d but " | |
910 | "should be %d", page->objects, max_objects); | |
911 | page->objects = max_objects; | |
912 | slab_fix(s, "Number of objects adjusted."); | |
913 | } | |
39b26464 | 914 | if (page->inuse != page->objects - nr) { |
70d71228 | 915 | slab_err(s, page, "Wrong object count. Counter is %d but " |
39b26464 CL |
916 | "counted were %d", page->inuse, page->objects - nr); |
917 | page->inuse = page->objects - nr; | |
24922684 | 918 | slab_fix(s, "Object count adjusted."); |
81819f0f CL |
919 | } |
920 | return search == NULL; | |
921 | } | |
922 | ||
0121c619 CL |
923 | static void trace(struct kmem_cache *s, struct page *page, void *object, |
924 | int alloc) | |
3ec09742 CL |
925 | { |
926 | if (s->flags & SLAB_TRACE) { | |
f9f58285 | 927 | pr_info("TRACE %s %s 0x%p inuse=%d fp=0x%p\n", |
3ec09742 CL |
928 | s->name, |
929 | alloc ? "alloc" : "free", | |
930 | object, page->inuse, | |
931 | page->freelist); | |
932 | ||
933 | if (!alloc) | |
d0e0ac97 CG |
934 | print_section("Object ", (void *)object, |
935 | s->object_size); | |
3ec09742 CL |
936 | |
937 | dump_stack(); | |
938 | } | |
939 | } | |
940 | ||
643b1138 | 941 | /* |
672bba3a | 942 | * Tracking of fully allocated slabs for debugging purposes. |
643b1138 | 943 | */ |
5cc6eee8 CL |
944 | static void add_full(struct kmem_cache *s, |
945 | struct kmem_cache_node *n, struct page *page) | |
643b1138 | 946 | { |
5cc6eee8 CL |
947 | if (!(s->flags & SLAB_STORE_USER)) |
948 | return; | |
949 | ||
255d0884 | 950 | lockdep_assert_held(&n->list_lock); |
643b1138 | 951 | list_add(&page->lru, &n->full); |
643b1138 CL |
952 | } |
953 | ||
c65c1877 | 954 | static void remove_full(struct kmem_cache *s, struct kmem_cache_node *n, struct page *page) |
643b1138 | 955 | { |
643b1138 CL |
956 | if (!(s->flags & SLAB_STORE_USER)) |
957 | return; | |
958 | ||
255d0884 | 959 | lockdep_assert_held(&n->list_lock); |
643b1138 | 960 | list_del(&page->lru); |
643b1138 CL |
961 | } |
962 | ||
0f389ec6 CL |
963 | /* Tracking of the number of slabs for debugging purposes */ |
964 | static inline unsigned long slabs_node(struct kmem_cache *s, int node) | |
965 | { | |
966 | struct kmem_cache_node *n = get_node(s, node); | |
967 | ||
968 | return atomic_long_read(&n->nr_slabs); | |
969 | } | |
970 | ||
26c02cf0 AB |
971 | static inline unsigned long node_nr_slabs(struct kmem_cache_node *n) |
972 | { | |
973 | return atomic_long_read(&n->nr_slabs); | |
974 | } | |
975 | ||
205ab99d | 976 | static inline void inc_slabs_node(struct kmem_cache *s, int node, int objects) |
0f389ec6 CL |
977 | { |
978 | struct kmem_cache_node *n = get_node(s, node); | |
979 | ||
980 | /* | |
981 | * May be called early in order to allocate a slab for the | |
982 | * kmem_cache_node structure. Solve the chicken-egg | |
983 | * dilemma by deferring the increment of the count during | |
984 | * bootstrap (see early_kmem_cache_node_alloc). | |
985 | */ | |
338b2642 | 986 | if (likely(n)) { |
0f389ec6 | 987 | atomic_long_inc(&n->nr_slabs); |
205ab99d CL |
988 | atomic_long_add(objects, &n->total_objects); |
989 | } | |
0f389ec6 | 990 | } |
205ab99d | 991 | static inline void dec_slabs_node(struct kmem_cache *s, int node, int objects) |
0f389ec6 CL |
992 | { |
993 | struct kmem_cache_node *n = get_node(s, node); | |
994 | ||
995 | atomic_long_dec(&n->nr_slabs); | |
205ab99d | 996 | atomic_long_sub(objects, &n->total_objects); |
0f389ec6 CL |
997 | } |
998 | ||
999 | /* Object debug checks for alloc/free paths */ | |
3ec09742 CL |
1000 | static void setup_object_debug(struct kmem_cache *s, struct page *page, |
1001 | void *object) | |
1002 | { | |
1003 | if (!(s->flags & (SLAB_STORE_USER|SLAB_RED_ZONE|__OBJECT_POISON))) | |
1004 | return; | |
1005 | ||
f7cb1933 | 1006 | init_object(s, object, SLUB_RED_INACTIVE); |
3ec09742 CL |
1007 | init_tracking(s, object); |
1008 | } | |
1009 | ||
d0e0ac97 CG |
1010 | static noinline int alloc_debug_processing(struct kmem_cache *s, |
1011 | struct page *page, | |
ce71e27c | 1012 | void *object, unsigned long addr) |
81819f0f CL |
1013 | { |
1014 | if (!check_slab(s, page)) | |
1015 | goto bad; | |
1016 | ||
81819f0f CL |
1017 | if (!check_valid_pointer(s, page, object)) { |
1018 | object_err(s, page, object, "Freelist Pointer check fails"); | |
70d71228 | 1019 | goto bad; |
81819f0f CL |
1020 | } |
1021 | ||
f7cb1933 | 1022 | if (!check_object(s, page, object, SLUB_RED_INACTIVE)) |
81819f0f | 1023 | goto bad; |
81819f0f | 1024 | |
3ec09742 CL |
1025 | /* Success perform special debug activities for allocs */ |
1026 | if (s->flags & SLAB_STORE_USER) | |
1027 | set_track(s, object, TRACK_ALLOC, addr); | |
1028 | trace(s, page, object, 1); | |
f7cb1933 | 1029 | init_object(s, object, SLUB_RED_ACTIVE); |
81819f0f | 1030 | return 1; |
3ec09742 | 1031 | |
81819f0f CL |
1032 | bad: |
1033 | if (PageSlab(page)) { | |
1034 | /* | |
1035 | * If this is a slab page then lets do the best we can | |
1036 | * to avoid issues in the future. Marking all objects | |
672bba3a | 1037 | * as used avoids touching the remaining objects. |
81819f0f | 1038 | */ |
24922684 | 1039 | slab_fix(s, "Marking all objects used"); |
39b26464 | 1040 | page->inuse = page->objects; |
a973e9dd | 1041 | page->freelist = NULL; |
81819f0f CL |
1042 | } |
1043 | return 0; | |
1044 | } | |
1045 | ||
81084651 | 1046 | /* Supports checking bulk free of a constructed freelist */ |
19c7ff9e | 1047 | static noinline struct kmem_cache_node *free_debug_processing( |
81084651 JDB |
1048 | struct kmem_cache *s, struct page *page, |
1049 | void *head, void *tail, int bulk_cnt, | |
19c7ff9e | 1050 | unsigned long addr, unsigned long *flags) |
81819f0f | 1051 | { |
19c7ff9e | 1052 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); |
81084651 JDB |
1053 | void *object = head; |
1054 | int cnt = 0; | |
5c2e4bbb | 1055 | |
19c7ff9e | 1056 | spin_lock_irqsave(&n->list_lock, *flags); |
881db7fb CL |
1057 | slab_lock(page); |
1058 | ||
81819f0f CL |
1059 | if (!check_slab(s, page)) |
1060 | goto fail; | |
1061 | ||
81084651 JDB |
1062 | next_object: |
1063 | cnt++; | |
1064 | ||
81819f0f | 1065 | if (!check_valid_pointer(s, page, object)) { |
70d71228 | 1066 | slab_err(s, page, "Invalid object pointer 0x%p", object); |
81819f0f CL |
1067 | goto fail; |
1068 | } | |
1069 | ||
1070 | if (on_freelist(s, page, object)) { | |
24922684 | 1071 | object_err(s, page, object, "Object already free"); |
81819f0f CL |
1072 | goto fail; |
1073 | } | |
1074 | ||
f7cb1933 | 1075 | if (!check_object(s, page, object, SLUB_RED_ACTIVE)) |
5c2e4bbb | 1076 | goto out; |
81819f0f | 1077 | |
1b4f59e3 | 1078 | if (unlikely(s != page->slab_cache)) { |
3adbefee | 1079 | if (!PageSlab(page)) { |
70d71228 CL |
1080 | slab_err(s, page, "Attempt to free object(0x%p) " |
1081 | "outside of slab", object); | |
1b4f59e3 | 1082 | } else if (!page->slab_cache) { |
f9f58285 FF |
1083 | pr_err("SLUB <none>: no slab for object 0x%p.\n", |
1084 | object); | |
70d71228 | 1085 | dump_stack(); |
06428780 | 1086 | } else |
24922684 CL |
1087 | object_err(s, page, object, |
1088 | "page slab pointer corrupt."); | |
81819f0f CL |
1089 | goto fail; |
1090 | } | |
3ec09742 | 1091 | |
3ec09742 CL |
1092 | if (s->flags & SLAB_STORE_USER) |
1093 | set_track(s, object, TRACK_FREE, addr); | |
1094 | trace(s, page, object, 0); | |
81084651 | 1095 | /* Freepointer not overwritten by init_object(), SLAB_POISON moved it */ |
f7cb1933 | 1096 | init_object(s, object, SLUB_RED_INACTIVE); |
81084651 JDB |
1097 | |
1098 | /* Reached end of constructed freelist yet? */ | |
1099 | if (object != tail) { | |
1100 | object = get_freepointer(s, object); | |
1101 | goto next_object; | |
1102 | } | |
5c2e4bbb | 1103 | out: |
81084651 JDB |
1104 | if (cnt != bulk_cnt) |
1105 | slab_err(s, page, "Bulk freelist count(%d) invalid(%d)\n", | |
1106 | bulk_cnt, cnt); | |
1107 | ||
881db7fb | 1108 | slab_unlock(page); |
19c7ff9e CL |
1109 | /* |
1110 | * Keep node_lock to preserve integrity | |
1111 | * until the object is actually freed | |
1112 | */ | |
1113 | return n; | |
3ec09742 | 1114 | |
81819f0f | 1115 | fail: |
19c7ff9e CL |
1116 | slab_unlock(page); |
1117 | spin_unlock_irqrestore(&n->list_lock, *flags); | |
24922684 | 1118 | slab_fix(s, "Object at 0x%p not freed", object); |
19c7ff9e | 1119 | return NULL; |
81819f0f CL |
1120 | } |
1121 | ||
41ecc55b CL |
1122 | static int __init setup_slub_debug(char *str) |
1123 | { | |
f0630fff CL |
1124 | slub_debug = DEBUG_DEFAULT_FLAGS; |
1125 | if (*str++ != '=' || !*str) | |
1126 | /* | |
1127 | * No options specified. Switch on full debugging. | |
1128 | */ | |
1129 | goto out; | |
1130 | ||
1131 | if (*str == ',') | |
1132 | /* | |
1133 | * No options but restriction on slabs. This means full | |
1134 | * debugging for slabs matching a pattern. | |
1135 | */ | |
1136 | goto check_slabs; | |
1137 | ||
1138 | slub_debug = 0; | |
1139 | if (*str == '-') | |
1140 | /* | |
1141 | * Switch off all debugging measures. | |
1142 | */ | |
1143 | goto out; | |
1144 | ||
1145 | /* | |
1146 | * Determine which debug features should be switched on | |
1147 | */ | |
06428780 | 1148 | for (; *str && *str != ','; str++) { |
f0630fff CL |
1149 | switch (tolower(*str)) { |
1150 | case 'f': | |
1151 | slub_debug |= SLAB_DEBUG_FREE; | |
1152 | break; | |
1153 | case 'z': | |
1154 | slub_debug |= SLAB_RED_ZONE; | |
1155 | break; | |
1156 | case 'p': | |
1157 | slub_debug |= SLAB_POISON; | |
1158 | break; | |
1159 | case 'u': | |
1160 | slub_debug |= SLAB_STORE_USER; | |
1161 | break; | |
1162 | case 't': | |
1163 | slub_debug |= SLAB_TRACE; | |
1164 | break; | |
4c13dd3b DM |
1165 | case 'a': |
1166 | slub_debug |= SLAB_FAILSLAB; | |
1167 | break; | |
08303a73 CA |
1168 | case 'o': |
1169 | /* | |
1170 | * Avoid enabling debugging on caches if its minimum | |
1171 | * order would increase as a result. | |
1172 | */ | |
1173 | disable_higher_order_debug = 1; | |
1174 | break; | |
f0630fff | 1175 | default: |
f9f58285 FF |
1176 | pr_err("slub_debug option '%c' unknown. skipped\n", |
1177 | *str); | |
f0630fff | 1178 | } |
41ecc55b CL |
1179 | } |
1180 | ||
f0630fff | 1181 | check_slabs: |
41ecc55b CL |
1182 | if (*str == ',') |
1183 | slub_debug_slabs = str + 1; | |
f0630fff | 1184 | out: |
41ecc55b CL |
1185 | return 1; |
1186 | } | |
1187 | ||
1188 | __setup("slub_debug", setup_slub_debug); | |
1189 | ||
423c929c | 1190 | unsigned long kmem_cache_flags(unsigned long object_size, |
ba0268a8 | 1191 | unsigned long flags, const char *name, |
51cc5068 | 1192 | void (*ctor)(void *)) |
41ecc55b CL |
1193 | { |
1194 | /* | |
e153362a | 1195 | * Enable debugging if selected on the kernel commandline. |
41ecc55b | 1196 | */ |
c6f58d9b CL |
1197 | if (slub_debug && (!slub_debug_slabs || (name && |
1198 | !strncmp(slub_debug_slabs, name, strlen(slub_debug_slabs))))) | |
3de47213 | 1199 | flags |= slub_debug; |
ba0268a8 CL |
1200 | |
1201 | return flags; | |
41ecc55b | 1202 | } |
b4a64718 | 1203 | #else /* !CONFIG_SLUB_DEBUG */ |
3ec09742 CL |
1204 | static inline void setup_object_debug(struct kmem_cache *s, |
1205 | struct page *page, void *object) {} | |
41ecc55b | 1206 | |
3ec09742 | 1207 | static inline int alloc_debug_processing(struct kmem_cache *s, |
ce71e27c | 1208 | struct page *page, void *object, unsigned long addr) { return 0; } |
41ecc55b | 1209 | |
19c7ff9e | 1210 | static inline struct kmem_cache_node *free_debug_processing( |
81084651 JDB |
1211 | struct kmem_cache *s, struct page *page, |
1212 | void *head, void *tail, int bulk_cnt, | |
19c7ff9e | 1213 | unsigned long addr, unsigned long *flags) { return NULL; } |
41ecc55b | 1214 | |
41ecc55b CL |
1215 | static inline int slab_pad_check(struct kmem_cache *s, struct page *page) |
1216 | { return 1; } | |
1217 | static inline int check_object(struct kmem_cache *s, struct page *page, | |
f7cb1933 | 1218 | void *object, u8 val) { return 1; } |
5cc6eee8 CL |
1219 | static inline void add_full(struct kmem_cache *s, struct kmem_cache_node *n, |
1220 | struct page *page) {} | |
c65c1877 PZ |
1221 | static inline void remove_full(struct kmem_cache *s, struct kmem_cache_node *n, |
1222 | struct page *page) {} | |
423c929c | 1223 | unsigned long kmem_cache_flags(unsigned long object_size, |
ba0268a8 | 1224 | unsigned long flags, const char *name, |
51cc5068 | 1225 | void (*ctor)(void *)) |
ba0268a8 CL |
1226 | { |
1227 | return flags; | |
1228 | } | |
41ecc55b | 1229 | #define slub_debug 0 |
0f389ec6 | 1230 | |
fdaa45e9 IM |
1231 | #define disable_higher_order_debug 0 |
1232 | ||
0f389ec6 CL |
1233 | static inline unsigned long slabs_node(struct kmem_cache *s, int node) |
1234 | { return 0; } | |
26c02cf0 AB |
1235 | static inline unsigned long node_nr_slabs(struct kmem_cache_node *n) |
1236 | { return 0; } | |
205ab99d CL |
1237 | static inline void inc_slabs_node(struct kmem_cache *s, int node, |
1238 | int objects) {} | |
1239 | static inline void dec_slabs_node(struct kmem_cache *s, int node, | |
1240 | int objects) {} | |
7d550c56 | 1241 | |
02e72cc6 AR |
1242 | #endif /* CONFIG_SLUB_DEBUG */ |
1243 | ||
1244 | /* | |
1245 | * Hooks for other subsystems that check memory allocations. In a typical | |
1246 | * production configuration these hooks all should produce no code at all. | |
1247 | */ | |
d56791b3 RB |
1248 | static inline void kmalloc_large_node_hook(void *ptr, size_t size, gfp_t flags) |
1249 | { | |
1250 | kmemleak_alloc(ptr, size, 1, flags); | |
0316bec2 | 1251 | kasan_kmalloc_large(ptr, size); |
d56791b3 RB |
1252 | } |
1253 | ||
1254 | static inline void kfree_hook(const void *x) | |
1255 | { | |
1256 | kmemleak_free(x); | |
0316bec2 | 1257 | kasan_kfree_large(x); |
d56791b3 RB |
1258 | } |
1259 | ||
d56791b3 RB |
1260 | static inline void slab_free_hook(struct kmem_cache *s, void *x) |
1261 | { | |
1262 | kmemleak_free_recursive(x, s->flags); | |
7d550c56 | 1263 | |
02e72cc6 AR |
1264 | /* |
1265 | * Trouble is that we may no longer disable interrupts in the fast path | |
1266 | * So in order to make the debug calls that expect irqs to be | |
1267 | * disabled we need to disable interrupts temporarily. | |
1268 | */ | |
1269 | #if defined(CONFIG_KMEMCHECK) || defined(CONFIG_LOCKDEP) | |
1270 | { | |
1271 | unsigned long flags; | |
1272 | ||
1273 | local_irq_save(flags); | |
1274 | kmemcheck_slab_free(s, x, s->object_size); | |
1275 | debug_check_no_locks_freed(x, s->object_size); | |
1276 | local_irq_restore(flags); | |
1277 | } | |
1278 | #endif | |
1279 | if (!(s->flags & SLAB_DEBUG_OBJECTS)) | |
1280 | debug_check_no_obj_freed(x, s->object_size); | |
0316bec2 AR |
1281 | |
1282 | kasan_slab_free(s, x); | |
02e72cc6 | 1283 | } |
205ab99d | 1284 | |
81084651 JDB |
1285 | static inline void slab_free_freelist_hook(struct kmem_cache *s, |
1286 | void *head, void *tail) | |
1287 | { | |
1288 | /* | |
1289 | * Compiler cannot detect this function can be removed if slab_free_hook() | |
1290 | * evaluates to nothing. Thus, catch all relevant config debug options here. | |
1291 | */ | |
1292 | #if defined(CONFIG_KMEMCHECK) || \ | |
1293 | defined(CONFIG_LOCKDEP) || \ | |
1294 | defined(CONFIG_DEBUG_KMEMLEAK) || \ | |
1295 | defined(CONFIG_DEBUG_OBJECTS_FREE) || \ | |
1296 | defined(CONFIG_KASAN) | |
1297 | ||
1298 | void *object = head; | |
1299 | void *tail_obj = tail ? : head; | |
1300 | ||
1301 | do { | |
1302 | slab_free_hook(s, object); | |
1303 | } while ((object != tail_obj) && | |
1304 | (object = get_freepointer(s, object))); | |
1305 | #endif | |
1306 | } | |
1307 | ||
588f8ba9 TG |
1308 | static void setup_object(struct kmem_cache *s, struct page *page, |
1309 | void *object) | |
1310 | { | |
1311 | setup_object_debug(s, page, object); | |
1312 | if (unlikely(s->ctor)) { | |
1313 | kasan_unpoison_object_data(s, object); | |
1314 | s->ctor(object); | |
1315 | kasan_poison_object_data(s, object); | |
1316 | } | |
1317 | } | |
1318 | ||
81819f0f CL |
1319 | /* |
1320 | * Slab allocation and freeing | |
1321 | */ | |
5dfb4175 VD |
1322 | static inline struct page *alloc_slab_page(struct kmem_cache *s, |
1323 | gfp_t flags, int node, struct kmem_cache_order_objects oo) | |
65c3376a | 1324 | { |
5dfb4175 | 1325 | struct page *page; |
65c3376a CL |
1326 | int order = oo_order(oo); |
1327 | ||
b1eeab67 VN |
1328 | flags |= __GFP_NOTRACK; |
1329 | ||
2154a336 | 1330 | if (node == NUMA_NO_NODE) |
5dfb4175 | 1331 | page = alloc_pages(flags, order); |
65c3376a | 1332 | else |
96db800f | 1333 | page = __alloc_pages_node(node, flags, order); |
5dfb4175 | 1334 | |
f3ccb2c4 VD |
1335 | if (page && memcg_charge_slab(page, flags, order, s)) { |
1336 | __free_pages(page, order); | |
1337 | page = NULL; | |
1338 | } | |
5dfb4175 VD |
1339 | |
1340 | return page; | |
65c3376a CL |
1341 | } |
1342 | ||
81819f0f CL |
1343 | static struct page *allocate_slab(struct kmem_cache *s, gfp_t flags, int node) |
1344 | { | |
06428780 | 1345 | struct page *page; |
834f3d11 | 1346 | struct kmem_cache_order_objects oo = s->oo; |
ba52270d | 1347 | gfp_t alloc_gfp; |
588f8ba9 TG |
1348 | void *start, *p; |
1349 | int idx, order; | |
81819f0f | 1350 | |
7e0528da CL |
1351 | flags &= gfp_allowed_mask; |
1352 | ||
d0164adc | 1353 | if (gfpflags_allow_blocking(flags)) |
7e0528da CL |
1354 | local_irq_enable(); |
1355 | ||
b7a49f0d | 1356 | flags |= s->allocflags; |
e12ba74d | 1357 | |
ba52270d PE |
1358 | /* |
1359 | * Let the initial higher-order allocation fail under memory pressure | |
1360 | * so we fall-back to the minimum order allocation. | |
1361 | */ | |
1362 | alloc_gfp = (flags | __GFP_NOWARN | __GFP_NORETRY) & ~__GFP_NOFAIL; | |
d0164adc MG |
1363 | if ((alloc_gfp & __GFP_DIRECT_RECLAIM) && oo_order(oo) > oo_order(s->min)) |
1364 | alloc_gfp = (alloc_gfp | __GFP_NOMEMALLOC) & ~__GFP_DIRECT_RECLAIM; | |
ba52270d | 1365 | |
5dfb4175 | 1366 | page = alloc_slab_page(s, alloc_gfp, node, oo); |
65c3376a CL |
1367 | if (unlikely(!page)) { |
1368 | oo = s->min; | |
80c3a998 | 1369 | alloc_gfp = flags; |
65c3376a CL |
1370 | /* |
1371 | * Allocation may have failed due to fragmentation. | |
1372 | * Try a lower order alloc if possible | |
1373 | */ | |
5dfb4175 | 1374 | page = alloc_slab_page(s, alloc_gfp, node, oo); |
588f8ba9 TG |
1375 | if (unlikely(!page)) |
1376 | goto out; | |
1377 | stat(s, ORDER_FALLBACK); | |
65c3376a | 1378 | } |
5a896d9e | 1379 | |
588f8ba9 TG |
1380 | if (kmemcheck_enabled && |
1381 | !(s->flags & (SLAB_NOTRACK | DEBUG_DEFAULT_FLAGS))) { | |
b1eeab67 VN |
1382 | int pages = 1 << oo_order(oo); |
1383 | ||
80c3a998 | 1384 | kmemcheck_alloc_shadow(page, oo_order(oo), alloc_gfp, node); |
b1eeab67 VN |
1385 | |
1386 | /* | |
1387 | * Objects from caches that have a constructor don't get | |
1388 | * cleared when they're allocated, so we need to do it here. | |
1389 | */ | |
1390 | if (s->ctor) | |
1391 | kmemcheck_mark_uninitialized_pages(page, pages); | |
1392 | else | |
1393 | kmemcheck_mark_unallocated_pages(page, pages); | |
5a896d9e VN |
1394 | } |
1395 | ||
834f3d11 | 1396 | page->objects = oo_objects(oo); |
81819f0f | 1397 | |
1f458cbf | 1398 | order = compound_order(page); |
1b4f59e3 | 1399 | page->slab_cache = s; |
c03f94cc | 1400 | __SetPageSlab(page); |
2f064f34 | 1401 | if (page_is_pfmemalloc(page)) |
072bb0aa | 1402 | SetPageSlabPfmemalloc(page); |
81819f0f CL |
1403 | |
1404 | start = page_address(page); | |
81819f0f CL |
1405 | |
1406 | if (unlikely(s->flags & SLAB_POISON)) | |
1f458cbf | 1407 | memset(start, POISON_INUSE, PAGE_SIZE << order); |
81819f0f | 1408 | |
0316bec2 AR |
1409 | kasan_poison_slab(page); |
1410 | ||
54266640 WY |
1411 | for_each_object_idx(p, idx, s, start, page->objects) { |
1412 | setup_object(s, page, p); | |
1413 | if (likely(idx < page->objects)) | |
1414 | set_freepointer(s, p, p + s->size); | |
1415 | else | |
1416 | set_freepointer(s, p, NULL); | |
81819f0f | 1417 | } |
81819f0f CL |
1418 | |
1419 | page->freelist = start; | |
e6e82ea1 | 1420 | page->inuse = page->objects; |
8cb0a506 | 1421 | page->frozen = 1; |
588f8ba9 | 1422 | |
81819f0f | 1423 | out: |
d0164adc | 1424 | if (gfpflags_allow_blocking(flags)) |
588f8ba9 TG |
1425 | local_irq_disable(); |
1426 | if (!page) | |
1427 | return NULL; | |
1428 | ||
1429 | mod_zone_page_state(page_zone(page), | |
1430 | (s->flags & SLAB_RECLAIM_ACCOUNT) ? | |
1431 | NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE, | |
1432 | 1 << oo_order(oo)); | |
1433 | ||
1434 | inc_slabs_node(s, page_to_nid(page), page->objects); | |
1435 | ||
81819f0f CL |
1436 | return page; |
1437 | } | |
1438 | ||
588f8ba9 TG |
1439 | static struct page *new_slab(struct kmem_cache *s, gfp_t flags, int node) |
1440 | { | |
1441 | if (unlikely(flags & GFP_SLAB_BUG_MASK)) { | |
1442 | pr_emerg("gfp: %u\n", flags & GFP_SLAB_BUG_MASK); | |
1443 | BUG(); | |
1444 | } | |
1445 | ||
1446 | return allocate_slab(s, | |
1447 | flags & (GFP_RECLAIM_MASK | GFP_CONSTRAINT_MASK), node); | |
1448 | } | |
1449 | ||
81819f0f CL |
1450 | static void __free_slab(struct kmem_cache *s, struct page *page) |
1451 | { | |
834f3d11 CL |
1452 | int order = compound_order(page); |
1453 | int pages = 1 << order; | |
81819f0f | 1454 | |
af537b0a | 1455 | if (kmem_cache_debug(s)) { |
81819f0f CL |
1456 | void *p; |
1457 | ||
1458 | slab_pad_check(s, page); | |
224a88be CL |
1459 | for_each_object(p, s, page_address(page), |
1460 | page->objects) | |
f7cb1933 | 1461 | check_object(s, page, p, SLUB_RED_INACTIVE); |
81819f0f CL |
1462 | } |
1463 | ||
b1eeab67 | 1464 | kmemcheck_free_shadow(page, compound_order(page)); |
5a896d9e | 1465 | |
81819f0f CL |
1466 | mod_zone_page_state(page_zone(page), |
1467 | (s->flags & SLAB_RECLAIM_ACCOUNT) ? | |
1468 | NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE, | |
06428780 | 1469 | -pages); |
81819f0f | 1470 | |
072bb0aa | 1471 | __ClearPageSlabPfmemalloc(page); |
49bd5221 | 1472 | __ClearPageSlab(page); |
1f458cbf | 1473 | |
22b751c3 | 1474 | page_mapcount_reset(page); |
1eb5ac64 NP |
1475 | if (current->reclaim_state) |
1476 | current->reclaim_state->reclaimed_slab += pages; | |
f3ccb2c4 | 1477 | __free_kmem_pages(page, order); |
81819f0f CL |
1478 | } |
1479 | ||
da9a638c LJ |
1480 | #define need_reserve_slab_rcu \ |
1481 | (sizeof(((struct page *)NULL)->lru) < sizeof(struct rcu_head)) | |
1482 | ||
81819f0f CL |
1483 | static void rcu_free_slab(struct rcu_head *h) |
1484 | { | |
1485 | struct page *page; | |
1486 | ||
da9a638c LJ |
1487 | if (need_reserve_slab_rcu) |
1488 | page = virt_to_head_page(h); | |
1489 | else | |
1490 | page = container_of((struct list_head *)h, struct page, lru); | |
1491 | ||
1b4f59e3 | 1492 | __free_slab(page->slab_cache, page); |
81819f0f CL |
1493 | } |
1494 | ||
1495 | static void free_slab(struct kmem_cache *s, struct page *page) | |
1496 | { | |
1497 | if (unlikely(s->flags & SLAB_DESTROY_BY_RCU)) { | |
da9a638c LJ |
1498 | struct rcu_head *head; |
1499 | ||
1500 | if (need_reserve_slab_rcu) { | |
1501 | int order = compound_order(page); | |
1502 | int offset = (PAGE_SIZE << order) - s->reserved; | |
1503 | ||
1504 | VM_BUG_ON(s->reserved != sizeof(*head)); | |
1505 | head = page_address(page) + offset; | |
1506 | } else { | |
bc4f610d | 1507 | head = &page->rcu_head; |
da9a638c | 1508 | } |
81819f0f CL |
1509 | |
1510 | call_rcu(head, rcu_free_slab); | |
1511 | } else | |
1512 | __free_slab(s, page); | |
1513 | } | |
1514 | ||
1515 | static void discard_slab(struct kmem_cache *s, struct page *page) | |
1516 | { | |
205ab99d | 1517 | dec_slabs_node(s, page_to_nid(page), page->objects); |
81819f0f CL |
1518 | free_slab(s, page); |
1519 | } | |
1520 | ||
1521 | /* | |
5cc6eee8 | 1522 | * Management of partially allocated slabs. |
81819f0f | 1523 | */ |
1e4dd946 SR |
1524 | static inline void |
1525 | __add_partial(struct kmem_cache_node *n, struct page *page, int tail) | |
81819f0f | 1526 | { |
e95eed57 | 1527 | n->nr_partial++; |
136333d1 | 1528 | if (tail == DEACTIVATE_TO_TAIL) |
7c2e132c CL |
1529 | list_add_tail(&page->lru, &n->partial); |
1530 | else | |
1531 | list_add(&page->lru, &n->partial); | |
81819f0f CL |
1532 | } |
1533 | ||
1e4dd946 SR |
1534 | static inline void add_partial(struct kmem_cache_node *n, |
1535 | struct page *page, int tail) | |
62e346a8 | 1536 | { |
c65c1877 | 1537 | lockdep_assert_held(&n->list_lock); |
1e4dd946 SR |
1538 | __add_partial(n, page, tail); |
1539 | } | |
c65c1877 | 1540 | |
1e4dd946 SR |
1541 | static inline void remove_partial(struct kmem_cache_node *n, |
1542 | struct page *page) | |
1543 | { | |
1544 | lockdep_assert_held(&n->list_lock); | |
52b4b950 DS |
1545 | list_del(&page->lru); |
1546 | n->nr_partial--; | |
1e4dd946 SR |
1547 | } |
1548 | ||
81819f0f | 1549 | /* |
7ced3719 CL |
1550 | * Remove slab from the partial list, freeze it and |
1551 | * return the pointer to the freelist. | |
81819f0f | 1552 | * |
497b66f2 | 1553 | * Returns a list of objects or NULL if it fails. |
81819f0f | 1554 | */ |
497b66f2 | 1555 | static inline void *acquire_slab(struct kmem_cache *s, |
acd19fd1 | 1556 | struct kmem_cache_node *n, struct page *page, |
633b0764 | 1557 | int mode, int *objects) |
81819f0f | 1558 | { |
2cfb7455 CL |
1559 | void *freelist; |
1560 | unsigned long counters; | |
1561 | struct page new; | |
1562 | ||
c65c1877 PZ |
1563 | lockdep_assert_held(&n->list_lock); |
1564 | ||
2cfb7455 CL |
1565 | /* |
1566 | * Zap the freelist and set the frozen bit. | |
1567 | * The old freelist is the list of objects for the | |
1568 | * per cpu allocation list. | |
1569 | */ | |
7ced3719 CL |
1570 | freelist = page->freelist; |
1571 | counters = page->counters; | |
1572 | new.counters = counters; | |
633b0764 | 1573 | *objects = new.objects - new.inuse; |
23910c50 | 1574 | if (mode) { |
7ced3719 | 1575 | new.inuse = page->objects; |
23910c50 PE |
1576 | new.freelist = NULL; |
1577 | } else { | |
1578 | new.freelist = freelist; | |
1579 | } | |
2cfb7455 | 1580 | |
a0132ac0 | 1581 | VM_BUG_ON(new.frozen); |
7ced3719 | 1582 | new.frozen = 1; |
2cfb7455 | 1583 | |
7ced3719 | 1584 | if (!__cmpxchg_double_slab(s, page, |
2cfb7455 | 1585 | freelist, counters, |
02d7633f | 1586 | new.freelist, new.counters, |
7ced3719 | 1587 | "acquire_slab")) |
7ced3719 | 1588 | return NULL; |
2cfb7455 CL |
1589 | |
1590 | remove_partial(n, page); | |
7ced3719 | 1591 | WARN_ON(!freelist); |
49e22585 | 1592 | return freelist; |
81819f0f CL |
1593 | } |
1594 | ||
633b0764 | 1595 | static void put_cpu_partial(struct kmem_cache *s, struct page *page, int drain); |
8ba00bb6 | 1596 | static inline bool pfmemalloc_match(struct page *page, gfp_t gfpflags); |
49e22585 | 1597 | |
81819f0f | 1598 | /* |
672bba3a | 1599 | * Try to allocate a partial slab from a specific node. |
81819f0f | 1600 | */ |
8ba00bb6 JK |
1601 | static void *get_partial_node(struct kmem_cache *s, struct kmem_cache_node *n, |
1602 | struct kmem_cache_cpu *c, gfp_t flags) | |
81819f0f | 1603 | { |
49e22585 CL |
1604 | struct page *page, *page2; |
1605 | void *object = NULL; | |
633b0764 JK |
1606 | int available = 0; |
1607 | int objects; | |
81819f0f CL |
1608 | |
1609 | /* | |
1610 | * Racy check. If we mistakenly see no partial slabs then we | |
1611 | * just allocate an empty slab. If we mistakenly try to get a | |
672bba3a CL |
1612 | * partial slab and there is none available then get_partials() |
1613 | * will return NULL. | |
81819f0f CL |
1614 | */ |
1615 | if (!n || !n->nr_partial) | |
1616 | return NULL; | |
1617 | ||
1618 | spin_lock(&n->list_lock); | |
49e22585 | 1619 | list_for_each_entry_safe(page, page2, &n->partial, lru) { |
8ba00bb6 | 1620 | void *t; |
49e22585 | 1621 | |
8ba00bb6 JK |
1622 | if (!pfmemalloc_match(page, flags)) |
1623 | continue; | |
1624 | ||
633b0764 | 1625 | t = acquire_slab(s, n, page, object == NULL, &objects); |
49e22585 CL |
1626 | if (!t) |
1627 | break; | |
1628 | ||
633b0764 | 1629 | available += objects; |
12d79634 | 1630 | if (!object) { |
49e22585 | 1631 | c->page = page; |
49e22585 | 1632 | stat(s, ALLOC_FROM_PARTIAL); |
49e22585 | 1633 | object = t; |
49e22585 | 1634 | } else { |
633b0764 | 1635 | put_cpu_partial(s, page, 0); |
8028dcea | 1636 | stat(s, CPU_PARTIAL_NODE); |
49e22585 | 1637 | } |
345c905d JK |
1638 | if (!kmem_cache_has_cpu_partial(s) |
1639 | || available > s->cpu_partial / 2) | |
49e22585 CL |
1640 | break; |
1641 | ||
497b66f2 | 1642 | } |
81819f0f | 1643 | spin_unlock(&n->list_lock); |
497b66f2 | 1644 | return object; |
81819f0f CL |
1645 | } |
1646 | ||
1647 | /* | |
672bba3a | 1648 | * Get a page from somewhere. Search in increasing NUMA distances. |
81819f0f | 1649 | */ |
de3ec035 | 1650 | static void *get_any_partial(struct kmem_cache *s, gfp_t flags, |
acd19fd1 | 1651 | struct kmem_cache_cpu *c) |
81819f0f CL |
1652 | { |
1653 | #ifdef CONFIG_NUMA | |
1654 | struct zonelist *zonelist; | |
dd1a239f | 1655 | struct zoneref *z; |
54a6eb5c MG |
1656 | struct zone *zone; |
1657 | enum zone_type high_zoneidx = gfp_zone(flags); | |
497b66f2 | 1658 | void *object; |
cc9a6c87 | 1659 | unsigned int cpuset_mems_cookie; |
81819f0f CL |
1660 | |
1661 | /* | |
672bba3a CL |
1662 | * The defrag ratio allows a configuration of the tradeoffs between |
1663 | * inter node defragmentation and node local allocations. A lower | |
1664 | * defrag_ratio increases the tendency to do local allocations | |
1665 | * instead of attempting to obtain partial slabs from other nodes. | |
81819f0f | 1666 | * |
672bba3a CL |
1667 | * If the defrag_ratio is set to 0 then kmalloc() always |
1668 | * returns node local objects. If the ratio is higher then kmalloc() | |
1669 | * may return off node objects because partial slabs are obtained | |
1670 | * from other nodes and filled up. | |
81819f0f | 1671 | * |
6446faa2 | 1672 | * If /sys/kernel/slab/xx/defrag_ratio is set to 100 (which makes |
672bba3a CL |
1673 | * defrag_ratio = 1000) then every (well almost) allocation will |
1674 | * first attempt to defrag slab caches on other nodes. This means | |
1675 | * scanning over all nodes to look for partial slabs which may be | |
1676 | * expensive if we do it every time we are trying to find a slab | |
1677 | * with available objects. | |
81819f0f | 1678 | */ |
9824601e CL |
1679 | if (!s->remote_node_defrag_ratio || |
1680 | get_cycles() % 1024 > s->remote_node_defrag_ratio) | |
81819f0f CL |
1681 | return NULL; |
1682 | ||
cc9a6c87 | 1683 | do { |
d26914d1 | 1684 | cpuset_mems_cookie = read_mems_allowed_begin(); |
2a389610 | 1685 | zonelist = node_zonelist(mempolicy_slab_node(), flags); |
cc9a6c87 MG |
1686 | for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) { |
1687 | struct kmem_cache_node *n; | |
1688 | ||
1689 | n = get_node(s, zone_to_nid(zone)); | |
1690 | ||
dee2f8aa | 1691 | if (n && cpuset_zone_allowed(zone, flags) && |
cc9a6c87 | 1692 | n->nr_partial > s->min_partial) { |
8ba00bb6 | 1693 | object = get_partial_node(s, n, c, flags); |
cc9a6c87 MG |
1694 | if (object) { |
1695 | /* | |
d26914d1 MG |
1696 | * Don't check read_mems_allowed_retry() |
1697 | * here - if mems_allowed was updated in | |
1698 | * parallel, that was a harmless race | |
1699 | * between allocation and the cpuset | |
1700 | * update | |
cc9a6c87 | 1701 | */ |
cc9a6c87 MG |
1702 | return object; |
1703 | } | |
c0ff7453 | 1704 | } |
81819f0f | 1705 | } |
d26914d1 | 1706 | } while (read_mems_allowed_retry(cpuset_mems_cookie)); |
81819f0f CL |
1707 | #endif |
1708 | return NULL; | |
1709 | } | |
1710 | ||
1711 | /* | |
1712 | * Get a partial page, lock it and return it. | |
1713 | */ | |
497b66f2 | 1714 | static void *get_partial(struct kmem_cache *s, gfp_t flags, int node, |
acd19fd1 | 1715 | struct kmem_cache_cpu *c) |
81819f0f | 1716 | { |
497b66f2 | 1717 | void *object; |
a561ce00 JK |
1718 | int searchnode = node; |
1719 | ||
1720 | if (node == NUMA_NO_NODE) | |
1721 | searchnode = numa_mem_id(); | |
1722 | else if (!node_present_pages(node)) | |
1723 | searchnode = node_to_mem_node(node); | |
81819f0f | 1724 | |
8ba00bb6 | 1725 | object = get_partial_node(s, get_node(s, searchnode), c, flags); |
497b66f2 CL |
1726 | if (object || node != NUMA_NO_NODE) |
1727 | return object; | |
81819f0f | 1728 | |
acd19fd1 | 1729 | return get_any_partial(s, flags, c); |
81819f0f CL |
1730 | } |
1731 | ||
8a5ec0ba CL |
1732 | #ifdef CONFIG_PREEMPT |
1733 | /* | |
1734 | * Calculate the next globally unique transaction for disambiguiation | |
1735 | * during cmpxchg. The transactions start with the cpu number and are then | |
1736 | * incremented by CONFIG_NR_CPUS. | |
1737 | */ | |
1738 | #define TID_STEP roundup_pow_of_two(CONFIG_NR_CPUS) | |
1739 | #else | |
1740 | /* | |
1741 | * No preemption supported therefore also no need to check for | |
1742 | * different cpus. | |
1743 | */ | |
1744 | #define TID_STEP 1 | |
1745 | #endif | |
1746 | ||
1747 | static inline unsigned long next_tid(unsigned long tid) | |
1748 | { | |
1749 | return tid + TID_STEP; | |
1750 | } | |
1751 | ||
1752 | static inline unsigned int tid_to_cpu(unsigned long tid) | |
1753 | { | |
1754 | return tid % TID_STEP; | |
1755 | } | |
1756 | ||
1757 | static inline unsigned long tid_to_event(unsigned long tid) | |
1758 | { | |
1759 | return tid / TID_STEP; | |
1760 | } | |
1761 | ||
1762 | static inline unsigned int init_tid(int cpu) | |
1763 | { | |
1764 | return cpu; | |
1765 | } | |
1766 | ||
1767 | static inline void note_cmpxchg_failure(const char *n, | |
1768 | const struct kmem_cache *s, unsigned long tid) | |
1769 | { | |
1770 | #ifdef SLUB_DEBUG_CMPXCHG | |
1771 | unsigned long actual_tid = __this_cpu_read(s->cpu_slab->tid); | |
1772 | ||
f9f58285 | 1773 | pr_info("%s %s: cmpxchg redo ", n, s->name); |
8a5ec0ba CL |
1774 | |
1775 | #ifdef CONFIG_PREEMPT | |
1776 | if (tid_to_cpu(tid) != tid_to_cpu(actual_tid)) | |
f9f58285 | 1777 | pr_warn("due to cpu change %d -> %d\n", |
8a5ec0ba CL |
1778 | tid_to_cpu(tid), tid_to_cpu(actual_tid)); |
1779 | else | |
1780 | #endif | |
1781 | if (tid_to_event(tid) != tid_to_event(actual_tid)) | |
f9f58285 | 1782 | pr_warn("due to cpu running other code. Event %ld->%ld\n", |
8a5ec0ba CL |
1783 | tid_to_event(tid), tid_to_event(actual_tid)); |
1784 | else | |
f9f58285 | 1785 | pr_warn("for unknown reason: actual=%lx was=%lx target=%lx\n", |
8a5ec0ba CL |
1786 | actual_tid, tid, next_tid(tid)); |
1787 | #endif | |
4fdccdfb | 1788 | stat(s, CMPXCHG_DOUBLE_CPU_FAIL); |
8a5ec0ba CL |
1789 | } |
1790 | ||
788e1aad | 1791 | static void init_kmem_cache_cpus(struct kmem_cache *s) |
8a5ec0ba | 1792 | { |
8a5ec0ba CL |
1793 | int cpu; |
1794 | ||
1795 | for_each_possible_cpu(cpu) | |
1796 | per_cpu_ptr(s->cpu_slab, cpu)->tid = init_tid(cpu); | |
8a5ec0ba | 1797 | } |
2cfb7455 | 1798 | |
81819f0f CL |
1799 | /* |
1800 | * Remove the cpu slab | |
1801 | */ | |
d0e0ac97 CG |
1802 | static void deactivate_slab(struct kmem_cache *s, struct page *page, |
1803 | void *freelist) | |
81819f0f | 1804 | { |
2cfb7455 | 1805 | enum slab_modes { M_NONE, M_PARTIAL, M_FULL, M_FREE }; |
2cfb7455 CL |
1806 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); |
1807 | int lock = 0; | |
1808 | enum slab_modes l = M_NONE, m = M_NONE; | |
2cfb7455 | 1809 | void *nextfree; |
136333d1 | 1810 | int tail = DEACTIVATE_TO_HEAD; |
2cfb7455 CL |
1811 | struct page new; |
1812 | struct page old; | |
1813 | ||
1814 | if (page->freelist) { | |
84e554e6 | 1815 | stat(s, DEACTIVATE_REMOTE_FREES); |
136333d1 | 1816 | tail = DEACTIVATE_TO_TAIL; |
2cfb7455 CL |
1817 | } |
1818 | ||
894b8788 | 1819 | /* |
2cfb7455 CL |
1820 | * Stage one: Free all available per cpu objects back |
1821 | * to the page freelist while it is still frozen. Leave the | |
1822 | * last one. | |
1823 | * | |
1824 | * There is no need to take the list->lock because the page | |
1825 | * is still frozen. | |
1826 | */ | |
1827 | while (freelist && (nextfree = get_freepointer(s, freelist))) { | |
1828 | void *prior; | |
1829 | unsigned long counters; | |
1830 | ||
1831 | do { | |
1832 | prior = page->freelist; | |
1833 | counters = page->counters; | |
1834 | set_freepointer(s, freelist, prior); | |
1835 | new.counters = counters; | |
1836 | new.inuse--; | |
a0132ac0 | 1837 | VM_BUG_ON(!new.frozen); |
2cfb7455 | 1838 | |
1d07171c | 1839 | } while (!__cmpxchg_double_slab(s, page, |
2cfb7455 CL |
1840 | prior, counters, |
1841 | freelist, new.counters, | |
1842 | "drain percpu freelist")); | |
1843 | ||
1844 | freelist = nextfree; | |
1845 | } | |
1846 | ||
894b8788 | 1847 | /* |
2cfb7455 CL |
1848 | * Stage two: Ensure that the page is unfrozen while the |
1849 | * list presence reflects the actual number of objects | |
1850 | * during unfreeze. | |
1851 | * | |
1852 | * We setup the list membership and then perform a cmpxchg | |
1853 | * with the count. If there is a mismatch then the page | |
1854 | * is not unfrozen but the page is on the wrong list. | |
1855 | * | |
1856 | * Then we restart the process which may have to remove | |
1857 | * the page from the list that we just put it on again | |
1858 | * because the number of objects in the slab may have | |
1859 | * changed. | |
894b8788 | 1860 | */ |
2cfb7455 | 1861 | redo: |
894b8788 | 1862 | |
2cfb7455 CL |
1863 | old.freelist = page->freelist; |
1864 | old.counters = page->counters; | |
a0132ac0 | 1865 | VM_BUG_ON(!old.frozen); |
7c2e132c | 1866 | |
2cfb7455 CL |
1867 | /* Determine target state of the slab */ |
1868 | new.counters = old.counters; | |
1869 | if (freelist) { | |
1870 | new.inuse--; | |
1871 | set_freepointer(s, freelist, old.freelist); | |
1872 | new.freelist = freelist; | |
1873 | } else | |
1874 | new.freelist = old.freelist; | |
1875 | ||
1876 | new.frozen = 0; | |
1877 | ||
8a5b20ae | 1878 | if (!new.inuse && n->nr_partial >= s->min_partial) |
2cfb7455 CL |
1879 | m = M_FREE; |
1880 | else if (new.freelist) { | |
1881 | m = M_PARTIAL; | |
1882 | if (!lock) { | |
1883 | lock = 1; | |
1884 | /* | |
1885 | * Taking the spinlock removes the possiblity | |
1886 | * that acquire_slab() will see a slab page that | |
1887 | * is frozen | |
1888 | */ | |
1889 | spin_lock(&n->list_lock); | |
1890 | } | |
1891 | } else { | |
1892 | m = M_FULL; | |
1893 | if (kmem_cache_debug(s) && !lock) { | |
1894 | lock = 1; | |
1895 | /* | |
1896 | * This also ensures that the scanning of full | |
1897 | * slabs from diagnostic functions will not see | |
1898 | * any frozen slabs. | |
1899 | */ | |
1900 | spin_lock(&n->list_lock); | |
1901 | } | |
1902 | } | |
1903 | ||
1904 | if (l != m) { | |
1905 | ||
1906 | if (l == M_PARTIAL) | |
1907 | ||
1908 | remove_partial(n, page); | |
1909 | ||
1910 | else if (l == M_FULL) | |
894b8788 | 1911 | |
c65c1877 | 1912 | remove_full(s, n, page); |
2cfb7455 CL |
1913 | |
1914 | if (m == M_PARTIAL) { | |
1915 | ||
1916 | add_partial(n, page, tail); | |
136333d1 | 1917 | stat(s, tail); |
2cfb7455 CL |
1918 | |
1919 | } else if (m == M_FULL) { | |
894b8788 | 1920 | |
2cfb7455 CL |
1921 | stat(s, DEACTIVATE_FULL); |
1922 | add_full(s, n, page); | |
1923 | ||
1924 | } | |
1925 | } | |
1926 | ||
1927 | l = m; | |
1d07171c | 1928 | if (!__cmpxchg_double_slab(s, page, |
2cfb7455 CL |
1929 | old.freelist, old.counters, |
1930 | new.freelist, new.counters, | |
1931 | "unfreezing slab")) | |
1932 | goto redo; | |
1933 | ||
2cfb7455 CL |
1934 | if (lock) |
1935 | spin_unlock(&n->list_lock); | |
1936 | ||
1937 | if (m == M_FREE) { | |
1938 | stat(s, DEACTIVATE_EMPTY); | |
1939 | discard_slab(s, page); | |
1940 | stat(s, FREE_SLAB); | |
894b8788 | 1941 | } |
81819f0f CL |
1942 | } |
1943 | ||
d24ac77f JK |
1944 | /* |
1945 | * Unfreeze all the cpu partial slabs. | |
1946 | * | |
59a09917 CL |
1947 | * This function must be called with interrupts disabled |
1948 | * for the cpu using c (or some other guarantee must be there | |
1949 | * to guarantee no concurrent accesses). | |
d24ac77f | 1950 | */ |
59a09917 CL |
1951 | static void unfreeze_partials(struct kmem_cache *s, |
1952 | struct kmem_cache_cpu *c) | |
49e22585 | 1953 | { |
345c905d | 1954 | #ifdef CONFIG_SLUB_CPU_PARTIAL |
43d77867 | 1955 | struct kmem_cache_node *n = NULL, *n2 = NULL; |
9ada1934 | 1956 | struct page *page, *discard_page = NULL; |
49e22585 CL |
1957 | |
1958 | while ((page = c->partial)) { | |
49e22585 CL |
1959 | struct page new; |
1960 | struct page old; | |
1961 | ||
1962 | c->partial = page->next; | |
43d77867 JK |
1963 | |
1964 | n2 = get_node(s, page_to_nid(page)); | |
1965 | if (n != n2) { | |
1966 | if (n) | |
1967 | spin_unlock(&n->list_lock); | |
1968 | ||
1969 | n = n2; | |
1970 | spin_lock(&n->list_lock); | |
1971 | } | |
49e22585 CL |
1972 | |
1973 | do { | |
1974 | ||
1975 | old.freelist = page->freelist; | |
1976 | old.counters = page->counters; | |
a0132ac0 | 1977 | VM_BUG_ON(!old.frozen); |
49e22585 CL |
1978 | |
1979 | new.counters = old.counters; | |
1980 | new.freelist = old.freelist; | |
1981 | ||
1982 | new.frozen = 0; | |
1983 | ||
d24ac77f | 1984 | } while (!__cmpxchg_double_slab(s, page, |
49e22585 CL |
1985 | old.freelist, old.counters, |
1986 | new.freelist, new.counters, | |
1987 | "unfreezing slab")); | |
1988 | ||
8a5b20ae | 1989 | if (unlikely(!new.inuse && n->nr_partial >= s->min_partial)) { |
9ada1934 SL |
1990 | page->next = discard_page; |
1991 | discard_page = page; | |
43d77867 JK |
1992 | } else { |
1993 | add_partial(n, page, DEACTIVATE_TO_TAIL); | |
1994 | stat(s, FREE_ADD_PARTIAL); | |
49e22585 CL |
1995 | } |
1996 | } | |
1997 | ||
1998 | if (n) | |
1999 | spin_unlock(&n->list_lock); | |
9ada1934 SL |
2000 | |
2001 | while (discard_page) { | |
2002 | page = discard_page; | |
2003 | discard_page = discard_page->next; | |
2004 | ||
2005 | stat(s, DEACTIVATE_EMPTY); | |
2006 | discard_slab(s, page); | |
2007 | stat(s, FREE_SLAB); | |
2008 | } | |
345c905d | 2009 | #endif |
49e22585 CL |
2010 | } |
2011 | ||
2012 | /* | |
2013 | * Put a page that was just frozen (in __slab_free) into a partial page | |
2014 | * slot if available. This is done without interrupts disabled and without | |
2015 | * preemption disabled. The cmpxchg is racy and may put the partial page | |
2016 | * onto a random cpus partial slot. | |
2017 | * | |
2018 | * If we did not find a slot then simply move all the partials to the | |
2019 | * per node partial list. | |
2020 | */ | |
633b0764 | 2021 | static void put_cpu_partial(struct kmem_cache *s, struct page *page, int drain) |
49e22585 | 2022 | { |
345c905d | 2023 | #ifdef CONFIG_SLUB_CPU_PARTIAL |
49e22585 CL |
2024 | struct page *oldpage; |
2025 | int pages; | |
2026 | int pobjects; | |
2027 | ||
d6e0b7fa | 2028 | preempt_disable(); |
49e22585 CL |
2029 | do { |
2030 | pages = 0; | |
2031 | pobjects = 0; | |
2032 | oldpage = this_cpu_read(s->cpu_slab->partial); | |
2033 | ||
2034 | if (oldpage) { | |
2035 | pobjects = oldpage->pobjects; | |
2036 | pages = oldpage->pages; | |
2037 | if (drain && pobjects > s->cpu_partial) { | |
2038 | unsigned long flags; | |
2039 | /* | |
2040 | * partial array is full. Move the existing | |
2041 | * set to the per node partial list. | |
2042 | */ | |
2043 | local_irq_save(flags); | |
59a09917 | 2044 | unfreeze_partials(s, this_cpu_ptr(s->cpu_slab)); |
49e22585 | 2045 | local_irq_restore(flags); |
e24fc410 | 2046 | oldpage = NULL; |
49e22585 CL |
2047 | pobjects = 0; |
2048 | pages = 0; | |
8028dcea | 2049 | stat(s, CPU_PARTIAL_DRAIN); |
49e22585 CL |
2050 | } |
2051 | } | |
2052 | ||
2053 | pages++; | |
2054 | pobjects += page->objects - page->inuse; | |
2055 | ||
2056 | page->pages = pages; | |
2057 | page->pobjects = pobjects; | |
2058 | page->next = oldpage; | |
2059 | ||
d0e0ac97 CG |
2060 | } while (this_cpu_cmpxchg(s->cpu_slab->partial, oldpage, page) |
2061 | != oldpage); | |
d6e0b7fa VD |
2062 | if (unlikely(!s->cpu_partial)) { |
2063 | unsigned long flags; | |
2064 | ||
2065 | local_irq_save(flags); | |
2066 | unfreeze_partials(s, this_cpu_ptr(s->cpu_slab)); | |
2067 | local_irq_restore(flags); | |
2068 | } | |
2069 | preempt_enable(); | |
345c905d | 2070 | #endif |
49e22585 CL |
2071 | } |
2072 | ||
dfb4f096 | 2073 | static inline void flush_slab(struct kmem_cache *s, struct kmem_cache_cpu *c) |
81819f0f | 2074 | { |
84e554e6 | 2075 | stat(s, CPUSLAB_FLUSH); |
c17dda40 CL |
2076 | deactivate_slab(s, c->page, c->freelist); |
2077 | ||
2078 | c->tid = next_tid(c->tid); | |
2079 | c->page = NULL; | |
2080 | c->freelist = NULL; | |
81819f0f CL |
2081 | } |
2082 | ||
2083 | /* | |
2084 | * Flush cpu slab. | |
6446faa2 | 2085 | * |
81819f0f CL |
2086 | * Called from IPI handler with interrupts disabled. |
2087 | */ | |
0c710013 | 2088 | static inline void __flush_cpu_slab(struct kmem_cache *s, int cpu) |
81819f0f | 2089 | { |
9dfc6e68 | 2090 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu); |
81819f0f | 2091 | |
49e22585 CL |
2092 | if (likely(c)) { |
2093 | if (c->page) | |
2094 | flush_slab(s, c); | |
2095 | ||
59a09917 | 2096 | unfreeze_partials(s, c); |
49e22585 | 2097 | } |
81819f0f CL |
2098 | } |
2099 | ||
2100 | static void flush_cpu_slab(void *d) | |
2101 | { | |
2102 | struct kmem_cache *s = d; | |
81819f0f | 2103 | |
dfb4f096 | 2104 | __flush_cpu_slab(s, smp_processor_id()); |
81819f0f CL |
2105 | } |
2106 | ||
a8364d55 GBY |
2107 | static bool has_cpu_slab(int cpu, void *info) |
2108 | { | |
2109 | struct kmem_cache *s = info; | |
2110 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu); | |
2111 | ||
02e1a9cd | 2112 | return c->page || c->partial; |
a8364d55 GBY |
2113 | } |
2114 | ||
81819f0f CL |
2115 | static void flush_all(struct kmem_cache *s) |
2116 | { | |
a8364d55 | 2117 | on_each_cpu_cond(has_cpu_slab, flush_cpu_slab, s, 1, GFP_ATOMIC); |
81819f0f CL |
2118 | } |
2119 | ||
dfb4f096 CL |
2120 | /* |
2121 | * Check if the objects in a per cpu structure fit numa | |
2122 | * locality expectations. | |
2123 | */ | |
57d437d2 | 2124 | static inline int node_match(struct page *page, int node) |
dfb4f096 CL |
2125 | { |
2126 | #ifdef CONFIG_NUMA | |
4d7868e6 | 2127 | if (!page || (node != NUMA_NO_NODE && page_to_nid(page) != node)) |
dfb4f096 CL |
2128 | return 0; |
2129 | #endif | |
2130 | return 1; | |
2131 | } | |
2132 | ||
9a02d699 | 2133 | #ifdef CONFIG_SLUB_DEBUG |
781b2ba6 PE |
2134 | static int count_free(struct page *page) |
2135 | { | |
2136 | return page->objects - page->inuse; | |
2137 | } | |
2138 | ||
9a02d699 DR |
2139 | static inline unsigned long node_nr_objs(struct kmem_cache_node *n) |
2140 | { | |
2141 | return atomic_long_read(&n->total_objects); | |
2142 | } | |
2143 | #endif /* CONFIG_SLUB_DEBUG */ | |
2144 | ||
2145 | #if defined(CONFIG_SLUB_DEBUG) || defined(CONFIG_SYSFS) | |
781b2ba6 PE |
2146 | static unsigned long count_partial(struct kmem_cache_node *n, |
2147 | int (*get_count)(struct page *)) | |
2148 | { | |
2149 | unsigned long flags; | |
2150 | unsigned long x = 0; | |
2151 | struct page *page; | |
2152 | ||
2153 | spin_lock_irqsave(&n->list_lock, flags); | |
2154 | list_for_each_entry(page, &n->partial, lru) | |
2155 | x += get_count(page); | |
2156 | spin_unlock_irqrestore(&n->list_lock, flags); | |
2157 | return x; | |
2158 | } | |
9a02d699 | 2159 | #endif /* CONFIG_SLUB_DEBUG || CONFIG_SYSFS */ |
26c02cf0 | 2160 | |
781b2ba6 PE |
2161 | static noinline void |
2162 | slab_out_of_memory(struct kmem_cache *s, gfp_t gfpflags, int nid) | |
2163 | { | |
9a02d699 DR |
2164 | #ifdef CONFIG_SLUB_DEBUG |
2165 | static DEFINE_RATELIMIT_STATE(slub_oom_rs, DEFAULT_RATELIMIT_INTERVAL, | |
2166 | DEFAULT_RATELIMIT_BURST); | |
781b2ba6 | 2167 | int node; |
fa45dc25 | 2168 | struct kmem_cache_node *n; |
781b2ba6 | 2169 | |
9a02d699 DR |
2170 | if ((gfpflags & __GFP_NOWARN) || !__ratelimit(&slub_oom_rs)) |
2171 | return; | |
2172 | ||
f9f58285 | 2173 | pr_warn("SLUB: Unable to allocate memory on node %d (gfp=0x%x)\n", |
781b2ba6 | 2174 | nid, gfpflags); |
f9f58285 FF |
2175 | pr_warn(" cache: %s, object size: %d, buffer size: %d, default order: %d, min order: %d\n", |
2176 | s->name, s->object_size, s->size, oo_order(s->oo), | |
2177 | oo_order(s->min)); | |
781b2ba6 | 2178 | |
3b0efdfa | 2179 | if (oo_order(s->min) > get_order(s->object_size)) |
f9f58285 FF |
2180 | pr_warn(" %s debugging increased min order, use slub_debug=O to disable.\n", |
2181 | s->name); | |
fa5ec8a1 | 2182 | |
fa45dc25 | 2183 | for_each_kmem_cache_node(s, node, n) { |
781b2ba6 PE |
2184 | unsigned long nr_slabs; |
2185 | unsigned long nr_objs; | |
2186 | unsigned long nr_free; | |
2187 | ||
26c02cf0 AB |
2188 | nr_free = count_partial(n, count_free); |
2189 | nr_slabs = node_nr_slabs(n); | |
2190 | nr_objs = node_nr_objs(n); | |
781b2ba6 | 2191 | |
f9f58285 | 2192 | pr_warn(" node %d: slabs: %ld, objs: %ld, free: %ld\n", |
781b2ba6 PE |
2193 | node, nr_slabs, nr_objs, nr_free); |
2194 | } | |
9a02d699 | 2195 | #endif |
781b2ba6 PE |
2196 | } |
2197 | ||
497b66f2 CL |
2198 | static inline void *new_slab_objects(struct kmem_cache *s, gfp_t flags, |
2199 | int node, struct kmem_cache_cpu **pc) | |
2200 | { | |
6faa6833 | 2201 | void *freelist; |
188fd063 CL |
2202 | struct kmem_cache_cpu *c = *pc; |
2203 | struct page *page; | |
497b66f2 | 2204 | |
188fd063 | 2205 | freelist = get_partial(s, flags, node, c); |
497b66f2 | 2206 | |
188fd063 CL |
2207 | if (freelist) |
2208 | return freelist; | |
2209 | ||
2210 | page = new_slab(s, flags, node); | |
497b66f2 | 2211 | if (page) { |
7c8e0181 | 2212 | c = raw_cpu_ptr(s->cpu_slab); |
497b66f2 CL |
2213 | if (c->page) |
2214 | flush_slab(s, c); | |
2215 | ||
2216 | /* | |
2217 | * No other reference to the page yet so we can | |
2218 | * muck around with it freely without cmpxchg | |
2219 | */ | |
6faa6833 | 2220 | freelist = page->freelist; |
497b66f2 CL |
2221 | page->freelist = NULL; |
2222 | ||
2223 | stat(s, ALLOC_SLAB); | |
497b66f2 CL |
2224 | c->page = page; |
2225 | *pc = c; | |
2226 | } else | |
6faa6833 | 2227 | freelist = NULL; |
497b66f2 | 2228 | |
6faa6833 | 2229 | return freelist; |
497b66f2 CL |
2230 | } |
2231 | ||
072bb0aa MG |
2232 | static inline bool pfmemalloc_match(struct page *page, gfp_t gfpflags) |
2233 | { | |
2234 | if (unlikely(PageSlabPfmemalloc(page))) | |
2235 | return gfp_pfmemalloc_allowed(gfpflags); | |
2236 | ||
2237 | return true; | |
2238 | } | |
2239 | ||
213eeb9f | 2240 | /* |
d0e0ac97 CG |
2241 | * Check the page->freelist of a page and either transfer the freelist to the |
2242 | * per cpu freelist or deactivate the page. | |
213eeb9f CL |
2243 | * |
2244 | * The page is still frozen if the return value is not NULL. | |
2245 | * | |
2246 | * If this function returns NULL then the page has been unfrozen. | |
d24ac77f JK |
2247 | * |
2248 | * This function must be called with interrupt disabled. | |
213eeb9f CL |
2249 | */ |
2250 | static inline void *get_freelist(struct kmem_cache *s, struct page *page) | |
2251 | { | |
2252 | struct page new; | |
2253 | unsigned long counters; | |
2254 | void *freelist; | |
2255 | ||
2256 | do { | |
2257 | freelist = page->freelist; | |
2258 | counters = page->counters; | |
6faa6833 | 2259 | |
213eeb9f | 2260 | new.counters = counters; |
a0132ac0 | 2261 | VM_BUG_ON(!new.frozen); |
213eeb9f CL |
2262 | |
2263 | new.inuse = page->objects; | |
2264 | new.frozen = freelist != NULL; | |
2265 | ||
d24ac77f | 2266 | } while (!__cmpxchg_double_slab(s, page, |
213eeb9f CL |
2267 | freelist, counters, |
2268 | NULL, new.counters, | |
2269 | "get_freelist")); | |
2270 | ||
2271 | return freelist; | |
2272 | } | |
2273 | ||
81819f0f | 2274 | /* |
894b8788 CL |
2275 | * Slow path. The lockless freelist is empty or we need to perform |
2276 | * debugging duties. | |
2277 | * | |
894b8788 CL |
2278 | * Processing is still very fast if new objects have been freed to the |
2279 | * regular freelist. In that case we simply take over the regular freelist | |
2280 | * as the lockless freelist and zap the regular freelist. | |
81819f0f | 2281 | * |
894b8788 CL |
2282 | * If that is not working then we fall back to the partial lists. We take the |
2283 | * first element of the freelist as the object to allocate now and move the | |
2284 | * rest of the freelist to the lockless freelist. | |
81819f0f | 2285 | * |
894b8788 | 2286 | * And if we were unable to get a new slab from the partial slab lists then |
6446faa2 CL |
2287 | * we need to allocate a new slab. This is the slowest path since it involves |
2288 | * a call to the page allocator and the setup of a new slab. | |
a380a3c7 CL |
2289 | * |
2290 | * Version of __slab_alloc to use when we know that interrupts are | |
2291 | * already disabled (which is the case for bulk allocation). | |
81819f0f | 2292 | */ |
a380a3c7 | 2293 | static void *___slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node, |
ce71e27c | 2294 | unsigned long addr, struct kmem_cache_cpu *c) |
81819f0f | 2295 | { |
6faa6833 | 2296 | void *freelist; |
f6e7def7 | 2297 | struct page *page; |
81819f0f | 2298 | |
f6e7def7 CL |
2299 | page = c->page; |
2300 | if (!page) | |
81819f0f | 2301 | goto new_slab; |
49e22585 | 2302 | redo: |
6faa6833 | 2303 | |
57d437d2 | 2304 | if (unlikely(!node_match(page, node))) { |
a561ce00 JK |
2305 | int searchnode = node; |
2306 | ||
2307 | if (node != NUMA_NO_NODE && !node_present_pages(node)) | |
2308 | searchnode = node_to_mem_node(node); | |
2309 | ||
2310 | if (unlikely(!node_match(page, searchnode))) { | |
2311 | stat(s, ALLOC_NODE_MISMATCH); | |
2312 | deactivate_slab(s, page, c->freelist); | |
2313 | c->page = NULL; | |
2314 | c->freelist = NULL; | |
2315 | goto new_slab; | |
2316 | } | |
fc59c053 | 2317 | } |
6446faa2 | 2318 | |
072bb0aa MG |
2319 | /* |
2320 | * By rights, we should be searching for a slab page that was | |
2321 | * PFMEMALLOC but right now, we are losing the pfmemalloc | |
2322 | * information when the page leaves the per-cpu allocator | |
2323 | */ | |
2324 | if (unlikely(!pfmemalloc_match(page, gfpflags))) { | |
2325 | deactivate_slab(s, page, c->freelist); | |
2326 | c->page = NULL; | |
2327 | c->freelist = NULL; | |
2328 | goto new_slab; | |
2329 | } | |
2330 | ||
73736e03 | 2331 | /* must check again c->freelist in case of cpu migration or IRQ */ |
6faa6833 CL |
2332 | freelist = c->freelist; |
2333 | if (freelist) | |
73736e03 | 2334 | goto load_freelist; |
03e404af | 2335 | |
f6e7def7 | 2336 | freelist = get_freelist(s, page); |
6446faa2 | 2337 | |
6faa6833 | 2338 | if (!freelist) { |
03e404af CL |
2339 | c->page = NULL; |
2340 | stat(s, DEACTIVATE_BYPASS); | |
fc59c053 | 2341 | goto new_slab; |
03e404af | 2342 | } |
6446faa2 | 2343 | |
84e554e6 | 2344 | stat(s, ALLOC_REFILL); |
6446faa2 | 2345 | |
894b8788 | 2346 | load_freelist: |
507effea CL |
2347 | /* |
2348 | * freelist is pointing to the list of objects to be used. | |
2349 | * page is pointing to the page from which the objects are obtained. | |
2350 | * That page must be frozen for per cpu allocations to work. | |
2351 | */ | |
a0132ac0 | 2352 | VM_BUG_ON(!c->page->frozen); |
6faa6833 | 2353 | c->freelist = get_freepointer(s, freelist); |
8a5ec0ba | 2354 | c->tid = next_tid(c->tid); |
6faa6833 | 2355 | return freelist; |
81819f0f | 2356 | |
81819f0f | 2357 | new_slab: |
2cfb7455 | 2358 | |
49e22585 | 2359 | if (c->partial) { |
f6e7def7 CL |
2360 | page = c->page = c->partial; |
2361 | c->partial = page->next; | |
49e22585 CL |
2362 | stat(s, CPU_PARTIAL_ALLOC); |
2363 | c->freelist = NULL; | |
2364 | goto redo; | |
81819f0f CL |
2365 | } |
2366 | ||
188fd063 | 2367 | freelist = new_slab_objects(s, gfpflags, node, &c); |
01ad8a7b | 2368 | |
f4697436 | 2369 | if (unlikely(!freelist)) { |
9a02d699 | 2370 | slab_out_of_memory(s, gfpflags, node); |
f4697436 | 2371 | return NULL; |
81819f0f | 2372 | } |
2cfb7455 | 2373 | |
f6e7def7 | 2374 | page = c->page; |
5091b74a | 2375 | if (likely(!kmem_cache_debug(s) && pfmemalloc_match(page, gfpflags))) |
4b6f0750 | 2376 | goto load_freelist; |
2cfb7455 | 2377 | |
497b66f2 | 2378 | /* Only entered in the debug case */ |
d0e0ac97 CG |
2379 | if (kmem_cache_debug(s) && |
2380 | !alloc_debug_processing(s, page, freelist, addr)) | |
497b66f2 | 2381 | goto new_slab; /* Slab failed checks. Next slab needed */ |
894b8788 | 2382 | |
f6e7def7 | 2383 | deactivate_slab(s, page, get_freepointer(s, freelist)); |
c17dda40 CL |
2384 | c->page = NULL; |
2385 | c->freelist = NULL; | |
6faa6833 | 2386 | return freelist; |
894b8788 CL |
2387 | } |
2388 | ||
a380a3c7 CL |
2389 | /* |
2390 | * Another one that disabled interrupt and compensates for possible | |
2391 | * cpu changes by refetching the per cpu area pointer. | |
2392 | */ | |
2393 | static void *__slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node, | |
2394 | unsigned long addr, struct kmem_cache_cpu *c) | |
2395 | { | |
2396 | void *p; | |
2397 | unsigned long flags; | |
2398 | ||
2399 | local_irq_save(flags); | |
2400 | #ifdef CONFIG_PREEMPT | |
2401 | /* | |
2402 | * We may have been preempted and rescheduled on a different | |
2403 | * cpu before disabling interrupts. Need to reload cpu area | |
2404 | * pointer. | |
2405 | */ | |
2406 | c = this_cpu_ptr(s->cpu_slab); | |
2407 | #endif | |
2408 | ||
2409 | p = ___slab_alloc(s, gfpflags, node, addr, c); | |
2410 | local_irq_restore(flags); | |
2411 | return p; | |
2412 | } | |
2413 | ||
894b8788 CL |
2414 | /* |
2415 | * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc) | |
2416 | * have the fastpath folded into their functions. So no function call | |
2417 | * overhead for requests that can be satisfied on the fastpath. | |
2418 | * | |
2419 | * The fastpath works by first checking if the lockless freelist can be used. | |
2420 | * If not then __slab_alloc is called for slow processing. | |
2421 | * | |
2422 | * Otherwise we can simply pick the next object from the lockless free list. | |
2423 | */ | |
2b847c3c | 2424 | static __always_inline void *slab_alloc_node(struct kmem_cache *s, |
ce71e27c | 2425 | gfp_t gfpflags, int node, unsigned long addr) |
894b8788 | 2426 | { |
03ec0ed5 | 2427 | void *object; |
dfb4f096 | 2428 | struct kmem_cache_cpu *c; |
57d437d2 | 2429 | struct page *page; |
8a5ec0ba | 2430 | unsigned long tid; |
1f84260c | 2431 | |
8135be5a VD |
2432 | s = slab_pre_alloc_hook(s, gfpflags); |
2433 | if (!s) | |
773ff60e | 2434 | return NULL; |
8a5ec0ba | 2435 | redo: |
8a5ec0ba CL |
2436 | /* |
2437 | * Must read kmem_cache cpu data via this cpu ptr. Preemption is | |
2438 | * enabled. We may switch back and forth between cpus while | |
2439 | * reading from one cpu area. That does not matter as long | |
2440 | * as we end up on the original cpu again when doing the cmpxchg. | |
7cccd80b | 2441 | * |
9aabf810 JK |
2442 | * We should guarantee that tid and kmem_cache are retrieved on |
2443 | * the same cpu. It could be different if CONFIG_PREEMPT so we need | |
2444 | * to check if it is matched or not. | |
8a5ec0ba | 2445 | */ |
9aabf810 JK |
2446 | do { |
2447 | tid = this_cpu_read(s->cpu_slab->tid); | |
2448 | c = raw_cpu_ptr(s->cpu_slab); | |
859b7a0e MR |
2449 | } while (IS_ENABLED(CONFIG_PREEMPT) && |
2450 | unlikely(tid != READ_ONCE(c->tid))); | |
9aabf810 JK |
2451 | |
2452 | /* | |
2453 | * Irqless object alloc/free algorithm used here depends on sequence | |
2454 | * of fetching cpu_slab's data. tid should be fetched before anything | |
2455 | * on c to guarantee that object and page associated with previous tid | |
2456 | * won't be used with current tid. If we fetch tid first, object and | |
2457 | * page could be one associated with next tid and our alloc/free | |
2458 | * request will be failed. In this case, we will retry. So, no problem. | |
2459 | */ | |
2460 | barrier(); | |
8a5ec0ba | 2461 | |
8a5ec0ba CL |
2462 | /* |
2463 | * The transaction ids are globally unique per cpu and per operation on | |
2464 | * a per cpu queue. Thus they can be guarantee that the cmpxchg_double | |
2465 | * occurs on the right processor and that there was no operation on the | |
2466 | * linked list in between. | |
2467 | */ | |
8a5ec0ba | 2468 | |
9dfc6e68 | 2469 | object = c->freelist; |
57d437d2 | 2470 | page = c->page; |
8eae1492 | 2471 | if (unlikely(!object || !node_match(page, node))) { |
dfb4f096 | 2472 | object = __slab_alloc(s, gfpflags, node, addr, c); |
8eae1492 DH |
2473 | stat(s, ALLOC_SLOWPATH); |
2474 | } else { | |
0ad9500e ED |
2475 | void *next_object = get_freepointer_safe(s, object); |
2476 | ||
8a5ec0ba | 2477 | /* |
25985edc | 2478 | * The cmpxchg will only match if there was no additional |
8a5ec0ba CL |
2479 | * operation and if we are on the right processor. |
2480 | * | |
d0e0ac97 CG |
2481 | * The cmpxchg does the following atomically (without lock |
2482 | * semantics!) | |
8a5ec0ba CL |
2483 | * 1. Relocate first pointer to the current per cpu area. |
2484 | * 2. Verify that tid and freelist have not been changed | |
2485 | * 3. If they were not changed replace tid and freelist | |
2486 | * | |
d0e0ac97 CG |
2487 | * Since this is without lock semantics the protection is only |
2488 | * against code executing on this cpu *not* from access by | |
2489 | * other cpus. | |
8a5ec0ba | 2490 | */ |
933393f5 | 2491 | if (unlikely(!this_cpu_cmpxchg_double( |
8a5ec0ba CL |
2492 | s->cpu_slab->freelist, s->cpu_slab->tid, |
2493 | object, tid, | |
0ad9500e | 2494 | next_object, next_tid(tid)))) { |
8a5ec0ba CL |
2495 | |
2496 | note_cmpxchg_failure("slab_alloc", s, tid); | |
2497 | goto redo; | |
2498 | } | |
0ad9500e | 2499 | prefetch_freepointer(s, next_object); |
84e554e6 | 2500 | stat(s, ALLOC_FASTPATH); |
894b8788 | 2501 | } |
8a5ec0ba | 2502 | |
74e2134f | 2503 | if (unlikely(gfpflags & __GFP_ZERO) && object) |
3b0efdfa | 2504 | memset(object, 0, s->object_size); |
d07dbea4 | 2505 | |
03ec0ed5 | 2506 | slab_post_alloc_hook(s, gfpflags, 1, &object); |
5a896d9e | 2507 | |
894b8788 | 2508 | return object; |
81819f0f CL |
2509 | } |
2510 | ||
2b847c3c EG |
2511 | static __always_inline void *slab_alloc(struct kmem_cache *s, |
2512 | gfp_t gfpflags, unsigned long addr) | |
2513 | { | |
2514 | return slab_alloc_node(s, gfpflags, NUMA_NO_NODE, addr); | |
2515 | } | |
2516 | ||
81819f0f CL |
2517 | void *kmem_cache_alloc(struct kmem_cache *s, gfp_t gfpflags) |
2518 | { | |
2b847c3c | 2519 | void *ret = slab_alloc(s, gfpflags, _RET_IP_); |
5b882be4 | 2520 | |
d0e0ac97 CG |
2521 | trace_kmem_cache_alloc(_RET_IP_, ret, s->object_size, |
2522 | s->size, gfpflags); | |
5b882be4 EGM |
2523 | |
2524 | return ret; | |
81819f0f CL |
2525 | } |
2526 | EXPORT_SYMBOL(kmem_cache_alloc); | |
2527 | ||
0f24f128 | 2528 | #ifdef CONFIG_TRACING |
4a92379b RK |
2529 | void *kmem_cache_alloc_trace(struct kmem_cache *s, gfp_t gfpflags, size_t size) |
2530 | { | |
2b847c3c | 2531 | void *ret = slab_alloc(s, gfpflags, _RET_IP_); |
4a92379b | 2532 | trace_kmalloc(_RET_IP_, ret, size, s->size, gfpflags); |
0316bec2 | 2533 | kasan_kmalloc(s, ret, size); |
4a92379b RK |
2534 | return ret; |
2535 | } | |
2536 | EXPORT_SYMBOL(kmem_cache_alloc_trace); | |
5b882be4 EGM |
2537 | #endif |
2538 | ||
81819f0f CL |
2539 | #ifdef CONFIG_NUMA |
2540 | void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t gfpflags, int node) | |
2541 | { | |
2b847c3c | 2542 | void *ret = slab_alloc_node(s, gfpflags, node, _RET_IP_); |
5b882be4 | 2543 | |
ca2b84cb | 2544 | trace_kmem_cache_alloc_node(_RET_IP_, ret, |
3b0efdfa | 2545 | s->object_size, s->size, gfpflags, node); |
5b882be4 EGM |
2546 | |
2547 | return ret; | |
81819f0f CL |
2548 | } |
2549 | EXPORT_SYMBOL(kmem_cache_alloc_node); | |
81819f0f | 2550 | |
0f24f128 | 2551 | #ifdef CONFIG_TRACING |
4a92379b | 2552 | void *kmem_cache_alloc_node_trace(struct kmem_cache *s, |
5b882be4 | 2553 | gfp_t gfpflags, |
4a92379b | 2554 | int node, size_t size) |
5b882be4 | 2555 | { |
2b847c3c | 2556 | void *ret = slab_alloc_node(s, gfpflags, node, _RET_IP_); |
4a92379b RK |
2557 | |
2558 | trace_kmalloc_node(_RET_IP_, ret, | |
2559 | size, s->size, gfpflags, node); | |
0316bec2 AR |
2560 | |
2561 | kasan_kmalloc(s, ret, size); | |
4a92379b | 2562 | return ret; |
5b882be4 | 2563 | } |
4a92379b | 2564 | EXPORT_SYMBOL(kmem_cache_alloc_node_trace); |
5b882be4 | 2565 | #endif |
5d1f57e4 | 2566 | #endif |
5b882be4 | 2567 | |
81819f0f | 2568 | /* |
94e4d712 | 2569 | * Slow path handling. This may still be called frequently since objects |
894b8788 | 2570 | * have a longer lifetime than the cpu slabs in most processing loads. |
81819f0f | 2571 | * |
894b8788 CL |
2572 | * So we still attempt to reduce cache line usage. Just take the slab |
2573 | * lock and free the item. If there is no additional partial page | |
2574 | * handling required then we can return immediately. | |
81819f0f | 2575 | */ |
894b8788 | 2576 | static void __slab_free(struct kmem_cache *s, struct page *page, |
81084651 JDB |
2577 | void *head, void *tail, int cnt, |
2578 | unsigned long addr) | |
2579 | ||
81819f0f CL |
2580 | { |
2581 | void *prior; | |
2cfb7455 | 2582 | int was_frozen; |
2cfb7455 CL |
2583 | struct page new; |
2584 | unsigned long counters; | |
2585 | struct kmem_cache_node *n = NULL; | |
61728d1e | 2586 | unsigned long uninitialized_var(flags); |
81819f0f | 2587 | |
8a5ec0ba | 2588 | stat(s, FREE_SLOWPATH); |
81819f0f | 2589 | |
19c7ff9e | 2590 | if (kmem_cache_debug(s) && |
81084651 JDB |
2591 | !(n = free_debug_processing(s, page, head, tail, cnt, |
2592 | addr, &flags))) | |
80f08c19 | 2593 | return; |
6446faa2 | 2594 | |
2cfb7455 | 2595 | do { |
837d678d JK |
2596 | if (unlikely(n)) { |
2597 | spin_unlock_irqrestore(&n->list_lock, flags); | |
2598 | n = NULL; | |
2599 | } | |
2cfb7455 CL |
2600 | prior = page->freelist; |
2601 | counters = page->counters; | |
81084651 | 2602 | set_freepointer(s, tail, prior); |
2cfb7455 CL |
2603 | new.counters = counters; |
2604 | was_frozen = new.frozen; | |
81084651 | 2605 | new.inuse -= cnt; |
837d678d | 2606 | if ((!new.inuse || !prior) && !was_frozen) { |
49e22585 | 2607 | |
c65c1877 | 2608 | if (kmem_cache_has_cpu_partial(s) && !prior) { |
49e22585 CL |
2609 | |
2610 | /* | |
d0e0ac97 CG |
2611 | * Slab was on no list before and will be |
2612 | * partially empty | |
2613 | * We can defer the list move and instead | |
2614 | * freeze it. | |
49e22585 CL |
2615 | */ |
2616 | new.frozen = 1; | |
2617 | ||
c65c1877 | 2618 | } else { /* Needs to be taken off a list */ |
49e22585 | 2619 | |
b455def2 | 2620 | n = get_node(s, page_to_nid(page)); |
49e22585 CL |
2621 | /* |
2622 | * Speculatively acquire the list_lock. | |
2623 | * If the cmpxchg does not succeed then we may | |
2624 | * drop the list_lock without any processing. | |
2625 | * | |
2626 | * Otherwise the list_lock will synchronize with | |
2627 | * other processors updating the list of slabs. | |
2628 | */ | |
2629 | spin_lock_irqsave(&n->list_lock, flags); | |
2630 | ||
2631 | } | |
2cfb7455 | 2632 | } |
81819f0f | 2633 | |
2cfb7455 CL |
2634 | } while (!cmpxchg_double_slab(s, page, |
2635 | prior, counters, | |
81084651 | 2636 | head, new.counters, |
2cfb7455 | 2637 | "__slab_free")); |
81819f0f | 2638 | |
2cfb7455 | 2639 | if (likely(!n)) { |
49e22585 CL |
2640 | |
2641 | /* | |
2642 | * If we just froze the page then put it onto the | |
2643 | * per cpu partial list. | |
2644 | */ | |
8028dcea | 2645 | if (new.frozen && !was_frozen) { |
49e22585 | 2646 | put_cpu_partial(s, page, 1); |
8028dcea AS |
2647 | stat(s, CPU_PARTIAL_FREE); |
2648 | } | |
49e22585 | 2649 | /* |
2cfb7455 CL |
2650 | * The list lock was not taken therefore no list |
2651 | * activity can be necessary. | |
2652 | */ | |
b455def2 L |
2653 | if (was_frozen) |
2654 | stat(s, FREE_FROZEN); | |
2655 | return; | |
2656 | } | |
81819f0f | 2657 | |
8a5b20ae | 2658 | if (unlikely(!new.inuse && n->nr_partial >= s->min_partial)) |
837d678d JK |
2659 | goto slab_empty; |
2660 | ||
81819f0f | 2661 | /* |
837d678d JK |
2662 | * Objects left in the slab. If it was not on the partial list before |
2663 | * then add it. | |
81819f0f | 2664 | */ |
345c905d JK |
2665 | if (!kmem_cache_has_cpu_partial(s) && unlikely(!prior)) { |
2666 | if (kmem_cache_debug(s)) | |
c65c1877 | 2667 | remove_full(s, n, page); |
837d678d JK |
2668 | add_partial(n, page, DEACTIVATE_TO_TAIL); |
2669 | stat(s, FREE_ADD_PARTIAL); | |
8ff12cfc | 2670 | } |
80f08c19 | 2671 | spin_unlock_irqrestore(&n->list_lock, flags); |
81819f0f CL |
2672 | return; |
2673 | ||
2674 | slab_empty: | |
a973e9dd | 2675 | if (prior) { |
81819f0f | 2676 | /* |
6fbabb20 | 2677 | * Slab on the partial list. |
81819f0f | 2678 | */ |
5cc6eee8 | 2679 | remove_partial(n, page); |
84e554e6 | 2680 | stat(s, FREE_REMOVE_PARTIAL); |
c65c1877 | 2681 | } else { |
6fbabb20 | 2682 | /* Slab must be on the full list */ |
c65c1877 PZ |
2683 | remove_full(s, n, page); |
2684 | } | |
2cfb7455 | 2685 | |
80f08c19 | 2686 | spin_unlock_irqrestore(&n->list_lock, flags); |
84e554e6 | 2687 | stat(s, FREE_SLAB); |
81819f0f | 2688 | discard_slab(s, page); |
81819f0f CL |
2689 | } |
2690 | ||
894b8788 CL |
2691 | /* |
2692 | * Fastpath with forced inlining to produce a kfree and kmem_cache_free that | |
2693 | * can perform fastpath freeing without additional function calls. | |
2694 | * | |
2695 | * The fastpath is only possible if we are freeing to the current cpu slab | |
2696 | * of this processor. This typically the case if we have just allocated | |
2697 | * the item before. | |
2698 | * | |
2699 | * If fastpath is not possible then fall back to __slab_free where we deal | |
2700 | * with all sorts of special processing. | |
81084651 JDB |
2701 | * |
2702 | * Bulk free of a freelist with several objects (all pointing to the | |
2703 | * same page) possible by specifying head and tail ptr, plus objects | |
2704 | * count (cnt). Bulk free indicated by tail pointer being set. | |
894b8788 | 2705 | */ |
81084651 JDB |
2706 | static __always_inline void slab_free(struct kmem_cache *s, struct page *page, |
2707 | void *head, void *tail, int cnt, | |
2708 | unsigned long addr) | |
894b8788 | 2709 | { |
81084651 | 2710 | void *tail_obj = tail ? : head; |
dfb4f096 | 2711 | struct kmem_cache_cpu *c; |
8a5ec0ba | 2712 | unsigned long tid; |
1f84260c | 2713 | |
81084651 | 2714 | slab_free_freelist_hook(s, head, tail); |
c016b0bd | 2715 | |
8a5ec0ba CL |
2716 | redo: |
2717 | /* | |
2718 | * Determine the currently cpus per cpu slab. | |
2719 | * The cpu may change afterward. However that does not matter since | |
2720 | * data is retrieved via this pointer. If we are on the same cpu | |
2ae44005 | 2721 | * during the cmpxchg then the free will succeed. |
8a5ec0ba | 2722 | */ |
9aabf810 JK |
2723 | do { |
2724 | tid = this_cpu_read(s->cpu_slab->tid); | |
2725 | c = raw_cpu_ptr(s->cpu_slab); | |
859b7a0e MR |
2726 | } while (IS_ENABLED(CONFIG_PREEMPT) && |
2727 | unlikely(tid != READ_ONCE(c->tid))); | |
c016b0bd | 2728 | |
9aabf810 JK |
2729 | /* Same with comment on barrier() in slab_alloc_node() */ |
2730 | barrier(); | |
c016b0bd | 2731 | |
442b06bc | 2732 | if (likely(page == c->page)) { |
81084651 | 2733 | set_freepointer(s, tail_obj, c->freelist); |
8a5ec0ba | 2734 | |
933393f5 | 2735 | if (unlikely(!this_cpu_cmpxchg_double( |
8a5ec0ba CL |
2736 | s->cpu_slab->freelist, s->cpu_slab->tid, |
2737 | c->freelist, tid, | |
81084651 | 2738 | head, next_tid(tid)))) { |
8a5ec0ba CL |
2739 | |
2740 | note_cmpxchg_failure("slab_free", s, tid); | |
2741 | goto redo; | |
2742 | } | |
84e554e6 | 2743 | stat(s, FREE_FASTPATH); |
894b8788 | 2744 | } else |
81084651 | 2745 | __slab_free(s, page, head, tail_obj, cnt, addr); |
894b8788 | 2746 | |
894b8788 CL |
2747 | } |
2748 | ||
81819f0f CL |
2749 | void kmem_cache_free(struct kmem_cache *s, void *x) |
2750 | { | |
b9ce5ef4 GC |
2751 | s = cache_from_obj(s, x); |
2752 | if (!s) | |
79576102 | 2753 | return; |
81084651 | 2754 | slab_free(s, virt_to_head_page(x), x, NULL, 1, _RET_IP_); |
ca2b84cb | 2755 | trace_kmem_cache_free(_RET_IP_, x); |
81819f0f CL |
2756 | } |
2757 | EXPORT_SYMBOL(kmem_cache_free); | |
2758 | ||
d0ecd894 | 2759 | struct detached_freelist { |
fbd02630 | 2760 | struct page *page; |
d0ecd894 JDB |
2761 | void *tail; |
2762 | void *freelist; | |
2763 | int cnt; | |
376bf125 | 2764 | struct kmem_cache *s; |
d0ecd894 | 2765 | }; |
fbd02630 | 2766 | |
d0ecd894 JDB |
2767 | /* |
2768 | * This function progressively scans the array with free objects (with | |
2769 | * a limited look ahead) and extract objects belonging to the same | |
2770 | * page. It builds a detached freelist directly within the given | |
2771 | * page/objects. This can happen without any need for | |
2772 | * synchronization, because the objects are owned by running process. | |
2773 | * The freelist is build up as a single linked list in the objects. | |
2774 | * The idea is, that this detached freelist can then be bulk | |
2775 | * transferred to the real freelist(s), but only requiring a single | |
2776 | * synchronization primitive. Look ahead in the array is limited due | |
2777 | * to performance reasons. | |
2778 | */ | |
376bf125 JDB |
2779 | static inline |
2780 | int build_detached_freelist(struct kmem_cache *s, size_t size, | |
2781 | void **p, struct detached_freelist *df) | |
d0ecd894 JDB |
2782 | { |
2783 | size_t first_skipped_index = 0; | |
2784 | int lookahead = 3; | |
2785 | void *object; | |
ca257195 | 2786 | struct page *page; |
fbd02630 | 2787 | |
d0ecd894 JDB |
2788 | /* Always re-init detached_freelist */ |
2789 | df->page = NULL; | |
fbd02630 | 2790 | |
d0ecd894 JDB |
2791 | do { |
2792 | object = p[--size]; | |
ca257195 | 2793 | /* Do we need !ZERO_OR_NULL_PTR(object) here? (for kfree) */ |
d0ecd894 | 2794 | } while (!object && size); |
3eed034d | 2795 | |
d0ecd894 JDB |
2796 | if (!object) |
2797 | return 0; | |
fbd02630 | 2798 | |
ca257195 JDB |
2799 | page = virt_to_head_page(object); |
2800 | if (!s) { | |
2801 | /* Handle kalloc'ed objects */ | |
2802 | if (unlikely(!PageSlab(page))) { | |
2803 | BUG_ON(!PageCompound(page)); | |
2804 | kfree_hook(object); | |
2805 | __free_kmem_pages(page, compound_order(page)); | |
2806 | p[size] = NULL; /* mark object processed */ | |
2807 | return size; | |
2808 | } | |
2809 | /* Derive kmem_cache from object */ | |
2810 | df->s = page->slab_cache; | |
2811 | } else { | |
2812 | df->s = cache_from_obj(s, object); /* Support for memcg */ | |
2813 | } | |
376bf125 | 2814 | |
d0ecd894 | 2815 | /* Start new detached freelist */ |
ca257195 | 2816 | df->page = page; |
376bf125 | 2817 | set_freepointer(df->s, object, NULL); |
d0ecd894 JDB |
2818 | df->tail = object; |
2819 | df->freelist = object; | |
2820 | p[size] = NULL; /* mark object processed */ | |
2821 | df->cnt = 1; | |
2822 | ||
2823 | while (size) { | |
2824 | object = p[--size]; | |
2825 | if (!object) | |
2826 | continue; /* Skip processed objects */ | |
2827 | ||
2828 | /* df->page is always set at this point */ | |
2829 | if (df->page == virt_to_head_page(object)) { | |
2830 | /* Opportunity build freelist */ | |
376bf125 | 2831 | set_freepointer(df->s, object, df->freelist); |
d0ecd894 JDB |
2832 | df->freelist = object; |
2833 | df->cnt++; | |
2834 | p[size] = NULL; /* mark object processed */ | |
2835 | ||
2836 | continue; | |
fbd02630 | 2837 | } |
d0ecd894 JDB |
2838 | |
2839 | /* Limit look ahead search */ | |
2840 | if (!--lookahead) | |
2841 | break; | |
2842 | ||
2843 | if (!first_skipped_index) | |
2844 | first_skipped_index = size + 1; | |
fbd02630 | 2845 | } |
d0ecd894 JDB |
2846 | |
2847 | return first_skipped_index; | |
2848 | } | |
2849 | ||
d0ecd894 | 2850 | /* Note that interrupts must be enabled when calling this function. */ |
376bf125 | 2851 | void kmem_cache_free_bulk(struct kmem_cache *s, size_t size, void **p) |
d0ecd894 JDB |
2852 | { |
2853 | if (WARN_ON(!size)) | |
2854 | return; | |
2855 | ||
2856 | do { | |
2857 | struct detached_freelist df; | |
2858 | ||
2859 | size = build_detached_freelist(s, size, p, &df); | |
2860 | if (unlikely(!df.page)) | |
2861 | continue; | |
2862 | ||
376bf125 | 2863 | slab_free(df.s, df.page, df.freelist, df.tail, df.cnt,_RET_IP_); |
d0ecd894 | 2864 | } while (likely(size)); |
484748f0 CL |
2865 | } |
2866 | EXPORT_SYMBOL(kmem_cache_free_bulk); | |
2867 | ||
994eb764 | 2868 | /* Note that interrupts must be enabled when calling this function. */ |
865762a8 JDB |
2869 | int kmem_cache_alloc_bulk(struct kmem_cache *s, gfp_t flags, size_t size, |
2870 | void **p) | |
484748f0 | 2871 | { |
994eb764 JDB |
2872 | struct kmem_cache_cpu *c; |
2873 | int i; | |
2874 | ||
03ec0ed5 JDB |
2875 | /* memcg and kmem_cache debug support */ |
2876 | s = slab_pre_alloc_hook(s, flags); | |
2877 | if (unlikely(!s)) | |
2878 | return false; | |
994eb764 JDB |
2879 | /* |
2880 | * Drain objects in the per cpu slab, while disabling local | |
2881 | * IRQs, which protects against PREEMPT and interrupts | |
2882 | * handlers invoking normal fastpath. | |
2883 | */ | |
2884 | local_irq_disable(); | |
2885 | c = this_cpu_ptr(s->cpu_slab); | |
2886 | ||
2887 | for (i = 0; i < size; i++) { | |
2888 | void *object = c->freelist; | |
2889 | ||
ebe909e0 | 2890 | if (unlikely(!object)) { |
ebe909e0 JDB |
2891 | /* |
2892 | * Invoking slow path likely have side-effect | |
2893 | * of re-populating per CPU c->freelist | |
2894 | */ | |
87098373 | 2895 | p[i] = ___slab_alloc(s, flags, NUMA_NO_NODE, |
ebe909e0 | 2896 | _RET_IP_, c); |
87098373 CL |
2897 | if (unlikely(!p[i])) |
2898 | goto error; | |
2899 | ||
ebe909e0 JDB |
2900 | c = this_cpu_ptr(s->cpu_slab); |
2901 | continue; /* goto for-loop */ | |
2902 | } | |
994eb764 JDB |
2903 | c->freelist = get_freepointer(s, object); |
2904 | p[i] = object; | |
2905 | } | |
2906 | c->tid = next_tid(c->tid); | |
2907 | local_irq_enable(); | |
2908 | ||
2909 | /* Clear memory outside IRQ disabled fastpath loop */ | |
2910 | if (unlikely(flags & __GFP_ZERO)) { | |
2911 | int j; | |
2912 | ||
2913 | for (j = 0; j < i; j++) | |
2914 | memset(p[j], 0, s->object_size); | |
2915 | } | |
2916 | ||
03ec0ed5 JDB |
2917 | /* memcg and kmem_cache debug support */ |
2918 | slab_post_alloc_hook(s, flags, size, p); | |
865762a8 | 2919 | return i; |
87098373 | 2920 | error: |
87098373 | 2921 | local_irq_enable(); |
03ec0ed5 JDB |
2922 | slab_post_alloc_hook(s, flags, i, p); |
2923 | __kmem_cache_free_bulk(s, i, p); | |
865762a8 | 2924 | return 0; |
484748f0 CL |
2925 | } |
2926 | EXPORT_SYMBOL(kmem_cache_alloc_bulk); | |
2927 | ||
2928 | ||
81819f0f | 2929 | /* |
672bba3a CL |
2930 | * Object placement in a slab is made very easy because we always start at |
2931 | * offset 0. If we tune the size of the object to the alignment then we can | |
2932 | * get the required alignment by putting one properly sized object after | |
2933 | * another. | |
81819f0f CL |
2934 | * |
2935 | * Notice that the allocation order determines the sizes of the per cpu | |
2936 | * caches. Each processor has always one slab available for allocations. | |
2937 | * Increasing the allocation order reduces the number of times that slabs | |
672bba3a | 2938 | * must be moved on and off the partial lists and is therefore a factor in |
81819f0f | 2939 | * locking overhead. |
81819f0f CL |
2940 | */ |
2941 | ||
2942 | /* | |
2943 | * Mininum / Maximum order of slab pages. This influences locking overhead | |
2944 | * and slab fragmentation. A higher order reduces the number of partial slabs | |
2945 | * and increases the number of allocations possible without having to | |
2946 | * take the list_lock. | |
2947 | */ | |
2948 | static int slub_min_order; | |
114e9e89 | 2949 | static int slub_max_order = PAGE_ALLOC_COSTLY_ORDER; |
9b2cd506 | 2950 | static int slub_min_objects; |
81819f0f | 2951 | |
81819f0f CL |
2952 | /* |
2953 | * Calculate the order of allocation given an slab object size. | |
2954 | * | |
672bba3a CL |
2955 | * The order of allocation has significant impact on performance and other |
2956 | * system components. Generally order 0 allocations should be preferred since | |
2957 | * order 0 does not cause fragmentation in the page allocator. Larger objects | |
2958 | * be problematic to put into order 0 slabs because there may be too much | |
c124f5b5 | 2959 | * unused space left. We go to a higher order if more than 1/16th of the slab |
672bba3a CL |
2960 | * would be wasted. |
2961 | * | |
2962 | * In order to reach satisfactory performance we must ensure that a minimum | |
2963 | * number of objects is in one slab. Otherwise we may generate too much | |
2964 | * activity on the partial lists which requires taking the list_lock. This is | |
2965 | * less a concern for large slabs though which are rarely used. | |
81819f0f | 2966 | * |
672bba3a CL |
2967 | * slub_max_order specifies the order where we begin to stop considering the |
2968 | * number of objects in a slab as critical. If we reach slub_max_order then | |
2969 | * we try to keep the page order as low as possible. So we accept more waste | |
2970 | * of space in favor of a small page order. | |
81819f0f | 2971 | * |
672bba3a CL |
2972 | * Higher order allocations also allow the placement of more objects in a |
2973 | * slab and thereby reduce object handling overhead. If the user has | |
2974 | * requested a higher mininum order then we start with that one instead of | |
2975 | * the smallest order which will fit the object. | |
81819f0f | 2976 | */ |
5e6d444e | 2977 | static inline int slab_order(int size, int min_objects, |
ab9a0f19 | 2978 | int max_order, int fract_leftover, int reserved) |
81819f0f CL |
2979 | { |
2980 | int order; | |
2981 | int rem; | |
6300ea75 | 2982 | int min_order = slub_min_order; |
81819f0f | 2983 | |
ab9a0f19 | 2984 | if (order_objects(min_order, size, reserved) > MAX_OBJS_PER_PAGE) |
210b5c06 | 2985 | return get_order(size * MAX_OBJS_PER_PAGE) - 1; |
39b26464 | 2986 | |
9f835703 | 2987 | for (order = max(min_order, get_order(min_objects * size + reserved)); |
5e6d444e | 2988 | order <= max_order; order++) { |
81819f0f | 2989 | |
5e6d444e | 2990 | unsigned long slab_size = PAGE_SIZE << order; |
81819f0f | 2991 | |
ab9a0f19 | 2992 | rem = (slab_size - reserved) % size; |
81819f0f | 2993 | |
5e6d444e | 2994 | if (rem <= slab_size / fract_leftover) |
81819f0f | 2995 | break; |
81819f0f | 2996 | } |
672bba3a | 2997 | |
81819f0f CL |
2998 | return order; |
2999 | } | |
3000 | ||
ab9a0f19 | 3001 | static inline int calculate_order(int size, int reserved) |
5e6d444e CL |
3002 | { |
3003 | int order; | |
3004 | int min_objects; | |
3005 | int fraction; | |
e8120ff1 | 3006 | int max_objects; |
5e6d444e CL |
3007 | |
3008 | /* | |
3009 | * Attempt to find best configuration for a slab. This | |
3010 | * works by first attempting to generate a layout with | |
3011 | * the best configuration and backing off gradually. | |
3012 | * | |
422ff4d7 | 3013 | * First we increase the acceptable waste in a slab. Then |
5e6d444e CL |
3014 | * we reduce the minimum objects required in a slab. |
3015 | */ | |
3016 | min_objects = slub_min_objects; | |
9b2cd506 CL |
3017 | if (!min_objects) |
3018 | min_objects = 4 * (fls(nr_cpu_ids) + 1); | |
ab9a0f19 | 3019 | max_objects = order_objects(slub_max_order, size, reserved); |
e8120ff1 ZY |
3020 | min_objects = min(min_objects, max_objects); |
3021 | ||
5e6d444e | 3022 | while (min_objects > 1) { |
c124f5b5 | 3023 | fraction = 16; |
5e6d444e CL |
3024 | while (fraction >= 4) { |
3025 | order = slab_order(size, min_objects, | |
ab9a0f19 | 3026 | slub_max_order, fraction, reserved); |
5e6d444e CL |
3027 | if (order <= slub_max_order) |
3028 | return order; | |
3029 | fraction /= 2; | |
3030 | } | |
5086c389 | 3031 | min_objects--; |
5e6d444e CL |
3032 | } |
3033 | ||
3034 | /* | |
3035 | * We were unable to place multiple objects in a slab. Now | |
3036 | * lets see if we can place a single object there. | |
3037 | */ | |
ab9a0f19 | 3038 | order = slab_order(size, 1, slub_max_order, 1, reserved); |
5e6d444e CL |
3039 | if (order <= slub_max_order) |
3040 | return order; | |
3041 | ||
3042 | /* | |
3043 | * Doh this slab cannot be placed using slub_max_order. | |
3044 | */ | |
ab9a0f19 | 3045 | order = slab_order(size, 1, MAX_ORDER, 1, reserved); |
818cf590 | 3046 | if (order < MAX_ORDER) |
5e6d444e CL |
3047 | return order; |
3048 | return -ENOSYS; | |
3049 | } | |
3050 | ||
5595cffc | 3051 | static void |
4053497d | 3052 | init_kmem_cache_node(struct kmem_cache_node *n) |
81819f0f CL |
3053 | { |
3054 | n->nr_partial = 0; | |
81819f0f CL |
3055 | spin_lock_init(&n->list_lock); |
3056 | INIT_LIST_HEAD(&n->partial); | |
8ab1372f | 3057 | #ifdef CONFIG_SLUB_DEBUG |
0f389ec6 | 3058 | atomic_long_set(&n->nr_slabs, 0); |
02b71b70 | 3059 | atomic_long_set(&n->total_objects, 0); |
643b1138 | 3060 | INIT_LIST_HEAD(&n->full); |
8ab1372f | 3061 | #endif |
81819f0f CL |
3062 | } |
3063 | ||
55136592 | 3064 | static inline int alloc_kmem_cache_cpus(struct kmem_cache *s) |
4c93c355 | 3065 | { |
6c182dc0 | 3066 | BUILD_BUG_ON(PERCPU_DYNAMIC_EARLY_SIZE < |
95a05b42 | 3067 | KMALLOC_SHIFT_HIGH * sizeof(struct kmem_cache_cpu)); |
4c93c355 | 3068 | |
8a5ec0ba | 3069 | /* |
d4d84fef CM |
3070 | * Must align to double word boundary for the double cmpxchg |
3071 | * instructions to work; see __pcpu_double_call_return_bool(). | |
8a5ec0ba | 3072 | */ |
d4d84fef CM |
3073 | s->cpu_slab = __alloc_percpu(sizeof(struct kmem_cache_cpu), |
3074 | 2 * sizeof(void *)); | |
8a5ec0ba CL |
3075 | |
3076 | if (!s->cpu_slab) | |
3077 | return 0; | |
3078 | ||
3079 | init_kmem_cache_cpus(s); | |
4c93c355 | 3080 | |
8a5ec0ba | 3081 | return 1; |
4c93c355 | 3082 | } |
4c93c355 | 3083 | |
51df1142 CL |
3084 | static struct kmem_cache *kmem_cache_node; |
3085 | ||
81819f0f CL |
3086 | /* |
3087 | * No kmalloc_node yet so do it by hand. We know that this is the first | |
3088 | * slab on the node for this slabcache. There are no concurrent accesses | |
3089 | * possible. | |
3090 | * | |
721ae22a ZYW |
3091 | * Note that this function only works on the kmem_cache_node |
3092 | * when allocating for the kmem_cache_node. This is used for bootstrapping | |
4c93c355 | 3093 | * memory on a fresh node that has no slab structures yet. |
81819f0f | 3094 | */ |
55136592 | 3095 | static void early_kmem_cache_node_alloc(int node) |
81819f0f CL |
3096 | { |
3097 | struct page *page; | |
3098 | struct kmem_cache_node *n; | |
3099 | ||
51df1142 | 3100 | BUG_ON(kmem_cache_node->size < sizeof(struct kmem_cache_node)); |
81819f0f | 3101 | |
51df1142 | 3102 | page = new_slab(kmem_cache_node, GFP_NOWAIT, node); |
81819f0f CL |
3103 | |
3104 | BUG_ON(!page); | |
a2f92ee7 | 3105 | if (page_to_nid(page) != node) { |
f9f58285 FF |
3106 | pr_err("SLUB: Unable to allocate memory from node %d\n", node); |
3107 | pr_err("SLUB: Allocating a useless per node structure in order to be able to continue\n"); | |
a2f92ee7 CL |
3108 | } |
3109 | ||
81819f0f CL |
3110 | n = page->freelist; |
3111 | BUG_ON(!n); | |
51df1142 | 3112 | page->freelist = get_freepointer(kmem_cache_node, n); |
e6e82ea1 | 3113 | page->inuse = 1; |
8cb0a506 | 3114 | page->frozen = 0; |
51df1142 | 3115 | kmem_cache_node->node[node] = n; |
8ab1372f | 3116 | #ifdef CONFIG_SLUB_DEBUG |
f7cb1933 | 3117 | init_object(kmem_cache_node, n, SLUB_RED_ACTIVE); |
51df1142 | 3118 | init_tracking(kmem_cache_node, n); |
8ab1372f | 3119 | #endif |
0316bec2 | 3120 | kasan_kmalloc(kmem_cache_node, n, sizeof(struct kmem_cache_node)); |
4053497d | 3121 | init_kmem_cache_node(n); |
51df1142 | 3122 | inc_slabs_node(kmem_cache_node, node, page->objects); |
6446faa2 | 3123 | |
67b6c900 | 3124 | /* |
1e4dd946 SR |
3125 | * No locks need to be taken here as it has just been |
3126 | * initialized and there is no concurrent access. | |
67b6c900 | 3127 | */ |
1e4dd946 | 3128 | __add_partial(n, page, DEACTIVATE_TO_HEAD); |
81819f0f CL |
3129 | } |
3130 | ||
3131 | static void free_kmem_cache_nodes(struct kmem_cache *s) | |
3132 | { | |
3133 | int node; | |
fa45dc25 | 3134 | struct kmem_cache_node *n; |
81819f0f | 3135 | |
fa45dc25 CL |
3136 | for_each_kmem_cache_node(s, node, n) { |
3137 | kmem_cache_free(kmem_cache_node, n); | |
81819f0f CL |
3138 | s->node[node] = NULL; |
3139 | } | |
3140 | } | |
3141 | ||
52b4b950 DS |
3142 | void __kmem_cache_release(struct kmem_cache *s) |
3143 | { | |
3144 | free_percpu(s->cpu_slab); | |
3145 | free_kmem_cache_nodes(s); | |
3146 | } | |
3147 | ||
55136592 | 3148 | static int init_kmem_cache_nodes(struct kmem_cache *s) |
81819f0f CL |
3149 | { |
3150 | int node; | |
81819f0f | 3151 | |
f64dc58c | 3152 | for_each_node_state(node, N_NORMAL_MEMORY) { |
81819f0f CL |
3153 | struct kmem_cache_node *n; |
3154 | ||
73367bd8 | 3155 | if (slab_state == DOWN) { |
55136592 | 3156 | early_kmem_cache_node_alloc(node); |
73367bd8 AD |
3157 | continue; |
3158 | } | |
51df1142 | 3159 | n = kmem_cache_alloc_node(kmem_cache_node, |
55136592 | 3160 | GFP_KERNEL, node); |
81819f0f | 3161 | |
73367bd8 AD |
3162 | if (!n) { |
3163 | free_kmem_cache_nodes(s); | |
3164 | return 0; | |
81819f0f | 3165 | } |
73367bd8 | 3166 | |
81819f0f | 3167 | s->node[node] = n; |
4053497d | 3168 | init_kmem_cache_node(n); |
81819f0f CL |
3169 | } |
3170 | return 1; | |
3171 | } | |
81819f0f | 3172 | |
c0bdb232 | 3173 | static void set_min_partial(struct kmem_cache *s, unsigned long min) |
3b89d7d8 DR |
3174 | { |
3175 | if (min < MIN_PARTIAL) | |
3176 | min = MIN_PARTIAL; | |
3177 | else if (min > MAX_PARTIAL) | |
3178 | min = MAX_PARTIAL; | |
3179 | s->min_partial = min; | |
3180 | } | |
3181 | ||
81819f0f CL |
3182 | /* |
3183 | * calculate_sizes() determines the order and the distribution of data within | |
3184 | * a slab object. | |
3185 | */ | |
06b285dc | 3186 | static int calculate_sizes(struct kmem_cache *s, int forced_order) |
81819f0f CL |
3187 | { |
3188 | unsigned long flags = s->flags; | |
3b0efdfa | 3189 | unsigned long size = s->object_size; |
834f3d11 | 3190 | int order; |
81819f0f | 3191 | |
d8b42bf5 CL |
3192 | /* |
3193 | * Round up object size to the next word boundary. We can only | |
3194 | * place the free pointer at word boundaries and this determines | |
3195 | * the possible location of the free pointer. | |
3196 | */ | |
3197 | size = ALIGN(size, sizeof(void *)); | |
3198 | ||
3199 | #ifdef CONFIG_SLUB_DEBUG | |
81819f0f CL |
3200 | /* |
3201 | * Determine if we can poison the object itself. If the user of | |
3202 | * the slab may touch the object after free or before allocation | |
3203 | * then we should never poison the object itself. | |
3204 | */ | |
3205 | if ((flags & SLAB_POISON) && !(flags & SLAB_DESTROY_BY_RCU) && | |
c59def9f | 3206 | !s->ctor) |
81819f0f CL |
3207 | s->flags |= __OBJECT_POISON; |
3208 | else | |
3209 | s->flags &= ~__OBJECT_POISON; | |
3210 | ||
81819f0f CL |
3211 | |
3212 | /* | |
672bba3a | 3213 | * If we are Redzoning then check if there is some space between the |
81819f0f | 3214 | * end of the object and the free pointer. If not then add an |
672bba3a | 3215 | * additional word to have some bytes to store Redzone information. |
81819f0f | 3216 | */ |
3b0efdfa | 3217 | if ((flags & SLAB_RED_ZONE) && size == s->object_size) |
81819f0f | 3218 | size += sizeof(void *); |
41ecc55b | 3219 | #endif |
81819f0f CL |
3220 | |
3221 | /* | |
672bba3a CL |
3222 | * With that we have determined the number of bytes in actual use |
3223 | * by the object. This is the potential offset to the free pointer. | |
81819f0f CL |
3224 | */ |
3225 | s->inuse = size; | |
3226 | ||
3227 | if (((flags & (SLAB_DESTROY_BY_RCU | SLAB_POISON)) || | |
c59def9f | 3228 | s->ctor)) { |
81819f0f CL |
3229 | /* |
3230 | * Relocate free pointer after the object if it is not | |
3231 | * permitted to overwrite the first word of the object on | |
3232 | * kmem_cache_free. | |
3233 | * | |
3234 | * This is the case if we do RCU, have a constructor or | |
3235 | * destructor or are poisoning the objects. | |
3236 | */ | |
3237 | s->offset = size; | |
3238 | size += sizeof(void *); | |
3239 | } | |
3240 | ||
c12b3c62 | 3241 | #ifdef CONFIG_SLUB_DEBUG |
81819f0f CL |
3242 | if (flags & SLAB_STORE_USER) |
3243 | /* | |
3244 | * Need to store information about allocs and frees after | |
3245 | * the object. | |
3246 | */ | |
3247 | size += 2 * sizeof(struct track); | |
3248 | ||
be7b3fbc | 3249 | if (flags & SLAB_RED_ZONE) |
81819f0f CL |
3250 | /* |
3251 | * Add some empty padding so that we can catch | |
3252 | * overwrites from earlier objects rather than let | |
3253 | * tracking information or the free pointer be | |
0211a9c8 | 3254 | * corrupted if a user writes before the start |
81819f0f CL |
3255 | * of the object. |
3256 | */ | |
3257 | size += sizeof(void *); | |
41ecc55b | 3258 | #endif |
672bba3a | 3259 | |
81819f0f CL |
3260 | /* |
3261 | * SLUB stores one object immediately after another beginning from | |
3262 | * offset 0. In order to align the objects we have to simply size | |
3263 | * each object to conform to the alignment. | |
3264 | */ | |
45906855 | 3265 | size = ALIGN(size, s->align); |
81819f0f | 3266 | s->size = size; |
06b285dc CL |
3267 | if (forced_order >= 0) |
3268 | order = forced_order; | |
3269 | else | |
ab9a0f19 | 3270 | order = calculate_order(size, s->reserved); |
81819f0f | 3271 | |
834f3d11 | 3272 | if (order < 0) |
81819f0f CL |
3273 | return 0; |
3274 | ||
b7a49f0d | 3275 | s->allocflags = 0; |
834f3d11 | 3276 | if (order) |
b7a49f0d CL |
3277 | s->allocflags |= __GFP_COMP; |
3278 | ||
3279 | if (s->flags & SLAB_CACHE_DMA) | |
2c59dd65 | 3280 | s->allocflags |= GFP_DMA; |
b7a49f0d CL |
3281 | |
3282 | if (s->flags & SLAB_RECLAIM_ACCOUNT) | |
3283 | s->allocflags |= __GFP_RECLAIMABLE; | |
3284 | ||
81819f0f CL |
3285 | /* |
3286 | * Determine the number of objects per slab | |
3287 | */ | |
ab9a0f19 LJ |
3288 | s->oo = oo_make(order, size, s->reserved); |
3289 | s->min = oo_make(get_order(size), size, s->reserved); | |
205ab99d CL |
3290 | if (oo_objects(s->oo) > oo_objects(s->max)) |
3291 | s->max = s->oo; | |
81819f0f | 3292 | |
834f3d11 | 3293 | return !!oo_objects(s->oo); |
81819f0f CL |
3294 | } |
3295 | ||
8a13a4cc | 3296 | static int kmem_cache_open(struct kmem_cache *s, unsigned long flags) |
81819f0f | 3297 | { |
8a13a4cc | 3298 | s->flags = kmem_cache_flags(s->size, flags, s->name, s->ctor); |
ab9a0f19 | 3299 | s->reserved = 0; |
81819f0f | 3300 | |
da9a638c LJ |
3301 | if (need_reserve_slab_rcu && (s->flags & SLAB_DESTROY_BY_RCU)) |
3302 | s->reserved = sizeof(struct rcu_head); | |
81819f0f | 3303 | |
06b285dc | 3304 | if (!calculate_sizes(s, -1)) |
81819f0f | 3305 | goto error; |
3de47213 DR |
3306 | if (disable_higher_order_debug) { |
3307 | /* | |
3308 | * Disable debugging flags that store metadata if the min slab | |
3309 | * order increased. | |
3310 | */ | |
3b0efdfa | 3311 | if (get_order(s->size) > get_order(s->object_size)) { |
3de47213 DR |
3312 | s->flags &= ~DEBUG_METADATA_FLAGS; |
3313 | s->offset = 0; | |
3314 | if (!calculate_sizes(s, -1)) | |
3315 | goto error; | |
3316 | } | |
3317 | } | |
81819f0f | 3318 | |
2565409f HC |
3319 | #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \ |
3320 | defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE) | |
b789ef51 CL |
3321 | if (system_has_cmpxchg_double() && (s->flags & SLAB_DEBUG_FLAGS) == 0) |
3322 | /* Enable fast mode */ | |
3323 | s->flags |= __CMPXCHG_DOUBLE; | |
3324 | #endif | |
3325 | ||
3b89d7d8 DR |
3326 | /* |
3327 | * The larger the object size is, the more pages we want on the partial | |
3328 | * list to avoid pounding the page allocator excessively. | |
3329 | */ | |
49e22585 CL |
3330 | set_min_partial(s, ilog2(s->size) / 2); |
3331 | ||
3332 | /* | |
3333 | * cpu_partial determined the maximum number of objects kept in the | |
3334 | * per cpu partial lists of a processor. | |
3335 | * | |
3336 | * Per cpu partial lists mainly contain slabs that just have one | |
3337 | * object freed. If they are used for allocation then they can be | |
3338 | * filled up again with minimal effort. The slab will never hit the | |
3339 | * per node partial lists and therefore no locking will be required. | |
3340 | * | |
3341 | * This setting also determines | |
3342 | * | |
3343 | * A) The number of objects from per cpu partial slabs dumped to the | |
3344 | * per node list when we reach the limit. | |
9f264904 | 3345 | * B) The number of objects in cpu partial slabs to extract from the |
d0e0ac97 CG |
3346 | * per node list when we run out of per cpu objects. We only fetch |
3347 | * 50% to keep some capacity around for frees. | |
49e22585 | 3348 | */ |
345c905d | 3349 | if (!kmem_cache_has_cpu_partial(s)) |
8f1e33da CL |
3350 | s->cpu_partial = 0; |
3351 | else if (s->size >= PAGE_SIZE) | |
49e22585 CL |
3352 | s->cpu_partial = 2; |
3353 | else if (s->size >= 1024) | |
3354 | s->cpu_partial = 6; | |
3355 | else if (s->size >= 256) | |
3356 | s->cpu_partial = 13; | |
3357 | else | |
3358 | s->cpu_partial = 30; | |
3359 | ||
81819f0f | 3360 | #ifdef CONFIG_NUMA |
e2cb96b7 | 3361 | s->remote_node_defrag_ratio = 1000; |
81819f0f | 3362 | #endif |
55136592 | 3363 | if (!init_kmem_cache_nodes(s)) |
dfb4f096 | 3364 | goto error; |
81819f0f | 3365 | |
55136592 | 3366 | if (alloc_kmem_cache_cpus(s)) |
278b1bb1 | 3367 | return 0; |
ff12059e | 3368 | |
4c93c355 | 3369 | free_kmem_cache_nodes(s); |
81819f0f CL |
3370 | error: |
3371 | if (flags & SLAB_PANIC) | |
3372 | panic("Cannot create slab %s size=%lu realsize=%u " | |
3373 | "order=%u offset=%u flags=%lx\n", | |
d0e0ac97 CG |
3374 | s->name, (unsigned long)s->size, s->size, |
3375 | oo_order(s->oo), s->offset, flags); | |
278b1bb1 | 3376 | return -EINVAL; |
81819f0f | 3377 | } |
81819f0f | 3378 | |
33b12c38 CL |
3379 | static void list_slab_objects(struct kmem_cache *s, struct page *page, |
3380 | const char *text) | |
3381 | { | |
3382 | #ifdef CONFIG_SLUB_DEBUG | |
3383 | void *addr = page_address(page); | |
3384 | void *p; | |
a5dd5c11 NK |
3385 | unsigned long *map = kzalloc(BITS_TO_LONGS(page->objects) * |
3386 | sizeof(long), GFP_ATOMIC); | |
bbd7d57b ED |
3387 | if (!map) |
3388 | return; | |
945cf2b6 | 3389 | slab_err(s, page, text, s->name); |
33b12c38 | 3390 | slab_lock(page); |
33b12c38 | 3391 | |
5f80b13a | 3392 | get_map(s, page, map); |
33b12c38 CL |
3393 | for_each_object(p, s, addr, page->objects) { |
3394 | ||
3395 | if (!test_bit(slab_index(p, s, addr), map)) { | |
f9f58285 | 3396 | pr_err("INFO: Object 0x%p @offset=%tu\n", p, p - addr); |
33b12c38 CL |
3397 | print_tracking(s, p); |
3398 | } | |
3399 | } | |
3400 | slab_unlock(page); | |
bbd7d57b | 3401 | kfree(map); |
33b12c38 CL |
3402 | #endif |
3403 | } | |
3404 | ||
81819f0f | 3405 | /* |
599870b1 | 3406 | * Attempt to free all partial slabs on a node. |
52b4b950 DS |
3407 | * This is called from __kmem_cache_shutdown(). We must take list_lock |
3408 | * because sysfs file might still access partial list after the shutdowning. | |
81819f0f | 3409 | */ |
599870b1 | 3410 | static void free_partial(struct kmem_cache *s, struct kmem_cache_node *n) |
81819f0f | 3411 | { |
81819f0f CL |
3412 | struct page *page, *h; |
3413 | ||
52b4b950 DS |
3414 | BUG_ON(irqs_disabled()); |
3415 | spin_lock_irq(&n->list_lock); | |
33b12c38 | 3416 | list_for_each_entry_safe(page, h, &n->partial, lru) { |
81819f0f | 3417 | if (!page->inuse) { |
52b4b950 | 3418 | remove_partial(n, page); |
81819f0f | 3419 | discard_slab(s, page); |
33b12c38 CL |
3420 | } else { |
3421 | list_slab_objects(s, page, | |
52b4b950 | 3422 | "Objects remaining in %s on __kmem_cache_shutdown()"); |
599870b1 | 3423 | } |
33b12c38 | 3424 | } |
52b4b950 | 3425 | spin_unlock_irq(&n->list_lock); |
81819f0f CL |
3426 | } |
3427 | ||
3428 | /* | |
672bba3a | 3429 | * Release all resources used by a slab cache. |
81819f0f | 3430 | */ |
52b4b950 | 3431 | int __kmem_cache_shutdown(struct kmem_cache *s) |
81819f0f CL |
3432 | { |
3433 | int node; | |
fa45dc25 | 3434 | struct kmem_cache_node *n; |
81819f0f CL |
3435 | |
3436 | flush_all(s); | |
81819f0f | 3437 | /* Attempt to free all objects */ |
fa45dc25 | 3438 | for_each_kmem_cache_node(s, node, n) { |
599870b1 CL |
3439 | free_partial(s, n); |
3440 | if (n->nr_partial || slabs_node(s, node)) | |
81819f0f CL |
3441 | return 1; |
3442 | } | |
81819f0f CL |
3443 | return 0; |
3444 | } | |
3445 | ||
81819f0f CL |
3446 | /******************************************************************** |
3447 | * Kmalloc subsystem | |
3448 | *******************************************************************/ | |
3449 | ||
81819f0f CL |
3450 | static int __init setup_slub_min_order(char *str) |
3451 | { | |
06428780 | 3452 | get_option(&str, &slub_min_order); |
81819f0f CL |
3453 | |
3454 | return 1; | |
3455 | } | |
3456 | ||
3457 | __setup("slub_min_order=", setup_slub_min_order); | |
3458 | ||
3459 | static int __init setup_slub_max_order(char *str) | |
3460 | { | |
06428780 | 3461 | get_option(&str, &slub_max_order); |
818cf590 | 3462 | slub_max_order = min(slub_max_order, MAX_ORDER - 1); |
81819f0f CL |
3463 | |
3464 | return 1; | |
3465 | } | |
3466 | ||
3467 | __setup("slub_max_order=", setup_slub_max_order); | |
3468 | ||
3469 | static int __init setup_slub_min_objects(char *str) | |
3470 | { | |
06428780 | 3471 | get_option(&str, &slub_min_objects); |
81819f0f CL |
3472 | |
3473 | return 1; | |
3474 | } | |
3475 | ||
3476 | __setup("slub_min_objects=", setup_slub_min_objects); | |
3477 | ||
81819f0f CL |
3478 | void *__kmalloc(size_t size, gfp_t flags) |
3479 | { | |
aadb4bc4 | 3480 | struct kmem_cache *s; |
5b882be4 | 3481 | void *ret; |
81819f0f | 3482 | |
95a05b42 | 3483 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) |
eada35ef | 3484 | return kmalloc_large(size, flags); |
aadb4bc4 | 3485 | |
2c59dd65 | 3486 | s = kmalloc_slab(size, flags); |
aadb4bc4 CL |
3487 | |
3488 | if (unlikely(ZERO_OR_NULL_PTR(s))) | |
6cb8f913 CL |
3489 | return s; |
3490 | ||
2b847c3c | 3491 | ret = slab_alloc(s, flags, _RET_IP_); |
5b882be4 | 3492 | |
ca2b84cb | 3493 | trace_kmalloc(_RET_IP_, ret, size, s->size, flags); |
5b882be4 | 3494 | |
0316bec2 AR |
3495 | kasan_kmalloc(s, ret, size); |
3496 | ||
5b882be4 | 3497 | return ret; |
81819f0f CL |
3498 | } |
3499 | EXPORT_SYMBOL(__kmalloc); | |
3500 | ||
5d1f57e4 | 3501 | #ifdef CONFIG_NUMA |
f619cfe1 CL |
3502 | static void *kmalloc_large_node(size_t size, gfp_t flags, int node) |
3503 | { | |
b1eeab67 | 3504 | struct page *page; |
e4f7c0b4 | 3505 | void *ptr = NULL; |
f619cfe1 | 3506 | |
52383431 VD |
3507 | flags |= __GFP_COMP | __GFP_NOTRACK; |
3508 | page = alloc_kmem_pages_node(node, flags, get_order(size)); | |
f619cfe1 | 3509 | if (page) |
e4f7c0b4 CM |
3510 | ptr = page_address(page); |
3511 | ||
d56791b3 | 3512 | kmalloc_large_node_hook(ptr, size, flags); |
e4f7c0b4 | 3513 | return ptr; |
f619cfe1 CL |
3514 | } |
3515 | ||
81819f0f CL |
3516 | void *__kmalloc_node(size_t size, gfp_t flags, int node) |
3517 | { | |
aadb4bc4 | 3518 | struct kmem_cache *s; |
5b882be4 | 3519 | void *ret; |
81819f0f | 3520 | |
95a05b42 | 3521 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) { |
5b882be4 EGM |
3522 | ret = kmalloc_large_node(size, flags, node); |
3523 | ||
ca2b84cb EGM |
3524 | trace_kmalloc_node(_RET_IP_, ret, |
3525 | size, PAGE_SIZE << get_order(size), | |
3526 | flags, node); | |
5b882be4 EGM |
3527 | |
3528 | return ret; | |
3529 | } | |
aadb4bc4 | 3530 | |
2c59dd65 | 3531 | s = kmalloc_slab(size, flags); |
aadb4bc4 CL |
3532 | |
3533 | if (unlikely(ZERO_OR_NULL_PTR(s))) | |
6cb8f913 CL |
3534 | return s; |
3535 | ||
2b847c3c | 3536 | ret = slab_alloc_node(s, flags, node, _RET_IP_); |
5b882be4 | 3537 | |
ca2b84cb | 3538 | trace_kmalloc_node(_RET_IP_, ret, size, s->size, flags, node); |
5b882be4 | 3539 | |
0316bec2 AR |
3540 | kasan_kmalloc(s, ret, size); |
3541 | ||
5b882be4 | 3542 | return ret; |
81819f0f CL |
3543 | } |
3544 | EXPORT_SYMBOL(__kmalloc_node); | |
3545 | #endif | |
3546 | ||
0316bec2 | 3547 | static size_t __ksize(const void *object) |
81819f0f | 3548 | { |
272c1d21 | 3549 | struct page *page; |
81819f0f | 3550 | |
ef8b4520 | 3551 | if (unlikely(object == ZERO_SIZE_PTR)) |
272c1d21 CL |
3552 | return 0; |
3553 | ||
294a80a8 | 3554 | page = virt_to_head_page(object); |
294a80a8 | 3555 | |
76994412 PE |
3556 | if (unlikely(!PageSlab(page))) { |
3557 | WARN_ON(!PageCompound(page)); | |
294a80a8 | 3558 | return PAGE_SIZE << compound_order(page); |
76994412 | 3559 | } |
81819f0f | 3560 | |
1b4f59e3 | 3561 | return slab_ksize(page->slab_cache); |
81819f0f | 3562 | } |
0316bec2 AR |
3563 | |
3564 | size_t ksize(const void *object) | |
3565 | { | |
3566 | size_t size = __ksize(object); | |
3567 | /* We assume that ksize callers could use whole allocated area, | |
3568 | so we need unpoison this area. */ | |
3569 | kasan_krealloc(object, size); | |
3570 | return size; | |
3571 | } | |
b1aabecd | 3572 | EXPORT_SYMBOL(ksize); |
81819f0f CL |
3573 | |
3574 | void kfree(const void *x) | |
3575 | { | |
81819f0f | 3576 | struct page *page; |
5bb983b0 | 3577 | void *object = (void *)x; |
81819f0f | 3578 | |
2121db74 PE |
3579 | trace_kfree(_RET_IP_, x); |
3580 | ||
2408c550 | 3581 | if (unlikely(ZERO_OR_NULL_PTR(x))) |
81819f0f CL |
3582 | return; |
3583 | ||
b49af68f | 3584 | page = virt_to_head_page(x); |
aadb4bc4 | 3585 | if (unlikely(!PageSlab(page))) { |
0937502a | 3586 | BUG_ON(!PageCompound(page)); |
d56791b3 | 3587 | kfree_hook(x); |
52383431 | 3588 | __free_kmem_pages(page, compound_order(page)); |
aadb4bc4 CL |
3589 | return; |
3590 | } | |
81084651 | 3591 | slab_free(page->slab_cache, page, object, NULL, 1, _RET_IP_); |
81819f0f CL |
3592 | } |
3593 | EXPORT_SYMBOL(kfree); | |
3594 | ||
832f37f5 VD |
3595 | #define SHRINK_PROMOTE_MAX 32 |
3596 | ||
2086d26a | 3597 | /* |
832f37f5 VD |
3598 | * kmem_cache_shrink discards empty slabs and promotes the slabs filled |
3599 | * up most to the head of the partial lists. New allocations will then | |
3600 | * fill those up and thus they can be removed from the partial lists. | |
672bba3a CL |
3601 | * |
3602 | * The slabs with the least items are placed last. This results in them | |
3603 | * being allocated from last increasing the chance that the last objects | |
3604 | * are freed in them. | |
2086d26a | 3605 | */ |
d6e0b7fa | 3606 | int __kmem_cache_shrink(struct kmem_cache *s, bool deactivate) |
2086d26a CL |
3607 | { |
3608 | int node; | |
3609 | int i; | |
3610 | struct kmem_cache_node *n; | |
3611 | struct page *page; | |
3612 | struct page *t; | |
832f37f5 VD |
3613 | struct list_head discard; |
3614 | struct list_head promote[SHRINK_PROMOTE_MAX]; | |
2086d26a | 3615 | unsigned long flags; |
ce3712d7 | 3616 | int ret = 0; |
2086d26a | 3617 | |
d6e0b7fa VD |
3618 | if (deactivate) { |
3619 | /* | |
3620 | * Disable empty slabs caching. Used to avoid pinning offline | |
3621 | * memory cgroups by kmem pages that can be freed. | |
3622 | */ | |
3623 | s->cpu_partial = 0; | |
3624 | s->min_partial = 0; | |
3625 | ||
3626 | /* | |
3627 | * s->cpu_partial is checked locklessly (see put_cpu_partial), | |
3628 | * so we have to make sure the change is visible. | |
3629 | */ | |
3630 | kick_all_cpus_sync(); | |
3631 | } | |
3632 | ||
2086d26a | 3633 | flush_all(s); |
fa45dc25 | 3634 | for_each_kmem_cache_node(s, node, n) { |
832f37f5 VD |
3635 | INIT_LIST_HEAD(&discard); |
3636 | for (i = 0; i < SHRINK_PROMOTE_MAX; i++) | |
3637 | INIT_LIST_HEAD(promote + i); | |
2086d26a CL |
3638 | |
3639 | spin_lock_irqsave(&n->list_lock, flags); | |
3640 | ||
3641 | /* | |
832f37f5 | 3642 | * Build lists of slabs to discard or promote. |
2086d26a | 3643 | * |
672bba3a CL |
3644 | * Note that concurrent frees may occur while we hold the |
3645 | * list_lock. page->inuse here is the upper limit. | |
2086d26a CL |
3646 | */ |
3647 | list_for_each_entry_safe(page, t, &n->partial, lru) { | |
832f37f5 VD |
3648 | int free = page->objects - page->inuse; |
3649 | ||
3650 | /* Do not reread page->inuse */ | |
3651 | barrier(); | |
3652 | ||
3653 | /* We do not keep full slabs on the list */ | |
3654 | BUG_ON(free <= 0); | |
3655 | ||
3656 | if (free == page->objects) { | |
3657 | list_move(&page->lru, &discard); | |
69cb8e6b | 3658 | n->nr_partial--; |
832f37f5 VD |
3659 | } else if (free <= SHRINK_PROMOTE_MAX) |
3660 | list_move(&page->lru, promote + free - 1); | |
2086d26a CL |
3661 | } |
3662 | ||
2086d26a | 3663 | /* |
832f37f5 VD |
3664 | * Promote the slabs filled up most to the head of the |
3665 | * partial list. | |
2086d26a | 3666 | */ |
832f37f5 VD |
3667 | for (i = SHRINK_PROMOTE_MAX - 1; i >= 0; i--) |
3668 | list_splice(promote + i, &n->partial); | |
2086d26a | 3669 | |
2086d26a | 3670 | spin_unlock_irqrestore(&n->list_lock, flags); |
69cb8e6b CL |
3671 | |
3672 | /* Release empty slabs */ | |
832f37f5 | 3673 | list_for_each_entry_safe(page, t, &discard, lru) |
69cb8e6b | 3674 | discard_slab(s, page); |
ce3712d7 VD |
3675 | |
3676 | if (slabs_node(s, node)) | |
3677 | ret = 1; | |
2086d26a CL |
3678 | } |
3679 | ||
ce3712d7 | 3680 | return ret; |
2086d26a | 3681 | } |
2086d26a | 3682 | |
b9049e23 YG |
3683 | static int slab_mem_going_offline_callback(void *arg) |
3684 | { | |
3685 | struct kmem_cache *s; | |
3686 | ||
18004c5d | 3687 | mutex_lock(&slab_mutex); |
b9049e23 | 3688 | list_for_each_entry(s, &slab_caches, list) |
d6e0b7fa | 3689 | __kmem_cache_shrink(s, false); |
18004c5d | 3690 | mutex_unlock(&slab_mutex); |
b9049e23 YG |
3691 | |
3692 | return 0; | |
3693 | } | |
3694 | ||
3695 | static void slab_mem_offline_callback(void *arg) | |
3696 | { | |
3697 | struct kmem_cache_node *n; | |
3698 | struct kmem_cache *s; | |
3699 | struct memory_notify *marg = arg; | |
3700 | int offline_node; | |
3701 | ||
b9d5ab25 | 3702 | offline_node = marg->status_change_nid_normal; |
b9049e23 YG |
3703 | |
3704 | /* | |
3705 | * If the node still has available memory. we need kmem_cache_node | |
3706 | * for it yet. | |
3707 | */ | |
3708 | if (offline_node < 0) | |
3709 | return; | |
3710 | ||
18004c5d | 3711 | mutex_lock(&slab_mutex); |
b9049e23 YG |
3712 | list_for_each_entry(s, &slab_caches, list) { |
3713 | n = get_node(s, offline_node); | |
3714 | if (n) { | |
3715 | /* | |
3716 | * if n->nr_slabs > 0, slabs still exist on the node | |
3717 | * that is going down. We were unable to free them, | |
c9404c9c | 3718 | * and offline_pages() function shouldn't call this |
b9049e23 YG |
3719 | * callback. So, we must fail. |
3720 | */ | |
0f389ec6 | 3721 | BUG_ON(slabs_node(s, offline_node)); |
b9049e23 YG |
3722 | |
3723 | s->node[offline_node] = NULL; | |
8de66a0c | 3724 | kmem_cache_free(kmem_cache_node, n); |
b9049e23 YG |
3725 | } |
3726 | } | |
18004c5d | 3727 | mutex_unlock(&slab_mutex); |
b9049e23 YG |
3728 | } |
3729 | ||
3730 | static int slab_mem_going_online_callback(void *arg) | |
3731 | { | |
3732 | struct kmem_cache_node *n; | |
3733 | struct kmem_cache *s; | |
3734 | struct memory_notify *marg = arg; | |
b9d5ab25 | 3735 | int nid = marg->status_change_nid_normal; |
b9049e23 YG |
3736 | int ret = 0; |
3737 | ||
3738 | /* | |
3739 | * If the node's memory is already available, then kmem_cache_node is | |
3740 | * already created. Nothing to do. | |
3741 | */ | |
3742 | if (nid < 0) | |
3743 | return 0; | |
3744 | ||
3745 | /* | |
0121c619 | 3746 | * We are bringing a node online. No memory is available yet. We must |
b9049e23 YG |
3747 | * allocate a kmem_cache_node structure in order to bring the node |
3748 | * online. | |
3749 | */ | |
18004c5d | 3750 | mutex_lock(&slab_mutex); |
b9049e23 YG |
3751 | list_for_each_entry(s, &slab_caches, list) { |
3752 | /* | |
3753 | * XXX: kmem_cache_alloc_node will fallback to other nodes | |
3754 | * since memory is not yet available from the node that | |
3755 | * is brought up. | |
3756 | */ | |
8de66a0c | 3757 | n = kmem_cache_alloc(kmem_cache_node, GFP_KERNEL); |
b9049e23 YG |
3758 | if (!n) { |
3759 | ret = -ENOMEM; | |
3760 | goto out; | |
3761 | } | |
4053497d | 3762 | init_kmem_cache_node(n); |
b9049e23 YG |
3763 | s->node[nid] = n; |
3764 | } | |
3765 | out: | |
18004c5d | 3766 | mutex_unlock(&slab_mutex); |
b9049e23 YG |
3767 | return ret; |
3768 | } | |
3769 | ||
3770 | static int slab_memory_callback(struct notifier_block *self, | |
3771 | unsigned long action, void *arg) | |
3772 | { | |
3773 | int ret = 0; | |
3774 | ||
3775 | switch (action) { | |
3776 | case MEM_GOING_ONLINE: | |
3777 | ret = slab_mem_going_online_callback(arg); | |
3778 | break; | |
3779 | case MEM_GOING_OFFLINE: | |
3780 | ret = slab_mem_going_offline_callback(arg); | |
3781 | break; | |
3782 | case MEM_OFFLINE: | |
3783 | case MEM_CANCEL_ONLINE: | |
3784 | slab_mem_offline_callback(arg); | |
3785 | break; | |
3786 | case MEM_ONLINE: | |
3787 | case MEM_CANCEL_OFFLINE: | |
3788 | break; | |
3789 | } | |
dc19f9db KH |
3790 | if (ret) |
3791 | ret = notifier_from_errno(ret); | |
3792 | else | |
3793 | ret = NOTIFY_OK; | |
b9049e23 YG |
3794 | return ret; |
3795 | } | |
3796 | ||
3ac38faa AM |
3797 | static struct notifier_block slab_memory_callback_nb = { |
3798 | .notifier_call = slab_memory_callback, | |
3799 | .priority = SLAB_CALLBACK_PRI, | |
3800 | }; | |
b9049e23 | 3801 | |
81819f0f CL |
3802 | /******************************************************************** |
3803 | * Basic setup of slabs | |
3804 | *******************************************************************/ | |
3805 | ||
51df1142 CL |
3806 | /* |
3807 | * Used for early kmem_cache structures that were allocated using | |
dffb4d60 CL |
3808 | * the page allocator. Allocate them properly then fix up the pointers |
3809 | * that may be pointing to the wrong kmem_cache structure. | |
51df1142 CL |
3810 | */ |
3811 | ||
dffb4d60 | 3812 | static struct kmem_cache * __init bootstrap(struct kmem_cache *static_cache) |
51df1142 CL |
3813 | { |
3814 | int node; | |
dffb4d60 | 3815 | struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT); |
fa45dc25 | 3816 | struct kmem_cache_node *n; |
51df1142 | 3817 | |
dffb4d60 | 3818 | memcpy(s, static_cache, kmem_cache->object_size); |
51df1142 | 3819 | |
7d557b3c GC |
3820 | /* |
3821 | * This runs very early, and only the boot processor is supposed to be | |
3822 | * up. Even if it weren't true, IRQs are not up so we couldn't fire | |
3823 | * IPIs around. | |
3824 | */ | |
3825 | __flush_cpu_slab(s, smp_processor_id()); | |
fa45dc25 | 3826 | for_each_kmem_cache_node(s, node, n) { |
51df1142 CL |
3827 | struct page *p; |
3828 | ||
fa45dc25 CL |
3829 | list_for_each_entry(p, &n->partial, lru) |
3830 | p->slab_cache = s; | |
51df1142 | 3831 | |
607bf324 | 3832 | #ifdef CONFIG_SLUB_DEBUG |
fa45dc25 CL |
3833 | list_for_each_entry(p, &n->full, lru) |
3834 | p->slab_cache = s; | |
51df1142 | 3835 | #endif |
51df1142 | 3836 | } |
f7ce3190 | 3837 | slab_init_memcg_params(s); |
dffb4d60 CL |
3838 | list_add(&s->list, &slab_caches); |
3839 | return s; | |
51df1142 CL |
3840 | } |
3841 | ||
81819f0f CL |
3842 | void __init kmem_cache_init(void) |
3843 | { | |
dffb4d60 CL |
3844 | static __initdata struct kmem_cache boot_kmem_cache, |
3845 | boot_kmem_cache_node; | |
51df1142 | 3846 | |
fc8d8620 SG |
3847 | if (debug_guardpage_minorder()) |
3848 | slub_max_order = 0; | |
3849 | ||
dffb4d60 CL |
3850 | kmem_cache_node = &boot_kmem_cache_node; |
3851 | kmem_cache = &boot_kmem_cache; | |
51df1142 | 3852 | |
dffb4d60 CL |
3853 | create_boot_cache(kmem_cache_node, "kmem_cache_node", |
3854 | sizeof(struct kmem_cache_node), SLAB_HWCACHE_ALIGN); | |
b9049e23 | 3855 | |
3ac38faa | 3856 | register_hotmemory_notifier(&slab_memory_callback_nb); |
81819f0f CL |
3857 | |
3858 | /* Able to allocate the per node structures */ | |
3859 | slab_state = PARTIAL; | |
3860 | ||
dffb4d60 CL |
3861 | create_boot_cache(kmem_cache, "kmem_cache", |
3862 | offsetof(struct kmem_cache, node) + | |
3863 | nr_node_ids * sizeof(struct kmem_cache_node *), | |
3864 | SLAB_HWCACHE_ALIGN); | |
8a13a4cc | 3865 | |
dffb4d60 | 3866 | kmem_cache = bootstrap(&boot_kmem_cache); |
81819f0f | 3867 | |
51df1142 CL |
3868 | /* |
3869 | * Allocate kmem_cache_node properly from the kmem_cache slab. | |
3870 | * kmem_cache_node is separately allocated so no need to | |
3871 | * update any list pointers. | |
3872 | */ | |
dffb4d60 | 3873 | kmem_cache_node = bootstrap(&boot_kmem_cache_node); |
51df1142 CL |
3874 | |
3875 | /* Now we can use the kmem_cache to allocate kmalloc slabs */ | |
34cc6990 | 3876 | setup_kmalloc_cache_index_table(); |
f97d5f63 | 3877 | create_kmalloc_caches(0); |
81819f0f CL |
3878 | |
3879 | #ifdef CONFIG_SMP | |
3880 | register_cpu_notifier(&slab_notifier); | |
9dfc6e68 | 3881 | #endif |
81819f0f | 3882 | |
f9f58285 | 3883 | pr_info("SLUB: HWalign=%d, Order=%d-%d, MinObjects=%d, CPUs=%d, Nodes=%d\n", |
f97d5f63 | 3884 | cache_line_size(), |
81819f0f CL |
3885 | slub_min_order, slub_max_order, slub_min_objects, |
3886 | nr_cpu_ids, nr_node_ids); | |
3887 | } | |
3888 | ||
7e85ee0c PE |
3889 | void __init kmem_cache_init_late(void) |
3890 | { | |
7e85ee0c PE |
3891 | } |
3892 | ||
2633d7a0 | 3893 | struct kmem_cache * |
a44cb944 VD |
3894 | __kmem_cache_alias(const char *name, size_t size, size_t align, |
3895 | unsigned long flags, void (*ctor)(void *)) | |
81819f0f | 3896 | { |
426589f5 | 3897 | struct kmem_cache *s, *c; |
81819f0f | 3898 | |
a44cb944 | 3899 | s = find_mergeable(size, align, flags, name, ctor); |
81819f0f CL |
3900 | if (s) { |
3901 | s->refcount++; | |
84d0ddd6 | 3902 | |
81819f0f CL |
3903 | /* |
3904 | * Adjust the object sizes so that we clear | |
3905 | * the complete object on kzalloc. | |
3906 | */ | |
3b0efdfa | 3907 | s->object_size = max(s->object_size, (int)size); |
81819f0f | 3908 | s->inuse = max_t(int, s->inuse, ALIGN(size, sizeof(void *))); |
6446faa2 | 3909 | |
426589f5 | 3910 | for_each_memcg_cache(c, s) { |
84d0ddd6 VD |
3911 | c->object_size = s->object_size; |
3912 | c->inuse = max_t(int, c->inuse, | |
3913 | ALIGN(size, sizeof(void *))); | |
3914 | } | |
3915 | ||
7b8f3b66 | 3916 | if (sysfs_slab_alias(s, name)) { |
7b8f3b66 | 3917 | s->refcount--; |
cbb79694 | 3918 | s = NULL; |
7b8f3b66 | 3919 | } |
a0e1d1be | 3920 | } |
6446faa2 | 3921 | |
cbb79694 CL |
3922 | return s; |
3923 | } | |
84c1cf62 | 3924 | |
8a13a4cc | 3925 | int __kmem_cache_create(struct kmem_cache *s, unsigned long flags) |
cbb79694 | 3926 | { |
aac3a166 PE |
3927 | int err; |
3928 | ||
3929 | err = kmem_cache_open(s, flags); | |
3930 | if (err) | |
3931 | return err; | |
20cea968 | 3932 | |
45530c44 CL |
3933 | /* Mutex is not taken during early boot */ |
3934 | if (slab_state <= UP) | |
3935 | return 0; | |
3936 | ||
107dab5c | 3937 | memcg_propagate_slab_attrs(s); |
aac3a166 | 3938 | err = sysfs_slab_add(s); |
aac3a166 | 3939 | if (err) |
52b4b950 | 3940 | __kmem_cache_release(s); |
20cea968 | 3941 | |
aac3a166 | 3942 | return err; |
81819f0f | 3943 | } |
81819f0f | 3944 | |
81819f0f | 3945 | #ifdef CONFIG_SMP |
81819f0f | 3946 | /* |
672bba3a CL |
3947 | * Use the cpu notifier to insure that the cpu slabs are flushed when |
3948 | * necessary. | |
81819f0f | 3949 | */ |
0db0628d | 3950 | static int slab_cpuup_callback(struct notifier_block *nfb, |
81819f0f CL |
3951 | unsigned long action, void *hcpu) |
3952 | { | |
3953 | long cpu = (long)hcpu; | |
5b95a4ac CL |
3954 | struct kmem_cache *s; |
3955 | unsigned long flags; | |
81819f0f CL |
3956 | |
3957 | switch (action) { | |
3958 | case CPU_UP_CANCELED: | |
8bb78442 | 3959 | case CPU_UP_CANCELED_FROZEN: |
81819f0f | 3960 | case CPU_DEAD: |
8bb78442 | 3961 | case CPU_DEAD_FROZEN: |
18004c5d | 3962 | mutex_lock(&slab_mutex); |
5b95a4ac CL |
3963 | list_for_each_entry(s, &slab_caches, list) { |
3964 | local_irq_save(flags); | |
3965 | __flush_cpu_slab(s, cpu); | |
3966 | local_irq_restore(flags); | |
3967 | } | |
18004c5d | 3968 | mutex_unlock(&slab_mutex); |
81819f0f CL |
3969 | break; |
3970 | default: | |
3971 | break; | |
3972 | } | |
3973 | return NOTIFY_OK; | |
3974 | } | |
3975 | ||
0db0628d | 3976 | static struct notifier_block slab_notifier = { |
3adbefee | 3977 | .notifier_call = slab_cpuup_callback |
06428780 | 3978 | }; |
81819f0f CL |
3979 | |
3980 | #endif | |
3981 | ||
ce71e27c | 3982 | void *__kmalloc_track_caller(size_t size, gfp_t gfpflags, unsigned long caller) |
81819f0f | 3983 | { |
aadb4bc4 | 3984 | struct kmem_cache *s; |
94b528d0 | 3985 | void *ret; |
aadb4bc4 | 3986 | |
95a05b42 | 3987 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) |
eada35ef PE |
3988 | return kmalloc_large(size, gfpflags); |
3989 | ||
2c59dd65 | 3990 | s = kmalloc_slab(size, gfpflags); |
81819f0f | 3991 | |
2408c550 | 3992 | if (unlikely(ZERO_OR_NULL_PTR(s))) |
6cb8f913 | 3993 | return s; |
81819f0f | 3994 | |
2b847c3c | 3995 | ret = slab_alloc(s, gfpflags, caller); |
94b528d0 | 3996 | |
25985edc | 3997 | /* Honor the call site pointer we received. */ |
ca2b84cb | 3998 | trace_kmalloc(caller, ret, size, s->size, gfpflags); |
94b528d0 EGM |
3999 | |
4000 | return ret; | |
81819f0f CL |
4001 | } |
4002 | ||
5d1f57e4 | 4003 | #ifdef CONFIG_NUMA |
81819f0f | 4004 | void *__kmalloc_node_track_caller(size_t size, gfp_t gfpflags, |
ce71e27c | 4005 | int node, unsigned long caller) |
81819f0f | 4006 | { |
aadb4bc4 | 4007 | struct kmem_cache *s; |
94b528d0 | 4008 | void *ret; |
aadb4bc4 | 4009 | |
95a05b42 | 4010 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) { |
d3e14aa3 XF |
4011 | ret = kmalloc_large_node(size, gfpflags, node); |
4012 | ||
4013 | trace_kmalloc_node(caller, ret, | |
4014 | size, PAGE_SIZE << get_order(size), | |
4015 | gfpflags, node); | |
4016 | ||
4017 | return ret; | |
4018 | } | |
eada35ef | 4019 | |
2c59dd65 | 4020 | s = kmalloc_slab(size, gfpflags); |
81819f0f | 4021 | |
2408c550 | 4022 | if (unlikely(ZERO_OR_NULL_PTR(s))) |
6cb8f913 | 4023 | return s; |
81819f0f | 4024 | |
2b847c3c | 4025 | ret = slab_alloc_node(s, gfpflags, node, caller); |
94b528d0 | 4026 | |
25985edc | 4027 | /* Honor the call site pointer we received. */ |
ca2b84cb | 4028 | trace_kmalloc_node(caller, ret, size, s->size, gfpflags, node); |
94b528d0 EGM |
4029 | |
4030 | return ret; | |
81819f0f | 4031 | } |
5d1f57e4 | 4032 | #endif |
81819f0f | 4033 | |
ab4d5ed5 | 4034 | #ifdef CONFIG_SYSFS |
205ab99d CL |
4035 | static int count_inuse(struct page *page) |
4036 | { | |
4037 | return page->inuse; | |
4038 | } | |
4039 | ||
4040 | static int count_total(struct page *page) | |
4041 | { | |
4042 | return page->objects; | |
4043 | } | |
ab4d5ed5 | 4044 | #endif |
205ab99d | 4045 | |
ab4d5ed5 | 4046 | #ifdef CONFIG_SLUB_DEBUG |
434e245d CL |
4047 | static int validate_slab(struct kmem_cache *s, struct page *page, |
4048 | unsigned long *map) | |
53e15af0 CL |
4049 | { |
4050 | void *p; | |
a973e9dd | 4051 | void *addr = page_address(page); |
53e15af0 CL |
4052 | |
4053 | if (!check_slab(s, page) || | |
4054 | !on_freelist(s, page, NULL)) | |
4055 | return 0; | |
4056 | ||
4057 | /* Now we know that a valid freelist exists */ | |
39b26464 | 4058 | bitmap_zero(map, page->objects); |
53e15af0 | 4059 | |
5f80b13a CL |
4060 | get_map(s, page, map); |
4061 | for_each_object(p, s, addr, page->objects) { | |
4062 | if (test_bit(slab_index(p, s, addr), map)) | |
4063 | if (!check_object(s, page, p, SLUB_RED_INACTIVE)) | |
4064 | return 0; | |
53e15af0 CL |
4065 | } |
4066 | ||
224a88be | 4067 | for_each_object(p, s, addr, page->objects) |
7656c72b | 4068 | if (!test_bit(slab_index(p, s, addr), map)) |
37d57443 | 4069 | if (!check_object(s, page, p, SLUB_RED_ACTIVE)) |
53e15af0 CL |
4070 | return 0; |
4071 | return 1; | |
4072 | } | |
4073 | ||
434e245d CL |
4074 | static void validate_slab_slab(struct kmem_cache *s, struct page *page, |
4075 | unsigned long *map) | |
53e15af0 | 4076 | { |
881db7fb CL |
4077 | slab_lock(page); |
4078 | validate_slab(s, page, map); | |
4079 | slab_unlock(page); | |
53e15af0 CL |
4080 | } |
4081 | ||
434e245d CL |
4082 | static int validate_slab_node(struct kmem_cache *s, |
4083 | struct kmem_cache_node *n, unsigned long *map) | |
53e15af0 CL |
4084 | { |
4085 | unsigned long count = 0; | |
4086 | struct page *page; | |
4087 | unsigned long flags; | |
4088 | ||
4089 | spin_lock_irqsave(&n->list_lock, flags); | |
4090 | ||
4091 | list_for_each_entry(page, &n->partial, lru) { | |
434e245d | 4092 | validate_slab_slab(s, page, map); |
53e15af0 CL |
4093 | count++; |
4094 | } | |
4095 | if (count != n->nr_partial) | |
f9f58285 FF |
4096 | pr_err("SLUB %s: %ld partial slabs counted but counter=%ld\n", |
4097 | s->name, count, n->nr_partial); | |
53e15af0 CL |
4098 | |
4099 | if (!(s->flags & SLAB_STORE_USER)) | |
4100 | goto out; | |
4101 | ||
4102 | list_for_each_entry(page, &n->full, lru) { | |
434e245d | 4103 | validate_slab_slab(s, page, map); |
53e15af0 CL |
4104 | count++; |
4105 | } | |
4106 | if (count != atomic_long_read(&n->nr_slabs)) | |
f9f58285 FF |
4107 | pr_err("SLUB: %s %ld slabs counted but counter=%ld\n", |
4108 | s->name, count, atomic_long_read(&n->nr_slabs)); | |
53e15af0 CL |
4109 | |
4110 | out: | |
4111 | spin_unlock_irqrestore(&n->list_lock, flags); | |
4112 | return count; | |
4113 | } | |
4114 | ||
434e245d | 4115 | static long validate_slab_cache(struct kmem_cache *s) |
53e15af0 CL |
4116 | { |
4117 | int node; | |
4118 | unsigned long count = 0; | |
205ab99d | 4119 | unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) * |
434e245d | 4120 | sizeof(unsigned long), GFP_KERNEL); |
fa45dc25 | 4121 | struct kmem_cache_node *n; |
434e245d CL |
4122 | |
4123 | if (!map) | |
4124 | return -ENOMEM; | |
53e15af0 CL |
4125 | |
4126 | flush_all(s); | |
fa45dc25 | 4127 | for_each_kmem_cache_node(s, node, n) |
434e245d | 4128 | count += validate_slab_node(s, n, map); |
434e245d | 4129 | kfree(map); |
53e15af0 CL |
4130 | return count; |
4131 | } | |
88a420e4 | 4132 | /* |
672bba3a | 4133 | * Generate lists of code addresses where slabcache objects are allocated |
88a420e4 CL |
4134 | * and freed. |
4135 | */ | |
4136 | ||
4137 | struct location { | |
4138 | unsigned long count; | |
ce71e27c | 4139 | unsigned long addr; |
45edfa58 CL |
4140 | long long sum_time; |
4141 | long min_time; | |
4142 | long max_time; | |
4143 | long min_pid; | |
4144 | long max_pid; | |
174596a0 | 4145 | DECLARE_BITMAP(cpus, NR_CPUS); |
45edfa58 | 4146 | nodemask_t nodes; |
88a420e4 CL |
4147 | }; |
4148 | ||
4149 | struct loc_track { | |
4150 | unsigned long max; | |
4151 | unsigned long count; | |
4152 | struct location *loc; | |
4153 | }; | |
4154 | ||
4155 | static void free_loc_track(struct loc_track *t) | |
4156 | { | |
4157 | if (t->max) | |
4158 | free_pages((unsigned long)t->loc, | |
4159 | get_order(sizeof(struct location) * t->max)); | |
4160 | } | |
4161 | ||
68dff6a9 | 4162 | static int alloc_loc_track(struct loc_track *t, unsigned long max, gfp_t flags) |
88a420e4 CL |
4163 | { |
4164 | struct location *l; | |
4165 | int order; | |
4166 | ||
88a420e4 CL |
4167 | order = get_order(sizeof(struct location) * max); |
4168 | ||
68dff6a9 | 4169 | l = (void *)__get_free_pages(flags, order); |
88a420e4 CL |
4170 | if (!l) |
4171 | return 0; | |
4172 | ||
4173 | if (t->count) { | |
4174 | memcpy(l, t->loc, sizeof(struct location) * t->count); | |
4175 | free_loc_track(t); | |
4176 | } | |
4177 | t->max = max; | |
4178 | t->loc = l; | |
4179 | return 1; | |
4180 | } | |
4181 | ||
4182 | static int add_location(struct loc_track *t, struct kmem_cache *s, | |
45edfa58 | 4183 | const struct track *track) |
88a420e4 CL |
4184 | { |
4185 | long start, end, pos; | |
4186 | struct location *l; | |
ce71e27c | 4187 | unsigned long caddr; |
45edfa58 | 4188 | unsigned long age = jiffies - track->when; |
88a420e4 CL |
4189 | |
4190 | start = -1; | |
4191 | end = t->count; | |
4192 | ||
4193 | for ( ; ; ) { | |
4194 | pos = start + (end - start + 1) / 2; | |
4195 | ||
4196 | /* | |
4197 | * There is nothing at "end". If we end up there | |
4198 | * we need to add something to before end. | |
4199 | */ | |
4200 | if (pos == end) | |
4201 | break; | |
4202 | ||
4203 | caddr = t->loc[pos].addr; | |
45edfa58 CL |
4204 | if (track->addr == caddr) { |
4205 | ||
4206 | l = &t->loc[pos]; | |
4207 | l->count++; | |
4208 | if (track->when) { | |
4209 | l->sum_time += age; | |
4210 | if (age < l->min_time) | |
4211 | l->min_time = age; | |
4212 | if (age > l->max_time) | |
4213 | l->max_time = age; | |
4214 | ||
4215 | if (track->pid < l->min_pid) | |
4216 | l->min_pid = track->pid; | |
4217 | if (track->pid > l->max_pid) | |
4218 | l->max_pid = track->pid; | |
4219 | ||
174596a0 RR |
4220 | cpumask_set_cpu(track->cpu, |
4221 | to_cpumask(l->cpus)); | |
45edfa58 CL |
4222 | } |
4223 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
88a420e4 CL |
4224 | return 1; |
4225 | } | |
4226 | ||
45edfa58 | 4227 | if (track->addr < caddr) |
88a420e4 CL |
4228 | end = pos; |
4229 | else | |
4230 | start = pos; | |
4231 | } | |
4232 | ||
4233 | /* | |
672bba3a | 4234 | * Not found. Insert new tracking element. |
88a420e4 | 4235 | */ |
68dff6a9 | 4236 | if (t->count >= t->max && !alloc_loc_track(t, 2 * t->max, GFP_ATOMIC)) |
88a420e4 CL |
4237 | return 0; |
4238 | ||
4239 | l = t->loc + pos; | |
4240 | if (pos < t->count) | |
4241 | memmove(l + 1, l, | |
4242 | (t->count - pos) * sizeof(struct location)); | |
4243 | t->count++; | |
4244 | l->count = 1; | |
45edfa58 CL |
4245 | l->addr = track->addr; |
4246 | l->sum_time = age; | |
4247 | l->min_time = age; | |
4248 | l->max_time = age; | |
4249 | l->min_pid = track->pid; | |
4250 | l->max_pid = track->pid; | |
174596a0 RR |
4251 | cpumask_clear(to_cpumask(l->cpus)); |
4252 | cpumask_set_cpu(track->cpu, to_cpumask(l->cpus)); | |
45edfa58 CL |
4253 | nodes_clear(l->nodes); |
4254 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
88a420e4 CL |
4255 | return 1; |
4256 | } | |
4257 | ||
4258 | static void process_slab(struct loc_track *t, struct kmem_cache *s, | |
bbd7d57b | 4259 | struct page *page, enum track_item alloc, |
a5dd5c11 | 4260 | unsigned long *map) |
88a420e4 | 4261 | { |
a973e9dd | 4262 | void *addr = page_address(page); |
88a420e4 CL |
4263 | void *p; |
4264 | ||
39b26464 | 4265 | bitmap_zero(map, page->objects); |
5f80b13a | 4266 | get_map(s, page, map); |
88a420e4 | 4267 | |
224a88be | 4268 | for_each_object(p, s, addr, page->objects) |
45edfa58 CL |
4269 | if (!test_bit(slab_index(p, s, addr), map)) |
4270 | add_location(t, s, get_track(s, p, alloc)); | |
88a420e4 CL |
4271 | } |
4272 | ||
4273 | static int list_locations(struct kmem_cache *s, char *buf, | |
4274 | enum track_item alloc) | |
4275 | { | |
e374d483 | 4276 | int len = 0; |
88a420e4 | 4277 | unsigned long i; |
68dff6a9 | 4278 | struct loc_track t = { 0, 0, NULL }; |
88a420e4 | 4279 | int node; |
bbd7d57b ED |
4280 | unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) * |
4281 | sizeof(unsigned long), GFP_KERNEL); | |
fa45dc25 | 4282 | struct kmem_cache_node *n; |
88a420e4 | 4283 | |
bbd7d57b ED |
4284 | if (!map || !alloc_loc_track(&t, PAGE_SIZE / sizeof(struct location), |
4285 | GFP_TEMPORARY)) { | |
4286 | kfree(map); | |
68dff6a9 | 4287 | return sprintf(buf, "Out of memory\n"); |
bbd7d57b | 4288 | } |
88a420e4 CL |
4289 | /* Push back cpu slabs */ |
4290 | flush_all(s); | |
4291 | ||
fa45dc25 | 4292 | for_each_kmem_cache_node(s, node, n) { |
88a420e4 CL |
4293 | unsigned long flags; |
4294 | struct page *page; | |
4295 | ||
9e86943b | 4296 | if (!atomic_long_read(&n->nr_slabs)) |
88a420e4 CL |
4297 | continue; |
4298 | ||
4299 | spin_lock_irqsave(&n->list_lock, flags); | |
4300 | list_for_each_entry(page, &n->partial, lru) | |
bbd7d57b | 4301 | process_slab(&t, s, page, alloc, map); |
88a420e4 | 4302 | list_for_each_entry(page, &n->full, lru) |
bbd7d57b | 4303 | process_slab(&t, s, page, alloc, map); |
88a420e4 CL |
4304 | spin_unlock_irqrestore(&n->list_lock, flags); |
4305 | } | |
4306 | ||
4307 | for (i = 0; i < t.count; i++) { | |
45edfa58 | 4308 | struct location *l = &t.loc[i]; |
88a420e4 | 4309 | |
9c246247 | 4310 | if (len > PAGE_SIZE - KSYM_SYMBOL_LEN - 100) |
88a420e4 | 4311 | break; |
e374d483 | 4312 | len += sprintf(buf + len, "%7ld ", l->count); |
45edfa58 CL |
4313 | |
4314 | if (l->addr) | |
62c70bce | 4315 | len += sprintf(buf + len, "%pS", (void *)l->addr); |
88a420e4 | 4316 | else |
e374d483 | 4317 | len += sprintf(buf + len, "<not-available>"); |
45edfa58 CL |
4318 | |
4319 | if (l->sum_time != l->min_time) { | |
e374d483 | 4320 | len += sprintf(buf + len, " age=%ld/%ld/%ld", |
f8bd2258 RZ |
4321 | l->min_time, |
4322 | (long)div_u64(l->sum_time, l->count), | |
4323 | l->max_time); | |
45edfa58 | 4324 | } else |
e374d483 | 4325 | len += sprintf(buf + len, " age=%ld", |
45edfa58 CL |
4326 | l->min_time); |
4327 | ||
4328 | if (l->min_pid != l->max_pid) | |
e374d483 | 4329 | len += sprintf(buf + len, " pid=%ld-%ld", |
45edfa58 CL |
4330 | l->min_pid, l->max_pid); |
4331 | else | |
e374d483 | 4332 | len += sprintf(buf + len, " pid=%ld", |
45edfa58 CL |
4333 | l->min_pid); |
4334 | ||
174596a0 RR |
4335 | if (num_online_cpus() > 1 && |
4336 | !cpumask_empty(to_cpumask(l->cpus)) && | |
5024c1d7 TH |
4337 | len < PAGE_SIZE - 60) |
4338 | len += scnprintf(buf + len, PAGE_SIZE - len - 50, | |
4339 | " cpus=%*pbl", | |
4340 | cpumask_pr_args(to_cpumask(l->cpus))); | |
45edfa58 | 4341 | |
62bc62a8 | 4342 | if (nr_online_nodes > 1 && !nodes_empty(l->nodes) && |
5024c1d7 TH |
4343 | len < PAGE_SIZE - 60) |
4344 | len += scnprintf(buf + len, PAGE_SIZE - len - 50, | |
4345 | " nodes=%*pbl", | |
4346 | nodemask_pr_args(&l->nodes)); | |
45edfa58 | 4347 | |
e374d483 | 4348 | len += sprintf(buf + len, "\n"); |
88a420e4 CL |
4349 | } |
4350 | ||
4351 | free_loc_track(&t); | |
bbd7d57b | 4352 | kfree(map); |
88a420e4 | 4353 | if (!t.count) |
e374d483 HH |
4354 | len += sprintf(buf, "No data\n"); |
4355 | return len; | |
88a420e4 | 4356 | } |
ab4d5ed5 | 4357 | #endif |
88a420e4 | 4358 | |
a5a84755 | 4359 | #ifdef SLUB_RESILIENCY_TEST |
c07b8183 | 4360 | static void __init resiliency_test(void) |
a5a84755 CL |
4361 | { |
4362 | u8 *p; | |
4363 | ||
95a05b42 | 4364 | BUILD_BUG_ON(KMALLOC_MIN_SIZE > 16 || KMALLOC_SHIFT_HIGH < 10); |
a5a84755 | 4365 | |
f9f58285 FF |
4366 | pr_err("SLUB resiliency testing\n"); |
4367 | pr_err("-----------------------\n"); | |
4368 | pr_err("A. Corruption after allocation\n"); | |
a5a84755 CL |
4369 | |
4370 | p = kzalloc(16, GFP_KERNEL); | |
4371 | p[16] = 0x12; | |
f9f58285 FF |
4372 | pr_err("\n1. kmalloc-16: Clobber Redzone/next pointer 0x12->0x%p\n\n", |
4373 | p + 16); | |
a5a84755 CL |
4374 | |
4375 | validate_slab_cache(kmalloc_caches[4]); | |
4376 | ||
4377 | /* Hmmm... The next two are dangerous */ | |
4378 | p = kzalloc(32, GFP_KERNEL); | |
4379 | p[32 + sizeof(void *)] = 0x34; | |
f9f58285 FF |
4380 | pr_err("\n2. kmalloc-32: Clobber next pointer/next slab 0x34 -> -0x%p\n", |
4381 | p); | |
4382 | pr_err("If allocated object is overwritten then not detectable\n\n"); | |
a5a84755 CL |
4383 | |
4384 | validate_slab_cache(kmalloc_caches[5]); | |
4385 | p = kzalloc(64, GFP_KERNEL); | |
4386 | p += 64 + (get_cycles() & 0xff) * sizeof(void *); | |
4387 | *p = 0x56; | |
f9f58285 FF |
4388 | pr_err("\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n", |
4389 | p); | |
4390 | pr_err("If allocated object is overwritten then not detectable\n\n"); | |
a5a84755 CL |
4391 | validate_slab_cache(kmalloc_caches[6]); |
4392 | ||
f9f58285 | 4393 | pr_err("\nB. Corruption after free\n"); |
a5a84755 CL |
4394 | p = kzalloc(128, GFP_KERNEL); |
4395 | kfree(p); | |
4396 | *p = 0x78; | |
f9f58285 | 4397 | pr_err("1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p); |
a5a84755 CL |
4398 | validate_slab_cache(kmalloc_caches[7]); |
4399 | ||
4400 | p = kzalloc(256, GFP_KERNEL); | |
4401 | kfree(p); | |
4402 | p[50] = 0x9a; | |
f9f58285 | 4403 | pr_err("\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n", p); |
a5a84755 CL |
4404 | validate_slab_cache(kmalloc_caches[8]); |
4405 | ||
4406 | p = kzalloc(512, GFP_KERNEL); | |
4407 | kfree(p); | |
4408 | p[512] = 0xab; | |
f9f58285 | 4409 | pr_err("\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p); |
a5a84755 CL |
4410 | validate_slab_cache(kmalloc_caches[9]); |
4411 | } | |
4412 | #else | |
4413 | #ifdef CONFIG_SYSFS | |
4414 | static void resiliency_test(void) {}; | |
4415 | #endif | |
4416 | #endif | |
4417 | ||
ab4d5ed5 | 4418 | #ifdef CONFIG_SYSFS |
81819f0f | 4419 | enum slab_stat_type { |
205ab99d CL |
4420 | SL_ALL, /* All slabs */ |
4421 | SL_PARTIAL, /* Only partially allocated slabs */ | |
4422 | SL_CPU, /* Only slabs used for cpu caches */ | |
4423 | SL_OBJECTS, /* Determine allocated objects not slabs */ | |
4424 | SL_TOTAL /* Determine object capacity not slabs */ | |
81819f0f CL |
4425 | }; |
4426 | ||
205ab99d | 4427 | #define SO_ALL (1 << SL_ALL) |
81819f0f CL |
4428 | #define SO_PARTIAL (1 << SL_PARTIAL) |
4429 | #define SO_CPU (1 << SL_CPU) | |
4430 | #define SO_OBJECTS (1 << SL_OBJECTS) | |
205ab99d | 4431 | #define SO_TOTAL (1 << SL_TOTAL) |
81819f0f | 4432 | |
62e5c4b4 CG |
4433 | static ssize_t show_slab_objects(struct kmem_cache *s, |
4434 | char *buf, unsigned long flags) | |
81819f0f CL |
4435 | { |
4436 | unsigned long total = 0; | |
81819f0f CL |
4437 | int node; |
4438 | int x; | |
4439 | unsigned long *nodes; | |
81819f0f | 4440 | |
e35e1a97 | 4441 | nodes = kzalloc(sizeof(unsigned long) * nr_node_ids, GFP_KERNEL); |
62e5c4b4 CG |
4442 | if (!nodes) |
4443 | return -ENOMEM; | |
81819f0f | 4444 | |
205ab99d CL |
4445 | if (flags & SO_CPU) { |
4446 | int cpu; | |
81819f0f | 4447 | |
205ab99d | 4448 | for_each_possible_cpu(cpu) { |
d0e0ac97 CG |
4449 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, |
4450 | cpu); | |
ec3ab083 | 4451 | int node; |
49e22585 | 4452 | struct page *page; |
dfb4f096 | 4453 | |
4db0c3c2 | 4454 | page = READ_ONCE(c->page); |
ec3ab083 CL |
4455 | if (!page) |
4456 | continue; | |
205ab99d | 4457 | |
ec3ab083 CL |
4458 | node = page_to_nid(page); |
4459 | if (flags & SO_TOTAL) | |
4460 | x = page->objects; | |
4461 | else if (flags & SO_OBJECTS) | |
4462 | x = page->inuse; | |
4463 | else | |
4464 | x = 1; | |
49e22585 | 4465 | |
ec3ab083 CL |
4466 | total += x; |
4467 | nodes[node] += x; | |
4468 | ||
4db0c3c2 | 4469 | page = READ_ONCE(c->partial); |
49e22585 | 4470 | if (page) { |
8afb1474 LZ |
4471 | node = page_to_nid(page); |
4472 | if (flags & SO_TOTAL) | |
4473 | WARN_ON_ONCE(1); | |
4474 | else if (flags & SO_OBJECTS) | |
4475 | WARN_ON_ONCE(1); | |
4476 | else | |
4477 | x = page->pages; | |
bc6697d8 ED |
4478 | total += x; |
4479 | nodes[node] += x; | |
49e22585 | 4480 | } |
81819f0f CL |
4481 | } |
4482 | } | |
4483 | ||
bfc8c901 | 4484 | get_online_mems(); |
ab4d5ed5 | 4485 | #ifdef CONFIG_SLUB_DEBUG |
205ab99d | 4486 | if (flags & SO_ALL) { |
fa45dc25 CL |
4487 | struct kmem_cache_node *n; |
4488 | ||
4489 | for_each_kmem_cache_node(s, node, n) { | |
205ab99d | 4490 | |
d0e0ac97 CG |
4491 | if (flags & SO_TOTAL) |
4492 | x = atomic_long_read(&n->total_objects); | |
4493 | else if (flags & SO_OBJECTS) | |
4494 | x = atomic_long_read(&n->total_objects) - | |
4495 | count_partial(n, count_free); | |
81819f0f | 4496 | else |
205ab99d | 4497 | x = atomic_long_read(&n->nr_slabs); |
81819f0f CL |
4498 | total += x; |
4499 | nodes[node] += x; | |
4500 | } | |
4501 | ||
ab4d5ed5 CL |
4502 | } else |
4503 | #endif | |
4504 | if (flags & SO_PARTIAL) { | |
fa45dc25 | 4505 | struct kmem_cache_node *n; |
81819f0f | 4506 | |
fa45dc25 | 4507 | for_each_kmem_cache_node(s, node, n) { |
205ab99d CL |
4508 | if (flags & SO_TOTAL) |
4509 | x = count_partial(n, count_total); | |
4510 | else if (flags & SO_OBJECTS) | |
4511 | x = count_partial(n, count_inuse); | |
81819f0f | 4512 | else |
205ab99d | 4513 | x = n->nr_partial; |
81819f0f CL |
4514 | total += x; |
4515 | nodes[node] += x; | |
4516 | } | |
4517 | } | |
81819f0f CL |
4518 | x = sprintf(buf, "%lu", total); |
4519 | #ifdef CONFIG_NUMA | |
fa45dc25 | 4520 | for (node = 0; node < nr_node_ids; node++) |
81819f0f CL |
4521 | if (nodes[node]) |
4522 | x += sprintf(buf + x, " N%d=%lu", | |
4523 | node, nodes[node]); | |
4524 | #endif | |
bfc8c901 | 4525 | put_online_mems(); |
81819f0f CL |
4526 | kfree(nodes); |
4527 | return x + sprintf(buf + x, "\n"); | |
4528 | } | |
4529 | ||
ab4d5ed5 | 4530 | #ifdef CONFIG_SLUB_DEBUG |
81819f0f CL |
4531 | static int any_slab_objects(struct kmem_cache *s) |
4532 | { | |
4533 | int node; | |
fa45dc25 | 4534 | struct kmem_cache_node *n; |
81819f0f | 4535 | |
fa45dc25 | 4536 | for_each_kmem_cache_node(s, node, n) |
4ea33e2d | 4537 | if (atomic_long_read(&n->total_objects)) |
81819f0f | 4538 | return 1; |
fa45dc25 | 4539 | |
81819f0f CL |
4540 | return 0; |
4541 | } | |
ab4d5ed5 | 4542 | #endif |
81819f0f CL |
4543 | |
4544 | #define to_slab_attr(n) container_of(n, struct slab_attribute, attr) | |
497888cf | 4545 | #define to_slab(n) container_of(n, struct kmem_cache, kobj) |
81819f0f CL |
4546 | |
4547 | struct slab_attribute { | |
4548 | struct attribute attr; | |
4549 | ssize_t (*show)(struct kmem_cache *s, char *buf); | |
4550 | ssize_t (*store)(struct kmem_cache *s, const char *x, size_t count); | |
4551 | }; | |
4552 | ||
4553 | #define SLAB_ATTR_RO(_name) \ | |
ab067e99 VK |
4554 | static struct slab_attribute _name##_attr = \ |
4555 | __ATTR(_name, 0400, _name##_show, NULL) | |
81819f0f CL |
4556 | |
4557 | #define SLAB_ATTR(_name) \ | |
4558 | static struct slab_attribute _name##_attr = \ | |
ab067e99 | 4559 | __ATTR(_name, 0600, _name##_show, _name##_store) |
81819f0f | 4560 | |
81819f0f CL |
4561 | static ssize_t slab_size_show(struct kmem_cache *s, char *buf) |
4562 | { | |
4563 | return sprintf(buf, "%d\n", s->size); | |
4564 | } | |
4565 | SLAB_ATTR_RO(slab_size); | |
4566 | ||
4567 | static ssize_t align_show(struct kmem_cache *s, char *buf) | |
4568 | { | |
4569 | return sprintf(buf, "%d\n", s->align); | |
4570 | } | |
4571 | SLAB_ATTR_RO(align); | |
4572 | ||
4573 | static ssize_t object_size_show(struct kmem_cache *s, char *buf) | |
4574 | { | |
3b0efdfa | 4575 | return sprintf(buf, "%d\n", s->object_size); |
81819f0f CL |
4576 | } |
4577 | SLAB_ATTR_RO(object_size); | |
4578 | ||
4579 | static ssize_t objs_per_slab_show(struct kmem_cache *s, char *buf) | |
4580 | { | |
834f3d11 | 4581 | return sprintf(buf, "%d\n", oo_objects(s->oo)); |
81819f0f CL |
4582 | } |
4583 | SLAB_ATTR_RO(objs_per_slab); | |
4584 | ||
06b285dc CL |
4585 | static ssize_t order_store(struct kmem_cache *s, |
4586 | const char *buf, size_t length) | |
4587 | { | |
0121c619 CL |
4588 | unsigned long order; |
4589 | int err; | |
4590 | ||
3dbb95f7 | 4591 | err = kstrtoul(buf, 10, &order); |
0121c619 CL |
4592 | if (err) |
4593 | return err; | |
06b285dc CL |
4594 | |
4595 | if (order > slub_max_order || order < slub_min_order) | |
4596 | return -EINVAL; | |
4597 | ||
4598 | calculate_sizes(s, order); | |
4599 | return length; | |
4600 | } | |
4601 | ||
81819f0f CL |
4602 | static ssize_t order_show(struct kmem_cache *s, char *buf) |
4603 | { | |
834f3d11 | 4604 | return sprintf(buf, "%d\n", oo_order(s->oo)); |
81819f0f | 4605 | } |
06b285dc | 4606 | SLAB_ATTR(order); |
81819f0f | 4607 | |
73d342b1 DR |
4608 | static ssize_t min_partial_show(struct kmem_cache *s, char *buf) |
4609 | { | |
4610 | return sprintf(buf, "%lu\n", s->min_partial); | |
4611 | } | |
4612 | ||
4613 | static ssize_t min_partial_store(struct kmem_cache *s, const char *buf, | |
4614 | size_t length) | |
4615 | { | |
4616 | unsigned long min; | |
4617 | int err; | |
4618 | ||
3dbb95f7 | 4619 | err = kstrtoul(buf, 10, &min); |
73d342b1 DR |
4620 | if (err) |
4621 | return err; | |
4622 | ||
c0bdb232 | 4623 | set_min_partial(s, min); |
73d342b1 DR |
4624 | return length; |
4625 | } | |
4626 | SLAB_ATTR(min_partial); | |
4627 | ||
49e22585 CL |
4628 | static ssize_t cpu_partial_show(struct kmem_cache *s, char *buf) |
4629 | { | |
4630 | return sprintf(buf, "%u\n", s->cpu_partial); | |
4631 | } | |
4632 | ||
4633 | static ssize_t cpu_partial_store(struct kmem_cache *s, const char *buf, | |
4634 | size_t length) | |
4635 | { | |
4636 | unsigned long objects; | |
4637 | int err; | |
4638 | ||
3dbb95f7 | 4639 | err = kstrtoul(buf, 10, &objects); |
49e22585 CL |
4640 | if (err) |
4641 | return err; | |
345c905d | 4642 | if (objects && !kmem_cache_has_cpu_partial(s)) |
74ee4ef1 | 4643 | return -EINVAL; |
49e22585 CL |
4644 | |
4645 | s->cpu_partial = objects; | |
4646 | flush_all(s); | |
4647 | return length; | |
4648 | } | |
4649 | SLAB_ATTR(cpu_partial); | |
4650 | ||
81819f0f CL |
4651 | static ssize_t ctor_show(struct kmem_cache *s, char *buf) |
4652 | { | |
62c70bce JP |
4653 | if (!s->ctor) |
4654 | return 0; | |
4655 | return sprintf(buf, "%pS\n", s->ctor); | |
81819f0f CL |
4656 | } |
4657 | SLAB_ATTR_RO(ctor); | |
4658 | ||
81819f0f CL |
4659 | static ssize_t aliases_show(struct kmem_cache *s, char *buf) |
4660 | { | |
4307c14f | 4661 | return sprintf(buf, "%d\n", s->refcount < 0 ? 0 : s->refcount - 1); |
81819f0f CL |
4662 | } |
4663 | SLAB_ATTR_RO(aliases); | |
4664 | ||
81819f0f CL |
4665 | static ssize_t partial_show(struct kmem_cache *s, char *buf) |
4666 | { | |
d9acf4b7 | 4667 | return show_slab_objects(s, buf, SO_PARTIAL); |
81819f0f CL |
4668 | } |
4669 | SLAB_ATTR_RO(partial); | |
4670 | ||
4671 | static ssize_t cpu_slabs_show(struct kmem_cache *s, char *buf) | |
4672 | { | |
d9acf4b7 | 4673 | return show_slab_objects(s, buf, SO_CPU); |
81819f0f CL |
4674 | } |
4675 | SLAB_ATTR_RO(cpu_slabs); | |
4676 | ||
4677 | static ssize_t objects_show(struct kmem_cache *s, char *buf) | |
4678 | { | |
205ab99d | 4679 | return show_slab_objects(s, buf, SO_ALL|SO_OBJECTS); |
81819f0f CL |
4680 | } |
4681 | SLAB_ATTR_RO(objects); | |
4682 | ||
205ab99d CL |
4683 | static ssize_t objects_partial_show(struct kmem_cache *s, char *buf) |
4684 | { | |
4685 | return show_slab_objects(s, buf, SO_PARTIAL|SO_OBJECTS); | |
4686 | } | |
4687 | SLAB_ATTR_RO(objects_partial); | |
4688 | ||
49e22585 CL |
4689 | static ssize_t slabs_cpu_partial_show(struct kmem_cache *s, char *buf) |
4690 | { | |
4691 | int objects = 0; | |
4692 | int pages = 0; | |
4693 | int cpu; | |
4694 | int len; | |
4695 | ||
4696 | for_each_online_cpu(cpu) { | |
4697 | struct page *page = per_cpu_ptr(s->cpu_slab, cpu)->partial; | |
4698 | ||
4699 | if (page) { | |
4700 | pages += page->pages; | |
4701 | objects += page->pobjects; | |
4702 | } | |
4703 | } | |
4704 | ||
4705 | len = sprintf(buf, "%d(%d)", objects, pages); | |
4706 | ||
4707 | #ifdef CONFIG_SMP | |
4708 | for_each_online_cpu(cpu) { | |
4709 | struct page *page = per_cpu_ptr(s->cpu_slab, cpu) ->partial; | |
4710 | ||
4711 | if (page && len < PAGE_SIZE - 20) | |
4712 | len += sprintf(buf + len, " C%d=%d(%d)", cpu, | |
4713 | page->pobjects, page->pages); | |
4714 | } | |
4715 | #endif | |
4716 | return len + sprintf(buf + len, "\n"); | |
4717 | } | |
4718 | SLAB_ATTR_RO(slabs_cpu_partial); | |
4719 | ||
a5a84755 CL |
4720 | static ssize_t reclaim_account_show(struct kmem_cache *s, char *buf) |
4721 | { | |
4722 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_RECLAIM_ACCOUNT)); | |
4723 | } | |
4724 | ||
4725 | static ssize_t reclaim_account_store(struct kmem_cache *s, | |
4726 | const char *buf, size_t length) | |
4727 | { | |
4728 | s->flags &= ~SLAB_RECLAIM_ACCOUNT; | |
4729 | if (buf[0] == '1') | |
4730 | s->flags |= SLAB_RECLAIM_ACCOUNT; | |
4731 | return length; | |
4732 | } | |
4733 | SLAB_ATTR(reclaim_account); | |
4734 | ||
4735 | static ssize_t hwcache_align_show(struct kmem_cache *s, char *buf) | |
4736 | { | |
4737 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_HWCACHE_ALIGN)); | |
4738 | } | |
4739 | SLAB_ATTR_RO(hwcache_align); | |
4740 | ||
4741 | #ifdef CONFIG_ZONE_DMA | |
4742 | static ssize_t cache_dma_show(struct kmem_cache *s, char *buf) | |
4743 | { | |
4744 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_CACHE_DMA)); | |
4745 | } | |
4746 | SLAB_ATTR_RO(cache_dma); | |
4747 | #endif | |
4748 | ||
4749 | static ssize_t destroy_by_rcu_show(struct kmem_cache *s, char *buf) | |
4750 | { | |
4751 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_DESTROY_BY_RCU)); | |
4752 | } | |
4753 | SLAB_ATTR_RO(destroy_by_rcu); | |
4754 | ||
ab9a0f19 LJ |
4755 | static ssize_t reserved_show(struct kmem_cache *s, char *buf) |
4756 | { | |
4757 | return sprintf(buf, "%d\n", s->reserved); | |
4758 | } | |
4759 | SLAB_ATTR_RO(reserved); | |
4760 | ||
ab4d5ed5 | 4761 | #ifdef CONFIG_SLUB_DEBUG |
a5a84755 CL |
4762 | static ssize_t slabs_show(struct kmem_cache *s, char *buf) |
4763 | { | |
4764 | return show_slab_objects(s, buf, SO_ALL); | |
4765 | } | |
4766 | SLAB_ATTR_RO(slabs); | |
4767 | ||
205ab99d CL |
4768 | static ssize_t total_objects_show(struct kmem_cache *s, char *buf) |
4769 | { | |
4770 | return show_slab_objects(s, buf, SO_ALL|SO_TOTAL); | |
4771 | } | |
4772 | SLAB_ATTR_RO(total_objects); | |
4773 | ||
81819f0f CL |
4774 | static ssize_t sanity_checks_show(struct kmem_cache *s, char *buf) |
4775 | { | |
4776 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_DEBUG_FREE)); | |
4777 | } | |
4778 | ||
4779 | static ssize_t sanity_checks_store(struct kmem_cache *s, | |
4780 | const char *buf, size_t length) | |
4781 | { | |
4782 | s->flags &= ~SLAB_DEBUG_FREE; | |
b789ef51 CL |
4783 | if (buf[0] == '1') { |
4784 | s->flags &= ~__CMPXCHG_DOUBLE; | |
81819f0f | 4785 | s->flags |= SLAB_DEBUG_FREE; |
b789ef51 | 4786 | } |
81819f0f CL |
4787 | return length; |
4788 | } | |
4789 | SLAB_ATTR(sanity_checks); | |
4790 | ||
4791 | static ssize_t trace_show(struct kmem_cache *s, char *buf) | |
4792 | { | |
4793 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_TRACE)); | |
4794 | } | |
4795 | ||
4796 | static ssize_t trace_store(struct kmem_cache *s, const char *buf, | |
4797 | size_t length) | |
4798 | { | |
c9e16131 CL |
4799 | /* |
4800 | * Tracing a merged cache is going to give confusing results | |
4801 | * as well as cause other issues like converting a mergeable | |
4802 | * cache into an umergeable one. | |
4803 | */ | |
4804 | if (s->refcount > 1) | |
4805 | return -EINVAL; | |
4806 | ||
81819f0f | 4807 | s->flags &= ~SLAB_TRACE; |
b789ef51 CL |
4808 | if (buf[0] == '1') { |
4809 | s->flags &= ~__CMPXCHG_DOUBLE; | |
81819f0f | 4810 | s->flags |= SLAB_TRACE; |
b789ef51 | 4811 | } |
81819f0f CL |
4812 | return length; |
4813 | } | |
4814 | SLAB_ATTR(trace); | |
4815 | ||
81819f0f CL |
4816 | static ssize_t red_zone_show(struct kmem_cache *s, char *buf) |
4817 | { | |
4818 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_RED_ZONE)); | |
4819 | } | |
4820 | ||
4821 | static ssize_t red_zone_store(struct kmem_cache *s, | |
4822 | const char *buf, size_t length) | |
4823 | { | |
4824 | if (any_slab_objects(s)) | |
4825 | return -EBUSY; | |
4826 | ||
4827 | s->flags &= ~SLAB_RED_ZONE; | |
b789ef51 CL |
4828 | if (buf[0] == '1') { |
4829 | s->flags &= ~__CMPXCHG_DOUBLE; | |
81819f0f | 4830 | s->flags |= SLAB_RED_ZONE; |
b789ef51 | 4831 | } |
06b285dc | 4832 | calculate_sizes(s, -1); |
81819f0f CL |
4833 | return length; |
4834 | } | |
4835 | SLAB_ATTR(red_zone); | |
4836 | ||
4837 | static ssize_t poison_show(struct kmem_cache *s, char *buf) | |
4838 | { | |
4839 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_POISON)); | |
4840 | } | |
4841 | ||
4842 | static ssize_t poison_store(struct kmem_cache *s, | |
4843 | const char *buf, size_t length) | |
4844 | { | |
4845 | if (any_slab_objects(s)) | |
4846 | return -EBUSY; | |
4847 | ||
4848 | s->flags &= ~SLAB_POISON; | |
b789ef51 CL |
4849 | if (buf[0] == '1') { |
4850 | s->flags &= ~__CMPXCHG_DOUBLE; | |
81819f0f | 4851 | s->flags |= SLAB_POISON; |
b789ef51 | 4852 | } |
06b285dc | 4853 | calculate_sizes(s, -1); |
81819f0f CL |
4854 | return length; |
4855 | } | |
4856 | SLAB_ATTR(poison); | |
4857 | ||
4858 | static ssize_t store_user_show(struct kmem_cache *s, char *buf) | |
4859 | { | |
4860 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_STORE_USER)); | |
4861 | } | |
4862 | ||
4863 | static ssize_t store_user_store(struct kmem_cache *s, | |
4864 | const char *buf, size_t length) | |
4865 | { | |
4866 | if (any_slab_objects(s)) | |
4867 | return -EBUSY; | |
4868 | ||
4869 | s->flags &= ~SLAB_STORE_USER; | |
b789ef51 CL |
4870 | if (buf[0] == '1') { |
4871 | s->flags &= ~__CMPXCHG_DOUBLE; | |
81819f0f | 4872 | s->flags |= SLAB_STORE_USER; |
b789ef51 | 4873 | } |
06b285dc | 4874 | calculate_sizes(s, -1); |
81819f0f CL |
4875 | return length; |
4876 | } | |
4877 | SLAB_ATTR(store_user); | |
4878 | ||
53e15af0 CL |
4879 | static ssize_t validate_show(struct kmem_cache *s, char *buf) |
4880 | { | |
4881 | return 0; | |
4882 | } | |
4883 | ||
4884 | static ssize_t validate_store(struct kmem_cache *s, | |
4885 | const char *buf, size_t length) | |
4886 | { | |
434e245d CL |
4887 | int ret = -EINVAL; |
4888 | ||
4889 | if (buf[0] == '1') { | |
4890 | ret = validate_slab_cache(s); | |
4891 | if (ret >= 0) | |
4892 | ret = length; | |
4893 | } | |
4894 | return ret; | |
53e15af0 CL |
4895 | } |
4896 | SLAB_ATTR(validate); | |
a5a84755 CL |
4897 | |
4898 | static ssize_t alloc_calls_show(struct kmem_cache *s, char *buf) | |
4899 | { | |
4900 | if (!(s->flags & SLAB_STORE_USER)) | |
4901 | return -ENOSYS; | |
4902 | return list_locations(s, buf, TRACK_ALLOC); | |
4903 | } | |
4904 | SLAB_ATTR_RO(alloc_calls); | |
4905 | ||
4906 | static ssize_t free_calls_show(struct kmem_cache *s, char *buf) | |
4907 | { | |
4908 | if (!(s->flags & SLAB_STORE_USER)) | |
4909 | return -ENOSYS; | |
4910 | return list_locations(s, buf, TRACK_FREE); | |
4911 | } | |
4912 | SLAB_ATTR_RO(free_calls); | |
4913 | #endif /* CONFIG_SLUB_DEBUG */ | |
4914 | ||
4915 | #ifdef CONFIG_FAILSLAB | |
4916 | static ssize_t failslab_show(struct kmem_cache *s, char *buf) | |
4917 | { | |
4918 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_FAILSLAB)); | |
4919 | } | |
4920 | ||
4921 | static ssize_t failslab_store(struct kmem_cache *s, const char *buf, | |
4922 | size_t length) | |
4923 | { | |
c9e16131 CL |
4924 | if (s->refcount > 1) |
4925 | return -EINVAL; | |
4926 | ||
a5a84755 CL |
4927 | s->flags &= ~SLAB_FAILSLAB; |
4928 | if (buf[0] == '1') | |
4929 | s->flags |= SLAB_FAILSLAB; | |
4930 | return length; | |
4931 | } | |
4932 | SLAB_ATTR(failslab); | |
ab4d5ed5 | 4933 | #endif |
53e15af0 | 4934 | |
2086d26a CL |
4935 | static ssize_t shrink_show(struct kmem_cache *s, char *buf) |
4936 | { | |
4937 | return 0; | |
4938 | } | |
4939 | ||
4940 | static ssize_t shrink_store(struct kmem_cache *s, | |
4941 | const char *buf, size_t length) | |
4942 | { | |
832f37f5 VD |
4943 | if (buf[0] == '1') |
4944 | kmem_cache_shrink(s); | |
4945 | else | |
2086d26a CL |
4946 | return -EINVAL; |
4947 | return length; | |
4948 | } | |
4949 | SLAB_ATTR(shrink); | |
4950 | ||
81819f0f | 4951 | #ifdef CONFIG_NUMA |
9824601e | 4952 | static ssize_t remote_node_defrag_ratio_show(struct kmem_cache *s, char *buf) |
81819f0f | 4953 | { |
9824601e | 4954 | return sprintf(buf, "%d\n", s->remote_node_defrag_ratio / 10); |
81819f0f CL |
4955 | } |
4956 | ||
9824601e | 4957 | static ssize_t remote_node_defrag_ratio_store(struct kmem_cache *s, |
81819f0f CL |
4958 | const char *buf, size_t length) |
4959 | { | |
0121c619 CL |
4960 | unsigned long ratio; |
4961 | int err; | |
4962 | ||
3dbb95f7 | 4963 | err = kstrtoul(buf, 10, &ratio); |
0121c619 CL |
4964 | if (err) |
4965 | return err; | |
4966 | ||
e2cb96b7 | 4967 | if (ratio <= 100) |
0121c619 | 4968 | s->remote_node_defrag_ratio = ratio * 10; |
81819f0f | 4969 | |
81819f0f CL |
4970 | return length; |
4971 | } | |
9824601e | 4972 | SLAB_ATTR(remote_node_defrag_ratio); |
81819f0f CL |
4973 | #endif |
4974 | ||
8ff12cfc | 4975 | #ifdef CONFIG_SLUB_STATS |
8ff12cfc CL |
4976 | static int show_stat(struct kmem_cache *s, char *buf, enum stat_item si) |
4977 | { | |
4978 | unsigned long sum = 0; | |
4979 | int cpu; | |
4980 | int len; | |
4981 | int *data = kmalloc(nr_cpu_ids * sizeof(int), GFP_KERNEL); | |
4982 | ||
4983 | if (!data) | |
4984 | return -ENOMEM; | |
4985 | ||
4986 | for_each_online_cpu(cpu) { | |
9dfc6e68 | 4987 | unsigned x = per_cpu_ptr(s->cpu_slab, cpu)->stat[si]; |
8ff12cfc CL |
4988 | |
4989 | data[cpu] = x; | |
4990 | sum += x; | |
4991 | } | |
4992 | ||
4993 | len = sprintf(buf, "%lu", sum); | |
4994 | ||
50ef37b9 | 4995 | #ifdef CONFIG_SMP |
8ff12cfc CL |
4996 | for_each_online_cpu(cpu) { |
4997 | if (data[cpu] && len < PAGE_SIZE - 20) | |
50ef37b9 | 4998 | len += sprintf(buf + len, " C%d=%u", cpu, data[cpu]); |
8ff12cfc | 4999 | } |
50ef37b9 | 5000 | #endif |
8ff12cfc CL |
5001 | kfree(data); |
5002 | return len + sprintf(buf + len, "\n"); | |
5003 | } | |
5004 | ||
78eb00cc DR |
5005 | static void clear_stat(struct kmem_cache *s, enum stat_item si) |
5006 | { | |
5007 | int cpu; | |
5008 | ||
5009 | for_each_online_cpu(cpu) | |
9dfc6e68 | 5010 | per_cpu_ptr(s->cpu_slab, cpu)->stat[si] = 0; |
78eb00cc DR |
5011 | } |
5012 | ||
8ff12cfc CL |
5013 | #define STAT_ATTR(si, text) \ |
5014 | static ssize_t text##_show(struct kmem_cache *s, char *buf) \ | |
5015 | { \ | |
5016 | return show_stat(s, buf, si); \ | |
5017 | } \ | |
78eb00cc DR |
5018 | static ssize_t text##_store(struct kmem_cache *s, \ |
5019 | const char *buf, size_t length) \ | |
5020 | { \ | |
5021 | if (buf[0] != '0') \ | |
5022 | return -EINVAL; \ | |
5023 | clear_stat(s, si); \ | |
5024 | return length; \ | |
5025 | } \ | |
5026 | SLAB_ATTR(text); \ | |
8ff12cfc CL |
5027 | |
5028 | STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath); | |
5029 | STAT_ATTR(ALLOC_SLOWPATH, alloc_slowpath); | |
5030 | STAT_ATTR(FREE_FASTPATH, free_fastpath); | |
5031 | STAT_ATTR(FREE_SLOWPATH, free_slowpath); | |
5032 | STAT_ATTR(FREE_FROZEN, free_frozen); | |
5033 | STAT_ATTR(FREE_ADD_PARTIAL, free_add_partial); | |
5034 | STAT_ATTR(FREE_REMOVE_PARTIAL, free_remove_partial); | |
5035 | STAT_ATTR(ALLOC_FROM_PARTIAL, alloc_from_partial); | |
5036 | STAT_ATTR(ALLOC_SLAB, alloc_slab); | |
5037 | STAT_ATTR(ALLOC_REFILL, alloc_refill); | |
e36a2652 | 5038 | STAT_ATTR(ALLOC_NODE_MISMATCH, alloc_node_mismatch); |
8ff12cfc CL |
5039 | STAT_ATTR(FREE_SLAB, free_slab); |
5040 | STAT_ATTR(CPUSLAB_FLUSH, cpuslab_flush); | |
5041 | STAT_ATTR(DEACTIVATE_FULL, deactivate_full); | |
5042 | STAT_ATTR(DEACTIVATE_EMPTY, deactivate_empty); | |
5043 | STAT_ATTR(DEACTIVATE_TO_HEAD, deactivate_to_head); | |
5044 | STAT_ATTR(DEACTIVATE_TO_TAIL, deactivate_to_tail); | |
5045 | STAT_ATTR(DEACTIVATE_REMOTE_FREES, deactivate_remote_frees); | |
03e404af | 5046 | STAT_ATTR(DEACTIVATE_BYPASS, deactivate_bypass); |
65c3376a | 5047 | STAT_ATTR(ORDER_FALLBACK, order_fallback); |
b789ef51 CL |
5048 | STAT_ATTR(CMPXCHG_DOUBLE_CPU_FAIL, cmpxchg_double_cpu_fail); |
5049 | STAT_ATTR(CMPXCHG_DOUBLE_FAIL, cmpxchg_double_fail); | |
49e22585 CL |
5050 | STAT_ATTR(CPU_PARTIAL_ALLOC, cpu_partial_alloc); |
5051 | STAT_ATTR(CPU_PARTIAL_FREE, cpu_partial_free); | |
8028dcea AS |
5052 | STAT_ATTR(CPU_PARTIAL_NODE, cpu_partial_node); |
5053 | STAT_ATTR(CPU_PARTIAL_DRAIN, cpu_partial_drain); | |
8ff12cfc CL |
5054 | #endif |
5055 | ||
06428780 | 5056 | static struct attribute *slab_attrs[] = { |
81819f0f CL |
5057 | &slab_size_attr.attr, |
5058 | &object_size_attr.attr, | |
5059 | &objs_per_slab_attr.attr, | |
5060 | &order_attr.attr, | |
73d342b1 | 5061 | &min_partial_attr.attr, |
49e22585 | 5062 | &cpu_partial_attr.attr, |
81819f0f | 5063 | &objects_attr.attr, |
205ab99d | 5064 | &objects_partial_attr.attr, |
81819f0f CL |
5065 | &partial_attr.attr, |
5066 | &cpu_slabs_attr.attr, | |
5067 | &ctor_attr.attr, | |
81819f0f CL |
5068 | &aliases_attr.attr, |
5069 | &align_attr.attr, | |
81819f0f CL |
5070 | &hwcache_align_attr.attr, |
5071 | &reclaim_account_attr.attr, | |
5072 | &destroy_by_rcu_attr.attr, | |
a5a84755 | 5073 | &shrink_attr.attr, |
ab9a0f19 | 5074 | &reserved_attr.attr, |
49e22585 | 5075 | &slabs_cpu_partial_attr.attr, |
ab4d5ed5 | 5076 | #ifdef CONFIG_SLUB_DEBUG |
a5a84755 CL |
5077 | &total_objects_attr.attr, |
5078 | &slabs_attr.attr, | |
5079 | &sanity_checks_attr.attr, | |
5080 | &trace_attr.attr, | |
81819f0f CL |
5081 | &red_zone_attr.attr, |
5082 | &poison_attr.attr, | |
5083 | &store_user_attr.attr, | |
53e15af0 | 5084 | &validate_attr.attr, |
88a420e4 CL |
5085 | &alloc_calls_attr.attr, |
5086 | &free_calls_attr.attr, | |
ab4d5ed5 | 5087 | #endif |
81819f0f CL |
5088 | #ifdef CONFIG_ZONE_DMA |
5089 | &cache_dma_attr.attr, | |
5090 | #endif | |
5091 | #ifdef CONFIG_NUMA | |
9824601e | 5092 | &remote_node_defrag_ratio_attr.attr, |
8ff12cfc CL |
5093 | #endif |
5094 | #ifdef CONFIG_SLUB_STATS | |
5095 | &alloc_fastpath_attr.attr, | |
5096 | &alloc_slowpath_attr.attr, | |
5097 | &free_fastpath_attr.attr, | |
5098 | &free_slowpath_attr.attr, | |
5099 | &free_frozen_attr.attr, | |
5100 | &free_add_partial_attr.attr, | |
5101 | &free_remove_partial_attr.attr, | |
5102 | &alloc_from_partial_attr.attr, | |
5103 | &alloc_slab_attr.attr, | |
5104 | &alloc_refill_attr.attr, | |
e36a2652 | 5105 | &alloc_node_mismatch_attr.attr, |
8ff12cfc CL |
5106 | &free_slab_attr.attr, |
5107 | &cpuslab_flush_attr.attr, | |
5108 | &deactivate_full_attr.attr, | |
5109 | &deactivate_empty_attr.attr, | |
5110 | &deactivate_to_head_attr.attr, | |
5111 | &deactivate_to_tail_attr.attr, | |
5112 | &deactivate_remote_frees_attr.attr, | |
03e404af | 5113 | &deactivate_bypass_attr.attr, |
65c3376a | 5114 | &order_fallback_attr.attr, |
b789ef51 CL |
5115 | &cmpxchg_double_fail_attr.attr, |
5116 | &cmpxchg_double_cpu_fail_attr.attr, | |
49e22585 CL |
5117 | &cpu_partial_alloc_attr.attr, |
5118 | &cpu_partial_free_attr.attr, | |
8028dcea AS |
5119 | &cpu_partial_node_attr.attr, |
5120 | &cpu_partial_drain_attr.attr, | |
81819f0f | 5121 | #endif |
4c13dd3b DM |
5122 | #ifdef CONFIG_FAILSLAB |
5123 | &failslab_attr.attr, | |
5124 | #endif | |
5125 | ||
81819f0f CL |
5126 | NULL |
5127 | }; | |
5128 | ||
5129 | static struct attribute_group slab_attr_group = { | |
5130 | .attrs = slab_attrs, | |
5131 | }; | |
5132 | ||
5133 | static ssize_t slab_attr_show(struct kobject *kobj, | |
5134 | struct attribute *attr, | |
5135 | char *buf) | |
5136 | { | |
5137 | struct slab_attribute *attribute; | |
5138 | struct kmem_cache *s; | |
5139 | int err; | |
5140 | ||
5141 | attribute = to_slab_attr(attr); | |
5142 | s = to_slab(kobj); | |
5143 | ||
5144 | if (!attribute->show) | |
5145 | return -EIO; | |
5146 | ||
5147 | err = attribute->show(s, buf); | |
5148 | ||
5149 | return err; | |
5150 | } | |
5151 | ||
5152 | static ssize_t slab_attr_store(struct kobject *kobj, | |
5153 | struct attribute *attr, | |
5154 | const char *buf, size_t len) | |
5155 | { | |
5156 | struct slab_attribute *attribute; | |
5157 | struct kmem_cache *s; | |
5158 | int err; | |
5159 | ||
5160 | attribute = to_slab_attr(attr); | |
5161 | s = to_slab(kobj); | |
5162 | ||
5163 | if (!attribute->store) | |
5164 | return -EIO; | |
5165 | ||
5166 | err = attribute->store(s, buf, len); | |
127424c8 | 5167 | #ifdef CONFIG_MEMCG |
107dab5c | 5168 | if (slab_state >= FULL && err >= 0 && is_root_cache(s)) { |
426589f5 | 5169 | struct kmem_cache *c; |
81819f0f | 5170 | |
107dab5c GC |
5171 | mutex_lock(&slab_mutex); |
5172 | if (s->max_attr_size < len) | |
5173 | s->max_attr_size = len; | |
5174 | ||
ebe945c2 GC |
5175 | /* |
5176 | * This is a best effort propagation, so this function's return | |
5177 | * value will be determined by the parent cache only. This is | |
5178 | * basically because not all attributes will have a well | |
5179 | * defined semantics for rollbacks - most of the actions will | |
5180 | * have permanent effects. | |
5181 | * | |
5182 | * Returning the error value of any of the children that fail | |
5183 | * is not 100 % defined, in the sense that users seeing the | |
5184 | * error code won't be able to know anything about the state of | |
5185 | * the cache. | |
5186 | * | |
5187 | * Only returning the error code for the parent cache at least | |
5188 | * has well defined semantics. The cache being written to | |
5189 | * directly either failed or succeeded, in which case we loop | |
5190 | * through the descendants with best-effort propagation. | |
5191 | */ | |
426589f5 VD |
5192 | for_each_memcg_cache(c, s) |
5193 | attribute->store(c, buf, len); | |
107dab5c GC |
5194 | mutex_unlock(&slab_mutex); |
5195 | } | |
5196 | #endif | |
81819f0f CL |
5197 | return err; |
5198 | } | |
5199 | ||
107dab5c GC |
5200 | static void memcg_propagate_slab_attrs(struct kmem_cache *s) |
5201 | { | |
127424c8 | 5202 | #ifdef CONFIG_MEMCG |
107dab5c GC |
5203 | int i; |
5204 | char *buffer = NULL; | |
93030d83 | 5205 | struct kmem_cache *root_cache; |
107dab5c | 5206 | |
93030d83 | 5207 | if (is_root_cache(s)) |
107dab5c GC |
5208 | return; |
5209 | ||
f7ce3190 | 5210 | root_cache = s->memcg_params.root_cache; |
93030d83 | 5211 | |
107dab5c GC |
5212 | /* |
5213 | * This mean this cache had no attribute written. Therefore, no point | |
5214 | * in copying default values around | |
5215 | */ | |
93030d83 | 5216 | if (!root_cache->max_attr_size) |
107dab5c GC |
5217 | return; |
5218 | ||
5219 | for (i = 0; i < ARRAY_SIZE(slab_attrs); i++) { | |
5220 | char mbuf[64]; | |
5221 | char *buf; | |
5222 | struct slab_attribute *attr = to_slab_attr(slab_attrs[i]); | |
5223 | ||
5224 | if (!attr || !attr->store || !attr->show) | |
5225 | continue; | |
5226 | ||
5227 | /* | |
5228 | * It is really bad that we have to allocate here, so we will | |
5229 | * do it only as a fallback. If we actually allocate, though, | |
5230 | * we can just use the allocated buffer until the end. | |
5231 | * | |
5232 | * Most of the slub attributes will tend to be very small in | |
5233 | * size, but sysfs allows buffers up to a page, so they can | |
5234 | * theoretically happen. | |
5235 | */ | |
5236 | if (buffer) | |
5237 | buf = buffer; | |
93030d83 | 5238 | else if (root_cache->max_attr_size < ARRAY_SIZE(mbuf)) |
107dab5c GC |
5239 | buf = mbuf; |
5240 | else { | |
5241 | buffer = (char *) get_zeroed_page(GFP_KERNEL); | |
5242 | if (WARN_ON(!buffer)) | |
5243 | continue; | |
5244 | buf = buffer; | |
5245 | } | |
5246 | ||
93030d83 | 5247 | attr->show(root_cache, buf); |
107dab5c GC |
5248 | attr->store(s, buf, strlen(buf)); |
5249 | } | |
5250 | ||
5251 | if (buffer) | |
5252 | free_page((unsigned long)buffer); | |
5253 | #endif | |
5254 | } | |
5255 | ||
41a21285 CL |
5256 | static void kmem_cache_release(struct kobject *k) |
5257 | { | |
5258 | slab_kmem_cache_release(to_slab(k)); | |
5259 | } | |
5260 | ||
52cf25d0 | 5261 | static const struct sysfs_ops slab_sysfs_ops = { |
81819f0f CL |
5262 | .show = slab_attr_show, |
5263 | .store = slab_attr_store, | |
5264 | }; | |
5265 | ||
5266 | static struct kobj_type slab_ktype = { | |
5267 | .sysfs_ops = &slab_sysfs_ops, | |
41a21285 | 5268 | .release = kmem_cache_release, |
81819f0f CL |
5269 | }; |
5270 | ||
5271 | static int uevent_filter(struct kset *kset, struct kobject *kobj) | |
5272 | { | |
5273 | struct kobj_type *ktype = get_ktype(kobj); | |
5274 | ||
5275 | if (ktype == &slab_ktype) | |
5276 | return 1; | |
5277 | return 0; | |
5278 | } | |
5279 | ||
9cd43611 | 5280 | static const struct kset_uevent_ops slab_uevent_ops = { |
81819f0f CL |
5281 | .filter = uevent_filter, |
5282 | }; | |
5283 | ||
27c3a314 | 5284 | static struct kset *slab_kset; |
81819f0f | 5285 | |
9a41707b VD |
5286 | static inline struct kset *cache_kset(struct kmem_cache *s) |
5287 | { | |
127424c8 | 5288 | #ifdef CONFIG_MEMCG |
9a41707b | 5289 | if (!is_root_cache(s)) |
f7ce3190 | 5290 | return s->memcg_params.root_cache->memcg_kset; |
9a41707b VD |
5291 | #endif |
5292 | return slab_kset; | |
5293 | } | |
5294 | ||
81819f0f CL |
5295 | #define ID_STR_LENGTH 64 |
5296 | ||
5297 | /* Create a unique string id for a slab cache: | |
6446faa2 CL |
5298 | * |
5299 | * Format :[flags-]size | |
81819f0f CL |
5300 | */ |
5301 | static char *create_unique_id(struct kmem_cache *s) | |
5302 | { | |
5303 | char *name = kmalloc(ID_STR_LENGTH, GFP_KERNEL); | |
5304 | char *p = name; | |
5305 | ||
5306 | BUG_ON(!name); | |
5307 | ||
5308 | *p++ = ':'; | |
5309 | /* | |
5310 | * First flags affecting slabcache operations. We will only | |
5311 | * get here for aliasable slabs so we do not need to support | |
5312 | * too many flags. The flags here must cover all flags that | |
5313 | * are matched during merging to guarantee that the id is | |
5314 | * unique. | |
5315 | */ | |
5316 | if (s->flags & SLAB_CACHE_DMA) | |
5317 | *p++ = 'd'; | |
5318 | if (s->flags & SLAB_RECLAIM_ACCOUNT) | |
5319 | *p++ = 'a'; | |
5320 | if (s->flags & SLAB_DEBUG_FREE) | |
5321 | *p++ = 'F'; | |
5a896d9e VN |
5322 | if (!(s->flags & SLAB_NOTRACK)) |
5323 | *p++ = 't'; | |
230e9fc2 VD |
5324 | if (s->flags & SLAB_ACCOUNT) |
5325 | *p++ = 'A'; | |
81819f0f CL |
5326 | if (p != name + 1) |
5327 | *p++ = '-'; | |
5328 | p += sprintf(p, "%07d", s->size); | |
2633d7a0 | 5329 | |
81819f0f CL |
5330 | BUG_ON(p > name + ID_STR_LENGTH - 1); |
5331 | return name; | |
5332 | } | |
5333 | ||
5334 | static int sysfs_slab_add(struct kmem_cache *s) | |
5335 | { | |
5336 | int err; | |
5337 | const char *name; | |
45530c44 | 5338 | int unmergeable = slab_unmergeable(s); |
81819f0f | 5339 | |
81819f0f CL |
5340 | if (unmergeable) { |
5341 | /* | |
5342 | * Slabcache can never be merged so we can use the name proper. | |
5343 | * This is typically the case for debug situations. In that | |
5344 | * case we can catch duplicate names easily. | |
5345 | */ | |
27c3a314 | 5346 | sysfs_remove_link(&slab_kset->kobj, s->name); |
81819f0f CL |
5347 | name = s->name; |
5348 | } else { | |
5349 | /* | |
5350 | * Create a unique name for the slab as a target | |
5351 | * for the symlinks. | |
5352 | */ | |
5353 | name = create_unique_id(s); | |
5354 | } | |
5355 | ||
9a41707b | 5356 | s->kobj.kset = cache_kset(s); |
26e4f205 | 5357 | err = kobject_init_and_add(&s->kobj, &slab_ktype, NULL, "%s", name); |
54b6a731 | 5358 | if (err) |
80da026a | 5359 | goto out; |
81819f0f CL |
5360 | |
5361 | err = sysfs_create_group(&s->kobj, &slab_attr_group); | |
54b6a731 DJ |
5362 | if (err) |
5363 | goto out_del_kobj; | |
9a41707b | 5364 | |
127424c8 | 5365 | #ifdef CONFIG_MEMCG |
9a41707b VD |
5366 | if (is_root_cache(s)) { |
5367 | s->memcg_kset = kset_create_and_add("cgroup", NULL, &s->kobj); | |
5368 | if (!s->memcg_kset) { | |
54b6a731 DJ |
5369 | err = -ENOMEM; |
5370 | goto out_del_kobj; | |
9a41707b VD |
5371 | } |
5372 | } | |
5373 | #endif | |
5374 | ||
81819f0f CL |
5375 | kobject_uevent(&s->kobj, KOBJ_ADD); |
5376 | if (!unmergeable) { | |
5377 | /* Setup first alias */ | |
5378 | sysfs_slab_alias(s, s->name); | |
81819f0f | 5379 | } |
54b6a731 DJ |
5380 | out: |
5381 | if (!unmergeable) | |
5382 | kfree(name); | |
5383 | return err; | |
5384 | out_del_kobj: | |
5385 | kobject_del(&s->kobj); | |
54b6a731 | 5386 | goto out; |
81819f0f CL |
5387 | } |
5388 | ||
41a21285 | 5389 | void sysfs_slab_remove(struct kmem_cache *s) |
81819f0f | 5390 | { |
97d06609 | 5391 | if (slab_state < FULL) |
2bce6485 CL |
5392 | /* |
5393 | * Sysfs has not been setup yet so no need to remove the | |
5394 | * cache from sysfs. | |
5395 | */ | |
5396 | return; | |
5397 | ||
127424c8 | 5398 | #ifdef CONFIG_MEMCG |
9a41707b VD |
5399 | kset_unregister(s->memcg_kset); |
5400 | #endif | |
81819f0f CL |
5401 | kobject_uevent(&s->kobj, KOBJ_REMOVE); |
5402 | kobject_del(&s->kobj); | |
151c602f | 5403 | kobject_put(&s->kobj); |
81819f0f CL |
5404 | } |
5405 | ||
5406 | /* | |
5407 | * Need to buffer aliases during bootup until sysfs becomes | |
9f6c708e | 5408 | * available lest we lose that information. |
81819f0f CL |
5409 | */ |
5410 | struct saved_alias { | |
5411 | struct kmem_cache *s; | |
5412 | const char *name; | |
5413 | struct saved_alias *next; | |
5414 | }; | |
5415 | ||
5af328a5 | 5416 | static struct saved_alias *alias_list; |
81819f0f CL |
5417 | |
5418 | static int sysfs_slab_alias(struct kmem_cache *s, const char *name) | |
5419 | { | |
5420 | struct saved_alias *al; | |
5421 | ||
97d06609 | 5422 | if (slab_state == FULL) { |
81819f0f CL |
5423 | /* |
5424 | * If we have a leftover link then remove it. | |
5425 | */ | |
27c3a314 GKH |
5426 | sysfs_remove_link(&slab_kset->kobj, name); |
5427 | return sysfs_create_link(&slab_kset->kobj, &s->kobj, name); | |
81819f0f CL |
5428 | } |
5429 | ||
5430 | al = kmalloc(sizeof(struct saved_alias), GFP_KERNEL); | |
5431 | if (!al) | |
5432 | return -ENOMEM; | |
5433 | ||
5434 | al->s = s; | |
5435 | al->name = name; | |
5436 | al->next = alias_list; | |
5437 | alias_list = al; | |
5438 | return 0; | |
5439 | } | |
5440 | ||
5441 | static int __init slab_sysfs_init(void) | |
5442 | { | |
5b95a4ac | 5443 | struct kmem_cache *s; |
81819f0f CL |
5444 | int err; |
5445 | ||
18004c5d | 5446 | mutex_lock(&slab_mutex); |
2bce6485 | 5447 | |
0ff21e46 | 5448 | slab_kset = kset_create_and_add("slab", &slab_uevent_ops, kernel_kobj); |
27c3a314 | 5449 | if (!slab_kset) { |
18004c5d | 5450 | mutex_unlock(&slab_mutex); |
f9f58285 | 5451 | pr_err("Cannot register slab subsystem.\n"); |
81819f0f CL |
5452 | return -ENOSYS; |
5453 | } | |
5454 | ||
97d06609 | 5455 | slab_state = FULL; |
26a7bd03 | 5456 | |
5b95a4ac | 5457 | list_for_each_entry(s, &slab_caches, list) { |
26a7bd03 | 5458 | err = sysfs_slab_add(s); |
5d540fb7 | 5459 | if (err) |
f9f58285 FF |
5460 | pr_err("SLUB: Unable to add boot slab %s to sysfs\n", |
5461 | s->name); | |
26a7bd03 | 5462 | } |
81819f0f CL |
5463 | |
5464 | while (alias_list) { | |
5465 | struct saved_alias *al = alias_list; | |
5466 | ||
5467 | alias_list = alias_list->next; | |
5468 | err = sysfs_slab_alias(al->s, al->name); | |
5d540fb7 | 5469 | if (err) |
f9f58285 FF |
5470 | pr_err("SLUB: Unable to add boot slab alias %s to sysfs\n", |
5471 | al->name); | |
81819f0f CL |
5472 | kfree(al); |
5473 | } | |
5474 | ||
18004c5d | 5475 | mutex_unlock(&slab_mutex); |
81819f0f CL |
5476 | resiliency_test(); |
5477 | return 0; | |
5478 | } | |
5479 | ||
5480 | __initcall(slab_sysfs_init); | |
ab4d5ed5 | 5481 | #endif /* CONFIG_SYSFS */ |
57ed3eda PE |
5482 | |
5483 | /* | |
5484 | * The /proc/slabinfo ABI | |
5485 | */ | |
158a9624 | 5486 | #ifdef CONFIG_SLABINFO |
0d7561c6 | 5487 | void get_slabinfo(struct kmem_cache *s, struct slabinfo *sinfo) |
57ed3eda | 5488 | { |
57ed3eda | 5489 | unsigned long nr_slabs = 0; |
205ab99d CL |
5490 | unsigned long nr_objs = 0; |
5491 | unsigned long nr_free = 0; | |
57ed3eda | 5492 | int node; |
fa45dc25 | 5493 | struct kmem_cache_node *n; |
57ed3eda | 5494 | |
fa45dc25 | 5495 | for_each_kmem_cache_node(s, node, n) { |
c17fd13e WL |
5496 | nr_slabs += node_nr_slabs(n); |
5497 | nr_objs += node_nr_objs(n); | |
205ab99d | 5498 | nr_free += count_partial(n, count_free); |
57ed3eda PE |
5499 | } |
5500 | ||
0d7561c6 GC |
5501 | sinfo->active_objs = nr_objs - nr_free; |
5502 | sinfo->num_objs = nr_objs; | |
5503 | sinfo->active_slabs = nr_slabs; | |
5504 | sinfo->num_slabs = nr_slabs; | |
5505 | sinfo->objects_per_slab = oo_objects(s->oo); | |
5506 | sinfo->cache_order = oo_order(s->oo); | |
57ed3eda PE |
5507 | } |
5508 | ||
0d7561c6 | 5509 | void slabinfo_show_stats(struct seq_file *m, struct kmem_cache *s) |
7b3c3a50 | 5510 | { |
7b3c3a50 AD |
5511 | } |
5512 | ||
b7454ad3 GC |
5513 | ssize_t slabinfo_write(struct file *file, const char __user *buffer, |
5514 | size_t count, loff_t *ppos) | |
7b3c3a50 | 5515 | { |
b7454ad3 | 5516 | return -EIO; |
7b3c3a50 | 5517 | } |
158a9624 | 5518 | #endif /* CONFIG_SLABINFO */ |