]>
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 | |
d86bd1be JK |
127 | static inline void *fixup_red_left(struct kmem_cache *s, void *p) |
128 | { | |
129 | if (kmem_cache_debug(s) && s->flags & SLAB_RED_ZONE) | |
130 | p += s->red_left_pad; | |
131 | ||
132 | return p; | |
133 | } | |
134 | ||
345c905d JK |
135 | static inline bool kmem_cache_has_cpu_partial(struct kmem_cache *s) |
136 | { | |
137 | #ifdef CONFIG_SLUB_CPU_PARTIAL | |
138 | return !kmem_cache_debug(s); | |
139 | #else | |
140 | return false; | |
141 | #endif | |
142 | } | |
143 | ||
81819f0f CL |
144 | /* |
145 | * Issues still to be resolved: | |
146 | * | |
81819f0f CL |
147 | * - Support PAGE_ALLOC_DEBUG. Should be easy to do. |
148 | * | |
81819f0f CL |
149 | * - Variable sizing of the per node arrays |
150 | */ | |
151 | ||
152 | /* Enable to test recovery from slab corruption on boot */ | |
153 | #undef SLUB_RESILIENCY_TEST | |
154 | ||
b789ef51 CL |
155 | /* Enable to log cmpxchg failures */ |
156 | #undef SLUB_DEBUG_CMPXCHG | |
157 | ||
2086d26a CL |
158 | /* |
159 | * Mininum number of partial slabs. These will be left on the partial | |
160 | * lists even if they are empty. kmem_cache_shrink may reclaim them. | |
161 | */ | |
76be8950 | 162 | #define MIN_PARTIAL 5 |
e95eed57 | 163 | |
2086d26a CL |
164 | /* |
165 | * Maximum number of desirable partial slabs. | |
166 | * The existence of more partial slabs makes kmem_cache_shrink | |
721ae22a | 167 | * sort the partial list by the number of objects in use. |
2086d26a CL |
168 | */ |
169 | #define MAX_PARTIAL 10 | |
170 | ||
becfda68 | 171 | #define DEBUG_DEFAULT_FLAGS (SLAB_CONSISTENCY_CHECKS | SLAB_RED_ZONE | \ |
81819f0f | 172 | SLAB_POISON | SLAB_STORE_USER) |
672bba3a | 173 | |
149daaf3 LA |
174 | /* |
175 | * These debug flags cannot use CMPXCHG because there might be consistency | |
176 | * issues when checking or reading debug information | |
177 | */ | |
178 | #define SLAB_NO_CMPXCHG (SLAB_CONSISTENCY_CHECKS | SLAB_STORE_USER | \ | |
179 | SLAB_TRACE) | |
180 | ||
181 | ||
fa5ec8a1 | 182 | /* |
3de47213 DR |
183 | * Debugging flags that require metadata to be stored in the slab. These get |
184 | * disabled when slub_debug=O is used and a cache's min order increases with | |
185 | * metadata. | |
fa5ec8a1 | 186 | */ |
3de47213 | 187 | #define DEBUG_METADATA_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER) |
fa5ec8a1 | 188 | |
210b5c06 CG |
189 | #define OO_SHIFT 16 |
190 | #define OO_MASK ((1 << OO_SHIFT) - 1) | |
50d5c41c | 191 | #define MAX_OBJS_PER_PAGE 32767 /* since page.objects is u15 */ |
210b5c06 | 192 | |
81819f0f | 193 | /* Internal SLUB flags */ |
f90ec390 | 194 | #define __OBJECT_POISON 0x80000000UL /* Poison object */ |
b789ef51 | 195 | #define __CMPXCHG_DOUBLE 0x40000000UL /* Use cmpxchg_double */ |
81819f0f | 196 | |
81819f0f CL |
197 | #ifdef CONFIG_SMP |
198 | static struct notifier_block slab_notifier; | |
199 | #endif | |
200 | ||
02cbc874 CL |
201 | /* |
202 | * Tracking user of a slab. | |
203 | */ | |
d6543e39 | 204 | #define TRACK_ADDRS_COUNT 16 |
02cbc874 | 205 | struct track { |
ce71e27c | 206 | unsigned long addr; /* Called from address */ |
d6543e39 BG |
207 | #ifdef CONFIG_STACKTRACE |
208 | unsigned long addrs[TRACK_ADDRS_COUNT]; /* Called from address */ | |
209 | #endif | |
02cbc874 CL |
210 | int cpu; /* Was running on cpu */ |
211 | int pid; /* Pid context */ | |
212 | unsigned long when; /* When did the operation occur */ | |
213 | }; | |
214 | ||
215 | enum track_item { TRACK_ALLOC, TRACK_FREE }; | |
216 | ||
ab4d5ed5 | 217 | #ifdef CONFIG_SYSFS |
81819f0f CL |
218 | static int sysfs_slab_add(struct kmem_cache *); |
219 | static int sysfs_slab_alias(struct kmem_cache *, const char *); | |
107dab5c | 220 | static void memcg_propagate_slab_attrs(struct kmem_cache *s); |
81819f0f | 221 | #else |
0c710013 CL |
222 | static inline int sysfs_slab_add(struct kmem_cache *s) { return 0; } |
223 | static inline int sysfs_slab_alias(struct kmem_cache *s, const char *p) | |
224 | { return 0; } | |
107dab5c | 225 | static inline void memcg_propagate_slab_attrs(struct kmem_cache *s) { } |
81819f0f CL |
226 | #endif |
227 | ||
4fdccdfb | 228 | static inline void stat(const struct kmem_cache *s, enum stat_item si) |
8ff12cfc CL |
229 | { |
230 | #ifdef CONFIG_SLUB_STATS | |
88da03a6 CL |
231 | /* |
232 | * The rmw is racy on a preemptible kernel but this is acceptable, so | |
233 | * avoid this_cpu_add()'s irq-disable overhead. | |
234 | */ | |
235 | raw_cpu_inc(s->cpu_slab->stat[si]); | |
8ff12cfc CL |
236 | #endif |
237 | } | |
238 | ||
81819f0f CL |
239 | /******************************************************************** |
240 | * Core slab cache functions | |
241 | *******************************************************************/ | |
242 | ||
7656c72b CL |
243 | static inline void *get_freepointer(struct kmem_cache *s, void *object) |
244 | { | |
245 | return *(void **)(object + s->offset); | |
246 | } | |
247 | ||
0ad9500e ED |
248 | static void prefetch_freepointer(const struct kmem_cache *s, void *object) |
249 | { | |
250 | prefetch(object + s->offset); | |
251 | } | |
252 | ||
1393d9a1 CL |
253 | static inline void *get_freepointer_safe(struct kmem_cache *s, void *object) |
254 | { | |
255 | void *p; | |
256 | ||
922d566c JK |
257 | if (!debug_pagealloc_enabled()) |
258 | return get_freepointer(s, object); | |
259 | ||
1393d9a1 | 260 | probe_kernel_read(&p, (void **)(object + s->offset), sizeof(p)); |
1393d9a1 CL |
261 | return p; |
262 | } | |
263 | ||
7656c72b CL |
264 | static inline void set_freepointer(struct kmem_cache *s, void *object, void *fp) |
265 | { | |
266 | *(void **)(object + s->offset) = fp; | |
267 | } | |
268 | ||
269 | /* Loop over all objects in a slab */ | |
224a88be | 270 | #define for_each_object(__p, __s, __addr, __objects) \ |
d86bd1be JK |
271 | for (__p = fixup_red_left(__s, __addr); \ |
272 | __p < (__addr) + (__objects) * (__s)->size; \ | |
273 | __p += (__s)->size) | |
7656c72b | 274 | |
54266640 | 275 | #define for_each_object_idx(__p, __idx, __s, __addr, __objects) \ |
d86bd1be JK |
276 | for (__p = fixup_red_left(__s, __addr), __idx = 1; \ |
277 | __idx <= __objects; \ | |
278 | __p += (__s)->size, __idx++) | |
54266640 | 279 | |
7656c72b CL |
280 | /* Determine object index from a given position */ |
281 | static inline int slab_index(void *p, struct kmem_cache *s, void *addr) | |
282 | { | |
283 | return (p - addr) / s->size; | |
284 | } | |
285 | ||
ab9a0f19 LJ |
286 | static inline int order_objects(int order, unsigned long size, int reserved) |
287 | { | |
288 | return ((PAGE_SIZE << order) - reserved) / size; | |
289 | } | |
290 | ||
834f3d11 | 291 | static inline struct kmem_cache_order_objects oo_make(int order, |
ab9a0f19 | 292 | unsigned long size, int reserved) |
834f3d11 CL |
293 | { |
294 | struct kmem_cache_order_objects x = { | |
ab9a0f19 | 295 | (order << OO_SHIFT) + order_objects(order, size, reserved) |
834f3d11 CL |
296 | }; |
297 | ||
298 | return x; | |
299 | } | |
300 | ||
301 | static inline int oo_order(struct kmem_cache_order_objects x) | |
302 | { | |
210b5c06 | 303 | return x.x >> OO_SHIFT; |
834f3d11 CL |
304 | } |
305 | ||
306 | static inline int oo_objects(struct kmem_cache_order_objects x) | |
307 | { | |
210b5c06 | 308 | return x.x & OO_MASK; |
834f3d11 CL |
309 | } |
310 | ||
881db7fb CL |
311 | /* |
312 | * Per slab locking using the pagelock | |
313 | */ | |
314 | static __always_inline void slab_lock(struct page *page) | |
315 | { | |
48c935ad | 316 | VM_BUG_ON_PAGE(PageTail(page), page); |
881db7fb CL |
317 | bit_spin_lock(PG_locked, &page->flags); |
318 | } | |
319 | ||
320 | static __always_inline void slab_unlock(struct page *page) | |
321 | { | |
48c935ad | 322 | VM_BUG_ON_PAGE(PageTail(page), page); |
881db7fb CL |
323 | __bit_spin_unlock(PG_locked, &page->flags); |
324 | } | |
325 | ||
a0320865 DH |
326 | static inline void set_page_slub_counters(struct page *page, unsigned long counters_new) |
327 | { | |
328 | struct page tmp; | |
329 | tmp.counters = counters_new; | |
330 | /* | |
331 | * page->counters can cover frozen/inuse/objects as well | |
332 | * as page->_count. If we assign to ->counters directly | |
333 | * we run the risk of losing updates to page->_count, so | |
334 | * be careful and only assign to the fields we need. | |
335 | */ | |
336 | page->frozen = tmp.frozen; | |
337 | page->inuse = tmp.inuse; | |
338 | page->objects = tmp.objects; | |
339 | } | |
340 | ||
1d07171c CL |
341 | /* Interrupts must be disabled (for the fallback code to work right) */ |
342 | static inline bool __cmpxchg_double_slab(struct kmem_cache *s, struct page *page, | |
343 | void *freelist_old, unsigned long counters_old, | |
344 | void *freelist_new, unsigned long counters_new, | |
345 | const char *n) | |
346 | { | |
347 | VM_BUG_ON(!irqs_disabled()); | |
2565409f HC |
348 | #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \ |
349 | defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE) | |
1d07171c | 350 | if (s->flags & __CMPXCHG_DOUBLE) { |
cdcd6298 | 351 | if (cmpxchg_double(&page->freelist, &page->counters, |
0aa9a13d DC |
352 | freelist_old, counters_old, |
353 | freelist_new, counters_new)) | |
6f6528a1 | 354 | return true; |
1d07171c CL |
355 | } else |
356 | #endif | |
357 | { | |
358 | slab_lock(page); | |
d0e0ac97 CG |
359 | if (page->freelist == freelist_old && |
360 | page->counters == counters_old) { | |
1d07171c | 361 | page->freelist = freelist_new; |
a0320865 | 362 | set_page_slub_counters(page, counters_new); |
1d07171c | 363 | slab_unlock(page); |
6f6528a1 | 364 | return true; |
1d07171c CL |
365 | } |
366 | slab_unlock(page); | |
367 | } | |
368 | ||
369 | cpu_relax(); | |
370 | stat(s, CMPXCHG_DOUBLE_FAIL); | |
371 | ||
372 | #ifdef SLUB_DEBUG_CMPXCHG | |
f9f58285 | 373 | pr_info("%s %s: cmpxchg double redo ", n, s->name); |
1d07171c CL |
374 | #endif |
375 | ||
6f6528a1 | 376 | return false; |
1d07171c CL |
377 | } |
378 | ||
b789ef51 CL |
379 | static inline bool cmpxchg_double_slab(struct kmem_cache *s, struct page *page, |
380 | void *freelist_old, unsigned long counters_old, | |
381 | void *freelist_new, unsigned long counters_new, | |
382 | const char *n) | |
383 | { | |
2565409f HC |
384 | #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \ |
385 | defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE) | |
b789ef51 | 386 | if (s->flags & __CMPXCHG_DOUBLE) { |
cdcd6298 | 387 | if (cmpxchg_double(&page->freelist, &page->counters, |
0aa9a13d DC |
388 | freelist_old, counters_old, |
389 | freelist_new, counters_new)) | |
6f6528a1 | 390 | return true; |
b789ef51 CL |
391 | } else |
392 | #endif | |
393 | { | |
1d07171c CL |
394 | unsigned long flags; |
395 | ||
396 | local_irq_save(flags); | |
881db7fb | 397 | slab_lock(page); |
d0e0ac97 CG |
398 | if (page->freelist == freelist_old && |
399 | page->counters == counters_old) { | |
b789ef51 | 400 | page->freelist = freelist_new; |
a0320865 | 401 | set_page_slub_counters(page, counters_new); |
881db7fb | 402 | slab_unlock(page); |
1d07171c | 403 | local_irq_restore(flags); |
6f6528a1 | 404 | return true; |
b789ef51 | 405 | } |
881db7fb | 406 | slab_unlock(page); |
1d07171c | 407 | local_irq_restore(flags); |
b789ef51 CL |
408 | } |
409 | ||
410 | cpu_relax(); | |
411 | stat(s, CMPXCHG_DOUBLE_FAIL); | |
412 | ||
413 | #ifdef SLUB_DEBUG_CMPXCHG | |
f9f58285 | 414 | pr_info("%s %s: cmpxchg double redo ", n, s->name); |
b789ef51 CL |
415 | #endif |
416 | ||
6f6528a1 | 417 | return false; |
b789ef51 CL |
418 | } |
419 | ||
41ecc55b | 420 | #ifdef CONFIG_SLUB_DEBUG |
5f80b13a CL |
421 | /* |
422 | * Determine a map of object in use on a page. | |
423 | * | |
881db7fb | 424 | * Node listlock must be held to guarantee that the page does |
5f80b13a CL |
425 | * not vanish from under us. |
426 | */ | |
427 | static void get_map(struct kmem_cache *s, struct page *page, unsigned long *map) | |
428 | { | |
429 | void *p; | |
430 | void *addr = page_address(page); | |
431 | ||
432 | for (p = page->freelist; p; p = get_freepointer(s, p)) | |
433 | set_bit(slab_index(p, s, addr), map); | |
434 | } | |
435 | ||
d86bd1be JK |
436 | static inline int size_from_object(struct kmem_cache *s) |
437 | { | |
438 | if (s->flags & SLAB_RED_ZONE) | |
439 | return s->size - s->red_left_pad; | |
440 | ||
441 | return s->size; | |
442 | } | |
443 | ||
444 | static inline void *restore_red_left(struct kmem_cache *s, void *p) | |
445 | { | |
446 | if (s->flags & SLAB_RED_ZONE) | |
447 | p -= s->red_left_pad; | |
448 | ||
449 | return p; | |
450 | } | |
451 | ||
41ecc55b CL |
452 | /* |
453 | * Debug settings: | |
454 | */ | |
89d3c87e | 455 | #if defined(CONFIG_SLUB_DEBUG_ON) |
f0630fff | 456 | static int slub_debug = DEBUG_DEFAULT_FLAGS; |
89d3c87e AR |
457 | #elif defined(CONFIG_KASAN) |
458 | static int slub_debug = SLAB_STORE_USER; | |
f0630fff | 459 | #else |
41ecc55b | 460 | static int slub_debug; |
f0630fff | 461 | #endif |
41ecc55b CL |
462 | |
463 | static char *slub_debug_slabs; | |
fa5ec8a1 | 464 | static int disable_higher_order_debug; |
41ecc55b | 465 | |
a79316c6 AR |
466 | /* |
467 | * slub is about to manipulate internal object metadata. This memory lies | |
468 | * outside the range of the allocated object, so accessing it would normally | |
469 | * be reported by kasan as a bounds error. metadata_access_enable() is used | |
470 | * to tell kasan that these accesses are OK. | |
471 | */ | |
472 | static inline void metadata_access_enable(void) | |
473 | { | |
474 | kasan_disable_current(); | |
475 | } | |
476 | ||
477 | static inline void metadata_access_disable(void) | |
478 | { | |
479 | kasan_enable_current(); | |
480 | } | |
481 | ||
81819f0f CL |
482 | /* |
483 | * Object debugging | |
484 | */ | |
d86bd1be JK |
485 | |
486 | /* Verify that a pointer has an address that is valid within a slab page */ | |
487 | static inline int check_valid_pointer(struct kmem_cache *s, | |
488 | struct page *page, void *object) | |
489 | { | |
490 | void *base; | |
491 | ||
492 | if (!object) | |
493 | return 1; | |
494 | ||
495 | base = page_address(page); | |
496 | object = restore_red_left(s, object); | |
497 | if (object < base || object >= base + page->objects * s->size || | |
498 | (object - base) % s->size) { | |
499 | return 0; | |
500 | } | |
501 | ||
502 | return 1; | |
503 | } | |
504 | ||
81819f0f CL |
505 | static void print_section(char *text, u8 *addr, unsigned int length) |
506 | { | |
a79316c6 | 507 | metadata_access_enable(); |
ffc79d28 SAS |
508 | print_hex_dump(KERN_ERR, text, DUMP_PREFIX_ADDRESS, 16, 1, addr, |
509 | length, 1); | |
a79316c6 | 510 | metadata_access_disable(); |
81819f0f CL |
511 | } |
512 | ||
81819f0f CL |
513 | static struct track *get_track(struct kmem_cache *s, void *object, |
514 | enum track_item alloc) | |
515 | { | |
516 | struct track *p; | |
517 | ||
518 | if (s->offset) | |
519 | p = object + s->offset + sizeof(void *); | |
520 | else | |
521 | p = object + s->inuse; | |
522 | ||
523 | return p + alloc; | |
524 | } | |
525 | ||
526 | static void set_track(struct kmem_cache *s, void *object, | |
ce71e27c | 527 | enum track_item alloc, unsigned long addr) |
81819f0f | 528 | { |
1a00df4a | 529 | struct track *p = get_track(s, object, alloc); |
81819f0f | 530 | |
81819f0f | 531 | if (addr) { |
d6543e39 BG |
532 | #ifdef CONFIG_STACKTRACE |
533 | struct stack_trace trace; | |
534 | int i; | |
535 | ||
536 | trace.nr_entries = 0; | |
537 | trace.max_entries = TRACK_ADDRS_COUNT; | |
538 | trace.entries = p->addrs; | |
539 | trace.skip = 3; | |
a79316c6 | 540 | metadata_access_enable(); |
d6543e39 | 541 | save_stack_trace(&trace); |
a79316c6 | 542 | metadata_access_disable(); |
d6543e39 BG |
543 | |
544 | /* See rant in lockdep.c */ | |
545 | if (trace.nr_entries != 0 && | |
546 | trace.entries[trace.nr_entries - 1] == ULONG_MAX) | |
547 | trace.nr_entries--; | |
548 | ||
549 | for (i = trace.nr_entries; i < TRACK_ADDRS_COUNT; i++) | |
550 | p->addrs[i] = 0; | |
551 | #endif | |
81819f0f CL |
552 | p->addr = addr; |
553 | p->cpu = smp_processor_id(); | |
88e4ccf2 | 554 | p->pid = current->pid; |
81819f0f CL |
555 | p->when = jiffies; |
556 | } else | |
557 | memset(p, 0, sizeof(struct track)); | |
558 | } | |
559 | ||
81819f0f CL |
560 | static void init_tracking(struct kmem_cache *s, void *object) |
561 | { | |
24922684 CL |
562 | if (!(s->flags & SLAB_STORE_USER)) |
563 | return; | |
564 | ||
ce71e27c EGM |
565 | set_track(s, object, TRACK_FREE, 0UL); |
566 | set_track(s, object, TRACK_ALLOC, 0UL); | |
81819f0f CL |
567 | } |
568 | ||
569 | static void print_track(const char *s, struct track *t) | |
570 | { | |
571 | if (!t->addr) | |
572 | return; | |
573 | ||
f9f58285 FF |
574 | pr_err("INFO: %s in %pS age=%lu cpu=%u pid=%d\n", |
575 | s, (void *)t->addr, jiffies - t->when, t->cpu, t->pid); | |
d6543e39 BG |
576 | #ifdef CONFIG_STACKTRACE |
577 | { | |
578 | int i; | |
579 | for (i = 0; i < TRACK_ADDRS_COUNT; i++) | |
580 | if (t->addrs[i]) | |
f9f58285 | 581 | pr_err("\t%pS\n", (void *)t->addrs[i]); |
d6543e39 BG |
582 | else |
583 | break; | |
584 | } | |
585 | #endif | |
24922684 CL |
586 | } |
587 | ||
588 | static void print_tracking(struct kmem_cache *s, void *object) | |
589 | { | |
590 | if (!(s->flags & SLAB_STORE_USER)) | |
591 | return; | |
592 | ||
593 | print_track("Allocated", get_track(s, object, TRACK_ALLOC)); | |
594 | print_track("Freed", get_track(s, object, TRACK_FREE)); | |
595 | } | |
596 | ||
597 | static void print_page_info(struct page *page) | |
598 | { | |
f9f58285 | 599 | pr_err("INFO: Slab 0x%p objects=%u used=%u fp=0x%p flags=0x%04lx\n", |
d0e0ac97 | 600 | page, page->objects, page->inuse, page->freelist, page->flags); |
24922684 CL |
601 | |
602 | } | |
603 | ||
604 | static void slab_bug(struct kmem_cache *s, char *fmt, ...) | |
605 | { | |
ecc42fbe | 606 | struct va_format vaf; |
24922684 | 607 | va_list args; |
24922684 CL |
608 | |
609 | va_start(args, fmt); | |
ecc42fbe FF |
610 | vaf.fmt = fmt; |
611 | vaf.va = &args; | |
f9f58285 | 612 | pr_err("=============================================================================\n"); |
ecc42fbe | 613 | pr_err("BUG %s (%s): %pV\n", s->name, print_tainted(), &vaf); |
f9f58285 | 614 | pr_err("-----------------------------------------------------------------------------\n\n"); |
645df230 | 615 | |
373d4d09 | 616 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); |
ecc42fbe | 617 | va_end(args); |
81819f0f CL |
618 | } |
619 | ||
24922684 CL |
620 | static void slab_fix(struct kmem_cache *s, char *fmt, ...) |
621 | { | |
ecc42fbe | 622 | struct va_format vaf; |
24922684 | 623 | va_list args; |
24922684 CL |
624 | |
625 | va_start(args, fmt); | |
ecc42fbe FF |
626 | vaf.fmt = fmt; |
627 | vaf.va = &args; | |
628 | pr_err("FIX %s: %pV\n", s->name, &vaf); | |
24922684 | 629 | va_end(args); |
24922684 CL |
630 | } |
631 | ||
632 | static void print_trailer(struct kmem_cache *s, struct page *page, u8 *p) | |
81819f0f CL |
633 | { |
634 | unsigned int off; /* Offset of last byte */ | |
a973e9dd | 635 | u8 *addr = page_address(page); |
24922684 CL |
636 | |
637 | print_tracking(s, p); | |
638 | ||
639 | print_page_info(page); | |
640 | ||
f9f58285 FF |
641 | pr_err("INFO: Object 0x%p @offset=%tu fp=0x%p\n\n", |
642 | p, p - addr, get_freepointer(s, p)); | |
24922684 | 643 | |
d86bd1be JK |
644 | if (s->flags & SLAB_RED_ZONE) |
645 | print_section("Redzone ", p - s->red_left_pad, s->red_left_pad); | |
646 | else if (p > addr + 16) | |
ffc79d28 | 647 | print_section("Bytes b4 ", p - 16, 16); |
81819f0f | 648 | |
3b0efdfa | 649 | print_section("Object ", p, min_t(unsigned long, s->object_size, |
ffc79d28 | 650 | PAGE_SIZE)); |
81819f0f | 651 | if (s->flags & SLAB_RED_ZONE) |
3b0efdfa CL |
652 | print_section("Redzone ", p + s->object_size, |
653 | s->inuse - s->object_size); | |
81819f0f | 654 | |
81819f0f CL |
655 | if (s->offset) |
656 | off = s->offset + sizeof(void *); | |
657 | else | |
658 | off = s->inuse; | |
659 | ||
24922684 | 660 | if (s->flags & SLAB_STORE_USER) |
81819f0f | 661 | off += 2 * sizeof(struct track); |
81819f0f | 662 | |
d86bd1be | 663 | if (off != size_from_object(s)) |
81819f0f | 664 | /* Beginning of the filler is the free pointer */ |
d86bd1be | 665 | print_section("Padding ", p + off, size_from_object(s) - off); |
24922684 CL |
666 | |
667 | dump_stack(); | |
81819f0f CL |
668 | } |
669 | ||
75c66def | 670 | void object_err(struct kmem_cache *s, struct page *page, |
81819f0f CL |
671 | u8 *object, char *reason) |
672 | { | |
3dc50637 | 673 | slab_bug(s, "%s", reason); |
24922684 | 674 | print_trailer(s, page, object); |
81819f0f CL |
675 | } |
676 | ||
d0e0ac97 CG |
677 | static void slab_err(struct kmem_cache *s, struct page *page, |
678 | const char *fmt, ...) | |
81819f0f CL |
679 | { |
680 | va_list args; | |
681 | char buf[100]; | |
682 | ||
24922684 CL |
683 | va_start(args, fmt); |
684 | vsnprintf(buf, sizeof(buf), fmt, args); | |
81819f0f | 685 | va_end(args); |
3dc50637 | 686 | slab_bug(s, "%s", buf); |
24922684 | 687 | print_page_info(page); |
81819f0f CL |
688 | dump_stack(); |
689 | } | |
690 | ||
f7cb1933 | 691 | static void init_object(struct kmem_cache *s, void *object, u8 val) |
81819f0f CL |
692 | { |
693 | u8 *p = object; | |
694 | ||
d86bd1be JK |
695 | if (s->flags & SLAB_RED_ZONE) |
696 | memset(p - s->red_left_pad, val, s->red_left_pad); | |
697 | ||
81819f0f | 698 | if (s->flags & __OBJECT_POISON) { |
3b0efdfa CL |
699 | memset(p, POISON_FREE, s->object_size - 1); |
700 | p[s->object_size - 1] = POISON_END; | |
81819f0f CL |
701 | } |
702 | ||
703 | if (s->flags & SLAB_RED_ZONE) | |
3b0efdfa | 704 | memset(p + s->object_size, val, s->inuse - s->object_size); |
81819f0f CL |
705 | } |
706 | ||
24922684 CL |
707 | static void restore_bytes(struct kmem_cache *s, char *message, u8 data, |
708 | void *from, void *to) | |
709 | { | |
710 | slab_fix(s, "Restoring 0x%p-0x%p=0x%x\n", from, to - 1, data); | |
711 | memset(from, data, to - from); | |
712 | } | |
713 | ||
714 | static int check_bytes_and_report(struct kmem_cache *s, struct page *page, | |
715 | u8 *object, char *what, | |
06428780 | 716 | u8 *start, unsigned int value, unsigned int bytes) |
24922684 CL |
717 | { |
718 | u8 *fault; | |
719 | u8 *end; | |
720 | ||
a79316c6 | 721 | metadata_access_enable(); |
79824820 | 722 | fault = memchr_inv(start, value, bytes); |
a79316c6 | 723 | metadata_access_disable(); |
24922684 CL |
724 | if (!fault) |
725 | return 1; | |
726 | ||
727 | end = start + bytes; | |
728 | while (end > fault && end[-1] == value) | |
729 | end--; | |
730 | ||
731 | slab_bug(s, "%s overwritten", what); | |
f9f58285 | 732 | pr_err("INFO: 0x%p-0x%p. First byte 0x%x instead of 0x%x\n", |
24922684 CL |
733 | fault, end - 1, fault[0], value); |
734 | print_trailer(s, page, object); | |
735 | ||
736 | restore_bytes(s, what, value, fault, end); | |
737 | return 0; | |
81819f0f CL |
738 | } |
739 | ||
81819f0f CL |
740 | /* |
741 | * Object layout: | |
742 | * | |
743 | * object address | |
744 | * Bytes of the object to be managed. | |
745 | * If the freepointer may overlay the object then the free | |
746 | * pointer is the first word of the object. | |
672bba3a | 747 | * |
81819f0f CL |
748 | * Poisoning uses 0x6b (POISON_FREE) and the last byte is |
749 | * 0xa5 (POISON_END) | |
750 | * | |
3b0efdfa | 751 | * object + s->object_size |
81819f0f | 752 | * Padding to reach word boundary. This is also used for Redzoning. |
672bba3a | 753 | * Padding is extended by another word if Redzoning is enabled and |
3b0efdfa | 754 | * object_size == inuse. |
672bba3a | 755 | * |
81819f0f CL |
756 | * We fill with 0xbb (RED_INACTIVE) for inactive objects and with |
757 | * 0xcc (RED_ACTIVE) for objects in use. | |
758 | * | |
759 | * object + s->inuse | |
672bba3a CL |
760 | * Meta data starts here. |
761 | * | |
81819f0f CL |
762 | * A. Free pointer (if we cannot overwrite object on free) |
763 | * B. Tracking data for SLAB_STORE_USER | |
672bba3a | 764 | * C. Padding to reach required alignment boundary or at mininum |
6446faa2 | 765 | * one word if debugging is on to be able to detect writes |
672bba3a CL |
766 | * before the word boundary. |
767 | * | |
768 | * Padding is done using 0x5a (POISON_INUSE) | |
81819f0f CL |
769 | * |
770 | * object + s->size | |
672bba3a | 771 | * Nothing is used beyond s->size. |
81819f0f | 772 | * |
3b0efdfa | 773 | * If slabcaches are merged then the object_size and inuse boundaries are mostly |
672bba3a | 774 | * ignored. And therefore no slab options that rely on these boundaries |
81819f0f CL |
775 | * may be used with merged slabcaches. |
776 | */ | |
777 | ||
81819f0f CL |
778 | static int check_pad_bytes(struct kmem_cache *s, struct page *page, u8 *p) |
779 | { | |
780 | unsigned long off = s->inuse; /* The end of info */ | |
781 | ||
782 | if (s->offset) | |
783 | /* Freepointer is placed after the object. */ | |
784 | off += sizeof(void *); | |
785 | ||
786 | if (s->flags & SLAB_STORE_USER) | |
787 | /* We also have user information there */ | |
788 | off += 2 * sizeof(struct track); | |
789 | ||
d86bd1be | 790 | if (size_from_object(s) == off) |
81819f0f CL |
791 | return 1; |
792 | ||
24922684 | 793 | return check_bytes_and_report(s, page, p, "Object padding", |
d86bd1be | 794 | p + off, POISON_INUSE, size_from_object(s) - off); |
81819f0f CL |
795 | } |
796 | ||
39b26464 | 797 | /* Check the pad bytes at the end of a slab page */ |
81819f0f CL |
798 | static int slab_pad_check(struct kmem_cache *s, struct page *page) |
799 | { | |
24922684 CL |
800 | u8 *start; |
801 | u8 *fault; | |
802 | u8 *end; | |
803 | int length; | |
804 | int remainder; | |
81819f0f CL |
805 | |
806 | if (!(s->flags & SLAB_POISON)) | |
807 | return 1; | |
808 | ||
a973e9dd | 809 | start = page_address(page); |
ab9a0f19 | 810 | length = (PAGE_SIZE << compound_order(page)) - s->reserved; |
39b26464 CL |
811 | end = start + length; |
812 | remainder = length % s->size; | |
81819f0f CL |
813 | if (!remainder) |
814 | return 1; | |
815 | ||
a79316c6 | 816 | metadata_access_enable(); |
79824820 | 817 | fault = memchr_inv(end - remainder, POISON_INUSE, remainder); |
a79316c6 | 818 | metadata_access_disable(); |
24922684 CL |
819 | if (!fault) |
820 | return 1; | |
821 | while (end > fault && end[-1] == POISON_INUSE) | |
822 | end--; | |
823 | ||
824 | slab_err(s, page, "Padding overwritten. 0x%p-0x%p", fault, end - 1); | |
ffc79d28 | 825 | print_section("Padding ", end - remainder, remainder); |
24922684 | 826 | |
8a3d271d | 827 | restore_bytes(s, "slab padding", POISON_INUSE, end - remainder, end); |
24922684 | 828 | return 0; |
81819f0f CL |
829 | } |
830 | ||
831 | static int check_object(struct kmem_cache *s, struct page *page, | |
f7cb1933 | 832 | void *object, u8 val) |
81819f0f CL |
833 | { |
834 | u8 *p = object; | |
3b0efdfa | 835 | u8 *endobject = object + s->object_size; |
81819f0f CL |
836 | |
837 | if (s->flags & SLAB_RED_ZONE) { | |
d86bd1be JK |
838 | if (!check_bytes_and_report(s, page, object, "Redzone", |
839 | object - s->red_left_pad, val, s->red_left_pad)) | |
840 | return 0; | |
841 | ||
24922684 | 842 | if (!check_bytes_and_report(s, page, object, "Redzone", |
3b0efdfa | 843 | endobject, val, s->inuse - s->object_size)) |
81819f0f | 844 | return 0; |
81819f0f | 845 | } else { |
3b0efdfa | 846 | if ((s->flags & SLAB_POISON) && s->object_size < s->inuse) { |
3adbefee | 847 | check_bytes_and_report(s, page, p, "Alignment padding", |
d0e0ac97 CG |
848 | endobject, POISON_INUSE, |
849 | s->inuse - s->object_size); | |
3adbefee | 850 | } |
81819f0f CL |
851 | } |
852 | ||
853 | if (s->flags & SLAB_POISON) { | |
f7cb1933 | 854 | if (val != SLUB_RED_ACTIVE && (s->flags & __OBJECT_POISON) && |
24922684 | 855 | (!check_bytes_and_report(s, page, p, "Poison", p, |
3b0efdfa | 856 | POISON_FREE, s->object_size - 1) || |
24922684 | 857 | !check_bytes_and_report(s, page, p, "Poison", |
3b0efdfa | 858 | p + s->object_size - 1, POISON_END, 1))) |
81819f0f | 859 | return 0; |
81819f0f CL |
860 | /* |
861 | * check_pad_bytes cleans up on its own. | |
862 | */ | |
863 | check_pad_bytes(s, page, p); | |
864 | } | |
865 | ||
f7cb1933 | 866 | if (!s->offset && val == SLUB_RED_ACTIVE) |
81819f0f CL |
867 | /* |
868 | * Object and freepointer overlap. Cannot check | |
869 | * freepointer while object is allocated. | |
870 | */ | |
871 | return 1; | |
872 | ||
873 | /* Check free pointer validity */ | |
874 | if (!check_valid_pointer(s, page, get_freepointer(s, p))) { | |
875 | object_err(s, page, p, "Freepointer corrupt"); | |
876 | /* | |
9f6c708e | 877 | * No choice but to zap it and thus lose the remainder |
81819f0f | 878 | * of the free objects in this slab. May cause |
672bba3a | 879 | * another error because the object count is now wrong. |
81819f0f | 880 | */ |
a973e9dd | 881 | set_freepointer(s, p, NULL); |
81819f0f CL |
882 | return 0; |
883 | } | |
884 | return 1; | |
885 | } | |
886 | ||
887 | static int check_slab(struct kmem_cache *s, struct page *page) | |
888 | { | |
39b26464 CL |
889 | int maxobj; |
890 | ||
81819f0f CL |
891 | VM_BUG_ON(!irqs_disabled()); |
892 | ||
893 | if (!PageSlab(page)) { | |
24922684 | 894 | slab_err(s, page, "Not a valid slab page"); |
81819f0f CL |
895 | return 0; |
896 | } | |
39b26464 | 897 | |
ab9a0f19 | 898 | maxobj = order_objects(compound_order(page), s->size, s->reserved); |
39b26464 CL |
899 | if (page->objects > maxobj) { |
900 | slab_err(s, page, "objects %u > max %u", | |
f6edde9c | 901 | page->objects, maxobj); |
39b26464 CL |
902 | return 0; |
903 | } | |
904 | if (page->inuse > page->objects) { | |
24922684 | 905 | slab_err(s, page, "inuse %u > max %u", |
f6edde9c | 906 | page->inuse, page->objects); |
81819f0f CL |
907 | return 0; |
908 | } | |
909 | /* Slab_pad_check fixes things up after itself */ | |
910 | slab_pad_check(s, page); | |
911 | return 1; | |
912 | } | |
913 | ||
914 | /* | |
672bba3a CL |
915 | * Determine if a certain object on a page is on the freelist. Must hold the |
916 | * slab lock to guarantee that the chains are in a consistent state. | |
81819f0f CL |
917 | */ |
918 | static int on_freelist(struct kmem_cache *s, struct page *page, void *search) | |
919 | { | |
920 | int nr = 0; | |
881db7fb | 921 | void *fp; |
81819f0f | 922 | void *object = NULL; |
f6edde9c | 923 | int max_objects; |
81819f0f | 924 | |
881db7fb | 925 | fp = page->freelist; |
39b26464 | 926 | while (fp && nr <= page->objects) { |
81819f0f CL |
927 | if (fp == search) |
928 | return 1; | |
929 | if (!check_valid_pointer(s, page, fp)) { | |
930 | if (object) { | |
931 | object_err(s, page, object, | |
932 | "Freechain corrupt"); | |
a973e9dd | 933 | set_freepointer(s, object, NULL); |
81819f0f | 934 | } else { |
24922684 | 935 | slab_err(s, page, "Freepointer corrupt"); |
a973e9dd | 936 | page->freelist = NULL; |
39b26464 | 937 | page->inuse = page->objects; |
24922684 | 938 | slab_fix(s, "Freelist cleared"); |
81819f0f CL |
939 | return 0; |
940 | } | |
941 | break; | |
942 | } | |
943 | object = fp; | |
944 | fp = get_freepointer(s, object); | |
945 | nr++; | |
946 | } | |
947 | ||
ab9a0f19 | 948 | max_objects = order_objects(compound_order(page), s->size, s->reserved); |
210b5c06 CG |
949 | if (max_objects > MAX_OBJS_PER_PAGE) |
950 | max_objects = MAX_OBJS_PER_PAGE; | |
224a88be CL |
951 | |
952 | if (page->objects != max_objects) { | |
756a025f JP |
953 | slab_err(s, page, "Wrong number of objects. Found %d but should be %d", |
954 | page->objects, max_objects); | |
224a88be CL |
955 | page->objects = max_objects; |
956 | slab_fix(s, "Number of objects adjusted."); | |
957 | } | |
39b26464 | 958 | if (page->inuse != page->objects - nr) { |
756a025f JP |
959 | slab_err(s, page, "Wrong object count. Counter is %d but counted were %d", |
960 | page->inuse, page->objects - nr); | |
39b26464 | 961 | page->inuse = page->objects - nr; |
24922684 | 962 | slab_fix(s, "Object count adjusted."); |
81819f0f CL |
963 | } |
964 | return search == NULL; | |
965 | } | |
966 | ||
0121c619 CL |
967 | static void trace(struct kmem_cache *s, struct page *page, void *object, |
968 | int alloc) | |
3ec09742 CL |
969 | { |
970 | if (s->flags & SLAB_TRACE) { | |
f9f58285 | 971 | pr_info("TRACE %s %s 0x%p inuse=%d fp=0x%p\n", |
3ec09742 CL |
972 | s->name, |
973 | alloc ? "alloc" : "free", | |
974 | object, page->inuse, | |
975 | page->freelist); | |
976 | ||
977 | if (!alloc) | |
d0e0ac97 CG |
978 | print_section("Object ", (void *)object, |
979 | s->object_size); | |
3ec09742 CL |
980 | |
981 | dump_stack(); | |
982 | } | |
983 | } | |
984 | ||
643b1138 | 985 | /* |
672bba3a | 986 | * Tracking of fully allocated slabs for debugging purposes. |
643b1138 | 987 | */ |
5cc6eee8 CL |
988 | static void add_full(struct kmem_cache *s, |
989 | struct kmem_cache_node *n, struct page *page) | |
643b1138 | 990 | { |
5cc6eee8 CL |
991 | if (!(s->flags & SLAB_STORE_USER)) |
992 | return; | |
993 | ||
255d0884 | 994 | lockdep_assert_held(&n->list_lock); |
643b1138 | 995 | list_add(&page->lru, &n->full); |
643b1138 CL |
996 | } |
997 | ||
c65c1877 | 998 | static void remove_full(struct kmem_cache *s, struct kmem_cache_node *n, struct page *page) |
643b1138 | 999 | { |
643b1138 CL |
1000 | if (!(s->flags & SLAB_STORE_USER)) |
1001 | return; | |
1002 | ||
255d0884 | 1003 | lockdep_assert_held(&n->list_lock); |
643b1138 | 1004 | list_del(&page->lru); |
643b1138 CL |
1005 | } |
1006 | ||
0f389ec6 CL |
1007 | /* Tracking of the number of slabs for debugging purposes */ |
1008 | static inline unsigned long slabs_node(struct kmem_cache *s, int node) | |
1009 | { | |
1010 | struct kmem_cache_node *n = get_node(s, node); | |
1011 | ||
1012 | return atomic_long_read(&n->nr_slabs); | |
1013 | } | |
1014 | ||
26c02cf0 AB |
1015 | static inline unsigned long node_nr_slabs(struct kmem_cache_node *n) |
1016 | { | |
1017 | return atomic_long_read(&n->nr_slabs); | |
1018 | } | |
1019 | ||
205ab99d | 1020 | static inline void inc_slabs_node(struct kmem_cache *s, int node, int objects) |
0f389ec6 CL |
1021 | { |
1022 | struct kmem_cache_node *n = get_node(s, node); | |
1023 | ||
1024 | /* | |
1025 | * May be called early in order to allocate a slab for the | |
1026 | * kmem_cache_node structure. Solve the chicken-egg | |
1027 | * dilemma by deferring the increment of the count during | |
1028 | * bootstrap (see early_kmem_cache_node_alloc). | |
1029 | */ | |
338b2642 | 1030 | if (likely(n)) { |
0f389ec6 | 1031 | atomic_long_inc(&n->nr_slabs); |
205ab99d CL |
1032 | atomic_long_add(objects, &n->total_objects); |
1033 | } | |
0f389ec6 | 1034 | } |
205ab99d | 1035 | static inline void dec_slabs_node(struct kmem_cache *s, int node, int objects) |
0f389ec6 CL |
1036 | { |
1037 | struct kmem_cache_node *n = get_node(s, node); | |
1038 | ||
1039 | atomic_long_dec(&n->nr_slabs); | |
205ab99d | 1040 | atomic_long_sub(objects, &n->total_objects); |
0f389ec6 CL |
1041 | } |
1042 | ||
1043 | /* Object debug checks for alloc/free paths */ | |
3ec09742 CL |
1044 | static void setup_object_debug(struct kmem_cache *s, struct page *page, |
1045 | void *object) | |
1046 | { | |
1047 | if (!(s->flags & (SLAB_STORE_USER|SLAB_RED_ZONE|__OBJECT_POISON))) | |
1048 | return; | |
1049 | ||
f7cb1933 | 1050 | init_object(s, object, SLUB_RED_INACTIVE); |
3ec09742 CL |
1051 | init_tracking(s, object); |
1052 | } | |
1053 | ||
becfda68 | 1054 | static inline int alloc_consistency_checks(struct kmem_cache *s, |
d0e0ac97 | 1055 | struct page *page, |
ce71e27c | 1056 | void *object, unsigned long addr) |
81819f0f CL |
1057 | { |
1058 | if (!check_slab(s, page)) | |
becfda68 | 1059 | return 0; |
81819f0f | 1060 | |
81819f0f CL |
1061 | if (!check_valid_pointer(s, page, object)) { |
1062 | object_err(s, page, object, "Freelist Pointer check fails"); | |
becfda68 | 1063 | return 0; |
81819f0f CL |
1064 | } |
1065 | ||
f7cb1933 | 1066 | if (!check_object(s, page, object, SLUB_RED_INACTIVE)) |
becfda68 LA |
1067 | return 0; |
1068 | ||
1069 | return 1; | |
1070 | } | |
1071 | ||
1072 | static noinline int alloc_debug_processing(struct kmem_cache *s, | |
1073 | struct page *page, | |
1074 | void *object, unsigned long addr) | |
1075 | { | |
1076 | if (s->flags & SLAB_CONSISTENCY_CHECKS) { | |
1077 | if (!alloc_consistency_checks(s, page, object, addr)) | |
1078 | goto bad; | |
1079 | } | |
81819f0f | 1080 | |
3ec09742 CL |
1081 | /* Success perform special debug activities for allocs */ |
1082 | if (s->flags & SLAB_STORE_USER) | |
1083 | set_track(s, object, TRACK_ALLOC, addr); | |
1084 | trace(s, page, object, 1); | |
f7cb1933 | 1085 | init_object(s, object, SLUB_RED_ACTIVE); |
81819f0f | 1086 | return 1; |
3ec09742 | 1087 | |
81819f0f CL |
1088 | bad: |
1089 | if (PageSlab(page)) { | |
1090 | /* | |
1091 | * If this is a slab page then lets do the best we can | |
1092 | * to avoid issues in the future. Marking all objects | |
672bba3a | 1093 | * as used avoids touching the remaining objects. |
81819f0f | 1094 | */ |
24922684 | 1095 | slab_fix(s, "Marking all objects used"); |
39b26464 | 1096 | page->inuse = page->objects; |
a973e9dd | 1097 | page->freelist = NULL; |
81819f0f CL |
1098 | } |
1099 | return 0; | |
1100 | } | |
1101 | ||
becfda68 LA |
1102 | static inline int free_consistency_checks(struct kmem_cache *s, |
1103 | struct page *page, void *object, unsigned long addr) | |
81819f0f | 1104 | { |
81819f0f | 1105 | if (!check_valid_pointer(s, page, object)) { |
70d71228 | 1106 | slab_err(s, page, "Invalid object pointer 0x%p", object); |
becfda68 | 1107 | return 0; |
81819f0f CL |
1108 | } |
1109 | ||
1110 | if (on_freelist(s, page, object)) { | |
24922684 | 1111 | object_err(s, page, object, "Object already free"); |
becfda68 | 1112 | return 0; |
81819f0f CL |
1113 | } |
1114 | ||
f7cb1933 | 1115 | if (!check_object(s, page, object, SLUB_RED_ACTIVE)) |
becfda68 | 1116 | return 0; |
81819f0f | 1117 | |
1b4f59e3 | 1118 | if (unlikely(s != page->slab_cache)) { |
3adbefee | 1119 | if (!PageSlab(page)) { |
756a025f JP |
1120 | slab_err(s, page, "Attempt to free object(0x%p) outside of slab", |
1121 | object); | |
1b4f59e3 | 1122 | } else if (!page->slab_cache) { |
f9f58285 FF |
1123 | pr_err("SLUB <none>: no slab for object 0x%p.\n", |
1124 | object); | |
70d71228 | 1125 | dump_stack(); |
06428780 | 1126 | } else |
24922684 CL |
1127 | object_err(s, page, object, |
1128 | "page slab pointer corrupt."); | |
becfda68 LA |
1129 | return 0; |
1130 | } | |
1131 | return 1; | |
1132 | } | |
1133 | ||
1134 | /* Supports checking bulk free of a constructed freelist */ | |
1135 | static noinline int free_debug_processing( | |
1136 | struct kmem_cache *s, struct page *page, | |
1137 | void *head, void *tail, int bulk_cnt, | |
1138 | unsigned long addr) | |
1139 | { | |
1140 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); | |
1141 | void *object = head; | |
1142 | int cnt = 0; | |
1143 | unsigned long uninitialized_var(flags); | |
1144 | int ret = 0; | |
1145 | ||
1146 | spin_lock_irqsave(&n->list_lock, flags); | |
1147 | slab_lock(page); | |
1148 | ||
1149 | if (s->flags & SLAB_CONSISTENCY_CHECKS) { | |
1150 | if (!check_slab(s, page)) | |
1151 | goto out; | |
1152 | } | |
1153 | ||
1154 | next_object: | |
1155 | cnt++; | |
1156 | ||
1157 | if (s->flags & SLAB_CONSISTENCY_CHECKS) { | |
1158 | if (!free_consistency_checks(s, page, object, addr)) | |
1159 | goto out; | |
81819f0f | 1160 | } |
3ec09742 | 1161 | |
3ec09742 CL |
1162 | if (s->flags & SLAB_STORE_USER) |
1163 | set_track(s, object, TRACK_FREE, addr); | |
1164 | trace(s, page, object, 0); | |
81084651 | 1165 | /* Freepointer not overwritten by init_object(), SLAB_POISON moved it */ |
f7cb1933 | 1166 | init_object(s, object, SLUB_RED_INACTIVE); |
81084651 JDB |
1167 | |
1168 | /* Reached end of constructed freelist yet? */ | |
1169 | if (object != tail) { | |
1170 | object = get_freepointer(s, object); | |
1171 | goto next_object; | |
1172 | } | |
804aa132 LA |
1173 | ret = 1; |
1174 | ||
5c2e4bbb | 1175 | out: |
81084651 JDB |
1176 | if (cnt != bulk_cnt) |
1177 | slab_err(s, page, "Bulk freelist count(%d) invalid(%d)\n", | |
1178 | bulk_cnt, cnt); | |
1179 | ||
881db7fb | 1180 | slab_unlock(page); |
282acb43 | 1181 | spin_unlock_irqrestore(&n->list_lock, flags); |
804aa132 LA |
1182 | if (!ret) |
1183 | slab_fix(s, "Object at 0x%p not freed", object); | |
1184 | return ret; | |
81819f0f CL |
1185 | } |
1186 | ||
41ecc55b CL |
1187 | static int __init setup_slub_debug(char *str) |
1188 | { | |
f0630fff CL |
1189 | slub_debug = DEBUG_DEFAULT_FLAGS; |
1190 | if (*str++ != '=' || !*str) | |
1191 | /* | |
1192 | * No options specified. Switch on full debugging. | |
1193 | */ | |
1194 | goto out; | |
1195 | ||
1196 | if (*str == ',') | |
1197 | /* | |
1198 | * No options but restriction on slabs. This means full | |
1199 | * debugging for slabs matching a pattern. | |
1200 | */ | |
1201 | goto check_slabs; | |
1202 | ||
1203 | slub_debug = 0; | |
1204 | if (*str == '-') | |
1205 | /* | |
1206 | * Switch off all debugging measures. | |
1207 | */ | |
1208 | goto out; | |
1209 | ||
1210 | /* | |
1211 | * Determine which debug features should be switched on | |
1212 | */ | |
06428780 | 1213 | for (; *str && *str != ','; str++) { |
f0630fff CL |
1214 | switch (tolower(*str)) { |
1215 | case 'f': | |
becfda68 | 1216 | slub_debug |= SLAB_CONSISTENCY_CHECKS; |
f0630fff CL |
1217 | break; |
1218 | case 'z': | |
1219 | slub_debug |= SLAB_RED_ZONE; | |
1220 | break; | |
1221 | case 'p': | |
1222 | slub_debug |= SLAB_POISON; | |
1223 | break; | |
1224 | case 'u': | |
1225 | slub_debug |= SLAB_STORE_USER; | |
1226 | break; | |
1227 | case 't': | |
1228 | slub_debug |= SLAB_TRACE; | |
1229 | break; | |
4c13dd3b DM |
1230 | case 'a': |
1231 | slub_debug |= SLAB_FAILSLAB; | |
1232 | break; | |
08303a73 CA |
1233 | case 'o': |
1234 | /* | |
1235 | * Avoid enabling debugging on caches if its minimum | |
1236 | * order would increase as a result. | |
1237 | */ | |
1238 | disable_higher_order_debug = 1; | |
1239 | break; | |
f0630fff | 1240 | default: |
f9f58285 FF |
1241 | pr_err("slub_debug option '%c' unknown. skipped\n", |
1242 | *str); | |
f0630fff | 1243 | } |
41ecc55b CL |
1244 | } |
1245 | ||
f0630fff | 1246 | check_slabs: |
41ecc55b CL |
1247 | if (*str == ',') |
1248 | slub_debug_slabs = str + 1; | |
f0630fff | 1249 | out: |
41ecc55b CL |
1250 | return 1; |
1251 | } | |
1252 | ||
1253 | __setup("slub_debug", setup_slub_debug); | |
1254 | ||
423c929c | 1255 | unsigned long kmem_cache_flags(unsigned long object_size, |
ba0268a8 | 1256 | unsigned long flags, const char *name, |
51cc5068 | 1257 | void (*ctor)(void *)) |
41ecc55b CL |
1258 | { |
1259 | /* | |
e153362a | 1260 | * Enable debugging if selected on the kernel commandline. |
41ecc55b | 1261 | */ |
c6f58d9b CL |
1262 | if (slub_debug && (!slub_debug_slabs || (name && |
1263 | !strncmp(slub_debug_slabs, name, strlen(slub_debug_slabs))))) | |
3de47213 | 1264 | flags |= slub_debug; |
ba0268a8 CL |
1265 | |
1266 | return flags; | |
41ecc55b | 1267 | } |
b4a64718 | 1268 | #else /* !CONFIG_SLUB_DEBUG */ |
3ec09742 CL |
1269 | static inline void setup_object_debug(struct kmem_cache *s, |
1270 | struct page *page, void *object) {} | |
41ecc55b | 1271 | |
3ec09742 | 1272 | static inline int alloc_debug_processing(struct kmem_cache *s, |
ce71e27c | 1273 | struct page *page, void *object, unsigned long addr) { return 0; } |
41ecc55b | 1274 | |
282acb43 | 1275 | static inline int free_debug_processing( |
81084651 JDB |
1276 | struct kmem_cache *s, struct page *page, |
1277 | void *head, void *tail, int bulk_cnt, | |
282acb43 | 1278 | unsigned long addr) { return 0; } |
41ecc55b | 1279 | |
41ecc55b CL |
1280 | static inline int slab_pad_check(struct kmem_cache *s, struct page *page) |
1281 | { return 1; } | |
1282 | static inline int check_object(struct kmem_cache *s, struct page *page, | |
f7cb1933 | 1283 | void *object, u8 val) { return 1; } |
5cc6eee8 CL |
1284 | static inline void add_full(struct kmem_cache *s, struct kmem_cache_node *n, |
1285 | struct page *page) {} | |
c65c1877 PZ |
1286 | static inline void remove_full(struct kmem_cache *s, struct kmem_cache_node *n, |
1287 | struct page *page) {} | |
423c929c | 1288 | unsigned long kmem_cache_flags(unsigned long object_size, |
ba0268a8 | 1289 | unsigned long flags, const char *name, |
51cc5068 | 1290 | void (*ctor)(void *)) |
ba0268a8 CL |
1291 | { |
1292 | return flags; | |
1293 | } | |
41ecc55b | 1294 | #define slub_debug 0 |
0f389ec6 | 1295 | |
fdaa45e9 IM |
1296 | #define disable_higher_order_debug 0 |
1297 | ||
0f389ec6 CL |
1298 | static inline unsigned long slabs_node(struct kmem_cache *s, int node) |
1299 | { return 0; } | |
26c02cf0 AB |
1300 | static inline unsigned long node_nr_slabs(struct kmem_cache_node *n) |
1301 | { return 0; } | |
205ab99d CL |
1302 | static inline void inc_slabs_node(struct kmem_cache *s, int node, |
1303 | int objects) {} | |
1304 | static inline void dec_slabs_node(struct kmem_cache *s, int node, | |
1305 | int objects) {} | |
7d550c56 | 1306 | |
02e72cc6 AR |
1307 | #endif /* CONFIG_SLUB_DEBUG */ |
1308 | ||
1309 | /* | |
1310 | * Hooks for other subsystems that check memory allocations. In a typical | |
1311 | * production configuration these hooks all should produce no code at all. | |
1312 | */ | |
d56791b3 RB |
1313 | static inline void kmalloc_large_node_hook(void *ptr, size_t size, gfp_t flags) |
1314 | { | |
1315 | kmemleak_alloc(ptr, size, 1, flags); | |
505f5dcb | 1316 | kasan_kmalloc_large(ptr, size, flags); |
d56791b3 RB |
1317 | } |
1318 | ||
1319 | static inline void kfree_hook(const void *x) | |
1320 | { | |
1321 | kmemleak_free(x); | |
0316bec2 | 1322 | kasan_kfree_large(x); |
d56791b3 RB |
1323 | } |
1324 | ||
d56791b3 RB |
1325 | static inline void slab_free_hook(struct kmem_cache *s, void *x) |
1326 | { | |
1327 | kmemleak_free_recursive(x, s->flags); | |
7d550c56 | 1328 | |
02e72cc6 AR |
1329 | /* |
1330 | * Trouble is that we may no longer disable interrupts in the fast path | |
1331 | * So in order to make the debug calls that expect irqs to be | |
1332 | * disabled we need to disable interrupts temporarily. | |
1333 | */ | |
1334 | #if defined(CONFIG_KMEMCHECK) || defined(CONFIG_LOCKDEP) | |
1335 | { | |
1336 | unsigned long flags; | |
1337 | ||
1338 | local_irq_save(flags); | |
1339 | kmemcheck_slab_free(s, x, s->object_size); | |
1340 | debug_check_no_locks_freed(x, s->object_size); | |
1341 | local_irq_restore(flags); | |
1342 | } | |
1343 | #endif | |
1344 | if (!(s->flags & SLAB_DEBUG_OBJECTS)) | |
1345 | debug_check_no_obj_freed(x, s->object_size); | |
0316bec2 AR |
1346 | |
1347 | kasan_slab_free(s, x); | |
02e72cc6 | 1348 | } |
205ab99d | 1349 | |
81084651 JDB |
1350 | static inline void slab_free_freelist_hook(struct kmem_cache *s, |
1351 | void *head, void *tail) | |
1352 | { | |
1353 | /* | |
1354 | * Compiler cannot detect this function can be removed if slab_free_hook() | |
1355 | * evaluates to nothing. Thus, catch all relevant config debug options here. | |
1356 | */ | |
1357 | #if defined(CONFIG_KMEMCHECK) || \ | |
1358 | defined(CONFIG_LOCKDEP) || \ | |
1359 | defined(CONFIG_DEBUG_KMEMLEAK) || \ | |
1360 | defined(CONFIG_DEBUG_OBJECTS_FREE) || \ | |
1361 | defined(CONFIG_KASAN) | |
1362 | ||
1363 | void *object = head; | |
1364 | void *tail_obj = tail ? : head; | |
1365 | ||
1366 | do { | |
1367 | slab_free_hook(s, object); | |
1368 | } while ((object != tail_obj) && | |
1369 | (object = get_freepointer(s, object))); | |
1370 | #endif | |
1371 | } | |
1372 | ||
588f8ba9 TG |
1373 | static void setup_object(struct kmem_cache *s, struct page *page, |
1374 | void *object) | |
1375 | { | |
1376 | setup_object_debug(s, page, object); | |
1377 | if (unlikely(s->ctor)) { | |
1378 | kasan_unpoison_object_data(s, object); | |
1379 | s->ctor(object); | |
1380 | kasan_poison_object_data(s, object); | |
1381 | } | |
1382 | } | |
1383 | ||
81819f0f CL |
1384 | /* |
1385 | * Slab allocation and freeing | |
1386 | */ | |
5dfb4175 VD |
1387 | static inline struct page *alloc_slab_page(struct kmem_cache *s, |
1388 | gfp_t flags, int node, struct kmem_cache_order_objects oo) | |
65c3376a | 1389 | { |
5dfb4175 | 1390 | struct page *page; |
65c3376a CL |
1391 | int order = oo_order(oo); |
1392 | ||
b1eeab67 VN |
1393 | flags |= __GFP_NOTRACK; |
1394 | ||
2154a336 | 1395 | if (node == NUMA_NO_NODE) |
5dfb4175 | 1396 | page = alloc_pages(flags, order); |
65c3376a | 1397 | else |
96db800f | 1398 | page = __alloc_pages_node(node, flags, order); |
5dfb4175 | 1399 | |
f3ccb2c4 VD |
1400 | if (page && memcg_charge_slab(page, flags, order, s)) { |
1401 | __free_pages(page, order); | |
1402 | page = NULL; | |
1403 | } | |
5dfb4175 VD |
1404 | |
1405 | return page; | |
65c3376a CL |
1406 | } |
1407 | ||
81819f0f CL |
1408 | static struct page *allocate_slab(struct kmem_cache *s, gfp_t flags, int node) |
1409 | { | |
06428780 | 1410 | struct page *page; |
834f3d11 | 1411 | struct kmem_cache_order_objects oo = s->oo; |
ba52270d | 1412 | gfp_t alloc_gfp; |
588f8ba9 TG |
1413 | void *start, *p; |
1414 | int idx, order; | |
81819f0f | 1415 | |
7e0528da CL |
1416 | flags &= gfp_allowed_mask; |
1417 | ||
d0164adc | 1418 | if (gfpflags_allow_blocking(flags)) |
7e0528da CL |
1419 | local_irq_enable(); |
1420 | ||
b7a49f0d | 1421 | flags |= s->allocflags; |
e12ba74d | 1422 | |
ba52270d PE |
1423 | /* |
1424 | * Let the initial higher-order allocation fail under memory pressure | |
1425 | * so we fall-back to the minimum order allocation. | |
1426 | */ | |
1427 | alloc_gfp = (flags | __GFP_NOWARN | __GFP_NORETRY) & ~__GFP_NOFAIL; | |
d0164adc | 1428 | if ((alloc_gfp & __GFP_DIRECT_RECLAIM) && oo_order(oo) > oo_order(s->min)) |
444eb2a4 | 1429 | alloc_gfp = (alloc_gfp | __GFP_NOMEMALLOC) & ~(__GFP_RECLAIM|__GFP_NOFAIL); |
ba52270d | 1430 | |
5dfb4175 | 1431 | page = alloc_slab_page(s, alloc_gfp, node, oo); |
65c3376a CL |
1432 | if (unlikely(!page)) { |
1433 | oo = s->min; | |
80c3a998 | 1434 | alloc_gfp = flags; |
65c3376a CL |
1435 | /* |
1436 | * Allocation may have failed due to fragmentation. | |
1437 | * Try a lower order alloc if possible | |
1438 | */ | |
5dfb4175 | 1439 | page = alloc_slab_page(s, alloc_gfp, node, oo); |
588f8ba9 TG |
1440 | if (unlikely(!page)) |
1441 | goto out; | |
1442 | stat(s, ORDER_FALLBACK); | |
65c3376a | 1443 | } |
5a896d9e | 1444 | |
588f8ba9 TG |
1445 | if (kmemcheck_enabled && |
1446 | !(s->flags & (SLAB_NOTRACK | DEBUG_DEFAULT_FLAGS))) { | |
b1eeab67 VN |
1447 | int pages = 1 << oo_order(oo); |
1448 | ||
80c3a998 | 1449 | kmemcheck_alloc_shadow(page, oo_order(oo), alloc_gfp, node); |
b1eeab67 VN |
1450 | |
1451 | /* | |
1452 | * Objects from caches that have a constructor don't get | |
1453 | * cleared when they're allocated, so we need to do it here. | |
1454 | */ | |
1455 | if (s->ctor) | |
1456 | kmemcheck_mark_uninitialized_pages(page, pages); | |
1457 | else | |
1458 | kmemcheck_mark_unallocated_pages(page, pages); | |
5a896d9e VN |
1459 | } |
1460 | ||
834f3d11 | 1461 | page->objects = oo_objects(oo); |
81819f0f | 1462 | |
1f458cbf | 1463 | order = compound_order(page); |
1b4f59e3 | 1464 | page->slab_cache = s; |
c03f94cc | 1465 | __SetPageSlab(page); |
2f064f34 | 1466 | if (page_is_pfmemalloc(page)) |
072bb0aa | 1467 | SetPageSlabPfmemalloc(page); |
81819f0f CL |
1468 | |
1469 | start = page_address(page); | |
81819f0f CL |
1470 | |
1471 | if (unlikely(s->flags & SLAB_POISON)) | |
1f458cbf | 1472 | memset(start, POISON_INUSE, PAGE_SIZE << order); |
81819f0f | 1473 | |
0316bec2 AR |
1474 | kasan_poison_slab(page); |
1475 | ||
54266640 WY |
1476 | for_each_object_idx(p, idx, s, start, page->objects) { |
1477 | setup_object(s, page, p); | |
1478 | if (likely(idx < page->objects)) | |
1479 | set_freepointer(s, p, p + s->size); | |
1480 | else | |
1481 | set_freepointer(s, p, NULL); | |
81819f0f | 1482 | } |
81819f0f | 1483 | |
d86bd1be | 1484 | page->freelist = fixup_red_left(s, start); |
e6e82ea1 | 1485 | page->inuse = page->objects; |
8cb0a506 | 1486 | page->frozen = 1; |
588f8ba9 | 1487 | |
81819f0f | 1488 | out: |
d0164adc | 1489 | if (gfpflags_allow_blocking(flags)) |
588f8ba9 TG |
1490 | local_irq_disable(); |
1491 | if (!page) | |
1492 | return NULL; | |
1493 | ||
1494 | mod_zone_page_state(page_zone(page), | |
1495 | (s->flags & SLAB_RECLAIM_ACCOUNT) ? | |
1496 | NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE, | |
1497 | 1 << oo_order(oo)); | |
1498 | ||
1499 | inc_slabs_node(s, page_to_nid(page), page->objects); | |
1500 | ||
81819f0f CL |
1501 | return page; |
1502 | } | |
1503 | ||
588f8ba9 TG |
1504 | static struct page *new_slab(struct kmem_cache *s, gfp_t flags, int node) |
1505 | { | |
1506 | if (unlikely(flags & GFP_SLAB_BUG_MASK)) { | |
1507 | pr_emerg("gfp: %u\n", flags & GFP_SLAB_BUG_MASK); | |
1508 | BUG(); | |
1509 | } | |
1510 | ||
1511 | return allocate_slab(s, | |
1512 | flags & (GFP_RECLAIM_MASK | GFP_CONSTRAINT_MASK), node); | |
1513 | } | |
1514 | ||
81819f0f CL |
1515 | static void __free_slab(struct kmem_cache *s, struct page *page) |
1516 | { | |
834f3d11 CL |
1517 | int order = compound_order(page); |
1518 | int pages = 1 << order; | |
81819f0f | 1519 | |
becfda68 | 1520 | if (s->flags & SLAB_CONSISTENCY_CHECKS) { |
81819f0f CL |
1521 | void *p; |
1522 | ||
1523 | slab_pad_check(s, page); | |
224a88be CL |
1524 | for_each_object(p, s, page_address(page), |
1525 | page->objects) | |
f7cb1933 | 1526 | check_object(s, page, p, SLUB_RED_INACTIVE); |
81819f0f CL |
1527 | } |
1528 | ||
b1eeab67 | 1529 | kmemcheck_free_shadow(page, compound_order(page)); |
5a896d9e | 1530 | |
81819f0f CL |
1531 | mod_zone_page_state(page_zone(page), |
1532 | (s->flags & SLAB_RECLAIM_ACCOUNT) ? | |
1533 | NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE, | |
06428780 | 1534 | -pages); |
81819f0f | 1535 | |
072bb0aa | 1536 | __ClearPageSlabPfmemalloc(page); |
49bd5221 | 1537 | __ClearPageSlab(page); |
1f458cbf | 1538 | |
22b751c3 | 1539 | page_mapcount_reset(page); |
1eb5ac64 NP |
1540 | if (current->reclaim_state) |
1541 | current->reclaim_state->reclaimed_slab += pages; | |
27ee57c9 VD |
1542 | memcg_uncharge_slab(page, order, s); |
1543 | __free_pages(page, order); | |
81819f0f CL |
1544 | } |
1545 | ||
da9a638c LJ |
1546 | #define need_reserve_slab_rcu \ |
1547 | (sizeof(((struct page *)NULL)->lru) < sizeof(struct rcu_head)) | |
1548 | ||
81819f0f CL |
1549 | static void rcu_free_slab(struct rcu_head *h) |
1550 | { | |
1551 | struct page *page; | |
1552 | ||
da9a638c LJ |
1553 | if (need_reserve_slab_rcu) |
1554 | page = virt_to_head_page(h); | |
1555 | else | |
1556 | page = container_of((struct list_head *)h, struct page, lru); | |
1557 | ||
1b4f59e3 | 1558 | __free_slab(page->slab_cache, page); |
81819f0f CL |
1559 | } |
1560 | ||
1561 | static void free_slab(struct kmem_cache *s, struct page *page) | |
1562 | { | |
1563 | if (unlikely(s->flags & SLAB_DESTROY_BY_RCU)) { | |
da9a638c LJ |
1564 | struct rcu_head *head; |
1565 | ||
1566 | if (need_reserve_slab_rcu) { | |
1567 | int order = compound_order(page); | |
1568 | int offset = (PAGE_SIZE << order) - s->reserved; | |
1569 | ||
1570 | VM_BUG_ON(s->reserved != sizeof(*head)); | |
1571 | head = page_address(page) + offset; | |
1572 | } else { | |
bc4f610d | 1573 | head = &page->rcu_head; |
da9a638c | 1574 | } |
81819f0f CL |
1575 | |
1576 | call_rcu(head, rcu_free_slab); | |
1577 | } else | |
1578 | __free_slab(s, page); | |
1579 | } | |
1580 | ||
1581 | static void discard_slab(struct kmem_cache *s, struct page *page) | |
1582 | { | |
205ab99d | 1583 | dec_slabs_node(s, page_to_nid(page), page->objects); |
81819f0f CL |
1584 | free_slab(s, page); |
1585 | } | |
1586 | ||
1587 | /* | |
5cc6eee8 | 1588 | * Management of partially allocated slabs. |
81819f0f | 1589 | */ |
1e4dd946 SR |
1590 | static inline void |
1591 | __add_partial(struct kmem_cache_node *n, struct page *page, int tail) | |
81819f0f | 1592 | { |
e95eed57 | 1593 | n->nr_partial++; |
136333d1 | 1594 | if (tail == DEACTIVATE_TO_TAIL) |
7c2e132c CL |
1595 | list_add_tail(&page->lru, &n->partial); |
1596 | else | |
1597 | list_add(&page->lru, &n->partial); | |
81819f0f CL |
1598 | } |
1599 | ||
1e4dd946 SR |
1600 | static inline void add_partial(struct kmem_cache_node *n, |
1601 | struct page *page, int tail) | |
62e346a8 | 1602 | { |
c65c1877 | 1603 | lockdep_assert_held(&n->list_lock); |
1e4dd946 SR |
1604 | __add_partial(n, page, tail); |
1605 | } | |
c65c1877 | 1606 | |
1e4dd946 SR |
1607 | static inline void remove_partial(struct kmem_cache_node *n, |
1608 | struct page *page) | |
1609 | { | |
1610 | lockdep_assert_held(&n->list_lock); | |
52b4b950 DS |
1611 | list_del(&page->lru); |
1612 | n->nr_partial--; | |
1e4dd946 SR |
1613 | } |
1614 | ||
81819f0f | 1615 | /* |
7ced3719 CL |
1616 | * Remove slab from the partial list, freeze it and |
1617 | * return the pointer to the freelist. | |
81819f0f | 1618 | * |
497b66f2 | 1619 | * Returns a list of objects or NULL if it fails. |
81819f0f | 1620 | */ |
497b66f2 | 1621 | static inline void *acquire_slab(struct kmem_cache *s, |
acd19fd1 | 1622 | struct kmem_cache_node *n, struct page *page, |
633b0764 | 1623 | int mode, int *objects) |
81819f0f | 1624 | { |
2cfb7455 CL |
1625 | void *freelist; |
1626 | unsigned long counters; | |
1627 | struct page new; | |
1628 | ||
c65c1877 PZ |
1629 | lockdep_assert_held(&n->list_lock); |
1630 | ||
2cfb7455 CL |
1631 | /* |
1632 | * Zap the freelist and set the frozen bit. | |
1633 | * The old freelist is the list of objects for the | |
1634 | * per cpu allocation list. | |
1635 | */ | |
7ced3719 CL |
1636 | freelist = page->freelist; |
1637 | counters = page->counters; | |
1638 | new.counters = counters; | |
633b0764 | 1639 | *objects = new.objects - new.inuse; |
23910c50 | 1640 | if (mode) { |
7ced3719 | 1641 | new.inuse = page->objects; |
23910c50 PE |
1642 | new.freelist = NULL; |
1643 | } else { | |
1644 | new.freelist = freelist; | |
1645 | } | |
2cfb7455 | 1646 | |
a0132ac0 | 1647 | VM_BUG_ON(new.frozen); |
7ced3719 | 1648 | new.frozen = 1; |
2cfb7455 | 1649 | |
7ced3719 | 1650 | if (!__cmpxchg_double_slab(s, page, |
2cfb7455 | 1651 | freelist, counters, |
02d7633f | 1652 | new.freelist, new.counters, |
7ced3719 | 1653 | "acquire_slab")) |
7ced3719 | 1654 | return NULL; |
2cfb7455 CL |
1655 | |
1656 | remove_partial(n, page); | |
7ced3719 | 1657 | WARN_ON(!freelist); |
49e22585 | 1658 | return freelist; |
81819f0f CL |
1659 | } |
1660 | ||
633b0764 | 1661 | static void put_cpu_partial(struct kmem_cache *s, struct page *page, int drain); |
8ba00bb6 | 1662 | static inline bool pfmemalloc_match(struct page *page, gfp_t gfpflags); |
49e22585 | 1663 | |
81819f0f | 1664 | /* |
672bba3a | 1665 | * Try to allocate a partial slab from a specific node. |
81819f0f | 1666 | */ |
8ba00bb6 JK |
1667 | static void *get_partial_node(struct kmem_cache *s, struct kmem_cache_node *n, |
1668 | struct kmem_cache_cpu *c, gfp_t flags) | |
81819f0f | 1669 | { |
49e22585 CL |
1670 | struct page *page, *page2; |
1671 | void *object = NULL; | |
633b0764 JK |
1672 | int available = 0; |
1673 | int objects; | |
81819f0f CL |
1674 | |
1675 | /* | |
1676 | * Racy check. If we mistakenly see no partial slabs then we | |
1677 | * just allocate an empty slab. If we mistakenly try to get a | |
672bba3a CL |
1678 | * partial slab and there is none available then get_partials() |
1679 | * will return NULL. | |
81819f0f CL |
1680 | */ |
1681 | if (!n || !n->nr_partial) | |
1682 | return NULL; | |
1683 | ||
1684 | spin_lock(&n->list_lock); | |
49e22585 | 1685 | list_for_each_entry_safe(page, page2, &n->partial, lru) { |
8ba00bb6 | 1686 | void *t; |
49e22585 | 1687 | |
8ba00bb6 JK |
1688 | if (!pfmemalloc_match(page, flags)) |
1689 | continue; | |
1690 | ||
633b0764 | 1691 | t = acquire_slab(s, n, page, object == NULL, &objects); |
49e22585 CL |
1692 | if (!t) |
1693 | break; | |
1694 | ||
633b0764 | 1695 | available += objects; |
12d79634 | 1696 | if (!object) { |
49e22585 | 1697 | c->page = page; |
49e22585 | 1698 | stat(s, ALLOC_FROM_PARTIAL); |
49e22585 | 1699 | object = t; |
49e22585 | 1700 | } else { |
633b0764 | 1701 | put_cpu_partial(s, page, 0); |
8028dcea | 1702 | stat(s, CPU_PARTIAL_NODE); |
49e22585 | 1703 | } |
345c905d JK |
1704 | if (!kmem_cache_has_cpu_partial(s) |
1705 | || available > s->cpu_partial / 2) | |
49e22585 CL |
1706 | break; |
1707 | ||
497b66f2 | 1708 | } |
81819f0f | 1709 | spin_unlock(&n->list_lock); |
497b66f2 | 1710 | return object; |
81819f0f CL |
1711 | } |
1712 | ||
1713 | /* | |
672bba3a | 1714 | * Get a page from somewhere. Search in increasing NUMA distances. |
81819f0f | 1715 | */ |
de3ec035 | 1716 | static void *get_any_partial(struct kmem_cache *s, gfp_t flags, |
acd19fd1 | 1717 | struct kmem_cache_cpu *c) |
81819f0f CL |
1718 | { |
1719 | #ifdef CONFIG_NUMA | |
1720 | struct zonelist *zonelist; | |
dd1a239f | 1721 | struct zoneref *z; |
54a6eb5c MG |
1722 | struct zone *zone; |
1723 | enum zone_type high_zoneidx = gfp_zone(flags); | |
497b66f2 | 1724 | void *object; |
cc9a6c87 | 1725 | unsigned int cpuset_mems_cookie; |
81819f0f CL |
1726 | |
1727 | /* | |
672bba3a CL |
1728 | * The defrag ratio allows a configuration of the tradeoffs between |
1729 | * inter node defragmentation and node local allocations. A lower | |
1730 | * defrag_ratio increases the tendency to do local allocations | |
1731 | * instead of attempting to obtain partial slabs from other nodes. | |
81819f0f | 1732 | * |
672bba3a CL |
1733 | * If the defrag_ratio is set to 0 then kmalloc() always |
1734 | * returns node local objects. If the ratio is higher then kmalloc() | |
1735 | * may return off node objects because partial slabs are obtained | |
1736 | * from other nodes and filled up. | |
81819f0f | 1737 | * |
6446faa2 | 1738 | * If /sys/kernel/slab/xx/defrag_ratio is set to 100 (which makes |
672bba3a CL |
1739 | * defrag_ratio = 1000) then every (well almost) allocation will |
1740 | * first attempt to defrag slab caches on other nodes. This means | |
1741 | * scanning over all nodes to look for partial slabs which may be | |
1742 | * expensive if we do it every time we are trying to find a slab | |
1743 | * with available objects. | |
81819f0f | 1744 | */ |
9824601e CL |
1745 | if (!s->remote_node_defrag_ratio || |
1746 | get_cycles() % 1024 > s->remote_node_defrag_ratio) | |
81819f0f CL |
1747 | return NULL; |
1748 | ||
cc9a6c87 | 1749 | do { |
d26914d1 | 1750 | cpuset_mems_cookie = read_mems_allowed_begin(); |
2a389610 | 1751 | zonelist = node_zonelist(mempolicy_slab_node(), flags); |
cc9a6c87 MG |
1752 | for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) { |
1753 | struct kmem_cache_node *n; | |
1754 | ||
1755 | n = get_node(s, zone_to_nid(zone)); | |
1756 | ||
dee2f8aa | 1757 | if (n && cpuset_zone_allowed(zone, flags) && |
cc9a6c87 | 1758 | n->nr_partial > s->min_partial) { |
8ba00bb6 | 1759 | object = get_partial_node(s, n, c, flags); |
cc9a6c87 MG |
1760 | if (object) { |
1761 | /* | |
d26914d1 MG |
1762 | * Don't check read_mems_allowed_retry() |
1763 | * here - if mems_allowed was updated in | |
1764 | * parallel, that was a harmless race | |
1765 | * between allocation and the cpuset | |
1766 | * update | |
cc9a6c87 | 1767 | */ |
cc9a6c87 MG |
1768 | return object; |
1769 | } | |
c0ff7453 | 1770 | } |
81819f0f | 1771 | } |
d26914d1 | 1772 | } while (read_mems_allowed_retry(cpuset_mems_cookie)); |
81819f0f CL |
1773 | #endif |
1774 | return NULL; | |
1775 | } | |
1776 | ||
1777 | /* | |
1778 | * Get a partial page, lock it and return it. | |
1779 | */ | |
497b66f2 | 1780 | static void *get_partial(struct kmem_cache *s, gfp_t flags, int node, |
acd19fd1 | 1781 | struct kmem_cache_cpu *c) |
81819f0f | 1782 | { |
497b66f2 | 1783 | void *object; |
a561ce00 JK |
1784 | int searchnode = node; |
1785 | ||
1786 | if (node == NUMA_NO_NODE) | |
1787 | searchnode = numa_mem_id(); | |
1788 | else if (!node_present_pages(node)) | |
1789 | searchnode = node_to_mem_node(node); | |
81819f0f | 1790 | |
8ba00bb6 | 1791 | object = get_partial_node(s, get_node(s, searchnode), c, flags); |
497b66f2 CL |
1792 | if (object || node != NUMA_NO_NODE) |
1793 | return object; | |
81819f0f | 1794 | |
acd19fd1 | 1795 | return get_any_partial(s, flags, c); |
81819f0f CL |
1796 | } |
1797 | ||
8a5ec0ba CL |
1798 | #ifdef CONFIG_PREEMPT |
1799 | /* | |
1800 | * Calculate the next globally unique transaction for disambiguiation | |
1801 | * during cmpxchg. The transactions start with the cpu number and are then | |
1802 | * incremented by CONFIG_NR_CPUS. | |
1803 | */ | |
1804 | #define TID_STEP roundup_pow_of_two(CONFIG_NR_CPUS) | |
1805 | #else | |
1806 | /* | |
1807 | * No preemption supported therefore also no need to check for | |
1808 | * different cpus. | |
1809 | */ | |
1810 | #define TID_STEP 1 | |
1811 | #endif | |
1812 | ||
1813 | static inline unsigned long next_tid(unsigned long tid) | |
1814 | { | |
1815 | return tid + TID_STEP; | |
1816 | } | |
1817 | ||
1818 | static inline unsigned int tid_to_cpu(unsigned long tid) | |
1819 | { | |
1820 | return tid % TID_STEP; | |
1821 | } | |
1822 | ||
1823 | static inline unsigned long tid_to_event(unsigned long tid) | |
1824 | { | |
1825 | return tid / TID_STEP; | |
1826 | } | |
1827 | ||
1828 | static inline unsigned int init_tid(int cpu) | |
1829 | { | |
1830 | return cpu; | |
1831 | } | |
1832 | ||
1833 | static inline void note_cmpxchg_failure(const char *n, | |
1834 | const struct kmem_cache *s, unsigned long tid) | |
1835 | { | |
1836 | #ifdef SLUB_DEBUG_CMPXCHG | |
1837 | unsigned long actual_tid = __this_cpu_read(s->cpu_slab->tid); | |
1838 | ||
f9f58285 | 1839 | pr_info("%s %s: cmpxchg redo ", n, s->name); |
8a5ec0ba CL |
1840 | |
1841 | #ifdef CONFIG_PREEMPT | |
1842 | if (tid_to_cpu(tid) != tid_to_cpu(actual_tid)) | |
f9f58285 | 1843 | pr_warn("due to cpu change %d -> %d\n", |
8a5ec0ba CL |
1844 | tid_to_cpu(tid), tid_to_cpu(actual_tid)); |
1845 | else | |
1846 | #endif | |
1847 | if (tid_to_event(tid) != tid_to_event(actual_tid)) | |
f9f58285 | 1848 | pr_warn("due to cpu running other code. Event %ld->%ld\n", |
8a5ec0ba CL |
1849 | tid_to_event(tid), tid_to_event(actual_tid)); |
1850 | else | |
f9f58285 | 1851 | pr_warn("for unknown reason: actual=%lx was=%lx target=%lx\n", |
8a5ec0ba CL |
1852 | actual_tid, tid, next_tid(tid)); |
1853 | #endif | |
4fdccdfb | 1854 | stat(s, CMPXCHG_DOUBLE_CPU_FAIL); |
8a5ec0ba CL |
1855 | } |
1856 | ||
788e1aad | 1857 | static void init_kmem_cache_cpus(struct kmem_cache *s) |
8a5ec0ba | 1858 | { |
8a5ec0ba CL |
1859 | int cpu; |
1860 | ||
1861 | for_each_possible_cpu(cpu) | |
1862 | per_cpu_ptr(s->cpu_slab, cpu)->tid = init_tid(cpu); | |
8a5ec0ba | 1863 | } |
2cfb7455 | 1864 | |
81819f0f CL |
1865 | /* |
1866 | * Remove the cpu slab | |
1867 | */ | |
d0e0ac97 CG |
1868 | static void deactivate_slab(struct kmem_cache *s, struct page *page, |
1869 | void *freelist) | |
81819f0f | 1870 | { |
2cfb7455 | 1871 | enum slab_modes { M_NONE, M_PARTIAL, M_FULL, M_FREE }; |
2cfb7455 CL |
1872 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); |
1873 | int lock = 0; | |
1874 | enum slab_modes l = M_NONE, m = M_NONE; | |
2cfb7455 | 1875 | void *nextfree; |
136333d1 | 1876 | int tail = DEACTIVATE_TO_HEAD; |
2cfb7455 CL |
1877 | struct page new; |
1878 | struct page old; | |
1879 | ||
1880 | if (page->freelist) { | |
84e554e6 | 1881 | stat(s, DEACTIVATE_REMOTE_FREES); |
136333d1 | 1882 | tail = DEACTIVATE_TO_TAIL; |
2cfb7455 CL |
1883 | } |
1884 | ||
894b8788 | 1885 | /* |
2cfb7455 CL |
1886 | * Stage one: Free all available per cpu objects back |
1887 | * to the page freelist while it is still frozen. Leave the | |
1888 | * last one. | |
1889 | * | |
1890 | * There is no need to take the list->lock because the page | |
1891 | * is still frozen. | |
1892 | */ | |
1893 | while (freelist && (nextfree = get_freepointer(s, freelist))) { | |
1894 | void *prior; | |
1895 | unsigned long counters; | |
1896 | ||
1897 | do { | |
1898 | prior = page->freelist; | |
1899 | counters = page->counters; | |
1900 | set_freepointer(s, freelist, prior); | |
1901 | new.counters = counters; | |
1902 | new.inuse--; | |
a0132ac0 | 1903 | VM_BUG_ON(!new.frozen); |
2cfb7455 | 1904 | |
1d07171c | 1905 | } while (!__cmpxchg_double_slab(s, page, |
2cfb7455 CL |
1906 | prior, counters, |
1907 | freelist, new.counters, | |
1908 | "drain percpu freelist")); | |
1909 | ||
1910 | freelist = nextfree; | |
1911 | } | |
1912 | ||
894b8788 | 1913 | /* |
2cfb7455 CL |
1914 | * Stage two: Ensure that the page is unfrozen while the |
1915 | * list presence reflects the actual number of objects | |
1916 | * during unfreeze. | |
1917 | * | |
1918 | * We setup the list membership and then perform a cmpxchg | |
1919 | * with the count. If there is a mismatch then the page | |
1920 | * is not unfrozen but the page is on the wrong list. | |
1921 | * | |
1922 | * Then we restart the process which may have to remove | |
1923 | * the page from the list that we just put it on again | |
1924 | * because the number of objects in the slab may have | |
1925 | * changed. | |
894b8788 | 1926 | */ |
2cfb7455 | 1927 | redo: |
894b8788 | 1928 | |
2cfb7455 CL |
1929 | old.freelist = page->freelist; |
1930 | old.counters = page->counters; | |
a0132ac0 | 1931 | VM_BUG_ON(!old.frozen); |
7c2e132c | 1932 | |
2cfb7455 CL |
1933 | /* Determine target state of the slab */ |
1934 | new.counters = old.counters; | |
1935 | if (freelist) { | |
1936 | new.inuse--; | |
1937 | set_freepointer(s, freelist, old.freelist); | |
1938 | new.freelist = freelist; | |
1939 | } else | |
1940 | new.freelist = old.freelist; | |
1941 | ||
1942 | new.frozen = 0; | |
1943 | ||
8a5b20ae | 1944 | if (!new.inuse && n->nr_partial >= s->min_partial) |
2cfb7455 CL |
1945 | m = M_FREE; |
1946 | else if (new.freelist) { | |
1947 | m = M_PARTIAL; | |
1948 | if (!lock) { | |
1949 | lock = 1; | |
1950 | /* | |
1951 | * Taking the spinlock removes the possiblity | |
1952 | * that acquire_slab() will see a slab page that | |
1953 | * is frozen | |
1954 | */ | |
1955 | spin_lock(&n->list_lock); | |
1956 | } | |
1957 | } else { | |
1958 | m = M_FULL; | |
1959 | if (kmem_cache_debug(s) && !lock) { | |
1960 | lock = 1; | |
1961 | /* | |
1962 | * This also ensures that the scanning of full | |
1963 | * slabs from diagnostic functions will not see | |
1964 | * any frozen slabs. | |
1965 | */ | |
1966 | spin_lock(&n->list_lock); | |
1967 | } | |
1968 | } | |
1969 | ||
1970 | if (l != m) { | |
1971 | ||
1972 | if (l == M_PARTIAL) | |
1973 | ||
1974 | remove_partial(n, page); | |
1975 | ||
1976 | else if (l == M_FULL) | |
894b8788 | 1977 | |
c65c1877 | 1978 | remove_full(s, n, page); |
2cfb7455 CL |
1979 | |
1980 | if (m == M_PARTIAL) { | |
1981 | ||
1982 | add_partial(n, page, tail); | |
136333d1 | 1983 | stat(s, tail); |
2cfb7455 CL |
1984 | |
1985 | } else if (m == M_FULL) { | |
894b8788 | 1986 | |
2cfb7455 CL |
1987 | stat(s, DEACTIVATE_FULL); |
1988 | add_full(s, n, page); | |
1989 | ||
1990 | } | |
1991 | } | |
1992 | ||
1993 | l = m; | |
1d07171c | 1994 | if (!__cmpxchg_double_slab(s, page, |
2cfb7455 CL |
1995 | old.freelist, old.counters, |
1996 | new.freelist, new.counters, | |
1997 | "unfreezing slab")) | |
1998 | goto redo; | |
1999 | ||
2cfb7455 CL |
2000 | if (lock) |
2001 | spin_unlock(&n->list_lock); | |
2002 | ||
2003 | if (m == M_FREE) { | |
2004 | stat(s, DEACTIVATE_EMPTY); | |
2005 | discard_slab(s, page); | |
2006 | stat(s, FREE_SLAB); | |
894b8788 | 2007 | } |
81819f0f CL |
2008 | } |
2009 | ||
d24ac77f JK |
2010 | /* |
2011 | * Unfreeze all the cpu partial slabs. | |
2012 | * | |
59a09917 CL |
2013 | * This function must be called with interrupts disabled |
2014 | * for the cpu using c (or some other guarantee must be there | |
2015 | * to guarantee no concurrent accesses). | |
d24ac77f | 2016 | */ |
59a09917 CL |
2017 | static void unfreeze_partials(struct kmem_cache *s, |
2018 | struct kmem_cache_cpu *c) | |
49e22585 | 2019 | { |
345c905d | 2020 | #ifdef CONFIG_SLUB_CPU_PARTIAL |
43d77867 | 2021 | struct kmem_cache_node *n = NULL, *n2 = NULL; |
9ada1934 | 2022 | struct page *page, *discard_page = NULL; |
49e22585 CL |
2023 | |
2024 | while ((page = c->partial)) { | |
49e22585 CL |
2025 | struct page new; |
2026 | struct page old; | |
2027 | ||
2028 | c->partial = page->next; | |
43d77867 JK |
2029 | |
2030 | n2 = get_node(s, page_to_nid(page)); | |
2031 | if (n != n2) { | |
2032 | if (n) | |
2033 | spin_unlock(&n->list_lock); | |
2034 | ||
2035 | n = n2; | |
2036 | spin_lock(&n->list_lock); | |
2037 | } | |
49e22585 CL |
2038 | |
2039 | do { | |
2040 | ||
2041 | old.freelist = page->freelist; | |
2042 | old.counters = page->counters; | |
a0132ac0 | 2043 | VM_BUG_ON(!old.frozen); |
49e22585 CL |
2044 | |
2045 | new.counters = old.counters; | |
2046 | new.freelist = old.freelist; | |
2047 | ||
2048 | new.frozen = 0; | |
2049 | ||
d24ac77f | 2050 | } while (!__cmpxchg_double_slab(s, page, |
49e22585 CL |
2051 | old.freelist, old.counters, |
2052 | new.freelist, new.counters, | |
2053 | "unfreezing slab")); | |
2054 | ||
8a5b20ae | 2055 | if (unlikely(!new.inuse && n->nr_partial >= s->min_partial)) { |
9ada1934 SL |
2056 | page->next = discard_page; |
2057 | discard_page = page; | |
43d77867 JK |
2058 | } else { |
2059 | add_partial(n, page, DEACTIVATE_TO_TAIL); | |
2060 | stat(s, FREE_ADD_PARTIAL); | |
49e22585 CL |
2061 | } |
2062 | } | |
2063 | ||
2064 | if (n) | |
2065 | spin_unlock(&n->list_lock); | |
9ada1934 SL |
2066 | |
2067 | while (discard_page) { | |
2068 | page = discard_page; | |
2069 | discard_page = discard_page->next; | |
2070 | ||
2071 | stat(s, DEACTIVATE_EMPTY); | |
2072 | discard_slab(s, page); | |
2073 | stat(s, FREE_SLAB); | |
2074 | } | |
345c905d | 2075 | #endif |
49e22585 CL |
2076 | } |
2077 | ||
2078 | /* | |
2079 | * Put a page that was just frozen (in __slab_free) into a partial page | |
2080 | * slot if available. This is done without interrupts disabled and without | |
2081 | * preemption disabled. The cmpxchg is racy and may put the partial page | |
2082 | * onto a random cpus partial slot. | |
2083 | * | |
2084 | * If we did not find a slot then simply move all the partials to the | |
2085 | * per node partial list. | |
2086 | */ | |
633b0764 | 2087 | static void put_cpu_partial(struct kmem_cache *s, struct page *page, int drain) |
49e22585 | 2088 | { |
345c905d | 2089 | #ifdef CONFIG_SLUB_CPU_PARTIAL |
49e22585 CL |
2090 | struct page *oldpage; |
2091 | int pages; | |
2092 | int pobjects; | |
2093 | ||
d6e0b7fa | 2094 | preempt_disable(); |
49e22585 CL |
2095 | do { |
2096 | pages = 0; | |
2097 | pobjects = 0; | |
2098 | oldpage = this_cpu_read(s->cpu_slab->partial); | |
2099 | ||
2100 | if (oldpage) { | |
2101 | pobjects = oldpage->pobjects; | |
2102 | pages = oldpage->pages; | |
2103 | if (drain && pobjects > s->cpu_partial) { | |
2104 | unsigned long flags; | |
2105 | /* | |
2106 | * partial array is full. Move the existing | |
2107 | * set to the per node partial list. | |
2108 | */ | |
2109 | local_irq_save(flags); | |
59a09917 | 2110 | unfreeze_partials(s, this_cpu_ptr(s->cpu_slab)); |
49e22585 | 2111 | local_irq_restore(flags); |
e24fc410 | 2112 | oldpage = NULL; |
49e22585 CL |
2113 | pobjects = 0; |
2114 | pages = 0; | |
8028dcea | 2115 | stat(s, CPU_PARTIAL_DRAIN); |
49e22585 CL |
2116 | } |
2117 | } | |
2118 | ||
2119 | pages++; | |
2120 | pobjects += page->objects - page->inuse; | |
2121 | ||
2122 | page->pages = pages; | |
2123 | page->pobjects = pobjects; | |
2124 | page->next = oldpage; | |
2125 | ||
d0e0ac97 CG |
2126 | } while (this_cpu_cmpxchg(s->cpu_slab->partial, oldpage, page) |
2127 | != oldpage); | |
d6e0b7fa VD |
2128 | if (unlikely(!s->cpu_partial)) { |
2129 | unsigned long flags; | |
2130 | ||
2131 | local_irq_save(flags); | |
2132 | unfreeze_partials(s, this_cpu_ptr(s->cpu_slab)); | |
2133 | local_irq_restore(flags); | |
2134 | } | |
2135 | preempt_enable(); | |
345c905d | 2136 | #endif |
49e22585 CL |
2137 | } |
2138 | ||
dfb4f096 | 2139 | static inline void flush_slab(struct kmem_cache *s, struct kmem_cache_cpu *c) |
81819f0f | 2140 | { |
84e554e6 | 2141 | stat(s, CPUSLAB_FLUSH); |
c17dda40 CL |
2142 | deactivate_slab(s, c->page, c->freelist); |
2143 | ||
2144 | c->tid = next_tid(c->tid); | |
2145 | c->page = NULL; | |
2146 | c->freelist = NULL; | |
81819f0f CL |
2147 | } |
2148 | ||
2149 | /* | |
2150 | * Flush cpu slab. | |
6446faa2 | 2151 | * |
81819f0f CL |
2152 | * Called from IPI handler with interrupts disabled. |
2153 | */ | |
0c710013 | 2154 | static inline void __flush_cpu_slab(struct kmem_cache *s, int cpu) |
81819f0f | 2155 | { |
9dfc6e68 | 2156 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu); |
81819f0f | 2157 | |
49e22585 CL |
2158 | if (likely(c)) { |
2159 | if (c->page) | |
2160 | flush_slab(s, c); | |
2161 | ||
59a09917 | 2162 | unfreeze_partials(s, c); |
49e22585 | 2163 | } |
81819f0f CL |
2164 | } |
2165 | ||
2166 | static void flush_cpu_slab(void *d) | |
2167 | { | |
2168 | struct kmem_cache *s = d; | |
81819f0f | 2169 | |
dfb4f096 | 2170 | __flush_cpu_slab(s, smp_processor_id()); |
81819f0f CL |
2171 | } |
2172 | ||
a8364d55 GBY |
2173 | static bool has_cpu_slab(int cpu, void *info) |
2174 | { | |
2175 | struct kmem_cache *s = info; | |
2176 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu); | |
2177 | ||
02e1a9cd | 2178 | return c->page || c->partial; |
a8364d55 GBY |
2179 | } |
2180 | ||
81819f0f CL |
2181 | static void flush_all(struct kmem_cache *s) |
2182 | { | |
a8364d55 | 2183 | on_each_cpu_cond(has_cpu_slab, flush_cpu_slab, s, 1, GFP_ATOMIC); |
81819f0f CL |
2184 | } |
2185 | ||
dfb4f096 CL |
2186 | /* |
2187 | * Check if the objects in a per cpu structure fit numa | |
2188 | * locality expectations. | |
2189 | */ | |
57d437d2 | 2190 | static inline int node_match(struct page *page, int node) |
dfb4f096 CL |
2191 | { |
2192 | #ifdef CONFIG_NUMA | |
4d7868e6 | 2193 | if (!page || (node != NUMA_NO_NODE && page_to_nid(page) != node)) |
dfb4f096 CL |
2194 | return 0; |
2195 | #endif | |
2196 | return 1; | |
2197 | } | |
2198 | ||
9a02d699 | 2199 | #ifdef CONFIG_SLUB_DEBUG |
781b2ba6 PE |
2200 | static int count_free(struct page *page) |
2201 | { | |
2202 | return page->objects - page->inuse; | |
2203 | } | |
2204 | ||
9a02d699 DR |
2205 | static inline unsigned long node_nr_objs(struct kmem_cache_node *n) |
2206 | { | |
2207 | return atomic_long_read(&n->total_objects); | |
2208 | } | |
2209 | #endif /* CONFIG_SLUB_DEBUG */ | |
2210 | ||
2211 | #if defined(CONFIG_SLUB_DEBUG) || defined(CONFIG_SYSFS) | |
781b2ba6 PE |
2212 | static unsigned long count_partial(struct kmem_cache_node *n, |
2213 | int (*get_count)(struct page *)) | |
2214 | { | |
2215 | unsigned long flags; | |
2216 | unsigned long x = 0; | |
2217 | struct page *page; | |
2218 | ||
2219 | spin_lock_irqsave(&n->list_lock, flags); | |
2220 | list_for_each_entry(page, &n->partial, lru) | |
2221 | x += get_count(page); | |
2222 | spin_unlock_irqrestore(&n->list_lock, flags); | |
2223 | return x; | |
2224 | } | |
9a02d699 | 2225 | #endif /* CONFIG_SLUB_DEBUG || CONFIG_SYSFS */ |
26c02cf0 | 2226 | |
781b2ba6 PE |
2227 | static noinline void |
2228 | slab_out_of_memory(struct kmem_cache *s, gfp_t gfpflags, int nid) | |
2229 | { | |
9a02d699 DR |
2230 | #ifdef CONFIG_SLUB_DEBUG |
2231 | static DEFINE_RATELIMIT_STATE(slub_oom_rs, DEFAULT_RATELIMIT_INTERVAL, | |
2232 | DEFAULT_RATELIMIT_BURST); | |
781b2ba6 | 2233 | int node; |
fa45dc25 | 2234 | struct kmem_cache_node *n; |
781b2ba6 | 2235 | |
9a02d699 DR |
2236 | if ((gfpflags & __GFP_NOWARN) || !__ratelimit(&slub_oom_rs)) |
2237 | return; | |
2238 | ||
5b3810e5 VB |
2239 | pr_warn("SLUB: Unable to allocate memory on node %d, gfp=%#x(%pGg)\n", |
2240 | nid, gfpflags, &gfpflags); | |
f9f58285 FF |
2241 | pr_warn(" cache: %s, object size: %d, buffer size: %d, default order: %d, min order: %d\n", |
2242 | s->name, s->object_size, s->size, oo_order(s->oo), | |
2243 | oo_order(s->min)); | |
781b2ba6 | 2244 | |
3b0efdfa | 2245 | if (oo_order(s->min) > get_order(s->object_size)) |
f9f58285 FF |
2246 | pr_warn(" %s debugging increased min order, use slub_debug=O to disable.\n", |
2247 | s->name); | |
fa5ec8a1 | 2248 | |
fa45dc25 | 2249 | for_each_kmem_cache_node(s, node, n) { |
781b2ba6 PE |
2250 | unsigned long nr_slabs; |
2251 | unsigned long nr_objs; | |
2252 | unsigned long nr_free; | |
2253 | ||
26c02cf0 AB |
2254 | nr_free = count_partial(n, count_free); |
2255 | nr_slabs = node_nr_slabs(n); | |
2256 | nr_objs = node_nr_objs(n); | |
781b2ba6 | 2257 | |
f9f58285 | 2258 | pr_warn(" node %d: slabs: %ld, objs: %ld, free: %ld\n", |
781b2ba6 PE |
2259 | node, nr_slabs, nr_objs, nr_free); |
2260 | } | |
9a02d699 | 2261 | #endif |
781b2ba6 PE |
2262 | } |
2263 | ||
497b66f2 CL |
2264 | static inline void *new_slab_objects(struct kmem_cache *s, gfp_t flags, |
2265 | int node, struct kmem_cache_cpu **pc) | |
2266 | { | |
6faa6833 | 2267 | void *freelist; |
188fd063 CL |
2268 | struct kmem_cache_cpu *c = *pc; |
2269 | struct page *page; | |
497b66f2 | 2270 | |
188fd063 | 2271 | freelist = get_partial(s, flags, node, c); |
497b66f2 | 2272 | |
188fd063 CL |
2273 | if (freelist) |
2274 | return freelist; | |
2275 | ||
2276 | page = new_slab(s, flags, node); | |
497b66f2 | 2277 | if (page) { |
7c8e0181 | 2278 | c = raw_cpu_ptr(s->cpu_slab); |
497b66f2 CL |
2279 | if (c->page) |
2280 | flush_slab(s, c); | |
2281 | ||
2282 | /* | |
2283 | * No other reference to the page yet so we can | |
2284 | * muck around with it freely without cmpxchg | |
2285 | */ | |
6faa6833 | 2286 | freelist = page->freelist; |
497b66f2 CL |
2287 | page->freelist = NULL; |
2288 | ||
2289 | stat(s, ALLOC_SLAB); | |
497b66f2 CL |
2290 | c->page = page; |
2291 | *pc = c; | |
2292 | } else | |
6faa6833 | 2293 | freelist = NULL; |
497b66f2 | 2294 | |
6faa6833 | 2295 | return freelist; |
497b66f2 CL |
2296 | } |
2297 | ||
072bb0aa MG |
2298 | static inline bool pfmemalloc_match(struct page *page, gfp_t gfpflags) |
2299 | { | |
2300 | if (unlikely(PageSlabPfmemalloc(page))) | |
2301 | return gfp_pfmemalloc_allowed(gfpflags); | |
2302 | ||
2303 | return true; | |
2304 | } | |
2305 | ||
213eeb9f | 2306 | /* |
d0e0ac97 CG |
2307 | * Check the page->freelist of a page and either transfer the freelist to the |
2308 | * per cpu freelist or deactivate the page. | |
213eeb9f CL |
2309 | * |
2310 | * The page is still frozen if the return value is not NULL. | |
2311 | * | |
2312 | * If this function returns NULL then the page has been unfrozen. | |
d24ac77f JK |
2313 | * |
2314 | * This function must be called with interrupt disabled. | |
213eeb9f CL |
2315 | */ |
2316 | static inline void *get_freelist(struct kmem_cache *s, struct page *page) | |
2317 | { | |
2318 | struct page new; | |
2319 | unsigned long counters; | |
2320 | void *freelist; | |
2321 | ||
2322 | do { | |
2323 | freelist = page->freelist; | |
2324 | counters = page->counters; | |
6faa6833 | 2325 | |
213eeb9f | 2326 | new.counters = counters; |
a0132ac0 | 2327 | VM_BUG_ON(!new.frozen); |
213eeb9f CL |
2328 | |
2329 | new.inuse = page->objects; | |
2330 | new.frozen = freelist != NULL; | |
2331 | ||
d24ac77f | 2332 | } while (!__cmpxchg_double_slab(s, page, |
213eeb9f CL |
2333 | freelist, counters, |
2334 | NULL, new.counters, | |
2335 | "get_freelist")); | |
2336 | ||
2337 | return freelist; | |
2338 | } | |
2339 | ||
81819f0f | 2340 | /* |
894b8788 CL |
2341 | * Slow path. The lockless freelist is empty or we need to perform |
2342 | * debugging duties. | |
2343 | * | |
894b8788 CL |
2344 | * Processing is still very fast if new objects have been freed to the |
2345 | * regular freelist. In that case we simply take over the regular freelist | |
2346 | * as the lockless freelist and zap the regular freelist. | |
81819f0f | 2347 | * |
894b8788 CL |
2348 | * If that is not working then we fall back to the partial lists. We take the |
2349 | * first element of the freelist as the object to allocate now and move the | |
2350 | * rest of the freelist to the lockless freelist. | |
81819f0f | 2351 | * |
894b8788 | 2352 | * And if we were unable to get a new slab from the partial slab lists then |
6446faa2 CL |
2353 | * we need to allocate a new slab. This is the slowest path since it involves |
2354 | * a call to the page allocator and the setup of a new slab. | |
a380a3c7 CL |
2355 | * |
2356 | * Version of __slab_alloc to use when we know that interrupts are | |
2357 | * already disabled (which is the case for bulk allocation). | |
81819f0f | 2358 | */ |
a380a3c7 | 2359 | static void *___slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node, |
ce71e27c | 2360 | unsigned long addr, struct kmem_cache_cpu *c) |
81819f0f | 2361 | { |
6faa6833 | 2362 | void *freelist; |
f6e7def7 | 2363 | struct page *page; |
81819f0f | 2364 | |
f6e7def7 CL |
2365 | page = c->page; |
2366 | if (!page) | |
81819f0f | 2367 | goto new_slab; |
49e22585 | 2368 | redo: |
6faa6833 | 2369 | |
57d437d2 | 2370 | if (unlikely(!node_match(page, node))) { |
a561ce00 JK |
2371 | int searchnode = node; |
2372 | ||
2373 | if (node != NUMA_NO_NODE && !node_present_pages(node)) | |
2374 | searchnode = node_to_mem_node(node); | |
2375 | ||
2376 | if (unlikely(!node_match(page, searchnode))) { | |
2377 | stat(s, ALLOC_NODE_MISMATCH); | |
2378 | deactivate_slab(s, page, c->freelist); | |
2379 | c->page = NULL; | |
2380 | c->freelist = NULL; | |
2381 | goto new_slab; | |
2382 | } | |
fc59c053 | 2383 | } |
6446faa2 | 2384 | |
072bb0aa MG |
2385 | /* |
2386 | * By rights, we should be searching for a slab page that was | |
2387 | * PFMEMALLOC but right now, we are losing the pfmemalloc | |
2388 | * information when the page leaves the per-cpu allocator | |
2389 | */ | |
2390 | if (unlikely(!pfmemalloc_match(page, gfpflags))) { | |
2391 | deactivate_slab(s, page, c->freelist); | |
2392 | c->page = NULL; | |
2393 | c->freelist = NULL; | |
2394 | goto new_slab; | |
2395 | } | |
2396 | ||
73736e03 | 2397 | /* must check again c->freelist in case of cpu migration or IRQ */ |
6faa6833 CL |
2398 | freelist = c->freelist; |
2399 | if (freelist) | |
73736e03 | 2400 | goto load_freelist; |
03e404af | 2401 | |
f6e7def7 | 2402 | freelist = get_freelist(s, page); |
6446faa2 | 2403 | |
6faa6833 | 2404 | if (!freelist) { |
03e404af CL |
2405 | c->page = NULL; |
2406 | stat(s, DEACTIVATE_BYPASS); | |
fc59c053 | 2407 | goto new_slab; |
03e404af | 2408 | } |
6446faa2 | 2409 | |
84e554e6 | 2410 | stat(s, ALLOC_REFILL); |
6446faa2 | 2411 | |
894b8788 | 2412 | load_freelist: |
507effea CL |
2413 | /* |
2414 | * freelist is pointing to the list of objects to be used. | |
2415 | * page is pointing to the page from which the objects are obtained. | |
2416 | * That page must be frozen for per cpu allocations to work. | |
2417 | */ | |
a0132ac0 | 2418 | VM_BUG_ON(!c->page->frozen); |
6faa6833 | 2419 | c->freelist = get_freepointer(s, freelist); |
8a5ec0ba | 2420 | c->tid = next_tid(c->tid); |
6faa6833 | 2421 | return freelist; |
81819f0f | 2422 | |
81819f0f | 2423 | new_slab: |
2cfb7455 | 2424 | |
49e22585 | 2425 | if (c->partial) { |
f6e7def7 CL |
2426 | page = c->page = c->partial; |
2427 | c->partial = page->next; | |
49e22585 CL |
2428 | stat(s, CPU_PARTIAL_ALLOC); |
2429 | c->freelist = NULL; | |
2430 | goto redo; | |
81819f0f CL |
2431 | } |
2432 | ||
188fd063 | 2433 | freelist = new_slab_objects(s, gfpflags, node, &c); |
01ad8a7b | 2434 | |
f4697436 | 2435 | if (unlikely(!freelist)) { |
9a02d699 | 2436 | slab_out_of_memory(s, gfpflags, node); |
f4697436 | 2437 | return NULL; |
81819f0f | 2438 | } |
2cfb7455 | 2439 | |
f6e7def7 | 2440 | page = c->page; |
5091b74a | 2441 | if (likely(!kmem_cache_debug(s) && pfmemalloc_match(page, gfpflags))) |
4b6f0750 | 2442 | goto load_freelist; |
2cfb7455 | 2443 | |
497b66f2 | 2444 | /* Only entered in the debug case */ |
d0e0ac97 CG |
2445 | if (kmem_cache_debug(s) && |
2446 | !alloc_debug_processing(s, page, freelist, addr)) | |
497b66f2 | 2447 | goto new_slab; /* Slab failed checks. Next slab needed */ |
894b8788 | 2448 | |
f6e7def7 | 2449 | deactivate_slab(s, page, get_freepointer(s, freelist)); |
c17dda40 CL |
2450 | c->page = NULL; |
2451 | c->freelist = NULL; | |
6faa6833 | 2452 | return freelist; |
894b8788 CL |
2453 | } |
2454 | ||
a380a3c7 CL |
2455 | /* |
2456 | * Another one that disabled interrupt and compensates for possible | |
2457 | * cpu changes by refetching the per cpu area pointer. | |
2458 | */ | |
2459 | static void *__slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node, | |
2460 | unsigned long addr, struct kmem_cache_cpu *c) | |
2461 | { | |
2462 | void *p; | |
2463 | unsigned long flags; | |
2464 | ||
2465 | local_irq_save(flags); | |
2466 | #ifdef CONFIG_PREEMPT | |
2467 | /* | |
2468 | * We may have been preempted and rescheduled on a different | |
2469 | * cpu before disabling interrupts. Need to reload cpu area | |
2470 | * pointer. | |
2471 | */ | |
2472 | c = this_cpu_ptr(s->cpu_slab); | |
2473 | #endif | |
2474 | ||
2475 | p = ___slab_alloc(s, gfpflags, node, addr, c); | |
2476 | local_irq_restore(flags); | |
2477 | return p; | |
2478 | } | |
2479 | ||
894b8788 CL |
2480 | /* |
2481 | * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc) | |
2482 | * have the fastpath folded into their functions. So no function call | |
2483 | * overhead for requests that can be satisfied on the fastpath. | |
2484 | * | |
2485 | * The fastpath works by first checking if the lockless freelist can be used. | |
2486 | * If not then __slab_alloc is called for slow processing. | |
2487 | * | |
2488 | * Otherwise we can simply pick the next object from the lockless free list. | |
2489 | */ | |
2b847c3c | 2490 | static __always_inline void *slab_alloc_node(struct kmem_cache *s, |
ce71e27c | 2491 | gfp_t gfpflags, int node, unsigned long addr) |
894b8788 | 2492 | { |
03ec0ed5 | 2493 | void *object; |
dfb4f096 | 2494 | struct kmem_cache_cpu *c; |
57d437d2 | 2495 | struct page *page; |
8a5ec0ba | 2496 | unsigned long tid; |
1f84260c | 2497 | |
8135be5a VD |
2498 | s = slab_pre_alloc_hook(s, gfpflags); |
2499 | if (!s) | |
773ff60e | 2500 | return NULL; |
8a5ec0ba | 2501 | redo: |
8a5ec0ba CL |
2502 | /* |
2503 | * Must read kmem_cache cpu data via this cpu ptr. Preemption is | |
2504 | * enabled. We may switch back and forth between cpus while | |
2505 | * reading from one cpu area. That does not matter as long | |
2506 | * as we end up on the original cpu again when doing the cmpxchg. | |
7cccd80b | 2507 | * |
9aabf810 JK |
2508 | * We should guarantee that tid and kmem_cache are retrieved on |
2509 | * the same cpu. It could be different if CONFIG_PREEMPT so we need | |
2510 | * to check if it is matched or not. | |
8a5ec0ba | 2511 | */ |
9aabf810 JK |
2512 | do { |
2513 | tid = this_cpu_read(s->cpu_slab->tid); | |
2514 | c = raw_cpu_ptr(s->cpu_slab); | |
859b7a0e MR |
2515 | } while (IS_ENABLED(CONFIG_PREEMPT) && |
2516 | unlikely(tid != READ_ONCE(c->tid))); | |
9aabf810 JK |
2517 | |
2518 | /* | |
2519 | * Irqless object alloc/free algorithm used here depends on sequence | |
2520 | * of fetching cpu_slab's data. tid should be fetched before anything | |
2521 | * on c to guarantee that object and page associated with previous tid | |
2522 | * won't be used with current tid. If we fetch tid first, object and | |
2523 | * page could be one associated with next tid and our alloc/free | |
2524 | * request will be failed. In this case, we will retry. So, no problem. | |
2525 | */ | |
2526 | barrier(); | |
8a5ec0ba | 2527 | |
8a5ec0ba CL |
2528 | /* |
2529 | * The transaction ids are globally unique per cpu and per operation on | |
2530 | * a per cpu queue. Thus they can be guarantee that the cmpxchg_double | |
2531 | * occurs on the right processor and that there was no operation on the | |
2532 | * linked list in between. | |
2533 | */ | |
8a5ec0ba | 2534 | |
9dfc6e68 | 2535 | object = c->freelist; |
57d437d2 | 2536 | page = c->page; |
8eae1492 | 2537 | if (unlikely(!object || !node_match(page, node))) { |
dfb4f096 | 2538 | object = __slab_alloc(s, gfpflags, node, addr, c); |
8eae1492 DH |
2539 | stat(s, ALLOC_SLOWPATH); |
2540 | } else { | |
0ad9500e ED |
2541 | void *next_object = get_freepointer_safe(s, object); |
2542 | ||
8a5ec0ba | 2543 | /* |
25985edc | 2544 | * The cmpxchg will only match if there was no additional |
8a5ec0ba CL |
2545 | * operation and if we are on the right processor. |
2546 | * | |
d0e0ac97 CG |
2547 | * The cmpxchg does the following atomically (without lock |
2548 | * semantics!) | |
8a5ec0ba CL |
2549 | * 1. Relocate first pointer to the current per cpu area. |
2550 | * 2. Verify that tid and freelist have not been changed | |
2551 | * 3. If they were not changed replace tid and freelist | |
2552 | * | |
d0e0ac97 CG |
2553 | * Since this is without lock semantics the protection is only |
2554 | * against code executing on this cpu *not* from access by | |
2555 | * other cpus. | |
8a5ec0ba | 2556 | */ |
933393f5 | 2557 | if (unlikely(!this_cpu_cmpxchg_double( |
8a5ec0ba CL |
2558 | s->cpu_slab->freelist, s->cpu_slab->tid, |
2559 | object, tid, | |
0ad9500e | 2560 | next_object, next_tid(tid)))) { |
8a5ec0ba CL |
2561 | |
2562 | note_cmpxchg_failure("slab_alloc", s, tid); | |
2563 | goto redo; | |
2564 | } | |
0ad9500e | 2565 | prefetch_freepointer(s, next_object); |
84e554e6 | 2566 | stat(s, ALLOC_FASTPATH); |
894b8788 | 2567 | } |
8a5ec0ba | 2568 | |
74e2134f | 2569 | if (unlikely(gfpflags & __GFP_ZERO) && object) |
3b0efdfa | 2570 | memset(object, 0, s->object_size); |
d07dbea4 | 2571 | |
03ec0ed5 | 2572 | slab_post_alloc_hook(s, gfpflags, 1, &object); |
5a896d9e | 2573 | |
894b8788 | 2574 | return object; |
81819f0f CL |
2575 | } |
2576 | ||
2b847c3c EG |
2577 | static __always_inline void *slab_alloc(struct kmem_cache *s, |
2578 | gfp_t gfpflags, unsigned long addr) | |
2579 | { | |
2580 | return slab_alloc_node(s, gfpflags, NUMA_NO_NODE, addr); | |
2581 | } | |
2582 | ||
81819f0f CL |
2583 | void *kmem_cache_alloc(struct kmem_cache *s, gfp_t gfpflags) |
2584 | { | |
2b847c3c | 2585 | void *ret = slab_alloc(s, gfpflags, _RET_IP_); |
5b882be4 | 2586 | |
d0e0ac97 CG |
2587 | trace_kmem_cache_alloc(_RET_IP_, ret, s->object_size, |
2588 | s->size, gfpflags); | |
5b882be4 EGM |
2589 | |
2590 | return ret; | |
81819f0f CL |
2591 | } |
2592 | EXPORT_SYMBOL(kmem_cache_alloc); | |
2593 | ||
0f24f128 | 2594 | #ifdef CONFIG_TRACING |
4a92379b RK |
2595 | void *kmem_cache_alloc_trace(struct kmem_cache *s, gfp_t gfpflags, size_t size) |
2596 | { | |
2b847c3c | 2597 | void *ret = slab_alloc(s, gfpflags, _RET_IP_); |
4a92379b | 2598 | trace_kmalloc(_RET_IP_, ret, size, s->size, gfpflags); |
505f5dcb | 2599 | kasan_kmalloc(s, ret, size, gfpflags); |
4a92379b RK |
2600 | return ret; |
2601 | } | |
2602 | EXPORT_SYMBOL(kmem_cache_alloc_trace); | |
5b882be4 EGM |
2603 | #endif |
2604 | ||
81819f0f CL |
2605 | #ifdef CONFIG_NUMA |
2606 | void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t gfpflags, int node) | |
2607 | { | |
2b847c3c | 2608 | void *ret = slab_alloc_node(s, gfpflags, node, _RET_IP_); |
5b882be4 | 2609 | |
ca2b84cb | 2610 | trace_kmem_cache_alloc_node(_RET_IP_, ret, |
3b0efdfa | 2611 | s->object_size, s->size, gfpflags, node); |
5b882be4 EGM |
2612 | |
2613 | return ret; | |
81819f0f CL |
2614 | } |
2615 | EXPORT_SYMBOL(kmem_cache_alloc_node); | |
81819f0f | 2616 | |
0f24f128 | 2617 | #ifdef CONFIG_TRACING |
4a92379b | 2618 | void *kmem_cache_alloc_node_trace(struct kmem_cache *s, |
5b882be4 | 2619 | gfp_t gfpflags, |
4a92379b | 2620 | int node, size_t size) |
5b882be4 | 2621 | { |
2b847c3c | 2622 | void *ret = slab_alloc_node(s, gfpflags, node, _RET_IP_); |
4a92379b RK |
2623 | |
2624 | trace_kmalloc_node(_RET_IP_, ret, | |
2625 | size, s->size, gfpflags, node); | |
0316bec2 | 2626 | |
505f5dcb | 2627 | kasan_kmalloc(s, ret, size, gfpflags); |
4a92379b | 2628 | return ret; |
5b882be4 | 2629 | } |
4a92379b | 2630 | EXPORT_SYMBOL(kmem_cache_alloc_node_trace); |
5b882be4 | 2631 | #endif |
5d1f57e4 | 2632 | #endif |
5b882be4 | 2633 | |
81819f0f | 2634 | /* |
94e4d712 | 2635 | * Slow path handling. This may still be called frequently since objects |
894b8788 | 2636 | * have a longer lifetime than the cpu slabs in most processing loads. |
81819f0f | 2637 | * |
894b8788 CL |
2638 | * So we still attempt to reduce cache line usage. Just take the slab |
2639 | * lock and free the item. If there is no additional partial page | |
2640 | * handling required then we can return immediately. | |
81819f0f | 2641 | */ |
894b8788 | 2642 | static void __slab_free(struct kmem_cache *s, struct page *page, |
81084651 JDB |
2643 | void *head, void *tail, int cnt, |
2644 | unsigned long addr) | |
2645 | ||
81819f0f CL |
2646 | { |
2647 | void *prior; | |
2cfb7455 | 2648 | int was_frozen; |
2cfb7455 CL |
2649 | struct page new; |
2650 | unsigned long counters; | |
2651 | struct kmem_cache_node *n = NULL; | |
61728d1e | 2652 | unsigned long uninitialized_var(flags); |
81819f0f | 2653 | |
8a5ec0ba | 2654 | stat(s, FREE_SLOWPATH); |
81819f0f | 2655 | |
19c7ff9e | 2656 | if (kmem_cache_debug(s) && |
282acb43 | 2657 | !free_debug_processing(s, page, head, tail, cnt, addr)) |
80f08c19 | 2658 | return; |
6446faa2 | 2659 | |
2cfb7455 | 2660 | do { |
837d678d JK |
2661 | if (unlikely(n)) { |
2662 | spin_unlock_irqrestore(&n->list_lock, flags); | |
2663 | n = NULL; | |
2664 | } | |
2cfb7455 CL |
2665 | prior = page->freelist; |
2666 | counters = page->counters; | |
81084651 | 2667 | set_freepointer(s, tail, prior); |
2cfb7455 CL |
2668 | new.counters = counters; |
2669 | was_frozen = new.frozen; | |
81084651 | 2670 | new.inuse -= cnt; |
837d678d | 2671 | if ((!new.inuse || !prior) && !was_frozen) { |
49e22585 | 2672 | |
c65c1877 | 2673 | if (kmem_cache_has_cpu_partial(s) && !prior) { |
49e22585 CL |
2674 | |
2675 | /* | |
d0e0ac97 CG |
2676 | * Slab was on no list before and will be |
2677 | * partially empty | |
2678 | * We can defer the list move and instead | |
2679 | * freeze it. | |
49e22585 CL |
2680 | */ |
2681 | new.frozen = 1; | |
2682 | ||
c65c1877 | 2683 | } else { /* Needs to be taken off a list */ |
49e22585 | 2684 | |
b455def2 | 2685 | n = get_node(s, page_to_nid(page)); |
49e22585 CL |
2686 | /* |
2687 | * Speculatively acquire the list_lock. | |
2688 | * If the cmpxchg does not succeed then we may | |
2689 | * drop the list_lock without any processing. | |
2690 | * | |
2691 | * Otherwise the list_lock will synchronize with | |
2692 | * other processors updating the list of slabs. | |
2693 | */ | |
2694 | spin_lock_irqsave(&n->list_lock, flags); | |
2695 | ||
2696 | } | |
2cfb7455 | 2697 | } |
81819f0f | 2698 | |
2cfb7455 CL |
2699 | } while (!cmpxchg_double_slab(s, page, |
2700 | prior, counters, | |
81084651 | 2701 | head, new.counters, |
2cfb7455 | 2702 | "__slab_free")); |
81819f0f | 2703 | |
2cfb7455 | 2704 | if (likely(!n)) { |
49e22585 CL |
2705 | |
2706 | /* | |
2707 | * If we just froze the page then put it onto the | |
2708 | * per cpu partial list. | |
2709 | */ | |
8028dcea | 2710 | if (new.frozen && !was_frozen) { |
49e22585 | 2711 | put_cpu_partial(s, page, 1); |
8028dcea AS |
2712 | stat(s, CPU_PARTIAL_FREE); |
2713 | } | |
49e22585 | 2714 | /* |
2cfb7455 CL |
2715 | * The list lock was not taken therefore no list |
2716 | * activity can be necessary. | |
2717 | */ | |
b455def2 L |
2718 | if (was_frozen) |
2719 | stat(s, FREE_FROZEN); | |
2720 | return; | |
2721 | } | |
81819f0f | 2722 | |
8a5b20ae | 2723 | if (unlikely(!new.inuse && n->nr_partial >= s->min_partial)) |
837d678d JK |
2724 | goto slab_empty; |
2725 | ||
81819f0f | 2726 | /* |
837d678d JK |
2727 | * Objects left in the slab. If it was not on the partial list before |
2728 | * then add it. | |
81819f0f | 2729 | */ |
345c905d JK |
2730 | if (!kmem_cache_has_cpu_partial(s) && unlikely(!prior)) { |
2731 | if (kmem_cache_debug(s)) | |
c65c1877 | 2732 | remove_full(s, n, page); |
837d678d JK |
2733 | add_partial(n, page, DEACTIVATE_TO_TAIL); |
2734 | stat(s, FREE_ADD_PARTIAL); | |
8ff12cfc | 2735 | } |
80f08c19 | 2736 | spin_unlock_irqrestore(&n->list_lock, flags); |
81819f0f CL |
2737 | return; |
2738 | ||
2739 | slab_empty: | |
a973e9dd | 2740 | if (prior) { |
81819f0f | 2741 | /* |
6fbabb20 | 2742 | * Slab on the partial list. |
81819f0f | 2743 | */ |
5cc6eee8 | 2744 | remove_partial(n, page); |
84e554e6 | 2745 | stat(s, FREE_REMOVE_PARTIAL); |
c65c1877 | 2746 | } else { |
6fbabb20 | 2747 | /* Slab must be on the full list */ |
c65c1877 PZ |
2748 | remove_full(s, n, page); |
2749 | } | |
2cfb7455 | 2750 | |
80f08c19 | 2751 | spin_unlock_irqrestore(&n->list_lock, flags); |
84e554e6 | 2752 | stat(s, FREE_SLAB); |
81819f0f | 2753 | discard_slab(s, page); |
81819f0f CL |
2754 | } |
2755 | ||
894b8788 CL |
2756 | /* |
2757 | * Fastpath with forced inlining to produce a kfree and kmem_cache_free that | |
2758 | * can perform fastpath freeing without additional function calls. | |
2759 | * | |
2760 | * The fastpath is only possible if we are freeing to the current cpu slab | |
2761 | * of this processor. This typically the case if we have just allocated | |
2762 | * the item before. | |
2763 | * | |
2764 | * If fastpath is not possible then fall back to __slab_free where we deal | |
2765 | * with all sorts of special processing. | |
81084651 JDB |
2766 | * |
2767 | * Bulk free of a freelist with several objects (all pointing to the | |
2768 | * same page) possible by specifying head and tail ptr, plus objects | |
2769 | * count (cnt). Bulk free indicated by tail pointer being set. | |
894b8788 | 2770 | */ |
81084651 JDB |
2771 | static __always_inline void slab_free(struct kmem_cache *s, struct page *page, |
2772 | void *head, void *tail, int cnt, | |
2773 | unsigned long addr) | |
894b8788 | 2774 | { |
81084651 | 2775 | void *tail_obj = tail ? : head; |
dfb4f096 | 2776 | struct kmem_cache_cpu *c; |
8a5ec0ba | 2777 | unsigned long tid; |
1f84260c | 2778 | |
81084651 | 2779 | slab_free_freelist_hook(s, head, tail); |
c016b0bd | 2780 | |
8a5ec0ba CL |
2781 | redo: |
2782 | /* | |
2783 | * Determine the currently cpus per cpu slab. | |
2784 | * The cpu may change afterward. However that does not matter since | |
2785 | * data is retrieved via this pointer. If we are on the same cpu | |
2ae44005 | 2786 | * during the cmpxchg then the free will succeed. |
8a5ec0ba | 2787 | */ |
9aabf810 JK |
2788 | do { |
2789 | tid = this_cpu_read(s->cpu_slab->tid); | |
2790 | c = raw_cpu_ptr(s->cpu_slab); | |
859b7a0e MR |
2791 | } while (IS_ENABLED(CONFIG_PREEMPT) && |
2792 | unlikely(tid != READ_ONCE(c->tid))); | |
c016b0bd | 2793 | |
9aabf810 JK |
2794 | /* Same with comment on barrier() in slab_alloc_node() */ |
2795 | barrier(); | |
c016b0bd | 2796 | |
442b06bc | 2797 | if (likely(page == c->page)) { |
81084651 | 2798 | set_freepointer(s, tail_obj, c->freelist); |
8a5ec0ba | 2799 | |
933393f5 | 2800 | if (unlikely(!this_cpu_cmpxchg_double( |
8a5ec0ba CL |
2801 | s->cpu_slab->freelist, s->cpu_slab->tid, |
2802 | c->freelist, tid, | |
81084651 | 2803 | head, next_tid(tid)))) { |
8a5ec0ba CL |
2804 | |
2805 | note_cmpxchg_failure("slab_free", s, tid); | |
2806 | goto redo; | |
2807 | } | |
84e554e6 | 2808 | stat(s, FREE_FASTPATH); |
894b8788 | 2809 | } else |
81084651 | 2810 | __slab_free(s, page, head, tail_obj, cnt, addr); |
894b8788 | 2811 | |
894b8788 CL |
2812 | } |
2813 | ||
81819f0f CL |
2814 | void kmem_cache_free(struct kmem_cache *s, void *x) |
2815 | { | |
b9ce5ef4 GC |
2816 | s = cache_from_obj(s, x); |
2817 | if (!s) | |
79576102 | 2818 | return; |
81084651 | 2819 | slab_free(s, virt_to_head_page(x), x, NULL, 1, _RET_IP_); |
ca2b84cb | 2820 | trace_kmem_cache_free(_RET_IP_, x); |
81819f0f CL |
2821 | } |
2822 | EXPORT_SYMBOL(kmem_cache_free); | |
2823 | ||
d0ecd894 | 2824 | struct detached_freelist { |
fbd02630 | 2825 | struct page *page; |
d0ecd894 JDB |
2826 | void *tail; |
2827 | void *freelist; | |
2828 | int cnt; | |
376bf125 | 2829 | struct kmem_cache *s; |
d0ecd894 | 2830 | }; |
fbd02630 | 2831 | |
d0ecd894 JDB |
2832 | /* |
2833 | * This function progressively scans the array with free objects (with | |
2834 | * a limited look ahead) and extract objects belonging to the same | |
2835 | * page. It builds a detached freelist directly within the given | |
2836 | * page/objects. This can happen without any need for | |
2837 | * synchronization, because the objects are owned by running process. | |
2838 | * The freelist is build up as a single linked list in the objects. | |
2839 | * The idea is, that this detached freelist can then be bulk | |
2840 | * transferred to the real freelist(s), but only requiring a single | |
2841 | * synchronization primitive. Look ahead in the array is limited due | |
2842 | * to performance reasons. | |
2843 | */ | |
376bf125 JDB |
2844 | static inline |
2845 | int build_detached_freelist(struct kmem_cache *s, size_t size, | |
2846 | void **p, struct detached_freelist *df) | |
d0ecd894 JDB |
2847 | { |
2848 | size_t first_skipped_index = 0; | |
2849 | int lookahead = 3; | |
2850 | void *object; | |
ca257195 | 2851 | struct page *page; |
fbd02630 | 2852 | |
d0ecd894 JDB |
2853 | /* Always re-init detached_freelist */ |
2854 | df->page = NULL; | |
fbd02630 | 2855 | |
d0ecd894 JDB |
2856 | do { |
2857 | object = p[--size]; | |
ca257195 | 2858 | /* Do we need !ZERO_OR_NULL_PTR(object) here? (for kfree) */ |
d0ecd894 | 2859 | } while (!object && size); |
3eed034d | 2860 | |
d0ecd894 JDB |
2861 | if (!object) |
2862 | return 0; | |
fbd02630 | 2863 | |
ca257195 JDB |
2864 | page = virt_to_head_page(object); |
2865 | if (!s) { | |
2866 | /* Handle kalloc'ed objects */ | |
2867 | if (unlikely(!PageSlab(page))) { | |
2868 | BUG_ON(!PageCompound(page)); | |
2869 | kfree_hook(object); | |
2870 | __free_kmem_pages(page, compound_order(page)); | |
2871 | p[size] = NULL; /* mark object processed */ | |
2872 | return size; | |
2873 | } | |
2874 | /* Derive kmem_cache from object */ | |
2875 | df->s = page->slab_cache; | |
2876 | } else { | |
2877 | df->s = cache_from_obj(s, object); /* Support for memcg */ | |
2878 | } | |
376bf125 | 2879 | |
d0ecd894 | 2880 | /* Start new detached freelist */ |
ca257195 | 2881 | df->page = page; |
376bf125 | 2882 | set_freepointer(df->s, object, NULL); |
d0ecd894 JDB |
2883 | df->tail = object; |
2884 | df->freelist = object; | |
2885 | p[size] = NULL; /* mark object processed */ | |
2886 | df->cnt = 1; | |
2887 | ||
2888 | while (size) { | |
2889 | object = p[--size]; | |
2890 | if (!object) | |
2891 | continue; /* Skip processed objects */ | |
2892 | ||
2893 | /* df->page is always set at this point */ | |
2894 | if (df->page == virt_to_head_page(object)) { | |
2895 | /* Opportunity build freelist */ | |
376bf125 | 2896 | set_freepointer(df->s, object, df->freelist); |
d0ecd894 JDB |
2897 | df->freelist = object; |
2898 | df->cnt++; | |
2899 | p[size] = NULL; /* mark object processed */ | |
2900 | ||
2901 | continue; | |
fbd02630 | 2902 | } |
d0ecd894 JDB |
2903 | |
2904 | /* Limit look ahead search */ | |
2905 | if (!--lookahead) | |
2906 | break; | |
2907 | ||
2908 | if (!first_skipped_index) | |
2909 | first_skipped_index = size + 1; | |
fbd02630 | 2910 | } |
d0ecd894 JDB |
2911 | |
2912 | return first_skipped_index; | |
2913 | } | |
2914 | ||
d0ecd894 | 2915 | /* Note that interrupts must be enabled when calling this function. */ |
376bf125 | 2916 | void kmem_cache_free_bulk(struct kmem_cache *s, size_t size, void **p) |
d0ecd894 JDB |
2917 | { |
2918 | if (WARN_ON(!size)) | |
2919 | return; | |
2920 | ||
2921 | do { | |
2922 | struct detached_freelist df; | |
2923 | ||
2924 | size = build_detached_freelist(s, size, p, &df); | |
2925 | if (unlikely(!df.page)) | |
2926 | continue; | |
2927 | ||
376bf125 | 2928 | slab_free(df.s, df.page, df.freelist, df.tail, df.cnt,_RET_IP_); |
d0ecd894 | 2929 | } while (likely(size)); |
484748f0 CL |
2930 | } |
2931 | EXPORT_SYMBOL(kmem_cache_free_bulk); | |
2932 | ||
994eb764 | 2933 | /* Note that interrupts must be enabled when calling this function. */ |
865762a8 JDB |
2934 | int kmem_cache_alloc_bulk(struct kmem_cache *s, gfp_t flags, size_t size, |
2935 | void **p) | |
484748f0 | 2936 | { |
994eb764 JDB |
2937 | struct kmem_cache_cpu *c; |
2938 | int i; | |
2939 | ||
03ec0ed5 JDB |
2940 | /* memcg and kmem_cache debug support */ |
2941 | s = slab_pre_alloc_hook(s, flags); | |
2942 | if (unlikely(!s)) | |
2943 | return false; | |
994eb764 JDB |
2944 | /* |
2945 | * Drain objects in the per cpu slab, while disabling local | |
2946 | * IRQs, which protects against PREEMPT and interrupts | |
2947 | * handlers invoking normal fastpath. | |
2948 | */ | |
2949 | local_irq_disable(); | |
2950 | c = this_cpu_ptr(s->cpu_slab); | |
2951 | ||
2952 | for (i = 0; i < size; i++) { | |
2953 | void *object = c->freelist; | |
2954 | ||
ebe909e0 | 2955 | if (unlikely(!object)) { |
ebe909e0 JDB |
2956 | /* |
2957 | * Invoking slow path likely have side-effect | |
2958 | * of re-populating per CPU c->freelist | |
2959 | */ | |
87098373 | 2960 | p[i] = ___slab_alloc(s, flags, NUMA_NO_NODE, |
ebe909e0 | 2961 | _RET_IP_, c); |
87098373 CL |
2962 | if (unlikely(!p[i])) |
2963 | goto error; | |
2964 | ||
ebe909e0 JDB |
2965 | c = this_cpu_ptr(s->cpu_slab); |
2966 | continue; /* goto for-loop */ | |
2967 | } | |
994eb764 JDB |
2968 | c->freelist = get_freepointer(s, object); |
2969 | p[i] = object; | |
2970 | } | |
2971 | c->tid = next_tid(c->tid); | |
2972 | local_irq_enable(); | |
2973 | ||
2974 | /* Clear memory outside IRQ disabled fastpath loop */ | |
2975 | if (unlikely(flags & __GFP_ZERO)) { | |
2976 | int j; | |
2977 | ||
2978 | for (j = 0; j < i; j++) | |
2979 | memset(p[j], 0, s->object_size); | |
2980 | } | |
2981 | ||
03ec0ed5 JDB |
2982 | /* memcg and kmem_cache debug support */ |
2983 | slab_post_alloc_hook(s, flags, size, p); | |
865762a8 | 2984 | return i; |
87098373 | 2985 | error: |
87098373 | 2986 | local_irq_enable(); |
03ec0ed5 JDB |
2987 | slab_post_alloc_hook(s, flags, i, p); |
2988 | __kmem_cache_free_bulk(s, i, p); | |
865762a8 | 2989 | return 0; |
484748f0 CL |
2990 | } |
2991 | EXPORT_SYMBOL(kmem_cache_alloc_bulk); | |
2992 | ||
2993 | ||
81819f0f | 2994 | /* |
672bba3a CL |
2995 | * Object placement in a slab is made very easy because we always start at |
2996 | * offset 0. If we tune the size of the object to the alignment then we can | |
2997 | * get the required alignment by putting one properly sized object after | |
2998 | * another. | |
81819f0f CL |
2999 | * |
3000 | * Notice that the allocation order determines the sizes of the per cpu | |
3001 | * caches. Each processor has always one slab available for allocations. | |
3002 | * Increasing the allocation order reduces the number of times that slabs | |
672bba3a | 3003 | * must be moved on and off the partial lists and is therefore a factor in |
81819f0f | 3004 | * locking overhead. |
81819f0f CL |
3005 | */ |
3006 | ||
3007 | /* | |
3008 | * Mininum / Maximum order of slab pages. This influences locking overhead | |
3009 | * and slab fragmentation. A higher order reduces the number of partial slabs | |
3010 | * and increases the number of allocations possible without having to | |
3011 | * take the list_lock. | |
3012 | */ | |
3013 | static int slub_min_order; | |
114e9e89 | 3014 | static int slub_max_order = PAGE_ALLOC_COSTLY_ORDER; |
9b2cd506 | 3015 | static int slub_min_objects; |
81819f0f | 3016 | |
81819f0f CL |
3017 | /* |
3018 | * Calculate the order of allocation given an slab object size. | |
3019 | * | |
672bba3a CL |
3020 | * The order of allocation has significant impact on performance and other |
3021 | * system components. Generally order 0 allocations should be preferred since | |
3022 | * order 0 does not cause fragmentation in the page allocator. Larger objects | |
3023 | * be problematic to put into order 0 slabs because there may be too much | |
c124f5b5 | 3024 | * unused space left. We go to a higher order if more than 1/16th of the slab |
672bba3a CL |
3025 | * would be wasted. |
3026 | * | |
3027 | * In order to reach satisfactory performance we must ensure that a minimum | |
3028 | * number of objects is in one slab. Otherwise we may generate too much | |
3029 | * activity on the partial lists which requires taking the list_lock. This is | |
3030 | * less a concern for large slabs though which are rarely used. | |
81819f0f | 3031 | * |
672bba3a CL |
3032 | * slub_max_order specifies the order where we begin to stop considering the |
3033 | * number of objects in a slab as critical. If we reach slub_max_order then | |
3034 | * we try to keep the page order as low as possible. So we accept more waste | |
3035 | * of space in favor of a small page order. | |
81819f0f | 3036 | * |
672bba3a CL |
3037 | * Higher order allocations also allow the placement of more objects in a |
3038 | * slab and thereby reduce object handling overhead. If the user has | |
3039 | * requested a higher mininum order then we start with that one instead of | |
3040 | * the smallest order which will fit the object. | |
81819f0f | 3041 | */ |
5e6d444e | 3042 | static inline int slab_order(int size, int min_objects, |
ab9a0f19 | 3043 | int max_order, int fract_leftover, int reserved) |
81819f0f CL |
3044 | { |
3045 | int order; | |
3046 | int rem; | |
6300ea75 | 3047 | int min_order = slub_min_order; |
81819f0f | 3048 | |
ab9a0f19 | 3049 | if (order_objects(min_order, size, reserved) > MAX_OBJS_PER_PAGE) |
210b5c06 | 3050 | return get_order(size * MAX_OBJS_PER_PAGE) - 1; |
39b26464 | 3051 | |
9f835703 | 3052 | for (order = max(min_order, get_order(min_objects * size + reserved)); |
5e6d444e | 3053 | order <= max_order; order++) { |
81819f0f | 3054 | |
5e6d444e | 3055 | unsigned long slab_size = PAGE_SIZE << order; |
81819f0f | 3056 | |
ab9a0f19 | 3057 | rem = (slab_size - reserved) % size; |
81819f0f | 3058 | |
5e6d444e | 3059 | if (rem <= slab_size / fract_leftover) |
81819f0f | 3060 | break; |
81819f0f | 3061 | } |
672bba3a | 3062 | |
81819f0f CL |
3063 | return order; |
3064 | } | |
3065 | ||
ab9a0f19 | 3066 | static inline int calculate_order(int size, int reserved) |
5e6d444e CL |
3067 | { |
3068 | int order; | |
3069 | int min_objects; | |
3070 | int fraction; | |
e8120ff1 | 3071 | int max_objects; |
5e6d444e CL |
3072 | |
3073 | /* | |
3074 | * Attempt to find best configuration for a slab. This | |
3075 | * works by first attempting to generate a layout with | |
3076 | * the best configuration and backing off gradually. | |
3077 | * | |
422ff4d7 | 3078 | * First we increase the acceptable waste in a slab. Then |
5e6d444e CL |
3079 | * we reduce the minimum objects required in a slab. |
3080 | */ | |
3081 | min_objects = slub_min_objects; | |
9b2cd506 CL |
3082 | if (!min_objects) |
3083 | min_objects = 4 * (fls(nr_cpu_ids) + 1); | |
ab9a0f19 | 3084 | max_objects = order_objects(slub_max_order, size, reserved); |
e8120ff1 ZY |
3085 | min_objects = min(min_objects, max_objects); |
3086 | ||
5e6d444e | 3087 | while (min_objects > 1) { |
c124f5b5 | 3088 | fraction = 16; |
5e6d444e CL |
3089 | while (fraction >= 4) { |
3090 | order = slab_order(size, min_objects, | |
ab9a0f19 | 3091 | slub_max_order, fraction, reserved); |
5e6d444e CL |
3092 | if (order <= slub_max_order) |
3093 | return order; | |
3094 | fraction /= 2; | |
3095 | } | |
5086c389 | 3096 | min_objects--; |
5e6d444e CL |
3097 | } |
3098 | ||
3099 | /* | |
3100 | * We were unable to place multiple objects in a slab. Now | |
3101 | * lets see if we can place a single object there. | |
3102 | */ | |
ab9a0f19 | 3103 | order = slab_order(size, 1, slub_max_order, 1, reserved); |
5e6d444e CL |
3104 | if (order <= slub_max_order) |
3105 | return order; | |
3106 | ||
3107 | /* | |
3108 | * Doh this slab cannot be placed using slub_max_order. | |
3109 | */ | |
ab9a0f19 | 3110 | order = slab_order(size, 1, MAX_ORDER, 1, reserved); |
818cf590 | 3111 | if (order < MAX_ORDER) |
5e6d444e CL |
3112 | return order; |
3113 | return -ENOSYS; | |
3114 | } | |
3115 | ||
5595cffc | 3116 | static void |
4053497d | 3117 | init_kmem_cache_node(struct kmem_cache_node *n) |
81819f0f CL |
3118 | { |
3119 | n->nr_partial = 0; | |
81819f0f CL |
3120 | spin_lock_init(&n->list_lock); |
3121 | INIT_LIST_HEAD(&n->partial); | |
8ab1372f | 3122 | #ifdef CONFIG_SLUB_DEBUG |
0f389ec6 | 3123 | atomic_long_set(&n->nr_slabs, 0); |
02b71b70 | 3124 | atomic_long_set(&n->total_objects, 0); |
643b1138 | 3125 | INIT_LIST_HEAD(&n->full); |
8ab1372f | 3126 | #endif |
81819f0f CL |
3127 | } |
3128 | ||
55136592 | 3129 | static inline int alloc_kmem_cache_cpus(struct kmem_cache *s) |
4c93c355 | 3130 | { |
6c182dc0 | 3131 | BUILD_BUG_ON(PERCPU_DYNAMIC_EARLY_SIZE < |
95a05b42 | 3132 | KMALLOC_SHIFT_HIGH * sizeof(struct kmem_cache_cpu)); |
4c93c355 | 3133 | |
8a5ec0ba | 3134 | /* |
d4d84fef CM |
3135 | * Must align to double word boundary for the double cmpxchg |
3136 | * instructions to work; see __pcpu_double_call_return_bool(). | |
8a5ec0ba | 3137 | */ |
d4d84fef CM |
3138 | s->cpu_slab = __alloc_percpu(sizeof(struct kmem_cache_cpu), |
3139 | 2 * sizeof(void *)); | |
8a5ec0ba CL |
3140 | |
3141 | if (!s->cpu_slab) | |
3142 | return 0; | |
3143 | ||
3144 | init_kmem_cache_cpus(s); | |
4c93c355 | 3145 | |
8a5ec0ba | 3146 | return 1; |
4c93c355 | 3147 | } |
4c93c355 | 3148 | |
51df1142 CL |
3149 | static struct kmem_cache *kmem_cache_node; |
3150 | ||
81819f0f CL |
3151 | /* |
3152 | * No kmalloc_node yet so do it by hand. We know that this is the first | |
3153 | * slab on the node for this slabcache. There are no concurrent accesses | |
3154 | * possible. | |
3155 | * | |
721ae22a ZYW |
3156 | * Note that this function only works on the kmem_cache_node |
3157 | * when allocating for the kmem_cache_node. This is used for bootstrapping | |
4c93c355 | 3158 | * memory on a fresh node that has no slab structures yet. |
81819f0f | 3159 | */ |
55136592 | 3160 | static void early_kmem_cache_node_alloc(int node) |
81819f0f CL |
3161 | { |
3162 | struct page *page; | |
3163 | struct kmem_cache_node *n; | |
3164 | ||
51df1142 | 3165 | BUG_ON(kmem_cache_node->size < sizeof(struct kmem_cache_node)); |
81819f0f | 3166 | |
51df1142 | 3167 | page = new_slab(kmem_cache_node, GFP_NOWAIT, node); |
81819f0f CL |
3168 | |
3169 | BUG_ON(!page); | |
a2f92ee7 | 3170 | if (page_to_nid(page) != node) { |
f9f58285 FF |
3171 | pr_err("SLUB: Unable to allocate memory from node %d\n", node); |
3172 | pr_err("SLUB: Allocating a useless per node structure in order to be able to continue\n"); | |
a2f92ee7 CL |
3173 | } |
3174 | ||
81819f0f CL |
3175 | n = page->freelist; |
3176 | BUG_ON(!n); | |
51df1142 | 3177 | page->freelist = get_freepointer(kmem_cache_node, n); |
e6e82ea1 | 3178 | page->inuse = 1; |
8cb0a506 | 3179 | page->frozen = 0; |
51df1142 | 3180 | kmem_cache_node->node[node] = n; |
8ab1372f | 3181 | #ifdef CONFIG_SLUB_DEBUG |
f7cb1933 | 3182 | init_object(kmem_cache_node, n, SLUB_RED_ACTIVE); |
51df1142 | 3183 | init_tracking(kmem_cache_node, n); |
8ab1372f | 3184 | #endif |
505f5dcb AP |
3185 | kasan_kmalloc(kmem_cache_node, n, sizeof(struct kmem_cache_node), |
3186 | GFP_KERNEL); | |
4053497d | 3187 | init_kmem_cache_node(n); |
51df1142 | 3188 | inc_slabs_node(kmem_cache_node, node, page->objects); |
6446faa2 | 3189 | |
67b6c900 | 3190 | /* |
1e4dd946 SR |
3191 | * No locks need to be taken here as it has just been |
3192 | * initialized and there is no concurrent access. | |
67b6c900 | 3193 | */ |
1e4dd946 | 3194 | __add_partial(n, page, DEACTIVATE_TO_HEAD); |
81819f0f CL |
3195 | } |
3196 | ||
3197 | static void free_kmem_cache_nodes(struct kmem_cache *s) | |
3198 | { | |
3199 | int node; | |
fa45dc25 | 3200 | struct kmem_cache_node *n; |
81819f0f | 3201 | |
fa45dc25 CL |
3202 | for_each_kmem_cache_node(s, node, n) { |
3203 | kmem_cache_free(kmem_cache_node, n); | |
81819f0f CL |
3204 | s->node[node] = NULL; |
3205 | } | |
3206 | } | |
3207 | ||
52b4b950 DS |
3208 | void __kmem_cache_release(struct kmem_cache *s) |
3209 | { | |
3210 | free_percpu(s->cpu_slab); | |
3211 | free_kmem_cache_nodes(s); | |
3212 | } | |
3213 | ||
55136592 | 3214 | static int init_kmem_cache_nodes(struct kmem_cache *s) |
81819f0f CL |
3215 | { |
3216 | int node; | |
81819f0f | 3217 | |
f64dc58c | 3218 | for_each_node_state(node, N_NORMAL_MEMORY) { |
81819f0f CL |
3219 | struct kmem_cache_node *n; |
3220 | ||
73367bd8 | 3221 | if (slab_state == DOWN) { |
55136592 | 3222 | early_kmem_cache_node_alloc(node); |
73367bd8 AD |
3223 | continue; |
3224 | } | |
51df1142 | 3225 | n = kmem_cache_alloc_node(kmem_cache_node, |
55136592 | 3226 | GFP_KERNEL, node); |
81819f0f | 3227 | |
73367bd8 AD |
3228 | if (!n) { |
3229 | free_kmem_cache_nodes(s); | |
3230 | return 0; | |
81819f0f | 3231 | } |
73367bd8 | 3232 | |
81819f0f | 3233 | s->node[node] = n; |
4053497d | 3234 | init_kmem_cache_node(n); |
81819f0f CL |
3235 | } |
3236 | return 1; | |
3237 | } | |
81819f0f | 3238 | |
c0bdb232 | 3239 | static void set_min_partial(struct kmem_cache *s, unsigned long min) |
3b89d7d8 DR |
3240 | { |
3241 | if (min < MIN_PARTIAL) | |
3242 | min = MIN_PARTIAL; | |
3243 | else if (min > MAX_PARTIAL) | |
3244 | min = MAX_PARTIAL; | |
3245 | s->min_partial = min; | |
3246 | } | |
3247 | ||
81819f0f CL |
3248 | /* |
3249 | * calculate_sizes() determines the order and the distribution of data within | |
3250 | * a slab object. | |
3251 | */ | |
06b285dc | 3252 | static int calculate_sizes(struct kmem_cache *s, int forced_order) |
81819f0f CL |
3253 | { |
3254 | unsigned long flags = s->flags; | |
3b0efdfa | 3255 | unsigned long size = s->object_size; |
834f3d11 | 3256 | int order; |
81819f0f | 3257 | |
d8b42bf5 CL |
3258 | /* |
3259 | * Round up object size to the next word boundary. We can only | |
3260 | * place the free pointer at word boundaries and this determines | |
3261 | * the possible location of the free pointer. | |
3262 | */ | |
3263 | size = ALIGN(size, sizeof(void *)); | |
3264 | ||
3265 | #ifdef CONFIG_SLUB_DEBUG | |
81819f0f CL |
3266 | /* |
3267 | * Determine if we can poison the object itself. If the user of | |
3268 | * the slab may touch the object after free or before allocation | |
3269 | * then we should never poison the object itself. | |
3270 | */ | |
3271 | if ((flags & SLAB_POISON) && !(flags & SLAB_DESTROY_BY_RCU) && | |
c59def9f | 3272 | !s->ctor) |
81819f0f CL |
3273 | s->flags |= __OBJECT_POISON; |
3274 | else | |
3275 | s->flags &= ~__OBJECT_POISON; | |
3276 | ||
81819f0f CL |
3277 | |
3278 | /* | |
672bba3a | 3279 | * If we are Redzoning then check if there is some space between the |
81819f0f | 3280 | * end of the object and the free pointer. If not then add an |
672bba3a | 3281 | * additional word to have some bytes to store Redzone information. |
81819f0f | 3282 | */ |
3b0efdfa | 3283 | if ((flags & SLAB_RED_ZONE) && size == s->object_size) |
81819f0f | 3284 | size += sizeof(void *); |
41ecc55b | 3285 | #endif |
81819f0f CL |
3286 | |
3287 | /* | |
672bba3a CL |
3288 | * With that we have determined the number of bytes in actual use |
3289 | * by the object. This is the potential offset to the free pointer. | |
81819f0f CL |
3290 | */ |
3291 | s->inuse = size; | |
3292 | ||
3293 | if (((flags & (SLAB_DESTROY_BY_RCU | SLAB_POISON)) || | |
c59def9f | 3294 | s->ctor)) { |
81819f0f CL |
3295 | /* |
3296 | * Relocate free pointer after the object if it is not | |
3297 | * permitted to overwrite the first word of the object on | |
3298 | * kmem_cache_free. | |
3299 | * | |
3300 | * This is the case if we do RCU, have a constructor or | |
3301 | * destructor or are poisoning the objects. | |
3302 | */ | |
3303 | s->offset = size; | |
3304 | size += sizeof(void *); | |
3305 | } | |
3306 | ||
c12b3c62 | 3307 | #ifdef CONFIG_SLUB_DEBUG |
81819f0f CL |
3308 | if (flags & SLAB_STORE_USER) |
3309 | /* | |
3310 | * Need to store information about allocs and frees after | |
3311 | * the object. | |
3312 | */ | |
3313 | size += 2 * sizeof(struct track); | |
3314 | ||
d86bd1be | 3315 | if (flags & SLAB_RED_ZONE) { |
81819f0f CL |
3316 | /* |
3317 | * Add some empty padding so that we can catch | |
3318 | * overwrites from earlier objects rather than let | |
3319 | * tracking information or the free pointer be | |
0211a9c8 | 3320 | * corrupted if a user writes before the start |
81819f0f CL |
3321 | * of the object. |
3322 | */ | |
3323 | size += sizeof(void *); | |
d86bd1be JK |
3324 | |
3325 | s->red_left_pad = sizeof(void *); | |
3326 | s->red_left_pad = ALIGN(s->red_left_pad, s->align); | |
3327 | size += s->red_left_pad; | |
3328 | } | |
41ecc55b | 3329 | #endif |
672bba3a | 3330 | |
81819f0f CL |
3331 | /* |
3332 | * SLUB stores one object immediately after another beginning from | |
3333 | * offset 0. In order to align the objects we have to simply size | |
3334 | * each object to conform to the alignment. | |
3335 | */ | |
45906855 | 3336 | size = ALIGN(size, s->align); |
81819f0f | 3337 | s->size = size; |
06b285dc CL |
3338 | if (forced_order >= 0) |
3339 | order = forced_order; | |
3340 | else | |
ab9a0f19 | 3341 | order = calculate_order(size, s->reserved); |
81819f0f | 3342 | |
834f3d11 | 3343 | if (order < 0) |
81819f0f CL |
3344 | return 0; |
3345 | ||
b7a49f0d | 3346 | s->allocflags = 0; |
834f3d11 | 3347 | if (order) |
b7a49f0d CL |
3348 | s->allocflags |= __GFP_COMP; |
3349 | ||
3350 | if (s->flags & SLAB_CACHE_DMA) | |
2c59dd65 | 3351 | s->allocflags |= GFP_DMA; |
b7a49f0d CL |
3352 | |
3353 | if (s->flags & SLAB_RECLAIM_ACCOUNT) | |
3354 | s->allocflags |= __GFP_RECLAIMABLE; | |
3355 | ||
81819f0f CL |
3356 | /* |
3357 | * Determine the number of objects per slab | |
3358 | */ | |
ab9a0f19 LJ |
3359 | s->oo = oo_make(order, size, s->reserved); |
3360 | s->min = oo_make(get_order(size), size, s->reserved); | |
205ab99d CL |
3361 | if (oo_objects(s->oo) > oo_objects(s->max)) |
3362 | s->max = s->oo; | |
81819f0f | 3363 | |
834f3d11 | 3364 | return !!oo_objects(s->oo); |
81819f0f CL |
3365 | } |
3366 | ||
8a13a4cc | 3367 | static int kmem_cache_open(struct kmem_cache *s, unsigned long flags) |
81819f0f | 3368 | { |
8a13a4cc | 3369 | s->flags = kmem_cache_flags(s->size, flags, s->name, s->ctor); |
ab9a0f19 | 3370 | s->reserved = 0; |
81819f0f | 3371 | |
da9a638c LJ |
3372 | if (need_reserve_slab_rcu && (s->flags & SLAB_DESTROY_BY_RCU)) |
3373 | s->reserved = sizeof(struct rcu_head); | |
81819f0f | 3374 | |
06b285dc | 3375 | if (!calculate_sizes(s, -1)) |
81819f0f | 3376 | goto error; |
3de47213 DR |
3377 | if (disable_higher_order_debug) { |
3378 | /* | |
3379 | * Disable debugging flags that store metadata if the min slab | |
3380 | * order increased. | |
3381 | */ | |
3b0efdfa | 3382 | if (get_order(s->size) > get_order(s->object_size)) { |
3de47213 DR |
3383 | s->flags &= ~DEBUG_METADATA_FLAGS; |
3384 | s->offset = 0; | |
3385 | if (!calculate_sizes(s, -1)) | |
3386 | goto error; | |
3387 | } | |
3388 | } | |
81819f0f | 3389 | |
2565409f HC |
3390 | #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \ |
3391 | defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE) | |
149daaf3 | 3392 | if (system_has_cmpxchg_double() && (s->flags & SLAB_NO_CMPXCHG) == 0) |
b789ef51 CL |
3393 | /* Enable fast mode */ |
3394 | s->flags |= __CMPXCHG_DOUBLE; | |
3395 | #endif | |
3396 | ||
3b89d7d8 DR |
3397 | /* |
3398 | * The larger the object size is, the more pages we want on the partial | |
3399 | * list to avoid pounding the page allocator excessively. | |
3400 | */ | |
49e22585 CL |
3401 | set_min_partial(s, ilog2(s->size) / 2); |
3402 | ||
3403 | /* | |
3404 | * cpu_partial determined the maximum number of objects kept in the | |
3405 | * per cpu partial lists of a processor. | |
3406 | * | |
3407 | * Per cpu partial lists mainly contain slabs that just have one | |
3408 | * object freed. If they are used for allocation then they can be | |
3409 | * filled up again with minimal effort. The slab will never hit the | |
3410 | * per node partial lists and therefore no locking will be required. | |
3411 | * | |
3412 | * This setting also determines | |
3413 | * | |
3414 | * A) The number of objects from per cpu partial slabs dumped to the | |
3415 | * per node list when we reach the limit. | |
9f264904 | 3416 | * B) The number of objects in cpu partial slabs to extract from the |
d0e0ac97 CG |
3417 | * per node list when we run out of per cpu objects. We only fetch |
3418 | * 50% to keep some capacity around for frees. | |
49e22585 | 3419 | */ |
345c905d | 3420 | if (!kmem_cache_has_cpu_partial(s)) |
8f1e33da CL |
3421 | s->cpu_partial = 0; |
3422 | else if (s->size >= PAGE_SIZE) | |
49e22585 CL |
3423 | s->cpu_partial = 2; |
3424 | else if (s->size >= 1024) | |
3425 | s->cpu_partial = 6; | |
3426 | else if (s->size >= 256) | |
3427 | s->cpu_partial = 13; | |
3428 | else | |
3429 | s->cpu_partial = 30; | |
3430 | ||
81819f0f | 3431 | #ifdef CONFIG_NUMA |
e2cb96b7 | 3432 | s->remote_node_defrag_ratio = 1000; |
81819f0f | 3433 | #endif |
55136592 | 3434 | if (!init_kmem_cache_nodes(s)) |
dfb4f096 | 3435 | goto error; |
81819f0f | 3436 | |
55136592 | 3437 | if (alloc_kmem_cache_cpus(s)) |
278b1bb1 | 3438 | return 0; |
ff12059e | 3439 | |
4c93c355 | 3440 | free_kmem_cache_nodes(s); |
81819f0f CL |
3441 | error: |
3442 | if (flags & SLAB_PANIC) | |
756a025f JP |
3443 | panic("Cannot create slab %s size=%lu realsize=%u order=%u offset=%u flags=%lx\n", |
3444 | s->name, (unsigned long)s->size, s->size, | |
3445 | oo_order(s->oo), s->offset, flags); | |
278b1bb1 | 3446 | return -EINVAL; |
81819f0f | 3447 | } |
81819f0f | 3448 | |
33b12c38 CL |
3449 | static void list_slab_objects(struct kmem_cache *s, struct page *page, |
3450 | const char *text) | |
3451 | { | |
3452 | #ifdef CONFIG_SLUB_DEBUG | |
3453 | void *addr = page_address(page); | |
3454 | void *p; | |
a5dd5c11 NK |
3455 | unsigned long *map = kzalloc(BITS_TO_LONGS(page->objects) * |
3456 | sizeof(long), GFP_ATOMIC); | |
bbd7d57b ED |
3457 | if (!map) |
3458 | return; | |
945cf2b6 | 3459 | slab_err(s, page, text, s->name); |
33b12c38 | 3460 | slab_lock(page); |
33b12c38 | 3461 | |
5f80b13a | 3462 | get_map(s, page, map); |
33b12c38 CL |
3463 | for_each_object(p, s, addr, page->objects) { |
3464 | ||
3465 | if (!test_bit(slab_index(p, s, addr), map)) { | |
f9f58285 | 3466 | pr_err("INFO: Object 0x%p @offset=%tu\n", p, p - addr); |
33b12c38 CL |
3467 | print_tracking(s, p); |
3468 | } | |
3469 | } | |
3470 | slab_unlock(page); | |
bbd7d57b | 3471 | kfree(map); |
33b12c38 CL |
3472 | #endif |
3473 | } | |
3474 | ||
81819f0f | 3475 | /* |
599870b1 | 3476 | * Attempt to free all partial slabs on a node. |
52b4b950 DS |
3477 | * This is called from __kmem_cache_shutdown(). We must take list_lock |
3478 | * because sysfs file might still access partial list after the shutdowning. | |
81819f0f | 3479 | */ |
599870b1 | 3480 | static void free_partial(struct kmem_cache *s, struct kmem_cache_node *n) |
81819f0f | 3481 | { |
81819f0f CL |
3482 | struct page *page, *h; |
3483 | ||
52b4b950 DS |
3484 | BUG_ON(irqs_disabled()); |
3485 | spin_lock_irq(&n->list_lock); | |
33b12c38 | 3486 | list_for_each_entry_safe(page, h, &n->partial, lru) { |
81819f0f | 3487 | if (!page->inuse) { |
52b4b950 | 3488 | remove_partial(n, page); |
81819f0f | 3489 | discard_slab(s, page); |
33b12c38 CL |
3490 | } else { |
3491 | list_slab_objects(s, page, | |
52b4b950 | 3492 | "Objects remaining in %s on __kmem_cache_shutdown()"); |
599870b1 | 3493 | } |
33b12c38 | 3494 | } |
52b4b950 | 3495 | spin_unlock_irq(&n->list_lock); |
81819f0f CL |
3496 | } |
3497 | ||
3498 | /* | |
672bba3a | 3499 | * Release all resources used by a slab cache. |
81819f0f | 3500 | */ |
52b4b950 | 3501 | int __kmem_cache_shutdown(struct kmem_cache *s) |
81819f0f CL |
3502 | { |
3503 | int node; | |
fa45dc25 | 3504 | struct kmem_cache_node *n; |
81819f0f CL |
3505 | |
3506 | flush_all(s); | |
81819f0f | 3507 | /* Attempt to free all objects */ |
fa45dc25 | 3508 | for_each_kmem_cache_node(s, node, n) { |
599870b1 CL |
3509 | free_partial(s, n); |
3510 | if (n->nr_partial || slabs_node(s, node)) | |
81819f0f CL |
3511 | return 1; |
3512 | } | |
81819f0f CL |
3513 | return 0; |
3514 | } | |
3515 | ||
81819f0f CL |
3516 | /******************************************************************** |
3517 | * Kmalloc subsystem | |
3518 | *******************************************************************/ | |
3519 | ||
81819f0f CL |
3520 | static int __init setup_slub_min_order(char *str) |
3521 | { | |
06428780 | 3522 | get_option(&str, &slub_min_order); |
81819f0f CL |
3523 | |
3524 | return 1; | |
3525 | } | |
3526 | ||
3527 | __setup("slub_min_order=", setup_slub_min_order); | |
3528 | ||
3529 | static int __init setup_slub_max_order(char *str) | |
3530 | { | |
06428780 | 3531 | get_option(&str, &slub_max_order); |
818cf590 | 3532 | slub_max_order = min(slub_max_order, MAX_ORDER - 1); |
81819f0f CL |
3533 | |
3534 | return 1; | |
3535 | } | |
3536 | ||
3537 | __setup("slub_max_order=", setup_slub_max_order); | |
3538 | ||
3539 | static int __init setup_slub_min_objects(char *str) | |
3540 | { | |
06428780 | 3541 | get_option(&str, &slub_min_objects); |
81819f0f CL |
3542 | |
3543 | return 1; | |
3544 | } | |
3545 | ||
3546 | __setup("slub_min_objects=", setup_slub_min_objects); | |
3547 | ||
81819f0f CL |
3548 | void *__kmalloc(size_t size, gfp_t flags) |
3549 | { | |
aadb4bc4 | 3550 | struct kmem_cache *s; |
5b882be4 | 3551 | void *ret; |
81819f0f | 3552 | |
95a05b42 | 3553 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) |
eada35ef | 3554 | return kmalloc_large(size, flags); |
aadb4bc4 | 3555 | |
2c59dd65 | 3556 | s = kmalloc_slab(size, flags); |
aadb4bc4 CL |
3557 | |
3558 | if (unlikely(ZERO_OR_NULL_PTR(s))) | |
6cb8f913 CL |
3559 | return s; |
3560 | ||
2b847c3c | 3561 | ret = slab_alloc(s, flags, _RET_IP_); |
5b882be4 | 3562 | |
ca2b84cb | 3563 | trace_kmalloc(_RET_IP_, ret, size, s->size, flags); |
5b882be4 | 3564 | |
505f5dcb | 3565 | kasan_kmalloc(s, ret, size, flags); |
0316bec2 | 3566 | |
5b882be4 | 3567 | return ret; |
81819f0f CL |
3568 | } |
3569 | EXPORT_SYMBOL(__kmalloc); | |
3570 | ||
5d1f57e4 | 3571 | #ifdef CONFIG_NUMA |
f619cfe1 CL |
3572 | static void *kmalloc_large_node(size_t size, gfp_t flags, int node) |
3573 | { | |
b1eeab67 | 3574 | struct page *page; |
e4f7c0b4 | 3575 | void *ptr = NULL; |
f619cfe1 | 3576 | |
52383431 VD |
3577 | flags |= __GFP_COMP | __GFP_NOTRACK; |
3578 | page = alloc_kmem_pages_node(node, flags, get_order(size)); | |
f619cfe1 | 3579 | if (page) |
e4f7c0b4 CM |
3580 | ptr = page_address(page); |
3581 | ||
d56791b3 | 3582 | kmalloc_large_node_hook(ptr, size, flags); |
e4f7c0b4 | 3583 | return ptr; |
f619cfe1 CL |
3584 | } |
3585 | ||
81819f0f CL |
3586 | void *__kmalloc_node(size_t size, gfp_t flags, int node) |
3587 | { | |
aadb4bc4 | 3588 | struct kmem_cache *s; |
5b882be4 | 3589 | void *ret; |
81819f0f | 3590 | |
95a05b42 | 3591 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) { |
5b882be4 EGM |
3592 | ret = kmalloc_large_node(size, flags, node); |
3593 | ||
ca2b84cb EGM |
3594 | trace_kmalloc_node(_RET_IP_, ret, |
3595 | size, PAGE_SIZE << get_order(size), | |
3596 | flags, node); | |
5b882be4 EGM |
3597 | |
3598 | return ret; | |
3599 | } | |
aadb4bc4 | 3600 | |
2c59dd65 | 3601 | s = kmalloc_slab(size, flags); |
aadb4bc4 CL |
3602 | |
3603 | if (unlikely(ZERO_OR_NULL_PTR(s))) | |
6cb8f913 CL |
3604 | return s; |
3605 | ||
2b847c3c | 3606 | ret = slab_alloc_node(s, flags, node, _RET_IP_); |
5b882be4 | 3607 | |
ca2b84cb | 3608 | trace_kmalloc_node(_RET_IP_, ret, size, s->size, flags, node); |
5b882be4 | 3609 | |
505f5dcb | 3610 | kasan_kmalloc(s, ret, size, flags); |
0316bec2 | 3611 | |
5b882be4 | 3612 | return ret; |
81819f0f CL |
3613 | } |
3614 | EXPORT_SYMBOL(__kmalloc_node); | |
3615 | #endif | |
3616 | ||
0316bec2 | 3617 | static size_t __ksize(const void *object) |
81819f0f | 3618 | { |
272c1d21 | 3619 | struct page *page; |
81819f0f | 3620 | |
ef8b4520 | 3621 | if (unlikely(object == ZERO_SIZE_PTR)) |
272c1d21 CL |
3622 | return 0; |
3623 | ||
294a80a8 | 3624 | page = virt_to_head_page(object); |
294a80a8 | 3625 | |
76994412 PE |
3626 | if (unlikely(!PageSlab(page))) { |
3627 | WARN_ON(!PageCompound(page)); | |
294a80a8 | 3628 | return PAGE_SIZE << compound_order(page); |
76994412 | 3629 | } |
81819f0f | 3630 | |
1b4f59e3 | 3631 | return slab_ksize(page->slab_cache); |
81819f0f | 3632 | } |
0316bec2 AR |
3633 | |
3634 | size_t ksize(const void *object) | |
3635 | { | |
3636 | size_t size = __ksize(object); | |
3637 | /* We assume that ksize callers could use whole allocated area, | |
3638 | so we need unpoison this area. */ | |
505f5dcb | 3639 | kasan_krealloc(object, size, GFP_NOWAIT); |
0316bec2 AR |
3640 | return size; |
3641 | } | |
b1aabecd | 3642 | EXPORT_SYMBOL(ksize); |
81819f0f CL |
3643 | |
3644 | void kfree(const void *x) | |
3645 | { | |
81819f0f | 3646 | struct page *page; |
5bb983b0 | 3647 | void *object = (void *)x; |
81819f0f | 3648 | |
2121db74 PE |
3649 | trace_kfree(_RET_IP_, x); |
3650 | ||
2408c550 | 3651 | if (unlikely(ZERO_OR_NULL_PTR(x))) |
81819f0f CL |
3652 | return; |
3653 | ||
b49af68f | 3654 | page = virt_to_head_page(x); |
aadb4bc4 | 3655 | if (unlikely(!PageSlab(page))) { |
0937502a | 3656 | BUG_ON(!PageCompound(page)); |
d56791b3 | 3657 | kfree_hook(x); |
52383431 | 3658 | __free_kmem_pages(page, compound_order(page)); |
aadb4bc4 CL |
3659 | return; |
3660 | } | |
81084651 | 3661 | slab_free(page->slab_cache, page, object, NULL, 1, _RET_IP_); |
81819f0f CL |
3662 | } |
3663 | EXPORT_SYMBOL(kfree); | |
3664 | ||
832f37f5 VD |
3665 | #define SHRINK_PROMOTE_MAX 32 |
3666 | ||
2086d26a | 3667 | /* |
832f37f5 VD |
3668 | * kmem_cache_shrink discards empty slabs and promotes the slabs filled |
3669 | * up most to the head of the partial lists. New allocations will then | |
3670 | * fill those up and thus they can be removed from the partial lists. | |
672bba3a CL |
3671 | * |
3672 | * The slabs with the least items are placed last. This results in them | |
3673 | * being allocated from last increasing the chance that the last objects | |
3674 | * are freed in them. | |
2086d26a | 3675 | */ |
d6e0b7fa | 3676 | int __kmem_cache_shrink(struct kmem_cache *s, bool deactivate) |
2086d26a CL |
3677 | { |
3678 | int node; | |
3679 | int i; | |
3680 | struct kmem_cache_node *n; | |
3681 | struct page *page; | |
3682 | struct page *t; | |
832f37f5 VD |
3683 | struct list_head discard; |
3684 | struct list_head promote[SHRINK_PROMOTE_MAX]; | |
2086d26a | 3685 | unsigned long flags; |
ce3712d7 | 3686 | int ret = 0; |
2086d26a | 3687 | |
d6e0b7fa VD |
3688 | if (deactivate) { |
3689 | /* | |
3690 | * Disable empty slabs caching. Used to avoid pinning offline | |
3691 | * memory cgroups by kmem pages that can be freed. | |
3692 | */ | |
3693 | s->cpu_partial = 0; | |
3694 | s->min_partial = 0; | |
3695 | ||
3696 | /* | |
3697 | * s->cpu_partial is checked locklessly (see put_cpu_partial), | |
3698 | * so we have to make sure the change is visible. | |
3699 | */ | |
3700 | kick_all_cpus_sync(); | |
3701 | } | |
3702 | ||
2086d26a | 3703 | flush_all(s); |
fa45dc25 | 3704 | for_each_kmem_cache_node(s, node, n) { |
832f37f5 VD |
3705 | INIT_LIST_HEAD(&discard); |
3706 | for (i = 0; i < SHRINK_PROMOTE_MAX; i++) | |
3707 | INIT_LIST_HEAD(promote + i); | |
2086d26a CL |
3708 | |
3709 | spin_lock_irqsave(&n->list_lock, flags); | |
3710 | ||
3711 | /* | |
832f37f5 | 3712 | * Build lists of slabs to discard or promote. |
2086d26a | 3713 | * |
672bba3a CL |
3714 | * Note that concurrent frees may occur while we hold the |
3715 | * list_lock. page->inuse here is the upper limit. | |
2086d26a CL |
3716 | */ |
3717 | list_for_each_entry_safe(page, t, &n->partial, lru) { | |
832f37f5 VD |
3718 | int free = page->objects - page->inuse; |
3719 | ||
3720 | /* Do not reread page->inuse */ | |
3721 | barrier(); | |
3722 | ||
3723 | /* We do not keep full slabs on the list */ | |
3724 | BUG_ON(free <= 0); | |
3725 | ||
3726 | if (free == page->objects) { | |
3727 | list_move(&page->lru, &discard); | |
69cb8e6b | 3728 | n->nr_partial--; |
832f37f5 VD |
3729 | } else if (free <= SHRINK_PROMOTE_MAX) |
3730 | list_move(&page->lru, promote + free - 1); | |
2086d26a CL |
3731 | } |
3732 | ||
2086d26a | 3733 | /* |
832f37f5 VD |
3734 | * Promote the slabs filled up most to the head of the |
3735 | * partial list. | |
2086d26a | 3736 | */ |
832f37f5 VD |
3737 | for (i = SHRINK_PROMOTE_MAX - 1; i >= 0; i--) |
3738 | list_splice(promote + i, &n->partial); | |
2086d26a | 3739 | |
2086d26a | 3740 | spin_unlock_irqrestore(&n->list_lock, flags); |
69cb8e6b CL |
3741 | |
3742 | /* Release empty slabs */ | |
832f37f5 | 3743 | list_for_each_entry_safe(page, t, &discard, lru) |
69cb8e6b | 3744 | discard_slab(s, page); |
ce3712d7 VD |
3745 | |
3746 | if (slabs_node(s, node)) | |
3747 | ret = 1; | |
2086d26a CL |
3748 | } |
3749 | ||
ce3712d7 | 3750 | return ret; |
2086d26a | 3751 | } |
2086d26a | 3752 | |
b9049e23 YG |
3753 | static int slab_mem_going_offline_callback(void *arg) |
3754 | { | |
3755 | struct kmem_cache *s; | |
3756 | ||
18004c5d | 3757 | mutex_lock(&slab_mutex); |
b9049e23 | 3758 | list_for_each_entry(s, &slab_caches, list) |
d6e0b7fa | 3759 | __kmem_cache_shrink(s, false); |
18004c5d | 3760 | mutex_unlock(&slab_mutex); |
b9049e23 YG |
3761 | |
3762 | return 0; | |
3763 | } | |
3764 | ||
3765 | static void slab_mem_offline_callback(void *arg) | |
3766 | { | |
3767 | struct kmem_cache_node *n; | |
3768 | struct kmem_cache *s; | |
3769 | struct memory_notify *marg = arg; | |
3770 | int offline_node; | |
3771 | ||
b9d5ab25 | 3772 | offline_node = marg->status_change_nid_normal; |
b9049e23 YG |
3773 | |
3774 | /* | |
3775 | * If the node still has available memory. we need kmem_cache_node | |
3776 | * for it yet. | |
3777 | */ | |
3778 | if (offline_node < 0) | |
3779 | return; | |
3780 | ||
18004c5d | 3781 | mutex_lock(&slab_mutex); |
b9049e23 YG |
3782 | list_for_each_entry(s, &slab_caches, list) { |
3783 | n = get_node(s, offline_node); | |
3784 | if (n) { | |
3785 | /* | |
3786 | * if n->nr_slabs > 0, slabs still exist on the node | |
3787 | * that is going down. We were unable to free them, | |
c9404c9c | 3788 | * and offline_pages() function shouldn't call this |
b9049e23 YG |
3789 | * callback. So, we must fail. |
3790 | */ | |
0f389ec6 | 3791 | BUG_ON(slabs_node(s, offline_node)); |
b9049e23 YG |
3792 | |
3793 | s->node[offline_node] = NULL; | |
8de66a0c | 3794 | kmem_cache_free(kmem_cache_node, n); |
b9049e23 YG |
3795 | } |
3796 | } | |
18004c5d | 3797 | mutex_unlock(&slab_mutex); |
b9049e23 YG |
3798 | } |
3799 | ||
3800 | static int slab_mem_going_online_callback(void *arg) | |
3801 | { | |
3802 | struct kmem_cache_node *n; | |
3803 | struct kmem_cache *s; | |
3804 | struct memory_notify *marg = arg; | |
b9d5ab25 | 3805 | int nid = marg->status_change_nid_normal; |
b9049e23 YG |
3806 | int ret = 0; |
3807 | ||
3808 | /* | |
3809 | * If the node's memory is already available, then kmem_cache_node is | |
3810 | * already created. Nothing to do. | |
3811 | */ | |
3812 | if (nid < 0) | |
3813 | return 0; | |
3814 | ||
3815 | /* | |
0121c619 | 3816 | * We are bringing a node online. No memory is available yet. We must |
b9049e23 YG |
3817 | * allocate a kmem_cache_node structure in order to bring the node |
3818 | * online. | |
3819 | */ | |
18004c5d | 3820 | mutex_lock(&slab_mutex); |
b9049e23 YG |
3821 | list_for_each_entry(s, &slab_caches, list) { |
3822 | /* | |
3823 | * XXX: kmem_cache_alloc_node will fallback to other nodes | |
3824 | * since memory is not yet available from the node that | |
3825 | * is brought up. | |
3826 | */ | |
8de66a0c | 3827 | n = kmem_cache_alloc(kmem_cache_node, GFP_KERNEL); |
b9049e23 YG |
3828 | if (!n) { |
3829 | ret = -ENOMEM; | |
3830 | goto out; | |
3831 | } | |
4053497d | 3832 | init_kmem_cache_node(n); |
b9049e23 YG |
3833 | s->node[nid] = n; |
3834 | } | |
3835 | out: | |
18004c5d | 3836 | mutex_unlock(&slab_mutex); |
b9049e23 YG |
3837 | return ret; |
3838 | } | |
3839 | ||
3840 | static int slab_memory_callback(struct notifier_block *self, | |
3841 | unsigned long action, void *arg) | |
3842 | { | |
3843 | int ret = 0; | |
3844 | ||
3845 | switch (action) { | |
3846 | case MEM_GOING_ONLINE: | |
3847 | ret = slab_mem_going_online_callback(arg); | |
3848 | break; | |
3849 | case MEM_GOING_OFFLINE: | |
3850 | ret = slab_mem_going_offline_callback(arg); | |
3851 | break; | |
3852 | case MEM_OFFLINE: | |
3853 | case MEM_CANCEL_ONLINE: | |
3854 | slab_mem_offline_callback(arg); | |
3855 | break; | |
3856 | case MEM_ONLINE: | |
3857 | case MEM_CANCEL_OFFLINE: | |
3858 | break; | |
3859 | } | |
dc19f9db KH |
3860 | if (ret) |
3861 | ret = notifier_from_errno(ret); | |
3862 | else | |
3863 | ret = NOTIFY_OK; | |
b9049e23 YG |
3864 | return ret; |
3865 | } | |
3866 | ||
3ac38faa AM |
3867 | static struct notifier_block slab_memory_callback_nb = { |
3868 | .notifier_call = slab_memory_callback, | |
3869 | .priority = SLAB_CALLBACK_PRI, | |
3870 | }; | |
b9049e23 | 3871 | |
81819f0f CL |
3872 | /******************************************************************** |
3873 | * Basic setup of slabs | |
3874 | *******************************************************************/ | |
3875 | ||
51df1142 CL |
3876 | /* |
3877 | * Used for early kmem_cache structures that were allocated using | |
dffb4d60 CL |
3878 | * the page allocator. Allocate them properly then fix up the pointers |
3879 | * that may be pointing to the wrong kmem_cache structure. | |
51df1142 CL |
3880 | */ |
3881 | ||
dffb4d60 | 3882 | static struct kmem_cache * __init bootstrap(struct kmem_cache *static_cache) |
51df1142 CL |
3883 | { |
3884 | int node; | |
dffb4d60 | 3885 | struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT); |
fa45dc25 | 3886 | struct kmem_cache_node *n; |
51df1142 | 3887 | |
dffb4d60 | 3888 | memcpy(s, static_cache, kmem_cache->object_size); |
51df1142 | 3889 | |
7d557b3c GC |
3890 | /* |
3891 | * This runs very early, and only the boot processor is supposed to be | |
3892 | * up. Even if it weren't true, IRQs are not up so we couldn't fire | |
3893 | * IPIs around. | |
3894 | */ | |
3895 | __flush_cpu_slab(s, smp_processor_id()); | |
fa45dc25 | 3896 | for_each_kmem_cache_node(s, node, n) { |
51df1142 CL |
3897 | struct page *p; |
3898 | ||
fa45dc25 CL |
3899 | list_for_each_entry(p, &n->partial, lru) |
3900 | p->slab_cache = s; | |
51df1142 | 3901 | |
607bf324 | 3902 | #ifdef CONFIG_SLUB_DEBUG |
fa45dc25 CL |
3903 | list_for_each_entry(p, &n->full, lru) |
3904 | p->slab_cache = s; | |
51df1142 | 3905 | #endif |
51df1142 | 3906 | } |
f7ce3190 | 3907 | slab_init_memcg_params(s); |
dffb4d60 CL |
3908 | list_add(&s->list, &slab_caches); |
3909 | return s; | |
51df1142 CL |
3910 | } |
3911 | ||
81819f0f CL |
3912 | void __init kmem_cache_init(void) |
3913 | { | |
dffb4d60 CL |
3914 | static __initdata struct kmem_cache boot_kmem_cache, |
3915 | boot_kmem_cache_node; | |
51df1142 | 3916 | |
fc8d8620 SG |
3917 | if (debug_guardpage_minorder()) |
3918 | slub_max_order = 0; | |
3919 | ||
dffb4d60 CL |
3920 | kmem_cache_node = &boot_kmem_cache_node; |
3921 | kmem_cache = &boot_kmem_cache; | |
51df1142 | 3922 | |
dffb4d60 CL |
3923 | create_boot_cache(kmem_cache_node, "kmem_cache_node", |
3924 | sizeof(struct kmem_cache_node), SLAB_HWCACHE_ALIGN); | |
b9049e23 | 3925 | |
3ac38faa | 3926 | register_hotmemory_notifier(&slab_memory_callback_nb); |
81819f0f CL |
3927 | |
3928 | /* Able to allocate the per node structures */ | |
3929 | slab_state = PARTIAL; | |
3930 | ||
dffb4d60 CL |
3931 | create_boot_cache(kmem_cache, "kmem_cache", |
3932 | offsetof(struct kmem_cache, node) + | |
3933 | nr_node_ids * sizeof(struct kmem_cache_node *), | |
3934 | SLAB_HWCACHE_ALIGN); | |
8a13a4cc | 3935 | |
dffb4d60 | 3936 | kmem_cache = bootstrap(&boot_kmem_cache); |
81819f0f | 3937 | |
51df1142 CL |
3938 | /* |
3939 | * Allocate kmem_cache_node properly from the kmem_cache slab. | |
3940 | * kmem_cache_node is separately allocated so no need to | |
3941 | * update any list pointers. | |
3942 | */ | |
dffb4d60 | 3943 | kmem_cache_node = bootstrap(&boot_kmem_cache_node); |
51df1142 CL |
3944 | |
3945 | /* Now we can use the kmem_cache to allocate kmalloc slabs */ | |
34cc6990 | 3946 | setup_kmalloc_cache_index_table(); |
f97d5f63 | 3947 | create_kmalloc_caches(0); |
81819f0f CL |
3948 | |
3949 | #ifdef CONFIG_SMP | |
3950 | register_cpu_notifier(&slab_notifier); | |
9dfc6e68 | 3951 | #endif |
81819f0f | 3952 | |
f9f58285 | 3953 | pr_info("SLUB: HWalign=%d, Order=%d-%d, MinObjects=%d, CPUs=%d, Nodes=%d\n", |
f97d5f63 | 3954 | cache_line_size(), |
81819f0f CL |
3955 | slub_min_order, slub_max_order, slub_min_objects, |
3956 | nr_cpu_ids, nr_node_ids); | |
3957 | } | |
3958 | ||
7e85ee0c PE |
3959 | void __init kmem_cache_init_late(void) |
3960 | { | |
7e85ee0c PE |
3961 | } |
3962 | ||
2633d7a0 | 3963 | struct kmem_cache * |
a44cb944 VD |
3964 | __kmem_cache_alias(const char *name, size_t size, size_t align, |
3965 | unsigned long flags, void (*ctor)(void *)) | |
81819f0f | 3966 | { |
426589f5 | 3967 | struct kmem_cache *s, *c; |
81819f0f | 3968 | |
a44cb944 | 3969 | s = find_mergeable(size, align, flags, name, ctor); |
81819f0f CL |
3970 | if (s) { |
3971 | s->refcount++; | |
84d0ddd6 | 3972 | |
81819f0f CL |
3973 | /* |
3974 | * Adjust the object sizes so that we clear | |
3975 | * the complete object on kzalloc. | |
3976 | */ | |
3b0efdfa | 3977 | s->object_size = max(s->object_size, (int)size); |
81819f0f | 3978 | s->inuse = max_t(int, s->inuse, ALIGN(size, sizeof(void *))); |
6446faa2 | 3979 | |
426589f5 | 3980 | for_each_memcg_cache(c, s) { |
84d0ddd6 VD |
3981 | c->object_size = s->object_size; |
3982 | c->inuse = max_t(int, c->inuse, | |
3983 | ALIGN(size, sizeof(void *))); | |
3984 | } | |
3985 | ||
7b8f3b66 | 3986 | if (sysfs_slab_alias(s, name)) { |
7b8f3b66 | 3987 | s->refcount--; |
cbb79694 | 3988 | s = NULL; |
7b8f3b66 | 3989 | } |
a0e1d1be | 3990 | } |
6446faa2 | 3991 | |
cbb79694 CL |
3992 | return s; |
3993 | } | |
84c1cf62 | 3994 | |
8a13a4cc | 3995 | int __kmem_cache_create(struct kmem_cache *s, unsigned long flags) |
cbb79694 | 3996 | { |
aac3a166 PE |
3997 | int err; |
3998 | ||
3999 | err = kmem_cache_open(s, flags); | |
4000 | if (err) | |
4001 | return err; | |
20cea968 | 4002 | |
45530c44 CL |
4003 | /* Mutex is not taken during early boot */ |
4004 | if (slab_state <= UP) | |
4005 | return 0; | |
4006 | ||
107dab5c | 4007 | memcg_propagate_slab_attrs(s); |
aac3a166 | 4008 | err = sysfs_slab_add(s); |
aac3a166 | 4009 | if (err) |
52b4b950 | 4010 | __kmem_cache_release(s); |
20cea968 | 4011 | |
aac3a166 | 4012 | return err; |
81819f0f | 4013 | } |
81819f0f | 4014 | |
81819f0f | 4015 | #ifdef CONFIG_SMP |
81819f0f | 4016 | /* |
672bba3a CL |
4017 | * Use the cpu notifier to insure that the cpu slabs are flushed when |
4018 | * necessary. | |
81819f0f | 4019 | */ |
0db0628d | 4020 | static int slab_cpuup_callback(struct notifier_block *nfb, |
81819f0f CL |
4021 | unsigned long action, void *hcpu) |
4022 | { | |
4023 | long cpu = (long)hcpu; | |
5b95a4ac CL |
4024 | struct kmem_cache *s; |
4025 | unsigned long flags; | |
81819f0f CL |
4026 | |
4027 | switch (action) { | |
4028 | case CPU_UP_CANCELED: | |
8bb78442 | 4029 | case CPU_UP_CANCELED_FROZEN: |
81819f0f | 4030 | case CPU_DEAD: |
8bb78442 | 4031 | case CPU_DEAD_FROZEN: |
18004c5d | 4032 | mutex_lock(&slab_mutex); |
5b95a4ac CL |
4033 | list_for_each_entry(s, &slab_caches, list) { |
4034 | local_irq_save(flags); | |
4035 | __flush_cpu_slab(s, cpu); | |
4036 | local_irq_restore(flags); | |
4037 | } | |
18004c5d | 4038 | mutex_unlock(&slab_mutex); |
81819f0f CL |
4039 | break; |
4040 | default: | |
4041 | break; | |
4042 | } | |
4043 | return NOTIFY_OK; | |
4044 | } | |
4045 | ||
0db0628d | 4046 | static struct notifier_block slab_notifier = { |
3adbefee | 4047 | .notifier_call = slab_cpuup_callback |
06428780 | 4048 | }; |
81819f0f CL |
4049 | |
4050 | #endif | |
4051 | ||
ce71e27c | 4052 | void *__kmalloc_track_caller(size_t size, gfp_t gfpflags, unsigned long caller) |
81819f0f | 4053 | { |
aadb4bc4 | 4054 | struct kmem_cache *s; |
94b528d0 | 4055 | void *ret; |
aadb4bc4 | 4056 | |
95a05b42 | 4057 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) |
eada35ef PE |
4058 | return kmalloc_large(size, gfpflags); |
4059 | ||
2c59dd65 | 4060 | s = kmalloc_slab(size, gfpflags); |
81819f0f | 4061 | |
2408c550 | 4062 | if (unlikely(ZERO_OR_NULL_PTR(s))) |
6cb8f913 | 4063 | return s; |
81819f0f | 4064 | |
2b847c3c | 4065 | ret = slab_alloc(s, gfpflags, caller); |
94b528d0 | 4066 | |
25985edc | 4067 | /* Honor the call site pointer we received. */ |
ca2b84cb | 4068 | trace_kmalloc(caller, ret, size, s->size, gfpflags); |
94b528d0 EGM |
4069 | |
4070 | return ret; | |
81819f0f CL |
4071 | } |
4072 | ||
5d1f57e4 | 4073 | #ifdef CONFIG_NUMA |
81819f0f | 4074 | void *__kmalloc_node_track_caller(size_t size, gfp_t gfpflags, |
ce71e27c | 4075 | int node, unsigned long caller) |
81819f0f | 4076 | { |
aadb4bc4 | 4077 | struct kmem_cache *s; |
94b528d0 | 4078 | void *ret; |
aadb4bc4 | 4079 | |
95a05b42 | 4080 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) { |
d3e14aa3 XF |
4081 | ret = kmalloc_large_node(size, gfpflags, node); |
4082 | ||
4083 | trace_kmalloc_node(caller, ret, | |
4084 | size, PAGE_SIZE << get_order(size), | |
4085 | gfpflags, node); | |
4086 | ||
4087 | return ret; | |
4088 | } | |
eada35ef | 4089 | |
2c59dd65 | 4090 | s = kmalloc_slab(size, gfpflags); |
81819f0f | 4091 | |
2408c550 | 4092 | if (unlikely(ZERO_OR_NULL_PTR(s))) |
6cb8f913 | 4093 | return s; |
81819f0f | 4094 | |
2b847c3c | 4095 | ret = slab_alloc_node(s, gfpflags, node, caller); |
94b528d0 | 4096 | |
25985edc | 4097 | /* Honor the call site pointer we received. */ |
ca2b84cb | 4098 | trace_kmalloc_node(caller, ret, size, s->size, gfpflags, node); |
94b528d0 EGM |
4099 | |
4100 | return ret; | |
81819f0f | 4101 | } |
5d1f57e4 | 4102 | #endif |
81819f0f | 4103 | |
ab4d5ed5 | 4104 | #ifdef CONFIG_SYSFS |
205ab99d CL |
4105 | static int count_inuse(struct page *page) |
4106 | { | |
4107 | return page->inuse; | |
4108 | } | |
4109 | ||
4110 | static int count_total(struct page *page) | |
4111 | { | |
4112 | return page->objects; | |
4113 | } | |
ab4d5ed5 | 4114 | #endif |
205ab99d | 4115 | |
ab4d5ed5 | 4116 | #ifdef CONFIG_SLUB_DEBUG |
434e245d CL |
4117 | static int validate_slab(struct kmem_cache *s, struct page *page, |
4118 | unsigned long *map) | |
53e15af0 CL |
4119 | { |
4120 | void *p; | |
a973e9dd | 4121 | void *addr = page_address(page); |
53e15af0 CL |
4122 | |
4123 | if (!check_slab(s, page) || | |
4124 | !on_freelist(s, page, NULL)) | |
4125 | return 0; | |
4126 | ||
4127 | /* Now we know that a valid freelist exists */ | |
39b26464 | 4128 | bitmap_zero(map, page->objects); |
53e15af0 | 4129 | |
5f80b13a CL |
4130 | get_map(s, page, map); |
4131 | for_each_object(p, s, addr, page->objects) { | |
4132 | if (test_bit(slab_index(p, s, addr), map)) | |
4133 | if (!check_object(s, page, p, SLUB_RED_INACTIVE)) | |
4134 | return 0; | |
53e15af0 CL |
4135 | } |
4136 | ||
224a88be | 4137 | for_each_object(p, s, addr, page->objects) |
7656c72b | 4138 | if (!test_bit(slab_index(p, s, addr), map)) |
37d57443 | 4139 | if (!check_object(s, page, p, SLUB_RED_ACTIVE)) |
53e15af0 CL |
4140 | return 0; |
4141 | return 1; | |
4142 | } | |
4143 | ||
434e245d CL |
4144 | static void validate_slab_slab(struct kmem_cache *s, struct page *page, |
4145 | unsigned long *map) | |
53e15af0 | 4146 | { |
881db7fb CL |
4147 | slab_lock(page); |
4148 | validate_slab(s, page, map); | |
4149 | slab_unlock(page); | |
53e15af0 CL |
4150 | } |
4151 | ||
434e245d CL |
4152 | static int validate_slab_node(struct kmem_cache *s, |
4153 | struct kmem_cache_node *n, unsigned long *map) | |
53e15af0 CL |
4154 | { |
4155 | unsigned long count = 0; | |
4156 | struct page *page; | |
4157 | unsigned long flags; | |
4158 | ||
4159 | spin_lock_irqsave(&n->list_lock, flags); | |
4160 | ||
4161 | list_for_each_entry(page, &n->partial, lru) { | |
434e245d | 4162 | validate_slab_slab(s, page, map); |
53e15af0 CL |
4163 | count++; |
4164 | } | |
4165 | if (count != n->nr_partial) | |
f9f58285 FF |
4166 | pr_err("SLUB %s: %ld partial slabs counted but counter=%ld\n", |
4167 | s->name, count, n->nr_partial); | |
53e15af0 CL |
4168 | |
4169 | if (!(s->flags & SLAB_STORE_USER)) | |
4170 | goto out; | |
4171 | ||
4172 | list_for_each_entry(page, &n->full, lru) { | |
434e245d | 4173 | validate_slab_slab(s, page, map); |
53e15af0 CL |
4174 | count++; |
4175 | } | |
4176 | if (count != atomic_long_read(&n->nr_slabs)) | |
f9f58285 FF |
4177 | pr_err("SLUB: %s %ld slabs counted but counter=%ld\n", |
4178 | s->name, count, atomic_long_read(&n->nr_slabs)); | |
53e15af0 CL |
4179 | |
4180 | out: | |
4181 | spin_unlock_irqrestore(&n->list_lock, flags); | |
4182 | return count; | |
4183 | } | |
4184 | ||
434e245d | 4185 | static long validate_slab_cache(struct kmem_cache *s) |
53e15af0 CL |
4186 | { |
4187 | int node; | |
4188 | unsigned long count = 0; | |
205ab99d | 4189 | unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) * |
434e245d | 4190 | sizeof(unsigned long), GFP_KERNEL); |
fa45dc25 | 4191 | struct kmem_cache_node *n; |
434e245d CL |
4192 | |
4193 | if (!map) | |
4194 | return -ENOMEM; | |
53e15af0 CL |
4195 | |
4196 | flush_all(s); | |
fa45dc25 | 4197 | for_each_kmem_cache_node(s, node, n) |
434e245d | 4198 | count += validate_slab_node(s, n, map); |
434e245d | 4199 | kfree(map); |
53e15af0 CL |
4200 | return count; |
4201 | } | |
88a420e4 | 4202 | /* |
672bba3a | 4203 | * Generate lists of code addresses where slabcache objects are allocated |
88a420e4 CL |
4204 | * and freed. |
4205 | */ | |
4206 | ||
4207 | struct location { | |
4208 | unsigned long count; | |
ce71e27c | 4209 | unsigned long addr; |
45edfa58 CL |
4210 | long long sum_time; |
4211 | long min_time; | |
4212 | long max_time; | |
4213 | long min_pid; | |
4214 | long max_pid; | |
174596a0 | 4215 | DECLARE_BITMAP(cpus, NR_CPUS); |
45edfa58 | 4216 | nodemask_t nodes; |
88a420e4 CL |
4217 | }; |
4218 | ||
4219 | struct loc_track { | |
4220 | unsigned long max; | |
4221 | unsigned long count; | |
4222 | struct location *loc; | |
4223 | }; | |
4224 | ||
4225 | static void free_loc_track(struct loc_track *t) | |
4226 | { | |
4227 | if (t->max) | |
4228 | free_pages((unsigned long)t->loc, | |
4229 | get_order(sizeof(struct location) * t->max)); | |
4230 | } | |
4231 | ||
68dff6a9 | 4232 | static int alloc_loc_track(struct loc_track *t, unsigned long max, gfp_t flags) |
88a420e4 CL |
4233 | { |
4234 | struct location *l; | |
4235 | int order; | |
4236 | ||
88a420e4 CL |
4237 | order = get_order(sizeof(struct location) * max); |
4238 | ||
68dff6a9 | 4239 | l = (void *)__get_free_pages(flags, order); |
88a420e4 CL |
4240 | if (!l) |
4241 | return 0; | |
4242 | ||
4243 | if (t->count) { | |
4244 | memcpy(l, t->loc, sizeof(struct location) * t->count); | |
4245 | free_loc_track(t); | |
4246 | } | |
4247 | t->max = max; | |
4248 | t->loc = l; | |
4249 | return 1; | |
4250 | } | |
4251 | ||
4252 | static int add_location(struct loc_track *t, struct kmem_cache *s, | |
45edfa58 | 4253 | const struct track *track) |
88a420e4 CL |
4254 | { |
4255 | long start, end, pos; | |
4256 | struct location *l; | |
ce71e27c | 4257 | unsigned long caddr; |
45edfa58 | 4258 | unsigned long age = jiffies - track->when; |
88a420e4 CL |
4259 | |
4260 | start = -1; | |
4261 | end = t->count; | |
4262 | ||
4263 | for ( ; ; ) { | |
4264 | pos = start + (end - start + 1) / 2; | |
4265 | ||
4266 | /* | |
4267 | * There is nothing at "end". If we end up there | |
4268 | * we need to add something to before end. | |
4269 | */ | |
4270 | if (pos == end) | |
4271 | break; | |
4272 | ||
4273 | caddr = t->loc[pos].addr; | |
45edfa58 CL |
4274 | if (track->addr == caddr) { |
4275 | ||
4276 | l = &t->loc[pos]; | |
4277 | l->count++; | |
4278 | if (track->when) { | |
4279 | l->sum_time += age; | |
4280 | if (age < l->min_time) | |
4281 | l->min_time = age; | |
4282 | if (age > l->max_time) | |
4283 | l->max_time = age; | |
4284 | ||
4285 | if (track->pid < l->min_pid) | |
4286 | l->min_pid = track->pid; | |
4287 | if (track->pid > l->max_pid) | |
4288 | l->max_pid = track->pid; | |
4289 | ||
174596a0 RR |
4290 | cpumask_set_cpu(track->cpu, |
4291 | to_cpumask(l->cpus)); | |
45edfa58 CL |
4292 | } |
4293 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
88a420e4 CL |
4294 | return 1; |
4295 | } | |
4296 | ||
45edfa58 | 4297 | if (track->addr < caddr) |
88a420e4 CL |
4298 | end = pos; |
4299 | else | |
4300 | start = pos; | |
4301 | } | |
4302 | ||
4303 | /* | |
672bba3a | 4304 | * Not found. Insert new tracking element. |
88a420e4 | 4305 | */ |
68dff6a9 | 4306 | if (t->count >= t->max && !alloc_loc_track(t, 2 * t->max, GFP_ATOMIC)) |
88a420e4 CL |
4307 | return 0; |
4308 | ||
4309 | l = t->loc + pos; | |
4310 | if (pos < t->count) | |
4311 | memmove(l + 1, l, | |
4312 | (t->count - pos) * sizeof(struct location)); | |
4313 | t->count++; | |
4314 | l->count = 1; | |
45edfa58 CL |
4315 | l->addr = track->addr; |
4316 | l->sum_time = age; | |
4317 | l->min_time = age; | |
4318 | l->max_time = age; | |
4319 | l->min_pid = track->pid; | |
4320 | l->max_pid = track->pid; | |
174596a0 RR |
4321 | cpumask_clear(to_cpumask(l->cpus)); |
4322 | cpumask_set_cpu(track->cpu, to_cpumask(l->cpus)); | |
45edfa58 CL |
4323 | nodes_clear(l->nodes); |
4324 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
88a420e4 CL |
4325 | return 1; |
4326 | } | |
4327 | ||
4328 | static void process_slab(struct loc_track *t, struct kmem_cache *s, | |
bbd7d57b | 4329 | struct page *page, enum track_item alloc, |
a5dd5c11 | 4330 | unsigned long *map) |
88a420e4 | 4331 | { |
a973e9dd | 4332 | void *addr = page_address(page); |
88a420e4 CL |
4333 | void *p; |
4334 | ||
39b26464 | 4335 | bitmap_zero(map, page->objects); |
5f80b13a | 4336 | get_map(s, page, map); |
88a420e4 | 4337 | |
224a88be | 4338 | for_each_object(p, s, addr, page->objects) |
45edfa58 CL |
4339 | if (!test_bit(slab_index(p, s, addr), map)) |
4340 | add_location(t, s, get_track(s, p, alloc)); | |
88a420e4 CL |
4341 | } |
4342 | ||
4343 | static int list_locations(struct kmem_cache *s, char *buf, | |
4344 | enum track_item alloc) | |
4345 | { | |
e374d483 | 4346 | int len = 0; |
88a420e4 | 4347 | unsigned long i; |
68dff6a9 | 4348 | struct loc_track t = { 0, 0, NULL }; |
88a420e4 | 4349 | int node; |
bbd7d57b ED |
4350 | unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) * |
4351 | sizeof(unsigned long), GFP_KERNEL); | |
fa45dc25 | 4352 | struct kmem_cache_node *n; |
88a420e4 | 4353 | |
bbd7d57b ED |
4354 | if (!map || !alloc_loc_track(&t, PAGE_SIZE / sizeof(struct location), |
4355 | GFP_TEMPORARY)) { | |
4356 | kfree(map); | |
68dff6a9 | 4357 | return sprintf(buf, "Out of memory\n"); |
bbd7d57b | 4358 | } |
88a420e4 CL |
4359 | /* Push back cpu slabs */ |
4360 | flush_all(s); | |
4361 | ||
fa45dc25 | 4362 | for_each_kmem_cache_node(s, node, n) { |
88a420e4 CL |
4363 | unsigned long flags; |
4364 | struct page *page; | |
4365 | ||
9e86943b | 4366 | if (!atomic_long_read(&n->nr_slabs)) |
88a420e4 CL |
4367 | continue; |
4368 | ||
4369 | spin_lock_irqsave(&n->list_lock, flags); | |
4370 | list_for_each_entry(page, &n->partial, lru) | |
bbd7d57b | 4371 | process_slab(&t, s, page, alloc, map); |
88a420e4 | 4372 | list_for_each_entry(page, &n->full, lru) |
bbd7d57b | 4373 | process_slab(&t, s, page, alloc, map); |
88a420e4 CL |
4374 | spin_unlock_irqrestore(&n->list_lock, flags); |
4375 | } | |
4376 | ||
4377 | for (i = 0; i < t.count; i++) { | |
45edfa58 | 4378 | struct location *l = &t.loc[i]; |
88a420e4 | 4379 | |
9c246247 | 4380 | if (len > PAGE_SIZE - KSYM_SYMBOL_LEN - 100) |
88a420e4 | 4381 | break; |
e374d483 | 4382 | len += sprintf(buf + len, "%7ld ", l->count); |
45edfa58 CL |
4383 | |
4384 | if (l->addr) | |
62c70bce | 4385 | len += sprintf(buf + len, "%pS", (void *)l->addr); |
88a420e4 | 4386 | else |
e374d483 | 4387 | len += sprintf(buf + len, "<not-available>"); |
45edfa58 CL |
4388 | |
4389 | if (l->sum_time != l->min_time) { | |
e374d483 | 4390 | len += sprintf(buf + len, " age=%ld/%ld/%ld", |
f8bd2258 RZ |
4391 | l->min_time, |
4392 | (long)div_u64(l->sum_time, l->count), | |
4393 | l->max_time); | |
45edfa58 | 4394 | } else |
e374d483 | 4395 | len += sprintf(buf + len, " age=%ld", |
45edfa58 CL |
4396 | l->min_time); |
4397 | ||
4398 | if (l->min_pid != l->max_pid) | |
e374d483 | 4399 | len += sprintf(buf + len, " pid=%ld-%ld", |
45edfa58 CL |
4400 | l->min_pid, l->max_pid); |
4401 | else | |
e374d483 | 4402 | len += sprintf(buf + len, " pid=%ld", |
45edfa58 CL |
4403 | l->min_pid); |
4404 | ||
174596a0 RR |
4405 | if (num_online_cpus() > 1 && |
4406 | !cpumask_empty(to_cpumask(l->cpus)) && | |
5024c1d7 TH |
4407 | len < PAGE_SIZE - 60) |
4408 | len += scnprintf(buf + len, PAGE_SIZE - len - 50, | |
4409 | " cpus=%*pbl", | |
4410 | cpumask_pr_args(to_cpumask(l->cpus))); | |
45edfa58 | 4411 | |
62bc62a8 | 4412 | if (nr_online_nodes > 1 && !nodes_empty(l->nodes) && |
5024c1d7 TH |
4413 | len < PAGE_SIZE - 60) |
4414 | len += scnprintf(buf + len, PAGE_SIZE - len - 50, | |
4415 | " nodes=%*pbl", | |
4416 | nodemask_pr_args(&l->nodes)); | |
45edfa58 | 4417 | |
e374d483 | 4418 | len += sprintf(buf + len, "\n"); |
88a420e4 CL |
4419 | } |
4420 | ||
4421 | free_loc_track(&t); | |
bbd7d57b | 4422 | kfree(map); |
88a420e4 | 4423 | if (!t.count) |
e374d483 HH |
4424 | len += sprintf(buf, "No data\n"); |
4425 | return len; | |
88a420e4 | 4426 | } |
ab4d5ed5 | 4427 | #endif |
88a420e4 | 4428 | |
a5a84755 | 4429 | #ifdef SLUB_RESILIENCY_TEST |
c07b8183 | 4430 | static void __init resiliency_test(void) |
a5a84755 CL |
4431 | { |
4432 | u8 *p; | |
4433 | ||
95a05b42 | 4434 | BUILD_BUG_ON(KMALLOC_MIN_SIZE > 16 || KMALLOC_SHIFT_HIGH < 10); |
a5a84755 | 4435 | |
f9f58285 FF |
4436 | pr_err("SLUB resiliency testing\n"); |
4437 | pr_err("-----------------------\n"); | |
4438 | pr_err("A. Corruption after allocation\n"); | |
a5a84755 CL |
4439 | |
4440 | p = kzalloc(16, GFP_KERNEL); | |
4441 | p[16] = 0x12; | |
f9f58285 FF |
4442 | pr_err("\n1. kmalloc-16: Clobber Redzone/next pointer 0x12->0x%p\n\n", |
4443 | p + 16); | |
a5a84755 CL |
4444 | |
4445 | validate_slab_cache(kmalloc_caches[4]); | |
4446 | ||
4447 | /* Hmmm... The next two are dangerous */ | |
4448 | p = kzalloc(32, GFP_KERNEL); | |
4449 | p[32 + sizeof(void *)] = 0x34; | |
f9f58285 FF |
4450 | pr_err("\n2. kmalloc-32: Clobber next pointer/next slab 0x34 -> -0x%p\n", |
4451 | p); | |
4452 | pr_err("If allocated object is overwritten then not detectable\n\n"); | |
a5a84755 CL |
4453 | |
4454 | validate_slab_cache(kmalloc_caches[5]); | |
4455 | p = kzalloc(64, GFP_KERNEL); | |
4456 | p += 64 + (get_cycles() & 0xff) * sizeof(void *); | |
4457 | *p = 0x56; | |
f9f58285 FF |
4458 | pr_err("\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n", |
4459 | p); | |
4460 | pr_err("If allocated object is overwritten then not detectable\n\n"); | |
a5a84755 CL |
4461 | validate_slab_cache(kmalloc_caches[6]); |
4462 | ||
f9f58285 | 4463 | pr_err("\nB. Corruption after free\n"); |
a5a84755 CL |
4464 | p = kzalloc(128, GFP_KERNEL); |
4465 | kfree(p); | |
4466 | *p = 0x78; | |
f9f58285 | 4467 | pr_err("1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p); |
a5a84755 CL |
4468 | validate_slab_cache(kmalloc_caches[7]); |
4469 | ||
4470 | p = kzalloc(256, GFP_KERNEL); | |
4471 | kfree(p); | |
4472 | p[50] = 0x9a; | |
f9f58285 | 4473 | pr_err("\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n", p); |
a5a84755 CL |
4474 | validate_slab_cache(kmalloc_caches[8]); |
4475 | ||
4476 | p = kzalloc(512, GFP_KERNEL); | |
4477 | kfree(p); | |
4478 | p[512] = 0xab; | |
f9f58285 | 4479 | pr_err("\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p); |
a5a84755 CL |
4480 | validate_slab_cache(kmalloc_caches[9]); |
4481 | } | |
4482 | #else | |
4483 | #ifdef CONFIG_SYSFS | |
4484 | static void resiliency_test(void) {}; | |
4485 | #endif | |
4486 | #endif | |
4487 | ||
ab4d5ed5 | 4488 | #ifdef CONFIG_SYSFS |
81819f0f | 4489 | enum slab_stat_type { |
205ab99d CL |
4490 | SL_ALL, /* All slabs */ |
4491 | SL_PARTIAL, /* Only partially allocated slabs */ | |
4492 | SL_CPU, /* Only slabs used for cpu caches */ | |
4493 | SL_OBJECTS, /* Determine allocated objects not slabs */ | |
4494 | SL_TOTAL /* Determine object capacity not slabs */ | |
81819f0f CL |
4495 | }; |
4496 | ||
205ab99d | 4497 | #define SO_ALL (1 << SL_ALL) |
81819f0f CL |
4498 | #define SO_PARTIAL (1 << SL_PARTIAL) |
4499 | #define SO_CPU (1 << SL_CPU) | |
4500 | #define SO_OBJECTS (1 << SL_OBJECTS) | |
205ab99d | 4501 | #define SO_TOTAL (1 << SL_TOTAL) |
81819f0f | 4502 | |
62e5c4b4 CG |
4503 | static ssize_t show_slab_objects(struct kmem_cache *s, |
4504 | char *buf, unsigned long flags) | |
81819f0f CL |
4505 | { |
4506 | unsigned long total = 0; | |
81819f0f CL |
4507 | int node; |
4508 | int x; | |
4509 | unsigned long *nodes; | |
81819f0f | 4510 | |
e35e1a97 | 4511 | nodes = kzalloc(sizeof(unsigned long) * nr_node_ids, GFP_KERNEL); |
62e5c4b4 CG |
4512 | if (!nodes) |
4513 | return -ENOMEM; | |
81819f0f | 4514 | |
205ab99d CL |
4515 | if (flags & SO_CPU) { |
4516 | int cpu; | |
81819f0f | 4517 | |
205ab99d | 4518 | for_each_possible_cpu(cpu) { |
d0e0ac97 CG |
4519 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, |
4520 | cpu); | |
ec3ab083 | 4521 | int node; |
49e22585 | 4522 | struct page *page; |
dfb4f096 | 4523 | |
4db0c3c2 | 4524 | page = READ_ONCE(c->page); |
ec3ab083 CL |
4525 | if (!page) |
4526 | continue; | |
205ab99d | 4527 | |
ec3ab083 CL |
4528 | node = page_to_nid(page); |
4529 | if (flags & SO_TOTAL) | |
4530 | x = page->objects; | |
4531 | else if (flags & SO_OBJECTS) | |
4532 | x = page->inuse; | |
4533 | else | |
4534 | x = 1; | |
49e22585 | 4535 | |
ec3ab083 CL |
4536 | total += x; |
4537 | nodes[node] += x; | |
4538 | ||
4db0c3c2 | 4539 | page = READ_ONCE(c->partial); |
49e22585 | 4540 | if (page) { |
8afb1474 LZ |
4541 | node = page_to_nid(page); |
4542 | if (flags & SO_TOTAL) | |
4543 | WARN_ON_ONCE(1); | |
4544 | else if (flags & SO_OBJECTS) | |
4545 | WARN_ON_ONCE(1); | |
4546 | else | |
4547 | x = page->pages; | |
bc6697d8 ED |
4548 | total += x; |
4549 | nodes[node] += x; | |
49e22585 | 4550 | } |
81819f0f CL |
4551 | } |
4552 | } | |
4553 | ||
bfc8c901 | 4554 | get_online_mems(); |
ab4d5ed5 | 4555 | #ifdef CONFIG_SLUB_DEBUG |
205ab99d | 4556 | if (flags & SO_ALL) { |
fa45dc25 CL |
4557 | struct kmem_cache_node *n; |
4558 | ||
4559 | for_each_kmem_cache_node(s, node, n) { | |
205ab99d | 4560 | |
d0e0ac97 CG |
4561 | if (flags & SO_TOTAL) |
4562 | x = atomic_long_read(&n->total_objects); | |
4563 | else if (flags & SO_OBJECTS) | |
4564 | x = atomic_long_read(&n->total_objects) - | |
4565 | count_partial(n, count_free); | |
81819f0f | 4566 | else |
205ab99d | 4567 | x = atomic_long_read(&n->nr_slabs); |
81819f0f CL |
4568 | total += x; |
4569 | nodes[node] += x; | |
4570 | } | |
4571 | ||
ab4d5ed5 CL |
4572 | } else |
4573 | #endif | |
4574 | if (flags & SO_PARTIAL) { | |
fa45dc25 | 4575 | struct kmem_cache_node *n; |
81819f0f | 4576 | |
fa45dc25 | 4577 | for_each_kmem_cache_node(s, node, n) { |
205ab99d CL |
4578 | if (flags & SO_TOTAL) |
4579 | x = count_partial(n, count_total); | |
4580 | else if (flags & SO_OBJECTS) | |
4581 | x = count_partial(n, count_inuse); | |
81819f0f | 4582 | else |
205ab99d | 4583 | x = n->nr_partial; |
81819f0f CL |
4584 | total += x; |
4585 | nodes[node] += x; | |
4586 | } | |
4587 | } | |
81819f0f CL |
4588 | x = sprintf(buf, "%lu", total); |
4589 | #ifdef CONFIG_NUMA | |
fa45dc25 | 4590 | for (node = 0; node < nr_node_ids; node++) |
81819f0f CL |
4591 | if (nodes[node]) |
4592 | x += sprintf(buf + x, " N%d=%lu", | |
4593 | node, nodes[node]); | |
4594 | #endif | |
bfc8c901 | 4595 | put_online_mems(); |
81819f0f CL |
4596 | kfree(nodes); |
4597 | return x + sprintf(buf + x, "\n"); | |
4598 | } | |
4599 | ||
ab4d5ed5 | 4600 | #ifdef CONFIG_SLUB_DEBUG |
81819f0f CL |
4601 | static int any_slab_objects(struct kmem_cache *s) |
4602 | { | |
4603 | int node; | |
fa45dc25 | 4604 | struct kmem_cache_node *n; |
81819f0f | 4605 | |
fa45dc25 | 4606 | for_each_kmem_cache_node(s, node, n) |
4ea33e2d | 4607 | if (atomic_long_read(&n->total_objects)) |
81819f0f | 4608 | return 1; |
fa45dc25 | 4609 | |
81819f0f CL |
4610 | return 0; |
4611 | } | |
ab4d5ed5 | 4612 | #endif |
81819f0f CL |
4613 | |
4614 | #define to_slab_attr(n) container_of(n, struct slab_attribute, attr) | |
497888cf | 4615 | #define to_slab(n) container_of(n, struct kmem_cache, kobj) |
81819f0f CL |
4616 | |
4617 | struct slab_attribute { | |
4618 | struct attribute attr; | |
4619 | ssize_t (*show)(struct kmem_cache *s, char *buf); | |
4620 | ssize_t (*store)(struct kmem_cache *s, const char *x, size_t count); | |
4621 | }; | |
4622 | ||
4623 | #define SLAB_ATTR_RO(_name) \ | |
ab067e99 VK |
4624 | static struct slab_attribute _name##_attr = \ |
4625 | __ATTR(_name, 0400, _name##_show, NULL) | |
81819f0f CL |
4626 | |
4627 | #define SLAB_ATTR(_name) \ | |
4628 | static struct slab_attribute _name##_attr = \ | |
ab067e99 | 4629 | __ATTR(_name, 0600, _name##_show, _name##_store) |
81819f0f | 4630 | |
81819f0f CL |
4631 | static ssize_t slab_size_show(struct kmem_cache *s, char *buf) |
4632 | { | |
4633 | return sprintf(buf, "%d\n", s->size); | |
4634 | } | |
4635 | SLAB_ATTR_RO(slab_size); | |
4636 | ||
4637 | static ssize_t align_show(struct kmem_cache *s, char *buf) | |
4638 | { | |
4639 | return sprintf(buf, "%d\n", s->align); | |
4640 | } | |
4641 | SLAB_ATTR_RO(align); | |
4642 | ||
4643 | static ssize_t object_size_show(struct kmem_cache *s, char *buf) | |
4644 | { | |
3b0efdfa | 4645 | return sprintf(buf, "%d\n", s->object_size); |
81819f0f CL |
4646 | } |
4647 | SLAB_ATTR_RO(object_size); | |
4648 | ||
4649 | static ssize_t objs_per_slab_show(struct kmem_cache *s, char *buf) | |
4650 | { | |
834f3d11 | 4651 | return sprintf(buf, "%d\n", oo_objects(s->oo)); |
81819f0f CL |
4652 | } |
4653 | SLAB_ATTR_RO(objs_per_slab); | |
4654 | ||
06b285dc CL |
4655 | static ssize_t order_store(struct kmem_cache *s, |
4656 | const char *buf, size_t length) | |
4657 | { | |
0121c619 CL |
4658 | unsigned long order; |
4659 | int err; | |
4660 | ||
3dbb95f7 | 4661 | err = kstrtoul(buf, 10, &order); |
0121c619 CL |
4662 | if (err) |
4663 | return err; | |
06b285dc CL |
4664 | |
4665 | if (order > slub_max_order || order < slub_min_order) | |
4666 | return -EINVAL; | |
4667 | ||
4668 | calculate_sizes(s, order); | |
4669 | return length; | |
4670 | } | |
4671 | ||
81819f0f CL |
4672 | static ssize_t order_show(struct kmem_cache *s, char *buf) |
4673 | { | |
834f3d11 | 4674 | return sprintf(buf, "%d\n", oo_order(s->oo)); |
81819f0f | 4675 | } |
06b285dc | 4676 | SLAB_ATTR(order); |
81819f0f | 4677 | |
73d342b1 DR |
4678 | static ssize_t min_partial_show(struct kmem_cache *s, char *buf) |
4679 | { | |
4680 | return sprintf(buf, "%lu\n", s->min_partial); | |
4681 | } | |
4682 | ||
4683 | static ssize_t min_partial_store(struct kmem_cache *s, const char *buf, | |
4684 | size_t length) | |
4685 | { | |
4686 | unsigned long min; | |
4687 | int err; | |
4688 | ||
3dbb95f7 | 4689 | err = kstrtoul(buf, 10, &min); |
73d342b1 DR |
4690 | if (err) |
4691 | return err; | |
4692 | ||
c0bdb232 | 4693 | set_min_partial(s, min); |
73d342b1 DR |
4694 | return length; |
4695 | } | |
4696 | SLAB_ATTR(min_partial); | |
4697 | ||
49e22585 CL |
4698 | static ssize_t cpu_partial_show(struct kmem_cache *s, char *buf) |
4699 | { | |
4700 | return sprintf(buf, "%u\n", s->cpu_partial); | |
4701 | } | |
4702 | ||
4703 | static ssize_t cpu_partial_store(struct kmem_cache *s, const char *buf, | |
4704 | size_t length) | |
4705 | { | |
4706 | unsigned long objects; | |
4707 | int err; | |
4708 | ||
3dbb95f7 | 4709 | err = kstrtoul(buf, 10, &objects); |
49e22585 CL |
4710 | if (err) |
4711 | return err; | |
345c905d | 4712 | if (objects && !kmem_cache_has_cpu_partial(s)) |
74ee4ef1 | 4713 | return -EINVAL; |
49e22585 CL |
4714 | |
4715 | s->cpu_partial = objects; | |
4716 | flush_all(s); | |
4717 | return length; | |
4718 | } | |
4719 | SLAB_ATTR(cpu_partial); | |
4720 | ||
81819f0f CL |
4721 | static ssize_t ctor_show(struct kmem_cache *s, char *buf) |
4722 | { | |
62c70bce JP |
4723 | if (!s->ctor) |
4724 | return 0; | |
4725 | return sprintf(buf, "%pS\n", s->ctor); | |
81819f0f CL |
4726 | } |
4727 | SLAB_ATTR_RO(ctor); | |
4728 | ||
81819f0f CL |
4729 | static ssize_t aliases_show(struct kmem_cache *s, char *buf) |
4730 | { | |
4307c14f | 4731 | return sprintf(buf, "%d\n", s->refcount < 0 ? 0 : s->refcount - 1); |
81819f0f CL |
4732 | } |
4733 | SLAB_ATTR_RO(aliases); | |
4734 | ||
81819f0f CL |
4735 | static ssize_t partial_show(struct kmem_cache *s, char *buf) |
4736 | { | |
d9acf4b7 | 4737 | return show_slab_objects(s, buf, SO_PARTIAL); |
81819f0f CL |
4738 | } |
4739 | SLAB_ATTR_RO(partial); | |
4740 | ||
4741 | static ssize_t cpu_slabs_show(struct kmem_cache *s, char *buf) | |
4742 | { | |
d9acf4b7 | 4743 | return show_slab_objects(s, buf, SO_CPU); |
81819f0f CL |
4744 | } |
4745 | SLAB_ATTR_RO(cpu_slabs); | |
4746 | ||
4747 | static ssize_t objects_show(struct kmem_cache *s, char *buf) | |
4748 | { | |
205ab99d | 4749 | return show_slab_objects(s, buf, SO_ALL|SO_OBJECTS); |
81819f0f CL |
4750 | } |
4751 | SLAB_ATTR_RO(objects); | |
4752 | ||
205ab99d CL |
4753 | static ssize_t objects_partial_show(struct kmem_cache *s, char *buf) |
4754 | { | |
4755 | return show_slab_objects(s, buf, SO_PARTIAL|SO_OBJECTS); | |
4756 | } | |
4757 | SLAB_ATTR_RO(objects_partial); | |
4758 | ||
49e22585 CL |
4759 | static ssize_t slabs_cpu_partial_show(struct kmem_cache *s, char *buf) |
4760 | { | |
4761 | int objects = 0; | |
4762 | int pages = 0; | |
4763 | int cpu; | |
4764 | int len; | |
4765 | ||
4766 | for_each_online_cpu(cpu) { | |
4767 | struct page *page = per_cpu_ptr(s->cpu_slab, cpu)->partial; | |
4768 | ||
4769 | if (page) { | |
4770 | pages += page->pages; | |
4771 | objects += page->pobjects; | |
4772 | } | |
4773 | } | |
4774 | ||
4775 | len = sprintf(buf, "%d(%d)", objects, pages); | |
4776 | ||
4777 | #ifdef CONFIG_SMP | |
4778 | for_each_online_cpu(cpu) { | |
4779 | struct page *page = per_cpu_ptr(s->cpu_slab, cpu) ->partial; | |
4780 | ||
4781 | if (page && len < PAGE_SIZE - 20) | |
4782 | len += sprintf(buf + len, " C%d=%d(%d)", cpu, | |
4783 | page->pobjects, page->pages); | |
4784 | } | |
4785 | #endif | |
4786 | return len + sprintf(buf + len, "\n"); | |
4787 | } | |
4788 | SLAB_ATTR_RO(slabs_cpu_partial); | |
4789 | ||
a5a84755 CL |
4790 | static ssize_t reclaim_account_show(struct kmem_cache *s, char *buf) |
4791 | { | |
4792 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_RECLAIM_ACCOUNT)); | |
4793 | } | |
4794 | ||
4795 | static ssize_t reclaim_account_store(struct kmem_cache *s, | |
4796 | const char *buf, size_t length) | |
4797 | { | |
4798 | s->flags &= ~SLAB_RECLAIM_ACCOUNT; | |
4799 | if (buf[0] == '1') | |
4800 | s->flags |= SLAB_RECLAIM_ACCOUNT; | |
4801 | return length; | |
4802 | } | |
4803 | SLAB_ATTR(reclaim_account); | |
4804 | ||
4805 | static ssize_t hwcache_align_show(struct kmem_cache *s, char *buf) | |
4806 | { | |
4807 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_HWCACHE_ALIGN)); | |
4808 | } | |
4809 | SLAB_ATTR_RO(hwcache_align); | |
4810 | ||
4811 | #ifdef CONFIG_ZONE_DMA | |
4812 | static ssize_t cache_dma_show(struct kmem_cache *s, char *buf) | |
4813 | { | |
4814 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_CACHE_DMA)); | |
4815 | } | |
4816 | SLAB_ATTR_RO(cache_dma); | |
4817 | #endif | |
4818 | ||
4819 | static ssize_t destroy_by_rcu_show(struct kmem_cache *s, char *buf) | |
4820 | { | |
4821 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_DESTROY_BY_RCU)); | |
4822 | } | |
4823 | SLAB_ATTR_RO(destroy_by_rcu); | |
4824 | ||
ab9a0f19 LJ |
4825 | static ssize_t reserved_show(struct kmem_cache *s, char *buf) |
4826 | { | |
4827 | return sprintf(buf, "%d\n", s->reserved); | |
4828 | } | |
4829 | SLAB_ATTR_RO(reserved); | |
4830 | ||
ab4d5ed5 | 4831 | #ifdef CONFIG_SLUB_DEBUG |
a5a84755 CL |
4832 | static ssize_t slabs_show(struct kmem_cache *s, char *buf) |
4833 | { | |
4834 | return show_slab_objects(s, buf, SO_ALL); | |
4835 | } | |
4836 | SLAB_ATTR_RO(slabs); | |
4837 | ||
205ab99d CL |
4838 | static ssize_t total_objects_show(struct kmem_cache *s, char *buf) |
4839 | { | |
4840 | return show_slab_objects(s, buf, SO_ALL|SO_TOTAL); | |
4841 | } | |
4842 | SLAB_ATTR_RO(total_objects); | |
4843 | ||
81819f0f CL |
4844 | static ssize_t sanity_checks_show(struct kmem_cache *s, char *buf) |
4845 | { | |
becfda68 | 4846 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_CONSISTENCY_CHECKS)); |
81819f0f CL |
4847 | } |
4848 | ||
4849 | static ssize_t sanity_checks_store(struct kmem_cache *s, | |
4850 | const char *buf, size_t length) | |
4851 | { | |
becfda68 | 4852 | s->flags &= ~SLAB_CONSISTENCY_CHECKS; |
b789ef51 CL |
4853 | if (buf[0] == '1') { |
4854 | s->flags &= ~__CMPXCHG_DOUBLE; | |
becfda68 | 4855 | s->flags |= SLAB_CONSISTENCY_CHECKS; |
b789ef51 | 4856 | } |
81819f0f CL |
4857 | return length; |
4858 | } | |
4859 | SLAB_ATTR(sanity_checks); | |
4860 | ||
4861 | static ssize_t trace_show(struct kmem_cache *s, char *buf) | |
4862 | { | |
4863 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_TRACE)); | |
4864 | } | |
4865 | ||
4866 | static ssize_t trace_store(struct kmem_cache *s, const char *buf, | |
4867 | size_t length) | |
4868 | { | |
c9e16131 CL |
4869 | /* |
4870 | * Tracing a merged cache is going to give confusing results | |
4871 | * as well as cause other issues like converting a mergeable | |
4872 | * cache into an umergeable one. | |
4873 | */ | |
4874 | if (s->refcount > 1) | |
4875 | return -EINVAL; | |
4876 | ||
81819f0f | 4877 | s->flags &= ~SLAB_TRACE; |
b789ef51 CL |
4878 | if (buf[0] == '1') { |
4879 | s->flags &= ~__CMPXCHG_DOUBLE; | |
81819f0f | 4880 | s->flags |= SLAB_TRACE; |
b789ef51 | 4881 | } |
81819f0f CL |
4882 | return length; |
4883 | } | |
4884 | SLAB_ATTR(trace); | |
4885 | ||
81819f0f CL |
4886 | static ssize_t red_zone_show(struct kmem_cache *s, char *buf) |
4887 | { | |
4888 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_RED_ZONE)); | |
4889 | } | |
4890 | ||
4891 | static ssize_t red_zone_store(struct kmem_cache *s, | |
4892 | const char *buf, size_t length) | |
4893 | { | |
4894 | if (any_slab_objects(s)) | |
4895 | return -EBUSY; | |
4896 | ||
4897 | s->flags &= ~SLAB_RED_ZONE; | |
b789ef51 | 4898 | if (buf[0] == '1') { |
81819f0f | 4899 | s->flags |= SLAB_RED_ZONE; |
b789ef51 | 4900 | } |
06b285dc | 4901 | calculate_sizes(s, -1); |
81819f0f CL |
4902 | return length; |
4903 | } | |
4904 | SLAB_ATTR(red_zone); | |
4905 | ||
4906 | static ssize_t poison_show(struct kmem_cache *s, char *buf) | |
4907 | { | |
4908 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_POISON)); | |
4909 | } | |
4910 | ||
4911 | static ssize_t poison_store(struct kmem_cache *s, | |
4912 | const char *buf, size_t length) | |
4913 | { | |
4914 | if (any_slab_objects(s)) | |
4915 | return -EBUSY; | |
4916 | ||
4917 | s->flags &= ~SLAB_POISON; | |
b789ef51 | 4918 | if (buf[0] == '1') { |
81819f0f | 4919 | s->flags |= SLAB_POISON; |
b789ef51 | 4920 | } |
06b285dc | 4921 | calculate_sizes(s, -1); |
81819f0f CL |
4922 | return length; |
4923 | } | |
4924 | SLAB_ATTR(poison); | |
4925 | ||
4926 | static ssize_t store_user_show(struct kmem_cache *s, char *buf) | |
4927 | { | |
4928 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_STORE_USER)); | |
4929 | } | |
4930 | ||
4931 | static ssize_t store_user_store(struct kmem_cache *s, | |
4932 | const char *buf, size_t length) | |
4933 | { | |
4934 | if (any_slab_objects(s)) | |
4935 | return -EBUSY; | |
4936 | ||
4937 | s->flags &= ~SLAB_STORE_USER; | |
b789ef51 CL |
4938 | if (buf[0] == '1') { |
4939 | s->flags &= ~__CMPXCHG_DOUBLE; | |
81819f0f | 4940 | s->flags |= SLAB_STORE_USER; |
b789ef51 | 4941 | } |
06b285dc | 4942 | calculate_sizes(s, -1); |
81819f0f CL |
4943 | return length; |
4944 | } | |
4945 | SLAB_ATTR(store_user); | |
4946 | ||
53e15af0 CL |
4947 | static ssize_t validate_show(struct kmem_cache *s, char *buf) |
4948 | { | |
4949 | return 0; | |
4950 | } | |
4951 | ||
4952 | static ssize_t validate_store(struct kmem_cache *s, | |
4953 | const char *buf, size_t length) | |
4954 | { | |
434e245d CL |
4955 | int ret = -EINVAL; |
4956 | ||
4957 | if (buf[0] == '1') { | |
4958 | ret = validate_slab_cache(s); | |
4959 | if (ret >= 0) | |
4960 | ret = length; | |
4961 | } | |
4962 | return ret; | |
53e15af0 CL |
4963 | } |
4964 | SLAB_ATTR(validate); | |
a5a84755 CL |
4965 | |
4966 | static ssize_t alloc_calls_show(struct kmem_cache *s, char *buf) | |
4967 | { | |
4968 | if (!(s->flags & SLAB_STORE_USER)) | |
4969 | return -ENOSYS; | |
4970 | return list_locations(s, buf, TRACK_ALLOC); | |
4971 | } | |
4972 | SLAB_ATTR_RO(alloc_calls); | |
4973 | ||
4974 | static ssize_t free_calls_show(struct kmem_cache *s, char *buf) | |
4975 | { | |
4976 | if (!(s->flags & SLAB_STORE_USER)) | |
4977 | return -ENOSYS; | |
4978 | return list_locations(s, buf, TRACK_FREE); | |
4979 | } | |
4980 | SLAB_ATTR_RO(free_calls); | |
4981 | #endif /* CONFIG_SLUB_DEBUG */ | |
4982 | ||
4983 | #ifdef CONFIG_FAILSLAB | |
4984 | static ssize_t failslab_show(struct kmem_cache *s, char *buf) | |
4985 | { | |
4986 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_FAILSLAB)); | |
4987 | } | |
4988 | ||
4989 | static ssize_t failslab_store(struct kmem_cache *s, const char *buf, | |
4990 | size_t length) | |
4991 | { | |
c9e16131 CL |
4992 | if (s->refcount > 1) |
4993 | return -EINVAL; | |
4994 | ||
a5a84755 CL |
4995 | s->flags &= ~SLAB_FAILSLAB; |
4996 | if (buf[0] == '1') | |
4997 | s->flags |= SLAB_FAILSLAB; | |
4998 | return length; | |
4999 | } | |
5000 | SLAB_ATTR(failslab); | |
ab4d5ed5 | 5001 | #endif |
53e15af0 | 5002 | |
2086d26a CL |
5003 | static ssize_t shrink_show(struct kmem_cache *s, char *buf) |
5004 | { | |
5005 | return 0; | |
5006 | } | |
5007 | ||
5008 | static ssize_t shrink_store(struct kmem_cache *s, | |
5009 | const char *buf, size_t length) | |
5010 | { | |
832f37f5 VD |
5011 | if (buf[0] == '1') |
5012 | kmem_cache_shrink(s); | |
5013 | else | |
2086d26a CL |
5014 | return -EINVAL; |
5015 | return length; | |
5016 | } | |
5017 | SLAB_ATTR(shrink); | |
5018 | ||
81819f0f | 5019 | #ifdef CONFIG_NUMA |
9824601e | 5020 | static ssize_t remote_node_defrag_ratio_show(struct kmem_cache *s, char *buf) |
81819f0f | 5021 | { |
9824601e | 5022 | return sprintf(buf, "%d\n", s->remote_node_defrag_ratio / 10); |
81819f0f CL |
5023 | } |
5024 | ||
9824601e | 5025 | static ssize_t remote_node_defrag_ratio_store(struct kmem_cache *s, |
81819f0f CL |
5026 | const char *buf, size_t length) |
5027 | { | |
0121c619 CL |
5028 | unsigned long ratio; |
5029 | int err; | |
5030 | ||
3dbb95f7 | 5031 | err = kstrtoul(buf, 10, &ratio); |
0121c619 CL |
5032 | if (err) |
5033 | return err; | |
5034 | ||
e2cb96b7 | 5035 | if (ratio <= 100) |
0121c619 | 5036 | s->remote_node_defrag_ratio = ratio * 10; |
81819f0f | 5037 | |
81819f0f CL |
5038 | return length; |
5039 | } | |
9824601e | 5040 | SLAB_ATTR(remote_node_defrag_ratio); |
81819f0f CL |
5041 | #endif |
5042 | ||
8ff12cfc | 5043 | #ifdef CONFIG_SLUB_STATS |
8ff12cfc CL |
5044 | static int show_stat(struct kmem_cache *s, char *buf, enum stat_item si) |
5045 | { | |
5046 | unsigned long sum = 0; | |
5047 | int cpu; | |
5048 | int len; | |
5049 | int *data = kmalloc(nr_cpu_ids * sizeof(int), GFP_KERNEL); | |
5050 | ||
5051 | if (!data) | |
5052 | return -ENOMEM; | |
5053 | ||
5054 | for_each_online_cpu(cpu) { | |
9dfc6e68 | 5055 | unsigned x = per_cpu_ptr(s->cpu_slab, cpu)->stat[si]; |
8ff12cfc CL |
5056 | |
5057 | data[cpu] = x; | |
5058 | sum += x; | |
5059 | } | |
5060 | ||
5061 | len = sprintf(buf, "%lu", sum); | |
5062 | ||
50ef37b9 | 5063 | #ifdef CONFIG_SMP |
8ff12cfc CL |
5064 | for_each_online_cpu(cpu) { |
5065 | if (data[cpu] && len < PAGE_SIZE - 20) | |
50ef37b9 | 5066 | len += sprintf(buf + len, " C%d=%u", cpu, data[cpu]); |
8ff12cfc | 5067 | } |
50ef37b9 | 5068 | #endif |
8ff12cfc CL |
5069 | kfree(data); |
5070 | return len + sprintf(buf + len, "\n"); | |
5071 | } | |
5072 | ||
78eb00cc DR |
5073 | static void clear_stat(struct kmem_cache *s, enum stat_item si) |
5074 | { | |
5075 | int cpu; | |
5076 | ||
5077 | for_each_online_cpu(cpu) | |
9dfc6e68 | 5078 | per_cpu_ptr(s->cpu_slab, cpu)->stat[si] = 0; |
78eb00cc DR |
5079 | } |
5080 | ||
8ff12cfc CL |
5081 | #define STAT_ATTR(si, text) \ |
5082 | static ssize_t text##_show(struct kmem_cache *s, char *buf) \ | |
5083 | { \ | |
5084 | return show_stat(s, buf, si); \ | |
5085 | } \ | |
78eb00cc DR |
5086 | static ssize_t text##_store(struct kmem_cache *s, \ |
5087 | const char *buf, size_t length) \ | |
5088 | { \ | |
5089 | if (buf[0] != '0') \ | |
5090 | return -EINVAL; \ | |
5091 | clear_stat(s, si); \ | |
5092 | return length; \ | |
5093 | } \ | |
5094 | SLAB_ATTR(text); \ | |
8ff12cfc CL |
5095 | |
5096 | STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath); | |
5097 | STAT_ATTR(ALLOC_SLOWPATH, alloc_slowpath); | |
5098 | STAT_ATTR(FREE_FASTPATH, free_fastpath); | |
5099 | STAT_ATTR(FREE_SLOWPATH, free_slowpath); | |
5100 | STAT_ATTR(FREE_FROZEN, free_frozen); | |
5101 | STAT_ATTR(FREE_ADD_PARTIAL, free_add_partial); | |
5102 | STAT_ATTR(FREE_REMOVE_PARTIAL, free_remove_partial); | |
5103 | STAT_ATTR(ALLOC_FROM_PARTIAL, alloc_from_partial); | |
5104 | STAT_ATTR(ALLOC_SLAB, alloc_slab); | |
5105 | STAT_ATTR(ALLOC_REFILL, alloc_refill); | |
e36a2652 | 5106 | STAT_ATTR(ALLOC_NODE_MISMATCH, alloc_node_mismatch); |
8ff12cfc CL |
5107 | STAT_ATTR(FREE_SLAB, free_slab); |
5108 | STAT_ATTR(CPUSLAB_FLUSH, cpuslab_flush); | |
5109 | STAT_ATTR(DEACTIVATE_FULL, deactivate_full); | |
5110 | STAT_ATTR(DEACTIVATE_EMPTY, deactivate_empty); | |
5111 | STAT_ATTR(DEACTIVATE_TO_HEAD, deactivate_to_head); | |
5112 | STAT_ATTR(DEACTIVATE_TO_TAIL, deactivate_to_tail); | |
5113 | STAT_ATTR(DEACTIVATE_REMOTE_FREES, deactivate_remote_frees); | |
03e404af | 5114 | STAT_ATTR(DEACTIVATE_BYPASS, deactivate_bypass); |
65c3376a | 5115 | STAT_ATTR(ORDER_FALLBACK, order_fallback); |
b789ef51 CL |
5116 | STAT_ATTR(CMPXCHG_DOUBLE_CPU_FAIL, cmpxchg_double_cpu_fail); |
5117 | STAT_ATTR(CMPXCHG_DOUBLE_FAIL, cmpxchg_double_fail); | |
49e22585 CL |
5118 | STAT_ATTR(CPU_PARTIAL_ALLOC, cpu_partial_alloc); |
5119 | STAT_ATTR(CPU_PARTIAL_FREE, cpu_partial_free); | |
8028dcea AS |
5120 | STAT_ATTR(CPU_PARTIAL_NODE, cpu_partial_node); |
5121 | STAT_ATTR(CPU_PARTIAL_DRAIN, cpu_partial_drain); | |
8ff12cfc CL |
5122 | #endif |
5123 | ||
06428780 | 5124 | static struct attribute *slab_attrs[] = { |
81819f0f CL |
5125 | &slab_size_attr.attr, |
5126 | &object_size_attr.attr, | |
5127 | &objs_per_slab_attr.attr, | |
5128 | &order_attr.attr, | |
73d342b1 | 5129 | &min_partial_attr.attr, |
49e22585 | 5130 | &cpu_partial_attr.attr, |
81819f0f | 5131 | &objects_attr.attr, |
205ab99d | 5132 | &objects_partial_attr.attr, |
81819f0f CL |
5133 | &partial_attr.attr, |
5134 | &cpu_slabs_attr.attr, | |
5135 | &ctor_attr.attr, | |
81819f0f CL |
5136 | &aliases_attr.attr, |
5137 | &align_attr.attr, | |
81819f0f CL |
5138 | &hwcache_align_attr.attr, |
5139 | &reclaim_account_attr.attr, | |
5140 | &destroy_by_rcu_attr.attr, | |
a5a84755 | 5141 | &shrink_attr.attr, |
ab9a0f19 | 5142 | &reserved_attr.attr, |
49e22585 | 5143 | &slabs_cpu_partial_attr.attr, |
ab4d5ed5 | 5144 | #ifdef CONFIG_SLUB_DEBUG |
a5a84755 CL |
5145 | &total_objects_attr.attr, |
5146 | &slabs_attr.attr, | |
5147 | &sanity_checks_attr.attr, | |
5148 | &trace_attr.attr, | |
81819f0f CL |
5149 | &red_zone_attr.attr, |
5150 | &poison_attr.attr, | |
5151 | &store_user_attr.attr, | |
53e15af0 | 5152 | &validate_attr.attr, |
88a420e4 CL |
5153 | &alloc_calls_attr.attr, |
5154 | &free_calls_attr.attr, | |
ab4d5ed5 | 5155 | #endif |
81819f0f CL |
5156 | #ifdef CONFIG_ZONE_DMA |
5157 | &cache_dma_attr.attr, | |
5158 | #endif | |
5159 | #ifdef CONFIG_NUMA | |
9824601e | 5160 | &remote_node_defrag_ratio_attr.attr, |
8ff12cfc CL |
5161 | #endif |
5162 | #ifdef CONFIG_SLUB_STATS | |
5163 | &alloc_fastpath_attr.attr, | |
5164 | &alloc_slowpath_attr.attr, | |
5165 | &free_fastpath_attr.attr, | |
5166 | &free_slowpath_attr.attr, | |
5167 | &free_frozen_attr.attr, | |
5168 | &free_add_partial_attr.attr, | |
5169 | &free_remove_partial_attr.attr, | |
5170 | &alloc_from_partial_attr.attr, | |
5171 | &alloc_slab_attr.attr, | |
5172 | &alloc_refill_attr.attr, | |
e36a2652 | 5173 | &alloc_node_mismatch_attr.attr, |
8ff12cfc CL |
5174 | &free_slab_attr.attr, |
5175 | &cpuslab_flush_attr.attr, | |
5176 | &deactivate_full_attr.attr, | |
5177 | &deactivate_empty_attr.attr, | |
5178 | &deactivate_to_head_attr.attr, | |
5179 | &deactivate_to_tail_attr.attr, | |
5180 | &deactivate_remote_frees_attr.attr, | |
03e404af | 5181 | &deactivate_bypass_attr.attr, |
65c3376a | 5182 | &order_fallback_attr.attr, |
b789ef51 CL |
5183 | &cmpxchg_double_fail_attr.attr, |
5184 | &cmpxchg_double_cpu_fail_attr.attr, | |
49e22585 CL |
5185 | &cpu_partial_alloc_attr.attr, |
5186 | &cpu_partial_free_attr.attr, | |
8028dcea AS |
5187 | &cpu_partial_node_attr.attr, |
5188 | &cpu_partial_drain_attr.attr, | |
81819f0f | 5189 | #endif |
4c13dd3b DM |
5190 | #ifdef CONFIG_FAILSLAB |
5191 | &failslab_attr.attr, | |
5192 | #endif | |
5193 | ||
81819f0f CL |
5194 | NULL |
5195 | }; | |
5196 | ||
5197 | static struct attribute_group slab_attr_group = { | |
5198 | .attrs = slab_attrs, | |
5199 | }; | |
5200 | ||
5201 | static ssize_t slab_attr_show(struct kobject *kobj, | |
5202 | struct attribute *attr, | |
5203 | char *buf) | |
5204 | { | |
5205 | struct slab_attribute *attribute; | |
5206 | struct kmem_cache *s; | |
5207 | int err; | |
5208 | ||
5209 | attribute = to_slab_attr(attr); | |
5210 | s = to_slab(kobj); | |
5211 | ||
5212 | if (!attribute->show) | |
5213 | return -EIO; | |
5214 | ||
5215 | err = attribute->show(s, buf); | |
5216 | ||
5217 | return err; | |
5218 | } | |
5219 | ||
5220 | static ssize_t slab_attr_store(struct kobject *kobj, | |
5221 | struct attribute *attr, | |
5222 | const char *buf, size_t len) | |
5223 | { | |
5224 | struct slab_attribute *attribute; | |
5225 | struct kmem_cache *s; | |
5226 | int err; | |
5227 | ||
5228 | attribute = to_slab_attr(attr); | |
5229 | s = to_slab(kobj); | |
5230 | ||
5231 | if (!attribute->store) | |
5232 | return -EIO; | |
5233 | ||
5234 | err = attribute->store(s, buf, len); | |
127424c8 | 5235 | #ifdef CONFIG_MEMCG |
107dab5c | 5236 | if (slab_state >= FULL && err >= 0 && is_root_cache(s)) { |
426589f5 | 5237 | struct kmem_cache *c; |
81819f0f | 5238 | |
107dab5c GC |
5239 | mutex_lock(&slab_mutex); |
5240 | if (s->max_attr_size < len) | |
5241 | s->max_attr_size = len; | |
5242 | ||
ebe945c2 GC |
5243 | /* |
5244 | * This is a best effort propagation, so this function's return | |
5245 | * value will be determined by the parent cache only. This is | |
5246 | * basically because not all attributes will have a well | |
5247 | * defined semantics for rollbacks - most of the actions will | |
5248 | * have permanent effects. | |
5249 | * | |
5250 | * Returning the error value of any of the children that fail | |
5251 | * is not 100 % defined, in the sense that users seeing the | |
5252 | * error code won't be able to know anything about the state of | |
5253 | * the cache. | |
5254 | * | |
5255 | * Only returning the error code for the parent cache at least | |
5256 | * has well defined semantics. The cache being written to | |
5257 | * directly either failed or succeeded, in which case we loop | |
5258 | * through the descendants with best-effort propagation. | |
5259 | */ | |
426589f5 VD |
5260 | for_each_memcg_cache(c, s) |
5261 | attribute->store(c, buf, len); | |
107dab5c GC |
5262 | mutex_unlock(&slab_mutex); |
5263 | } | |
5264 | #endif | |
81819f0f CL |
5265 | return err; |
5266 | } | |
5267 | ||
107dab5c GC |
5268 | static void memcg_propagate_slab_attrs(struct kmem_cache *s) |
5269 | { | |
127424c8 | 5270 | #ifdef CONFIG_MEMCG |
107dab5c GC |
5271 | int i; |
5272 | char *buffer = NULL; | |
93030d83 | 5273 | struct kmem_cache *root_cache; |
107dab5c | 5274 | |
93030d83 | 5275 | if (is_root_cache(s)) |
107dab5c GC |
5276 | return; |
5277 | ||
f7ce3190 | 5278 | root_cache = s->memcg_params.root_cache; |
93030d83 | 5279 | |
107dab5c GC |
5280 | /* |
5281 | * This mean this cache had no attribute written. Therefore, no point | |
5282 | * in copying default values around | |
5283 | */ | |
93030d83 | 5284 | if (!root_cache->max_attr_size) |
107dab5c GC |
5285 | return; |
5286 | ||
5287 | for (i = 0; i < ARRAY_SIZE(slab_attrs); i++) { | |
5288 | char mbuf[64]; | |
5289 | char *buf; | |
5290 | struct slab_attribute *attr = to_slab_attr(slab_attrs[i]); | |
5291 | ||
5292 | if (!attr || !attr->store || !attr->show) | |
5293 | continue; | |
5294 | ||
5295 | /* | |
5296 | * It is really bad that we have to allocate here, so we will | |
5297 | * do it only as a fallback. If we actually allocate, though, | |
5298 | * we can just use the allocated buffer until the end. | |
5299 | * | |
5300 | * Most of the slub attributes will tend to be very small in | |
5301 | * size, but sysfs allows buffers up to a page, so they can | |
5302 | * theoretically happen. | |
5303 | */ | |
5304 | if (buffer) | |
5305 | buf = buffer; | |
93030d83 | 5306 | else if (root_cache->max_attr_size < ARRAY_SIZE(mbuf)) |
107dab5c GC |
5307 | buf = mbuf; |
5308 | else { | |
5309 | buffer = (char *) get_zeroed_page(GFP_KERNEL); | |
5310 | if (WARN_ON(!buffer)) | |
5311 | continue; | |
5312 | buf = buffer; | |
5313 | } | |
5314 | ||
93030d83 | 5315 | attr->show(root_cache, buf); |
107dab5c GC |
5316 | attr->store(s, buf, strlen(buf)); |
5317 | } | |
5318 | ||
5319 | if (buffer) | |
5320 | free_page((unsigned long)buffer); | |
5321 | #endif | |
5322 | } | |
5323 | ||
41a21285 CL |
5324 | static void kmem_cache_release(struct kobject *k) |
5325 | { | |
5326 | slab_kmem_cache_release(to_slab(k)); | |
5327 | } | |
5328 | ||
52cf25d0 | 5329 | static const struct sysfs_ops slab_sysfs_ops = { |
81819f0f CL |
5330 | .show = slab_attr_show, |
5331 | .store = slab_attr_store, | |
5332 | }; | |
5333 | ||
5334 | static struct kobj_type slab_ktype = { | |
5335 | .sysfs_ops = &slab_sysfs_ops, | |
41a21285 | 5336 | .release = kmem_cache_release, |
81819f0f CL |
5337 | }; |
5338 | ||
5339 | static int uevent_filter(struct kset *kset, struct kobject *kobj) | |
5340 | { | |
5341 | struct kobj_type *ktype = get_ktype(kobj); | |
5342 | ||
5343 | if (ktype == &slab_ktype) | |
5344 | return 1; | |
5345 | return 0; | |
5346 | } | |
5347 | ||
9cd43611 | 5348 | static const struct kset_uevent_ops slab_uevent_ops = { |
81819f0f CL |
5349 | .filter = uevent_filter, |
5350 | }; | |
5351 | ||
27c3a314 | 5352 | static struct kset *slab_kset; |
81819f0f | 5353 | |
9a41707b VD |
5354 | static inline struct kset *cache_kset(struct kmem_cache *s) |
5355 | { | |
127424c8 | 5356 | #ifdef CONFIG_MEMCG |
9a41707b | 5357 | if (!is_root_cache(s)) |
f7ce3190 | 5358 | return s->memcg_params.root_cache->memcg_kset; |
9a41707b VD |
5359 | #endif |
5360 | return slab_kset; | |
5361 | } | |
5362 | ||
81819f0f CL |
5363 | #define ID_STR_LENGTH 64 |
5364 | ||
5365 | /* Create a unique string id for a slab cache: | |
6446faa2 CL |
5366 | * |
5367 | * Format :[flags-]size | |
81819f0f CL |
5368 | */ |
5369 | static char *create_unique_id(struct kmem_cache *s) | |
5370 | { | |
5371 | char *name = kmalloc(ID_STR_LENGTH, GFP_KERNEL); | |
5372 | char *p = name; | |
5373 | ||
5374 | BUG_ON(!name); | |
5375 | ||
5376 | *p++ = ':'; | |
5377 | /* | |
5378 | * First flags affecting slabcache operations. We will only | |
5379 | * get here for aliasable slabs so we do not need to support | |
5380 | * too many flags. The flags here must cover all flags that | |
5381 | * are matched during merging to guarantee that the id is | |
5382 | * unique. | |
5383 | */ | |
5384 | if (s->flags & SLAB_CACHE_DMA) | |
5385 | *p++ = 'd'; | |
5386 | if (s->flags & SLAB_RECLAIM_ACCOUNT) | |
5387 | *p++ = 'a'; | |
becfda68 | 5388 | if (s->flags & SLAB_CONSISTENCY_CHECKS) |
81819f0f | 5389 | *p++ = 'F'; |
5a896d9e VN |
5390 | if (!(s->flags & SLAB_NOTRACK)) |
5391 | *p++ = 't'; | |
230e9fc2 VD |
5392 | if (s->flags & SLAB_ACCOUNT) |
5393 | *p++ = 'A'; | |
81819f0f CL |
5394 | if (p != name + 1) |
5395 | *p++ = '-'; | |
5396 | p += sprintf(p, "%07d", s->size); | |
2633d7a0 | 5397 | |
81819f0f CL |
5398 | BUG_ON(p > name + ID_STR_LENGTH - 1); |
5399 | return name; | |
5400 | } | |
5401 | ||
5402 | static int sysfs_slab_add(struct kmem_cache *s) | |
5403 | { | |
5404 | int err; | |
5405 | const char *name; | |
45530c44 | 5406 | int unmergeable = slab_unmergeable(s); |
81819f0f | 5407 | |
81819f0f CL |
5408 | if (unmergeable) { |
5409 | /* | |
5410 | * Slabcache can never be merged so we can use the name proper. | |
5411 | * This is typically the case for debug situations. In that | |
5412 | * case we can catch duplicate names easily. | |
5413 | */ | |
27c3a314 | 5414 | sysfs_remove_link(&slab_kset->kobj, s->name); |
81819f0f CL |
5415 | name = s->name; |
5416 | } else { | |
5417 | /* | |
5418 | * Create a unique name for the slab as a target | |
5419 | * for the symlinks. | |
5420 | */ | |
5421 | name = create_unique_id(s); | |
5422 | } | |
5423 | ||
9a41707b | 5424 | s->kobj.kset = cache_kset(s); |
26e4f205 | 5425 | err = kobject_init_and_add(&s->kobj, &slab_ktype, NULL, "%s", name); |
54b6a731 | 5426 | if (err) |
80da026a | 5427 | goto out; |
81819f0f CL |
5428 | |
5429 | err = sysfs_create_group(&s->kobj, &slab_attr_group); | |
54b6a731 DJ |
5430 | if (err) |
5431 | goto out_del_kobj; | |
9a41707b | 5432 | |
127424c8 | 5433 | #ifdef CONFIG_MEMCG |
9a41707b VD |
5434 | if (is_root_cache(s)) { |
5435 | s->memcg_kset = kset_create_and_add("cgroup", NULL, &s->kobj); | |
5436 | if (!s->memcg_kset) { | |
54b6a731 DJ |
5437 | err = -ENOMEM; |
5438 | goto out_del_kobj; | |
9a41707b VD |
5439 | } |
5440 | } | |
5441 | #endif | |
5442 | ||
81819f0f CL |
5443 | kobject_uevent(&s->kobj, KOBJ_ADD); |
5444 | if (!unmergeable) { | |
5445 | /* Setup first alias */ | |
5446 | sysfs_slab_alias(s, s->name); | |
81819f0f | 5447 | } |
54b6a731 DJ |
5448 | out: |
5449 | if (!unmergeable) | |
5450 | kfree(name); | |
5451 | return err; | |
5452 | out_del_kobj: | |
5453 | kobject_del(&s->kobj); | |
54b6a731 | 5454 | goto out; |
81819f0f CL |
5455 | } |
5456 | ||
41a21285 | 5457 | void sysfs_slab_remove(struct kmem_cache *s) |
81819f0f | 5458 | { |
97d06609 | 5459 | if (slab_state < FULL) |
2bce6485 CL |
5460 | /* |
5461 | * Sysfs has not been setup yet so no need to remove the | |
5462 | * cache from sysfs. | |
5463 | */ | |
5464 | return; | |
5465 | ||
127424c8 | 5466 | #ifdef CONFIG_MEMCG |
9a41707b VD |
5467 | kset_unregister(s->memcg_kset); |
5468 | #endif | |
81819f0f CL |
5469 | kobject_uevent(&s->kobj, KOBJ_REMOVE); |
5470 | kobject_del(&s->kobj); | |
151c602f | 5471 | kobject_put(&s->kobj); |
81819f0f CL |
5472 | } |
5473 | ||
5474 | /* | |
5475 | * Need to buffer aliases during bootup until sysfs becomes | |
9f6c708e | 5476 | * available lest we lose that information. |
81819f0f CL |
5477 | */ |
5478 | struct saved_alias { | |
5479 | struct kmem_cache *s; | |
5480 | const char *name; | |
5481 | struct saved_alias *next; | |
5482 | }; | |
5483 | ||
5af328a5 | 5484 | static struct saved_alias *alias_list; |
81819f0f CL |
5485 | |
5486 | static int sysfs_slab_alias(struct kmem_cache *s, const char *name) | |
5487 | { | |
5488 | struct saved_alias *al; | |
5489 | ||
97d06609 | 5490 | if (slab_state == FULL) { |
81819f0f CL |
5491 | /* |
5492 | * If we have a leftover link then remove it. | |
5493 | */ | |
27c3a314 GKH |
5494 | sysfs_remove_link(&slab_kset->kobj, name); |
5495 | return sysfs_create_link(&slab_kset->kobj, &s->kobj, name); | |
81819f0f CL |
5496 | } |
5497 | ||
5498 | al = kmalloc(sizeof(struct saved_alias), GFP_KERNEL); | |
5499 | if (!al) | |
5500 | return -ENOMEM; | |
5501 | ||
5502 | al->s = s; | |
5503 | al->name = name; | |
5504 | al->next = alias_list; | |
5505 | alias_list = al; | |
5506 | return 0; | |
5507 | } | |
5508 | ||
5509 | static int __init slab_sysfs_init(void) | |
5510 | { | |
5b95a4ac | 5511 | struct kmem_cache *s; |
81819f0f CL |
5512 | int err; |
5513 | ||
18004c5d | 5514 | mutex_lock(&slab_mutex); |
2bce6485 | 5515 | |
0ff21e46 | 5516 | slab_kset = kset_create_and_add("slab", &slab_uevent_ops, kernel_kobj); |
27c3a314 | 5517 | if (!slab_kset) { |
18004c5d | 5518 | mutex_unlock(&slab_mutex); |
f9f58285 | 5519 | pr_err("Cannot register slab subsystem.\n"); |
81819f0f CL |
5520 | return -ENOSYS; |
5521 | } | |
5522 | ||
97d06609 | 5523 | slab_state = FULL; |
26a7bd03 | 5524 | |
5b95a4ac | 5525 | list_for_each_entry(s, &slab_caches, list) { |
26a7bd03 | 5526 | err = sysfs_slab_add(s); |
5d540fb7 | 5527 | if (err) |
f9f58285 FF |
5528 | pr_err("SLUB: Unable to add boot slab %s to sysfs\n", |
5529 | s->name); | |
26a7bd03 | 5530 | } |
81819f0f CL |
5531 | |
5532 | while (alias_list) { | |
5533 | struct saved_alias *al = alias_list; | |
5534 | ||
5535 | alias_list = alias_list->next; | |
5536 | err = sysfs_slab_alias(al->s, al->name); | |
5d540fb7 | 5537 | if (err) |
f9f58285 FF |
5538 | pr_err("SLUB: Unable to add boot slab alias %s to sysfs\n", |
5539 | al->name); | |
81819f0f CL |
5540 | kfree(al); |
5541 | } | |
5542 | ||
18004c5d | 5543 | mutex_unlock(&slab_mutex); |
81819f0f CL |
5544 | resiliency_test(); |
5545 | return 0; | |
5546 | } | |
5547 | ||
5548 | __initcall(slab_sysfs_init); | |
ab4d5ed5 | 5549 | #endif /* CONFIG_SYSFS */ |
57ed3eda PE |
5550 | |
5551 | /* | |
5552 | * The /proc/slabinfo ABI | |
5553 | */ | |
158a9624 | 5554 | #ifdef CONFIG_SLABINFO |
0d7561c6 | 5555 | void get_slabinfo(struct kmem_cache *s, struct slabinfo *sinfo) |
57ed3eda | 5556 | { |
57ed3eda | 5557 | unsigned long nr_slabs = 0; |
205ab99d CL |
5558 | unsigned long nr_objs = 0; |
5559 | unsigned long nr_free = 0; | |
57ed3eda | 5560 | int node; |
fa45dc25 | 5561 | struct kmem_cache_node *n; |
57ed3eda | 5562 | |
fa45dc25 | 5563 | for_each_kmem_cache_node(s, node, n) { |
c17fd13e WL |
5564 | nr_slabs += node_nr_slabs(n); |
5565 | nr_objs += node_nr_objs(n); | |
205ab99d | 5566 | nr_free += count_partial(n, count_free); |
57ed3eda PE |
5567 | } |
5568 | ||
0d7561c6 GC |
5569 | sinfo->active_objs = nr_objs - nr_free; |
5570 | sinfo->num_objs = nr_objs; | |
5571 | sinfo->active_slabs = nr_slabs; | |
5572 | sinfo->num_slabs = nr_slabs; | |
5573 | sinfo->objects_per_slab = oo_objects(s->oo); | |
5574 | sinfo->cache_order = oo_order(s->oo); | |
57ed3eda PE |
5575 | } |
5576 | ||
0d7561c6 | 5577 | void slabinfo_show_stats(struct seq_file *m, struct kmem_cache *s) |
7b3c3a50 | 5578 | { |
7b3c3a50 AD |
5579 | } |
5580 | ||
b7454ad3 GC |
5581 | ssize_t slabinfo_write(struct file *file, const char __user *buffer, |
5582 | size_t count, loff_t *ppos) | |
7b3c3a50 | 5583 | { |
b7454ad3 | 5584 | return -EIO; |
7b3c3a50 | 5585 | } |
158a9624 | 5586 | #endif /* CONFIG_SLABINFO */ |