Commit | Line | Data |
---|---|---|
b2441318 | 1 | // SPDX-License-Identifier: GPL-2.0 |
81819f0f CL |
2 | /* |
3 | * SLUB: A slab allocator that limits cache line use instead of queuing | |
4 | * objects in per cpu and per node lists. | |
5 | * | |
dc84207d | 6 | * The allocator synchronizes using per slab locks or atomic operations |
881db7fb | 7 | * and only uses a centralized lock to manage a pool of partial slabs. |
81819f0f | 8 | * |
cde53535 | 9 | * (C) 2007 SGI, Christoph Lameter |
881db7fb | 10 | * (C) 2011 Linux Foundation, Christoph Lameter |
81819f0f CL |
11 | */ |
12 | ||
13 | #include <linux/mm.h> | |
1eb5ac64 | 14 | #include <linux/swap.h> /* struct reclaim_state */ |
81819f0f CL |
15 | #include <linux/module.h> |
16 | #include <linux/bit_spinlock.h> | |
17 | #include <linux/interrupt.h> | |
1b3865d0 | 18 | #include <linux/swab.h> |
81819f0f CL |
19 | #include <linux/bitops.h> |
20 | #include <linux/slab.h> | |
97d06609 | 21 | #include "slab.h" |
7b3c3a50 | 22 | #include <linux/proc_fs.h> |
81819f0f | 23 | #include <linux/seq_file.h> |
a79316c6 | 24 | #include <linux/kasan.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> |
b89fb5ef | 31 | #include <linux/kfence.h> |
b9049e23 | 32 | #include <linux/memory.h> |
f8bd2258 | 33 | #include <linux/math64.h> |
773ff60e | 34 | #include <linux/fault-inject.h> |
bfa71457 | 35 | #include <linux/stacktrace.h> |
4de900b4 | 36 | #include <linux/prefetch.h> |
2633d7a0 | 37 | #include <linux/memcontrol.h> |
2482ddec | 38 | #include <linux/random.h> |
1f9f78b1 | 39 | #include <kunit/test.h> |
81819f0f | 40 | |
64dd6849 | 41 | #include <linux/debugfs.h> |
4a92379b RK |
42 | #include <trace/events/kmem.h> |
43 | ||
072bb0aa MG |
44 | #include "internal.h" |
45 | ||
81819f0f CL |
46 | /* |
47 | * Lock order: | |
18004c5d | 48 | * 1. slab_mutex (Global Mutex) |
bd0e7491 VB |
49 | * 2. node->list_lock (Spinlock) |
50 | * 3. kmem_cache->cpu_slab->lock (Local lock) | |
51 | * 4. slab_lock(page) (Only on some arches or for debugging) | |
52 | * 5. object_map_lock (Only for debugging) | |
81819f0f | 53 | * |
18004c5d | 54 | * slab_mutex |
881db7fb | 55 | * |
18004c5d | 56 | * The role of the slab_mutex is to protect the list of all the slabs |
881db7fb | 57 | * and to synchronize major metadata changes to slab cache structures. |
bd0e7491 VB |
58 | * Also synchronizes memory hotplug callbacks. |
59 | * | |
60 | * slab_lock | |
61 | * | |
62 | * The slab_lock is a wrapper around the page lock, thus it is a bit | |
63 | * spinlock. | |
881db7fb CL |
64 | * |
65 | * The slab_lock is only used for debugging and on arches that do not | |
b7ccc7f8 | 66 | * have the ability to do a cmpxchg_double. It only protects: |
881db7fb | 67 | * A. page->freelist -> List of object free in a page |
b7ccc7f8 MW |
68 | * B. page->inuse -> Number of objects in use |
69 | * C. page->objects -> Number of objects in page | |
70 | * D. page->frozen -> frozen state | |
881db7fb | 71 | * |
bd0e7491 VB |
72 | * Frozen slabs |
73 | * | |
881db7fb | 74 | * If a slab is frozen then it is exempt from list management. It is not |
632b2ef0 LX |
75 | * on any list except per cpu partial list. The processor that froze the |
76 | * slab is the one who can perform list operations on the page. Other | |
77 | * processors may put objects onto the freelist but the processor that | |
78 | * froze the slab is the only one that can retrieve the objects from the | |
79 | * page's freelist. | |
81819f0f | 80 | * |
bd0e7491 VB |
81 | * list_lock |
82 | * | |
81819f0f CL |
83 | * The list_lock protects the partial and full list on each node and |
84 | * the partial slab counter. If taken then no new slabs may be added or | |
85 | * removed from the lists nor make the number of partial slabs be modified. | |
86 | * (Note that the total number of slabs is an atomic value that may be | |
87 | * modified without taking the list lock). | |
88 | * | |
89 | * The list_lock is a centralized lock and thus we avoid taking it as | |
90 | * much as possible. As long as SLUB does not have to handle partial | |
91 | * slabs, operations can continue without any centralized lock. F.e. | |
92 | * allocating a long series of objects that fill up slabs does not require | |
93 | * the list lock. | |
bd0e7491 VB |
94 | * |
95 | * cpu_slab->lock local lock | |
96 | * | |
97 | * This locks protect slowpath manipulation of all kmem_cache_cpu fields | |
98 | * except the stat counters. This is a percpu structure manipulated only by | |
99 | * the local cpu, so the lock protects against being preempted or interrupted | |
100 | * by an irq. Fast path operations rely on lockless operations instead. | |
101 | * On PREEMPT_RT, the local lock does not actually disable irqs (and thus | |
102 | * prevent the lockless operations), so fastpath operations also need to take | |
103 | * the lock and are no longer lockless. | |
104 | * | |
105 | * lockless fastpaths | |
106 | * | |
107 | * The fast path allocation (slab_alloc_node()) and freeing (do_slab_free()) | |
108 | * are fully lockless when satisfied from the percpu slab (and when | |
109 | * cmpxchg_double is possible to use, otherwise slab_lock is taken). | |
110 | * They also don't disable preemption or migration or irqs. They rely on | |
111 | * the transaction id (tid) field to detect being preempted or moved to | |
112 | * another cpu. | |
113 | * | |
114 | * irq, preemption, migration considerations | |
115 | * | |
116 | * Interrupts are disabled as part of list_lock or local_lock operations, or | |
117 | * around the slab_lock operation, in order to make the slab allocator safe | |
118 | * to use in the context of an irq. | |
119 | * | |
120 | * In addition, preemption (or migration on PREEMPT_RT) is disabled in the | |
121 | * allocation slowpath, bulk allocation, and put_cpu_partial(), so that the | |
122 | * local cpu doesn't change in the process and e.g. the kmem_cache_cpu pointer | |
123 | * doesn't have to be revalidated in each section protected by the local lock. | |
81819f0f CL |
124 | * |
125 | * SLUB assigns one slab for allocation to each processor. | |
126 | * Allocations only occur from these slabs called cpu slabs. | |
127 | * | |
672bba3a CL |
128 | * Slabs with free elements are kept on a partial list and during regular |
129 | * operations no list for full slabs is used. If an object in a full slab is | |
81819f0f | 130 | * freed then the slab will show up again on the partial lists. |
672bba3a CL |
131 | * We track full slabs for debugging purposes though because otherwise we |
132 | * cannot scan all objects. | |
81819f0f CL |
133 | * |
134 | * Slabs are freed when they become empty. Teardown and setup is | |
135 | * minimal so we rely on the page allocators per cpu caches for | |
136 | * fast frees and allocs. | |
137 | * | |
aed68148 | 138 | * page->frozen The slab is frozen and exempt from list processing. |
4b6f0750 CL |
139 | * This means that the slab is dedicated to a purpose |
140 | * such as satisfying allocations for a specific | |
141 | * processor. Objects may be freed in the slab while | |
142 | * it is frozen but slab_free will then skip the usual | |
143 | * list operations. It is up to the processor holding | |
144 | * the slab to integrate the slab into the slab lists | |
145 | * when the slab is no longer needed. | |
146 | * | |
147 | * One use of this flag is to mark slabs that are | |
148 | * used for allocations. Then such a slab becomes a cpu | |
149 | * slab. The cpu slab may be equipped with an additional | |
dfb4f096 | 150 | * freelist that allows lockless access to |
894b8788 CL |
151 | * free objects in addition to the regular freelist |
152 | * that requires the slab lock. | |
81819f0f | 153 | * |
aed68148 | 154 | * SLAB_DEBUG_FLAGS Slab requires special handling due to debug |
81819f0f | 155 | * options set. This moves slab handling out of |
894b8788 | 156 | * the fast path and disables lockless freelists. |
81819f0f CL |
157 | */ |
158 | ||
25c00c50 VB |
159 | /* |
160 | * We could simply use migrate_disable()/enable() but as long as it's a | |
161 | * function call even on !PREEMPT_RT, use inline preempt_disable() there. | |
162 | */ | |
163 | #ifndef CONFIG_PREEMPT_RT | |
164 | #define slub_get_cpu_ptr(var) get_cpu_ptr(var) | |
165 | #define slub_put_cpu_ptr(var) put_cpu_ptr(var) | |
166 | #else | |
167 | #define slub_get_cpu_ptr(var) \ | |
168 | ({ \ | |
169 | migrate_disable(); \ | |
170 | this_cpu_ptr(var); \ | |
171 | }) | |
172 | #define slub_put_cpu_ptr(var) \ | |
173 | do { \ | |
174 | (void)(var); \ | |
175 | migrate_enable(); \ | |
176 | } while (0) | |
177 | #endif | |
178 | ||
ca0cab65 VB |
179 | #ifdef CONFIG_SLUB_DEBUG |
180 | #ifdef CONFIG_SLUB_DEBUG_ON | |
181 | DEFINE_STATIC_KEY_TRUE(slub_debug_enabled); | |
182 | #else | |
183 | DEFINE_STATIC_KEY_FALSE(slub_debug_enabled); | |
184 | #endif | |
79270291 | 185 | #endif /* CONFIG_SLUB_DEBUG */ |
ca0cab65 | 186 | |
59052e89 VB |
187 | static inline bool kmem_cache_debug(struct kmem_cache *s) |
188 | { | |
189 | return kmem_cache_debug_flags(s, SLAB_DEBUG_FLAGS); | |
af537b0a | 190 | } |
5577bd8a | 191 | |
117d54df | 192 | void *fixup_red_left(struct kmem_cache *s, void *p) |
d86bd1be | 193 | { |
59052e89 | 194 | if (kmem_cache_debug_flags(s, SLAB_RED_ZONE)) |
d86bd1be JK |
195 | p += s->red_left_pad; |
196 | ||
197 | return p; | |
198 | } | |
199 | ||
345c905d JK |
200 | static inline bool kmem_cache_has_cpu_partial(struct kmem_cache *s) |
201 | { | |
202 | #ifdef CONFIG_SLUB_CPU_PARTIAL | |
203 | return !kmem_cache_debug(s); | |
204 | #else | |
205 | return false; | |
206 | #endif | |
207 | } | |
208 | ||
81819f0f CL |
209 | /* |
210 | * Issues still to be resolved: | |
211 | * | |
81819f0f CL |
212 | * - Support PAGE_ALLOC_DEBUG. Should be easy to do. |
213 | * | |
81819f0f CL |
214 | * - Variable sizing of the per node arrays |
215 | */ | |
216 | ||
b789ef51 CL |
217 | /* Enable to log cmpxchg failures */ |
218 | #undef SLUB_DEBUG_CMPXCHG | |
219 | ||
2086d26a | 220 | /* |
dc84207d | 221 | * Minimum number of partial slabs. These will be left on the partial |
2086d26a CL |
222 | * lists even if they are empty. kmem_cache_shrink may reclaim them. |
223 | */ | |
76be8950 | 224 | #define MIN_PARTIAL 5 |
e95eed57 | 225 | |
2086d26a CL |
226 | /* |
227 | * Maximum number of desirable partial slabs. | |
228 | * The existence of more partial slabs makes kmem_cache_shrink | |
721ae22a | 229 | * sort the partial list by the number of objects in use. |
2086d26a CL |
230 | */ |
231 | #define MAX_PARTIAL 10 | |
232 | ||
becfda68 | 233 | #define DEBUG_DEFAULT_FLAGS (SLAB_CONSISTENCY_CHECKS | SLAB_RED_ZONE | \ |
81819f0f | 234 | SLAB_POISON | SLAB_STORE_USER) |
672bba3a | 235 | |
149daaf3 LA |
236 | /* |
237 | * These debug flags cannot use CMPXCHG because there might be consistency | |
238 | * issues when checking or reading debug information | |
239 | */ | |
240 | #define SLAB_NO_CMPXCHG (SLAB_CONSISTENCY_CHECKS | SLAB_STORE_USER | \ | |
241 | SLAB_TRACE) | |
242 | ||
243 | ||
fa5ec8a1 | 244 | /* |
3de47213 DR |
245 | * Debugging flags that require metadata to be stored in the slab. These get |
246 | * disabled when slub_debug=O is used and a cache's min order increases with | |
247 | * metadata. | |
fa5ec8a1 | 248 | */ |
3de47213 | 249 | #define DEBUG_METADATA_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER) |
fa5ec8a1 | 250 | |
210b5c06 CG |
251 | #define OO_SHIFT 16 |
252 | #define OO_MASK ((1 << OO_SHIFT) - 1) | |
50d5c41c | 253 | #define MAX_OBJS_PER_PAGE 32767 /* since page.objects is u15 */ |
210b5c06 | 254 | |
81819f0f | 255 | /* Internal SLUB flags */ |
d50112ed | 256 | /* Poison object */ |
4fd0b46e | 257 | #define __OBJECT_POISON ((slab_flags_t __force)0x80000000U) |
d50112ed | 258 | /* Use cmpxchg_double */ |
4fd0b46e | 259 | #define __CMPXCHG_DOUBLE ((slab_flags_t __force)0x40000000U) |
81819f0f | 260 | |
02cbc874 CL |
261 | /* |
262 | * Tracking user of a slab. | |
263 | */ | |
d6543e39 | 264 | #define TRACK_ADDRS_COUNT 16 |
02cbc874 | 265 | struct track { |
ce71e27c | 266 | unsigned long addr; /* Called from address */ |
ae14c63a LT |
267 | #ifdef CONFIG_STACKTRACE |
268 | unsigned long addrs[TRACK_ADDRS_COUNT]; /* Called from address */ | |
d6543e39 | 269 | #endif |
02cbc874 CL |
270 | int cpu; /* Was running on cpu */ |
271 | int pid; /* Pid context */ | |
272 | unsigned long when; /* When did the operation occur */ | |
273 | }; | |
274 | ||
275 | enum track_item { TRACK_ALLOC, TRACK_FREE }; | |
276 | ||
ab4d5ed5 | 277 | #ifdef CONFIG_SYSFS |
81819f0f CL |
278 | static int sysfs_slab_add(struct kmem_cache *); |
279 | static int sysfs_slab_alias(struct kmem_cache *, const char *); | |
81819f0f | 280 | #else |
0c710013 CL |
281 | static inline int sysfs_slab_add(struct kmem_cache *s) { return 0; } |
282 | static inline int sysfs_slab_alias(struct kmem_cache *s, const char *p) | |
283 | { return 0; } | |
81819f0f CL |
284 | #endif |
285 | ||
64dd6849 FM |
286 | #if defined(CONFIG_DEBUG_FS) && defined(CONFIG_SLUB_DEBUG) |
287 | static void debugfs_slab_add(struct kmem_cache *); | |
288 | #else | |
289 | static inline void debugfs_slab_add(struct kmem_cache *s) { } | |
290 | #endif | |
291 | ||
4fdccdfb | 292 | static inline void stat(const struct kmem_cache *s, enum stat_item si) |
8ff12cfc CL |
293 | { |
294 | #ifdef CONFIG_SLUB_STATS | |
88da03a6 CL |
295 | /* |
296 | * The rmw is racy on a preemptible kernel but this is acceptable, so | |
297 | * avoid this_cpu_add()'s irq-disable overhead. | |
298 | */ | |
299 | raw_cpu_inc(s->cpu_slab->stat[si]); | |
8ff12cfc CL |
300 | #endif |
301 | } | |
302 | ||
7e1fa93d VB |
303 | /* |
304 | * Tracks for which NUMA nodes we have kmem_cache_nodes allocated. | |
305 | * Corresponds to node_state[N_NORMAL_MEMORY], but can temporarily | |
306 | * differ during memory hotplug/hotremove operations. | |
307 | * Protected by slab_mutex. | |
308 | */ | |
309 | static nodemask_t slab_nodes; | |
310 | ||
81819f0f CL |
311 | /******************************************************************** |
312 | * Core slab cache functions | |
313 | *******************************************************************/ | |
314 | ||
2482ddec KC |
315 | /* |
316 | * Returns freelist pointer (ptr). With hardening, this is obfuscated | |
317 | * with an XOR of the address where the pointer is held and a per-cache | |
318 | * random number. | |
319 | */ | |
320 | static inline void *freelist_ptr(const struct kmem_cache *s, void *ptr, | |
321 | unsigned long ptr_addr) | |
322 | { | |
323 | #ifdef CONFIG_SLAB_FREELIST_HARDENED | |
d36a63a9 | 324 | /* |
aa1ef4d7 | 325 | * When CONFIG_KASAN_SW/HW_TAGS is enabled, ptr_addr might be tagged. |
d36a63a9 AK |
326 | * Normally, this doesn't cause any issues, as both set_freepointer() |
327 | * and get_freepointer() are called with a pointer with the same tag. | |
328 | * However, there are some issues with CONFIG_SLUB_DEBUG code. For | |
329 | * example, when __free_slub() iterates over objects in a cache, it | |
330 | * passes untagged pointers to check_object(). check_object() in turns | |
331 | * calls get_freepointer() with an untagged pointer, which causes the | |
332 | * freepointer to be restored incorrectly. | |
333 | */ | |
334 | return (void *)((unsigned long)ptr ^ s->random ^ | |
1ad53d9f | 335 | swab((unsigned long)kasan_reset_tag((void *)ptr_addr))); |
2482ddec KC |
336 | #else |
337 | return ptr; | |
338 | #endif | |
339 | } | |
340 | ||
341 | /* Returns the freelist pointer recorded at location ptr_addr. */ | |
342 | static inline void *freelist_dereference(const struct kmem_cache *s, | |
343 | void *ptr_addr) | |
344 | { | |
345 | return freelist_ptr(s, (void *)*(unsigned long *)(ptr_addr), | |
346 | (unsigned long)ptr_addr); | |
347 | } | |
348 | ||
7656c72b CL |
349 | static inline void *get_freepointer(struct kmem_cache *s, void *object) |
350 | { | |
aa1ef4d7 | 351 | object = kasan_reset_tag(object); |
2482ddec | 352 | return freelist_dereference(s, object + s->offset); |
7656c72b CL |
353 | } |
354 | ||
0ad9500e ED |
355 | static void prefetch_freepointer(const struct kmem_cache *s, void *object) |
356 | { | |
04b4b006 | 357 | prefetchw(object + s->offset); |
0ad9500e ED |
358 | } |
359 | ||
1393d9a1 CL |
360 | static inline void *get_freepointer_safe(struct kmem_cache *s, void *object) |
361 | { | |
2482ddec | 362 | unsigned long freepointer_addr; |
1393d9a1 CL |
363 | void *p; |
364 | ||
8e57f8ac | 365 | if (!debug_pagealloc_enabled_static()) |
922d566c JK |
366 | return get_freepointer(s, object); |
367 | ||
f70b0049 | 368 | object = kasan_reset_tag(object); |
2482ddec | 369 | freepointer_addr = (unsigned long)object + s->offset; |
fe557319 | 370 | copy_from_kernel_nofault(&p, (void **)freepointer_addr, sizeof(p)); |
2482ddec | 371 | return freelist_ptr(s, p, freepointer_addr); |
1393d9a1 CL |
372 | } |
373 | ||
7656c72b CL |
374 | static inline void set_freepointer(struct kmem_cache *s, void *object, void *fp) |
375 | { | |
2482ddec KC |
376 | unsigned long freeptr_addr = (unsigned long)object + s->offset; |
377 | ||
ce6fa91b AP |
378 | #ifdef CONFIG_SLAB_FREELIST_HARDENED |
379 | BUG_ON(object == fp); /* naive detection of double free or corruption */ | |
380 | #endif | |
381 | ||
aa1ef4d7 | 382 | freeptr_addr = (unsigned long)kasan_reset_tag((void *)freeptr_addr); |
2482ddec | 383 | *(void **)freeptr_addr = freelist_ptr(s, fp, freeptr_addr); |
7656c72b CL |
384 | } |
385 | ||
386 | /* Loop over all objects in a slab */ | |
224a88be | 387 | #define for_each_object(__p, __s, __addr, __objects) \ |
d86bd1be JK |
388 | for (__p = fixup_red_left(__s, __addr); \ |
389 | __p < (__addr) + (__objects) * (__s)->size; \ | |
390 | __p += (__s)->size) | |
7656c72b | 391 | |
9736d2a9 | 392 | static inline unsigned int order_objects(unsigned int order, unsigned int size) |
ab9a0f19 | 393 | { |
9736d2a9 | 394 | return ((unsigned int)PAGE_SIZE << order) / size; |
ab9a0f19 LJ |
395 | } |
396 | ||
19af27af | 397 | static inline struct kmem_cache_order_objects oo_make(unsigned int order, |
9736d2a9 | 398 | unsigned int size) |
834f3d11 CL |
399 | { |
400 | struct kmem_cache_order_objects x = { | |
9736d2a9 | 401 | (order << OO_SHIFT) + order_objects(order, size) |
834f3d11 CL |
402 | }; |
403 | ||
404 | return x; | |
405 | } | |
406 | ||
19af27af | 407 | static inline unsigned int oo_order(struct kmem_cache_order_objects x) |
834f3d11 | 408 | { |
210b5c06 | 409 | return x.x >> OO_SHIFT; |
834f3d11 CL |
410 | } |
411 | ||
19af27af | 412 | static inline unsigned int oo_objects(struct kmem_cache_order_objects x) |
834f3d11 | 413 | { |
210b5c06 | 414 | return x.x & OO_MASK; |
834f3d11 CL |
415 | } |
416 | ||
b47291ef VB |
417 | #ifdef CONFIG_SLUB_CPU_PARTIAL |
418 | static void slub_set_cpu_partial(struct kmem_cache *s, unsigned int nr_objects) | |
419 | { | |
420 | unsigned int nr_pages; | |
421 | ||
422 | s->cpu_partial = nr_objects; | |
423 | ||
424 | /* | |
425 | * We take the number of objects but actually limit the number of | |
426 | * pages on the per cpu partial list, in order to limit excessive | |
427 | * growth of the list. For simplicity we assume that the pages will | |
428 | * be half-full. | |
429 | */ | |
430 | nr_pages = DIV_ROUND_UP(nr_objects * 2, oo_objects(s->oo)); | |
431 | s->cpu_partial_pages = nr_pages; | |
432 | } | |
433 | #else | |
434 | static inline void | |
435 | slub_set_cpu_partial(struct kmem_cache *s, unsigned int nr_objects) | |
436 | { | |
437 | } | |
438 | #endif /* CONFIG_SLUB_CPU_PARTIAL */ | |
439 | ||
881db7fb CL |
440 | /* |
441 | * Per slab locking using the pagelock | |
442 | */ | |
0393895b | 443 | static __always_inline void __slab_lock(struct slab *slab) |
881db7fb | 444 | { |
0393895b VB |
445 | struct page *page = slab_page(slab); |
446 | ||
48c935ad | 447 | VM_BUG_ON_PAGE(PageTail(page), page); |
881db7fb CL |
448 | bit_spin_lock(PG_locked, &page->flags); |
449 | } | |
450 | ||
0393895b | 451 | static __always_inline void __slab_unlock(struct slab *slab) |
881db7fb | 452 | { |
0393895b VB |
453 | struct page *page = slab_page(slab); |
454 | ||
48c935ad | 455 | VM_BUG_ON_PAGE(PageTail(page), page); |
881db7fb CL |
456 | __bit_spin_unlock(PG_locked, &page->flags); |
457 | } | |
458 | ||
a2b4ae8b VB |
459 | static __always_inline void slab_lock(struct page *page, unsigned long *flags) |
460 | { | |
461 | if (IS_ENABLED(CONFIG_PREEMPT_RT)) | |
462 | local_irq_save(*flags); | |
0393895b | 463 | __slab_lock(page_slab(page)); |
a2b4ae8b VB |
464 | } |
465 | ||
466 | static __always_inline void slab_unlock(struct page *page, unsigned long *flags) | |
467 | { | |
0393895b | 468 | __slab_unlock(page_slab(page)); |
a2b4ae8b VB |
469 | if (IS_ENABLED(CONFIG_PREEMPT_RT)) |
470 | local_irq_restore(*flags); | |
471 | } | |
472 | ||
473 | /* | |
474 | * Interrupts must be disabled (for the fallback code to work right), typically | |
475 | * by an _irqsave() lock variant. Except on PREEMPT_RT where locks are different | |
476 | * so we disable interrupts as part of slab_[un]lock(). | |
477 | */ | |
1d07171c CL |
478 | static inline bool __cmpxchg_double_slab(struct kmem_cache *s, struct page *page, |
479 | void *freelist_old, unsigned long counters_old, | |
480 | void *freelist_new, unsigned long counters_new, | |
481 | const char *n) | |
482 | { | |
a2b4ae8b VB |
483 | if (!IS_ENABLED(CONFIG_PREEMPT_RT)) |
484 | lockdep_assert_irqs_disabled(); | |
2565409f HC |
485 | #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \ |
486 | defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE) | |
1d07171c | 487 | if (s->flags & __CMPXCHG_DOUBLE) { |
cdcd6298 | 488 | if (cmpxchg_double(&page->freelist, &page->counters, |
0aa9a13d DC |
489 | freelist_old, counters_old, |
490 | freelist_new, counters_new)) | |
6f6528a1 | 491 | return true; |
1d07171c CL |
492 | } else |
493 | #endif | |
494 | { | |
a2b4ae8b VB |
495 | /* init to 0 to prevent spurious warnings */ |
496 | unsigned long flags = 0; | |
497 | ||
498 | slab_lock(page, &flags); | |
d0e0ac97 CG |
499 | if (page->freelist == freelist_old && |
500 | page->counters == counters_old) { | |
1d07171c | 501 | page->freelist = freelist_new; |
7d27a04b | 502 | page->counters = counters_new; |
a2b4ae8b | 503 | slab_unlock(page, &flags); |
6f6528a1 | 504 | return true; |
1d07171c | 505 | } |
a2b4ae8b | 506 | slab_unlock(page, &flags); |
1d07171c CL |
507 | } |
508 | ||
509 | cpu_relax(); | |
510 | stat(s, CMPXCHG_DOUBLE_FAIL); | |
511 | ||
512 | #ifdef SLUB_DEBUG_CMPXCHG | |
f9f58285 | 513 | pr_info("%s %s: cmpxchg double redo ", n, s->name); |
1d07171c CL |
514 | #endif |
515 | ||
6f6528a1 | 516 | return false; |
1d07171c CL |
517 | } |
518 | ||
b789ef51 CL |
519 | static inline bool cmpxchg_double_slab(struct kmem_cache *s, struct page *page, |
520 | void *freelist_old, unsigned long counters_old, | |
521 | void *freelist_new, unsigned long counters_new, | |
522 | const char *n) | |
523 | { | |
2565409f HC |
524 | #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \ |
525 | defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE) | |
b789ef51 | 526 | if (s->flags & __CMPXCHG_DOUBLE) { |
cdcd6298 | 527 | if (cmpxchg_double(&page->freelist, &page->counters, |
0aa9a13d DC |
528 | freelist_old, counters_old, |
529 | freelist_new, counters_new)) | |
6f6528a1 | 530 | return true; |
b789ef51 CL |
531 | } else |
532 | #endif | |
533 | { | |
1d07171c CL |
534 | unsigned long flags; |
535 | ||
536 | local_irq_save(flags); | |
0393895b | 537 | __slab_lock(page_slab(page)); |
d0e0ac97 CG |
538 | if (page->freelist == freelist_old && |
539 | page->counters == counters_old) { | |
b789ef51 | 540 | page->freelist = freelist_new; |
7d27a04b | 541 | page->counters = counters_new; |
0393895b | 542 | __slab_unlock(page_slab(page)); |
1d07171c | 543 | local_irq_restore(flags); |
6f6528a1 | 544 | return true; |
b789ef51 | 545 | } |
0393895b | 546 | __slab_unlock(page_slab(page)); |
1d07171c | 547 | local_irq_restore(flags); |
b789ef51 CL |
548 | } |
549 | ||
550 | cpu_relax(); | |
551 | stat(s, CMPXCHG_DOUBLE_FAIL); | |
552 | ||
553 | #ifdef SLUB_DEBUG_CMPXCHG | |
f9f58285 | 554 | pr_info("%s %s: cmpxchg double redo ", n, s->name); |
b789ef51 CL |
555 | #endif |
556 | ||
6f6528a1 | 557 | return false; |
b789ef51 CL |
558 | } |
559 | ||
41ecc55b | 560 | #ifdef CONFIG_SLUB_DEBUG |
90e9f6a6 | 561 | static unsigned long object_map[BITS_TO_LONGS(MAX_OBJS_PER_PAGE)]; |
94ef0304 | 562 | static DEFINE_RAW_SPINLOCK(object_map_lock); |
90e9f6a6 | 563 | |
b3fd64e1 VB |
564 | static void __fill_map(unsigned long *obj_map, struct kmem_cache *s, |
565 | struct page *page) | |
566 | { | |
567 | void *addr = page_address(page); | |
568 | void *p; | |
569 | ||
570 | bitmap_zero(obj_map, page->objects); | |
571 | ||
572 | for (p = page->freelist; p; p = get_freepointer(s, p)) | |
573 | set_bit(__obj_to_index(s, addr, p), obj_map); | |
574 | } | |
575 | ||
1f9f78b1 OG |
576 | #if IS_ENABLED(CONFIG_KUNIT) |
577 | static bool slab_add_kunit_errors(void) | |
578 | { | |
579 | struct kunit_resource *resource; | |
580 | ||
581 | if (likely(!current->kunit_test)) | |
582 | return false; | |
583 | ||
584 | resource = kunit_find_named_resource(current->kunit_test, "slab_errors"); | |
585 | if (!resource) | |
586 | return false; | |
587 | ||
588 | (*(int *)resource->data)++; | |
589 | kunit_put_resource(resource); | |
590 | return true; | |
591 | } | |
592 | #else | |
593 | static inline bool slab_add_kunit_errors(void) { return false; } | |
594 | #endif | |
595 | ||
5f80b13a CL |
596 | /* |
597 | * Determine a map of object in use on a page. | |
598 | * | |
881db7fb | 599 | * Node listlock must be held to guarantee that the page does |
5f80b13a CL |
600 | * not vanish from under us. |
601 | */ | |
90e9f6a6 | 602 | static unsigned long *get_map(struct kmem_cache *s, struct page *page) |
31364c2e | 603 | __acquires(&object_map_lock) |
5f80b13a | 604 | { |
90e9f6a6 YZ |
605 | VM_BUG_ON(!irqs_disabled()); |
606 | ||
94ef0304 | 607 | raw_spin_lock(&object_map_lock); |
90e9f6a6 | 608 | |
b3fd64e1 | 609 | __fill_map(object_map, s, page); |
90e9f6a6 YZ |
610 | |
611 | return object_map; | |
612 | } | |
613 | ||
81aba9e0 | 614 | static void put_map(unsigned long *map) __releases(&object_map_lock) |
90e9f6a6 YZ |
615 | { |
616 | VM_BUG_ON(map != object_map); | |
94ef0304 | 617 | raw_spin_unlock(&object_map_lock); |
5f80b13a CL |
618 | } |
619 | ||
870b1fbb | 620 | static inline unsigned int size_from_object(struct kmem_cache *s) |
d86bd1be JK |
621 | { |
622 | if (s->flags & SLAB_RED_ZONE) | |
623 | return s->size - s->red_left_pad; | |
624 | ||
625 | return s->size; | |
626 | } | |
627 | ||
628 | static inline void *restore_red_left(struct kmem_cache *s, void *p) | |
629 | { | |
630 | if (s->flags & SLAB_RED_ZONE) | |
631 | p -= s->red_left_pad; | |
632 | ||
633 | return p; | |
634 | } | |
635 | ||
41ecc55b CL |
636 | /* |
637 | * Debug settings: | |
638 | */ | |
89d3c87e | 639 | #if defined(CONFIG_SLUB_DEBUG_ON) |
d50112ed | 640 | static slab_flags_t slub_debug = DEBUG_DEFAULT_FLAGS; |
f0630fff | 641 | #else |
d50112ed | 642 | static slab_flags_t slub_debug; |
f0630fff | 643 | #endif |
41ecc55b | 644 | |
e17f1dfb | 645 | static char *slub_debug_string; |
fa5ec8a1 | 646 | static int disable_higher_order_debug; |
41ecc55b | 647 | |
a79316c6 AR |
648 | /* |
649 | * slub is about to manipulate internal object metadata. This memory lies | |
650 | * outside the range of the allocated object, so accessing it would normally | |
651 | * be reported by kasan as a bounds error. metadata_access_enable() is used | |
652 | * to tell kasan that these accesses are OK. | |
653 | */ | |
654 | static inline void metadata_access_enable(void) | |
655 | { | |
656 | kasan_disable_current(); | |
657 | } | |
658 | ||
659 | static inline void metadata_access_disable(void) | |
660 | { | |
661 | kasan_enable_current(); | |
662 | } | |
663 | ||
81819f0f CL |
664 | /* |
665 | * Object debugging | |
666 | */ | |
d86bd1be JK |
667 | |
668 | /* Verify that a pointer has an address that is valid within a slab page */ | |
669 | static inline int check_valid_pointer(struct kmem_cache *s, | |
670 | struct page *page, void *object) | |
671 | { | |
672 | void *base; | |
673 | ||
674 | if (!object) | |
675 | return 1; | |
676 | ||
677 | base = page_address(page); | |
338cfaad | 678 | object = kasan_reset_tag(object); |
d86bd1be JK |
679 | object = restore_red_left(s, object); |
680 | if (object < base || object >= base + page->objects * s->size || | |
681 | (object - base) % s->size) { | |
682 | return 0; | |
683 | } | |
684 | ||
685 | return 1; | |
686 | } | |
687 | ||
aa2efd5e DT |
688 | static void print_section(char *level, char *text, u8 *addr, |
689 | unsigned int length) | |
81819f0f | 690 | { |
a79316c6 | 691 | metadata_access_enable(); |
340caf17 KYL |
692 | print_hex_dump(level, text, DUMP_PREFIX_ADDRESS, |
693 | 16, 1, kasan_reset_tag((void *)addr), length, 1); | |
a79316c6 | 694 | metadata_access_disable(); |
81819f0f CL |
695 | } |
696 | ||
cbfc35a4 WL |
697 | /* |
698 | * See comment in calculate_sizes(). | |
699 | */ | |
700 | static inline bool freeptr_outside_object(struct kmem_cache *s) | |
701 | { | |
702 | return s->offset >= s->inuse; | |
703 | } | |
704 | ||
705 | /* | |
706 | * Return offset of the end of info block which is inuse + free pointer if | |
707 | * not overlapping with object. | |
708 | */ | |
709 | static inline unsigned int get_info_end(struct kmem_cache *s) | |
710 | { | |
711 | if (freeptr_outside_object(s)) | |
712 | return s->inuse + sizeof(void *); | |
713 | else | |
714 | return s->inuse; | |
715 | } | |
716 | ||
81819f0f CL |
717 | static struct track *get_track(struct kmem_cache *s, void *object, |
718 | enum track_item alloc) | |
719 | { | |
720 | struct track *p; | |
721 | ||
cbfc35a4 | 722 | p = object + get_info_end(s); |
81819f0f | 723 | |
aa1ef4d7 | 724 | return kasan_reset_tag(p + alloc); |
81819f0f CL |
725 | } |
726 | ||
727 | static void set_track(struct kmem_cache *s, void *object, | |
ce71e27c | 728 | enum track_item alloc, unsigned long addr) |
81819f0f | 729 | { |
1a00df4a | 730 | struct track *p = get_track(s, object, alloc); |
81819f0f | 731 | |
81819f0f | 732 | if (addr) { |
ae14c63a LT |
733 | #ifdef CONFIG_STACKTRACE |
734 | unsigned int nr_entries; | |
735 | ||
736 | metadata_access_enable(); | |
737 | nr_entries = stack_trace_save(kasan_reset_tag(p->addrs), | |
738 | TRACK_ADDRS_COUNT, 3); | |
739 | metadata_access_disable(); | |
740 | ||
741 | if (nr_entries < TRACK_ADDRS_COUNT) | |
742 | p->addrs[nr_entries] = 0; | |
d6543e39 | 743 | #endif |
81819f0f CL |
744 | p->addr = addr; |
745 | p->cpu = smp_processor_id(); | |
88e4ccf2 | 746 | p->pid = current->pid; |
81819f0f | 747 | p->when = jiffies; |
b8ca7ff7 | 748 | } else { |
81819f0f | 749 | memset(p, 0, sizeof(struct track)); |
b8ca7ff7 | 750 | } |
81819f0f CL |
751 | } |
752 | ||
81819f0f CL |
753 | static void init_tracking(struct kmem_cache *s, void *object) |
754 | { | |
24922684 CL |
755 | if (!(s->flags & SLAB_STORE_USER)) |
756 | return; | |
757 | ||
ce71e27c EGM |
758 | set_track(s, object, TRACK_FREE, 0UL); |
759 | set_track(s, object, TRACK_ALLOC, 0UL); | |
81819f0f CL |
760 | } |
761 | ||
86609d33 | 762 | static void print_track(const char *s, struct track *t, unsigned long pr_time) |
81819f0f CL |
763 | { |
764 | if (!t->addr) | |
765 | return; | |
766 | ||
96b94abc | 767 | pr_err("%s in %pS age=%lu cpu=%u pid=%d\n", |
86609d33 | 768 | s, (void *)t->addr, pr_time - t->when, t->cpu, t->pid); |
ae14c63a | 769 | #ifdef CONFIG_STACKTRACE |
d6543e39 | 770 | { |
ae14c63a LT |
771 | int i; |
772 | for (i = 0; i < TRACK_ADDRS_COUNT; i++) | |
773 | if (t->addrs[i]) | |
774 | pr_err("\t%pS\n", (void *)t->addrs[i]); | |
775 | else | |
776 | break; | |
d6543e39 BG |
777 | } |
778 | #endif | |
24922684 CL |
779 | } |
780 | ||
e42f174e | 781 | void print_tracking(struct kmem_cache *s, void *object) |
24922684 | 782 | { |
86609d33 | 783 | unsigned long pr_time = jiffies; |
24922684 CL |
784 | if (!(s->flags & SLAB_STORE_USER)) |
785 | return; | |
786 | ||
86609d33 CP |
787 | print_track("Allocated", get_track(s, object, TRACK_ALLOC), pr_time); |
788 | print_track("Freed", get_track(s, object, TRACK_FREE), pr_time); | |
24922684 CL |
789 | } |
790 | ||
fb012e27 | 791 | static void print_slab_info(const struct slab *slab) |
24922684 | 792 | { |
fb012e27 | 793 | struct folio *folio = (struct folio *)slab_folio(slab); |
24922684 | 794 | |
fb012e27 MWO |
795 | pr_err("Slab 0x%p objects=%u used=%u fp=0x%p flags=%pGp\n", |
796 | slab, slab->objects, slab->inuse, slab->freelist, | |
797 | folio_flags(folio, 0)); | |
24922684 CL |
798 | } |
799 | ||
800 | static void slab_bug(struct kmem_cache *s, char *fmt, ...) | |
801 | { | |
ecc42fbe | 802 | struct va_format vaf; |
24922684 | 803 | va_list args; |
24922684 CL |
804 | |
805 | va_start(args, fmt); | |
ecc42fbe FF |
806 | vaf.fmt = fmt; |
807 | vaf.va = &args; | |
f9f58285 | 808 | pr_err("=============================================================================\n"); |
ecc42fbe | 809 | pr_err("BUG %s (%s): %pV\n", s->name, print_tainted(), &vaf); |
f9f58285 | 810 | pr_err("-----------------------------------------------------------------------------\n\n"); |
ecc42fbe | 811 | va_end(args); |
81819f0f CL |
812 | } |
813 | ||
582d1212 | 814 | __printf(2, 3) |
24922684 CL |
815 | static void slab_fix(struct kmem_cache *s, char *fmt, ...) |
816 | { | |
ecc42fbe | 817 | struct va_format vaf; |
24922684 | 818 | va_list args; |
24922684 | 819 | |
1f9f78b1 OG |
820 | if (slab_add_kunit_errors()) |
821 | return; | |
822 | ||
24922684 | 823 | va_start(args, fmt); |
ecc42fbe FF |
824 | vaf.fmt = fmt; |
825 | vaf.va = &args; | |
826 | pr_err("FIX %s: %pV\n", s->name, &vaf); | |
24922684 | 827 | va_end(args); |
24922684 CL |
828 | } |
829 | ||
830 | static void print_trailer(struct kmem_cache *s, struct page *page, u8 *p) | |
81819f0f CL |
831 | { |
832 | unsigned int off; /* Offset of last byte */ | |
a973e9dd | 833 | u8 *addr = page_address(page); |
24922684 CL |
834 | |
835 | print_tracking(s, p); | |
836 | ||
fb012e27 | 837 | print_slab_info(page_slab(page)); |
24922684 | 838 | |
96b94abc | 839 | pr_err("Object 0x%p @offset=%tu fp=0x%p\n\n", |
f9f58285 | 840 | p, p - addr, get_freepointer(s, p)); |
24922684 | 841 | |
d86bd1be | 842 | if (s->flags & SLAB_RED_ZONE) |
8669dbab | 843 | print_section(KERN_ERR, "Redzone ", p - s->red_left_pad, |
aa2efd5e | 844 | s->red_left_pad); |
d86bd1be | 845 | else if (p > addr + 16) |
aa2efd5e | 846 | print_section(KERN_ERR, "Bytes b4 ", p - 16, 16); |
81819f0f | 847 | |
8669dbab | 848 | print_section(KERN_ERR, "Object ", p, |
1b473f29 | 849 | min_t(unsigned int, s->object_size, PAGE_SIZE)); |
81819f0f | 850 | if (s->flags & SLAB_RED_ZONE) |
8669dbab | 851 | print_section(KERN_ERR, "Redzone ", p + s->object_size, |
3b0efdfa | 852 | s->inuse - s->object_size); |
81819f0f | 853 | |
cbfc35a4 | 854 | off = get_info_end(s); |
81819f0f | 855 | |
24922684 | 856 | if (s->flags & SLAB_STORE_USER) |
81819f0f | 857 | off += 2 * sizeof(struct track); |
81819f0f | 858 | |
80a9201a AP |
859 | off += kasan_metadata_size(s); |
860 | ||
d86bd1be | 861 | if (off != size_from_object(s)) |
81819f0f | 862 | /* Beginning of the filler is the free pointer */ |
8669dbab | 863 | print_section(KERN_ERR, "Padding ", p + off, |
aa2efd5e | 864 | size_from_object(s) - off); |
24922684 CL |
865 | |
866 | dump_stack(); | |
81819f0f CL |
867 | } |
868 | ||
ae16d059 | 869 | static void object_err(struct kmem_cache *s, struct page *page, |
81819f0f CL |
870 | u8 *object, char *reason) |
871 | { | |
1f9f78b1 OG |
872 | if (slab_add_kunit_errors()) |
873 | return; | |
874 | ||
3dc50637 | 875 | slab_bug(s, "%s", reason); |
24922684 | 876 | print_trailer(s, page, object); |
65ebdeef | 877 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); |
81819f0f CL |
878 | } |
879 | ||
ae16d059 VB |
880 | static bool freelist_corrupted(struct kmem_cache *s, struct page *page, |
881 | void **freelist, void *nextfree) | |
882 | { | |
883 | if ((s->flags & SLAB_CONSISTENCY_CHECKS) && | |
884 | !check_valid_pointer(s, page, nextfree) && freelist) { | |
885 | object_err(s, page, *freelist, "Freechain corrupt"); | |
886 | *freelist = NULL; | |
887 | slab_fix(s, "Isolate corrupted freechain"); | |
888 | return true; | |
889 | } | |
890 | ||
891 | return false; | |
892 | } | |
893 | ||
a38965bf | 894 | static __printf(3, 4) void slab_err(struct kmem_cache *s, struct page *page, |
d0e0ac97 | 895 | const char *fmt, ...) |
81819f0f CL |
896 | { |
897 | va_list args; | |
898 | char buf[100]; | |
899 | ||
1f9f78b1 OG |
900 | if (slab_add_kunit_errors()) |
901 | return; | |
902 | ||
24922684 CL |
903 | va_start(args, fmt); |
904 | vsnprintf(buf, sizeof(buf), fmt, args); | |
81819f0f | 905 | va_end(args); |
3dc50637 | 906 | slab_bug(s, "%s", buf); |
fb012e27 | 907 | print_slab_info(page_slab(page)); |
81819f0f | 908 | dump_stack(); |
65ebdeef | 909 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); |
81819f0f CL |
910 | } |
911 | ||
f7cb1933 | 912 | static void init_object(struct kmem_cache *s, void *object, u8 val) |
81819f0f | 913 | { |
aa1ef4d7 | 914 | u8 *p = kasan_reset_tag(object); |
81819f0f | 915 | |
d86bd1be JK |
916 | if (s->flags & SLAB_RED_ZONE) |
917 | memset(p - s->red_left_pad, val, s->red_left_pad); | |
918 | ||
81819f0f | 919 | if (s->flags & __OBJECT_POISON) { |
3b0efdfa CL |
920 | memset(p, POISON_FREE, s->object_size - 1); |
921 | p[s->object_size - 1] = POISON_END; | |
81819f0f CL |
922 | } |
923 | ||
924 | if (s->flags & SLAB_RED_ZONE) | |
3b0efdfa | 925 | memset(p + s->object_size, val, s->inuse - s->object_size); |
81819f0f CL |
926 | } |
927 | ||
24922684 CL |
928 | static void restore_bytes(struct kmem_cache *s, char *message, u8 data, |
929 | void *from, void *to) | |
930 | { | |
582d1212 | 931 | slab_fix(s, "Restoring %s 0x%p-0x%p=0x%x", message, from, to - 1, data); |
24922684 CL |
932 | memset(from, data, to - from); |
933 | } | |
934 | ||
935 | static int check_bytes_and_report(struct kmem_cache *s, struct page *page, | |
936 | u8 *object, char *what, | |
06428780 | 937 | u8 *start, unsigned int value, unsigned int bytes) |
24922684 CL |
938 | { |
939 | u8 *fault; | |
940 | u8 *end; | |
e1b70dd1 | 941 | u8 *addr = page_address(page); |
24922684 | 942 | |
a79316c6 | 943 | metadata_access_enable(); |
aa1ef4d7 | 944 | fault = memchr_inv(kasan_reset_tag(start), value, bytes); |
a79316c6 | 945 | metadata_access_disable(); |
24922684 CL |
946 | if (!fault) |
947 | return 1; | |
948 | ||
949 | end = start + bytes; | |
950 | while (end > fault && end[-1] == value) | |
951 | end--; | |
952 | ||
1f9f78b1 OG |
953 | if (slab_add_kunit_errors()) |
954 | goto skip_bug_print; | |
955 | ||
24922684 | 956 | slab_bug(s, "%s overwritten", what); |
96b94abc | 957 | pr_err("0x%p-0x%p @offset=%tu. First byte 0x%x instead of 0x%x\n", |
e1b70dd1 MC |
958 | fault, end - 1, fault - addr, |
959 | fault[0], value); | |
24922684 | 960 | print_trailer(s, page, object); |
65ebdeef | 961 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); |
24922684 | 962 | |
1f9f78b1 | 963 | skip_bug_print: |
24922684 CL |
964 | restore_bytes(s, what, value, fault, end); |
965 | return 0; | |
81819f0f CL |
966 | } |
967 | ||
81819f0f CL |
968 | /* |
969 | * Object layout: | |
970 | * | |
971 | * object address | |
972 | * Bytes of the object to be managed. | |
973 | * If the freepointer may overlay the object then the free | |
cbfc35a4 | 974 | * pointer is at the middle of the object. |
672bba3a | 975 | * |
81819f0f CL |
976 | * Poisoning uses 0x6b (POISON_FREE) and the last byte is |
977 | * 0xa5 (POISON_END) | |
978 | * | |
3b0efdfa | 979 | * object + s->object_size |
81819f0f | 980 | * Padding to reach word boundary. This is also used for Redzoning. |
672bba3a | 981 | * Padding is extended by another word if Redzoning is enabled and |
3b0efdfa | 982 | * object_size == inuse. |
672bba3a | 983 | * |
81819f0f CL |
984 | * We fill with 0xbb (RED_INACTIVE) for inactive objects and with |
985 | * 0xcc (RED_ACTIVE) for objects in use. | |
986 | * | |
987 | * object + s->inuse | |
672bba3a CL |
988 | * Meta data starts here. |
989 | * | |
81819f0f CL |
990 | * A. Free pointer (if we cannot overwrite object on free) |
991 | * B. Tracking data for SLAB_STORE_USER | |
dc84207d | 992 | * C. Padding to reach required alignment boundary or at minimum |
6446faa2 | 993 | * one word if debugging is on to be able to detect writes |
672bba3a CL |
994 | * before the word boundary. |
995 | * | |
996 | * Padding is done using 0x5a (POISON_INUSE) | |
81819f0f CL |
997 | * |
998 | * object + s->size | |
672bba3a | 999 | * Nothing is used beyond s->size. |
81819f0f | 1000 | * |
3b0efdfa | 1001 | * If slabcaches are merged then the object_size and inuse boundaries are mostly |
672bba3a | 1002 | * ignored. And therefore no slab options that rely on these boundaries |
81819f0f CL |
1003 | * may be used with merged slabcaches. |
1004 | */ | |
1005 | ||
81819f0f CL |
1006 | static int check_pad_bytes(struct kmem_cache *s, struct page *page, u8 *p) |
1007 | { | |
cbfc35a4 | 1008 | unsigned long off = get_info_end(s); /* The end of info */ |
81819f0f CL |
1009 | |
1010 | if (s->flags & SLAB_STORE_USER) | |
1011 | /* We also have user information there */ | |
1012 | off += 2 * sizeof(struct track); | |
1013 | ||
80a9201a AP |
1014 | off += kasan_metadata_size(s); |
1015 | ||
d86bd1be | 1016 | if (size_from_object(s) == off) |
81819f0f CL |
1017 | return 1; |
1018 | ||
24922684 | 1019 | return check_bytes_and_report(s, page, p, "Object padding", |
d86bd1be | 1020 | p + off, POISON_INUSE, size_from_object(s) - off); |
81819f0f CL |
1021 | } |
1022 | ||
39b26464 | 1023 | /* Check the pad bytes at the end of a slab page */ |
81819f0f CL |
1024 | static int slab_pad_check(struct kmem_cache *s, struct page *page) |
1025 | { | |
24922684 CL |
1026 | u8 *start; |
1027 | u8 *fault; | |
1028 | u8 *end; | |
5d682681 | 1029 | u8 *pad; |
24922684 CL |
1030 | int length; |
1031 | int remainder; | |
81819f0f CL |
1032 | |
1033 | if (!(s->flags & SLAB_POISON)) | |
1034 | return 1; | |
1035 | ||
a973e9dd | 1036 | start = page_address(page); |
a50b854e | 1037 | length = page_size(page); |
39b26464 CL |
1038 | end = start + length; |
1039 | remainder = length % s->size; | |
81819f0f CL |
1040 | if (!remainder) |
1041 | return 1; | |
1042 | ||
5d682681 | 1043 | pad = end - remainder; |
a79316c6 | 1044 | metadata_access_enable(); |
aa1ef4d7 | 1045 | fault = memchr_inv(kasan_reset_tag(pad), POISON_INUSE, remainder); |
a79316c6 | 1046 | metadata_access_disable(); |
24922684 CL |
1047 | if (!fault) |
1048 | return 1; | |
1049 | while (end > fault && end[-1] == POISON_INUSE) | |
1050 | end--; | |
1051 | ||
e1b70dd1 MC |
1052 | slab_err(s, page, "Padding overwritten. 0x%p-0x%p @offset=%tu", |
1053 | fault, end - 1, fault - start); | |
5d682681 | 1054 | print_section(KERN_ERR, "Padding ", pad, remainder); |
24922684 | 1055 | |
5d682681 | 1056 | restore_bytes(s, "slab padding", POISON_INUSE, fault, end); |
24922684 | 1057 | return 0; |
81819f0f CL |
1058 | } |
1059 | ||
1060 | static int check_object(struct kmem_cache *s, struct page *page, | |
f7cb1933 | 1061 | void *object, u8 val) |
81819f0f CL |
1062 | { |
1063 | u8 *p = object; | |
3b0efdfa | 1064 | u8 *endobject = object + s->object_size; |
81819f0f CL |
1065 | |
1066 | if (s->flags & SLAB_RED_ZONE) { | |
8669dbab | 1067 | if (!check_bytes_and_report(s, page, object, "Left Redzone", |
d86bd1be JK |
1068 | object - s->red_left_pad, val, s->red_left_pad)) |
1069 | return 0; | |
1070 | ||
8669dbab | 1071 | if (!check_bytes_and_report(s, page, object, "Right Redzone", |
3b0efdfa | 1072 | endobject, val, s->inuse - s->object_size)) |
81819f0f | 1073 | return 0; |
81819f0f | 1074 | } else { |
3b0efdfa | 1075 | if ((s->flags & SLAB_POISON) && s->object_size < s->inuse) { |
3adbefee | 1076 | check_bytes_and_report(s, page, p, "Alignment padding", |
d0e0ac97 CG |
1077 | endobject, POISON_INUSE, |
1078 | s->inuse - s->object_size); | |
3adbefee | 1079 | } |
81819f0f CL |
1080 | } |
1081 | ||
1082 | if (s->flags & SLAB_POISON) { | |
f7cb1933 | 1083 | if (val != SLUB_RED_ACTIVE && (s->flags & __OBJECT_POISON) && |
24922684 | 1084 | (!check_bytes_and_report(s, page, p, "Poison", p, |
3b0efdfa | 1085 | POISON_FREE, s->object_size - 1) || |
8669dbab | 1086 | !check_bytes_and_report(s, page, p, "End Poison", |
3b0efdfa | 1087 | p + s->object_size - 1, POISON_END, 1))) |
81819f0f | 1088 | return 0; |
81819f0f CL |
1089 | /* |
1090 | * check_pad_bytes cleans up on its own. | |
1091 | */ | |
1092 | check_pad_bytes(s, page, p); | |
1093 | } | |
1094 | ||
cbfc35a4 | 1095 | if (!freeptr_outside_object(s) && val == SLUB_RED_ACTIVE) |
81819f0f CL |
1096 | /* |
1097 | * Object and freepointer overlap. Cannot check | |
1098 | * freepointer while object is allocated. | |
1099 | */ | |
1100 | return 1; | |
1101 | ||
1102 | /* Check free pointer validity */ | |
1103 | if (!check_valid_pointer(s, page, get_freepointer(s, p))) { | |
1104 | object_err(s, page, p, "Freepointer corrupt"); | |
1105 | /* | |
9f6c708e | 1106 | * No choice but to zap it and thus lose the remainder |
81819f0f | 1107 | * of the free objects in this slab. May cause |
672bba3a | 1108 | * another error because the object count is now wrong. |
81819f0f | 1109 | */ |
a973e9dd | 1110 | set_freepointer(s, p, NULL); |
81819f0f CL |
1111 | return 0; |
1112 | } | |
1113 | return 1; | |
1114 | } | |
1115 | ||
1116 | static int check_slab(struct kmem_cache *s, struct page *page) | |
1117 | { | |
39b26464 CL |
1118 | int maxobj; |
1119 | ||
81819f0f | 1120 | if (!PageSlab(page)) { |
24922684 | 1121 | slab_err(s, page, "Not a valid slab page"); |
81819f0f CL |
1122 | return 0; |
1123 | } | |
39b26464 | 1124 | |
9736d2a9 | 1125 | maxobj = order_objects(compound_order(page), s->size); |
39b26464 CL |
1126 | if (page->objects > maxobj) { |
1127 | slab_err(s, page, "objects %u > max %u", | |
f6edde9c | 1128 | page->objects, maxobj); |
39b26464 CL |
1129 | return 0; |
1130 | } | |
1131 | if (page->inuse > page->objects) { | |
24922684 | 1132 | slab_err(s, page, "inuse %u > max %u", |
f6edde9c | 1133 | page->inuse, page->objects); |
81819f0f CL |
1134 | return 0; |
1135 | } | |
1136 | /* Slab_pad_check fixes things up after itself */ | |
1137 | slab_pad_check(s, page); | |
1138 | return 1; | |
1139 | } | |
1140 | ||
1141 | /* | |
672bba3a CL |
1142 | * Determine if a certain object on a page is on the freelist. Must hold the |
1143 | * slab lock to guarantee that the chains are in a consistent state. | |
81819f0f CL |
1144 | */ |
1145 | static int on_freelist(struct kmem_cache *s, struct page *page, void *search) | |
1146 | { | |
1147 | int nr = 0; | |
881db7fb | 1148 | void *fp; |
81819f0f | 1149 | void *object = NULL; |
f6edde9c | 1150 | int max_objects; |
81819f0f | 1151 | |
881db7fb | 1152 | fp = page->freelist; |
39b26464 | 1153 | while (fp && nr <= page->objects) { |
81819f0f CL |
1154 | if (fp == search) |
1155 | return 1; | |
1156 | if (!check_valid_pointer(s, page, fp)) { | |
1157 | if (object) { | |
1158 | object_err(s, page, object, | |
1159 | "Freechain corrupt"); | |
a973e9dd | 1160 | set_freepointer(s, object, NULL); |
81819f0f | 1161 | } else { |
24922684 | 1162 | slab_err(s, page, "Freepointer corrupt"); |
a973e9dd | 1163 | page->freelist = NULL; |
39b26464 | 1164 | page->inuse = page->objects; |
24922684 | 1165 | slab_fix(s, "Freelist cleared"); |
81819f0f CL |
1166 | return 0; |
1167 | } | |
1168 | break; | |
1169 | } | |
1170 | object = fp; | |
1171 | fp = get_freepointer(s, object); | |
1172 | nr++; | |
1173 | } | |
1174 | ||
9736d2a9 | 1175 | max_objects = order_objects(compound_order(page), s->size); |
210b5c06 CG |
1176 | if (max_objects > MAX_OBJS_PER_PAGE) |
1177 | max_objects = MAX_OBJS_PER_PAGE; | |
224a88be CL |
1178 | |
1179 | if (page->objects != max_objects) { | |
756a025f JP |
1180 | slab_err(s, page, "Wrong number of objects. Found %d but should be %d", |
1181 | page->objects, max_objects); | |
224a88be | 1182 | page->objects = max_objects; |
582d1212 | 1183 | slab_fix(s, "Number of objects adjusted"); |
224a88be | 1184 | } |
39b26464 | 1185 | if (page->inuse != page->objects - nr) { |
756a025f JP |
1186 | slab_err(s, page, "Wrong object count. Counter is %d but counted were %d", |
1187 | page->inuse, page->objects - nr); | |
39b26464 | 1188 | page->inuse = page->objects - nr; |
582d1212 | 1189 | slab_fix(s, "Object count adjusted"); |
81819f0f CL |
1190 | } |
1191 | return search == NULL; | |
1192 | } | |
1193 | ||
0121c619 CL |
1194 | static void trace(struct kmem_cache *s, struct page *page, void *object, |
1195 | int alloc) | |
3ec09742 CL |
1196 | { |
1197 | if (s->flags & SLAB_TRACE) { | |
f9f58285 | 1198 | pr_info("TRACE %s %s 0x%p inuse=%d fp=0x%p\n", |
3ec09742 CL |
1199 | s->name, |
1200 | alloc ? "alloc" : "free", | |
1201 | object, page->inuse, | |
1202 | page->freelist); | |
1203 | ||
1204 | if (!alloc) | |
aa2efd5e | 1205 | print_section(KERN_INFO, "Object ", (void *)object, |
d0e0ac97 | 1206 | s->object_size); |
3ec09742 CL |
1207 | |
1208 | dump_stack(); | |
1209 | } | |
1210 | } | |
1211 | ||
643b1138 | 1212 | /* |
672bba3a | 1213 | * Tracking of fully allocated slabs for debugging purposes. |
643b1138 | 1214 | */ |
5cc6eee8 CL |
1215 | static void add_full(struct kmem_cache *s, |
1216 | struct kmem_cache_node *n, struct page *page) | |
643b1138 | 1217 | { |
5cc6eee8 CL |
1218 | if (!(s->flags & SLAB_STORE_USER)) |
1219 | return; | |
1220 | ||
255d0884 | 1221 | lockdep_assert_held(&n->list_lock); |
916ac052 | 1222 | list_add(&page->slab_list, &n->full); |
643b1138 CL |
1223 | } |
1224 | ||
c65c1877 | 1225 | static void remove_full(struct kmem_cache *s, struct kmem_cache_node *n, struct page *page) |
643b1138 | 1226 | { |
643b1138 CL |
1227 | if (!(s->flags & SLAB_STORE_USER)) |
1228 | return; | |
1229 | ||
255d0884 | 1230 | lockdep_assert_held(&n->list_lock); |
916ac052 | 1231 | list_del(&page->slab_list); |
643b1138 CL |
1232 | } |
1233 | ||
0f389ec6 CL |
1234 | /* Tracking of the number of slabs for debugging purposes */ |
1235 | static inline unsigned long slabs_node(struct kmem_cache *s, int node) | |
1236 | { | |
1237 | struct kmem_cache_node *n = get_node(s, node); | |
1238 | ||
1239 | return atomic_long_read(&n->nr_slabs); | |
1240 | } | |
1241 | ||
26c02cf0 AB |
1242 | static inline unsigned long node_nr_slabs(struct kmem_cache_node *n) |
1243 | { | |
1244 | return atomic_long_read(&n->nr_slabs); | |
1245 | } | |
1246 | ||
205ab99d | 1247 | static inline void inc_slabs_node(struct kmem_cache *s, int node, int objects) |
0f389ec6 CL |
1248 | { |
1249 | struct kmem_cache_node *n = get_node(s, node); | |
1250 | ||
1251 | /* | |
1252 | * May be called early in order to allocate a slab for the | |
1253 | * kmem_cache_node structure. Solve the chicken-egg | |
1254 | * dilemma by deferring the increment of the count during | |
1255 | * bootstrap (see early_kmem_cache_node_alloc). | |
1256 | */ | |
338b2642 | 1257 | if (likely(n)) { |
0f389ec6 | 1258 | atomic_long_inc(&n->nr_slabs); |
205ab99d CL |
1259 | atomic_long_add(objects, &n->total_objects); |
1260 | } | |
0f389ec6 | 1261 | } |
205ab99d | 1262 | static inline void dec_slabs_node(struct kmem_cache *s, int node, int objects) |
0f389ec6 CL |
1263 | { |
1264 | struct kmem_cache_node *n = get_node(s, node); | |
1265 | ||
1266 | atomic_long_dec(&n->nr_slabs); | |
205ab99d | 1267 | atomic_long_sub(objects, &n->total_objects); |
0f389ec6 CL |
1268 | } |
1269 | ||
1270 | /* Object debug checks for alloc/free paths */ | |
3ec09742 CL |
1271 | static void setup_object_debug(struct kmem_cache *s, struct page *page, |
1272 | void *object) | |
1273 | { | |
8fc8d666 | 1274 | if (!kmem_cache_debug_flags(s, SLAB_STORE_USER|SLAB_RED_ZONE|__OBJECT_POISON)) |
3ec09742 CL |
1275 | return; |
1276 | ||
f7cb1933 | 1277 | init_object(s, object, SLUB_RED_INACTIVE); |
3ec09742 CL |
1278 | init_tracking(s, object); |
1279 | } | |
1280 | ||
a50b854e MWO |
1281 | static |
1282 | void setup_page_debug(struct kmem_cache *s, struct page *page, void *addr) | |
a7101224 | 1283 | { |
8fc8d666 | 1284 | if (!kmem_cache_debug_flags(s, SLAB_POISON)) |
a7101224 AK |
1285 | return; |
1286 | ||
1287 | metadata_access_enable(); | |
aa1ef4d7 | 1288 | memset(kasan_reset_tag(addr), POISON_INUSE, page_size(page)); |
a7101224 AK |
1289 | metadata_access_disable(); |
1290 | } | |
1291 | ||
becfda68 | 1292 | static inline int alloc_consistency_checks(struct kmem_cache *s, |
278d7756 | 1293 | struct page *page, void *object) |
81819f0f CL |
1294 | { |
1295 | if (!check_slab(s, page)) | |
becfda68 | 1296 | return 0; |
81819f0f | 1297 | |
81819f0f CL |
1298 | if (!check_valid_pointer(s, page, object)) { |
1299 | object_err(s, page, object, "Freelist Pointer check fails"); | |
becfda68 | 1300 | return 0; |
81819f0f CL |
1301 | } |
1302 | ||
f7cb1933 | 1303 | if (!check_object(s, page, object, SLUB_RED_INACTIVE)) |
becfda68 LA |
1304 | return 0; |
1305 | ||
1306 | return 1; | |
1307 | } | |
1308 | ||
1309 | static noinline int alloc_debug_processing(struct kmem_cache *s, | |
1310 | struct page *page, | |
1311 | void *object, unsigned long addr) | |
1312 | { | |
1313 | if (s->flags & SLAB_CONSISTENCY_CHECKS) { | |
278d7756 | 1314 | if (!alloc_consistency_checks(s, page, object)) |
becfda68 LA |
1315 | goto bad; |
1316 | } | |
81819f0f | 1317 | |
3ec09742 CL |
1318 | /* Success perform special debug activities for allocs */ |
1319 | if (s->flags & SLAB_STORE_USER) | |
1320 | set_track(s, object, TRACK_ALLOC, addr); | |
1321 | trace(s, page, object, 1); | |
f7cb1933 | 1322 | init_object(s, object, SLUB_RED_ACTIVE); |
81819f0f | 1323 | return 1; |
3ec09742 | 1324 | |
81819f0f CL |
1325 | bad: |
1326 | if (PageSlab(page)) { | |
1327 | /* | |
1328 | * If this is a slab page then lets do the best we can | |
1329 | * to avoid issues in the future. Marking all objects | |
672bba3a | 1330 | * as used avoids touching the remaining objects. |
81819f0f | 1331 | */ |
24922684 | 1332 | slab_fix(s, "Marking all objects used"); |
39b26464 | 1333 | page->inuse = page->objects; |
a973e9dd | 1334 | page->freelist = NULL; |
81819f0f CL |
1335 | } |
1336 | return 0; | |
1337 | } | |
1338 | ||
becfda68 LA |
1339 | static inline int free_consistency_checks(struct kmem_cache *s, |
1340 | struct page *page, void *object, unsigned long addr) | |
81819f0f | 1341 | { |
81819f0f | 1342 | if (!check_valid_pointer(s, page, object)) { |
70d71228 | 1343 | slab_err(s, page, "Invalid object pointer 0x%p", object); |
becfda68 | 1344 | return 0; |
81819f0f CL |
1345 | } |
1346 | ||
1347 | if (on_freelist(s, page, object)) { | |
24922684 | 1348 | object_err(s, page, object, "Object already free"); |
becfda68 | 1349 | return 0; |
81819f0f CL |
1350 | } |
1351 | ||
f7cb1933 | 1352 | if (!check_object(s, page, object, SLUB_RED_ACTIVE)) |
becfda68 | 1353 | return 0; |
81819f0f | 1354 | |
1b4f59e3 | 1355 | if (unlikely(s != page->slab_cache)) { |
3adbefee | 1356 | if (!PageSlab(page)) { |
756a025f JP |
1357 | slab_err(s, page, "Attempt to free object(0x%p) outside of slab", |
1358 | object); | |
1b4f59e3 | 1359 | } else if (!page->slab_cache) { |
f9f58285 FF |
1360 | pr_err("SLUB <none>: no slab for object 0x%p.\n", |
1361 | object); | |
70d71228 | 1362 | dump_stack(); |
06428780 | 1363 | } else |
24922684 CL |
1364 | object_err(s, page, object, |
1365 | "page slab pointer corrupt."); | |
becfda68 LA |
1366 | return 0; |
1367 | } | |
1368 | return 1; | |
1369 | } | |
1370 | ||
1371 | /* Supports checking bulk free of a constructed freelist */ | |
1372 | static noinline int free_debug_processing( | |
1373 | struct kmem_cache *s, struct page *page, | |
1374 | void *head, void *tail, int bulk_cnt, | |
1375 | unsigned long addr) | |
1376 | { | |
1377 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); | |
1378 | void *object = head; | |
1379 | int cnt = 0; | |
a2b4ae8b | 1380 | unsigned long flags, flags2; |
becfda68 LA |
1381 | int ret = 0; |
1382 | ||
1383 | spin_lock_irqsave(&n->list_lock, flags); | |
a2b4ae8b | 1384 | slab_lock(page, &flags2); |
becfda68 LA |
1385 | |
1386 | if (s->flags & SLAB_CONSISTENCY_CHECKS) { | |
1387 | if (!check_slab(s, page)) | |
1388 | goto out; | |
1389 | } | |
1390 | ||
1391 | next_object: | |
1392 | cnt++; | |
1393 | ||
1394 | if (s->flags & SLAB_CONSISTENCY_CHECKS) { | |
1395 | if (!free_consistency_checks(s, page, object, addr)) | |
1396 | goto out; | |
81819f0f | 1397 | } |
3ec09742 | 1398 | |
3ec09742 CL |
1399 | if (s->flags & SLAB_STORE_USER) |
1400 | set_track(s, object, TRACK_FREE, addr); | |
1401 | trace(s, page, object, 0); | |
81084651 | 1402 | /* Freepointer not overwritten by init_object(), SLAB_POISON moved it */ |
f7cb1933 | 1403 | init_object(s, object, SLUB_RED_INACTIVE); |
81084651 JDB |
1404 | |
1405 | /* Reached end of constructed freelist yet? */ | |
1406 | if (object != tail) { | |
1407 | object = get_freepointer(s, object); | |
1408 | goto next_object; | |
1409 | } | |
804aa132 LA |
1410 | ret = 1; |
1411 | ||
5c2e4bbb | 1412 | out: |
81084651 JDB |
1413 | if (cnt != bulk_cnt) |
1414 | slab_err(s, page, "Bulk freelist count(%d) invalid(%d)\n", | |
1415 | bulk_cnt, cnt); | |
1416 | ||
a2b4ae8b | 1417 | slab_unlock(page, &flags2); |
282acb43 | 1418 | spin_unlock_irqrestore(&n->list_lock, flags); |
804aa132 LA |
1419 | if (!ret) |
1420 | slab_fix(s, "Object at 0x%p not freed", object); | |
1421 | return ret; | |
81819f0f CL |
1422 | } |
1423 | ||
e17f1dfb VB |
1424 | /* |
1425 | * Parse a block of slub_debug options. Blocks are delimited by ';' | |
1426 | * | |
1427 | * @str: start of block | |
1428 | * @flags: returns parsed flags, or DEBUG_DEFAULT_FLAGS if none specified | |
1429 | * @slabs: return start of list of slabs, or NULL when there's no list | |
1430 | * @init: assume this is initial parsing and not per-kmem-create parsing | |
1431 | * | |
1432 | * returns the start of next block if there's any, or NULL | |
1433 | */ | |
1434 | static char * | |
1435 | parse_slub_debug_flags(char *str, slab_flags_t *flags, char **slabs, bool init) | |
41ecc55b | 1436 | { |
e17f1dfb | 1437 | bool higher_order_disable = false; |
f0630fff | 1438 | |
e17f1dfb VB |
1439 | /* Skip any completely empty blocks */ |
1440 | while (*str && *str == ';') | |
1441 | str++; | |
1442 | ||
1443 | if (*str == ',') { | |
f0630fff CL |
1444 | /* |
1445 | * No options but restriction on slabs. This means full | |
1446 | * debugging for slabs matching a pattern. | |
1447 | */ | |
e17f1dfb | 1448 | *flags = DEBUG_DEFAULT_FLAGS; |
f0630fff | 1449 | goto check_slabs; |
e17f1dfb VB |
1450 | } |
1451 | *flags = 0; | |
f0630fff | 1452 | |
e17f1dfb VB |
1453 | /* Determine which debug features should be switched on */ |
1454 | for (; *str && *str != ',' && *str != ';'; str++) { | |
f0630fff | 1455 | switch (tolower(*str)) { |
e17f1dfb VB |
1456 | case '-': |
1457 | *flags = 0; | |
1458 | break; | |
f0630fff | 1459 | case 'f': |
e17f1dfb | 1460 | *flags |= SLAB_CONSISTENCY_CHECKS; |
f0630fff CL |
1461 | break; |
1462 | case 'z': | |
e17f1dfb | 1463 | *flags |= SLAB_RED_ZONE; |
f0630fff CL |
1464 | break; |
1465 | case 'p': | |
e17f1dfb | 1466 | *flags |= SLAB_POISON; |
f0630fff CL |
1467 | break; |
1468 | case 'u': | |
e17f1dfb | 1469 | *flags |= SLAB_STORE_USER; |
f0630fff CL |
1470 | break; |
1471 | case 't': | |
e17f1dfb | 1472 | *flags |= SLAB_TRACE; |
f0630fff | 1473 | break; |
4c13dd3b | 1474 | case 'a': |
e17f1dfb | 1475 | *flags |= SLAB_FAILSLAB; |
4c13dd3b | 1476 | break; |
08303a73 CA |
1477 | case 'o': |
1478 | /* | |
1479 | * Avoid enabling debugging on caches if its minimum | |
1480 | * order would increase as a result. | |
1481 | */ | |
e17f1dfb | 1482 | higher_order_disable = true; |
08303a73 | 1483 | break; |
f0630fff | 1484 | default: |
e17f1dfb VB |
1485 | if (init) |
1486 | pr_err("slub_debug option '%c' unknown. skipped\n", *str); | |
f0630fff | 1487 | } |
41ecc55b | 1488 | } |
f0630fff | 1489 | check_slabs: |
41ecc55b | 1490 | if (*str == ',') |
e17f1dfb VB |
1491 | *slabs = ++str; |
1492 | else | |
1493 | *slabs = NULL; | |
1494 | ||
1495 | /* Skip over the slab list */ | |
1496 | while (*str && *str != ';') | |
1497 | str++; | |
1498 | ||
1499 | /* Skip any completely empty blocks */ | |
1500 | while (*str && *str == ';') | |
1501 | str++; | |
1502 | ||
1503 | if (init && higher_order_disable) | |
1504 | disable_higher_order_debug = 1; | |
1505 | ||
1506 | if (*str) | |
1507 | return str; | |
1508 | else | |
1509 | return NULL; | |
1510 | } | |
1511 | ||
1512 | static int __init setup_slub_debug(char *str) | |
1513 | { | |
1514 | slab_flags_t flags; | |
a7f1d485 | 1515 | slab_flags_t global_flags; |
e17f1dfb VB |
1516 | char *saved_str; |
1517 | char *slab_list; | |
1518 | bool global_slub_debug_changed = false; | |
1519 | bool slab_list_specified = false; | |
1520 | ||
a7f1d485 | 1521 | global_flags = DEBUG_DEFAULT_FLAGS; |
e17f1dfb VB |
1522 | if (*str++ != '=' || !*str) |
1523 | /* | |
1524 | * No options specified. Switch on full debugging. | |
1525 | */ | |
1526 | goto out; | |
1527 | ||
1528 | saved_str = str; | |
1529 | while (str) { | |
1530 | str = parse_slub_debug_flags(str, &flags, &slab_list, true); | |
1531 | ||
1532 | if (!slab_list) { | |
a7f1d485 | 1533 | global_flags = flags; |
e17f1dfb VB |
1534 | global_slub_debug_changed = true; |
1535 | } else { | |
1536 | slab_list_specified = true; | |
1537 | } | |
1538 | } | |
1539 | ||
1540 | /* | |
1541 | * For backwards compatibility, a single list of flags with list of | |
a7f1d485 VB |
1542 | * slabs means debugging is only changed for those slabs, so the global |
1543 | * slub_debug should be unchanged (0 or DEBUG_DEFAULT_FLAGS, depending | |
1544 | * on CONFIG_SLUB_DEBUG_ON). We can extended that to multiple lists as | |
e17f1dfb VB |
1545 | * long as there is no option specifying flags without a slab list. |
1546 | */ | |
1547 | if (slab_list_specified) { | |
1548 | if (!global_slub_debug_changed) | |
a7f1d485 | 1549 | global_flags = slub_debug; |
e17f1dfb VB |
1550 | slub_debug_string = saved_str; |
1551 | } | |
f0630fff | 1552 | out: |
a7f1d485 | 1553 | slub_debug = global_flags; |
ca0cab65 VB |
1554 | if (slub_debug != 0 || slub_debug_string) |
1555 | static_branch_enable(&slub_debug_enabled); | |
02ac47d0 SB |
1556 | else |
1557 | static_branch_disable(&slub_debug_enabled); | |
6471384a AP |
1558 | if ((static_branch_unlikely(&init_on_alloc) || |
1559 | static_branch_unlikely(&init_on_free)) && | |
1560 | (slub_debug & SLAB_POISON)) | |
1561 | pr_info("mem auto-init: SLAB_POISON will take precedence over init_on_alloc/init_on_free\n"); | |
41ecc55b CL |
1562 | return 1; |
1563 | } | |
1564 | ||
1565 | __setup("slub_debug", setup_slub_debug); | |
1566 | ||
c5fd3ca0 AT |
1567 | /* |
1568 | * kmem_cache_flags - apply debugging options to the cache | |
1569 | * @object_size: the size of an object without meta data | |
1570 | * @flags: flags to set | |
1571 | * @name: name of the cache | |
c5fd3ca0 AT |
1572 | * |
1573 | * Debug option(s) are applied to @flags. In addition to the debug | |
1574 | * option(s), if a slab name (or multiple) is specified i.e. | |
1575 | * slub_debug=<Debug-Options>,<slab name1>,<slab name2> ... | |
1576 | * then only the select slabs will receive the debug option(s). | |
1577 | */ | |
0293d1fd | 1578 | slab_flags_t kmem_cache_flags(unsigned int object_size, |
37540008 | 1579 | slab_flags_t flags, const char *name) |
41ecc55b | 1580 | { |
c5fd3ca0 AT |
1581 | char *iter; |
1582 | size_t len; | |
e17f1dfb VB |
1583 | char *next_block; |
1584 | slab_flags_t block_flags; | |
ca220593 JB |
1585 | slab_flags_t slub_debug_local = slub_debug; |
1586 | ||
1587 | /* | |
1588 | * If the slab cache is for debugging (e.g. kmemleak) then | |
1589 | * don't store user (stack trace) information by default, | |
1590 | * but let the user enable it via the command line below. | |
1591 | */ | |
1592 | if (flags & SLAB_NOLEAKTRACE) | |
1593 | slub_debug_local &= ~SLAB_STORE_USER; | |
c5fd3ca0 | 1594 | |
c5fd3ca0 | 1595 | len = strlen(name); |
e17f1dfb VB |
1596 | next_block = slub_debug_string; |
1597 | /* Go through all blocks of debug options, see if any matches our slab's name */ | |
1598 | while (next_block) { | |
1599 | next_block = parse_slub_debug_flags(next_block, &block_flags, &iter, false); | |
1600 | if (!iter) | |
1601 | continue; | |
1602 | /* Found a block that has a slab list, search it */ | |
1603 | while (*iter) { | |
1604 | char *end, *glob; | |
1605 | size_t cmplen; | |
1606 | ||
1607 | end = strchrnul(iter, ','); | |
1608 | if (next_block && next_block < end) | |
1609 | end = next_block - 1; | |
1610 | ||
1611 | glob = strnchr(iter, end - iter, '*'); | |
1612 | if (glob) | |
1613 | cmplen = glob - iter; | |
1614 | else | |
1615 | cmplen = max_t(size_t, len, (end - iter)); | |
c5fd3ca0 | 1616 | |
e17f1dfb VB |
1617 | if (!strncmp(name, iter, cmplen)) { |
1618 | flags |= block_flags; | |
1619 | return flags; | |
1620 | } | |
c5fd3ca0 | 1621 | |
e17f1dfb VB |
1622 | if (!*end || *end == ';') |
1623 | break; | |
1624 | iter = end + 1; | |
c5fd3ca0 | 1625 | } |
c5fd3ca0 | 1626 | } |
ba0268a8 | 1627 | |
ca220593 | 1628 | return flags | slub_debug_local; |
41ecc55b | 1629 | } |
b4a64718 | 1630 | #else /* !CONFIG_SLUB_DEBUG */ |
3ec09742 CL |
1631 | static inline void setup_object_debug(struct kmem_cache *s, |
1632 | struct page *page, void *object) {} | |
a50b854e MWO |
1633 | static inline |
1634 | void setup_page_debug(struct kmem_cache *s, struct page *page, void *addr) {} | |
41ecc55b | 1635 | |
3ec09742 | 1636 | static inline int alloc_debug_processing(struct kmem_cache *s, |
ce71e27c | 1637 | struct page *page, void *object, unsigned long addr) { return 0; } |
41ecc55b | 1638 | |
282acb43 | 1639 | static inline int free_debug_processing( |
81084651 JDB |
1640 | struct kmem_cache *s, struct page *page, |
1641 | void *head, void *tail, int bulk_cnt, | |
282acb43 | 1642 | unsigned long addr) { return 0; } |
41ecc55b | 1643 | |
41ecc55b CL |
1644 | static inline int slab_pad_check(struct kmem_cache *s, struct page *page) |
1645 | { return 1; } | |
1646 | static inline int check_object(struct kmem_cache *s, struct page *page, | |
f7cb1933 | 1647 | void *object, u8 val) { return 1; } |
5cc6eee8 CL |
1648 | static inline void add_full(struct kmem_cache *s, struct kmem_cache_node *n, |
1649 | struct page *page) {} | |
c65c1877 PZ |
1650 | static inline void remove_full(struct kmem_cache *s, struct kmem_cache_node *n, |
1651 | struct page *page) {} | |
0293d1fd | 1652 | slab_flags_t kmem_cache_flags(unsigned int object_size, |
37540008 | 1653 | slab_flags_t flags, const char *name) |
ba0268a8 CL |
1654 | { |
1655 | return flags; | |
1656 | } | |
41ecc55b | 1657 | #define slub_debug 0 |
0f389ec6 | 1658 | |
fdaa45e9 IM |
1659 | #define disable_higher_order_debug 0 |
1660 | ||
0f389ec6 CL |
1661 | static inline unsigned long slabs_node(struct kmem_cache *s, int node) |
1662 | { return 0; } | |
26c02cf0 AB |
1663 | static inline unsigned long node_nr_slabs(struct kmem_cache_node *n) |
1664 | { return 0; } | |
205ab99d CL |
1665 | static inline void inc_slabs_node(struct kmem_cache *s, int node, |
1666 | int objects) {} | |
1667 | static inline void dec_slabs_node(struct kmem_cache *s, int node, | |
1668 | int objects) {} | |
7d550c56 | 1669 | |
52f23478 | 1670 | static bool freelist_corrupted(struct kmem_cache *s, struct page *page, |
dc07a728 | 1671 | void **freelist, void *nextfree) |
52f23478 DZ |
1672 | { |
1673 | return false; | |
1674 | } | |
02e72cc6 AR |
1675 | #endif /* CONFIG_SLUB_DEBUG */ |
1676 | ||
1677 | /* | |
1678 | * Hooks for other subsystems that check memory allocations. In a typical | |
1679 | * production configuration these hooks all should produce no code at all. | |
1680 | */ | |
0116523c | 1681 | static inline void *kmalloc_large_node_hook(void *ptr, size_t size, gfp_t flags) |
d56791b3 | 1682 | { |
53128245 | 1683 | ptr = kasan_kmalloc_large(ptr, size, flags); |
a2f77575 | 1684 | /* As ptr might get tagged, call kmemleak hook after KASAN. */ |
d56791b3 | 1685 | kmemleak_alloc(ptr, size, 1, flags); |
53128245 | 1686 | return ptr; |
d56791b3 RB |
1687 | } |
1688 | ||
ee3ce779 | 1689 | static __always_inline void kfree_hook(void *x) |
d56791b3 RB |
1690 | { |
1691 | kmemleak_free(x); | |
027b37b5 | 1692 | kasan_kfree_large(x); |
d56791b3 RB |
1693 | } |
1694 | ||
d57a964e AK |
1695 | static __always_inline bool slab_free_hook(struct kmem_cache *s, |
1696 | void *x, bool init) | |
d56791b3 RB |
1697 | { |
1698 | kmemleak_free_recursive(x, s->flags); | |
7d550c56 | 1699 | |
84048039 | 1700 | debug_check_no_locks_freed(x, s->object_size); |
02e72cc6 | 1701 | |
02e72cc6 AR |
1702 | if (!(s->flags & SLAB_DEBUG_OBJECTS)) |
1703 | debug_check_no_obj_freed(x, s->object_size); | |
0316bec2 | 1704 | |
cfbe1636 ME |
1705 | /* Use KCSAN to help debug racy use-after-free. */ |
1706 | if (!(s->flags & SLAB_TYPESAFE_BY_RCU)) | |
1707 | __kcsan_check_access(x, s->object_size, | |
1708 | KCSAN_ACCESS_WRITE | KCSAN_ACCESS_ASSERT); | |
1709 | ||
d57a964e AK |
1710 | /* |
1711 | * As memory initialization might be integrated into KASAN, | |
1712 | * kasan_slab_free and initialization memset's must be | |
1713 | * kept together to avoid discrepancies in behavior. | |
1714 | * | |
1715 | * The initialization memset's clear the object and the metadata, | |
1716 | * but don't touch the SLAB redzone. | |
1717 | */ | |
1718 | if (init) { | |
1719 | int rsize; | |
1720 | ||
1721 | if (!kasan_has_integrated_init()) | |
1722 | memset(kasan_reset_tag(x), 0, s->object_size); | |
1723 | rsize = (s->flags & SLAB_RED_ZONE) ? s->red_left_pad : 0; | |
1724 | memset((char *)kasan_reset_tag(x) + s->inuse, 0, | |
1725 | s->size - s->inuse - rsize); | |
1726 | } | |
1727 | /* KASAN might put x into memory quarantine, delaying its reuse. */ | |
1728 | return kasan_slab_free(s, x, init); | |
02e72cc6 | 1729 | } |
205ab99d | 1730 | |
c3895391 | 1731 | static inline bool slab_free_freelist_hook(struct kmem_cache *s, |
899447f6 ML |
1732 | void **head, void **tail, |
1733 | int *cnt) | |
81084651 | 1734 | { |
6471384a AP |
1735 | |
1736 | void *object; | |
1737 | void *next = *head; | |
1738 | void *old_tail = *tail ? *tail : *head; | |
6471384a | 1739 | |
b89fb5ef | 1740 | if (is_kfence_address(next)) { |
d57a964e | 1741 | slab_free_hook(s, next, false); |
b89fb5ef AP |
1742 | return true; |
1743 | } | |
1744 | ||
aea4df4c LA |
1745 | /* Head and tail of the reconstructed freelist */ |
1746 | *head = NULL; | |
1747 | *tail = NULL; | |
1b7e816f | 1748 | |
aea4df4c LA |
1749 | do { |
1750 | object = next; | |
1751 | next = get_freepointer(s, object); | |
1752 | ||
c3895391 | 1753 | /* If object's reuse doesn't have to be delayed */ |
d57a964e | 1754 | if (!slab_free_hook(s, object, slab_want_init_on_free(s))) { |
c3895391 AK |
1755 | /* Move object to the new freelist */ |
1756 | set_freepointer(s, object, *head); | |
1757 | *head = object; | |
1758 | if (!*tail) | |
1759 | *tail = object; | |
899447f6 ML |
1760 | } else { |
1761 | /* | |
1762 | * Adjust the reconstructed freelist depth | |
1763 | * accordingly if object's reuse is delayed. | |
1764 | */ | |
1765 | --(*cnt); | |
c3895391 AK |
1766 | } |
1767 | } while (object != old_tail); | |
1768 | ||
1769 | if (*head == *tail) | |
1770 | *tail = NULL; | |
1771 | ||
1772 | return *head != NULL; | |
81084651 JDB |
1773 | } |
1774 | ||
4d176711 | 1775 | static void *setup_object(struct kmem_cache *s, struct page *page, |
588f8ba9 TG |
1776 | void *object) |
1777 | { | |
1778 | setup_object_debug(s, page, object); | |
4d176711 | 1779 | object = kasan_init_slab_obj(s, object); |
588f8ba9 TG |
1780 | if (unlikely(s->ctor)) { |
1781 | kasan_unpoison_object_data(s, object); | |
1782 | s->ctor(object); | |
1783 | kasan_poison_object_data(s, object); | |
1784 | } | |
4d176711 | 1785 | return object; |
588f8ba9 TG |
1786 | } |
1787 | ||
81819f0f CL |
1788 | /* |
1789 | * Slab allocation and freeing | |
1790 | */ | |
5dfb4175 VD |
1791 | static inline struct page *alloc_slab_page(struct kmem_cache *s, |
1792 | gfp_t flags, int node, struct kmem_cache_order_objects oo) | |
65c3376a | 1793 | { |
5dfb4175 | 1794 | struct page *page; |
19af27af | 1795 | unsigned int order = oo_order(oo); |
65c3376a | 1796 | |
2154a336 | 1797 | if (node == NUMA_NO_NODE) |
5dfb4175 | 1798 | page = alloc_pages(flags, order); |
65c3376a | 1799 | else |
96db800f | 1800 | page = __alloc_pages_node(node, flags, order); |
5dfb4175 | 1801 | |
5dfb4175 | 1802 | return page; |
65c3376a CL |
1803 | } |
1804 | ||
210e7a43 TG |
1805 | #ifdef CONFIG_SLAB_FREELIST_RANDOM |
1806 | /* Pre-initialize the random sequence cache */ | |
1807 | static int init_cache_random_seq(struct kmem_cache *s) | |
1808 | { | |
19af27af | 1809 | unsigned int count = oo_objects(s->oo); |
210e7a43 | 1810 | int err; |
210e7a43 | 1811 | |
a810007a SR |
1812 | /* Bailout if already initialised */ |
1813 | if (s->random_seq) | |
1814 | return 0; | |
1815 | ||
210e7a43 TG |
1816 | err = cache_random_seq_create(s, count, GFP_KERNEL); |
1817 | if (err) { | |
1818 | pr_err("SLUB: Unable to initialize free list for %s\n", | |
1819 | s->name); | |
1820 | return err; | |
1821 | } | |
1822 | ||
1823 | /* Transform to an offset on the set of pages */ | |
1824 | if (s->random_seq) { | |
19af27af AD |
1825 | unsigned int i; |
1826 | ||
210e7a43 TG |
1827 | for (i = 0; i < count; i++) |
1828 | s->random_seq[i] *= s->size; | |
1829 | } | |
1830 | return 0; | |
1831 | } | |
1832 | ||
1833 | /* Initialize each random sequence freelist per cache */ | |
1834 | static void __init init_freelist_randomization(void) | |
1835 | { | |
1836 | struct kmem_cache *s; | |
1837 | ||
1838 | mutex_lock(&slab_mutex); | |
1839 | ||
1840 | list_for_each_entry(s, &slab_caches, list) | |
1841 | init_cache_random_seq(s); | |
1842 | ||
1843 | mutex_unlock(&slab_mutex); | |
1844 | } | |
1845 | ||
1846 | /* Get the next entry on the pre-computed freelist randomized */ | |
1847 | static void *next_freelist_entry(struct kmem_cache *s, struct page *page, | |
1848 | unsigned long *pos, void *start, | |
1849 | unsigned long page_limit, | |
1850 | unsigned long freelist_count) | |
1851 | { | |
1852 | unsigned int idx; | |
1853 | ||
1854 | /* | |
1855 | * If the target page allocation failed, the number of objects on the | |
1856 | * page might be smaller than the usual size defined by the cache. | |
1857 | */ | |
1858 | do { | |
1859 | idx = s->random_seq[*pos]; | |
1860 | *pos += 1; | |
1861 | if (*pos >= freelist_count) | |
1862 | *pos = 0; | |
1863 | } while (unlikely(idx >= page_limit)); | |
1864 | ||
1865 | return (char *)start + idx; | |
1866 | } | |
1867 | ||
1868 | /* Shuffle the single linked freelist based on a random pre-computed sequence */ | |
1869 | static bool shuffle_freelist(struct kmem_cache *s, struct page *page) | |
1870 | { | |
1871 | void *start; | |
1872 | void *cur; | |
1873 | void *next; | |
1874 | unsigned long idx, pos, page_limit, freelist_count; | |
1875 | ||
1876 | if (page->objects < 2 || !s->random_seq) | |
1877 | return false; | |
1878 | ||
1879 | freelist_count = oo_objects(s->oo); | |
1880 | pos = get_random_int() % freelist_count; | |
1881 | ||
1882 | page_limit = page->objects * s->size; | |
1883 | start = fixup_red_left(s, page_address(page)); | |
1884 | ||
1885 | /* First entry is used as the base of the freelist */ | |
1886 | cur = next_freelist_entry(s, page, &pos, start, page_limit, | |
1887 | freelist_count); | |
4d176711 | 1888 | cur = setup_object(s, page, cur); |
210e7a43 TG |
1889 | page->freelist = cur; |
1890 | ||
1891 | for (idx = 1; idx < page->objects; idx++) { | |
210e7a43 TG |
1892 | next = next_freelist_entry(s, page, &pos, start, page_limit, |
1893 | freelist_count); | |
4d176711 | 1894 | next = setup_object(s, page, next); |
210e7a43 TG |
1895 | set_freepointer(s, cur, next); |
1896 | cur = next; | |
1897 | } | |
210e7a43 TG |
1898 | set_freepointer(s, cur, NULL); |
1899 | ||
1900 | return true; | |
1901 | } | |
1902 | #else | |
1903 | static inline int init_cache_random_seq(struct kmem_cache *s) | |
1904 | { | |
1905 | return 0; | |
1906 | } | |
1907 | static inline void init_freelist_randomization(void) { } | |
1908 | static inline bool shuffle_freelist(struct kmem_cache *s, struct page *page) | |
1909 | { | |
1910 | return false; | |
1911 | } | |
1912 | #endif /* CONFIG_SLAB_FREELIST_RANDOM */ | |
1913 | ||
81819f0f CL |
1914 | static struct page *allocate_slab(struct kmem_cache *s, gfp_t flags, int node) |
1915 | { | |
06428780 | 1916 | struct page *page; |
834f3d11 | 1917 | struct kmem_cache_order_objects oo = s->oo; |
ba52270d | 1918 | gfp_t alloc_gfp; |
4d176711 | 1919 | void *start, *p, *next; |
a50b854e | 1920 | int idx; |
210e7a43 | 1921 | bool shuffle; |
81819f0f | 1922 | |
7e0528da CL |
1923 | flags &= gfp_allowed_mask; |
1924 | ||
b7a49f0d | 1925 | flags |= s->allocflags; |
e12ba74d | 1926 | |
ba52270d PE |
1927 | /* |
1928 | * Let the initial higher-order allocation fail under memory pressure | |
1929 | * so we fall-back to the minimum order allocation. | |
1930 | */ | |
1931 | alloc_gfp = (flags | __GFP_NOWARN | __GFP_NORETRY) & ~__GFP_NOFAIL; | |
d0164adc | 1932 | if ((alloc_gfp & __GFP_DIRECT_RECLAIM) && oo_order(oo) > oo_order(s->min)) |
444eb2a4 | 1933 | alloc_gfp = (alloc_gfp | __GFP_NOMEMALLOC) & ~(__GFP_RECLAIM|__GFP_NOFAIL); |
ba52270d | 1934 | |
5dfb4175 | 1935 | page = alloc_slab_page(s, alloc_gfp, node, oo); |
65c3376a CL |
1936 | if (unlikely(!page)) { |
1937 | oo = s->min; | |
80c3a998 | 1938 | alloc_gfp = flags; |
65c3376a CL |
1939 | /* |
1940 | * Allocation may have failed due to fragmentation. | |
1941 | * Try a lower order alloc if possible | |
1942 | */ | |
5dfb4175 | 1943 | page = alloc_slab_page(s, alloc_gfp, node, oo); |
588f8ba9 TG |
1944 | if (unlikely(!page)) |
1945 | goto out; | |
1946 | stat(s, ORDER_FALLBACK); | |
65c3376a | 1947 | } |
5a896d9e | 1948 | |
834f3d11 | 1949 | page->objects = oo_objects(oo); |
81819f0f | 1950 | |
b918653b | 1951 | account_slab(page_slab(page), oo_order(oo), s, flags); |
1f3147b4 | 1952 | |
1b4f59e3 | 1953 | page->slab_cache = s; |
c03f94cc | 1954 | __SetPageSlab(page); |
2f064f34 | 1955 | if (page_is_pfmemalloc(page)) |
072bb0aa | 1956 | SetPageSlabPfmemalloc(page); |
81819f0f | 1957 | |
a7101224 | 1958 | kasan_poison_slab(page); |
81819f0f | 1959 | |
a7101224 | 1960 | start = page_address(page); |
81819f0f | 1961 | |
a50b854e | 1962 | setup_page_debug(s, page, start); |
0316bec2 | 1963 | |
210e7a43 TG |
1964 | shuffle = shuffle_freelist(s, page); |
1965 | ||
1966 | if (!shuffle) { | |
4d176711 AK |
1967 | start = fixup_red_left(s, start); |
1968 | start = setup_object(s, page, start); | |
1969 | page->freelist = start; | |
18e50661 AK |
1970 | for (idx = 0, p = start; idx < page->objects - 1; idx++) { |
1971 | next = p + s->size; | |
1972 | next = setup_object(s, page, next); | |
1973 | set_freepointer(s, p, next); | |
1974 | p = next; | |
1975 | } | |
1976 | set_freepointer(s, p, NULL); | |
81819f0f | 1977 | } |
81819f0f | 1978 | |
e6e82ea1 | 1979 | page->inuse = page->objects; |
8cb0a506 | 1980 | page->frozen = 1; |
588f8ba9 | 1981 | |
81819f0f | 1982 | out: |
588f8ba9 TG |
1983 | if (!page) |
1984 | return NULL; | |
1985 | ||
588f8ba9 TG |
1986 | inc_slabs_node(s, page_to_nid(page), page->objects); |
1987 | ||
81819f0f CL |
1988 | return page; |
1989 | } | |
1990 | ||
588f8ba9 TG |
1991 | static struct page *new_slab(struct kmem_cache *s, gfp_t flags, int node) |
1992 | { | |
44405099 LL |
1993 | if (unlikely(flags & GFP_SLAB_BUG_MASK)) |
1994 | flags = kmalloc_fix_flags(flags); | |
588f8ba9 | 1995 | |
53a0de06 VB |
1996 | WARN_ON_ONCE(s->ctor && (flags & __GFP_ZERO)); |
1997 | ||
588f8ba9 TG |
1998 | return allocate_slab(s, |
1999 | flags & (GFP_RECLAIM_MASK | GFP_CONSTRAINT_MASK), node); | |
2000 | } | |
2001 | ||
81819f0f CL |
2002 | static void __free_slab(struct kmem_cache *s, struct page *page) |
2003 | { | |
834f3d11 CL |
2004 | int order = compound_order(page); |
2005 | int pages = 1 << order; | |
81819f0f | 2006 | |
8fc8d666 | 2007 | if (kmem_cache_debug_flags(s, SLAB_CONSISTENCY_CHECKS)) { |
81819f0f CL |
2008 | void *p; |
2009 | ||
2010 | slab_pad_check(s, page); | |
224a88be CL |
2011 | for_each_object(p, s, page_address(page), |
2012 | page->objects) | |
f7cb1933 | 2013 | check_object(s, page, p, SLUB_RED_INACTIVE); |
81819f0f CL |
2014 | } |
2015 | ||
072bb0aa | 2016 | __ClearPageSlabPfmemalloc(page); |
49bd5221 | 2017 | __ClearPageSlab(page); |
0c06dd75 VB |
2018 | /* In union with page->mapping where page allocator expects NULL */ |
2019 | page->slab_cache = NULL; | |
1eb5ac64 NP |
2020 | if (current->reclaim_state) |
2021 | current->reclaim_state->reclaimed_slab += pages; | |
b918653b | 2022 | unaccount_slab(page_slab(page), order, s); |
27ee57c9 | 2023 | __free_pages(page, order); |
81819f0f CL |
2024 | } |
2025 | ||
2026 | static void rcu_free_slab(struct rcu_head *h) | |
2027 | { | |
bf68c214 | 2028 | struct page *page = container_of(h, struct page, rcu_head); |
da9a638c | 2029 | |
1b4f59e3 | 2030 | __free_slab(page->slab_cache, page); |
81819f0f CL |
2031 | } |
2032 | ||
2033 | static void free_slab(struct kmem_cache *s, struct page *page) | |
2034 | { | |
5f0d5a3a | 2035 | if (unlikely(s->flags & SLAB_TYPESAFE_BY_RCU)) { |
bf68c214 | 2036 | call_rcu(&page->rcu_head, rcu_free_slab); |
81819f0f CL |
2037 | } else |
2038 | __free_slab(s, page); | |
2039 | } | |
2040 | ||
2041 | static void discard_slab(struct kmem_cache *s, struct page *page) | |
2042 | { | |
205ab99d | 2043 | dec_slabs_node(s, page_to_nid(page), page->objects); |
81819f0f CL |
2044 | free_slab(s, page); |
2045 | } | |
2046 | ||
2047 | /* | |
5cc6eee8 | 2048 | * Management of partially allocated slabs. |
81819f0f | 2049 | */ |
1e4dd946 SR |
2050 | static inline void |
2051 | __add_partial(struct kmem_cache_node *n, struct page *page, int tail) | |
81819f0f | 2052 | { |
e95eed57 | 2053 | n->nr_partial++; |
136333d1 | 2054 | if (tail == DEACTIVATE_TO_TAIL) |
916ac052 | 2055 | list_add_tail(&page->slab_list, &n->partial); |
7c2e132c | 2056 | else |
916ac052 | 2057 | list_add(&page->slab_list, &n->partial); |
81819f0f CL |
2058 | } |
2059 | ||
1e4dd946 SR |
2060 | static inline void add_partial(struct kmem_cache_node *n, |
2061 | struct page *page, int tail) | |
62e346a8 | 2062 | { |
c65c1877 | 2063 | lockdep_assert_held(&n->list_lock); |
1e4dd946 SR |
2064 | __add_partial(n, page, tail); |
2065 | } | |
c65c1877 | 2066 | |
1e4dd946 SR |
2067 | static inline void remove_partial(struct kmem_cache_node *n, |
2068 | struct page *page) | |
2069 | { | |
2070 | lockdep_assert_held(&n->list_lock); | |
916ac052 | 2071 | list_del(&page->slab_list); |
52b4b950 | 2072 | n->nr_partial--; |
1e4dd946 SR |
2073 | } |
2074 | ||
81819f0f | 2075 | /* |
7ced3719 CL |
2076 | * Remove slab from the partial list, freeze it and |
2077 | * return the pointer to the freelist. | |
81819f0f | 2078 | * |
497b66f2 | 2079 | * Returns a list of objects or NULL if it fails. |
81819f0f | 2080 | */ |
497b66f2 | 2081 | static inline void *acquire_slab(struct kmem_cache *s, |
acd19fd1 | 2082 | struct kmem_cache_node *n, struct page *page, |
b47291ef | 2083 | int mode) |
81819f0f | 2084 | { |
2cfb7455 CL |
2085 | void *freelist; |
2086 | unsigned long counters; | |
2087 | struct page new; | |
2088 | ||
c65c1877 PZ |
2089 | lockdep_assert_held(&n->list_lock); |
2090 | ||
2cfb7455 CL |
2091 | /* |
2092 | * Zap the freelist and set the frozen bit. | |
2093 | * The old freelist is the list of objects for the | |
2094 | * per cpu allocation list. | |
2095 | */ | |
7ced3719 CL |
2096 | freelist = page->freelist; |
2097 | counters = page->counters; | |
2098 | new.counters = counters; | |
23910c50 | 2099 | if (mode) { |
7ced3719 | 2100 | new.inuse = page->objects; |
23910c50 PE |
2101 | new.freelist = NULL; |
2102 | } else { | |
2103 | new.freelist = freelist; | |
2104 | } | |
2cfb7455 | 2105 | |
a0132ac0 | 2106 | VM_BUG_ON(new.frozen); |
7ced3719 | 2107 | new.frozen = 1; |
2cfb7455 | 2108 | |
7ced3719 | 2109 | if (!__cmpxchg_double_slab(s, page, |
2cfb7455 | 2110 | freelist, counters, |
02d7633f | 2111 | new.freelist, new.counters, |
7ced3719 | 2112 | "acquire_slab")) |
7ced3719 | 2113 | return NULL; |
2cfb7455 CL |
2114 | |
2115 | remove_partial(n, page); | |
7ced3719 | 2116 | WARN_ON(!freelist); |
49e22585 | 2117 | return freelist; |
81819f0f CL |
2118 | } |
2119 | ||
e0a043aa | 2120 | #ifdef CONFIG_SLUB_CPU_PARTIAL |
633b0764 | 2121 | static void put_cpu_partial(struct kmem_cache *s, struct page *page, int drain); |
e0a043aa VB |
2122 | #else |
2123 | static inline void put_cpu_partial(struct kmem_cache *s, struct page *page, | |
2124 | int drain) { } | |
2125 | #endif | |
8ba00bb6 | 2126 | static inline bool pfmemalloc_match(struct page *page, gfp_t gfpflags); |
49e22585 | 2127 | |
81819f0f | 2128 | /* |
672bba3a | 2129 | * Try to allocate a partial slab from a specific node. |
81819f0f | 2130 | */ |
8ba00bb6 | 2131 | static void *get_partial_node(struct kmem_cache *s, struct kmem_cache_node *n, |
4b1f449d | 2132 | struct page **ret_page, gfp_t gfpflags) |
81819f0f | 2133 | { |
49e22585 CL |
2134 | struct page *page, *page2; |
2135 | void *object = NULL; | |
4b1f449d | 2136 | unsigned long flags; |
b47291ef | 2137 | unsigned int partial_pages = 0; |
81819f0f CL |
2138 | |
2139 | /* | |
2140 | * Racy check. If we mistakenly see no partial slabs then we | |
2141 | * just allocate an empty slab. If we mistakenly try to get a | |
70b6d25e | 2142 | * partial slab and there is none available then get_partial() |
672bba3a | 2143 | * will return NULL. |
81819f0f CL |
2144 | */ |
2145 | if (!n || !n->nr_partial) | |
2146 | return NULL; | |
2147 | ||
4b1f449d | 2148 | spin_lock_irqsave(&n->list_lock, flags); |
916ac052 | 2149 | list_for_each_entry_safe(page, page2, &n->partial, slab_list) { |
8ba00bb6 | 2150 | void *t; |
49e22585 | 2151 | |
4b1f449d | 2152 | if (!pfmemalloc_match(page, gfpflags)) |
8ba00bb6 JK |
2153 | continue; |
2154 | ||
b47291ef | 2155 | t = acquire_slab(s, n, page, object == NULL); |
49e22585 | 2156 | if (!t) |
9b1ea29b | 2157 | break; |
49e22585 | 2158 | |
12d79634 | 2159 | if (!object) { |
75c8ff28 | 2160 | *ret_page = page; |
49e22585 | 2161 | stat(s, ALLOC_FROM_PARTIAL); |
49e22585 | 2162 | object = t; |
49e22585 | 2163 | } else { |
633b0764 | 2164 | put_cpu_partial(s, page, 0); |
8028dcea | 2165 | stat(s, CPU_PARTIAL_NODE); |
b47291ef | 2166 | partial_pages++; |
49e22585 | 2167 | } |
b47291ef | 2168 | #ifdef CONFIG_SLUB_CPU_PARTIAL |
345c905d | 2169 | if (!kmem_cache_has_cpu_partial(s) |
b47291ef | 2170 | || partial_pages > s->cpu_partial_pages / 2) |
49e22585 | 2171 | break; |
b47291ef VB |
2172 | #else |
2173 | break; | |
2174 | #endif | |
49e22585 | 2175 | |
497b66f2 | 2176 | } |
4b1f449d | 2177 | spin_unlock_irqrestore(&n->list_lock, flags); |
497b66f2 | 2178 | return object; |
81819f0f CL |
2179 | } |
2180 | ||
2181 | /* | |
672bba3a | 2182 | * Get a page from somewhere. Search in increasing NUMA distances. |
81819f0f | 2183 | */ |
de3ec035 | 2184 | static void *get_any_partial(struct kmem_cache *s, gfp_t flags, |
75c8ff28 | 2185 | struct page **ret_page) |
81819f0f CL |
2186 | { |
2187 | #ifdef CONFIG_NUMA | |
2188 | struct zonelist *zonelist; | |
dd1a239f | 2189 | struct zoneref *z; |
54a6eb5c | 2190 | struct zone *zone; |
97a225e6 | 2191 | enum zone_type highest_zoneidx = gfp_zone(flags); |
497b66f2 | 2192 | void *object; |
cc9a6c87 | 2193 | unsigned int cpuset_mems_cookie; |
81819f0f CL |
2194 | |
2195 | /* | |
672bba3a CL |
2196 | * The defrag ratio allows a configuration of the tradeoffs between |
2197 | * inter node defragmentation and node local allocations. A lower | |
2198 | * defrag_ratio increases the tendency to do local allocations | |
2199 | * instead of attempting to obtain partial slabs from other nodes. | |
81819f0f | 2200 | * |
672bba3a CL |
2201 | * If the defrag_ratio is set to 0 then kmalloc() always |
2202 | * returns node local objects. If the ratio is higher then kmalloc() | |
2203 | * may return off node objects because partial slabs are obtained | |
2204 | * from other nodes and filled up. | |
81819f0f | 2205 | * |
43efd3ea LP |
2206 | * If /sys/kernel/slab/xx/remote_node_defrag_ratio is set to 100 |
2207 | * (which makes defrag_ratio = 1000) then every (well almost) | |
2208 | * allocation will first attempt to defrag slab caches on other nodes. | |
2209 | * This means scanning over all nodes to look for partial slabs which | |
2210 | * may be expensive if we do it every time we are trying to find a slab | |
672bba3a | 2211 | * with available objects. |
81819f0f | 2212 | */ |
9824601e CL |
2213 | if (!s->remote_node_defrag_ratio || |
2214 | get_cycles() % 1024 > s->remote_node_defrag_ratio) | |
81819f0f CL |
2215 | return NULL; |
2216 | ||
cc9a6c87 | 2217 | do { |
d26914d1 | 2218 | cpuset_mems_cookie = read_mems_allowed_begin(); |
2a389610 | 2219 | zonelist = node_zonelist(mempolicy_slab_node(), flags); |
97a225e6 | 2220 | for_each_zone_zonelist(zone, z, zonelist, highest_zoneidx) { |
cc9a6c87 MG |
2221 | struct kmem_cache_node *n; |
2222 | ||
2223 | n = get_node(s, zone_to_nid(zone)); | |
2224 | ||
dee2f8aa | 2225 | if (n && cpuset_zone_allowed(zone, flags) && |
cc9a6c87 | 2226 | n->nr_partial > s->min_partial) { |
75c8ff28 | 2227 | object = get_partial_node(s, n, ret_page, flags); |
cc9a6c87 MG |
2228 | if (object) { |
2229 | /* | |
d26914d1 MG |
2230 | * Don't check read_mems_allowed_retry() |
2231 | * here - if mems_allowed was updated in | |
2232 | * parallel, that was a harmless race | |
2233 | * between allocation and the cpuset | |
2234 | * update | |
cc9a6c87 | 2235 | */ |
cc9a6c87 MG |
2236 | return object; |
2237 | } | |
c0ff7453 | 2238 | } |
81819f0f | 2239 | } |
d26914d1 | 2240 | } while (read_mems_allowed_retry(cpuset_mems_cookie)); |
6dfd1b65 | 2241 | #endif /* CONFIG_NUMA */ |
81819f0f CL |
2242 | return NULL; |
2243 | } | |
2244 | ||
2245 | /* | |
2246 | * Get a partial page, lock it and return it. | |
2247 | */ | |
497b66f2 | 2248 | static void *get_partial(struct kmem_cache *s, gfp_t flags, int node, |
75c8ff28 | 2249 | struct page **ret_page) |
81819f0f | 2250 | { |
497b66f2 | 2251 | void *object; |
a561ce00 JK |
2252 | int searchnode = node; |
2253 | ||
2254 | if (node == NUMA_NO_NODE) | |
2255 | searchnode = numa_mem_id(); | |
81819f0f | 2256 | |
75c8ff28 | 2257 | object = get_partial_node(s, get_node(s, searchnode), ret_page, flags); |
497b66f2 CL |
2258 | if (object || node != NUMA_NO_NODE) |
2259 | return object; | |
81819f0f | 2260 | |
75c8ff28 | 2261 | return get_any_partial(s, flags, ret_page); |
81819f0f CL |
2262 | } |
2263 | ||
923717cb | 2264 | #ifdef CONFIG_PREEMPTION |
8a5ec0ba | 2265 | /* |
0d645ed1 | 2266 | * Calculate the next globally unique transaction for disambiguation |
8a5ec0ba CL |
2267 | * during cmpxchg. The transactions start with the cpu number and are then |
2268 | * incremented by CONFIG_NR_CPUS. | |
2269 | */ | |
2270 | #define TID_STEP roundup_pow_of_two(CONFIG_NR_CPUS) | |
2271 | #else | |
2272 | /* | |
2273 | * No preemption supported therefore also no need to check for | |
2274 | * different cpus. | |
2275 | */ | |
2276 | #define TID_STEP 1 | |
2277 | #endif | |
2278 | ||
2279 | static inline unsigned long next_tid(unsigned long tid) | |
2280 | { | |
2281 | return tid + TID_STEP; | |
2282 | } | |
2283 | ||
9d5f0be0 | 2284 | #ifdef SLUB_DEBUG_CMPXCHG |
8a5ec0ba CL |
2285 | static inline unsigned int tid_to_cpu(unsigned long tid) |
2286 | { | |
2287 | return tid % TID_STEP; | |
2288 | } | |
2289 | ||
2290 | static inline unsigned long tid_to_event(unsigned long tid) | |
2291 | { | |
2292 | return tid / TID_STEP; | |
2293 | } | |
9d5f0be0 | 2294 | #endif |
8a5ec0ba CL |
2295 | |
2296 | static inline unsigned int init_tid(int cpu) | |
2297 | { | |
2298 | return cpu; | |
2299 | } | |
2300 | ||
2301 | static inline void note_cmpxchg_failure(const char *n, | |
2302 | const struct kmem_cache *s, unsigned long tid) | |
2303 | { | |
2304 | #ifdef SLUB_DEBUG_CMPXCHG | |
2305 | unsigned long actual_tid = __this_cpu_read(s->cpu_slab->tid); | |
2306 | ||
f9f58285 | 2307 | pr_info("%s %s: cmpxchg redo ", n, s->name); |
8a5ec0ba | 2308 | |
923717cb | 2309 | #ifdef CONFIG_PREEMPTION |
8a5ec0ba | 2310 | if (tid_to_cpu(tid) != tid_to_cpu(actual_tid)) |
f9f58285 | 2311 | pr_warn("due to cpu change %d -> %d\n", |
8a5ec0ba CL |
2312 | tid_to_cpu(tid), tid_to_cpu(actual_tid)); |
2313 | else | |
2314 | #endif | |
2315 | if (tid_to_event(tid) != tid_to_event(actual_tid)) | |
f9f58285 | 2316 | pr_warn("due to cpu running other code. Event %ld->%ld\n", |
8a5ec0ba CL |
2317 | tid_to_event(tid), tid_to_event(actual_tid)); |
2318 | else | |
f9f58285 | 2319 | pr_warn("for unknown reason: actual=%lx was=%lx target=%lx\n", |
8a5ec0ba CL |
2320 | actual_tid, tid, next_tid(tid)); |
2321 | #endif | |
4fdccdfb | 2322 | stat(s, CMPXCHG_DOUBLE_CPU_FAIL); |
8a5ec0ba CL |
2323 | } |
2324 | ||
788e1aad | 2325 | static void init_kmem_cache_cpus(struct kmem_cache *s) |
8a5ec0ba | 2326 | { |
8a5ec0ba | 2327 | int cpu; |
bd0e7491 | 2328 | struct kmem_cache_cpu *c; |
8a5ec0ba | 2329 | |
bd0e7491 VB |
2330 | for_each_possible_cpu(cpu) { |
2331 | c = per_cpu_ptr(s->cpu_slab, cpu); | |
2332 | local_lock_init(&c->lock); | |
2333 | c->tid = init_tid(cpu); | |
2334 | } | |
8a5ec0ba | 2335 | } |
2cfb7455 | 2336 | |
81819f0f | 2337 | /* |
a019d201 VB |
2338 | * Finishes removing the cpu slab. Merges cpu's freelist with page's freelist, |
2339 | * unfreezes the slabs and puts it on the proper list. | |
2340 | * Assumes the slab has been already safely taken away from kmem_cache_cpu | |
2341 | * by the caller. | |
81819f0f | 2342 | */ |
d0e0ac97 | 2343 | static void deactivate_slab(struct kmem_cache *s, struct page *page, |
a019d201 | 2344 | void *freelist) |
81819f0f | 2345 | { |
2cfb7455 | 2346 | enum slab_modes { M_NONE, M_PARTIAL, M_FULL, M_FREE }; |
2cfb7455 | 2347 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); |
d930ff03 | 2348 | int lock = 0, free_delta = 0; |
2cfb7455 | 2349 | enum slab_modes l = M_NONE, m = M_NONE; |
d930ff03 | 2350 | void *nextfree, *freelist_iter, *freelist_tail; |
136333d1 | 2351 | int tail = DEACTIVATE_TO_HEAD; |
3406e91b | 2352 | unsigned long flags = 0; |
2cfb7455 CL |
2353 | struct page new; |
2354 | struct page old; | |
2355 | ||
2356 | if (page->freelist) { | |
84e554e6 | 2357 | stat(s, DEACTIVATE_REMOTE_FREES); |
136333d1 | 2358 | tail = DEACTIVATE_TO_TAIL; |
2cfb7455 CL |
2359 | } |
2360 | ||
894b8788 | 2361 | /* |
d930ff03 VB |
2362 | * Stage one: Count the objects on cpu's freelist as free_delta and |
2363 | * remember the last object in freelist_tail for later splicing. | |
2cfb7455 | 2364 | */ |
d930ff03 VB |
2365 | freelist_tail = NULL; |
2366 | freelist_iter = freelist; | |
2367 | while (freelist_iter) { | |
2368 | nextfree = get_freepointer(s, freelist_iter); | |
2cfb7455 | 2369 | |
52f23478 DZ |
2370 | /* |
2371 | * If 'nextfree' is invalid, it is possible that the object at | |
d930ff03 VB |
2372 | * 'freelist_iter' is already corrupted. So isolate all objects |
2373 | * starting at 'freelist_iter' by skipping them. | |
52f23478 | 2374 | */ |
d930ff03 | 2375 | if (freelist_corrupted(s, page, &freelist_iter, nextfree)) |
52f23478 DZ |
2376 | break; |
2377 | ||
d930ff03 VB |
2378 | freelist_tail = freelist_iter; |
2379 | free_delta++; | |
2cfb7455 | 2380 | |
d930ff03 | 2381 | freelist_iter = nextfree; |
2cfb7455 CL |
2382 | } |
2383 | ||
894b8788 | 2384 | /* |
d930ff03 VB |
2385 | * Stage two: Unfreeze the page while splicing the per-cpu |
2386 | * freelist to the head of page's freelist. | |
2387 | * | |
2388 | * Ensure that the page is unfrozen while the list presence | |
2389 | * reflects the actual number of objects during unfreeze. | |
2cfb7455 CL |
2390 | * |
2391 | * We setup the list membership and then perform a cmpxchg | |
2392 | * with the count. If there is a mismatch then the page | |
2393 | * is not unfrozen but the page is on the wrong list. | |
2394 | * | |
2395 | * Then we restart the process which may have to remove | |
2396 | * the page from the list that we just put it on again | |
2397 | * because the number of objects in the slab may have | |
2398 | * changed. | |
894b8788 | 2399 | */ |
2cfb7455 | 2400 | redo: |
894b8788 | 2401 | |
d930ff03 VB |
2402 | old.freelist = READ_ONCE(page->freelist); |
2403 | old.counters = READ_ONCE(page->counters); | |
a0132ac0 | 2404 | VM_BUG_ON(!old.frozen); |
7c2e132c | 2405 | |
2cfb7455 CL |
2406 | /* Determine target state of the slab */ |
2407 | new.counters = old.counters; | |
d930ff03 VB |
2408 | if (freelist_tail) { |
2409 | new.inuse -= free_delta; | |
2410 | set_freepointer(s, freelist_tail, old.freelist); | |
2cfb7455 CL |
2411 | new.freelist = freelist; |
2412 | } else | |
2413 | new.freelist = old.freelist; | |
2414 | ||
2415 | new.frozen = 0; | |
2416 | ||
8a5b20ae | 2417 | if (!new.inuse && n->nr_partial >= s->min_partial) |
2cfb7455 CL |
2418 | m = M_FREE; |
2419 | else if (new.freelist) { | |
2420 | m = M_PARTIAL; | |
2421 | if (!lock) { | |
2422 | lock = 1; | |
2423 | /* | |
8bb4e7a2 | 2424 | * Taking the spinlock removes the possibility |
2cfb7455 CL |
2425 | * that acquire_slab() will see a slab page that |
2426 | * is frozen | |
2427 | */ | |
3406e91b | 2428 | spin_lock_irqsave(&n->list_lock, flags); |
2cfb7455 CL |
2429 | } |
2430 | } else { | |
2431 | m = M_FULL; | |
965c4848 | 2432 | if (kmem_cache_debug_flags(s, SLAB_STORE_USER) && !lock) { |
2cfb7455 CL |
2433 | lock = 1; |
2434 | /* | |
2435 | * This also ensures that the scanning of full | |
2436 | * slabs from diagnostic functions will not see | |
2437 | * any frozen slabs. | |
2438 | */ | |
3406e91b | 2439 | spin_lock_irqsave(&n->list_lock, flags); |
2cfb7455 CL |
2440 | } |
2441 | } | |
2442 | ||
2443 | if (l != m) { | |
2cfb7455 | 2444 | if (l == M_PARTIAL) |
2cfb7455 | 2445 | remove_partial(n, page); |
2cfb7455 | 2446 | else if (l == M_FULL) |
c65c1877 | 2447 | remove_full(s, n, page); |
2cfb7455 | 2448 | |
88349a28 | 2449 | if (m == M_PARTIAL) |
2cfb7455 | 2450 | add_partial(n, page, tail); |
88349a28 | 2451 | else if (m == M_FULL) |
2cfb7455 | 2452 | add_full(s, n, page); |
2cfb7455 CL |
2453 | } |
2454 | ||
2455 | l = m; | |
3406e91b | 2456 | if (!cmpxchg_double_slab(s, page, |
2cfb7455 CL |
2457 | old.freelist, old.counters, |
2458 | new.freelist, new.counters, | |
2459 | "unfreezing slab")) | |
2460 | goto redo; | |
2461 | ||
2cfb7455 | 2462 | if (lock) |
3406e91b | 2463 | spin_unlock_irqrestore(&n->list_lock, flags); |
2cfb7455 | 2464 | |
88349a28 WY |
2465 | if (m == M_PARTIAL) |
2466 | stat(s, tail); | |
2467 | else if (m == M_FULL) | |
2468 | stat(s, DEACTIVATE_FULL); | |
2469 | else if (m == M_FREE) { | |
2cfb7455 CL |
2470 | stat(s, DEACTIVATE_EMPTY); |
2471 | discard_slab(s, page); | |
2472 | stat(s, FREE_SLAB); | |
894b8788 | 2473 | } |
81819f0f CL |
2474 | } |
2475 | ||
345c905d | 2476 | #ifdef CONFIG_SLUB_CPU_PARTIAL |
fc1455f4 VB |
2477 | static void __unfreeze_partials(struct kmem_cache *s, struct page *partial_page) |
2478 | { | |
43d77867 | 2479 | struct kmem_cache_node *n = NULL, *n2 = NULL; |
fc1455f4 | 2480 | struct page *page, *discard_page = NULL; |
7cf9f3ba | 2481 | unsigned long flags = 0; |
49e22585 | 2482 | |
c2f973ba | 2483 | while (partial_page) { |
49e22585 CL |
2484 | struct page new; |
2485 | struct page old; | |
2486 | ||
c2f973ba VB |
2487 | page = partial_page; |
2488 | partial_page = page->next; | |
43d77867 JK |
2489 | |
2490 | n2 = get_node(s, page_to_nid(page)); | |
2491 | if (n != n2) { | |
2492 | if (n) | |
7cf9f3ba | 2493 | spin_unlock_irqrestore(&n->list_lock, flags); |
43d77867 JK |
2494 | |
2495 | n = n2; | |
7cf9f3ba | 2496 | spin_lock_irqsave(&n->list_lock, flags); |
43d77867 | 2497 | } |
49e22585 CL |
2498 | |
2499 | do { | |
2500 | ||
2501 | old.freelist = page->freelist; | |
2502 | old.counters = page->counters; | |
a0132ac0 | 2503 | VM_BUG_ON(!old.frozen); |
49e22585 CL |
2504 | |
2505 | new.counters = old.counters; | |
2506 | new.freelist = old.freelist; | |
2507 | ||
2508 | new.frozen = 0; | |
2509 | ||
d24ac77f | 2510 | } while (!__cmpxchg_double_slab(s, page, |
49e22585 CL |
2511 | old.freelist, old.counters, |
2512 | new.freelist, new.counters, | |
2513 | "unfreezing slab")); | |
2514 | ||
8a5b20ae | 2515 | if (unlikely(!new.inuse && n->nr_partial >= s->min_partial)) { |
9ada1934 SL |
2516 | page->next = discard_page; |
2517 | discard_page = page; | |
43d77867 JK |
2518 | } else { |
2519 | add_partial(n, page, DEACTIVATE_TO_TAIL); | |
2520 | stat(s, FREE_ADD_PARTIAL); | |
49e22585 CL |
2521 | } |
2522 | } | |
2523 | ||
2524 | if (n) | |
7cf9f3ba | 2525 | spin_unlock_irqrestore(&n->list_lock, flags); |
8de06a6f | 2526 | |
9ada1934 SL |
2527 | while (discard_page) { |
2528 | page = discard_page; | |
2529 | discard_page = discard_page->next; | |
2530 | ||
2531 | stat(s, DEACTIVATE_EMPTY); | |
2532 | discard_slab(s, page); | |
2533 | stat(s, FREE_SLAB); | |
2534 | } | |
fc1455f4 | 2535 | } |
f3ab8b6b | 2536 | |
fc1455f4 VB |
2537 | /* |
2538 | * Unfreeze all the cpu partial slabs. | |
2539 | */ | |
2540 | static void unfreeze_partials(struct kmem_cache *s) | |
2541 | { | |
2542 | struct page *partial_page; | |
2543 | unsigned long flags; | |
2544 | ||
bd0e7491 | 2545 | local_lock_irqsave(&s->cpu_slab->lock, flags); |
fc1455f4 VB |
2546 | partial_page = this_cpu_read(s->cpu_slab->partial); |
2547 | this_cpu_write(s->cpu_slab->partial, NULL); | |
bd0e7491 | 2548 | local_unlock_irqrestore(&s->cpu_slab->lock, flags); |
fc1455f4 VB |
2549 | |
2550 | if (partial_page) | |
2551 | __unfreeze_partials(s, partial_page); | |
2552 | } | |
2553 | ||
2554 | static void unfreeze_partials_cpu(struct kmem_cache *s, | |
2555 | struct kmem_cache_cpu *c) | |
2556 | { | |
2557 | struct page *partial_page; | |
2558 | ||
2559 | partial_page = slub_percpu_partial(c); | |
2560 | c->partial = NULL; | |
2561 | ||
2562 | if (partial_page) | |
2563 | __unfreeze_partials(s, partial_page); | |
49e22585 CL |
2564 | } |
2565 | ||
2566 | /* | |
9234bae9 WY |
2567 | * Put a page that was just frozen (in __slab_free|get_partial_node) into a |
2568 | * partial page slot if available. | |
49e22585 CL |
2569 | * |
2570 | * If we did not find a slot then simply move all the partials to the | |
2571 | * per node partial list. | |
2572 | */ | |
633b0764 | 2573 | static void put_cpu_partial(struct kmem_cache *s, struct page *page, int drain) |
49e22585 CL |
2574 | { |
2575 | struct page *oldpage; | |
e0a043aa VB |
2576 | struct page *page_to_unfreeze = NULL; |
2577 | unsigned long flags; | |
2578 | int pages = 0; | |
49e22585 | 2579 | |
bd0e7491 | 2580 | local_lock_irqsave(&s->cpu_slab->lock, flags); |
49e22585 | 2581 | |
e0a043aa VB |
2582 | oldpage = this_cpu_read(s->cpu_slab->partial); |
2583 | ||
2584 | if (oldpage) { | |
b47291ef | 2585 | if (drain && oldpage->pages >= s->cpu_partial_pages) { |
e0a043aa VB |
2586 | /* |
2587 | * Partial array is full. Move the existing set to the | |
2588 | * per node partial list. Postpone the actual unfreezing | |
2589 | * outside of the critical section. | |
2590 | */ | |
2591 | page_to_unfreeze = oldpage; | |
2592 | oldpage = NULL; | |
2593 | } else { | |
49e22585 | 2594 | pages = oldpage->pages; |
49e22585 | 2595 | } |
e0a043aa | 2596 | } |
49e22585 | 2597 | |
e0a043aa | 2598 | pages++; |
49e22585 | 2599 | |
e0a043aa | 2600 | page->pages = pages; |
e0a043aa | 2601 | page->next = oldpage; |
49e22585 | 2602 | |
e0a043aa VB |
2603 | this_cpu_write(s->cpu_slab->partial, page); |
2604 | ||
bd0e7491 | 2605 | local_unlock_irqrestore(&s->cpu_slab->lock, flags); |
e0a043aa VB |
2606 | |
2607 | if (page_to_unfreeze) { | |
2608 | __unfreeze_partials(s, page_to_unfreeze); | |
2609 | stat(s, CPU_PARTIAL_DRAIN); | |
2610 | } | |
49e22585 CL |
2611 | } |
2612 | ||
e0a043aa VB |
2613 | #else /* CONFIG_SLUB_CPU_PARTIAL */ |
2614 | ||
2615 | static inline void unfreeze_partials(struct kmem_cache *s) { } | |
2616 | static inline void unfreeze_partials_cpu(struct kmem_cache *s, | |
2617 | struct kmem_cache_cpu *c) { } | |
2618 | ||
2619 | #endif /* CONFIG_SLUB_CPU_PARTIAL */ | |
2620 | ||
dfb4f096 | 2621 | static inline void flush_slab(struct kmem_cache *s, struct kmem_cache_cpu *c) |
81819f0f | 2622 | { |
5a836bf6 SAS |
2623 | unsigned long flags; |
2624 | struct page *page; | |
2625 | void *freelist; | |
2626 | ||
bd0e7491 | 2627 | local_lock_irqsave(&s->cpu_slab->lock, flags); |
5a836bf6 SAS |
2628 | |
2629 | page = c->page; | |
2630 | freelist = c->freelist; | |
c17dda40 | 2631 | |
a019d201 VB |
2632 | c->page = NULL; |
2633 | c->freelist = NULL; | |
c17dda40 | 2634 | c->tid = next_tid(c->tid); |
a019d201 | 2635 | |
bd0e7491 | 2636 | local_unlock_irqrestore(&s->cpu_slab->lock, flags); |
a019d201 | 2637 | |
5a836bf6 SAS |
2638 | if (page) { |
2639 | deactivate_slab(s, page, freelist); | |
2640 | stat(s, CPUSLAB_FLUSH); | |
2641 | } | |
81819f0f CL |
2642 | } |
2643 | ||
0c710013 | 2644 | static inline void __flush_cpu_slab(struct kmem_cache *s, int cpu) |
81819f0f | 2645 | { |
9dfc6e68 | 2646 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu); |
08beb547 VB |
2647 | void *freelist = c->freelist; |
2648 | struct page *page = c->page; | |
81819f0f | 2649 | |
08beb547 VB |
2650 | c->page = NULL; |
2651 | c->freelist = NULL; | |
2652 | c->tid = next_tid(c->tid); | |
2653 | ||
2654 | if (page) { | |
2655 | deactivate_slab(s, page, freelist); | |
2656 | stat(s, CPUSLAB_FLUSH); | |
2657 | } | |
49e22585 | 2658 | |
fc1455f4 | 2659 | unfreeze_partials_cpu(s, c); |
81819f0f CL |
2660 | } |
2661 | ||
5a836bf6 SAS |
2662 | struct slub_flush_work { |
2663 | struct work_struct work; | |
2664 | struct kmem_cache *s; | |
2665 | bool skip; | |
2666 | }; | |
2667 | ||
fc1455f4 VB |
2668 | /* |
2669 | * Flush cpu slab. | |
2670 | * | |
5a836bf6 | 2671 | * Called from CPU work handler with migration disabled. |
fc1455f4 | 2672 | */ |
5a836bf6 | 2673 | static void flush_cpu_slab(struct work_struct *w) |
81819f0f | 2674 | { |
5a836bf6 SAS |
2675 | struct kmem_cache *s; |
2676 | struct kmem_cache_cpu *c; | |
2677 | struct slub_flush_work *sfw; | |
2678 | ||
2679 | sfw = container_of(w, struct slub_flush_work, work); | |
2680 | ||
2681 | s = sfw->s; | |
2682 | c = this_cpu_ptr(s->cpu_slab); | |
fc1455f4 VB |
2683 | |
2684 | if (c->page) | |
2685 | flush_slab(s, c); | |
81819f0f | 2686 | |
fc1455f4 | 2687 | unfreeze_partials(s); |
81819f0f CL |
2688 | } |
2689 | ||
5a836bf6 | 2690 | static bool has_cpu_slab(int cpu, struct kmem_cache *s) |
a8364d55 | 2691 | { |
a8364d55 GBY |
2692 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu); |
2693 | ||
a93cf07b | 2694 | return c->page || slub_percpu_partial(c); |
a8364d55 GBY |
2695 | } |
2696 | ||
5a836bf6 SAS |
2697 | static DEFINE_MUTEX(flush_lock); |
2698 | static DEFINE_PER_CPU(struct slub_flush_work, slub_flush); | |
2699 | ||
2700 | static void flush_all_cpus_locked(struct kmem_cache *s) | |
2701 | { | |
2702 | struct slub_flush_work *sfw; | |
2703 | unsigned int cpu; | |
2704 | ||
2705 | lockdep_assert_cpus_held(); | |
2706 | mutex_lock(&flush_lock); | |
2707 | ||
2708 | for_each_online_cpu(cpu) { | |
2709 | sfw = &per_cpu(slub_flush, cpu); | |
2710 | if (!has_cpu_slab(cpu, s)) { | |
2711 | sfw->skip = true; | |
2712 | continue; | |
2713 | } | |
2714 | INIT_WORK(&sfw->work, flush_cpu_slab); | |
2715 | sfw->skip = false; | |
2716 | sfw->s = s; | |
2717 | schedule_work_on(cpu, &sfw->work); | |
2718 | } | |
2719 | ||
2720 | for_each_online_cpu(cpu) { | |
2721 | sfw = &per_cpu(slub_flush, cpu); | |
2722 | if (sfw->skip) | |
2723 | continue; | |
2724 | flush_work(&sfw->work); | |
2725 | } | |
2726 | ||
2727 | mutex_unlock(&flush_lock); | |
2728 | } | |
2729 | ||
81819f0f CL |
2730 | static void flush_all(struct kmem_cache *s) |
2731 | { | |
5a836bf6 SAS |
2732 | cpus_read_lock(); |
2733 | flush_all_cpus_locked(s); | |
2734 | cpus_read_unlock(); | |
81819f0f CL |
2735 | } |
2736 | ||
a96a87bf SAS |
2737 | /* |
2738 | * Use the cpu notifier to insure that the cpu slabs are flushed when | |
2739 | * necessary. | |
2740 | */ | |
2741 | static int slub_cpu_dead(unsigned int cpu) | |
2742 | { | |
2743 | struct kmem_cache *s; | |
a96a87bf SAS |
2744 | |
2745 | mutex_lock(&slab_mutex); | |
0e7ac738 | 2746 | list_for_each_entry(s, &slab_caches, list) |
a96a87bf | 2747 | __flush_cpu_slab(s, cpu); |
a96a87bf SAS |
2748 | mutex_unlock(&slab_mutex); |
2749 | return 0; | |
2750 | } | |
2751 | ||
dfb4f096 CL |
2752 | /* |
2753 | * Check if the objects in a per cpu structure fit numa | |
2754 | * locality expectations. | |
2755 | */ | |
57d437d2 | 2756 | static inline int node_match(struct page *page, int node) |
dfb4f096 CL |
2757 | { |
2758 | #ifdef CONFIG_NUMA | |
6159d0f5 | 2759 | if (node != NUMA_NO_NODE && page_to_nid(page) != node) |
dfb4f096 CL |
2760 | return 0; |
2761 | #endif | |
2762 | return 1; | |
2763 | } | |
2764 | ||
9a02d699 | 2765 | #ifdef CONFIG_SLUB_DEBUG |
781b2ba6 PE |
2766 | static int count_free(struct page *page) |
2767 | { | |
2768 | return page->objects - page->inuse; | |
2769 | } | |
2770 | ||
9a02d699 DR |
2771 | static inline unsigned long node_nr_objs(struct kmem_cache_node *n) |
2772 | { | |
2773 | return atomic_long_read(&n->total_objects); | |
2774 | } | |
2775 | #endif /* CONFIG_SLUB_DEBUG */ | |
2776 | ||
2777 | #if defined(CONFIG_SLUB_DEBUG) || defined(CONFIG_SYSFS) | |
781b2ba6 PE |
2778 | static unsigned long count_partial(struct kmem_cache_node *n, |
2779 | int (*get_count)(struct page *)) | |
2780 | { | |
2781 | unsigned long flags; | |
2782 | unsigned long x = 0; | |
2783 | struct page *page; | |
2784 | ||
2785 | spin_lock_irqsave(&n->list_lock, flags); | |
916ac052 | 2786 | list_for_each_entry(page, &n->partial, slab_list) |
781b2ba6 PE |
2787 | x += get_count(page); |
2788 | spin_unlock_irqrestore(&n->list_lock, flags); | |
2789 | return x; | |
2790 | } | |
9a02d699 | 2791 | #endif /* CONFIG_SLUB_DEBUG || CONFIG_SYSFS */ |
26c02cf0 | 2792 | |
781b2ba6 PE |
2793 | static noinline void |
2794 | slab_out_of_memory(struct kmem_cache *s, gfp_t gfpflags, int nid) | |
2795 | { | |
9a02d699 DR |
2796 | #ifdef CONFIG_SLUB_DEBUG |
2797 | static DEFINE_RATELIMIT_STATE(slub_oom_rs, DEFAULT_RATELIMIT_INTERVAL, | |
2798 | DEFAULT_RATELIMIT_BURST); | |
781b2ba6 | 2799 | int node; |
fa45dc25 | 2800 | struct kmem_cache_node *n; |
781b2ba6 | 2801 | |
9a02d699 DR |
2802 | if ((gfpflags & __GFP_NOWARN) || !__ratelimit(&slub_oom_rs)) |
2803 | return; | |
2804 | ||
5b3810e5 VB |
2805 | pr_warn("SLUB: Unable to allocate memory on node %d, gfp=%#x(%pGg)\n", |
2806 | nid, gfpflags, &gfpflags); | |
19af27af | 2807 | pr_warn(" cache: %s, object size: %u, buffer size: %u, default order: %u, min order: %u\n", |
f9f58285 FF |
2808 | s->name, s->object_size, s->size, oo_order(s->oo), |
2809 | oo_order(s->min)); | |
781b2ba6 | 2810 | |
3b0efdfa | 2811 | if (oo_order(s->min) > get_order(s->object_size)) |
f9f58285 FF |
2812 | pr_warn(" %s debugging increased min order, use slub_debug=O to disable.\n", |
2813 | s->name); | |
fa5ec8a1 | 2814 | |
fa45dc25 | 2815 | for_each_kmem_cache_node(s, node, n) { |
781b2ba6 PE |
2816 | unsigned long nr_slabs; |
2817 | unsigned long nr_objs; | |
2818 | unsigned long nr_free; | |
2819 | ||
26c02cf0 AB |
2820 | nr_free = count_partial(n, count_free); |
2821 | nr_slabs = node_nr_slabs(n); | |
2822 | nr_objs = node_nr_objs(n); | |
781b2ba6 | 2823 | |
f9f58285 | 2824 | pr_warn(" node %d: slabs: %ld, objs: %ld, free: %ld\n", |
781b2ba6 PE |
2825 | node, nr_slabs, nr_objs, nr_free); |
2826 | } | |
9a02d699 | 2827 | #endif |
781b2ba6 PE |
2828 | } |
2829 | ||
072bb0aa MG |
2830 | static inline bool pfmemalloc_match(struct page *page, gfp_t gfpflags) |
2831 | { | |
2832 | if (unlikely(PageSlabPfmemalloc(page))) | |
2833 | return gfp_pfmemalloc_allowed(gfpflags); | |
2834 | ||
2835 | return true; | |
2836 | } | |
2837 | ||
0b303fb4 VB |
2838 | /* |
2839 | * A variant of pfmemalloc_match() that tests page flags without asserting | |
2840 | * PageSlab. Intended for opportunistic checks before taking a lock and | |
2841 | * rechecking that nobody else freed the page under us. | |
2842 | */ | |
2843 | static inline bool pfmemalloc_match_unsafe(struct page *page, gfp_t gfpflags) | |
2844 | { | |
2845 | if (unlikely(__PageSlabPfmemalloc(page))) | |
2846 | return gfp_pfmemalloc_allowed(gfpflags); | |
2847 | ||
2848 | return true; | |
2849 | } | |
2850 | ||
213eeb9f | 2851 | /* |
d0e0ac97 CG |
2852 | * Check the page->freelist of a page and either transfer the freelist to the |
2853 | * per cpu freelist or deactivate the page. | |
213eeb9f CL |
2854 | * |
2855 | * The page is still frozen if the return value is not NULL. | |
2856 | * | |
2857 | * If this function returns NULL then the page has been unfrozen. | |
2858 | */ | |
2859 | static inline void *get_freelist(struct kmem_cache *s, struct page *page) | |
2860 | { | |
2861 | struct page new; | |
2862 | unsigned long counters; | |
2863 | void *freelist; | |
2864 | ||
bd0e7491 VB |
2865 | lockdep_assert_held(this_cpu_ptr(&s->cpu_slab->lock)); |
2866 | ||
213eeb9f CL |
2867 | do { |
2868 | freelist = page->freelist; | |
2869 | counters = page->counters; | |
6faa6833 | 2870 | |
213eeb9f | 2871 | new.counters = counters; |
a0132ac0 | 2872 | VM_BUG_ON(!new.frozen); |
213eeb9f CL |
2873 | |
2874 | new.inuse = page->objects; | |
2875 | new.frozen = freelist != NULL; | |
2876 | ||
d24ac77f | 2877 | } while (!__cmpxchg_double_slab(s, page, |
213eeb9f CL |
2878 | freelist, counters, |
2879 | NULL, new.counters, | |
2880 | "get_freelist")); | |
2881 | ||
2882 | return freelist; | |
2883 | } | |
2884 | ||
81819f0f | 2885 | /* |
894b8788 CL |
2886 | * Slow path. The lockless freelist is empty or we need to perform |
2887 | * debugging duties. | |
2888 | * | |
894b8788 CL |
2889 | * Processing is still very fast if new objects have been freed to the |
2890 | * regular freelist. In that case we simply take over the regular freelist | |
2891 | * as the lockless freelist and zap the regular freelist. | |
81819f0f | 2892 | * |
894b8788 CL |
2893 | * If that is not working then we fall back to the partial lists. We take the |
2894 | * first element of the freelist as the object to allocate now and move the | |
2895 | * rest of the freelist to the lockless freelist. | |
81819f0f | 2896 | * |
894b8788 | 2897 | * And if we were unable to get a new slab from the partial slab lists then |
6446faa2 CL |
2898 | * we need to allocate a new slab. This is the slowest path since it involves |
2899 | * a call to the page allocator and the setup of a new slab. | |
a380a3c7 | 2900 | * |
e500059b | 2901 | * Version of __slab_alloc to use when we know that preemption is |
a380a3c7 | 2902 | * already disabled (which is the case for bulk allocation). |
81819f0f | 2903 | */ |
a380a3c7 | 2904 | static void *___slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node, |
ce71e27c | 2905 | unsigned long addr, struct kmem_cache_cpu *c) |
81819f0f | 2906 | { |
6faa6833 | 2907 | void *freelist; |
f6e7def7 | 2908 | struct page *page; |
e500059b | 2909 | unsigned long flags; |
81819f0f | 2910 | |
9f986d99 AW |
2911 | stat(s, ALLOC_SLOWPATH); |
2912 | ||
0b303fb4 VB |
2913 | reread_page: |
2914 | ||
2915 | page = READ_ONCE(c->page); | |
0715e6c5 VB |
2916 | if (!page) { |
2917 | /* | |
2918 | * if the node is not online or has no normal memory, just | |
2919 | * ignore the node constraint | |
2920 | */ | |
2921 | if (unlikely(node != NUMA_NO_NODE && | |
7e1fa93d | 2922 | !node_isset(node, slab_nodes))) |
0715e6c5 | 2923 | node = NUMA_NO_NODE; |
81819f0f | 2924 | goto new_slab; |
0715e6c5 | 2925 | } |
49e22585 | 2926 | redo: |
6faa6833 | 2927 | |
57d437d2 | 2928 | if (unlikely(!node_match(page, node))) { |
0715e6c5 VB |
2929 | /* |
2930 | * same as above but node_match() being false already | |
2931 | * implies node != NUMA_NO_NODE | |
2932 | */ | |
7e1fa93d | 2933 | if (!node_isset(node, slab_nodes)) { |
0715e6c5 VB |
2934 | node = NUMA_NO_NODE; |
2935 | goto redo; | |
2936 | } else { | |
a561ce00 | 2937 | stat(s, ALLOC_NODE_MISMATCH); |
0b303fb4 | 2938 | goto deactivate_slab; |
a561ce00 | 2939 | } |
fc59c053 | 2940 | } |
6446faa2 | 2941 | |
072bb0aa MG |
2942 | /* |
2943 | * By rights, we should be searching for a slab page that was | |
2944 | * PFMEMALLOC but right now, we are losing the pfmemalloc | |
2945 | * information when the page leaves the per-cpu allocator | |
2946 | */ | |
0b303fb4 VB |
2947 | if (unlikely(!pfmemalloc_match_unsafe(page, gfpflags))) |
2948 | goto deactivate_slab; | |
072bb0aa | 2949 | |
25c00c50 | 2950 | /* must check again c->page in case we got preempted and it changed */ |
bd0e7491 | 2951 | local_lock_irqsave(&s->cpu_slab->lock, flags); |
0b303fb4 | 2952 | if (unlikely(page != c->page)) { |
bd0e7491 | 2953 | local_unlock_irqrestore(&s->cpu_slab->lock, flags); |
0b303fb4 VB |
2954 | goto reread_page; |
2955 | } | |
6faa6833 CL |
2956 | freelist = c->freelist; |
2957 | if (freelist) | |
73736e03 | 2958 | goto load_freelist; |
03e404af | 2959 | |
f6e7def7 | 2960 | freelist = get_freelist(s, page); |
6446faa2 | 2961 | |
6faa6833 | 2962 | if (!freelist) { |
03e404af | 2963 | c->page = NULL; |
bd0e7491 | 2964 | local_unlock_irqrestore(&s->cpu_slab->lock, flags); |
03e404af | 2965 | stat(s, DEACTIVATE_BYPASS); |
fc59c053 | 2966 | goto new_slab; |
03e404af | 2967 | } |
6446faa2 | 2968 | |
84e554e6 | 2969 | stat(s, ALLOC_REFILL); |
6446faa2 | 2970 | |
894b8788 | 2971 | load_freelist: |
0b303fb4 | 2972 | |
bd0e7491 | 2973 | lockdep_assert_held(this_cpu_ptr(&s->cpu_slab->lock)); |
0b303fb4 | 2974 | |
507effea CL |
2975 | /* |
2976 | * freelist is pointing to the list of objects to be used. | |
2977 | * page is pointing to the page from which the objects are obtained. | |
2978 | * That page must be frozen for per cpu allocations to work. | |
2979 | */ | |
a0132ac0 | 2980 | VM_BUG_ON(!c->page->frozen); |
6faa6833 | 2981 | c->freelist = get_freepointer(s, freelist); |
8a5ec0ba | 2982 | c->tid = next_tid(c->tid); |
bd0e7491 | 2983 | local_unlock_irqrestore(&s->cpu_slab->lock, flags); |
6faa6833 | 2984 | return freelist; |
81819f0f | 2985 | |
0b303fb4 VB |
2986 | deactivate_slab: |
2987 | ||
bd0e7491 | 2988 | local_lock_irqsave(&s->cpu_slab->lock, flags); |
0b303fb4 | 2989 | if (page != c->page) { |
bd0e7491 | 2990 | local_unlock_irqrestore(&s->cpu_slab->lock, flags); |
0b303fb4 VB |
2991 | goto reread_page; |
2992 | } | |
a019d201 VB |
2993 | freelist = c->freelist; |
2994 | c->page = NULL; | |
2995 | c->freelist = NULL; | |
bd0e7491 | 2996 | local_unlock_irqrestore(&s->cpu_slab->lock, flags); |
cfdf836e | 2997 | deactivate_slab(s, page, freelist); |
0b303fb4 | 2998 | |
81819f0f | 2999 | new_slab: |
2cfb7455 | 3000 | |
a93cf07b | 3001 | if (slub_percpu_partial(c)) { |
bd0e7491 | 3002 | local_lock_irqsave(&s->cpu_slab->lock, flags); |
fa417ab7 | 3003 | if (unlikely(c->page)) { |
bd0e7491 | 3004 | local_unlock_irqrestore(&s->cpu_slab->lock, flags); |
fa417ab7 VB |
3005 | goto reread_page; |
3006 | } | |
4b1f449d | 3007 | if (unlikely(!slub_percpu_partial(c))) { |
bd0e7491 | 3008 | local_unlock_irqrestore(&s->cpu_slab->lock, flags); |
25c00c50 VB |
3009 | /* we were preempted and partial list got empty */ |
3010 | goto new_objects; | |
4b1f449d | 3011 | } |
fa417ab7 | 3012 | |
a93cf07b WY |
3013 | page = c->page = slub_percpu_partial(c); |
3014 | slub_set_percpu_partial(c, page); | |
bd0e7491 | 3015 | local_unlock_irqrestore(&s->cpu_slab->lock, flags); |
49e22585 | 3016 | stat(s, CPU_PARTIAL_ALLOC); |
49e22585 | 3017 | goto redo; |
81819f0f CL |
3018 | } |
3019 | ||
fa417ab7 VB |
3020 | new_objects: |
3021 | ||
75c8ff28 | 3022 | freelist = get_partial(s, gfpflags, node, &page); |
3f2b77e3 | 3023 | if (freelist) |
2a904905 VB |
3024 | goto check_new_page; |
3025 | ||
25c00c50 | 3026 | slub_put_cpu_ptr(s->cpu_slab); |
53a0de06 | 3027 | page = new_slab(s, gfpflags, node); |
25c00c50 | 3028 | c = slub_get_cpu_ptr(s->cpu_slab); |
01ad8a7b | 3029 | |
53a0de06 | 3030 | if (unlikely(!page)) { |
9a02d699 | 3031 | slab_out_of_memory(s, gfpflags, node); |
f4697436 | 3032 | return NULL; |
81819f0f | 3033 | } |
2cfb7455 | 3034 | |
53a0de06 VB |
3035 | /* |
3036 | * No other reference to the page yet so we can | |
3037 | * muck around with it freely without cmpxchg | |
3038 | */ | |
3039 | freelist = page->freelist; | |
3040 | page->freelist = NULL; | |
3041 | ||
3042 | stat(s, ALLOC_SLAB); | |
53a0de06 | 3043 | |
2a904905 | 3044 | check_new_page: |
2cfb7455 | 3045 | |
1572df7c | 3046 | if (kmem_cache_debug(s)) { |
fa417ab7 | 3047 | if (!alloc_debug_processing(s, page, freelist, addr)) { |
1572df7c VB |
3048 | /* Slab failed checks. Next slab needed */ |
3049 | goto new_slab; | |
fa417ab7 | 3050 | } else { |
1572df7c VB |
3051 | /* |
3052 | * For debug case, we don't load freelist so that all | |
3053 | * allocations go through alloc_debug_processing() | |
3054 | */ | |
3055 | goto return_single; | |
fa417ab7 | 3056 | } |
1572df7c VB |
3057 | } |
3058 | ||
3059 | if (unlikely(!pfmemalloc_match(page, gfpflags))) | |
3060 | /* | |
3061 | * For !pfmemalloc_match() case we don't load freelist so that | |
3062 | * we don't make further mismatched allocations easier. | |
3063 | */ | |
3064 | goto return_single; | |
3065 | ||
cfdf836e VB |
3066 | retry_load_page: |
3067 | ||
bd0e7491 | 3068 | local_lock_irqsave(&s->cpu_slab->lock, flags); |
cfdf836e VB |
3069 | if (unlikely(c->page)) { |
3070 | void *flush_freelist = c->freelist; | |
3071 | struct page *flush_page = c->page; | |
3072 | ||
3073 | c->page = NULL; | |
3074 | c->freelist = NULL; | |
3075 | c->tid = next_tid(c->tid); | |
3076 | ||
bd0e7491 | 3077 | local_unlock_irqrestore(&s->cpu_slab->lock, flags); |
cfdf836e VB |
3078 | |
3079 | deactivate_slab(s, flush_page, flush_freelist); | |
3080 | ||
3081 | stat(s, CPUSLAB_FLUSH); | |
3082 | ||
3083 | goto retry_load_page; | |
3084 | } | |
3f2b77e3 VB |
3085 | c->page = page; |
3086 | ||
1572df7c VB |
3087 | goto load_freelist; |
3088 | ||
3089 | return_single: | |
894b8788 | 3090 | |
a019d201 | 3091 | deactivate_slab(s, page, get_freepointer(s, freelist)); |
6faa6833 | 3092 | return freelist; |
894b8788 CL |
3093 | } |
3094 | ||
a380a3c7 | 3095 | /* |
e500059b VB |
3096 | * A wrapper for ___slab_alloc() for contexts where preemption is not yet |
3097 | * disabled. Compensates for possible cpu changes by refetching the per cpu area | |
3098 | * pointer. | |
a380a3c7 CL |
3099 | */ |
3100 | static void *__slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node, | |
3101 | unsigned long addr, struct kmem_cache_cpu *c) | |
3102 | { | |
3103 | void *p; | |
a380a3c7 | 3104 | |
e500059b | 3105 | #ifdef CONFIG_PREEMPT_COUNT |
a380a3c7 CL |
3106 | /* |
3107 | * We may have been preempted and rescheduled on a different | |
e500059b | 3108 | * cpu before disabling preemption. Need to reload cpu area |
a380a3c7 CL |
3109 | * pointer. |
3110 | */ | |
25c00c50 | 3111 | c = slub_get_cpu_ptr(s->cpu_slab); |
a380a3c7 CL |
3112 | #endif |
3113 | ||
3114 | p = ___slab_alloc(s, gfpflags, node, addr, c); | |
e500059b | 3115 | #ifdef CONFIG_PREEMPT_COUNT |
25c00c50 | 3116 | slub_put_cpu_ptr(s->cpu_slab); |
e500059b | 3117 | #endif |
a380a3c7 CL |
3118 | return p; |
3119 | } | |
3120 | ||
0f181f9f AP |
3121 | /* |
3122 | * If the object has been wiped upon free, make sure it's fully initialized by | |
3123 | * zeroing out freelist pointer. | |
3124 | */ | |
3125 | static __always_inline void maybe_wipe_obj_freeptr(struct kmem_cache *s, | |
3126 | void *obj) | |
3127 | { | |
3128 | if (unlikely(slab_want_init_on_free(s)) && obj) | |
ce5716c6 AK |
3129 | memset((void *)((char *)kasan_reset_tag(obj) + s->offset), |
3130 | 0, sizeof(void *)); | |
0f181f9f AP |
3131 | } |
3132 | ||
894b8788 CL |
3133 | /* |
3134 | * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc) | |
3135 | * have the fastpath folded into their functions. So no function call | |
3136 | * overhead for requests that can be satisfied on the fastpath. | |
3137 | * | |
3138 | * The fastpath works by first checking if the lockless freelist can be used. | |
3139 | * If not then __slab_alloc is called for slow processing. | |
3140 | * | |
3141 | * Otherwise we can simply pick the next object from the lockless free list. | |
3142 | */ | |
2b847c3c | 3143 | static __always_inline void *slab_alloc_node(struct kmem_cache *s, |
b89fb5ef | 3144 | gfp_t gfpflags, int node, unsigned long addr, size_t orig_size) |
894b8788 | 3145 | { |
03ec0ed5 | 3146 | void *object; |
dfb4f096 | 3147 | struct kmem_cache_cpu *c; |
57d437d2 | 3148 | struct page *page; |
8a5ec0ba | 3149 | unsigned long tid; |
964d4bd3 | 3150 | struct obj_cgroup *objcg = NULL; |
da844b78 | 3151 | bool init = false; |
1f84260c | 3152 | |
964d4bd3 | 3153 | s = slab_pre_alloc_hook(s, &objcg, 1, gfpflags); |
8135be5a | 3154 | if (!s) |
773ff60e | 3155 | return NULL; |
b89fb5ef AP |
3156 | |
3157 | object = kfence_alloc(s, orig_size, gfpflags); | |
3158 | if (unlikely(object)) | |
3159 | goto out; | |
3160 | ||
8a5ec0ba | 3161 | redo: |
8a5ec0ba CL |
3162 | /* |
3163 | * Must read kmem_cache cpu data via this cpu ptr. Preemption is | |
3164 | * enabled. We may switch back and forth between cpus while | |
3165 | * reading from one cpu area. That does not matter as long | |
3166 | * as we end up on the original cpu again when doing the cmpxchg. | |
7cccd80b | 3167 | * |
9b4bc85a VB |
3168 | * We must guarantee that tid and kmem_cache_cpu are retrieved on the |
3169 | * same cpu. We read first the kmem_cache_cpu pointer and use it to read | |
3170 | * the tid. If we are preempted and switched to another cpu between the | |
3171 | * two reads, it's OK as the two are still associated with the same cpu | |
3172 | * and cmpxchg later will validate the cpu. | |
8a5ec0ba | 3173 | */ |
9b4bc85a VB |
3174 | c = raw_cpu_ptr(s->cpu_slab); |
3175 | tid = READ_ONCE(c->tid); | |
9aabf810 JK |
3176 | |
3177 | /* | |
3178 | * Irqless object alloc/free algorithm used here depends on sequence | |
3179 | * of fetching cpu_slab's data. tid should be fetched before anything | |
3180 | * on c to guarantee that object and page associated with previous tid | |
3181 | * won't be used with current tid. If we fetch tid first, object and | |
3182 | * page could be one associated with next tid and our alloc/free | |
3183 | * request will be failed. In this case, we will retry. So, no problem. | |
3184 | */ | |
3185 | barrier(); | |
8a5ec0ba | 3186 | |
8a5ec0ba CL |
3187 | /* |
3188 | * The transaction ids are globally unique per cpu and per operation on | |
3189 | * a per cpu queue. Thus they can be guarantee that the cmpxchg_double | |
3190 | * occurs on the right processor and that there was no operation on the | |
3191 | * linked list in between. | |
3192 | */ | |
8a5ec0ba | 3193 | |
9dfc6e68 | 3194 | object = c->freelist; |
57d437d2 | 3195 | page = c->page; |
bd0e7491 VB |
3196 | /* |
3197 | * We cannot use the lockless fastpath on PREEMPT_RT because if a | |
3198 | * slowpath has taken the local_lock_irqsave(), it is not protected | |
3199 | * against a fast path operation in an irq handler. So we need to take | |
3200 | * the slow path which uses local_lock. It is still relatively fast if | |
3201 | * there is a suitable cpu freelist. | |
3202 | */ | |
3203 | if (IS_ENABLED(CONFIG_PREEMPT_RT) || | |
3204 | unlikely(!object || !page || !node_match(page, node))) { | |
dfb4f096 | 3205 | object = __slab_alloc(s, gfpflags, node, addr, c); |
8eae1492 | 3206 | } else { |
0ad9500e ED |
3207 | void *next_object = get_freepointer_safe(s, object); |
3208 | ||
8a5ec0ba | 3209 | /* |
25985edc | 3210 | * The cmpxchg will only match if there was no additional |
8a5ec0ba CL |
3211 | * operation and if we are on the right processor. |
3212 | * | |
d0e0ac97 CG |
3213 | * The cmpxchg does the following atomically (without lock |
3214 | * semantics!) | |
8a5ec0ba CL |
3215 | * 1. Relocate first pointer to the current per cpu area. |
3216 | * 2. Verify that tid and freelist have not been changed | |
3217 | * 3. If they were not changed replace tid and freelist | |
3218 | * | |
d0e0ac97 CG |
3219 | * Since this is without lock semantics the protection is only |
3220 | * against code executing on this cpu *not* from access by | |
3221 | * other cpus. | |
8a5ec0ba | 3222 | */ |
933393f5 | 3223 | if (unlikely(!this_cpu_cmpxchg_double( |
8a5ec0ba CL |
3224 | s->cpu_slab->freelist, s->cpu_slab->tid, |
3225 | object, tid, | |
0ad9500e | 3226 | next_object, next_tid(tid)))) { |
8a5ec0ba CL |
3227 | |
3228 | note_cmpxchg_failure("slab_alloc", s, tid); | |
3229 | goto redo; | |
3230 | } | |
0ad9500e | 3231 | prefetch_freepointer(s, next_object); |
84e554e6 | 3232 | stat(s, ALLOC_FASTPATH); |
894b8788 | 3233 | } |
0f181f9f | 3234 | |
ce5716c6 | 3235 | maybe_wipe_obj_freeptr(s, object); |
da844b78 | 3236 | init = slab_want_init_on_alloc(gfpflags, s); |
d07dbea4 | 3237 | |
b89fb5ef | 3238 | out: |
da844b78 | 3239 | slab_post_alloc_hook(s, objcg, gfpflags, 1, &object, init); |
5a896d9e | 3240 | |
894b8788 | 3241 | return object; |
81819f0f CL |
3242 | } |
3243 | ||
2b847c3c | 3244 | static __always_inline void *slab_alloc(struct kmem_cache *s, |
b89fb5ef | 3245 | gfp_t gfpflags, unsigned long addr, size_t orig_size) |
2b847c3c | 3246 | { |
b89fb5ef | 3247 | return slab_alloc_node(s, gfpflags, NUMA_NO_NODE, addr, orig_size); |
2b847c3c EG |
3248 | } |
3249 | ||
81819f0f CL |
3250 | void *kmem_cache_alloc(struct kmem_cache *s, gfp_t gfpflags) |
3251 | { | |
b89fb5ef | 3252 | void *ret = slab_alloc(s, gfpflags, _RET_IP_, s->object_size); |
5b882be4 | 3253 | |
d0e0ac97 CG |
3254 | trace_kmem_cache_alloc(_RET_IP_, ret, s->object_size, |
3255 | s->size, gfpflags); | |
5b882be4 EGM |
3256 | |
3257 | return ret; | |
81819f0f CL |
3258 | } |
3259 | EXPORT_SYMBOL(kmem_cache_alloc); | |
3260 | ||
0f24f128 | 3261 | #ifdef CONFIG_TRACING |
4a92379b RK |
3262 | void *kmem_cache_alloc_trace(struct kmem_cache *s, gfp_t gfpflags, size_t size) |
3263 | { | |
b89fb5ef | 3264 | void *ret = slab_alloc(s, gfpflags, _RET_IP_, size); |
4a92379b | 3265 | trace_kmalloc(_RET_IP_, ret, size, s->size, gfpflags); |
0116523c | 3266 | ret = kasan_kmalloc(s, ret, size, gfpflags); |
4a92379b RK |
3267 | return ret; |
3268 | } | |
3269 | EXPORT_SYMBOL(kmem_cache_alloc_trace); | |
5b882be4 EGM |
3270 | #endif |
3271 | ||
81819f0f CL |
3272 | #ifdef CONFIG_NUMA |
3273 | void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t gfpflags, int node) | |
3274 | { | |
b89fb5ef | 3275 | void *ret = slab_alloc_node(s, gfpflags, node, _RET_IP_, s->object_size); |
5b882be4 | 3276 | |
ca2b84cb | 3277 | trace_kmem_cache_alloc_node(_RET_IP_, ret, |
3b0efdfa | 3278 | s->object_size, s->size, gfpflags, node); |
5b882be4 EGM |
3279 | |
3280 | return ret; | |
81819f0f CL |
3281 | } |
3282 | EXPORT_SYMBOL(kmem_cache_alloc_node); | |
81819f0f | 3283 | |
0f24f128 | 3284 | #ifdef CONFIG_TRACING |
4a92379b | 3285 | void *kmem_cache_alloc_node_trace(struct kmem_cache *s, |
5b882be4 | 3286 | gfp_t gfpflags, |
4a92379b | 3287 | int node, size_t size) |
5b882be4 | 3288 | { |
b89fb5ef | 3289 | void *ret = slab_alloc_node(s, gfpflags, node, _RET_IP_, size); |
4a92379b RK |
3290 | |
3291 | trace_kmalloc_node(_RET_IP_, ret, | |
3292 | size, s->size, gfpflags, node); | |
0316bec2 | 3293 | |
0116523c | 3294 | ret = kasan_kmalloc(s, ret, size, gfpflags); |
4a92379b | 3295 | return ret; |
5b882be4 | 3296 | } |
4a92379b | 3297 | EXPORT_SYMBOL(kmem_cache_alloc_node_trace); |
5b882be4 | 3298 | #endif |
6dfd1b65 | 3299 | #endif /* CONFIG_NUMA */ |
5b882be4 | 3300 | |
81819f0f | 3301 | /* |
94e4d712 | 3302 | * Slow path handling. This may still be called frequently since objects |
894b8788 | 3303 | * have a longer lifetime than the cpu slabs in most processing loads. |
81819f0f | 3304 | * |
894b8788 CL |
3305 | * So we still attempt to reduce cache line usage. Just take the slab |
3306 | * lock and free the item. If there is no additional partial page | |
3307 | * handling required then we can return immediately. | |
81819f0f | 3308 | */ |
894b8788 | 3309 | static void __slab_free(struct kmem_cache *s, struct page *page, |
81084651 JDB |
3310 | void *head, void *tail, int cnt, |
3311 | unsigned long addr) | |
3312 | ||
81819f0f CL |
3313 | { |
3314 | void *prior; | |
2cfb7455 | 3315 | int was_frozen; |
2cfb7455 CL |
3316 | struct page new; |
3317 | unsigned long counters; | |
3318 | struct kmem_cache_node *n = NULL; | |
3f649ab7 | 3319 | unsigned long flags; |
81819f0f | 3320 | |
8a5ec0ba | 3321 | stat(s, FREE_SLOWPATH); |
81819f0f | 3322 | |
b89fb5ef AP |
3323 | if (kfence_free(head)) |
3324 | return; | |
3325 | ||
19c7ff9e | 3326 | if (kmem_cache_debug(s) && |
282acb43 | 3327 | !free_debug_processing(s, page, head, tail, cnt, addr)) |
80f08c19 | 3328 | return; |
6446faa2 | 3329 | |
2cfb7455 | 3330 | do { |
837d678d JK |
3331 | if (unlikely(n)) { |
3332 | spin_unlock_irqrestore(&n->list_lock, flags); | |
3333 | n = NULL; | |
3334 | } | |
2cfb7455 CL |
3335 | prior = page->freelist; |
3336 | counters = page->counters; | |
81084651 | 3337 | set_freepointer(s, tail, prior); |
2cfb7455 CL |
3338 | new.counters = counters; |
3339 | was_frozen = new.frozen; | |
81084651 | 3340 | new.inuse -= cnt; |
837d678d | 3341 | if ((!new.inuse || !prior) && !was_frozen) { |
49e22585 | 3342 | |
c65c1877 | 3343 | if (kmem_cache_has_cpu_partial(s) && !prior) { |
49e22585 CL |
3344 | |
3345 | /* | |
d0e0ac97 CG |
3346 | * Slab was on no list before and will be |
3347 | * partially empty | |
3348 | * We can defer the list move and instead | |
3349 | * freeze it. | |
49e22585 CL |
3350 | */ |
3351 | new.frozen = 1; | |
3352 | ||
c65c1877 | 3353 | } else { /* Needs to be taken off a list */ |
49e22585 | 3354 | |
b455def2 | 3355 | n = get_node(s, page_to_nid(page)); |
49e22585 CL |
3356 | /* |
3357 | * Speculatively acquire the list_lock. | |
3358 | * If the cmpxchg does not succeed then we may | |
3359 | * drop the list_lock without any processing. | |
3360 | * | |
3361 | * Otherwise the list_lock will synchronize with | |
3362 | * other processors updating the list of slabs. | |
3363 | */ | |
3364 | spin_lock_irqsave(&n->list_lock, flags); | |
3365 | ||
3366 | } | |
2cfb7455 | 3367 | } |
81819f0f | 3368 | |
2cfb7455 CL |
3369 | } while (!cmpxchg_double_slab(s, page, |
3370 | prior, counters, | |
81084651 | 3371 | head, new.counters, |
2cfb7455 | 3372 | "__slab_free")); |
81819f0f | 3373 | |
2cfb7455 | 3374 | if (likely(!n)) { |
49e22585 | 3375 | |
c270cf30 AW |
3376 | if (likely(was_frozen)) { |
3377 | /* | |
3378 | * The list lock was not taken therefore no list | |
3379 | * activity can be necessary. | |
3380 | */ | |
3381 | stat(s, FREE_FROZEN); | |
3382 | } else if (new.frozen) { | |
3383 | /* | |
3384 | * If we just froze the page then put it onto the | |
3385 | * per cpu partial list. | |
3386 | */ | |
49e22585 | 3387 | put_cpu_partial(s, page, 1); |
8028dcea AS |
3388 | stat(s, CPU_PARTIAL_FREE); |
3389 | } | |
c270cf30 | 3390 | |
b455def2 L |
3391 | return; |
3392 | } | |
81819f0f | 3393 | |
8a5b20ae | 3394 | if (unlikely(!new.inuse && n->nr_partial >= s->min_partial)) |
837d678d JK |
3395 | goto slab_empty; |
3396 | ||
81819f0f | 3397 | /* |
837d678d JK |
3398 | * Objects left in the slab. If it was not on the partial list before |
3399 | * then add it. | |
81819f0f | 3400 | */ |
345c905d | 3401 | if (!kmem_cache_has_cpu_partial(s) && unlikely(!prior)) { |
a4d3f891 | 3402 | remove_full(s, n, page); |
837d678d JK |
3403 | add_partial(n, page, DEACTIVATE_TO_TAIL); |
3404 | stat(s, FREE_ADD_PARTIAL); | |
8ff12cfc | 3405 | } |
80f08c19 | 3406 | spin_unlock_irqrestore(&n->list_lock, flags); |
81819f0f CL |
3407 | return; |
3408 | ||
3409 | slab_empty: | |
a973e9dd | 3410 | if (prior) { |
81819f0f | 3411 | /* |
6fbabb20 | 3412 | * Slab on the partial list. |
81819f0f | 3413 | */ |
5cc6eee8 | 3414 | remove_partial(n, page); |
84e554e6 | 3415 | stat(s, FREE_REMOVE_PARTIAL); |
c65c1877 | 3416 | } else { |
6fbabb20 | 3417 | /* Slab must be on the full list */ |
c65c1877 PZ |
3418 | remove_full(s, n, page); |
3419 | } | |
2cfb7455 | 3420 | |
80f08c19 | 3421 | spin_unlock_irqrestore(&n->list_lock, flags); |
84e554e6 | 3422 | stat(s, FREE_SLAB); |
81819f0f | 3423 | discard_slab(s, page); |
81819f0f CL |
3424 | } |
3425 | ||
894b8788 CL |
3426 | /* |
3427 | * Fastpath with forced inlining to produce a kfree and kmem_cache_free that | |
3428 | * can perform fastpath freeing without additional function calls. | |
3429 | * | |
3430 | * The fastpath is only possible if we are freeing to the current cpu slab | |
3431 | * of this processor. This typically the case if we have just allocated | |
3432 | * the item before. | |
3433 | * | |
3434 | * If fastpath is not possible then fall back to __slab_free where we deal | |
3435 | * with all sorts of special processing. | |
81084651 JDB |
3436 | * |
3437 | * Bulk free of a freelist with several objects (all pointing to the | |
3438 | * same page) possible by specifying head and tail ptr, plus objects | |
3439 | * count (cnt). Bulk free indicated by tail pointer being set. | |
894b8788 | 3440 | */ |
80a9201a AP |
3441 | static __always_inline void do_slab_free(struct kmem_cache *s, |
3442 | struct page *page, void *head, void *tail, | |
3443 | int cnt, unsigned long addr) | |
894b8788 | 3444 | { |
81084651 | 3445 | void *tail_obj = tail ? : head; |
dfb4f096 | 3446 | struct kmem_cache_cpu *c; |
8a5ec0ba | 3447 | unsigned long tid; |
964d4bd3 | 3448 | |
3ddd6026 ML |
3449 | /* memcg_slab_free_hook() is already called for bulk free. */ |
3450 | if (!tail) | |
3451 | memcg_slab_free_hook(s, &head, 1); | |
8a5ec0ba CL |
3452 | redo: |
3453 | /* | |
3454 | * Determine the currently cpus per cpu slab. | |
3455 | * The cpu may change afterward. However that does not matter since | |
3456 | * data is retrieved via this pointer. If we are on the same cpu | |
2ae44005 | 3457 | * during the cmpxchg then the free will succeed. |
8a5ec0ba | 3458 | */ |
9b4bc85a VB |
3459 | c = raw_cpu_ptr(s->cpu_slab); |
3460 | tid = READ_ONCE(c->tid); | |
c016b0bd | 3461 | |
9aabf810 JK |
3462 | /* Same with comment on barrier() in slab_alloc_node() */ |
3463 | barrier(); | |
c016b0bd | 3464 | |
442b06bc | 3465 | if (likely(page == c->page)) { |
bd0e7491 | 3466 | #ifndef CONFIG_PREEMPT_RT |
5076190d LT |
3467 | void **freelist = READ_ONCE(c->freelist); |
3468 | ||
3469 | set_freepointer(s, tail_obj, freelist); | |
8a5ec0ba | 3470 | |
933393f5 | 3471 | if (unlikely(!this_cpu_cmpxchg_double( |
8a5ec0ba | 3472 | s->cpu_slab->freelist, s->cpu_slab->tid, |
5076190d | 3473 | freelist, tid, |
81084651 | 3474 | head, next_tid(tid)))) { |
8a5ec0ba CL |
3475 | |
3476 | note_cmpxchg_failure("slab_free", s, tid); | |
3477 | goto redo; | |
3478 | } | |
bd0e7491 VB |
3479 | #else /* CONFIG_PREEMPT_RT */ |
3480 | /* | |
3481 | * We cannot use the lockless fastpath on PREEMPT_RT because if | |
3482 | * a slowpath has taken the local_lock_irqsave(), it is not | |
3483 | * protected against a fast path operation in an irq handler. So | |
3484 | * we need to take the local_lock. We shouldn't simply defer to | |
3485 | * __slab_free() as that wouldn't use the cpu freelist at all. | |
3486 | */ | |
3487 | void **freelist; | |
3488 | ||
3489 | local_lock(&s->cpu_slab->lock); | |
3490 | c = this_cpu_ptr(s->cpu_slab); | |
3491 | if (unlikely(page != c->page)) { | |
3492 | local_unlock(&s->cpu_slab->lock); | |
3493 | goto redo; | |
3494 | } | |
3495 | tid = c->tid; | |
3496 | freelist = c->freelist; | |
3497 | ||
3498 | set_freepointer(s, tail_obj, freelist); | |
3499 | c->freelist = head; | |
3500 | c->tid = next_tid(tid); | |
3501 | ||
3502 | local_unlock(&s->cpu_slab->lock); | |
3503 | #endif | |
84e554e6 | 3504 | stat(s, FREE_FASTPATH); |
894b8788 | 3505 | } else |
81084651 | 3506 | __slab_free(s, page, head, tail_obj, cnt, addr); |
894b8788 | 3507 | |
894b8788 CL |
3508 | } |
3509 | ||
80a9201a AP |
3510 | static __always_inline void slab_free(struct kmem_cache *s, struct page *page, |
3511 | void *head, void *tail, int cnt, | |
3512 | unsigned long addr) | |
3513 | { | |
80a9201a | 3514 | /* |
c3895391 AK |
3515 | * With KASAN enabled slab_free_freelist_hook modifies the freelist |
3516 | * to remove objects, whose reuse must be delayed. | |
80a9201a | 3517 | */ |
899447f6 | 3518 | if (slab_free_freelist_hook(s, &head, &tail, &cnt)) |
c3895391 | 3519 | do_slab_free(s, page, head, tail, cnt, addr); |
80a9201a AP |
3520 | } |
3521 | ||
2bd926b4 | 3522 | #ifdef CONFIG_KASAN_GENERIC |
80a9201a AP |
3523 | void ___cache_free(struct kmem_cache *cache, void *x, unsigned long addr) |
3524 | { | |
d835eef4 | 3525 | do_slab_free(cache, slab_page(virt_to_slab(x)), x, NULL, 1, addr); |
80a9201a AP |
3526 | } |
3527 | #endif | |
3528 | ||
81819f0f CL |
3529 | void kmem_cache_free(struct kmem_cache *s, void *x) |
3530 | { | |
b9ce5ef4 GC |
3531 | s = cache_from_obj(s, x); |
3532 | if (!s) | |
79576102 | 3533 | return; |
3544de8e | 3534 | trace_kmem_cache_free(_RET_IP_, x, s->name); |
d835eef4 | 3535 | slab_free(s, slab_page(virt_to_slab(x)), x, NULL, 1, _RET_IP_); |
81819f0f CL |
3536 | } |
3537 | EXPORT_SYMBOL(kmem_cache_free); | |
3538 | ||
d0ecd894 | 3539 | struct detached_freelist { |
cc465c3b | 3540 | struct slab *slab; |
d0ecd894 JDB |
3541 | void *tail; |
3542 | void *freelist; | |
3543 | int cnt; | |
376bf125 | 3544 | struct kmem_cache *s; |
d0ecd894 | 3545 | }; |
fbd02630 | 3546 | |
d835eef4 | 3547 | static inline void free_large_kmalloc(struct folio *folio, void *object) |
f227f0fa | 3548 | { |
d835eef4 | 3549 | unsigned int order = folio_order(folio); |
f227f0fa | 3550 | |
d835eef4 | 3551 | if (WARN_ON_ONCE(order == 0)) |
d0fe47c6 KW |
3552 | pr_warn_once("object pointer: 0x%p\n", object); |
3553 | ||
1ed7ce57 | 3554 | kfree_hook(object); |
d835eef4 MWO |
3555 | mod_lruvec_page_state(folio_page(folio, 0), NR_SLAB_UNRECLAIMABLE_B, |
3556 | -(PAGE_SIZE << order)); | |
3557 | __free_pages(folio_page(folio, 0), order); | |
f227f0fa SB |
3558 | } |
3559 | ||
d0ecd894 JDB |
3560 | /* |
3561 | * This function progressively scans the array with free objects (with | |
3562 | * a limited look ahead) and extract objects belonging to the same | |
cc465c3b MWO |
3563 | * slab. It builds a detached freelist directly within the given |
3564 | * slab/objects. This can happen without any need for | |
d0ecd894 JDB |
3565 | * synchronization, because the objects are owned by running process. |
3566 | * The freelist is build up as a single linked list in the objects. | |
3567 | * The idea is, that this detached freelist can then be bulk | |
3568 | * transferred to the real freelist(s), but only requiring a single | |
3569 | * synchronization primitive. Look ahead in the array is limited due | |
3570 | * to performance reasons. | |
3571 | */ | |
376bf125 JDB |
3572 | static inline |
3573 | int build_detached_freelist(struct kmem_cache *s, size_t size, | |
3574 | void **p, struct detached_freelist *df) | |
d0ecd894 JDB |
3575 | { |
3576 | size_t first_skipped_index = 0; | |
3577 | int lookahead = 3; | |
3578 | void *object; | |
cc465c3b MWO |
3579 | struct folio *folio; |
3580 | struct slab *slab; | |
fbd02630 | 3581 | |
d0ecd894 | 3582 | /* Always re-init detached_freelist */ |
cc465c3b | 3583 | df->slab = NULL; |
fbd02630 | 3584 | |
d0ecd894 JDB |
3585 | do { |
3586 | object = p[--size]; | |
ca257195 | 3587 | /* Do we need !ZERO_OR_NULL_PTR(object) here? (for kfree) */ |
d0ecd894 | 3588 | } while (!object && size); |
3eed034d | 3589 | |
d0ecd894 JDB |
3590 | if (!object) |
3591 | return 0; | |
fbd02630 | 3592 | |
cc465c3b | 3593 | folio = virt_to_folio(object); |
ca257195 JDB |
3594 | if (!s) { |
3595 | /* Handle kalloc'ed objects */ | |
cc465c3b | 3596 | if (unlikely(!folio_test_slab(folio))) { |
d835eef4 | 3597 | free_large_kmalloc(folio, object); |
ca257195 JDB |
3598 | p[size] = NULL; /* mark object processed */ |
3599 | return size; | |
3600 | } | |
3601 | /* Derive kmem_cache from object */ | |
cc465c3b MWO |
3602 | slab = folio_slab(folio); |
3603 | df->s = slab->slab_cache; | |
ca257195 | 3604 | } else { |
cc465c3b | 3605 | slab = folio_slab(folio); |
ca257195 JDB |
3606 | df->s = cache_from_obj(s, object); /* Support for memcg */ |
3607 | } | |
376bf125 | 3608 | |
b89fb5ef | 3609 | if (is_kfence_address(object)) { |
d57a964e | 3610 | slab_free_hook(df->s, object, false); |
b89fb5ef AP |
3611 | __kfence_free(object); |
3612 | p[size] = NULL; /* mark object processed */ | |
3613 | return size; | |
3614 | } | |
3615 | ||
d0ecd894 | 3616 | /* Start new detached freelist */ |
cc465c3b | 3617 | df->slab = slab; |
376bf125 | 3618 | set_freepointer(df->s, object, NULL); |
d0ecd894 JDB |
3619 | df->tail = object; |
3620 | df->freelist = object; | |
3621 | p[size] = NULL; /* mark object processed */ | |
3622 | df->cnt = 1; | |
3623 | ||
3624 | while (size) { | |
3625 | object = p[--size]; | |
3626 | if (!object) | |
3627 | continue; /* Skip processed objects */ | |
3628 | ||
cc465c3b MWO |
3629 | /* df->slab is always set at this point */ |
3630 | if (df->slab == virt_to_slab(object)) { | |
d0ecd894 | 3631 | /* Opportunity build freelist */ |
376bf125 | 3632 | set_freepointer(df->s, object, df->freelist); |
d0ecd894 JDB |
3633 | df->freelist = object; |
3634 | df->cnt++; | |
3635 | p[size] = NULL; /* mark object processed */ | |
3636 | ||
3637 | continue; | |
fbd02630 | 3638 | } |
d0ecd894 JDB |
3639 | |
3640 | /* Limit look ahead search */ | |
3641 | if (!--lookahead) | |
3642 | break; | |
3643 | ||
3644 | if (!first_skipped_index) | |
3645 | first_skipped_index = size + 1; | |
fbd02630 | 3646 | } |
d0ecd894 JDB |
3647 | |
3648 | return first_skipped_index; | |
3649 | } | |
3650 | ||
d0ecd894 | 3651 | /* Note that interrupts must be enabled when calling this function. */ |
376bf125 | 3652 | void kmem_cache_free_bulk(struct kmem_cache *s, size_t size, void **p) |
d0ecd894 JDB |
3653 | { |
3654 | if (WARN_ON(!size)) | |
3655 | return; | |
3656 | ||
d1b2cf6c | 3657 | memcg_slab_free_hook(s, p, size); |
d0ecd894 JDB |
3658 | do { |
3659 | struct detached_freelist df; | |
3660 | ||
3661 | size = build_detached_freelist(s, size, p, &df); | |
cc465c3b | 3662 | if (!df.slab) |
d0ecd894 JDB |
3663 | continue; |
3664 | ||
cc465c3b | 3665 | slab_free(df.s, slab_page(df.slab), df.freelist, df.tail, df.cnt, _RET_IP_); |
d0ecd894 | 3666 | } while (likely(size)); |
484748f0 CL |
3667 | } |
3668 | EXPORT_SYMBOL(kmem_cache_free_bulk); | |
3669 | ||
994eb764 | 3670 | /* Note that interrupts must be enabled when calling this function. */ |
865762a8 JDB |
3671 | int kmem_cache_alloc_bulk(struct kmem_cache *s, gfp_t flags, size_t size, |
3672 | void **p) | |
484748f0 | 3673 | { |
994eb764 JDB |
3674 | struct kmem_cache_cpu *c; |
3675 | int i; | |
964d4bd3 | 3676 | struct obj_cgroup *objcg = NULL; |
994eb764 | 3677 | |
03ec0ed5 | 3678 | /* memcg and kmem_cache debug support */ |
964d4bd3 | 3679 | s = slab_pre_alloc_hook(s, &objcg, size, flags); |
03ec0ed5 JDB |
3680 | if (unlikely(!s)) |
3681 | return false; | |
994eb764 JDB |
3682 | /* |
3683 | * Drain objects in the per cpu slab, while disabling local | |
3684 | * IRQs, which protects against PREEMPT and interrupts | |
3685 | * handlers invoking normal fastpath. | |
3686 | */ | |
25c00c50 | 3687 | c = slub_get_cpu_ptr(s->cpu_slab); |
bd0e7491 | 3688 | local_lock_irq(&s->cpu_slab->lock); |
994eb764 JDB |
3689 | |
3690 | for (i = 0; i < size; i++) { | |
b89fb5ef | 3691 | void *object = kfence_alloc(s, s->object_size, flags); |
994eb764 | 3692 | |
b89fb5ef AP |
3693 | if (unlikely(object)) { |
3694 | p[i] = object; | |
3695 | continue; | |
3696 | } | |
3697 | ||
3698 | object = c->freelist; | |
ebe909e0 | 3699 | if (unlikely(!object)) { |
fd4d9c7d JH |
3700 | /* |
3701 | * We may have removed an object from c->freelist using | |
3702 | * the fastpath in the previous iteration; in that case, | |
3703 | * c->tid has not been bumped yet. | |
3704 | * Since ___slab_alloc() may reenable interrupts while | |
3705 | * allocating memory, we should bump c->tid now. | |
3706 | */ | |
3707 | c->tid = next_tid(c->tid); | |
3708 | ||
bd0e7491 | 3709 | local_unlock_irq(&s->cpu_slab->lock); |
e500059b | 3710 | |
ebe909e0 JDB |
3711 | /* |
3712 | * Invoking slow path likely have side-effect | |
3713 | * of re-populating per CPU c->freelist | |
3714 | */ | |
87098373 | 3715 | p[i] = ___slab_alloc(s, flags, NUMA_NO_NODE, |
ebe909e0 | 3716 | _RET_IP_, c); |
87098373 CL |
3717 | if (unlikely(!p[i])) |
3718 | goto error; | |
3719 | ||
ebe909e0 | 3720 | c = this_cpu_ptr(s->cpu_slab); |
0f181f9f AP |
3721 | maybe_wipe_obj_freeptr(s, p[i]); |
3722 | ||
bd0e7491 | 3723 | local_lock_irq(&s->cpu_slab->lock); |
e500059b | 3724 | |
ebe909e0 JDB |
3725 | continue; /* goto for-loop */ |
3726 | } | |
994eb764 JDB |
3727 | c->freelist = get_freepointer(s, object); |
3728 | p[i] = object; | |
0f181f9f | 3729 | maybe_wipe_obj_freeptr(s, p[i]); |
994eb764 JDB |
3730 | } |
3731 | c->tid = next_tid(c->tid); | |
bd0e7491 | 3732 | local_unlock_irq(&s->cpu_slab->lock); |
25c00c50 | 3733 | slub_put_cpu_ptr(s->cpu_slab); |
994eb764 | 3734 | |
da844b78 AK |
3735 | /* |
3736 | * memcg and kmem_cache debug support and memory initialization. | |
3737 | * Done outside of the IRQ disabled fastpath loop. | |
3738 | */ | |
3739 | slab_post_alloc_hook(s, objcg, flags, size, p, | |
3740 | slab_want_init_on_alloc(flags, s)); | |
865762a8 | 3741 | return i; |
87098373 | 3742 | error: |
25c00c50 | 3743 | slub_put_cpu_ptr(s->cpu_slab); |
da844b78 | 3744 | slab_post_alloc_hook(s, objcg, flags, i, p, false); |
03ec0ed5 | 3745 | __kmem_cache_free_bulk(s, i, p); |
865762a8 | 3746 | return 0; |
484748f0 CL |
3747 | } |
3748 | EXPORT_SYMBOL(kmem_cache_alloc_bulk); | |
3749 | ||
3750 | ||
81819f0f | 3751 | /* |
672bba3a CL |
3752 | * Object placement in a slab is made very easy because we always start at |
3753 | * offset 0. If we tune the size of the object to the alignment then we can | |
3754 | * get the required alignment by putting one properly sized object after | |
3755 | * another. | |
81819f0f CL |
3756 | * |
3757 | * Notice that the allocation order determines the sizes of the per cpu | |
3758 | * caches. Each processor has always one slab available for allocations. | |
3759 | * Increasing the allocation order reduces the number of times that slabs | |
672bba3a | 3760 | * must be moved on and off the partial lists and is therefore a factor in |
81819f0f | 3761 | * locking overhead. |
81819f0f CL |
3762 | */ |
3763 | ||
3764 | /* | |
f0953a1b | 3765 | * Minimum / Maximum order of slab pages. This influences locking overhead |
81819f0f CL |
3766 | * and slab fragmentation. A higher order reduces the number of partial slabs |
3767 | * and increases the number of allocations possible without having to | |
3768 | * take the list_lock. | |
3769 | */ | |
19af27af AD |
3770 | static unsigned int slub_min_order; |
3771 | static unsigned int slub_max_order = PAGE_ALLOC_COSTLY_ORDER; | |
3772 | static unsigned int slub_min_objects; | |
81819f0f | 3773 | |
81819f0f CL |
3774 | /* |
3775 | * Calculate the order of allocation given an slab object size. | |
3776 | * | |
672bba3a CL |
3777 | * The order of allocation has significant impact on performance and other |
3778 | * system components. Generally order 0 allocations should be preferred since | |
3779 | * order 0 does not cause fragmentation in the page allocator. Larger objects | |
3780 | * be problematic to put into order 0 slabs because there may be too much | |
c124f5b5 | 3781 | * unused space left. We go to a higher order if more than 1/16th of the slab |
672bba3a CL |
3782 | * would be wasted. |
3783 | * | |
3784 | * In order to reach satisfactory performance we must ensure that a minimum | |
3785 | * number of objects is in one slab. Otherwise we may generate too much | |
3786 | * activity on the partial lists which requires taking the list_lock. This is | |
3787 | * less a concern for large slabs though which are rarely used. | |
81819f0f | 3788 | * |
672bba3a CL |
3789 | * slub_max_order specifies the order where we begin to stop considering the |
3790 | * number of objects in a slab as critical. If we reach slub_max_order then | |
3791 | * we try to keep the page order as low as possible. So we accept more waste | |
3792 | * of space in favor of a small page order. | |
81819f0f | 3793 | * |
672bba3a CL |
3794 | * Higher order allocations also allow the placement of more objects in a |
3795 | * slab and thereby reduce object handling overhead. If the user has | |
dc84207d | 3796 | * requested a higher minimum order then we start with that one instead of |
672bba3a | 3797 | * the smallest order which will fit the object. |
81819f0f | 3798 | */ |
d122019b | 3799 | static inline unsigned int calc_slab_order(unsigned int size, |
19af27af | 3800 | unsigned int min_objects, unsigned int max_order, |
9736d2a9 | 3801 | unsigned int fract_leftover) |
81819f0f | 3802 | { |
19af27af AD |
3803 | unsigned int min_order = slub_min_order; |
3804 | unsigned int order; | |
81819f0f | 3805 | |
9736d2a9 | 3806 | if (order_objects(min_order, size) > MAX_OBJS_PER_PAGE) |
210b5c06 | 3807 | return get_order(size * MAX_OBJS_PER_PAGE) - 1; |
39b26464 | 3808 | |
9736d2a9 | 3809 | for (order = max(min_order, (unsigned int)get_order(min_objects * size)); |
5e6d444e | 3810 | order <= max_order; order++) { |
81819f0f | 3811 | |
19af27af AD |
3812 | unsigned int slab_size = (unsigned int)PAGE_SIZE << order; |
3813 | unsigned int rem; | |
81819f0f | 3814 | |
9736d2a9 | 3815 | rem = slab_size % size; |
81819f0f | 3816 | |
5e6d444e | 3817 | if (rem <= slab_size / fract_leftover) |
81819f0f | 3818 | break; |
81819f0f | 3819 | } |
672bba3a | 3820 | |
81819f0f CL |
3821 | return order; |
3822 | } | |
3823 | ||
9736d2a9 | 3824 | static inline int calculate_order(unsigned int size) |
5e6d444e | 3825 | { |
19af27af AD |
3826 | unsigned int order; |
3827 | unsigned int min_objects; | |
3828 | unsigned int max_objects; | |
3286222f | 3829 | unsigned int nr_cpus; |
5e6d444e CL |
3830 | |
3831 | /* | |
3832 | * Attempt to find best configuration for a slab. This | |
3833 | * works by first attempting to generate a layout with | |
3834 | * the best configuration and backing off gradually. | |
3835 | * | |
422ff4d7 | 3836 | * First we increase the acceptable waste in a slab. Then |
5e6d444e CL |
3837 | * we reduce the minimum objects required in a slab. |
3838 | */ | |
3839 | min_objects = slub_min_objects; | |
3286222f VB |
3840 | if (!min_objects) { |
3841 | /* | |
3842 | * Some architectures will only update present cpus when | |
3843 | * onlining them, so don't trust the number if it's just 1. But | |
3844 | * we also don't want to use nr_cpu_ids always, as on some other | |
3845 | * architectures, there can be many possible cpus, but never | |
3846 | * onlined. Here we compromise between trying to avoid too high | |
3847 | * order on systems that appear larger than they are, and too | |
3848 | * low order on systems that appear smaller than they are. | |
3849 | */ | |
3850 | nr_cpus = num_present_cpus(); | |
3851 | if (nr_cpus <= 1) | |
3852 | nr_cpus = nr_cpu_ids; | |
3853 | min_objects = 4 * (fls(nr_cpus) + 1); | |
3854 | } | |
9736d2a9 | 3855 | max_objects = order_objects(slub_max_order, size); |
e8120ff1 ZY |
3856 | min_objects = min(min_objects, max_objects); |
3857 | ||
5e6d444e | 3858 | while (min_objects > 1) { |
19af27af AD |
3859 | unsigned int fraction; |
3860 | ||
c124f5b5 | 3861 | fraction = 16; |
5e6d444e | 3862 | while (fraction >= 4) { |
d122019b | 3863 | order = calc_slab_order(size, min_objects, |
9736d2a9 | 3864 | slub_max_order, fraction); |
5e6d444e CL |
3865 | if (order <= slub_max_order) |
3866 | return order; | |
3867 | fraction /= 2; | |
3868 | } | |
5086c389 | 3869 | min_objects--; |
5e6d444e CL |
3870 | } |
3871 | ||
3872 | /* | |
3873 | * We were unable to place multiple objects in a slab. Now | |
3874 | * lets see if we can place a single object there. | |
3875 | */ | |
d122019b | 3876 | order = calc_slab_order(size, 1, slub_max_order, 1); |
5e6d444e CL |
3877 | if (order <= slub_max_order) |
3878 | return order; | |
3879 | ||
3880 | /* | |
3881 | * Doh this slab cannot be placed using slub_max_order. | |
3882 | */ | |
d122019b | 3883 | order = calc_slab_order(size, 1, MAX_ORDER, 1); |
818cf590 | 3884 | if (order < MAX_ORDER) |
5e6d444e CL |
3885 | return order; |
3886 | return -ENOSYS; | |
3887 | } | |
3888 | ||
5595cffc | 3889 | static void |
4053497d | 3890 | init_kmem_cache_node(struct kmem_cache_node *n) |
81819f0f CL |
3891 | { |
3892 | n->nr_partial = 0; | |
81819f0f CL |
3893 | spin_lock_init(&n->list_lock); |
3894 | INIT_LIST_HEAD(&n->partial); | |
8ab1372f | 3895 | #ifdef CONFIG_SLUB_DEBUG |
0f389ec6 | 3896 | atomic_long_set(&n->nr_slabs, 0); |
02b71b70 | 3897 | atomic_long_set(&n->total_objects, 0); |
643b1138 | 3898 | INIT_LIST_HEAD(&n->full); |
8ab1372f | 3899 | #endif |
81819f0f CL |
3900 | } |
3901 | ||
55136592 | 3902 | static inline int alloc_kmem_cache_cpus(struct kmem_cache *s) |
4c93c355 | 3903 | { |
6c182dc0 | 3904 | BUILD_BUG_ON(PERCPU_DYNAMIC_EARLY_SIZE < |
95a05b42 | 3905 | KMALLOC_SHIFT_HIGH * sizeof(struct kmem_cache_cpu)); |
4c93c355 | 3906 | |
8a5ec0ba | 3907 | /* |
d4d84fef CM |
3908 | * Must align to double word boundary for the double cmpxchg |
3909 | * instructions to work; see __pcpu_double_call_return_bool(). | |
8a5ec0ba | 3910 | */ |
d4d84fef CM |
3911 | s->cpu_slab = __alloc_percpu(sizeof(struct kmem_cache_cpu), |
3912 | 2 * sizeof(void *)); | |
8a5ec0ba CL |
3913 | |
3914 | if (!s->cpu_slab) | |
3915 | return 0; | |
3916 | ||
3917 | init_kmem_cache_cpus(s); | |
4c93c355 | 3918 | |
8a5ec0ba | 3919 | return 1; |
4c93c355 | 3920 | } |
4c93c355 | 3921 | |
51df1142 CL |
3922 | static struct kmem_cache *kmem_cache_node; |
3923 | ||
81819f0f CL |
3924 | /* |
3925 | * No kmalloc_node yet so do it by hand. We know that this is the first | |
3926 | * slab on the node for this slabcache. There are no concurrent accesses | |
3927 | * possible. | |
3928 | * | |
721ae22a ZYW |
3929 | * Note that this function only works on the kmem_cache_node |
3930 | * when allocating for the kmem_cache_node. This is used for bootstrapping | |
4c93c355 | 3931 | * memory on a fresh node that has no slab structures yet. |
81819f0f | 3932 | */ |
55136592 | 3933 | static void early_kmem_cache_node_alloc(int node) |
81819f0f CL |
3934 | { |
3935 | struct page *page; | |
3936 | struct kmem_cache_node *n; | |
3937 | ||
51df1142 | 3938 | BUG_ON(kmem_cache_node->size < sizeof(struct kmem_cache_node)); |
81819f0f | 3939 | |
51df1142 | 3940 | page = new_slab(kmem_cache_node, GFP_NOWAIT, node); |
81819f0f CL |
3941 | |
3942 | BUG_ON(!page); | |
a2f92ee7 | 3943 | if (page_to_nid(page) != node) { |
f9f58285 FF |
3944 | pr_err("SLUB: Unable to allocate memory from node %d\n", node); |
3945 | pr_err("SLUB: Allocating a useless per node structure in order to be able to continue\n"); | |
a2f92ee7 CL |
3946 | } |
3947 | ||
81819f0f CL |
3948 | n = page->freelist; |
3949 | BUG_ON(!n); | |
8ab1372f | 3950 | #ifdef CONFIG_SLUB_DEBUG |
f7cb1933 | 3951 | init_object(kmem_cache_node, n, SLUB_RED_ACTIVE); |
51df1142 | 3952 | init_tracking(kmem_cache_node, n); |
8ab1372f | 3953 | #endif |
da844b78 | 3954 | n = kasan_slab_alloc(kmem_cache_node, n, GFP_KERNEL, false); |
12b22386 AK |
3955 | page->freelist = get_freepointer(kmem_cache_node, n); |
3956 | page->inuse = 1; | |
3957 | page->frozen = 0; | |
3958 | kmem_cache_node->node[node] = n; | |
4053497d | 3959 | init_kmem_cache_node(n); |
51df1142 | 3960 | inc_slabs_node(kmem_cache_node, node, page->objects); |
6446faa2 | 3961 | |
67b6c900 | 3962 | /* |
1e4dd946 SR |
3963 | * No locks need to be taken here as it has just been |
3964 | * initialized and there is no concurrent access. | |
67b6c900 | 3965 | */ |
1e4dd946 | 3966 | __add_partial(n, page, DEACTIVATE_TO_HEAD); |
81819f0f CL |
3967 | } |
3968 | ||
3969 | static void free_kmem_cache_nodes(struct kmem_cache *s) | |
3970 | { | |
3971 | int node; | |
fa45dc25 | 3972 | struct kmem_cache_node *n; |
81819f0f | 3973 | |
fa45dc25 | 3974 | for_each_kmem_cache_node(s, node, n) { |
81819f0f | 3975 | s->node[node] = NULL; |
ea37df54 | 3976 | kmem_cache_free(kmem_cache_node, n); |
81819f0f CL |
3977 | } |
3978 | } | |
3979 | ||
52b4b950 DS |
3980 | void __kmem_cache_release(struct kmem_cache *s) |
3981 | { | |
210e7a43 | 3982 | cache_random_seq_destroy(s); |
52b4b950 DS |
3983 | free_percpu(s->cpu_slab); |
3984 | free_kmem_cache_nodes(s); | |
3985 | } | |
3986 | ||
55136592 | 3987 | static int init_kmem_cache_nodes(struct kmem_cache *s) |
81819f0f CL |
3988 | { |
3989 | int node; | |
81819f0f | 3990 | |
7e1fa93d | 3991 | for_each_node_mask(node, slab_nodes) { |
81819f0f CL |
3992 | struct kmem_cache_node *n; |
3993 | ||
73367bd8 | 3994 | if (slab_state == DOWN) { |
55136592 | 3995 | early_kmem_cache_node_alloc(node); |
73367bd8 AD |
3996 | continue; |
3997 | } | |
51df1142 | 3998 | n = kmem_cache_alloc_node(kmem_cache_node, |
55136592 | 3999 | GFP_KERNEL, node); |
81819f0f | 4000 | |
73367bd8 AD |
4001 | if (!n) { |
4002 | free_kmem_cache_nodes(s); | |
4003 | return 0; | |
81819f0f | 4004 | } |
73367bd8 | 4005 | |
4053497d | 4006 | init_kmem_cache_node(n); |
ea37df54 | 4007 | s->node[node] = n; |
81819f0f CL |
4008 | } |
4009 | return 1; | |
4010 | } | |
81819f0f | 4011 | |
c0bdb232 | 4012 | static void set_min_partial(struct kmem_cache *s, unsigned long min) |
3b89d7d8 DR |
4013 | { |
4014 | if (min < MIN_PARTIAL) | |
4015 | min = MIN_PARTIAL; | |
4016 | else if (min > MAX_PARTIAL) | |
4017 | min = MAX_PARTIAL; | |
4018 | s->min_partial = min; | |
4019 | } | |
4020 | ||
e6d0e1dc WY |
4021 | static void set_cpu_partial(struct kmem_cache *s) |
4022 | { | |
4023 | #ifdef CONFIG_SLUB_CPU_PARTIAL | |
b47291ef VB |
4024 | unsigned int nr_objects; |
4025 | ||
e6d0e1dc WY |
4026 | /* |
4027 | * cpu_partial determined the maximum number of objects kept in the | |
4028 | * per cpu partial lists of a processor. | |
4029 | * | |
4030 | * Per cpu partial lists mainly contain slabs that just have one | |
4031 | * object freed. If they are used for allocation then they can be | |
4032 | * filled up again with minimal effort. The slab will never hit the | |
4033 | * per node partial lists and therefore no locking will be required. | |
4034 | * | |
b47291ef VB |
4035 | * For backwards compatibility reasons, this is determined as number |
4036 | * of objects, even though we now limit maximum number of pages, see | |
4037 | * slub_set_cpu_partial() | |
e6d0e1dc WY |
4038 | */ |
4039 | if (!kmem_cache_has_cpu_partial(s)) | |
b47291ef | 4040 | nr_objects = 0; |
e6d0e1dc | 4041 | else if (s->size >= PAGE_SIZE) |
b47291ef | 4042 | nr_objects = 6; |
e6d0e1dc | 4043 | else if (s->size >= 1024) |
23e98ad1 | 4044 | nr_objects = 24; |
e6d0e1dc | 4045 | else if (s->size >= 256) |
23e98ad1 | 4046 | nr_objects = 52; |
e6d0e1dc | 4047 | else |
23e98ad1 | 4048 | nr_objects = 120; |
b47291ef VB |
4049 | |
4050 | slub_set_cpu_partial(s, nr_objects); | |
e6d0e1dc WY |
4051 | #endif |
4052 | } | |
4053 | ||
81819f0f CL |
4054 | /* |
4055 | * calculate_sizes() determines the order and the distribution of data within | |
4056 | * a slab object. | |
4057 | */ | |
06b285dc | 4058 | static int calculate_sizes(struct kmem_cache *s, int forced_order) |
81819f0f | 4059 | { |
d50112ed | 4060 | slab_flags_t flags = s->flags; |
be4a7988 | 4061 | unsigned int size = s->object_size; |
19af27af | 4062 | unsigned int order; |
81819f0f | 4063 | |
d8b42bf5 CL |
4064 | /* |
4065 | * Round up object size to the next word boundary. We can only | |
4066 | * place the free pointer at word boundaries and this determines | |
4067 | * the possible location of the free pointer. | |
4068 | */ | |
4069 | size = ALIGN(size, sizeof(void *)); | |
4070 | ||
4071 | #ifdef CONFIG_SLUB_DEBUG | |
81819f0f CL |
4072 | /* |
4073 | * Determine if we can poison the object itself. If the user of | |
4074 | * the slab may touch the object after free or before allocation | |
4075 | * then we should never poison the object itself. | |
4076 | */ | |
5f0d5a3a | 4077 | if ((flags & SLAB_POISON) && !(flags & SLAB_TYPESAFE_BY_RCU) && |
c59def9f | 4078 | !s->ctor) |
81819f0f CL |
4079 | s->flags |= __OBJECT_POISON; |
4080 | else | |
4081 | s->flags &= ~__OBJECT_POISON; | |
4082 | ||
81819f0f CL |
4083 | |
4084 | /* | |
672bba3a | 4085 | * If we are Redzoning then check if there is some space between the |
81819f0f | 4086 | * end of the object and the free pointer. If not then add an |
672bba3a | 4087 | * additional word to have some bytes to store Redzone information. |
81819f0f | 4088 | */ |
3b0efdfa | 4089 | if ((flags & SLAB_RED_ZONE) && size == s->object_size) |
81819f0f | 4090 | size += sizeof(void *); |
41ecc55b | 4091 | #endif |
81819f0f CL |
4092 | |
4093 | /* | |
672bba3a | 4094 | * With that we have determined the number of bytes in actual use |
e41a49fa | 4095 | * by the object and redzoning. |
81819f0f CL |
4096 | */ |
4097 | s->inuse = size; | |
4098 | ||
74c1d3e0 KC |
4099 | if ((flags & (SLAB_TYPESAFE_BY_RCU | SLAB_POISON)) || |
4100 | ((flags & SLAB_RED_ZONE) && s->object_size < sizeof(void *)) || | |
4101 | s->ctor) { | |
81819f0f CL |
4102 | /* |
4103 | * Relocate free pointer after the object if it is not | |
4104 | * permitted to overwrite the first word of the object on | |
4105 | * kmem_cache_free. | |
4106 | * | |
4107 | * This is the case if we do RCU, have a constructor or | |
74c1d3e0 KC |
4108 | * destructor, are poisoning the objects, or are |
4109 | * redzoning an object smaller than sizeof(void *). | |
cbfc35a4 WL |
4110 | * |
4111 | * The assumption that s->offset >= s->inuse means free | |
4112 | * pointer is outside of the object is used in the | |
4113 | * freeptr_outside_object() function. If that is no | |
4114 | * longer true, the function needs to be modified. | |
81819f0f CL |
4115 | */ |
4116 | s->offset = size; | |
4117 | size += sizeof(void *); | |
e41a49fa | 4118 | } else { |
3202fa62 KC |
4119 | /* |
4120 | * Store freelist pointer near middle of object to keep | |
4121 | * it away from the edges of the object to avoid small | |
4122 | * sized over/underflows from neighboring allocations. | |
4123 | */ | |
e41a49fa | 4124 | s->offset = ALIGN_DOWN(s->object_size / 2, sizeof(void *)); |
81819f0f CL |
4125 | } |
4126 | ||
c12b3c62 | 4127 | #ifdef CONFIG_SLUB_DEBUG |
81819f0f CL |
4128 | if (flags & SLAB_STORE_USER) |
4129 | /* | |
4130 | * Need to store information about allocs and frees after | |
4131 | * the object. | |
4132 | */ | |
4133 | size += 2 * sizeof(struct track); | |
80a9201a | 4134 | #endif |
81819f0f | 4135 | |
80a9201a AP |
4136 | kasan_cache_create(s, &size, &s->flags); |
4137 | #ifdef CONFIG_SLUB_DEBUG | |
d86bd1be | 4138 | if (flags & SLAB_RED_ZONE) { |
81819f0f CL |
4139 | /* |
4140 | * Add some empty padding so that we can catch | |
4141 | * overwrites from earlier objects rather than let | |
4142 | * tracking information or the free pointer be | |
0211a9c8 | 4143 | * corrupted if a user writes before the start |
81819f0f CL |
4144 | * of the object. |
4145 | */ | |
4146 | size += sizeof(void *); | |
d86bd1be JK |
4147 | |
4148 | s->red_left_pad = sizeof(void *); | |
4149 | s->red_left_pad = ALIGN(s->red_left_pad, s->align); | |
4150 | size += s->red_left_pad; | |
4151 | } | |
41ecc55b | 4152 | #endif |
672bba3a | 4153 | |
81819f0f CL |
4154 | /* |
4155 | * SLUB stores one object immediately after another beginning from | |
4156 | * offset 0. In order to align the objects we have to simply size | |
4157 | * each object to conform to the alignment. | |
4158 | */ | |
45906855 | 4159 | size = ALIGN(size, s->align); |
81819f0f | 4160 | s->size = size; |
4138fdfc | 4161 | s->reciprocal_size = reciprocal_value(size); |
06b285dc CL |
4162 | if (forced_order >= 0) |
4163 | order = forced_order; | |
4164 | else | |
9736d2a9 | 4165 | order = calculate_order(size); |
81819f0f | 4166 | |
19af27af | 4167 | if ((int)order < 0) |
81819f0f CL |
4168 | return 0; |
4169 | ||
b7a49f0d | 4170 | s->allocflags = 0; |
834f3d11 | 4171 | if (order) |
b7a49f0d CL |
4172 | s->allocflags |= __GFP_COMP; |
4173 | ||
4174 | if (s->flags & SLAB_CACHE_DMA) | |
2c59dd65 | 4175 | s->allocflags |= GFP_DMA; |
b7a49f0d | 4176 | |
6d6ea1e9 NB |
4177 | if (s->flags & SLAB_CACHE_DMA32) |
4178 | s->allocflags |= GFP_DMA32; | |
4179 | ||
b7a49f0d CL |
4180 | if (s->flags & SLAB_RECLAIM_ACCOUNT) |
4181 | s->allocflags |= __GFP_RECLAIMABLE; | |
4182 | ||
81819f0f CL |
4183 | /* |
4184 | * Determine the number of objects per slab | |
4185 | */ | |
9736d2a9 MW |
4186 | s->oo = oo_make(order, size); |
4187 | s->min = oo_make(get_order(size), size); | |
205ab99d CL |
4188 | if (oo_objects(s->oo) > oo_objects(s->max)) |
4189 | s->max = s->oo; | |
81819f0f | 4190 | |
834f3d11 | 4191 | return !!oo_objects(s->oo); |
81819f0f CL |
4192 | } |
4193 | ||
d50112ed | 4194 | static int kmem_cache_open(struct kmem_cache *s, slab_flags_t flags) |
81819f0f | 4195 | { |
37540008 | 4196 | s->flags = kmem_cache_flags(s->size, flags, s->name); |
2482ddec KC |
4197 | #ifdef CONFIG_SLAB_FREELIST_HARDENED |
4198 | s->random = get_random_long(); | |
4199 | #endif | |
81819f0f | 4200 | |
06b285dc | 4201 | if (!calculate_sizes(s, -1)) |
81819f0f | 4202 | goto error; |
3de47213 DR |
4203 | if (disable_higher_order_debug) { |
4204 | /* | |
4205 | * Disable debugging flags that store metadata if the min slab | |
4206 | * order increased. | |
4207 | */ | |
3b0efdfa | 4208 | if (get_order(s->size) > get_order(s->object_size)) { |
3de47213 DR |
4209 | s->flags &= ~DEBUG_METADATA_FLAGS; |
4210 | s->offset = 0; | |
4211 | if (!calculate_sizes(s, -1)) | |
4212 | goto error; | |
4213 | } | |
4214 | } | |
81819f0f | 4215 | |
2565409f HC |
4216 | #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \ |
4217 | defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE) | |
149daaf3 | 4218 | if (system_has_cmpxchg_double() && (s->flags & SLAB_NO_CMPXCHG) == 0) |
b789ef51 CL |
4219 | /* Enable fast mode */ |
4220 | s->flags |= __CMPXCHG_DOUBLE; | |
4221 | #endif | |
4222 | ||
3b89d7d8 DR |
4223 | /* |
4224 | * The larger the object size is, the more pages we want on the partial | |
4225 | * list to avoid pounding the page allocator excessively. | |
4226 | */ | |
49e22585 CL |
4227 | set_min_partial(s, ilog2(s->size) / 2); |
4228 | ||
e6d0e1dc | 4229 | set_cpu_partial(s); |
49e22585 | 4230 | |
81819f0f | 4231 | #ifdef CONFIG_NUMA |
e2cb96b7 | 4232 | s->remote_node_defrag_ratio = 1000; |
81819f0f | 4233 | #endif |
210e7a43 TG |
4234 | |
4235 | /* Initialize the pre-computed randomized freelist if slab is up */ | |
4236 | if (slab_state >= UP) { | |
4237 | if (init_cache_random_seq(s)) | |
4238 | goto error; | |
4239 | } | |
4240 | ||
55136592 | 4241 | if (!init_kmem_cache_nodes(s)) |
dfb4f096 | 4242 | goto error; |
81819f0f | 4243 | |
55136592 | 4244 | if (alloc_kmem_cache_cpus(s)) |
278b1bb1 | 4245 | return 0; |
ff12059e | 4246 | |
81819f0f | 4247 | error: |
9037c576 | 4248 | __kmem_cache_release(s); |
278b1bb1 | 4249 | return -EINVAL; |
81819f0f | 4250 | } |
81819f0f | 4251 | |
33b12c38 | 4252 | static void list_slab_objects(struct kmem_cache *s, struct page *page, |
55860d96 | 4253 | const char *text) |
33b12c38 CL |
4254 | { |
4255 | #ifdef CONFIG_SLUB_DEBUG | |
4256 | void *addr = page_address(page); | |
a2b4ae8b | 4257 | unsigned long flags; |
55860d96 | 4258 | unsigned long *map; |
33b12c38 | 4259 | void *p; |
aa456c7a | 4260 | |
945cf2b6 | 4261 | slab_err(s, page, text, s->name); |
a2b4ae8b | 4262 | slab_lock(page, &flags); |
33b12c38 | 4263 | |
90e9f6a6 | 4264 | map = get_map(s, page); |
33b12c38 CL |
4265 | for_each_object(p, s, addr, page->objects) { |
4266 | ||
4138fdfc | 4267 | if (!test_bit(__obj_to_index(s, addr, p), map)) { |
96b94abc | 4268 | pr_err("Object 0x%p @offset=%tu\n", p, p - addr); |
33b12c38 CL |
4269 | print_tracking(s, p); |
4270 | } | |
4271 | } | |
55860d96 | 4272 | put_map(map); |
a2b4ae8b | 4273 | slab_unlock(page, &flags); |
33b12c38 CL |
4274 | #endif |
4275 | } | |
4276 | ||
81819f0f | 4277 | /* |
599870b1 | 4278 | * Attempt to free all partial slabs on a node. |
52b4b950 DS |
4279 | * This is called from __kmem_cache_shutdown(). We must take list_lock |
4280 | * because sysfs file might still access partial list after the shutdowning. | |
81819f0f | 4281 | */ |
599870b1 | 4282 | static void free_partial(struct kmem_cache *s, struct kmem_cache_node *n) |
81819f0f | 4283 | { |
60398923 | 4284 | LIST_HEAD(discard); |
81819f0f CL |
4285 | struct page *page, *h; |
4286 | ||
52b4b950 DS |
4287 | BUG_ON(irqs_disabled()); |
4288 | spin_lock_irq(&n->list_lock); | |
916ac052 | 4289 | list_for_each_entry_safe(page, h, &n->partial, slab_list) { |
81819f0f | 4290 | if (!page->inuse) { |
52b4b950 | 4291 | remove_partial(n, page); |
916ac052 | 4292 | list_add(&page->slab_list, &discard); |
33b12c38 CL |
4293 | } else { |
4294 | list_slab_objects(s, page, | |
55860d96 | 4295 | "Objects remaining in %s on __kmem_cache_shutdown()"); |
599870b1 | 4296 | } |
33b12c38 | 4297 | } |
52b4b950 | 4298 | spin_unlock_irq(&n->list_lock); |
60398923 | 4299 | |
916ac052 | 4300 | list_for_each_entry_safe(page, h, &discard, slab_list) |
60398923 | 4301 | discard_slab(s, page); |
81819f0f CL |
4302 | } |
4303 | ||
f9e13c0a SB |
4304 | bool __kmem_cache_empty(struct kmem_cache *s) |
4305 | { | |
4306 | int node; | |
4307 | struct kmem_cache_node *n; | |
4308 | ||
4309 | for_each_kmem_cache_node(s, node, n) | |
4310 | if (n->nr_partial || slabs_node(s, node)) | |
4311 | return false; | |
4312 | return true; | |
4313 | } | |
4314 | ||
81819f0f | 4315 | /* |
672bba3a | 4316 | * Release all resources used by a slab cache. |
81819f0f | 4317 | */ |
52b4b950 | 4318 | int __kmem_cache_shutdown(struct kmem_cache *s) |
81819f0f CL |
4319 | { |
4320 | int node; | |
fa45dc25 | 4321 | struct kmem_cache_node *n; |
81819f0f | 4322 | |
5a836bf6 | 4323 | flush_all_cpus_locked(s); |
81819f0f | 4324 | /* Attempt to free all objects */ |
fa45dc25 | 4325 | for_each_kmem_cache_node(s, node, n) { |
599870b1 CL |
4326 | free_partial(s, n); |
4327 | if (n->nr_partial || slabs_node(s, node)) | |
81819f0f CL |
4328 | return 1; |
4329 | } | |
81819f0f CL |
4330 | return 0; |
4331 | } | |
4332 | ||
5bb1bb35 | 4333 | #ifdef CONFIG_PRINTK |
7213230a | 4334 | void kmem_obj_info(struct kmem_obj_info *kpp, void *object, struct slab *slab) |
8e7f37f2 PM |
4335 | { |
4336 | void *base; | |
4337 | int __maybe_unused i; | |
4338 | unsigned int objnr; | |
4339 | void *objp; | |
4340 | void *objp0; | |
7213230a | 4341 | struct kmem_cache *s = slab->slab_cache; |
8e7f37f2 PM |
4342 | struct track __maybe_unused *trackp; |
4343 | ||
4344 | kpp->kp_ptr = object; | |
7213230a | 4345 | kpp->kp_slab = slab; |
8e7f37f2 | 4346 | kpp->kp_slab_cache = s; |
7213230a | 4347 | base = slab_address(slab); |
8e7f37f2 PM |
4348 | objp0 = kasan_reset_tag(object); |
4349 | #ifdef CONFIG_SLUB_DEBUG | |
4350 | objp = restore_red_left(s, objp0); | |
4351 | #else | |
4352 | objp = objp0; | |
4353 | #endif | |
7213230a | 4354 | objnr = obj_to_index(s, slab_page(slab), objp); |
8e7f37f2 PM |
4355 | kpp->kp_data_offset = (unsigned long)((char *)objp0 - (char *)objp); |
4356 | objp = base + s->size * objnr; | |
4357 | kpp->kp_objp = objp; | |
7213230a MWO |
4358 | if (WARN_ON_ONCE(objp < base || objp >= base + slab->objects * s->size |
4359 | || (objp - base) % s->size) || | |
8e7f37f2 PM |
4360 | !(s->flags & SLAB_STORE_USER)) |
4361 | return; | |
4362 | #ifdef CONFIG_SLUB_DEBUG | |
0cbc124b | 4363 | objp = fixup_red_left(s, objp); |
8e7f37f2 PM |
4364 | trackp = get_track(s, objp, TRACK_ALLOC); |
4365 | kpp->kp_ret = (void *)trackp->addr; | |
ae14c63a LT |
4366 | #ifdef CONFIG_STACKTRACE |
4367 | for (i = 0; i < KS_ADDRS_COUNT && i < TRACK_ADDRS_COUNT; i++) { | |
4368 | kpp->kp_stack[i] = (void *)trackp->addrs[i]; | |
4369 | if (!kpp->kp_stack[i]) | |
4370 | break; | |
4371 | } | |
78869146 | 4372 | |
ae14c63a LT |
4373 | trackp = get_track(s, objp, TRACK_FREE); |
4374 | for (i = 0; i < KS_ADDRS_COUNT && i < TRACK_ADDRS_COUNT; i++) { | |
4375 | kpp->kp_free_stack[i] = (void *)trackp->addrs[i]; | |
4376 | if (!kpp->kp_free_stack[i]) | |
4377 | break; | |
e548eaa1 | 4378 | } |
8e7f37f2 PM |
4379 | #endif |
4380 | #endif | |
4381 | } | |
5bb1bb35 | 4382 | #endif |
8e7f37f2 | 4383 | |
81819f0f CL |
4384 | /******************************************************************** |
4385 | * Kmalloc subsystem | |
4386 | *******************************************************************/ | |
4387 | ||
81819f0f CL |
4388 | static int __init setup_slub_min_order(char *str) |
4389 | { | |
19af27af | 4390 | get_option(&str, (int *)&slub_min_order); |
81819f0f CL |
4391 | |
4392 | return 1; | |
4393 | } | |
4394 | ||
4395 | __setup("slub_min_order=", setup_slub_min_order); | |
4396 | ||
4397 | static int __init setup_slub_max_order(char *str) | |
4398 | { | |
19af27af AD |
4399 | get_option(&str, (int *)&slub_max_order); |
4400 | slub_max_order = min(slub_max_order, (unsigned int)MAX_ORDER - 1); | |
81819f0f CL |
4401 | |
4402 | return 1; | |
4403 | } | |
4404 | ||
4405 | __setup("slub_max_order=", setup_slub_max_order); | |
4406 | ||
4407 | static int __init setup_slub_min_objects(char *str) | |
4408 | { | |
19af27af | 4409 | get_option(&str, (int *)&slub_min_objects); |
81819f0f CL |
4410 | |
4411 | return 1; | |
4412 | } | |
4413 | ||
4414 | __setup("slub_min_objects=", setup_slub_min_objects); | |
4415 | ||
81819f0f CL |
4416 | void *__kmalloc(size_t size, gfp_t flags) |
4417 | { | |
aadb4bc4 | 4418 | struct kmem_cache *s; |
5b882be4 | 4419 | void *ret; |
81819f0f | 4420 | |
95a05b42 | 4421 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) |
eada35ef | 4422 | return kmalloc_large(size, flags); |
aadb4bc4 | 4423 | |
2c59dd65 | 4424 | s = kmalloc_slab(size, flags); |
aadb4bc4 CL |
4425 | |
4426 | if (unlikely(ZERO_OR_NULL_PTR(s))) | |
6cb8f913 CL |
4427 | return s; |
4428 | ||
b89fb5ef | 4429 | ret = slab_alloc(s, flags, _RET_IP_, size); |
5b882be4 | 4430 | |
ca2b84cb | 4431 | trace_kmalloc(_RET_IP_, ret, size, s->size, flags); |
5b882be4 | 4432 | |
0116523c | 4433 | ret = kasan_kmalloc(s, ret, size, flags); |
0316bec2 | 4434 | |
5b882be4 | 4435 | return ret; |
81819f0f CL |
4436 | } |
4437 | EXPORT_SYMBOL(__kmalloc); | |
4438 | ||
5d1f57e4 | 4439 | #ifdef CONFIG_NUMA |
f619cfe1 CL |
4440 | static void *kmalloc_large_node(size_t size, gfp_t flags, int node) |
4441 | { | |
b1eeab67 | 4442 | struct page *page; |
e4f7c0b4 | 4443 | void *ptr = NULL; |
6a486c0a | 4444 | unsigned int order = get_order(size); |
f619cfe1 | 4445 | |
75f296d9 | 4446 | flags |= __GFP_COMP; |
6a486c0a VB |
4447 | page = alloc_pages_node(node, flags, order); |
4448 | if (page) { | |
e4f7c0b4 | 4449 | ptr = page_address(page); |
96403bfe MS |
4450 | mod_lruvec_page_state(page, NR_SLAB_UNRECLAIMABLE_B, |
4451 | PAGE_SIZE << order); | |
6a486c0a | 4452 | } |
e4f7c0b4 | 4453 | |
0116523c | 4454 | return kmalloc_large_node_hook(ptr, size, flags); |
f619cfe1 CL |
4455 | } |
4456 | ||
81819f0f CL |
4457 | void *__kmalloc_node(size_t size, gfp_t flags, int node) |
4458 | { | |
aadb4bc4 | 4459 | struct kmem_cache *s; |
5b882be4 | 4460 | void *ret; |
81819f0f | 4461 | |
95a05b42 | 4462 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) { |
5b882be4 EGM |
4463 | ret = kmalloc_large_node(size, flags, node); |
4464 | ||
ca2b84cb EGM |
4465 | trace_kmalloc_node(_RET_IP_, ret, |
4466 | size, PAGE_SIZE << get_order(size), | |
4467 | flags, node); | |
5b882be4 EGM |
4468 | |
4469 | return ret; | |
4470 | } | |
aadb4bc4 | 4471 | |
2c59dd65 | 4472 | s = kmalloc_slab(size, flags); |
aadb4bc4 CL |
4473 | |
4474 | if (unlikely(ZERO_OR_NULL_PTR(s))) | |
6cb8f913 CL |
4475 | return s; |
4476 | ||
b89fb5ef | 4477 | ret = slab_alloc_node(s, flags, node, _RET_IP_, size); |
5b882be4 | 4478 | |
ca2b84cb | 4479 | trace_kmalloc_node(_RET_IP_, ret, size, s->size, flags, node); |
5b882be4 | 4480 | |
0116523c | 4481 | ret = kasan_kmalloc(s, ret, size, flags); |
0316bec2 | 4482 | |
5b882be4 | 4483 | return ret; |
81819f0f CL |
4484 | } |
4485 | EXPORT_SYMBOL(__kmalloc_node); | |
6dfd1b65 | 4486 | #endif /* CONFIG_NUMA */ |
81819f0f | 4487 | |
ed18adc1 KC |
4488 | #ifdef CONFIG_HARDENED_USERCOPY |
4489 | /* | |
afcc90f8 KC |
4490 | * Rejects incorrectly sized objects and objects that are to be copied |
4491 | * to/from userspace but do not fall entirely within the containing slab | |
4492 | * cache's usercopy region. | |
ed18adc1 KC |
4493 | * |
4494 | * Returns NULL if check passes, otherwise const char * to name of cache | |
4495 | * to indicate an error. | |
4496 | */ | |
0b3eb091 MWO |
4497 | void __check_heap_object(const void *ptr, unsigned long n, |
4498 | const struct slab *slab, bool to_user) | |
ed18adc1 KC |
4499 | { |
4500 | struct kmem_cache *s; | |
44065b2e | 4501 | unsigned int offset; |
b89fb5ef | 4502 | bool is_kfence = is_kfence_address(ptr); |
ed18adc1 | 4503 | |
96fedce2 AK |
4504 | ptr = kasan_reset_tag(ptr); |
4505 | ||
ed18adc1 | 4506 | /* Find object and usable object size. */ |
0b3eb091 | 4507 | s = slab->slab_cache; |
ed18adc1 KC |
4508 | |
4509 | /* Reject impossible pointers. */ | |
0b3eb091 | 4510 | if (ptr < slab_address(slab)) |
f4e6e289 KC |
4511 | usercopy_abort("SLUB object not in SLUB page?!", NULL, |
4512 | to_user, 0, n); | |
ed18adc1 KC |
4513 | |
4514 | /* Find offset within object. */ | |
b89fb5ef AP |
4515 | if (is_kfence) |
4516 | offset = ptr - kfence_object_start(ptr); | |
4517 | else | |
0b3eb091 | 4518 | offset = (ptr - slab_address(slab)) % s->size; |
ed18adc1 KC |
4519 | |
4520 | /* Adjust for redzone and reject if within the redzone. */ | |
b89fb5ef | 4521 | if (!is_kfence && kmem_cache_debug_flags(s, SLAB_RED_ZONE)) { |
ed18adc1 | 4522 | if (offset < s->red_left_pad) |
f4e6e289 KC |
4523 | usercopy_abort("SLUB object in left red zone", |
4524 | s->name, to_user, offset, n); | |
ed18adc1 KC |
4525 | offset -= s->red_left_pad; |
4526 | } | |
4527 | ||
afcc90f8 KC |
4528 | /* Allow address range falling entirely within usercopy region. */ |
4529 | if (offset >= s->useroffset && | |
4530 | offset - s->useroffset <= s->usersize && | |
4531 | n <= s->useroffset - offset + s->usersize) | |
f4e6e289 | 4532 | return; |
ed18adc1 | 4533 | |
f4e6e289 | 4534 | usercopy_abort("SLUB object", s->name, to_user, offset, n); |
ed18adc1 KC |
4535 | } |
4536 | #endif /* CONFIG_HARDENED_USERCOPY */ | |
4537 | ||
10d1f8cb | 4538 | size_t __ksize(const void *object) |
81819f0f | 4539 | { |
0c24811b | 4540 | struct folio *folio; |
81819f0f | 4541 | |
ef8b4520 | 4542 | if (unlikely(object == ZERO_SIZE_PTR)) |
272c1d21 CL |
4543 | return 0; |
4544 | ||
0c24811b | 4545 | folio = virt_to_folio(object); |
294a80a8 | 4546 | |
0c24811b MWO |
4547 | if (unlikely(!folio_test_slab(folio))) |
4548 | return folio_size(folio); | |
81819f0f | 4549 | |
0c24811b | 4550 | return slab_ksize(folio_slab(folio)->slab_cache); |
81819f0f | 4551 | } |
10d1f8cb | 4552 | EXPORT_SYMBOL(__ksize); |
81819f0f CL |
4553 | |
4554 | void kfree(const void *x) | |
4555 | { | |
d835eef4 MWO |
4556 | struct folio *folio; |
4557 | struct slab *slab; | |
5bb983b0 | 4558 | void *object = (void *)x; |
81819f0f | 4559 | |
2121db74 PE |
4560 | trace_kfree(_RET_IP_, x); |
4561 | ||
2408c550 | 4562 | if (unlikely(ZERO_OR_NULL_PTR(x))) |
81819f0f CL |
4563 | return; |
4564 | ||
d835eef4 MWO |
4565 | folio = virt_to_folio(x); |
4566 | if (unlikely(!folio_test_slab(folio))) { | |
4567 | free_large_kmalloc(folio, object); | |
aadb4bc4 CL |
4568 | return; |
4569 | } | |
d835eef4 MWO |
4570 | slab = folio_slab(folio); |
4571 | slab_free(slab->slab_cache, slab_page(slab), object, NULL, 1, _RET_IP_); | |
81819f0f CL |
4572 | } |
4573 | EXPORT_SYMBOL(kfree); | |
4574 | ||
832f37f5 VD |
4575 | #define SHRINK_PROMOTE_MAX 32 |
4576 | ||
2086d26a | 4577 | /* |
832f37f5 VD |
4578 | * kmem_cache_shrink discards empty slabs and promotes the slabs filled |
4579 | * up most to the head of the partial lists. New allocations will then | |
4580 | * fill those up and thus they can be removed from the partial lists. | |
672bba3a CL |
4581 | * |
4582 | * The slabs with the least items are placed last. This results in them | |
4583 | * being allocated from last increasing the chance that the last objects | |
4584 | * are freed in them. | |
2086d26a | 4585 | */ |
5a836bf6 | 4586 | static int __kmem_cache_do_shrink(struct kmem_cache *s) |
2086d26a CL |
4587 | { |
4588 | int node; | |
4589 | int i; | |
4590 | struct kmem_cache_node *n; | |
4591 | struct page *page; | |
4592 | struct page *t; | |
832f37f5 VD |
4593 | struct list_head discard; |
4594 | struct list_head promote[SHRINK_PROMOTE_MAX]; | |
2086d26a | 4595 | unsigned long flags; |
ce3712d7 | 4596 | int ret = 0; |
2086d26a | 4597 | |
fa45dc25 | 4598 | for_each_kmem_cache_node(s, node, n) { |
832f37f5 VD |
4599 | INIT_LIST_HEAD(&discard); |
4600 | for (i = 0; i < SHRINK_PROMOTE_MAX; i++) | |
4601 | INIT_LIST_HEAD(promote + i); | |
2086d26a CL |
4602 | |
4603 | spin_lock_irqsave(&n->list_lock, flags); | |
4604 | ||
4605 | /* | |
832f37f5 | 4606 | * Build lists of slabs to discard or promote. |
2086d26a | 4607 | * |
672bba3a CL |
4608 | * Note that concurrent frees may occur while we hold the |
4609 | * list_lock. page->inuse here is the upper limit. | |
2086d26a | 4610 | */ |
916ac052 | 4611 | list_for_each_entry_safe(page, t, &n->partial, slab_list) { |
832f37f5 VD |
4612 | int free = page->objects - page->inuse; |
4613 | ||
4614 | /* Do not reread page->inuse */ | |
4615 | barrier(); | |
4616 | ||
4617 | /* We do not keep full slabs on the list */ | |
4618 | BUG_ON(free <= 0); | |
4619 | ||
4620 | if (free == page->objects) { | |
916ac052 | 4621 | list_move(&page->slab_list, &discard); |
69cb8e6b | 4622 | n->nr_partial--; |
832f37f5 | 4623 | } else if (free <= SHRINK_PROMOTE_MAX) |
916ac052 | 4624 | list_move(&page->slab_list, promote + free - 1); |
2086d26a CL |
4625 | } |
4626 | ||
2086d26a | 4627 | /* |
832f37f5 VD |
4628 | * Promote the slabs filled up most to the head of the |
4629 | * partial list. | |
2086d26a | 4630 | */ |
832f37f5 VD |
4631 | for (i = SHRINK_PROMOTE_MAX - 1; i >= 0; i--) |
4632 | list_splice(promote + i, &n->partial); | |
2086d26a | 4633 | |
2086d26a | 4634 | spin_unlock_irqrestore(&n->list_lock, flags); |
69cb8e6b CL |
4635 | |
4636 | /* Release empty slabs */ | |
916ac052 | 4637 | list_for_each_entry_safe(page, t, &discard, slab_list) |
69cb8e6b | 4638 | discard_slab(s, page); |
ce3712d7 VD |
4639 | |
4640 | if (slabs_node(s, node)) | |
4641 | ret = 1; | |
2086d26a CL |
4642 | } |
4643 | ||
ce3712d7 | 4644 | return ret; |
2086d26a | 4645 | } |
2086d26a | 4646 | |
5a836bf6 SAS |
4647 | int __kmem_cache_shrink(struct kmem_cache *s) |
4648 | { | |
4649 | flush_all(s); | |
4650 | return __kmem_cache_do_shrink(s); | |
4651 | } | |
4652 | ||
b9049e23 YG |
4653 | static int slab_mem_going_offline_callback(void *arg) |
4654 | { | |
4655 | struct kmem_cache *s; | |
4656 | ||
18004c5d | 4657 | mutex_lock(&slab_mutex); |
5a836bf6 SAS |
4658 | list_for_each_entry(s, &slab_caches, list) { |
4659 | flush_all_cpus_locked(s); | |
4660 | __kmem_cache_do_shrink(s); | |
4661 | } | |
18004c5d | 4662 | mutex_unlock(&slab_mutex); |
b9049e23 YG |
4663 | |
4664 | return 0; | |
4665 | } | |
4666 | ||
4667 | static void slab_mem_offline_callback(void *arg) | |
4668 | { | |
b9049e23 YG |
4669 | struct memory_notify *marg = arg; |
4670 | int offline_node; | |
4671 | ||
b9d5ab25 | 4672 | offline_node = marg->status_change_nid_normal; |
b9049e23 YG |
4673 | |
4674 | /* | |
4675 | * If the node still has available memory. we need kmem_cache_node | |
4676 | * for it yet. | |
4677 | */ | |
4678 | if (offline_node < 0) | |
4679 | return; | |
4680 | ||
18004c5d | 4681 | mutex_lock(&slab_mutex); |
7e1fa93d | 4682 | node_clear(offline_node, slab_nodes); |
666716fd VB |
4683 | /* |
4684 | * We no longer free kmem_cache_node structures here, as it would be | |
4685 | * racy with all get_node() users, and infeasible to protect them with | |
4686 | * slab_mutex. | |
4687 | */ | |
18004c5d | 4688 | mutex_unlock(&slab_mutex); |
b9049e23 YG |
4689 | } |
4690 | ||
4691 | static int slab_mem_going_online_callback(void *arg) | |
4692 | { | |
4693 | struct kmem_cache_node *n; | |
4694 | struct kmem_cache *s; | |
4695 | struct memory_notify *marg = arg; | |
b9d5ab25 | 4696 | int nid = marg->status_change_nid_normal; |
b9049e23 YG |
4697 | int ret = 0; |
4698 | ||
4699 | /* | |
4700 | * If the node's memory is already available, then kmem_cache_node is | |
4701 | * already created. Nothing to do. | |
4702 | */ | |
4703 | if (nid < 0) | |
4704 | return 0; | |
4705 | ||
4706 | /* | |
0121c619 | 4707 | * We are bringing a node online. No memory is available yet. We must |
b9049e23 YG |
4708 | * allocate a kmem_cache_node structure in order to bring the node |
4709 | * online. | |
4710 | */ | |
18004c5d | 4711 | mutex_lock(&slab_mutex); |
b9049e23 | 4712 | list_for_each_entry(s, &slab_caches, list) { |
666716fd VB |
4713 | /* |
4714 | * The structure may already exist if the node was previously | |
4715 | * onlined and offlined. | |
4716 | */ | |
4717 | if (get_node(s, nid)) | |
4718 | continue; | |
b9049e23 YG |
4719 | /* |
4720 | * XXX: kmem_cache_alloc_node will fallback to other nodes | |
4721 | * since memory is not yet available from the node that | |
4722 | * is brought up. | |
4723 | */ | |
8de66a0c | 4724 | n = kmem_cache_alloc(kmem_cache_node, GFP_KERNEL); |
b9049e23 YG |
4725 | if (!n) { |
4726 | ret = -ENOMEM; | |
4727 | goto out; | |
4728 | } | |
4053497d | 4729 | init_kmem_cache_node(n); |
b9049e23 YG |
4730 | s->node[nid] = n; |
4731 | } | |
7e1fa93d VB |
4732 | /* |
4733 | * Any cache created after this point will also have kmem_cache_node | |
4734 | * initialized for the new node. | |
4735 | */ | |
4736 | node_set(nid, slab_nodes); | |
b9049e23 | 4737 | out: |
18004c5d | 4738 | mutex_unlock(&slab_mutex); |
b9049e23 YG |
4739 | return ret; |
4740 | } | |
4741 | ||
4742 | static int slab_memory_callback(struct notifier_block *self, | |
4743 | unsigned long action, void *arg) | |
4744 | { | |
4745 | int ret = 0; | |
4746 | ||
4747 | switch (action) { | |
4748 | case MEM_GOING_ONLINE: | |
4749 | ret = slab_mem_going_online_callback(arg); | |
4750 | break; | |
4751 | case MEM_GOING_OFFLINE: | |
4752 | ret = slab_mem_going_offline_callback(arg); | |
4753 | break; | |
4754 | case MEM_OFFLINE: | |
4755 | case MEM_CANCEL_ONLINE: | |
4756 | slab_mem_offline_callback(arg); | |
4757 | break; | |
4758 | case MEM_ONLINE: | |
4759 | case MEM_CANCEL_OFFLINE: | |
4760 | break; | |
4761 | } | |
dc19f9db KH |
4762 | if (ret) |
4763 | ret = notifier_from_errno(ret); | |
4764 | else | |
4765 | ret = NOTIFY_OK; | |
b9049e23 YG |
4766 | return ret; |
4767 | } | |
4768 | ||
3ac38faa AM |
4769 | static struct notifier_block slab_memory_callback_nb = { |
4770 | .notifier_call = slab_memory_callback, | |
4771 | .priority = SLAB_CALLBACK_PRI, | |
4772 | }; | |
b9049e23 | 4773 | |
81819f0f CL |
4774 | /******************************************************************** |
4775 | * Basic setup of slabs | |
4776 | *******************************************************************/ | |
4777 | ||
51df1142 CL |
4778 | /* |
4779 | * Used for early kmem_cache structures that were allocated using | |
dffb4d60 CL |
4780 | * the page allocator. Allocate them properly then fix up the pointers |
4781 | * that may be pointing to the wrong kmem_cache structure. | |
51df1142 CL |
4782 | */ |
4783 | ||
dffb4d60 | 4784 | static struct kmem_cache * __init bootstrap(struct kmem_cache *static_cache) |
51df1142 CL |
4785 | { |
4786 | int node; | |
dffb4d60 | 4787 | struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT); |
fa45dc25 | 4788 | struct kmem_cache_node *n; |
51df1142 | 4789 | |
dffb4d60 | 4790 | memcpy(s, static_cache, kmem_cache->object_size); |
51df1142 | 4791 | |
7d557b3c GC |
4792 | /* |
4793 | * This runs very early, and only the boot processor is supposed to be | |
4794 | * up. Even if it weren't true, IRQs are not up so we couldn't fire | |
4795 | * IPIs around. | |
4796 | */ | |
4797 | __flush_cpu_slab(s, smp_processor_id()); | |
fa45dc25 | 4798 | for_each_kmem_cache_node(s, node, n) { |
51df1142 CL |
4799 | struct page *p; |
4800 | ||
916ac052 | 4801 | list_for_each_entry(p, &n->partial, slab_list) |
fa45dc25 | 4802 | p->slab_cache = s; |
51df1142 | 4803 | |
607bf324 | 4804 | #ifdef CONFIG_SLUB_DEBUG |
916ac052 | 4805 | list_for_each_entry(p, &n->full, slab_list) |
fa45dc25 | 4806 | p->slab_cache = s; |
51df1142 | 4807 | #endif |
51df1142 | 4808 | } |
dffb4d60 CL |
4809 | list_add(&s->list, &slab_caches); |
4810 | return s; | |
51df1142 CL |
4811 | } |
4812 | ||
81819f0f CL |
4813 | void __init kmem_cache_init(void) |
4814 | { | |
dffb4d60 CL |
4815 | static __initdata struct kmem_cache boot_kmem_cache, |
4816 | boot_kmem_cache_node; | |
7e1fa93d | 4817 | int node; |
51df1142 | 4818 | |
fc8d8620 SG |
4819 | if (debug_guardpage_minorder()) |
4820 | slub_max_order = 0; | |
4821 | ||
79270291 SB |
4822 | /* Print slub debugging pointers without hashing */ |
4823 | if (__slub_debug_enabled()) | |
4824 | no_hash_pointers_enable(NULL); | |
4825 | ||
dffb4d60 CL |
4826 | kmem_cache_node = &boot_kmem_cache_node; |
4827 | kmem_cache = &boot_kmem_cache; | |
51df1142 | 4828 | |
7e1fa93d VB |
4829 | /* |
4830 | * Initialize the nodemask for which we will allocate per node | |
4831 | * structures. Here we don't need taking slab_mutex yet. | |
4832 | */ | |
4833 | for_each_node_state(node, N_NORMAL_MEMORY) | |
4834 | node_set(node, slab_nodes); | |
4835 | ||
dffb4d60 | 4836 | create_boot_cache(kmem_cache_node, "kmem_cache_node", |
8eb8284b | 4837 | sizeof(struct kmem_cache_node), SLAB_HWCACHE_ALIGN, 0, 0); |
b9049e23 | 4838 | |
3ac38faa | 4839 | register_hotmemory_notifier(&slab_memory_callback_nb); |
81819f0f CL |
4840 | |
4841 | /* Able to allocate the per node structures */ | |
4842 | slab_state = PARTIAL; | |
4843 | ||
dffb4d60 CL |
4844 | create_boot_cache(kmem_cache, "kmem_cache", |
4845 | offsetof(struct kmem_cache, node) + | |
4846 | nr_node_ids * sizeof(struct kmem_cache_node *), | |
8eb8284b | 4847 | SLAB_HWCACHE_ALIGN, 0, 0); |
8a13a4cc | 4848 | |
dffb4d60 | 4849 | kmem_cache = bootstrap(&boot_kmem_cache); |
dffb4d60 | 4850 | kmem_cache_node = bootstrap(&boot_kmem_cache_node); |
51df1142 CL |
4851 | |
4852 | /* Now we can use the kmem_cache to allocate kmalloc slabs */ | |
34cc6990 | 4853 | setup_kmalloc_cache_index_table(); |
f97d5f63 | 4854 | create_kmalloc_caches(0); |
81819f0f | 4855 | |
210e7a43 TG |
4856 | /* Setup random freelists for each cache */ |
4857 | init_freelist_randomization(); | |
4858 | ||
a96a87bf SAS |
4859 | cpuhp_setup_state_nocalls(CPUHP_SLUB_DEAD, "slub:dead", NULL, |
4860 | slub_cpu_dead); | |
81819f0f | 4861 | |
b9726c26 | 4862 | pr_info("SLUB: HWalign=%d, Order=%u-%u, MinObjects=%u, CPUs=%u, Nodes=%u\n", |
f97d5f63 | 4863 | cache_line_size(), |
81819f0f CL |
4864 | slub_min_order, slub_max_order, slub_min_objects, |
4865 | nr_cpu_ids, nr_node_ids); | |
4866 | } | |
4867 | ||
7e85ee0c PE |
4868 | void __init kmem_cache_init_late(void) |
4869 | { | |
7e85ee0c PE |
4870 | } |
4871 | ||
2633d7a0 | 4872 | struct kmem_cache * |
f4957d5b | 4873 | __kmem_cache_alias(const char *name, unsigned int size, unsigned int align, |
d50112ed | 4874 | slab_flags_t flags, void (*ctor)(void *)) |
81819f0f | 4875 | { |
10befea9 | 4876 | struct kmem_cache *s; |
81819f0f | 4877 | |
a44cb944 | 4878 | s = find_mergeable(size, align, flags, name, ctor); |
81819f0f CL |
4879 | if (s) { |
4880 | s->refcount++; | |
84d0ddd6 | 4881 | |
81819f0f CL |
4882 | /* |
4883 | * Adjust the object sizes so that we clear | |
4884 | * the complete object on kzalloc. | |
4885 | */ | |
1b473f29 | 4886 | s->object_size = max(s->object_size, size); |
52ee6d74 | 4887 | s->inuse = max(s->inuse, ALIGN(size, sizeof(void *))); |
6446faa2 | 4888 | |
7b8f3b66 | 4889 | if (sysfs_slab_alias(s, name)) { |
7b8f3b66 | 4890 | s->refcount--; |
cbb79694 | 4891 | s = NULL; |
7b8f3b66 | 4892 | } |
a0e1d1be | 4893 | } |
6446faa2 | 4894 | |
cbb79694 CL |
4895 | return s; |
4896 | } | |
84c1cf62 | 4897 | |
d50112ed | 4898 | int __kmem_cache_create(struct kmem_cache *s, slab_flags_t flags) |
cbb79694 | 4899 | { |
aac3a166 PE |
4900 | int err; |
4901 | ||
4902 | err = kmem_cache_open(s, flags); | |
4903 | if (err) | |
4904 | return err; | |
20cea968 | 4905 | |
45530c44 CL |
4906 | /* Mutex is not taken during early boot */ |
4907 | if (slab_state <= UP) | |
4908 | return 0; | |
4909 | ||
aac3a166 | 4910 | err = sysfs_slab_add(s); |
67823a54 | 4911 | if (err) { |
52b4b950 | 4912 | __kmem_cache_release(s); |
67823a54 ML |
4913 | return err; |
4914 | } | |
20cea968 | 4915 | |
64dd6849 FM |
4916 | if (s->flags & SLAB_STORE_USER) |
4917 | debugfs_slab_add(s); | |
4918 | ||
67823a54 | 4919 | return 0; |
81819f0f | 4920 | } |
81819f0f | 4921 | |
ce71e27c | 4922 | void *__kmalloc_track_caller(size_t size, gfp_t gfpflags, unsigned long caller) |
81819f0f | 4923 | { |
aadb4bc4 | 4924 | struct kmem_cache *s; |
94b528d0 | 4925 | void *ret; |
aadb4bc4 | 4926 | |
95a05b42 | 4927 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) |
eada35ef PE |
4928 | return kmalloc_large(size, gfpflags); |
4929 | ||
2c59dd65 | 4930 | s = kmalloc_slab(size, gfpflags); |
81819f0f | 4931 | |
2408c550 | 4932 | if (unlikely(ZERO_OR_NULL_PTR(s))) |
6cb8f913 | 4933 | return s; |
81819f0f | 4934 | |
b89fb5ef | 4935 | ret = slab_alloc(s, gfpflags, caller, size); |
94b528d0 | 4936 | |
25985edc | 4937 | /* Honor the call site pointer we received. */ |
ca2b84cb | 4938 | trace_kmalloc(caller, ret, size, s->size, gfpflags); |
94b528d0 EGM |
4939 | |
4940 | return ret; | |
81819f0f | 4941 | } |
fd7cb575 | 4942 | EXPORT_SYMBOL(__kmalloc_track_caller); |
81819f0f | 4943 | |
5d1f57e4 | 4944 | #ifdef CONFIG_NUMA |
81819f0f | 4945 | void *__kmalloc_node_track_caller(size_t size, gfp_t gfpflags, |
ce71e27c | 4946 | int node, unsigned long caller) |
81819f0f | 4947 | { |
aadb4bc4 | 4948 | struct kmem_cache *s; |
94b528d0 | 4949 | void *ret; |
aadb4bc4 | 4950 | |
95a05b42 | 4951 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) { |
d3e14aa3 XF |
4952 | ret = kmalloc_large_node(size, gfpflags, node); |
4953 | ||
4954 | trace_kmalloc_node(caller, ret, | |
4955 | size, PAGE_SIZE << get_order(size), | |
4956 | gfpflags, node); | |
4957 | ||
4958 | return ret; | |
4959 | } | |
eada35ef | 4960 | |
2c59dd65 | 4961 | s = kmalloc_slab(size, gfpflags); |
81819f0f | 4962 | |
2408c550 | 4963 | if (unlikely(ZERO_OR_NULL_PTR(s))) |
6cb8f913 | 4964 | return s; |
81819f0f | 4965 | |
b89fb5ef | 4966 | ret = slab_alloc_node(s, gfpflags, node, caller, size); |
94b528d0 | 4967 | |
25985edc | 4968 | /* Honor the call site pointer we received. */ |
ca2b84cb | 4969 | trace_kmalloc_node(caller, ret, size, s->size, gfpflags, node); |
94b528d0 EGM |
4970 | |
4971 | return ret; | |
81819f0f | 4972 | } |
fd7cb575 | 4973 | EXPORT_SYMBOL(__kmalloc_node_track_caller); |
5d1f57e4 | 4974 | #endif |
81819f0f | 4975 | |
ab4d5ed5 | 4976 | #ifdef CONFIG_SYSFS |
205ab99d CL |
4977 | static int count_inuse(struct page *page) |
4978 | { | |
4979 | return page->inuse; | |
4980 | } | |
4981 | ||
4982 | static int count_total(struct page *page) | |
4983 | { | |
4984 | return page->objects; | |
4985 | } | |
ab4d5ed5 | 4986 | #endif |
205ab99d | 4987 | |
ab4d5ed5 | 4988 | #ifdef CONFIG_SLUB_DEBUG |
0a19e7dd VB |
4989 | static void validate_slab(struct kmem_cache *s, struct page *page, |
4990 | unsigned long *obj_map) | |
53e15af0 CL |
4991 | { |
4992 | void *p; | |
a973e9dd | 4993 | void *addr = page_address(page); |
a2b4ae8b | 4994 | unsigned long flags; |
90e9f6a6 | 4995 | |
a2b4ae8b | 4996 | slab_lock(page, &flags); |
53e15af0 | 4997 | |
dd98afd4 | 4998 | if (!check_slab(s, page) || !on_freelist(s, page, NULL)) |
90e9f6a6 | 4999 | goto unlock; |
53e15af0 CL |
5000 | |
5001 | /* Now we know that a valid freelist exists */ | |
0a19e7dd | 5002 | __fill_map(obj_map, s, page); |
5f80b13a | 5003 | for_each_object(p, s, addr, page->objects) { |
0a19e7dd | 5004 | u8 val = test_bit(__obj_to_index(s, addr, p), obj_map) ? |
dd98afd4 | 5005 | SLUB_RED_INACTIVE : SLUB_RED_ACTIVE; |
53e15af0 | 5006 | |
dd98afd4 YZ |
5007 | if (!check_object(s, page, p, val)) |
5008 | break; | |
5009 | } | |
90e9f6a6 | 5010 | unlock: |
a2b4ae8b | 5011 | slab_unlock(page, &flags); |
53e15af0 CL |
5012 | } |
5013 | ||
434e245d | 5014 | static int validate_slab_node(struct kmem_cache *s, |
0a19e7dd | 5015 | struct kmem_cache_node *n, unsigned long *obj_map) |
53e15af0 CL |
5016 | { |
5017 | unsigned long count = 0; | |
5018 | struct page *page; | |
5019 | unsigned long flags; | |
5020 | ||
5021 | spin_lock_irqsave(&n->list_lock, flags); | |
5022 | ||
916ac052 | 5023 | list_for_each_entry(page, &n->partial, slab_list) { |
0a19e7dd | 5024 | validate_slab(s, page, obj_map); |
53e15af0 CL |
5025 | count++; |
5026 | } | |
1f9f78b1 | 5027 | if (count != n->nr_partial) { |
f9f58285 FF |
5028 | pr_err("SLUB %s: %ld partial slabs counted but counter=%ld\n", |
5029 | s->name, count, n->nr_partial); | |
1f9f78b1 OG |
5030 | slab_add_kunit_errors(); |
5031 | } | |
53e15af0 CL |
5032 | |
5033 | if (!(s->flags & SLAB_STORE_USER)) | |
5034 | goto out; | |
5035 | ||
916ac052 | 5036 | list_for_each_entry(page, &n->full, slab_list) { |
0a19e7dd | 5037 | validate_slab(s, page, obj_map); |
53e15af0 CL |
5038 | count++; |
5039 | } | |
1f9f78b1 | 5040 | if (count != atomic_long_read(&n->nr_slabs)) { |
f9f58285 FF |
5041 | pr_err("SLUB: %s %ld slabs counted but counter=%ld\n", |
5042 | s->name, count, atomic_long_read(&n->nr_slabs)); | |
1f9f78b1 OG |
5043 | slab_add_kunit_errors(); |
5044 | } | |
53e15af0 CL |
5045 | |
5046 | out: | |
5047 | spin_unlock_irqrestore(&n->list_lock, flags); | |
5048 | return count; | |
5049 | } | |
5050 | ||
1f9f78b1 | 5051 | long validate_slab_cache(struct kmem_cache *s) |
53e15af0 CL |
5052 | { |
5053 | int node; | |
5054 | unsigned long count = 0; | |
fa45dc25 | 5055 | struct kmem_cache_node *n; |
0a19e7dd VB |
5056 | unsigned long *obj_map; |
5057 | ||
5058 | obj_map = bitmap_alloc(oo_objects(s->oo), GFP_KERNEL); | |
5059 | if (!obj_map) | |
5060 | return -ENOMEM; | |
53e15af0 CL |
5061 | |
5062 | flush_all(s); | |
fa45dc25 | 5063 | for_each_kmem_cache_node(s, node, n) |
0a19e7dd VB |
5064 | count += validate_slab_node(s, n, obj_map); |
5065 | ||
5066 | bitmap_free(obj_map); | |
90e9f6a6 | 5067 | |
53e15af0 CL |
5068 | return count; |
5069 | } | |
1f9f78b1 OG |
5070 | EXPORT_SYMBOL(validate_slab_cache); |
5071 | ||
64dd6849 | 5072 | #ifdef CONFIG_DEBUG_FS |
88a420e4 | 5073 | /* |
672bba3a | 5074 | * Generate lists of code addresses where slabcache objects are allocated |
88a420e4 CL |
5075 | * and freed. |
5076 | */ | |
5077 | ||
5078 | struct location { | |
5079 | unsigned long count; | |
ce71e27c | 5080 | unsigned long addr; |
45edfa58 CL |
5081 | long long sum_time; |
5082 | long min_time; | |
5083 | long max_time; | |
5084 | long min_pid; | |
5085 | long max_pid; | |
174596a0 | 5086 | DECLARE_BITMAP(cpus, NR_CPUS); |
45edfa58 | 5087 | nodemask_t nodes; |
88a420e4 CL |
5088 | }; |
5089 | ||
5090 | struct loc_track { | |
5091 | unsigned long max; | |
5092 | unsigned long count; | |
5093 | struct location *loc; | |
005a79e5 | 5094 | loff_t idx; |
88a420e4 CL |
5095 | }; |
5096 | ||
64dd6849 FM |
5097 | static struct dentry *slab_debugfs_root; |
5098 | ||
88a420e4 CL |
5099 | static void free_loc_track(struct loc_track *t) |
5100 | { | |
5101 | if (t->max) | |
5102 | free_pages((unsigned long)t->loc, | |
5103 | get_order(sizeof(struct location) * t->max)); | |
5104 | } | |
5105 | ||
68dff6a9 | 5106 | static int alloc_loc_track(struct loc_track *t, unsigned long max, gfp_t flags) |
88a420e4 CL |
5107 | { |
5108 | struct location *l; | |
5109 | int order; | |
5110 | ||
88a420e4 CL |
5111 | order = get_order(sizeof(struct location) * max); |
5112 | ||
68dff6a9 | 5113 | l = (void *)__get_free_pages(flags, order); |
88a420e4 CL |
5114 | if (!l) |
5115 | return 0; | |
5116 | ||
5117 | if (t->count) { | |
5118 | memcpy(l, t->loc, sizeof(struct location) * t->count); | |
5119 | free_loc_track(t); | |
5120 | } | |
5121 | t->max = max; | |
5122 | t->loc = l; | |
5123 | return 1; | |
5124 | } | |
5125 | ||
5126 | static int add_location(struct loc_track *t, struct kmem_cache *s, | |
45edfa58 | 5127 | const struct track *track) |
88a420e4 CL |
5128 | { |
5129 | long start, end, pos; | |
5130 | struct location *l; | |
ce71e27c | 5131 | unsigned long caddr; |
45edfa58 | 5132 | unsigned long age = jiffies - track->when; |
88a420e4 CL |
5133 | |
5134 | start = -1; | |
5135 | end = t->count; | |
5136 | ||
5137 | for ( ; ; ) { | |
5138 | pos = start + (end - start + 1) / 2; | |
5139 | ||
5140 | /* | |
5141 | * There is nothing at "end". If we end up there | |
5142 | * we need to add something to before end. | |
5143 | */ | |
5144 | if (pos == end) | |
5145 | break; | |
5146 | ||
5147 | caddr = t->loc[pos].addr; | |
45edfa58 CL |
5148 | if (track->addr == caddr) { |
5149 | ||
5150 | l = &t->loc[pos]; | |
5151 | l->count++; | |
5152 | if (track->when) { | |
5153 | l->sum_time += age; | |
5154 | if (age < l->min_time) | |
5155 | l->min_time = age; | |
5156 | if (age > l->max_time) | |
5157 | l->max_time = age; | |
5158 | ||
5159 | if (track->pid < l->min_pid) | |
5160 | l->min_pid = track->pid; | |
5161 | if (track->pid > l->max_pid) | |
5162 | l->max_pid = track->pid; | |
5163 | ||
174596a0 RR |
5164 | cpumask_set_cpu(track->cpu, |
5165 | to_cpumask(l->cpus)); | |
45edfa58 CL |
5166 | } |
5167 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
88a420e4 CL |
5168 | return 1; |
5169 | } | |
5170 | ||
45edfa58 | 5171 | if (track->addr < caddr) |
88a420e4 CL |
5172 | end = pos; |
5173 | else | |
5174 | start = pos; | |
5175 | } | |
5176 | ||
5177 | /* | |
672bba3a | 5178 | * Not found. Insert new tracking element. |
88a420e4 | 5179 | */ |
68dff6a9 | 5180 | if (t->count >= t->max && !alloc_loc_track(t, 2 * t->max, GFP_ATOMIC)) |
88a420e4 CL |
5181 | return 0; |
5182 | ||
5183 | l = t->loc + pos; | |
5184 | if (pos < t->count) | |
5185 | memmove(l + 1, l, | |
5186 | (t->count - pos) * sizeof(struct location)); | |
5187 | t->count++; | |
5188 | l->count = 1; | |
45edfa58 CL |
5189 | l->addr = track->addr; |
5190 | l->sum_time = age; | |
5191 | l->min_time = age; | |
5192 | l->max_time = age; | |
5193 | l->min_pid = track->pid; | |
5194 | l->max_pid = track->pid; | |
174596a0 RR |
5195 | cpumask_clear(to_cpumask(l->cpus)); |
5196 | cpumask_set_cpu(track->cpu, to_cpumask(l->cpus)); | |
45edfa58 CL |
5197 | nodes_clear(l->nodes); |
5198 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
88a420e4 CL |
5199 | return 1; |
5200 | } | |
5201 | ||
5202 | static void process_slab(struct loc_track *t, struct kmem_cache *s, | |
b3fd64e1 VB |
5203 | struct page *page, enum track_item alloc, |
5204 | unsigned long *obj_map) | |
88a420e4 | 5205 | { |
a973e9dd | 5206 | void *addr = page_address(page); |
88a420e4 CL |
5207 | void *p; |
5208 | ||
b3fd64e1 VB |
5209 | __fill_map(obj_map, s, page); |
5210 | ||
224a88be | 5211 | for_each_object(p, s, addr, page->objects) |
b3fd64e1 | 5212 | if (!test_bit(__obj_to_index(s, addr, p), obj_map)) |
45edfa58 | 5213 | add_location(t, s, get_track(s, p, alloc)); |
88a420e4 | 5214 | } |
64dd6849 | 5215 | #endif /* CONFIG_DEBUG_FS */ |
6dfd1b65 | 5216 | #endif /* CONFIG_SLUB_DEBUG */ |
88a420e4 | 5217 | |
ab4d5ed5 | 5218 | #ifdef CONFIG_SYSFS |
81819f0f | 5219 | enum slab_stat_type { |
205ab99d CL |
5220 | SL_ALL, /* All slabs */ |
5221 | SL_PARTIAL, /* Only partially allocated slabs */ | |
5222 | SL_CPU, /* Only slabs used for cpu caches */ | |
5223 | SL_OBJECTS, /* Determine allocated objects not slabs */ | |
5224 | SL_TOTAL /* Determine object capacity not slabs */ | |
81819f0f CL |
5225 | }; |
5226 | ||
205ab99d | 5227 | #define SO_ALL (1 << SL_ALL) |
81819f0f CL |
5228 | #define SO_PARTIAL (1 << SL_PARTIAL) |
5229 | #define SO_CPU (1 << SL_CPU) | |
5230 | #define SO_OBJECTS (1 << SL_OBJECTS) | |
205ab99d | 5231 | #define SO_TOTAL (1 << SL_TOTAL) |
81819f0f | 5232 | |
62e5c4b4 | 5233 | static ssize_t show_slab_objects(struct kmem_cache *s, |
bf16d19a | 5234 | char *buf, unsigned long flags) |
81819f0f CL |
5235 | { |
5236 | unsigned long total = 0; | |
81819f0f CL |
5237 | int node; |
5238 | int x; | |
5239 | unsigned long *nodes; | |
bf16d19a | 5240 | int len = 0; |
81819f0f | 5241 | |
6396bb22 | 5242 | nodes = kcalloc(nr_node_ids, sizeof(unsigned long), GFP_KERNEL); |
62e5c4b4 CG |
5243 | if (!nodes) |
5244 | return -ENOMEM; | |
81819f0f | 5245 | |
205ab99d CL |
5246 | if (flags & SO_CPU) { |
5247 | int cpu; | |
81819f0f | 5248 | |
205ab99d | 5249 | for_each_possible_cpu(cpu) { |
d0e0ac97 CG |
5250 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, |
5251 | cpu); | |
ec3ab083 | 5252 | int node; |
49e22585 | 5253 | struct page *page; |
dfb4f096 | 5254 | |
4db0c3c2 | 5255 | page = READ_ONCE(c->page); |
ec3ab083 CL |
5256 | if (!page) |
5257 | continue; | |
205ab99d | 5258 | |
ec3ab083 CL |
5259 | node = page_to_nid(page); |
5260 | if (flags & SO_TOTAL) | |
5261 | x = page->objects; | |
5262 | else if (flags & SO_OBJECTS) | |
5263 | x = page->inuse; | |
5264 | else | |
5265 | x = 1; | |
49e22585 | 5266 | |
ec3ab083 CL |
5267 | total += x; |
5268 | nodes[node] += x; | |
5269 | ||
a93cf07b | 5270 | page = slub_percpu_partial_read_once(c); |
49e22585 | 5271 | if (page) { |
8afb1474 LZ |
5272 | node = page_to_nid(page); |
5273 | if (flags & SO_TOTAL) | |
5274 | WARN_ON_ONCE(1); | |
5275 | else if (flags & SO_OBJECTS) | |
5276 | WARN_ON_ONCE(1); | |
5277 | else | |
5278 | x = page->pages; | |
bc6697d8 ED |
5279 | total += x; |
5280 | nodes[node] += x; | |
49e22585 | 5281 | } |
81819f0f CL |
5282 | } |
5283 | } | |
5284 | ||
e4f8e513 QC |
5285 | /* |
5286 | * It is impossible to take "mem_hotplug_lock" here with "kernfs_mutex" | |
5287 | * already held which will conflict with an existing lock order: | |
5288 | * | |
5289 | * mem_hotplug_lock->slab_mutex->kernfs_mutex | |
5290 | * | |
5291 | * We don't really need mem_hotplug_lock (to hold off | |
5292 | * slab_mem_going_offline_callback) here because slab's memory hot | |
5293 | * unplug code doesn't destroy the kmem_cache->node[] data. | |
5294 | */ | |
5295 | ||
ab4d5ed5 | 5296 | #ifdef CONFIG_SLUB_DEBUG |
205ab99d | 5297 | if (flags & SO_ALL) { |
fa45dc25 CL |
5298 | struct kmem_cache_node *n; |
5299 | ||
5300 | for_each_kmem_cache_node(s, node, n) { | |
205ab99d | 5301 | |
d0e0ac97 CG |
5302 | if (flags & SO_TOTAL) |
5303 | x = atomic_long_read(&n->total_objects); | |
5304 | else if (flags & SO_OBJECTS) | |
5305 | x = atomic_long_read(&n->total_objects) - | |
5306 | count_partial(n, count_free); | |
81819f0f | 5307 | else |
205ab99d | 5308 | x = atomic_long_read(&n->nr_slabs); |
81819f0f CL |
5309 | total += x; |
5310 | nodes[node] += x; | |
5311 | } | |
5312 | ||
ab4d5ed5 CL |
5313 | } else |
5314 | #endif | |
5315 | if (flags & SO_PARTIAL) { | |
fa45dc25 | 5316 | struct kmem_cache_node *n; |
81819f0f | 5317 | |
fa45dc25 | 5318 | for_each_kmem_cache_node(s, node, n) { |
205ab99d CL |
5319 | if (flags & SO_TOTAL) |
5320 | x = count_partial(n, count_total); | |
5321 | else if (flags & SO_OBJECTS) | |
5322 | x = count_partial(n, count_inuse); | |
81819f0f | 5323 | else |
205ab99d | 5324 | x = n->nr_partial; |
81819f0f CL |
5325 | total += x; |
5326 | nodes[node] += x; | |
5327 | } | |
5328 | } | |
bf16d19a JP |
5329 | |
5330 | len += sysfs_emit_at(buf, len, "%lu", total); | |
81819f0f | 5331 | #ifdef CONFIG_NUMA |
bf16d19a | 5332 | for (node = 0; node < nr_node_ids; node++) { |
81819f0f | 5333 | if (nodes[node]) |
bf16d19a JP |
5334 | len += sysfs_emit_at(buf, len, " N%d=%lu", |
5335 | node, nodes[node]); | |
5336 | } | |
81819f0f | 5337 | #endif |
bf16d19a | 5338 | len += sysfs_emit_at(buf, len, "\n"); |
81819f0f | 5339 | kfree(nodes); |
bf16d19a JP |
5340 | |
5341 | return len; | |
81819f0f CL |
5342 | } |
5343 | ||
81819f0f | 5344 | #define to_slab_attr(n) container_of(n, struct slab_attribute, attr) |
497888cf | 5345 | #define to_slab(n) container_of(n, struct kmem_cache, kobj) |
81819f0f CL |
5346 | |
5347 | struct slab_attribute { | |
5348 | struct attribute attr; | |
5349 | ssize_t (*show)(struct kmem_cache *s, char *buf); | |
5350 | ssize_t (*store)(struct kmem_cache *s, const char *x, size_t count); | |
5351 | }; | |
5352 | ||
5353 | #define SLAB_ATTR_RO(_name) \ | |
ab067e99 VK |
5354 | static struct slab_attribute _name##_attr = \ |
5355 | __ATTR(_name, 0400, _name##_show, NULL) | |
81819f0f CL |
5356 | |
5357 | #define SLAB_ATTR(_name) \ | |
5358 | static struct slab_attribute _name##_attr = \ | |
ab067e99 | 5359 | __ATTR(_name, 0600, _name##_show, _name##_store) |
81819f0f | 5360 | |
81819f0f CL |
5361 | static ssize_t slab_size_show(struct kmem_cache *s, char *buf) |
5362 | { | |
bf16d19a | 5363 | return sysfs_emit(buf, "%u\n", s->size); |
81819f0f CL |
5364 | } |
5365 | SLAB_ATTR_RO(slab_size); | |
5366 | ||
5367 | static ssize_t align_show(struct kmem_cache *s, char *buf) | |
5368 | { | |
bf16d19a | 5369 | return sysfs_emit(buf, "%u\n", s->align); |
81819f0f CL |
5370 | } |
5371 | SLAB_ATTR_RO(align); | |
5372 | ||
5373 | static ssize_t object_size_show(struct kmem_cache *s, char *buf) | |
5374 | { | |
bf16d19a | 5375 | return sysfs_emit(buf, "%u\n", s->object_size); |
81819f0f CL |
5376 | } |
5377 | SLAB_ATTR_RO(object_size); | |
5378 | ||
5379 | static ssize_t objs_per_slab_show(struct kmem_cache *s, char *buf) | |
5380 | { | |
bf16d19a | 5381 | return sysfs_emit(buf, "%u\n", oo_objects(s->oo)); |
81819f0f CL |
5382 | } |
5383 | SLAB_ATTR_RO(objs_per_slab); | |
5384 | ||
5385 | static ssize_t order_show(struct kmem_cache *s, char *buf) | |
5386 | { | |
bf16d19a | 5387 | return sysfs_emit(buf, "%u\n", oo_order(s->oo)); |
81819f0f | 5388 | } |
32a6f409 | 5389 | SLAB_ATTR_RO(order); |
81819f0f | 5390 | |
73d342b1 DR |
5391 | static ssize_t min_partial_show(struct kmem_cache *s, char *buf) |
5392 | { | |
bf16d19a | 5393 | return sysfs_emit(buf, "%lu\n", s->min_partial); |
73d342b1 DR |
5394 | } |
5395 | ||
5396 | static ssize_t min_partial_store(struct kmem_cache *s, const char *buf, | |
5397 | size_t length) | |
5398 | { | |
5399 | unsigned long min; | |
5400 | int err; | |
5401 | ||
3dbb95f7 | 5402 | err = kstrtoul(buf, 10, &min); |
73d342b1 DR |
5403 | if (err) |
5404 | return err; | |
5405 | ||
c0bdb232 | 5406 | set_min_partial(s, min); |
73d342b1 DR |
5407 | return length; |
5408 | } | |
5409 | SLAB_ATTR(min_partial); | |
5410 | ||
49e22585 CL |
5411 | static ssize_t cpu_partial_show(struct kmem_cache *s, char *buf) |
5412 | { | |
b47291ef VB |
5413 | unsigned int nr_partial = 0; |
5414 | #ifdef CONFIG_SLUB_CPU_PARTIAL | |
5415 | nr_partial = s->cpu_partial; | |
5416 | #endif | |
5417 | ||
5418 | return sysfs_emit(buf, "%u\n", nr_partial); | |
49e22585 CL |
5419 | } |
5420 | ||
5421 | static ssize_t cpu_partial_store(struct kmem_cache *s, const char *buf, | |
5422 | size_t length) | |
5423 | { | |
e5d9998f | 5424 | unsigned int objects; |
49e22585 CL |
5425 | int err; |
5426 | ||
e5d9998f | 5427 | err = kstrtouint(buf, 10, &objects); |
49e22585 CL |
5428 | if (err) |
5429 | return err; | |
345c905d | 5430 | if (objects && !kmem_cache_has_cpu_partial(s)) |
74ee4ef1 | 5431 | return -EINVAL; |
49e22585 | 5432 | |
e6d0e1dc | 5433 | slub_set_cpu_partial(s, objects); |
49e22585 CL |
5434 | flush_all(s); |
5435 | return length; | |
5436 | } | |
5437 | SLAB_ATTR(cpu_partial); | |
5438 | ||
81819f0f CL |
5439 | static ssize_t ctor_show(struct kmem_cache *s, char *buf) |
5440 | { | |
62c70bce JP |
5441 | if (!s->ctor) |
5442 | return 0; | |
bf16d19a | 5443 | return sysfs_emit(buf, "%pS\n", s->ctor); |
81819f0f CL |
5444 | } |
5445 | SLAB_ATTR_RO(ctor); | |
5446 | ||
81819f0f CL |
5447 | static ssize_t aliases_show(struct kmem_cache *s, char *buf) |
5448 | { | |
bf16d19a | 5449 | return sysfs_emit(buf, "%d\n", s->refcount < 0 ? 0 : s->refcount - 1); |
81819f0f CL |
5450 | } |
5451 | SLAB_ATTR_RO(aliases); | |
5452 | ||
81819f0f CL |
5453 | static ssize_t partial_show(struct kmem_cache *s, char *buf) |
5454 | { | |
d9acf4b7 | 5455 | return show_slab_objects(s, buf, SO_PARTIAL); |
81819f0f CL |
5456 | } |
5457 | SLAB_ATTR_RO(partial); | |
5458 | ||
5459 | static ssize_t cpu_slabs_show(struct kmem_cache *s, char *buf) | |
5460 | { | |
d9acf4b7 | 5461 | return show_slab_objects(s, buf, SO_CPU); |
81819f0f CL |
5462 | } |
5463 | SLAB_ATTR_RO(cpu_slabs); | |
5464 | ||
5465 | static ssize_t objects_show(struct kmem_cache *s, char *buf) | |
5466 | { | |
205ab99d | 5467 | return show_slab_objects(s, buf, SO_ALL|SO_OBJECTS); |
81819f0f CL |
5468 | } |
5469 | SLAB_ATTR_RO(objects); | |
5470 | ||
205ab99d CL |
5471 | static ssize_t objects_partial_show(struct kmem_cache *s, char *buf) |
5472 | { | |
5473 | return show_slab_objects(s, buf, SO_PARTIAL|SO_OBJECTS); | |
5474 | } | |
5475 | SLAB_ATTR_RO(objects_partial); | |
5476 | ||
49e22585 CL |
5477 | static ssize_t slabs_cpu_partial_show(struct kmem_cache *s, char *buf) |
5478 | { | |
5479 | int objects = 0; | |
5480 | int pages = 0; | |
5481 | int cpu; | |
bf16d19a | 5482 | int len = 0; |
49e22585 CL |
5483 | |
5484 | for_each_online_cpu(cpu) { | |
a93cf07b WY |
5485 | struct page *page; |
5486 | ||
5487 | page = slub_percpu_partial(per_cpu_ptr(s->cpu_slab, cpu)); | |
49e22585 | 5488 | |
b47291ef | 5489 | if (page) |
49e22585 | 5490 | pages += page->pages; |
49e22585 CL |
5491 | } |
5492 | ||
b47291ef VB |
5493 | /* Approximate half-full pages , see slub_set_cpu_partial() */ |
5494 | objects = (pages * oo_objects(s->oo)) / 2; | |
bf16d19a | 5495 | len += sysfs_emit_at(buf, len, "%d(%d)", objects, pages); |
49e22585 CL |
5496 | |
5497 | #ifdef CONFIG_SMP | |
5498 | for_each_online_cpu(cpu) { | |
a93cf07b WY |
5499 | struct page *page; |
5500 | ||
5501 | page = slub_percpu_partial(per_cpu_ptr(s->cpu_slab, cpu)); | |
b47291ef VB |
5502 | if (page) { |
5503 | pages = READ_ONCE(page->pages); | |
5504 | objects = (pages * oo_objects(s->oo)) / 2; | |
bf16d19a | 5505 | len += sysfs_emit_at(buf, len, " C%d=%d(%d)", |
b47291ef VB |
5506 | cpu, objects, pages); |
5507 | } | |
49e22585 CL |
5508 | } |
5509 | #endif | |
bf16d19a JP |
5510 | len += sysfs_emit_at(buf, len, "\n"); |
5511 | ||
5512 | return len; | |
49e22585 CL |
5513 | } |
5514 | SLAB_ATTR_RO(slabs_cpu_partial); | |
5515 | ||
a5a84755 CL |
5516 | static ssize_t reclaim_account_show(struct kmem_cache *s, char *buf) |
5517 | { | |
bf16d19a | 5518 | return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_RECLAIM_ACCOUNT)); |
a5a84755 | 5519 | } |
8f58119a | 5520 | SLAB_ATTR_RO(reclaim_account); |
a5a84755 CL |
5521 | |
5522 | static ssize_t hwcache_align_show(struct kmem_cache *s, char *buf) | |
5523 | { | |
bf16d19a | 5524 | return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_HWCACHE_ALIGN)); |
a5a84755 CL |
5525 | } |
5526 | SLAB_ATTR_RO(hwcache_align); | |
5527 | ||
5528 | #ifdef CONFIG_ZONE_DMA | |
5529 | static ssize_t cache_dma_show(struct kmem_cache *s, char *buf) | |
5530 | { | |
bf16d19a | 5531 | return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_CACHE_DMA)); |
a5a84755 CL |
5532 | } |
5533 | SLAB_ATTR_RO(cache_dma); | |
5534 | #endif | |
5535 | ||
8eb8284b DW |
5536 | static ssize_t usersize_show(struct kmem_cache *s, char *buf) |
5537 | { | |
bf16d19a | 5538 | return sysfs_emit(buf, "%u\n", s->usersize); |
8eb8284b DW |
5539 | } |
5540 | SLAB_ATTR_RO(usersize); | |
5541 | ||
a5a84755 CL |
5542 | static ssize_t destroy_by_rcu_show(struct kmem_cache *s, char *buf) |
5543 | { | |
bf16d19a | 5544 | return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_TYPESAFE_BY_RCU)); |
a5a84755 CL |
5545 | } |
5546 | SLAB_ATTR_RO(destroy_by_rcu); | |
5547 | ||
ab4d5ed5 | 5548 | #ifdef CONFIG_SLUB_DEBUG |
a5a84755 CL |
5549 | static ssize_t slabs_show(struct kmem_cache *s, char *buf) |
5550 | { | |
5551 | return show_slab_objects(s, buf, SO_ALL); | |
5552 | } | |
5553 | SLAB_ATTR_RO(slabs); | |
5554 | ||
205ab99d CL |
5555 | static ssize_t total_objects_show(struct kmem_cache *s, char *buf) |
5556 | { | |
5557 | return show_slab_objects(s, buf, SO_ALL|SO_TOTAL); | |
5558 | } | |
5559 | SLAB_ATTR_RO(total_objects); | |
5560 | ||
81819f0f CL |
5561 | static ssize_t sanity_checks_show(struct kmem_cache *s, char *buf) |
5562 | { | |
bf16d19a | 5563 | return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_CONSISTENCY_CHECKS)); |
81819f0f | 5564 | } |
060807f8 | 5565 | SLAB_ATTR_RO(sanity_checks); |
81819f0f CL |
5566 | |
5567 | static ssize_t trace_show(struct kmem_cache *s, char *buf) | |
5568 | { | |
bf16d19a | 5569 | return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_TRACE)); |
81819f0f | 5570 | } |
060807f8 | 5571 | SLAB_ATTR_RO(trace); |
81819f0f | 5572 | |
81819f0f CL |
5573 | static ssize_t red_zone_show(struct kmem_cache *s, char *buf) |
5574 | { | |
bf16d19a | 5575 | return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_RED_ZONE)); |
81819f0f CL |
5576 | } |
5577 | ||
ad38b5b1 | 5578 | SLAB_ATTR_RO(red_zone); |
81819f0f CL |
5579 | |
5580 | static ssize_t poison_show(struct kmem_cache *s, char *buf) | |
5581 | { | |
bf16d19a | 5582 | return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_POISON)); |
81819f0f CL |
5583 | } |
5584 | ||
ad38b5b1 | 5585 | SLAB_ATTR_RO(poison); |
81819f0f CL |
5586 | |
5587 | static ssize_t store_user_show(struct kmem_cache *s, char *buf) | |
5588 | { | |
bf16d19a | 5589 | return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_STORE_USER)); |
81819f0f CL |
5590 | } |
5591 | ||
ad38b5b1 | 5592 | SLAB_ATTR_RO(store_user); |
81819f0f | 5593 | |
53e15af0 CL |
5594 | static ssize_t validate_show(struct kmem_cache *s, char *buf) |
5595 | { | |
5596 | return 0; | |
5597 | } | |
5598 | ||
5599 | static ssize_t validate_store(struct kmem_cache *s, | |
5600 | const char *buf, size_t length) | |
5601 | { | |
434e245d CL |
5602 | int ret = -EINVAL; |
5603 | ||
5604 | if (buf[0] == '1') { | |
5605 | ret = validate_slab_cache(s); | |
5606 | if (ret >= 0) | |
5607 | ret = length; | |
5608 | } | |
5609 | return ret; | |
53e15af0 CL |
5610 | } |
5611 | SLAB_ATTR(validate); | |
a5a84755 | 5612 | |
a5a84755 CL |
5613 | #endif /* CONFIG_SLUB_DEBUG */ |
5614 | ||
5615 | #ifdef CONFIG_FAILSLAB | |
5616 | static ssize_t failslab_show(struct kmem_cache *s, char *buf) | |
5617 | { | |
bf16d19a | 5618 | return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_FAILSLAB)); |
a5a84755 | 5619 | } |
060807f8 | 5620 | SLAB_ATTR_RO(failslab); |
ab4d5ed5 | 5621 | #endif |
53e15af0 | 5622 | |
2086d26a CL |
5623 | static ssize_t shrink_show(struct kmem_cache *s, char *buf) |
5624 | { | |
5625 | return 0; | |
5626 | } | |
5627 | ||
5628 | static ssize_t shrink_store(struct kmem_cache *s, | |
5629 | const char *buf, size_t length) | |
5630 | { | |
832f37f5 | 5631 | if (buf[0] == '1') |
10befea9 | 5632 | kmem_cache_shrink(s); |
832f37f5 | 5633 | else |
2086d26a CL |
5634 | return -EINVAL; |
5635 | return length; | |
5636 | } | |
5637 | SLAB_ATTR(shrink); | |
5638 | ||
81819f0f | 5639 | #ifdef CONFIG_NUMA |
9824601e | 5640 | static ssize_t remote_node_defrag_ratio_show(struct kmem_cache *s, char *buf) |
81819f0f | 5641 | { |
bf16d19a | 5642 | return sysfs_emit(buf, "%u\n", s->remote_node_defrag_ratio / 10); |
81819f0f CL |
5643 | } |
5644 | ||
9824601e | 5645 | static ssize_t remote_node_defrag_ratio_store(struct kmem_cache *s, |
81819f0f CL |
5646 | const char *buf, size_t length) |
5647 | { | |
eb7235eb | 5648 | unsigned int ratio; |
0121c619 CL |
5649 | int err; |
5650 | ||
eb7235eb | 5651 | err = kstrtouint(buf, 10, &ratio); |
0121c619 CL |
5652 | if (err) |
5653 | return err; | |
eb7235eb AD |
5654 | if (ratio > 100) |
5655 | return -ERANGE; | |
0121c619 | 5656 | |
eb7235eb | 5657 | s->remote_node_defrag_ratio = ratio * 10; |
81819f0f | 5658 | |
81819f0f CL |
5659 | return length; |
5660 | } | |
9824601e | 5661 | SLAB_ATTR(remote_node_defrag_ratio); |
81819f0f CL |
5662 | #endif |
5663 | ||
8ff12cfc | 5664 | #ifdef CONFIG_SLUB_STATS |
8ff12cfc CL |
5665 | static int show_stat(struct kmem_cache *s, char *buf, enum stat_item si) |
5666 | { | |
5667 | unsigned long sum = 0; | |
5668 | int cpu; | |
bf16d19a | 5669 | int len = 0; |
6da2ec56 | 5670 | int *data = kmalloc_array(nr_cpu_ids, sizeof(int), GFP_KERNEL); |
8ff12cfc CL |
5671 | |
5672 | if (!data) | |
5673 | return -ENOMEM; | |
5674 | ||
5675 | for_each_online_cpu(cpu) { | |
9dfc6e68 | 5676 | unsigned x = per_cpu_ptr(s->cpu_slab, cpu)->stat[si]; |
8ff12cfc CL |
5677 | |
5678 | data[cpu] = x; | |
5679 | sum += x; | |
5680 | } | |
5681 | ||
bf16d19a | 5682 | len += sysfs_emit_at(buf, len, "%lu", sum); |
8ff12cfc | 5683 | |
50ef37b9 | 5684 | #ifdef CONFIG_SMP |
8ff12cfc | 5685 | for_each_online_cpu(cpu) { |
bf16d19a JP |
5686 | if (data[cpu]) |
5687 | len += sysfs_emit_at(buf, len, " C%d=%u", | |
5688 | cpu, data[cpu]); | |
8ff12cfc | 5689 | } |
50ef37b9 | 5690 | #endif |
8ff12cfc | 5691 | kfree(data); |
bf16d19a JP |
5692 | len += sysfs_emit_at(buf, len, "\n"); |
5693 | ||
5694 | return len; | |
8ff12cfc CL |
5695 | } |
5696 | ||
78eb00cc DR |
5697 | static void clear_stat(struct kmem_cache *s, enum stat_item si) |
5698 | { | |
5699 | int cpu; | |
5700 | ||
5701 | for_each_online_cpu(cpu) | |
9dfc6e68 | 5702 | per_cpu_ptr(s->cpu_slab, cpu)->stat[si] = 0; |
78eb00cc DR |
5703 | } |
5704 | ||
8ff12cfc CL |
5705 | #define STAT_ATTR(si, text) \ |
5706 | static ssize_t text##_show(struct kmem_cache *s, char *buf) \ | |
5707 | { \ | |
5708 | return show_stat(s, buf, si); \ | |
5709 | } \ | |
78eb00cc DR |
5710 | static ssize_t text##_store(struct kmem_cache *s, \ |
5711 | const char *buf, size_t length) \ | |
5712 | { \ | |
5713 | if (buf[0] != '0') \ | |
5714 | return -EINVAL; \ | |
5715 | clear_stat(s, si); \ | |
5716 | return length; \ | |
5717 | } \ | |
5718 | SLAB_ATTR(text); \ | |
8ff12cfc CL |
5719 | |
5720 | STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath); | |
5721 | STAT_ATTR(ALLOC_SLOWPATH, alloc_slowpath); | |
5722 | STAT_ATTR(FREE_FASTPATH, free_fastpath); | |
5723 | STAT_ATTR(FREE_SLOWPATH, free_slowpath); | |
5724 | STAT_ATTR(FREE_FROZEN, free_frozen); | |
5725 | STAT_ATTR(FREE_ADD_PARTIAL, free_add_partial); | |
5726 | STAT_ATTR(FREE_REMOVE_PARTIAL, free_remove_partial); | |
5727 | STAT_ATTR(ALLOC_FROM_PARTIAL, alloc_from_partial); | |
5728 | STAT_ATTR(ALLOC_SLAB, alloc_slab); | |
5729 | STAT_ATTR(ALLOC_REFILL, alloc_refill); | |
e36a2652 | 5730 | STAT_ATTR(ALLOC_NODE_MISMATCH, alloc_node_mismatch); |
8ff12cfc CL |
5731 | STAT_ATTR(FREE_SLAB, free_slab); |
5732 | STAT_ATTR(CPUSLAB_FLUSH, cpuslab_flush); | |
5733 | STAT_ATTR(DEACTIVATE_FULL, deactivate_full); | |
5734 | STAT_ATTR(DEACTIVATE_EMPTY, deactivate_empty); | |
5735 | STAT_ATTR(DEACTIVATE_TO_HEAD, deactivate_to_head); | |
5736 | STAT_ATTR(DEACTIVATE_TO_TAIL, deactivate_to_tail); | |
5737 | STAT_ATTR(DEACTIVATE_REMOTE_FREES, deactivate_remote_frees); | |
03e404af | 5738 | STAT_ATTR(DEACTIVATE_BYPASS, deactivate_bypass); |
65c3376a | 5739 | STAT_ATTR(ORDER_FALLBACK, order_fallback); |
b789ef51 CL |
5740 | STAT_ATTR(CMPXCHG_DOUBLE_CPU_FAIL, cmpxchg_double_cpu_fail); |
5741 | STAT_ATTR(CMPXCHG_DOUBLE_FAIL, cmpxchg_double_fail); | |
49e22585 CL |
5742 | STAT_ATTR(CPU_PARTIAL_ALLOC, cpu_partial_alloc); |
5743 | STAT_ATTR(CPU_PARTIAL_FREE, cpu_partial_free); | |
8028dcea AS |
5744 | STAT_ATTR(CPU_PARTIAL_NODE, cpu_partial_node); |
5745 | STAT_ATTR(CPU_PARTIAL_DRAIN, cpu_partial_drain); | |
6dfd1b65 | 5746 | #endif /* CONFIG_SLUB_STATS */ |
8ff12cfc | 5747 | |
06428780 | 5748 | static struct attribute *slab_attrs[] = { |
81819f0f CL |
5749 | &slab_size_attr.attr, |
5750 | &object_size_attr.attr, | |
5751 | &objs_per_slab_attr.attr, | |
5752 | &order_attr.attr, | |
73d342b1 | 5753 | &min_partial_attr.attr, |
49e22585 | 5754 | &cpu_partial_attr.attr, |
81819f0f | 5755 | &objects_attr.attr, |
205ab99d | 5756 | &objects_partial_attr.attr, |
81819f0f CL |
5757 | &partial_attr.attr, |
5758 | &cpu_slabs_attr.attr, | |
5759 | &ctor_attr.attr, | |
81819f0f CL |
5760 | &aliases_attr.attr, |
5761 | &align_attr.attr, | |
81819f0f CL |
5762 | &hwcache_align_attr.attr, |
5763 | &reclaim_account_attr.attr, | |
5764 | &destroy_by_rcu_attr.attr, | |
a5a84755 | 5765 | &shrink_attr.attr, |
49e22585 | 5766 | &slabs_cpu_partial_attr.attr, |
ab4d5ed5 | 5767 | #ifdef CONFIG_SLUB_DEBUG |
a5a84755 CL |
5768 | &total_objects_attr.attr, |
5769 | &slabs_attr.attr, | |
5770 | &sanity_checks_attr.attr, | |
5771 | &trace_attr.attr, | |
81819f0f CL |
5772 | &red_zone_attr.attr, |
5773 | &poison_attr.attr, | |
5774 | &store_user_attr.attr, | |
53e15af0 | 5775 | &validate_attr.attr, |
ab4d5ed5 | 5776 | #endif |
81819f0f CL |
5777 | #ifdef CONFIG_ZONE_DMA |
5778 | &cache_dma_attr.attr, | |
5779 | #endif | |
5780 | #ifdef CONFIG_NUMA | |
9824601e | 5781 | &remote_node_defrag_ratio_attr.attr, |
8ff12cfc CL |
5782 | #endif |
5783 | #ifdef CONFIG_SLUB_STATS | |
5784 | &alloc_fastpath_attr.attr, | |
5785 | &alloc_slowpath_attr.attr, | |
5786 | &free_fastpath_attr.attr, | |
5787 | &free_slowpath_attr.attr, | |
5788 | &free_frozen_attr.attr, | |
5789 | &free_add_partial_attr.attr, | |
5790 | &free_remove_partial_attr.attr, | |
5791 | &alloc_from_partial_attr.attr, | |
5792 | &alloc_slab_attr.attr, | |
5793 | &alloc_refill_attr.attr, | |
e36a2652 | 5794 | &alloc_node_mismatch_attr.attr, |
8ff12cfc CL |
5795 | &free_slab_attr.attr, |
5796 | &cpuslab_flush_attr.attr, | |
5797 | &deactivate_full_attr.attr, | |
5798 | &deactivate_empty_attr.attr, | |
5799 | &deactivate_to_head_attr.attr, | |
5800 | &deactivate_to_tail_attr.attr, | |
5801 | &deactivate_remote_frees_attr.attr, | |
03e404af | 5802 | &deactivate_bypass_attr.attr, |
65c3376a | 5803 | &order_fallback_attr.attr, |
b789ef51 CL |
5804 | &cmpxchg_double_fail_attr.attr, |
5805 | &cmpxchg_double_cpu_fail_attr.attr, | |
49e22585 CL |
5806 | &cpu_partial_alloc_attr.attr, |
5807 | &cpu_partial_free_attr.attr, | |
8028dcea AS |
5808 | &cpu_partial_node_attr.attr, |
5809 | &cpu_partial_drain_attr.attr, | |
81819f0f | 5810 | #endif |
4c13dd3b DM |
5811 | #ifdef CONFIG_FAILSLAB |
5812 | &failslab_attr.attr, | |
5813 | #endif | |
8eb8284b | 5814 | &usersize_attr.attr, |
4c13dd3b | 5815 | |
81819f0f CL |
5816 | NULL |
5817 | }; | |
5818 | ||
1fdaaa23 | 5819 | static const struct attribute_group slab_attr_group = { |
81819f0f CL |
5820 | .attrs = slab_attrs, |
5821 | }; | |
5822 | ||
5823 | static ssize_t slab_attr_show(struct kobject *kobj, | |
5824 | struct attribute *attr, | |
5825 | char *buf) | |
5826 | { | |
5827 | struct slab_attribute *attribute; | |
5828 | struct kmem_cache *s; | |
5829 | int err; | |
5830 | ||
5831 | attribute = to_slab_attr(attr); | |
5832 | s = to_slab(kobj); | |
5833 | ||
5834 | if (!attribute->show) | |
5835 | return -EIO; | |
5836 | ||
5837 | err = attribute->show(s, buf); | |
5838 | ||
5839 | return err; | |
5840 | } | |
5841 | ||
5842 | static ssize_t slab_attr_store(struct kobject *kobj, | |
5843 | struct attribute *attr, | |
5844 | const char *buf, size_t len) | |
5845 | { | |
5846 | struct slab_attribute *attribute; | |
5847 | struct kmem_cache *s; | |
5848 | int err; | |
5849 | ||
5850 | attribute = to_slab_attr(attr); | |
5851 | s = to_slab(kobj); | |
5852 | ||
5853 | if (!attribute->store) | |
5854 | return -EIO; | |
5855 | ||
5856 | err = attribute->store(s, buf, len); | |
81819f0f CL |
5857 | return err; |
5858 | } | |
5859 | ||
41a21285 CL |
5860 | static void kmem_cache_release(struct kobject *k) |
5861 | { | |
5862 | slab_kmem_cache_release(to_slab(k)); | |
5863 | } | |
5864 | ||
52cf25d0 | 5865 | static const struct sysfs_ops slab_sysfs_ops = { |
81819f0f CL |
5866 | .show = slab_attr_show, |
5867 | .store = slab_attr_store, | |
5868 | }; | |
5869 | ||
5870 | static struct kobj_type slab_ktype = { | |
5871 | .sysfs_ops = &slab_sysfs_ops, | |
41a21285 | 5872 | .release = kmem_cache_release, |
81819f0f CL |
5873 | }; |
5874 | ||
27c3a314 | 5875 | static struct kset *slab_kset; |
81819f0f | 5876 | |
9a41707b VD |
5877 | static inline struct kset *cache_kset(struct kmem_cache *s) |
5878 | { | |
9a41707b VD |
5879 | return slab_kset; |
5880 | } | |
5881 | ||
81819f0f CL |
5882 | #define ID_STR_LENGTH 64 |
5883 | ||
5884 | /* Create a unique string id for a slab cache: | |
6446faa2 CL |
5885 | * |
5886 | * Format :[flags-]size | |
81819f0f CL |
5887 | */ |
5888 | static char *create_unique_id(struct kmem_cache *s) | |
5889 | { | |
5890 | char *name = kmalloc(ID_STR_LENGTH, GFP_KERNEL); | |
5891 | char *p = name; | |
5892 | ||
5893 | BUG_ON(!name); | |
5894 | ||
5895 | *p++ = ':'; | |
5896 | /* | |
5897 | * First flags affecting slabcache operations. We will only | |
5898 | * get here for aliasable slabs so we do not need to support | |
5899 | * too many flags. The flags here must cover all flags that | |
5900 | * are matched during merging to guarantee that the id is | |
5901 | * unique. | |
5902 | */ | |
5903 | if (s->flags & SLAB_CACHE_DMA) | |
5904 | *p++ = 'd'; | |
6d6ea1e9 NB |
5905 | if (s->flags & SLAB_CACHE_DMA32) |
5906 | *p++ = 'D'; | |
81819f0f CL |
5907 | if (s->flags & SLAB_RECLAIM_ACCOUNT) |
5908 | *p++ = 'a'; | |
becfda68 | 5909 | if (s->flags & SLAB_CONSISTENCY_CHECKS) |
81819f0f | 5910 | *p++ = 'F'; |
230e9fc2 VD |
5911 | if (s->flags & SLAB_ACCOUNT) |
5912 | *p++ = 'A'; | |
81819f0f CL |
5913 | if (p != name + 1) |
5914 | *p++ = '-'; | |
44065b2e | 5915 | p += sprintf(p, "%07u", s->size); |
2633d7a0 | 5916 | |
81819f0f CL |
5917 | BUG_ON(p > name + ID_STR_LENGTH - 1); |
5918 | return name; | |
5919 | } | |
5920 | ||
5921 | static int sysfs_slab_add(struct kmem_cache *s) | |
5922 | { | |
5923 | int err; | |
5924 | const char *name; | |
1663f26d | 5925 | struct kset *kset = cache_kset(s); |
45530c44 | 5926 | int unmergeable = slab_unmergeable(s); |
81819f0f | 5927 | |
1663f26d TH |
5928 | if (!kset) { |
5929 | kobject_init(&s->kobj, &slab_ktype); | |
5930 | return 0; | |
5931 | } | |
5932 | ||
11066386 MC |
5933 | if (!unmergeable && disable_higher_order_debug && |
5934 | (slub_debug & DEBUG_METADATA_FLAGS)) | |
5935 | unmergeable = 1; | |
5936 | ||
81819f0f CL |
5937 | if (unmergeable) { |
5938 | /* | |
5939 | * Slabcache can never be merged so we can use the name proper. | |
5940 | * This is typically the case for debug situations. In that | |
5941 | * case we can catch duplicate names easily. | |
5942 | */ | |
27c3a314 | 5943 | sysfs_remove_link(&slab_kset->kobj, s->name); |
81819f0f CL |
5944 | name = s->name; |
5945 | } else { | |
5946 | /* | |
5947 | * Create a unique name for the slab as a target | |
5948 | * for the symlinks. | |
5949 | */ | |
5950 | name = create_unique_id(s); | |
5951 | } | |
5952 | ||
1663f26d | 5953 | s->kobj.kset = kset; |
26e4f205 | 5954 | err = kobject_init_and_add(&s->kobj, &slab_ktype, NULL, "%s", name); |
757fed1d | 5955 | if (err) |
80da026a | 5956 | goto out; |
81819f0f CL |
5957 | |
5958 | err = sysfs_create_group(&s->kobj, &slab_attr_group); | |
54b6a731 DJ |
5959 | if (err) |
5960 | goto out_del_kobj; | |
9a41707b | 5961 | |
81819f0f CL |
5962 | if (!unmergeable) { |
5963 | /* Setup first alias */ | |
5964 | sysfs_slab_alias(s, s->name); | |
81819f0f | 5965 | } |
54b6a731 DJ |
5966 | out: |
5967 | if (!unmergeable) | |
5968 | kfree(name); | |
5969 | return err; | |
5970 | out_del_kobj: | |
5971 | kobject_del(&s->kobj); | |
54b6a731 | 5972 | goto out; |
81819f0f CL |
5973 | } |
5974 | ||
d50d82fa MP |
5975 | void sysfs_slab_unlink(struct kmem_cache *s) |
5976 | { | |
5977 | if (slab_state >= FULL) | |
5978 | kobject_del(&s->kobj); | |
5979 | } | |
5980 | ||
bf5eb3de TH |
5981 | void sysfs_slab_release(struct kmem_cache *s) |
5982 | { | |
5983 | if (slab_state >= FULL) | |
5984 | kobject_put(&s->kobj); | |
81819f0f CL |
5985 | } |
5986 | ||
5987 | /* | |
5988 | * Need to buffer aliases during bootup until sysfs becomes | |
9f6c708e | 5989 | * available lest we lose that information. |
81819f0f CL |
5990 | */ |
5991 | struct saved_alias { | |
5992 | struct kmem_cache *s; | |
5993 | const char *name; | |
5994 | struct saved_alias *next; | |
5995 | }; | |
5996 | ||
5af328a5 | 5997 | static struct saved_alias *alias_list; |
81819f0f CL |
5998 | |
5999 | static int sysfs_slab_alias(struct kmem_cache *s, const char *name) | |
6000 | { | |
6001 | struct saved_alias *al; | |
6002 | ||
97d06609 | 6003 | if (slab_state == FULL) { |
81819f0f CL |
6004 | /* |
6005 | * If we have a leftover link then remove it. | |
6006 | */ | |
27c3a314 GKH |
6007 | sysfs_remove_link(&slab_kset->kobj, name); |
6008 | return sysfs_create_link(&slab_kset->kobj, &s->kobj, name); | |
81819f0f CL |
6009 | } |
6010 | ||
6011 | al = kmalloc(sizeof(struct saved_alias), GFP_KERNEL); | |
6012 | if (!al) | |
6013 | return -ENOMEM; | |
6014 | ||
6015 | al->s = s; | |
6016 | al->name = name; | |
6017 | al->next = alias_list; | |
6018 | alias_list = al; | |
6019 | return 0; | |
6020 | } | |
6021 | ||
6022 | static int __init slab_sysfs_init(void) | |
6023 | { | |
5b95a4ac | 6024 | struct kmem_cache *s; |
81819f0f CL |
6025 | int err; |
6026 | ||
18004c5d | 6027 | mutex_lock(&slab_mutex); |
2bce6485 | 6028 | |
d7660ce5 | 6029 | slab_kset = kset_create_and_add("slab", NULL, kernel_kobj); |
27c3a314 | 6030 | if (!slab_kset) { |
18004c5d | 6031 | mutex_unlock(&slab_mutex); |
f9f58285 | 6032 | pr_err("Cannot register slab subsystem.\n"); |
81819f0f CL |
6033 | return -ENOSYS; |
6034 | } | |
6035 | ||
97d06609 | 6036 | slab_state = FULL; |
26a7bd03 | 6037 | |
5b95a4ac | 6038 | list_for_each_entry(s, &slab_caches, list) { |
26a7bd03 | 6039 | err = sysfs_slab_add(s); |
5d540fb7 | 6040 | if (err) |
f9f58285 FF |
6041 | pr_err("SLUB: Unable to add boot slab %s to sysfs\n", |
6042 | s->name); | |
26a7bd03 | 6043 | } |
81819f0f CL |
6044 | |
6045 | while (alias_list) { | |
6046 | struct saved_alias *al = alias_list; | |
6047 | ||
6048 | alias_list = alias_list->next; | |
6049 | err = sysfs_slab_alias(al->s, al->name); | |
5d540fb7 | 6050 | if (err) |
f9f58285 FF |
6051 | pr_err("SLUB: Unable to add boot slab alias %s to sysfs\n", |
6052 | al->name); | |
81819f0f CL |
6053 | kfree(al); |
6054 | } | |
6055 | ||
18004c5d | 6056 | mutex_unlock(&slab_mutex); |
81819f0f CL |
6057 | return 0; |
6058 | } | |
6059 | ||
6060 | __initcall(slab_sysfs_init); | |
ab4d5ed5 | 6061 | #endif /* CONFIG_SYSFS */ |
57ed3eda | 6062 | |
64dd6849 FM |
6063 | #if defined(CONFIG_SLUB_DEBUG) && defined(CONFIG_DEBUG_FS) |
6064 | static int slab_debugfs_show(struct seq_file *seq, void *v) | |
6065 | { | |
64dd6849 | 6066 | struct loc_track *t = seq->private; |
005a79e5 GS |
6067 | struct location *l; |
6068 | unsigned long idx; | |
64dd6849 | 6069 | |
005a79e5 | 6070 | idx = (unsigned long) t->idx; |
64dd6849 FM |
6071 | if (idx < t->count) { |
6072 | l = &t->loc[idx]; | |
6073 | ||
6074 | seq_printf(seq, "%7ld ", l->count); | |
6075 | ||
6076 | if (l->addr) | |
6077 | seq_printf(seq, "%pS", (void *)l->addr); | |
6078 | else | |
6079 | seq_puts(seq, "<not-available>"); | |
6080 | ||
6081 | if (l->sum_time != l->min_time) { | |
6082 | seq_printf(seq, " age=%ld/%llu/%ld", | |
6083 | l->min_time, div_u64(l->sum_time, l->count), | |
6084 | l->max_time); | |
6085 | } else | |
6086 | seq_printf(seq, " age=%ld", l->min_time); | |
6087 | ||
6088 | if (l->min_pid != l->max_pid) | |
6089 | seq_printf(seq, " pid=%ld-%ld", l->min_pid, l->max_pid); | |
6090 | else | |
6091 | seq_printf(seq, " pid=%ld", | |
6092 | l->min_pid); | |
6093 | ||
6094 | if (num_online_cpus() > 1 && !cpumask_empty(to_cpumask(l->cpus))) | |
6095 | seq_printf(seq, " cpus=%*pbl", | |
6096 | cpumask_pr_args(to_cpumask(l->cpus))); | |
6097 | ||
6098 | if (nr_online_nodes > 1 && !nodes_empty(l->nodes)) | |
6099 | seq_printf(seq, " nodes=%*pbl", | |
6100 | nodemask_pr_args(&l->nodes)); | |
6101 | ||
6102 | seq_puts(seq, "\n"); | |
6103 | } | |
6104 | ||
6105 | if (!idx && !t->count) | |
6106 | seq_puts(seq, "No data\n"); | |
6107 | ||
6108 | return 0; | |
6109 | } | |
6110 | ||
6111 | static void slab_debugfs_stop(struct seq_file *seq, void *v) | |
6112 | { | |
6113 | } | |
6114 | ||
6115 | static void *slab_debugfs_next(struct seq_file *seq, void *v, loff_t *ppos) | |
6116 | { | |
6117 | struct loc_track *t = seq->private; | |
6118 | ||
005a79e5 | 6119 | t->idx = ++(*ppos); |
64dd6849 | 6120 | if (*ppos <= t->count) |
005a79e5 | 6121 | return ppos; |
64dd6849 FM |
6122 | |
6123 | return NULL; | |
6124 | } | |
6125 | ||
6126 | static void *slab_debugfs_start(struct seq_file *seq, loff_t *ppos) | |
6127 | { | |
005a79e5 GS |
6128 | struct loc_track *t = seq->private; |
6129 | ||
6130 | t->idx = *ppos; | |
64dd6849 FM |
6131 | return ppos; |
6132 | } | |
6133 | ||
6134 | static const struct seq_operations slab_debugfs_sops = { | |
6135 | .start = slab_debugfs_start, | |
6136 | .next = slab_debugfs_next, | |
6137 | .stop = slab_debugfs_stop, | |
6138 | .show = slab_debugfs_show, | |
6139 | }; | |
6140 | ||
6141 | static int slab_debug_trace_open(struct inode *inode, struct file *filep) | |
6142 | { | |
6143 | ||
6144 | struct kmem_cache_node *n; | |
6145 | enum track_item alloc; | |
6146 | int node; | |
6147 | struct loc_track *t = __seq_open_private(filep, &slab_debugfs_sops, | |
6148 | sizeof(struct loc_track)); | |
6149 | struct kmem_cache *s = file_inode(filep)->i_private; | |
b3fd64e1 VB |
6150 | unsigned long *obj_map; |
6151 | ||
2127d225 ML |
6152 | if (!t) |
6153 | return -ENOMEM; | |
6154 | ||
b3fd64e1 | 6155 | obj_map = bitmap_alloc(oo_objects(s->oo), GFP_KERNEL); |
2127d225 ML |
6156 | if (!obj_map) { |
6157 | seq_release_private(inode, filep); | |
b3fd64e1 | 6158 | return -ENOMEM; |
2127d225 | 6159 | } |
64dd6849 FM |
6160 | |
6161 | if (strcmp(filep->f_path.dentry->d_name.name, "alloc_traces") == 0) | |
6162 | alloc = TRACK_ALLOC; | |
6163 | else | |
6164 | alloc = TRACK_FREE; | |
6165 | ||
b3fd64e1 VB |
6166 | if (!alloc_loc_track(t, PAGE_SIZE / sizeof(struct location), GFP_KERNEL)) { |
6167 | bitmap_free(obj_map); | |
2127d225 | 6168 | seq_release_private(inode, filep); |
64dd6849 | 6169 | return -ENOMEM; |
b3fd64e1 | 6170 | } |
64dd6849 | 6171 | |
64dd6849 FM |
6172 | for_each_kmem_cache_node(s, node, n) { |
6173 | unsigned long flags; | |
6174 | struct page *page; | |
6175 | ||
6176 | if (!atomic_long_read(&n->nr_slabs)) | |
6177 | continue; | |
6178 | ||
6179 | spin_lock_irqsave(&n->list_lock, flags); | |
6180 | list_for_each_entry(page, &n->partial, slab_list) | |
b3fd64e1 | 6181 | process_slab(t, s, page, alloc, obj_map); |
64dd6849 | 6182 | list_for_each_entry(page, &n->full, slab_list) |
b3fd64e1 | 6183 | process_slab(t, s, page, alloc, obj_map); |
64dd6849 FM |
6184 | spin_unlock_irqrestore(&n->list_lock, flags); |
6185 | } | |
6186 | ||
b3fd64e1 | 6187 | bitmap_free(obj_map); |
64dd6849 FM |
6188 | return 0; |
6189 | } | |
6190 | ||
6191 | static int slab_debug_trace_release(struct inode *inode, struct file *file) | |
6192 | { | |
6193 | struct seq_file *seq = file->private_data; | |
6194 | struct loc_track *t = seq->private; | |
6195 | ||
6196 | free_loc_track(t); | |
6197 | return seq_release_private(inode, file); | |
6198 | } | |
6199 | ||
6200 | static const struct file_operations slab_debugfs_fops = { | |
6201 | .open = slab_debug_trace_open, | |
6202 | .read = seq_read, | |
6203 | .llseek = seq_lseek, | |
6204 | .release = slab_debug_trace_release, | |
6205 | }; | |
6206 | ||
6207 | static void debugfs_slab_add(struct kmem_cache *s) | |
6208 | { | |
6209 | struct dentry *slab_cache_dir; | |
6210 | ||
6211 | if (unlikely(!slab_debugfs_root)) | |
6212 | return; | |
6213 | ||
6214 | slab_cache_dir = debugfs_create_dir(s->name, slab_debugfs_root); | |
6215 | ||
6216 | debugfs_create_file("alloc_traces", 0400, | |
6217 | slab_cache_dir, s, &slab_debugfs_fops); | |
6218 | ||
6219 | debugfs_create_file("free_traces", 0400, | |
6220 | slab_cache_dir, s, &slab_debugfs_fops); | |
6221 | } | |
6222 | ||
6223 | void debugfs_slab_release(struct kmem_cache *s) | |
6224 | { | |
6225 | debugfs_remove_recursive(debugfs_lookup(s->name, slab_debugfs_root)); | |
6226 | } | |
6227 | ||
6228 | static int __init slab_debugfs_init(void) | |
6229 | { | |
6230 | struct kmem_cache *s; | |
6231 | ||
6232 | slab_debugfs_root = debugfs_create_dir("slab", NULL); | |
6233 | ||
6234 | list_for_each_entry(s, &slab_caches, list) | |
6235 | if (s->flags & SLAB_STORE_USER) | |
6236 | debugfs_slab_add(s); | |
6237 | ||
6238 | return 0; | |
6239 | ||
6240 | } | |
6241 | __initcall(slab_debugfs_init); | |
6242 | #endif | |
57ed3eda PE |
6243 | /* |
6244 | * The /proc/slabinfo ABI | |
6245 | */ | |
5b365771 | 6246 | #ifdef CONFIG_SLUB_DEBUG |
0d7561c6 | 6247 | void get_slabinfo(struct kmem_cache *s, struct slabinfo *sinfo) |
57ed3eda | 6248 | { |
57ed3eda | 6249 | unsigned long nr_slabs = 0; |
205ab99d CL |
6250 | unsigned long nr_objs = 0; |
6251 | unsigned long nr_free = 0; | |
57ed3eda | 6252 | int node; |
fa45dc25 | 6253 | struct kmem_cache_node *n; |
57ed3eda | 6254 | |
fa45dc25 | 6255 | for_each_kmem_cache_node(s, node, n) { |
c17fd13e WL |
6256 | nr_slabs += node_nr_slabs(n); |
6257 | nr_objs += node_nr_objs(n); | |
205ab99d | 6258 | nr_free += count_partial(n, count_free); |
57ed3eda PE |
6259 | } |
6260 | ||
0d7561c6 GC |
6261 | sinfo->active_objs = nr_objs - nr_free; |
6262 | sinfo->num_objs = nr_objs; | |
6263 | sinfo->active_slabs = nr_slabs; | |
6264 | sinfo->num_slabs = nr_slabs; | |
6265 | sinfo->objects_per_slab = oo_objects(s->oo); | |
6266 | sinfo->cache_order = oo_order(s->oo); | |
57ed3eda PE |
6267 | } |
6268 | ||
0d7561c6 | 6269 | void slabinfo_show_stats(struct seq_file *m, struct kmem_cache *s) |
7b3c3a50 | 6270 | { |
7b3c3a50 AD |
6271 | } |
6272 | ||
b7454ad3 GC |
6273 | ssize_t slabinfo_write(struct file *file, const char __user *buffer, |
6274 | size_t count, loff_t *ppos) | |
7b3c3a50 | 6275 | { |
b7454ad3 | 6276 | return -EIO; |
7b3c3a50 | 6277 | } |
5b365771 | 6278 | #endif /* CONFIG_SLUB_DEBUG */ |