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> | |
5cf909c5 | 29 | #include <linux/stackdepot.h> |
3ac7fe5a | 30 | #include <linux/debugobjects.h> |
81819f0f | 31 | #include <linux/kallsyms.h> |
b89fb5ef | 32 | #include <linux/kfence.h> |
b9049e23 | 33 | #include <linux/memory.h> |
f8bd2258 | 34 | #include <linux/math64.h> |
773ff60e | 35 | #include <linux/fault-inject.h> |
bfa71457 | 36 | #include <linux/stacktrace.h> |
4de900b4 | 37 | #include <linux/prefetch.h> |
2633d7a0 | 38 | #include <linux/memcontrol.h> |
2482ddec | 39 | #include <linux/random.h> |
1f9f78b1 | 40 | #include <kunit/test.h> |
553c0369 | 41 | #include <linux/sort.h> |
81819f0f | 42 | |
64dd6849 | 43 | #include <linux/debugfs.h> |
4a92379b RK |
44 | #include <trace/events/kmem.h> |
45 | ||
072bb0aa MG |
46 | #include "internal.h" |
47 | ||
81819f0f CL |
48 | /* |
49 | * Lock order: | |
18004c5d | 50 | * 1. slab_mutex (Global Mutex) |
bd0e7491 VB |
51 | * 2. node->list_lock (Spinlock) |
52 | * 3. kmem_cache->cpu_slab->lock (Local lock) | |
c2092c12 | 53 | * 4. slab_lock(slab) (Only on some arches or for debugging) |
bd0e7491 | 54 | * 5. object_map_lock (Only for debugging) |
81819f0f | 55 | * |
18004c5d | 56 | * slab_mutex |
881db7fb | 57 | * |
18004c5d | 58 | * The role of the slab_mutex is to protect the list of all the slabs |
881db7fb | 59 | * and to synchronize major metadata changes to slab cache structures. |
bd0e7491 VB |
60 | * Also synchronizes memory hotplug callbacks. |
61 | * | |
62 | * slab_lock | |
63 | * | |
64 | * The slab_lock is a wrapper around the page lock, thus it is a bit | |
65 | * spinlock. | |
881db7fb CL |
66 | * |
67 | * The slab_lock is only used for debugging and on arches that do not | |
b7ccc7f8 | 68 | * have the ability to do a cmpxchg_double. It only protects: |
c2092c12 VB |
69 | * A. slab->freelist -> List of free objects in a slab |
70 | * B. slab->inuse -> Number of objects in use | |
71 | * C. slab->objects -> Number of objects in slab | |
72 | * D. slab->frozen -> frozen state | |
881db7fb | 73 | * |
bd0e7491 VB |
74 | * Frozen slabs |
75 | * | |
881db7fb | 76 | * If a slab is frozen then it is exempt from list management. It is not |
632b2ef0 | 77 | * on any list except per cpu partial list. The processor that froze the |
c2092c12 | 78 | * slab is the one who can perform list operations on the slab. Other |
632b2ef0 LX |
79 | * processors may put objects onto the freelist but the processor that |
80 | * froze the slab is the only one that can retrieve the objects from the | |
c2092c12 | 81 | * slab's freelist. |
81819f0f | 82 | * |
bd0e7491 VB |
83 | * list_lock |
84 | * | |
81819f0f CL |
85 | * The list_lock protects the partial and full list on each node and |
86 | * the partial slab counter. If taken then no new slabs may be added or | |
87 | * removed from the lists nor make the number of partial slabs be modified. | |
88 | * (Note that the total number of slabs is an atomic value that may be | |
89 | * modified without taking the list lock). | |
90 | * | |
91 | * The list_lock is a centralized lock and thus we avoid taking it as | |
92 | * much as possible. As long as SLUB does not have to handle partial | |
93 | * slabs, operations can continue without any centralized lock. F.e. | |
94 | * allocating a long series of objects that fill up slabs does not require | |
95 | * the list lock. | |
bd0e7491 VB |
96 | * |
97 | * cpu_slab->lock local lock | |
98 | * | |
99 | * This locks protect slowpath manipulation of all kmem_cache_cpu fields | |
100 | * except the stat counters. This is a percpu structure manipulated only by | |
101 | * the local cpu, so the lock protects against being preempted or interrupted | |
102 | * by an irq. Fast path operations rely on lockless operations instead. | |
103 | * On PREEMPT_RT, the local lock does not actually disable irqs (and thus | |
104 | * prevent the lockless operations), so fastpath operations also need to take | |
105 | * the lock and are no longer lockless. | |
106 | * | |
107 | * lockless fastpaths | |
108 | * | |
109 | * The fast path allocation (slab_alloc_node()) and freeing (do_slab_free()) | |
110 | * are fully lockless when satisfied from the percpu slab (and when | |
111 | * cmpxchg_double is possible to use, otherwise slab_lock is taken). | |
112 | * They also don't disable preemption or migration or irqs. They rely on | |
113 | * the transaction id (tid) field to detect being preempted or moved to | |
114 | * another cpu. | |
115 | * | |
116 | * irq, preemption, migration considerations | |
117 | * | |
118 | * Interrupts are disabled as part of list_lock or local_lock operations, or | |
119 | * around the slab_lock operation, in order to make the slab allocator safe | |
120 | * to use in the context of an irq. | |
121 | * | |
122 | * In addition, preemption (or migration on PREEMPT_RT) is disabled in the | |
123 | * allocation slowpath, bulk allocation, and put_cpu_partial(), so that the | |
124 | * local cpu doesn't change in the process and e.g. the kmem_cache_cpu pointer | |
125 | * doesn't have to be revalidated in each section protected by the local lock. | |
81819f0f CL |
126 | * |
127 | * SLUB assigns one slab for allocation to each processor. | |
128 | * Allocations only occur from these slabs called cpu slabs. | |
129 | * | |
672bba3a CL |
130 | * Slabs with free elements are kept on a partial list and during regular |
131 | * operations no list for full slabs is used. If an object in a full slab is | |
81819f0f | 132 | * freed then the slab will show up again on the partial lists. |
672bba3a CL |
133 | * We track full slabs for debugging purposes though because otherwise we |
134 | * cannot scan all objects. | |
81819f0f CL |
135 | * |
136 | * Slabs are freed when they become empty. Teardown and setup is | |
137 | * minimal so we rely on the page allocators per cpu caches for | |
138 | * fast frees and allocs. | |
139 | * | |
c2092c12 | 140 | * slab->frozen The slab is frozen and exempt from list processing. |
4b6f0750 CL |
141 | * This means that the slab is dedicated to a purpose |
142 | * such as satisfying allocations for a specific | |
143 | * processor. Objects may be freed in the slab while | |
144 | * it is frozen but slab_free will then skip the usual | |
145 | * list operations. It is up to the processor holding | |
146 | * the slab to integrate the slab into the slab lists | |
147 | * when the slab is no longer needed. | |
148 | * | |
149 | * One use of this flag is to mark slabs that are | |
150 | * used for allocations. Then such a slab becomes a cpu | |
151 | * slab. The cpu slab may be equipped with an additional | |
dfb4f096 | 152 | * freelist that allows lockless access to |
894b8788 CL |
153 | * free objects in addition to the regular freelist |
154 | * that requires the slab lock. | |
81819f0f | 155 | * |
aed68148 | 156 | * SLAB_DEBUG_FLAGS Slab requires special handling due to debug |
81819f0f | 157 | * options set. This moves slab handling out of |
894b8788 | 158 | * the fast path and disables lockless freelists. |
81819f0f CL |
159 | */ |
160 | ||
25c00c50 VB |
161 | /* |
162 | * We could simply use migrate_disable()/enable() but as long as it's a | |
163 | * function call even on !PREEMPT_RT, use inline preempt_disable() there. | |
164 | */ | |
165 | #ifndef CONFIG_PREEMPT_RT | |
166 | #define slub_get_cpu_ptr(var) get_cpu_ptr(var) | |
167 | #define slub_put_cpu_ptr(var) put_cpu_ptr(var) | |
168 | #else | |
169 | #define slub_get_cpu_ptr(var) \ | |
170 | ({ \ | |
171 | migrate_disable(); \ | |
172 | this_cpu_ptr(var); \ | |
173 | }) | |
174 | #define slub_put_cpu_ptr(var) \ | |
175 | do { \ | |
176 | (void)(var); \ | |
177 | migrate_enable(); \ | |
178 | } while (0) | |
179 | #endif | |
180 | ||
ca0cab65 VB |
181 | #ifdef CONFIG_SLUB_DEBUG |
182 | #ifdef CONFIG_SLUB_DEBUG_ON | |
183 | DEFINE_STATIC_KEY_TRUE(slub_debug_enabled); | |
184 | #else | |
185 | DEFINE_STATIC_KEY_FALSE(slub_debug_enabled); | |
186 | #endif | |
79270291 | 187 | #endif /* CONFIG_SLUB_DEBUG */ |
ca0cab65 | 188 | |
59052e89 VB |
189 | static inline bool kmem_cache_debug(struct kmem_cache *s) |
190 | { | |
191 | return kmem_cache_debug_flags(s, SLAB_DEBUG_FLAGS); | |
af537b0a | 192 | } |
5577bd8a | 193 | |
117d54df | 194 | void *fixup_red_left(struct kmem_cache *s, void *p) |
d86bd1be | 195 | { |
59052e89 | 196 | if (kmem_cache_debug_flags(s, SLAB_RED_ZONE)) |
d86bd1be JK |
197 | p += s->red_left_pad; |
198 | ||
199 | return p; | |
200 | } | |
201 | ||
345c905d JK |
202 | static inline bool kmem_cache_has_cpu_partial(struct kmem_cache *s) |
203 | { | |
204 | #ifdef CONFIG_SLUB_CPU_PARTIAL | |
205 | return !kmem_cache_debug(s); | |
206 | #else | |
207 | return false; | |
208 | #endif | |
209 | } | |
210 | ||
81819f0f CL |
211 | /* |
212 | * Issues still to be resolved: | |
213 | * | |
81819f0f CL |
214 | * - Support PAGE_ALLOC_DEBUG. Should be easy to do. |
215 | * | |
81819f0f CL |
216 | * - Variable sizing of the per node arrays |
217 | */ | |
218 | ||
b789ef51 CL |
219 | /* Enable to log cmpxchg failures */ |
220 | #undef SLUB_DEBUG_CMPXCHG | |
221 | ||
2086d26a | 222 | /* |
dc84207d | 223 | * Minimum number of partial slabs. These will be left on the partial |
2086d26a CL |
224 | * lists even if they are empty. kmem_cache_shrink may reclaim them. |
225 | */ | |
76be8950 | 226 | #define MIN_PARTIAL 5 |
e95eed57 | 227 | |
2086d26a CL |
228 | /* |
229 | * Maximum number of desirable partial slabs. | |
230 | * The existence of more partial slabs makes kmem_cache_shrink | |
721ae22a | 231 | * sort the partial list by the number of objects in use. |
2086d26a CL |
232 | */ |
233 | #define MAX_PARTIAL 10 | |
234 | ||
becfda68 | 235 | #define DEBUG_DEFAULT_FLAGS (SLAB_CONSISTENCY_CHECKS | SLAB_RED_ZONE | \ |
81819f0f | 236 | SLAB_POISON | SLAB_STORE_USER) |
672bba3a | 237 | |
149daaf3 LA |
238 | /* |
239 | * These debug flags cannot use CMPXCHG because there might be consistency | |
240 | * issues when checking or reading debug information | |
241 | */ | |
242 | #define SLAB_NO_CMPXCHG (SLAB_CONSISTENCY_CHECKS | SLAB_STORE_USER | \ | |
243 | SLAB_TRACE) | |
244 | ||
245 | ||
fa5ec8a1 | 246 | /* |
3de47213 DR |
247 | * Debugging flags that require metadata to be stored in the slab. These get |
248 | * disabled when slub_debug=O is used and a cache's min order increases with | |
249 | * metadata. | |
fa5ec8a1 | 250 | */ |
3de47213 | 251 | #define DEBUG_METADATA_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER) |
fa5ec8a1 | 252 | |
210b5c06 CG |
253 | #define OO_SHIFT 16 |
254 | #define OO_MASK ((1 << OO_SHIFT) - 1) | |
c2092c12 | 255 | #define MAX_OBJS_PER_PAGE 32767 /* since slab.objects is u15 */ |
210b5c06 | 256 | |
81819f0f | 257 | /* Internal SLUB flags */ |
d50112ed | 258 | /* Poison object */ |
4fd0b46e | 259 | #define __OBJECT_POISON ((slab_flags_t __force)0x80000000U) |
d50112ed | 260 | /* Use cmpxchg_double */ |
4fd0b46e | 261 | #define __CMPXCHG_DOUBLE ((slab_flags_t __force)0x40000000U) |
81819f0f | 262 | |
02cbc874 CL |
263 | /* |
264 | * Tracking user of a slab. | |
265 | */ | |
d6543e39 | 266 | #define TRACK_ADDRS_COUNT 16 |
02cbc874 | 267 | struct track { |
ce71e27c | 268 | unsigned long addr; /* Called from address */ |
5cf909c5 OG |
269 | #ifdef CONFIG_STACKDEPOT |
270 | depot_stack_handle_t handle; | |
d6543e39 | 271 | #endif |
02cbc874 CL |
272 | int cpu; /* Was running on cpu */ |
273 | int pid; /* Pid context */ | |
274 | unsigned long when; /* When did the operation occur */ | |
275 | }; | |
276 | ||
277 | enum track_item { TRACK_ALLOC, TRACK_FREE }; | |
278 | ||
ab4d5ed5 | 279 | #ifdef CONFIG_SYSFS |
81819f0f CL |
280 | static int sysfs_slab_add(struct kmem_cache *); |
281 | static int sysfs_slab_alias(struct kmem_cache *, const char *); | |
81819f0f | 282 | #else |
0c710013 CL |
283 | static inline int sysfs_slab_add(struct kmem_cache *s) { return 0; } |
284 | static inline int sysfs_slab_alias(struct kmem_cache *s, const char *p) | |
285 | { return 0; } | |
81819f0f CL |
286 | #endif |
287 | ||
64dd6849 FM |
288 | #if defined(CONFIG_DEBUG_FS) && defined(CONFIG_SLUB_DEBUG) |
289 | static void debugfs_slab_add(struct kmem_cache *); | |
290 | #else | |
291 | static inline void debugfs_slab_add(struct kmem_cache *s) { } | |
292 | #endif | |
293 | ||
4fdccdfb | 294 | static inline void stat(const struct kmem_cache *s, enum stat_item si) |
8ff12cfc CL |
295 | { |
296 | #ifdef CONFIG_SLUB_STATS | |
88da03a6 CL |
297 | /* |
298 | * The rmw is racy on a preemptible kernel but this is acceptable, so | |
299 | * avoid this_cpu_add()'s irq-disable overhead. | |
300 | */ | |
301 | raw_cpu_inc(s->cpu_slab->stat[si]); | |
8ff12cfc CL |
302 | #endif |
303 | } | |
304 | ||
7e1fa93d VB |
305 | /* |
306 | * Tracks for which NUMA nodes we have kmem_cache_nodes allocated. | |
307 | * Corresponds to node_state[N_NORMAL_MEMORY], but can temporarily | |
308 | * differ during memory hotplug/hotremove operations. | |
309 | * Protected by slab_mutex. | |
310 | */ | |
311 | static nodemask_t slab_nodes; | |
312 | ||
81819f0f CL |
313 | /******************************************************************** |
314 | * Core slab cache functions | |
315 | *******************************************************************/ | |
316 | ||
2482ddec KC |
317 | /* |
318 | * Returns freelist pointer (ptr). With hardening, this is obfuscated | |
319 | * with an XOR of the address where the pointer is held and a per-cache | |
320 | * random number. | |
321 | */ | |
322 | static inline void *freelist_ptr(const struct kmem_cache *s, void *ptr, | |
323 | unsigned long ptr_addr) | |
324 | { | |
325 | #ifdef CONFIG_SLAB_FREELIST_HARDENED | |
d36a63a9 | 326 | /* |
aa1ef4d7 | 327 | * When CONFIG_KASAN_SW/HW_TAGS is enabled, ptr_addr might be tagged. |
d36a63a9 AK |
328 | * Normally, this doesn't cause any issues, as both set_freepointer() |
329 | * and get_freepointer() are called with a pointer with the same tag. | |
330 | * However, there are some issues with CONFIG_SLUB_DEBUG code. For | |
331 | * example, when __free_slub() iterates over objects in a cache, it | |
332 | * passes untagged pointers to check_object(). check_object() in turns | |
333 | * calls get_freepointer() with an untagged pointer, which causes the | |
334 | * freepointer to be restored incorrectly. | |
335 | */ | |
336 | return (void *)((unsigned long)ptr ^ s->random ^ | |
1ad53d9f | 337 | swab((unsigned long)kasan_reset_tag((void *)ptr_addr))); |
2482ddec KC |
338 | #else |
339 | return ptr; | |
340 | #endif | |
341 | } | |
342 | ||
343 | /* Returns the freelist pointer recorded at location ptr_addr. */ | |
344 | static inline void *freelist_dereference(const struct kmem_cache *s, | |
345 | void *ptr_addr) | |
346 | { | |
347 | return freelist_ptr(s, (void *)*(unsigned long *)(ptr_addr), | |
348 | (unsigned long)ptr_addr); | |
349 | } | |
350 | ||
7656c72b CL |
351 | static inline void *get_freepointer(struct kmem_cache *s, void *object) |
352 | { | |
aa1ef4d7 | 353 | object = kasan_reset_tag(object); |
2482ddec | 354 | return freelist_dereference(s, object + s->offset); |
7656c72b CL |
355 | } |
356 | ||
0ad9500e ED |
357 | static void prefetch_freepointer(const struct kmem_cache *s, void *object) |
358 | { | |
04b4b006 | 359 | prefetchw(object + s->offset); |
0ad9500e ED |
360 | } |
361 | ||
1393d9a1 CL |
362 | static inline void *get_freepointer_safe(struct kmem_cache *s, void *object) |
363 | { | |
2482ddec | 364 | unsigned long freepointer_addr; |
1393d9a1 CL |
365 | void *p; |
366 | ||
8e57f8ac | 367 | if (!debug_pagealloc_enabled_static()) |
922d566c JK |
368 | return get_freepointer(s, object); |
369 | ||
f70b0049 | 370 | object = kasan_reset_tag(object); |
2482ddec | 371 | freepointer_addr = (unsigned long)object + s->offset; |
fe557319 | 372 | copy_from_kernel_nofault(&p, (void **)freepointer_addr, sizeof(p)); |
2482ddec | 373 | return freelist_ptr(s, p, freepointer_addr); |
1393d9a1 CL |
374 | } |
375 | ||
7656c72b CL |
376 | static inline void set_freepointer(struct kmem_cache *s, void *object, void *fp) |
377 | { | |
2482ddec KC |
378 | unsigned long freeptr_addr = (unsigned long)object + s->offset; |
379 | ||
ce6fa91b AP |
380 | #ifdef CONFIG_SLAB_FREELIST_HARDENED |
381 | BUG_ON(object == fp); /* naive detection of double free or corruption */ | |
382 | #endif | |
383 | ||
aa1ef4d7 | 384 | freeptr_addr = (unsigned long)kasan_reset_tag((void *)freeptr_addr); |
2482ddec | 385 | *(void **)freeptr_addr = freelist_ptr(s, fp, freeptr_addr); |
7656c72b CL |
386 | } |
387 | ||
388 | /* Loop over all objects in a slab */ | |
224a88be | 389 | #define for_each_object(__p, __s, __addr, __objects) \ |
d86bd1be JK |
390 | for (__p = fixup_red_left(__s, __addr); \ |
391 | __p < (__addr) + (__objects) * (__s)->size; \ | |
392 | __p += (__s)->size) | |
7656c72b | 393 | |
9736d2a9 | 394 | static inline unsigned int order_objects(unsigned int order, unsigned int size) |
ab9a0f19 | 395 | { |
9736d2a9 | 396 | return ((unsigned int)PAGE_SIZE << order) / size; |
ab9a0f19 LJ |
397 | } |
398 | ||
19af27af | 399 | static inline struct kmem_cache_order_objects oo_make(unsigned int order, |
9736d2a9 | 400 | unsigned int size) |
834f3d11 CL |
401 | { |
402 | struct kmem_cache_order_objects x = { | |
9736d2a9 | 403 | (order << OO_SHIFT) + order_objects(order, size) |
834f3d11 CL |
404 | }; |
405 | ||
406 | return x; | |
407 | } | |
408 | ||
19af27af | 409 | static inline unsigned int oo_order(struct kmem_cache_order_objects x) |
834f3d11 | 410 | { |
210b5c06 | 411 | return x.x >> OO_SHIFT; |
834f3d11 CL |
412 | } |
413 | ||
19af27af | 414 | static inline unsigned int oo_objects(struct kmem_cache_order_objects x) |
834f3d11 | 415 | { |
210b5c06 | 416 | return x.x & OO_MASK; |
834f3d11 CL |
417 | } |
418 | ||
b47291ef VB |
419 | #ifdef CONFIG_SLUB_CPU_PARTIAL |
420 | static void slub_set_cpu_partial(struct kmem_cache *s, unsigned int nr_objects) | |
421 | { | |
bb192ed9 | 422 | unsigned int nr_slabs; |
b47291ef VB |
423 | |
424 | s->cpu_partial = nr_objects; | |
425 | ||
426 | /* | |
427 | * We take the number of objects but actually limit the number of | |
c2092c12 VB |
428 | * slabs on the per cpu partial list, in order to limit excessive |
429 | * growth of the list. For simplicity we assume that the slabs will | |
b47291ef VB |
430 | * be half-full. |
431 | */ | |
bb192ed9 VB |
432 | nr_slabs = DIV_ROUND_UP(nr_objects * 2, oo_objects(s->oo)); |
433 | s->cpu_partial_slabs = nr_slabs; | |
b47291ef VB |
434 | } |
435 | #else | |
436 | static inline void | |
437 | slub_set_cpu_partial(struct kmem_cache *s, unsigned int nr_objects) | |
438 | { | |
439 | } | |
440 | #endif /* CONFIG_SLUB_CPU_PARTIAL */ | |
441 | ||
881db7fb CL |
442 | /* |
443 | * Per slab locking using the pagelock | |
444 | */ | |
0393895b | 445 | static __always_inline void __slab_lock(struct slab *slab) |
881db7fb | 446 | { |
0393895b VB |
447 | struct page *page = slab_page(slab); |
448 | ||
48c935ad | 449 | VM_BUG_ON_PAGE(PageTail(page), page); |
881db7fb CL |
450 | bit_spin_lock(PG_locked, &page->flags); |
451 | } | |
452 | ||
0393895b | 453 | static __always_inline void __slab_unlock(struct slab *slab) |
881db7fb | 454 | { |
0393895b VB |
455 | struct page *page = slab_page(slab); |
456 | ||
48c935ad | 457 | VM_BUG_ON_PAGE(PageTail(page), page); |
881db7fb CL |
458 | __bit_spin_unlock(PG_locked, &page->flags); |
459 | } | |
460 | ||
bb192ed9 | 461 | static __always_inline void slab_lock(struct slab *slab, unsigned long *flags) |
a2b4ae8b VB |
462 | { |
463 | if (IS_ENABLED(CONFIG_PREEMPT_RT)) | |
464 | local_irq_save(*flags); | |
bb192ed9 | 465 | __slab_lock(slab); |
a2b4ae8b VB |
466 | } |
467 | ||
bb192ed9 | 468 | static __always_inline void slab_unlock(struct slab *slab, unsigned long *flags) |
a2b4ae8b | 469 | { |
bb192ed9 | 470 | __slab_unlock(slab); |
a2b4ae8b VB |
471 | if (IS_ENABLED(CONFIG_PREEMPT_RT)) |
472 | local_irq_restore(*flags); | |
473 | } | |
474 | ||
475 | /* | |
476 | * Interrupts must be disabled (for the fallback code to work right), typically | |
477 | * by an _irqsave() lock variant. Except on PREEMPT_RT where locks are different | |
478 | * so we disable interrupts as part of slab_[un]lock(). | |
479 | */ | |
bb192ed9 | 480 | static inline bool __cmpxchg_double_slab(struct kmem_cache *s, struct slab *slab, |
1d07171c CL |
481 | void *freelist_old, unsigned long counters_old, |
482 | void *freelist_new, unsigned long counters_new, | |
483 | const char *n) | |
484 | { | |
a2b4ae8b VB |
485 | if (!IS_ENABLED(CONFIG_PREEMPT_RT)) |
486 | lockdep_assert_irqs_disabled(); | |
2565409f HC |
487 | #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \ |
488 | defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE) | |
1d07171c | 489 | if (s->flags & __CMPXCHG_DOUBLE) { |
bb192ed9 | 490 | if (cmpxchg_double(&slab->freelist, &slab->counters, |
0aa9a13d DC |
491 | freelist_old, counters_old, |
492 | freelist_new, counters_new)) | |
6f6528a1 | 493 | return true; |
1d07171c CL |
494 | } else |
495 | #endif | |
496 | { | |
a2b4ae8b VB |
497 | /* init to 0 to prevent spurious warnings */ |
498 | unsigned long flags = 0; | |
499 | ||
bb192ed9 VB |
500 | slab_lock(slab, &flags); |
501 | if (slab->freelist == freelist_old && | |
502 | slab->counters == counters_old) { | |
503 | slab->freelist = freelist_new; | |
504 | slab->counters = counters_new; | |
505 | slab_unlock(slab, &flags); | |
6f6528a1 | 506 | return true; |
1d07171c | 507 | } |
bb192ed9 | 508 | slab_unlock(slab, &flags); |
1d07171c CL |
509 | } |
510 | ||
511 | cpu_relax(); | |
512 | stat(s, CMPXCHG_DOUBLE_FAIL); | |
513 | ||
514 | #ifdef SLUB_DEBUG_CMPXCHG | |
f9f58285 | 515 | pr_info("%s %s: cmpxchg double redo ", n, s->name); |
1d07171c CL |
516 | #endif |
517 | ||
6f6528a1 | 518 | return false; |
1d07171c CL |
519 | } |
520 | ||
bb192ed9 | 521 | static inline bool cmpxchg_double_slab(struct kmem_cache *s, struct slab *slab, |
b789ef51 CL |
522 | void *freelist_old, unsigned long counters_old, |
523 | void *freelist_new, unsigned long counters_new, | |
524 | const char *n) | |
525 | { | |
2565409f HC |
526 | #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \ |
527 | defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE) | |
b789ef51 | 528 | if (s->flags & __CMPXCHG_DOUBLE) { |
bb192ed9 | 529 | if (cmpxchg_double(&slab->freelist, &slab->counters, |
0aa9a13d DC |
530 | freelist_old, counters_old, |
531 | freelist_new, counters_new)) | |
6f6528a1 | 532 | return true; |
b789ef51 CL |
533 | } else |
534 | #endif | |
535 | { | |
1d07171c CL |
536 | unsigned long flags; |
537 | ||
538 | local_irq_save(flags); | |
bb192ed9 VB |
539 | __slab_lock(slab); |
540 | if (slab->freelist == freelist_old && | |
541 | slab->counters == counters_old) { | |
542 | slab->freelist = freelist_new; | |
543 | slab->counters = counters_new; | |
544 | __slab_unlock(slab); | |
1d07171c | 545 | local_irq_restore(flags); |
6f6528a1 | 546 | return true; |
b789ef51 | 547 | } |
bb192ed9 | 548 | __slab_unlock(slab); |
1d07171c | 549 | local_irq_restore(flags); |
b789ef51 CL |
550 | } |
551 | ||
552 | cpu_relax(); | |
553 | stat(s, CMPXCHG_DOUBLE_FAIL); | |
554 | ||
555 | #ifdef SLUB_DEBUG_CMPXCHG | |
f9f58285 | 556 | pr_info("%s %s: cmpxchg double redo ", n, s->name); |
b789ef51 CL |
557 | #endif |
558 | ||
6f6528a1 | 559 | return false; |
b789ef51 CL |
560 | } |
561 | ||
41ecc55b | 562 | #ifdef CONFIG_SLUB_DEBUG |
90e9f6a6 | 563 | static unsigned long object_map[BITS_TO_LONGS(MAX_OBJS_PER_PAGE)]; |
94ef0304 | 564 | static DEFINE_RAW_SPINLOCK(object_map_lock); |
90e9f6a6 | 565 | |
b3fd64e1 | 566 | static void __fill_map(unsigned long *obj_map, struct kmem_cache *s, |
bb192ed9 | 567 | struct slab *slab) |
b3fd64e1 | 568 | { |
bb192ed9 | 569 | void *addr = slab_address(slab); |
b3fd64e1 VB |
570 | void *p; |
571 | ||
bb192ed9 | 572 | bitmap_zero(obj_map, slab->objects); |
b3fd64e1 | 573 | |
bb192ed9 | 574 | for (p = slab->freelist; p; p = get_freepointer(s, p)) |
b3fd64e1 VB |
575 | set_bit(__obj_to_index(s, addr, p), obj_map); |
576 | } | |
577 | ||
1f9f78b1 OG |
578 | #if IS_ENABLED(CONFIG_KUNIT) |
579 | static bool slab_add_kunit_errors(void) | |
580 | { | |
581 | struct kunit_resource *resource; | |
582 | ||
583 | if (likely(!current->kunit_test)) | |
584 | return false; | |
585 | ||
586 | resource = kunit_find_named_resource(current->kunit_test, "slab_errors"); | |
587 | if (!resource) | |
588 | return false; | |
589 | ||
590 | (*(int *)resource->data)++; | |
591 | kunit_put_resource(resource); | |
592 | return true; | |
593 | } | |
594 | #else | |
595 | static inline bool slab_add_kunit_errors(void) { return false; } | |
596 | #endif | |
597 | ||
5f80b13a | 598 | /* |
c2092c12 | 599 | * Determine a map of objects in use in a slab. |
5f80b13a | 600 | * |
c2092c12 | 601 | * Node listlock must be held to guarantee that the slab does |
5f80b13a CL |
602 | * not vanish from under us. |
603 | */ | |
bb192ed9 | 604 | static unsigned long *get_map(struct kmem_cache *s, struct slab *slab) |
31364c2e | 605 | __acquires(&object_map_lock) |
5f80b13a | 606 | { |
90e9f6a6 YZ |
607 | VM_BUG_ON(!irqs_disabled()); |
608 | ||
94ef0304 | 609 | raw_spin_lock(&object_map_lock); |
90e9f6a6 | 610 | |
bb192ed9 | 611 | __fill_map(object_map, s, slab); |
90e9f6a6 YZ |
612 | |
613 | return object_map; | |
614 | } | |
615 | ||
81aba9e0 | 616 | static void put_map(unsigned long *map) __releases(&object_map_lock) |
90e9f6a6 YZ |
617 | { |
618 | VM_BUG_ON(map != object_map); | |
94ef0304 | 619 | raw_spin_unlock(&object_map_lock); |
5f80b13a CL |
620 | } |
621 | ||
870b1fbb | 622 | static inline unsigned int size_from_object(struct kmem_cache *s) |
d86bd1be JK |
623 | { |
624 | if (s->flags & SLAB_RED_ZONE) | |
625 | return s->size - s->red_left_pad; | |
626 | ||
627 | return s->size; | |
628 | } | |
629 | ||
630 | static inline void *restore_red_left(struct kmem_cache *s, void *p) | |
631 | { | |
632 | if (s->flags & SLAB_RED_ZONE) | |
633 | p -= s->red_left_pad; | |
634 | ||
635 | return p; | |
636 | } | |
637 | ||
41ecc55b CL |
638 | /* |
639 | * Debug settings: | |
640 | */ | |
89d3c87e | 641 | #if defined(CONFIG_SLUB_DEBUG_ON) |
d50112ed | 642 | static slab_flags_t slub_debug = DEBUG_DEFAULT_FLAGS; |
f0630fff | 643 | #else |
d50112ed | 644 | static slab_flags_t slub_debug; |
f0630fff | 645 | #endif |
41ecc55b | 646 | |
e17f1dfb | 647 | static char *slub_debug_string; |
fa5ec8a1 | 648 | static int disable_higher_order_debug; |
41ecc55b | 649 | |
a79316c6 AR |
650 | /* |
651 | * slub is about to manipulate internal object metadata. This memory lies | |
652 | * outside the range of the allocated object, so accessing it would normally | |
653 | * be reported by kasan as a bounds error. metadata_access_enable() is used | |
654 | * to tell kasan that these accesses are OK. | |
655 | */ | |
656 | static inline void metadata_access_enable(void) | |
657 | { | |
658 | kasan_disable_current(); | |
659 | } | |
660 | ||
661 | static inline void metadata_access_disable(void) | |
662 | { | |
663 | kasan_enable_current(); | |
664 | } | |
665 | ||
81819f0f CL |
666 | /* |
667 | * Object debugging | |
668 | */ | |
d86bd1be JK |
669 | |
670 | /* Verify that a pointer has an address that is valid within a slab page */ | |
671 | static inline int check_valid_pointer(struct kmem_cache *s, | |
bb192ed9 | 672 | struct slab *slab, void *object) |
d86bd1be JK |
673 | { |
674 | void *base; | |
675 | ||
676 | if (!object) | |
677 | return 1; | |
678 | ||
bb192ed9 | 679 | base = slab_address(slab); |
338cfaad | 680 | object = kasan_reset_tag(object); |
d86bd1be | 681 | object = restore_red_left(s, object); |
bb192ed9 | 682 | if (object < base || object >= base + slab->objects * s->size || |
d86bd1be JK |
683 | (object - base) % s->size) { |
684 | return 0; | |
685 | } | |
686 | ||
687 | return 1; | |
688 | } | |
689 | ||
aa2efd5e DT |
690 | static void print_section(char *level, char *text, u8 *addr, |
691 | unsigned int length) | |
81819f0f | 692 | { |
a79316c6 | 693 | metadata_access_enable(); |
340caf17 KYL |
694 | print_hex_dump(level, text, DUMP_PREFIX_ADDRESS, |
695 | 16, 1, kasan_reset_tag((void *)addr), length, 1); | |
a79316c6 | 696 | metadata_access_disable(); |
81819f0f CL |
697 | } |
698 | ||
cbfc35a4 WL |
699 | /* |
700 | * See comment in calculate_sizes(). | |
701 | */ | |
702 | static inline bool freeptr_outside_object(struct kmem_cache *s) | |
703 | { | |
704 | return s->offset >= s->inuse; | |
705 | } | |
706 | ||
707 | /* | |
708 | * Return offset of the end of info block which is inuse + free pointer if | |
709 | * not overlapping with object. | |
710 | */ | |
711 | static inline unsigned int get_info_end(struct kmem_cache *s) | |
712 | { | |
713 | if (freeptr_outside_object(s)) | |
714 | return s->inuse + sizeof(void *); | |
715 | else | |
716 | return s->inuse; | |
717 | } | |
718 | ||
81819f0f CL |
719 | static struct track *get_track(struct kmem_cache *s, void *object, |
720 | enum track_item alloc) | |
721 | { | |
722 | struct track *p; | |
723 | ||
cbfc35a4 | 724 | p = object + get_info_end(s); |
81819f0f | 725 | |
aa1ef4d7 | 726 | return kasan_reset_tag(p + alloc); |
81819f0f CL |
727 | } |
728 | ||
5cf909c5 | 729 | #ifdef CONFIG_STACKDEPOT |
c4cf6785 SAS |
730 | static noinline depot_stack_handle_t set_track_prepare(void) |
731 | { | |
732 | depot_stack_handle_t handle; | |
5cf909c5 | 733 | unsigned long entries[TRACK_ADDRS_COUNT]; |
0cd1a029 | 734 | unsigned int nr_entries; |
ae14c63a | 735 | |
5cf909c5 | 736 | nr_entries = stack_trace_save(entries, ARRAY_SIZE(entries), 3); |
c4cf6785 SAS |
737 | handle = stack_depot_save(entries, nr_entries, GFP_NOWAIT); |
738 | ||
739 | return handle; | |
740 | } | |
741 | #else | |
742 | static inline depot_stack_handle_t set_track_prepare(void) | |
743 | { | |
744 | return 0; | |
745 | } | |
d6543e39 | 746 | #endif |
5cf909c5 | 747 | |
c4cf6785 SAS |
748 | static void set_track_update(struct kmem_cache *s, void *object, |
749 | enum track_item alloc, unsigned long addr, | |
750 | depot_stack_handle_t handle) | |
751 | { | |
752 | struct track *p = get_track(s, object, alloc); | |
753 | ||
754 | #ifdef CONFIG_STACKDEPOT | |
755 | p->handle = handle; | |
756 | #endif | |
0cd1a029 VB |
757 | p->addr = addr; |
758 | p->cpu = smp_processor_id(); | |
759 | p->pid = current->pid; | |
760 | p->when = jiffies; | |
81819f0f CL |
761 | } |
762 | ||
c4cf6785 SAS |
763 | static __always_inline void set_track(struct kmem_cache *s, void *object, |
764 | enum track_item alloc, unsigned long addr) | |
765 | { | |
766 | depot_stack_handle_t handle = set_track_prepare(); | |
767 | ||
768 | set_track_update(s, object, alloc, addr, handle); | |
769 | } | |
770 | ||
81819f0f CL |
771 | static void init_tracking(struct kmem_cache *s, void *object) |
772 | { | |
0cd1a029 VB |
773 | struct track *p; |
774 | ||
24922684 CL |
775 | if (!(s->flags & SLAB_STORE_USER)) |
776 | return; | |
777 | ||
0cd1a029 VB |
778 | p = get_track(s, object, TRACK_ALLOC); |
779 | memset(p, 0, 2*sizeof(struct track)); | |
81819f0f CL |
780 | } |
781 | ||
86609d33 | 782 | static void print_track(const char *s, struct track *t, unsigned long pr_time) |
81819f0f | 783 | { |
5cf909c5 OG |
784 | depot_stack_handle_t handle __maybe_unused; |
785 | ||
81819f0f CL |
786 | if (!t->addr) |
787 | return; | |
788 | ||
96b94abc | 789 | pr_err("%s in %pS age=%lu cpu=%u pid=%d\n", |
86609d33 | 790 | s, (void *)t->addr, pr_time - t->when, t->cpu, t->pid); |
5cf909c5 OG |
791 | #ifdef CONFIG_STACKDEPOT |
792 | handle = READ_ONCE(t->handle); | |
793 | if (handle) | |
794 | stack_depot_print(handle); | |
795 | else | |
796 | pr_err("object allocation/free stack trace missing\n"); | |
d6543e39 | 797 | #endif |
24922684 CL |
798 | } |
799 | ||
e42f174e | 800 | void print_tracking(struct kmem_cache *s, void *object) |
24922684 | 801 | { |
86609d33 | 802 | unsigned long pr_time = jiffies; |
24922684 CL |
803 | if (!(s->flags & SLAB_STORE_USER)) |
804 | return; | |
805 | ||
86609d33 CP |
806 | print_track("Allocated", get_track(s, object, TRACK_ALLOC), pr_time); |
807 | print_track("Freed", get_track(s, object, TRACK_FREE), pr_time); | |
24922684 CL |
808 | } |
809 | ||
fb012e27 | 810 | static void print_slab_info(const struct slab *slab) |
24922684 | 811 | { |
fb012e27 | 812 | struct folio *folio = (struct folio *)slab_folio(slab); |
24922684 | 813 | |
fb012e27 MWO |
814 | pr_err("Slab 0x%p objects=%u used=%u fp=0x%p flags=%pGp\n", |
815 | slab, slab->objects, slab->inuse, slab->freelist, | |
816 | folio_flags(folio, 0)); | |
24922684 CL |
817 | } |
818 | ||
819 | static void slab_bug(struct kmem_cache *s, char *fmt, ...) | |
820 | { | |
ecc42fbe | 821 | struct va_format vaf; |
24922684 | 822 | va_list args; |
24922684 CL |
823 | |
824 | va_start(args, fmt); | |
ecc42fbe FF |
825 | vaf.fmt = fmt; |
826 | vaf.va = &args; | |
f9f58285 | 827 | pr_err("=============================================================================\n"); |
ecc42fbe | 828 | pr_err("BUG %s (%s): %pV\n", s->name, print_tainted(), &vaf); |
f9f58285 | 829 | pr_err("-----------------------------------------------------------------------------\n\n"); |
ecc42fbe | 830 | va_end(args); |
81819f0f CL |
831 | } |
832 | ||
582d1212 | 833 | __printf(2, 3) |
24922684 CL |
834 | static void slab_fix(struct kmem_cache *s, char *fmt, ...) |
835 | { | |
ecc42fbe | 836 | struct va_format vaf; |
24922684 | 837 | va_list args; |
24922684 | 838 | |
1f9f78b1 OG |
839 | if (slab_add_kunit_errors()) |
840 | return; | |
841 | ||
24922684 | 842 | va_start(args, fmt); |
ecc42fbe FF |
843 | vaf.fmt = fmt; |
844 | vaf.va = &args; | |
845 | pr_err("FIX %s: %pV\n", s->name, &vaf); | |
24922684 | 846 | va_end(args); |
24922684 CL |
847 | } |
848 | ||
bb192ed9 | 849 | static void print_trailer(struct kmem_cache *s, struct slab *slab, u8 *p) |
81819f0f CL |
850 | { |
851 | unsigned int off; /* Offset of last byte */ | |
bb192ed9 | 852 | u8 *addr = slab_address(slab); |
24922684 CL |
853 | |
854 | print_tracking(s, p); | |
855 | ||
bb192ed9 | 856 | print_slab_info(slab); |
24922684 | 857 | |
96b94abc | 858 | pr_err("Object 0x%p @offset=%tu fp=0x%p\n\n", |
f9f58285 | 859 | p, p - addr, get_freepointer(s, p)); |
24922684 | 860 | |
d86bd1be | 861 | if (s->flags & SLAB_RED_ZONE) |
8669dbab | 862 | print_section(KERN_ERR, "Redzone ", p - s->red_left_pad, |
aa2efd5e | 863 | s->red_left_pad); |
d86bd1be | 864 | else if (p > addr + 16) |
aa2efd5e | 865 | print_section(KERN_ERR, "Bytes b4 ", p - 16, 16); |
81819f0f | 866 | |
8669dbab | 867 | print_section(KERN_ERR, "Object ", p, |
1b473f29 | 868 | min_t(unsigned int, s->object_size, PAGE_SIZE)); |
81819f0f | 869 | if (s->flags & SLAB_RED_ZONE) |
8669dbab | 870 | print_section(KERN_ERR, "Redzone ", p + s->object_size, |
3b0efdfa | 871 | s->inuse - s->object_size); |
81819f0f | 872 | |
cbfc35a4 | 873 | off = get_info_end(s); |
81819f0f | 874 | |
24922684 | 875 | if (s->flags & SLAB_STORE_USER) |
81819f0f | 876 | off += 2 * sizeof(struct track); |
81819f0f | 877 | |
80a9201a AP |
878 | off += kasan_metadata_size(s); |
879 | ||
d86bd1be | 880 | if (off != size_from_object(s)) |
81819f0f | 881 | /* Beginning of the filler is the free pointer */ |
8669dbab | 882 | print_section(KERN_ERR, "Padding ", p + off, |
aa2efd5e | 883 | size_from_object(s) - off); |
24922684 CL |
884 | |
885 | dump_stack(); | |
81819f0f CL |
886 | } |
887 | ||
bb192ed9 | 888 | static void object_err(struct kmem_cache *s, struct slab *slab, |
81819f0f CL |
889 | u8 *object, char *reason) |
890 | { | |
1f9f78b1 OG |
891 | if (slab_add_kunit_errors()) |
892 | return; | |
893 | ||
3dc50637 | 894 | slab_bug(s, "%s", reason); |
bb192ed9 | 895 | print_trailer(s, slab, object); |
65ebdeef | 896 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); |
81819f0f CL |
897 | } |
898 | ||
bb192ed9 | 899 | static bool freelist_corrupted(struct kmem_cache *s, struct slab *slab, |
ae16d059 VB |
900 | void **freelist, void *nextfree) |
901 | { | |
902 | if ((s->flags & SLAB_CONSISTENCY_CHECKS) && | |
bb192ed9 VB |
903 | !check_valid_pointer(s, slab, nextfree) && freelist) { |
904 | object_err(s, slab, *freelist, "Freechain corrupt"); | |
ae16d059 VB |
905 | *freelist = NULL; |
906 | slab_fix(s, "Isolate corrupted freechain"); | |
907 | return true; | |
908 | } | |
909 | ||
910 | return false; | |
911 | } | |
912 | ||
bb192ed9 | 913 | static __printf(3, 4) void slab_err(struct kmem_cache *s, struct slab *slab, |
d0e0ac97 | 914 | const char *fmt, ...) |
81819f0f CL |
915 | { |
916 | va_list args; | |
917 | char buf[100]; | |
918 | ||
1f9f78b1 OG |
919 | if (slab_add_kunit_errors()) |
920 | return; | |
921 | ||
24922684 CL |
922 | va_start(args, fmt); |
923 | vsnprintf(buf, sizeof(buf), fmt, args); | |
81819f0f | 924 | va_end(args); |
3dc50637 | 925 | slab_bug(s, "%s", buf); |
bb192ed9 | 926 | print_slab_info(slab); |
81819f0f | 927 | dump_stack(); |
65ebdeef | 928 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); |
81819f0f CL |
929 | } |
930 | ||
f7cb1933 | 931 | static void init_object(struct kmem_cache *s, void *object, u8 val) |
81819f0f | 932 | { |
aa1ef4d7 | 933 | u8 *p = kasan_reset_tag(object); |
81819f0f | 934 | |
d86bd1be JK |
935 | if (s->flags & SLAB_RED_ZONE) |
936 | memset(p - s->red_left_pad, val, s->red_left_pad); | |
937 | ||
81819f0f | 938 | if (s->flags & __OBJECT_POISON) { |
3b0efdfa CL |
939 | memset(p, POISON_FREE, s->object_size - 1); |
940 | p[s->object_size - 1] = POISON_END; | |
81819f0f CL |
941 | } |
942 | ||
943 | if (s->flags & SLAB_RED_ZONE) | |
3b0efdfa | 944 | memset(p + s->object_size, val, s->inuse - s->object_size); |
81819f0f CL |
945 | } |
946 | ||
24922684 CL |
947 | static void restore_bytes(struct kmem_cache *s, char *message, u8 data, |
948 | void *from, void *to) | |
949 | { | |
582d1212 | 950 | slab_fix(s, "Restoring %s 0x%p-0x%p=0x%x", message, from, to - 1, data); |
24922684 CL |
951 | memset(from, data, to - from); |
952 | } | |
953 | ||
bb192ed9 | 954 | static int check_bytes_and_report(struct kmem_cache *s, struct slab *slab, |
24922684 | 955 | u8 *object, char *what, |
06428780 | 956 | u8 *start, unsigned int value, unsigned int bytes) |
24922684 CL |
957 | { |
958 | u8 *fault; | |
959 | u8 *end; | |
bb192ed9 | 960 | u8 *addr = slab_address(slab); |
24922684 | 961 | |
a79316c6 | 962 | metadata_access_enable(); |
aa1ef4d7 | 963 | fault = memchr_inv(kasan_reset_tag(start), value, bytes); |
a79316c6 | 964 | metadata_access_disable(); |
24922684 CL |
965 | if (!fault) |
966 | return 1; | |
967 | ||
968 | end = start + bytes; | |
969 | while (end > fault && end[-1] == value) | |
970 | end--; | |
971 | ||
1f9f78b1 OG |
972 | if (slab_add_kunit_errors()) |
973 | goto skip_bug_print; | |
974 | ||
24922684 | 975 | slab_bug(s, "%s overwritten", what); |
96b94abc | 976 | pr_err("0x%p-0x%p @offset=%tu. First byte 0x%x instead of 0x%x\n", |
e1b70dd1 MC |
977 | fault, end - 1, fault - addr, |
978 | fault[0], value); | |
bb192ed9 | 979 | print_trailer(s, slab, object); |
65ebdeef | 980 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); |
24922684 | 981 | |
1f9f78b1 | 982 | skip_bug_print: |
24922684 CL |
983 | restore_bytes(s, what, value, fault, end); |
984 | return 0; | |
81819f0f CL |
985 | } |
986 | ||
81819f0f CL |
987 | /* |
988 | * Object layout: | |
989 | * | |
990 | * object address | |
991 | * Bytes of the object to be managed. | |
992 | * If the freepointer may overlay the object then the free | |
cbfc35a4 | 993 | * pointer is at the middle of the object. |
672bba3a | 994 | * |
81819f0f CL |
995 | * Poisoning uses 0x6b (POISON_FREE) and the last byte is |
996 | * 0xa5 (POISON_END) | |
997 | * | |
3b0efdfa | 998 | * object + s->object_size |
81819f0f | 999 | * Padding to reach word boundary. This is also used for Redzoning. |
672bba3a | 1000 | * Padding is extended by another word if Redzoning is enabled and |
3b0efdfa | 1001 | * object_size == inuse. |
672bba3a | 1002 | * |
81819f0f CL |
1003 | * We fill with 0xbb (RED_INACTIVE) for inactive objects and with |
1004 | * 0xcc (RED_ACTIVE) for objects in use. | |
1005 | * | |
1006 | * object + s->inuse | |
672bba3a CL |
1007 | * Meta data starts here. |
1008 | * | |
81819f0f CL |
1009 | * A. Free pointer (if we cannot overwrite object on free) |
1010 | * B. Tracking data for SLAB_STORE_USER | |
dc84207d | 1011 | * C. Padding to reach required alignment boundary or at minimum |
6446faa2 | 1012 | * one word if debugging is on to be able to detect writes |
672bba3a CL |
1013 | * before the word boundary. |
1014 | * | |
1015 | * Padding is done using 0x5a (POISON_INUSE) | |
81819f0f CL |
1016 | * |
1017 | * object + s->size | |
672bba3a | 1018 | * Nothing is used beyond s->size. |
81819f0f | 1019 | * |
3b0efdfa | 1020 | * If slabcaches are merged then the object_size and inuse boundaries are mostly |
672bba3a | 1021 | * ignored. And therefore no slab options that rely on these boundaries |
81819f0f CL |
1022 | * may be used with merged slabcaches. |
1023 | */ | |
1024 | ||
bb192ed9 | 1025 | static int check_pad_bytes(struct kmem_cache *s, struct slab *slab, u8 *p) |
81819f0f | 1026 | { |
cbfc35a4 | 1027 | unsigned long off = get_info_end(s); /* The end of info */ |
81819f0f CL |
1028 | |
1029 | if (s->flags & SLAB_STORE_USER) | |
1030 | /* We also have user information there */ | |
1031 | off += 2 * sizeof(struct track); | |
1032 | ||
80a9201a AP |
1033 | off += kasan_metadata_size(s); |
1034 | ||
d86bd1be | 1035 | if (size_from_object(s) == off) |
81819f0f CL |
1036 | return 1; |
1037 | ||
bb192ed9 | 1038 | return check_bytes_and_report(s, slab, p, "Object padding", |
d86bd1be | 1039 | p + off, POISON_INUSE, size_from_object(s) - off); |
81819f0f CL |
1040 | } |
1041 | ||
39b26464 | 1042 | /* Check the pad bytes at the end of a slab page */ |
a204e6d6 | 1043 | static void slab_pad_check(struct kmem_cache *s, struct slab *slab) |
81819f0f | 1044 | { |
24922684 CL |
1045 | u8 *start; |
1046 | u8 *fault; | |
1047 | u8 *end; | |
5d682681 | 1048 | u8 *pad; |
24922684 CL |
1049 | int length; |
1050 | int remainder; | |
81819f0f CL |
1051 | |
1052 | if (!(s->flags & SLAB_POISON)) | |
a204e6d6 | 1053 | return; |
81819f0f | 1054 | |
bb192ed9 VB |
1055 | start = slab_address(slab); |
1056 | length = slab_size(slab); | |
39b26464 CL |
1057 | end = start + length; |
1058 | remainder = length % s->size; | |
81819f0f | 1059 | if (!remainder) |
a204e6d6 | 1060 | return; |
81819f0f | 1061 | |
5d682681 | 1062 | pad = end - remainder; |
a79316c6 | 1063 | metadata_access_enable(); |
aa1ef4d7 | 1064 | fault = memchr_inv(kasan_reset_tag(pad), POISON_INUSE, remainder); |
a79316c6 | 1065 | metadata_access_disable(); |
24922684 | 1066 | if (!fault) |
a204e6d6 | 1067 | return; |
24922684 CL |
1068 | while (end > fault && end[-1] == POISON_INUSE) |
1069 | end--; | |
1070 | ||
bb192ed9 | 1071 | slab_err(s, slab, "Padding overwritten. 0x%p-0x%p @offset=%tu", |
e1b70dd1 | 1072 | fault, end - 1, fault - start); |
5d682681 | 1073 | print_section(KERN_ERR, "Padding ", pad, remainder); |
24922684 | 1074 | |
5d682681 | 1075 | restore_bytes(s, "slab padding", POISON_INUSE, fault, end); |
81819f0f CL |
1076 | } |
1077 | ||
bb192ed9 | 1078 | static int check_object(struct kmem_cache *s, struct slab *slab, |
f7cb1933 | 1079 | void *object, u8 val) |
81819f0f CL |
1080 | { |
1081 | u8 *p = object; | |
3b0efdfa | 1082 | u8 *endobject = object + s->object_size; |
81819f0f CL |
1083 | |
1084 | if (s->flags & SLAB_RED_ZONE) { | |
bb192ed9 | 1085 | if (!check_bytes_and_report(s, slab, object, "Left Redzone", |
d86bd1be JK |
1086 | object - s->red_left_pad, val, s->red_left_pad)) |
1087 | return 0; | |
1088 | ||
bb192ed9 | 1089 | if (!check_bytes_and_report(s, slab, object, "Right Redzone", |
3b0efdfa | 1090 | endobject, val, s->inuse - s->object_size)) |
81819f0f | 1091 | return 0; |
81819f0f | 1092 | } else { |
3b0efdfa | 1093 | if ((s->flags & SLAB_POISON) && s->object_size < s->inuse) { |
bb192ed9 | 1094 | check_bytes_and_report(s, slab, p, "Alignment padding", |
d0e0ac97 CG |
1095 | endobject, POISON_INUSE, |
1096 | s->inuse - s->object_size); | |
3adbefee | 1097 | } |
81819f0f CL |
1098 | } |
1099 | ||
1100 | if (s->flags & SLAB_POISON) { | |
f7cb1933 | 1101 | if (val != SLUB_RED_ACTIVE && (s->flags & __OBJECT_POISON) && |
bb192ed9 | 1102 | (!check_bytes_and_report(s, slab, p, "Poison", p, |
3b0efdfa | 1103 | POISON_FREE, s->object_size - 1) || |
bb192ed9 | 1104 | !check_bytes_and_report(s, slab, p, "End Poison", |
3b0efdfa | 1105 | p + s->object_size - 1, POISON_END, 1))) |
81819f0f | 1106 | return 0; |
81819f0f CL |
1107 | /* |
1108 | * check_pad_bytes cleans up on its own. | |
1109 | */ | |
bb192ed9 | 1110 | check_pad_bytes(s, slab, p); |
81819f0f CL |
1111 | } |
1112 | ||
cbfc35a4 | 1113 | if (!freeptr_outside_object(s) && val == SLUB_RED_ACTIVE) |
81819f0f CL |
1114 | /* |
1115 | * Object and freepointer overlap. Cannot check | |
1116 | * freepointer while object is allocated. | |
1117 | */ | |
1118 | return 1; | |
1119 | ||
1120 | /* Check free pointer validity */ | |
bb192ed9 VB |
1121 | if (!check_valid_pointer(s, slab, get_freepointer(s, p))) { |
1122 | object_err(s, slab, p, "Freepointer corrupt"); | |
81819f0f | 1123 | /* |
9f6c708e | 1124 | * No choice but to zap it and thus lose the remainder |
81819f0f | 1125 | * of the free objects in this slab. May cause |
672bba3a | 1126 | * another error because the object count is now wrong. |
81819f0f | 1127 | */ |
a973e9dd | 1128 | set_freepointer(s, p, NULL); |
81819f0f CL |
1129 | return 0; |
1130 | } | |
1131 | return 1; | |
1132 | } | |
1133 | ||
bb192ed9 | 1134 | static int check_slab(struct kmem_cache *s, struct slab *slab) |
81819f0f | 1135 | { |
39b26464 CL |
1136 | int maxobj; |
1137 | ||
bb192ed9 VB |
1138 | if (!folio_test_slab(slab_folio(slab))) { |
1139 | slab_err(s, slab, "Not a valid slab page"); | |
81819f0f CL |
1140 | return 0; |
1141 | } | |
39b26464 | 1142 | |
bb192ed9 VB |
1143 | maxobj = order_objects(slab_order(slab), s->size); |
1144 | if (slab->objects > maxobj) { | |
1145 | slab_err(s, slab, "objects %u > max %u", | |
1146 | slab->objects, maxobj); | |
39b26464 CL |
1147 | return 0; |
1148 | } | |
bb192ed9 VB |
1149 | if (slab->inuse > slab->objects) { |
1150 | slab_err(s, slab, "inuse %u > max %u", | |
1151 | slab->inuse, slab->objects); | |
81819f0f CL |
1152 | return 0; |
1153 | } | |
1154 | /* Slab_pad_check fixes things up after itself */ | |
bb192ed9 | 1155 | slab_pad_check(s, slab); |
81819f0f CL |
1156 | return 1; |
1157 | } | |
1158 | ||
1159 | /* | |
c2092c12 | 1160 | * Determine if a certain object in a slab is on the freelist. Must hold the |
672bba3a | 1161 | * slab lock to guarantee that the chains are in a consistent state. |
81819f0f | 1162 | */ |
bb192ed9 | 1163 | static int on_freelist(struct kmem_cache *s, struct slab *slab, void *search) |
81819f0f CL |
1164 | { |
1165 | int nr = 0; | |
881db7fb | 1166 | void *fp; |
81819f0f | 1167 | void *object = NULL; |
f6edde9c | 1168 | int max_objects; |
81819f0f | 1169 | |
bb192ed9 VB |
1170 | fp = slab->freelist; |
1171 | while (fp && nr <= slab->objects) { | |
81819f0f CL |
1172 | if (fp == search) |
1173 | return 1; | |
bb192ed9 | 1174 | if (!check_valid_pointer(s, slab, fp)) { |
81819f0f | 1175 | if (object) { |
bb192ed9 | 1176 | object_err(s, slab, object, |
81819f0f | 1177 | "Freechain corrupt"); |
a973e9dd | 1178 | set_freepointer(s, object, NULL); |
81819f0f | 1179 | } else { |
bb192ed9 VB |
1180 | slab_err(s, slab, "Freepointer corrupt"); |
1181 | slab->freelist = NULL; | |
1182 | slab->inuse = slab->objects; | |
24922684 | 1183 | slab_fix(s, "Freelist cleared"); |
81819f0f CL |
1184 | return 0; |
1185 | } | |
1186 | break; | |
1187 | } | |
1188 | object = fp; | |
1189 | fp = get_freepointer(s, object); | |
1190 | nr++; | |
1191 | } | |
1192 | ||
bb192ed9 | 1193 | max_objects = order_objects(slab_order(slab), s->size); |
210b5c06 CG |
1194 | if (max_objects > MAX_OBJS_PER_PAGE) |
1195 | max_objects = MAX_OBJS_PER_PAGE; | |
224a88be | 1196 | |
bb192ed9 VB |
1197 | if (slab->objects != max_objects) { |
1198 | slab_err(s, slab, "Wrong number of objects. Found %d but should be %d", | |
1199 | slab->objects, max_objects); | |
1200 | slab->objects = max_objects; | |
582d1212 | 1201 | slab_fix(s, "Number of objects adjusted"); |
224a88be | 1202 | } |
bb192ed9 VB |
1203 | if (slab->inuse != slab->objects - nr) { |
1204 | slab_err(s, slab, "Wrong object count. Counter is %d but counted were %d", | |
1205 | slab->inuse, slab->objects - nr); | |
1206 | slab->inuse = slab->objects - nr; | |
582d1212 | 1207 | slab_fix(s, "Object count adjusted"); |
81819f0f CL |
1208 | } |
1209 | return search == NULL; | |
1210 | } | |
1211 | ||
bb192ed9 | 1212 | static void trace(struct kmem_cache *s, struct slab *slab, void *object, |
0121c619 | 1213 | int alloc) |
3ec09742 CL |
1214 | { |
1215 | if (s->flags & SLAB_TRACE) { | |
f9f58285 | 1216 | pr_info("TRACE %s %s 0x%p inuse=%d fp=0x%p\n", |
3ec09742 CL |
1217 | s->name, |
1218 | alloc ? "alloc" : "free", | |
bb192ed9 VB |
1219 | object, slab->inuse, |
1220 | slab->freelist); | |
3ec09742 CL |
1221 | |
1222 | if (!alloc) | |
aa2efd5e | 1223 | print_section(KERN_INFO, "Object ", (void *)object, |
d0e0ac97 | 1224 | s->object_size); |
3ec09742 CL |
1225 | |
1226 | dump_stack(); | |
1227 | } | |
1228 | } | |
1229 | ||
643b1138 | 1230 | /* |
672bba3a | 1231 | * Tracking of fully allocated slabs for debugging purposes. |
643b1138 | 1232 | */ |
5cc6eee8 | 1233 | static void add_full(struct kmem_cache *s, |
bb192ed9 | 1234 | struct kmem_cache_node *n, struct slab *slab) |
643b1138 | 1235 | { |
5cc6eee8 CL |
1236 | if (!(s->flags & SLAB_STORE_USER)) |
1237 | return; | |
1238 | ||
255d0884 | 1239 | lockdep_assert_held(&n->list_lock); |
bb192ed9 | 1240 | list_add(&slab->slab_list, &n->full); |
643b1138 CL |
1241 | } |
1242 | ||
bb192ed9 | 1243 | static void remove_full(struct kmem_cache *s, struct kmem_cache_node *n, struct slab *slab) |
643b1138 | 1244 | { |
643b1138 CL |
1245 | if (!(s->flags & SLAB_STORE_USER)) |
1246 | return; | |
1247 | ||
255d0884 | 1248 | lockdep_assert_held(&n->list_lock); |
bb192ed9 | 1249 | list_del(&slab->slab_list); |
643b1138 CL |
1250 | } |
1251 | ||
0f389ec6 CL |
1252 | /* Tracking of the number of slabs for debugging purposes */ |
1253 | static inline unsigned long slabs_node(struct kmem_cache *s, int node) | |
1254 | { | |
1255 | struct kmem_cache_node *n = get_node(s, node); | |
1256 | ||
1257 | return atomic_long_read(&n->nr_slabs); | |
1258 | } | |
1259 | ||
26c02cf0 AB |
1260 | static inline unsigned long node_nr_slabs(struct kmem_cache_node *n) |
1261 | { | |
1262 | return atomic_long_read(&n->nr_slabs); | |
1263 | } | |
1264 | ||
205ab99d | 1265 | static inline void inc_slabs_node(struct kmem_cache *s, int node, int objects) |
0f389ec6 CL |
1266 | { |
1267 | struct kmem_cache_node *n = get_node(s, node); | |
1268 | ||
1269 | /* | |
1270 | * May be called early in order to allocate a slab for the | |
1271 | * kmem_cache_node structure. Solve the chicken-egg | |
1272 | * dilemma by deferring the increment of the count during | |
1273 | * bootstrap (see early_kmem_cache_node_alloc). | |
1274 | */ | |
338b2642 | 1275 | if (likely(n)) { |
0f389ec6 | 1276 | atomic_long_inc(&n->nr_slabs); |
205ab99d CL |
1277 | atomic_long_add(objects, &n->total_objects); |
1278 | } | |
0f389ec6 | 1279 | } |
205ab99d | 1280 | static inline void dec_slabs_node(struct kmem_cache *s, int node, int objects) |
0f389ec6 CL |
1281 | { |
1282 | struct kmem_cache_node *n = get_node(s, node); | |
1283 | ||
1284 | atomic_long_dec(&n->nr_slabs); | |
205ab99d | 1285 | atomic_long_sub(objects, &n->total_objects); |
0f389ec6 CL |
1286 | } |
1287 | ||
1288 | /* Object debug checks for alloc/free paths */ | |
c0f81a94 | 1289 | static void setup_object_debug(struct kmem_cache *s, void *object) |
3ec09742 | 1290 | { |
8fc8d666 | 1291 | if (!kmem_cache_debug_flags(s, SLAB_STORE_USER|SLAB_RED_ZONE|__OBJECT_POISON)) |
3ec09742 CL |
1292 | return; |
1293 | ||
f7cb1933 | 1294 | init_object(s, object, SLUB_RED_INACTIVE); |
3ec09742 CL |
1295 | init_tracking(s, object); |
1296 | } | |
1297 | ||
a50b854e | 1298 | static |
bb192ed9 | 1299 | void setup_slab_debug(struct kmem_cache *s, struct slab *slab, void *addr) |
a7101224 | 1300 | { |
8fc8d666 | 1301 | if (!kmem_cache_debug_flags(s, SLAB_POISON)) |
a7101224 AK |
1302 | return; |
1303 | ||
1304 | metadata_access_enable(); | |
bb192ed9 | 1305 | memset(kasan_reset_tag(addr), POISON_INUSE, slab_size(slab)); |
a7101224 AK |
1306 | metadata_access_disable(); |
1307 | } | |
1308 | ||
becfda68 | 1309 | static inline int alloc_consistency_checks(struct kmem_cache *s, |
bb192ed9 | 1310 | struct slab *slab, void *object) |
81819f0f | 1311 | { |
bb192ed9 | 1312 | if (!check_slab(s, slab)) |
becfda68 | 1313 | return 0; |
81819f0f | 1314 | |
bb192ed9 VB |
1315 | if (!check_valid_pointer(s, slab, object)) { |
1316 | object_err(s, slab, object, "Freelist Pointer check fails"); | |
becfda68 | 1317 | return 0; |
81819f0f CL |
1318 | } |
1319 | ||
bb192ed9 | 1320 | if (!check_object(s, slab, object, SLUB_RED_INACTIVE)) |
becfda68 LA |
1321 | return 0; |
1322 | ||
1323 | return 1; | |
1324 | } | |
1325 | ||
1326 | static noinline int alloc_debug_processing(struct kmem_cache *s, | |
bb192ed9 | 1327 | struct slab *slab, |
becfda68 LA |
1328 | void *object, unsigned long addr) |
1329 | { | |
1330 | if (s->flags & SLAB_CONSISTENCY_CHECKS) { | |
bb192ed9 | 1331 | if (!alloc_consistency_checks(s, slab, object)) |
becfda68 LA |
1332 | goto bad; |
1333 | } | |
81819f0f | 1334 | |
3ec09742 CL |
1335 | /* Success perform special debug activities for allocs */ |
1336 | if (s->flags & SLAB_STORE_USER) | |
1337 | set_track(s, object, TRACK_ALLOC, addr); | |
bb192ed9 | 1338 | trace(s, slab, object, 1); |
f7cb1933 | 1339 | init_object(s, object, SLUB_RED_ACTIVE); |
81819f0f | 1340 | return 1; |
3ec09742 | 1341 | |
81819f0f | 1342 | bad: |
bb192ed9 | 1343 | if (folio_test_slab(slab_folio(slab))) { |
81819f0f CL |
1344 | /* |
1345 | * If this is a slab page then lets do the best we can | |
1346 | * to avoid issues in the future. Marking all objects | |
672bba3a | 1347 | * as used avoids touching the remaining objects. |
81819f0f | 1348 | */ |
24922684 | 1349 | slab_fix(s, "Marking all objects used"); |
bb192ed9 VB |
1350 | slab->inuse = slab->objects; |
1351 | slab->freelist = NULL; | |
81819f0f CL |
1352 | } |
1353 | return 0; | |
1354 | } | |
1355 | ||
becfda68 | 1356 | static inline int free_consistency_checks(struct kmem_cache *s, |
bb192ed9 | 1357 | struct slab *slab, void *object, unsigned long addr) |
81819f0f | 1358 | { |
bb192ed9 VB |
1359 | if (!check_valid_pointer(s, slab, object)) { |
1360 | slab_err(s, slab, "Invalid object pointer 0x%p", object); | |
becfda68 | 1361 | return 0; |
81819f0f CL |
1362 | } |
1363 | ||
bb192ed9 VB |
1364 | if (on_freelist(s, slab, object)) { |
1365 | object_err(s, slab, object, "Object already free"); | |
becfda68 | 1366 | return 0; |
81819f0f CL |
1367 | } |
1368 | ||
bb192ed9 | 1369 | if (!check_object(s, slab, object, SLUB_RED_ACTIVE)) |
becfda68 | 1370 | return 0; |
81819f0f | 1371 | |
bb192ed9 VB |
1372 | if (unlikely(s != slab->slab_cache)) { |
1373 | if (!folio_test_slab(slab_folio(slab))) { | |
1374 | slab_err(s, slab, "Attempt to free object(0x%p) outside of slab", | |
756a025f | 1375 | object); |
bb192ed9 | 1376 | } else if (!slab->slab_cache) { |
f9f58285 FF |
1377 | pr_err("SLUB <none>: no slab for object 0x%p.\n", |
1378 | object); | |
70d71228 | 1379 | dump_stack(); |
06428780 | 1380 | } else |
bb192ed9 | 1381 | object_err(s, slab, object, |
24922684 | 1382 | "page slab pointer corrupt."); |
becfda68 LA |
1383 | return 0; |
1384 | } | |
1385 | return 1; | |
1386 | } | |
1387 | ||
1388 | /* Supports checking bulk free of a constructed freelist */ | |
1389 | static noinline int free_debug_processing( | |
bb192ed9 | 1390 | struct kmem_cache *s, struct slab *slab, |
becfda68 LA |
1391 | void *head, void *tail, int bulk_cnt, |
1392 | unsigned long addr) | |
1393 | { | |
bb192ed9 | 1394 | struct kmem_cache_node *n = get_node(s, slab_nid(slab)); |
becfda68 LA |
1395 | void *object = head; |
1396 | int cnt = 0; | |
a2b4ae8b | 1397 | unsigned long flags, flags2; |
becfda68 | 1398 | int ret = 0; |
c4cf6785 SAS |
1399 | depot_stack_handle_t handle = 0; |
1400 | ||
1401 | if (s->flags & SLAB_STORE_USER) | |
1402 | handle = set_track_prepare(); | |
becfda68 LA |
1403 | |
1404 | spin_lock_irqsave(&n->list_lock, flags); | |
bb192ed9 | 1405 | slab_lock(slab, &flags2); |
becfda68 LA |
1406 | |
1407 | if (s->flags & SLAB_CONSISTENCY_CHECKS) { | |
bb192ed9 | 1408 | if (!check_slab(s, slab)) |
becfda68 LA |
1409 | goto out; |
1410 | } | |
1411 | ||
1412 | next_object: | |
1413 | cnt++; | |
1414 | ||
1415 | if (s->flags & SLAB_CONSISTENCY_CHECKS) { | |
bb192ed9 | 1416 | if (!free_consistency_checks(s, slab, object, addr)) |
becfda68 | 1417 | goto out; |
81819f0f | 1418 | } |
3ec09742 | 1419 | |
3ec09742 | 1420 | if (s->flags & SLAB_STORE_USER) |
c4cf6785 | 1421 | set_track_update(s, object, TRACK_FREE, addr, handle); |
bb192ed9 | 1422 | trace(s, slab, object, 0); |
81084651 | 1423 | /* Freepointer not overwritten by init_object(), SLAB_POISON moved it */ |
f7cb1933 | 1424 | init_object(s, object, SLUB_RED_INACTIVE); |
81084651 JDB |
1425 | |
1426 | /* Reached end of constructed freelist yet? */ | |
1427 | if (object != tail) { | |
1428 | object = get_freepointer(s, object); | |
1429 | goto next_object; | |
1430 | } | |
804aa132 LA |
1431 | ret = 1; |
1432 | ||
5c2e4bbb | 1433 | out: |
81084651 | 1434 | if (cnt != bulk_cnt) |
bb192ed9 | 1435 | slab_err(s, slab, "Bulk freelist count(%d) invalid(%d)\n", |
81084651 JDB |
1436 | bulk_cnt, cnt); |
1437 | ||
bb192ed9 | 1438 | slab_unlock(slab, &flags2); |
282acb43 | 1439 | spin_unlock_irqrestore(&n->list_lock, flags); |
804aa132 LA |
1440 | if (!ret) |
1441 | slab_fix(s, "Object at 0x%p not freed", object); | |
1442 | return ret; | |
81819f0f CL |
1443 | } |
1444 | ||
e17f1dfb VB |
1445 | /* |
1446 | * Parse a block of slub_debug options. Blocks are delimited by ';' | |
1447 | * | |
1448 | * @str: start of block | |
1449 | * @flags: returns parsed flags, or DEBUG_DEFAULT_FLAGS if none specified | |
1450 | * @slabs: return start of list of slabs, or NULL when there's no list | |
1451 | * @init: assume this is initial parsing and not per-kmem-create parsing | |
1452 | * | |
1453 | * returns the start of next block if there's any, or NULL | |
1454 | */ | |
1455 | static char * | |
1456 | parse_slub_debug_flags(char *str, slab_flags_t *flags, char **slabs, bool init) | |
41ecc55b | 1457 | { |
e17f1dfb | 1458 | bool higher_order_disable = false; |
f0630fff | 1459 | |
e17f1dfb VB |
1460 | /* Skip any completely empty blocks */ |
1461 | while (*str && *str == ';') | |
1462 | str++; | |
1463 | ||
1464 | if (*str == ',') { | |
f0630fff CL |
1465 | /* |
1466 | * No options but restriction on slabs. This means full | |
1467 | * debugging for slabs matching a pattern. | |
1468 | */ | |
e17f1dfb | 1469 | *flags = DEBUG_DEFAULT_FLAGS; |
f0630fff | 1470 | goto check_slabs; |
e17f1dfb VB |
1471 | } |
1472 | *flags = 0; | |
f0630fff | 1473 | |
e17f1dfb VB |
1474 | /* Determine which debug features should be switched on */ |
1475 | for (; *str && *str != ',' && *str != ';'; str++) { | |
f0630fff | 1476 | switch (tolower(*str)) { |
e17f1dfb VB |
1477 | case '-': |
1478 | *flags = 0; | |
1479 | break; | |
f0630fff | 1480 | case 'f': |
e17f1dfb | 1481 | *flags |= SLAB_CONSISTENCY_CHECKS; |
f0630fff CL |
1482 | break; |
1483 | case 'z': | |
e17f1dfb | 1484 | *flags |= SLAB_RED_ZONE; |
f0630fff CL |
1485 | break; |
1486 | case 'p': | |
e17f1dfb | 1487 | *flags |= SLAB_POISON; |
f0630fff CL |
1488 | break; |
1489 | case 'u': | |
e17f1dfb | 1490 | *flags |= SLAB_STORE_USER; |
f0630fff CL |
1491 | break; |
1492 | case 't': | |
e17f1dfb | 1493 | *flags |= SLAB_TRACE; |
f0630fff | 1494 | break; |
4c13dd3b | 1495 | case 'a': |
e17f1dfb | 1496 | *flags |= SLAB_FAILSLAB; |
4c13dd3b | 1497 | break; |
08303a73 CA |
1498 | case 'o': |
1499 | /* | |
1500 | * Avoid enabling debugging on caches if its minimum | |
1501 | * order would increase as a result. | |
1502 | */ | |
e17f1dfb | 1503 | higher_order_disable = true; |
08303a73 | 1504 | break; |
f0630fff | 1505 | default: |
e17f1dfb VB |
1506 | if (init) |
1507 | pr_err("slub_debug option '%c' unknown. skipped\n", *str); | |
f0630fff | 1508 | } |
41ecc55b | 1509 | } |
f0630fff | 1510 | check_slabs: |
41ecc55b | 1511 | if (*str == ',') |
e17f1dfb VB |
1512 | *slabs = ++str; |
1513 | else | |
1514 | *slabs = NULL; | |
1515 | ||
1516 | /* Skip over the slab list */ | |
1517 | while (*str && *str != ';') | |
1518 | str++; | |
1519 | ||
1520 | /* Skip any completely empty blocks */ | |
1521 | while (*str && *str == ';') | |
1522 | str++; | |
1523 | ||
1524 | if (init && higher_order_disable) | |
1525 | disable_higher_order_debug = 1; | |
1526 | ||
1527 | if (*str) | |
1528 | return str; | |
1529 | else | |
1530 | return NULL; | |
1531 | } | |
1532 | ||
1533 | static int __init setup_slub_debug(char *str) | |
1534 | { | |
1535 | slab_flags_t flags; | |
a7f1d485 | 1536 | slab_flags_t global_flags; |
e17f1dfb VB |
1537 | char *saved_str; |
1538 | char *slab_list; | |
1539 | bool global_slub_debug_changed = false; | |
1540 | bool slab_list_specified = false; | |
1541 | ||
a7f1d485 | 1542 | global_flags = DEBUG_DEFAULT_FLAGS; |
e17f1dfb VB |
1543 | if (*str++ != '=' || !*str) |
1544 | /* | |
1545 | * No options specified. Switch on full debugging. | |
1546 | */ | |
1547 | goto out; | |
1548 | ||
1549 | saved_str = str; | |
1550 | while (str) { | |
1551 | str = parse_slub_debug_flags(str, &flags, &slab_list, true); | |
1552 | ||
1553 | if (!slab_list) { | |
a7f1d485 | 1554 | global_flags = flags; |
e17f1dfb VB |
1555 | global_slub_debug_changed = true; |
1556 | } else { | |
1557 | slab_list_specified = true; | |
5cf909c5 OG |
1558 | if (flags & SLAB_STORE_USER) |
1559 | stack_depot_want_early_init(); | |
e17f1dfb VB |
1560 | } |
1561 | } | |
1562 | ||
1563 | /* | |
1564 | * For backwards compatibility, a single list of flags with list of | |
a7f1d485 VB |
1565 | * slabs means debugging is only changed for those slabs, so the global |
1566 | * slub_debug should be unchanged (0 or DEBUG_DEFAULT_FLAGS, depending | |
1567 | * on CONFIG_SLUB_DEBUG_ON). We can extended that to multiple lists as | |
e17f1dfb VB |
1568 | * long as there is no option specifying flags without a slab list. |
1569 | */ | |
1570 | if (slab_list_specified) { | |
1571 | if (!global_slub_debug_changed) | |
a7f1d485 | 1572 | global_flags = slub_debug; |
e17f1dfb VB |
1573 | slub_debug_string = saved_str; |
1574 | } | |
f0630fff | 1575 | out: |
a7f1d485 | 1576 | slub_debug = global_flags; |
5cf909c5 OG |
1577 | if (slub_debug & SLAB_STORE_USER) |
1578 | stack_depot_want_early_init(); | |
ca0cab65 VB |
1579 | if (slub_debug != 0 || slub_debug_string) |
1580 | static_branch_enable(&slub_debug_enabled); | |
02ac47d0 SB |
1581 | else |
1582 | static_branch_disable(&slub_debug_enabled); | |
6471384a AP |
1583 | if ((static_branch_unlikely(&init_on_alloc) || |
1584 | static_branch_unlikely(&init_on_free)) && | |
1585 | (slub_debug & SLAB_POISON)) | |
1586 | pr_info("mem auto-init: SLAB_POISON will take precedence over init_on_alloc/init_on_free\n"); | |
41ecc55b CL |
1587 | return 1; |
1588 | } | |
1589 | ||
1590 | __setup("slub_debug", setup_slub_debug); | |
1591 | ||
c5fd3ca0 AT |
1592 | /* |
1593 | * kmem_cache_flags - apply debugging options to the cache | |
1594 | * @object_size: the size of an object without meta data | |
1595 | * @flags: flags to set | |
1596 | * @name: name of the cache | |
c5fd3ca0 AT |
1597 | * |
1598 | * Debug option(s) are applied to @flags. In addition to the debug | |
1599 | * option(s), if a slab name (or multiple) is specified i.e. | |
1600 | * slub_debug=<Debug-Options>,<slab name1>,<slab name2> ... | |
1601 | * then only the select slabs will receive the debug option(s). | |
1602 | */ | |
0293d1fd | 1603 | slab_flags_t kmem_cache_flags(unsigned int object_size, |
37540008 | 1604 | slab_flags_t flags, const char *name) |
41ecc55b | 1605 | { |
c5fd3ca0 AT |
1606 | char *iter; |
1607 | size_t len; | |
e17f1dfb VB |
1608 | char *next_block; |
1609 | slab_flags_t block_flags; | |
ca220593 JB |
1610 | slab_flags_t slub_debug_local = slub_debug; |
1611 | ||
a285909f HY |
1612 | if (flags & SLAB_NO_USER_FLAGS) |
1613 | return flags; | |
1614 | ||
ca220593 JB |
1615 | /* |
1616 | * If the slab cache is for debugging (e.g. kmemleak) then | |
1617 | * don't store user (stack trace) information by default, | |
1618 | * but let the user enable it via the command line below. | |
1619 | */ | |
1620 | if (flags & SLAB_NOLEAKTRACE) | |
1621 | slub_debug_local &= ~SLAB_STORE_USER; | |
c5fd3ca0 | 1622 | |
c5fd3ca0 | 1623 | len = strlen(name); |
e17f1dfb VB |
1624 | next_block = slub_debug_string; |
1625 | /* Go through all blocks of debug options, see if any matches our slab's name */ | |
1626 | while (next_block) { | |
1627 | next_block = parse_slub_debug_flags(next_block, &block_flags, &iter, false); | |
1628 | if (!iter) | |
1629 | continue; | |
1630 | /* Found a block that has a slab list, search it */ | |
1631 | while (*iter) { | |
1632 | char *end, *glob; | |
1633 | size_t cmplen; | |
1634 | ||
1635 | end = strchrnul(iter, ','); | |
1636 | if (next_block && next_block < end) | |
1637 | end = next_block - 1; | |
1638 | ||
1639 | glob = strnchr(iter, end - iter, '*'); | |
1640 | if (glob) | |
1641 | cmplen = glob - iter; | |
1642 | else | |
1643 | cmplen = max_t(size_t, len, (end - iter)); | |
c5fd3ca0 | 1644 | |
e17f1dfb VB |
1645 | if (!strncmp(name, iter, cmplen)) { |
1646 | flags |= block_flags; | |
1647 | return flags; | |
1648 | } | |
c5fd3ca0 | 1649 | |
e17f1dfb VB |
1650 | if (!*end || *end == ';') |
1651 | break; | |
1652 | iter = end + 1; | |
c5fd3ca0 | 1653 | } |
c5fd3ca0 | 1654 | } |
ba0268a8 | 1655 | |
ca220593 | 1656 | return flags | slub_debug_local; |
41ecc55b | 1657 | } |
b4a64718 | 1658 | #else /* !CONFIG_SLUB_DEBUG */ |
c0f81a94 | 1659 | static inline void setup_object_debug(struct kmem_cache *s, void *object) {} |
a50b854e | 1660 | static inline |
bb192ed9 | 1661 | void setup_slab_debug(struct kmem_cache *s, struct slab *slab, void *addr) {} |
41ecc55b | 1662 | |
3ec09742 | 1663 | static inline int alloc_debug_processing(struct kmem_cache *s, |
bb192ed9 | 1664 | struct slab *slab, void *object, unsigned long addr) { return 0; } |
41ecc55b | 1665 | |
282acb43 | 1666 | static inline int free_debug_processing( |
bb192ed9 | 1667 | struct kmem_cache *s, struct slab *slab, |
81084651 | 1668 | void *head, void *tail, int bulk_cnt, |
282acb43 | 1669 | unsigned long addr) { return 0; } |
41ecc55b | 1670 | |
a204e6d6 | 1671 | static inline void slab_pad_check(struct kmem_cache *s, struct slab *slab) {} |
bb192ed9 | 1672 | static inline int check_object(struct kmem_cache *s, struct slab *slab, |
f7cb1933 | 1673 | void *object, u8 val) { return 1; } |
5cc6eee8 | 1674 | static inline void add_full(struct kmem_cache *s, struct kmem_cache_node *n, |
bb192ed9 | 1675 | struct slab *slab) {} |
c65c1877 | 1676 | static inline void remove_full(struct kmem_cache *s, struct kmem_cache_node *n, |
bb192ed9 | 1677 | struct slab *slab) {} |
0293d1fd | 1678 | slab_flags_t kmem_cache_flags(unsigned int object_size, |
37540008 | 1679 | slab_flags_t flags, const char *name) |
ba0268a8 CL |
1680 | { |
1681 | return flags; | |
1682 | } | |
41ecc55b | 1683 | #define slub_debug 0 |
0f389ec6 | 1684 | |
fdaa45e9 IM |
1685 | #define disable_higher_order_debug 0 |
1686 | ||
0f389ec6 CL |
1687 | static inline unsigned long slabs_node(struct kmem_cache *s, int node) |
1688 | { return 0; } | |
26c02cf0 AB |
1689 | static inline unsigned long node_nr_slabs(struct kmem_cache_node *n) |
1690 | { return 0; } | |
205ab99d CL |
1691 | static inline void inc_slabs_node(struct kmem_cache *s, int node, |
1692 | int objects) {} | |
1693 | static inline void dec_slabs_node(struct kmem_cache *s, int node, | |
1694 | int objects) {} | |
7d550c56 | 1695 | |
bb192ed9 | 1696 | static bool freelist_corrupted(struct kmem_cache *s, struct slab *slab, |
dc07a728 | 1697 | void **freelist, void *nextfree) |
52f23478 DZ |
1698 | { |
1699 | return false; | |
1700 | } | |
02e72cc6 AR |
1701 | #endif /* CONFIG_SLUB_DEBUG */ |
1702 | ||
1703 | /* | |
1704 | * Hooks for other subsystems that check memory allocations. In a typical | |
1705 | * production configuration these hooks all should produce no code at all. | |
1706 | */ | |
ee3ce779 | 1707 | static __always_inline void kfree_hook(void *x) |
d56791b3 RB |
1708 | { |
1709 | kmemleak_free(x); | |
027b37b5 | 1710 | kasan_kfree_large(x); |
d56791b3 RB |
1711 | } |
1712 | ||
d57a964e AK |
1713 | static __always_inline bool slab_free_hook(struct kmem_cache *s, |
1714 | void *x, bool init) | |
d56791b3 RB |
1715 | { |
1716 | kmemleak_free_recursive(x, s->flags); | |
7d550c56 | 1717 | |
84048039 | 1718 | debug_check_no_locks_freed(x, s->object_size); |
02e72cc6 | 1719 | |
02e72cc6 AR |
1720 | if (!(s->flags & SLAB_DEBUG_OBJECTS)) |
1721 | debug_check_no_obj_freed(x, s->object_size); | |
0316bec2 | 1722 | |
cfbe1636 ME |
1723 | /* Use KCSAN to help debug racy use-after-free. */ |
1724 | if (!(s->flags & SLAB_TYPESAFE_BY_RCU)) | |
1725 | __kcsan_check_access(x, s->object_size, | |
1726 | KCSAN_ACCESS_WRITE | KCSAN_ACCESS_ASSERT); | |
1727 | ||
d57a964e AK |
1728 | /* |
1729 | * As memory initialization might be integrated into KASAN, | |
1730 | * kasan_slab_free and initialization memset's must be | |
1731 | * kept together to avoid discrepancies in behavior. | |
1732 | * | |
1733 | * The initialization memset's clear the object and the metadata, | |
1734 | * but don't touch the SLAB redzone. | |
1735 | */ | |
1736 | if (init) { | |
1737 | int rsize; | |
1738 | ||
1739 | if (!kasan_has_integrated_init()) | |
1740 | memset(kasan_reset_tag(x), 0, s->object_size); | |
1741 | rsize = (s->flags & SLAB_RED_ZONE) ? s->red_left_pad : 0; | |
1742 | memset((char *)kasan_reset_tag(x) + s->inuse, 0, | |
1743 | s->size - s->inuse - rsize); | |
1744 | } | |
1745 | /* KASAN might put x into memory quarantine, delaying its reuse. */ | |
1746 | return kasan_slab_free(s, x, init); | |
02e72cc6 | 1747 | } |
205ab99d | 1748 | |
c3895391 | 1749 | static inline bool slab_free_freelist_hook(struct kmem_cache *s, |
899447f6 ML |
1750 | void **head, void **tail, |
1751 | int *cnt) | |
81084651 | 1752 | { |
6471384a AP |
1753 | |
1754 | void *object; | |
1755 | void *next = *head; | |
1756 | void *old_tail = *tail ? *tail : *head; | |
6471384a | 1757 | |
b89fb5ef | 1758 | if (is_kfence_address(next)) { |
d57a964e | 1759 | slab_free_hook(s, next, false); |
b89fb5ef AP |
1760 | return true; |
1761 | } | |
1762 | ||
aea4df4c LA |
1763 | /* Head and tail of the reconstructed freelist */ |
1764 | *head = NULL; | |
1765 | *tail = NULL; | |
1b7e816f | 1766 | |
aea4df4c LA |
1767 | do { |
1768 | object = next; | |
1769 | next = get_freepointer(s, object); | |
1770 | ||
c3895391 | 1771 | /* If object's reuse doesn't have to be delayed */ |
d57a964e | 1772 | if (!slab_free_hook(s, object, slab_want_init_on_free(s))) { |
c3895391 AK |
1773 | /* Move object to the new freelist */ |
1774 | set_freepointer(s, object, *head); | |
1775 | *head = object; | |
1776 | if (!*tail) | |
1777 | *tail = object; | |
899447f6 ML |
1778 | } else { |
1779 | /* | |
1780 | * Adjust the reconstructed freelist depth | |
1781 | * accordingly if object's reuse is delayed. | |
1782 | */ | |
1783 | --(*cnt); | |
c3895391 AK |
1784 | } |
1785 | } while (object != old_tail); | |
1786 | ||
1787 | if (*head == *tail) | |
1788 | *tail = NULL; | |
1789 | ||
1790 | return *head != NULL; | |
81084651 JDB |
1791 | } |
1792 | ||
c0f81a94 | 1793 | static void *setup_object(struct kmem_cache *s, void *object) |
588f8ba9 | 1794 | { |
c0f81a94 | 1795 | setup_object_debug(s, object); |
4d176711 | 1796 | object = kasan_init_slab_obj(s, object); |
588f8ba9 TG |
1797 | if (unlikely(s->ctor)) { |
1798 | kasan_unpoison_object_data(s, object); | |
1799 | s->ctor(object); | |
1800 | kasan_poison_object_data(s, object); | |
1801 | } | |
4d176711 | 1802 | return object; |
588f8ba9 TG |
1803 | } |
1804 | ||
81819f0f CL |
1805 | /* |
1806 | * Slab allocation and freeing | |
1807 | */ | |
a485e1da XS |
1808 | static inline struct slab *alloc_slab_page(gfp_t flags, int node, |
1809 | struct kmem_cache_order_objects oo) | |
65c3376a | 1810 | { |
45387b8c VB |
1811 | struct folio *folio; |
1812 | struct slab *slab; | |
19af27af | 1813 | unsigned int order = oo_order(oo); |
65c3376a | 1814 | |
2154a336 | 1815 | if (node == NUMA_NO_NODE) |
45387b8c | 1816 | folio = (struct folio *)alloc_pages(flags, order); |
65c3376a | 1817 | else |
45387b8c | 1818 | folio = (struct folio *)__alloc_pages_node(node, flags, order); |
5dfb4175 | 1819 | |
45387b8c VB |
1820 | if (!folio) |
1821 | return NULL; | |
1822 | ||
1823 | slab = folio_slab(folio); | |
1824 | __folio_set_slab(folio); | |
1825 | if (page_is_pfmemalloc(folio_page(folio, 0))) | |
1826 | slab_set_pfmemalloc(slab); | |
1827 | ||
1828 | return slab; | |
65c3376a CL |
1829 | } |
1830 | ||
210e7a43 TG |
1831 | #ifdef CONFIG_SLAB_FREELIST_RANDOM |
1832 | /* Pre-initialize the random sequence cache */ | |
1833 | static int init_cache_random_seq(struct kmem_cache *s) | |
1834 | { | |
19af27af | 1835 | unsigned int count = oo_objects(s->oo); |
210e7a43 | 1836 | int err; |
210e7a43 | 1837 | |
a810007a SR |
1838 | /* Bailout if already initialised */ |
1839 | if (s->random_seq) | |
1840 | return 0; | |
1841 | ||
210e7a43 TG |
1842 | err = cache_random_seq_create(s, count, GFP_KERNEL); |
1843 | if (err) { | |
1844 | pr_err("SLUB: Unable to initialize free list for %s\n", | |
1845 | s->name); | |
1846 | return err; | |
1847 | } | |
1848 | ||
1849 | /* Transform to an offset on the set of pages */ | |
1850 | if (s->random_seq) { | |
19af27af AD |
1851 | unsigned int i; |
1852 | ||
210e7a43 TG |
1853 | for (i = 0; i < count; i++) |
1854 | s->random_seq[i] *= s->size; | |
1855 | } | |
1856 | return 0; | |
1857 | } | |
1858 | ||
1859 | /* Initialize each random sequence freelist per cache */ | |
1860 | static void __init init_freelist_randomization(void) | |
1861 | { | |
1862 | struct kmem_cache *s; | |
1863 | ||
1864 | mutex_lock(&slab_mutex); | |
1865 | ||
1866 | list_for_each_entry(s, &slab_caches, list) | |
1867 | init_cache_random_seq(s); | |
1868 | ||
1869 | mutex_unlock(&slab_mutex); | |
1870 | } | |
1871 | ||
1872 | /* Get the next entry on the pre-computed freelist randomized */ | |
bb192ed9 | 1873 | static void *next_freelist_entry(struct kmem_cache *s, struct slab *slab, |
210e7a43 TG |
1874 | unsigned long *pos, void *start, |
1875 | unsigned long page_limit, | |
1876 | unsigned long freelist_count) | |
1877 | { | |
1878 | unsigned int idx; | |
1879 | ||
1880 | /* | |
1881 | * If the target page allocation failed, the number of objects on the | |
1882 | * page might be smaller than the usual size defined by the cache. | |
1883 | */ | |
1884 | do { | |
1885 | idx = s->random_seq[*pos]; | |
1886 | *pos += 1; | |
1887 | if (*pos >= freelist_count) | |
1888 | *pos = 0; | |
1889 | } while (unlikely(idx >= page_limit)); | |
1890 | ||
1891 | return (char *)start + idx; | |
1892 | } | |
1893 | ||
1894 | /* Shuffle the single linked freelist based on a random pre-computed sequence */ | |
bb192ed9 | 1895 | static bool shuffle_freelist(struct kmem_cache *s, struct slab *slab) |
210e7a43 TG |
1896 | { |
1897 | void *start; | |
1898 | void *cur; | |
1899 | void *next; | |
1900 | unsigned long idx, pos, page_limit, freelist_count; | |
1901 | ||
bb192ed9 | 1902 | if (slab->objects < 2 || !s->random_seq) |
210e7a43 TG |
1903 | return false; |
1904 | ||
1905 | freelist_count = oo_objects(s->oo); | |
1906 | pos = get_random_int() % freelist_count; | |
1907 | ||
bb192ed9 VB |
1908 | page_limit = slab->objects * s->size; |
1909 | start = fixup_red_left(s, slab_address(slab)); | |
210e7a43 TG |
1910 | |
1911 | /* First entry is used as the base of the freelist */ | |
bb192ed9 | 1912 | cur = next_freelist_entry(s, slab, &pos, start, page_limit, |
210e7a43 | 1913 | freelist_count); |
c0f81a94 | 1914 | cur = setup_object(s, cur); |
bb192ed9 | 1915 | slab->freelist = cur; |
210e7a43 | 1916 | |
bb192ed9 VB |
1917 | for (idx = 1; idx < slab->objects; idx++) { |
1918 | next = next_freelist_entry(s, slab, &pos, start, page_limit, | |
210e7a43 | 1919 | freelist_count); |
c0f81a94 | 1920 | next = setup_object(s, next); |
210e7a43 TG |
1921 | set_freepointer(s, cur, next); |
1922 | cur = next; | |
1923 | } | |
210e7a43 TG |
1924 | set_freepointer(s, cur, NULL); |
1925 | ||
1926 | return true; | |
1927 | } | |
1928 | #else | |
1929 | static inline int init_cache_random_seq(struct kmem_cache *s) | |
1930 | { | |
1931 | return 0; | |
1932 | } | |
1933 | static inline void init_freelist_randomization(void) { } | |
bb192ed9 | 1934 | static inline bool shuffle_freelist(struct kmem_cache *s, struct slab *slab) |
210e7a43 TG |
1935 | { |
1936 | return false; | |
1937 | } | |
1938 | #endif /* CONFIG_SLAB_FREELIST_RANDOM */ | |
1939 | ||
bb192ed9 | 1940 | static struct slab *allocate_slab(struct kmem_cache *s, gfp_t flags, int node) |
81819f0f | 1941 | { |
bb192ed9 | 1942 | struct slab *slab; |
834f3d11 | 1943 | struct kmem_cache_order_objects oo = s->oo; |
ba52270d | 1944 | gfp_t alloc_gfp; |
4d176711 | 1945 | void *start, *p, *next; |
a50b854e | 1946 | int idx; |
210e7a43 | 1947 | bool shuffle; |
81819f0f | 1948 | |
7e0528da CL |
1949 | flags &= gfp_allowed_mask; |
1950 | ||
b7a49f0d | 1951 | flags |= s->allocflags; |
e12ba74d | 1952 | |
ba52270d PE |
1953 | /* |
1954 | * Let the initial higher-order allocation fail under memory pressure | |
1955 | * so we fall-back to the minimum order allocation. | |
1956 | */ | |
1957 | alloc_gfp = (flags | __GFP_NOWARN | __GFP_NORETRY) & ~__GFP_NOFAIL; | |
d0164adc | 1958 | if ((alloc_gfp & __GFP_DIRECT_RECLAIM) && oo_order(oo) > oo_order(s->min)) |
27c08f75 | 1959 | alloc_gfp = (alloc_gfp | __GFP_NOMEMALLOC) & ~__GFP_RECLAIM; |
ba52270d | 1960 | |
a485e1da | 1961 | slab = alloc_slab_page(alloc_gfp, node, oo); |
bb192ed9 | 1962 | if (unlikely(!slab)) { |
65c3376a | 1963 | oo = s->min; |
80c3a998 | 1964 | alloc_gfp = flags; |
65c3376a CL |
1965 | /* |
1966 | * Allocation may have failed due to fragmentation. | |
1967 | * Try a lower order alloc if possible | |
1968 | */ | |
a485e1da | 1969 | slab = alloc_slab_page(alloc_gfp, node, oo); |
bb192ed9 | 1970 | if (unlikely(!slab)) |
588f8ba9 TG |
1971 | goto out; |
1972 | stat(s, ORDER_FALLBACK); | |
65c3376a | 1973 | } |
5a896d9e | 1974 | |
bb192ed9 | 1975 | slab->objects = oo_objects(oo); |
81819f0f | 1976 | |
bb192ed9 | 1977 | account_slab(slab, oo_order(oo), s, flags); |
1f3147b4 | 1978 | |
bb192ed9 | 1979 | slab->slab_cache = s; |
81819f0f | 1980 | |
6e48a966 | 1981 | kasan_poison_slab(slab); |
81819f0f | 1982 | |
bb192ed9 | 1983 | start = slab_address(slab); |
81819f0f | 1984 | |
bb192ed9 | 1985 | setup_slab_debug(s, slab, start); |
0316bec2 | 1986 | |
bb192ed9 | 1987 | shuffle = shuffle_freelist(s, slab); |
210e7a43 TG |
1988 | |
1989 | if (!shuffle) { | |
4d176711 | 1990 | start = fixup_red_left(s, start); |
c0f81a94 | 1991 | start = setup_object(s, start); |
bb192ed9 VB |
1992 | slab->freelist = start; |
1993 | for (idx = 0, p = start; idx < slab->objects - 1; idx++) { | |
18e50661 | 1994 | next = p + s->size; |
c0f81a94 | 1995 | next = setup_object(s, next); |
18e50661 AK |
1996 | set_freepointer(s, p, next); |
1997 | p = next; | |
1998 | } | |
1999 | set_freepointer(s, p, NULL); | |
81819f0f | 2000 | } |
81819f0f | 2001 | |
bb192ed9 VB |
2002 | slab->inuse = slab->objects; |
2003 | slab->frozen = 1; | |
588f8ba9 | 2004 | |
81819f0f | 2005 | out: |
bb192ed9 | 2006 | if (!slab) |
588f8ba9 TG |
2007 | return NULL; |
2008 | ||
bb192ed9 | 2009 | inc_slabs_node(s, slab_nid(slab), slab->objects); |
588f8ba9 | 2010 | |
bb192ed9 | 2011 | return slab; |
81819f0f CL |
2012 | } |
2013 | ||
bb192ed9 | 2014 | static struct slab *new_slab(struct kmem_cache *s, gfp_t flags, int node) |
588f8ba9 | 2015 | { |
44405099 LL |
2016 | if (unlikely(flags & GFP_SLAB_BUG_MASK)) |
2017 | flags = kmalloc_fix_flags(flags); | |
588f8ba9 | 2018 | |
53a0de06 VB |
2019 | WARN_ON_ONCE(s->ctor && (flags & __GFP_ZERO)); |
2020 | ||
588f8ba9 TG |
2021 | return allocate_slab(s, |
2022 | flags & (GFP_RECLAIM_MASK | GFP_CONSTRAINT_MASK), node); | |
2023 | } | |
2024 | ||
4020b4a2 | 2025 | static void __free_slab(struct kmem_cache *s, struct slab *slab) |
81819f0f | 2026 | { |
4020b4a2 VB |
2027 | struct folio *folio = slab_folio(slab); |
2028 | int order = folio_order(folio); | |
834f3d11 | 2029 | int pages = 1 << order; |
81819f0f | 2030 | |
8fc8d666 | 2031 | if (kmem_cache_debug_flags(s, SLAB_CONSISTENCY_CHECKS)) { |
81819f0f CL |
2032 | void *p; |
2033 | ||
bb192ed9 | 2034 | slab_pad_check(s, slab); |
4020b4a2 | 2035 | for_each_object(p, s, slab_address(slab), slab->objects) |
bb192ed9 | 2036 | check_object(s, slab, p, SLUB_RED_INACTIVE); |
81819f0f CL |
2037 | } |
2038 | ||
4020b4a2 VB |
2039 | __slab_clear_pfmemalloc(slab); |
2040 | __folio_clear_slab(folio); | |
2041 | folio->mapping = NULL; | |
1eb5ac64 NP |
2042 | if (current->reclaim_state) |
2043 | current->reclaim_state->reclaimed_slab += pages; | |
4020b4a2 VB |
2044 | unaccount_slab(slab, order, s); |
2045 | __free_pages(folio_page(folio, 0), order); | |
81819f0f CL |
2046 | } |
2047 | ||
2048 | static void rcu_free_slab(struct rcu_head *h) | |
2049 | { | |
bb192ed9 | 2050 | struct slab *slab = container_of(h, struct slab, rcu_head); |
da9a638c | 2051 | |
bb192ed9 | 2052 | __free_slab(slab->slab_cache, slab); |
81819f0f CL |
2053 | } |
2054 | ||
bb192ed9 | 2055 | static void free_slab(struct kmem_cache *s, struct slab *slab) |
81819f0f | 2056 | { |
5f0d5a3a | 2057 | if (unlikely(s->flags & SLAB_TYPESAFE_BY_RCU)) { |
bb192ed9 | 2058 | call_rcu(&slab->rcu_head, rcu_free_slab); |
81819f0f | 2059 | } else |
bb192ed9 | 2060 | __free_slab(s, slab); |
81819f0f CL |
2061 | } |
2062 | ||
bb192ed9 | 2063 | static void discard_slab(struct kmem_cache *s, struct slab *slab) |
81819f0f | 2064 | { |
bb192ed9 VB |
2065 | dec_slabs_node(s, slab_nid(slab), slab->objects); |
2066 | free_slab(s, slab); | |
81819f0f CL |
2067 | } |
2068 | ||
2069 | /* | |
5cc6eee8 | 2070 | * Management of partially allocated slabs. |
81819f0f | 2071 | */ |
1e4dd946 | 2072 | static inline void |
bb192ed9 | 2073 | __add_partial(struct kmem_cache_node *n, struct slab *slab, int tail) |
81819f0f | 2074 | { |
e95eed57 | 2075 | n->nr_partial++; |
136333d1 | 2076 | if (tail == DEACTIVATE_TO_TAIL) |
bb192ed9 | 2077 | list_add_tail(&slab->slab_list, &n->partial); |
7c2e132c | 2078 | else |
bb192ed9 | 2079 | list_add(&slab->slab_list, &n->partial); |
81819f0f CL |
2080 | } |
2081 | ||
1e4dd946 | 2082 | static inline void add_partial(struct kmem_cache_node *n, |
bb192ed9 | 2083 | struct slab *slab, int tail) |
62e346a8 | 2084 | { |
c65c1877 | 2085 | lockdep_assert_held(&n->list_lock); |
bb192ed9 | 2086 | __add_partial(n, slab, tail); |
1e4dd946 | 2087 | } |
c65c1877 | 2088 | |
1e4dd946 | 2089 | static inline void remove_partial(struct kmem_cache_node *n, |
bb192ed9 | 2090 | struct slab *slab) |
1e4dd946 SR |
2091 | { |
2092 | lockdep_assert_held(&n->list_lock); | |
bb192ed9 | 2093 | list_del(&slab->slab_list); |
52b4b950 | 2094 | n->nr_partial--; |
1e4dd946 SR |
2095 | } |
2096 | ||
81819f0f | 2097 | /* |
7ced3719 CL |
2098 | * Remove slab from the partial list, freeze it and |
2099 | * return the pointer to the freelist. | |
81819f0f | 2100 | * |
497b66f2 | 2101 | * Returns a list of objects or NULL if it fails. |
81819f0f | 2102 | */ |
497b66f2 | 2103 | static inline void *acquire_slab(struct kmem_cache *s, |
bb192ed9 | 2104 | struct kmem_cache_node *n, struct slab *slab, |
b47291ef | 2105 | int mode) |
81819f0f | 2106 | { |
2cfb7455 CL |
2107 | void *freelist; |
2108 | unsigned long counters; | |
bb192ed9 | 2109 | struct slab new; |
2cfb7455 | 2110 | |
c65c1877 PZ |
2111 | lockdep_assert_held(&n->list_lock); |
2112 | ||
2cfb7455 CL |
2113 | /* |
2114 | * Zap the freelist and set the frozen bit. | |
2115 | * The old freelist is the list of objects for the | |
2116 | * per cpu allocation list. | |
2117 | */ | |
bb192ed9 VB |
2118 | freelist = slab->freelist; |
2119 | counters = slab->counters; | |
7ced3719 | 2120 | new.counters = counters; |
23910c50 | 2121 | if (mode) { |
bb192ed9 | 2122 | new.inuse = slab->objects; |
23910c50 PE |
2123 | new.freelist = NULL; |
2124 | } else { | |
2125 | new.freelist = freelist; | |
2126 | } | |
2cfb7455 | 2127 | |
a0132ac0 | 2128 | VM_BUG_ON(new.frozen); |
7ced3719 | 2129 | new.frozen = 1; |
2cfb7455 | 2130 | |
bb192ed9 | 2131 | if (!__cmpxchg_double_slab(s, slab, |
2cfb7455 | 2132 | freelist, counters, |
02d7633f | 2133 | new.freelist, new.counters, |
7ced3719 | 2134 | "acquire_slab")) |
7ced3719 | 2135 | return NULL; |
2cfb7455 | 2136 | |
bb192ed9 | 2137 | remove_partial(n, slab); |
7ced3719 | 2138 | WARN_ON(!freelist); |
49e22585 | 2139 | return freelist; |
81819f0f CL |
2140 | } |
2141 | ||
e0a043aa | 2142 | #ifdef CONFIG_SLUB_CPU_PARTIAL |
bb192ed9 | 2143 | static void put_cpu_partial(struct kmem_cache *s, struct slab *slab, int drain); |
e0a043aa | 2144 | #else |
bb192ed9 | 2145 | static inline void put_cpu_partial(struct kmem_cache *s, struct slab *slab, |
e0a043aa VB |
2146 | int drain) { } |
2147 | #endif | |
01b34d16 | 2148 | static inline bool pfmemalloc_match(struct slab *slab, gfp_t gfpflags); |
49e22585 | 2149 | |
81819f0f | 2150 | /* |
672bba3a | 2151 | * Try to allocate a partial slab from a specific node. |
81819f0f | 2152 | */ |
8ba00bb6 | 2153 | static void *get_partial_node(struct kmem_cache *s, struct kmem_cache_node *n, |
bb192ed9 | 2154 | struct slab **ret_slab, gfp_t gfpflags) |
81819f0f | 2155 | { |
bb192ed9 | 2156 | struct slab *slab, *slab2; |
49e22585 | 2157 | void *object = NULL; |
4b1f449d | 2158 | unsigned long flags; |
bb192ed9 | 2159 | unsigned int partial_slabs = 0; |
81819f0f CL |
2160 | |
2161 | /* | |
2162 | * Racy check. If we mistakenly see no partial slabs then we | |
2163 | * just allocate an empty slab. If we mistakenly try to get a | |
70b6d25e | 2164 | * partial slab and there is none available then get_partial() |
672bba3a | 2165 | * will return NULL. |
81819f0f CL |
2166 | */ |
2167 | if (!n || !n->nr_partial) | |
2168 | return NULL; | |
2169 | ||
4b1f449d | 2170 | spin_lock_irqsave(&n->list_lock, flags); |
bb192ed9 | 2171 | list_for_each_entry_safe(slab, slab2, &n->partial, slab_list) { |
8ba00bb6 | 2172 | void *t; |
49e22585 | 2173 | |
bb192ed9 | 2174 | if (!pfmemalloc_match(slab, gfpflags)) |
8ba00bb6 JK |
2175 | continue; |
2176 | ||
bb192ed9 | 2177 | t = acquire_slab(s, n, slab, object == NULL); |
49e22585 | 2178 | if (!t) |
9b1ea29b | 2179 | break; |
49e22585 | 2180 | |
12d79634 | 2181 | if (!object) { |
bb192ed9 | 2182 | *ret_slab = slab; |
49e22585 | 2183 | stat(s, ALLOC_FROM_PARTIAL); |
49e22585 | 2184 | object = t; |
49e22585 | 2185 | } else { |
bb192ed9 | 2186 | put_cpu_partial(s, slab, 0); |
8028dcea | 2187 | stat(s, CPU_PARTIAL_NODE); |
bb192ed9 | 2188 | partial_slabs++; |
49e22585 | 2189 | } |
b47291ef | 2190 | #ifdef CONFIG_SLUB_CPU_PARTIAL |
345c905d | 2191 | if (!kmem_cache_has_cpu_partial(s) |
bb192ed9 | 2192 | || partial_slabs > s->cpu_partial_slabs / 2) |
49e22585 | 2193 | break; |
b47291ef VB |
2194 | #else |
2195 | break; | |
2196 | #endif | |
49e22585 | 2197 | |
497b66f2 | 2198 | } |
4b1f449d | 2199 | spin_unlock_irqrestore(&n->list_lock, flags); |
497b66f2 | 2200 | return object; |
81819f0f CL |
2201 | } |
2202 | ||
2203 | /* | |
c2092c12 | 2204 | * Get a slab from somewhere. Search in increasing NUMA distances. |
81819f0f | 2205 | */ |
de3ec035 | 2206 | static void *get_any_partial(struct kmem_cache *s, gfp_t flags, |
bb192ed9 | 2207 | struct slab **ret_slab) |
81819f0f CL |
2208 | { |
2209 | #ifdef CONFIG_NUMA | |
2210 | struct zonelist *zonelist; | |
dd1a239f | 2211 | struct zoneref *z; |
54a6eb5c | 2212 | struct zone *zone; |
97a225e6 | 2213 | enum zone_type highest_zoneidx = gfp_zone(flags); |
497b66f2 | 2214 | void *object; |
cc9a6c87 | 2215 | unsigned int cpuset_mems_cookie; |
81819f0f CL |
2216 | |
2217 | /* | |
672bba3a CL |
2218 | * The defrag ratio allows a configuration of the tradeoffs between |
2219 | * inter node defragmentation and node local allocations. A lower | |
2220 | * defrag_ratio increases the tendency to do local allocations | |
2221 | * instead of attempting to obtain partial slabs from other nodes. | |
81819f0f | 2222 | * |
672bba3a CL |
2223 | * If the defrag_ratio is set to 0 then kmalloc() always |
2224 | * returns node local objects. If the ratio is higher then kmalloc() | |
2225 | * may return off node objects because partial slabs are obtained | |
2226 | * from other nodes and filled up. | |
81819f0f | 2227 | * |
43efd3ea LP |
2228 | * If /sys/kernel/slab/xx/remote_node_defrag_ratio is set to 100 |
2229 | * (which makes defrag_ratio = 1000) then every (well almost) | |
2230 | * allocation will first attempt to defrag slab caches on other nodes. | |
2231 | * This means scanning over all nodes to look for partial slabs which | |
2232 | * may be expensive if we do it every time we are trying to find a slab | |
672bba3a | 2233 | * with available objects. |
81819f0f | 2234 | */ |
9824601e CL |
2235 | if (!s->remote_node_defrag_ratio || |
2236 | get_cycles() % 1024 > s->remote_node_defrag_ratio) | |
81819f0f CL |
2237 | return NULL; |
2238 | ||
cc9a6c87 | 2239 | do { |
d26914d1 | 2240 | cpuset_mems_cookie = read_mems_allowed_begin(); |
2a389610 | 2241 | zonelist = node_zonelist(mempolicy_slab_node(), flags); |
97a225e6 | 2242 | for_each_zone_zonelist(zone, z, zonelist, highest_zoneidx) { |
cc9a6c87 MG |
2243 | struct kmem_cache_node *n; |
2244 | ||
2245 | n = get_node(s, zone_to_nid(zone)); | |
2246 | ||
dee2f8aa | 2247 | if (n && cpuset_zone_allowed(zone, flags) && |
cc9a6c87 | 2248 | n->nr_partial > s->min_partial) { |
bb192ed9 | 2249 | object = get_partial_node(s, n, ret_slab, flags); |
cc9a6c87 MG |
2250 | if (object) { |
2251 | /* | |
d26914d1 MG |
2252 | * Don't check read_mems_allowed_retry() |
2253 | * here - if mems_allowed was updated in | |
2254 | * parallel, that was a harmless race | |
2255 | * between allocation and the cpuset | |
2256 | * update | |
cc9a6c87 | 2257 | */ |
cc9a6c87 MG |
2258 | return object; |
2259 | } | |
c0ff7453 | 2260 | } |
81819f0f | 2261 | } |
d26914d1 | 2262 | } while (read_mems_allowed_retry(cpuset_mems_cookie)); |
6dfd1b65 | 2263 | #endif /* CONFIG_NUMA */ |
81819f0f CL |
2264 | return NULL; |
2265 | } | |
2266 | ||
2267 | /* | |
c2092c12 | 2268 | * Get a partial slab, lock it and return it. |
81819f0f | 2269 | */ |
497b66f2 | 2270 | static void *get_partial(struct kmem_cache *s, gfp_t flags, int node, |
bb192ed9 | 2271 | struct slab **ret_slab) |
81819f0f | 2272 | { |
497b66f2 | 2273 | void *object; |
a561ce00 JK |
2274 | int searchnode = node; |
2275 | ||
2276 | if (node == NUMA_NO_NODE) | |
2277 | searchnode = numa_mem_id(); | |
81819f0f | 2278 | |
bb192ed9 | 2279 | object = get_partial_node(s, get_node(s, searchnode), ret_slab, flags); |
497b66f2 CL |
2280 | if (object || node != NUMA_NO_NODE) |
2281 | return object; | |
81819f0f | 2282 | |
bb192ed9 | 2283 | return get_any_partial(s, flags, ret_slab); |
81819f0f CL |
2284 | } |
2285 | ||
923717cb | 2286 | #ifdef CONFIG_PREEMPTION |
8a5ec0ba | 2287 | /* |
0d645ed1 | 2288 | * Calculate the next globally unique transaction for disambiguation |
8a5ec0ba CL |
2289 | * during cmpxchg. The transactions start with the cpu number and are then |
2290 | * incremented by CONFIG_NR_CPUS. | |
2291 | */ | |
2292 | #define TID_STEP roundup_pow_of_two(CONFIG_NR_CPUS) | |
2293 | #else | |
2294 | /* | |
2295 | * No preemption supported therefore also no need to check for | |
2296 | * different cpus. | |
2297 | */ | |
2298 | #define TID_STEP 1 | |
2299 | #endif | |
2300 | ||
2301 | static inline unsigned long next_tid(unsigned long tid) | |
2302 | { | |
2303 | return tid + TID_STEP; | |
2304 | } | |
2305 | ||
9d5f0be0 | 2306 | #ifdef SLUB_DEBUG_CMPXCHG |
8a5ec0ba CL |
2307 | static inline unsigned int tid_to_cpu(unsigned long tid) |
2308 | { | |
2309 | return tid % TID_STEP; | |
2310 | } | |
2311 | ||
2312 | static inline unsigned long tid_to_event(unsigned long tid) | |
2313 | { | |
2314 | return tid / TID_STEP; | |
2315 | } | |
9d5f0be0 | 2316 | #endif |
8a5ec0ba CL |
2317 | |
2318 | static inline unsigned int init_tid(int cpu) | |
2319 | { | |
2320 | return cpu; | |
2321 | } | |
2322 | ||
2323 | static inline void note_cmpxchg_failure(const char *n, | |
2324 | const struct kmem_cache *s, unsigned long tid) | |
2325 | { | |
2326 | #ifdef SLUB_DEBUG_CMPXCHG | |
2327 | unsigned long actual_tid = __this_cpu_read(s->cpu_slab->tid); | |
2328 | ||
f9f58285 | 2329 | pr_info("%s %s: cmpxchg redo ", n, s->name); |
8a5ec0ba | 2330 | |
923717cb | 2331 | #ifdef CONFIG_PREEMPTION |
8a5ec0ba | 2332 | if (tid_to_cpu(tid) != tid_to_cpu(actual_tid)) |
f9f58285 | 2333 | pr_warn("due to cpu change %d -> %d\n", |
8a5ec0ba CL |
2334 | tid_to_cpu(tid), tid_to_cpu(actual_tid)); |
2335 | else | |
2336 | #endif | |
2337 | if (tid_to_event(tid) != tid_to_event(actual_tid)) | |
f9f58285 | 2338 | pr_warn("due to cpu running other code. Event %ld->%ld\n", |
8a5ec0ba CL |
2339 | tid_to_event(tid), tid_to_event(actual_tid)); |
2340 | else | |
f9f58285 | 2341 | pr_warn("for unknown reason: actual=%lx was=%lx target=%lx\n", |
8a5ec0ba CL |
2342 | actual_tid, tid, next_tid(tid)); |
2343 | #endif | |
4fdccdfb | 2344 | stat(s, CMPXCHG_DOUBLE_CPU_FAIL); |
8a5ec0ba CL |
2345 | } |
2346 | ||
788e1aad | 2347 | static void init_kmem_cache_cpus(struct kmem_cache *s) |
8a5ec0ba | 2348 | { |
8a5ec0ba | 2349 | int cpu; |
bd0e7491 | 2350 | struct kmem_cache_cpu *c; |
8a5ec0ba | 2351 | |
bd0e7491 VB |
2352 | for_each_possible_cpu(cpu) { |
2353 | c = per_cpu_ptr(s->cpu_slab, cpu); | |
2354 | local_lock_init(&c->lock); | |
2355 | c->tid = init_tid(cpu); | |
2356 | } | |
8a5ec0ba | 2357 | } |
2cfb7455 | 2358 | |
81819f0f | 2359 | /* |
c2092c12 | 2360 | * Finishes removing the cpu slab. Merges cpu's freelist with slab's freelist, |
a019d201 VB |
2361 | * unfreezes the slabs and puts it on the proper list. |
2362 | * Assumes the slab has been already safely taken away from kmem_cache_cpu | |
2363 | * by the caller. | |
81819f0f | 2364 | */ |
bb192ed9 | 2365 | static void deactivate_slab(struct kmem_cache *s, struct slab *slab, |
a019d201 | 2366 | void *freelist) |
81819f0f | 2367 | { |
6d3a16d0 | 2368 | enum slab_modes { M_NONE, M_PARTIAL, M_FULL, M_FREE, M_FULL_NOLIST }; |
bb192ed9 | 2369 | struct kmem_cache_node *n = get_node(s, slab_nid(slab)); |
6d3a16d0 HY |
2370 | int free_delta = 0; |
2371 | enum slab_modes mode = M_NONE; | |
d930ff03 | 2372 | void *nextfree, *freelist_iter, *freelist_tail; |
136333d1 | 2373 | int tail = DEACTIVATE_TO_HEAD; |
3406e91b | 2374 | unsigned long flags = 0; |
bb192ed9 VB |
2375 | struct slab new; |
2376 | struct slab old; | |
2cfb7455 | 2377 | |
bb192ed9 | 2378 | if (slab->freelist) { |
84e554e6 | 2379 | stat(s, DEACTIVATE_REMOTE_FREES); |
136333d1 | 2380 | tail = DEACTIVATE_TO_TAIL; |
2cfb7455 CL |
2381 | } |
2382 | ||
894b8788 | 2383 | /* |
d930ff03 VB |
2384 | * Stage one: Count the objects on cpu's freelist as free_delta and |
2385 | * remember the last object in freelist_tail for later splicing. | |
2cfb7455 | 2386 | */ |
d930ff03 VB |
2387 | freelist_tail = NULL; |
2388 | freelist_iter = freelist; | |
2389 | while (freelist_iter) { | |
2390 | nextfree = get_freepointer(s, freelist_iter); | |
2cfb7455 | 2391 | |
52f23478 DZ |
2392 | /* |
2393 | * If 'nextfree' is invalid, it is possible that the object at | |
d930ff03 VB |
2394 | * 'freelist_iter' is already corrupted. So isolate all objects |
2395 | * starting at 'freelist_iter' by skipping them. | |
52f23478 | 2396 | */ |
bb192ed9 | 2397 | if (freelist_corrupted(s, slab, &freelist_iter, nextfree)) |
52f23478 DZ |
2398 | break; |
2399 | ||
d930ff03 VB |
2400 | freelist_tail = freelist_iter; |
2401 | free_delta++; | |
2cfb7455 | 2402 | |
d930ff03 | 2403 | freelist_iter = nextfree; |
2cfb7455 CL |
2404 | } |
2405 | ||
894b8788 | 2406 | /* |
c2092c12 VB |
2407 | * Stage two: Unfreeze the slab while splicing the per-cpu |
2408 | * freelist to the head of slab's freelist. | |
d930ff03 | 2409 | * |
c2092c12 | 2410 | * Ensure that the slab is unfrozen while the list presence |
d930ff03 | 2411 | * reflects the actual number of objects during unfreeze. |
2cfb7455 | 2412 | * |
6d3a16d0 HY |
2413 | * We first perform cmpxchg holding lock and insert to list |
2414 | * when it succeed. If there is mismatch then the slab is not | |
2415 | * unfrozen and number of objects in the slab may have changed. | |
2416 | * Then release lock and retry cmpxchg again. | |
894b8788 | 2417 | */ |
2cfb7455 | 2418 | redo: |
894b8788 | 2419 | |
bb192ed9 VB |
2420 | old.freelist = READ_ONCE(slab->freelist); |
2421 | old.counters = READ_ONCE(slab->counters); | |
a0132ac0 | 2422 | VM_BUG_ON(!old.frozen); |
7c2e132c | 2423 | |
2cfb7455 CL |
2424 | /* Determine target state of the slab */ |
2425 | new.counters = old.counters; | |
d930ff03 VB |
2426 | if (freelist_tail) { |
2427 | new.inuse -= free_delta; | |
2428 | set_freepointer(s, freelist_tail, old.freelist); | |
2cfb7455 CL |
2429 | new.freelist = freelist; |
2430 | } else | |
2431 | new.freelist = old.freelist; | |
2432 | ||
2433 | new.frozen = 0; | |
2434 | ||
6d3a16d0 HY |
2435 | if (!new.inuse && n->nr_partial >= s->min_partial) { |
2436 | mode = M_FREE; | |
2437 | } else if (new.freelist) { | |
2438 | mode = M_PARTIAL; | |
2439 | /* | |
2440 | * Taking the spinlock removes the possibility that | |
2441 | * acquire_slab() will see a slab that is frozen | |
2442 | */ | |
2443 | spin_lock_irqsave(&n->list_lock, flags); | |
2444 | } else if (kmem_cache_debug_flags(s, SLAB_STORE_USER)) { | |
2445 | mode = M_FULL; | |
2446 | /* | |
2447 | * This also ensures that the scanning of full | |
2448 | * slabs from diagnostic functions will not see | |
2449 | * any frozen slabs. | |
2450 | */ | |
2451 | spin_lock_irqsave(&n->list_lock, flags); | |
2cfb7455 | 2452 | } else { |
6d3a16d0 | 2453 | mode = M_FULL_NOLIST; |
2cfb7455 CL |
2454 | } |
2455 | ||
2cfb7455 | 2456 | |
bb192ed9 | 2457 | if (!cmpxchg_double_slab(s, slab, |
2cfb7455 CL |
2458 | old.freelist, old.counters, |
2459 | new.freelist, new.counters, | |
6d3a16d0 HY |
2460 | "unfreezing slab")) { |
2461 | if (mode == M_PARTIAL || mode == M_FULL) | |
2462 | spin_unlock_irqrestore(&n->list_lock, flags); | |
2cfb7455 | 2463 | goto redo; |
6d3a16d0 | 2464 | } |
2cfb7455 | 2465 | |
2cfb7455 | 2466 | |
6d3a16d0 HY |
2467 | if (mode == M_PARTIAL) { |
2468 | add_partial(n, slab, tail); | |
2469 | spin_unlock_irqrestore(&n->list_lock, flags); | |
88349a28 | 2470 | stat(s, tail); |
6d3a16d0 | 2471 | } else if (mode == M_FREE) { |
2cfb7455 | 2472 | stat(s, DEACTIVATE_EMPTY); |
bb192ed9 | 2473 | discard_slab(s, slab); |
2cfb7455 | 2474 | stat(s, FREE_SLAB); |
6d3a16d0 HY |
2475 | } else if (mode == M_FULL) { |
2476 | add_full(s, n, slab); | |
2477 | spin_unlock_irqrestore(&n->list_lock, flags); | |
2478 | stat(s, DEACTIVATE_FULL); | |
2479 | } else if (mode == M_FULL_NOLIST) { | |
2480 | stat(s, DEACTIVATE_FULL); | |
894b8788 | 2481 | } |
81819f0f CL |
2482 | } |
2483 | ||
345c905d | 2484 | #ifdef CONFIG_SLUB_CPU_PARTIAL |
bb192ed9 | 2485 | static void __unfreeze_partials(struct kmem_cache *s, struct slab *partial_slab) |
fc1455f4 | 2486 | { |
43d77867 | 2487 | struct kmem_cache_node *n = NULL, *n2 = NULL; |
bb192ed9 | 2488 | struct slab *slab, *slab_to_discard = NULL; |
7cf9f3ba | 2489 | unsigned long flags = 0; |
49e22585 | 2490 | |
bb192ed9 VB |
2491 | while (partial_slab) { |
2492 | struct slab new; | |
2493 | struct slab old; | |
49e22585 | 2494 | |
bb192ed9 VB |
2495 | slab = partial_slab; |
2496 | partial_slab = slab->next; | |
43d77867 | 2497 | |
bb192ed9 | 2498 | n2 = get_node(s, slab_nid(slab)); |
43d77867 JK |
2499 | if (n != n2) { |
2500 | if (n) | |
7cf9f3ba | 2501 | spin_unlock_irqrestore(&n->list_lock, flags); |
43d77867 JK |
2502 | |
2503 | n = n2; | |
7cf9f3ba | 2504 | spin_lock_irqsave(&n->list_lock, flags); |
43d77867 | 2505 | } |
49e22585 CL |
2506 | |
2507 | do { | |
2508 | ||
bb192ed9 VB |
2509 | old.freelist = slab->freelist; |
2510 | old.counters = slab->counters; | |
a0132ac0 | 2511 | VM_BUG_ON(!old.frozen); |
49e22585 CL |
2512 | |
2513 | new.counters = old.counters; | |
2514 | new.freelist = old.freelist; | |
2515 | ||
2516 | new.frozen = 0; | |
2517 | ||
bb192ed9 | 2518 | } while (!__cmpxchg_double_slab(s, slab, |
49e22585 CL |
2519 | old.freelist, old.counters, |
2520 | new.freelist, new.counters, | |
2521 | "unfreezing slab")); | |
2522 | ||
8a5b20ae | 2523 | if (unlikely(!new.inuse && n->nr_partial >= s->min_partial)) { |
bb192ed9 VB |
2524 | slab->next = slab_to_discard; |
2525 | slab_to_discard = slab; | |
43d77867 | 2526 | } else { |
bb192ed9 | 2527 | add_partial(n, slab, DEACTIVATE_TO_TAIL); |
43d77867 | 2528 | stat(s, FREE_ADD_PARTIAL); |
49e22585 CL |
2529 | } |
2530 | } | |
2531 | ||
2532 | if (n) | |
7cf9f3ba | 2533 | spin_unlock_irqrestore(&n->list_lock, flags); |
8de06a6f | 2534 | |
bb192ed9 VB |
2535 | while (slab_to_discard) { |
2536 | slab = slab_to_discard; | |
2537 | slab_to_discard = slab_to_discard->next; | |
9ada1934 SL |
2538 | |
2539 | stat(s, DEACTIVATE_EMPTY); | |
bb192ed9 | 2540 | discard_slab(s, slab); |
9ada1934 SL |
2541 | stat(s, FREE_SLAB); |
2542 | } | |
fc1455f4 | 2543 | } |
f3ab8b6b | 2544 | |
fc1455f4 VB |
2545 | /* |
2546 | * Unfreeze all the cpu partial slabs. | |
2547 | */ | |
2548 | static void unfreeze_partials(struct kmem_cache *s) | |
2549 | { | |
bb192ed9 | 2550 | struct slab *partial_slab; |
fc1455f4 VB |
2551 | unsigned long flags; |
2552 | ||
bd0e7491 | 2553 | local_lock_irqsave(&s->cpu_slab->lock, flags); |
bb192ed9 | 2554 | partial_slab = this_cpu_read(s->cpu_slab->partial); |
fc1455f4 | 2555 | this_cpu_write(s->cpu_slab->partial, NULL); |
bd0e7491 | 2556 | local_unlock_irqrestore(&s->cpu_slab->lock, flags); |
fc1455f4 | 2557 | |
bb192ed9 VB |
2558 | if (partial_slab) |
2559 | __unfreeze_partials(s, partial_slab); | |
fc1455f4 VB |
2560 | } |
2561 | ||
2562 | static void unfreeze_partials_cpu(struct kmem_cache *s, | |
2563 | struct kmem_cache_cpu *c) | |
2564 | { | |
bb192ed9 | 2565 | struct slab *partial_slab; |
fc1455f4 | 2566 | |
bb192ed9 | 2567 | partial_slab = slub_percpu_partial(c); |
fc1455f4 VB |
2568 | c->partial = NULL; |
2569 | ||
bb192ed9 VB |
2570 | if (partial_slab) |
2571 | __unfreeze_partials(s, partial_slab); | |
49e22585 CL |
2572 | } |
2573 | ||
2574 | /* | |
c2092c12 VB |
2575 | * Put a slab that was just frozen (in __slab_free|get_partial_node) into a |
2576 | * partial slab slot if available. | |
49e22585 CL |
2577 | * |
2578 | * If we did not find a slot then simply move all the partials to the | |
2579 | * per node partial list. | |
2580 | */ | |
bb192ed9 | 2581 | static void put_cpu_partial(struct kmem_cache *s, struct slab *slab, int drain) |
49e22585 | 2582 | { |
bb192ed9 VB |
2583 | struct slab *oldslab; |
2584 | struct slab *slab_to_unfreeze = NULL; | |
e0a043aa | 2585 | unsigned long flags; |
bb192ed9 | 2586 | int slabs = 0; |
49e22585 | 2587 | |
bd0e7491 | 2588 | local_lock_irqsave(&s->cpu_slab->lock, flags); |
49e22585 | 2589 | |
bb192ed9 | 2590 | oldslab = this_cpu_read(s->cpu_slab->partial); |
e0a043aa | 2591 | |
bb192ed9 VB |
2592 | if (oldslab) { |
2593 | if (drain && oldslab->slabs >= s->cpu_partial_slabs) { | |
e0a043aa VB |
2594 | /* |
2595 | * Partial array is full. Move the existing set to the | |
2596 | * per node partial list. Postpone the actual unfreezing | |
2597 | * outside of the critical section. | |
2598 | */ | |
bb192ed9 VB |
2599 | slab_to_unfreeze = oldslab; |
2600 | oldslab = NULL; | |
e0a043aa | 2601 | } else { |
bb192ed9 | 2602 | slabs = oldslab->slabs; |
49e22585 | 2603 | } |
e0a043aa | 2604 | } |
49e22585 | 2605 | |
bb192ed9 | 2606 | slabs++; |
49e22585 | 2607 | |
bb192ed9 VB |
2608 | slab->slabs = slabs; |
2609 | slab->next = oldslab; | |
49e22585 | 2610 | |
bb192ed9 | 2611 | this_cpu_write(s->cpu_slab->partial, slab); |
e0a043aa | 2612 | |
bd0e7491 | 2613 | local_unlock_irqrestore(&s->cpu_slab->lock, flags); |
e0a043aa | 2614 | |
bb192ed9 VB |
2615 | if (slab_to_unfreeze) { |
2616 | __unfreeze_partials(s, slab_to_unfreeze); | |
e0a043aa VB |
2617 | stat(s, CPU_PARTIAL_DRAIN); |
2618 | } | |
49e22585 CL |
2619 | } |
2620 | ||
e0a043aa VB |
2621 | #else /* CONFIG_SLUB_CPU_PARTIAL */ |
2622 | ||
2623 | static inline void unfreeze_partials(struct kmem_cache *s) { } | |
2624 | static inline void unfreeze_partials_cpu(struct kmem_cache *s, | |
2625 | struct kmem_cache_cpu *c) { } | |
2626 | ||
2627 | #endif /* CONFIG_SLUB_CPU_PARTIAL */ | |
2628 | ||
dfb4f096 | 2629 | static inline void flush_slab(struct kmem_cache *s, struct kmem_cache_cpu *c) |
81819f0f | 2630 | { |
5a836bf6 | 2631 | unsigned long flags; |
bb192ed9 | 2632 | struct slab *slab; |
5a836bf6 SAS |
2633 | void *freelist; |
2634 | ||
bd0e7491 | 2635 | local_lock_irqsave(&s->cpu_slab->lock, flags); |
5a836bf6 | 2636 | |
bb192ed9 | 2637 | slab = c->slab; |
5a836bf6 | 2638 | freelist = c->freelist; |
c17dda40 | 2639 | |
bb192ed9 | 2640 | c->slab = NULL; |
a019d201 | 2641 | c->freelist = NULL; |
c17dda40 | 2642 | c->tid = next_tid(c->tid); |
a019d201 | 2643 | |
bd0e7491 | 2644 | local_unlock_irqrestore(&s->cpu_slab->lock, flags); |
a019d201 | 2645 | |
bb192ed9 VB |
2646 | if (slab) { |
2647 | deactivate_slab(s, slab, freelist); | |
5a836bf6 SAS |
2648 | stat(s, CPUSLAB_FLUSH); |
2649 | } | |
81819f0f CL |
2650 | } |
2651 | ||
0c710013 | 2652 | static inline void __flush_cpu_slab(struct kmem_cache *s, int cpu) |
81819f0f | 2653 | { |
9dfc6e68 | 2654 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu); |
08beb547 | 2655 | void *freelist = c->freelist; |
bb192ed9 | 2656 | struct slab *slab = c->slab; |
81819f0f | 2657 | |
bb192ed9 | 2658 | c->slab = NULL; |
08beb547 VB |
2659 | c->freelist = NULL; |
2660 | c->tid = next_tid(c->tid); | |
2661 | ||
bb192ed9 VB |
2662 | if (slab) { |
2663 | deactivate_slab(s, slab, freelist); | |
08beb547 VB |
2664 | stat(s, CPUSLAB_FLUSH); |
2665 | } | |
49e22585 | 2666 | |
fc1455f4 | 2667 | unfreeze_partials_cpu(s, c); |
81819f0f CL |
2668 | } |
2669 | ||
5a836bf6 SAS |
2670 | struct slub_flush_work { |
2671 | struct work_struct work; | |
2672 | struct kmem_cache *s; | |
2673 | bool skip; | |
2674 | }; | |
2675 | ||
fc1455f4 VB |
2676 | /* |
2677 | * Flush cpu slab. | |
2678 | * | |
5a836bf6 | 2679 | * Called from CPU work handler with migration disabled. |
fc1455f4 | 2680 | */ |
5a836bf6 | 2681 | static void flush_cpu_slab(struct work_struct *w) |
81819f0f | 2682 | { |
5a836bf6 SAS |
2683 | struct kmem_cache *s; |
2684 | struct kmem_cache_cpu *c; | |
2685 | struct slub_flush_work *sfw; | |
2686 | ||
2687 | sfw = container_of(w, struct slub_flush_work, work); | |
2688 | ||
2689 | s = sfw->s; | |
2690 | c = this_cpu_ptr(s->cpu_slab); | |
fc1455f4 | 2691 | |
bb192ed9 | 2692 | if (c->slab) |
fc1455f4 | 2693 | flush_slab(s, c); |
81819f0f | 2694 | |
fc1455f4 | 2695 | unfreeze_partials(s); |
81819f0f CL |
2696 | } |
2697 | ||
5a836bf6 | 2698 | static bool has_cpu_slab(int cpu, struct kmem_cache *s) |
a8364d55 | 2699 | { |
a8364d55 GBY |
2700 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu); |
2701 | ||
bb192ed9 | 2702 | return c->slab || slub_percpu_partial(c); |
a8364d55 GBY |
2703 | } |
2704 | ||
5a836bf6 SAS |
2705 | static DEFINE_MUTEX(flush_lock); |
2706 | static DEFINE_PER_CPU(struct slub_flush_work, slub_flush); | |
2707 | ||
2708 | static void flush_all_cpus_locked(struct kmem_cache *s) | |
2709 | { | |
2710 | struct slub_flush_work *sfw; | |
2711 | unsigned int cpu; | |
2712 | ||
2713 | lockdep_assert_cpus_held(); | |
2714 | mutex_lock(&flush_lock); | |
2715 | ||
2716 | for_each_online_cpu(cpu) { | |
2717 | sfw = &per_cpu(slub_flush, cpu); | |
2718 | if (!has_cpu_slab(cpu, s)) { | |
2719 | sfw->skip = true; | |
2720 | continue; | |
2721 | } | |
2722 | INIT_WORK(&sfw->work, flush_cpu_slab); | |
2723 | sfw->skip = false; | |
2724 | sfw->s = s; | |
2725 | schedule_work_on(cpu, &sfw->work); | |
2726 | } | |
2727 | ||
2728 | for_each_online_cpu(cpu) { | |
2729 | sfw = &per_cpu(slub_flush, cpu); | |
2730 | if (sfw->skip) | |
2731 | continue; | |
2732 | flush_work(&sfw->work); | |
2733 | } | |
2734 | ||
2735 | mutex_unlock(&flush_lock); | |
2736 | } | |
2737 | ||
81819f0f CL |
2738 | static void flush_all(struct kmem_cache *s) |
2739 | { | |
5a836bf6 SAS |
2740 | cpus_read_lock(); |
2741 | flush_all_cpus_locked(s); | |
2742 | cpus_read_unlock(); | |
81819f0f CL |
2743 | } |
2744 | ||
a96a87bf SAS |
2745 | /* |
2746 | * Use the cpu notifier to insure that the cpu slabs are flushed when | |
2747 | * necessary. | |
2748 | */ | |
2749 | static int slub_cpu_dead(unsigned int cpu) | |
2750 | { | |
2751 | struct kmem_cache *s; | |
a96a87bf SAS |
2752 | |
2753 | mutex_lock(&slab_mutex); | |
0e7ac738 | 2754 | list_for_each_entry(s, &slab_caches, list) |
a96a87bf | 2755 | __flush_cpu_slab(s, cpu); |
a96a87bf SAS |
2756 | mutex_unlock(&slab_mutex); |
2757 | return 0; | |
2758 | } | |
2759 | ||
dfb4f096 CL |
2760 | /* |
2761 | * Check if the objects in a per cpu structure fit numa | |
2762 | * locality expectations. | |
2763 | */ | |
bb192ed9 | 2764 | static inline int node_match(struct slab *slab, int node) |
dfb4f096 CL |
2765 | { |
2766 | #ifdef CONFIG_NUMA | |
bb192ed9 | 2767 | if (node != NUMA_NO_NODE && slab_nid(slab) != node) |
dfb4f096 CL |
2768 | return 0; |
2769 | #endif | |
2770 | return 1; | |
2771 | } | |
2772 | ||
9a02d699 | 2773 | #ifdef CONFIG_SLUB_DEBUG |
bb192ed9 | 2774 | static int count_free(struct slab *slab) |
781b2ba6 | 2775 | { |
bb192ed9 | 2776 | return slab->objects - slab->inuse; |
781b2ba6 PE |
2777 | } |
2778 | ||
9a02d699 DR |
2779 | static inline unsigned long node_nr_objs(struct kmem_cache_node *n) |
2780 | { | |
2781 | return atomic_long_read(&n->total_objects); | |
2782 | } | |
2783 | #endif /* CONFIG_SLUB_DEBUG */ | |
2784 | ||
2785 | #if defined(CONFIG_SLUB_DEBUG) || defined(CONFIG_SYSFS) | |
781b2ba6 | 2786 | static unsigned long count_partial(struct kmem_cache_node *n, |
bb192ed9 | 2787 | int (*get_count)(struct slab *)) |
781b2ba6 PE |
2788 | { |
2789 | unsigned long flags; | |
2790 | unsigned long x = 0; | |
bb192ed9 | 2791 | struct slab *slab; |
781b2ba6 PE |
2792 | |
2793 | spin_lock_irqsave(&n->list_lock, flags); | |
bb192ed9 VB |
2794 | list_for_each_entry(slab, &n->partial, slab_list) |
2795 | x += get_count(slab); | |
781b2ba6 PE |
2796 | spin_unlock_irqrestore(&n->list_lock, flags); |
2797 | return x; | |
2798 | } | |
9a02d699 | 2799 | #endif /* CONFIG_SLUB_DEBUG || CONFIG_SYSFS */ |
26c02cf0 | 2800 | |
781b2ba6 PE |
2801 | static noinline void |
2802 | slab_out_of_memory(struct kmem_cache *s, gfp_t gfpflags, int nid) | |
2803 | { | |
9a02d699 DR |
2804 | #ifdef CONFIG_SLUB_DEBUG |
2805 | static DEFINE_RATELIMIT_STATE(slub_oom_rs, DEFAULT_RATELIMIT_INTERVAL, | |
2806 | DEFAULT_RATELIMIT_BURST); | |
781b2ba6 | 2807 | int node; |
fa45dc25 | 2808 | struct kmem_cache_node *n; |
781b2ba6 | 2809 | |
9a02d699 DR |
2810 | if ((gfpflags & __GFP_NOWARN) || !__ratelimit(&slub_oom_rs)) |
2811 | return; | |
2812 | ||
5b3810e5 VB |
2813 | pr_warn("SLUB: Unable to allocate memory on node %d, gfp=%#x(%pGg)\n", |
2814 | nid, gfpflags, &gfpflags); | |
19af27af | 2815 | pr_warn(" cache: %s, object size: %u, buffer size: %u, default order: %u, min order: %u\n", |
f9f58285 FF |
2816 | s->name, s->object_size, s->size, oo_order(s->oo), |
2817 | oo_order(s->min)); | |
781b2ba6 | 2818 | |
3b0efdfa | 2819 | if (oo_order(s->min) > get_order(s->object_size)) |
f9f58285 FF |
2820 | pr_warn(" %s debugging increased min order, use slub_debug=O to disable.\n", |
2821 | s->name); | |
fa5ec8a1 | 2822 | |
fa45dc25 | 2823 | for_each_kmem_cache_node(s, node, n) { |
781b2ba6 PE |
2824 | unsigned long nr_slabs; |
2825 | unsigned long nr_objs; | |
2826 | unsigned long nr_free; | |
2827 | ||
26c02cf0 AB |
2828 | nr_free = count_partial(n, count_free); |
2829 | nr_slabs = node_nr_slabs(n); | |
2830 | nr_objs = node_nr_objs(n); | |
781b2ba6 | 2831 | |
f9f58285 | 2832 | pr_warn(" node %d: slabs: %ld, objs: %ld, free: %ld\n", |
781b2ba6 PE |
2833 | node, nr_slabs, nr_objs, nr_free); |
2834 | } | |
9a02d699 | 2835 | #endif |
781b2ba6 PE |
2836 | } |
2837 | ||
01b34d16 | 2838 | static inline bool pfmemalloc_match(struct slab *slab, gfp_t gfpflags) |
072bb0aa | 2839 | { |
01b34d16 | 2840 | if (unlikely(slab_test_pfmemalloc(slab))) |
0b303fb4 VB |
2841 | return gfp_pfmemalloc_allowed(gfpflags); |
2842 | ||
2843 | return true; | |
2844 | } | |
2845 | ||
213eeb9f | 2846 | /* |
c2092c12 VB |
2847 | * Check the slab->freelist and either transfer the freelist to the |
2848 | * per cpu freelist or deactivate the slab. | |
213eeb9f | 2849 | * |
c2092c12 | 2850 | * The slab is still frozen if the return value is not NULL. |
213eeb9f | 2851 | * |
c2092c12 | 2852 | * If this function returns NULL then the slab has been unfrozen. |
213eeb9f | 2853 | */ |
bb192ed9 | 2854 | static inline void *get_freelist(struct kmem_cache *s, struct slab *slab) |
213eeb9f | 2855 | { |
bb192ed9 | 2856 | struct slab new; |
213eeb9f CL |
2857 | unsigned long counters; |
2858 | void *freelist; | |
2859 | ||
bd0e7491 VB |
2860 | lockdep_assert_held(this_cpu_ptr(&s->cpu_slab->lock)); |
2861 | ||
213eeb9f | 2862 | do { |
bb192ed9 VB |
2863 | freelist = slab->freelist; |
2864 | counters = slab->counters; | |
6faa6833 | 2865 | |
213eeb9f | 2866 | new.counters = counters; |
a0132ac0 | 2867 | VM_BUG_ON(!new.frozen); |
213eeb9f | 2868 | |
bb192ed9 | 2869 | new.inuse = slab->objects; |
213eeb9f CL |
2870 | new.frozen = freelist != NULL; |
2871 | ||
bb192ed9 | 2872 | } while (!__cmpxchg_double_slab(s, slab, |
213eeb9f CL |
2873 | freelist, counters, |
2874 | NULL, new.counters, | |
2875 | "get_freelist")); | |
2876 | ||
2877 | return freelist; | |
2878 | } | |
2879 | ||
81819f0f | 2880 | /* |
894b8788 CL |
2881 | * Slow path. The lockless freelist is empty or we need to perform |
2882 | * debugging duties. | |
2883 | * | |
894b8788 CL |
2884 | * Processing is still very fast if new objects have been freed to the |
2885 | * regular freelist. In that case we simply take over the regular freelist | |
2886 | * as the lockless freelist and zap the regular freelist. | |
81819f0f | 2887 | * |
894b8788 CL |
2888 | * If that is not working then we fall back to the partial lists. We take the |
2889 | * first element of the freelist as the object to allocate now and move the | |
2890 | * rest of the freelist to the lockless freelist. | |
81819f0f | 2891 | * |
894b8788 | 2892 | * And if we were unable to get a new slab from the partial slab lists then |
6446faa2 CL |
2893 | * we need to allocate a new slab. This is the slowest path since it involves |
2894 | * a call to the page allocator and the setup of a new slab. | |
a380a3c7 | 2895 | * |
e500059b | 2896 | * Version of __slab_alloc to use when we know that preemption is |
a380a3c7 | 2897 | * already disabled (which is the case for bulk allocation). |
81819f0f | 2898 | */ |
a380a3c7 | 2899 | static void *___slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node, |
ce71e27c | 2900 | unsigned long addr, struct kmem_cache_cpu *c) |
81819f0f | 2901 | { |
6faa6833 | 2902 | void *freelist; |
bb192ed9 | 2903 | struct slab *slab; |
e500059b | 2904 | unsigned long flags; |
81819f0f | 2905 | |
9f986d99 AW |
2906 | stat(s, ALLOC_SLOWPATH); |
2907 | ||
c2092c12 | 2908 | reread_slab: |
0b303fb4 | 2909 | |
bb192ed9 VB |
2910 | slab = READ_ONCE(c->slab); |
2911 | if (!slab) { | |
0715e6c5 VB |
2912 | /* |
2913 | * if the node is not online or has no normal memory, just | |
2914 | * ignore the node constraint | |
2915 | */ | |
2916 | if (unlikely(node != NUMA_NO_NODE && | |
7e1fa93d | 2917 | !node_isset(node, slab_nodes))) |
0715e6c5 | 2918 | node = NUMA_NO_NODE; |
81819f0f | 2919 | goto new_slab; |
0715e6c5 | 2920 | } |
49e22585 | 2921 | redo: |
6faa6833 | 2922 | |
bb192ed9 | 2923 | if (unlikely(!node_match(slab, node))) { |
0715e6c5 VB |
2924 | /* |
2925 | * same as above but node_match() being false already | |
2926 | * implies node != NUMA_NO_NODE | |
2927 | */ | |
7e1fa93d | 2928 | if (!node_isset(node, slab_nodes)) { |
0715e6c5 | 2929 | node = NUMA_NO_NODE; |
0715e6c5 | 2930 | } else { |
a561ce00 | 2931 | stat(s, ALLOC_NODE_MISMATCH); |
0b303fb4 | 2932 | goto deactivate_slab; |
a561ce00 | 2933 | } |
fc59c053 | 2934 | } |
6446faa2 | 2935 | |
072bb0aa MG |
2936 | /* |
2937 | * By rights, we should be searching for a slab page that was | |
2938 | * PFMEMALLOC but right now, we are losing the pfmemalloc | |
2939 | * information when the page leaves the per-cpu allocator | |
2940 | */ | |
bb192ed9 | 2941 | if (unlikely(!pfmemalloc_match(slab, gfpflags))) |
0b303fb4 | 2942 | goto deactivate_slab; |
072bb0aa | 2943 | |
c2092c12 | 2944 | /* must check again c->slab in case we got preempted and it changed */ |
bd0e7491 | 2945 | local_lock_irqsave(&s->cpu_slab->lock, flags); |
bb192ed9 | 2946 | if (unlikely(slab != c->slab)) { |
bd0e7491 | 2947 | local_unlock_irqrestore(&s->cpu_slab->lock, flags); |
c2092c12 | 2948 | goto reread_slab; |
0b303fb4 | 2949 | } |
6faa6833 CL |
2950 | freelist = c->freelist; |
2951 | if (freelist) | |
73736e03 | 2952 | goto load_freelist; |
03e404af | 2953 | |
bb192ed9 | 2954 | freelist = get_freelist(s, slab); |
6446faa2 | 2955 | |
6faa6833 | 2956 | if (!freelist) { |
bb192ed9 | 2957 | c->slab = NULL; |
eeaa345e | 2958 | c->tid = next_tid(c->tid); |
bd0e7491 | 2959 | local_unlock_irqrestore(&s->cpu_slab->lock, flags); |
03e404af | 2960 | stat(s, DEACTIVATE_BYPASS); |
fc59c053 | 2961 | goto new_slab; |
03e404af | 2962 | } |
6446faa2 | 2963 | |
84e554e6 | 2964 | stat(s, ALLOC_REFILL); |
6446faa2 | 2965 | |
894b8788 | 2966 | load_freelist: |
0b303fb4 | 2967 | |
bd0e7491 | 2968 | lockdep_assert_held(this_cpu_ptr(&s->cpu_slab->lock)); |
0b303fb4 | 2969 | |
507effea CL |
2970 | /* |
2971 | * freelist is pointing to the list of objects to be used. | |
c2092c12 VB |
2972 | * slab is pointing to the slab from which the objects are obtained. |
2973 | * That slab must be frozen for per cpu allocations to work. | |
507effea | 2974 | */ |
bb192ed9 | 2975 | VM_BUG_ON(!c->slab->frozen); |
6faa6833 | 2976 | c->freelist = get_freepointer(s, freelist); |
8a5ec0ba | 2977 | c->tid = next_tid(c->tid); |
bd0e7491 | 2978 | local_unlock_irqrestore(&s->cpu_slab->lock, flags); |
6faa6833 | 2979 | return freelist; |
81819f0f | 2980 | |
0b303fb4 VB |
2981 | deactivate_slab: |
2982 | ||
bd0e7491 | 2983 | local_lock_irqsave(&s->cpu_slab->lock, flags); |
bb192ed9 | 2984 | if (slab != c->slab) { |
bd0e7491 | 2985 | local_unlock_irqrestore(&s->cpu_slab->lock, flags); |
c2092c12 | 2986 | goto reread_slab; |
0b303fb4 | 2987 | } |
a019d201 | 2988 | freelist = c->freelist; |
bb192ed9 | 2989 | c->slab = NULL; |
a019d201 | 2990 | c->freelist = NULL; |
eeaa345e | 2991 | c->tid = next_tid(c->tid); |
bd0e7491 | 2992 | local_unlock_irqrestore(&s->cpu_slab->lock, flags); |
bb192ed9 | 2993 | deactivate_slab(s, slab, freelist); |
0b303fb4 | 2994 | |
81819f0f | 2995 | new_slab: |
2cfb7455 | 2996 | |
a93cf07b | 2997 | if (slub_percpu_partial(c)) { |
bd0e7491 | 2998 | local_lock_irqsave(&s->cpu_slab->lock, flags); |
bb192ed9 | 2999 | if (unlikely(c->slab)) { |
bd0e7491 | 3000 | local_unlock_irqrestore(&s->cpu_slab->lock, flags); |
c2092c12 | 3001 | goto reread_slab; |
fa417ab7 | 3002 | } |
4b1f449d | 3003 | if (unlikely(!slub_percpu_partial(c))) { |
bd0e7491 | 3004 | local_unlock_irqrestore(&s->cpu_slab->lock, flags); |
25c00c50 VB |
3005 | /* we were preempted and partial list got empty */ |
3006 | goto new_objects; | |
4b1f449d | 3007 | } |
fa417ab7 | 3008 | |
bb192ed9 VB |
3009 | slab = c->slab = slub_percpu_partial(c); |
3010 | slub_set_percpu_partial(c, slab); | |
bd0e7491 | 3011 | local_unlock_irqrestore(&s->cpu_slab->lock, flags); |
49e22585 | 3012 | stat(s, CPU_PARTIAL_ALLOC); |
49e22585 | 3013 | goto redo; |
81819f0f CL |
3014 | } |
3015 | ||
fa417ab7 VB |
3016 | new_objects: |
3017 | ||
bb192ed9 | 3018 | freelist = get_partial(s, gfpflags, node, &slab); |
3f2b77e3 | 3019 | if (freelist) |
c2092c12 | 3020 | goto check_new_slab; |
2a904905 | 3021 | |
25c00c50 | 3022 | slub_put_cpu_ptr(s->cpu_slab); |
bb192ed9 | 3023 | slab = new_slab(s, gfpflags, node); |
25c00c50 | 3024 | c = slub_get_cpu_ptr(s->cpu_slab); |
01ad8a7b | 3025 | |
bb192ed9 | 3026 | if (unlikely(!slab)) { |
9a02d699 | 3027 | slab_out_of_memory(s, gfpflags, node); |
f4697436 | 3028 | return NULL; |
81819f0f | 3029 | } |
2cfb7455 | 3030 | |
53a0de06 | 3031 | /* |
c2092c12 | 3032 | * No other reference to the slab yet so we can |
53a0de06 VB |
3033 | * muck around with it freely without cmpxchg |
3034 | */ | |
bb192ed9 VB |
3035 | freelist = slab->freelist; |
3036 | slab->freelist = NULL; | |
53a0de06 VB |
3037 | |
3038 | stat(s, ALLOC_SLAB); | |
53a0de06 | 3039 | |
c2092c12 | 3040 | check_new_slab: |
2cfb7455 | 3041 | |
1572df7c | 3042 | if (kmem_cache_debug(s)) { |
bb192ed9 | 3043 | if (!alloc_debug_processing(s, slab, freelist, addr)) { |
1572df7c VB |
3044 | /* Slab failed checks. Next slab needed */ |
3045 | goto new_slab; | |
fa417ab7 | 3046 | } else { |
1572df7c VB |
3047 | /* |
3048 | * For debug case, we don't load freelist so that all | |
3049 | * allocations go through alloc_debug_processing() | |
3050 | */ | |
3051 | goto return_single; | |
fa417ab7 | 3052 | } |
1572df7c VB |
3053 | } |
3054 | ||
bb192ed9 | 3055 | if (unlikely(!pfmemalloc_match(slab, gfpflags))) |
1572df7c VB |
3056 | /* |
3057 | * For !pfmemalloc_match() case we don't load freelist so that | |
3058 | * we don't make further mismatched allocations easier. | |
3059 | */ | |
3060 | goto return_single; | |
3061 | ||
c2092c12 | 3062 | retry_load_slab: |
cfdf836e | 3063 | |
bd0e7491 | 3064 | local_lock_irqsave(&s->cpu_slab->lock, flags); |
bb192ed9 | 3065 | if (unlikely(c->slab)) { |
cfdf836e | 3066 | void *flush_freelist = c->freelist; |
bb192ed9 | 3067 | struct slab *flush_slab = c->slab; |
cfdf836e | 3068 | |
bb192ed9 | 3069 | c->slab = NULL; |
cfdf836e VB |
3070 | c->freelist = NULL; |
3071 | c->tid = next_tid(c->tid); | |
3072 | ||
bd0e7491 | 3073 | local_unlock_irqrestore(&s->cpu_slab->lock, flags); |
cfdf836e | 3074 | |
bb192ed9 | 3075 | deactivate_slab(s, flush_slab, flush_freelist); |
cfdf836e VB |
3076 | |
3077 | stat(s, CPUSLAB_FLUSH); | |
3078 | ||
c2092c12 | 3079 | goto retry_load_slab; |
cfdf836e | 3080 | } |
bb192ed9 | 3081 | c->slab = slab; |
3f2b77e3 | 3082 | |
1572df7c VB |
3083 | goto load_freelist; |
3084 | ||
3085 | return_single: | |
894b8788 | 3086 | |
bb192ed9 | 3087 | deactivate_slab(s, slab, get_freepointer(s, freelist)); |
6faa6833 | 3088 | return freelist; |
894b8788 CL |
3089 | } |
3090 | ||
a380a3c7 | 3091 | /* |
e500059b VB |
3092 | * A wrapper for ___slab_alloc() for contexts where preemption is not yet |
3093 | * disabled. Compensates for possible cpu changes by refetching the per cpu area | |
3094 | * pointer. | |
a380a3c7 CL |
3095 | */ |
3096 | static void *__slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node, | |
3097 | unsigned long addr, struct kmem_cache_cpu *c) | |
3098 | { | |
3099 | void *p; | |
a380a3c7 | 3100 | |
e500059b | 3101 | #ifdef CONFIG_PREEMPT_COUNT |
a380a3c7 CL |
3102 | /* |
3103 | * We may have been preempted and rescheduled on a different | |
e500059b | 3104 | * cpu before disabling preemption. Need to reload cpu area |
a380a3c7 CL |
3105 | * pointer. |
3106 | */ | |
25c00c50 | 3107 | c = slub_get_cpu_ptr(s->cpu_slab); |
a380a3c7 CL |
3108 | #endif |
3109 | ||
3110 | p = ___slab_alloc(s, gfpflags, node, addr, c); | |
e500059b | 3111 | #ifdef CONFIG_PREEMPT_COUNT |
25c00c50 | 3112 | slub_put_cpu_ptr(s->cpu_slab); |
e500059b | 3113 | #endif |
a380a3c7 CL |
3114 | return p; |
3115 | } | |
3116 | ||
0f181f9f AP |
3117 | /* |
3118 | * If the object has been wiped upon free, make sure it's fully initialized by | |
3119 | * zeroing out freelist pointer. | |
3120 | */ | |
3121 | static __always_inline void maybe_wipe_obj_freeptr(struct kmem_cache *s, | |
3122 | void *obj) | |
3123 | { | |
3124 | if (unlikely(slab_want_init_on_free(s)) && obj) | |
ce5716c6 AK |
3125 | memset((void *)((char *)kasan_reset_tag(obj) + s->offset), |
3126 | 0, sizeof(void *)); | |
0f181f9f AP |
3127 | } |
3128 | ||
894b8788 CL |
3129 | /* |
3130 | * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc) | |
3131 | * have the fastpath folded into their functions. So no function call | |
3132 | * overhead for requests that can be satisfied on the fastpath. | |
3133 | * | |
3134 | * The fastpath works by first checking if the lockless freelist can be used. | |
3135 | * If not then __slab_alloc is called for slow processing. | |
3136 | * | |
3137 | * Otherwise we can simply pick the next object from the lockless free list. | |
3138 | */ | |
88f2ef73 | 3139 | static __always_inline void *slab_alloc_node(struct kmem_cache *s, struct list_lru *lru, |
b89fb5ef | 3140 | gfp_t gfpflags, int node, unsigned long addr, size_t orig_size) |
894b8788 | 3141 | { |
03ec0ed5 | 3142 | void *object; |
dfb4f096 | 3143 | struct kmem_cache_cpu *c; |
bb192ed9 | 3144 | struct slab *slab; |
8a5ec0ba | 3145 | unsigned long tid; |
964d4bd3 | 3146 | struct obj_cgroup *objcg = NULL; |
da844b78 | 3147 | bool init = false; |
1f84260c | 3148 | |
88f2ef73 | 3149 | s = slab_pre_alloc_hook(s, lru, &objcg, 1, gfpflags); |
8135be5a | 3150 | if (!s) |
773ff60e | 3151 | return NULL; |
b89fb5ef AP |
3152 | |
3153 | object = kfence_alloc(s, orig_size, gfpflags); | |
3154 | if (unlikely(object)) | |
3155 | goto out; | |
3156 | ||
8a5ec0ba | 3157 | redo: |
8a5ec0ba CL |
3158 | /* |
3159 | * Must read kmem_cache cpu data via this cpu ptr. Preemption is | |
3160 | * enabled. We may switch back and forth between cpus while | |
3161 | * reading from one cpu area. That does not matter as long | |
3162 | * as we end up on the original cpu again when doing the cmpxchg. | |
7cccd80b | 3163 | * |
9b4bc85a VB |
3164 | * We must guarantee that tid and kmem_cache_cpu are retrieved on the |
3165 | * same cpu. We read first the kmem_cache_cpu pointer and use it to read | |
3166 | * the tid. If we are preempted and switched to another cpu between the | |
3167 | * two reads, it's OK as the two are still associated with the same cpu | |
3168 | * and cmpxchg later will validate the cpu. | |
8a5ec0ba | 3169 | */ |
9b4bc85a VB |
3170 | c = raw_cpu_ptr(s->cpu_slab); |
3171 | tid = READ_ONCE(c->tid); | |
9aabf810 JK |
3172 | |
3173 | /* | |
3174 | * Irqless object alloc/free algorithm used here depends on sequence | |
3175 | * of fetching cpu_slab's data. tid should be fetched before anything | |
c2092c12 | 3176 | * on c to guarantee that object and slab associated with previous tid |
9aabf810 | 3177 | * won't be used with current tid. If we fetch tid first, object and |
c2092c12 | 3178 | * slab could be one associated with next tid and our alloc/free |
9aabf810 JK |
3179 | * request will be failed. In this case, we will retry. So, no problem. |
3180 | */ | |
3181 | barrier(); | |
8a5ec0ba | 3182 | |
8a5ec0ba CL |
3183 | /* |
3184 | * The transaction ids are globally unique per cpu and per operation on | |
3185 | * a per cpu queue. Thus they can be guarantee that the cmpxchg_double | |
3186 | * occurs on the right processor and that there was no operation on the | |
3187 | * linked list in between. | |
3188 | */ | |
8a5ec0ba | 3189 | |
9dfc6e68 | 3190 | object = c->freelist; |
bb192ed9 | 3191 | slab = c->slab; |
bd0e7491 VB |
3192 | /* |
3193 | * We cannot use the lockless fastpath on PREEMPT_RT because if a | |
3194 | * slowpath has taken the local_lock_irqsave(), it is not protected | |
3195 | * against a fast path operation in an irq handler. So we need to take | |
3196 | * the slow path which uses local_lock. It is still relatively fast if | |
3197 | * there is a suitable cpu freelist. | |
3198 | */ | |
3199 | if (IS_ENABLED(CONFIG_PREEMPT_RT) || | |
bb192ed9 | 3200 | unlikely(!object || !slab || !node_match(slab, node))) { |
dfb4f096 | 3201 | object = __slab_alloc(s, gfpflags, node, addr, c); |
8eae1492 | 3202 | } else { |
0ad9500e ED |
3203 | void *next_object = get_freepointer_safe(s, object); |
3204 | ||
8a5ec0ba | 3205 | /* |
25985edc | 3206 | * The cmpxchg will only match if there was no additional |
8a5ec0ba CL |
3207 | * operation and if we are on the right processor. |
3208 | * | |
d0e0ac97 CG |
3209 | * The cmpxchg does the following atomically (without lock |
3210 | * semantics!) | |
8a5ec0ba CL |
3211 | * 1. Relocate first pointer to the current per cpu area. |
3212 | * 2. Verify that tid and freelist have not been changed | |
3213 | * 3. If they were not changed replace tid and freelist | |
3214 | * | |
d0e0ac97 CG |
3215 | * Since this is without lock semantics the protection is only |
3216 | * against code executing on this cpu *not* from access by | |
3217 | * other cpus. | |
8a5ec0ba | 3218 | */ |
933393f5 | 3219 | if (unlikely(!this_cpu_cmpxchg_double( |
8a5ec0ba CL |
3220 | s->cpu_slab->freelist, s->cpu_slab->tid, |
3221 | object, tid, | |
0ad9500e | 3222 | next_object, next_tid(tid)))) { |
8a5ec0ba CL |
3223 | |
3224 | note_cmpxchg_failure("slab_alloc", s, tid); | |
3225 | goto redo; | |
3226 | } | |
0ad9500e | 3227 | prefetch_freepointer(s, next_object); |
84e554e6 | 3228 | stat(s, ALLOC_FASTPATH); |
894b8788 | 3229 | } |
0f181f9f | 3230 | |
ce5716c6 | 3231 | maybe_wipe_obj_freeptr(s, object); |
da844b78 | 3232 | init = slab_want_init_on_alloc(gfpflags, s); |
d07dbea4 | 3233 | |
b89fb5ef | 3234 | out: |
da844b78 | 3235 | slab_post_alloc_hook(s, objcg, gfpflags, 1, &object, init); |
5a896d9e | 3236 | |
894b8788 | 3237 | return object; |
81819f0f CL |
3238 | } |
3239 | ||
88f2ef73 | 3240 | static __always_inline void *slab_alloc(struct kmem_cache *s, struct list_lru *lru, |
b89fb5ef | 3241 | gfp_t gfpflags, unsigned long addr, size_t orig_size) |
2b847c3c | 3242 | { |
88f2ef73 | 3243 | return slab_alloc_node(s, lru, gfpflags, NUMA_NO_NODE, addr, orig_size); |
2b847c3c EG |
3244 | } |
3245 | ||
88f2ef73 MS |
3246 | static __always_inline |
3247 | void *__kmem_cache_alloc_lru(struct kmem_cache *s, struct list_lru *lru, | |
3248 | gfp_t gfpflags) | |
81819f0f | 3249 | { |
88f2ef73 | 3250 | void *ret = slab_alloc(s, lru, gfpflags, _RET_IP_, s->object_size); |
5b882be4 | 3251 | |
b347aa7b | 3252 | trace_kmem_cache_alloc(_RET_IP_, ret, s, s->object_size, |
d0e0ac97 | 3253 | s->size, gfpflags); |
5b882be4 EGM |
3254 | |
3255 | return ret; | |
81819f0f | 3256 | } |
88f2ef73 MS |
3257 | |
3258 | void *kmem_cache_alloc(struct kmem_cache *s, gfp_t gfpflags) | |
3259 | { | |
3260 | return __kmem_cache_alloc_lru(s, NULL, gfpflags); | |
3261 | } | |
81819f0f CL |
3262 | EXPORT_SYMBOL(kmem_cache_alloc); |
3263 | ||
88f2ef73 MS |
3264 | void *kmem_cache_alloc_lru(struct kmem_cache *s, struct list_lru *lru, |
3265 | gfp_t gfpflags) | |
3266 | { | |
3267 | return __kmem_cache_alloc_lru(s, lru, gfpflags); | |
3268 | } | |
3269 | EXPORT_SYMBOL(kmem_cache_alloc_lru); | |
3270 | ||
0f24f128 | 3271 | #ifdef CONFIG_TRACING |
4a92379b RK |
3272 | void *kmem_cache_alloc_trace(struct kmem_cache *s, gfp_t gfpflags, size_t size) |
3273 | { | |
88f2ef73 | 3274 | void *ret = slab_alloc(s, NULL, gfpflags, _RET_IP_, size); |
b347aa7b | 3275 | trace_kmalloc(_RET_IP_, ret, s, size, s->size, gfpflags); |
0116523c | 3276 | ret = kasan_kmalloc(s, ret, size, gfpflags); |
4a92379b RK |
3277 | return ret; |
3278 | } | |
3279 | EXPORT_SYMBOL(kmem_cache_alloc_trace); | |
5b882be4 EGM |
3280 | #endif |
3281 | ||
81819f0f CL |
3282 | void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t gfpflags, int node) |
3283 | { | |
88f2ef73 | 3284 | void *ret = slab_alloc_node(s, NULL, gfpflags, node, _RET_IP_, s->object_size); |
5b882be4 | 3285 | |
b347aa7b | 3286 | trace_kmem_cache_alloc_node(_RET_IP_, ret, s, |
3b0efdfa | 3287 | s->object_size, s->size, gfpflags, node); |
5b882be4 EGM |
3288 | |
3289 | return ret; | |
81819f0f CL |
3290 | } |
3291 | EXPORT_SYMBOL(kmem_cache_alloc_node); | |
81819f0f | 3292 | |
0f24f128 | 3293 | #ifdef CONFIG_TRACING |
4a92379b | 3294 | void *kmem_cache_alloc_node_trace(struct kmem_cache *s, |
5b882be4 | 3295 | gfp_t gfpflags, |
4a92379b | 3296 | int node, size_t size) |
5b882be4 | 3297 | { |
88f2ef73 | 3298 | void *ret = slab_alloc_node(s, NULL, gfpflags, node, _RET_IP_, size); |
4a92379b | 3299 | |
b347aa7b | 3300 | trace_kmalloc_node(_RET_IP_, ret, s, |
4a92379b | 3301 | size, s->size, gfpflags, node); |
0316bec2 | 3302 | |
0116523c | 3303 | ret = kasan_kmalloc(s, ret, size, gfpflags); |
4a92379b | 3304 | return ret; |
5b882be4 | 3305 | } |
4a92379b | 3306 | EXPORT_SYMBOL(kmem_cache_alloc_node_trace); |
5b882be4 EGM |
3307 | #endif |
3308 | ||
81819f0f | 3309 | /* |
94e4d712 | 3310 | * Slow path handling. This may still be called frequently since objects |
894b8788 | 3311 | * have a longer lifetime than the cpu slabs in most processing loads. |
81819f0f | 3312 | * |
894b8788 | 3313 | * So we still attempt to reduce cache line usage. Just take the slab |
c2092c12 | 3314 | * lock and free the item. If there is no additional partial slab |
894b8788 | 3315 | * handling required then we can return immediately. |
81819f0f | 3316 | */ |
bb192ed9 | 3317 | static void __slab_free(struct kmem_cache *s, struct slab *slab, |
81084651 JDB |
3318 | void *head, void *tail, int cnt, |
3319 | unsigned long addr) | |
3320 | ||
81819f0f CL |
3321 | { |
3322 | void *prior; | |
2cfb7455 | 3323 | int was_frozen; |
bb192ed9 | 3324 | struct slab new; |
2cfb7455 CL |
3325 | unsigned long counters; |
3326 | struct kmem_cache_node *n = NULL; | |
3f649ab7 | 3327 | unsigned long flags; |
81819f0f | 3328 | |
8a5ec0ba | 3329 | stat(s, FREE_SLOWPATH); |
81819f0f | 3330 | |
b89fb5ef AP |
3331 | if (kfence_free(head)) |
3332 | return; | |
3333 | ||
19c7ff9e | 3334 | if (kmem_cache_debug(s) && |
bb192ed9 | 3335 | !free_debug_processing(s, slab, head, tail, cnt, addr)) |
80f08c19 | 3336 | return; |
6446faa2 | 3337 | |
2cfb7455 | 3338 | do { |
837d678d JK |
3339 | if (unlikely(n)) { |
3340 | spin_unlock_irqrestore(&n->list_lock, flags); | |
3341 | n = NULL; | |
3342 | } | |
bb192ed9 VB |
3343 | prior = slab->freelist; |
3344 | counters = slab->counters; | |
81084651 | 3345 | set_freepointer(s, tail, prior); |
2cfb7455 CL |
3346 | new.counters = counters; |
3347 | was_frozen = new.frozen; | |
81084651 | 3348 | new.inuse -= cnt; |
837d678d | 3349 | if ((!new.inuse || !prior) && !was_frozen) { |
49e22585 | 3350 | |
c65c1877 | 3351 | if (kmem_cache_has_cpu_partial(s) && !prior) { |
49e22585 CL |
3352 | |
3353 | /* | |
d0e0ac97 CG |
3354 | * Slab was on no list before and will be |
3355 | * partially empty | |
3356 | * We can defer the list move and instead | |
3357 | * freeze it. | |
49e22585 CL |
3358 | */ |
3359 | new.frozen = 1; | |
3360 | ||
c65c1877 | 3361 | } else { /* Needs to be taken off a list */ |
49e22585 | 3362 | |
bb192ed9 | 3363 | n = get_node(s, slab_nid(slab)); |
49e22585 CL |
3364 | /* |
3365 | * Speculatively acquire the list_lock. | |
3366 | * If the cmpxchg does not succeed then we may | |
3367 | * drop the list_lock without any processing. | |
3368 | * | |
3369 | * Otherwise the list_lock will synchronize with | |
3370 | * other processors updating the list of slabs. | |
3371 | */ | |
3372 | spin_lock_irqsave(&n->list_lock, flags); | |
3373 | ||
3374 | } | |
2cfb7455 | 3375 | } |
81819f0f | 3376 | |
bb192ed9 | 3377 | } while (!cmpxchg_double_slab(s, slab, |
2cfb7455 | 3378 | prior, counters, |
81084651 | 3379 | head, new.counters, |
2cfb7455 | 3380 | "__slab_free")); |
81819f0f | 3381 | |
2cfb7455 | 3382 | if (likely(!n)) { |
49e22585 | 3383 | |
c270cf30 AW |
3384 | if (likely(was_frozen)) { |
3385 | /* | |
3386 | * The list lock was not taken therefore no list | |
3387 | * activity can be necessary. | |
3388 | */ | |
3389 | stat(s, FREE_FROZEN); | |
3390 | } else if (new.frozen) { | |
3391 | /* | |
c2092c12 | 3392 | * If we just froze the slab then put it onto the |
c270cf30 AW |
3393 | * per cpu partial list. |
3394 | */ | |
bb192ed9 | 3395 | put_cpu_partial(s, slab, 1); |
8028dcea AS |
3396 | stat(s, CPU_PARTIAL_FREE); |
3397 | } | |
c270cf30 | 3398 | |
b455def2 L |
3399 | return; |
3400 | } | |
81819f0f | 3401 | |
8a5b20ae | 3402 | if (unlikely(!new.inuse && n->nr_partial >= s->min_partial)) |
837d678d JK |
3403 | goto slab_empty; |
3404 | ||
81819f0f | 3405 | /* |
837d678d JK |
3406 | * Objects left in the slab. If it was not on the partial list before |
3407 | * then add it. | |
81819f0f | 3408 | */ |
345c905d | 3409 | if (!kmem_cache_has_cpu_partial(s) && unlikely(!prior)) { |
bb192ed9 VB |
3410 | remove_full(s, n, slab); |
3411 | add_partial(n, slab, DEACTIVATE_TO_TAIL); | |
837d678d | 3412 | stat(s, FREE_ADD_PARTIAL); |
8ff12cfc | 3413 | } |
80f08c19 | 3414 | spin_unlock_irqrestore(&n->list_lock, flags); |
81819f0f CL |
3415 | return; |
3416 | ||
3417 | slab_empty: | |
a973e9dd | 3418 | if (prior) { |
81819f0f | 3419 | /* |
6fbabb20 | 3420 | * Slab on the partial list. |
81819f0f | 3421 | */ |
bb192ed9 | 3422 | remove_partial(n, slab); |
84e554e6 | 3423 | stat(s, FREE_REMOVE_PARTIAL); |
c65c1877 | 3424 | } else { |
6fbabb20 | 3425 | /* Slab must be on the full list */ |
bb192ed9 | 3426 | remove_full(s, n, slab); |
c65c1877 | 3427 | } |
2cfb7455 | 3428 | |
80f08c19 | 3429 | spin_unlock_irqrestore(&n->list_lock, flags); |
84e554e6 | 3430 | stat(s, FREE_SLAB); |
bb192ed9 | 3431 | discard_slab(s, slab); |
81819f0f CL |
3432 | } |
3433 | ||
894b8788 CL |
3434 | /* |
3435 | * Fastpath with forced inlining to produce a kfree and kmem_cache_free that | |
3436 | * can perform fastpath freeing without additional function calls. | |
3437 | * | |
3438 | * The fastpath is only possible if we are freeing to the current cpu slab | |
3439 | * of this processor. This typically the case if we have just allocated | |
3440 | * the item before. | |
3441 | * | |
3442 | * If fastpath is not possible then fall back to __slab_free where we deal | |
3443 | * with all sorts of special processing. | |
81084651 JDB |
3444 | * |
3445 | * Bulk free of a freelist with several objects (all pointing to the | |
c2092c12 | 3446 | * same slab) possible by specifying head and tail ptr, plus objects |
81084651 | 3447 | * count (cnt). Bulk free indicated by tail pointer being set. |
894b8788 | 3448 | */ |
80a9201a | 3449 | static __always_inline void do_slab_free(struct kmem_cache *s, |
bb192ed9 | 3450 | struct slab *slab, void *head, void *tail, |
80a9201a | 3451 | int cnt, unsigned long addr) |
894b8788 | 3452 | { |
81084651 | 3453 | void *tail_obj = tail ? : head; |
dfb4f096 | 3454 | struct kmem_cache_cpu *c; |
8a5ec0ba | 3455 | unsigned long tid; |
964d4bd3 | 3456 | |
8a5ec0ba CL |
3457 | redo: |
3458 | /* | |
3459 | * Determine the currently cpus per cpu slab. | |
3460 | * The cpu may change afterward. However that does not matter since | |
3461 | * data is retrieved via this pointer. If we are on the same cpu | |
2ae44005 | 3462 | * during the cmpxchg then the free will succeed. |
8a5ec0ba | 3463 | */ |
9b4bc85a VB |
3464 | c = raw_cpu_ptr(s->cpu_slab); |
3465 | tid = READ_ONCE(c->tid); | |
c016b0bd | 3466 | |
9aabf810 JK |
3467 | /* Same with comment on barrier() in slab_alloc_node() */ |
3468 | barrier(); | |
c016b0bd | 3469 | |
bb192ed9 | 3470 | if (likely(slab == c->slab)) { |
bd0e7491 | 3471 | #ifndef CONFIG_PREEMPT_RT |
5076190d LT |
3472 | void **freelist = READ_ONCE(c->freelist); |
3473 | ||
3474 | set_freepointer(s, tail_obj, freelist); | |
8a5ec0ba | 3475 | |
933393f5 | 3476 | if (unlikely(!this_cpu_cmpxchg_double( |
8a5ec0ba | 3477 | s->cpu_slab->freelist, s->cpu_slab->tid, |
5076190d | 3478 | freelist, tid, |
81084651 | 3479 | head, next_tid(tid)))) { |
8a5ec0ba CL |
3480 | |
3481 | note_cmpxchg_failure("slab_free", s, tid); | |
3482 | goto redo; | |
3483 | } | |
bd0e7491 VB |
3484 | #else /* CONFIG_PREEMPT_RT */ |
3485 | /* | |
3486 | * We cannot use the lockless fastpath on PREEMPT_RT because if | |
3487 | * a slowpath has taken the local_lock_irqsave(), it is not | |
3488 | * protected against a fast path operation in an irq handler. So | |
3489 | * we need to take the local_lock. We shouldn't simply defer to | |
3490 | * __slab_free() as that wouldn't use the cpu freelist at all. | |
3491 | */ | |
3492 | void **freelist; | |
3493 | ||
3494 | local_lock(&s->cpu_slab->lock); | |
3495 | c = this_cpu_ptr(s->cpu_slab); | |
bb192ed9 | 3496 | if (unlikely(slab != c->slab)) { |
bd0e7491 VB |
3497 | local_unlock(&s->cpu_slab->lock); |
3498 | goto redo; | |
3499 | } | |
3500 | tid = c->tid; | |
3501 | freelist = c->freelist; | |
3502 | ||
3503 | set_freepointer(s, tail_obj, freelist); | |
3504 | c->freelist = head; | |
3505 | c->tid = next_tid(tid); | |
3506 | ||
3507 | local_unlock(&s->cpu_slab->lock); | |
3508 | #endif | |
84e554e6 | 3509 | stat(s, FREE_FASTPATH); |
894b8788 | 3510 | } else |
bb192ed9 | 3511 | __slab_free(s, slab, head, tail_obj, cnt, addr); |
894b8788 | 3512 | |
894b8788 CL |
3513 | } |
3514 | ||
bb192ed9 | 3515 | static __always_inline void slab_free(struct kmem_cache *s, struct slab *slab, |
b77d5b1b | 3516 | void *head, void *tail, void **p, int cnt, |
80a9201a AP |
3517 | unsigned long addr) |
3518 | { | |
b77d5b1b | 3519 | memcg_slab_free_hook(s, slab, p, cnt); |
80a9201a | 3520 | /* |
c3895391 AK |
3521 | * With KASAN enabled slab_free_freelist_hook modifies the freelist |
3522 | * to remove objects, whose reuse must be delayed. | |
80a9201a | 3523 | */ |
899447f6 | 3524 | if (slab_free_freelist_hook(s, &head, &tail, &cnt)) |
bb192ed9 | 3525 | do_slab_free(s, slab, head, tail, cnt, addr); |
80a9201a AP |
3526 | } |
3527 | ||
2bd926b4 | 3528 | #ifdef CONFIG_KASAN_GENERIC |
80a9201a AP |
3529 | void ___cache_free(struct kmem_cache *cache, void *x, unsigned long addr) |
3530 | { | |
bb192ed9 | 3531 | do_slab_free(cache, virt_to_slab(x), x, NULL, 1, addr); |
80a9201a AP |
3532 | } |
3533 | #endif | |
3534 | ||
81819f0f CL |
3535 | void kmem_cache_free(struct kmem_cache *s, void *x) |
3536 | { | |
b9ce5ef4 GC |
3537 | s = cache_from_obj(s, x); |
3538 | if (!s) | |
79576102 | 3539 | return; |
3544de8e | 3540 | trace_kmem_cache_free(_RET_IP_, x, s->name); |
b77d5b1b | 3541 | slab_free(s, virt_to_slab(x), x, NULL, &x, 1, _RET_IP_); |
81819f0f CL |
3542 | } |
3543 | EXPORT_SYMBOL(kmem_cache_free); | |
3544 | ||
d0ecd894 | 3545 | struct detached_freelist { |
cc465c3b | 3546 | struct slab *slab; |
d0ecd894 JDB |
3547 | void *tail; |
3548 | void *freelist; | |
3549 | int cnt; | |
376bf125 | 3550 | struct kmem_cache *s; |
d0ecd894 | 3551 | }; |
fbd02630 | 3552 | |
d835eef4 | 3553 | static inline void free_large_kmalloc(struct folio *folio, void *object) |
f227f0fa | 3554 | { |
d835eef4 | 3555 | unsigned int order = folio_order(folio); |
f227f0fa | 3556 | |
d835eef4 | 3557 | if (WARN_ON_ONCE(order == 0)) |
d0fe47c6 KW |
3558 | pr_warn_once("object pointer: 0x%p\n", object); |
3559 | ||
1ed7ce57 | 3560 | kfree_hook(object); |
d835eef4 MWO |
3561 | mod_lruvec_page_state(folio_page(folio, 0), NR_SLAB_UNRECLAIMABLE_B, |
3562 | -(PAGE_SIZE << order)); | |
3563 | __free_pages(folio_page(folio, 0), order); | |
f227f0fa SB |
3564 | } |
3565 | ||
d0ecd894 JDB |
3566 | /* |
3567 | * This function progressively scans the array with free objects (with | |
3568 | * a limited look ahead) and extract objects belonging to the same | |
cc465c3b MWO |
3569 | * slab. It builds a detached freelist directly within the given |
3570 | * slab/objects. This can happen without any need for | |
d0ecd894 JDB |
3571 | * synchronization, because the objects are owned by running process. |
3572 | * The freelist is build up as a single linked list in the objects. | |
3573 | * The idea is, that this detached freelist can then be bulk | |
3574 | * transferred to the real freelist(s), but only requiring a single | |
3575 | * synchronization primitive. Look ahead in the array is limited due | |
3576 | * to performance reasons. | |
3577 | */ | |
376bf125 JDB |
3578 | static inline |
3579 | int build_detached_freelist(struct kmem_cache *s, size_t size, | |
3580 | void **p, struct detached_freelist *df) | |
d0ecd894 | 3581 | { |
d0ecd894 JDB |
3582 | int lookahead = 3; |
3583 | void *object; | |
cc465c3b | 3584 | struct folio *folio; |
b77d5b1b | 3585 | size_t same; |
fbd02630 | 3586 | |
b77d5b1b | 3587 | object = p[--size]; |
cc465c3b | 3588 | folio = virt_to_folio(object); |
ca257195 JDB |
3589 | if (!s) { |
3590 | /* Handle kalloc'ed objects */ | |
cc465c3b | 3591 | if (unlikely(!folio_test_slab(folio))) { |
d835eef4 | 3592 | free_large_kmalloc(folio, object); |
b77d5b1b | 3593 | df->slab = NULL; |
ca257195 JDB |
3594 | return size; |
3595 | } | |
3596 | /* Derive kmem_cache from object */ | |
b77d5b1b MS |
3597 | df->slab = folio_slab(folio); |
3598 | df->s = df->slab->slab_cache; | |
ca257195 | 3599 | } else { |
b77d5b1b | 3600 | df->slab = folio_slab(folio); |
ca257195 JDB |
3601 | df->s = cache_from_obj(s, object); /* Support for memcg */ |
3602 | } | |
376bf125 | 3603 | |
d0ecd894 | 3604 | /* Start new detached freelist */ |
d0ecd894 JDB |
3605 | df->tail = object; |
3606 | df->freelist = object; | |
d0ecd894 JDB |
3607 | df->cnt = 1; |
3608 | ||
b77d5b1b MS |
3609 | if (is_kfence_address(object)) |
3610 | return size; | |
3611 | ||
3612 | set_freepointer(df->s, object, NULL); | |
3613 | ||
3614 | same = size; | |
d0ecd894 JDB |
3615 | while (size) { |
3616 | object = p[--size]; | |
cc465c3b MWO |
3617 | /* df->slab is always set at this point */ |
3618 | if (df->slab == virt_to_slab(object)) { | |
d0ecd894 | 3619 | /* Opportunity build freelist */ |
376bf125 | 3620 | set_freepointer(df->s, object, df->freelist); |
d0ecd894 JDB |
3621 | df->freelist = object; |
3622 | df->cnt++; | |
b77d5b1b MS |
3623 | same--; |
3624 | if (size != same) | |
3625 | swap(p[size], p[same]); | |
d0ecd894 | 3626 | continue; |
fbd02630 | 3627 | } |
d0ecd894 JDB |
3628 | |
3629 | /* Limit look ahead search */ | |
3630 | if (!--lookahead) | |
3631 | break; | |
fbd02630 | 3632 | } |
d0ecd894 | 3633 | |
b77d5b1b | 3634 | return same; |
d0ecd894 JDB |
3635 | } |
3636 | ||
d0ecd894 | 3637 | /* Note that interrupts must be enabled when calling this function. */ |
376bf125 | 3638 | void kmem_cache_free_bulk(struct kmem_cache *s, size_t size, void **p) |
d0ecd894 | 3639 | { |
2055e67b | 3640 | if (!size) |
d0ecd894 JDB |
3641 | return; |
3642 | ||
3643 | do { | |
3644 | struct detached_freelist df; | |
3645 | ||
3646 | size = build_detached_freelist(s, size, p, &df); | |
cc465c3b | 3647 | if (!df.slab) |
d0ecd894 JDB |
3648 | continue; |
3649 | ||
b77d5b1b MS |
3650 | slab_free(df.s, df.slab, df.freelist, df.tail, &p[size], df.cnt, |
3651 | _RET_IP_); | |
d0ecd894 | 3652 | } while (likely(size)); |
484748f0 CL |
3653 | } |
3654 | EXPORT_SYMBOL(kmem_cache_free_bulk); | |
3655 | ||
994eb764 | 3656 | /* Note that interrupts must be enabled when calling this function. */ |
865762a8 JDB |
3657 | int kmem_cache_alloc_bulk(struct kmem_cache *s, gfp_t flags, size_t size, |
3658 | void **p) | |
484748f0 | 3659 | { |
994eb764 JDB |
3660 | struct kmem_cache_cpu *c; |
3661 | int i; | |
964d4bd3 | 3662 | struct obj_cgroup *objcg = NULL; |
994eb764 | 3663 | |
03ec0ed5 | 3664 | /* memcg and kmem_cache debug support */ |
88f2ef73 | 3665 | s = slab_pre_alloc_hook(s, NULL, &objcg, size, flags); |
03ec0ed5 JDB |
3666 | if (unlikely(!s)) |
3667 | return false; | |
994eb764 JDB |
3668 | /* |
3669 | * Drain objects in the per cpu slab, while disabling local | |
3670 | * IRQs, which protects against PREEMPT and interrupts | |
3671 | * handlers invoking normal fastpath. | |
3672 | */ | |
25c00c50 | 3673 | c = slub_get_cpu_ptr(s->cpu_slab); |
bd0e7491 | 3674 | local_lock_irq(&s->cpu_slab->lock); |
994eb764 JDB |
3675 | |
3676 | for (i = 0; i < size; i++) { | |
b89fb5ef | 3677 | void *object = kfence_alloc(s, s->object_size, flags); |
994eb764 | 3678 | |
b89fb5ef AP |
3679 | if (unlikely(object)) { |
3680 | p[i] = object; | |
3681 | continue; | |
3682 | } | |
3683 | ||
3684 | object = c->freelist; | |
ebe909e0 | 3685 | if (unlikely(!object)) { |
fd4d9c7d JH |
3686 | /* |
3687 | * We may have removed an object from c->freelist using | |
3688 | * the fastpath in the previous iteration; in that case, | |
3689 | * c->tid has not been bumped yet. | |
3690 | * Since ___slab_alloc() may reenable interrupts while | |
3691 | * allocating memory, we should bump c->tid now. | |
3692 | */ | |
3693 | c->tid = next_tid(c->tid); | |
3694 | ||
bd0e7491 | 3695 | local_unlock_irq(&s->cpu_slab->lock); |
e500059b | 3696 | |
ebe909e0 JDB |
3697 | /* |
3698 | * Invoking slow path likely have side-effect | |
3699 | * of re-populating per CPU c->freelist | |
3700 | */ | |
87098373 | 3701 | p[i] = ___slab_alloc(s, flags, NUMA_NO_NODE, |
ebe909e0 | 3702 | _RET_IP_, c); |
87098373 CL |
3703 | if (unlikely(!p[i])) |
3704 | goto error; | |
3705 | ||
ebe909e0 | 3706 | c = this_cpu_ptr(s->cpu_slab); |
0f181f9f AP |
3707 | maybe_wipe_obj_freeptr(s, p[i]); |
3708 | ||
bd0e7491 | 3709 | local_lock_irq(&s->cpu_slab->lock); |
e500059b | 3710 | |
ebe909e0 JDB |
3711 | continue; /* goto for-loop */ |
3712 | } | |
994eb764 JDB |
3713 | c->freelist = get_freepointer(s, object); |
3714 | p[i] = object; | |
0f181f9f | 3715 | maybe_wipe_obj_freeptr(s, p[i]); |
994eb764 JDB |
3716 | } |
3717 | c->tid = next_tid(c->tid); | |
bd0e7491 | 3718 | local_unlock_irq(&s->cpu_slab->lock); |
25c00c50 | 3719 | slub_put_cpu_ptr(s->cpu_slab); |
994eb764 | 3720 | |
da844b78 AK |
3721 | /* |
3722 | * memcg and kmem_cache debug support and memory initialization. | |
3723 | * Done outside of the IRQ disabled fastpath loop. | |
3724 | */ | |
3725 | slab_post_alloc_hook(s, objcg, flags, size, p, | |
3726 | slab_want_init_on_alloc(flags, s)); | |
865762a8 | 3727 | return i; |
87098373 | 3728 | error: |
25c00c50 | 3729 | slub_put_cpu_ptr(s->cpu_slab); |
da844b78 | 3730 | slab_post_alloc_hook(s, objcg, flags, i, p, false); |
2055e67b | 3731 | kmem_cache_free_bulk(s, i, p); |
865762a8 | 3732 | return 0; |
484748f0 CL |
3733 | } |
3734 | EXPORT_SYMBOL(kmem_cache_alloc_bulk); | |
3735 | ||
3736 | ||
81819f0f | 3737 | /* |
672bba3a CL |
3738 | * Object placement in a slab is made very easy because we always start at |
3739 | * offset 0. If we tune the size of the object to the alignment then we can | |
3740 | * get the required alignment by putting one properly sized object after | |
3741 | * another. | |
81819f0f CL |
3742 | * |
3743 | * Notice that the allocation order determines the sizes of the per cpu | |
3744 | * caches. Each processor has always one slab available for allocations. | |
3745 | * Increasing the allocation order reduces the number of times that slabs | |
672bba3a | 3746 | * must be moved on and off the partial lists and is therefore a factor in |
81819f0f | 3747 | * locking overhead. |
81819f0f CL |
3748 | */ |
3749 | ||
3750 | /* | |
f0953a1b | 3751 | * Minimum / Maximum order of slab pages. This influences locking overhead |
81819f0f CL |
3752 | * and slab fragmentation. A higher order reduces the number of partial slabs |
3753 | * and increases the number of allocations possible without having to | |
3754 | * take the list_lock. | |
3755 | */ | |
19af27af AD |
3756 | static unsigned int slub_min_order; |
3757 | static unsigned int slub_max_order = PAGE_ALLOC_COSTLY_ORDER; | |
3758 | static unsigned int slub_min_objects; | |
81819f0f | 3759 | |
81819f0f CL |
3760 | /* |
3761 | * Calculate the order of allocation given an slab object size. | |
3762 | * | |
672bba3a CL |
3763 | * The order of allocation has significant impact on performance and other |
3764 | * system components. Generally order 0 allocations should be preferred since | |
3765 | * order 0 does not cause fragmentation in the page allocator. Larger objects | |
3766 | * be problematic to put into order 0 slabs because there may be too much | |
c124f5b5 | 3767 | * unused space left. We go to a higher order if more than 1/16th of the slab |
672bba3a CL |
3768 | * would be wasted. |
3769 | * | |
3770 | * In order to reach satisfactory performance we must ensure that a minimum | |
3771 | * number of objects is in one slab. Otherwise we may generate too much | |
3772 | * activity on the partial lists which requires taking the list_lock. This is | |
3773 | * less a concern for large slabs though which are rarely used. | |
81819f0f | 3774 | * |
672bba3a CL |
3775 | * slub_max_order specifies the order where we begin to stop considering the |
3776 | * number of objects in a slab as critical. If we reach slub_max_order then | |
3777 | * we try to keep the page order as low as possible. So we accept more waste | |
3778 | * of space in favor of a small page order. | |
81819f0f | 3779 | * |
672bba3a CL |
3780 | * Higher order allocations also allow the placement of more objects in a |
3781 | * slab and thereby reduce object handling overhead. If the user has | |
dc84207d | 3782 | * requested a higher minimum order then we start with that one instead of |
672bba3a | 3783 | * the smallest order which will fit the object. |
81819f0f | 3784 | */ |
d122019b | 3785 | static inline unsigned int calc_slab_order(unsigned int size, |
19af27af | 3786 | unsigned int min_objects, unsigned int max_order, |
9736d2a9 | 3787 | unsigned int fract_leftover) |
81819f0f | 3788 | { |
19af27af AD |
3789 | unsigned int min_order = slub_min_order; |
3790 | unsigned int order; | |
81819f0f | 3791 | |
9736d2a9 | 3792 | if (order_objects(min_order, size) > MAX_OBJS_PER_PAGE) |
210b5c06 | 3793 | return get_order(size * MAX_OBJS_PER_PAGE) - 1; |
39b26464 | 3794 | |
9736d2a9 | 3795 | for (order = max(min_order, (unsigned int)get_order(min_objects * size)); |
5e6d444e | 3796 | order <= max_order; order++) { |
81819f0f | 3797 | |
19af27af AD |
3798 | unsigned int slab_size = (unsigned int)PAGE_SIZE << order; |
3799 | unsigned int rem; | |
81819f0f | 3800 | |
9736d2a9 | 3801 | rem = slab_size % size; |
81819f0f | 3802 | |
5e6d444e | 3803 | if (rem <= slab_size / fract_leftover) |
81819f0f | 3804 | break; |
81819f0f | 3805 | } |
672bba3a | 3806 | |
81819f0f CL |
3807 | return order; |
3808 | } | |
3809 | ||
9736d2a9 | 3810 | static inline int calculate_order(unsigned int size) |
5e6d444e | 3811 | { |
19af27af AD |
3812 | unsigned int order; |
3813 | unsigned int min_objects; | |
3814 | unsigned int max_objects; | |
3286222f | 3815 | unsigned int nr_cpus; |
5e6d444e CL |
3816 | |
3817 | /* | |
3818 | * Attempt to find best configuration for a slab. This | |
3819 | * works by first attempting to generate a layout with | |
3820 | * the best configuration and backing off gradually. | |
3821 | * | |
422ff4d7 | 3822 | * First we increase the acceptable waste in a slab. Then |
5e6d444e CL |
3823 | * we reduce the minimum objects required in a slab. |
3824 | */ | |
3825 | min_objects = slub_min_objects; | |
3286222f VB |
3826 | if (!min_objects) { |
3827 | /* | |
3828 | * Some architectures will only update present cpus when | |
3829 | * onlining them, so don't trust the number if it's just 1. But | |
3830 | * we also don't want to use nr_cpu_ids always, as on some other | |
3831 | * architectures, there can be many possible cpus, but never | |
3832 | * onlined. Here we compromise between trying to avoid too high | |
3833 | * order on systems that appear larger than they are, and too | |
3834 | * low order on systems that appear smaller than they are. | |
3835 | */ | |
3836 | nr_cpus = num_present_cpus(); | |
3837 | if (nr_cpus <= 1) | |
3838 | nr_cpus = nr_cpu_ids; | |
3839 | min_objects = 4 * (fls(nr_cpus) + 1); | |
3840 | } | |
9736d2a9 | 3841 | max_objects = order_objects(slub_max_order, size); |
e8120ff1 ZY |
3842 | min_objects = min(min_objects, max_objects); |
3843 | ||
5e6d444e | 3844 | while (min_objects > 1) { |
19af27af AD |
3845 | unsigned int fraction; |
3846 | ||
c124f5b5 | 3847 | fraction = 16; |
5e6d444e | 3848 | while (fraction >= 4) { |
d122019b | 3849 | order = calc_slab_order(size, min_objects, |
9736d2a9 | 3850 | slub_max_order, fraction); |
5e6d444e CL |
3851 | if (order <= slub_max_order) |
3852 | return order; | |
3853 | fraction /= 2; | |
3854 | } | |
5086c389 | 3855 | min_objects--; |
5e6d444e CL |
3856 | } |
3857 | ||
3858 | /* | |
3859 | * We were unable to place multiple objects in a slab. Now | |
3860 | * lets see if we can place a single object there. | |
3861 | */ | |
d122019b | 3862 | order = calc_slab_order(size, 1, slub_max_order, 1); |
5e6d444e CL |
3863 | if (order <= slub_max_order) |
3864 | return order; | |
3865 | ||
3866 | /* | |
3867 | * Doh this slab cannot be placed using slub_max_order. | |
3868 | */ | |
d122019b | 3869 | order = calc_slab_order(size, 1, MAX_ORDER, 1); |
818cf590 | 3870 | if (order < MAX_ORDER) |
5e6d444e CL |
3871 | return order; |
3872 | return -ENOSYS; | |
3873 | } | |
3874 | ||
5595cffc | 3875 | static void |
4053497d | 3876 | init_kmem_cache_node(struct kmem_cache_node *n) |
81819f0f CL |
3877 | { |
3878 | n->nr_partial = 0; | |
81819f0f CL |
3879 | spin_lock_init(&n->list_lock); |
3880 | INIT_LIST_HEAD(&n->partial); | |
8ab1372f | 3881 | #ifdef CONFIG_SLUB_DEBUG |
0f389ec6 | 3882 | atomic_long_set(&n->nr_slabs, 0); |
02b71b70 | 3883 | atomic_long_set(&n->total_objects, 0); |
643b1138 | 3884 | INIT_LIST_HEAD(&n->full); |
8ab1372f | 3885 | #endif |
81819f0f CL |
3886 | } |
3887 | ||
55136592 | 3888 | static inline int alloc_kmem_cache_cpus(struct kmem_cache *s) |
4c93c355 | 3889 | { |
6c182dc0 | 3890 | BUILD_BUG_ON(PERCPU_DYNAMIC_EARLY_SIZE < |
95a05b42 | 3891 | KMALLOC_SHIFT_HIGH * sizeof(struct kmem_cache_cpu)); |
4c93c355 | 3892 | |
8a5ec0ba | 3893 | /* |
d4d84fef CM |
3894 | * Must align to double word boundary for the double cmpxchg |
3895 | * instructions to work; see __pcpu_double_call_return_bool(). | |
8a5ec0ba | 3896 | */ |
d4d84fef CM |
3897 | s->cpu_slab = __alloc_percpu(sizeof(struct kmem_cache_cpu), |
3898 | 2 * sizeof(void *)); | |
8a5ec0ba CL |
3899 | |
3900 | if (!s->cpu_slab) | |
3901 | return 0; | |
3902 | ||
3903 | init_kmem_cache_cpus(s); | |
4c93c355 | 3904 | |
8a5ec0ba | 3905 | return 1; |
4c93c355 | 3906 | } |
4c93c355 | 3907 | |
51df1142 CL |
3908 | static struct kmem_cache *kmem_cache_node; |
3909 | ||
81819f0f CL |
3910 | /* |
3911 | * No kmalloc_node yet so do it by hand. We know that this is the first | |
3912 | * slab on the node for this slabcache. There are no concurrent accesses | |
3913 | * possible. | |
3914 | * | |
721ae22a ZYW |
3915 | * Note that this function only works on the kmem_cache_node |
3916 | * when allocating for the kmem_cache_node. This is used for bootstrapping | |
4c93c355 | 3917 | * memory on a fresh node that has no slab structures yet. |
81819f0f | 3918 | */ |
55136592 | 3919 | static void early_kmem_cache_node_alloc(int node) |
81819f0f | 3920 | { |
bb192ed9 | 3921 | struct slab *slab; |
81819f0f CL |
3922 | struct kmem_cache_node *n; |
3923 | ||
51df1142 | 3924 | BUG_ON(kmem_cache_node->size < sizeof(struct kmem_cache_node)); |
81819f0f | 3925 | |
bb192ed9 | 3926 | slab = new_slab(kmem_cache_node, GFP_NOWAIT, node); |
81819f0f | 3927 | |
bb192ed9 VB |
3928 | BUG_ON(!slab); |
3929 | if (slab_nid(slab) != node) { | |
f9f58285 FF |
3930 | pr_err("SLUB: Unable to allocate memory from node %d\n", node); |
3931 | pr_err("SLUB: Allocating a useless per node structure in order to be able to continue\n"); | |
a2f92ee7 CL |
3932 | } |
3933 | ||
bb192ed9 | 3934 | n = slab->freelist; |
81819f0f | 3935 | BUG_ON(!n); |
8ab1372f | 3936 | #ifdef CONFIG_SLUB_DEBUG |
f7cb1933 | 3937 | init_object(kmem_cache_node, n, SLUB_RED_ACTIVE); |
51df1142 | 3938 | init_tracking(kmem_cache_node, n); |
8ab1372f | 3939 | #endif |
da844b78 | 3940 | n = kasan_slab_alloc(kmem_cache_node, n, GFP_KERNEL, false); |
bb192ed9 VB |
3941 | slab->freelist = get_freepointer(kmem_cache_node, n); |
3942 | slab->inuse = 1; | |
3943 | slab->frozen = 0; | |
12b22386 | 3944 | kmem_cache_node->node[node] = n; |
4053497d | 3945 | init_kmem_cache_node(n); |
bb192ed9 | 3946 | inc_slabs_node(kmem_cache_node, node, slab->objects); |
6446faa2 | 3947 | |
67b6c900 | 3948 | /* |
1e4dd946 SR |
3949 | * No locks need to be taken here as it has just been |
3950 | * initialized and there is no concurrent access. | |
67b6c900 | 3951 | */ |
bb192ed9 | 3952 | __add_partial(n, slab, DEACTIVATE_TO_HEAD); |
81819f0f CL |
3953 | } |
3954 | ||
3955 | static void free_kmem_cache_nodes(struct kmem_cache *s) | |
3956 | { | |
3957 | int node; | |
fa45dc25 | 3958 | struct kmem_cache_node *n; |
81819f0f | 3959 | |
fa45dc25 | 3960 | for_each_kmem_cache_node(s, node, n) { |
81819f0f | 3961 | s->node[node] = NULL; |
ea37df54 | 3962 | kmem_cache_free(kmem_cache_node, n); |
81819f0f CL |
3963 | } |
3964 | } | |
3965 | ||
52b4b950 DS |
3966 | void __kmem_cache_release(struct kmem_cache *s) |
3967 | { | |
210e7a43 | 3968 | cache_random_seq_destroy(s); |
52b4b950 DS |
3969 | free_percpu(s->cpu_slab); |
3970 | free_kmem_cache_nodes(s); | |
3971 | } | |
3972 | ||
55136592 | 3973 | static int init_kmem_cache_nodes(struct kmem_cache *s) |
81819f0f CL |
3974 | { |
3975 | int node; | |
81819f0f | 3976 | |
7e1fa93d | 3977 | for_each_node_mask(node, slab_nodes) { |
81819f0f CL |
3978 | struct kmem_cache_node *n; |
3979 | ||
73367bd8 | 3980 | if (slab_state == DOWN) { |
55136592 | 3981 | early_kmem_cache_node_alloc(node); |
73367bd8 AD |
3982 | continue; |
3983 | } | |
51df1142 | 3984 | n = kmem_cache_alloc_node(kmem_cache_node, |
55136592 | 3985 | GFP_KERNEL, node); |
81819f0f | 3986 | |
73367bd8 AD |
3987 | if (!n) { |
3988 | free_kmem_cache_nodes(s); | |
3989 | return 0; | |
81819f0f | 3990 | } |
73367bd8 | 3991 | |
4053497d | 3992 | init_kmem_cache_node(n); |
ea37df54 | 3993 | s->node[node] = n; |
81819f0f CL |
3994 | } |
3995 | return 1; | |
3996 | } | |
81819f0f | 3997 | |
e6d0e1dc WY |
3998 | static void set_cpu_partial(struct kmem_cache *s) |
3999 | { | |
4000 | #ifdef CONFIG_SLUB_CPU_PARTIAL | |
b47291ef VB |
4001 | unsigned int nr_objects; |
4002 | ||
e6d0e1dc WY |
4003 | /* |
4004 | * cpu_partial determined the maximum number of objects kept in the | |
4005 | * per cpu partial lists of a processor. | |
4006 | * | |
4007 | * Per cpu partial lists mainly contain slabs that just have one | |
4008 | * object freed. If they are used for allocation then they can be | |
4009 | * filled up again with minimal effort. The slab will never hit the | |
4010 | * per node partial lists and therefore no locking will be required. | |
4011 | * | |
b47291ef VB |
4012 | * For backwards compatibility reasons, this is determined as number |
4013 | * of objects, even though we now limit maximum number of pages, see | |
4014 | * slub_set_cpu_partial() | |
e6d0e1dc WY |
4015 | */ |
4016 | if (!kmem_cache_has_cpu_partial(s)) | |
b47291ef | 4017 | nr_objects = 0; |
e6d0e1dc | 4018 | else if (s->size >= PAGE_SIZE) |
b47291ef | 4019 | nr_objects = 6; |
e6d0e1dc | 4020 | else if (s->size >= 1024) |
23e98ad1 | 4021 | nr_objects = 24; |
e6d0e1dc | 4022 | else if (s->size >= 256) |
23e98ad1 | 4023 | nr_objects = 52; |
e6d0e1dc | 4024 | else |
23e98ad1 | 4025 | nr_objects = 120; |
b47291ef VB |
4026 | |
4027 | slub_set_cpu_partial(s, nr_objects); | |
e6d0e1dc WY |
4028 | #endif |
4029 | } | |
4030 | ||
81819f0f CL |
4031 | /* |
4032 | * calculate_sizes() determines the order and the distribution of data within | |
4033 | * a slab object. | |
4034 | */ | |
ae44d81d | 4035 | static int calculate_sizes(struct kmem_cache *s) |
81819f0f | 4036 | { |
d50112ed | 4037 | slab_flags_t flags = s->flags; |
be4a7988 | 4038 | unsigned int size = s->object_size; |
19af27af | 4039 | unsigned int order; |
81819f0f | 4040 | |
d8b42bf5 CL |
4041 | /* |
4042 | * Round up object size to the next word boundary. We can only | |
4043 | * place the free pointer at word boundaries and this determines | |
4044 | * the possible location of the free pointer. | |
4045 | */ | |
4046 | size = ALIGN(size, sizeof(void *)); | |
4047 | ||
4048 | #ifdef CONFIG_SLUB_DEBUG | |
81819f0f CL |
4049 | /* |
4050 | * Determine if we can poison the object itself. If the user of | |
4051 | * the slab may touch the object after free or before allocation | |
4052 | * then we should never poison the object itself. | |
4053 | */ | |
5f0d5a3a | 4054 | if ((flags & SLAB_POISON) && !(flags & SLAB_TYPESAFE_BY_RCU) && |
c59def9f | 4055 | !s->ctor) |
81819f0f CL |
4056 | s->flags |= __OBJECT_POISON; |
4057 | else | |
4058 | s->flags &= ~__OBJECT_POISON; | |
4059 | ||
81819f0f CL |
4060 | |
4061 | /* | |
672bba3a | 4062 | * If we are Redzoning then check if there is some space between the |
81819f0f | 4063 | * end of the object and the free pointer. If not then add an |
672bba3a | 4064 | * additional word to have some bytes to store Redzone information. |
81819f0f | 4065 | */ |
3b0efdfa | 4066 | if ((flags & SLAB_RED_ZONE) && size == s->object_size) |
81819f0f | 4067 | size += sizeof(void *); |
41ecc55b | 4068 | #endif |
81819f0f CL |
4069 | |
4070 | /* | |
672bba3a | 4071 | * With that we have determined the number of bytes in actual use |
e41a49fa | 4072 | * by the object and redzoning. |
81819f0f CL |
4073 | */ |
4074 | s->inuse = size; | |
4075 | ||
74c1d3e0 KC |
4076 | if ((flags & (SLAB_TYPESAFE_BY_RCU | SLAB_POISON)) || |
4077 | ((flags & SLAB_RED_ZONE) && s->object_size < sizeof(void *)) || | |
4078 | s->ctor) { | |
81819f0f CL |
4079 | /* |
4080 | * Relocate free pointer after the object if it is not | |
4081 | * permitted to overwrite the first word of the object on | |
4082 | * kmem_cache_free. | |
4083 | * | |
4084 | * This is the case if we do RCU, have a constructor or | |
74c1d3e0 KC |
4085 | * destructor, are poisoning the objects, or are |
4086 | * redzoning an object smaller than sizeof(void *). | |
cbfc35a4 WL |
4087 | * |
4088 | * The assumption that s->offset >= s->inuse means free | |
4089 | * pointer is outside of the object is used in the | |
4090 | * freeptr_outside_object() function. If that is no | |
4091 | * longer true, the function needs to be modified. | |
81819f0f CL |
4092 | */ |
4093 | s->offset = size; | |
4094 | size += sizeof(void *); | |
e41a49fa | 4095 | } else { |
3202fa62 KC |
4096 | /* |
4097 | * Store freelist pointer near middle of object to keep | |
4098 | * it away from the edges of the object to avoid small | |
4099 | * sized over/underflows from neighboring allocations. | |
4100 | */ | |
e41a49fa | 4101 | s->offset = ALIGN_DOWN(s->object_size / 2, sizeof(void *)); |
81819f0f CL |
4102 | } |
4103 | ||
c12b3c62 | 4104 | #ifdef CONFIG_SLUB_DEBUG |
81819f0f CL |
4105 | if (flags & SLAB_STORE_USER) |
4106 | /* | |
4107 | * Need to store information about allocs and frees after | |
4108 | * the object. | |
4109 | */ | |
4110 | size += 2 * sizeof(struct track); | |
80a9201a | 4111 | #endif |
81819f0f | 4112 | |
80a9201a AP |
4113 | kasan_cache_create(s, &size, &s->flags); |
4114 | #ifdef CONFIG_SLUB_DEBUG | |
d86bd1be | 4115 | if (flags & SLAB_RED_ZONE) { |
81819f0f CL |
4116 | /* |
4117 | * Add some empty padding so that we can catch | |
4118 | * overwrites from earlier objects rather than let | |
4119 | * tracking information or the free pointer be | |
0211a9c8 | 4120 | * corrupted if a user writes before the start |
81819f0f CL |
4121 | * of the object. |
4122 | */ | |
4123 | size += sizeof(void *); | |
d86bd1be JK |
4124 | |
4125 | s->red_left_pad = sizeof(void *); | |
4126 | s->red_left_pad = ALIGN(s->red_left_pad, s->align); | |
4127 | size += s->red_left_pad; | |
4128 | } | |
41ecc55b | 4129 | #endif |
672bba3a | 4130 | |
81819f0f CL |
4131 | /* |
4132 | * SLUB stores one object immediately after another beginning from | |
4133 | * offset 0. In order to align the objects we have to simply size | |
4134 | * each object to conform to the alignment. | |
4135 | */ | |
45906855 | 4136 | size = ALIGN(size, s->align); |
81819f0f | 4137 | s->size = size; |
4138fdfc | 4138 | s->reciprocal_size = reciprocal_value(size); |
ae44d81d | 4139 | order = calculate_order(size); |
81819f0f | 4140 | |
19af27af | 4141 | if ((int)order < 0) |
81819f0f CL |
4142 | return 0; |
4143 | ||
b7a49f0d | 4144 | s->allocflags = 0; |
834f3d11 | 4145 | if (order) |
b7a49f0d CL |
4146 | s->allocflags |= __GFP_COMP; |
4147 | ||
4148 | if (s->flags & SLAB_CACHE_DMA) | |
2c59dd65 | 4149 | s->allocflags |= GFP_DMA; |
b7a49f0d | 4150 | |
6d6ea1e9 NB |
4151 | if (s->flags & SLAB_CACHE_DMA32) |
4152 | s->allocflags |= GFP_DMA32; | |
4153 | ||
b7a49f0d CL |
4154 | if (s->flags & SLAB_RECLAIM_ACCOUNT) |
4155 | s->allocflags |= __GFP_RECLAIMABLE; | |
4156 | ||
81819f0f CL |
4157 | /* |
4158 | * Determine the number of objects per slab | |
4159 | */ | |
9736d2a9 MW |
4160 | s->oo = oo_make(order, size); |
4161 | s->min = oo_make(get_order(size), size); | |
81819f0f | 4162 | |
834f3d11 | 4163 | return !!oo_objects(s->oo); |
81819f0f CL |
4164 | } |
4165 | ||
d50112ed | 4166 | static int kmem_cache_open(struct kmem_cache *s, slab_flags_t flags) |
81819f0f | 4167 | { |
37540008 | 4168 | s->flags = kmem_cache_flags(s->size, flags, s->name); |
2482ddec KC |
4169 | #ifdef CONFIG_SLAB_FREELIST_HARDENED |
4170 | s->random = get_random_long(); | |
4171 | #endif | |
81819f0f | 4172 | |
ae44d81d | 4173 | if (!calculate_sizes(s)) |
81819f0f | 4174 | goto error; |
3de47213 DR |
4175 | if (disable_higher_order_debug) { |
4176 | /* | |
4177 | * Disable debugging flags that store metadata if the min slab | |
4178 | * order increased. | |
4179 | */ | |
3b0efdfa | 4180 | if (get_order(s->size) > get_order(s->object_size)) { |
3de47213 DR |
4181 | s->flags &= ~DEBUG_METADATA_FLAGS; |
4182 | s->offset = 0; | |
ae44d81d | 4183 | if (!calculate_sizes(s)) |
3de47213 DR |
4184 | goto error; |
4185 | } | |
4186 | } | |
81819f0f | 4187 | |
2565409f HC |
4188 | #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \ |
4189 | defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE) | |
149daaf3 | 4190 | if (system_has_cmpxchg_double() && (s->flags & SLAB_NO_CMPXCHG) == 0) |
b789ef51 CL |
4191 | /* Enable fast mode */ |
4192 | s->flags |= __CMPXCHG_DOUBLE; | |
4193 | #endif | |
4194 | ||
3b89d7d8 | 4195 | /* |
c2092c12 | 4196 | * The larger the object size is, the more slabs we want on the partial |
3b89d7d8 DR |
4197 | * list to avoid pounding the page allocator excessively. |
4198 | */ | |
5182f3c9 HY |
4199 | s->min_partial = min_t(unsigned long, MAX_PARTIAL, ilog2(s->size) / 2); |
4200 | s->min_partial = max_t(unsigned long, MIN_PARTIAL, s->min_partial); | |
49e22585 | 4201 | |
e6d0e1dc | 4202 | set_cpu_partial(s); |
49e22585 | 4203 | |
81819f0f | 4204 | #ifdef CONFIG_NUMA |
e2cb96b7 | 4205 | s->remote_node_defrag_ratio = 1000; |
81819f0f | 4206 | #endif |
210e7a43 TG |
4207 | |
4208 | /* Initialize the pre-computed randomized freelist if slab is up */ | |
4209 | if (slab_state >= UP) { | |
4210 | if (init_cache_random_seq(s)) | |
4211 | goto error; | |
4212 | } | |
4213 | ||
55136592 | 4214 | if (!init_kmem_cache_nodes(s)) |
dfb4f096 | 4215 | goto error; |
81819f0f | 4216 | |
55136592 | 4217 | if (alloc_kmem_cache_cpus(s)) |
278b1bb1 | 4218 | return 0; |
ff12059e | 4219 | |
81819f0f | 4220 | error: |
9037c576 | 4221 | __kmem_cache_release(s); |
278b1bb1 | 4222 | return -EINVAL; |
81819f0f | 4223 | } |
81819f0f | 4224 | |
bb192ed9 | 4225 | static void list_slab_objects(struct kmem_cache *s, struct slab *slab, |
55860d96 | 4226 | const char *text) |
33b12c38 CL |
4227 | { |
4228 | #ifdef CONFIG_SLUB_DEBUG | |
bb192ed9 | 4229 | void *addr = slab_address(slab); |
a2b4ae8b | 4230 | unsigned long flags; |
55860d96 | 4231 | unsigned long *map; |
33b12c38 | 4232 | void *p; |
aa456c7a | 4233 | |
bb192ed9 VB |
4234 | slab_err(s, slab, text, s->name); |
4235 | slab_lock(slab, &flags); | |
33b12c38 | 4236 | |
bb192ed9 VB |
4237 | map = get_map(s, slab); |
4238 | for_each_object(p, s, addr, slab->objects) { | |
33b12c38 | 4239 | |
4138fdfc | 4240 | if (!test_bit(__obj_to_index(s, addr, p), map)) { |
96b94abc | 4241 | pr_err("Object 0x%p @offset=%tu\n", p, p - addr); |
33b12c38 CL |
4242 | print_tracking(s, p); |
4243 | } | |
4244 | } | |
55860d96 | 4245 | put_map(map); |
bb192ed9 | 4246 | slab_unlock(slab, &flags); |
33b12c38 CL |
4247 | #endif |
4248 | } | |
4249 | ||
81819f0f | 4250 | /* |
599870b1 | 4251 | * Attempt to free all partial slabs on a node. |
52b4b950 DS |
4252 | * This is called from __kmem_cache_shutdown(). We must take list_lock |
4253 | * because sysfs file might still access partial list after the shutdowning. | |
81819f0f | 4254 | */ |
599870b1 | 4255 | static void free_partial(struct kmem_cache *s, struct kmem_cache_node *n) |
81819f0f | 4256 | { |
60398923 | 4257 | LIST_HEAD(discard); |
bb192ed9 | 4258 | struct slab *slab, *h; |
81819f0f | 4259 | |
52b4b950 DS |
4260 | BUG_ON(irqs_disabled()); |
4261 | spin_lock_irq(&n->list_lock); | |
bb192ed9 VB |
4262 | list_for_each_entry_safe(slab, h, &n->partial, slab_list) { |
4263 | if (!slab->inuse) { | |
4264 | remove_partial(n, slab); | |
4265 | list_add(&slab->slab_list, &discard); | |
33b12c38 | 4266 | } else { |
bb192ed9 | 4267 | list_slab_objects(s, slab, |
55860d96 | 4268 | "Objects remaining in %s on __kmem_cache_shutdown()"); |
599870b1 | 4269 | } |
33b12c38 | 4270 | } |
52b4b950 | 4271 | spin_unlock_irq(&n->list_lock); |
60398923 | 4272 | |
bb192ed9 VB |
4273 | list_for_each_entry_safe(slab, h, &discard, slab_list) |
4274 | discard_slab(s, slab); | |
81819f0f CL |
4275 | } |
4276 | ||
f9e13c0a SB |
4277 | bool __kmem_cache_empty(struct kmem_cache *s) |
4278 | { | |
4279 | int node; | |
4280 | struct kmem_cache_node *n; | |
4281 | ||
4282 | for_each_kmem_cache_node(s, node, n) | |
4283 | if (n->nr_partial || slabs_node(s, node)) | |
4284 | return false; | |
4285 | return true; | |
4286 | } | |
4287 | ||
81819f0f | 4288 | /* |
672bba3a | 4289 | * Release all resources used by a slab cache. |
81819f0f | 4290 | */ |
52b4b950 | 4291 | int __kmem_cache_shutdown(struct kmem_cache *s) |
81819f0f CL |
4292 | { |
4293 | int node; | |
fa45dc25 | 4294 | struct kmem_cache_node *n; |
81819f0f | 4295 | |
5a836bf6 | 4296 | flush_all_cpus_locked(s); |
81819f0f | 4297 | /* Attempt to free all objects */ |
fa45dc25 | 4298 | for_each_kmem_cache_node(s, node, n) { |
599870b1 CL |
4299 | free_partial(s, n); |
4300 | if (n->nr_partial || slabs_node(s, node)) | |
81819f0f CL |
4301 | return 1; |
4302 | } | |
81819f0f CL |
4303 | return 0; |
4304 | } | |
4305 | ||
5bb1bb35 | 4306 | #ifdef CONFIG_PRINTK |
2dfe63e6 | 4307 | void __kmem_obj_info(struct kmem_obj_info *kpp, void *object, struct slab *slab) |
8e7f37f2 PM |
4308 | { |
4309 | void *base; | |
4310 | int __maybe_unused i; | |
4311 | unsigned int objnr; | |
4312 | void *objp; | |
4313 | void *objp0; | |
7213230a | 4314 | struct kmem_cache *s = slab->slab_cache; |
8e7f37f2 PM |
4315 | struct track __maybe_unused *trackp; |
4316 | ||
4317 | kpp->kp_ptr = object; | |
7213230a | 4318 | kpp->kp_slab = slab; |
8e7f37f2 | 4319 | kpp->kp_slab_cache = s; |
7213230a | 4320 | base = slab_address(slab); |
8e7f37f2 PM |
4321 | objp0 = kasan_reset_tag(object); |
4322 | #ifdef CONFIG_SLUB_DEBUG | |
4323 | objp = restore_red_left(s, objp0); | |
4324 | #else | |
4325 | objp = objp0; | |
4326 | #endif | |
40f3bf0c | 4327 | objnr = obj_to_index(s, slab, objp); |
8e7f37f2 PM |
4328 | kpp->kp_data_offset = (unsigned long)((char *)objp0 - (char *)objp); |
4329 | objp = base + s->size * objnr; | |
4330 | kpp->kp_objp = objp; | |
7213230a MWO |
4331 | if (WARN_ON_ONCE(objp < base || objp >= base + slab->objects * s->size |
4332 | || (objp - base) % s->size) || | |
8e7f37f2 PM |
4333 | !(s->flags & SLAB_STORE_USER)) |
4334 | return; | |
4335 | #ifdef CONFIG_SLUB_DEBUG | |
0cbc124b | 4336 | objp = fixup_red_left(s, objp); |
8e7f37f2 PM |
4337 | trackp = get_track(s, objp, TRACK_ALLOC); |
4338 | kpp->kp_ret = (void *)trackp->addr; | |
5cf909c5 OG |
4339 | #ifdef CONFIG_STACKDEPOT |
4340 | { | |
4341 | depot_stack_handle_t handle; | |
4342 | unsigned long *entries; | |
4343 | unsigned int nr_entries; | |
78869146 | 4344 | |
5cf909c5 OG |
4345 | handle = READ_ONCE(trackp->handle); |
4346 | if (handle) { | |
4347 | nr_entries = stack_depot_fetch(handle, &entries); | |
4348 | for (i = 0; i < KS_ADDRS_COUNT && i < nr_entries; i++) | |
4349 | kpp->kp_stack[i] = (void *)entries[i]; | |
4350 | } | |
78869146 | 4351 | |
5cf909c5 OG |
4352 | trackp = get_track(s, objp, TRACK_FREE); |
4353 | handle = READ_ONCE(trackp->handle); | |
4354 | if (handle) { | |
4355 | nr_entries = stack_depot_fetch(handle, &entries); | |
4356 | for (i = 0; i < KS_ADDRS_COUNT && i < nr_entries; i++) | |
4357 | kpp->kp_free_stack[i] = (void *)entries[i]; | |
4358 | } | |
e548eaa1 | 4359 | } |
8e7f37f2 PM |
4360 | #endif |
4361 | #endif | |
4362 | } | |
5bb1bb35 | 4363 | #endif |
8e7f37f2 | 4364 | |
81819f0f CL |
4365 | /******************************************************************** |
4366 | * Kmalloc subsystem | |
4367 | *******************************************************************/ | |
4368 | ||
81819f0f CL |
4369 | static int __init setup_slub_min_order(char *str) |
4370 | { | |
19af27af | 4371 | get_option(&str, (int *)&slub_min_order); |
81819f0f CL |
4372 | |
4373 | return 1; | |
4374 | } | |
4375 | ||
4376 | __setup("slub_min_order=", setup_slub_min_order); | |
4377 | ||
4378 | static int __init setup_slub_max_order(char *str) | |
4379 | { | |
19af27af AD |
4380 | get_option(&str, (int *)&slub_max_order); |
4381 | slub_max_order = min(slub_max_order, (unsigned int)MAX_ORDER - 1); | |
81819f0f CL |
4382 | |
4383 | return 1; | |
4384 | } | |
4385 | ||
4386 | __setup("slub_max_order=", setup_slub_max_order); | |
4387 | ||
4388 | static int __init setup_slub_min_objects(char *str) | |
4389 | { | |
19af27af | 4390 | get_option(&str, (int *)&slub_min_objects); |
81819f0f CL |
4391 | |
4392 | return 1; | |
4393 | } | |
4394 | ||
4395 | __setup("slub_min_objects=", setup_slub_min_objects); | |
4396 | ||
0f853b2e HY |
4397 | static __always_inline |
4398 | void *__do_kmalloc_node(size_t size, gfp_t flags, int node, unsigned long caller) | |
81819f0f | 4399 | { |
aadb4bc4 | 4400 | struct kmem_cache *s; |
5b882be4 | 4401 | void *ret; |
81819f0f | 4402 | |
95a05b42 | 4403 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) { |
bf37d791 | 4404 | ret = kmalloc_large_node_notrace(size, flags, node); |
5b882be4 | 4405 | |
0f853b2e | 4406 | trace_kmalloc_node(caller, ret, NULL, |
ca2b84cb EGM |
4407 | size, PAGE_SIZE << get_order(size), |
4408 | flags, node); | |
5b882be4 EGM |
4409 | |
4410 | return ret; | |
4411 | } | |
aadb4bc4 | 4412 | |
2c59dd65 | 4413 | s = kmalloc_slab(size, flags); |
aadb4bc4 CL |
4414 | |
4415 | if (unlikely(ZERO_OR_NULL_PTR(s))) | |
6cb8f913 CL |
4416 | return s; |
4417 | ||
0f853b2e | 4418 | ret = slab_alloc_node(s, NULL, flags, node, caller, size); |
5b882be4 | 4419 | |
0f853b2e | 4420 | trace_kmalloc_node(caller, ret, s, size, s->size, flags, node); |
5b882be4 | 4421 | |
0116523c | 4422 | ret = kasan_kmalloc(s, ret, size, flags); |
0316bec2 | 4423 | |
5b882be4 | 4424 | return ret; |
81819f0f | 4425 | } |
0f853b2e HY |
4426 | |
4427 | void *__kmalloc_node(size_t size, gfp_t flags, int node) | |
4428 | { | |
4429 | return __do_kmalloc_node(size, flags, node, _RET_IP_); | |
4430 | } | |
81819f0f | 4431 | EXPORT_SYMBOL(__kmalloc_node); |
81819f0f | 4432 | |
0f853b2e HY |
4433 | void *__kmalloc(size_t size, gfp_t flags) |
4434 | { | |
4435 | return __do_kmalloc_node(size, flags, NUMA_NO_NODE, _RET_IP_); | |
4436 | } | |
4437 | EXPORT_SYMBOL(__kmalloc); | |
4438 | ||
4439 | ||
ed18adc1 KC |
4440 | #ifdef CONFIG_HARDENED_USERCOPY |
4441 | /* | |
afcc90f8 KC |
4442 | * Rejects incorrectly sized objects and objects that are to be copied |
4443 | * to/from userspace but do not fall entirely within the containing slab | |
4444 | * cache's usercopy region. | |
ed18adc1 KC |
4445 | * |
4446 | * Returns NULL if check passes, otherwise const char * to name of cache | |
4447 | * to indicate an error. | |
4448 | */ | |
0b3eb091 MWO |
4449 | void __check_heap_object(const void *ptr, unsigned long n, |
4450 | const struct slab *slab, bool to_user) | |
ed18adc1 KC |
4451 | { |
4452 | struct kmem_cache *s; | |
44065b2e | 4453 | unsigned int offset; |
b89fb5ef | 4454 | bool is_kfence = is_kfence_address(ptr); |
ed18adc1 | 4455 | |
96fedce2 AK |
4456 | ptr = kasan_reset_tag(ptr); |
4457 | ||
ed18adc1 | 4458 | /* Find object and usable object size. */ |
0b3eb091 | 4459 | s = slab->slab_cache; |
ed18adc1 KC |
4460 | |
4461 | /* Reject impossible pointers. */ | |
0b3eb091 | 4462 | if (ptr < slab_address(slab)) |
f4e6e289 KC |
4463 | usercopy_abort("SLUB object not in SLUB page?!", NULL, |
4464 | to_user, 0, n); | |
ed18adc1 KC |
4465 | |
4466 | /* Find offset within object. */ | |
b89fb5ef AP |
4467 | if (is_kfence) |
4468 | offset = ptr - kfence_object_start(ptr); | |
4469 | else | |
0b3eb091 | 4470 | offset = (ptr - slab_address(slab)) % s->size; |
ed18adc1 KC |
4471 | |
4472 | /* Adjust for redzone and reject if within the redzone. */ | |
b89fb5ef | 4473 | if (!is_kfence && kmem_cache_debug_flags(s, SLAB_RED_ZONE)) { |
ed18adc1 | 4474 | if (offset < s->red_left_pad) |
f4e6e289 KC |
4475 | usercopy_abort("SLUB object in left red zone", |
4476 | s->name, to_user, offset, n); | |
ed18adc1 KC |
4477 | offset -= s->red_left_pad; |
4478 | } | |
4479 | ||
afcc90f8 KC |
4480 | /* Allow address range falling entirely within usercopy region. */ |
4481 | if (offset >= s->useroffset && | |
4482 | offset - s->useroffset <= s->usersize && | |
4483 | n <= s->useroffset - offset + s->usersize) | |
f4e6e289 | 4484 | return; |
ed18adc1 | 4485 | |
f4e6e289 | 4486 | usercopy_abort("SLUB object", s->name, to_user, offset, n); |
ed18adc1 KC |
4487 | } |
4488 | #endif /* CONFIG_HARDENED_USERCOPY */ | |
4489 | ||
10d1f8cb | 4490 | size_t __ksize(const void *object) |
81819f0f | 4491 | { |
0c24811b | 4492 | struct folio *folio; |
81819f0f | 4493 | |
ef8b4520 | 4494 | if (unlikely(object == ZERO_SIZE_PTR)) |
272c1d21 CL |
4495 | return 0; |
4496 | ||
0c24811b | 4497 | folio = virt_to_folio(object); |
294a80a8 | 4498 | |
0c24811b MWO |
4499 | if (unlikely(!folio_test_slab(folio))) |
4500 | return folio_size(folio); | |
81819f0f | 4501 | |
0c24811b | 4502 | return slab_ksize(folio_slab(folio)->slab_cache); |
81819f0f | 4503 | } |
10d1f8cb | 4504 | EXPORT_SYMBOL(__ksize); |
81819f0f CL |
4505 | |
4506 | void kfree(const void *x) | |
4507 | { | |
d835eef4 MWO |
4508 | struct folio *folio; |
4509 | struct slab *slab; | |
5bb983b0 | 4510 | void *object = (void *)x; |
81819f0f | 4511 | |
2121db74 PE |
4512 | trace_kfree(_RET_IP_, x); |
4513 | ||
2408c550 | 4514 | if (unlikely(ZERO_OR_NULL_PTR(x))) |
81819f0f CL |
4515 | return; |
4516 | ||
d835eef4 MWO |
4517 | folio = virt_to_folio(x); |
4518 | if (unlikely(!folio_test_slab(folio))) { | |
4519 | free_large_kmalloc(folio, object); | |
aadb4bc4 CL |
4520 | return; |
4521 | } | |
d835eef4 | 4522 | slab = folio_slab(folio); |
b77d5b1b | 4523 | slab_free(slab->slab_cache, slab, object, NULL, &object, 1, _RET_IP_); |
81819f0f CL |
4524 | } |
4525 | EXPORT_SYMBOL(kfree); | |
4526 | ||
832f37f5 VD |
4527 | #define SHRINK_PROMOTE_MAX 32 |
4528 | ||
2086d26a | 4529 | /* |
832f37f5 VD |
4530 | * kmem_cache_shrink discards empty slabs and promotes the slabs filled |
4531 | * up most to the head of the partial lists. New allocations will then | |
4532 | * fill those up and thus they can be removed from the partial lists. | |
672bba3a CL |
4533 | * |
4534 | * The slabs with the least items are placed last. This results in them | |
4535 | * being allocated from last increasing the chance that the last objects | |
4536 | * are freed in them. | |
2086d26a | 4537 | */ |
5a836bf6 | 4538 | static int __kmem_cache_do_shrink(struct kmem_cache *s) |
2086d26a CL |
4539 | { |
4540 | int node; | |
4541 | int i; | |
4542 | struct kmem_cache_node *n; | |
bb192ed9 VB |
4543 | struct slab *slab; |
4544 | struct slab *t; | |
832f37f5 VD |
4545 | struct list_head discard; |
4546 | struct list_head promote[SHRINK_PROMOTE_MAX]; | |
2086d26a | 4547 | unsigned long flags; |
ce3712d7 | 4548 | int ret = 0; |
2086d26a | 4549 | |
fa45dc25 | 4550 | for_each_kmem_cache_node(s, node, n) { |
832f37f5 VD |
4551 | INIT_LIST_HEAD(&discard); |
4552 | for (i = 0; i < SHRINK_PROMOTE_MAX; i++) | |
4553 | INIT_LIST_HEAD(promote + i); | |
2086d26a CL |
4554 | |
4555 | spin_lock_irqsave(&n->list_lock, flags); | |
4556 | ||
4557 | /* | |
832f37f5 | 4558 | * Build lists of slabs to discard or promote. |
2086d26a | 4559 | * |
672bba3a | 4560 | * Note that concurrent frees may occur while we hold the |
c2092c12 | 4561 | * list_lock. slab->inuse here is the upper limit. |
2086d26a | 4562 | */ |
bb192ed9 VB |
4563 | list_for_each_entry_safe(slab, t, &n->partial, slab_list) { |
4564 | int free = slab->objects - slab->inuse; | |
832f37f5 | 4565 | |
c2092c12 | 4566 | /* Do not reread slab->inuse */ |
832f37f5 VD |
4567 | barrier(); |
4568 | ||
4569 | /* We do not keep full slabs on the list */ | |
4570 | BUG_ON(free <= 0); | |
4571 | ||
bb192ed9 VB |
4572 | if (free == slab->objects) { |
4573 | list_move(&slab->slab_list, &discard); | |
69cb8e6b | 4574 | n->nr_partial--; |
832f37f5 | 4575 | } else if (free <= SHRINK_PROMOTE_MAX) |
bb192ed9 | 4576 | list_move(&slab->slab_list, promote + free - 1); |
2086d26a CL |
4577 | } |
4578 | ||
2086d26a | 4579 | /* |
832f37f5 VD |
4580 | * Promote the slabs filled up most to the head of the |
4581 | * partial list. | |
2086d26a | 4582 | */ |
832f37f5 VD |
4583 | for (i = SHRINK_PROMOTE_MAX - 1; i >= 0; i--) |
4584 | list_splice(promote + i, &n->partial); | |
2086d26a | 4585 | |
2086d26a | 4586 | spin_unlock_irqrestore(&n->list_lock, flags); |
69cb8e6b CL |
4587 | |
4588 | /* Release empty slabs */ | |
bb192ed9 VB |
4589 | list_for_each_entry_safe(slab, t, &discard, slab_list) |
4590 | discard_slab(s, slab); | |
ce3712d7 VD |
4591 | |
4592 | if (slabs_node(s, node)) | |
4593 | ret = 1; | |
2086d26a CL |
4594 | } |
4595 | ||
ce3712d7 | 4596 | return ret; |
2086d26a | 4597 | } |
2086d26a | 4598 | |
5a836bf6 SAS |
4599 | int __kmem_cache_shrink(struct kmem_cache *s) |
4600 | { | |
4601 | flush_all(s); | |
4602 | return __kmem_cache_do_shrink(s); | |
4603 | } | |
4604 | ||
b9049e23 YG |
4605 | static int slab_mem_going_offline_callback(void *arg) |
4606 | { | |
4607 | struct kmem_cache *s; | |
4608 | ||
18004c5d | 4609 | mutex_lock(&slab_mutex); |
5a836bf6 SAS |
4610 | list_for_each_entry(s, &slab_caches, list) { |
4611 | flush_all_cpus_locked(s); | |
4612 | __kmem_cache_do_shrink(s); | |
4613 | } | |
18004c5d | 4614 | mutex_unlock(&slab_mutex); |
b9049e23 YG |
4615 | |
4616 | return 0; | |
4617 | } | |
4618 | ||
4619 | static void slab_mem_offline_callback(void *arg) | |
4620 | { | |
b9049e23 YG |
4621 | struct memory_notify *marg = arg; |
4622 | int offline_node; | |
4623 | ||
b9d5ab25 | 4624 | offline_node = marg->status_change_nid_normal; |
b9049e23 YG |
4625 | |
4626 | /* | |
4627 | * If the node still has available memory. we need kmem_cache_node | |
4628 | * for it yet. | |
4629 | */ | |
4630 | if (offline_node < 0) | |
4631 | return; | |
4632 | ||
18004c5d | 4633 | mutex_lock(&slab_mutex); |
7e1fa93d | 4634 | node_clear(offline_node, slab_nodes); |
666716fd VB |
4635 | /* |
4636 | * We no longer free kmem_cache_node structures here, as it would be | |
4637 | * racy with all get_node() users, and infeasible to protect them with | |
4638 | * slab_mutex. | |
4639 | */ | |
18004c5d | 4640 | mutex_unlock(&slab_mutex); |
b9049e23 YG |
4641 | } |
4642 | ||
4643 | static int slab_mem_going_online_callback(void *arg) | |
4644 | { | |
4645 | struct kmem_cache_node *n; | |
4646 | struct kmem_cache *s; | |
4647 | struct memory_notify *marg = arg; | |
b9d5ab25 | 4648 | int nid = marg->status_change_nid_normal; |
b9049e23 YG |
4649 | int ret = 0; |
4650 | ||
4651 | /* | |
4652 | * If the node's memory is already available, then kmem_cache_node is | |
4653 | * already created. Nothing to do. | |
4654 | */ | |
4655 | if (nid < 0) | |
4656 | return 0; | |
4657 | ||
4658 | /* | |
0121c619 | 4659 | * We are bringing a node online. No memory is available yet. We must |
b9049e23 YG |
4660 | * allocate a kmem_cache_node structure in order to bring the node |
4661 | * online. | |
4662 | */ | |
18004c5d | 4663 | mutex_lock(&slab_mutex); |
b9049e23 | 4664 | list_for_each_entry(s, &slab_caches, list) { |
666716fd VB |
4665 | /* |
4666 | * The structure may already exist if the node was previously | |
4667 | * onlined and offlined. | |
4668 | */ | |
4669 | if (get_node(s, nid)) | |
4670 | continue; | |
b9049e23 YG |
4671 | /* |
4672 | * XXX: kmem_cache_alloc_node will fallback to other nodes | |
4673 | * since memory is not yet available from the node that | |
4674 | * is brought up. | |
4675 | */ | |
8de66a0c | 4676 | n = kmem_cache_alloc(kmem_cache_node, GFP_KERNEL); |
b9049e23 YG |
4677 | if (!n) { |
4678 | ret = -ENOMEM; | |
4679 | goto out; | |
4680 | } | |
4053497d | 4681 | init_kmem_cache_node(n); |
b9049e23 YG |
4682 | s->node[nid] = n; |
4683 | } | |
7e1fa93d VB |
4684 | /* |
4685 | * Any cache created after this point will also have kmem_cache_node | |
4686 | * initialized for the new node. | |
4687 | */ | |
4688 | node_set(nid, slab_nodes); | |
b9049e23 | 4689 | out: |
18004c5d | 4690 | mutex_unlock(&slab_mutex); |
b9049e23 YG |
4691 | return ret; |
4692 | } | |
4693 | ||
4694 | static int slab_memory_callback(struct notifier_block *self, | |
4695 | unsigned long action, void *arg) | |
4696 | { | |
4697 | int ret = 0; | |
4698 | ||
4699 | switch (action) { | |
4700 | case MEM_GOING_ONLINE: | |
4701 | ret = slab_mem_going_online_callback(arg); | |
4702 | break; | |
4703 | case MEM_GOING_OFFLINE: | |
4704 | ret = slab_mem_going_offline_callback(arg); | |
4705 | break; | |
4706 | case MEM_OFFLINE: | |
4707 | case MEM_CANCEL_ONLINE: | |
4708 | slab_mem_offline_callback(arg); | |
4709 | break; | |
4710 | case MEM_ONLINE: | |
4711 | case MEM_CANCEL_OFFLINE: | |
4712 | break; | |
4713 | } | |
dc19f9db KH |
4714 | if (ret) |
4715 | ret = notifier_from_errno(ret); | |
4716 | else | |
4717 | ret = NOTIFY_OK; | |
b9049e23 YG |
4718 | return ret; |
4719 | } | |
4720 | ||
3ac38faa AM |
4721 | static struct notifier_block slab_memory_callback_nb = { |
4722 | .notifier_call = slab_memory_callback, | |
4723 | .priority = SLAB_CALLBACK_PRI, | |
4724 | }; | |
b9049e23 | 4725 | |
81819f0f CL |
4726 | /******************************************************************** |
4727 | * Basic setup of slabs | |
4728 | *******************************************************************/ | |
4729 | ||
51df1142 CL |
4730 | /* |
4731 | * Used for early kmem_cache structures that were allocated using | |
dffb4d60 CL |
4732 | * the page allocator. Allocate them properly then fix up the pointers |
4733 | * that may be pointing to the wrong kmem_cache structure. | |
51df1142 CL |
4734 | */ |
4735 | ||
dffb4d60 | 4736 | static struct kmem_cache * __init bootstrap(struct kmem_cache *static_cache) |
51df1142 CL |
4737 | { |
4738 | int node; | |
dffb4d60 | 4739 | struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT); |
fa45dc25 | 4740 | struct kmem_cache_node *n; |
51df1142 | 4741 | |
dffb4d60 | 4742 | memcpy(s, static_cache, kmem_cache->object_size); |
51df1142 | 4743 | |
7d557b3c GC |
4744 | /* |
4745 | * This runs very early, and only the boot processor is supposed to be | |
4746 | * up. Even if it weren't true, IRQs are not up so we couldn't fire | |
4747 | * IPIs around. | |
4748 | */ | |
4749 | __flush_cpu_slab(s, smp_processor_id()); | |
fa45dc25 | 4750 | for_each_kmem_cache_node(s, node, n) { |
bb192ed9 | 4751 | struct slab *p; |
51df1142 | 4752 | |
916ac052 | 4753 | list_for_each_entry(p, &n->partial, slab_list) |
fa45dc25 | 4754 | p->slab_cache = s; |
51df1142 | 4755 | |
607bf324 | 4756 | #ifdef CONFIG_SLUB_DEBUG |
916ac052 | 4757 | list_for_each_entry(p, &n->full, slab_list) |
fa45dc25 | 4758 | p->slab_cache = s; |
51df1142 | 4759 | #endif |
51df1142 | 4760 | } |
dffb4d60 CL |
4761 | list_add(&s->list, &slab_caches); |
4762 | return s; | |
51df1142 CL |
4763 | } |
4764 | ||
81819f0f CL |
4765 | void __init kmem_cache_init(void) |
4766 | { | |
dffb4d60 CL |
4767 | static __initdata struct kmem_cache boot_kmem_cache, |
4768 | boot_kmem_cache_node; | |
7e1fa93d | 4769 | int node; |
51df1142 | 4770 | |
fc8d8620 SG |
4771 | if (debug_guardpage_minorder()) |
4772 | slub_max_order = 0; | |
4773 | ||
79270291 SB |
4774 | /* Print slub debugging pointers without hashing */ |
4775 | if (__slub_debug_enabled()) | |
4776 | no_hash_pointers_enable(NULL); | |
4777 | ||
dffb4d60 CL |
4778 | kmem_cache_node = &boot_kmem_cache_node; |
4779 | kmem_cache = &boot_kmem_cache; | |
51df1142 | 4780 | |
7e1fa93d VB |
4781 | /* |
4782 | * Initialize the nodemask for which we will allocate per node | |
4783 | * structures. Here we don't need taking slab_mutex yet. | |
4784 | */ | |
4785 | for_each_node_state(node, N_NORMAL_MEMORY) | |
4786 | node_set(node, slab_nodes); | |
4787 | ||
dffb4d60 | 4788 | create_boot_cache(kmem_cache_node, "kmem_cache_node", |
8eb8284b | 4789 | sizeof(struct kmem_cache_node), SLAB_HWCACHE_ALIGN, 0, 0); |
b9049e23 | 4790 | |
3ac38faa | 4791 | register_hotmemory_notifier(&slab_memory_callback_nb); |
81819f0f CL |
4792 | |
4793 | /* Able to allocate the per node structures */ | |
4794 | slab_state = PARTIAL; | |
4795 | ||
dffb4d60 CL |
4796 | create_boot_cache(kmem_cache, "kmem_cache", |
4797 | offsetof(struct kmem_cache, node) + | |
4798 | nr_node_ids * sizeof(struct kmem_cache_node *), | |
8eb8284b | 4799 | SLAB_HWCACHE_ALIGN, 0, 0); |
8a13a4cc | 4800 | |
dffb4d60 | 4801 | kmem_cache = bootstrap(&boot_kmem_cache); |
dffb4d60 | 4802 | kmem_cache_node = bootstrap(&boot_kmem_cache_node); |
51df1142 CL |
4803 | |
4804 | /* Now we can use the kmem_cache to allocate kmalloc slabs */ | |
34cc6990 | 4805 | setup_kmalloc_cache_index_table(); |
f97d5f63 | 4806 | create_kmalloc_caches(0); |
81819f0f | 4807 | |
210e7a43 TG |
4808 | /* Setup random freelists for each cache */ |
4809 | init_freelist_randomization(); | |
4810 | ||
a96a87bf SAS |
4811 | cpuhp_setup_state_nocalls(CPUHP_SLUB_DEAD, "slub:dead", NULL, |
4812 | slub_cpu_dead); | |
81819f0f | 4813 | |
b9726c26 | 4814 | pr_info("SLUB: HWalign=%d, Order=%u-%u, MinObjects=%u, CPUs=%u, Nodes=%u\n", |
f97d5f63 | 4815 | cache_line_size(), |
81819f0f CL |
4816 | slub_min_order, slub_max_order, slub_min_objects, |
4817 | nr_cpu_ids, nr_node_ids); | |
4818 | } | |
4819 | ||
7e85ee0c PE |
4820 | void __init kmem_cache_init_late(void) |
4821 | { | |
7e85ee0c PE |
4822 | } |
4823 | ||
2633d7a0 | 4824 | struct kmem_cache * |
f4957d5b | 4825 | __kmem_cache_alias(const char *name, unsigned int size, unsigned int align, |
d50112ed | 4826 | slab_flags_t flags, void (*ctor)(void *)) |
81819f0f | 4827 | { |
10befea9 | 4828 | struct kmem_cache *s; |
81819f0f | 4829 | |
a44cb944 | 4830 | s = find_mergeable(size, align, flags, name, ctor); |
81819f0f | 4831 | if (s) { |
efb93527 XS |
4832 | if (sysfs_slab_alias(s, name)) |
4833 | return NULL; | |
4834 | ||
81819f0f | 4835 | s->refcount++; |
84d0ddd6 | 4836 | |
81819f0f CL |
4837 | /* |
4838 | * Adjust the object sizes so that we clear | |
4839 | * the complete object on kzalloc. | |
4840 | */ | |
1b473f29 | 4841 | s->object_size = max(s->object_size, size); |
52ee6d74 | 4842 | s->inuse = max(s->inuse, ALIGN(size, sizeof(void *))); |
a0e1d1be | 4843 | } |
6446faa2 | 4844 | |
cbb79694 CL |
4845 | return s; |
4846 | } | |
84c1cf62 | 4847 | |
d50112ed | 4848 | int __kmem_cache_create(struct kmem_cache *s, slab_flags_t flags) |
cbb79694 | 4849 | { |
aac3a166 PE |
4850 | int err; |
4851 | ||
4852 | err = kmem_cache_open(s, flags); | |
4853 | if (err) | |
4854 | return err; | |
20cea968 | 4855 | |
45530c44 CL |
4856 | /* Mutex is not taken during early boot */ |
4857 | if (slab_state <= UP) | |
4858 | return 0; | |
4859 | ||
aac3a166 | 4860 | err = sysfs_slab_add(s); |
67823a54 | 4861 | if (err) { |
52b4b950 | 4862 | __kmem_cache_release(s); |
67823a54 ML |
4863 | return err; |
4864 | } | |
20cea968 | 4865 | |
64dd6849 FM |
4866 | if (s->flags & SLAB_STORE_USER) |
4867 | debugfs_slab_add(s); | |
4868 | ||
67823a54 | 4869 | return 0; |
81819f0f | 4870 | } |
81819f0f | 4871 | |
81819f0f | 4872 | void *__kmalloc_node_track_caller(size_t size, gfp_t gfpflags, |
0f853b2e | 4873 | int node, unsigned long caller) |
81819f0f | 4874 | { |
0f853b2e | 4875 | return __do_kmalloc_node(size, gfpflags, node, caller); |
81819f0f | 4876 | } |
fd7cb575 | 4877 | EXPORT_SYMBOL(__kmalloc_node_track_caller); |
81819f0f | 4878 | |
ab4d5ed5 | 4879 | #ifdef CONFIG_SYSFS |
bb192ed9 | 4880 | static int count_inuse(struct slab *slab) |
205ab99d | 4881 | { |
bb192ed9 | 4882 | return slab->inuse; |
205ab99d CL |
4883 | } |
4884 | ||
bb192ed9 | 4885 | static int count_total(struct slab *slab) |
205ab99d | 4886 | { |
bb192ed9 | 4887 | return slab->objects; |
205ab99d | 4888 | } |
ab4d5ed5 | 4889 | #endif |
205ab99d | 4890 | |
ab4d5ed5 | 4891 | #ifdef CONFIG_SLUB_DEBUG |
bb192ed9 | 4892 | static void validate_slab(struct kmem_cache *s, struct slab *slab, |
0a19e7dd | 4893 | unsigned long *obj_map) |
53e15af0 CL |
4894 | { |
4895 | void *p; | |
bb192ed9 | 4896 | void *addr = slab_address(slab); |
a2b4ae8b | 4897 | unsigned long flags; |
90e9f6a6 | 4898 | |
bb192ed9 | 4899 | slab_lock(slab, &flags); |
53e15af0 | 4900 | |
bb192ed9 | 4901 | if (!check_slab(s, slab) || !on_freelist(s, slab, NULL)) |
90e9f6a6 | 4902 | goto unlock; |
53e15af0 CL |
4903 | |
4904 | /* Now we know that a valid freelist exists */ | |
bb192ed9 VB |
4905 | __fill_map(obj_map, s, slab); |
4906 | for_each_object(p, s, addr, slab->objects) { | |
0a19e7dd | 4907 | u8 val = test_bit(__obj_to_index(s, addr, p), obj_map) ? |
dd98afd4 | 4908 | SLUB_RED_INACTIVE : SLUB_RED_ACTIVE; |
53e15af0 | 4909 | |
bb192ed9 | 4910 | if (!check_object(s, slab, p, val)) |
dd98afd4 YZ |
4911 | break; |
4912 | } | |
90e9f6a6 | 4913 | unlock: |
bb192ed9 | 4914 | slab_unlock(slab, &flags); |
53e15af0 CL |
4915 | } |
4916 | ||
434e245d | 4917 | static int validate_slab_node(struct kmem_cache *s, |
0a19e7dd | 4918 | struct kmem_cache_node *n, unsigned long *obj_map) |
53e15af0 CL |
4919 | { |
4920 | unsigned long count = 0; | |
bb192ed9 | 4921 | struct slab *slab; |
53e15af0 CL |
4922 | unsigned long flags; |
4923 | ||
4924 | spin_lock_irqsave(&n->list_lock, flags); | |
4925 | ||
bb192ed9 VB |
4926 | list_for_each_entry(slab, &n->partial, slab_list) { |
4927 | validate_slab(s, slab, obj_map); | |
53e15af0 CL |
4928 | count++; |
4929 | } | |
1f9f78b1 | 4930 | if (count != n->nr_partial) { |
f9f58285 FF |
4931 | pr_err("SLUB %s: %ld partial slabs counted but counter=%ld\n", |
4932 | s->name, count, n->nr_partial); | |
1f9f78b1 OG |
4933 | slab_add_kunit_errors(); |
4934 | } | |
53e15af0 CL |
4935 | |
4936 | if (!(s->flags & SLAB_STORE_USER)) | |
4937 | goto out; | |
4938 | ||
bb192ed9 VB |
4939 | list_for_each_entry(slab, &n->full, slab_list) { |
4940 | validate_slab(s, slab, obj_map); | |
53e15af0 CL |
4941 | count++; |
4942 | } | |
1f9f78b1 | 4943 | if (count != atomic_long_read(&n->nr_slabs)) { |
f9f58285 FF |
4944 | pr_err("SLUB: %s %ld slabs counted but counter=%ld\n", |
4945 | s->name, count, atomic_long_read(&n->nr_slabs)); | |
1f9f78b1 OG |
4946 | slab_add_kunit_errors(); |
4947 | } | |
53e15af0 CL |
4948 | |
4949 | out: | |
4950 | spin_unlock_irqrestore(&n->list_lock, flags); | |
4951 | return count; | |
4952 | } | |
4953 | ||
1f9f78b1 | 4954 | long validate_slab_cache(struct kmem_cache *s) |
53e15af0 CL |
4955 | { |
4956 | int node; | |
4957 | unsigned long count = 0; | |
fa45dc25 | 4958 | struct kmem_cache_node *n; |
0a19e7dd VB |
4959 | unsigned long *obj_map; |
4960 | ||
4961 | obj_map = bitmap_alloc(oo_objects(s->oo), GFP_KERNEL); | |
4962 | if (!obj_map) | |
4963 | return -ENOMEM; | |
53e15af0 CL |
4964 | |
4965 | flush_all(s); | |
fa45dc25 | 4966 | for_each_kmem_cache_node(s, node, n) |
0a19e7dd VB |
4967 | count += validate_slab_node(s, n, obj_map); |
4968 | ||
4969 | bitmap_free(obj_map); | |
90e9f6a6 | 4970 | |
53e15af0 CL |
4971 | return count; |
4972 | } | |
1f9f78b1 OG |
4973 | EXPORT_SYMBOL(validate_slab_cache); |
4974 | ||
64dd6849 | 4975 | #ifdef CONFIG_DEBUG_FS |
88a420e4 | 4976 | /* |
672bba3a | 4977 | * Generate lists of code addresses where slabcache objects are allocated |
88a420e4 CL |
4978 | * and freed. |
4979 | */ | |
4980 | ||
4981 | struct location { | |
8ea9fb92 | 4982 | depot_stack_handle_t handle; |
88a420e4 | 4983 | unsigned long count; |
ce71e27c | 4984 | unsigned long addr; |
45edfa58 CL |
4985 | long long sum_time; |
4986 | long min_time; | |
4987 | long max_time; | |
4988 | long min_pid; | |
4989 | long max_pid; | |
174596a0 | 4990 | DECLARE_BITMAP(cpus, NR_CPUS); |
45edfa58 | 4991 | nodemask_t nodes; |
88a420e4 CL |
4992 | }; |
4993 | ||
4994 | struct loc_track { | |
4995 | unsigned long max; | |
4996 | unsigned long count; | |
4997 | struct location *loc; | |
005a79e5 | 4998 | loff_t idx; |
88a420e4 CL |
4999 | }; |
5000 | ||
64dd6849 FM |
5001 | static struct dentry *slab_debugfs_root; |
5002 | ||
88a420e4 CL |
5003 | static void free_loc_track(struct loc_track *t) |
5004 | { | |
5005 | if (t->max) | |
5006 | free_pages((unsigned long)t->loc, | |
5007 | get_order(sizeof(struct location) * t->max)); | |
5008 | } | |
5009 | ||
68dff6a9 | 5010 | static int alloc_loc_track(struct loc_track *t, unsigned long max, gfp_t flags) |
88a420e4 CL |
5011 | { |
5012 | struct location *l; | |
5013 | int order; | |
5014 | ||
88a420e4 CL |
5015 | order = get_order(sizeof(struct location) * max); |
5016 | ||
68dff6a9 | 5017 | l = (void *)__get_free_pages(flags, order); |
88a420e4 CL |
5018 | if (!l) |
5019 | return 0; | |
5020 | ||
5021 | if (t->count) { | |
5022 | memcpy(l, t->loc, sizeof(struct location) * t->count); | |
5023 | free_loc_track(t); | |
5024 | } | |
5025 | t->max = max; | |
5026 | t->loc = l; | |
5027 | return 1; | |
5028 | } | |
5029 | ||
5030 | static int add_location(struct loc_track *t, struct kmem_cache *s, | |
45edfa58 | 5031 | const struct track *track) |
88a420e4 CL |
5032 | { |
5033 | long start, end, pos; | |
5034 | struct location *l; | |
8ea9fb92 | 5035 | unsigned long caddr, chandle; |
45edfa58 | 5036 | unsigned long age = jiffies - track->when; |
8ea9fb92 | 5037 | depot_stack_handle_t handle = 0; |
88a420e4 | 5038 | |
8ea9fb92 OG |
5039 | #ifdef CONFIG_STACKDEPOT |
5040 | handle = READ_ONCE(track->handle); | |
5041 | #endif | |
88a420e4 CL |
5042 | start = -1; |
5043 | end = t->count; | |
5044 | ||
5045 | for ( ; ; ) { | |
5046 | pos = start + (end - start + 1) / 2; | |
5047 | ||
5048 | /* | |
5049 | * There is nothing at "end". If we end up there | |
5050 | * we need to add something to before end. | |
5051 | */ | |
5052 | if (pos == end) | |
5053 | break; | |
5054 | ||
5055 | caddr = t->loc[pos].addr; | |
8ea9fb92 OG |
5056 | chandle = t->loc[pos].handle; |
5057 | if ((track->addr == caddr) && (handle == chandle)) { | |
45edfa58 CL |
5058 | |
5059 | l = &t->loc[pos]; | |
5060 | l->count++; | |
5061 | if (track->when) { | |
5062 | l->sum_time += age; | |
5063 | if (age < l->min_time) | |
5064 | l->min_time = age; | |
5065 | if (age > l->max_time) | |
5066 | l->max_time = age; | |
5067 | ||
5068 | if (track->pid < l->min_pid) | |
5069 | l->min_pid = track->pid; | |
5070 | if (track->pid > l->max_pid) | |
5071 | l->max_pid = track->pid; | |
5072 | ||
174596a0 RR |
5073 | cpumask_set_cpu(track->cpu, |
5074 | to_cpumask(l->cpus)); | |
45edfa58 CL |
5075 | } |
5076 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
88a420e4 CL |
5077 | return 1; |
5078 | } | |
5079 | ||
45edfa58 | 5080 | if (track->addr < caddr) |
88a420e4 | 5081 | end = pos; |
8ea9fb92 OG |
5082 | else if (track->addr == caddr && handle < chandle) |
5083 | end = pos; | |
88a420e4 CL |
5084 | else |
5085 | start = pos; | |
5086 | } | |
5087 | ||
5088 | /* | |
672bba3a | 5089 | * Not found. Insert new tracking element. |
88a420e4 | 5090 | */ |
68dff6a9 | 5091 | if (t->count >= t->max && !alloc_loc_track(t, 2 * t->max, GFP_ATOMIC)) |
88a420e4 CL |
5092 | return 0; |
5093 | ||
5094 | l = t->loc + pos; | |
5095 | if (pos < t->count) | |
5096 | memmove(l + 1, l, | |
5097 | (t->count - pos) * sizeof(struct location)); | |
5098 | t->count++; | |
5099 | l->count = 1; | |
45edfa58 CL |
5100 | l->addr = track->addr; |
5101 | l->sum_time = age; | |
5102 | l->min_time = age; | |
5103 | l->max_time = age; | |
5104 | l->min_pid = track->pid; | |
5105 | l->max_pid = track->pid; | |
8ea9fb92 | 5106 | l->handle = handle; |
174596a0 RR |
5107 | cpumask_clear(to_cpumask(l->cpus)); |
5108 | cpumask_set_cpu(track->cpu, to_cpumask(l->cpus)); | |
45edfa58 CL |
5109 | nodes_clear(l->nodes); |
5110 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
88a420e4 CL |
5111 | return 1; |
5112 | } | |
5113 | ||
5114 | static void process_slab(struct loc_track *t, struct kmem_cache *s, | |
bb192ed9 | 5115 | struct slab *slab, enum track_item alloc, |
b3fd64e1 | 5116 | unsigned long *obj_map) |
88a420e4 | 5117 | { |
bb192ed9 | 5118 | void *addr = slab_address(slab); |
88a420e4 CL |
5119 | void *p; |
5120 | ||
bb192ed9 | 5121 | __fill_map(obj_map, s, slab); |
b3fd64e1 | 5122 | |
bb192ed9 | 5123 | for_each_object(p, s, addr, slab->objects) |
b3fd64e1 | 5124 | if (!test_bit(__obj_to_index(s, addr, p), obj_map)) |
45edfa58 | 5125 | add_location(t, s, get_track(s, p, alloc)); |
88a420e4 | 5126 | } |
64dd6849 | 5127 | #endif /* CONFIG_DEBUG_FS */ |
6dfd1b65 | 5128 | #endif /* CONFIG_SLUB_DEBUG */ |
88a420e4 | 5129 | |
ab4d5ed5 | 5130 | #ifdef CONFIG_SYSFS |
81819f0f | 5131 | enum slab_stat_type { |
205ab99d CL |
5132 | SL_ALL, /* All slabs */ |
5133 | SL_PARTIAL, /* Only partially allocated slabs */ | |
5134 | SL_CPU, /* Only slabs used for cpu caches */ | |
5135 | SL_OBJECTS, /* Determine allocated objects not slabs */ | |
5136 | SL_TOTAL /* Determine object capacity not slabs */ | |
81819f0f CL |
5137 | }; |
5138 | ||
205ab99d | 5139 | #define SO_ALL (1 << SL_ALL) |
81819f0f CL |
5140 | #define SO_PARTIAL (1 << SL_PARTIAL) |
5141 | #define SO_CPU (1 << SL_CPU) | |
5142 | #define SO_OBJECTS (1 << SL_OBJECTS) | |
205ab99d | 5143 | #define SO_TOTAL (1 << SL_TOTAL) |
81819f0f | 5144 | |
62e5c4b4 | 5145 | static ssize_t show_slab_objects(struct kmem_cache *s, |
bf16d19a | 5146 | char *buf, unsigned long flags) |
81819f0f CL |
5147 | { |
5148 | unsigned long total = 0; | |
81819f0f CL |
5149 | int node; |
5150 | int x; | |
5151 | unsigned long *nodes; | |
bf16d19a | 5152 | int len = 0; |
81819f0f | 5153 | |
6396bb22 | 5154 | nodes = kcalloc(nr_node_ids, sizeof(unsigned long), GFP_KERNEL); |
62e5c4b4 CG |
5155 | if (!nodes) |
5156 | return -ENOMEM; | |
81819f0f | 5157 | |
205ab99d CL |
5158 | if (flags & SO_CPU) { |
5159 | int cpu; | |
81819f0f | 5160 | |
205ab99d | 5161 | for_each_possible_cpu(cpu) { |
d0e0ac97 CG |
5162 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, |
5163 | cpu); | |
ec3ab083 | 5164 | int node; |
bb192ed9 | 5165 | struct slab *slab; |
dfb4f096 | 5166 | |
bb192ed9 VB |
5167 | slab = READ_ONCE(c->slab); |
5168 | if (!slab) | |
ec3ab083 | 5169 | continue; |
205ab99d | 5170 | |
bb192ed9 | 5171 | node = slab_nid(slab); |
ec3ab083 | 5172 | if (flags & SO_TOTAL) |
bb192ed9 | 5173 | x = slab->objects; |
ec3ab083 | 5174 | else if (flags & SO_OBJECTS) |
bb192ed9 | 5175 | x = slab->inuse; |
ec3ab083 CL |
5176 | else |
5177 | x = 1; | |
49e22585 | 5178 | |
ec3ab083 CL |
5179 | total += x; |
5180 | nodes[node] += x; | |
5181 | ||
9c01e9af | 5182 | #ifdef CONFIG_SLUB_CPU_PARTIAL |
bb192ed9 VB |
5183 | slab = slub_percpu_partial_read_once(c); |
5184 | if (slab) { | |
5185 | node = slab_nid(slab); | |
8afb1474 LZ |
5186 | if (flags & SO_TOTAL) |
5187 | WARN_ON_ONCE(1); | |
5188 | else if (flags & SO_OBJECTS) | |
5189 | WARN_ON_ONCE(1); | |
5190 | else | |
bb192ed9 | 5191 | x = slab->slabs; |
bc6697d8 ED |
5192 | total += x; |
5193 | nodes[node] += x; | |
49e22585 | 5194 | } |
9c01e9af | 5195 | #endif |
81819f0f CL |
5196 | } |
5197 | } | |
5198 | ||
e4f8e513 QC |
5199 | /* |
5200 | * It is impossible to take "mem_hotplug_lock" here with "kernfs_mutex" | |
5201 | * already held which will conflict with an existing lock order: | |
5202 | * | |
5203 | * mem_hotplug_lock->slab_mutex->kernfs_mutex | |
5204 | * | |
5205 | * We don't really need mem_hotplug_lock (to hold off | |
5206 | * slab_mem_going_offline_callback) here because slab's memory hot | |
5207 | * unplug code doesn't destroy the kmem_cache->node[] data. | |
5208 | */ | |
5209 | ||
ab4d5ed5 | 5210 | #ifdef CONFIG_SLUB_DEBUG |
205ab99d | 5211 | if (flags & SO_ALL) { |
fa45dc25 CL |
5212 | struct kmem_cache_node *n; |
5213 | ||
5214 | for_each_kmem_cache_node(s, node, n) { | |
205ab99d | 5215 | |
d0e0ac97 CG |
5216 | if (flags & SO_TOTAL) |
5217 | x = atomic_long_read(&n->total_objects); | |
5218 | else if (flags & SO_OBJECTS) | |
5219 | x = atomic_long_read(&n->total_objects) - | |
5220 | count_partial(n, count_free); | |
81819f0f | 5221 | else |
205ab99d | 5222 | x = atomic_long_read(&n->nr_slabs); |
81819f0f CL |
5223 | total += x; |
5224 | nodes[node] += x; | |
5225 | } | |
5226 | ||
ab4d5ed5 CL |
5227 | } else |
5228 | #endif | |
5229 | if (flags & SO_PARTIAL) { | |
fa45dc25 | 5230 | struct kmem_cache_node *n; |
81819f0f | 5231 | |
fa45dc25 | 5232 | for_each_kmem_cache_node(s, node, n) { |
205ab99d CL |
5233 | if (flags & SO_TOTAL) |
5234 | x = count_partial(n, count_total); | |
5235 | else if (flags & SO_OBJECTS) | |
5236 | x = count_partial(n, count_inuse); | |
81819f0f | 5237 | else |
205ab99d | 5238 | x = n->nr_partial; |
81819f0f CL |
5239 | total += x; |
5240 | nodes[node] += x; | |
5241 | } | |
5242 | } | |
bf16d19a JP |
5243 | |
5244 | len += sysfs_emit_at(buf, len, "%lu", total); | |
81819f0f | 5245 | #ifdef CONFIG_NUMA |
bf16d19a | 5246 | for (node = 0; node < nr_node_ids; node++) { |
81819f0f | 5247 | if (nodes[node]) |
bf16d19a JP |
5248 | len += sysfs_emit_at(buf, len, " N%d=%lu", |
5249 | node, nodes[node]); | |
5250 | } | |
81819f0f | 5251 | #endif |
bf16d19a | 5252 | len += sysfs_emit_at(buf, len, "\n"); |
81819f0f | 5253 | kfree(nodes); |
bf16d19a JP |
5254 | |
5255 | return len; | |
81819f0f CL |
5256 | } |
5257 | ||
81819f0f | 5258 | #define to_slab_attr(n) container_of(n, struct slab_attribute, attr) |
497888cf | 5259 | #define to_slab(n) container_of(n, struct kmem_cache, kobj) |
81819f0f CL |
5260 | |
5261 | struct slab_attribute { | |
5262 | struct attribute attr; | |
5263 | ssize_t (*show)(struct kmem_cache *s, char *buf); | |
5264 | ssize_t (*store)(struct kmem_cache *s, const char *x, size_t count); | |
5265 | }; | |
5266 | ||
5267 | #define SLAB_ATTR_RO(_name) \ | |
d1d28bd9 | 5268 | static struct slab_attribute _name##_attr = __ATTR_RO_MODE(_name, 0400) |
81819f0f CL |
5269 | |
5270 | #define SLAB_ATTR(_name) \ | |
d1d28bd9 | 5271 | static struct slab_attribute _name##_attr = __ATTR_RW_MODE(_name, 0600) |
81819f0f | 5272 | |
81819f0f CL |
5273 | static ssize_t slab_size_show(struct kmem_cache *s, char *buf) |
5274 | { | |
bf16d19a | 5275 | return sysfs_emit(buf, "%u\n", s->size); |
81819f0f CL |
5276 | } |
5277 | SLAB_ATTR_RO(slab_size); | |
5278 | ||
5279 | static ssize_t align_show(struct kmem_cache *s, char *buf) | |
5280 | { | |
bf16d19a | 5281 | return sysfs_emit(buf, "%u\n", s->align); |
81819f0f CL |
5282 | } |
5283 | SLAB_ATTR_RO(align); | |
5284 | ||
5285 | static ssize_t object_size_show(struct kmem_cache *s, char *buf) | |
5286 | { | |
bf16d19a | 5287 | return sysfs_emit(buf, "%u\n", s->object_size); |
81819f0f CL |
5288 | } |
5289 | SLAB_ATTR_RO(object_size); | |
5290 | ||
5291 | static ssize_t objs_per_slab_show(struct kmem_cache *s, char *buf) | |
5292 | { | |
bf16d19a | 5293 | return sysfs_emit(buf, "%u\n", oo_objects(s->oo)); |
81819f0f CL |
5294 | } |
5295 | SLAB_ATTR_RO(objs_per_slab); | |
5296 | ||
5297 | static ssize_t order_show(struct kmem_cache *s, char *buf) | |
5298 | { | |
bf16d19a | 5299 | return sysfs_emit(buf, "%u\n", oo_order(s->oo)); |
81819f0f | 5300 | } |
32a6f409 | 5301 | SLAB_ATTR_RO(order); |
81819f0f | 5302 | |
73d342b1 DR |
5303 | static ssize_t min_partial_show(struct kmem_cache *s, char *buf) |
5304 | { | |
bf16d19a | 5305 | return sysfs_emit(buf, "%lu\n", s->min_partial); |
73d342b1 DR |
5306 | } |
5307 | ||
5308 | static ssize_t min_partial_store(struct kmem_cache *s, const char *buf, | |
5309 | size_t length) | |
5310 | { | |
5311 | unsigned long min; | |
5312 | int err; | |
5313 | ||
3dbb95f7 | 5314 | err = kstrtoul(buf, 10, &min); |
73d342b1 DR |
5315 | if (err) |
5316 | return err; | |
5317 | ||
5182f3c9 | 5318 | s->min_partial = min; |
73d342b1 DR |
5319 | return length; |
5320 | } | |
5321 | SLAB_ATTR(min_partial); | |
5322 | ||
49e22585 CL |
5323 | static ssize_t cpu_partial_show(struct kmem_cache *s, char *buf) |
5324 | { | |
b47291ef VB |
5325 | unsigned int nr_partial = 0; |
5326 | #ifdef CONFIG_SLUB_CPU_PARTIAL | |
5327 | nr_partial = s->cpu_partial; | |
5328 | #endif | |
5329 | ||
5330 | return sysfs_emit(buf, "%u\n", nr_partial); | |
49e22585 CL |
5331 | } |
5332 | ||
5333 | static ssize_t cpu_partial_store(struct kmem_cache *s, const char *buf, | |
5334 | size_t length) | |
5335 | { | |
e5d9998f | 5336 | unsigned int objects; |
49e22585 CL |
5337 | int err; |
5338 | ||
e5d9998f | 5339 | err = kstrtouint(buf, 10, &objects); |
49e22585 CL |
5340 | if (err) |
5341 | return err; | |
345c905d | 5342 | if (objects && !kmem_cache_has_cpu_partial(s)) |
74ee4ef1 | 5343 | return -EINVAL; |
49e22585 | 5344 | |
e6d0e1dc | 5345 | slub_set_cpu_partial(s, objects); |
49e22585 CL |
5346 | flush_all(s); |
5347 | return length; | |
5348 | } | |
5349 | SLAB_ATTR(cpu_partial); | |
5350 | ||
81819f0f CL |
5351 | static ssize_t ctor_show(struct kmem_cache *s, char *buf) |
5352 | { | |
62c70bce JP |
5353 | if (!s->ctor) |
5354 | return 0; | |
bf16d19a | 5355 | return sysfs_emit(buf, "%pS\n", s->ctor); |
81819f0f CL |
5356 | } |
5357 | SLAB_ATTR_RO(ctor); | |
5358 | ||
81819f0f CL |
5359 | static ssize_t aliases_show(struct kmem_cache *s, char *buf) |
5360 | { | |
bf16d19a | 5361 | return sysfs_emit(buf, "%d\n", s->refcount < 0 ? 0 : s->refcount - 1); |
81819f0f CL |
5362 | } |
5363 | SLAB_ATTR_RO(aliases); | |
5364 | ||
81819f0f CL |
5365 | static ssize_t partial_show(struct kmem_cache *s, char *buf) |
5366 | { | |
d9acf4b7 | 5367 | return show_slab_objects(s, buf, SO_PARTIAL); |
81819f0f CL |
5368 | } |
5369 | SLAB_ATTR_RO(partial); | |
5370 | ||
5371 | static ssize_t cpu_slabs_show(struct kmem_cache *s, char *buf) | |
5372 | { | |
d9acf4b7 | 5373 | return show_slab_objects(s, buf, SO_CPU); |
81819f0f CL |
5374 | } |
5375 | SLAB_ATTR_RO(cpu_slabs); | |
5376 | ||
5377 | static ssize_t objects_show(struct kmem_cache *s, char *buf) | |
5378 | { | |
205ab99d | 5379 | return show_slab_objects(s, buf, SO_ALL|SO_OBJECTS); |
81819f0f CL |
5380 | } |
5381 | SLAB_ATTR_RO(objects); | |
5382 | ||
205ab99d CL |
5383 | static ssize_t objects_partial_show(struct kmem_cache *s, char *buf) |
5384 | { | |
5385 | return show_slab_objects(s, buf, SO_PARTIAL|SO_OBJECTS); | |
5386 | } | |
5387 | SLAB_ATTR_RO(objects_partial); | |
5388 | ||
49e22585 CL |
5389 | static ssize_t slabs_cpu_partial_show(struct kmem_cache *s, char *buf) |
5390 | { | |
5391 | int objects = 0; | |
bb192ed9 | 5392 | int slabs = 0; |
9c01e9af | 5393 | int cpu __maybe_unused; |
bf16d19a | 5394 | int len = 0; |
49e22585 | 5395 | |
9c01e9af | 5396 | #ifdef CONFIG_SLUB_CPU_PARTIAL |
49e22585 | 5397 | for_each_online_cpu(cpu) { |
bb192ed9 | 5398 | struct slab *slab; |
a93cf07b | 5399 | |
bb192ed9 | 5400 | slab = slub_percpu_partial(per_cpu_ptr(s->cpu_slab, cpu)); |
49e22585 | 5401 | |
bb192ed9 VB |
5402 | if (slab) |
5403 | slabs += slab->slabs; | |
49e22585 | 5404 | } |
9c01e9af | 5405 | #endif |
49e22585 | 5406 | |
c2092c12 | 5407 | /* Approximate half-full slabs, see slub_set_cpu_partial() */ |
bb192ed9 VB |
5408 | objects = (slabs * oo_objects(s->oo)) / 2; |
5409 | len += sysfs_emit_at(buf, len, "%d(%d)", objects, slabs); | |
49e22585 | 5410 | |
9c01e9af | 5411 | #if defined(CONFIG_SLUB_CPU_PARTIAL) && defined(CONFIG_SMP) |
49e22585 | 5412 | for_each_online_cpu(cpu) { |
bb192ed9 | 5413 | struct slab *slab; |
a93cf07b | 5414 | |
bb192ed9 VB |
5415 | slab = slub_percpu_partial(per_cpu_ptr(s->cpu_slab, cpu)); |
5416 | if (slab) { | |
5417 | slabs = READ_ONCE(slab->slabs); | |
5418 | objects = (slabs * oo_objects(s->oo)) / 2; | |
bf16d19a | 5419 | len += sysfs_emit_at(buf, len, " C%d=%d(%d)", |
bb192ed9 | 5420 | cpu, objects, slabs); |
b47291ef | 5421 | } |
49e22585 CL |
5422 | } |
5423 | #endif | |
bf16d19a JP |
5424 | len += sysfs_emit_at(buf, len, "\n"); |
5425 | ||
5426 | return len; | |
49e22585 CL |
5427 | } |
5428 | SLAB_ATTR_RO(slabs_cpu_partial); | |
5429 | ||
a5a84755 CL |
5430 | static ssize_t reclaim_account_show(struct kmem_cache *s, char *buf) |
5431 | { | |
bf16d19a | 5432 | return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_RECLAIM_ACCOUNT)); |
a5a84755 | 5433 | } |
8f58119a | 5434 | SLAB_ATTR_RO(reclaim_account); |
a5a84755 CL |
5435 | |
5436 | static ssize_t hwcache_align_show(struct kmem_cache *s, char *buf) | |
5437 | { | |
bf16d19a | 5438 | return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_HWCACHE_ALIGN)); |
a5a84755 CL |
5439 | } |
5440 | SLAB_ATTR_RO(hwcache_align); | |
5441 | ||
5442 | #ifdef CONFIG_ZONE_DMA | |
5443 | static ssize_t cache_dma_show(struct kmem_cache *s, char *buf) | |
5444 | { | |
bf16d19a | 5445 | return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_CACHE_DMA)); |
a5a84755 CL |
5446 | } |
5447 | SLAB_ATTR_RO(cache_dma); | |
5448 | #endif | |
5449 | ||
8eb8284b DW |
5450 | static ssize_t usersize_show(struct kmem_cache *s, char *buf) |
5451 | { | |
bf16d19a | 5452 | return sysfs_emit(buf, "%u\n", s->usersize); |
8eb8284b DW |
5453 | } |
5454 | SLAB_ATTR_RO(usersize); | |
5455 | ||
a5a84755 CL |
5456 | static ssize_t destroy_by_rcu_show(struct kmem_cache *s, char *buf) |
5457 | { | |
bf16d19a | 5458 | return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_TYPESAFE_BY_RCU)); |
a5a84755 CL |
5459 | } |
5460 | SLAB_ATTR_RO(destroy_by_rcu); | |
5461 | ||
ab4d5ed5 | 5462 | #ifdef CONFIG_SLUB_DEBUG |
a5a84755 CL |
5463 | static ssize_t slabs_show(struct kmem_cache *s, char *buf) |
5464 | { | |
5465 | return show_slab_objects(s, buf, SO_ALL); | |
5466 | } | |
5467 | SLAB_ATTR_RO(slabs); | |
5468 | ||
205ab99d CL |
5469 | static ssize_t total_objects_show(struct kmem_cache *s, char *buf) |
5470 | { | |
5471 | return show_slab_objects(s, buf, SO_ALL|SO_TOTAL); | |
5472 | } | |
5473 | SLAB_ATTR_RO(total_objects); | |
5474 | ||
81819f0f CL |
5475 | static ssize_t sanity_checks_show(struct kmem_cache *s, char *buf) |
5476 | { | |
bf16d19a | 5477 | return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_CONSISTENCY_CHECKS)); |
81819f0f | 5478 | } |
060807f8 | 5479 | SLAB_ATTR_RO(sanity_checks); |
81819f0f CL |
5480 | |
5481 | static ssize_t trace_show(struct kmem_cache *s, char *buf) | |
5482 | { | |
bf16d19a | 5483 | return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_TRACE)); |
81819f0f | 5484 | } |
060807f8 | 5485 | SLAB_ATTR_RO(trace); |
81819f0f | 5486 | |
81819f0f CL |
5487 | static ssize_t red_zone_show(struct kmem_cache *s, char *buf) |
5488 | { | |
bf16d19a | 5489 | return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_RED_ZONE)); |
81819f0f CL |
5490 | } |
5491 | ||
ad38b5b1 | 5492 | SLAB_ATTR_RO(red_zone); |
81819f0f CL |
5493 | |
5494 | static ssize_t poison_show(struct kmem_cache *s, char *buf) | |
5495 | { | |
bf16d19a | 5496 | return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_POISON)); |
81819f0f CL |
5497 | } |
5498 | ||
ad38b5b1 | 5499 | SLAB_ATTR_RO(poison); |
81819f0f CL |
5500 | |
5501 | static ssize_t store_user_show(struct kmem_cache *s, char *buf) | |
5502 | { | |
bf16d19a | 5503 | return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_STORE_USER)); |
81819f0f CL |
5504 | } |
5505 | ||
ad38b5b1 | 5506 | SLAB_ATTR_RO(store_user); |
81819f0f | 5507 | |
53e15af0 CL |
5508 | static ssize_t validate_show(struct kmem_cache *s, char *buf) |
5509 | { | |
5510 | return 0; | |
5511 | } | |
5512 | ||
5513 | static ssize_t validate_store(struct kmem_cache *s, | |
5514 | const char *buf, size_t length) | |
5515 | { | |
434e245d CL |
5516 | int ret = -EINVAL; |
5517 | ||
5518 | if (buf[0] == '1') { | |
5519 | ret = validate_slab_cache(s); | |
5520 | if (ret >= 0) | |
5521 | ret = length; | |
5522 | } | |
5523 | return ret; | |
53e15af0 CL |
5524 | } |
5525 | SLAB_ATTR(validate); | |
a5a84755 | 5526 | |
a5a84755 CL |
5527 | #endif /* CONFIG_SLUB_DEBUG */ |
5528 | ||
5529 | #ifdef CONFIG_FAILSLAB | |
5530 | static ssize_t failslab_show(struct kmem_cache *s, char *buf) | |
5531 | { | |
bf16d19a | 5532 | return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_FAILSLAB)); |
a5a84755 | 5533 | } |
060807f8 | 5534 | SLAB_ATTR_RO(failslab); |
ab4d5ed5 | 5535 | #endif |
53e15af0 | 5536 | |
2086d26a CL |
5537 | static ssize_t shrink_show(struct kmem_cache *s, char *buf) |
5538 | { | |
5539 | return 0; | |
5540 | } | |
5541 | ||
5542 | static ssize_t shrink_store(struct kmem_cache *s, | |
5543 | const char *buf, size_t length) | |
5544 | { | |
832f37f5 | 5545 | if (buf[0] == '1') |
10befea9 | 5546 | kmem_cache_shrink(s); |
832f37f5 | 5547 | else |
2086d26a CL |
5548 | return -EINVAL; |
5549 | return length; | |
5550 | } | |
5551 | SLAB_ATTR(shrink); | |
5552 | ||
81819f0f | 5553 | #ifdef CONFIG_NUMA |
9824601e | 5554 | static ssize_t remote_node_defrag_ratio_show(struct kmem_cache *s, char *buf) |
81819f0f | 5555 | { |
bf16d19a | 5556 | return sysfs_emit(buf, "%u\n", s->remote_node_defrag_ratio / 10); |
81819f0f CL |
5557 | } |
5558 | ||
9824601e | 5559 | static ssize_t remote_node_defrag_ratio_store(struct kmem_cache *s, |
81819f0f CL |
5560 | const char *buf, size_t length) |
5561 | { | |
eb7235eb | 5562 | unsigned int ratio; |
0121c619 CL |
5563 | int err; |
5564 | ||
eb7235eb | 5565 | err = kstrtouint(buf, 10, &ratio); |
0121c619 CL |
5566 | if (err) |
5567 | return err; | |
eb7235eb AD |
5568 | if (ratio > 100) |
5569 | return -ERANGE; | |
0121c619 | 5570 | |
eb7235eb | 5571 | s->remote_node_defrag_ratio = ratio * 10; |
81819f0f | 5572 | |
81819f0f CL |
5573 | return length; |
5574 | } | |
9824601e | 5575 | SLAB_ATTR(remote_node_defrag_ratio); |
81819f0f CL |
5576 | #endif |
5577 | ||
8ff12cfc | 5578 | #ifdef CONFIG_SLUB_STATS |
8ff12cfc CL |
5579 | static int show_stat(struct kmem_cache *s, char *buf, enum stat_item si) |
5580 | { | |
5581 | unsigned long sum = 0; | |
5582 | int cpu; | |
bf16d19a | 5583 | int len = 0; |
6da2ec56 | 5584 | int *data = kmalloc_array(nr_cpu_ids, sizeof(int), GFP_KERNEL); |
8ff12cfc CL |
5585 | |
5586 | if (!data) | |
5587 | return -ENOMEM; | |
5588 | ||
5589 | for_each_online_cpu(cpu) { | |
9dfc6e68 | 5590 | unsigned x = per_cpu_ptr(s->cpu_slab, cpu)->stat[si]; |
8ff12cfc CL |
5591 | |
5592 | data[cpu] = x; | |
5593 | sum += x; | |
5594 | } | |
5595 | ||
bf16d19a | 5596 | len += sysfs_emit_at(buf, len, "%lu", sum); |
8ff12cfc | 5597 | |
50ef37b9 | 5598 | #ifdef CONFIG_SMP |
8ff12cfc | 5599 | for_each_online_cpu(cpu) { |
bf16d19a JP |
5600 | if (data[cpu]) |
5601 | len += sysfs_emit_at(buf, len, " C%d=%u", | |
5602 | cpu, data[cpu]); | |
8ff12cfc | 5603 | } |
50ef37b9 | 5604 | #endif |
8ff12cfc | 5605 | kfree(data); |
bf16d19a JP |
5606 | len += sysfs_emit_at(buf, len, "\n"); |
5607 | ||
5608 | return len; | |
8ff12cfc CL |
5609 | } |
5610 | ||
78eb00cc DR |
5611 | static void clear_stat(struct kmem_cache *s, enum stat_item si) |
5612 | { | |
5613 | int cpu; | |
5614 | ||
5615 | for_each_online_cpu(cpu) | |
9dfc6e68 | 5616 | per_cpu_ptr(s->cpu_slab, cpu)->stat[si] = 0; |
78eb00cc DR |
5617 | } |
5618 | ||
8ff12cfc CL |
5619 | #define STAT_ATTR(si, text) \ |
5620 | static ssize_t text##_show(struct kmem_cache *s, char *buf) \ | |
5621 | { \ | |
5622 | return show_stat(s, buf, si); \ | |
5623 | } \ | |
78eb00cc DR |
5624 | static ssize_t text##_store(struct kmem_cache *s, \ |
5625 | const char *buf, size_t length) \ | |
5626 | { \ | |
5627 | if (buf[0] != '0') \ | |
5628 | return -EINVAL; \ | |
5629 | clear_stat(s, si); \ | |
5630 | return length; \ | |
5631 | } \ | |
5632 | SLAB_ATTR(text); \ | |
8ff12cfc CL |
5633 | |
5634 | STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath); | |
5635 | STAT_ATTR(ALLOC_SLOWPATH, alloc_slowpath); | |
5636 | STAT_ATTR(FREE_FASTPATH, free_fastpath); | |
5637 | STAT_ATTR(FREE_SLOWPATH, free_slowpath); | |
5638 | STAT_ATTR(FREE_FROZEN, free_frozen); | |
5639 | STAT_ATTR(FREE_ADD_PARTIAL, free_add_partial); | |
5640 | STAT_ATTR(FREE_REMOVE_PARTIAL, free_remove_partial); | |
5641 | STAT_ATTR(ALLOC_FROM_PARTIAL, alloc_from_partial); | |
5642 | STAT_ATTR(ALLOC_SLAB, alloc_slab); | |
5643 | STAT_ATTR(ALLOC_REFILL, alloc_refill); | |
e36a2652 | 5644 | STAT_ATTR(ALLOC_NODE_MISMATCH, alloc_node_mismatch); |
8ff12cfc CL |
5645 | STAT_ATTR(FREE_SLAB, free_slab); |
5646 | STAT_ATTR(CPUSLAB_FLUSH, cpuslab_flush); | |
5647 | STAT_ATTR(DEACTIVATE_FULL, deactivate_full); | |
5648 | STAT_ATTR(DEACTIVATE_EMPTY, deactivate_empty); | |
5649 | STAT_ATTR(DEACTIVATE_TO_HEAD, deactivate_to_head); | |
5650 | STAT_ATTR(DEACTIVATE_TO_TAIL, deactivate_to_tail); | |
5651 | STAT_ATTR(DEACTIVATE_REMOTE_FREES, deactivate_remote_frees); | |
03e404af | 5652 | STAT_ATTR(DEACTIVATE_BYPASS, deactivate_bypass); |
65c3376a | 5653 | STAT_ATTR(ORDER_FALLBACK, order_fallback); |
b789ef51 CL |
5654 | STAT_ATTR(CMPXCHG_DOUBLE_CPU_FAIL, cmpxchg_double_cpu_fail); |
5655 | STAT_ATTR(CMPXCHG_DOUBLE_FAIL, cmpxchg_double_fail); | |
49e22585 CL |
5656 | STAT_ATTR(CPU_PARTIAL_ALLOC, cpu_partial_alloc); |
5657 | STAT_ATTR(CPU_PARTIAL_FREE, cpu_partial_free); | |
8028dcea AS |
5658 | STAT_ATTR(CPU_PARTIAL_NODE, cpu_partial_node); |
5659 | STAT_ATTR(CPU_PARTIAL_DRAIN, cpu_partial_drain); | |
6dfd1b65 | 5660 | #endif /* CONFIG_SLUB_STATS */ |
8ff12cfc | 5661 | |
06428780 | 5662 | static struct attribute *slab_attrs[] = { |
81819f0f CL |
5663 | &slab_size_attr.attr, |
5664 | &object_size_attr.attr, | |
5665 | &objs_per_slab_attr.attr, | |
5666 | &order_attr.attr, | |
73d342b1 | 5667 | &min_partial_attr.attr, |
49e22585 | 5668 | &cpu_partial_attr.attr, |
81819f0f | 5669 | &objects_attr.attr, |
205ab99d | 5670 | &objects_partial_attr.attr, |
81819f0f CL |
5671 | &partial_attr.attr, |
5672 | &cpu_slabs_attr.attr, | |
5673 | &ctor_attr.attr, | |
81819f0f CL |
5674 | &aliases_attr.attr, |
5675 | &align_attr.attr, | |
81819f0f CL |
5676 | &hwcache_align_attr.attr, |
5677 | &reclaim_account_attr.attr, | |
5678 | &destroy_by_rcu_attr.attr, | |
a5a84755 | 5679 | &shrink_attr.attr, |
49e22585 | 5680 | &slabs_cpu_partial_attr.attr, |
ab4d5ed5 | 5681 | #ifdef CONFIG_SLUB_DEBUG |
a5a84755 CL |
5682 | &total_objects_attr.attr, |
5683 | &slabs_attr.attr, | |
5684 | &sanity_checks_attr.attr, | |
5685 | &trace_attr.attr, | |
81819f0f CL |
5686 | &red_zone_attr.attr, |
5687 | &poison_attr.attr, | |
5688 | &store_user_attr.attr, | |
53e15af0 | 5689 | &validate_attr.attr, |
ab4d5ed5 | 5690 | #endif |
81819f0f CL |
5691 | #ifdef CONFIG_ZONE_DMA |
5692 | &cache_dma_attr.attr, | |
5693 | #endif | |
5694 | #ifdef CONFIG_NUMA | |
9824601e | 5695 | &remote_node_defrag_ratio_attr.attr, |
8ff12cfc CL |
5696 | #endif |
5697 | #ifdef CONFIG_SLUB_STATS | |
5698 | &alloc_fastpath_attr.attr, | |
5699 | &alloc_slowpath_attr.attr, | |
5700 | &free_fastpath_attr.attr, | |
5701 | &free_slowpath_attr.attr, | |
5702 | &free_frozen_attr.attr, | |
5703 | &free_add_partial_attr.attr, | |
5704 | &free_remove_partial_attr.attr, | |
5705 | &alloc_from_partial_attr.attr, | |
5706 | &alloc_slab_attr.attr, | |
5707 | &alloc_refill_attr.attr, | |
e36a2652 | 5708 | &alloc_node_mismatch_attr.attr, |
8ff12cfc CL |
5709 | &free_slab_attr.attr, |
5710 | &cpuslab_flush_attr.attr, | |
5711 | &deactivate_full_attr.attr, | |
5712 | &deactivate_empty_attr.attr, | |
5713 | &deactivate_to_head_attr.attr, | |
5714 | &deactivate_to_tail_attr.attr, | |
5715 | &deactivate_remote_frees_attr.attr, | |
03e404af | 5716 | &deactivate_bypass_attr.attr, |
65c3376a | 5717 | &order_fallback_attr.attr, |
b789ef51 CL |
5718 | &cmpxchg_double_fail_attr.attr, |
5719 | &cmpxchg_double_cpu_fail_attr.attr, | |
49e22585 CL |
5720 | &cpu_partial_alloc_attr.attr, |
5721 | &cpu_partial_free_attr.attr, | |
8028dcea AS |
5722 | &cpu_partial_node_attr.attr, |
5723 | &cpu_partial_drain_attr.attr, | |
81819f0f | 5724 | #endif |
4c13dd3b DM |
5725 | #ifdef CONFIG_FAILSLAB |
5726 | &failslab_attr.attr, | |
5727 | #endif | |
8eb8284b | 5728 | &usersize_attr.attr, |
4c13dd3b | 5729 | |
81819f0f CL |
5730 | NULL |
5731 | }; | |
5732 | ||
1fdaaa23 | 5733 | static const struct attribute_group slab_attr_group = { |
81819f0f CL |
5734 | .attrs = slab_attrs, |
5735 | }; | |
5736 | ||
5737 | static ssize_t slab_attr_show(struct kobject *kobj, | |
5738 | struct attribute *attr, | |
5739 | char *buf) | |
5740 | { | |
5741 | struct slab_attribute *attribute; | |
5742 | struct kmem_cache *s; | |
5743 | int err; | |
5744 | ||
5745 | attribute = to_slab_attr(attr); | |
5746 | s = to_slab(kobj); | |
5747 | ||
5748 | if (!attribute->show) | |
5749 | return -EIO; | |
5750 | ||
5751 | err = attribute->show(s, buf); | |
5752 | ||
5753 | return err; | |
5754 | } | |
5755 | ||
5756 | static ssize_t slab_attr_store(struct kobject *kobj, | |
5757 | struct attribute *attr, | |
5758 | const char *buf, size_t len) | |
5759 | { | |
5760 | struct slab_attribute *attribute; | |
5761 | struct kmem_cache *s; | |
5762 | int err; | |
5763 | ||
5764 | attribute = to_slab_attr(attr); | |
5765 | s = to_slab(kobj); | |
5766 | ||
5767 | if (!attribute->store) | |
5768 | return -EIO; | |
5769 | ||
5770 | err = attribute->store(s, buf, len); | |
81819f0f CL |
5771 | return err; |
5772 | } | |
5773 | ||
41a21285 CL |
5774 | static void kmem_cache_release(struct kobject *k) |
5775 | { | |
5776 | slab_kmem_cache_release(to_slab(k)); | |
5777 | } | |
5778 | ||
52cf25d0 | 5779 | static const struct sysfs_ops slab_sysfs_ops = { |
81819f0f CL |
5780 | .show = slab_attr_show, |
5781 | .store = slab_attr_store, | |
5782 | }; | |
5783 | ||
5784 | static struct kobj_type slab_ktype = { | |
5785 | .sysfs_ops = &slab_sysfs_ops, | |
41a21285 | 5786 | .release = kmem_cache_release, |
81819f0f CL |
5787 | }; |
5788 | ||
27c3a314 | 5789 | static struct kset *slab_kset; |
81819f0f | 5790 | |
9a41707b VD |
5791 | static inline struct kset *cache_kset(struct kmem_cache *s) |
5792 | { | |
9a41707b VD |
5793 | return slab_kset; |
5794 | } | |
5795 | ||
81819f0f CL |
5796 | #define ID_STR_LENGTH 64 |
5797 | ||
5798 | /* Create a unique string id for a slab cache: | |
6446faa2 CL |
5799 | * |
5800 | * Format :[flags-]size | |
81819f0f CL |
5801 | */ |
5802 | static char *create_unique_id(struct kmem_cache *s) | |
5803 | { | |
5804 | char *name = kmalloc(ID_STR_LENGTH, GFP_KERNEL); | |
5805 | char *p = name; | |
5806 | ||
5807 | BUG_ON(!name); | |
5808 | ||
5809 | *p++ = ':'; | |
5810 | /* | |
5811 | * First flags affecting slabcache operations. We will only | |
5812 | * get here for aliasable slabs so we do not need to support | |
5813 | * too many flags. The flags here must cover all flags that | |
5814 | * are matched during merging to guarantee that the id is | |
5815 | * unique. | |
5816 | */ | |
5817 | if (s->flags & SLAB_CACHE_DMA) | |
5818 | *p++ = 'd'; | |
6d6ea1e9 NB |
5819 | if (s->flags & SLAB_CACHE_DMA32) |
5820 | *p++ = 'D'; | |
81819f0f CL |
5821 | if (s->flags & SLAB_RECLAIM_ACCOUNT) |
5822 | *p++ = 'a'; | |
becfda68 | 5823 | if (s->flags & SLAB_CONSISTENCY_CHECKS) |
81819f0f | 5824 | *p++ = 'F'; |
230e9fc2 VD |
5825 | if (s->flags & SLAB_ACCOUNT) |
5826 | *p++ = 'A'; | |
81819f0f CL |
5827 | if (p != name + 1) |
5828 | *p++ = '-'; | |
44065b2e | 5829 | p += sprintf(p, "%07u", s->size); |
2633d7a0 | 5830 | |
81819f0f CL |
5831 | BUG_ON(p > name + ID_STR_LENGTH - 1); |
5832 | return name; | |
5833 | } | |
5834 | ||
5835 | static int sysfs_slab_add(struct kmem_cache *s) | |
5836 | { | |
5837 | int err; | |
5838 | const char *name; | |
1663f26d | 5839 | struct kset *kset = cache_kset(s); |
45530c44 | 5840 | int unmergeable = slab_unmergeable(s); |
81819f0f | 5841 | |
1663f26d TH |
5842 | if (!kset) { |
5843 | kobject_init(&s->kobj, &slab_ktype); | |
5844 | return 0; | |
5845 | } | |
5846 | ||
11066386 MC |
5847 | if (!unmergeable && disable_higher_order_debug && |
5848 | (slub_debug & DEBUG_METADATA_FLAGS)) | |
5849 | unmergeable = 1; | |
5850 | ||
81819f0f CL |
5851 | if (unmergeable) { |
5852 | /* | |
5853 | * Slabcache can never be merged so we can use the name proper. | |
5854 | * This is typically the case for debug situations. In that | |
5855 | * case we can catch duplicate names easily. | |
5856 | */ | |
27c3a314 | 5857 | sysfs_remove_link(&slab_kset->kobj, s->name); |
81819f0f CL |
5858 | name = s->name; |
5859 | } else { | |
5860 | /* | |
5861 | * Create a unique name for the slab as a target | |
5862 | * for the symlinks. | |
5863 | */ | |
5864 | name = create_unique_id(s); | |
5865 | } | |
5866 | ||
1663f26d | 5867 | s->kobj.kset = kset; |
26e4f205 | 5868 | err = kobject_init_and_add(&s->kobj, &slab_ktype, NULL, "%s", name); |
757fed1d | 5869 | if (err) |
80da026a | 5870 | goto out; |
81819f0f CL |
5871 | |
5872 | err = sysfs_create_group(&s->kobj, &slab_attr_group); | |
54b6a731 DJ |
5873 | if (err) |
5874 | goto out_del_kobj; | |
9a41707b | 5875 | |
81819f0f CL |
5876 | if (!unmergeable) { |
5877 | /* Setup first alias */ | |
5878 | sysfs_slab_alias(s, s->name); | |
81819f0f | 5879 | } |
54b6a731 DJ |
5880 | out: |
5881 | if (!unmergeable) | |
5882 | kfree(name); | |
5883 | return err; | |
5884 | out_del_kobj: | |
5885 | kobject_del(&s->kobj); | |
54b6a731 | 5886 | goto out; |
81819f0f CL |
5887 | } |
5888 | ||
d50d82fa MP |
5889 | void sysfs_slab_unlink(struct kmem_cache *s) |
5890 | { | |
5891 | if (slab_state >= FULL) | |
5892 | kobject_del(&s->kobj); | |
5893 | } | |
5894 | ||
bf5eb3de TH |
5895 | void sysfs_slab_release(struct kmem_cache *s) |
5896 | { | |
5897 | if (slab_state >= FULL) | |
5898 | kobject_put(&s->kobj); | |
81819f0f CL |
5899 | } |
5900 | ||
5901 | /* | |
5902 | * Need to buffer aliases during bootup until sysfs becomes | |
9f6c708e | 5903 | * available lest we lose that information. |
81819f0f CL |
5904 | */ |
5905 | struct saved_alias { | |
5906 | struct kmem_cache *s; | |
5907 | const char *name; | |
5908 | struct saved_alias *next; | |
5909 | }; | |
5910 | ||
5af328a5 | 5911 | static struct saved_alias *alias_list; |
81819f0f CL |
5912 | |
5913 | static int sysfs_slab_alias(struct kmem_cache *s, const char *name) | |
5914 | { | |
5915 | struct saved_alias *al; | |
5916 | ||
97d06609 | 5917 | if (slab_state == FULL) { |
81819f0f CL |
5918 | /* |
5919 | * If we have a leftover link then remove it. | |
5920 | */ | |
27c3a314 GKH |
5921 | sysfs_remove_link(&slab_kset->kobj, name); |
5922 | return sysfs_create_link(&slab_kset->kobj, &s->kobj, name); | |
81819f0f CL |
5923 | } |
5924 | ||
5925 | al = kmalloc(sizeof(struct saved_alias), GFP_KERNEL); | |
5926 | if (!al) | |
5927 | return -ENOMEM; | |
5928 | ||
5929 | al->s = s; | |
5930 | al->name = name; | |
5931 | al->next = alias_list; | |
5932 | alias_list = al; | |
5933 | return 0; | |
5934 | } | |
5935 | ||
5936 | static int __init slab_sysfs_init(void) | |
5937 | { | |
5b95a4ac | 5938 | struct kmem_cache *s; |
81819f0f CL |
5939 | int err; |
5940 | ||
18004c5d | 5941 | mutex_lock(&slab_mutex); |
2bce6485 | 5942 | |
d7660ce5 | 5943 | slab_kset = kset_create_and_add("slab", NULL, kernel_kobj); |
27c3a314 | 5944 | if (!slab_kset) { |
18004c5d | 5945 | mutex_unlock(&slab_mutex); |
f9f58285 | 5946 | pr_err("Cannot register slab subsystem.\n"); |
81819f0f CL |
5947 | return -ENOSYS; |
5948 | } | |
5949 | ||
97d06609 | 5950 | slab_state = FULL; |
26a7bd03 | 5951 | |
5b95a4ac | 5952 | list_for_each_entry(s, &slab_caches, list) { |
26a7bd03 | 5953 | err = sysfs_slab_add(s); |
5d540fb7 | 5954 | if (err) |
f9f58285 FF |
5955 | pr_err("SLUB: Unable to add boot slab %s to sysfs\n", |
5956 | s->name); | |
26a7bd03 | 5957 | } |
81819f0f CL |
5958 | |
5959 | while (alias_list) { | |
5960 | struct saved_alias *al = alias_list; | |
5961 | ||
5962 | alias_list = alias_list->next; | |
5963 | err = sysfs_slab_alias(al->s, al->name); | |
5d540fb7 | 5964 | if (err) |
f9f58285 FF |
5965 | pr_err("SLUB: Unable to add boot slab alias %s to sysfs\n", |
5966 | al->name); | |
81819f0f CL |
5967 | kfree(al); |
5968 | } | |
5969 | ||
18004c5d | 5970 | mutex_unlock(&slab_mutex); |
81819f0f CL |
5971 | return 0; |
5972 | } | |
5973 | ||
5974 | __initcall(slab_sysfs_init); | |
ab4d5ed5 | 5975 | #endif /* CONFIG_SYSFS */ |
57ed3eda | 5976 | |
64dd6849 FM |
5977 | #if defined(CONFIG_SLUB_DEBUG) && defined(CONFIG_DEBUG_FS) |
5978 | static int slab_debugfs_show(struct seq_file *seq, void *v) | |
5979 | { | |
64dd6849 | 5980 | struct loc_track *t = seq->private; |
005a79e5 GS |
5981 | struct location *l; |
5982 | unsigned long idx; | |
64dd6849 | 5983 | |
005a79e5 | 5984 | idx = (unsigned long) t->idx; |
64dd6849 FM |
5985 | if (idx < t->count) { |
5986 | l = &t->loc[idx]; | |
5987 | ||
5988 | seq_printf(seq, "%7ld ", l->count); | |
5989 | ||
5990 | if (l->addr) | |
5991 | seq_printf(seq, "%pS", (void *)l->addr); | |
5992 | else | |
5993 | seq_puts(seq, "<not-available>"); | |
5994 | ||
5995 | if (l->sum_time != l->min_time) { | |
5996 | seq_printf(seq, " age=%ld/%llu/%ld", | |
5997 | l->min_time, div_u64(l->sum_time, l->count), | |
5998 | l->max_time); | |
5999 | } else | |
6000 | seq_printf(seq, " age=%ld", l->min_time); | |
6001 | ||
6002 | if (l->min_pid != l->max_pid) | |
6003 | seq_printf(seq, " pid=%ld-%ld", l->min_pid, l->max_pid); | |
6004 | else | |
6005 | seq_printf(seq, " pid=%ld", | |
6006 | l->min_pid); | |
6007 | ||
6008 | if (num_online_cpus() > 1 && !cpumask_empty(to_cpumask(l->cpus))) | |
6009 | seq_printf(seq, " cpus=%*pbl", | |
6010 | cpumask_pr_args(to_cpumask(l->cpus))); | |
6011 | ||
6012 | if (nr_online_nodes > 1 && !nodes_empty(l->nodes)) | |
6013 | seq_printf(seq, " nodes=%*pbl", | |
6014 | nodemask_pr_args(&l->nodes)); | |
6015 | ||
8ea9fb92 OG |
6016 | #ifdef CONFIG_STACKDEPOT |
6017 | { | |
6018 | depot_stack_handle_t handle; | |
6019 | unsigned long *entries; | |
6020 | unsigned int nr_entries, j; | |
6021 | ||
6022 | handle = READ_ONCE(l->handle); | |
6023 | if (handle) { | |
6024 | nr_entries = stack_depot_fetch(handle, &entries); | |
6025 | seq_puts(seq, "\n"); | |
6026 | for (j = 0; j < nr_entries; j++) | |
6027 | seq_printf(seq, " %pS\n", (void *)entries[j]); | |
6028 | } | |
6029 | } | |
6030 | #endif | |
64dd6849 FM |
6031 | seq_puts(seq, "\n"); |
6032 | } | |
6033 | ||
6034 | if (!idx && !t->count) | |
6035 | seq_puts(seq, "No data\n"); | |
6036 | ||
6037 | return 0; | |
6038 | } | |
6039 | ||
6040 | static void slab_debugfs_stop(struct seq_file *seq, void *v) | |
6041 | { | |
6042 | } | |
6043 | ||
6044 | static void *slab_debugfs_next(struct seq_file *seq, void *v, loff_t *ppos) | |
6045 | { | |
6046 | struct loc_track *t = seq->private; | |
6047 | ||
005a79e5 | 6048 | t->idx = ++(*ppos); |
64dd6849 | 6049 | if (*ppos <= t->count) |
005a79e5 | 6050 | return ppos; |
64dd6849 FM |
6051 | |
6052 | return NULL; | |
6053 | } | |
6054 | ||
553c0369 OG |
6055 | static int cmp_loc_by_count(const void *a, const void *b, const void *data) |
6056 | { | |
6057 | struct location *loc1 = (struct location *)a; | |
6058 | struct location *loc2 = (struct location *)b; | |
6059 | ||
6060 | if (loc1->count > loc2->count) | |
6061 | return -1; | |
6062 | else | |
6063 | return 1; | |
6064 | } | |
6065 | ||
64dd6849 FM |
6066 | static void *slab_debugfs_start(struct seq_file *seq, loff_t *ppos) |
6067 | { | |
005a79e5 GS |
6068 | struct loc_track *t = seq->private; |
6069 | ||
6070 | t->idx = *ppos; | |
64dd6849 FM |
6071 | return ppos; |
6072 | } | |
6073 | ||
6074 | static const struct seq_operations slab_debugfs_sops = { | |
6075 | .start = slab_debugfs_start, | |
6076 | .next = slab_debugfs_next, | |
6077 | .stop = slab_debugfs_stop, | |
6078 | .show = slab_debugfs_show, | |
6079 | }; | |
6080 | ||
6081 | static int slab_debug_trace_open(struct inode *inode, struct file *filep) | |
6082 | { | |
6083 | ||
6084 | struct kmem_cache_node *n; | |
6085 | enum track_item alloc; | |
6086 | int node; | |
6087 | struct loc_track *t = __seq_open_private(filep, &slab_debugfs_sops, | |
6088 | sizeof(struct loc_track)); | |
6089 | struct kmem_cache *s = file_inode(filep)->i_private; | |
b3fd64e1 VB |
6090 | unsigned long *obj_map; |
6091 | ||
2127d225 ML |
6092 | if (!t) |
6093 | return -ENOMEM; | |
6094 | ||
b3fd64e1 | 6095 | obj_map = bitmap_alloc(oo_objects(s->oo), GFP_KERNEL); |
2127d225 ML |
6096 | if (!obj_map) { |
6097 | seq_release_private(inode, filep); | |
b3fd64e1 | 6098 | return -ENOMEM; |
2127d225 | 6099 | } |
64dd6849 FM |
6100 | |
6101 | if (strcmp(filep->f_path.dentry->d_name.name, "alloc_traces") == 0) | |
6102 | alloc = TRACK_ALLOC; | |
6103 | else | |
6104 | alloc = TRACK_FREE; | |
6105 | ||
b3fd64e1 VB |
6106 | if (!alloc_loc_track(t, PAGE_SIZE / sizeof(struct location), GFP_KERNEL)) { |
6107 | bitmap_free(obj_map); | |
2127d225 | 6108 | seq_release_private(inode, filep); |
64dd6849 | 6109 | return -ENOMEM; |
b3fd64e1 | 6110 | } |
64dd6849 | 6111 | |
64dd6849 FM |
6112 | for_each_kmem_cache_node(s, node, n) { |
6113 | unsigned long flags; | |
bb192ed9 | 6114 | struct slab *slab; |
64dd6849 FM |
6115 | |
6116 | if (!atomic_long_read(&n->nr_slabs)) | |
6117 | continue; | |
6118 | ||
6119 | spin_lock_irqsave(&n->list_lock, flags); | |
bb192ed9 VB |
6120 | list_for_each_entry(slab, &n->partial, slab_list) |
6121 | process_slab(t, s, slab, alloc, obj_map); | |
6122 | list_for_each_entry(slab, &n->full, slab_list) | |
6123 | process_slab(t, s, slab, alloc, obj_map); | |
64dd6849 FM |
6124 | spin_unlock_irqrestore(&n->list_lock, flags); |
6125 | } | |
6126 | ||
553c0369 OG |
6127 | /* Sort locations by count */ |
6128 | sort_r(t->loc, t->count, sizeof(struct location), | |
6129 | cmp_loc_by_count, NULL, NULL); | |
6130 | ||
b3fd64e1 | 6131 | bitmap_free(obj_map); |
64dd6849 FM |
6132 | return 0; |
6133 | } | |
6134 | ||
6135 | static int slab_debug_trace_release(struct inode *inode, struct file *file) | |
6136 | { | |
6137 | struct seq_file *seq = file->private_data; | |
6138 | struct loc_track *t = seq->private; | |
6139 | ||
6140 | free_loc_track(t); | |
6141 | return seq_release_private(inode, file); | |
6142 | } | |
6143 | ||
6144 | static const struct file_operations slab_debugfs_fops = { | |
6145 | .open = slab_debug_trace_open, | |
6146 | .read = seq_read, | |
6147 | .llseek = seq_lseek, | |
6148 | .release = slab_debug_trace_release, | |
6149 | }; | |
6150 | ||
6151 | static void debugfs_slab_add(struct kmem_cache *s) | |
6152 | { | |
6153 | struct dentry *slab_cache_dir; | |
6154 | ||
6155 | if (unlikely(!slab_debugfs_root)) | |
6156 | return; | |
6157 | ||
6158 | slab_cache_dir = debugfs_create_dir(s->name, slab_debugfs_root); | |
6159 | ||
6160 | debugfs_create_file("alloc_traces", 0400, | |
6161 | slab_cache_dir, s, &slab_debugfs_fops); | |
6162 | ||
6163 | debugfs_create_file("free_traces", 0400, | |
6164 | slab_cache_dir, s, &slab_debugfs_fops); | |
6165 | } | |
6166 | ||
6167 | void debugfs_slab_release(struct kmem_cache *s) | |
6168 | { | |
6169 | debugfs_remove_recursive(debugfs_lookup(s->name, slab_debugfs_root)); | |
6170 | } | |
6171 | ||
6172 | static int __init slab_debugfs_init(void) | |
6173 | { | |
6174 | struct kmem_cache *s; | |
6175 | ||
6176 | slab_debugfs_root = debugfs_create_dir("slab", NULL); | |
6177 | ||
6178 | list_for_each_entry(s, &slab_caches, list) | |
6179 | if (s->flags & SLAB_STORE_USER) | |
6180 | debugfs_slab_add(s); | |
6181 | ||
6182 | return 0; | |
6183 | ||
6184 | } | |
6185 | __initcall(slab_debugfs_init); | |
6186 | #endif | |
57ed3eda PE |
6187 | /* |
6188 | * The /proc/slabinfo ABI | |
6189 | */ | |
5b365771 | 6190 | #ifdef CONFIG_SLUB_DEBUG |
0d7561c6 | 6191 | void get_slabinfo(struct kmem_cache *s, struct slabinfo *sinfo) |
57ed3eda | 6192 | { |
57ed3eda | 6193 | unsigned long nr_slabs = 0; |
205ab99d CL |
6194 | unsigned long nr_objs = 0; |
6195 | unsigned long nr_free = 0; | |
57ed3eda | 6196 | int node; |
fa45dc25 | 6197 | struct kmem_cache_node *n; |
57ed3eda | 6198 | |
fa45dc25 | 6199 | for_each_kmem_cache_node(s, node, n) { |
c17fd13e WL |
6200 | nr_slabs += node_nr_slabs(n); |
6201 | nr_objs += node_nr_objs(n); | |
205ab99d | 6202 | nr_free += count_partial(n, count_free); |
57ed3eda PE |
6203 | } |
6204 | ||
0d7561c6 GC |
6205 | sinfo->active_objs = nr_objs - nr_free; |
6206 | sinfo->num_objs = nr_objs; | |
6207 | sinfo->active_slabs = nr_slabs; | |
6208 | sinfo->num_slabs = nr_slabs; | |
6209 | sinfo->objects_per_slab = oo_objects(s->oo); | |
6210 | sinfo->cache_order = oo_order(s->oo); | |
57ed3eda PE |
6211 | } |
6212 | ||
0d7561c6 | 6213 | void slabinfo_show_stats(struct seq_file *m, struct kmem_cache *s) |
7b3c3a50 | 6214 | { |
7b3c3a50 AD |
6215 | } |
6216 | ||
b7454ad3 GC |
6217 | ssize_t slabinfo_write(struct file *file, const char __user *buffer, |
6218 | size_t count, loff_t *ppos) | |
7b3c3a50 | 6219 | { |
b7454ad3 | 6220 | return -EIO; |
7b3c3a50 | 6221 | } |
5b365771 | 6222 | #endif /* CONFIG_SLUB_DEBUG */ |