]>
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> | |
18 | #include <linux/bitops.h> | |
19 | #include <linux/slab.h> | |
97d06609 | 20 | #include "slab.h" |
7b3c3a50 | 21 | #include <linux/proc_fs.h> |
81819f0f | 22 | #include <linux/seq_file.h> |
a79316c6 | 23 | #include <linux/kasan.h> |
81819f0f CL |
24 | #include <linux/cpu.h> |
25 | #include <linux/cpuset.h> | |
26 | #include <linux/mempolicy.h> | |
27 | #include <linux/ctype.h> | |
3ac7fe5a | 28 | #include <linux/debugobjects.h> |
81819f0f | 29 | #include <linux/kallsyms.h> |
b89fb5ef | 30 | #include <linux/kfence.h> |
b9049e23 | 31 | #include <linux/memory.h> |
f8bd2258 | 32 | #include <linux/math64.h> |
773ff60e | 33 | #include <linux/fault-inject.h> |
bfa71457 | 34 | #include <linux/stacktrace.h> |
4de900b4 | 35 | #include <linux/prefetch.h> |
2633d7a0 | 36 | #include <linux/memcontrol.h> |
2482ddec | 37 | #include <linux/random.h> |
81819f0f | 38 | |
4a92379b RK |
39 | #include <trace/events/kmem.h> |
40 | ||
072bb0aa MG |
41 | #include "internal.h" |
42 | ||
81819f0f CL |
43 | /* |
44 | * Lock order: | |
18004c5d | 45 | * 1. slab_mutex (Global Mutex) |
881db7fb CL |
46 | * 2. node->list_lock |
47 | * 3. slab_lock(page) (Only on some arches and for debugging) | |
81819f0f | 48 | * |
18004c5d | 49 | * slab_mutex |
881db7fb | 50 | * |
18004c5d | 51 | * The role of the slab_mutex is to protect the list of all the slabs |
881db7fb CL |
52 | * and to synchronize major metadata changes to slab cache structures. |
53 | * | |
54 | * The slab_lock is only used for debugging and on arches that do not | |
b7ccc7f8 | 55 | * have the ability to do a cmpxchg_double. It only protects: |
881db7fb | 56 | * A. page->freelist -> List of object free in a page |
b7ccc7f8 MW |
57 | * B. page->inuse -> Number of objects in use |
58 | * C. page->objects -> Number of objects in page | |
59 | * D. page->frozen -> frozen state | |
881db7fb CL |
60 | * |
61 | * If a slab is frozen then it is exempt from list management. It is not | |
632b2ef0 LX |
62 | * on any list except per cpu partial list. The processor that froze the |
63 | * slab is the one who can perform list operations on the page. Other | |
64 | * processors may put objects onto the freelist but the processor that | |
65 | * froze the slab is the only one that can retrieve the objects from the | |
66 | * page's freelist. | |
81819f0f CL |
67 | * |
68 | * The list_lock protects the partial and full list on each node and | |
69 | * the partial slab counter. If taken then no new slabs may be added or | |
70 | * removed from the lists nor make the number of partial slabs be modified. | |
71 | * (Note that the total number of slabs is an atomic value that may be | |
72 | * modified without taking the list lock). | |
73 | * | |
74 | * The list_lock is a centralized lock and thus we avoid taking it as | |
75 | * much as possible. As long as SLUB does not have to handle partial | |
76 | * slabs, operations can continue without any centralized lock. F.e. | |
77 | * allocating a long series of objects that fill up slabs does not require | |
78 | * the list lock. | |
81819f0f CL |
79 | * Interrupts are disabled during allocation and deallocation in order to |
80 | * make the slab allocator safe to use in the context of an irq. In addition | |
81 | * interrupts are disabled to ensure that the processor does not change | |
82 | * while handling per_cpu slabs, due to kernel preemption. | |
83 | * | |
84 | * SLUB assigns one slab for allocation to each processor. | |
85 | * Allocations only occur from these slabs called cpu slabs. | |
86 | * | |
672bba3a CL |
87 | * Slabs with free elements are kept on a partial list and during regular |
88 | * operations no list for full slabs is used. If an object in a full slab is | |
81819f0f | 89 | * freed then the slab will show up again on the partial lists. |
672bba3a CL |
90 | * We track full slabs for debugging purposes though because otherwise we |
91 | * cannot scan all objects. | |
81819f0f CL |
92 | * |
93 | * Slabs are freed when they become empty. Teardown and setup is | |
94 | * minimal so we rely on the page allocators per cpu caches for | |
95 | * fast frees and allocs. | |
96 | * | |
aed68148 | 97 | * page->frozen The slab is frozen and exempt from list processing. |
4b6f0750 CL |
98 | * This means that the slab is dedicated to a purpose |
99 | * such as satisfying allocations for a specific | |
100 | * processor. Objects may be freed in the slab while | |
101 | * it is frozen but slab_free will then skip the usual | |
102 | * list operations. It is up to the processor holding | |
103 | * the slab to integrate the slab into the slab lists | |
104 | * when the slab is no longer needed. | |
105 | * | |
106 | * One use of this flag is to mark slabs that are | |
107 | * used for allocations. Then such a slab becomes a cpu | |
108 | * slab. The cpu slab may be equipped with an additional | |
dfb4f096 | 109 | * freelist that allows lockless access to |
894b8788 CL |
110 | * free objects in addition to the regular freelist |
111 | * that requires the slab lock. | |
81819f0f | 112 | * |
aed68148 | 113 | * SLAB_DEBUG_FLAGS Slab requires special handling due to debug |
81819f0f | 114 | * options set. This moves slab handling out of |
894b8788 | 115 | * the fast path and disables lockless freelists. |
81819f0f CL |
116 | */ |
117 | ||
ca0cab65 VB |
118 | #ifdef CONFIG_SLUB_DEBUG |
119 | #ifdef CONFIG_SLUB_DEBUG_ON | |
120 | DEFINE_STATIC_KEY_TRUE(slub_debug_enabled); | |
121 | #else | |
122 | DEFINE_STATIC_KEY_FALSE(slub_debug_enabled); | |
123 | #endif | |
124 | #endif | |
125 | ||
59052e89 VB |
126 | static inline bool kmem_cache_debug(struct kmem_cache *s) |
127 | { | |
128 | return kmem_cache_debug_flags(s, SLAB_DEBUG_FLAGS); | |
af537b0a | 129 | } |
5577bd8a | 130 | |
117d54df | 131 | void *fixup_red_left(struct kmem_cache *s, void *p) |
d86bd1be | 132 | { |
59052e89 | 133 | if (kmem_cache_debug_flags(s, SLAB_RED_ZONE)) |
d86bd1be JK |
134 | p += s->red_left_pad; |
135 | ||
136 | return p; | |
137 | } | |
138 | ||
345c905d JK |
139 | static inline bool kmem_cache_has_cpu_partial(struct kmem_cache *s) |
140 | { | |
141 | #ifdef CONFIG_SLUB_CPU_PARTIAL | |
142 | return !kmem_cache_debug(s); | |
143 | #else | |
144 | return false; | |
145 | #endif | |
146 | } | |
147 | ||
81819f0f CL |
148 | /* |
149 | * Issues still to be resolved: | |
150 | * | |
81819f0f CL |
151 | * - Support PAGE_ALLOC_DEBUG. Should be easy to do. |
152 | * | |
81819f0f CL |
153 | * - Variable sizing of the per node arrays |
154 | */ | |
155 | ||
156 | /* Enable to test recovery from slab corruption on boot */ | |
157 | #undef SLUB_RESILIENCY_TEST | |
158 | ||
b789ef51 CL |
159 | /* Enable to log cmpxchg failures */ |
160 | #undef SLUB_DEBUG_CMPXCHG | |
161 | ||
2086d26a | 162 | /* |
dc84207d | 163 | * Minimum number of partial slabs. These will be left on the partial |
2086d26a CL |
164 | * lists even if they are empty. kmem_cache_shrink may reclaim them. |
165 | */ | |
76be8950 | 166 | #define MIN_PARTIAL 5 |
e95eed57 | 167 | |
2086d26a CL |
168 | /* |
169 | * Maximum number of desirable partial slabs. | |
170 | * The existence of more partial slabs makes kmem_cache_shrink | |
721ae22a | 171 | * sort the partial list by the number of objects in use. |
2086d26a CL |
172 | */ |
173 | #define MAX_PARTIAL 10 | |
174 | ||
becfda68 | 175 | #define DEBUG_DEFAULT_FLAGS (SLAB_CONSISTENCY_CHECKS | SLAB_RED_ZONE | \ |
81819f0f | 176 | SLAB_POISON | SLAB_STORE_USER) |
672bba3a | 177 | |
149daaf3 LA |
178 | /* |
179 | * These debug flags cannot use CMPXCHG because there might be consistency | |
180 | * issues when checking or reading debug information | |
181 | */ | |
182 | #define SLAB_NO_CMPXCHG (SLAB_CONSISTENCY_CHECKS | SLAB_STORE_USER | \ | |
183 | SLAB_TRACE) | |
184 | ||
185 | ||
fa5ec8a1 | 186 | /* |
3de47213 DR |
187 | * Debugging flags that require metadata to be stored in the slab. These get |
188 | * disabled when slub_debug=O is used and a cache's min order increases with | |
189 | * metadata. | |
fa5ec8a1 | 190 | */ |
3de47213 | 191 | #define DEBUG_METADATA_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER) |
fa5ec8a1 | 192 | |
210b5c06 CG |
193 | #define OO_SHIFT 16 |
194 | #define OO_MASK ((1 << OO_SHIFT) - 1) | |
50d5c41c | 195 | #define MAX_OBJS_PER_PAGE 32767 /* since page.objects is u15 */ |
210b5c06 | 196 | |
81819f0f | 197 | /* Internal SLUB flags */ |
d50112ed | 198 | /* Poison object */ |
4fd0b46e | 199 | #define __OBJECT_POISON ((slab_flags_t __force)0x80000000U) |
d50112ed | 200 | /* Use cmpxchg_double */ |
4fd0b46e | 201 | #define __CMPXCHG_DOUBLE ((slab_flags_t __force)0x40000000U) |
81819f0f | 202 | |
02cbc874 CL |
203 | /* |
204 | * Tracking user of a slab. | |
205 | */ | |
d6543e39 | 206 | #define TRACK_ADDRS_COUNT 16 |
02cbc874 | 207 | struct track { |
ce71e27c | 208 | unsigned long addr; /* Called from address */ |
d6543e39 BG |
209 | #ifdef CONFIG_STACKTRACE |
210 | unsigned long addrs[TRACK_ADDRS_COUNT]; /* Called from address */ | |
211 | #endif | |
02cbc874 CL |
212 | int cpu; /* Was running on cpu */ |
213 | int pid; /* Pid context */ | |
214 | unsigned long when; /* When did the operation occur */ | |
215 | }; | |
216 | ||
217 | enum track_item { TRACK_ALLOC, TRACK_FREE }; | |
218 | ||
ab4d5ed5 | 219 | #ifdef CONFIG_SYSFS |
81819f0f CL |
220 | static int sysfs_slab_add(struct kmem_cache *); |
221 | static int sysfs_slab_alias(struct kmem_cache *, const char *); | |
81819f0f | 222 | #else |
0c710013 CL |
223 | static inline int sysfs_slab_add(struct kmem_cache *s) { return 0; } |
224 | static inline int sysfs_slab_alias(struct kmem_cache *s, const char *p) | |
225 | { return 0; } | |
81819f0f CL |
226 | #endif |
227 | ||
4fdccdfb | 228 | static inline void stat(const struct kmem_cache *s, enum stat_item si) |
8ff12cfc CL |
229 | { |
230 | #ifdef CONFIG_SLUB_STATS | |
88da03a6 CL |
231 | /* |
232 | * The rmw is racy on a preemptible kernel but this is acceptable, so | |
233 | * avoid this_cpu_add()'s irq-disable overhead. | |
234 | */ | |
235 | raw_cpu_inc(s->cpu_slab->stat[si]); | |
8ff12cfc CL |
236 | #endif |
237 | } | |
238 | ||
7e1fa93d VB |
239 | /* |
240 | * Tracks for which NUMA nodes we have kmem_cache_nodes allocated. | |
241 | * Corresponds to node_state[N_NORMAL_MEMORY], but can temporarily | |
242 | * differ during memory hotplug/hotremove operations. | |
243 | * Protected by slab_mutex. | |
244 | */ | |
245 | static nodemask_t slab_nodes; | |
246 | ||
81819f0f CL |
247 | /******************************************************************** |
248 | * Core slab cache functions | |
249 | *******************************************************************/ | |
250 | ||
2482ddec KC |
251 | /* |
252 | * Returns freelist pointer (ptr). With hardening, this is obfuscated | |
253 | * with an XOR of the address where the pointer is held and a per-cache | |
254 | * random number. | |
255 | */ | |
256 | static inline void *freelist_ptr(const struct kmem_cache *s, void *ptr, | |
257 | unsigned long ptr_addr) | |
258 | { | |
259 | #ifdef CONFIG_SLAB_FREELIST_HARDENED | |
d36a63a9 | 260 | /* |
aa1ef4d7 | 261 | * When CONFIG_KASAN_SW/HW_TAGS is enabled, ptr_addr might be tagged. |
d36a63a9 AK |
262 | * Normally, this doesn't cause any issues, as both set_freepointer() |
263 | * and get_freepointer() are called with a pointer with the same tag. | |
264 | * However, there are some issues with CONFIG_SLUB_DEBUG code. For | |
265 | * example, when __free_slub() iterates over objects in a cache, it | |
266 | * passes untagged pointers to check_object(). check_object() in turns | |
267 | * calls get_freepointer() with an untagged pointer, which causes the | |
268 | * freepointer to be restored incorrectly. | |
269 | */ | |
270 | return (void *)((unsigned long)ptr ^ s->random ^ | |
1ad53d9f | 271 | swab((unsigned long)kasan_reset_tag((void *)ptr_addr))); |
2482ddec KC |
272 | #else |
273 | return ptr; | |
274 | #endif | |
275 | } | |
276 | ||
277 | /* Returns the freelist pointer recorded at location ptr_addr. */ | |
278 | static inline void *freelist_dereference(const struct kmem_cache *s, | |
279 | void *ptr_addr) | |
280 | { | |
281 | return freelist_ptr(s, (void *)*(unsigned long *)(ptr_addr), | |
282 | (unsigned long)ptr_addr); | |
283 | } | |
284 | ||
7656c72b CL |
285 | static inline void *get_freepointer(struct kmem_cache *s, void *object) |
286 | { | |
aa1ef4d7 | 287 | object = kasan_reset_tag(object); |
2482ddec | 288 | return freelist_dereference(s, object + s->offset); |
7656c72b CL |
289 | } |
290 | ||
0ad9500e ED |
291 | static void prefetch_freepointer(const struct kmem_cache *s, void *object) |
292 | { | |
0882ff91 | 293 | prefetch(object + s->offset); |
0ad9500e ED |
294 | } |
295 | ||
1393d9a1 CL |
296 | static inline void *get_freepointer_safe(struct kmem_cache *s, void *object) |
297 | { | |
2482ddec | 298 | unsigned long freepointer_addr; |
1393d9a1 CL |
299 | void *p; |
300 | ||
8e57f8ac | 301 | if (!debug_pagealloc_enabled_static()) |
922d566c JK |
302 | return get_freepointer(s, object); |
303 | ||
f70b0049 | 304 | object = kasan_reset_tag(object); |
2482ddec | 305 | freepointer_addr = (unsigned long)object + s->offset; |
fe557319 | 306 | copy_from_kernel_nofault(&p, (void **)freepointer_addr, sizeof(p)); |
2482ddec | 307 | return freelist_ptr(s, p, freepointer_addr); |
1393d9a1 CL |
308 | } |
309 | ||
7656c72b CL |
310 | static inline void set_freepointer(struct kmem_cache *s, void *object, void *fp) |
311 | { | |
2482ddec KC |
312 | unsigned long freeptr_addr = (unsigned long)object + s->offset; |
313 | ||
ce6fa91b AP |
314 | #ifdef CONFIG_SLAB_FREELIST_HARDENED |
315 | BUG_ON(object == fp); /* naive detection of double free or corruption */ | |
316 | #endif | |
317 | ||
aa1ef4d7 | 318 | freeptr_addr = (unsigned long)kasan_reset_tag((void *)freeptr_addr); |
2482ddec | 319 | *(void **)freeptr_addr = freelist_ptr(s, fp, freeptr_addr); |
7656c72b CL |
320 | } |
321 | ||
322 | /* Loop over all objects in a slab */ | |
224a88be | 323 | #define for_each_object(__p, __s, __addr, __objects) \ |
d86bd1be JK |
324 | for (__p = fixup_red_left(__s, __addr); \ |
325 | __p < (__addr) + (__objects) * (__s)->size; \ | |
326 | __p += (__s)->size) | |
7656c72b | 327 | |
9736d2a9 | 328 | static inline unsigned int order_objects(unsigned int order, unsigned int size) |
ab9a0f19 | 329 | { |
9736d2a9 | 330 | return ((unsigned int)PAGE_SIZE << order) / size; |
ab9a0f19 LJ |
331 | } |
332 | ||
19af27af | 333 | static inline struct kmem_cache_order_objects oo_make(unsigned int order, |
9736d2a9 | 334 | unsigned int size) |
834f3d11 CL |
335 | { |
336 | struct kmem_cache_order_objects x = { | |
9736d2a9 | 337 | (order << OO_SHIFT) + order_objects(order, size) |
834f3d11 CL |
338 | }; |
339 | ||
340 | return x; | |
341 | } | |
342 | ||
19af27af | 343 | static inline unsigned int oo_order(struct kmem_cache_order_objects x) |
834f3d11 | 344 | { |
210b5c06 | 345 | return x.x >> OO_SHIFT; |
834f3d11 CL |
346 | } |
347 | ||
19af27af | 348 | static inline unsigned int oo_objects(struct kmem_cache_order_objects x) |
834f3d11 | 349 | { |
210b5c06 | 350 | return x.x & OO_MASK; |
834f3d11 CL |
351 | } |
352 | ||
881db7fb CL |
353 | /* |
354 | * Per slab locking using the pagelock | |
355 | */ | |
356 | static __always_inline void slab_lock(struct page *page) | |
357 | { | |
48c935ad | 358 | VM_BUG_ON_PAGE(PageTail(page), page); |
881db7fb CL |
359 | bit_spin_lock(PG_locked, &page->flags); |
360 | } | |
361 | ||
362 | static __always_inline void slab_unlock(struct page *page) | |
363 | { | |
48c935ad | 364 | VM_BUG_ON_PAGE(PageTail(page), page); |
881db7fb CL |
365 | __bit_spin_unlock(PG_locked, &page->flags); |
366 | } | |
367 | ||
1d07171c CL |
368 | /* Interrupts must be disabled (for the fallback code to work right) */ |
369 | static inline bool __cmpxchg_double_slab(struct kmem_cache *s, struct page *page, | |
370 | void *freelist_old, unsigned long counters_old, | |
371 | void *freelist_new, unsigned long counters_new, | |
372 | const char *n) | |
373 | { | |
374 | VM_BUG_ON(!irqs_disabled()); | |
2565409f HC |
375 | #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \ |
376 | defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE) | |
1d07171c | 377 | if (s->flags & __CMPXCHG_DOUBLE) { |
cdcd6298 | 378 | if (cmpxchg_double(&page->freelist, &page->counters, |
0aa9a13d DC |
379 | freelist_old, counters_old, |
380 | freelist_new, counters_new)) | |
6f6528a1 | 381 | return true; |
1d07171c CL |
382 | } else |
383 | #endif | |
384 | { | |
385 | slab_lock(page); | |
d0e0ac97 CG |
386 | if (page->freelist == freelist_old && |
387 | page->counters == counters_old) { | |
1d07171c | 388 | page->freelist = freelist_new; |
7d27a04b | 389 | page->counters = counters_new; |
1d07171c | 390 | slab_unlock(page); |
6f6528a1 | 391 | return true; |
1d07171c CL |
392 | } |
393 | slab_unlock(page); | |
394 | } | |
395 | ||
396 | cpu_relax(); | |
397 | stat(s, CMPXCHG_DOUBLE_FAIL); | |
398 | ||
399 | #ifdef SLUB_DEBUG_CMPXCHG | |
f9f58285 | 400 | pr_info("%s %s: cmpxchg double redo ", n, s->name); |
1d07171c CL |
401 | #endif |
402 | ||
6f6528a1 | 403 | return false; |
1d07171c CL |
404 | } |
405 | ||
b789ef51 CL |
406 | static inline bool cmpxchg_double_slab(struct kmem_cache *s, struct page *page, |
407 | void *freelist_old, unsigned long counters_old, | |
408 | void *freelist_new, unsigned long counters_new, | |
409 | const char *n) | |
410 | { | |
2565409f HC |
411 | #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \ |
412 | defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE) | |
b789ef51 | 413 | if (s->flags & __CMPXCHG_DOUBLE) { |
cdcd6298 | 414 | if (cmpxchg_double(&page->freelist, &page->counters, |
0aa9a13d DC |
415 | freelist_old, counters_old, |
416 | freelist_new, counters_new)) | |
6f6528a1 | 417 | return true; |
b789ef51 CL |
418 | } else |
419 | #endif | |
420 | { | |
1d07171c CL |
421 | unsigned long flags; |
422 | ||
423 | local_irq_save(flags); | |
881db7fb | 424 | slab_lock(page); |
d0e0ac97 CG |
425 | if (page->freelist == freelist_old && |
426 | page->counters == counters_old) { | |
b789ef51 | 427 | page->freelist = freelist_new; |
7d27a04b | 428 | page->counters = counters_new; |
881db7fb | 429 | slab_unlock(page); |
1d07171c | 430 | local_irq_restore(flags); |
6f6528a1 | 431 | return true; |
b789ef51 | 432 | } |
881db7fb | 433 | slab_unlock(page); |
1d07171c | 434 | local_irq_restore(flags); |
b789ef51 CL |
435 | } |
436 | ||
437 | cpu_relax(); | |
438 | stat(s, CMPXCHG_DOUBLE_FAIL); | |
439 | ||
440 | #ifdef SLUB_DEBUG_CMPXCHG | |
f9f58285 | 441 | pr_info("%s %s: cmpxchg double redo ", n, s->name); |
b789ef51 CL |
442 | #endif |
443 | ||
6f6528a1 | 444 | return false; |
b789ef51 CL |
445 | } |
446 | ||
41ecc55b | 447 | #ifdef CONFIG_SLUB_DEBUG |
90e9f6a6 YZ |
448 | static unsigned long object_map[BITS_TO_LONGS(MAX_OBJS_PER_PAGE)]; |
449 | static DEFINE_SPINLOCK(object_map_lock); | |
450 | ||
5f80b13a CL |
451 | /* |
452 | * Determine a map of object in use on a page. | |
453 | * | |
881db7fb | 454 | * Node listlock must be held to guarantee that the page does |
5f80b13a CL |
455 | * not vanish from under us. |
456 | */ | |
90e9f6a6 | 457 | static unsigned long *get_map(struct kmem_cache *s, struct page *page) |
31364c2e | 458 | __acquires(&object_map_lock) |
5f80b13a CL |
459 | { |
460 | void *p; | |
461 | void *addr = page_address(page); | |
462 | ||
90e9f6a6 YZ |
463 | VM_BUG_ON(!irqs_disabled()); |
464 | ||
465 | spin_lock(&object_map_lock); | |
466 | ||
467 | bitmap_zero(object_map, page->objects); | |
468 | ||
5f80b13a | 469 | for (p = page->freelist; p; p = get_freepointer(s, p)) |
4138fdfc | 470 | set_bit(__obj_to_index(s, addr, p), object_map); |
90e9f6a6 YZ |
471 | |
472 | return object_map; | |
473 | } | |
474 | ||
81aba9e0 | 475 | static void put_map(unsigned long *map) __releases(&object_map_lock) |
90e9f6a6 YZ |
476 | { |
477 | VM_BUG_ON(map != object_map); | |
90e9f6a6 | 478 | spin_unlock(&object_map_lock); |
5f80b13a CL |
479 | } |
480 | ||
870b1fbb | 481 | static inline unsigned int size_from_object(struct kmem_cache *s) |
d86bd1be JK |
482 | { |
483 | if (s->flags & SLAB_RED_ZONE) | |
484 | return s->size - s->red_left_pad; | |
485 | ||
486 | return s->size; | |
487 | } | |
488 | ||
489 | static inline void *restore_red_left(struct kmem_cache *s, void *p) | |
490 | { | |
491 | if (s->flags & SLAB_RED_ZONE) | |
492 | p -= s->red_left_pad; | |
493 | ||
494 | return p; | |
495 | } | |
496 | ||
41ecc55b CL |
497 | /* |
498 | * Debug settings: | |
499 | */ | |
89d3c87e | 500 | #if defined(CONFIG_SLUB_DEBUG_ON) |
d50112ed | 501 | static slab_flags_t slub_debug = DEBUG_DEFAULT_FLAGS; |
f0630fff | 502 | #else |
d50112ed | 503 | static slab_flags_t slub_debug; |
f0630fff | 504 | #endif |
41ecc55b | 505 | |
e17f1dfb | 506 | static char *slub_debug_string; |
fa5ec8a1 | 507 | static int disable_higher_order_debug; |
41ecc55b | 508 | |
a79316c6 AR |
509 | /* |
510 | * slub is about to manipulate internal object metadata. This memory lies | |
511 | * outside the range of the allocated object, so accessing it would normally | |
512 | * be reported by kasan as a bounds error. metadata_access_enable() is used | |
513 | * to tell kasan that these accesses are OK. | |
514 | */ | |
515 | static inline void metadata_access_enable(void) | |
516 | { | |
517 | kasan_disable_current(); | |
518 | } | |
519 | ||
520 | static inline void metadata_access_disable(void) | |
521 | { | |
522 | kasan_enable_current(); | |
523 | } | |
524 | ||
81819f0f CL |
525 | /* |
526 | * Object debugging | |
527 | */ | |
d86bd1be JK |
528 | |
529 | /* Verify that a pointer has an address that is valid within a slab page */ | |
530 | static inline int check_valid_pointer(struct kmem_cache *s, | |
531 | struct page *page, void *object) | |
532 | { | |
533 | void *base; | |
534 | ||
535 | if (!object) | |
536 | return 1; | |
537 | ||
538 | base = page_address(page); | |
338cfaad | 539 | object = kasan_reset_tag(object); |
d86bd1be JK |
540 | object = restore_red_left(s, object); |
541 | if (object < base || object >= base + page->objects * s->size || | |
542 | (object - base) % s->size) { | |
543 | return 0; | |
544 | } | |
545 | ||
546 | return 1; | |
547 | } | |
548 | ||
aa2efd5e DT |
549 | static void print_section(char *level, char *text, u8 *addr, |
550 | unsigned int length) | |
81819f0f | 551 | { |
a79316c6 | 552 | metadata_access_enable(); |
aa1ef4d7 AK |
553 | print_hex_dump(level, kasan_reset_tag(text), DUMP_PREFIX_ADDRESS, |
554 | 16, 1, addr, length, 1); | |
a79316c6 | 555 | metadata_access_disable(); |
81819f0f CL |
556 | } |
557 | ||
cbfc35a4 WL |
558 | /* |
559 | * See comment in calculate_sizes(). | |
560 | */ | |
561 | static inline bool freeptr_outside_object(struct kmem_cache *s) | |
562 | { | |
563 | return s->offset >= s->inuse; | |
564 | } | |
565 | ||
566 | /* | |
567 | * Return offset of the end of info block which is inuse + free pointer if | |
568 | * not overlapping with object. | |
569 | */ | |
570 | static inline unsigned int get_info_end(struct kmem_cache *s) | |
571 | { | |
572 | if (freeptr_outside_object(s)) | |
573 | return s->inuse + sizeof(void *); | |
574 | else | |
575 | return s->inuse; | |
576 | } | |
577 | ||
81819f0f CL |
578 | static struct track *get_track(struct kmem_cache *s, void *object, |
579 | enum track_item alloc) | |
580 | { | |
581 | struct track *p; | |
582 | ||
cbfc35a4 | 583 | p = object + get_info_end(s); |
81819f0f | 584 | |
aa1ef4d7 | 585 | return kasan_reset_tag(p + alloc); |
81819f0f CL |
586 | } |
587 | ||
588 | static void set_track(struct kmem_cache *s, void *object, | |
ce71e27c | 589 | enum track_item alloc, unsigned long addr) |
81819f0f | 590 | { |
1a00df4a | 591 | struct track *p = get_track(s, object, alloc); |
81819f0f | 592 | |
81819f0f | 593 | if (addr) { |
d6543e39 | 594 | #ifdef CONFIG_STACKTRACE |
79716799 | 595 | unsigned int nr_entries; |
d6543e39 | 596 | |
a79316c6 | 597 | metadata_access_enable(); |
aa1ef4d7 AK |
598 | nr_entries = stack_trace_save(kasan_reset_tag(p->addrs), |
599 | TRACK_ADDRS_COUNT, 3); | |
a79316c6 | 600 | metadata_access_disable(); |
d6543e39 | 601 | |
79716799 TG |
602 | if (nr_entries < TRACK_ADDRS_COUNT) |
603 | p->addrs[nr_entries] = 0; | |
d6543e39 | 604 | #endif |
81819f0f CL |
605 | p->addr = addr; |
606 | p->cpu = smp_processor_id(); | |
88e4ccf2 | 607 | p->pid = current->pid; |
81819f0f | 608 | p->when = jiffies; |
b8ca7ff7 | 609 | } else { |
81819f0f | 610 | memset(p, 0, sizeof(struct track)); |
b8ca7ff7 | 611 | } |
81819f0f CL |
612 | } |
613 | ||
81819f0f CL |
614 | static void init_tracking(struct kmem_cache *s, void *object) |
615 | { | |
24922684 CL |
616 | if (!(s->flags & SLAB_STORE_USER)) |
617 | return; | |
618 | ||
ce71e27c EGM |
619 | set_track(s, object, TRACK_FREE, 0UL); |
620 | set_track(s, object, TRACK_ALLOC, 0UL); | |
81819f0f CL |
621 | } |
622 | ||
86609d33 | 623 | static void print_track(const char *s, struct track *t, unsigned long pr_time) |
81819f0f CL |
624 | { |
625 | if (!t->addr) | |
626 | return; | |
627 | ||
96b94abc | 628 | pr_err("%s in %pS age=%lu cpu=%u pid=%d\n", |
86609d33 | 629 | s, (void *)t->addr, pr_time - t->when, t->cpu, t->pid); |
d6543e39 BG |
630 | #ifdef CONFIG_STACKTRACE |
631 | { | |
632 | int i; | |
633 | for (i = 0; i < TRACK_ADDRS_COUNT; i++) | |
634 | if (t->addrs[i]) | |
f9f58285 | 635 | pr_err("\t%pS\n", (void *)t->addrs[i]); |
d6543e39 BG |
636 | else |
637 | break; | |
638 | } | |
639 | #endif | |
24922684 CL |
640 | } |
641 | ||
e42f174e | 642 | void print_tracking(struct kmem_cache *s, void *object) |
24922684 | 643 | { |
86609d33 | 644 | unsigned long pr_time = jiffies; |
24922684 CL |
645 | if (!(s->flags & SLAB_STORE_USER)) |
646 | return; | |
647 | ||
86609d33 CP |
648 | print_track("Allocated", get_track(s, object, TRACK_ALLOC), pr_time); |
649 | print_track("Freed", get_track(s, object, TRACK_FREE), pr_time); | |
24922684 CL |
650 | } |
651 | ||
652 | static void print_page_info(struct page *page) | |
653 | { | |
96b94abc | 654 | pr_err("Slab 0x%p objects=%u used=%u fp=0x%p flags=%#lx(%pGp)\n", |
4a8ef190 YS |
655 | page, page->objects, page->inuse, page->freelist, |
656 | page->flags, &page->flags); | |
24922684 CL |
657 | |
658 | } | |
659 | ||
660 | static void slab_bug(struct kmem_cache *s, char *fmt, ...) | |
661 | { | |
ecc42fbe | 662 | struct va_format vaf; |
24922684 | 663 | va_list args; |
24922684 CL |
664 | |
665 | va_start(args, fmt); | |
ecc42fbe FF |
666 | vaf.fmt = fmt; |
667 | vaf.va = &args; | |
f9f58285 | 668 | pr_err("=============================================================================\n"); |
ecc42fbe | 669 | pr_err("BUG %s (%s): %pV\n", s->name, print_tainted(), &vaf); |
f9f58285 | 670 | pr_err("-----------------------------------------------------------------------------\n\n"); |
645df230 | 671 | |
373d4d09 | 672 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); |
ecc42fbe | 673 | va_end(args); |
81819f0f CL |
674 | } |
675 | ||
24922684 CL |
676 | static void slab_fix(struct kmem_cache *s, char *fmt, ...) |
677 | { | |
ecc42fbe | 678 | struct va_format vaf; |
24922684 | 679 | va_list args; |
24922684 CL |
680 | |
681 | va_start(args, fmt); | |
ecc42fbe FF |
682 | vaf.fmt = fmt; |
683 | vaf.va = &args; | |
684 | pr_err("FIX %s: %pV\n", s->name, &vaf); | |
24922684 | 685 | va_end(args); |
24922684 CL |
686 | } |
687 | ||
52f23478 | 688 | static bool freelist_corrupted(struct kmem_cache *s, struct page *page, |
dc07a728 | 689 | void **freelist, void *nextfree) |
52f23478 DZ |
690 | { |
691 | if ((s->flags & SLAB_CONSISTENCY_CHECKS) && | |
dc07a728 ER |
692 | !check_valid_pointer(s, page, nextfree) && freelist) { |
693 | object_err(s, page, *freelist, "Freechain corrupt"); | |
694 | *freelist = NULL; | |
52f23478 DZ |
695 | slab_fix(s, "Isolate corrupted freechain"); |
696 | return true; | |
697 | } | |
698 | ||
699 | return false; | |
700 | } | |
701 | ||
24922684 | 702 | static void print_trailer(struct kmem_cache *s, struct page *page, u8 *p) |
81819f0f CL |
703 | { |
704 | unsigned int off; /* Offset of last byte */ | |
a973e9dd | 705 | u8 *addr = page_address(page); |
24922684 CL |
706 | |
707 | print_tracking(s, p); | |
708 | ||
709 | print_page_info(page); | |
710 | ||
96b94abc | 711 | pr_err("Object 0x%p @offset=%tu fp=0x%p\n\n", |
f9f58285 | 712 | p, p - addr, get_freepointer(s, p)); |
24922684 | 713 | |
d86bd1be | 714 | if (s->flags & SLAB_RED_ZONE) |
8669dbab | 715 | print_section(KERN_ERR, "Redzone ", p - s->red_left_pad, |
aa2efd5e | 716 | s->red_left_pad); |
d86bd1be | 717 | else if (p > addr + 16) |
aa2efd5e | 718 | print_section(KERN_ERR, "Bytes b4 ", p - 16, 16); |
81819f0f | 719 | |
8669dbab | 720 | print_section(KERN_ERR, "Object ", p, |
1b473f29 | 721 | min_t(unsigned int, s->object_size, PAGE_SIZE)); |
81819f0f | 722 | if (s->flags & SLAB_RED_ZONE) |
8669dbab | 723 | print_section(KERN_ERR, "Redzone ", p + s->object_size, |
3b0efdfa | 724 | s->inuse - s->object_size); |
81819f0f | 725 | |
cbfc35a4 | 726 | off = get_info_end(s); |
81819f0f | 727 | |
24922684 | 728 | if (s->flags & SLAB_STORE_USER) |
81819f0f | 729 | off += 2 * sizeof(struct track); |
81819f0f | 730 | |
80a9201a AP |
731 | off += kasan_metadata_size(s); |
732 | ||
d86bd1be | 733 | if (off != size_from_object(s)) |
81819f0f | 734 | /* Beginning of the filler is the free pointer */ |
8669dbab | 735 | print_section(KERN_ERR, "Padding ", p + off, |
aa2efd5e | 736 | size_from_object(s) - off); |
24922684 CL |
737 | |
738 | dump_stack(); | |
81819f0f CL |
739 | } |
740 | ||
75c66def | 741 | void object_err(struct kmem_cache *s, struct page *page, |
81819f0f CL |
742 | u8 *object, char *reason) |
743 | { | |
3dc50637 | 744 | slab_bug(s, "%s", reason); |
24922684 | 745 | print_trailer(s, page, object); |
81819f0f CL |
746 | } |
747 | ||
a38965bf | 748 | static __printf(3, 4) void slab_err(struct kmem_cache *s, struct page *page, |
d0e0ac97 | 749 | const char *fmt, ...) |
81819f0f CL |
750 | { |
751 | va_list args; | |
752 | char buf[100]; | |
753 | ||
24922684 CL |
754 | va_start(args, fmt); |
755 | vsnprintf(buf, sizeof(buf), fmt, args); | |
81819f0f | 756 | va_end(args); |
3dc50637 | 757 | slab_bug(s, "%s", buf); |
24922684 | 758 | print_page_info(page); |
81819f0f CL |
759 | dump_stack(); |
760 | } | |
761 | ||
f7cb1933 | 762 | static void init_object(struct kmem_cache *s, void *object, u8 val) |
81819f0f | 763 | { |
aa1ef4d7 | 764 | u8 *p = kasan_reset_tag(object); |
81819f0f | 765 | |
d86bd1be JK |
766 | if (s->flags & SLAB_RED_ZONE) |
767 | memset(p - s->red_left_pad, val, s->red_left_pad); | |
768 | ||
81819f0f | 769 | if (s->flags & __OBJECT_POISON) { |
3b0efdfa CL |
770 | memset(p, POISON_FREE, s->object_size - 1); |
771 | p[s->object_size - 1] = POISON_END; | |
81819f0f CL |
772 | } |
773 | ||
774 | if (s->flags & SLAB_RED_ZONE) | |
3b0efdfa | 775 | memset(p + s->object_size, val, s->inuse - s->object_size); |
81819f0f CL |
776 | } |
777 | ||
24922684 CL |
778 | static void restore_bytes(struct kmem_cache *s, char *message, u8 data, |
779 | void *from, void *to) | |
780 | { | |
781 | slab_fix(s, "Restoring 0x%p-0x%p=0x%x\n", from, to - 1, data); | |
782 | memset(from, data, to - from); | |
783 | } | |
784 | ||
785 | static int check_bytes_and_report(struct kmem_cache *s, struct page *page, | |
786 | u8 *object, char *what, | |
06428780 | 787 | u8 *start, unsigned int value, unsigned int bytes) |
24922684 CL |
788 | { |
789 | u8 *fault; | |
790 | u8 *end; | |
e1b70dd1 | 791 | u8 *addr = page_address(page); |
24922684 | 792 | |
a79316c6 | 793 | metadata_access_enable(); |
aa1ef4d7 | 794 | fault = memchr_inv(kasan_reset_tag(start), value, bytes); |
a79316c6 | 795 | metadata_access_disable(); |
24922684 CL |
796 | if (!fault) |
797 | return 1; | |
798 | ||
799 | end = start + bytes; | |
800 | while (end > fault && end[-1] == value) | |
801 | end--; | |
802 | ||
803 | slab_bug(s, "%s overwritten", what); | |
96b94abc | 804 | pr_err("0x%p-0x%p @offset=%tu. First byte 0x%x instead of 0x%x\n", |
e1b70dd1 MC |
805 | fault, end - 1, fault - addr, |
806 | fault[0], value); | |
24922684 CL |
807 | print_trailer(s, page, object); |
808 | ||
809 | restore_bytes(s, what, value, fault, end); | |
810 | return 0; | |
81819f0f CL |
811 | } |
812 | ||
81819f0f CL |
813 | /* |
814 | * Object layout: | |
815 | * | |
816 | * object address | |
817 | * Bytes of the object to be managed. | |
818 | * If the freepointer may overlay the object then the free | |
cbfc35a4 | 819 | * pointer is at the middle of the object. |
672bba3a | 820 | * |
81819f0f CL |
821 | * Poisoning uses 0x6b (POISON_FREE) and the last byte is |
822 | * 0xa5 (POISON_END) | |
823 | * | |
3b0efdfa | 824 | * object + s->object_size |
81819f0f | 825 | * Padding to reach word boundary. This is also used for Redzoning. |
672bba3a | 826 | * Padding is extended by another word if Redzoning is enabled and |
3b0efdfa | 827 | * object_size == inuse. |
672bba3a | 828 | * |
81819f0f CL |
829 | * We fill with 0xbb (RED_INACTIVE) for inactive objects and with |
830 | * 0xcc (RED_ACTIVE) for objects in use. | |
831 | * | |
832 | * object + s->inuse | |
672bba3a CL |
833 | * Meta data starts here. |
834 | * | |
81819f0f CL |
835 | * A. Free pointer (if we cannot overwrite object on free) |
836 | * B. Tracking data for SLAB_STORE_USER | |
dc84207d | 837 | * C. Padding to reach required alignment boundary or at minimum |
6446faa2 | 838 | * one word if debugging is on to be able to detect writes |
672bba3a CL |
839 | * before the word boundary. |
840 | * | |
841 | * Padding is done using 0x5a (POISON_INUSE) | |
81819f0f CL |
842 | * |
843 | * object + s->size | |
672bba3a | 844 | * Nothing is used beyond s->size. |
81819f0f | 845 | * |
3b0efdfa | 846 | * If slabcaches are merged then the object_size and inuse boundaries are mostly |
672bba3a | 847 | * ignored. And therefore no slab options that rely on these boundaries |
81819f0f CL |
848 | * may be used with merged slabcaches. |
849 | */ | |
850 | ||
81819f0f CL |
851 | static int check_pad_bytes(struct kmem_cache *s, struct page *page, u8 *p) |
852 | { | |
cbfc35a4 | 853 | unsigned long off = get_info_end(s); /* The end of info */ |
81819f0f CL |
854 | |
855 | if (s->flags & SLAB_STORE_USER) | |
856 | /* We also have user information there */ | |
857 | off += 2 * sizeof(struct track); | |
858 | ||
80a9201a AP |
859 | off += kasan_metadata_size(s); |
860 | ||
d86bd1be | 861 | if (size_from_object(s) == off) |
81819f0f CL |
862 | return 1; |
863 | ||
24922684 | 864 | return check_bytes_and_report(s, page, p, "Object padding", |
d86bd1be | 865 | p + off, POISON_INUSE, size_from_object(s) - off); |
81819f0f CL |
866 | } |
867 | ||
39b26464 | 868 | /* Check the pad bytes at the end of a slab page */ |
81819f0f CL |
869 | static int slab_pad_check(struct kmem_cache *s, struct page *page) |
870 | { | |
24922684 CL |
871 | u8 *start; |
872 | u8 *fault; | |
873 | u8 *end; | |
5d682681 | 874 | u8 *pad; |
24922684 CL |
875 | int length; |
876 | int remainder; | |
81819f0f CL |
877 | |
878 | if (!(s->flags & SLAB_POISON)) | |
879 | return 1; | |
880 | ||
a973e9dd | 881 | start = page_address(page); |
a50b854e | 882 | length = page_size(page); |
39b26464 CL |
883 | end = start + length; |
884 | remainder = length % s->size; | |
81819f0f CL |
885 | if (!remainder) |
886 | return 1; | |
887 | ||
5d682681 | 888 | pad = end - remainder; |
a79316c6 | 889 | metadata_access_enable(); |
aa1ef4d7 | 890 | fault = memchr_inv(kasan_reset_tag(pad), POISON_INUSE, remainder); |
a79316c6 | 891 | metadata_access_disable(); |
24922684 CL |
892 | if (!fault) |
893 | return 1; | |
894 | while (end > fault && end[-1] == POISON_INUSE) | |
895 | end--; | |
896 | ||
e1b70dd1 MC |
897 | slab_err(s, page, "Padding overwritten. 0x%p-0x%p @offset=%tu", |
898 | fault, end - 1, fault - start); | |
5d682681 | 899 | print_section(KERN_ERR, "Padding ", pad, remainder); |
24922684 | 900 | |
5d682681 | 901 | restore_bytes(s, "slab padding", POISON_INUSE, fault, end); |
24922684 | 902 | return 0; |
81819f0f CL |
903 | } |
904 | ||
905 | static int check_object(struct kmem_cache *s, struct page *page, | |
f7cb1933 | 906 | void *object, u8 val) |
81819f0f CL |
907 | { |
908 | u8 *p = object; | |
3b0efdfa | 909 | u8 *endobject = object + s->object_size; |
81819f0f CL |
910 | |
911 | if (s->flags & SLAB_RED_ZONE) { | |
8669dbab | 912 | if (!check_bytes_and_report(s, page, object, "Left Redzone", |
d86bd1be JK |
913 | object - s->red_left_pad, val, s->red_left_pad)) |
914 | return 0; | |
915 | ||
8669dbab | 916 | if (!check_bytes_and_report(s, page, object, "Right Redzone", |
3b0efdfa | 917 | endobject, val, s->inuse - s->object_size)) |
81819f0f | 918 | return 0; |
81819f0f | 919 | } else { |
3b0efdfa | 920 | if ((s->flags & SLAB_POISON) && s->object_size < s->inuse) { |
3adbefee | 921 | check_bytes_and_report(s, page, p, "Alignment padding", |
d0e0ac97 CG |
922 | endobject, POISON_INUSE, |
923 | s->inuse - s->object_size); | |
3adbefee | 924 | } |
81819f0f CL |
925 | } |
926 | ||
927 | if (s->flags & SLAB_POISON) { | |
f7cb1933 | 928 | if (val != SLUB_RED_ACTIVE && (s->flags & __OBJECT_POISON) && |
24922684 | 929 | (!check_bytes_and_report(s, page, p, "Poison", p, |
3b0efdfa | 930 | POISON_FREE, s->object_size - 1) || |
8669dbab | 931 | !check_bytes_and_report(s, page, p, "End Poison", |
3b0efdfa | 932 | p + s->object_size - 1, POISON_END, 1))) |
81819f0f | 933 | return 0; |
81819f0f CL |
934 | /* |
935 | * check_pad_bytes cleans up on its own. | |
936 | */ | |
937 | check_pad_bytes(s, page, p); | |
938 | } | |
939 | ||
cbfc35a4 | 940 | if (!freeptr_outside_object(s) && val == SLUB_RED_ACTIVE) |
81819f0f CL |
941 | /* |
942 | * Object and freepointer overlap. Cannot check | |
943 | * freepointer while object is allocated. | |
944 | */ | |
945 | return 1; | |
946 | ||
947 | /* Check free pointer validity */ | |
948 | if (!check_valid_pointer(s, page, get_freepointer(s, p))) { | |
949 | object_err(s, page, p, "Freepointer corrupt"); | |
950 | /* | |
9f6c708e | 951 | * No choice but to zap it and thus lose the remainder |
81819f0f | 952 | * of the free objects in this slab. May cause |
672bba3a | 953 | * another error because the object count is now wrong. |
81819f0f | 954 | */ |
a973e9dd | 955 | set_freepointer(s, p, NULL); |
81819f0f CL |
956 | return 0; |
957 | } | |
958 | return 1; | |
959 | } | |
960 | ||
961 | static int check_slab(struct kmem_cache *s, struct page *page) | |
962 | { | |
39b26464 CL |
963 | int maxobj; |
964 | ||
81819f0f CL |
965 | VM_BUG_ON(!irqs_disabled()); |
966 | ||
967 | if (!PageSlab(page)) { | |
24922684 | 968 | slab_err(s, page, "Not a valid slab page"); |
81819f0f CL |
969 | return 0; |
970 | } | |
39b26464 | 971 | |
9736d2a9 | 972 | maxobj = order_objects(compound_order(page), s->size); |
39b26464 CL |
973 | if (page->objects > maxobj) { |
974 | slab_err(s, page, "objects %u > max %u", | |
f6edde9c | 975 | page->objects, maxobj); |
39b26464 CL |
976 | return 0; |
977 | } | |
978 | if (page->inuse > page->objects) { | |
24922684 | 979 | slab_err(s, page, "inuse %u > max %u", |
f6edde9c | 980 | page->inuse, page->objects); |
81819f0f CL |
981 | return 0; |
982 | } | |
983 | /* Slab_pad_check fixes things up after itself */ | |
984 | slab_pad_check(s, page); | |
985 | return 1; | |
986 | } | |
987 | ||
988 | /* | |
672bba3a CL |
989 | * Determine if a certain object on a page is on the freelist. Must hold the |
990 | * slab lock to guarantee that the chains are in a consistent state. | |
81819f0f CL |
991 | */ |
992 | static int on_freelist(struct kmem_cache *s, struct page *page, void *search) | |
993 | { | |
994 | int nr = 0; | |
881db7fb | 995 | void *fp; |
81819f0f | 996 | void *object = NULL; |
f6edde9c | 997 | int max_objects; |
81819f0f | 998 | |
881db7fb | 999 | fp = page->freelist; |
39b26464 | 1000 | while (fp && nr <= page->objects) { |
81819f0f CL |
1001 | if (fp == search) |
1002 | return 1; | |
1003 | if (!check_valid_pointer(s, page, fp)) { | |
1004 | if (object) { | |
1005 | object_err(s, page, object, | |
1006 | "Freechain corrupt"); | |
a973e9dd | 1007 | set_freepointer(s, object, NULL); |
81819f0f | 1008 | } else { |
24922684 | 1009 | slab_err(s, page, "Freepointer corrupt"); |
a973e9dd | 1010 | page->freelist = NULL; |
39b26464 | 1011 | page->inuse = page->objects; |
24922684 | 1012 | slab_fix(s, "Freelist cleared"); |
81819f0f CL |
1013 | return 0; |
1014 | } | |
1015 | break; | |
1016 | } | |
1017 | object = fp; | |
1018 | fp = get_freepointer(s, object); | |
1019 | nr++; | |
1020 | } | |
1021 | ||
9736d2a9 | 1022 | max_objects = order_objects(compound_order(page), s->size); |
210b5c06 CG |
1023 | if (max_objects > MAX_OBJS_PER_PAGE) |
1024 | max_objects = MAX_OBJS_PER_PAGE; | |
224a88be CL |
1025 | |
1026 | if (page->objects != max_objects) { | |
756a025f JP |
1027 | slab_err(s, page, "Wrong number of objects. Found %d but should be %d", |
1028 | page->objects, max_objects); | |
224a88be CL |
1029 | page->objects = max_objects; |
1030 | slab_fix(s, "Number of objects adjusted."); | |
1031 | } | |
39b26464 | 1032 | if (page->inuse != page->objects - nr) { |
756a025f JP |
1033 | slab_err(s, page, "Wrong object count. Counter is %d but counted were %d", |
1034 | page->inuse, page->objects - nr); | |
39b26464 | 1035 | page->inuse = page->objects - nr; |
24922684 | 1036 | slab_fix(s, "Object count adjusted."); |
81819f0f CL |
1037 | } |
1038 | return search == NULL; | |
1039 | } | |
1040 | ||
0121c619 CL |
1041 | static void trace(struct kmem_cache *s, struct page *page, void *object, |
1042 | int alloc) | |
3ec09742 CL |
1043 | { |
1044 | if (s->flags & SLAB_TRACE) { | |
f9f58285 | 1045 | pr_info("TRACE %s %s 0x%p inuse=%d fp=0x%p\n", |
3ec09742 CL |
1046 | s->name, |
1047 | alloc ? "alloc" : "free", | |
1048 | object, page->inuse, | |
1049 | page->freelist); | |
1050 | ||
1051 | if (!alloc) | |
aa2efd5e | 1052 | print_section(KERN_INFO, "Object ", (void *)object, |
d0e0ac97 | 1053 | s->object_size); |
3ec09742 CL |
1054 | |
1055 | dump_stack(); | |
1056 | } | |
1057 | } | |
1058 | ||
643b1138 | 1059 | /* |
672bba3a | 1060 | * Tracking of fully allocated slabs for debugging purposes. |
643b1138 | 1061 | */ |
5cc6eee8 CL |
1062 | static void add_full(struct kmem_cache *s, |
1063 | struct kmem_cache_node *n, struct page *page) | |
643b1138 | 1064 | { |
5cc6eee8 CL |
1065 | if (!(s->flags & SLAB_STORE_USER)) |
1066 | return; | |
1067 | ||
255d0884 | 1068 | lockdep_assert_held(&n->list_lock); |
916ac052 | 1069 | list_add(&page->slab_list, &n->full); |
643b1138 CL |
1070 | } |
1071 | ||
c65c1877 | 1072 | static void remove_full(struct kmem_cache *s, struct kmem_cache_node *n, struct page *page) |
643b1138 | 1073 | { |
643b1138 CL |
1074 | if (!(s->flags & SLAB_STORE_USER)) |
1075 | return; | |
1076 | ||
255d0884 | 1077 | lockdep_assert_held(&n->list_lock); |
916ac052 | 1078 | list_del(&page->slab_list); |
643b1138 CL |
1079 | } |
1080 | ||
0f389ec6 CL |
1081 | /* Tracking of the number of slabs for debugging purposes */ |
1082 | static inline unsigned long slabs_node(struct kmem_cache *s, int node) | |
1083 | { | |
1084 | struct kmem_cache_node *n = get_node(s, node); | |
1085 | ||
1086 | return atomic_long_read(&n->nr_slabs); | |
1087 | } | |
1088 | ||
26c02cf0 AB |
1089 | static inline unsigned long node_nr_slabs(struct kmem_cache_node *n) |
1090 | { | |
1091 | return atomic_long_read(&n->nr_slabs); | |
1092 | } | |
1093 | ||
205ab99d | 1094 | static inline void inc_slabs_node(struct kmem_cache *s, int node, int objects) |
0f389ec6 CL |
1095 | { |
1096 | struct kmem_cache_node *n = get_node(s, node); | |
1097 | ||
1098 | /* | |
1099 | * May be called early in order to allocate a slab for the | |
1100 | * kmem_cache_node structure. Solve the chicken-egg | |
1101 | * dilemma by deferring the increment of the count during | |
1102 | * bootstrap (see early_kmem_cache_node_alloc). | |
1103 | */ | |
338b2642 | 1104 | if (likely(n)) { |
0f389ec6 | 1105 | atomic_long_inc(&n->nr_slabs); |
205ab99d CL |
1106 | atomic_long_add(objects, &n->total_objects); |
1107 | } | |
0f389ec6 | 1108 | } |
205ab99d | 1109 | static inline void dec_slabs_node(struct kmem_cache *s, int node, int objects) |
0f389ec6 CL |
1110 | { |
1111 | struct kmem_cache_node *n = get_node(s, node); | |
1112 | ||
1113 | atomic_long_dec(&n->nr_slabs); | |
205ab99d | 1114 | atomic_long_sub(objects, &n->total_objects); |
0f389ec6 CL |
1115 | } |
1116 | ||
1117 | /* Object debug checks for alloc/free paths */ | |
3ec09742 CL |
1118 | static void setup_object_debug(struct kmem_cache *s, struct page *page, |
1119 | void *object) | |
1120 | { | |
8fc8d666 | 1121 | if (!kmem_cache_debug_flags(s, SLAB_STORE_USER|SLAB_RED_ZONE|__OBJECT_POISON)) |
3ec09742 CL |
1122 | return; |
1123 | ||
f7cb1933 | 1124 | init_object(s, object, SLUB_RED_INACTIVE); |
3ec09742 CL |
1125 | init_tracking(s, object); |
1126 | } | |
1127 | ||
a50b854e MWO |
1128 | static |
1129 | void setup_page_debug(struct kmem_cache *s, struct page *page, void *addr) | |
a7101224 | 1130 | { |
8fc8d666 | 1131 | if (!kmem_cache_debug_flags(s, SLAB_POISON)) |
a7101224 AK |
1132 | return; |
1133 | ||
1134 | metadata_access_enable(); | |
aa1ef4d7 | 1135 | memset(kasan_reset_tag(addr), POISON_INUSE, page_size(page)); |
a7101224 AK |
1136 | metadata_access_disable(); |
1137 | } | |
1138 | ||
becfda68 | 1139 | static inline int alloc_consistency_checks(struct kmem_cache *s, |
278d7756 | 1140 | struct page *page, void *object) |
81819f0f CL |
1141 | { |
1142 | if (!check_slab(s, page)) | |
becfda68 | 1143 | return 0; |
81819f0f | 1144 | |
81819f0f CL |
1145 | if (!check_valid_pointer(s, page, object)) { |
1146 | object_err(s, page, object, "Freelist Pointer check fails"); | |
becfda68 | 1147 | return 0; |
81819f0f CL |
1148 | } |
1149 | ||
f7cb1933 | 1150 | if (!check_object(s, page, object, SLUB_RED_INACTIVE)) |
becfda68 LA |
1151 | return 0; |
1152 | ||
1153 | return 1; | |
1154 | } | |
1155 | ||
1156 | static noinline int alloc_debug_processing(struct kmem_cache *s, | |
1157 | struct page *page, | |
1158 | void *object, unsigned long addr) | |
1159 | { | |
1160 | if (s->flags & SLAB_CONSISTENCY_CHECKS) { | |
278d7756 | 1161 | if (!alloc_consistency_checks(s, page, object)) |
becfda68 LA |
1162 | goto bad; |
1163 | } | |
81819f0f | 1164 | |
3ec09742 CL |
1165 | /* Success perform special debug activities for allocs */ |
1166 | if (s->flags & SLAB_STORE_USER) | |
1167 | set_track(s, object, TRACK_ALLOC, addr); | |
1168 | trace(s, page, object, 1); | |
f7cb1933 | 1169 | init_object(s, object, SLUB_RED_ACTIVE); |
81819f0f | 1170 | return 1; |
3ec09742 | 1171 | |
81819f0f CL |
1172 | bad: |
1173 | if (PageSlab(page)) { | |
1174 | /* | |
1175 | * If this is a slab page then lets do the best we can | |
1176 | * to avoid issues in the future. Marking all objects | |
672bba3a | 1177 | * as used avoids touching the remaining objects. |
81819f0f | 1178 | */ |
24922684 | 1179 | slab_fix(s, "Marking all objects used"); |
39b26464 | 1180 | page->inuse = page->objects; |
a973e9dd | 1181 | page->freelist = NULL; |
81819f0f CL |
1182 | } |
1183 | return 0; | |
1184 | } | |
1185 | ||
becfda68 LA |
1186 | static inline int free_consistency_checks(struct kmem_cache *s, |
1187 | struct page *page, void *object, unsigned long addr) | |
81819f0f | 1188 | { |
81819f0f | 1189 | if (!check_valid_pointer(s, page, object)) { |
70d71228 | 1190 | slab_err(s, page, "Invalid object pointer 0x%p", object); |
becfda68 | 1191 | return 0; |
81819f0f CL |
1192 | } |
1193 | ||
1194 | if (on_freelist(s, page, object)) { | |
24922684 | 1195 | object_err(s, page, object, "Object already free"); |
becfda68 | 1196 | return 0; |
81819f0f CL |
1197 | } |
1198 | ||
f7cb1933 | 1199 | if (!check_object(s, page, object, SLUB_RED_ACTIVE)) |
becfda68 | 1200 | return 0; |
81819f0f | 1201 | |
1b4f59e3 | 1202 | if (unlikely(s != page->slab_cache)) { |
3adbefee | 1203 | if (!PageSlab(page)) { |
756a025f JP |
1204 | slab_err(s, page, "Attempt to free object(0x%p) outside of slab", |
1205 | object); | |
1b4f59e3 | 1206 | } else if (!page->slab_cache) { |
f9f58285 FF |
1207 | pr_err("SLUB <none>: no slab for object 0x%p.\n", |
1208 | object); | |
70d71228 | 1209 | dump_stack(); |
06428780 | 1210 | } else |
24922684 CL |
1211 | object_err(s, page, object, |
1212 | "page slab pointer corrupt."); | |
becfda68 LA |
1213 | return 0; |
1214 | } | |
1215 | return 1; | |
1216 | } | |
1217 | ||
1218 | /* Supports checking bulk free of a constructed freelist */ | |
1219 | static noinline int free_debug_processing( | |
1220 | struct kmem_cache *s, struct page *page, | |
1221 | void *head, void *tail, int bulk_cnt, | |
1222 | unsigned long addr) | |
1223 | { | |
1224 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); | |
1225 | void *object = head; | |
1226 | int cnt = 0; | |
3f649ab7 | 1227 | unsigned long flags; |
becfda68 LA |
1228 | int ret = 0; |
1229 | ||
1230 | spin_lock_irqsave(&n->list_lock, flags); | |
1231 | slab_lock(page); | |
1232 | ||
1233 | if (s->flags & SLAB_CONSISTENCY_CHECKS) { | |
1234 | if (!check_slab(s, page)) | |
1235 | goto out; | |
1236 | } | |
1237 | ||
1238 | next_object: | |
1239 | cnt++; | |
1240 | ||
1241 | if (s->flags & SLAB_CONSISTENCY_CHECKS) { | |
1242 | if (!free_consistency_checks(s, page, object, addr)) | |
1243 | goto out; | |
81819f0f | 1244 | } |
3ec09742 | 1245 | |
3ec09742 CL |
1246 | if (s->flags & SLAB_STORE_USER) |
1247 | set_track(s, object, TRACK_FREE, addr); | |
1248 | trace(s, page, object, 0); | |
81084651 | 1249 | /* Freepointer not overwritten by init_object(), SLAB_POISON moved it */ |
f7cb1933 | 1250 | init_object(s, object, SLUB_RED_INACTIVE); |
81084651 JDB |
1251 | |
1252 | /* Reached end of constructed freelist yet? */ | |
1253 | if (object != tail) { | |
1254 | object = get_freepointer(s, object); | |
1255 | goto next_object; | |
1256 | } | |
804aa132 LA |
1257 | ret = 1; |
1258 | ||
5c2e4bbb | 1259 | out: |
81084651 JDB |
1260 | if (cnt != bulk_cnt) |
1261 | slab_err(s, page, "Bulk freelist count(%d) invalid(%d)\n", | |
1262 | bulk_cnt, cnt); | |
1263 | ||
881db7fb | 1264 | slab_unlock(page); |
282acb43 | 1265 | spin_unlock_irqrestore(&n->list_lock, flags); |
804aa132 LA |
1266 | if (!ret) |
1267 | slab_fix(s, "Object at 0x%p not freed", object); | |
1268 | return ret; | |
81819f0f CL |
1269 | } |
1270 | ||
e17f1dfb VB |
1271 | /* |
1272 | * Parse a block of slub_debug options. Blocks are delimited by ';' | |
1273 | * | |
1274 | * @str: start of block | |
1275 | * @flags: returns parsed flags, or DEBUG_DEFAULT_FLAGS if none specified | |
1276 | * @slabs: return start of list of slabs, or NULL when there's no list | |
1277 | * @init: assume this is initial parsing and not per-kmem-create parsing | |
1278 | * | |
1279 | * returns the start of next block if there's any, or NULL | |
1280 | */ | |
1281 | static char * | |
1282 | parse_slub_debug_flags(char *str, slab_flags_t *flags, char **slabs, bool init) | |
41ecc55b | 1283 | { |
e17f1dfb | 1284 | bool higher_order_disable = false; |
f0630fff | 1285 | |
e17f1dfb VB |
1286 | /* Skip any completely empty blocks */ |
1287 | while (*str && *str == ';') | |
1288 | str++; | |
1289 | ||
1290 | if (*str == ',') { | |
f0630fff CL |
1291 | /* |
1292 | * No options but restriction on slabs. This means full | |
1293 | * debugging for slabs matching a pattern. | |
1294 | */ | |
e17f1dfb | 1295 | *flags = DEBUG_DEFAULT_FLAGS; |
f0630fff | 1296 | goto check_slabs; |
e17f1dfb VB |
1297 | } |
1298 | *flags = 0; | |
f0630fff | 1299 | |
e17f1dfb VB |
1300 | /* Determine which debug features should be switched on */ |
1301 | for (; *str && *str != ',' && *str != ';'; str++) { | |
f0630fff | 1302 | switch (tolower(*str)) { |
e17f1dfb VB |
1303 | case '-': |
1304 | *flags = 0; | |
1305 | break; | |
f0630fff | 1306 | case 'f': |
e17f1dfb | 1307 | *flags |= SLAB_CONSISTENCY_CHECKS; |
f0630fff CL |
1308 | break; |
1309 | case 'z': | |
e17f1dfb | 1310 | *flags |= SLAB_RED_ZONE; |
f0630fff CL |
1311 | break; |
1312 | case 'p': | |
e17f1dfb | 1313 | *flags |= SLAB_POISON; |
f0630fff CL |
1314 | break; |
1315 | case 'u': | |
e17f1dfb | 1316 | *flags |= SLAB_STORE_USER; |
f0630fff CL |
1317 | break; |
1318 | case 't': | |
e17f1dfb | 1319 | *flags |= SLAB_TRACE; |
f0630fff | 1320 | break; |
4c13dd3b | 1321 | case 'a': |
e17f1dfb | 1322 | *flags |= SLAB_FAILSLAB; |
4c13dd3b | 1323 | break; |
08303a73 CA |
1324 | case 'o': |
1325 | /* | |
1326 | * Avoid enabling debugging on caches if its minimum | |
1327 | * order would increase as a result. | |
1328 | */ | |
e17f1dfb | 1329 | higher_order_disable = true; |
08303a73 | 1330 | break; |
f0630fff | 1331 | default: |
e17f1dfb VB |
1332 | if (init) |
1333 | pr_err("slub_debug option '%c' unknown. skipped\n", *str); | |
f0630fff | 1334 | } |
41ecc55b | 1335 | } |
f0630fff | 1336 | check_slabs: |
41ecc55b | 1337 | if (*str == ',') |
e17f1dfb VB |
1338 | *slabs = ++str; |
1339 | else | |
1340 | *slabs = NULL; | |
1341 | ||
1342 | /* Skip over the slab list */ | |
1343 | while (*str && *str != ';') | |
1344 | str++; | |
1345 | ||
1346 | /* Skip any completely empty blocks */ | |
1347 | while (*str && *str == ';') | |
1348 | str++; | |
1349 | ||
1350 | if (init && higher_order_disable) | |
1351 | disable_higher_order_debug = 1; | |
1352 | ||
1353 | if (*str) | |
1354 | return str; | |
1355 | else | |
1356 | return NULL; | |
1357 | } | |
1358 | ||
1359 | static int __init setup_slub_debug(char *str) | |
1360 | { | |
1361 | slab_flags_t flags; | |
1362 | char *saved_str; | |
1363 | char *slab_list; | |
1364 | bool global_slub_debug_changed = false; | |
1365 | bool slab_list_specified = false; | |
1366 | ||
1367 | slub_debug = DEBUG_DEFAULT_FLAGS; | |
1368 | if (*str++ != '=' || !*str) | |
1369 | /* | |
1370 | * No options specified. Switch on full debugging. | |
1371 | */ | |
1372 | goto out; | |
1373 | ||
1374 | saved_str = str; | |
1375 | while (str) { | |
1376 | str = parse_slub_debug_flags(str, &flags, &slab_list, true); | |
1377 | ||
1378 | if (!slab_list) { | |
1379 | slub_debug = flags; | |
1380 | global_slub_debug_changed = true; | |
1381 | } else { | |
1382 | slab_list_specified = true; | |
1383 | } | |
1384 | } | |
1385 | ||
1386 | /* | |
1387 | * For backwards compatibility, a single list of flags with list of | |
1388 | * slabs means debugging is only enabled for those slabs, so the global | |
1389 | * slub_debug should be 0. We can extended that to multiple lists as | |
1390 | * long as there is no option specifying flags without a slab list. | |
1391 | */ | |
1392 | if (slab_list_specified) { | |
1393 | if (!global_slub_debug_changed) | |
1394 | slub_debug = 0; | |
1395 | slub_debug_string = saved_str; | |
1396 | } | |
f0630fff | 1397 | out: |
ca0cab65 VB |
1398 | if (slub_debug != 0 || slub_debug_string) |
1399 | static_branch_enable(&slub_debug_enabled); | |
6471384a AP |
1400 | if ((static_branch_unlikely(&init_on_alloc) || |
1401 | static_branch_unlikely(&init_on_free)) && | |
1402 | (slub_debug & SLAB_POISON)) | |
1403 | pr_info("mem auto-init: SLAB_POISON will take precedence over init_on_alloc/init_on_free\n"); | |
41ecc55b CL |
1404 | return 1; |
1405 | } | |
1406 | ||
1407 | __setup("slub_debug", setup_slub_debug); | |
1408 | ||
c5fd3ca0 AT |
1409 | /* |
1410 | * kmem_cache_flags - apply debugging options to the cache | |
1411 | * @object_size: the size of an object without meta data | |
1412 | * @flags: flags to set | |
1413 | * @name: name of the cache | |
c5fd3ca0 AT |
1414 | * |
1415 | * Debug option(s) are applied to @flags. In addition to the debug | |
1416 | * option(s), if a slab name (or multiple) is specified i.e. | |
1417 | * slub_debug=<Debug-Options>,<slab name1>,<slab name2> ... | |
1418 | * then only the select slabs will receive the debug option(s). | |
1419 | */ | |
0293d1fd | 1420 | slab_flags_t kmem_cache_flags(unsigned int object_size, |
37540008 | 1421 | slab_flags_t flags, const char *name) |
41ecc55b | 1422 | { |
c5fd3ca0 AT |
1423 | char *iter; |
1424 | size_t len; | |
e17f1dfb VB |
1425 | char *next_block; |
1426 | slab_flags_t block_flags; | |
ca220593 JB |
1427 | slab_flags_t slub_debug_local = slub_debug; |
1428 | ||
1429 | /* | |
1430 | * If the slab cache is for debugging (e.g. kmemleak) then | |
1431 | * don't store user (stack trace) information by default, | |
1432 | * but let the user enable it via the command line below. | |
1433 | */ | |
1434 | if (flags & SLAB_NOLEAKTRACE) | |
1435 | slub_debug_local &= ~SLAB_STORE_USER; | |
c5fd3ca0 | 1436 | |
c5fd3ca0 | 1437 | len = strlen(name); |
e17f1dfb VB |
1438 | next_block = slub_debug_string; |
1439 | /* Go through all blocks of debug options, see if any matches our slab's name */ | |
1440 | while (next_block) { | |
1441 | next_block = parse_slub_debug_flags(next_block, &block_flags, &iter, false); | |
1442 | if (!iter) | |
1443 | continue; | |
1444 | /* Found a block that has a slab list, search it */ | |
1445 | while (*iter) { | |
1446 | char *end, *glob; | |
1447 | size_t cmplen; | |
1448 | ||
1449 | end = strchrnul(iter, ','); | |
1450 | if (next_block && next_block < end) | |
1451 | end = next_block - 1; | |
1452 | ||
1453 | glob = strnchr(iter, end - iter, '*'); | |
1454 | if (glob) | |
1455 | cmplen = glob - iter; | |
1456 | else | |
1457 | cmplen = max_t(size_t, len, (end - iter)); | |
c5fd3ca0 | 1458 | |
e17f1dfb VB |
1459 | if (!strncmp(name, iter, cmplen)) { |
1460 | flags |= block_flags; | |
1461 | return flags; | |
1462 | } | |
c5fd3ca0 | 1463 | |
e17f1dfb VB |
1464 | if (!*end || *end == ';') |
1465 | break; | |
1466 | iter = end + 1; | |
c5fd3ca0 | 1467 | } |
c5fd3ca0 | 1468 | } |
ba0268a8 | 1469 | |
ca220593 | 1470 | return flags | slub_debug_local; |
41ecc55b | 1471 | } |
b4a64718 | 1472 | #else /* !CONFIG_SLUB_DEBUG */ |
3ec09742 CL |
1473 | static inline void setup_object_debug(struct kmem_cache *s, |
1474 | struct page *page, void *object) {} | |
a50b854e MWO |
1475 | static inline |
1476 | void setup_page_debug(struct kmem_cache *s, struct page *page, void *addr) {} | |
41ecc55b | 1477 | |
3ec09742 | 1478 | static inline int alloc_debug_processing(struct kmem_cache *s, |
ce71e27c | 1479 | struct page *page, void *object, unsigned long addr) { return 0; } |
41ecc55b | 1480 | |
282acb43 | 1481 | static inline int free_debug_processing( |
81084651 JDB |
1482 | struct kmem_cache *s, struct page *page, |
1483 | void *head, void *tail, int bulk_cnt, | |
282acb43 | 1484 | unsigned long addr) { return 0; } |
41ecc55b | 1485 | |
41ecc55b CL |
1486 | static inline int slab_pad_check(struct kmem_cache *s, struct page *page) |
1487 | { return 1; } | |
1488 | static inline int check_object(struct kmem_cache *s, struct page *page, | |
f7cb1933 | 1489 | void *object, u8 val) { return 1; } |
5cc6eee8 CL |
1490 | static inline void add_full(struct kmem_cache *s, struct kmem_cache_node *n, |
1491 | struct page *page) {} | |
c65c1877 PZ |
1492 | static inline void remove_full(struct kmem_cache *s, struct kmem_cache_node *n, |
1493 | struct page *page) {} | |
0293d1fd | 1494 | slab_flags_t kmem_cache_flags(unsigned int object_size, |
37540008 | 1495 | slab_flags_t flags, const char *name) |
ba0268a8 CL |
1496 | { |
1497 | return flags; | |
1498 | } | |
41ecc55b | 1499 | #define slub_debug 0 |
0f389ec6 | 1500 | |
fdaa45e9 IM |
1501 | #define disable_higher_order_debug 0 |
1502 | ||
0f389ec6 CL |
1503 | static inline unsigned long slabs_node(struct kmem_cache *s, int node) |
1504 | { return 0; } | |
26c02cf0 AB |
1505 | static inline unsigned long node_nr_slabs(struct kmem_cache_node *n) |
1506 | { return 0; } | |
205ab99d CL |
1507 | static inline void inc_slabs_node(struct kmem_cache *s, int node, |
1508 | int objects) {} | |
1509 | static inline void dec_slabs_node(struct kmem_cache *s, int node, | |
1510 | int objects) {} | |
7d550c56 | 1511 | |
52f23478 | 1512 | static bool freelist_corrupted(struct kmem_cache *s, struct page *page, |
dc07a728 | 1513 | void **freelist, void *nextfree) |
52f23478 DZ |
1514 | { |
1515 | return false; | |
1516 | } | |
02e72cc6 AR |
1517 | #endif /* CONFIG_SLUB_DEBUG */ |
1518 | ||
1519 | /* | |
1520 | * Hooks for other subsystems that check memory allocations. In a typical | |
1521 | * production configuration these hooks all should produce no code at all. | |
1522 | */ | |
0116523c | 1523 | static inline void *kmalloc_large_node_hook(void *ptr, size_t size, gfp_t flags) |
d56791b3 | 1524 | { |
53128245 | 1525 | ptr = kasan_kmalloc_large(ptr, size, flags); |
a2f77575 | 1526 | /* As ptr might get tagged, call kmemleak hook after KASAN. */ |
d56791b3 | 1527 | kmemleak_alloc(ptr, size, 1, flags); |
53128245 | 1528 | return ptr; |
d56791b3 RB |
1529 | } |
1530 | ||
ee3ce779 | 1531 | static __always_inline void kfree_hook(void *x) |
d56791b3 RB |
1532 | { |
1533 | kmemleak_free(x); | |
027b37b5 | 1534 | kasan_kfree_large(x); |
d56791b3 RB |
1535 | } |
1536 | ||
d57a964e AK |
1537 | static __always_inline bool slab_free_hook(struct kmem_cache *s, |
1538 | void *x, bool init) | |
d56791b3 RB |
1539 | { |
1540 | kmemleak_free_recursive(x, s->flags); | |
7d550c56 | 1541 | |
02e72cc6 AR |
1542 | /* |
1543 | * Trouble is that we may no longer disable interrupts in the fast path | |
1544 | * So in order to make the debug calls that expect irqs to be | |
1545 | * disabled we need to disable interrupts temporarily. | |
1546 | */ | |
4675ff05 | 1547 | #ifdef CONFIG_LOCKDEP |
02e72cc6 AR |
1548 | { |
1549 | unsigned long flags; | |
1550 | ||
1551 | local_irq_save(flags); | |
02e72cc6 AR |
1552 | debug_check_no_locks_freed(x, s->object_size); |
1553 | local_irq_restore(flags); | |
1554 | } | |
1555 | #endif | |
1556 | if (!(s->flags & SLAB_DEBUG_OBJECTS)) | |
1557 | debug_check_no_obj_freed(x, s->object_size); | |
0316bec2 | 1558 | |
cfbe1636 ME |
1559 | /* Use KCSAN to help debug racy use-after-free. */ |
1560 | if (!(s->flags & SLAB_TYPESAFE_BY_RCU)) | |
1561 | __kcsan_check_access(x, s->object_size, | |
1562 | KCSAN_ACCESS_WRITE | KCSAN_ACCESS_ASSERT); | |
1563 | ||
d57a964e AK |
1564 | /* |
1565 | * As memory initialization might be integrated into KASAN, | |
1566 | * kasan_slab_free and initialization memset's must be | |
1567 | * kept together to avoid discrepancies in behavior. | |
1568 | * | |
1569 | * The initialization memset's clear the object and the metadata, | |
1570 | * but don't touch the SLAB redzone. | |
1571 | */ | |
1572 | if (init) { | |
1573 | int rsize; | |
1574 | ||
1575 | if (!kasan_has_integrated_init()) | |
1576 | memset(kasan_reset_tag(x), 0, s->object_size); | |
1577 | rsize = (s->flags & SLAB_RED_ZONE) ? s->red_left_pad : 0; | |
1578 | memset((char *)kasan_reset_tag(x) + s->inuse, 0, | |
1579 | s->size - s->inuse - rsize); | |
1580 | } | |
1581 | /* KASAN might put x into memory quarantine, delaying its reuse. */ | |
1582 | return kasan_slab_free(s, x, init); | |
02e72cc6 | 1583 | } |
205ab99d | 1584 | |
c3895391 AK |
1585 | static inline bool slab_free_freelist_hook(struct kmem_cache *s, |
1586 | void **head, void **tail) | |
81084651 | 1587 | { |
6471384a AP |
1588 | |
1589 | void *object; | |
1590 | void *next = *head; | |
1591 | void *old_tail = *tail ? *tail : *head; | |
6471384a | 1592 | |
b89fb5ef | 1593 | if (is_kfence_address(next)) { |
d57a964e | 1594 | slab_free_hook(s, next, false); |
b89fb5ef AP |
1595 | return true; |
1596 | } | |
1597 | ||
aea4df4c LA |
1598 | /* Head and tail of the reconstructed freelist */ |
1599 | *head = NULL; | |
1600 | *tail = NULL; | |
1b7e816f | 1601 | |
aea4df4c LA |
1602 | do { |
1603 | object = next; | |
1604 | next = get_freepointer(s, object); | |
1605 | ||
c3895391 | 1606 | /* If object's reuse doesn't have to be delayed */ |
d57a964e | 1607 | if (!slab_free_hook(s, object, slab_want_init_on_free(s))) { |
c3895391 AK |
1608 | /* Move object to the new freelist */ |
1609 | set_freepointer(s, object, *head); | |
1610 | *head = object; | |
1611 | if (!*tail) | |
1612 | *tail = object; | |
1613 | } | |
1614 | } while (object != old_tail); | |
1615 | ||
1616 | if (*head == *tail) | |
1617 | *tail = NULL; | |
1618 | ||
1619 | return *head != NULL; | |
81084651 JDB |
1620 | } |
1621 | ||
4d176711 | 1622 | static void *setup_object(struct kmem_cache *s, struct page *page, |
588f8ba9 TG |
1623 | void *object) |
1624 | { | |
1625 | setup_object_debug(s, page, object); | |
4d176711 | 1626 | object = kasan_init_slab_obj(s, object); |
588f8ba9 TG |
1627 | if (unlikely(s->ctor)) { |
1628 | kasan_unpoison_object_data(s, object); | |
1629 | s->ctor(object); | |
1630 | kasan_poison_object_data(s, object); | |
1631 | } | |
4d176711 | 1632 | return object; |
588f8ba9 TG |
1633 | } |
1634 | ||
81819f0f CL |
1635 | /* |
1636 | * Slab allocation and freeing | |
1637 | */ | |
5dfb4175 VD |
1638 | static inline struct page *alloc_slab_page(struct kmem_cache *s, |
1639 | gfp_t flags, int node, struct kmem_cache_order_objects oo) | |
65c3376a | 1640 | { |
5dfb4175 | 1641 | struct page *page; |
19af27af | 1642 | unsigned int order = oo_order(oo); |
65c3376a | 1643 | |
2154a336 | 1644 | if (node == NUMA_NO_NODE) |
5dfb4175 | 1645 | page = alloc_pages(flags, order); |
65c3376a | 1646 | else |
96db800f | 1647 | page = __alloc_pages_node(node, flags, order); |
5dfb4175 | 1648 | |
5dfb4175 | 1649 | return page; |
65c3376a CL |
1650 | } |
1651 | ||
210e7a43 TG |
1652 | #ifdef CONFIG_SLAB_FREELIST_RANDOM |
1653 | /* Pre-initialize the random sequence cache */ | |
1654 | static int init_cache_random_seq(struct kmem_cache *s) | |
1655 | { | |
19af27af | 1656 | unsigned int count = oo_objects(s->oo); |
210e7a43 | 1657 | int err; |
210e7a43 | 1658 | |
a810007a SR |
1659 | /* Bailout if already initialised */ |
1660 | if (s->random_seq) | |
1661 | return 0; | |
1662 | ||
210e7a43 TG |
1663 | err = cache_random_seq_create(s, count, GFP_KERNEL); |
1664 | if (err) { | |
1665 | pr_err("SLUB: Unable to initialize free list for %s\n", | |
1666 | s->name); | |
1667 | return err; | |
1668 | } | |
1669 | ||
1670 | /* Transform to an offset on the set of pages */ | |
1671 | if (s->random_seq) { | |
19af27af AD |
1672 | unsigned int i; |
1673 | ||
210e7a43 TG |
1674 | for (i = 0; i < count; i++) |
1675 | s->random_seq[i] *= s->size; | |
1676 | } | |
1677 | return 0; | |
1678 | } | |
1679 | ||
1680 | /* Initialize each random sequence freelist per cache */ | |
1681 | static void __init init_freelist_randomization(void) | |
1682 | { | |
1683 | struct kmem_cache *s; | |
1684 | ||
1685 | mutex_lock(&slab_mutex); | |
1686 | ||
1687 | list_for_each_entry(s, &slab_caches, list) | |
1688 | init_cache_random_seq(s); | |
1689 | ||
1690 | mutex_unlock(&slab_mutex); | |
1691 | } | |
1692 | ||
1693 | /* Get the next entry on the pre-computed freelist randomized */ | |
1694 | static void *next_freelist_entry(struct kmem_cache *s, struct page *page, | |
1695 | unsigned long *pos, void *start, | |
1696 | unsigned long page_limit, | |
1697 | unsigned long freelist_count) | |
1698 | { | |
1699 | unsigned int idx; | |
1700 | ||
1701 | /* | |
1702 | * If the target page allocation failed, the number of objects on the | |
1703 | * page might be smaller than the usual size defined by the cache. | |
1704 | */ | |
1705 | do { | |
1706 | idx = s->random_seq[*pos]; | |
1707 | *pos += 1; | |
1708 | if (*pos >= freelist_count) | |
1709 | *pos = 0; | |
1710 | } while (unlikely(idx >= page_limit)); | |
1711 | ||
1712 | return (char *)start + idx; | |
1713 | } | |
1714 | ||
1715 | /* Shuffle the single linked freelist based on a random pre-computed sequence */ | |
1716 | static bool shuffle_freelist(struct kmem_cache *s, struct page *page) | |
1717 | { | |
1718 | void *start; | |
1719 | void *cur; | |
1720 | void *next; | |
1721 | unsigned long idx, pos, page_limit, freelist_count; | |
1722 | ||
1723 | if (page->objects < 2 || !s->random_seq) | |
1724 | return false; | |
1725 | ||
1726 | freelist_count = oo_objects(s->oo); | |
1727 | pos = get_random_int() % freelist_count; | |
1728 | ||
1729 | page_limit = page->objects * s->size; | |
1730 | start = fixup_red_left(s, page_address(page)); | |
1731 | ||
1732 | /* First entry is used as the base of the freelist */ | |
1733 | cur = next_freelist_entry(s, page, &pos, start, page_limit, | |
1734 | freelist_count); | |
4d176711 | 1735 | cur = setup_object(s, page, cur); |
210e7a43 TG |
1736 | page->freelist = cur; |
1737 | ||
1738 | for (idx = 1; idx < page->objects; idx++) { | |
210e7a43 TG |
1739 | next = next_freelist_entry(s, page, &pos, start, page_limit, |
1740 | freelist_count); | |
4d176711 | 1741 | next = setup_object(s, page, next); |
210e7a43 TG |
1742 | set_freepointer(s, cur, next); |
1743 | cur = next; | |
1744 | } | |
210e7a43 TG |
1745 | set_freepointer(s, cur, NULL); |
1746 | ||
1747 | return true; | |
1748 | } | |
1749 | #else | |
1750 | static inline int init_cache_random_seq(struct kmem_cache *s) | |
1751 | { | |
1752 | return 0; | |
1753 | } | |
1754 | static inline void init_freelist_randomization(void) { } | |
1755 | static inline bool shuffle_freelist(struct kmem_cache *s, struct page *page) | |
1756 | { | |
1757 | return false; | |
1758 | } | |
1759 | #endif /* CONFIG_SLAB_FREELIST_RANDOM */ | |
1760 | ||
81819f0f CL |
1761 | static struct page *allocate_slab(struct kmem_cache *s, gfp_t flags, int node) |
1762 | { | |
06428780 | 1763 | struct page *page; |
834f3d11 | 1764 | struct kmem_cache_order_objects oo = s->oo; |
ba52270d | 1765 | gfp_t alloc_gfp; |
4d176711 | 1766 | void *start, *p, *next; |
a50b854e | 1767 | int idx; |
210e7a43 | 1768 | bool shuffle; |
81819f0f | 1769 | |
7e0528da CL |
1770 | flags &= gfp_allowed_mask; |
1771 | ||
d0164adc | 1772 | if (gfpflags_allow_blocking(flags)) |
7e0528da CL |
1773 | local_irq_enable(); |
1774 | ||
b7a49f0d | 1775 | flags |= s->allocflags; |
e12ba74d | 1776 | |
ba52270d PE |
1777 | /* |
1778 | * Let the initial higher-order allocation fail under memory pressure | |
1779 | * so we fall-back to the minimum order allocation. | |
1780 | */ | |
1781 | alloc_gfp = (flags | __GFP_NOWARN | __GFP_NORETRY) & ~__GFP_NOFAIL; | |
d0164adc | 1782 | if ((alloc_gfp & __GFP_DIRECT_RECLAIM) && oo_order(oo) > oo_order(s->min)) |
444eb2a4 | 1783 | alloc_gfp = (alloc_gfp | __GFP_NOMEMALLOC) & ~(__GFP_RECLAIM|__GFP_NOFAIL); |
ba52270d | 1784 | |
5dfb4175 | 1785 | page = alloc_slab_page(s, alloc_gfp, node, oo); |
65c3376a CL |
1786 | if (unlikely(!page)) { |
1787 | oo = s->min; | |
80c3a998 | 1788 | alloc_gfp = flags; |
65c3376a CL |
1789 | /* |
1790 | * Allocation may have failed due to fragmentation. | |
1791 | * Try a lower order alloc if possible | |
1792 | */ | |
5dfb4175 | 1793 | page = alloc_slab_page(s, alloc_gfp, node, oo); |
588f8ba9 TG |
1794 | if (unlikely(!page)) |
1795 | goto out; | |
1796 | stat(s, ORDER_FALLBACK); | |
65c3376a | 1797 | } |
5a896d9e | 1798 | |
834f3d11 | 1799 | page->objects = oo_objects(oo); |
81819f0f | 1800 | |
2e9bd483 | 1801 | account_slab_page(page, oo_order(oo), s, flags); |
1f3147b4 | 1802 | |
1b4f59e3 | 1803 | page->slab_cache = s; |
c03f94cc | 1804 | __SetPageSlab(page); |
2f064f34 | 1805 | if (page_is_pfmemalloc(page)) |
072bb0aa | 1806 | SetPageSlabPfmemalloc(page); |
81819f0f | 1807 | |
a7101224 | 1808 | kasan_poison_slab(page); |
81819f0f | 1809 | |
a7101224 | 1810 | start = page_address(page); |
81819f0f | 1811 | |
a50b854e | 1812 | setup_page_debug(s, page, start); |
0316bec2 | 1813 | |
210e7a43 TG |
1814 | shuffle = shuffle_freelist(s, page); |
1815 | ||
1816 | if (!shuffle) { | |
4d176711 AK |
1817 | start = fixup_red_left(s, start); |
1818 | start = setup_object(s, page, start); | |
1819 | page->freelist = start; | |
18e50661 AK |
1820 | for (idx = 0, p = start; idx < page->objects - 1; idx++) { |
1821 | next = p + s->size; | |
1822 | next = setup_object(s, page, next); | |
1823 | set_freepointer(s, p, next); | |
1824 | p = next; | |
1825 | } | |
1826 | set_freepointer(s, p, NULL); | |
81819f0f | 1827 | } |
81819f0f | 1828 | |
e6e82ea1 | 1829 | page->inuse = page->objects; |
8cb0a506 | 1830 | page->frozen = 1; |
588f8ba9 | 1831 | |
81819f0f | 1832 | out: |
d0164adc | 1833 | if (gfpflags_allow_blocking(flags)) |
588f8ba9 TG |
1834 | local_irq_disable(); |
1835 | if (!page) | |
1836 | return NULL; | |
1837 | ||
588f8ba9 TG |
1838 | inc_slabs_node(s, page_to_nid(page), page->objects); |
1839 | ||
81819f0f CL |
1840 | return page; |
1841 | } | |
1842 | ||
588f8ba9 TG |
1843 | static struct page *new_slab(struct kmem_cache *s, gfp_t flags, int node) |
1844 | { | |
44405099 LL |
1845 | if (unlikely(flags & GFP_SLAB_BUG_MASK)) |
1846 | flags = kmalloc_fix_flags(flags); | |
588f8ba9 TG |
1847 | |
1848 | return allocate_slab(s, | |
1849 | flags & (GFP_RECLAIM_MASK | GFP_CONSTRAINT_MASK), node); | |
1850 | } | |
1851 | ||
81819f0f CL |
1852 | static void __free_slab(struct kmem_cache *s, struct page *page) |
1853 | { | |
834f3d11 CL |
1854 | int order = compound_order(page); |
1855 | int pages = 1 << order; | |
81819f0f | 1856 | |
8fc8d666 | 1857 | if (kmem_cache_debug_flags(s, SLAB_CONSISTENCY_CHECKS)) { |
81819f0f CL |
1858 | void *p; |
1859 | ||
1860 | slab_pad_check(s, page); | |
224a88be CL |
1861 | for_each_object(p, s, page_address(page), |
1862 | page->objects) | |
f7cb1933 | 1863 | check_object(s, page, p, SLUB_RED_INACTIVE); |
81819f0f CL |
1864 | } |
1865 | ||
072bb0aa | 1866 | __ClearPageSlabPfmemalloc(page); |
49bd5221 | 1867 | __ClearPageSlab(page); |
0c06dd75 VB |
1868 | /* In union with page->mapping where page allocator expects NULL */ |
1869 | page->slab_cache = NULL; | |
1eb5ac64 NP |
1870 | if (current->reclaim_state) |
1871 | current->reclaim_state->reclaimed_slab += pages; | |
74d555be | 1872 | unaccount_slab_page(page, order, s); |
27ee57c9 | 1873 | __free_pages(page, order); |
81819f0f CL |
1874 | } |
1875 | ||
1876 | static void rcu_free_slab(struct rcu_head *h) | |
1877 | { | |
bf68c214 | 1878 | struct page *page = container_of(h, struct page, rcu_head); |
da9a638c | 1879 | |
1b4f59e3 | 1880 | __free_slab(page->slab_cache, page); |
81819f0f CL |
1881 | } |
1882 | ||
1883 | static void free_slab(struct kmem_cache *s, struct page *page) | |
1884 | { | |
5f0d5a3a | 1885 | if (unlikely(s->flags & SLAB_TYPESAFE_BY_RCU)) { |
bf68c214 | 1886 | call_rcu(&page->rcu_head, rcu_free_slab); |
81819f0f CL |
1887 | } else |
1888 | __free_slab(s, page); | |
1889 | } | |
1890 | ||
1891 | static void discard_slab(struct kmem_cache *s, struct page *page) | |
1892 | { | |
205ab99d | 1893 | dec_slabs_node(s, page_to_nid(page), page->objects); |
81819f0f CL |
1894 | free_slab(s, page); |
1895 | } | |
1896 | ||
1897 | /* | |
5cc6eee8 | 1898 | * Management of partially allocated slabs. |
81819f0f | 1899 | */ |
1e4dd946 SR |
1900 | static inline void |
1901 | __add_partial(struct kmem_cache_node *n, struct page *page, int tail) | |
81819f0f | 1902 | { |
e95eed57 | 1903 | n->nr_partial++; |
136333d1 | 1904 | if (tail == DEACTIVATE_TO_TAIL) |
916ac052 | 1905 | list_add_tail(&page->slab_list, &n->partial); |
7c2e132c | 1906 | else |
916ac052 | 1907 | list_add(&page->slab_list, &n->partial); |
81819f0f CL |
1908 | } |
1909 | ||
1e4dd946 SR |
1910 | static inline void add_partial(struct kmem_cache_node *n, |
1911 | struct page *page, int tail) | |
62e346a8 | 1912 | { |
c65c1877 | 1913 | lockdep_assert_held(&n->list_lock); |
1e4dd946 SR |
1914 | __add_partial(n, page, tail); |
1915 | } | |
c65c1877 | 1916 | |
1e4dd946 SR |
1917 | static inline void remove_partial(struct kmem_cache_node *n, |
1918 | struct page *page) | |
1919 | { | |
1920 | lockdep_assert_held(&n->list_lock); | |
916ac052 | 1921 | list_del(&page->slab_list); |
52b4b950 | 1922 | n->nr_partial--; |
1e4dd946 SR |
1923 | } |
1924 | ||
81819f0f | 1925 | /* |
7ced3719 CL |
1926 | * Remove slab from the partial list, freeze it and |
1927 | * return the pointer to the freelist. | |
81819f0f | 1928 | * |
497b66f2 | 1929 | * Returns a list of objects or NULL if it fails. |
81819f0f | 1930 | */ |
497b66f2 | 1931 | static inline void *acquire_slab(struct kmem_cache *s, |
acd19fd1 | 1932 | struct kmem_cache_node *n, struct page *page, |
633b0764 | 1933 | int mode, int *objects) |
81819f0f | 1934 | { |
2cfb7455 CL |
1935 | void *freelist; |
1936 | unsigned long counters; | |
1937 | struct page new; | |
1938 | ||
c65c1877 PZ |
1939 | lockdep_assert_held(&n->list_lock); |
1940 | ||
2cfb7455 CL |
1941 | /* |
1942 | * Zap the freelist and set the frozen bit. | |
1943 | * The old freelist is the list of objects for the | |
1944 | * per cpu allocation list. | |
1945 | */ | |
7ced3719 CL |
1946 | freelist = page->freelist; |
1947 | counters = page->counters; | |
1948 | new.counters = counters; | |
633b0764 | 1949 | *objects = new.objects - new.inuse; |
23910c50 | 1950 | if (mode) { |
7ced3719 | 1951 | new.inuse = page->objects; |
23910c50 PE |
1952 | new.freelist = NULL; |
1953 | } else { | |
1954 | new.freelist = freelist; | |
1955 | } | |
2cfb7455 | 1956 | |
a0132ac0 | 1957 | VM_BUG_ON(new.frozen); |
7ced3719 | 1958 | new.frozen = 1; |
2cfb7455 | 1959 | |
7ced3719 | 1960 | if (!__cmpxchg_double_slab(s, page, |
2cfb7455 | 1961 | freelist, counters, |
02d7633f | 1962 | new.freelist, new.counters, |
7ced3719 | 1963 | "acquire_slab")) |
7ced3719 | 1964 | return NULL; |
2cfb7455 CL |
1965 | |
1966 | remove_partial(n, page); | |
7ced3719 | 1967 | WARN_ON(!freelist); |
49e22585 | 1968 | return freelist; |
81819f0f CL |
1969 | } |
1970 | ||
633b0764 | 1971 | static void put_cpu_partial(struct kmem_cache *s, struct page *page, int drain); |
8ba00bb6 | 1972 | static inline bool pfmemalloc_match(struct page *page, gfp_t gfpflags); |
49e22585 | 1973 | |
81819f0f | 1974 | /* |
672bba3a | 1975 | * Try to allocate a partial slab from a specific node. |
81819f0f | 1976 | */ |
8ba00bb6 JK |
1977 | static void *get_partial_node(struct kmem_cache *s, struct kmem_cache_node *n, |
1978 | struct kmem_cache_cpu *c, gfp_t flags) | |
81819f0f | 1979 | { |
49e22585 CL |
1980 | struct page *page, *page2; |
1981 | void *object = NULL; | |
e5d9998f | 1982 | unsigned int available = 0; |
633b0764 | 1983 | int objects; |
81819f0f CL |
1984 | |
1985 | /* | |
1986 | * Racy check. If we mistakenly see no partial slabs then we | |
1987 | * just allocate an empty slab. If we mistakenly try to get a | |
70b6d25e | 1988 | * partial slab and there is none available then get_partial() |
672bba3a | 1989 | * will return NULL. |
81819f0f CL |
1990 | */ |
1991 | if (!n || !n->nr_partial) | |
1992 | return NULL; | |
1993 | ||
1994 | spin_lock(&n->list_lock); | |
916ac052 | 1995 | list_for_each_entry_safe(page, page2, &n->partial, slab_list) { |
8ba00bb6 | 1996 | void *t; |
49e22585 | 1997 | |
8ba00bb6 JK |
1998 | if (!pfmemalloc_match(page, flags)) |
1999 | continue; | |
2000 | ||
633b0764 | 2001 | t = acquire_slab(s, n, page, object == NULL, &objects); |
49e22585 | 2002 | if (!t) |
9b1ea29b | 2003 | break; |
49e22585 | 2004 | |
633b0764 | 2005 | available += objects; |
12d79634 | 2006 | if (!object) { |
49e22585 | 2007 | c->page = page; |
49e22585 | 2008 | stat(s, ALLOC_FROM_PARTIAL); |
49e22585 | 2009 | object = t; |
49e22585 | 2010 | } else { |
633b0764 | 2011 | put_cpu_partial(s, page, 0); |
8028dcea | 2012 | stat(s, CPU_PARTIAL_NODE); |
49e22585 | 2013 | } |
345c905d | 2014 | if (!kmem_cache_has_cpu_partial(s) |
e6d0e1dc | 2015 | || available > slub_cpu_partial(s) / 2) |
49e22585 CL |
2016 | break; |
2017 | ||
497b66f2 | 2018 | } |
81819f0f | 2019 | spin_unlock(&n->list_lock); |
497b66f2 | 2020 | return object; |
81819f0f CL |
2021 | } |
2022 | ||
2023 | /* | |
672bba3a | 2024 | * Get a page from somewhere. Search in increasing NUMA distances. |
81819f0f | 2025 | */ |
de3ec035 | 2026 | static void *get_any_partial(struct kmem_cache *s, gfp_t flags, |
acd19fd1 | 2027 | struct kmem_cache_cpu *c) |
81819f0f CL |
2028 | { |
2029 | #ifdef CONFIG_NUMA | |
2030 | struct zonelist *zonelist; | |
dd1a239f | 2031 | struct zoneref *z; |
54a6eb5c | 2032 | struct zone *zone; |
97a225e6 | 2033 | enum zone_type highest_zoneidx = gfp_zone(flags); |
497b66f2 | 2034 | void *object; |
cc9a6c87 | 2035 | unsigned int cpuset_mems_cookie; |
81819f0f CL |
2036 | |
2037 | /* | |
672bba3a CL |
2038 | * The defrag ratio allows a configuration of the tradeoffs between |
2039 | * inter node defragmentation and node local allocations. A lower | |
2040 | * defrag_ratio increases the tendency to do local allocations | |
2041 | * instead of attempting to obtain partial slabs from other nodes. | |
81819f0f | 2042 | * |
672bba3a CL |
2043 | * If the defrag_ratio is set to 0 then kmalloc() always |
2044 | * returns node local objects. If the ratio is higher then kmalloc() | |
2045 | * may return off node objects because partial slabs are obtained | |
2046 | * from other nodes and filled up. | |
81819f0f | 2047 | * |
43efd3ea LP |
2048 | * If /sys/kernel/slab/xx/remote_node_defrag_ratio is set to 100 |
2049 | * (which makes defrag_ratio = 1000) then every (well almost) | |
2050 | * allocation will first attempt to defrag slab caches on other nodes. | |
2051 | * This means scanning over all nodes to look for partial slabs which | |
2052 | * may be expensive if we do it every time we are trying to find a slab | |
672bba3a | 2053 | * with available objects. |
81819f0f | 2054 | */ |
9824601e CL |
2055 | if (!s->remote_node_defrag_ratio || |
2056 | get_cycles() % 1024 > s->remote_node_defrag_ratio) | |
81819f0f CL |
2057 | return NULL; |
2058 | ||
cc9a6c87 | 2059 | do { |
d26914d1 | 2060 | cpuset_mems_cookie = read_mems_allowed_begin(); |
2a389610 | 2061 | zonelist = node_zonelist(mempolicy_slab_node(), flags); |
97a225e6 | 2062 | for_each_zone_zonelist(zone, z, zonelist, highest_zoneidx) { |
cc9a6c87 MG |
2063 | struct kmem_cache_node *n; |
2064 | ||
2065 | n = get_node(s, zone_to_nid(zone)); | |
2066 | ||
dee2f8aa | 2067 | if (n && cpuset_zone_allowed(zone, flags) && |
cc9a6c87 | 2068 | n->nr_partial > s->min_partial) { |
8ba00bb6 | 2069 | object = get_partial_node(s, n, c, flags); |
cc9a6c87 MG |
2070 | if (object) { |
2071 | /* | |
d26914d1 MG |
2072 | * Don't check read_mems_allowed_retry() |
2073 | * here - if mems_allowed was updated in | |
2074 | * parallel, that was a harmless race | |
2075 | * between allocation and the cpuset | |
2076 | * update | |
cc9a6c87 | 2077 | */ |
cc9a6c87 MG |
2078 | return object; |
2079 | } | |
c0ff7453 | 2080 | } |
81819f0f | 2081 | } |
d26914d1 | 2082 | } while (read_mems_allowed_retry(cpuset_mems_cookie)); |
6dfd1b65 | 2083 | #endif /* CONFIG_NUMA */ |
81819f0f CL |
2084 | return NULL; |
2085 | } | |
2086 | ||
2087 | /* | |
2088 | * Get a partial page, lock it and return it. | |
2089 | */ | |
497b66f2 | 2090 | static void *get_partial(struct kmem_cache *s, gfp_t flags, int node, |
acd19fd1 | 2091 | struct kmem_cache_cpu *c) |
81819f0f | 2092 | { |
497b66f2 | 2093 | void *object; |
a561ce00 JK |
2094 | int searchnode = node; |
2095 | ||
2096 | if (node == NUMA_NO_NODE) | |
2097 | searchnode = numa_mem_id(); | |
81819f0f | 2098 | |
8ba00bb6 | 2099 | object = get_partial_node(s, get_node(s, searchnode), c, flags); |
497b66f2 CL |
2100 | if (object || node != NUMA_NO_NODE) |
2101 | return object; | |
81819f0f | 2102 | |
acd19fd1 | 2103 | return get_any_partial(s, flags, c); |
81819f0f CL |
2104 | } |
2105 | ||
923717cb | 2106 | #ifdef CONFIG_PREEMPTION |
8a5ec0ba | 2107 | /* |
0d645ed1 | 2108 | * Calculate the next globally unique transaction for disambiguation |
8a5ec0ba CL |
2109 | * during cmpxchg. The transactions start with the cpu number and are then |
2110 | * incremented by CONFIG_NR_CPUS. | |
2111 | */ | |
2112 | #define TID_STEP roundup_pow_of_two(CONFIG_NR_CPUS) | |
2113 | #else | |
2114 | /* | |
2115 | * No preemption supported therefore also no need to check for | |
2116 | * different cpus. | |
2117 | */ | |
2118 | #define TID_STEP 1 | |
2119 | #endif | |
2120 | ||
2121 | static inline unsigned long next_tid(unsigned long tid) | |
2122 | { | |
2123 | return tid + TID_STEP; | |
2124 | } | |
2125 | ||
9d5f0be0 | 2126 | #ifdef SLUB_DEBUG_CMPXCHG |
8a5ec0ba CL |
2127 | static inline unsigned int tid_to_cpu(unsigned long tid) |
2128 | { | |
2129 | return tid % TID_STEP; | |
2130 | } | |
2131 | ||
2132 | static inline unsigned long tid_to_event(unsigned long tid) | |
2133 | { | |
2134 | return tid / TID_STEP; | |
2135 | } | |
9d5f0be0 | 2136 | #endif |
8a5ec0ba CL |
2137 | |
2138 | static inline unsigned int init_tid(int cpu) | |
2139 | { | |
2140 | return cpu; | |
2141 | } | |
2142 | ||
2143 | static inline void note_cmpxchg_failure(const char *n, | |
2144 | const struct kmem_cache *s, unsigned long tid) | |
2145 | { | |
2146 | #ifdef SLUB_DEBUG_CMPXCHG | |
2147 | unsigned long actual_tid = __this_cpu_read(s->cpu_slab->tid); | |
2148 | ||
f9f58285 | 2149 | pr_info("%s %s: cmpxchg redo ", n, s->name); |
8a5ec0ba | 2150 | |
923717cb | 2151 | #ifdef CONFIG_PREEMPTION |
8a5ec0ba | 2152 | if (tid_to_cpu(tid) != tid_to_cpu(actual_tid)) |
f9f58285 | 2153 | pr_warn("due to cpu change %d -> %d\n", |
8a5ec0ba CL |
2154 | tid_to_cpu(tid), tid_to_cpu(actual_tid)); |
2155 | else | |
2156 | #endif | |
2157 | if (tid_to_event(tid) != tid_to_event(actual_tid)) | |
f9f58285 | 2158 | pr_warn("due to cpu running other code. Event %ld->%ld\n", |
8a5ec0ba CL |
2159 | tid_to_event(tid), tid_to_event(actual_tid)); |
2160 | else | |
f9f58285 | 2161 | pr_warn("for unknown reason: actual=%lx was=%lx target=%lx\n", |
8a5ec0ba CL |
2162 | actual_tid, tid, next_tid(tid)); |
2163 | #endif | |
4fdccdfb | 2164 | stat(s, CMPXCHG_DOUBLE_CPU_FAIL); |
8a5ec0ba CL |
2165 | } |
2166 | ||
788e1aad | 2167 | static void init_kmem_cache_cpus(struct kmem_cache *s) |
8a5ec0ba | 2168 | { |
8a5ec0ba CL |
2169 | int cpu; |
2170 | ||
2171 | for_each_possible_cpu(cpu) | |
2172 | per_cpu_ptr(s->cpu_slab, cpu)->tid = init_tid(cpu); | |
8a5ec0ba | 2173 | } |
2cfb7455 | 2174 | |
81819f0f CL |
2175 | /* |
2176 | * Remove the cpu slab | |
2177 | */ | |
d0e0ac97 | 2178 | static void deactivate_slab(struct kmem_cache *s, struct page *page, |
d4ff6d35 | 2179 | void *freelist, struct kmem_cache_cpu *c) |
81819f0f | 2180 | { |
2cfb7455 | 2181 | enum slab_modes { M_NONE, M_PARTIAL, M_FULL, M_FREE }; |
2cfb7455 | 2182 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); |
d930ff03 | 2183 | int lock = 0, free_delta = 0; |
2cfb7455 | 2184 | enum slab_modes l = M_NONE, m = M_NONE; |
d930ff03 | 2185 | void *nextfree, *freelist_iter, *freelist_tail; |
136333d1 | 2186 | int tail = DEACTIVATE_TO_HEAD; |
2cfb7455 CL |
2187 | struct page new; |
2188 | struct page old; | |
2189 | ||
2190 | if (page->freelist) { | |
84e554e6 | 2191 | stat(s, DEACTIVATE_REMOTE_FREES); |
136333d1 | 2192 | tail = DEACTIVATE_TO_TAIL; |
2cfb7455 CL |
2193 | } |
2194 | ||
894b8788 | 2195 | /* |
d930ff03 VB |
2196 | * Stage one: Count the objects on cpu's freelist as free_delta and |
2197 | * remember the last object in freelist_tail for later splicing. | |
2cfb7455 | 2198 | */ |
d930ff03 VB |
2199 | freelist_tail = NULL; |
2200 | freelist_iter = freelist; | |
2201 | while (freelist_iter) { | |
2202 | nextfree = get_freepointer(s, freelist_iter); | |
2cfb7455 | 2203 | |
52f23478 DZ |
2204 | /* |
2205 | * If 'nextfree' is invalid, it is possible that the object at | |
d930ff03 VB |
2206 | * 'freelist_iter' is already corrupted. So isolate all objects |
2207 | * starting at 'freelist_iter' by skipping them. | |
52f23478 | 2208 | */ |
d930ff03 | 2209 | if (freelist_corrupted(s, page, &freelist_iter, nextfree)) |
52f23478 DZ |
2210 | break; |
2211 | ||
d930ff03 VB |
2212 | freelist_tail = freelist_iter; |
2213 | free_delta++; | |
2cfb7455 | 2214 | |
d930ff03 | 2215 | freelist_iter = nextfree; |
2cfb7455 CL |
2216 | } |
2217 | ||
894b8788 | 2218 | /* |
d930ff03 VB |
2219 | * Stage two: Unfreeze the page while splicing the per-cpu |
2220 | * freelist to the head of page's freelist. | |
2221 | * | |
2222 | * Ensure that the page is unfrozen while the list presence | |
2223 | * reflects the actual number of objects during unfreeze. | |
2cfb7455 CL |
2224 | * |
2225 | * We setup the list membership and then perform a cmpxchg | |
2226 | * with the count. If there is a mismatch then the page | |
2227 | * is not unfrozen but the page is on the wrong list. | |
2228 | * | |
2229 | * Then we restart the process which may have to remove | |
2230 | * the page from the list that we just put it on again | |
2231 | * because the number of objects in the slab may have | |
2232 | * changed. | |
894b8788 | 2233 | */ |
2cfb7455 | 2234 | redo: |
894b8788 | 2235 | |
d930ff03 VB |
2236 | old.freelist = READ_ONCE(page->freelist); |
2237 | old.counters = READ_ONCE(page->counters); | |
a0132ac0 | 2238 | VM_BUG_ON(!old.frozen); |
7c2e132c | 2239 | |
2cfb7455 CL |
2240 | /* Determine target state of the slab */ |
2241 | new.counters = old.counters; | |
d930ff03 VB |
2242 | if (freelist_tail) { |
2243 | new.inuse -= free_delta; | |
2244 | set_freepointer(s, freelist_tail, old.freelist); | |
2cfb7455 CL |
2245 | new.freelist = freelist; |
2246 | } else | |
2247 | new.freelist = old.freelist; | |
2248 | ||
2249 | new.frozen = 0; | |
2250 | ||
8a5b20ae | 2251 | if (!new.inuse && n->nr_partial >= s->min_partial) |
2cfb7455 CL |
2252 | m = M_FREE; |
2253 | else if (new.freelist) { | |
2254 | m = M_PARTIAL; | |
2255 | if (!lock) { | |
2256 | lock = 1; | |
2257 | /* | |
8bb4e7a2 | 2258 | * Taking the spinlock removes the possibility |
2cfb7455 CL |
2259 | * that acquire_slab() will see a slab page that |
2260 | * is frozen | |
2261 | */ | |
2262 | spin_lock(&n->list_lock); | |
2263 | } | |
2264 | } else { | |
2265 | m = M_FULL; | |
965c4848 | 2266 | if (kmem_cache_debug_flags(s, SLAB_STORE_USER) && !lock) { |
2cfb7455 CL |
2267 | lock = 1; |
2268 | /* | |
2269 | * This also ensures that the scanning of full | |
2270 | * slabs from diagnostic functions will not see | |
2271 | * any frozen slabs. | |
2272 | */ | |
2273 | spin_lock(&n->list_lock); | |
2274 | } | |
2275 | } | |
2276 | ||
2277 | if (l != m) { | |
2cfb7455 | 2278 | if (l == M_PARTIAL) |
2cfb7455 | 2279 | remove_partial(n, page); |
2cfb7455 | 2280 | else if (l == M_FULL) |
c65c1877 | 2281 | remove_full(s, n, page); |
2cfb7455 | 2282 | |
88349a28 | 2283 | if (m == M_PARTIAL) |
2cfb7455 | 2284 | add_partial(n, page, tail); |
88349a28 | 2285 | else if (m == M_FULL) |
2cfb7455 | 2286 | add_full(s, n, page); |
2cfb7455 CL |
2287 | } |
2288 | ||
2289 | l = m; | |
1d07171c | 2290 | if (!__cmpxchg_double_slab(s, page, |
2cfb7455 CL |
2291 | old.freelist, old.counters, |
2292 | new.freelist, new.counters, | |
2293 | "unfreezing slab")) | |
2294 | goto redo; | |
2295 | ||
2cfb7455 CL |
2296 | if (lock) |
2297 | spin_unlock(&n->list_lock); | |
2298 | ||
88349a28 WY |
2299 | if (m == M_PARTIAL) |
2300 | stat(s, tail); | |
2301 | else if (m == M_FULL) | |
2302 | stat(s, DEACTIVATE_FULL); | |
2303 | else if (m == M_FREE) { | |
2cfb7455 CL |
2304 | stat(s, DEACTIVATE_EMPTY); |
2305 | discard_slab(s, page); | |
2306 | stat(s, FREE_SLAB); | |
894b8788 | 2307 | } |
d4ff6d35 WY |
2308 | |
2309 | c->page = NULL; | |
2310 | c->freelist = NULL; | |
81819f0f CL |
2311 | } |
2312 | ||
d24ac77f JK |
2313 | /* |
2314 | * Unfreeze all the cpu partial slabs. | |
2315 | * | |
59a09917 CL |
2316 | * This function must be called with interrupts disabled |
2317 | * for the cpu using c (or some other guarantee must be there | |
2318 | * to guarantee no concurrent accesses). | |
d24ac77f | 2319 | */ |
59a09917 CL |
2320 | static void unfreeze_partials(struct kmem_cache *s, |
2321 | struct kmem_cache_cpu *c) | |
49e22585 | 2322 | { |
345c905d | 2323 | #ifdef CONFIG_SLUB_CPU_PARTIAL |
43d77867 | 2324 | struct kmem_cache_node *n = NULL, *n2 = NULL; |
9ada1934 | 2325 | struct page *page, *discard_page = NULL; |
49e22585 | 2326 | |
4c7ba22e | 2327 | while ((page = slub_percpu_partial(c))) { |
49e22585 CL |
2328 | struct page new; |
2329 | struct page old; | |
2330 | ||
4c7ba22e | 2331 | slub_set_percpu_partial(c, page); |
43d77867 JK |
2332 | |
2333 | n2 = get_node(s, page_to_nid(page)); | |
2334 | if (n != n2) { | |
2335 | if (n) | |
2336 | spin_unlock(&n->list_lock); | |
2337 | ||
2338 | n = n2; | |
2339 | spin_lock(&n->list_lock); | |
2340 | } | |
49e22585 CL |
2341 | |
2342 | do { | |
2343 | ||
2344 | old.freelist = page->freelist; | |
2345 | old.counters = page->counters; | |
a0132ac0 | 2346 | VM_BUG_ON(!old.frozen); |
49e22585 CL |
2347 | |
2348 | new.counters = old.counters; | |
2349 | new.freelist = old.freelist; | |
2350 | ||
2351 | new.frozen = 0; | |
2352 | ||
d24ac77f | 2353 | } while (!__cmpxchg_double_slab(s, page, |
49e22585 CL |
2354 | old.freelist, old.counters, |
2355 | new.freelist, new.counters, | |
2356 | "unfreezing slab")); | |
2357 | ||
8a5b20ae | 2358 | if (unlikely(!new.inuse && n->nr_partial >= s->min_partial)) { |
9ada1934 SL |
2359 | page->next = discard_page; |
2360 | discard_page = page; | |
43d77867 JK |
2361 | } else { |
2362 | add_partial(n, page, DEACTIVATE_TO_TAIL); | |
2363 | stat(s, FREE_ADD_PARTIAL); | |
49e22585 CL |
2364 | } |
2365 | } | |
2366 | ||
2367 | if (n) | |
2368 | spin_unlock(&n->list_lock); | |
9ada1934 SL |
2369 | |
2370 | while (discard_page) { | |
2371 | page = discard_page; | |
2372 | discard_page = discard_page->next; | |
2373 | ||
2374 | stat(s, DEACTIVATE_EMPTY); | |
2375 | discard_slab(s, page); | |
2376 | stat(s, FREE_SLAB); | |
2377 | } | |
6dfd1b65 | 2378 | #endif /* CONFIG_SLUB_CPU_PARTIAL */ |
49e22585 CL |
2379 | } |
2380 | ||
2381 | /* | |
9234bae9 WY |
2382 | * Put a page that was just frozen (in __slab_free|get_partial_node) into a |
2383 | * partial page slot if available. | |
49e22585 CL |
2384 | * |
2385 | * If we did not find a slot then simply move all the partials to the | |
2386 | * per node partial list. | |
2387 | */ | |
633b0764 | 2388 | static void put_cpu_partial(struct kmem_cache *s, struct page *page, int drain) |
49e22585 | 2389 | { |
345c905d | 2390 | #ifdef CONFIG_SLUB_CPU_PARTIAL |
49e22585 CL |
2391 | struct page *oldpage; |
2392 | int pages; | |
2393 | int pobjects; | |
2394 | ||
d6e0b7fa | 2395 | preempt_disable(); |
49e22585 CL |
2396 | do { |
2397 | pages = 0; | |
2398 | pobjects = 0; | |
2399 | oldpage = this_cpu_read(s->cpu_slab->partial); | |
2400 | ||
2401 | if (oldpage) { | |
2402 | pobjects = oldpage->pobjects; | |
2403 | pages = oldpage->pages; | |
bbd4e305 | 2404 | if (drain && pobjects > slub_cpu_partial(s)) { |
49e22585 CL |
2405 | unsigned long flags; |
2406 | /* | |
2407 | * partial array is full. Move the existing | |
2408 | * set to the per node partial list. | |
2409 | */ | |
2410 | local_irq_save(flags); | |
59a09917 | 2411 | unfreeze_partials(s, this_cpu_ptr(s->cpu_slab)); |
49e22585 | 2412 | local_irq_restore(flags); |
e24fc410 | 2413 | oldpage = NULL; |
49e22585 CL |
2414 | pobjects = 0; |
2415 | pages = 0; | |
8028dcea | 2416 | stat(s, CPU_PARTIAL_DRAIN); |
49e22585 CL |
2417 | } |
2418 | } | |
2419 | ||
2420 | pages++; | |
2421 | pobjects += page->objects - page->inuse; | |
2422 | ||
2423 | page->pages = pages; | |
2424 | page->pobjects = pobjects; | |
2425 | page->next = oldpage; | |
2426 | ||
d0e0ac97 CG |
2427 | } while (this_cpu_cmpxchg(s->cpu_slab->partial, oldpage, page) |
2428 | != oldpage); | |
bbd4e305 | 2429 | if (unlikely(!slub_cpu_partial(s))) { |
d6e0b7fa VD |
2430 | unsigned long flags; |
2431 | ||
2432 | local_irq_save(flags); | |
2433 | unfreeze_partials(s, this_cpu_ptr(s->cpu_slab)); | |
2434 | local_irq_restore(flags); | |
2435 | } | |
2436 | preempt_enable(); | |
6dfd1b65 | 2437 | #endif /* CONFIG_SLUB_CPU_PARTIAL */ |
49e22585 CL |
2438 | } |
2439 | ||
dfb4f096 | 2440 | static inline void flush_slab(struct kmem_cache *s, struct kmem_cache_cpu *c) |
81819f0f | 2441 | { |
84e554e6 | 2442 | stat(s, CPUSLAB_FLUSH); |
d4ff6d35 | 2443 | deactivate_slab(s, c->page, c->freelist, c); |
c17dda40 CL |
2444 | |
2445 | c->tid = next_tid(c->tid); | |
81819f0f CL |
2446 | } |
2447 | ||
2448 | /* | |
2449 | * Flush cpu slab. | |
6446faa2 | 2450 | * |
81819f0f CL |
2451 | * Called from IPI handler with interrupts disabled. |
2452 | */ | |
0c710013 | 2453 | static inline void __flush_cpu_slab(struct kmem_cache *s, int cpu) |
81819f0f | 2454 | { |
9dfc6e68 | 2455 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu); |
81819f0f | 2456 | |
1265ef2d WY |
2457 | if (c->page) |
2458 | flush_slab(s, c); | |
49e22585 | 2459 | |
1265ef2d | 2460 | unfreeze_partials(s, c); |
81819f0f CL |
2461 | } |
2462 | ||
2463 | static void flush_cpu_slab(void *d) | |
2464 | { | |
2465 | struct kmem_cache *s = d; | |
81819f0f | 2466 | |
dfb4f096 | 2467 | __flush_cpu_slab(s, smp_processor_id()); |
81819f0f CL |
2468 | } |
2469 | ||
a8364d55 GBY |
2470 | static bool has_cpu_slab(int cpu, void *info) |
2471 | { | |
2472 | struct kmem_cache *s = info; | |
2473 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu); | |
2474 | ||
a93cf07b | 2475 | return c->page || slub_percpu_partial(c); |
a8364d55 GBY |
2476 | } |
2477 | ||
81819f0f CL |
2478 | static void flush_all(struct kmem_cache *s) |
2479 | { | |
cb923159 | 2480 | on_each_cpu_cond(has_cpu_slab, flush_cpu_slab, s, 1); |
81819f0f CL |
2481 | } |
2482 | ||
a96a87bf SAS |
2483 | /* |
2484 | * Use the cpu notifier to insure that the cpu slabs are flushed when | |
2485 | * necessary. | |
2486 | */ | |
2487 | static int slub_cpu_dead(unsigned int cpu) | |
2488 | { | |
2489 | struct kmem_cache *s; | |
2490 | unsigned long flags; | |
2491 | ||
2492 | mutex_lock(&slab_mutex); | |
2493 | list_for_each_entry(s, &slab_caches, list) { | |
2494 | local_irq_save(flags); | |
2495 | __flush_cpu_slab(s, cpu); | |
2496 | local_irq_restore(flags); | |
2497 | } | |
2498 | mutex_unlock(&slab_mutex); | |
2499 | return 0; | |
2500 | } | |
2501 | ||
dfb4f096 CL |
2502 | /* |
2503 | * Check if the objects in a per cpu structure fit numa | |
2504 | * locality expectations. | |
2505 | */ | |
57d437d2 | 2506 | static inline int node_match(struct page *page, int node) |
dfb4f096 CL |
2507 | { |
2508 | #ifdef CONFIG_NUMA | |
6159d0f5 | 2509 | if (node != NUMA_NO_NODE && page_to_nid(page) != node) |
dfb4f096 CL |
2510 | return 0; |
2511 | #endif | |
2512 | return 1; | |
2513 | } | |
2514 | ||
9a02d699 | 2515 | #ifdef CONFIG_SLUB_DEBUG |
781b2ba6 PE |
2516 | static int count_free(struct page *page) |
2517 | { | |
2518 | return page->objects - page->inuse; | |
2519 | } | |
2520 | ||
9a02d699 DR |
2521 | static inline unsigned long node_nr_objs(struct kmem_cache_node *n) |
2522 | { | |
2523 | return atomic_long_read(&n->total_objects); | |
2524 | } | |
2525 | #endif /* CONFIG_SLUB_DEBUG */ | |
2526 | ||
2527 | #if defined(CONFIG_SLUB_DEBUG) || defined(CONFIG_SYSFS) | |
781b2ba6 PE |
2528 | static unsigned long count_partial(struct kmem_cache_node *n, |
2529 | int (*get_count)(struct page *)) | |
2530 | { | |
2531 | unsigned long flags; | |
2532 | unsigned long x = 0; | |
2533 | struct page *page; | |
2534 | ||
2535 | spin_lock_irqsave(&n->list_lock, flags); | |
916ac052 | 2536 | list_for_each_entry(page, &n->partial, slab_list) |
781b2ba6 PE |
2537 | x += get_count(page); |
2538 | spin_unlock_irqrestore(&n->list_lock, flags); | |
2539 | return x; | |
2540 | } | |
9a02d699 | 2541 | #endif /* CONFIG_SLUB_DEBUG || CONFIG_SYSFS */ |
26c02cf0 | 2542 | |
781b2ba6 PE |
2543 | static noinline void |
2544 | slab_out_of_memory(struct kmem_cache *s, gfp_t gfpflags, int nid) | |
2545 | { | |
9a02d699 DR |
2546 | #ifdef CONFIG_SLUB_DEBUG |
2547 | static DEFINE_RATELIMIT_STATE(slub_oom_rs, DEFAULT_RATELIMIT_INTERVAL, | |
2548 | DEFAULT_RATELIMIT_BURST); | |
781b2ba6 | 2549 | int node; |
fa45dc25 | 2550 | struct kmem_cache_node *n; |
781b2ba6 | 2551 | |
9a02d699 DR |
2552 | if ((gfpflags & __GFP_NOWARN) || !__ratelimit(&slub_oom_rs)) |
2553 | return; | |
2554 | ||
5b3810e5 VB |
2555 | pr_warn("SLUB: Unable to allocate memory on node %d, gfp=%#x(%pGg)\n", |
2556 | nid, gfpflags, &gfpflags); | |
19af27af | 2557 | pr_warn(" cache: %s, object size: %u, buffer size: %u, default order: %u, min order: %u\n", |
f9f58285 FF |
2558 | s->name, s->object_size, s->size, oo_order(s->oo), |
2559 | oo_order(s->min)); | |
781b2ba6 | 2560 | |
3b0efdfa | 2561 | if (oo_order(s->min) > get_order(s->object_size)) |
f9f58285 FF |
2562 | pr_warn(" %s debugging increased min order, use slub_debug=O to disable.\n", |
2563 | s->name); | |
fa5ec8a1 | 2564 | |
fa45dc25 | 2565 | for_each_kmem_cache_node(s, node, n) { |
781b2ba6 PE |
2566 | unsigned long nr_slabs; |
2567 | unsigned long nr_objs; | |
2568 | unsigned long nr_free; | |
2569 | ||
26c02cf0 AB |
2570 | nr_free = count_partial(n, count_free); |
2571 | nr_slabs = node_nr_slabs(n); | |
2572 | nr_objs = node_nr_objs(n); | |
781b2ba6 | 2573 | |
f9f58285 | 2574 | pr_warn(" node %d: slabs: %ld, objs: %ld, free: %ld\n", |
781b2ba6 PE |
2575 | node, nr_slabs, nr_objs, nr_free); |
2576 | } | |
9a02d699 | 2577 | #endif |
781b2ba6 PE |
2578 | } |
2579 | ||
497b66f2 CL |
2580 | static inline void *new_slab_objects(struct kmem_cache *s, gfp_t flags, |
2581 | int node, struct kmem_cache_cpu **pc) | |
2582 | { | |
6faa6833 | 2583 | void *freelist; |
188fd063 CL |
2584 | struct kmem_cache_cpu *c = *pc; |
2585 | struct page *page; | |
497b66f2 | 2586 | |
128227e7 MW |
2587 | WARN_ON_ONCE(s->ctor && (flags & __GFP_ZERO)); |
2588 | ||
188fd063 | 2589 | freelist = get_partial(s, flags, node, c); |
497b66f2 | 2590 | |
188fd063 CL |
2591 | if (freelist) |
2592 | return freelist; | |
2593 | ||
2594 | page = new_slab(s, flags, node); | |
497b66f2 | 2595 | if (page) { |
7c8e0181 | 2596 | c = raw_cpu_ptr(s->cpu_slab); |
497b66f2 CL |
2597 | if (c->page) |
2598 | flush_slab(s, c); | |
2599 | ||
2600 | /* | |
2601 | * No other reference to the page yet so we can | |
2602 | * muck around with it freely without cmpxchg | |
2603 | */ | |
6faa6833 | 2604 | freelist = page->freelist; |
497b66f2 CL |
2605 | page->freelist = NULL; |
2606 | ||
2607 | stat(s, ALLOC_SLAB); | |
497b66f2 CL |
2608 | c->page = page; |
2609 | *pc = c; | |
edde82b6 | 2610 | } |
497b66f2 | 2611 | |
6faa6833 | 2612 | return freelist; |
497b66f2 CL |
2613 | } |
2614 | ||
072bb0aa MG |
2615 | static inline bool pfmemalloc_match(struct page *page, gfp_t gfpflags) |
2616 | { | |
2617 | if (unlikely(PageSlabPfmemalloc(page))) | |
2618 | return gfp_pfmemalloc_allowed(gfpflags); | |
2619 | ||
2620 | return true; | |
2621 | } | |
2622 | ||
213eeb9f | 2623 | /* |
d0e0ac97 CG |
2624 | * Check the page->freelist of a page and either transfer the freelist to the |
2625 | * per cpu freelist or deactivate the page. | |
213eeb9f CL |
2626 | * |
2627 | * The page is still frozen if the return value is not NULL. | |
2628 | * | |
2629 | * If this function returns NULL then the page has been unfrozen. | |
d24ac77f JK |
2630 | * |
2631 | * This function must be called with interrupt disabled. | |
213eeb9f CL |
2632 | */ |
2633 | static inline void *get_freelist(struct kmem_cache *s, struct page *page) | |
2634 | { | |
2635 | struct page new; | |
2636 | unsigned long counters; | |
2637 | void *freelist; | |
2638 | ||
2639 | do { | |
2640 | freelist = page->freelist; | |
2641 | counters = page->counters; | |
6faa6833 | 2642 | |
213eeb9f | 2643 | new.counters = counters; |
a0132ac0 | 2644 | VM_BUG_ON(!new.frozen); |
213eeb9f CL |
2645 | |
2646 | new.inuse = page->objects; | |
2647 | new.frozen = freelist != NULL; | |
2648 | ||
d24ac77f | 2649 | } while (!__cmpxchg_double_slab(s, page, |
213eeb9f CL |
2650 | freelist, counters, |
2651 | NULL, new.counters, | |
2652 | "get_freelist")); | |
2653 | ||
2654 | return freelist; | |
2655 | } | |
2656 | ||
81819f0f | 2657 | /* |
894b8788 CL |
2658 | * Slow path. The lockless freelist is empty or we need to perform |
2659 | * debugging duties. | |
2660 | * | |
894b8788 CL |
2661 | * Processing is still very fast if new objects have been freed to the |
2662 | * regular freelist. In that case we simply take over the regular freelist | |
2663 | * as the lockless freelist and zap the regular freelist. | |
81819f0f | 2664 | * |
894b8788 CL |
2665 | * If that is not working then we fall back to the partial lists. We take the |
2666 | * first element of the freelist as the object to allocate now and move the | |
2667 | * rest of the freelist to the lockless freelist. | |
81819f0f | 2668 | * |
894b8788 | 2669 | * And if we were unable to get a new slab from the partial slab lists then |
6446faa2 CL |
2670 | * we need to allocate a new slab. This is the slowest path since it involves |
2671 | * a call to the page allocator and the setup of a new slab. | |
a380a3c7 CL |
2672 | * |
2673 | * Version of __slab_alloc to use when we know that interrupts are | |
2674 | * already disabled (which is the case for bulk allocation). | |
81819f0f | 2675 | */ |
a380a3c7 | 2676 | static void *___slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node, |
ce71e27c | 2677 | unsigned long addr, struct kmem_cache_cpu *c) |
81819f0f | 2678 | { |
6faa6833 | 2679 | void *freelist; |
f6e7def7 | 2680 | struct page *page; |
81819f0f | 2681 | |
9f986d99 AW |
2682 | stat(s, ALLOC_SLOWPATH); |
2683 | ||
f6e7def7 | 2684 | page = c->page; |
0715e6c5 VB |
2685 | if (!page) { |
2686 | /* | |
2687 | * if the node is not online or has no normal memory, just | |
2688 | * ignore the node constraint | |
2689 | */ | |
2690 | if (unlikely(node != NUMA_NO_NODE && | |
7e1fa93d | 2691 | !node_isset(node, slab_nodes))) |
0715e6c5 | 2692 | node = NUMA_NO_NODE; |
81819f0f | 2693 | goto new_slab; |
0715e6c5 | 2694 | } |
49e22585 | 2695 | redo: |
6faa6833 | 2696 | |
57d437d2 | 2697 | if (unlikely(!node_match(page, node))) { |
0715e6c5 VB |
2698 | /* |
2699 | * same as above but node_match() being false already | |
2700 | * implies node != NUMA_NO_NODE | |
2701 | */ | |
7e1fa93d | 2702 | if (!node_isset(node, slab_nodes)) { |
0715e6c5 VB |
2703 | node = NUMA_NO_NODE; |
2704 | goto redo; | |
2705 | } else { | |
a561ce00 | 2706 | stat(s, ALLOC_NODE_MISMATCH); |
d4ff6d35 | 2707 | deactivate_slab(s, page, c->freelist, c); |
a561ce00 JK |
2708 | goto new_slab; |
2709 | } | |
fc59c053 | 2710 | } |
6446faa2 | 2711 | |
072bb0aa MG |
2712 | /* |
2713 | * By rights, we should be searching for a slab page that was | |
2714 | * PFMEMALLOC but right now, we are losing the pfmemalloc | |
2715 | * information when the page leaves the per-cpu allocator | |
2716 | */ | |
2717 | if (unlikely(!pfmemalloc_match(page, gfpflags))) { | |
d4ff6d35 | 2718 | deactivate_slab(s, page, c->freelist, c); |
072bb0aa MG |
2719 | goto new_slab; |
2720 | } | |
2721 | ||
73736e03 | 2722 | /* must check again c->freelist in case of cpu migration or IRQ */ |
6faa6833 CL |
2723 | freelist = c->freelist; |
2724 | if (freelist) | |
73736e03 | 2725 | goto load_freelist; |
03e404af | 2726 | |
f6e7def7 | 2727 | freelist = get_freelist(s, page); |
6446faa2 | 2728 | |
6faa6833 | 2729 | if (!freelist) { |
03e404af CL |
2730 | c->page = NULL; |
2731 | stat(s, DEACTIVATE_BYPASS); | |
fc59c053 | 2732 | goto new_slab; |
03e404af | 2733 | } |
6446faa2 | 2734 | |
84e554e6 | 2735 | stat(s, ALLOC_REFILL); |
6446faa2 | 2736 | |
894b8788 | 2737 | load_freelist: |
507effea CL |
2738 | /* |
2739 | * freelist is pointing to the list of objects to be used. | |
2740 | * page is pointing to the page from which the objects are obtained. | |
2741 | * That page must be frozen for per cpu allocations to work. | |
2742 | */ | |
a0132ac0 | 2743 | VM_BUG_ON(!c->page->frozen); |
6faa6833 | 2744 | c->freelist = get_freepointer(s, freelist); |
8a5ec0ba | 2745 | c->tid = next_tid(c->tid); |
6faa6833 | 2746 | return freelist; |
81819f0f | 2747 | |
81819f0f | 2748 | new_slab: |
2cfb7455 | 2749 | |
a93cf07b WY |
2750 | if (slub_percpu_partial(c)) { |
2751 | page = c->page = slub_percpu_partial(c); | |
2752 | slub_set_percpu_partial(c, page); | |
49e22585 | 2753 | stat(s, CPU_PARTIAL_ALLOC); |
49e22585 | 2754 | goto redo; |
81819f0f CL |
2755 | } |
2756 | ||
188fd063 | 2757 | freelist = new_slab_objects(s, gfpflags, node, &c); |
01ad8a7b | 2758 | |
f4697436 | 2759 | if (unlikely(!freelist)) { |
9a02d699 | 2760 | slab_out_of_memory(s, gfpflags, node); |
f4697436 | 2761 | return NULL; |
81819f0f | 2762 | } |
2cfb7455 | 2763 | |
f6e7def7 | 2764 | page = c->page; |
5091b74a | 2765 | if (likely(!kmem_cache_debug(s) && pfmemalloc_match(page, gfpflags))) |
4b6f0750 | 2766 | goto load_freelist; |
2cfb7455 | 2767 | |
497b66f2 | 2768 | /* Only entered in the debug case */ |
d0e0ac97 CG |
2769 | if (kmem_cache_debug(s) && |
2770 | !alloc_debug_processing(s, page, freelist, addr)) | |
497b66f2 | 2771 | goto new_slab; /* Slab failed checks. Next slab needed */ |
894b8788 | 2772 | |
d4ff6d35 | 2773 | deactivate_slab(s, page, get_freepointer(s, freelist), c); |
6faa6833 | 2774 | return freelist; |
894b8788 CL |
2775 | } |
2776 | ||
a380a3c7 CL |
2777 | /* |
2778 | * Another one that disabled interrupt and compensates for possible | |
2779 | * cpu changes by refetching the per cpu area pointer. | |
2780 | */ | |
2781 | static void *__slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node, | |
2782 | unsigned long addr, struct kmem_cache_cpu *c) | |
2783 | { | |
2784 | void *p; | |
2785 | unsigned long flags; | |
2786 | ||
2787 | local_irq_save(flags); | |
923717cb | 2788 | #ifdef CONFIG_PREEMPTION |
a380a3c7 CL |
2789 | /* |
2790 | * We may have been preempted and rescheduled on a different | |
2791 | * cpu before disabling interrupts. Need to reload cpu area | |
2792 | * pointer. | |
2793 | */ | |
2794 | c = this_cpu_ptr(s->cpu_slab); | |
2795 | #endif | |
2796 | ||
2797 | p = ___slab_alloc(s, gfpflags, node, addr, c); | |
2798 | local_irq_restore(flags); | |
2799 | return p; | |
2800 | } | |
2801 | ||
0f181f9f AP |
2802 | /* |
2803 | * If the object has been wiped upon free, make sure it's fully initialized by | |
2804 | * zeroing out freelist pointer. | |
2805 | */ | |
2806 | static __always_inline void maybe_wipe_obj_freeptr(struct kmem_cache *s, | |
2807 | void *obj) | |
2808 | { | |
2809 | if (unlikely(slab_want_init_on_free(s)) && obj) | |
ce5716c6 AK |
2810 | memset((void *)((char *)kasan_reset_tag(obj) + s->offset), |
2811 | 0, sizeof(void *)); | |
0f181f9f AP |
2812 | } |
2813 | ||
894b8788 CL |
2814 | /* |
2815 | * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc) | |
2816 | * have the fastpath folded into their functions. So no function call | |
2817 | * overhead for requests that can be satisfied on the fastpath. | |
2818 | * | |
2819 | * The fastpath works by first checking if the lockless freelist can be used. | |
2820 | * If not then __slab_alloc is called for slow processing. | |
2821 | * | |
2822 | * Otherwise we can simply pick the next object from the lockless free list. | |
2823 | */ | |
2b847c3c | 2824 | static __always_inline void *slab_alloc_node(struct kmem_cache *s, |
b89fb5ef | 2825 | gfp_t gfpflags, int node, unsigned long addr, size_t orig_size) |
894b8788 | 2826 | { |
03ec0ed5 | 2827 | void *object; |
dfb4f096 | 2828 | struct kmem_cache_cpu *c; |
57d437d2 | 2829 | struct page *page; |
8a5ec0ba | 2830 | unsigned long tid; |
964d4bd3 | 2831 | struct obj_cgroup *objcg = NULL; |
da844b78 | 2832 | bool init = false; |
1f84260c | 2833 | |
964d4bd3 | 2834 | s = slab_pre_alloc_hook(s, &objcg, 1, gfpflags); |
8135be5a | 2835 | if (!s) |
773ff60e | 2836 | return NULL; |
b89fb5ef AP |
2837 | |
2838 | object = kfence_alloc(s, orig_size, gfpflags); | |
2839 | if (unlikely(object)) | |
2840 | goto out; | |
2841 | ||
8a5ec0ba | 2842 | redo: |
8a5ec0ba CL |
2843 | /* |
2844 | * Must read kmem_cache cpu data via this cpu ptr. Preemption is | |
2845 | * enabled. We may switch back and forth between cpus while | |
2846 | * reading from one cpu area. That does not matter as long | |
2847 | * as we end up on the original cpu again when doing the cmpxchg. | |
7cccd80b | 2848 | * |
9aabf810 | 2849 | * We should guarantee that tid and kmem_cache are retrieved on |
923717cb | 2850 | * the same cpu. It could be different if CONFIG_PREEMPTION so we need |
9aabf810 | 2851 | * to check if it is matched or not. |
8a5ec0ba | 2852 | */ |
9aabf810 JK |
2853 | do { |
2854 | tid = this_cpu_read(s->cpu_slab->tid); | |
2855 | c = raw_cpu_ptr(s->cpu_slab); | |
923717cb | 2856 | } while (IS_ENABLED(CONFIG_PREEMPTION) && |
859b7a0e | 2857 | unlikely(tid != READ_ONCE(c->tid))); |
9aabf810 JK |
2858 | |
2859 | /* | |
2860 | * Irqless object alloc/free algorithm used here depends on sequence | |
2861 | * of fetching cpu_slab's data. tid should be fetched before anything | |
2862 | * on c to guarantee that object and page associated with previous tid | |
2863 | * won't be used with current tid. If we fetch tid first, object and | |
2864 | * page could be one associated with next tid and our alloc/free | |
2865 | * request will be failed. In this case, we will retry. So, no problem. | |
2866 | */ | |
2867 | barrier(); | |
8a5ec0ba | 2868 | |
8a5ec0ba CL |
2869 | /* |
2870 | * The transaction ids are globally unique per cpu and per operation on | |
2871 | * a per cpu queue. Thus they can be guarantee that the cmpxchg_double | |
2872 | * occurs on the right processor and that there was no operation on the | |
2873 | * linked list in between. | |
2874 | */ | |
8a5ec0ba | 2875 | |
9dfc6e68 | 2876 | object = c->freelist; |
57d437d2 | 2877 | page = c->page; |
22e4663e | 2878 | if (unlikely(!object || !page || !node_match(page, node))) { |
dfb4f096 | 2879 | object = __slab_alloc(s, gfpflags, node, addr, c); |
8eae1492 | 2880 | } else { |
0ad9500e ED |
2881 | void *next_object = get_freepointer_safe(s, object); |
2882 | ||
8a5ec0ba | 2883 | /* |
25985edc | 2884 | * The cmpxchg will only match if there was no additional |
8a5ec0ba CL |
2885 | * operation and if we are on the right processor. |
2886 | * | |
d0e0ac97 CG |
2887 | * The cmpxchg does the following atomically (without lock |
2888 | * semantics!) | |
8a5ec0ba CL |
2889 | * 1. Relocate first pointer to the current per cpu area. |
2890 | * 2. Verify that tid and freelist have not been changed | |
2891 | * 3. If they were not changed replace tid and freelist | |
2892 | * | |
d0e0ac97 CG |
2893 | * Since this is without lock semantics the protection is only |
2894 | * against code executing on this cpu *not* from access by | |
2895 | * other cpus. | |
8a5ec0ba | 2896 | */ |
933393f5 | 2897 | if (unlikely(!this_cpu_cmpxchg_double( |
8a5ec0ba CL |
2898 | s->cpu_slab->freelist, s->cpu_slab->tid, |
2899 | object, tid, | |
0ad9500e | 2900 | next_object, next_tid(tid)))) { |
8a5ec0ba CL |
2901 | |
2902 | note_cmpxchg_failure("slab_alloc", s, tid); | |
2903 | goto redo; | |
2904 | } | |
0ad9500e | 2905 | prefetch_freepointer(s, next_object); |
84e554e6 | 2906 | stat(s, ALLOC_FASTPATH); |
894b8788 | 2907 | } |
0f181f9f | 2908 | |
ce5716c6 | 2909 | maybe_wipe_obj_freeptr(s, object); |
da844b78 | 2910 | init = slab_want_init_on_alloc(gfpflags, s); |
d07dbea4 | 2911 | |
b89fb5ef | 2912 | out: |
da844b78 | 2913 | slab_post_alloc_hook(s, objcg, gfpflags, 1, &object, init); |
5a896d9e | 2914 | |
894b8788 | 2915 | return object; |
81819f0f CL |
2916 | } |
2917 | ||
2b847c3c | 2918 | static __always_inline void *slab_alloc(struct kmem_cache *s, |
b89fb5ef | 2919 | gfp_t gfpflags, unsigned long addr, size_t orig_size) |
2b847c3c | 2920 | { |
b89fb5ef | 2921 | return slab_alloc_node(s, gfpflags, NUMA_NO_NODE, addr, orig_size); |
2b847c3c EG |
2922 | } |
2923 | ||
81819f0f CL |
2924 | void *kmem_cache_alloc(struct kmem_cache *s, gfp_t gfpflags) |
2925 | { | |
b89fb5ef | 2926 | void *ret = slab_alloc(s, gfpflags, _RET_IP_, s->object_size); |
5b882be4 | 2927 | |
d0e0ac97 CG |
2928 | trace_kmem_cache_alloc(_RET_IP_, ret, s->object_size, |
2929 | s->size, gfpflags); | |
5b882be4 EGM |
2930 | |
2931 | return ret; | |
81819f0f CL |
2932 | } |
2933 | EXPORT_SYMBOL(kmem_cache_alloc); | |
2934 | ||
0f24f128 | 2935 | #ifdef CONFIG_TRACING |
4a92379b RK |
2936 | void *kmem_cache_alloc_trace(struct kmem_cache *s, gfp_t gfpflags, size_t size) |
2937 | { | |
b89fb5ef | 2938 | void *ret = slab_alloc(s, gfpflags, _RET_IP_, size); |
4a92379b | 2939 | trace_kmalloc(_RET_IP_, ret, size, s->size, gfpflags); |
0116523c | 2940 | ret = kasan_kmalloc(s, ret, size, gfpflags); |
4a92379b RK |
2941 | return ret; |
2942 | } | |
2943 | EXPORT_SYMBOL(kmem_cache_alloc_trace); | |
5b882be4 EGM |
2944 | #endif |
2945 | ||
81819f0f CL |
2946 | #ifdef CONFIG_NUMA |
2947 | void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t gfpflags, int node) | |
2948 | { | |
b89fb5ef | 2949 | void *ret = slab_alloc_node(s, gfpflags, node, _RET_IP_, s->object_size); |
5b882be4 | 2950 | |
ca2b84cb | 2951 | trace_kmem_cache_alloc_node(_RET_IP_, ret, |
3b0efdfa | 2952 | s->object_size, s->size, gfpflags, node); |
5b882be4 EGM |
2953 | |
2954 | return ret; | |
81819f0f CL |
2955 | } |
2956 | EXPORT_SYMBOL(kmem_cache_alloc_node); | |
81819f0f | 2957 | |
0f24f128 | 2958 | #ifdef CONFIG_TRACING |
4a92379b | 2959 | void *kmem_cache_alloc_node_trace(struct kmem_cache *s, |
5b882be4 | 2960 | gfp_t gfpflags, |
4a92379b | 2961 | int node, size_t size) |
5b882be4 | 2962 | { |
b89fb5ef | 2963 | void *ret = slab_alloc_node(s, gfpflags, node, _RET_IP_, size); |
4a92379b RK |
2964 | |
2965 | trace_kmalloc_node(_RET_IP_, ret, | |
2966 | size, s->size, gfpflags, node); | |
0316bec2 | 2967 | |
0116523c | 2968 | ret = kasan_kmalloc(s, ret, size, gfpflags); |
4a92379b | 2969 | return ret; |
5b882be4 | 2970 | } |
4a92379b | 2971 | EXPORT_SYMBOL(kmem_cache_alloc_node_trace); |
5b882be4 | 2972 | #endif |
6dfd1b65 | 2973 | #endif /* CONFIG_NUMA */ |
5b882be4 | 2974 | |
81819f0f | 2975 | /* |
94e4d712 | 2976 | * Slow path handling. This may still be called frequently since objects |
894b8788 | 2977 | * have a longer lifetime than the cpu slabs in most processing loads. |
81819f0f | 2978 | * |
894b8788 CL |
2979 | * So we still attempt to reduce cache line usage. Just take the slab |
2980 | * lock and free the item. If there is no additional partial page | |
2981 | * handling required then we can return immediately. | |
81819f0f | 2982 | */ |
894b8788 | 2983 | static void __slab_free(struct kmem_cache *s, struct page *page, |
81084651 JDB |
2984 | void *head, void *tail, int cnt, |
2985 | unsigned long addr) | |
2986 | ||
81819f0f CL |
2987 | { |
2988 | void *prior; | |
2cfb7455 | 2989 | int was_frozen; |
2cfb7455 CL |
2990 | struct page new; |
2991 | unsigned long counters; | |
2992 | struct kmem_cache_node *n = NULL; | |
3f649ab7 | 2993 | unsigned long flags; |
81819f0f | 2994 | |
8a5ec0ba | 2995 | stat(s, FREE_SLOWPATH); |
81819f0f | 2996 | |
b89fb5ef AP |
2997 | if (kfence_free(head)) |
2998 | return; | |
2999 | ||
19c7ff9e | 3000 | if (kmem_cache_debug(s) && |
282acb43 | 3001 | !free_debug_processing(s, page, head, tail, cnt, addr)) |
80f08c19 | 3002 | return; |
6446faa2 | 3003 | |
2cfb7455 | 3004 | do { |
837d678d JK |
3005 | if (unlikely(n)) { |
3006 | spin_unlock_irqrestore(&n->list_lock, flags); | |
3007 | n = NULL; | |
3008 | } | |
2cfb7455 CL |
3009 | prior = page->freelist; |
3010 | counters = page->counters; | |
81084651 | 3011 | set_freepointer(s, tail, prior); |
2cfb7455 CL |
3012 | new.counters = counters; |
3013 | was_frozen = new.frozen; | |
81084651 | 3014 | new.inuse -= cnt; |
837d678d | 3015 | if ((!new.inuse || !prior) && !was_frozen) { |
49e22585 | 3016 | |
c65c1877 | 3017 | if (kmem_cache_has_cpu_partial(s) && !prior) { |
49e22585 CL |
3018 | |
3019 | /* | |
d0e0ac97 CG |
3020 | * Slab was on no list before and will be |
3021 | * partially empty | |
3022 | * We can defer the list move and instead | |
3023 | * freeze it. | |
49e22585 CL |
3024 | */ |
3025 | new.frozen = 1; | |
3026 | ||
c65c1877 | 3027 | } else { /* Needs to be taken off a list */ |
49e22585 | 3028 | |
b455def2 | 3029 | n = get_node(s, page_to_nid(page)); |
49e22585 CL |
3030 | /* |
3031 | * Speculatively acquire the list_lock. | |
3032 | * If the cmpxchg does not succeed then we may | |
3033 | * drop the list_lock without any processing. | |
3034 | * | |
3035 | * Otherwise the list_lock will synchronize with | |
3036 | * other processors updating the list of slabs. | |
3037 | */ | |
3038 | spin_lock_irqsave(&n->list_lock, flags); | |
3039 | ||
3040 | } | |
2cfb7455 | 3041 | } |
81819f0f | 3042 | |
2cfb7455 CL |
3043 | } while (!cmpxchg_double_slab(s, page, |
3044 | prior, counters, | |
81084651 | 3045 | head, new.counters, |
2cfb7455 | 3046 | "__slab_free")); |
81819f0f | 3047 | |
2cfb7455 | 3048 | if (likely(!n)) { |
49e22585 | 3049 | |
c270cf30 AW |
3050 | if (likely(was_frozen)) { |
3051 | /* | |
3052 | * The list lock was not taken therefore no list | |
3053 | * activity can be necessary. | |
3054 | */ | |
3055 | stat(s, FREE_FROZEN); | |
3056 | } else if (new.frozen) { | |
3057 | /* | |
3058 | * If we just froze the page then put it onto the | |
3059 | * per cpu partial list. | |
3060 | */ | |
49e22585 | 3061 | put_cpu_partial(s, page, 1); |
8028dcea AS |
3062 | stat(s, CPU_PARTIAL_FREE); |
3063 | } | |
c270cf30 | 3064 | |
b455def2 L |
3065 | return; |
3066 | } | |
81819f0f | 3067 | |
8a5b20ae | 3068 | if (unlikely(!new.inuse && n->nr_partial >= s->min_partial)) |
837d678d JK |
3069 | goto slab_empty; |
3070 | ||
81819f0f | 3071 | /* |
837d678d JK |
3072 | * Objects left in the slab. If it was not on the partial list before |
3073 | * then add it. | |
81819f0f | 3074 | */ |
345c905d | 3075 | if (!kmem_cache_has_cpu_partial(s) && unlikely(!prior)) { |
a4d3f891 | 3076 | remove_full(s, n, page); |
837d678d JK |
3077 | add_partial(n, page, DEACTIVATE_TO_TAIL); |
3078 | stat(s, FREE_ADD_PARTIAL); | |
8ff12cfc | 3079 | } |
80f08c19 | 3080 | spin_unlock_irqrestore(&n->list_lock, flags); |
81819f0f CL |
3081 | return; |
3082 | ||
3083 | slab_empty: | |
a973e9dd | 3084 | if (prior) { |
81819f0f | 3085 | /* |
6fbabb20 | 3086 | * Slab on the partial list. |
81819f0f | 3087 | */ |
5cc6eee8 | 3088 | remove_partial(n, page); |
84e554e6 | 3089 | stat(s, FREE_REMOVE_PARTIAL); |
c65c1877 | 3090 | } else { |
6fbabb20 | 3091 | /* Slab must be on the full list */ |
c65c1877 PZ |
3092 | remove_full(s, n, page); |
3093 | } | |
2cfb7455 | 3094 | |
80f08c19 | 3095 | spin_unlock_irqrestore(&n->list_lock, flags); |
84e554e6 | 3096 | stat(s, FREE_SLAB); |
81819f0f | 3097 | discard_slab(s, page); |
81819f0f CL |
3098 | } |
3099 | ||
894b8788 CL |
3100 | /* |
3101 | * Fastpath with forced inlining to produce a kfree and kmem_cache_free that | |
3102 | * can perform fastpath freeing without additional function calls. | |
3103 | * | |
3104 | * The fastpath is only possible if we are freeing to the current cpu slab | |
3105 | * of this processor. This typically the case if we have just allocated | |
3106 | * the item before. | |
3107 | * | |
3108 | * If fastpath is not possible then fall back to __slab_free where we deal | |
3109 | * with all sorts of special processing. | |
81084651 JDB |
3110 | * |
3111 | * Bulk free of a freelist with several objects (all pointing to the | |
3112 | * same page) possible by specifying head and tail ptr, plus objects | |
3113 | * count (cnt). Bulk free indicated by tail pointer being set. | |
894b8788 | 3114 | */ |
80a9201a AP |
3115 | static __always_inline void do_slab_free(struct kmem_cache *s, |
3116 | struct page *page, void *head, void *tail, | |
3117 | int cnt, unsigned long addr) | |
894b8788 | 3118 | { |
81084651 | 3119 | void *tail_obj = tail ? : head; |
dfb4f096 | 3120 | struct kmem_cache_cpu *c; |
8a5ec0ba | 3121 | unsigned long tid; |
964d4bd3 | 3122 | |
d1b2cf6c | 3123 | memcg_slab_free_hook(s, &head, 1); |
8a5ec0ba CL |
3124 | redo: |
3125 | /* | |
3126 | * Determine the currently cpus per cpu slab. | |
3127 | * The cpu may change afterward. However that does not matter since | |
3128 | * data is retrieved via this pointer. If we are on the same cpu | |
2ae44005 | 3129 | * during the cmpxchg then the free will succeed. |
8a5ec0ba | 3130 | */ |
9aabf810 JK |
3131 | do { |
3132 | tid = this_cpu_read(s->cpu_slab->tid); | |
3133 | c = raw_cpu_ptr(s->cpu_slab); | |
923717cb | 3134 | } while (IS_ENABLED(CONFIG_PREEMPTION) && |
859b7a0e | 3135 | unlikely(tid != READ_ONCE(c->tid))); |
c016b0bd | 3136 | |
9aabf810 JK |
3137 | /* Same with comment on barrier() in slab_alloc_node() */ |
3138 | barrier(); | |
c016b0bd | 3139 | |
442b06bc | 3140 | if (likely(page == c->page)) { |
5076190d LT |
3141 | void **freelist = READ_ONCE(c->freelist); |
3142 | ||
3143 | set_freepointer(s, tail_obj, freelist); | |
8a5ec0ba | 3144 | |
933393f5 | 3145 | if (unlikely(!this_cpu_cmpxchg_double( |
8a5ec0ba | 3146 | s->cpu_slab->freelist, s->cpu_slab->tid, |
5076190d | 3147 | freelist, tid, |
81084651 | 3148 | head, next_tid(tid)))) { |
8a5ec0ba CL |
3149 | |
3150 | note_cmpxchg_failure("slab_free", s, tid); | |
3151 | goto redo; | |
3152 | } | |
84e554e6 | 3153 | stat(s, FREE_FASTPATH); |
894b8788 | 3154 | } else |
81084651 | 3155 | __slab_free(s, page, head, tail_obj, cnt, addr); |
894b8788 | 3156 | |
894b8788 CL |
3157 | } |
3158 | ||
80a9201a AP |
3159 | static __always_inline void slab_free(struct kmem_cache *s, struct page *page, |
3160 | void *head, void *tail, int cnt, | |
3161 | unsigned long addr) | |
3162 | { | |
80a9201a | 3163 | /* |
c3895391 AK |
3164 | * With KASAN enabled slab_free_freelist_hook modifies the freelist |
3165 | * to remove objects, whose reuse must be delayed. | |
80a9201a | 3166 | */ |
c3895391 AK |
3167 | if (slab_free_freelist_hook(s, &head, &tail)) |
3168 | do_slab_free(s, page, head, tail, cnt, addr); | |
80a9201a AP |
3169 | } |
3170 | ||
2bd926b4 | 3171 | #ifdef CONFIG_KASAN_GENERIC |
80a9201a AP |
3172 | void ___cache_free(struct kmem_cache *cache, void *x, unsigned long addr) |
3173 | { | |
3174 | do_slab_free(cache, virt_to_head_page(x), x, NULL, 1, addr); | |
3175 | } | |
3176 | #endif | |
3177 | ||
81819f0f CL |
3178 | void kmem_cache_free(struct kmem_cache *s, void *x) |
3179 | { | |
b9ce5ef4 GC |
3180 | s = cache_from_obj(s, x); |
3181 | if (!s) | |
79576102 | 3182 | return; |
81084651 | 3183 | slab_free(s, virt_to_head_page(x), x, NULL, 1, _RET_IP_); |
3544de8e | 3184 | trace_kmem_cache_free(_RET_IP_, x, s->name); |
81819f0f CL |
3185 | } |
3186 | EXPORT_SYMBOL(kmem_cache_free); | |
3187 | ||
d0ecd894 | 3188 | struct detached_freelist { |
fbd02630 | 3189 | struct page *page; |
d0ecd894 JDB |
3190 | void *tail; |
3191 | void *freelist; | |
3192 | int cnt; | |
376bf125 | 3193 | struct kmem_cache *s; |
d0ecd894 | 3194 | }; |
fbd02630 | 3195 | |
d0ecd894 JDB |
3196 | /* |
3197 | * This function progressively scans the array with free objects (with | |
3198 | * a limited look ahead) and extract objects belonging to the same | |
3199 | * page. It builds a detached freelist directly within the given | |
3200 | * page/objects. This can happen without any need for | |
3201 | * synchronization, because the objects are owned by running process. | |
3202 | * The freelist is build up as a single linked list in the objects. | |
3203 | * The idea is, that this detached freelist can then be bulk | |
3204 | * transferred to the real freelist(s), but only requiring a single | |
3205 | * synchronization primitive. Look ahead in the array is limited due | |
3206 | * to performance reasons. | |
3207 | */ | |
376bf125 JDB |
3208 | static inline |
3209 | int build_detached_freelist(struct kmem_cache *s, size_t size, | |
3210 | void **p, struct detached_freelist *df) | |
d0ecd894 JDB |
3211 | { |
3212 | size_t first_skipped_index = 0; | |
3213 | int lookahead = 3; | |
3214 | void *object; | |
ca257195 | 3215 | struct page *page; |
fbd02630 | 3216 | |
d0ecd894 JDB |
3217 | /* Always re-init detached_freelist */ |
3218 | df->page = NULL; | |
fbd02630 | 3219 | |
d0ecd894 JDB |
3220 | do { |
3221 | object = p[--size]; | |
ca257195 | 3222 | /* Do we need !ZERO_OR_NULL_PTR(object) here? (for kfree) */ |
d0ecd894 | 3223 | } while (!object && size); |
3eed034d | 3224 | |
d0ecd894 JDB |
3225 | if (!object) |
3226 | return 0; | |
fbd02630 | 3227 | |
ca257195 JDB |
3228 | page = virt_to_head_page(object); |
3229 | if (!s) { | |
3230 | /* Handle kalloc'ed objects */ | |
3231 | if (unlikely(!PageSlab(page))) { | |
3232 | BUG_ON(!PageCompound(page)); | |
3233 | kfree_hook(object); | |
4949148a | 3234 | __free_pages(page, compound_order(page)); |
ca257195 JDB |
3235 | p[size] = NULL; /* mark object processed */ |
3236 | return size; | |
3237 | } | |
3238 | /* Derive kmem_cache from object */ | |
3239 | df->s = page->slab_cache; | |
3240 | } else { | |
3241 | df->s = cache_from_obj(s, object); /* Support for memcg */ | |
3242 | } | |
376bf125 | 3243 | |
b89fb5ef | 3244 | if (is_kfence_address(object)) { |
d57a964e | 3245 | slab_free_hook(df->s, object, false); |
b89fb5ef AP |
3246 | __kfence_free(object); |
3247 | p[size] = NULL; /* mark object processed */ | |
3248 | return size; | |
3249 | } | |
3250 | ||
d0ecd894 | 3251 | /* Start new detached freelist */ |
ca257195 | 3252 | df->page = page; |
376bf125 | 3253 | set_freepointer(df->s, object, NULL); |
d0ecd894 JDB |
3254 | df->tail = object; |
3255 | df->freelist = object; | |
3256 | p[size] = NULL; /* mark object processed */ | |
3257 | df->cnt = 1; | |
3258 | ||
3259 | while (size) { | |
3260 | object = p[--size]; | |
3261 | if (!object) | |
3262 | continue; /* Skip processed objects */ | |
3263 | ||
3264 | /* df->page is always set at this point */ | |
3265 | if (df->page == virt_to_head_page(object)) { | |
3266 | /* Opportunity build freelist */ | |
376bf125 | 3267 | set_freepointer(df->s, object, df->freelist); |
d0ecd894 JDB |
3268 | df->freelist = object; |
3269 | df->cnt++; | |
3270 | p[size] = NULL; /* mark object processed */ | |
3271 | ||
3272 | continue; | |
fbd02630 | 3273 | } |
d0ecd894 JDB |
3274 | |
3275 | /* Limit look ahead search */ | |
3276 | if (!--lookahead) | |
3277 | break; | |
3278 | ||
3279 | if (!first_skipped_index) | |
3280 | first_skipped_index = size + 1; | |
fbd02630 | 3281 | } |
d0ecd894 JDB |
3282 | |
3283 | return first_skipped_index; | |
3284 | } | |
3285 | ||
d0ecd894 | 3286 | /* Note that interrupts must be enabled when calling this function. */ |
376bf125 | 3287 | void kmem_cache_free_bulk(struct kmem_cache *s, size_t size, void **p) |
d0ecd894 JDB |
3288 | { |
3289 | if (WARN_ON(!size)) | |
3290 | return; | |
3291 | ||
d1b2cf6c | 3292 | memcg_slab_free_hook(s, p, size); |
d0ecd894 JDB |
3293 | do { |
3294 | struct detached_freelist df; | |
3295 | ||
3296 | size = build_detached_freelist(s, size, p, &df); | |
84582c8a | 3297 | if (!df.page) |
d0ecd894 JDB |
3298 | continue; |
3299 | ||
457c82c3 | 3300 | slab_free(df.s, df.page, df.freelist, df.tail, df.cnt, _RET_IP_); |
d0ecd894 | 3301 | } while (likely(size)); |
484748f0 CL |
3302 | } |
3303 | EXPORT_SYMBOL(kmem_cache_free_bulk); | |
3304 | ||
994eb764 | 3305 | /* Note that interrupts must be enabled when calling this function. */ |
865762a8 JDB |
3306 | int kmem_cache_alloc_bulk(struct kmem_cache *s, gfp_t flags, size_t size, |
3307 | void **p) | |
484748f0 | 3308 | { |
994eb764 JDB |
3309 | struct kmem_cache_cpu *c; |
3310 | int i; | |
964d4bd3 | 3311 | struct obj_cgroup *objcg = NULL; |
994eb764 | 3312 | |
03ec0ed5 | 3313 | /* memcg and kmem_cache debug support */ |
964d4bd3 | 3314 | s = slab_pre_alloc_hook(s, &objcg, size, flags); |
03ec0ed5 JDB |
3315 | if (unlikely(!s)) |
3316 | return false; | |
994eb764 JDB |
3317 | /* |
3318 | * Drain objects in the per cpu slab, while disabling local | |
3319 | * IRQs, which protects against PREEMPT and interrupts | |
3320 | * handlers invoking normal fastpath. | |
3321 | */ | |
3322 | local_irq_disable(); | |
3323 | c = this_cpu_ptr(s->cpu_slab); | |
3324 | ||
3325 | for (i = 0; i < size; i++) { | |
b89fb5ef | 3326 | void *object = kfence_alloc(s, s->object_size, flags); |
994eb764 | 3327 | |
b89fb5ef AP |
3328 | if (unlikely(object)) { |
3329 | p[i] = object; | |
3330 | continue; | |
3331 | } | |
3332 | ||
3333 | object = c->freelist; | |
ebe909e0 | 3334 | if (unlikely(!object)) { |
fd4d9c7d JH |
3335 | /* |
3336 | * We may have removed an object from c->freelist using | |
3337 | * the fastpath in the previous iteration; in that case, | |
3338 | * c->tid has not been bumped yet. | |
3339 | * Since ___slab_alloc() may reenable interrupts while | |
3340 | * allocating memory, we should bump c->tid now. | |
3341 | */ | |
3342 | c->tid = next_tid(c->tid); | |
3343 | ||
ebe909e0 JDB |
3344 | /* |
3345 | * Invoking slow path likely have side-effect | |
3346 | * of re-populating per CPU c->freelist | |
3347 | */ | |
87098373 | 3348 | p[i] = ___slab_alloc(s, flags, NUMA_NO_NODE, |
ebe909e0 | 3349 | _RET_IP_, c); |
87098373 CL |
3350 | if (unlikely(!p[i])) |
3351 | goto error; | |
3352 | ||
ebe909e0 | 3353 | c = this_cpu_ptr(s->cpu_slab); |
0f181f9f AP |
3354 | maybe_wipe_obj_freeptr(s, p[i]); |
3355 | ||
ebe909e0 JDB |
3356 | continue; /* goto for-loop */ |
3357 | } | |
994eb764 JDB |
3358 | c->freelist = get_freepointer(s, object); |
3359 | p[i] = object; | |
0f181f9f | 3360 | maybe_wipe_obj_freeptr(s, p[i]); |
994eb764 JDB |
3361 | } |
3362 | c->tid = next_tid(c->tid); | |
3363 | local_irq_enable(); | |
3364 | ||
da844b78 AK |
3365 | /* |
3366 | * memcg and kmem_cache debug support and memory initialization. | |
3367 | * Done outside of the IRQ disabled fastpath loop. | |
3368 | */ | |
3369 | slab_post_alloc_hook(s, objcg, flags, size, p, | |
3370 | slab_want_init_on_alloc(flags, s)); | |
865762a8 | 3371 | return i; |
87098373 | 3372 | error: |
87098373 | 3373 | local_irq_enable(); |
da844b78 | 3374 | slab_post_alloc_hook(s, objcg, flags, i, p, false); |
03ec0ed5 | 3375 | __kmem_cache_free_bulk(s, i, p); |
865762a8 | 3376 | return 0; |
484748f0 CL |
3377 | } |
3378 | EXPORT_SYMBOL(kmem_cache_alloc_bulk); | |
3379 | ||
3380 | ||
81819f0f | 3381 | /* |
672bba3a CL |
3382 | * Object placement in a slab is made very easy because we always start at |
3383 | * offset 0. If we tune the size of the object to the alignment then we can | |
3384 | * get the required alignment by putting one properly sized object after | |
3385 | * another. | |
81819f0f CL |
3386 | * |
3387 | * Notice that the allocation order determines the sizes of the per cpu | |
3388 | * caches. Each processor has always one slab available for allocations. | |
3389 | * Increasing the allocation order reduces the number of times that slabs | |
672bba3a | 3390 | * must be moved on and off the partial lists and is therefore a factor in |
81819f0f | 3391 | * locking overhead. |
81819f0f CL |
3392 | */ |
3393 | ||
3394 | /* | |
f0953a1b | 3395 | * Minimum / Maximum order of slab pages. This influences locking overhead |
81819f0f CL |
3396 | * and slab fragmentation. A higher order reduces the number of partial slabs |
3397 | * and increases the number of allocations possible without having to | |
3398 | * take the list_lock. | |
3399 | */ | |
19af27af AD |
3400 | static unsigned int slub_min_order; |
3401 | static unsigned int slub_max_order = PAGE_ALLOC_COSTLY_ORDER; | |
3402 | static unsigned int slub_min_objects; | |
81819f0f | 3403 | |
81819f0f CL |
3404 | /* |
3405 | * Calculate the order of allocation given an slab object size. | |
3406 | * | |
672bba3a CL |
3407 | * The order of allocation has significant impact on performance and other |
3408 | * system components. Generally order 0 allocations should be preferred since | |
3409 | * order 0 does not cause fragmentation in the page allocator. Larger objects | |
3410 | * be problematic to put into order 0 slabs because there may be too much | |
c124f5b5 | 3411 | * unused space left. We go to a higher order if more than 1/16th of the slab |
672bba3a CL |
3412 | * would be wasted. |
3413 | * | |
3414 | * In order to reach satisfactory performance we must ensure that a minimum | |
3415 | * number of objects is in one slab. Otherwise we may generate too much | |
3416 | * activity on the partial lists which requires taking the list_lock. This is | |
3417 | * less a concern for large slabs though which are rarely used. | |
81819f0f | 3418 | * |
672bba3a CL |
3419 | * slub_max_order specifies the order where we begin to stop considering the |
3420 | * number of objects in a slab as critical. If we reach slub_max_order then | |
3421 | * we try to keep the page order as low as possible. So we accept more waste | |
3422 | * of space in favor of a small page order. | |
81819f0f | 3423 | * |
672bba3a CL |
3424 | * Higher order allocations also allow the placement of more objects in a |
3425 | * slab and thereby reduce object handling overhead. If the user has | |
dc84207d | 3426 | * requested a higher minimum order then we start with that one instead of |
672bba3a | 3427 | * the smallest order which will fit the object. |
81819f0f | 3428 | */ |
19af27af AD |
3429 | static inline unsigned int slab_order(unsigned int size, |
3430 | unsigned int min_objects, unsigned int max_order, | |
9736d2a9 | 3431 | unsigned int fract_leftover) |
81819f0f | 3432 | { |
19af27af AD |
3433 | unsigned int min_order = slub_min_order; |
3434 | unsigned int order; | |
81819f0f | 3435 | |
9736d2a9 | 3436 | if (order_objects(min_order, size) > MAX_OBJS_PER_PAGE) |
210b5c06 | 3437 | return get_order(size * MAX_OBJS_PER_PAGE) - 1; |
39b26464 | 3438 | |
9736d2a9 | 3439 | for (order = max(min_order, (unsigned int)get_order(min_objects * size)); |
5e6d444e | 3440 | order <= max_order; order++) { |
81819f0f | 3441 | |
19af27af AD |
3442 | unsigned int slab_size = (unsigned int)PAGE_SIZE << order; |
3443 | unsigned int rem; | |
81819f0f | 3444 | |
9736d2a9 | 3445 | rem = slab_size % size; |
81819f0f | 3446 | |
5e6d444e | 3447 | if (rem <= slab_size / fract_leftover) |
81819f0f | 3448 | break; |
81819f0f | 3449 | } |
672bba3a | 3450 | |
81819f0f CL |
3451 | return order; |
3452 | } | |
3453 | ||
9736d2a9 | 3454 | static inline int calculate_order(unsigned int size) |
5e6d444e | 3455 | { |
19af27af AD |
3456 | unsigned int order; |
3457 | unsigned int min_objects; | |
3458 | unsigned int max_objects; | |
3286222f | 3459 | unsigned int nr_cpus; |
5e6d444e CL |
3460 | |
3461 | /* | |
3462 | * Attempt to find best configuration for a slab. This | |
3463 | * works by first attempting to generate a layout with | |
3464 | * the best configuration and backing off gradually. | |
3465 | * | |
422ff4d7 | 3466 | * First we increase the acceptable waste in a slab. Then |
5e6d444e CL |
3467 | * we reduce the minimum objects required in a slab. |
3468 | */ | |
3469 | min_objects = slub_min_objects; | |
3286222f VB |
3470 | if (!min_objects) { |
3471 | /* | |
3472 | * Some architectures will only update present cpus when | |
3473 | * onlining them, so don't trust the number if it's just 1. But | |
3474 | * we also don't want to use nr_cpu_ids always, as on some other | |
3475 | * architectures, there can be many possible cpus, but never | |
3476 | * onlined. Here we compromise between trying to avoid too high | |
3477 | * order on systems that appear larger than they are, and too | |
3478 | * low order on systems that appear smaller than they are. | |
3479 | */ | |
3480 | nr_cpus = num_present_cpus(); | |
3481 | if (nr_cpus <= 1) | |
3482 | nr_cpus = nr_cpu_ids; | |
3483 | min_objects = 4 * (fls(nr_cpus) + 1); | |
3484 | } | |
9736d2a9 | 3485 | max_objects = order_objects(slub_max_order, size); |
e8120ff1 ZY |
3486 | min_objects = min(min_objects, max_objects); |
3487 | ||
5e6d444e | 3488 | while (min_objects > 1) { |
19af27af AD |
3489 | unsigned int fraction; |
3490 | ||
c124f5b5 | 3491 | fraction = 16; |
5e6d444e CL |
3492 | while (fraction >= 4) { |
3493 | order = slab_order(size, min_objects, | |
9736d2a9 | 3494 | slub_max_order, fraction); |
5e6d444e CL |
3495 | if (order <= slub_max_order) |
3496 | return order; | |
3497 | fraction /= 2; | |
3498 | } | |
5086c389 | 3499 | min_objects--; |
5e6d444e CL |
3500 | } |
3501 | ||
3502 | /* | |
3503 | * We were unable to place multiple objects in a slab. Now | |
3504 | * lets see if we can place a single object there. | |
3505 | */ | |
9736d2a9 | 3506 | order = slab_order(size, 1, slub_max_order, 1); |
5e6d444e CL |
3507 | if (order <= slub_max_order) |
3508 | return order; | |
3509 | ||
3510 | /* | |
3511 | * Doh this slab cannot be placed using slub_max_order. | |
3512 | */ | |
9736d2a9 | 3513 | order = slab_order(size, 1, MAX_ORDER, 1); |
818cf590 | 3514 | if (order < MAX_ORDER) |
5e6d444e CL |
3515 | return order; |
3516 | return -ENOSYS; | |
3517 | } | |
3518 | ||
5595cffc | 3519 | static void |
4053497d | 3520 | init_kmem_cache_node(struct kmem_cache_node *n) |
81819f0f CL |
3521 | { |
3522 | n->nr_partial = 0; | |
81819f0f CL |
3523 | spin_lock_init(&n->list_lock); |
3524 | INIT_LIST_HEAD(&n->partial); | |
8ab1372f | 3525 | #ifdef CONFIG_SLUB_DEBUG |
0f389ec6 | 3526 | atomic_long_set(&n->nr_slabs, 0); |
02b71b70 | 3527 | atomic_long_set(&n->total_objects, 0); |
643b1138 | 3528 | INIT_LIST_HEAD(&n->full); |
8ab1372f | 3529 | #endif |
81819f0f CL |
3530 | } |
3531 | ||
55136592 | 3532 | static inline int alloc_kmem_cache_cpus(struct kmem_cache *s) |
4c93c355 | 3533 | { |
6c182dc0 | 3534 | BUILD_BUG_ON(PERCPU_DYNAMIC_EARLY_SIZE < |
95a05b42 | 3535 | KMALLOC_SHIFT_HIGH * sizeof(struct kmem_cache_cpu)); |
4c93c355 | 3536 | |
8a5ec0ba | 3537 | /* |
d4d84fef CM |
3538 | * Must align to double word boundary for the double cmpxchg |
3539 | * instructions to work; see __pcpu_double_call_return_bool(). | |
8a5ec0ba | 3540 | */ |
d4d84fef CM |
3541 | s->cpu_slab = __alloc_percpu(sizeof(struct kmem_cache_cpu), |
3542 | 2 * sizeof(void *)); | |
8a5ec0ba CL |
3543 | |
3544 | if (!s->cpu_slab) | |
3545 | return 0; | |
3546 | ||
3547 | init_kmem_cache_cpus(s); | |
4c93c355 | 3548 | |
8a5ec0ba | 3549 | return 1; |
4c93c355 | 3550 | } |
4c93c355 | 3551 | |
51df1142 CL |
3552 | static struct kmem_cache *kmem_cache_node; |
3553 | ||
81819f0f CL |
3554 | /* |
3555 | * No kmalloc_node yet so do it by hand. We know that this is the first | |
3556 | * slab on the node for this slabcache. There are no concurrent accesses | |
3557 | * possible. | |
3558 | * | |
721ae22a ZYW |
3559 | * Note that this function only works on the kmem_cache_node |
3560 | * when allocating for the kmem_cache_node. This is used for bootstrapping | |
4c93c355 | 3561 | * memory on a fresh node that has no slab structures yet. |
81819f0f | 3562 | */ |
55136592 | 3563 | static void early_kmem_cache_node_alloc(int node) |
81819f0f CL |
3564 | { |
3565 | struct page *page; | |
3566 | struct kmem_cache_node *n; | |
3567 | ||
51df1142 | 3568 | BUG_ON(kmem_cache_node->size < sizeof(struct kmem_cache_node)); |
81819f0f | 3569 | |
51df1142 | 3570 | page = new_slab(kmem_cache_node, GFP_NOWAIT, node); |
81819f0f CL |
3571 | |
3572 | BUG_ON(!page); | |
a2f92ee7 | 3573 | if (page_to_nid(page) != node) { |
f9f58285 FF |
3574 | pr_err("SLUB: Unable to allocate memory from node %d\n", node); |
3575 | pr_err("SLUB: Allocating a useless per node structure in order to be able to continue\n"); | |
a2f92ee7 CL |
3576 | } |
3577 | ||
81819f0f CL |
3578 | n = page->freelist; |
3579 | BUG_ON(!n); | |
8ab1372f | 3580 | #ifdef CONFIG_SLUB_DEBUG |
f7cb1933 | 3581 | init_object(kmem_cache_node, n, SLUB_RED_ACTIVE); |
51df1142 | 3582 | init_tracking(kmem_cache_node, n); |
8ab1372f | 3583 | #endif |
da844b78 | 3584 | n = kasan_slab_alloc(kmem_cache_node, n, GFP_KERNEL, false); |
12b22386 AK |
3585 | page->freelist = get_freepointer(kmem_cache_node, n); |
3586 | page->inuse = 1; | |
3587 | page->frozen = 0; | |
3588 | kmem_cache_node->node[node] = n; | |
4053497d | 3589 | init_kmem_cache_node(n); |
51df1142 | 3590 | inc_slabs_node(kmem_cache_node, node, page->objects); |
6446faa2 | 3591 | |
67b6c900 | 3592 | /* |
1e4dd946 SR |
3593 | * No locks need to be taken here as it has just been |
3594 | * initialized and there is no concurrent access. | |
67b6c900 | 3595 | */ |
1e4dd946 | 3596 | __add_partial(n, page, DEACTIVATE_TO_HEAD); |
81819f0f CL |
3597 | } |
3598 | ||
3599 | static void free_kmem_cache_nodes(struct kmem_cache *s) | |
3600 | { | |
3601 | int node; | |
fa45dc25 | 3602 | struct kmem_cache_node *n; |
81819f0f | 3603 | |
fa45dc25 | 3604 | for_each_kmem_cache_node(s, node, n) { |
81819f0f | 3605 | s->node[node] = NULL; |
ea37df54 | 3606 | kmem_cache_free(kmem_cache_node, n); |
81819f0f CL |
3607 | } |
3608 | } | |
3609 | ||
52b4b950 DS |
3610 | void __kmem_cache_release(struct kmem_cache *s) |
3611 | { | |
210e7a43 | 3612 | cache_random_seq_destroy(s); |
52b4b950 DS |
3613 | free_percpu(s->cpu_slab); |
3614 | free_kmem_cache_nodes(s); | |
3615 | } | |
3616 | ||
55136592 | 3617 | static int init_kmem_cache_nodes(struct kmem_cache *s) |
81819f0f CL |
3618 | { |
3619 | int node; | |
81819f0f | 3620 | |
7e1fa93d | 3621 | for_each_node_mask(node, slab_nodes) { |
81819f0f CL |
3622 | struct kmem_cache_node *n; |
3623 | ||
73367bd8 | 3624 | if (slab_state == DOWN) { |
55136592 | 3625 | early_kmem_cache_node_alloc(node); |
73367bd8 AD |
3626 | continue; |
3627 | } | |
51df1142 | 3628 | n = kmem_cache_alloc_node(kmem_cache_node, |
55136592 | 3629 | GFP_KERNEL, node); |
81819f0f | 3630 | |
73367bd8 AD |
3631 | if (!n) { |
3632 | free_kmem_cache_nodes(s); | |
3633 | return 0; | |
81819f0f | 3634 | } |
73367bd8 | 3635 | |
4053497d | 3636 | init_kmem_cache_node(n); |
ea37df54 | 3637 | s->node[node] = n; |
81819f0f CL |
3638 | } |
3639 | return 1; | |
3640 | } | |
81819f0f | 3641 | |
c0bdb232 | 3642 | static void set_min_partial(struct kmem_cache *s, unsigned long min) |
3b89d7d8 DR |
3643 | { |
3644 | if (min < MIN_PARTIAL) | |
3645 | min = MIN_PARTIAL; | |
3646 | else if (min > MAX_PARTIAL) | |
3647 | min = MAX_PARTIAL; | |
3648 | s->min_partial = min; | |
3649 | } | |
3650 | ||
e6d0e1dc WY |
3651 | static void set_cpu_partial(struct kmem_cache *s) |
3652 | { | |
3653 | #ifdef CONFIG_SLUB_CPU_PARTIAL | |
3654 | /* | |
3655 | * cpu_partial determined the maximum number of objects kept in the | |
3656 | * per cpu partial lists of a processor. | |
3657 | * | |
3658 | * Per cpu partial lists mainly contain slabs that just have one | |
3659 | * object freed. If they are used for allocation then they can be | |
3660 | * filled up again with minimal effort. The slab will never hit the | |
3661 | * per node partial lists and therefore no locking will be required. | |
3662 | * | |
3663 | * This setting also determines | |
3664 | * | |
3665 | * A) The number of objects from per cpu partial slabs dumped to the | |
3666 | * per node list when we reach the limit. | |
3667 | * B) The number of objects in cpu partial slabs to extract from the | |
3668 | * per node list when we run out of per cpu objects. We only fetch | |
3669 | * 50% to keep some capacity around for frees. | |
3670 | */ | |
3671 | if (!kmem_cache_has_cpu_partial(s)) | |
bbd4e305 | 3672 | slub_set_cpu_partial(s, 0); |
e6d0e1dc | 3673 | else if (s->size >= PAGE_SIZE) |
bbd4e305 | 3674 | slub_set_cpu_partial(s, 2); |
e6d0e1dc | 3675 | else if (s->size >= 1024) |
bbd4e305 | 3676 | slub_set_cpu_partial(s, 6); |
e6d0e1dc | 3677 | else if (s->size >= 256) |
bbd4e305 | 3678 | slub_set_cpu_partial(s, 13); |
e6d0e1dc | 3679 | else |
bbd4e305 | 3680 | slub_set_cpu_partial(s, 30); |
e6d0e1dc WY |
3681 | #endif |
3682 | } | |
3683 | ||
81819f0f CL |
3684 | /* |
3685 | * calculate_sizes() determines the order and the distribution of data within | |
3686 | * a slab object. | |
3687 | */ | |
06b285dc | 3688 | static int calculate_sizes(struct kmem_cache *s, int forced_order) |
81819f0f | 3689 | { |
d50112ed | 3690 | slab_flags_t flags = s->flags; |
be4a7988 | 3691 | unsigned int size = s->object_size; |
19af27af | 3692 | unsigned int order; |
81819f0f | 3693 | |
d8b42bf5 CL |
3694 | /* |
3695 | * Round up object size to the next word boundary. We can only | |
3696 | * place the free pointer at word boundaries and this determines | |
3697 | * the possible location of the free pointer. | |
3698 | */ | |
3699 | size = ALIGN(size, sizeof(void *)); | |
3700 | ||
3701 | #ifdef CONFIG_SLUB_DEBUG | |
81819f0f CL |
3702 | /* |
3703 | * Determine if we can poison the object itself. If the user of | |
3704 | * the slab may touch the object after free or before allocation | |
3705 | * then we should never poison the object itself. | |
3706 | */ | |
5f0d5a3a | 3707 | if ((flags & SLAB_POISON) && !(flags & SLAB_TYPESAFE_BY_RCU) && |
c59def9f | 3708 | !s->ctor) |
81819f0f CL |
3709 | s->flags |= __OBJECT_POISON; |
3710 | else | |
3711 | s->flags &= ~__OBJECT_POISON; | |
3712 | ||
81819f0f CL |
3713 | |
3714 | /* | |
672bba3a | 3715 | * If we are Redzoning then check if there is some space between the |
81819f0f | 3716 | * end of the object and the free pointer. If not then add an |
672bba3a | 3717 | * additional word to have some bytes to store Redzone information. |
81819f0f | 3718 | */ |
3b0efdfa | 3719 | if ((flags & SLAB_RED_ZONE) && size == s->object_size) |
81819f0f | 3720 | size += sizeof(void *); |
41ecc55b | 3721 | #endif |
81819f0f CL |
3722 | |
3723 | /* | |
672bba3a | 3724 | * With that we have determined the number of bytes in actual use |
e41a49fa | 3725 | * by the object and redzoning. |
81819f0f CL |
3726 | */ |
3727 | s->inuse = size; | |
3728 | ||
74c1d3e0 KC |
3729 | if ((flags & (SLAB_TYPESAFE_BY_RCU | SLAB_POISON)) || |
3730 | ((flags & SLAB_RED_ZONE) && s->object_size < sizeof(void *)) || | |
3731 | s->ctor) { | |
81819f0f CL |
3732 | /* |
3733 | * Relocate free pointer after the object if it is not | |
3734 | * permitted to overwrite the first word of the object on | |
3735 | * kmem_cache_free. | |
3736 | * | |
3737 | * This is the case if we do RCU, have a constructor or | |
74c1d3e0 KC |
3738 | * destructor, are poisoning the objects, or are |
3739 | * redzoning an object smaller than sizeof(void *). | |
cbfc35a4 WL |
3740 | * |
3741 | * The assumption that s->offset >= s->inuse means free | |
3742 | * pointer is outside of the object is used in the | |
3743 | * freeptr_outside_object() function. If that is no | |
3744 | * longer true, the function needs to be modified. | |
81819f0f CL |
3745 | */ |
3746 | s->offset = size; | |
3747 | size += sizeof(void *); | |
e41a49fa | 3748 | } else { |
3202fa62 KC |
3749 | /* |
3750 | * Store freelist pointer near middle of object to keep | |
3751 | * it away from the edges of the object to avoid small | |
3752 | * sized over/underflows from neighboring allocations. | |
3753 | */ | |
e41a49fa | 3754 | s->offset = ALIGN_DOWN(s->object_size / 2, sizeof(void *)); |
81819f0f CL |
3755 | } |
3756 | ||
c12b3c62 | 3757 | #ifdef CONFIG_SLUB_DEBUG |
81819f0f CL |
3758 | if (flags & SLAB_STORE_USER) |
3759 | /* | |
3760 | * Need to store information about allocs and frees after | |
3761 | * the object. | |
3762 | */ | |
3763 | size += 2 * sizeof(struct track); | |
80a9201a | 3764 | #endif |
81819f0f | 3765 | |
80a9201a AP |
3766 | kasan_cache_create(s, &size, &s->flags); |
3767 | #ifdef CONFIG_SLUB_DEBUG | |
d86bd1be | 3768 | if (flags & SLAB_RED_ZONE) { |
81819f0f CL |
3769 | /* |
3770 | * Add some empty padding so that we can catch | |
3771 | * overwrites from earlier objects rather than let | |
3772 | * tracking information or the free pointer be | |
0211a9c8 | 3773 | * corrupted if a user writes before the start |
81819f0f CL |
3774 | * of the object. |
3775 | */ | |
3776 | size += sizeof(void *); | |
d86bd1be JK |
3777 | |
3778 | s->red_left_pad = sizeof(void *); | |
3779 | s->red_left_pad = ALIGN(s->red_left_pad, s->align); | |
3780 | size += s->red_left_pad; | |
3781 | } | |
41ecc55b | 3782 | #endif |
672bba3a | 3783 | |
81819f0f CL |
3784 | /* |
3785 | * SLUB stores one object immediately after another beginning from | |
3786 | * offset 0. In order to align the objects we have to simply size | |
3787 | * each object to conform to the alignment. | |
3788 | */ | |
45906855 | 3789 | size = ALIGN(size, s->align); |
81819f0f | 3790 | s->size = size; |
4138fdfc | 3791 | s->reciprocal_size = reciprocal_value(size); |
06b285dc CL |
3792 | if (forced_order >= 0) |
3793 | order = forced_order; | |
3794 | else | |
9736d2a9 | 3795 | order = calculate_order(size); |
81819f0f | 3796 | |
19af27af | 3797 | if ((int)order < 0) |
81819f0f CL |
3798 | return 0; |
3799 | ||
b7a49f0d | 3800 | s->allocflags = 0; |
834f3d11 | 3801 | if (order) |
b7a49f0d CL |
3802 | s->allocflags |= __GFP_COMP; |
3803 | ||
3804 | if (s->flags & SLAB_CACHE_DMA) | |
2c59dd65 | 3805 | s->allocflags |= GFP_DMA; |
b7a49f0d | 3806 | |
6d6ea1e9 NB |
3807 | if (s->flags & SLAB_CACHE_DMA32) |
3808 | s->allocflags |= GFP_DMA32; | |
3809 | ||
b7a49f0d CL |
3810 | if (s->flags & SLAB_RECLAIM_ACCOUNT) |
3811 | s->allocflags |= __GFP_RECLAIMABLE; | |
3812 | ||
81819f0f CL |
3813 | /* |
3814 | * Determine the number of objects per slab | |
3815 | */ | |
9736d2a9 MW |
3816 | s->oo = oo_make(order, size); |
3817 | s->min = oo_make(get_order(size), size); | |
205ab99d CL |
3818 | if (oo_objects(s->oo) > oo_objects(s->max)) |
3819 | s->max = s->oo; | |
81819f0f | 3820 | |
834f3d11 | 3821 | return !!oo_objects(s->oo); |
81819f0f CL |
3822 | } |
3823 | ||
d50112ed | 3824 | static int kmem_cache_open(struct kmem_cache *s, slab_flags_t flags) |
81819f0f | 3825 | { |
37540008 | 3826 | s->flags = kmem_cache_flags(s->size, flags, s->name); |
2482ddec KC |
3827 | #ifdef CONFIG_SLAB_FREELIST_HARDENED |
3828 | s->random = get_random_long(); | |
3829 | #endif | |
81819f0f | 3830 | |
06b285dc | 3831 | if (!calculate_sizes(s, -1)) |
81819f0f | 3832 | goto error; |
3de47213 DR |
3833 | if (disable_higher_order_debug) { |
3834 | /* | |
3835 | * Disable debugging flags that store metadata if the min slab | |
3836 | * order increased. | |
3837 | */ | |
3b0efdfa | 3838 | if (get_order(s->size) > get_order(s->object_size)) { |
3de47213 DR |
3839 | s->flags &= ~DEBUG_METADATA_FLAGS; |
3840 | s->offset = 0; | |
3841 | if (!calculate_sizes(s, -1)) | |
3842 | goto error; | |
3843 | } | |
3844 | } | |
81819f0f | 3845 | |
2565409f HC |
3846 | #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \ |
3847 | defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE) | |
149daaf3 | 3848 | if (system_has_cmpxchg_double() && (s->flags & SLAB_NO_CMPXCHG) == 0) |
b789ef51 CL |
3849 | /* Enable fast mode */ |
3850 | s->flags |= __CMPXCHG_DOUBLE; | |
3851 | #endif | |
3852 | ||
3b89d7d8 DR |
3853 | /* |
3854 | * The larger the object size is, the more pages we want on the partial | |
3855 | * list to avoid pounding the page allocator excessively. | |
3856 | */ | |
49e22585 CL |
3857 | set_min_partial(s, ilog2(s->size) / 2); |
3858 | ||
e6d0e1dc | 3859 | set_cpu_partial(s); |
49e22585 | 3860 | |
81819f0f | 3861 | #ifdef CONFIG_NUMA |
e2cb96b7 | 3862 | s->remote_node_defrag_ratio = 1000; |
81819f0f | 3863 | #endif |
210e7a43 TG |
3864 | |
3865 | /* Initialize the pre-computed randomized freelist if slab is up */ | |
3866 | if (slab_state >= UP) { | |
3867 | if (init_cache_random_seq(s)) | |
3868 | goto error; | |
3869 | } | |
3870 | ||
55136592 | 3871 | if (!init_kmem_cache_nodes(s)) |
dfb4f096 | 3872 | goto error; |
81819f0f | 3873 | |
55136592 | 3874 | if (alloc_kmem_cache_cpus(s)) |
278b1bb1 | 3875 | return 0; |
ff12059e | 3876 | |
4c93c355 | 3877 | free_kmem_cache_nodes(s); |
81819f0f | 3878 | error: |
278b1bb1 | 3879 | return -EINVAL; |
81819f0f | 3880 | } |
81819f0f | 3881 | |
33b12c38 | 3882 | static void list_slab_objects(struct kmem_cache *s, struct page *page, |
55860d96 | 3883 | const char *text) |
33b12c38 CL |
3884 | { |
3885 | #ifdef CONFIG_SLUB_DEBUG | |
3886 | void *addr = page_address(page); | |
55860d96 | 3887 | unsigned long *map; |
33b12c38 | 3888 | void *p; |
aa456c7a | 3889 | |
945cf2b6 | 3890 | slab_err(s, page, text, s->name); |
33b12c38 | 3891 | slab_lock(page); |
33b12c38 | 3892 | |
90e9f6a6 | 3893 | map = get_map(s, page); |
33b12c38 CL |
3894 | for_each_object(p, s, addr, page->objects) { |
3895 | ||
4138fdfc | 3896 | if (!test_bit(__obj_to_index(s, addr, p), map)) { |
96b94abc | 3897 | pr_err("Object 0x%p @offset=%tu\n", p, p - addr); |
33b12c38 CL |
3898 | print_tracking(s, p); |
3899 | } | |
3900 | } | |
55860d96 | 3901 | put_map(map); |
33b12c38 CL |
3902 | slab_unlock(page); |
3903 | #endif | |
3904 | } | |
3905 | ||
81819f0f | 3906 | /* |
599870b1 | 3907 | * Attempt to free all partial slabs on a node. |
52b4b950 DS |
3908 | * This is called from __kmem_cache_shutdown(). We must take list_lock |
3909 | * because sysfs file might still access partial list after the shutdowning. | |
81819f0f | 3910 | */ |
599870b1 | 3911 | static void free_partial(struct kmem_cache *s, struct kmem_cache_node *n) |
81819f0f | 3912 | { |
60398923 | 3913 | LIST_HEAD(discard); |
81819f0f CL |
3914 | struct page *page, *h; |
3915 | ||
52b4b950 DS |
3916 | BUG_ON(irqs_disabled()); |
3917 | spin_lock_irq(&n->list_lock); | |
916ac052 | 3918 | list_for_each_entry_safe(page, h, &n->partial, slab_list) { |
81819f0f | 3919 | if (!page->inuse) { |
52b4b950 | 3920 | remove_partial(n, page); |
916ac052 | 3921 | list_add(&page->slab_list, &discard); |
33b12c38 CL |
3922 | } else { |
3923 | list_slab_objects(s, page, | |
55860d96 | 3924 | "Objects remaining in %s on __kmem_cache_shutdown()"); |
599870b1 | 3925 | } |
33b12c38 | 3926 | } |
52b4b950 | 3927 | spin_unlock_irq(&n->list_lock); |
60398923 | 3928 | |
916ac052 | 3929 | list_for_each_entry_safe(page, h, &discard, slab_list) |
60398923 | 3930 | discard_slab(s, page); |
81819f0f CL |
3931 | } |
3932 | ||
f9e13c0a SB |
3933 | bool __kmem_cache_empty(struct kmem_cache *s) |
3934 | { | |
3935 | int node; | |
3936 | struct kmem_cache_node *n; | |
3937 | ||
3938 | for_each_kmem_cache_node(s, node, n) | |
3939 | if (n->nr_partial || slabs_node(s, node)) | |
3940 | return false; | |
3941 | return true; | |
3942 | } | |
3943 | ||
81819f0f | 3944 | /* |
672bba3a | 3945 | * Release all resources used by a slab cache. |
81819f0f | 3946 | */ |
52b4b950 | 3947 | int __kmem_cache_shutdown(struct kmem_cache *s) |
81819f0f CL |
3948 | { |
3949 | int node; | |
fa45dc25 | 3950 | struct kmem_cache_node *n; |
81819f0f CL |
3951 | |
3952 | flush_all(s); | |
81819f0f | 3953 | /* Attempt to free all objects */ |
fa45dc25 | 3954 | for_each_kmem_cache_node(s, node, n) { |
599870b1 CL |
3955 | free_partial(s, n); |
3956 | if (n->nr_partial || slabs_node(s, node)) | |
81819f0f CL |
3957 | return 1; |
3958 | } | |
81819f0f CL |
3959 | return 0; |
3960 | } | |
3961 | ||
5bb1bb35 | 3962 | #ifdef CONFIG_PRINTK |
8e7f37f2 PM |
3963 | void kmem_obj_info(struct kmem_obj_info *kpp, void *object, struct page *page) |
3964 | { | |
3965 | void *base; | |
3966 | int __maybe_unused i; | |
3967 | unsigned int objnr; | |
3968 | void *objp; | |
3969 | void *objp0; | |
3970 | struct kmem_cache *s = page->slab_cache; | |
3971 | struct track __maybe_unused *trackp; | |
3972 | ||
3973 | kpp->kp_ptr = object; | |
3974 | kpp->kp_page = page; | |
3975 | kpp->kp_slab_cache = s; | |
3976 | base = page_address(page); | |
3977 | objp0 = kasan_reset_tag(object); | |
3978 | #ifdef CONFIG_SLUB_DEBUG | |
3979 | objp = restore_red_left(s, objp0); | |
3980 | #else | |
3981 | objp = objp0; | |
3982 | #endif | |
3983 | objnr = obj_to_index(s, page, objp); | |
3984 | kpp->kp_data_offset = (unsigned long)((char *)objp0 - (char *)objp); | |
3985 | objp = base + s->size * objnr; | |
3986 | kpp->kp_objp = objp; | |
3987 | if (WARN_ON_ONCE(objp < base || objp >= base + page->objects * s->size || (objp - base) % s->size) || | |
3988 | !(s->flags & SLAB_STORE_USER)) | |
3989 | return; | |
3990 | #ifdef CONFIG_SLUB_DEBUG | |
3991 | trackp = get_track(s, objp, TRACK_ALLOC); | |
3992 | kpp->kp_ret = (void *)trackp->addr; | |
3993 | #ifdef CONFIG_STACKTRACE | |
3994 | for (i = 0; i < KS_ADDRS_COUNT && i < TRACK_ADDRS_COUNT; i++) { | |
3995 | kpp->kp_stack[i] = (void *)trackp->addrs[i]; | |
3996 | if (!kpp->kp_stack[i]) | |
3997 | break; | |
3998 | } | |
3999 | #endif | |
4000 | #endif | |
4001 | } | |
5bb1bb35 | 4002 | #endif |
8e7f37f2 | 4003 | |
81819f0f CL |
4004 | /******************************************************************** |
4005 | * Kmalloc subsystem | |
4006 | *******************************************************************/ | |
4007 | ||
81819f0f CL |
4008 | static int __init setup_slub_min_order(char *str) |
4009 | { | |
19af27af | 4010 | get_option(&str, (int *)&slub_min_order); |
81819f0f CL |
4011 | |
4012 | return 1; | |
4013 | } | |
4014 | ||
4015 | __setup("slub_min_order=", setup_slub_min_order); | |
4016 | ||
4017 | static int __init setup_slub_max_order(char *str) | |
4018 | { | |
19af27af AD |
4019 | get_option(&str, (int *)&slub_max_order); |
4020 | slub_max_order = min(slub_max_order, (unsigned int)MAX_ORDER - 1); | |
81819f0f CL |
4021 | |
4022 | return 1; | |
4023 | } | |
4024 | ||
4025 | __setup("slub_max_order=", setup_slub_max_order); | |
4026 | ||
4027 | static int __init setup_slub_min_objects(char *str) | |
4028 | { | |
19af27af | 4029 | get_option(&str, (int *)&slub_min_objects); |
81819f0f CL |
4030 | |
4031 | return 1; | |
4032 | } | |
4033 | ||
4034 | __setup("slub_min_objects=", setup_slub_min_objects); | |
4035 | ||
81819f0f CL |
4036 | void *__kmalloc(size_t size, gfp_t flags) |
4037 | { | |
aadb4bc4 | 4038 | struct kmem_cache *s; |
5b882be4 | 4039 | void *ret; |
81819f0f | 4040 | |
95a05b42 | 4041 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) |
eada35ef | 4042 | return kmalloc_large(size, flags); |
aadb4bc4 | 4043 | |
2c59dd65 | 4044 | s = kmalloc_slab(size, flags); |
aadb4bc4 CL |
4045 | |
4046 | if (unlikely(ZERO_OR_NULL_PTR(s))) | |
6cb8f913 CL |
4047 | return s; |
4048 | ||
b89fb5ef | 4049 | ret = slab_alloc(s, flags, _RET_IP_, size); |
5b882be4 | 4050 | |
ca2b84cb | 4051 | trace_kmalloc(_RET_IP_, ret, size, s->size, flags); |
5b882be4 | 4052 | |
0116523c | 4053 | ret = kasan_kmalloc(s, ret, size, flags); |
0316bec2 | 4054 | |
5b882be4 | 4055 | return ret; |
81819f0f CL |
4056 | } |
4057 | EXPORT_SYMBOL(__kmalloc); | |
4058 | ||
5d1f57e4 | 4059 | #ifdef CONFIG_NUMA |
f619cfe1 CL |
4060 | static void *kmalloc_large_node(size_t size, gfp_t flags, int node) |
4061 | { | |
b1eeab67 | 4062 | struct page *page; |
e4f7c0b4 | 4063 | void *ptr = NULL; |
6a486c0a | 4064 | unsigned int order = get_order(size); |
f619cfe1 | 4065 | |
75f296d9 | 4066 | flags |= __GFP_COMP; |
6a486c0a VB |
4067 | page = alloc_pages_node(node, flags, order); |
4068 | if (page) { | |
e4f7c0b4 | 4069 | ptr = page_address(page); |
96403bfe MS |
4070 | mod_lruvec_page_state(page, NR_SLAB_UNRECLAIMABLE_B, |
4071 | PAGE_SIZE << order); | |
6a486c0a | 4072 | } |
e4f7c0b4 | 4073 | |
0116523c | 4074 | return kmalloc_large_node_hook(ptr, size, flags); |
f619cfe1 CL |
4075 | } |
4076 | ||
81819f0f CL |
4077 | void *__kmalloc_node(size_t size, gfp_t flags, int node) |
4078 | { | |
aadb4bc4 | 4079 | struct kmem_cache *s; |
5b882be4 | 4080 | void *ret; |
81819f0f | 4081 | |
95a05b42 | 4082 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) { |
5b882be4 EGM |
4083 | ret = kmalloc_large_node(size, flags, node); |
4084 | ||
ca2b84cb EGM |
4085 | trace_kmalloc_node(_RET_IP_, ret, |
4086 | size, PAGE_SIZE << get_order(size), | |
4087 | flags, node); | |
5b882be4 EGM |
4088 | |
4089 | return ret; | |
4090 | } | |
aadb4bc4 | 4091 | |
2c59dd65 | 4092 | s = kmalloc_slab(size, flags); |
aadb4bc4 CL |
4093 | |
4094 | if (unlikely(ZERO_OR_NULL_PTR(s))) | |
6cb8f913 CL |
4095 | return s; |
4096 | ||
b89fb5ef | 4097 | ret = slab_alloc_node(s, flags, node, _RET_IP_, size); |
5b882be4 | 4098 | |
ca2b84cb | 4099 | trace_kmalloc_node(_RET_IP_, ret, size, s->size, flags, node); |
5b882be4 | 4100 | |
0116523c | 4101 | ret = kasan_kmalloc(s, ret, size, flags); |
0316bec2 | 4102 | |
5b882be4 | 4103 | return ret; |
81819f0f CL |
4104 | } |
4105 | EXPORT_SYMBOL(__kmalloc_node); | |
6dfd1b65 | 4106 | #endif /* CONFIG_NUMA */ |
81819f0f | 4107 | |
ed18adc1 KC |
4108 | #ifdef CONFIG_HARDENED_USERCOPY |
4109 | /* | |
afcc90f8 KC |
4110 | * Rejects incorrectly sized objects and objects that are to be copied |
4111 | * to/from userspace but do not fall entirely within the containing slab | |
4112 | * cache's usercopy region. | |
ed18adc1 KC |
4113 | * |
4114 | * Returns NULL if check passes, otherwise const char * to name of cache | |
4115 | * to indicate an error. | |
4116 | */ | |
f4e6e289 KC |
4117 | void __check_heap_object(const void *ptr, unsigned long n, struct page *page, |
4118 | bool to_user) | |
ed18adc1 KC |
4119 | { |
4120 | struct kmem_cache *s; | |
44065b2e | 4121 | unsigned int offset; |
ed18adc1 | 4122 | size_t object_size; |
b89fb5ef | 4123 | bool is_kfence = is_kfence_address(ptr); |
ed18adc1 | 4124 | |
96fedce2 AK |
4125 | ptr = kasan_reset_tag(ptr); |
4126 | ||
ed18adc1 KC |
4127 | /* Find object and usable object size. */ |
4128 | s = page->slab_cache; | |
ed18adc1 KC |
4129 | |
4130 | /* Reject impossible pointers. */ | |
4131 | if (ptr < page_address(page)) | |
f4e6e289 KC |
4132 | usercopy_abort("SLUB object not in SLUB page?!", NULL, |
4133 | to_user, 0, n); | |
ed18adc1 KC |
4134 | |
4135 | /* Find offset within object. */ | |
b89fb5ef AP |
4136 | if (is_kfence) |
4137 | offset = ptr - kfence_object_start(ptr); | |
4138 | else | |
4139 | offset = (ptr - page_address(page)) % s->size; | |
ed18adc1 KC |
4140 | |
4141 | /* Adjust for redzone and reject if within the redzone. */ | |
b89fb5ef | 4142 | if (!is_kfence && kmem_cache_debug_flags(s, SLAB_RED_ZONE)) { |
ed18adc1 | 4143 | if (offset < s->red_left_pad) |
f4e6e289 KC |
4144 | usercopy_abort("SLUB object in left red zone", |
4145 | s->name, to_user, offset, n); | |
ed18adc1 KC |
4146 | offset -= s->red_left_pad; |
4147 | } | |
4148 | ||
afcc90f8 KC |
4149 | /* Allow address range falling entirely within usercopy region. */ |
4150 | if (offset >= s->useroffset && | |
4151 | offset - s->useroffset <= s->usersize && | |
4152 | n <= s->useroffset - offset + s->usersize) | |
f4e6e289 | 4153 | return; |
ed18adc1 | 4154 | |
afcc90f8 KC |
4155 | /* |
4156 | * If the copy is still within the allocated object, produce | |
4157 | * a warning instead of rejecting the copy. This is intended | |
4158 | * to be a temporary method to find any missing usercopy | |
4159 | * whitelists. | |
4160 | */ | |
4161 | object_size = slab_ksize(s); | |
2d891fbc KC |
4162 | if (usercopy_fallback && |
4163 | offset <= object_size && n <= object_size - offset) { | |
afcc90f8 KC |
4164 | usercopy_warn("SLUB object", s->name, to_user, offset, n); |
4165 | return; | |
4166 | } | |
ed18adc1 | 4167 | |
f4e6e289 | 4168 | usercopy_abort("SLUB object", s->name, to_user, offset, n); |
ed18adc1 KC |
4169 | } |
4170 | #endif /* CONFIG_HARDENED_USERCOPY */ | |
4171 | ||
10d1f8cb | 4172 | size_t __ksize(const void *object) |
81819f0f | 4173 | { |
272c1d21 | 4174 | struct page *page; |
81819f0f | 4175 | |
ef8b4520 | 4176 | if (unlikely(object == ZERO_SIZE_PTR)) |
272c1d21 CL |
4177 | return 0; |
4178 | ||
294a80a8 | 4179 | page = virt_to_head_page(object); |
294a80a8 | 4180 | |
76994412 PE |
4181 | if (unlikely(!PageSlab(page))) { |
4182 | WARN_ON(!PageCompound(page)); | |
a50b854e | 4183 | return page_size(page); |
76994412 | 4184 | } |
81819f0f | 4185 | |
1b4f59e3 | 4186 | return slab_ksize(page->slab_cache); |
81819f0f | 4187 | } |
10d1f8cb | 4188 | EXPORT_SYMBOL(__ksize); |
81819f0f CL |
4189 | |
4190 | void kfree(const void *x) | |
4191 | { | |
81819f0f | 4192 | struct page *page; |
5bb983b0 | 4193 | void *object = (void *)x; |
81819f0f | 4194 | |
2121db74 PE |
4195 | trace_kfree(_RET_IP_, x); |
4196 | ||
2408c550 | 4197 | if (unlikely(ZERO_OR_NULL_PTR(x))) |
81819f0f CL |
4198 | return; |
4199 | ||
b49af68f | 4200 | page = virt_to_head_page(x); |
aadb4bc4 | 4201 | if (unlikely(!PageSlab(page))) { |
6a486c0a VB |
4202 | unsigned int order = compound_order(page); |
4203 | ||
0937502a | 4204 | BUG_ON(!PageCompound(page)); |
47adccce | 4205 | kfree_hook(object); |
96403bfe MS |
4206 | mod_lruvec_page_state(page, NR_SLAB_UNRECLAIMABLE_B, |
4207 | -(PAGE_SIZE << order)); | |
6a486c0a | 4208 | __free_pages(page, order); |
aadb4bc4 CL |
4209 | return; |
4210 | } | |
81084651 | 4211 | slab_free(page->slab_cache, page, object, NULL, 1, _RET_IP_); |
81819f0f CL |
4212 | } |
4213 | EXPORT_SYMBOL(kfree); | |
4214 | ||
832f37f5 VD |
4215 | #define SHRINK_PROMOTE_MAX 32 |
4216 | ||
2086d26a | 4217 | /* |
832f37f5 VD |
4218 | * kmem_cache_shrink discards empty slabs and promotes the slabs filled |
4219 | * up most to the head of the partial lists. New allocations will then | |
4220 | * fill those up and thus they can be removed from the partial lists. | |
672bba3a CL |
4221 | * |
4222 | * The slabs with the least items are placed last. This results in them | |
4223 | * being allocated from last increasing the chance that the last objects | |
4224 | * are freed in them. | |
2086d26a | 4225 | */ |
c9fc5864 | 4226 | int __kmem_cache_shrink(struct kmem_cache *s) |
2086d26a CL |
4227 | { |
4228 | int node; | |
4229 | int i; | |
4230 | struct kmem_cache_node *n; | |
4231 | struct page *page; | |
4232 | struct page *t; | |
832f37f5 VD |
4233 | struct list_head discard; |
4234 | struct list_head promote[SHRINK_PROMOTE_MAX]; | |
2086d26a | 4235 | unsigned long flags; |
ce3712d7 | 4236 | int ret = 0; |
2086d26a | 4237 | |
2086d26a | 4238 | flush_all(s); |
fa45dc25 | 4239 | for_each_kmem_cache_node(s, node, n) { |
832f37f5 VD |
4240 | INIT_LIST_HEAD(&discard); |
4241 | for (i = 0; i < SHRINK_PROMOTE_MAX; i++) | |
4242 | INIT_LIST_HEAD(promote + i); | |
2086d26a CL |
4243 | |
4244 | spin_lock_irqsave(&n->list_lock, flags); | |
4245 | ||
4246 | /* | |
832f37f5 | 4247 | * Build lists of slabs to discard or promote. |
2086d26a | 4248 | * |
672bba3a CL |
4249 | * Note that concurrent frees may occur while we hold the |
4250 | * list_lock. page->inuse here is the upper limit. | |
2086d26a | 4251 | */ |
916ac052 | 4252 | list_for_each_entry_safe(page, t, &n->partial, slab_list) { |
832f37f5 VD |
4253 | int free = page->objects - page->inuse; |
4254 | ||
4255 | /* Do not reread page->inuse */ | |
4256 | barrier(); | |
4257 | ||
4258 | /* We do not keep full slabs on the list */ | |
4259 | BUG_ON(free <= 0); | |
4260 | ||
4261 | if (free == page->objects) { | |
916ac052 | 4262 | list_move(&page->slab_list, &discard); |
69cb8e6b | 4263 | n->nr_partial--; |
832f37f5 | 4264 | } else if (free <= SHRINK_PROMOTE_MAX) |
916ac052 | 4265 | list_move(&page->slab_list, promote + free - 1); |
2086d26a CL |
4266 | } |
4267 | ||
2086d26a | 4268 | /* |
832f37f5 VD |
4269 | * Promote the slabs filled up most to the head of the |
4270 | * partial list. | |
2086d26a | 4271 | */ |
832f37f5 VD |
4272 | for (i = SHRINK_PROMOTE_MAX - 1; i >= 0; i--) |
4273 | list_splice(promote + i, &n->partial); | |
2086d26a | 4274 | |
2086d26a | 4275 | spin_unlock_irqrestore(&n->list_lock, flags); |
69cb8e6b CL |
4276 | |
4277 | /* Release empty slabs */ | |
916ac052 | 4278 | list_for_each_entry_safe(page, t, &discard, slab_list) |
69cb8e6b | 4279 | discard_slab(s, page); |
ce3712d7 VD |
4280 | |
4281 | if (slabs_node(s, node)) | |
4282 | ret = 1; | |
2086d26a CL |
4283 | } |
4284 | ||
ce3712d7 | 4285 | return ret; |
2086d26a | 4286 | } |
2086d26a | 4287 | |
b9049e23 YG |
4288 | static int slab_mem_going_offline_callback(void *arg) |
4289 | { | |
4290 | struct kmem_cache *s; | |
4291 | ||
18004c5d | 4292 | mutex_lock(&slab_mutex); |
b9049e23 | 4293 | list_for_each_entry(s, &slab_caches, list) |
c9fc5864 | 4294 | __kmem_cache_shrink(s); |
18004c5d | 4295 | mutex_unlock(&slab_mutex); |
b9049e23 YG |
4296 | |
4297 | return 0; | |
4298 | } | |
4299 | ||
4300 | static void slab_mem_offline_callback(void *arg) | |
4301 | { | |
b9049e23 YG |
4302 | struct memory_notify *marg = arg; |
4303 | int offline_node; | |
4304 | ||
b9d5ab25 | 4305 | offline_node = marg->status_change_nid_normal; |
b9049e23 YG |
4306 | |
4307 | /* | |
4308 | * If the node still has available memory. we need kmem_cache_node | |
4309 | * for it yet. | |
4310 | */ | |
4311 | if (offline_node < 0) | |
4312 | return; | |
4313 | ||
18004c5d | 4314 | mutex_lock(&slab_mutex); |
7e1fa93d | 4315 | node_clear(offline_node, slab_nodes); |
666716fd VB |
4316 | /* |
4317 | * We no longer free kmem_cache_node structures here, as it would be | |
4318 | * racy with all get_node() users, and infeasible to protect them with | |
4319 | * slab_mutex. | |
4320 | */ | |
18004c5d | 4321 | mutex_unlock(&slab_mutex); |
b9049e23 YG |
4322 | } |
4323 | ||
4324 | static int slab_mem_going_online_callback(void *arg) | |
4325 | { | |
4326 | struct kmem_cache_node *n; | |
4327 | struct kmem_cache *s; | |
4328 | struct memory_notify *marg = arg; | |
b9d5ab25 | 4329 | int nid = marg->status_change_nid_normal; |
b9049e23 YG |
4330 | int ret = 0; |
4331 | ||
4332 | /* | |
4333 | * If the node's memory is already available, then kmem_cache_node is | |
4334 | * already created. Nothing to do. | |
4335 | */ | |
4336 | if (nid < 0) | |
4337 | return 0; | |
4338 | ||
4339 | /* | |
0121c619 | 4340 | * We are bringing a node online. No memory is available yet. We must |
b9049e23 YG |
4341 | * allocate a kmem_cache_node structure in order to bring the node |
4342 | * online. | |
4343 | */ | |
18004c5d | 4344 | mutex_lock(&slab_mutex); |
b9049e23 | 4345 | list_for_each_entry(s, &slab_caches, list) { |
666716fd VB |
4346 | /* |
4347 | * The structure may already exist if the node was previously | |
4348 | * onlined and offlined. | |
4349 | */ | |
4350 | if (get_node(s, nid)) | |
4351 | continue; | |
b9049e23 YG |
4352 | /* |
4353 | * XXX: kmem_cache_alloc_node will fallback to other nodes | |
4354 | * since memory is not yet available from the node that | |
4355 | * is brought up. | |
4356 | */ | |
8de66a0c | 4357 | n = kmem_cache_alloc(kmem_cache_node, GFP_KERNEL); |
b9049e23 YG |
4358 | if (!n) { |
4359 | ret = -ENOMEM; | |
4360 | goto out; | |
4361 | } | |
4053497d | 4362 | init_kmem_cache_node(n); |
b9049e23 YG |
4363 | s->node[nid] = n; |
4364 | } | |
7e1fa93d VB |
4365 | /* |
4366 | * Any cache created after this point will also have kmem_cache_node | |
4367 | * initialized for the new node. | |
4368 | */ | |
4369 | node_set(nid, slab_nodes); | |
b9049e23 | 4370 | out: |
18004c5d | 4371 | mutex_unlock(&slab_mutex); |
b9049e23 YG |
4372 | return ret; |
4373 | } | |
4374 | ||
4375 | static int slab_memory_callback(struct notifier_block *self, | |
4376 | unsigned long action, void *arg) | |
4377 | { | |
4378 | int ret = 0; | |
4379 | ||
4380 | switch (action) { | |
4381 | case MEM_GOING_ONLINE: | |
4382 | ret = slab_mem_going_online_callback(arg); | |
4383 | break; | |
4384 | case MEM_GOING_OFFLINE: | |
4385 | ret = slab_mem_going_offline_callback(arg); | |
4386 | break; | |
4387 | case MEM_OFFLINE: | |
4388 | case MEM_CANCEL_ONLINE: | |
4389 | slab_mem_offline_callback(arg); | |
4390 | break; | |
4391 | case MEM_ONLINE: | |
4392 | case MEM_CANCEL_OFFLINE: | |
4393 | break; | |
4394 | } | |
dc19f9db KH |
4395 | if (ret) |
4396 | ret = notifier_from_errno(ret); | |
4397 | else | |
4398 | ret = NOTIFY_OK; | |
b9049e23 YG |
4399 | return ret; |
4400 | } | |
4401 | ||
3ac38faa AM |
4402 | static struct notifier_block slab_memory_callback_nb = { |
4403 | .notifier_call = slab_memory_callback, | |
4404 | .priority = SLAB_CALLBACK_PRI, | |
4405 | }; | |
b9049e23 | 4406 | |
81819f0f CL |
4407 | /******************************************************************** |
4408 | * Basic setup of slabs | |
4409 | *******************************************************************/ | |
4410 | ||
51df1142 CL |
4411 | /* |
4412 | * Used for early kmem_cache structures that were allocated using | |
dffb4d60 CL |
4413 | * the page allocator. Allocate them properly then fix up the pointers |
4414 | * that may be pointing to the wrong kmem_cache structure. | |
51df1142 CL |
4415 | */ |
4416 | ||
dffb4d60 | 4417 | static struct kmem_cache * __init bootstrap(struct kmem_cache *static_cache) |
51df1142 CL |
4418 | { |
4419 | int node; | |
dffb4d60 | 4420 | struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT); |
fa45dc25 | 4421 | struct kmem_cache_node *n; |
51df1142 | 4422 | |
dffb4d60 | 4423 | memcpy(s, static_cache, kmem_cache->object_size); |
51df1142 | 4424 | |
7d557b3c GC |
4425 | /* |
4426 | * This runs very early, and only the boot processor is supposed to be | |
4427 | * up. Even if it weren't true, IRQs are not up so we couldn't fire | |
4428 | * IPIs around. | |
4429 | */ | |
4430 | __flush_cpu_slab(s, smp_processor_id()); | |
fa45dc25 | 4431 | for_each_kmem_cache_node(s, node, n) { |
51df1142 CL |
4432 | struct page *p; |
4433 | ||
916ac052 | 4434 | list_for_each_entry(p, &n->partial, slab_list) |
fa45dc25 | 4435 | p->slab_cache = s; |
51df1142 | 4436 | |
607bf324 | 4437 | #ifdef CONFIG_SLUB_DEBUG |
916ac052 | 4438 | list_for_each_entry(p, &n->full, slab_list) |
fa45dc25 | 4439 | p->slab_cache = s; |
51df1142 | 4440 | #endif |
51df1142 | 4441 | } |
dffb4d60 CL |
4442 | list_add(&s->list, &slab_caches); |
4443 | return s; | |
51df1142 CL |
4444 | } |
4445 | ||
81819f0f CL |
4446 | void __init kmem_cache_init(void) |
4447 | { | |
dffb4d60 CL |
4448 | static __initdata struct kmem_cache boot_kmem_cache, |
4449 | boot_kmem_cache_node; | |
7e1fa93d | 4450 | int node; |
51df1142 | 4451 | |
fc8d8620 SG |
4452 | if (debug_guardpage_minorder()) |
4453 | slub_max_order = 0; | |
4454 | ||
dffb4d60 CL |
4455 | kmem_cache_node = &boot_kmem_cache_node; |
4456 | kmem_cache = &boot_kmem_cache; | |
51df1142 | 4457 | |
7e1fa93d VB |
4458 | /* |
4459 | * Initialize the nodemask for which we will allocate per node | |
4460 | * structures. Here we don't need taking slab_mutex yet. | |
4461 | */ | |
4462 | for_each_node_state(node, N_NORMAL_MEMORY) | |
4463 | node_set(node, slab_nodes); | |
4464 | ||
dffb4d60 | 4465 | create_boot_cache(kmem_cache_node, "kmem_cache_node", |
8eb8284b | 4466 | sizeof(struct kmem_cache_node), SLAB_HWCACHE_ALIGN, 0, 0); |
b9049e23 | 4467 | |
3ac38faa | 4468 | register_hotmemory_notifier(&slab_memory_callback_nb); |
81819f0f CL |
4469 | |
4470 | /* Able to allocate the per node structures */ | |
4471 | slab_state = PARTIAL; | |
4472 | ||
dffb4d60 CL |
4473 | create_boot_cache(kmem_cache, "kmem_cache", |
4474 | offsetof(struct kmem_cache, node) + | |
4475 | nr_node_ids * sizeof(struct kmem_cache_node *), | |
8eb8284b | 4476 | SLAB_HWCACHE_ALIGN, 0, 0); |
8a13a4cc | 4477 | |
dffb4d60 | 4478 | kmem_cache = bootstrap(&boot_kmem_cache); |
dffb4d60 | 4479 | kmem_cache_node = bootstrap(&boot_kmem_cache_node); |
51df1142 CL |
4480 | |
4481 | /* Now we can use the kmem_cache to allocate kmalloc slabs */ | |
34cc6990 | 4482 | setup_kmalloc_cache_index_table(); |
f97d5f63 | 4483 | create_kmalloc_caches(0); |
81819f0f | 4484 | |
210e7a43 TG |
4485 | /* Setup random freelists for each cache */ |
4486 | init_freelist_randomization(); | |
4487 | ||
a96a87bf SAS |
4488 | cpuhp_setup_state_nocalls(CPUHP_SLUB_DEAD, "slub:dead", NULL, |
4489 | slub_cpu_dead); | |
81819f0f | 4490 | |
b9726c26 | 4491 | pr_info("SLUB: HWalign=%d, Order=%u-%u, MinObjects=%u, CPUs=%u, Nodes=%u\n", |
f97d5f63 | 4492 | cache_line_size(), |
81819f0f CL |
4493 | slub_min_order, slub_max_order, slub_min_objects, |
4494 | nr_cpu_ids, nr_node_ids); | |
4495 | } | |
4496 | ||
7e85ee0c PE |
4497 | void __init kmem_cache_init_late(void) |
4498 | { | |
7e85ee0c PE |
4499 | } |
4500 | ||
2633d7a0 | 4501 | struct kmem_cache * |
f4957d5b | 4502 | __kmem_cache_alias(const char *name, unsigned int size, unsigned int align, |
d50112ed | 4503 | slab_flags_t flags, void (*ctor)(void *)) |
81819f0f | 4504 | { |
10befea9 | 4505 | struct kmem_cache *s; |
81819f0f | 4506 | |
a44cb944 | 4507 | s = find_mergeable(size, align, flags, name, ctor); |
81819f0f CL |
4508 | if (s) { |
4509 | s->refcount++; | |
84d0ddd6 | 4510 | |
81819f0f CL |
4511 | /* |
4512 | * Adjust the object sizes so that we clear | |
4513 | * the complete object on kzalloc. | |
4514 | */ | |
1b473f29 | 4515 | s->object_size = max(s->object_size, size); |
52ee6d74 | 4516 | s->inuse = max(s->inuse, ALIGN(size, sizeof(void *))); |
6446faa2 | 4517 | |
7b8f3b66 | 4518 | if (sysfs_slab_alias(s, name)) { |
7b8f3b66 | 4519 | s->refcount--; |
cbb79694 | 4520 | s = NULL; |
7b8f3b66 | 4521 | } |
a0e1d1be | 4522 | } |
6446faa2 | 4523 | |
cbb79694 CL |
4524 | return s; |
4525 | } | |
84c1cf62 | 4526 | |
d50112ed | 4527 | int __kmem_cache_create(struct kmem_cache *s, slab_flags_t flags) |
cbb79694 | 4528 | { |
aac3a166 PE |
4529 | int err; |
4530 | ||
4531 | err = kmem_cache_open(s, flags); | |
4532 | if (err) | |
4533 | return err; | |
20cea968 | 4534 | |
45530c44 CL |
4535 | /* Mutex is not taken during early boot */ |
4536 | if (slab_state <= UP) | |
4537 | return 0; | |
4538 | ||
aac3a166 | 4539 | err = sysfs_slab_add(s); |
aac3a166 | 4540 | if (err) |
52b4b950 | 4541 | __kmem_cache_release(s); |
20cea968 | 4542 | |
aac3a166 | 4543 | return err; |
81819f0f | 4544 | } |
81819f0f | 4545 | |
ce71e27c | 4546 | void *__kmalloc_track_caller(size_t size, gfp_t gfpflags, unsigned long caller) |
81819f0f | 4547 | { |
aadb4bc4 | 4548 | struct kmem_cache *s; |
94b528d0 | 4549 | void *ret; |
aadb4bc4 | 4550 | |
95a05b42 | 4551 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) |
eada35ef PE |
4552 | return kmalloc_large(size, gfpflags); |
4553 | ||
2c59dd65 | 4554 | s = kmalloc_slab(size, gfpflags); |
81819f0f | 4555 | |
2408c550 | 4556 | if (unlikely(ZERO_OR_NULL_PTR(s))) |
6cb8f913 | 4557 | return s; |
81819f0f | 4558 | |
b89fb5ef | 4559 | ret = slab_alloc(s, gfpflags, caller, size); |
94b528d0 | 4560 | |
25985edc | 4561 | /* Honor the call site pointer we received. */ |
ca2b84cb | 4562 | trace_kmalloc(caller, ret, size, s->size, gfpflags); |
94b528d0 EGM |
4563 | |
4564 | return ret; | |
81819f0f | 4565 | } |
fd7cb575 | 4566 | EXPORT_SYMBOL(__kmalloc_track_caller); |
81819f0f | 4567 | |
5d1f57e4 | 4568 | #ifdef CONFIG_NUMA |
81819f0f | 4569 | void *__kmalloc_node_track_caller(size_t size, gfp_t gfpflags, |
ce71e27c | 4570 | int node, unsigned long caller) |
81819f0f | 4571 | { |
aadb4bc4 | 4572 | struct kmem_cache *s; |
94b528d0 | 4573 | void *ret; |
aadb4bc4 | 4574 | |
95a05b42 | 4575 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) { |
d3e14aa3 XF |
4576 | ret = kmalloc_large_node(size, gfpflags, node); |
4577 | ||
4578 | trace_kmalloc_node(caller, ret, | |
4579 | size, PAGE_SIZE << get_order(size), | |
4580 | gfpflags, node); | |
4581 | ||
4582 | return ret; | |
4583 | } | |
eada35ef | 4584 | |
2c59dd65 | 4585 | s = kmalloc_slab(size, gfpflags); |
81819f0f | 4586 | |
2408c550 | 4587 | if (unlikely(ZERO_OR_NULL_PTR(s))) |
6cb8f913 | 4588 | return s; |
81819f0f | 4589 | |
b89fb5ef | 4590 | ret = slab_alloc_node(s, gfpflags, node, caller, size); |
94b528d0 | 4591 | |
25985edc | 4592 | /* Honor the call site pointer we received. */ |
ca2b84cb | 4593 | trace_kmalloc_node(caller, ret, size, s->size, gfpflags, node); |
94b528d0 EGM |
4594 | |
4595 | return ret; | |
81819f0f | 4596 | } |
fd7cb575 | 4597 | EXPORT_SYMBOL(__kmalloc_node_track_caller); |
5d1f57e4 | 4598 | #endif |
81819f0f | 4599 | |
ab4d5ed5 | 4600 | #ifdef CONFIG_SYSFS |
205ab99d CL |
4601 | static int count_inuse(struct page *page) |
4602 | { | |
4603 | return page->inuse; | |
4604 | } | |
4605 | ||
4606 | static int count_total(struct page *page) | |
4607 | { | |
4608 | return page->objects; | |
4609 | } | |
ab4d5ed5 | 4610 | #endif |
205ab99d | 4611 | |
ab4d5ed5 | 4612 | #ifdef CONFIG_SLUB_DEBUG |
90e9f6a6 | 4613 | static void validate_slab(struct kmem_cache *s, struct page *page) |
53e15af0 CL |
4614 | { |
4615 | void *p; | |
a973e9dd | 4616 | void *addr = page_address(page); |
90e9f6a6 YZ |
4617 | unsigned long *map; |
4618 | ||
4619 | slab_lock(page); | |
53e15af0 | 4620 | |
dd98afd4 | 4621 | if (!check_slab(s, page) || !on_freelist(s, page, NULL)) |
90e9f6a6 | 4622 | goto unlock; |
53e15af0 CL |
4623 | |
4624 | /* Now we know that a valid freelist exists */ | |
90e9f6a6 | 4625 | map = get_map(s, page); |
5f80b13a | 4626 | for_each_object(p, s, addr, page->objects) { |
4138fdfc | 4627 | u8 val = test_bit(__obj_to_index(s, addr, p), map) ? |
dd98afd4 | 4628 | SLUB_RED_INACTIVE : SLUB_RED_ACTIVE; |
53e15af0 | 4629 | |
dd98afd4 YZ |
4630 | if (!check_object(s, page, p, val)) |
4631 | break; | |
4632 | } | |
90e9f6a6 YZ |
4633 | put_map(map); |
4634 | unlock: | |
881db7fb | 4635 | slab_unlock(page); |
53e15af0 CL |
4636 | } |
4637 | ||
434e245d | 4638 | static int validate_slab_node(struct kmem_cache *s, |
90e9f6a6 | 4639 | struct kmem_cache_node *n) |
53e15af0 CL |
4640 | { |
4641 | unsigned long count = 0; | |
4642 | struct page *page; | |
4643 | unsigned long flags; | |
4644 | ||
4645 | spin_lock_irqsave(&n->list_lock, flags); | |
4646 | ||
916ac052 | 4647 | list_for_each_entry(page, &n->partial, slab_list) { |
90e9f6a6 | 4648 | validate_slab(s, page); |
53e15af0 CL |
4649 | count++; |
4650 | } | |
4651 | if (count != n->nr_partial) | |
f9f58285 FF |
4652 | pr_err("SLUB %s: %ld partial slabs counted but counter=%ld\n", |
4653 | s->name, count, n->nr_partial); | |
53e15af0 CL |
4654 | |
4655 | if (!(s->flags & SLAB_STORE_USER)) | |
4656 | goto out; | |
4657 | ||
916ac052 | 4658 | list_for_each_entry(page, &n->full, slab_list) { |
90e9f6a6 | 4659 | validate_slab(s, page); |
53e15af0 CL |
4660 | count++; |
4661 | } | |
4662 | if (count != atomic_long_read(&n->nr_slabs)) | |
f9f58285 FF |
4663 | pr_err("SLUB: %s %ld slabs counted but counter=%ld\n", |
4664 | s->name, count, atomic_long_read(&n->nr_slabs)); | |
53e15af0 CL |
4665 | |
4666 | out: | |
4667 | spin_unlock_irqrestore(&n->list_lock, flags); | |
4668 | return count; | |
4669 | } | |
4670 | ||
434e245d | 4671 | static long validate_slab_cache(struct kmem_cache *s) |
53e15af0 CL |
4672 | { |
4673 | int node; | |
4674 | unsigned long count = 0; | |
fa45dc25 | 4675 | struct kmem_cache_node *n; |
53e15af0 CL |
4676 | |
4677 | flush_all(s); | |
fa45dc25 | 4678 | for_each_kmem_cache_node(s, node, n) |
90e9f6a6 YZ |
4679 | count += validate_slab_node(s, n); |
4680 | ||
53e15af0 CL |
4681 | return count; |
4682 | } | |
88a420e4 | 4683 | /* |
672bba3a | 4684 | * Generate lists of code addresses where slabcache objects are allocated |
88a420e4 CL |
4685 | * and freed. |
4686 | */ | |
4687 | ||
4688 | struct location { | |
4689 | unsigned long count; | |
ce71e27c | 4690 | unsigned long addr; |
45edfa58 CL |
4691 | long long sum_time; |
4692 | long min_time; | |
4693 | long max_time; | |
4694 | long min_pid; | |
4695 | long max_pid; | |
174596a0 | 4696 | DECLARE_BITMAP(cpus, NR_CPUS); |
45edfa58 | 4697 | nodemask_t nodes; |
88a420e4 CL |
4698 | }; |
4699 | ||
4700 | struct loc_track { | |
4701 | unsigned long max; | |
4702 | unsigned long count; | |
4703 | struct location *loc; | |
4704 | }; | |
4705 | ||
4706 | static void free_loc_track(struct loc_track *t) | |
4707 | { | |
4708 | if (t->max) | |
4709 | free_pages((unsigned long)t->loc, | |
4710 | get_order(sizeof(struct location) * t->max)); | |
4711 | } | |
4712 | ||
68dff6a9 | 4713 | static int alloc_loc_track(struct loc_track *t, unsigned long max, gfp_t flags) |
88a420e4 CL |
4714 | { |
4715 | struct location *l; | |
4716 | int order; | |
4717 | ||
88a420e4 CL |
4718 | order = get_order(sizeof(struct location) * max); |
4719 | ||
68dff6a9 | 4720 | l = (void *)__get_free_pages(flags, order); |
88a420e4 CL |
4721 | if (!l) |
4722 | return 0; | |
4723 | ||
4724 | if (t->count) { | |
4725 | memcpy(l, t->loc, sizeof(struct location) * t->count); | |
4726 | free_loc_track(t); | |
4727 | } | |
4728 | t->max = max; | |
4729 | t->loc = l; | |
4730 | return 1; | |
4731 | } | |
4732 | ||
4733 | static int add_location(struct loc_track *t, struct kmem_cache *s, | |
45edfa58 | 4734 | const struct track *track) |
88a420e4 CL |
4735 | { |
4736 | long start, end, pos; | |
4737 | struct location *l; | |
ce71e27c | 4738 | unsigned long caddr; |
45edfa58 | 4739 | unsigned long age = jiffies - track->when; |
88a420e4 CL |
4740 | |
4741 | start = -1; | |
4742 | end = t->count; | |
4743 | ||
4744 | for ( ; ; ) { | |
4745 | pos = start + (end - start + 1) / 2; | |
4746 | ||
4747 | /* | |
4748 | * There is nothing at "end". If we end up there | |
4749 | * we need to add something to before end. | |
4750 | */ | |
4751 | if (pos == end) | |
4752 | break; | |
4753 | ||
4754 | caddr = t->loc[pos].addr; | |
45edfa58 CL |
4755 | if (track->addr == caddr) { |
4756 | ||
4757 | l = &t->loc[pos]; | |
4758 | l->count++; | |
4759 | if (track->when) { | |
4760 | l->sum_time += age; | |
4761 | if (age < l->min_time) | |
4762 | l->min_time = age; | |
4763 | if (age > l->max_time) | |
4764 | l->max_time = age; | |
4765 | ||
4766 | if (track->pid < l->min_pid) | |
4767 | l->min_pid = track->pid; | |
4768 | if (track->pid > l->max_pid) | |
4769 | l->max_pid = track->pid; | |
4770 | ||
174596a0 RR |
4771 | cpumask_set_cpu(track->cpu, |
4772 | to_cpumask(l->cpus)); | |
45edfa58 CL |
4773 | } |
4774 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
88a420e4 CL |
4775 | return 1; |
4776 | } | |
4777 | ||
45edfa58 | 4778 | if (track->addr < caddr) |
88a420e4 CL |
4779 | end = pos; |
4780 | else | |
4781 | start = pos; | |
4782 | } | |
4783 | ||
4784 | /* | |
672bba3a | 4785 | * Not found. Insert new tracking element. |
88a420e4 | 4786 | */ |
68dff6a9 | 4787 | if (t->count >= t->max && !alloc_loc_track(t, 2 * t->max, GFP_ATOMIC)) |
88a420e4 CL |
4788 | return 0; |
4789 | ||
4790 | l = t->loc + pos; | |
4791 | if (pos < t->count) | |
4792 | memmove(l + 1, l, | |
4793 | (t->count - pos) * sizeof(struct location)); | |
4794 | t->count++; | |
4795 | l->count = 1; | |
45edfa58 CL |
4796 | l->addr = track->addr; |
4797 | l->sum_time = age; | |
4798 | l->min_time = age; | |
4799 | l->max_time = age; | |
4800 | l->min_pid = track->pid; | |
4801 | l->max_pid = track->pid; | |
174596a0 RR |
4802 | cpumask_clear(to_cpumask(l->cpus)); |
4803 | cpumask_set_cpu(track->cpu, to_cpumask(l->cpus)); | |
45edfa58 CL |
4804 | nodes_clear(l->nodes); |
4805 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
88a420e4 CL |
4806 | return 1; |
4807 | } | |
4808 | ||
4809 | static void process_slab(struct loc_track *t, struct kmem_cache *s, | |
90e9f6a6 | 4810 | struct page *page, enum track_item alloc) |
88a420e4 | 4811 | { |
a973e9dd | 4812 | void *addr = page_address(page); |
88a420e4 | 4813 | void *p; |
90e9f6a6 | 4814 | unsigned long *map; |
88a420e4 | 4815 | |
90e9f6a6 | 4816 | map = get_map(s, page); |
224a88be | 4817 | for_each_object(p, s, addr, page->objects) |
4138fdfc | 4818 | if (!test_bit(__obj_to_index(s, addr, p), map)) |
45edfa58 | 4819 | add_location(t, s, get_track(s, p, alloc)); |
90e9f6a6 | 4820 | put_map(map); |
88a420e4 CL |
4821 | } |
4822 | ||
4823 | static int list_locations(struct kmem_cache *s, char *buf, | |
bf16d19a | 4824 | enum track_item alloc) |
88a420e4 | 4825 | { |
e374d483 | 4826 | int len = 0; |
88a420e4 | 4827 | unsigned long i; |
68dff6a9 | 4828 | struct loc_track t = { 0, 0, NULL }; |
88a420e4 | 4829 | int node; |
fa45dc25 | 4830 | struct kmem_cache_node *n; |
88a420e4 | 4831 | |
90e9f6a6 YZ |
4832 | if (!alloc_loc_track(&t, PAGE_SIZE / sizeof(struct location), |
4833 | GFP_KERNEL)) { | |
bf16d19a | 4834 | return sysfs_emit(buf, "Out of memory\n"); |
bbd7d57b | 4835 | } |
88a420e4 CL |
4836 | /* Push back cpu slabs */ |
4837 | flush_all(s); | |
4838 | ||
fa45dc25 | 4839 | for_each_kmem_cache_node(s, node, n) { |
88a420e4 CL |
4840 | unsigned long flags; |
4841 | struct page *page; | |
4842 | ||
9e86943b | 4843 | if (!atomic_long_read(&n->nr_slabs)) |
88a420e4 CL |
4844 | continue; |
4845 | ||
4846 | spin_lock_irqsave(&n->list_lock, flags); | |
916ac052 | 4847 | list_for_each_entry(page, &n->partial, slab_list) |
90e9f6a6 | 4848 | process_slab(&t, s, page, alloc); |
916ac052 | 4849 | list_for_each_entry(page, &n->full, slab_list) |
90e9f6a6 | 4850 | process_slab(&t, s, page, alloc); |
88a420e4 CL |
4851 | spin_unlock_irqrestore(&n->list_lock, flags); |
4852 | } | |
4853 | ||
4854 | for (i = 0; i < t.count; i++) { | |
45edfa58 | 4855 | struct location *l = &t.loc[i]; |
88a420e4 | 4856 | |
bf16d19a | 4857 | len += sysfs_emit_at(buf, len, "%7ld ", l->count); |
45edfa58 CL |
4858 | |
4859 | if (l->addr) | |
bf16d19a | 4860 | len += sysfs_emit_at(buf, len, "%pS", (void *)l->addr); |
88a420e4 | 4861 | else |
bf16d19a JP |
4862 | len += sysfs_emit_at(buf, len, "<not-available>"); |
4863 | ||
4864 | if (l->sum_time != l->min_time) | |
4865 | len += sysfs_emit_at(buf, len, " age=%ld/%ld/%ld", | |
4866 | l->min_time, | |
4867 | (long)div_u64(l->sum_time, | |
4868 | l->count), | |
4869 | l->max_time); | |
4870 | else | |
4871 | len += sysfs_emit_at(buf, len, " age=%ld", l->min_time); | |
45edfa58 CL |
4872 | |
4873 | if (l->min_pid != l->max_pid) | |
bf16d19a JP |
4874 | len += sysfs_emit_at(buf, len, " pid=%ld-%ld", |
4875 | l->min_pid, l->max_pid); | |
45edfa58 | 4876 | else |
bf16d19a JP |
4877 | len += sysfs_emit_at(buf, len, " pid=%ld", |
4878 | l->min_pid); | |
45edfa58 | 4879 | |
174596a0 | 4880 | if (num_online_cpus() > 1 && |
bf16d19a JP |
4881 | !cpumask_empty(to_cpumask(l->cpus))) |
4882 | len += sysfs_emit_at(buf, len, " cpus=%*pbl", | |
4883 | cpumask_pr_args(to_cpumask(l->cpus))); | |
4884 | ||
4885 | if (nr_online_nodes > 1 && !nodes_empty(l->nodes)) | |
4886 | len += sysfs_emit_at(buf, len, " nodes=%*pbl", | |
4887 | nodemask_pr_args(&l->nodes)); | |
4888 | ||
4889 | len += sysfs_emit_at(buf, len, "\n"); | |
88a420e4 CL |
4890 | } |
4891 | ||
4892 | free_loc_track(&t); | |
4893 | if (!t.count) | |
bf16d19a JP |
4894 | len += sysfs_emit_at(buf, len, "No data\n"); |
4895 | ||
e374d483 | 4896 | return len; |
88a420e4 | 4897 | } |
6dfd1b65 | 4898 | #endif /* CONFIG_SLUB_DEBUG */ |
88a420e4 | 4899 | |
a5a84755 | 4900 | #ifdef SLUB_RESILIENCY_TEST |
c07b8183 | 4901 | static void __init resiliency_test(void) |
a5a84755 CL |
4902 | { |
4903 | u8 *p; | |
cc252eae | 4904 | int type = KMALLOC_NORMAL; |
a5a84755 | 4905 | |
95a05b42 | 4906 | BUILD_BUG_ON(KMALLOC_MIN_SIZE > 16 || KMALLOC_SHIFT_HIGH < 10); |
a5a84755 | 4907 | |
f9f58285 FF |
4908 | pr_err("SLUB resiliency testing\n"); |
4909 | pr_err("-----------------------\n"); | |
4910 | pr_err("A. Corruption after allocation\n"); | |
a5a84755 CL |
4911 | |
4912 | p = kzalloc(16, GFP_KERNEL); | |
4913 | p[16] = 0x12; | |
f9f58285 FF |
4914 | pr_err("\n1. kmalloc-16: Clobber Redzone/next pointer 0x12->0x%p\n\n", |
4915 | p + 16); | |
a5a84755 | 4916 | |
cc252eae | 4917 | validate_slab_cache(kmalloc_caches[type][4]); |
a5a84755 CL |
4918 | |
4919 | /* Hmmm... The next two are dangerous */ | |
4920 | p = kzalloc(32, GFP_KERNEL); | |
4921 | p[32 + sizeof(void *)] = 0x34; | |
f9f58285 FF |
4922 | pr_err("\n2. kmalloc-32: Clobber next pointer/next slab 0x34 -> -0x%p\n", |
4923 | p); | |
4924 | pr_err("If allocated object is overwritten then not detectable\n\n"); | |
a5a84755 | 4925 | |
cc252eae | 4926 | validate_slab_cache(kmalloc_caches[type][5]); |
a5a84755 CL |
4927 | p = kzalloc(64, GFP_KERNEL); |
4928 | p += 64 + (get_cycles() & 0xff) * sizeof(void *); | |
4929 | *p = 0x56; | |
f9f58285 FF |
4930 | pr_err("\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n", |
4931 | p); | |
4932 | pr_err("If allocated object is overwritten then not detectable\n\n"); | |
cc252eae | 4933 | validate_slab_cache(kmalloc_caches[type][6]); |
a5a84755 | 4934 | |
f9f58285 | 4935 | pr_err("\nB. Corruption after free\n"); |
a5a84755 CL |
4936 | p = kzalloc(128, GFP_KERNEL); |
4937 | kfree(p); | |
4938 | *p = 0x78; | |
f9f58285 | 4939 | pr_err("1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p); |
cc252eae | 4940 | validate_slab_cache(kmalloc_caches[type][7]); |
a5a84755 CL |
4941 | |
4942 | p = kzalloc(256, GFP_KERNEL); | |
4943 | kfree(p); | |
4944 | p[50] = 0x9a; | |
f9f58285 | 4945 | pr_err("\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n", p); |
cc252eae | 4946 | validate_slab_cache(kmalloc_caches[type][8]); |
a5a84755 CL |
4947 | |
4948 | p = kzalloc(512, GFP_KERNEL); | |
4949 | kfree(p); | |
4950 | p[512] = 0xab; | |
f9f58285 | 4951 | pr_err("\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p); |
cc252eae | 4952 | validate_slab_cache(kmalloc_caches[type][9]); |
a5a84755 CL |
4953 | } |
4954 | #else | |
4955 | #ifdef CONFIG_SYSFS | |
4956 | static void resiliency_test(void) {}; | |
4957 | #endif | |
6dfd1b65 | 4958 | #endif /* SLUB_RESILIENCY_TEST */ |
a5a84755 | 4959 | |
ab4d5ed5 | 4960 | #ifdef CONFIG_SYSFS |
81819f0f | 4961 | enum slab_stat_type { |
205ab99d CL |
4962 | SL_ALL, /* All slabs */ |
4963 | SL_PARTIAL, /* Only partially allocated slabs */ | |
4964 | SL_CPU, /* Only slabs used for cpu caches */ | |
4965 | SL_OBJECTS, /* Determine allocated objects not slabs */ | |
4966 | SL_TOTAL /* Determine object capacity not slabs */ | |
81819f0f CL |
4967 | }; |
4968 | ||
205ab99d | 4969 | #define SO_ALL (1 << SL_ALL) |
81819f0f CL |
4970 | #define SO_PARTIAL (1 << SL_PARTIAL) |
4971 | #define SO_CPU (1 << SL_CPU) | |
4972 | #define SO_OBJECTS (1 << SL_OBJECTS) | |
205ab99d | 4973 | #define SO_TOTAL (1 << SL_TOTAL) |
81819f0f | 4974 | |
62e5c4b4 | 4975 | static ssize_t show_slab_objects(struct kmem_cache *s, |
bf16d19a | 4976 | char *buf, unsigned long flags) |
81819f0f CL |
4977 | { |
4978 | unsigned long total = 0; | |
81819f0f CL |
4979 | int node; |
4980 | int x; | |
4981 | unsigned long *nodes; | |
bf16d19a | 4982 | int len = 0; |
81819f0f | 4983 | |
6396bb22 | 4984 | nodes = kcalloc(nr_node_ids, sizeof(unsigned long), GFP_KERNEL); |
62e5c4b4 CG |
4985 | if (!nodes) |
4986 | return -ENOMEM; | |
81819f0f | 4987 | |
205ab99d CL |
4988 | if (flags & SO_CPU) { |
4989 | int cpu; | |
81819f0f | 4990 | |
205ab99d | 4991 | for_each_possible_cpu(cpu) { |
d0e0ac97 CG |
4992 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, |
4993 | cpu); | |
ec3ab083 | 4994 | int node; |
49e22585 | 4995 | struct page *page; |
dfb4f096 | 4996 | |
4db0c3c2 | 4997 | page = READ_ONCE(c->page); |
ec3ab083 CL |
4998 | if (!page) |
4999 | continue; | |
205ab99d | 5000 | |
ec3ab083 CL |
5001 | node = page_to_nid(page); |
5002 | if (flags & SO_TOTAL) | |
5003 | x = page->objects; | |
5004 | else if (flags & SO_OBJECTS) | |
5005 | x = page->inuse; | |
5006 | else | |
5007 | x = 1; | |
49e22585 | 5008 | |
ec3ab083 CL |
5009 | total += x; |
5010 | nodes[node] += x; | |
5011 | ||
a93cf07b | 5012 | page = slub_percpu_partial_read_once(c); |
49e22585 | 5013 | if (page) { |
8afb1474 LZ |
5014 | node = page_to_nid(page); |
5015 | if (flags & SO_TOTAL) | |
5016 | WARN_ON_ONCE(1); | |
5017 | else if (flags & SO_OBJECTS) | |
5018 | WARN_ON_ONCE(1); | |
5019 | else | |
5020 | x = page->pages; | |
bc6697d8 ED |
5021 | total += x; |
5022 | nodes[node] += x; | |
49e22585 | 5023 | } |
81819f0f CL |
5024 | } |
5025 | } | |
5026 | ||
e4f8e513 QC |
5027 | /* |
5028 | * It is impossible to take "mem_hotplug_lock" here with "kernfs_mutex" | |
5029 | * already held which will conflict with an existing lock order: | |
5030 | * | |
5031 | * mem_hotplug_lock->slab_mutex->kernfs_mutex | |
5032 | * | |
5033 | * We don't really need mem_hotplug_lock (to hold off | |
5034 | * slab_mem_going_offline_callback) here because slab's memory hot | |
5035 | * unplug code doesn't destroy the kmem_cache->node[] data. | |
5036 | */ | |
5037 | ||
ab4d5ed5 | 5038 | #ifdef CONFIG_SLUB_DEBUG |
205ab99d | 5039 | if (flags & SO_ALL) { |
fa45dc25 CL |
5040 | struct kmem_cache_node *n; |
5041 | ||
5042 | for_each_kmem_cache_node(s, node, n) { | |
205ab99d | 5043 | |
d0e0ac97 CG |
5044 | if (flags & SO_TOTAL) |
5045 | x = atomic_long_read(&n->total_objects); | |
5046 | else if (flags & SO_OBJECTS) | |
5047 | x = atomic_long_read(&n->total_objects) - | |
5048 | count_partial(n, count_free); | |
81819f0f | 5049 | else |
205ab99d | 5050 | x = atomic_long_read(&n->nr_slabs); |
81819f0f CL |
5051 | total += x; |
5052 | nodes[node] += x; | |
5053 | } | |
5054 | ||
ab4d5ed5 CL |
5055 | } else |
5056 | #endif | |
5057 | if (flags & SO_PARTIAL) { | |
fa45dc25 | 5058 | struct kmem_cache_node *n; |
81819f0f | 5059 | |
fa45dc25 | 5060 | for_each_kmem_cache_node(s, node, n) { |
205ab99d CL |
5061 | if (flags & SO_TOTAL) |
5062 | x = count_partial(n, count_total); | |
5063 | else if (flags & SO_OBJECTS) | |
5064 | x = count_partial(n, count_inuse); | |
81819f0f | 5065 | else |
205ab99d | 5066 | x = n->nr_partial; |
81819f0f CL |
5067 | total += x; |
5068 | nodes[node] += x; | |
5069 | } | |
5070 | } | |
bf16d19a JP |
5071 | |
5072 | len += sysfs_emit_at(buf, len, "%lu", total); | |
81819f0f | 5073 | #ifdef CONFIG_NUMA |
bf16d19a | 5074 | for (node = 0; node < nr_node_ids; node++) { |
81819f0f | 5075 | if (nodes[node]) |
bf16d19a JP |
5076 | len += sysfs_emit_at(buf, len, " N%d=%lu", |
5077 | node, nodes[node]); | |
5078 | } | |
81819f0f | 5079 | #endif |
bf16d19a | 5080 | len += sysfs_emit_at(buf, len, "\n"); |
81819f0f | 5081 | kfree(nodes); |
bf16d19a JP |
5082 | |
5083 | return len; | |
81819f0f CL |
5084 | } |
5085 | ||
81819f0f | 5086 | #define to_slab_attr(n) container_of(n, struct slab_attribute, attr) |
497888cf | 5087 | #define to_slab(n) container_of(n, struct kmem_cache, kobj) |
81819f0f CL |
5088 | |
5089 | struct slab_attribute { | |
5090 | struct attribute attr; | |
5091 | ssize_t (*show)(struct kmem_cache *s, char *buf); | |
5092 | ssize_t (*store)(struct kmem_cache *s, const char *x, size_t count); | |
5093 | }; | |
5094 | ||
5095 | #define SLAB_ATTR_RO(_name) \ | |
ab067e99 VK |
5096 | static struct slab_attribute _name##_attr = \ |
5097 | __ATTR(_name, 0400, _name##_show, NULL) | |
81819f0f CL |
5098 | |
5099 | #define SLAB_ATTR(_name) \ | |
5100 | static struct slab_attribute _name##_attr = \ | |
ab067e99 | 5101 | __ATTR(_name, 0600, _name##_show, _name##_store) |
81819f0f | 5102 | |
81819f0f CL |
5103 | static ssize_t slab_size_show(struct kmem_cache *s, char *buf) |
5104 | { | |
bf16d19a | 5105 | return sysfs_emit(buf, "%u\n", s->size); |
81819f0f CL |
5106 | } |
5107 | SLAB_ATTR_RO(slab_size); | |
5108 | ||
5109 | static ssize_t align_show(struct kmem_cache *s, char *buf) | |
5110 | { | |
bf16d19a | 5111 | return sysfs_emit(buf, "%u\n", s->align); |
81819f0f CL |
5112 | } |
5113 | SLAB_ATTR_RO(align); | |
5114 | ||
5115 | static ssize_t object_size_show(struct kmem_cache *s, char *buf) | |
5116 | { | |
bf16d19a | 5117 | return sysfs_emit(buf, "%u\n", s->object_size); |
81819f0f CL |
5118 | } |
5119 | SLAB_ATTR_RO(object_size); | |
5120 | ||
5121 | static ssize_t objs_per_slab_show(struct kmem_cache *s, char *buf) | |
5122 | { | |
bf16d19a | 5123 | return sysfs_emit(buf, "%u\n", oo_objects(s->oo)); |
81819f0f CL |
5124 | } |
5125 | SLAB_ATTR_RO(objs_per_slab); | |
5126 | ||
5127 | static ssize_t order_show(struct kmem_cache *s, char *buf) | |
5128 | { | |
bf16d19a | 5129 | return sysfs_emit(buf, "%u\n", oo_order(s->oo)); |
81819f0f | 5130 | } |
32a6f409 | 5131 | SLAB_ATTR_RO(order); |
81819f0f | 5132 | |
73d342b1 DR |
5133 | static ssize_t min_partial_show(struct kmem_cache *s, char *buf) |
5134 | { | |
bf16d19a | 5135 | return sysfs_emit(buf, "%lu\n", s->min_partial); |
73d342b1 DR |
5136 | } |
5137 | ||
5138 | static ssize_t min_partial_store(struct kmem_cache *s, const char *buf, | |
5139 | size_t length) | |
5140 | { | |
5141 | unsigned long min; | |
5142 | int err; | |
5143 | ||
3dbb95f7 | 5144 | err = kstrtoul(buf, 10, &min); |
73d342b1 DR |
5145 | if (err) |
5146 | return err; | |
5147 | ||
c0bdb232 | 5148 | set_min_partial(s, min); |
73d342b1 DR |
5149 | return length; |
5150 | } | |
5151 | SLAB_ATTR(min_partial); | |
5152 | ||
49e22585 CL |
5153 | static ssize_t cpu_partial_show(struct kmem_cache *s, char *buf) |
5154 | { | |
bf16d19a | 5155 | return sysfs_emit(buf, "%u\n", slub_cpu_partial(s)); |
49e22585 CL |
5156 | } |
5157 | ||
5158 | static ssize_t cpu_partial_store(struct kmem_cache *s, const char *buf, | |
5159 | size_t length) | |
5160 | { | |
e5d9998f | 5161 | unsigned int objects; |
49e22585 CL |
5162 | int err; |
5163 | ||
e5d9998f | 5164 | err = kstrtouint(buf, 10, &objects); |
49e22585 CL |
5165 | if (err) |
5166 | return err; | |
345c905d | 5167 | if (objects && !kmem_cache_has_cpu_partial(s)) |
74ee4ef1 | 5168 | return -EINVAL; |
49e22585 | 5169 | |
e6d0e1dc | 5170 | slub_set_cpu_partial(s, objects); |
49e22585 CL |
5171 | flush_all(s); |
5172 | return length; | |
5173 | } | |
5174 | SLAB_ATTR(cpu_partial); | |
5175 | ||
81819f0f CL |
5176 | static ssize_t ctor_show(struct kmem_cache *s, char *buf) |
5177 | { | |
62c70bce JP |
5178 | if (!s->ctor) |
5179 | return 0; | |
bf16d19a | 5180 | return sysfs_emit(buf, "%pS\n", s->ctor); |
81819f0f CL |
5181 | } |
5182 | SLAB_ATTR_RO(ctor); | |
5183 | ||
81819f0f CL |
5184 | static ssize_t aliases_show(struct kmem_cache *s, char *buf) |
5185 | { | |
bf16d19a | 5186 | return sysfs_emit(buf, "%d\n", s->refcount < 0 ? 0 : s->refcount - 1); |
81819f0f CL |
5187 | } |
5188 | SLAB_ATTR_RO(aliases); | |
5189 | ||
81819f0f CL |
5190 | static ssize_t partial_show(struct kmem_cache *s, char *buf) |
5191 | { | |
d9acf4b7 | 5192 | return show_slab_objects(s, buf, SO_PARTIAL); |
81819f0f CL |
5193 | } |
5194 | SLAB_ATTR_RO(partial); | |
5195 | ||
5196 | static ssize_t cpu_slabs_show(struct kmem_cache *s, char *buf) | |
5197 | { | |
d9acf4b7 | 5198 | return show_slab_objects(s, buf, SO_CPU); |
81819f0f CL |
5199 | } |
5200 | SLAB_ATTR_RO(cpu_slabs); | |
5201 | ||
5202 | static ssize_t objects_show(struct kmem_cache *s, char *buf) | |
5203 | { | |
205ab99d | 5204 | return show_slab_objects(s, buf, SO_ALL|SO_OBJECTS); |
81819f0f CL |
5205 | } |
5206 | SLAB_ATTR_RO(objects); | |
5207 | ||
205ab99d CL |
5208 | static ssize_t objects_partial_show(struct kmem_cache *s, char *buf) |
5209 | { | |
5210 | return show_slab_objects(s, buf, SO_PARTIAL|SO_OBJECTS); | |
5211 | } | |
5212 | SLAB_ATTR_RO(objects_partial); | |
5213 | ||
49e22585 CL |
5214 | static ssize_t slabs_cpu_partial_show(struct kmem_cache *s, char *buf) |
5215 | { | |
5216 | int objects = 0; | |
5217 | int pages = 0; | |
5218 | int cpu; | |
bf16d19a | 5219 | int len = 0; |
49e22585 CL |
5220 | |
5221 | for_each_online_cpu(cpu) { | |
a93cf07b WY |
5222 | struct page *page; |
5223 | ||
5224 | page = slub_percpu_partial(per_cpu_ptr(s->cpu_slab, cpu)); | |
49e22585 CL |
5225 | |
5226 | if (page) { | |
5227 | pages += page->pages; | |
5228 | objects += page->pobjects; | |
5229 | } | |
5230 | } | |
5231 | ||
bf16d19a | 5232 | len += sysfs_emit_at(buf, len, "%d(%d)", objects, pages); |
49e22585 CL |
5233 | |
5234 | #ifdef CONFIG_SMP | |
5235 | for_each_online_cpu(cpu) { | |
a93cf07b WY |
5236 | struct page *page; |
5237 | ||
5238 | page = slub_percpu_partial(per_cpu_ptr(s->cpu_slab, cpu)); | |
bf16d19a JP |
5239 | if (page) |
5240 | len += sysfs_emit_at(buf, len, " C%d=%d(%d)", | |
5241 | cpu, page->pobjects, page->pages); | |
49e22585 CL |
5242 | } |
5243 | #endif | |
bf16d19a JP |
5244 | len += sysfs_emit_at(buf, len, "\n"); |
5245 | ||
5246 | return len; | |
49e22585 CL |
5247 | } |
5248 | SLAB_ATTR_RO(slabs_cpu_partial); | |
5249 | ||
a5a84755 CL |
5250 | static ssize_t reclaim_account_show(struct kmem_cache *s, char *buf) |
5251 | { | |
bf16d19a | 5252 | return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_RECLAIM_ACCOUNT)); |
a5a84755 | 5253 | } |
8f58119a | 5254 | SLAB_ATTR_RO(reclaim_account); |
a5a84755 CL |
5255 | |
5256 | static ssize_t hwcache_align_show(struct kmem_cache *s, char *buf) | |
5257 | { | |
bf16d19a | 5258 | return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_HWCACHE_ALIGN)); |
a5a84755 CL |
5259 | } |
5260 | SLAB_ATTR_RO(hwcache_align); | |
5261 | ||
5262 | #ifdef CONFIG_ZONE_DMA | |
5263 | static ssize_t cache_dma_show(struct kmem_cache *s, char *buf) | |
5264 | { | |
bf16d19a | 5265 | return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_CACHE_DMA)); |
a5a84755 CL |
5266 | } |
5267 | SLAB_ATTR_RO(cache_dma); | |
5268 | #endif | |
5269 | ||
8eb8284b DW |
5270 | static ssize_t usersize_show(struct kmem_cache *s, char *buf) |
5271 | { | |
bf16d19a | 5272 | return sysfs_emit(buf, "%u\n", s->usersize); |
8eb8284b DW |
5273 | } |
5274 | SLAB_ATTR_RO(usersize); | |
5275 | ||
a5a84755 CL |
5276 | static ssize_t destroy_by_rcu_show(struct kmem_cache *s, char *buf) |
5277 | { | |
bf16d19a | 5278 | return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_TYPESAFE_BY_RCU)); |
a5a84755 CL |
5279 | } |
5280 | SLAB_ATTR_RO(destroy_by_rcu); | |
5281 | ||
ab4d5ed5 | 5282 | #ifdef CONFIG_SLUB_DEBUG |
a5a84755 CL |
5283 | static ssize_t slabs_show(struct kmem_cache *s, char *buf) |
5284 | { | |
5285 | return show_slab_objects(s, buf, SO_ALL); | |
5286 | } | |
5287 | SLAB_ATTR_RO(slabs); | |
5288 | ||
205ab99d CL |
5289 | static ssize_t total_objects_show(struct kmem_cache *s, char *buf) |
5290 | { | |
5291 | return show_slab_objects(s, buf, SO_ALL|SO_TOTAL); | |
5292 | } | |
5293 | SLAB_ATTR_RO(total_objects); | |
5294 | ||
81819f0f CL |
5295 | static ssize_t sanity_checks_show(struct kmem_cache *s, char *buf) |
5296 | { | |
bf16d19a | 5297 | return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_CONSISTENCY_CHECKS)); |
81819f0f | 5298 | } |
060807f8 | 5299 | SLAB_ATTR_RO(sanity_checks); |
81819f0f CL |
5300 | |
5301 | static ssize_t trace_show(struct kmem_cache *s, char *buf) | |
5302 | { | |
bf16d19a | 5303 | return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_TRACE)); |
81819f0f | 5304 | } |
060807f8 | 5305 | SLAB_ATTR_RO(trace); |
81819f0f | 5306 | |
81819f0f CL |
5307 | static ssize_t red_zone_show(struct kmem_cache *s, char *buf) |
5308 | { | |
bf16d19a | 5309 | return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_RED_ZONE)); |
81819f0f CL |
5310 | } |
5311 | ||
ad38b5b1 | 5312 | SLAB_ATTR_RO(red_zone); |
81819f0f CL |
5313 | |
5314 | static ssize_t poison_show(struct kmem_cache *s, char *buf) | |
5315 | { | |
bf16d19a | 5316 | return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_POISON)); |
81819f0f CL |
5317 | } |
5318 | ||
ad38b5b1 | 5319 | SLAB_ATTR_RO(poison); |
81819f0f CL |
5320 | |
5321 | static ssize_t store_user_show(struct kmem_cache *s, char *buf) | |
5322 | { | |
bf16d19a | 5323 | return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_STORE_USER)); |
81819f0f CL |
5324 | } |
5325 | ||
ad38b5b1 | 5326 | SLAB_ATTR_RO(store_user); |
81819f0f | 5327 | |
53e15af0 CL |
5328 | static ssize_t validate_show(struct kmem_cache *s, char *buf) |
5329 | { | |
5330 | return 0; | |
5331 | } | |
5332 | ||
5333 | static ssize_t validate_store(struct kmem_cache *s, | |
5334 | const char *buf, size_t length) | |
5335 | { | |
434e245d CL |
5336 | int ret = -EINVAL; |
5337 | ||
5338 | if (buf[0] == '1') { | |
5339 | ret = validate_slab_cache(s); | |
5340 | if (ret >= 0) | |
5341 | ret = length; | |
5342 | } | |
5343 | return ret; | |
53e15af0 CL |
5344 | } |
5345 | SLAB_ATTR(validate); | |
a5a84755 CL |
5346 | |
5347 | static ssize_t alloc_calls_show(struct kmem_cache *s, char *buf) | |
5348 | { | |
5349 | if (!(s->flags & SLAB_STORE_USER)) | |
5350 | return -ENOSYS; | |
5351 | return list_locations(s, buf, TRACK_ALLOC); | |
5352 | } | |
5353 | SLAB_ATTR_RO(alloc_calls); | |
5354 | ||
5355 | static ssize_t free_calls_show(struct kmem_cache *s, char *buf) | |
5356 | { | |
5357 | if (!(s->flags & SLAB_STORE_USER)) | |
5358 | return -ENOSYS; | |
5359 | return list_locations(s, buf, TRACK_FREE); | |
5360 | } | |
5361 | SLAB_ATTR_RO(free_calls); | |
5362 | #endif /* CONFIG_SLUB_DEBUG */ | |
5363 | ||
5364 | #ifdef CONFIG_FAILSLAB | |
5365 | static ssize_t failslab_show(struct kmem_cache *s, char *buf) | |
5366 | { | |
bf16d19a | 5367 | return sysfs_emit(buf, "%d\n", !!(s->flags & SLAB_FAILSLAB)); |
a5a84755 | 5368 | } |
060807f8 | 5369 | SLAB_ATTR_RO(failslab); |
ab4d5ed5 | 5370 | #endif |
53e15af0 | 5371 | |
2086d26a CL |
5372 | static ssize_t shrink_show(struct kmem_cache *s, char *buf) |
5373 | { | |
5374 | return 0; | |
5375 | } | |
5376 | ||
5377 | static ssize_t shrink_store(struct kmem_cache *s, | |
5378 | const char *buf, size_t length) | |
5379 | { | |
832f37f5 | 5380 | if (buf[0] == '1') |
10befea9 | 5381 | kmem_cache_shrink(s); |
832f37f5 | 5382 | else |
2086d26a CL |
5383 | return -EINVAL; |
5384 | return length; | |
5385 | } | |
5386 | SLAB_ATTR(shrink); | |
5387 | ||
81819f0f | 5388 | #ifdef CONFIG_NUMA |
9824601e | 5389 | static ssize_t remote_node_defrag_ratio_show(struct kmem_cache *s, char *buf) |
81819f0f | 5390 | { |
bf16d19a | 5391 | return sysfs_emit(buf, "%u\n", s->remote_node_defrag_ratio / 10); |
81819f0f CL |
5392 | } |
5393 | ||
9824601e | 5394 | static ssize_t remote_node_defrag_ratio_store(struct kmem_cache *s, |
81819f0f CL |
5395 | const char *buf, size_t length) |
5396 | { | |
eb7235eb | 5397 | unsigned int ratio; |
0121c619 CL |
5398 | int err; |
5399 | ||
eb7235eb | 5400 | err = kstrtouint(buf, 10, &ratio); |
0121c619 CL |
5401 | if (err) |
5402 | return err; | |
eb7235eb AD |
5403 | if (ratio > 100) |
5404 | return -ERANGE; | |
0121c619 | 5405 | |
eb7235eb | 5406 | s->remote_node_defrag_ratio = ratio * 10; |
81819f0f | 5407 | |
81819f0f CL |
5408 | return length; |
5409 | } | |
9824601e | 5410 | SLAB_ATTR(remote_node_defrag_ratio); |
81819f0f CL |
5411 | #endif |
5412 | ||
8ff12cfc | 5413 | #ifdef CONFIG_SLUB_STATS |
8ff12cfc CL |
5414 | static int show_stat(struct kmem_cache *s, char *buf, enum stat_item si) |
5415 | { | |
5416 | unsigned long sum = 0; | |
5417 | int cpu; | |
bf16d19a | 5418 | int len = 0; |
6da2ec56 | 5419 | int *data = kmalloc_array(nr_cpu_ids, sizeof(int), GFP_KERNEL); |
8ff12cfc CL |
5420 | |
5421 | if (!data) | |
5422 | return -ENOMEM; | |
5423 | ||
5424 | for_each_online_cpu(cpu) { | |
9dfc6e68 | 5425 | unsigned x = per_cpu_ptr(s->cpu_slab, cpu)->stat[si]; |
8ff12cfc CL |
5426 | |
5427 | data[cpu] = x; | |
5428 | sum += x; | |
5429 | } | |
5430 | ||
bf16d19a | 5431 | len += sysfs_emit_at(buf, len, "%lu", sum); |
8ff12cfc | 5432 | |
50ef37b9 | 5433 | #ifdef CONFIG_SMP |
8ff12cfc | 5434 | for_each_online_cpu(cpu) { |
bf16d19a JP |
5435 | if (data[cpu]) |
5436 | len += sysfs_emit_at(buf, len, " C%d=%u", | |
5437 | cpu, data[cpu]); | |
8ff12cfc | 5438 | } |
50ef37b9 | 5439 | #endif |
8ff12cfc | 5440 | kfree(data); |
bf16d19a JP |
5441 | len += sysfs_emit_at(buf, len, "\n"); |
5442 | ||
5443 | return len; | |
8ff12cfc CL |
5444 | } |
5445 | ||
78eb00cc DR |
5446 | static void clear_stat(struct kmem_cache *s, enum stat_item si) |
5447 | { | |
5448 | int cpu; | |
5449 | ||
5450 | for_each_online_cpu(cpu) | |
9dfc6e68 | 5451 | per_cpu_ptr(s->cpu_slab, cpu)->stat[si] = 0; |
78eb00cc DR |
5452 | } |
5453 | ||
8ff12cfc CL |
5454 | #define STAT_ATTR(si, text) \ |
5455 | static ssize_t text##_show(struct kmem_cache *s, char *buf) \ | |
5456 | { \ | |
5457 | return show_stat(s, buf, si); \ | |
5458 | } \ | |
78eb00cc DR |
5459 | static ssize_t text##_store(struct kmem_cache *s, \ |
5460 | const char *buf, size_t length) \ | |
5461 | { \ | |
5462 | if (buf[0] != '0') \ | |
5463 | return -EINVAL; \ | |
5464 | clear_stat(s, si); \ | |
5465 | return length; \ | |
5466 | } \ | |
5467 | SLAB_ATTR(text); \ | |
8ff12cfc CL |
5468 | |
5469 | STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath); | |
5470 | STAT_ATTR(ALLOC_SLOWPATH, alloc_slowpath); | |
5471 | STAT_ATTR(FREE_FASTPATH, free_fastpath); | |
5472 | STAT_ATTR(FREE_SLOWPATH, free_slowpath); | |
5473 | STAT_ATTR(FREE_FROZEN, free_frozen); | |
5474 | STAT_ATTR(FREE_ADD_PARTIAL, free_add_partial); | |
5475 | STAT_ATTR(FREE_REMOVE_PARTIAL, free_remove_partial); | |
5476 | STAT_ATTR(ALLOC_FROM_PARTIAL, alloc_from_partial); | |
5477 | STAT_ATTR(ALLOC_SLAB, alloc_slab); | |
5478 | STAT_ATTR(ALLOC_REFILL, alloc_refill); | |
e36a2652 | 5479 | STAT_ATTR(ALLOC_NODE_MISMATCH, alloc_node_mismatch); |
8ff12cfc CL |
5480 | STAT_ATTR(FREE_SLAB, free_slab); |
5481 | STAT_ATTR(CPUSLAB_FLUSH, cpuslab_flush); | |
5482 | STAT_ATTR(DEACTIVATE_FULL, deactivate_full); | |
5483 | STAT_ATTR(DEACTIVATE_EMPTY, deactivate_empty); | |
5484 | STAT_ATTR(DEACTIVATE_TO_HEAD, deactivate_to_head); | |
5485 | STAT_ATTR(DEACTIVATE_TO_TAIL, deactivate_to_tail); | |
5486 | STAT_ATTR(DEACTIVATE_REMOTE_FREES, deactivate_remote_frees); | |
03e404af | 5487 | STAT_ATTR(DEACTIVATE_BYPASS, deactivate_bypass); |
65c3376a | 5488 | STAT_ATTR(ORDER_FALLBACK, order_fallback); |
b789ef51 CL |
5489 | STAT_ATTR(CMPXCHG_DOUBLE_CPU_FAIL, cmpxchg_double_cpu_fail); |
5490 | STAT_ATTR(CMPXCHG_DOUBLE_FAIL, cmpxchg_double_fail); | |
49e22585 CL |
5491 | STAT_ATTR(CPU_PARTIAL_ALLOC, cpu_partial_alloc); |
5492 | STAT_ATTR(CPU_PARTIAL_FREE, cpu_partial_free); | |
8028dcea AS |
5493 | STAT_ATTR(CPU_PARTIAL_NODE, cpu_partial_node); |
5494 | STAT_ATTR(CPU_PARTIAL_DRAIN, cpu_partial_drain); | |
6dfd1b65 | 5495 | #endif /* CONFIG_SLUB_STATS */ |
8ff12cfc | 5496 | |
06428780 | 5497 | static struct attribute *slab_attrs[] = { |
81819f0f CL |
5498 | &slab_size_attr.attr, |
5499 | &object_size_attr.attr, | |
5500 | &objs_per_slab_attr.attr, | |
5501 | &order_attr.attr, | |
73d342b1 | 5502 | &min_partial_attr.attr, |
49e22585 | 5503 | &cpu_partial_attr.attr, |
81819f0f | 5504 | &objects_attr.attr, |
205ab99d | 5505 | &objects_partial_attr.attr, |
81819f0f CL |
5506 | &partial_attr.attr, |
5507 | &cpu_slabs_attr.attr, | |
5508 | &ctor_attr.attr, | |
81819f0f CL |
5509 | &aliases_attr.attr, |
5510 | &align_attr.attr, | |
81819f0f CL |
5511 | &hwcache_align_attr.attr, |
5512 | &reclaim_account_attr.attr, | |
5513 | &destroy_by_rcu_attr.attr, | |
a5a84755 | 5514 | &shrink_attr.attr, |
49e22585 | 5515 | &slabs_cpu_partial_attr.attr, |
ab4d5ed5 | 5516 | #ifdef CONFIG_SLUB_DEBUG |
a5a84755 CL |
5517 | &total_objects_attr.attr, |
5518 | &slabs_attr.attr, | |
5519 | &sanity_checks_attr.attr, | |
5520 | &trace_attr.attr, | |
81819f0f CL |
5521 | &red_zone_attr.attr, |
5522 | &poison_attr.attr, | |
5523 | &store_user_attr.attr, | |
53e15af0 | 5524 | &validate_attr.attr, |
88a420e4 CL |
5525 | &alloc_calls_attr.attr, |
5526 | &free_calls_attr.attr, | |
ab4d5ed5 | 5527 | #endif |
81819f0f CL |
5528 | #ifdef CONFIG_ZONE_DMA |
5529 | &cache_dma_attr.attr, | |
5530 | #endif | |
5531 | #ifdef CONFIG_NUMA | |
9824601e | 5532 | &remote_node_defrag_ratio_attr.attr, |
8ff12cfc CL |
5533 | #endif |
5534 | #ifdef CONFIG_SLUB_STATS | |
5535 | &alloc_fastpath_attr.attr, | |
5536 | &alloc_slowpath_attr.attr, | |
5537 | &free_fastpath_attr.attr, | |
5538 | &free_slowpath_attr.attr, | |
5539 | &free_frozen_attr.attr, | |
5540 | &free_add_partial_attr.attr, | |
5541 | &free_remove_partial_attr.attr, | |
5542 | &alloc_from_partial_attr.attr, | |
5543 | &alloc_slab_attr.attr, | |
5544 | &alloc_refill_attr.attr, | |
e36a2652 | 5545 | &alloc_node_mismatch_attr.attr, |
8ff12cfc CL |
5546 | &free_slab_attr.attr, |
5547 | &cpuslab_flush_attr.attr, | |
5548 | &deactivate_full_attr.attr, | |
5549 | &deactivate_empty_attr.attr, | |
5550 | &deactivate_to_head_attr.attr, | |
5551 | &deactivate_to_tail_attr.attr, | |
5552 | &deactivate_remote_frees_attr.attr, | |
03e404af | 5553 | &deactivate_bypass_attr.attr, |
65c3376a | 5554 | &order_fallback_attr.attr, |
b789ef51 CL |
5555 | &cmpxchg_double_fail_attr.attr, |
5556 | &cmpxchg_double_cpu_fail_attr.attr, | |
49e22585 CL |
5557 | &cpu_partial_alloc_attr.attr, |
5558 | &cpu_partial_free_attr.attr, | |
8028dcea AS |
5559 | &cpu_partial_node_attr.attr, |
5560 | &cpu_partial_drain_attr.attr, | |
81819f0f | 5561 | #endif |
4c13dd3b DM |
5562 | #ifdef CONFIG_FAILSLAB |
5563 | &failslab_attr.attr, | |
5564 | #endif | |
8eb8284b | 5565 | &usersize_attr.attr, |
4c13dd3b | 5566 | |
81819f0f CL |
5567 | NULL |
5568 | }; | |
5569 | ||
1fdaaa23 | 5570 | static const struct attribute_group slab_attr_group = { |
81819f0f CL |
5571 | .attrs = slab_attrs, |
5572 | }; | |
5573 | ||
5574 | static ssize_t slab_attr_show(struct kobject *kobj, | |
5575 | struct attribute *attr, | |
5576 | char *buf) | |
5577 | { | |
5578 | struct slab_attribute *attribute; | |
5579 | struct kmem_cache *s; | |
5580 | int err; | |
5581 | ||
5582 | attribute = to_slab_attr(attr); | |
5583 | s = to_slab(kobj); | |
5584 | ||
5585 | if (!attribute->show) | |
5586 | return -EIO; | |
5587 | ||
5588 | err = attribute->show(s, buf); | |
5589 | ||
5590 | return err; | |
5591 | } | |
5592 | ||
5593 | static ssize_t slab_attr_store(struct kobject *kobj, | |
5594 | struct attribute *attr, | |
5595 | const char *buf, size_t len) | |
5596 | { | |
5597 | struct slab_attribute *attribute; | |
5598 | struct kmem_cache *s; | |
5599 | int err; | |
5600 | ||
5601 | attribute = to_slab_attr(attr); | |
5602 | s = to_slab(kobj); | |
5603 | ||
5604 | if (!attribute->store) | |
5605 | return -EIO; | |
5606 | ||
5607 | err = attribute->store(s, buf, len); | |
81819f0f CL |
5608 | return err; |
5609 | } | |
5610 | ||
41a21285 CL |
5611 | static void kmem_cache_release(struct kobject *k) |
5612 | { | |
5613 | slab_kmem_cache_release(to_slab(k)); | |
5614 | } | |
5615 | ||
52cf25d0 | 5616 | static const struct sysfs_ops slab_sysfs_ops = { |
81819f0f CL |
5617 | .show = slab_attr_show, |
5618 | .store = slab_attr_store, | |
5619 | }; | |
5620 | ||
5621 | static struct kobj_type slab_ktype = { | |
5622 | .sysfs_ops = &slab_sysfs_ops, | |
41a21285 | 5623 | .release = kmem_cache_release, |
81819f0f CL |
5624 | }; |
5625 | ||
27c3a314 | 5626 | static struct kset *slab_kset; |
81819f0f | 5627 | |
9a41707b VD |
5628 | static inline struct kset *cache_kset(struct kmem_cache *s) |
5629 | { | |
9a41707b VD |
5630 | return slab_kset; |
5631 | } | |
5632 | ||
81819f0f CL |
5633 | #define ID_STR_LENGTH 64 |
5634 | ||
5635 | /* Create a unique string id for a slab cache: | |
6446faa2 CL |
5636 | * |
5637 | * Format :[flags-]size | |
81819f0f CL |
5638 | */ |
5639 | static char *create_unique_id(struct kmem_cache *s) | |
5640 | { | |
5641 | char *name = kmalloc(ID_STR_LENGTH, GFP_KERNEL); | |
5642 | char *p = name; | |
5643 | ||
5644 | BUG_ON(!name); | |
5645 | ||
5646 | *p++ = ':'; | |
5647 | /* | |
5648 | * First flags affecting slabcache operations. We will only | |
5649 | * get here for aliasable slabs so we do not need to support | |
5650 | * too many flags. The flags here must cover all flags that | |
5651 | * are matched during merging to guarantee that the id is | |
5652 | * unique. | |
5653 | */ | |
5654 | if (s->flags & SLAB_CACHE_DMA) | |
5655 | *p++ = 'd'; | |
6d6ea1e9 NB |
5656 | if (s->flags & SLAB_CACHE_DMA32) |
5657 | *p++ = 'D'; | |
81819f0f CL |
5658 | if (s->flags & SLAB_RECLAIM_ACCOUNT) |
5659 | *p++ = 'a'; | |
becfda68 | 5660 | if (s->flags & SLAB_CONSISTENCY_CHECKS) |
81819f0f | 5661 | *p++ = 'F'; |
230e9fc2 VD |
5662 | if (s->flags & SLAB_ACCOUNT) |
5663 | *p++ = 'A'; | |
81819f0f CL |
5664 | if (p != name + 1) |
5665 | *p++ = '-'; | |
44065b2e | 5666 | p += sprintf(p, "%07u", s->size); |
2633d7a0 | 5667 | |
81819f0f CL |
5668 | BUG_ON(p > name + ID_STR_LENGTH - 1); |
5669 | return name; | |
5670 | } | |
5671 | ||
5672 | static int sysfs_slab_add(struct kmem_cache *s) | |
5673 | { | |
5674 | int err; | |
5675 | const char *name; | |
1663f26d | 5676 | struct kset *kset = cache_kset(s); |
45530c44 | 5677 | int unmergeable = slab_unmergeable(s); |
81819f0f | 5678 | |
1663f26d TH |
5679 | if (!kset) { |
5680 | kobject_init(&s->kobj, &slab_ktype); | |
5681 | return 0; | |
5682 | } | |
5683 | ||
11066386 MC |
5684 | if (!unmergeable && disable_higher_order_debug && |
5685 | (slub_debug & DEBUG_METADATA_FLAGS)) | |
5686 | unmergeable = 1; | |
5687 | ||
81819f0f CL |
5688 | if (unmergeable) { |
5689 | /* | |
5690 | * Slabcache can never be merged so we can use the name proper. | |
5691 | * This is typically the case for debug situations. In that | |
5692 | * case we can catch duplicate names easily. | |
5693 | */ | |
27c3a314 | 5694 | sysfs_remove_link(&slab_kset->kobj, s->name); |
81819f0f CL |
5695 | name = s->name; |
5696 | } else { | |
5697 | /* | |
5698 | * Create a unique name for the slab as a target | |
5699 | * for the symlinks. | |
5700 | */ | |
5701 | name = create_unique_id(s); | |
5702 | } | |
5703 | ||
1663f26d | 5704 | s->kobj.kset = kset; |
26e4f205 | 5705 | err = kobject_init_and_add(&s->kobj, &slab_ktype, NULL, "%s", name); |
757fed1d | 5706 | if (err) |
80da026a | 5707 | goto out; |
81819f0f CL |
5708 | |
5709 | err = sysfs_create_group(&s->kobj, &slab_attr_group); | |
54b6a731 DJ |
5710 | if (err) |
5711 | goto out_del_kobj; | |
9a41707b | 5712 | |
81819f0f CL |
5713 | if (!unmergeable) { |
5714 | /* Setup first alias */ | |
5715 | sysfs_slab_alias(s, s->name); | |
81819f0f | 5716 | } |
54b6a731 DJ |
5717 | out: |
5718 | if (!unmergeable) | |
5719 | kfree(name); | |
5720 | return err; | |
5721 | out_del_kobj: | |
5722 | kobject_del(&s->kobj); | |
54b6a731 | 5723 | goto out; |
81819f0f CL |
5724 | } |
5725 | ||
d50d82fa MP |
5726 | void sysfs_slab_unlink(struct kmem_cache *s) |
5727 | { | |
5728 | if (slab_state >= FULL) | |
5729 | kobject_del(&s->kobj); | |
5730 | } | |
5731 | ||
bf5eb3de TH |
5732 | void sysfs_slab_release(struct kmem_cache *s) |
5733 | { | |
5734 | if (slab_state >= FULL) | |
5735 | kobject_put(&s->kobj); | |
81819f0f CL |
5736 | } |
5737 | ||
5738 | /* | |
5739 | * Need to buffer aliases during bootup until sysfs becomes | |
9f6c708e | 5740 | * available lest we lose that information. |
81819f0f CL |
5741 | */ |
5742 | struct saved_alias { | |
5743 | struct kmem_cache *s; | |
5744 | const char *name; | |
5745 | struct saved_alias *next; | |
5746 | }; | |
5747 | ||
5af328a5 | 5748 | static struct saved_alias *alias_list; |
81819f0f CL |
5749 | |
5750 | static int sysfs_slab_alias(struct kmem_cache *s, const char *name) | |
5751 | { | |
5752 | struct saved_alias *al; | |
5753 | ||
97d06609 | 5754 | if (slab_state == FULL) { |
81819f0f CL |
5755 | /* |
5756 | * If we have a leftover link then remove it. | |
5757 | */ | |
27c3a314 GKH |
5758 | sysfs_remove_link(&slab_kset->kobj, name); |
5759 | return sysfs_create_link(&slab_kset->kobj, &s->kobj, name); | |
81819f0f CL |
5760 | } |
5761 | ||
5762 | al = kmalloc(sizeof(struct saved_alias), GFP_KERNEL); | |
5763 | if (!al) | |
5764 | return -ENOMEM; | |
5765 | ||
5766 | al->s = s; | |
5767 | al->name = name; | |
5768 | al->next = alias_list; | |
5769 | alias_list = al; | |
5770 | return 0; | |
5771 | } | |
5772 | ||
5773 | static int __init slab_sysfs_init(void) | |
5774 | { | |
5b95a4ac | 5775 | struct kmem_cache *s; |
81819f0f CL |
5776 | int err; |
5777 | ||
18004c5d | 5778 | mutex_lock(&slab_mutex); |
2bce6485 | 5779 | |
d7660ce5 | 5780 | slab_kset = kset_create_and_add("slab", NULL, kernel_kobj); |
27c3a314 | 5781 | if (!slab_kset) { |
18004c5d | 5782 | mutex_unlock(&slab_mutex); |
f9f58285 | 5783 | pr_err("Cannot register slab subsystem.\n"); |
81819f0f CL |
5784 | return -ENOSYS; |
5785 | } | |
5786 | ||
97d06609 | 5787 | slab_state = FULL; |
26a7bd03 | 5788 | |
5b95a4ac | 5789 | list_for_each_entry(s, &slab_caches, list) { |
26a7bd03 | 5790 | err = sysfs_slab_add(s); |
5d540fb7 | 5791 | if (err) |
f9f58285 FF |
5792 | pr_err("SLUB: Unable to add boot slab %s to sysfs\n", |
5793 | s->name); | |
26a7bd03 | 5794 | } |
81819f0f CL |
5795 | |
5796 | while (alias_list) { | |
5797 | struct saved_alias *al = alias_list; | |
5798 | ||
5799 | alias_list = alias_list->next; | |
5800 | err = sysfs_slab_alias(al->s, al->name); | |
5d540fb7 | 5801 | if (err) |
f9f58285 FF |
5802 | pr_err("SLUB: Unable to add boot slab alias %s to sysfs\n", |
5803 | al->name); | |
81819f0f CL |
5804 | kfree(al); |
5805 | } | |
5806 | ||
18004c5d | 5807 | mutex_unlock(&slab_mutex); |
81819f0f CL |
5808 | resiliency_test(); |
5809 | return 0; | |
5810 | } | |
5811 | ||
5812 | __initcall(slab_sysfs_init); | |
ab4d5ed5 | 5813 | #endif /* CONFIG_SYSFS */ |
57ed3eda PE |
5814 | |
5815 | /* | |
5816 | * The /proc/slabinfo ABI | |
5817 | */ | |
5b365771 | 5818 | #ifdef CONFIG_SLUB_DEBUG |
0d7561c6 | 5819 | void get_slabinfo(struct kmem_cache *s, struct slabinfo *sinfo) |
57ed3eda | 5820 | { |
57ed3eda | 5821 | unsigned long nr_slabs = 0; |
205ab99d CL |
5822 | unsigned long nr_objs = 0; |
5823 | unsigned long nr_free = 0; | |
57ed3eda | 5824 | int node; |
fa45dc25 | 5825 | struct kmem_cache_node *n; |
57ed3eda | 5826 | |
fa45dc25 | 5827 | for_each_kmem_cache_node(s, node, n) { |
c17fd13e WL |
5828 | nr_slabs += node_nr_slabs(n); |
5829 | nr_objs += node_nr_objs(n); | |
205ab99d | 5830 | nr_free += count_partial(n, count_free); |
57ed3eda PE |
5831 | } |
5832 | ||
0d7561c6 GC |
5833 | sinfo->active_objs = nr_objs - nr_free; |
5834 | sinfo->num_objs = nr_objs; | |
5835 | sinfo->active_slabs = nr_slabs; | |
5836 | sinfo->num_slabs = nr_slabs; | |
5837 | sinfo->objects_per_slab = oo_objects(s->oo); | |
5838 | sinfo->cache_order = oo_order(s->oo); | |
57ed3eda PE |
5839 | } |
5840 | ||
0d7561c6 | 5841 | void slabinfo_show_stats(struct seq_file *m, struct kmem_cache *s) |
7b3c3a50 | 5842 | { |
7b3c3a50 AD |
5843 | } |
5844 | ||
b7454ad3 GC |
5845 | ssize_t slabinfo_write(struct file *file, const char __user *buffer, |
5846 | size_t count, loff_t *ppos) | |
7b3c3a50 | 5847 | { |
b7454ad3 | 5848 | return -EIO; |
7b3c3a50 | 5849 | } |
5b365771 | 5850 | #endif /* CONFIG_SLUB_DEBUG */ |