]>
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 | * | |
881db7fb CL |
6 | * The allocator synchronizes using per slab locks or atomic operatios |
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> |
3ac38faa | 22 | #include <linux/notifier.h> |
81819f0f | 23 | #include <linux/seq_file.h> |
a79316c6 | 24 | #include <linux/kasan.h> |
5a896d9e | 25 | #include <linux/kmemcheck.h> |
81819f0f CL |
26 | #include <linux/cpu.h> |
27 | #include <linux/cpuset.h> | |
28 | #include <linux/mempolicy.h> | |
29 | #include <linux/ctype.h> | |
3ac7fe5a | 30 | #include <linux/debugobjects.h> |
81819f0f | 31 | #include <linux/kallsyms.h> |
b9049e23 | 32 | #include <linux/memory.h> |
f8bd2258 | 33 | #include <linux/math64.h> |
773ff60e | 34 | #include <linux/fault-inject.h> |
bfa71457 | 35 | #include <linux/stacktrace.h> |
4de900b4 | 36 | #include <linux/prefetch.h> |
2633d7a0 | 37 | #include <linux/memcontrol.h> |
2482ddec | 38 | #include <linux/random.h> |
81819f0f | 39 | |
4a92379b RK |
40 | #include <trace/events/kmem.h> |
41 | ||
072bb0aa MG |
42 | #include "internal.h" |
43 | ||
81819f0f CL |
44 | /* |
45 | * Lock order: | |
18004c5d | 46 | * 1. slab_mutex (Global Mutex) |
881db7fb CL |
47 | * 2. node->list_lock |
48 | * 3. slab_lock(page) (Only on some arches and for debugging) | |
81819f0f | 49 | * |
18004c5d | 50 | * slab_mutex |
881db7fb | 51 | * |
18004c5d | 52 | * The role of the slab_mutex is to protect the list of all the slabs |
881db7fb CL |
53 | * and to synchronize major metadata changes to slab cache structures. |
54 | * | |
55 | * The slab_lock is only used for debugging and on arches that do not | |
56 | * have the ability to do a cmpxchg_double. It only protects the second | |
57 | * double word in the page struct. Meaning | |
58 | * A. page->freelist -> List of object free in a page | |
59 | * B. page->counters -> Counters of objects | |
60 | * C. page->frozen -> frozen state | |
61 | * | |
62 | * If a slab is frozen then it is exempt from list management. It is not | |
63 | * on any list. The processor that froze the slab is the one who can | |
64 | * perform list operations on the page. Other processors may put objects | |
65 | * onto the freelist but the processor that froze the slab is the only | |
66 | * one that can retrieve the objects from the 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 | * | |
97 | * Overloading of page flags that are otherwise used for LRU management. | |
98 | * | |
4b6f0750 CL |
99 | * PageActive The slab is frozen and exempt from list processing. |
100 | * This means that the slab is dedicated to a purpose | |
101 | * such as satisfying allocations for a specific | |
102 | * processor. Objects may be freed in the slab while | |
103 | * it is frozen but slab_free will then skip the usual | |
104 | * list operations. It is up to the processor holding | |
105 | * the slab to integrate the slab into the slab lists | |
106 | * when the slab is no longer needed. | |
107 | * | |
108 | * One use of this flag is to mark slabs that are | |
109 | * used for allocations. Then such a slab becomes a cpu | |
110 | * slab. The cpu slab may be equipped with an additional | |
dfb4f096 | 111 | * freelist that allows lockless access to |
894b8788 CL |
112 | * free objects in addition to the regular freelist |
113 | * that requires the slab lock. | |
81819f0f CL |
114 | * |
115 | * PageError Slab requires special handling due to debug | |
116 | * options set. This moves slab handling out of | |
894b8788 | 117 | * the fast path and disables lockless freelists. |
81819f0f CL |
118 | */ |
119 | ||
af537b0a CL |
120 | static inline int kmem_cache_debug(struct kmem_cache *s) |
121 | { | |
5577bd8a | 122 | #ifdef CONFIG_SLUB_DEBUG |
af537b0a | 123 | return unlikely(s->flags & SLAB_DEBUG_FLAGS); |
5577bd8a | 124 | #else |
af537b0a | 125 | return 0; |
5577bd8a | 126 | #endif |
af537b0a | 127 | } |
5577bd8a | 128 | |
117d54df | 129 | void *fixup_red_left(struct kmem_cache *s, void *p) |
d86bd1be JK |
130 | { |
131 | if (kmem_cache_debug(s) && s->flags & SLAB_RED_ZONE) | |
132 | p += s->red_left_pad; | |
133 | ||
134 | return p; | |
135 | } | |
136 | ||
345c905d JK |
137 | static inline bool kmem_cache_has_cpu_partial(struct kmem_cache *s) |
138 | { | |
139 | #ifdef CONFIG_SLUB_CPU_PARTIAL | |
140 | return !kmem_cache_debug(s); | |
141 | #else | |
142 | return false; | |
143 | #endif | |
144 | } | |
145 | ||
81819f0f CL |
146 | /* |
147 | * Issues still to be resolved: | |
148 | * | |
81819f0f CL |
149 | * - Support PAGE_ALLOC_DEBUG. Should be easy to do. |
150 | * | |
81819f0f CL |
151 | * - Variable sizing of the per node arrays |
152 | */ | |
153 | ||
154 | /* Enable to test recovery from slab corruption on boot */ | |
155 | #undef SLUB_RESILIENCY_TEST | |
156 | ||
b789ef51 CL |
157 | /* Enable to log cmpxchg failures */ |
158 | #undef SLUB_DEBUG_CMPXCHG | |
159 | ||
2086d26a CL |
160 | /* |
161 | * Mininum number of partial slabs. These will be left on the partial | |
162 | * lists even if they are empty. kmem_cache_shrink may reclaim them. | |
163 | */ | |
76be8950 | 164 | #define MIN_PARTIAL 5 |
e95eed57 | 165 | |
2086d26a CL |
166 | /* |
167 | * Maximum number of desirable partial slabs. | |
168 | * The existence of more partial slabs makes kmem_cache_shrink | |
721ae22a | 169 | * sort the partial list by the number of objects in use. |
2086d26a CL |
170 | */ |
171 | #define MAX_PARTIAL 10 | |
172 | ||
becfda68 | 173 | #define DEBUG_DEFAULT_FLAGS (SLAB_CONSISTENCY_CHECKS | SLAB_RED_ZONE | \ |
81819f0f | 174 | SLAB_POISON | SLAB_STORE_USER) |
672bba3a | 175 | |
149daaf3 LA |
176 | /* |
177 | * These debug flags cannot use CMPXCHG because there might be consistency | |
178 | * issues when checking or reading debug information | |
179 | */ | |
180 | #define SLAB_NO_CMPXCHG (SLAB_CONSISTENCY_CHECKS | SLAB_STORE_USER | \ | |
181 | SLAB_TRACE) | |
182 | ||
183 | ||
fa5ec8a1 | 184 | /* |
3de47213 DR |
185 | * Debugging flags that require metadata to be stored in the slab. These get |
186 | * disabled when slub_debug=O is used and a cache's min order increases with | |
187 | * metadata. | |
fa5ec8a1 | 188 | */ |
3de47213 | 189 | #define DEBUG_METADATA_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER) |
fa5ec8a1 | 190 | |
210b5c06 CG |
191 | #define OO_SHIFT 16 |
192 | #define OO_MASK ((1 << OO_SHIFT) - 1) | |
50d5c41c | 193 | #define MAX_OBJS_PER_PAGE 32767 /* since page.objects is u15 */ |
210b5c06 | 194 | |
81819f0f | 195 | /* Internal SLUB flags */ |
d50112ed | 196 | /* Poison object */ |
4fd0b46e | 197 | #define __OBJECT_POISON ((slab_flags_t __force)0x80000000U) |
d50112ed | 198 | /* Use cmpxchg_double */ |
4fd0b46e | 199 | #define __CMPXCHG_DOUBLE ((slab_flags_t __force)0x40000000U) |
81819f0f | 200 | |
02cbc874 CL |
201 | /* |
202 | * Tracking user of a slab. | |
203 | */ | |
d6543e39 | 204 | #define TRACK_ADDRS_COUNT 16 |
02cbc874 | 205 | struct track { |
ce71e27c | 206 | unsigned long addr; /* Called from address */ |
d6543e39 BG |
207 | #ifdef CONFIG_STACKTRACE |
208 | unsigned long addrs[TRACK_ADDRS_COUNT]; /* Called from address */ | |
209 | #endif | |
02cbc874 CL |
210 | int cpu; /* Was running on cpu */ |
211 | int pid; /* Pid context */ | |
212 | unsigned long when; /* When did the operation occur */ | |
213 | }; | |
214 | ||
215 | enum track_item { TRACK_ALLOC, TRACK_FREE }; | |
216 | ||
ab4d5ed5 | 217 | #ifdef CONFIG_SYSFS |
81819f0f CL |
218 | static int sysfs_slab_add(struct kmem_cache *); |
219 | static int sysfs_slab_alias(struct kmem_cache *, const char *); | |
107dab5c | 220 | static void memcg_propagate_slab_attrs(struct kmem_cache *s); |
bf5eb3de | 221 | static void sysfs_slab_remove(struct kmem_cache *s); |
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; } | |
107dab5c | 226 | static inline void memcg_propagate_slab_attrs(struct kmem_cache *s) { } |
bf5eb3de | 227 | static inline void sysfs_slab_remove(struct kmem_cache *s) { } |
81819f0f CL |
228 | #endif |
229 | ||
4fdccdfb | 230 | static inline void stat(const struct kmem_cache *s, enum stat_item si) |
8ff12cfc CL |
231 | { |
232 | #ifdef CONFIG_SLUB_STATS | |
88da03a6 CL |
233 | /* |
234 | * The rmw is racy on a preemptible kernel but this is acceptable, so | |
235 | * avoid this_cpu_add()'s irq-disable overhead. | |
236 | */ | |
237 | raw_cpu_inc(s->cpu_slab->stat[si]); | |
8ff12cfc CL |
238 | #endif |
239 | } | |
240 | ||
81819f0f CL |
241 | /******************************************************************** |
242 | * Core slab cache functions | |
243 | *******************************************************************/ | |
244 | ||
2482ddec KC |
245 | /* |
246 | * Returns freelist pointer (ptr). With hardening, this is obfuscated | |
247 | * with an XOR of the address where the pointer is held and a per-cache | |
248 | * random number. | |
249 | */ | |
250 | static inline void *freelist_ptr(const struct kmem_cache *s, void *ptr, | |
251 | unsigned long ptr_addr) | |
252 | { | |
253 | #ifdef CONFIG_SLAB_FREELIST_HARDENED | |
254 | return (void *)((unsigned long)ptr ^ s->random ^ ptr_addr); | |
255 | #else | |
256 | return ptr; | |
257 | #endif | |
258 | } | |
259 | ||
260 | /* Returns the freelist pointer recorded at location ptr_addr. */ | |
261 | static inline void *freelist_dereference(const struct kmem_cache *s, | |
262 | void *ptr_addr) | |
263 | { | |
264 | return freelist_ptr(s, (void *)*(unsigned long *)(ptr_addr), | |
265 | (unsigned long)ptr_addr); | |
266 | } | |
267 | ||
7656c72b CL |
268 | static inline void *get_freepointer(struct kmem_cache *s, void *object) |
269 | { | |
2482ddec | 270 | return freelist_dereference(s, object + s->offset); |
7656c72b CL |
271 | } |
272 | ||
0ad9500e ED |
273 | static void prefetch_freepointer(const struct kmem_cache *s, void *object) |
274 | { | |
2482ddec KC |
275 | if (object) |
276 | prefetch(freelist_dereference(s, object + s->offset)); | |
0ad9500e ED |
277 | } |
278 | ||
1393d9a1 CL |
279 | static inline void *get_freepointer_safe(struct kmem_cache *s, void *object) |
280 | { | |
2482ddec | 281 | unsigned long freepointer_addr; |
1393d9a1 CL |
282 | void *p; |
283 | ||
922d566c JK |
284 | if (!debug_pagealloc_enabled()) |
285 | return get_freepointer(s, object); | |
286 | ||
2482ddec KC |
287 | freepointer_addr = (unsigned long)object + s->offset; |
288 | probe_kernel_read(&p, (void **)freepointer_addr, sizeof(p)); | |
289 | return freelist_ptr(s, p, freepointer_addr); | |
1393d9a1 CL |
290 | } |
291 | ||
7656c72b CL |
292 | static inline void set_freepointer(struct kmem_cache *s, void *object, void *fp) |
293 | { | |
2482ddec KC |
294 | unsigned long freeptr_addr = (unsigned long)object + s->offset; |
295 | ||
ce6fa91b AP |
296 | #ifdef CONFIG_SLAB_FREELIST_HARDENED |
297 | BUG_ON(object == fp); /* naive detection of double free or corruption */ | |
298 | #endif | |
299 | ||
2482ddec | 300 | *(void **)freeptr_addr = freelist_ptr(s, fp, freeptr_addr); |
7656c72b CL |
301 | } |
302 | ||
303 | /* Loop over all objects in a slab */ | |
224a88be | 304 | #define for_each_object(__p, __s, __addr, __objects) \ |
d86bd1be JK |
305 | for (__p = fixup_red_left(__s, __addr); \ |
306 | __p < (__addr) + (__objects) * (__s)->size; \ | |
307 | __p += (__s)->size) | |
7656c72b | 308 | |
54266640 | 309 | #define for_each_object_idx(__p, __idx, __s, __addr, __objects) \ |
d86bd1be JK |
310 | for (__p = fixup_red_left(__s, __addr), __idx = 1; \ |
311 | __idx <= __objects; \ | |
312 | __p += (__s)->size, __idx++) | |
54266640 | 313 | |
7656c72b CL |
314 | /* Determine object index from a given position */ |
315 | static inline int slab_index(void *p, struct kmem_cache *s, void *addr) | |
316 | { | |
317 | return (p - addr) / s->size; | |
318 | } | |
319 | ||
ab9a0f19 LJ |
320 | static inline int order_objects(int order, unsigned long size, int reserved) |
321 | { | |
322 | return ((PAGE_SIZE << order) - reserved) / size; | |
323 | } | |
324 | ||
834f3d11 | 325 | static inline struct kmem_cache_order_objects oo_make(int order, |
ab9a0f19 | 326 | unsigned long size, int reserved) |
834f3d11 CL |
327 | { |
328 | struct kmem_cache_order_objects x = { | |
ab9a0f19 | 329 | (order << OO_SHIFT) + order_objects(order, size, reserved) |
834f3d11 CL |
330 | }; |
331 | ||
332 | return x; | |
333 | } | |
334 | ||
335 | static inline int oo_order(struct kmem_cache_order_objects x) | |
336 | { | |
210b5c06 | 337 | return x.x >> OO_SHIFT; |
834f3d11 CL |
338 | } |
339 | ||
340 | static inline int oo_objects(struct kmem_cache_order_objects x) | |
341 | { | |
210b5c06 | 342 | return x.x & OO_MASK; |
834f3d11 CL |
343 | } |
344 | ||
881db7fb CL |
345 | /* |
346 | * Per slab locking using the pagelock | |
347 | */ | |
348 | static __always_inline void slab_lock(struct page *page) | |
349 | { | |
48c935ad | 350 | VM_BUG_ON_PAGE(PageTail(page), page); |
881db7fb CL |
351 | bit_spin_lock(PG_locked, &page->flags); |
352 | } | |
353 | ||
354 | static __always_inline void slab_unlock(struct page *page) | |
355 | { | |
48c935ad | 356 | VM_BUG_ON_PAGE(PageTail(page), page); |
881db7fb CL |
357 | __bit_spin_unlock(PG_locked, &page->flags); |
358 | } | |
359 | ||
a0320865 DH |
360 | static inline void set_page_slub_counters(struct page *page, unsigned long counters_new) |
361 | { | |
362 | struct page tmp; | |
363 | tmp.counters = counters_new; | |
364 | /* | |
365 | * page->counters can cover frozen/inuse/objects as well | |
0139aa7b JK |
366 | * as page->_refcount. If we assign to ->counters directly |
367 | * we run the risk of losing updates to page->_refcount, so | |
a0320865 DH |
368 | * be careful and only assign to the fields we need. |
369 | */ | |
370 | page->frozen = tmp.frozen; | |
371 | page->inuse = tmp.inuse; | |
372 | page->objects = tmp.objects; | |
373 | } | |
374 | ||
1d07171c CL |
375 | /* Interrupts must be disabled (for the fallback code to work right) */ |
376 | static inline bool __cmpxchg_double_slab(struct kmem_cache *s, struct page *page, | |
377 | void *freelist_old, unsigned long counters_old, | |
378 | void *freelist_new, unsigned long counters_new, | |
379 | const char *n) | |
380 | { | |
381 | VM_BUG_ON(!irqs_disabled()); | |
2565409f HC |
382 | #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \ |
383 | defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE) | |
1d07171c | 384 | if (s->flags & __CMPXCHG_DOUBLE) { |
cdcd6298 | 385 | if (cmpxchg_double(&page->freelist, &page->counters, |
0aa9a13d DC |
386 | freelist_old, counters_old, |
387 | freelist_new, counters_new)) | |
6f6528a1 | 388 | return true; |
1d07171c CL |
389 | } else |
390 | #endif | |
391 | { | |
392 | slab_lock(page); | |
d0e0ac97 CG |
393 | if (page->freelist == freelist_old && |
394 | page->counters == counters_old) { | |
1d07171c | 395 | page->freelist = freelist_new; |
a0320865 | 396 | set_page_slub_counters(page, counters_new); |
1d07171c | 397 | slab_unlock(page); |
6f6528a1 | 398 | return true; |
1d07171c CL |
399 | } |
400 | slab_unlock(page); | |
401 | } | |
402 | ||
403 | cpu_relax(); | |
404 | stat(s, CMPXCHG_DOUBLE_FAIL); | |
405 | ||
406 | #ifdef SLUB_DEBUG_CMPXCHG | |
f9f58285 | 407 | pr_info("%s %s: cmpxchg double redo ", n, s->name); |
1d07171c CL |
408 | #endif |
409 | ||
6f6528a1 | 410 | return false; |
1d07171c CL |
411 | } |
412 | ||
b789ef51 CL |
413 | static inline bool cmpxchg_double_slab(struct kmem_cache *s, struct page *page, |
414 | void *freelist_old, unsigned long counters_old, | |
415 | void *freelist_new, unsigned long counters_new, | |
416 | const char *n) | |
417 | { | |
2565409f HC |
418 | #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \ |
419 | defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE) | |
b789ef51 | 420 | if (s->flags & __CMPXCHG_DOUBLE) { |
cdcd6298 | 421 | if (cmpxchg_double(&page->freelist, &page->counters, |
0aa9a13d DC |
422 | freelist_old, counters_old, |
423 | freelist_new, counters_new)) | |
6f6528a1 | 424 | return true; |
b789ef51 CL |
425 | } else |
426 | #endif | |
427 | { | |
1d07171c CL |
428 | unsigned long flags; |
429 | ||
430 | local_irq_save(flags); | |
881db7fb | 431 | slab_lock(page); |
d0e0ac97 CG |
432 | if (page->freelist == freelist_old && |
433 | page->counters == counters_old) { | |
b789ef51 | 434 | page->freelist = freelist_new; |
a0320865 | 435 | set_page_slub_counters(page, counters_new); |
881db7fb | 436 | slab_unlock(page); |
1d07171c | 437 | local_irq_restore(flags); |
6f6528a1 | 438 | return true; |
b789ef51 | 439 | } |
881db7fb | 440 | slab_unlock(page); |
1d07171c | 441 | local_irq_restore(flags); |
b789ef51 CL |
442 | } |
443 | ||
444 | cpu_relax(); | |
445 | stat(s, CMPXCHG_DOUBLE_FAIL); | |
446 | ||
447 | #ifdef SLUB_DEBUG_CMPXCHG | |
f9f58285 | 448 | pr_info("%s %s: cmpxchg double redo ", n, s->name); |
b789ef51 CL |
449 | #endif |
450 | ||
6f6528a1 | 451 | return false; |
b789ef51 CL |
452 | } |
453 | ||
41ecc55b | 454 | #ifdef CONFIG_SLUB_DEBUG |
5f80b13a CL |
455 | /* |
456 | * Determine a map of object in use on a page. | |
457 | * | |
881db7fb | 458 | * Node listlock must be held to guarantee that the page does |
5f80b13a CL |
459 | * not vanish from under us. |
460 | */ | |
461 | static void get_map(struct kmem_cache *s, struct page *page, unsigned long *map) | |
462 | { | |
463 | void *p; | |
464 | void *addr = page_address(page); | |
465 | ||
466 | for (p = page->freelist; p; p = get_freepointer(s, p)) | |
467 | set_bit(slab_index(p, s, addr), map); | |
468 | } | |
469 | ||
d86bd1be JK |
470 | static inline int size_from_object(struct kmem_cache *s) |
471 | { | |
472 | if (s->flags & SLAB_RED_ZONE) | |
473 | return s->size - s->red_left_pad; | |
474 | ||
475 | return s->size; | |
476 | } | |
477 | ||
478 | static inline void *restore_red_left(struct kmem_cache *s, void *p) | |
479 | { | |
480 | if (s->flags & SLAB_RED_ZONE) | |
481 | p -= s->red_left_pad; | |
482 | ||
483 | return p; | |
484 | } | |
485 | ||
41ecc55b CL |
486 | /* |
487 | * Debug settings: | |
488 | */ | |
89d3c87e | 489 | #if defined(CONFIG_SLUB_DEBUG_ON) |
d50112ed | 490 | static slab_flags_t slub_debug = DEBUG_DEFAULT_FLAGS; |
f0630fff | 491 | #else |
d50112ed | 492 | static slab_flags_t slub_debug; |
f0630fff | 493 | #endif |
41ecc55b CL |
494 | |
495 | static char *slub_debug_slabs; | |
fa5ec8a1 | 496 | static int disable_higher_order_debug; |
41ecc55b | 497 | |
a79316c6 AR |
498 | /* |
499 | * slub is about to manipulate internal object metadata. This memory lies | |
500 | * outside the range of the allocated object, so accessing it would normally | |
501 | * be reported by kasan as a bounds error. metadata_access_enable() is used | |
502 | * to tell kasan that these accesses are OK. | |
503 | */ | |
504 | static inline void metadata_access_enable(void) | |
505 | { | |
506 | kasan_disable_current(); | |
507 | } | |
508 | ||
509 | static inline void metadata_access_disable(void) | |
510 | { | |
511 | kasan_enable_current(); | |
512 | } | |
513 | ||
81819f0f CL |
514 | /* |
515 | * Object debugging | |
516 | */ | |
d86bd1be JK |
517 | |
518 | /* Verify that a pointer has an address that is valid within a slab page */ | |
519 | static inline int check_valid_pointer(struct kmem_cache *s, | |
520 | struct page *page, void *object) | |
521 | { | |
522 | void *base; | |
523 | ||
524 | if (!object) | |
525 | return 1; | |
526 | ||
527 | base = page_address(page); | |
528 | object = restore_red_left(s, object); | |
529 | if (object < base || object >= base + page->objects * s->size || | |
530 | (object - base) % s->size) { | |
531 | return 0; | |
532 | } | |
533 | ||
534 | return 1; | |
535 | } | |
536 | ||
aa2efd5e DT |
537 | static void print_section(char *level, char *text, u8 *addr, |
538 | unsigned int length) | |
81819f0f | 539 | { |
a79316c6 | 540 | metadata_access_enable(); |
aa2efd5e | 541 | print_hex_dump(level, text, DUMP_PREFIX_ADDRESS, 16, 1, addr, |
ffc79d28 | 542 | length, 1); |
a79316c6 | 543 | metadata_access_disable(); |
81819f0f CL |
544 | } |
545 | ||
81819f0f CL |
546 | static struct track *get_track(struct kmem_cache *s, void *object, |
547 | enum track_item alloc) | |
548 | { | |
549 | struct track *p; | |
550 | ||
551 | if (s->offset) | |
552 | p = object + s->offset + sizeof(void *); | |
553 | else | |
554 | p = object + s->inuse; | |
555 | ||
556 | return p + alloc; | |
557 | } | |
558 | ||
559 | static void set_track(struct kmem_cache *s, void *object, | |
ce71e27c | 560 | enum track_item alloc, unsigned long addr) |
81819f0f | 561 | { |
1a00df4a | 562 | struct track *p = get_track(s, object, alloc); |
81819f0f | 563 | |
81819f0f | 564 | if (addr) { |
d6543e39 BG |
565 | #ifdef CONFIG_STACKTRACE |
566 | struct stack_trace trace; | |
567 | int i; | |
568 | ||
569 | trace.nr_entries = 0; | |
570 | trace.max_entries = TRACK_ADDRS_COUNT; | |
571 | trace.entries = p->addrs; | |
572 | trace.skip = 3; | |
a79316c6 | 573 | metadata_access_enable(); |
d6543e39 | 574 | save_stack_trace(&trace); |
a79316c6 | 575 | metadata_access_disable(); |
d6543e39 BG |
576 | |
577 | /* See rant in lockdep.c */ | |
578 | if (trace.nr_entries != 0 && | |
579 | trace.entries[trace.nr_entries - 1] == ULONG_MAX) | |
580 | trace.nr_entries--; | |
581 | ||
582 | for (i = trace.nr_entries; i < TRACK_ADDRS_COUNT; i++) | |
583 | p->addrs[i] = 0; | |
584 | #endif | |
81819f0f CL |
585 | p->addr = addr; |
586 | p->cpu = smp_processor_id(); | |
88e4ccf2 | 587 | p->pid = current->pid; |
81819f0f CL |
588 | p->when = jiffies; |
589 | } else | |
590 | memset(p, 0, sizeof(struct track)); | |
591 | } | |
592 | ||
81819f0f CL |
593 | static void init_tracking(struct kmem_cache *s, void *object) |
594 | { | |
24922684 CL |
595 | if (!(s->flags & SLAB_STORE_USER)) |
596 | return; | |
597 | ||
ce71e27c EGM |
598 | set_track(s, object, TRACK_FREE, 0UL); |
599 | set_track(s, object, TRACK_ALLOC, 0UL); | |
81819f0f CL |
600 | } |
601 | ||
602 | static void print_track(const char *s, struct track *t) | |
603 | { | |
604 | if (!t->addr) | |
605 | return; | |
606 | ||
f9f58285 FF |
607 | pr_err("INFO: %s in %pS age=%lu cpu=%u pid=%d\n", |
608 | s, (void *)t->addr, jiffies - t->when, t->cpu, t->pid); | |
d6543e39 BG |
609 | #ifdef CONFIG_STACKTRACE |
610 | { | |
611 | int i; | |
612 | for (i = 0; i < TRACK_ADDRS_COUNT; i++) | |
613 | if (t->addrs[i]) | |
f9f58285 | 614 | pr_err("\t%pS\n", (void *)t->addrs[i]); |
d6543e39 BG |
615 | else |
616 | break; | |
617 | } | |
618 | #endif | |
24922684 CL |
619 | } |
620 | ||
621 | static void print_tracking(struct kmem_cache *s, void *object) | |
622 | { | |
623 | if (!(s->flags & SLAB_STORE_USER)) | |
624 | return; | |
625 | ||
626 | print_track("Allocated", get_track(s, object, TRACK_ALLOC)); | |
627 | print_track("Freed", get_track(s, object, TRACK_FREE)); | |
628 | } | |
629 | ||
630 | static void print_page_info(struct page *page) | |
631 | { | |
f9f58285 | 632 | pr_err("INFO: Slab 0x%p objects=%u used=%u fp=0x%p flags=0x%04lx\n", |
d0e0ac97 | 633 | page, page->objects, page->inuse, page->freelist, page->flags); |
24922684 CL |
634 | |
635 | } | |
636 | ||
637 | static void slab_bug(struct kmem_cache *s, char *fmt, ...) | |
638 | { | |
ecc42fbe | 639 | struct va_format vaf; |
24922684 | 640 | va_list args; |
24922684 CL |
641 | |
642 | va_start(args, fmt); | |
ecc42fbe FF |
643 | vaf.fmt = fmt; |
644 | vaf.va = &args; | |
f9f58285 | 645 | pr_err("=============================================================================\n"); |
ecc42fbe | 646 | pr_err("BUG %s (%s): %pV\n", s->name, print_tainted(), &vaf); |
f9f58285 | 647 | pr_err("-----------------------------------------------------------------------------\n\n"); |
645df230 | 648 | |
373d4d09 | 649 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); |
ecc42fbe | 650 | va_end(args); |
81819f0f CL |
651 | } |
652 | ||
24922684 CL |
653 | static void slab_fix(struct kmem_cache *s, char *fmt, ...) |
654 | { | |
ecc42fbe | 655 | struct va_format vaf; |
24922684 | 656 | va_list args; |
24922684 CL |
657 | |
658 | va_start(args, fmt); | |
ecc42fbe FF |
659 | vaf.fmt = fmt; |
660 | vaf.va = &args; | |
661 | pr_err("FIX %s: %pV\n", s->name, &vaf); | |
24922684 | 662 | va_end(args); |
24922684 CL |
663 | } |
664 | ||
665 | static void print_trailer(struct kmem_cache *s, struct page *page, u8 *p) | |
81819f0f CL |
666 | { |
667 | unsigned int off; /* Offset of last byte */ | |
a973e9dd | 668 | u8 *addr = page_address(page); |
24922684 CL |
669 | |
670 | print_tracking(s, p); | |
671 | ||
672 | print_page_info(page); | |
673 | ||
f9f58285 FF |
674 | pr_err("INFO: Object 0x%p @offset=%tu fp=0x%p\n\n", |
675 | p, p - addr, get_freepointer(s, p)); | |
24922684 | 676 | |
d86bd1be | 677 | if (s->flags & SLAB_RED_ZONE) |
aa2efd5e DT |
678 | print_section(KERN_ERR, "Redzone ", p - s->red_left_pad, |
679 | s->red_left_pad); | |
d86bd1be | 680 | else if (p > addr + 16) |
aa2efd5e | 681 | print_section(KERN_ERR, "Bytes b4 ", p - 16, 16); |
81819f0f | 682 | |
aa2efd5e DT |
683 | print_section(KERN_ERR, "Object ", p, |
684 | min_t(unsigned long, s->object_size, PAGE_SIZE)); | |
81819f0f | 685 | if (s->flags & SLAB_RED_ZONE) |
aa2efd5e | 686 | print_section(KERN_ERR, "Redzone ", p + s->object_size, |
3b0efdfa | 687 | s->inuse - s->object_size); |
81819f0f | 688 | |
81819f0f CL |
689 | if (s->offset) |
690 | off = s->offset + sizeof(void *); | |
691 | else | |
692 | off = s->inuse; | |
693 | ||
24922684 | 694 | if (s->flags & SLAB_STORE_USER) |
81819f0f | 695 | off += 2 * sizeof(struct track); |
81819f0f | 696 | |
80a9201a AP |
697 | off += kasan_metadata_size(s); |
698 | ||
d86bd1be | 699 | if (off != size_from_object(s)) |
81819f0f | 700 | /* Beginning of the filler is the free pointer */ |
aa2efd5e DT |
701 | print_section(KERN_ERR, "Padding ", p + off, |
702 | size_from_object(s) - off); | |
24922684 CL |
703 | |
704 | dump_stack(); | |
81819f0f CL |
705 | } |
706 | ||
75c66def | 707 | void object_err(struct kmem_cache *s, struct page *page, |
81819f0f CL |
708 | u8 *object, char *reason) |
709 | { | |
3dc50637 | 710 | slab_bug(s, "%s", reason); |
24922684 | 711 | print_trailer(s, page, object); |
81819f0f CL |
712 | } |
713 | ||
d0e0ac97 CG |
714 | static void slab_err(struct kmem_cache *s, struct page *page, |
715 | const char *fmt, ...) | |
81819f0f CL |
716 | { |
717 | va_list args; | |
718 | char buf[100]; | |
719 | ||
24922684 CL |
720 | va_start(args, fmt); |
721 | vsnprintf(buf, sizeof(buf), fmt, args); | |
81819f0f | 722 | va_end(args); |
3dc50637 | 723 | slab_bug(s, "%s", buf); |
24922684 | 724 | print_page_info(page); |
81819f0f CL |
725 | dump_stack(); |
726 | } | |
727 | ||
f7cb1933 | 728 | static void init_object(struct kmem_cache *s, void *object, u8 val) |
81819f0f CL |
729 | { |
730 | u8 *p = object; | |
731 | ||
d86bd1be JK |
732 | if (s->flags & SLAB_RED_ZONE) |
733 | memset(p - s->red_left_pad, val, s->red_left_pad); | |
734 | ||
81819f0f | 735 | if (s->flags & __OBJECT_POISON) { |
3b0efdfa CL |
736 | memset(p, POISON_FREE, s->object_size - 1); |
737 | p[s->object_size - 1] = POISON_END; | |
81819f0f CL |
738 | } |
739 | ||
740 | if (s->flags & SLAB_RED_ZONE) | |
3b0efdfa | 741 | memset(p + s->object_size, val, s->inuse - s->object_size); |
81819f0f CL |
742 | } |
743 | ||
24922684 CL |
744 | static void restore_bytes(struct kmem_cache *s, char *message, u8 data, |
745 | void *from, void *to) | |
746 | { | |
747 | slab_fix(s, "Restoring 0x%p-0x%p=0x%x\n", from, to - 1, data); | |
748 | memset(from, data, to - from); | |
749 | } | |
750 | ||
751 | static int check_bytes_and_report(struct kmem_cache *s, struct page *page, | |
752 | u8 *object, char *what, | |
06428780 | 753 | u8 *start, unsigned int value, unsigned int bytes) |
24922684 CL |
754 | { |
755 | u8 *fault; | |
756 | u8 *end; | |
757 | ||
a79316c6 | 758 | metadata_access_enable(); |
79824820 | 759 | fault = memchr_inv(start, value, bytes); |
a79316c6 | 760 | metadata_access_disable(); |
24922684 CL |
761 | if (!fault) |
762 | return 1; | |
763 | ||
764 | end = start + bytes; | |
765 | while (end > fault && end[-1] == value) | |
766 | end--; | |
767 | ||
768 | slab_bug(s, "%s overwritten", what); | |
f9f58285 | 769 | pr_err("INFO: 0x%p-0x%p. First byte 0x%x instead of 0x%x\n", |
24922684 CL |
770 | fault, end - 1, fault[0], value); |
771 | print_trailer(s, page, object); | |
772 | ||
773 | restore_bytes(s, what, value, fault, end); | |
774 | return 0; | |
81819f0f CL |
775 | } |
776 | ||
81819f0f CL |
777 | /* |
778 | * Object layout: | |
779 | * | |
780 | * object address | |
781 | * Bytes of the object to be managed. | |
782 | * If the freepointer may overlay the object then the free | |
783 | * pointer is the first word of the object. | |
672bba3a | 784 | * |
81819f0f CL |
785 | * Poisoning uses 0x6b (POISON_FREE) and the last byte is |
786 | * 0xa5 (POISON_END) | |
787 | * | |
3b0efdfa | 788 | * object + s->object_size |
81819f0f | 789 | * Padding to reach word boundary. This is also used for Redzoning. |
672bba3a | 790 | * Padding is extended by another word if Redzoning is enabled and |
3b0efdfa | 791 | * object_size == inuse. |
672bba3a | 792 | * |
81819f0f CL |
793 | * We fill with 0xbb (RED_INACTIVE) for inactive objects and with |
794 | * 0xcc (RED_ACTIVE) for objects in use. | |
795 | * | |
796 | * object + s->inuse | |
672bba3a CL |
797 | * Meta data starts here. |
798 | * | |
81819f0f CL |
799 | * A. Free pointer (if we cannot overwrite object on free) |
800 | * B. Tracking data for SLAB_STORE_USER | |
672bba3a | 801 | * C. Padding to reach required alignment boundary or at mininum |
6446faa2 | 802 | * one word if debugging is on to be able to detect writes |
672bba3a CL |
803 | * before the word boundary. |
804 | * | |
805 | * Padding is done using 0x5a (POISON_INUSE) | |
81819f0f CL |
806 | * |
807 | * object + s->size | |
672bba3a | 808 | * Nothing is used beyond s->size. |
81819f0f | 809 | * |
3b0efdfa | 810 | * If slabcaches are merged then the object_size and inuse boundaries are mostly |
672bba3a | 811 | * ignored. And therefore no slab options that rely on these boundaries |
81819f0f CL |
812 | * may be used with merged slabcaches. |
813 | */ | |
814 | ||
81819f0f CL |
815 | static int check_pad_bytes(struct kmem_cache *s, struct page *page, u8 *p) |
816 | { | |
817 | unsigned long off = s->inuse; /* The end of info */ | |
818 | ||
819 | if (s->offset) | |
820 | /* Freepointer is placed after the object. */ | |
821 | off += sizeof(void *); | |
822 | ||
823 | if (s->flags & SLAB_STORE_USER) | |
824 | /* We also have user information there */ | |
825 | off += 2 * sizeof(struct track); | |
826 | ||
80a9201a AP |
827 | off += kasan_metadata_size(s); |
828 | ||
d86bd1be | 829 | if (size_from_object(s) == off) |
81819f0f CL |
830 | return 1; |
831 | ||
24922684 | 832 | return check_bytes_and_report(s, page, p, "Object padding", |
d86bd1be | 833 | p + off, POISON_INUSE, size_from_object(s) - off); |
81819f0f CL |
834 | } |
835 | ||
39b26464 | 836 | /* Check the pad bytes at the end of a slab page */ |
81819f0f CL |
837 | static int slab_pad_check(struct kmem_cache *s, struct page *page) |
838 | { | |
24922684 CL |
839 | u8 *start; |
840 | u8 *fault; | |
841 | u8 *end; | |
842 | int length; | |
843 | int remainder; | |
81819f0f CL |
844 | |
845 | if (!(s->flags & SLAB_POISON)) | |
846 | return 1; | |
847 | ||
a973e9dd | 848 | start = page_address(page); |
ab9a0f19 | 849 | length = (PAGE_SIZE << compound_order(page)) - s->reserved; |
39b26464 CL |
850 | end = start + length; |
851 | remainder = length % s->size; | |
81819f0f CL |
852 | if (!remainder) |
853 | return 1; | |
854 | ||
a79316c6 | 855 | metadata_access_enable(); |
79824820 | 856 | fault = memchr_inv(end - remainder, POISON_INUSE, remainder); |
a79316c6 | 857 | metadata_access_disable(); |
24922684 CL |
858 | if (!fault) |
859 | return 1; | |
860 | while (end > fault && end[-1] == POISON_INUSE) | |
861 | end--; | |
862 | ||
863 | slab_err(s, page, "Padding overwritten. 0x%p-0x%p", fault, end - 1); | |
aa2efd5e | 864 | print_section(KERN_ERR, "Padding ", end - remainder, remainder); |
24922684 | 865 | |
8a3d271d | 866 | restore_bytes(s, "slab padding", POISON_INUSE, end - remainder, end); |
24922684 | 867 | return 0; |
81819f0f CL |
868 | } |
869 | ||
870 | static int check_object(struct kmem_cache *s, struct page *page, | |
f7cb1933 | 871 | void *object, u8 val) |
81819f0f CL |
872 | { |
873 | u8 *p = object; | |
3b0efdfa | 874 | u8 *endobject = object + s->object_size; |
81819f0f CL |
875 | |
876 | if (s->flags & SLAB_RED_ZONE) { | |
d86bd1be JK |
877 | if (!check_bytes_and_report(s, page, object, "Redzone", |
878 | object - s->red_left_pad, val, s->red_left_pad)) | |
879 | return 0; | |
880 | ||
24922684 | 881 | if (!check_bytes_and_report(s, page, object, "Redzone", |
3b0efdfa | 882 | endobject, val, s->inuse - s->object_size)) |
81819f0f | 883 | return 0; |
81819f0f | 884 | } else { |
3b0efdfa | 885 | if ((s->flags & SLAB_POISON) && s->object_size < s->inuse) { |
3adbefee | 886 | check_bytes_and_report(s, page, p, "Alignment padding", |
d0e0ac97 CG |
887 | endobject, POISON_INUSE, |
888 | s->inuse - s->object_size); | |
3adbefee | 889 | } |
81819f0f CL |
890 | } |
891 | ||
892 | if (s->flags & SLAB_POISON) { | |
f7cb1933 | 893 | if (val != SLUB_RED_ACTIVE && (s->flags & __OBJECT_POISON) && |
24922684 | 894 | (!check_bytes_and_report(s, page, p, "Poison", p, |
3b0efdfa | 895 | POISON_FREE, s->object_size - 1) || |
24922684 | 896 | !check_bytes_and_report(s, page, p, "Poison", |
3b0efdfa | 897 | p + s->object_size - 1, POISON_END, 1))) |
81819f0f | 898 | return 0; |
81819f0f CL |
899 | /* |
900 | * check_pad_bytes cleans up on its own. | |
901 | */ | |
902 | check_pad_bytes(s, page, p); | |
903 | } | |
904 | ||
f7cb1933 | 905 | if (!s->offset && val == SLUB_RED_ACTIVE) |
81819f0f CL |
906 | /* |
907 | * Object and freepointer overlap. Cannot check | |
908 | * freepointer while object is allocated. | |
909 | */ | |
910 | return 1; | |
911 | ||
912 | /* Check free pointer validity */ | |
913 | if (!check_valid_pointer(s, page, get_freepointer(s, p))) { | |
914 | object_err(s, page, p, "Freepointer corrupt"); | |
915 | /* | |
9f6c708e | 916 | * No choice but to zap it and thus lose the remainder |
81819f0f | 917 | * of the free objects in this slab. May cause |
672bba3a | 918 | * another error because the object count is now wrong. |
81819f0f | 919 | */ |
a973e9dd | 920 | set_freepointer(s, p, NULL); |
81819f0f CL |
921 | return 0; |
922 | } | |
923 | return 1; | |
924 | } | |
925 | ||
926 | static int check_slab(struct kmem_cache *s, struct page *page) | |
927 | { | |
39b26464 CL |
928 | int maxobj; |
929 | ||
81819f0f CL |
930 | VM_BUG_ON(!irqs_disabled()); |
931 | ||
932 | if (!PageSlab(page)) { | |
24922684 | 933 | slab_err(s, page, "Not a valid slab page"); |
81819f0f CL |
934 | return 0; |
935 | } | |
39b26464 | 936 | |
ab9a0f19 | 937 | maxobj = order_objects(compound_order(page), s->size, s->reserved); |
39b26464 CL |
938 | if (page->objects > maxobj) { |
939 | slab_err(s, page, "objects %u > max %u", | |
f6edde9c | 940 | page->objects, maxobj); |
39b26464 CL |
941 | return 0; |
942 | } | |
943 | if (page->inuse > page->objects) { | |
24922684 | 944 | slab_err(s, page, "inuse %u > max %u", |
f6edde9c | 945 | page->inuse, page->objects); |
81819f0f CL |
946 | return 0; |
947 | } | |
948 | /* Slab_pad_check fixes things up after itself */ | |
949 | slab_pad_check(s, page); | |
950 | return 1; | |
951 | } | |
952 | ||
953 | /* | |
672bba3a CL |
954 | * Determine if a certain object on a page is on the freelist. Must hold the |
955 | * slab lock to guarantee that the chains are in a consistent state. | |
81819f0f CL |
956 | */ |
957 | static int on_freelist(struct kmem_cache *s, struct page *page, void *search) | |
958 | { | |
959 | int nr = 0; | |
881db7fb | 960 | void *fp; |
81819f0f | 961 | void *object = NULL; |
f6edde9c | 962 | int max_objects; |
81819f0f | 963 | |
881db7fb | 964 | fp = page->freelist; |
39b26464 | 965 | while (fp && nr <= page->objects) { |
81819f0f CL |
966 | if (fp == search) |
967 | return 1; | |
968 | if (!check_valid_pointer(s, page, fp)) { | |
969 | if (object) { | |
970 | object_err(s, page, object, | |
971 | "Freechain corrupt"); | |
a973e9dd | 972 | set_freepointer(s, object, NULL); |
81819f0f | 973 | } else { |
24922684 | 974 | slab_err(s, page, "Freepointer corrupt"); |
a973e9dd | 975 | page->freelist = NULL; |
39b26464 | 976 | page->inuse = page->objects; |
24922684 | 977 | slab_fix(s, "Freelist cleared"); |
81819f0f CL |
978 | return 0; |
979 | } | |
980 | break; | |
981 | } | |
982 | object = fp; | |
983 | fp = get_freepointer(s, object); | |
984 | nr++; | |
985 | } | |
986 | ||
ab9a0f19 | 987 | max_objects = order_objects(compound_order(page), s->size, s->reserved); |
210b5c06 CG |
988 | if (max_objects > MAX_OBJS_PER_PAGE) |
989 | max_objects = MAX_OBJS_PER_PAGE; | |
224a88be CL |
990 | |
991 | if (page->objects != max_objects) { | |
756a025f JP |
992 | slab_err(s, page, "Wrong number of objects. Found %d but should be %d", |
993 | page->objects, max_objects); | |
224a88be CL |
994 | page->objects = max_objects; |
995 | slab_fix(s, "Number of objects adjusted."); | |
996 | } | |
39b26464 | 997 | if (page->inuse != page->objects - nr) { |
756a025f JP |
998 | slab_err(s, page, "Wrong object count. Counter is %d but counted were %d", |
999 | page->inuse, page->objects - nr); | |
39b26464 | 1000 | page->inuse = page->objects - nr; |
24922684 | 1001 | slab_fix(s, "Object count adjusted."); |
81819f0f CL |
1002 | } |
1003 | return search == NULL; | |
1004 | } | |
1005 | ||
0121c619 CL |
1006 | static void trace(struct kmem_cache *s, struct page *page, void *object, |
1007 | int alloc) | |
3ec09742 CL |
1008 | { |
1009 | if (s->flags & SLAB_TRACE) { | |
f9f58285 | 1010 | pr_info("TRACE %s %s 0x%p inuse=%d fp=0x%p\n", |
3ec09742 CL |
1011 | s->name, |
1012 | alloc ? "alloc" : "free", | |
1013 | object, page->inuse, | |
1014 | page->freelist); | |
1015 | ||
1016 | if (!alloc) | |
aa2efd5e | 1017 | print_section(KERN_INFO, "Object ", (void *)object, |
d0e0ac97 | 1018 | s->object_size); |
3ec09742 CL |
1019 | |
1020 | dump_stack(); | |
1021 | } | |
1022 | } | |
1023 | ||
643b1138 | 1024 | /* |
672bba3a | 1025 | * Tracking of fully allocated slabs for debugging purposes. |
643b1138 | 1026 | */ |
5cc6eee8 CL |
1027 | static void add_full(struct kmem_cache *s, |
1028 | struct kmem_cache_node *n, struct page *page) | |
643b1138 | 1029 | { |
5cc6eee8 CL |
1030 | if (!(s->flags & SLAB_STORE_USER)) |
1031 | return; | |
1032 | ||
255d0884 | 1033 | lockdep_assert_held(&n->list_lock); |
643b1138 | 1034 | list_add(&page->lru, &n->full); |
643b1138 CL |
1035 | } |
1036 | ||
c65c1877 | 1037 | static void remove_full(struct kmem_cache *s, struct kmem_cache_node *n, struct page *page) |
643b1138 | 1038 | { |
643b1138 CL |
1039 | if (!(s->flags & SLAB_STORE_USER)) |
1040 | return; | |
1041 | ||
255d0884 | 1042 | lockdep_assert_held(&n->list_lock); |
643b1138 | 1043 | list_del(&page->lru); |
643b1138 CL |
1044 | } |
1045 | ||
0f389ec6 CL |
1046 | /* Tracking of the number of slabs for debugging purposes */ |
1047 | static inline unsigned long slabs_node(struct kmem_cache *s, int node) | |
1048 | { | |
1049 | struct kmem_cache_node *n = get_node(s, node); | |
1050 | ||
1051 | return atomic_long_read(&n->nr_slabs); | |
1052 | } | |
1053 | ||
26c02cf0 AB |
1054 | static inline unsigned long node_nr_slabs(struct kmem_cache_node *n) |
1055 | { | |
1056 | return atomic_long_read(&n->nr_slabs); | |
1057 | } | |
1058 | ||
205ab99d | 1059 | static inline void inc_slabs_node(struct kmem_cache *s, int node, int objects) |
0f389ec6 CL |
1060 | { |
1061 | struct kmem_cache_node *n = get_node(s, node); | |
1062 | ||
1063 | /* | |
1064 | * May be called early in order to allocate a slab for the | |
1065 | * kmem_cache_node structure. Solve the chicken-egg | |
1066 | * dilemma by deferring the increment of the count during | |
1067 | * bootstrap (see early_kmem_cache_node_alloc). | |
1068 | */ | |
338b2642 | 1069 | if (likely(n)) { |
0f389ec6 | 1070 | atomic_long_inc(&n->nr_slabs); |
205ab99d CL |
1071 | atomic_long_add(objects, &n->total_objects); |
1072 | } | |
0f389ec6 | 1073 | } |
205ab99d | 1074 | static inline void dec_slabs_node(struct kmem_cache *s, int node, int objects) |
0f389ec6 CL |
1075 | { |
1076 | struct kmem_cache_node *n = get_node(s, node); | |
1077 | ||
1078 | atomic_long_dec(&n->nr_slabs); | |
205ab99d | 1079 | atomic_long_sub(objects, &n->total_objects); |
0f389ec6 CL |
1080 | } |
1081 | ||
1082 | /* Object debug checks for alloc/free paths */ | |
3ec09742 CL |
1083 | static void setup_object_debug(struct kmem_cache *s, struct page *page, |
1084 | void *object) | |
1085 | { | |
1086 | if (!(s->flags & (SLAB_STORE_USER|SLAB_RED_ZONE|__OBJECT_POISON))) | |
1087 | return; | |
1088 | ||
f7cb1933 | 1089 | init_object(s, object, SLUB_RED_INACTIVE); |
3ec09742 CL |
1090 | init_tracking(s, object); |
1091 | } | |
1092 | ||
becfda68 | 1093 | static inline int alloc_consistency_checks(struct kmem_cache *s, |
d0e0ac97 | 1094 | struct page *page, |
ce71e27c | 1095 | void *object, unsigned long addr) |
81819f0f CL |
1096 | { |
1097 | if (!check_slab(s, page)) | |
becfda68 | 1098 | return 0; |
81819f0f | 1099 | |
81819f0f CL |
1100 | if (!check_valid_pointer(s, page, object)) { |
1101 | object_err(s, page, object, "Freelist Pointer check fails"); | |
becfda68 | 1102 | return 0; |
81819f0f CL |
1103 | } |
1104 | ||
f7cb1933 | 1105 | if (!check_object(s, page, object, SLUB_RED_INACTIVE)) |
becfda68 LA |
1106 | return 0; |
1107 | ||
1108 | return 1; | |
1109 | } | |
1110 | ||
1111 | static noinline int alloc_debug_processing(struct kmem_cache *s, | |
1112 | struct page *page, | |
1113 | void *object, unsigned long addr) | |
1114 | { | |
1115 | if (s->flags & SLAB_CONSISTENCY_CHECKS) { | |
1116 | if (!alloc_consistency_checks(s, page, object, addr)) | |
1117 | goto bad; | |
1118 | } | |
81819f0f | 1119 | |
3ec09742 CL |
1120 | /* Success perform special debug activities for allocs */ |
1121 | if (s->flags & SLAB_STORE_USER) | |
1122 | set_track(s, object, TRACK_ALLOC, addr); | |
1123 | trace(s, page, object, 1); | |
f7cb1933 | 1124 | init_object(s, object, SLUB_RED_ACTIVE); |
81819f0f | 1125 | return 1; |
3ec09742 | 1126 | |
81819f0f CL |
1127 | bad: |
1128 | if (PageSlab(page)) { | |
1129 | /* | |
1130 | * If this is a slab page then lets do the best we can | |
1131 | * to avoid issues in the future. Marking all objects | |
672bba3a | 1132 | * as used avoids touching the remaining objects. |
81819f0f | 1133 | */ |
24922684 | 1134 | slab_fix(s, "Marking all objects used"); |
39b26464 | 1135 | page->inuse = page->objects; |
a973e9dd | 1136 | page->freelist = NULL; |
81819f0f CL |
1137 | } |
1138 | return 0; | |
1139 | } | |
1140 | ||
becfda68 LA |
1141 | static inline int free_consistency_checks(struct kmem_cache *s, |
1142 | struct page *page, void *object, unsigned long addr) | |
81819f0f | 1143 | { |
81819f0f | 1144 | if (!check_valid_pointer(s, page, object)) { |
70d71228 | 1145 | slab_err(s, page, "Invalid object pointer 0x%p", object); |
becfda68 | 1146 | return 0; |
81819f0f CL |
1147 | } |
1148 | ||
1149 | if (on_freelist(s, page, object)) { | |
24922684 | 1150 | object_err(s, page, object, "Object already free"); |
becfda68 | 1151 | return 0; |
81819f0f CL |
1152 | } |
1153 | ||
f7cb1933 | 1154 | if (!check_object(s, page, object, SLUB_RED_ACTIVE)) |
becfda68 | 1155 | return 0; |
81819f0f | 1156 | |
1b4f59e3 | 1157 | if (unlikely(s != page->slab_cache)) { |
3adbefee | 1158 | if (!PageSlab(page)) { |
756a025f JP |
1159 | slab_err(s, page, "Attempt to free object(0x%p) outside of slab", |
1160 | object); | |
1b4f59e3 | 1161 | } else if (!page->slab_cache) { |
f9f58285 FF |
1162 | pr_err("SLUB <none>: no slab for object 0x%p.\n", |
1163 | object); | |
70d71228 | 1164 | dump_stack(); |
06428780 | 1165 | } else |
24922684 CL |
1166 | object_err(s, page, object, |
1167 | "page slab pointer corrupt."); | |
becfda68 LA |
1168 | return 0; |
1169 | } | |
1170 | return 1; | |
1171 | } | |
1172 | ||
1173 | /* Supports checking bulk free of a constructed freelist */ | |
1174 | static noinline int free_debug_processing( | |
1175 | struct kmem_cache *s, struct page *page, | |
1176 | void *head, void *tail, int bulk_cnt, | |
1177 | unsigned long addr) | |
1178 | { | |
1179 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); | |
1180 | void *object = head; | |
1181 | int cnt = 0; | |
1182 | unsigned long uninitialized_var(flags); | |
1183 | int ret = 0; | |
1184 | ||
1185 | spin_lock_irqsave(&n->list_lock, flags); | |
1186 | slab_lock(page); | |
1187 | ||
1188 | if (s->flags & SLAB_CONSISTENCY_CHECKS) { | |
1189 | if (!check_slab(s, page)) | |
1190 | goto out; | |
1191 | } | |
1192 | ||
1193 | next_object: | |
1194 | cnt++; | |
1195 | ||
1196 | if (s->flags & SLAB_CONSISTENCY_CHECKS) { | |
1197 | if (!free_consistency_checks(s, page, object, addr)) | |
1198 | goto out; | |
81819f0f | 1199 | } |
3ec09742 | 1200 | |
3ec09742 CL |
1201 | if (s->flags & SLAB_STORE_USER) |
1202 | set_track(s, object, TRACK_FREE, addr); | |
1203 | trace(s, page, object, 0); | |
81084651 | 1204 | /* Freepointer not overwritten by init_object(), SLAB_POISON moved it */ |
f7cb1933 | 1205 | init_object(s, object, SLUB_RED_INACTIVE); |
81084651 JDB |
1206 | |
1207 | /* Reached end of constructed freelist yet? */ | |
1208 | if (object != tail) { | |
1209 | object = get_freepointer(s, object); | |
1210 | goto next_object; | |
1211 | } | |
804aa132 LA |
1212 | ret = 1; |
1213 | ||
5c2e4bbb | 1214 | out: |
81084651 JDB |
1215 | if (cnt != bulk_cnt) |
1216 | slab_err(s, page, "Bulk freelist count(%d) invalid(%d)\n", | |
1217 | bulk_cnt, cnt); | |
1218 | ||
881db7fb | 1219 | slab_unlock(page); |
282acb43 | 1220 | spin_unlock_irqrestore(&n->list_lock, flags); |
804aa132 LA |
1221 | if (!ret) |
1222 | slab_fix(s, "Object at 0x%p not freed", object); | |
1223 | return ret; | |
81819f0f CL |
1224 | } |
1225 | ||
41ecc55b CL |
1226 | static int __init setup_slub_debug(char *str) |
1227 | { | |
f0630fff CL |
1228 | slub_debug = DEBUG_DEFAULT_FLAGS; |
1229 | if (*str++ != '=' || !*str) | |
1230 | /* | |
1231 | * No options specified. Switch on full debugging. | |
1232 | */ | |
1233 | goto out; | |
1234 | ||
1235 | if (*str == ',') | |
1236 | /* | |
1237 | * No options but restriction on slabs. This means full | |
1238 | * debugging for slabs matching a pattern. | |
1239 | */ | |
1240 | goto check_slabs; | |
1241 | ||
1242 | slub_debug = 0; | |
1243 | if (*str == '-') | |
1244 | /* | |
1245 | * Switch off all debugging measures. | |
1246 | */ | |
1247 | goto out; | |
1248 | ||
1249 | /* | |
1250 | * Determine which debug features should be switched on | |
1251 | */ | |
06428780 | 1252 | for (; *str && *str != ','; str++) { |
f0630fff CL |
1253 | switch (tolower(*str)) { |
1254 | case 'f': | |
becfda68 | 1255 | slub_debug |= SLAB_CONSISTENCY_CHECKS; |
f0630fff CL |
1256 | break; |
1257 | case 'z': | |
1258 | slub_debug |= SLAB_RED_ZONE; | |
1259 | break; | |
1260 | case 'p': | |
1261 | slub_debug |= SLAB_POISON; | |
1262 | break; | |
1263 | case 'u': | |
1264 | slub_debug |= SLAB_STORE_USER; | |
1265 | break; | |
1266 | case 't': | |
1267 | slub_debug |= SLAB_TRACE; | |
1268 | break; | |
4c13dd3b DM |
1269 | case 'a': |
1270 | slub_debug |= SLAB_FAILSLAB; | |
1271 | break; | |
08303a73 CA |
1272 | case 'o': |
1273 | /* | |
1274 | * Avoid enabling debugging on caches if its minimum | |
1275 | * order would increase as a result. | |
1276 | */ | |
1277 | disable_higher_order_debug = 1; | |
1278 | break; | |
f0630fff | 1279 | default: |
f9f58285 FF |
1280 | pr_err("slub_debug option '%c' unknown. skipped\n", |
1281 | *str); | |
f0630fff | 1282 | } |
41ecc55b CL |
1283 | } |
1284 | ||
f0630fff | 1285 | check_slabs: |
41ecc55b CL |
1286 | if (*str == ',') |
1287 | slub_debug_slabs = str + 1; | |
f0630fff | 1288 | out: |
41ecc55b CL |
1289 | return 1; |
1290 | } | |
1291 | ||
1292 | __setup("slub_debug", setup_slub_debug); | |
1293 | ||
d50112ed AD |
1294 | slab_flags_t kmem_cache_flags(unsigned long object_size, |
1295 | slab_flags_t flags, const char *name, | |
51cc5068 | 1296 | void (*ctor)(void *)) |
41ecc55b CL |
1297 | { |
1298 | /* | |
e153362a | 1299 | * Enable debugging if selected on the kernel commandline. |
41ecc55b | 1300 | */ |
c6f58d9b CL |
1301 | if (slub_debug && (!slub_debug_slabs || (name && |
1302 | !strncmp(slub_debug_slabs, name, strlen(slub_debug_slabs))))) | |
3de47213 | 1303 | flags |= slub_debug; |
ba0268a8 CL |
1304 | |
1305 | return flags; | |
41ecc55b | 1306 | } |
b4a64718 | 1307 | #else /* !CONFIG_SLUB_DEBUG */ |
3ec09742 CL |
1308 | static inline void setup_object_debug(struct kmem_cache *s, |
1309 | struct page *page, void *object) {} | |
41ecc55b | 1310 | |
3ec09742 | 1311 | static inline int alloc_debug_processing(struct kmem_cache *s, |
ce71e27c | 1312 | struct page *page, void *object, unsigned long addr) { return 0; } |
41ecc55b | 1313 | |
282acb43 | 1314 | static inline int free_debug_processing( |
81084651 JDB |
1315 | struct kmem_cache *s, struct page *page, |
1316 | void *head, void *tail, int bulk_cnt, | |
282acb43 | 1317 | unsigned long addr) { return 0; } |
41ecc55b | 1318 | |
41ecc55b CL |
1319 | static inline int slab_pad_check(struct kmem_cache *s, struct page *page) |
1320 | { return 1; } | |
1321 | static inline int check_object(struct kmem_cache *s, struct page *page, | |
f7cb1933 | 1322 | void *object, u8 val) { return 1; } |
5cc6eee8 CL |
1323 | static inline void add_full(struct kmem_cache *s, struct kmem_cache_node *n, |
1324 | struct page *page) {} | |
c65c1877 PZ |
1325 | static inline void remove_full(struct kmem_cache *s, struct kmem_cache_node *n, |
1326 | struct page *page) {} | |
d50112ed AD |
1327 | slab_flags_t kmem_cache_flags(unsigned long object_size, |
1328 | slab_flags_t flags, const char *name, | |
51cc5068 | 1329 | void (*ctor)(void *)) |
ba0268a8 CL |
1330 | { |
1331 | return flags; | |
1332 | } | |
41ecc55b | 1333 | #define slub_debug 0 |
0f389ec6 | 1334 | |
fdaa45e9 IM |
1335 | #define disable_higher_order_debug 0 |
1336 | ||
0f389ec6 CL |
1337 | static inline unsigned long slabs_node(struct kmem_cache *s, int node) |
1338 | { return 0; } | |
26c02cf0 AB |
1339 | static inline unsigned long node_nr_slabs(struct kmem_cache_node *n) |
1340 | { return 0; } | |
205ab99d CL |
1341 | static inline void inc_slabs_node(struct kmem_cache *s, int node, |
1342 | int objects) {} | |
1343 | static inline void dec_slabs_node(struct kmem_cache *s, int node, | |
1344 | int objects) {} | |
7d550c56 | 1345 | |
02e72cc6 AR |
1346 | #endif /* CONFIG_SLUB_DEBUG */ |
1347 | ||
1348 | /* | |
1349 | * Hooks for other subsystems that check memory allocations. In a typical | |
1350 | * production configuration these hooks all should produce no code at all. | |
1351 | */ | |
d56791b3 RB |
1352 | static inline void kmalloc_large_node_hook(void *ptr, size_t size, gfp_t flags) |
1353 | { | |
1354 | kmemleak_alloc(ptr, size, 1, flags); | |
505f5dcb | 1355 | kasan_kmalloc_large(ptr, size, flags); |
d56791b3 RB |
1356 | } |
1357 | ||
1358 | static inline void kfree_hook(const void *x) | |
1359 | { | |
1360 | kmemleak_free(x); | |
0316bec2 | 1361 | kasan_kfree_large(x); |
d56791b3 RB |
1362 | } |
1363 | ||
80a9201a | 1364 | static inline void *slab_free_hook(struct kmem_cache *s, void *x) |
d56791b3 | 1365 | { |
80a9201a AP |
1366 | void *freeptr; |
1367 | ||
d56791b3 | 1368 | kmemleak_free_recursive(x, s->flags); |
7d550c56 | 1369 | |
02e72cc6 AR |
1370 | /* |
1371 | * Trouble is that we may no longer disable interrupts in the fast path | |
1372 | * So in order to make the debug calls that expect irqs to be | |
1373 | * disabled we need to disable interrupts temporarily. | |
1374 | */ | |
1375 | #if defined(CONFIG_KMEMCHECK) || defined(CONFIG_LOCKDEP) | |
1376 | { | |
1377 | unsigned long flags; | |
1378 | ||
1379 | local_irq_save(flags); | |
1380 | kmemcheck_slab_free(s, x, s->object_size); | |
1381 | debug_check_no_locks_freed(x, s->object_size); | |
1382 | local_irq_restore(flags); | |
1383 | } | |
1384 | #endif | |
1385 | if (!(s->flags & SLAB_DEBUG_OBJECTS)) | |
1386 | debug_check_no_obj_freed(x, s->object_size); | |
0316bec2 | 1387 | |
80a9201a AP |
1388 | freeptr = get_freepointer(s, x); |
1389 | /* | |
1390 | * kasan_slab_free() may put x into memory quarantine, delaying its | |
1391 | * reuse. In this case the object's freelist pointer is changed. | |
1392 | */ | |
0316bec2 | 1393 | kasan_slab_free(s, x); |
80a9201a | 1394 | return freeptr; |
02e72cc6 | 1395 | } |
205ab99d | 1396 | |
81084651 JDB |
1397 | static inline void slab_free_freelist_hook(struct kmem_cache *s, |
1398 | void *head, void *tail) | |
1399 | { | |
1400 | /* | |
1401 | * Compiler cannot detect this function can be removed if slab_free_hook() | |
1402 | * evaluates to nothing. Thus, catch all relevant config debug options here. | |
1403 | */ | |
1404 | #if defined(CONFIG_KMEMCHECK) || \ | |
1405 | defined(CONFIG_LOCKDEP) || \ | |
1406 | defined(CONFIG_DEBUG_KMEMLEAK) || \ | |
1407 | defined(CONFIG_DEBUG_OBJECTS_FREE) || \ | |
1408 | defined(CONFIG_KASAN) | |
1409 | ||
1410 | void *object = head; | |
1411 | void *tail_obj = tail ? : head; | |
80a9201a | 1412 | void *freeptr; |
81084651 JDB |
1413 | |
1414 | do { | |
80a9201a AP |
1415 | freeptr = slab_free_hook(s, object); |
1416 | } while ((object != tail_obj) && (object = freeptr)); | |
81084651 JDB |
1417 | #endif |
1418 | } | |
1419 | ||
588f8ba9 TG |
1420 | static void setup_object(struct kmem_cache *s, struct page *page, |
1421 | void *object) | |
1422 | { | |
1423 | setup_object_debug(s, page, object); | |
b3cbd9bf | 1424 | kasan_init_slab_obj(s, object); |
588f8ba9 TG |
1425 | if (unlikely(s->ctor)) { |
1426 | kasan_unpoison_object_data(s, object); | |
1427 | s->ctor(object); | |
1428 | kasan_poison_object_data(s, object); | |
1429 | } | |
1430 | } | |
1431 | ||
81819f0f CL |
1432 | /* |
1433 | * Slab allocation and freeing | |
1434 | */ | |
5dfb4175 VD |
1435 | static inline struct page *alloc_slab_page(struct kmem_cache *s, |
1436 | gfp_t flags, int node, struct kmem_cache_order_objects oo) | |
65c3376a | 1437 | { |
5dfb4175 | 1438 | struct page *page; |
65c3376a CL |
1439 | int order = oo_order(oo); |
1440 | ||
b1eeab67 VN |
1441 | flags |= __GFP_NOTRACK; |
1442 | ||
2154a336 | 1443 | if (node == NUMA_NO_NODE) |
5dfb4175 | 1444 | page = alloc_pages(flags, order); |
65c3376a | 1445 | else |
96db800f | 1446 | page = __alloc_pages_node(node, flags, order); |
5dfb4175 | 1447 | |
f3ccb2c4 VD |
1448 | if (page && memcg_charge_slab(page, flags, order, s)) { |
1449 | __free_pages(page, order); | |
1450 | page = NULL; | |
1451 | } | |
5dfb4175 VD |
1452 | |
1453 | return page; | |
65c3376a CL |
1454 | } |
1455 | ||
210e7a43 TG |
1456 | #ifdef CONFIG_SLAB_FREELIST_RANDOM |
1457 | /* Pre-initialize the random sequence cache */ | |
1458 | static int init_cache_random_seq(struct kmem_cache *s) | |
1459 | { | |
1460 | int err; | |
1461 | unsigned long i, count = oo_objects(s->oo); | |
1462 | ||
a810007a SR |
1463 | /* Bailout if already initialised */ |
1464 | if (s->random_seq) | |
1465 | return 0; | |
1466 | ||
210e7a43 TG |
1467 | err = cache_random_seq_create(s, count, GFP_KERNEL); |
1468 | if (err) { | |
1469 | pr_err("SLUB: Unable to initialize free list for %s\n", | |
1470 | s->name); | |
1471 | return err; | |
1472 | } | |
1473 | ||
1474 | /* Transform to an offset on the set of pages */ | |
1475 | if (s->random_seq) { | |
1476 | for (i = 0; i < count; i++) | |
1477 | s->random_seq[i] *= s->size; | |
1478 | } | |
1479 | return 0; | |
1480 | } | |
1481 | ||
1482 | /* Initialize each random sequence freelist per cache */ | |
1483 | static void __init init_freelist_randomization(void) | |
1484 | { | |
1485 | struct kmem_cache *s; | |
1486 | ||
1487 | mutex_lock(&slab_mutex); | |
1488 | ||
1489 | list_for_each_entry(s, &slab_caches, list) | |
1490 | init_cache_random_seq(s); | |
1491 | ||
1492 | mutex_unlock(&slab_mutex); | |
1493 | } | |
1494 | ||
1495 | /* Get the next entry on the pre-computed freelist randomized */ | |
1496 | static void *next_freelist_entry(struct kmem_cache *s, struct page *page, | |
1497 | unsigned long *pos, void *start, | |
1498 | unsigned long page_limit, | |
1499 | unsigned long freelist_count) | |
1500 | { | |
1501 | unsigned int idx; | |
1502 | ||
1503 | /* | |
1504 | * If the target page allocation failed, the number of objects on the | |
1505 | * page might be smaller than the usual size defined by the cache. | |
1506 | */ | |
1507 | do { | |
1508 | idx = s->random_seq[*pos]; | |
1509 | *pos += 1; | |
1510 | if (*pos >= freelist_count) | |
1511 | *pos = 0; | |
1512 | } while (unlikely(idx >= page_limit)); | |
1513 | ||
1514 | return (char *)start + idx; | |
1515 | } | |
1516 | ||
1517 | /* Shuffle the single linked freelist based on a random pre-computed sequence */ | |
1518 | static bool shuffle_freelist(struct kmem_cache *s, struct page *page) | |
1519 | { | |
1520 | void *start; | |
1521 | void *cur; | |
1522 | void *next; | |
1523 | unsigned long idx, pos, page_limit, freelist_count; | |
1524 | ||
1525 | if (page->objects < 2 || !s->random_seq) | |
1526 | return false; | |
1527 | ||
1528 | freelist_count = oo_objects(s->oo); | |
1529 | pos = get_random_int() % freelist_count; | |
1530 | ||
1531 | page_limit = page->objects * s->size; | |
1532 | start = fixup_red_left(s, page_address(page)); | |
1533 | ||
1534 | /* First entry is used as the base of the freelist */ | |
1535 | cur = next_freelist_entry(s, page, &pos, start, page_limit, | |
1536 | freelist_count); | |
1537 | page->freelist = cur; | |
1538 | ||
1539 | for (idx = 1; idx < page->objects; idx++) { | |
1540 | setup_object(s, page, cur); | |
1541 | next = next_freelist_entry(s, page, &pos, start, page_limit, | |
1542 | freelist_count); | |
1543 | set_freepointer(s, cur, next); | |
1544 | cur = next; | |
1545 | } | |
1546 | setup_object(s, page, cur); | |
1547 | set_freepointer(s, cur, NULL); | |
1548 | ||
1549 | return true; | |
1550 | } | |
1551 | #else | |
1552 | static inline int init_cache_random_seq(struct kmem_cache *s) | |
1553 | { | |
1554 | return 0; | |
1555 | } | |
1556 | static inline void init_freelist_randomization(void) { } | |
1557 | static inline bool shuffle_freelist(struct kmem_cache *s, struct page *page) | |
1558 | { | |
1559 | return false; | |
1560 | } | |
1561 | #endif /* CONFIG_SLAB_FREELIST_RANDOM */ | |
1562 | ||
81819f0f CL |
1563 | static struct page *allocate_slab(struct kmem_cache *s, gfp_t flags, int node) |
1564 | { | |
06428780 | 1565 | struct page *page; |
834f3d11 | 1566 | struct kmem_cache_order_objects oo = s->oo; |
ba52270d | 1567 | gfp_t alloc_gfp; |
588f8ba9 TG |
1568 | void *start, *p; |
1569 | int idx, order; | |
210e7a43 | 1570 | bool shuffle; |
81819f0f | 1571 | |
7e0528da CL |
1572 | flags &= gfp_allowed_mask; |
1573 | ||
d0164adc | 1574 | if (gfpflags_allow_blocking(flags)) |
7e0528da CL |
1575 | local_irq_enable(); |
1576 | ||
b7a49f0d | 1577 | flags |= s->allocflags; |
e12ba74d | 1578 | |
ba52270d PE |
1579 | /* |
1580 | * Let the initial higher-order allocation fail under memory pressure | |
1581 | * so we fall-back to the minimum order allocation. | |
1582 | */ | |
1583 | alloc_gfp = (flags | __GFP_NOWARN | __GFP_NORETRY) & ~__GFP_NOFAIL; | |
d0164adc | 1584 | if ((alloc_gfp & __GFP_DIRECT_RECLAIM) && oo_order(oo) > oo_order(s->min)) |
444eb2a4 | 1585 | alloc_gfp = (alloc_gfp | __GFP_NOMEMALLOC) & ~(__GFP_RECLAIM|__GFP_NOFAIL); |
ba52270d | 1586 | |
5dfb4175 | 1587 | page = alloc_slab_page(s, alloc_gfp, node, oo); |
65c3376a CL |
1588 | if (unlikely(!page)) { |
1589 | oo = s->min; | |
80c3a998 | 1590 | alloc_gfp = flags; |
65c3376a CL |
1591 | /* |
1592 | * Allocation may have failed due to fragmentation. | |
1593 | * Try a lower order alloc if possible | |
1594 | */ | |
5dfb4175 | 1595 | page = alloc_slab_page(s, alloc_gfp, node, oo); |
588f8ba9 TG |
1596 | if (unlikely(!page)) |
1597 | goto out; | |
1598 | stat(s, ORDER_FALLBACK); | |
65c3376a | 1599 | } |
5a896d9e | 1600 | |
588f8ba9 TG |
1601 | if (kmemcheck_enabled && |
1602 | !(s->flags & (SLAB_NOTRACK | DEBUG_DEFAULT_FLAGS))) { | |
b1eeab67 VN |
1603 | int pages = 1 << oo_order(oo); |
1604 | ||
80c3a998 | 1605 | kmemcheck_alloc_shadow(page, oo_order(oo), alloc_gfp, node); |
b1eeab67 VN |
1606 | |
1607 | /* | |
1608 | * Objects from caches that have a constructor don't get | |
1609 | * cleared when they're allocated, so we need to do it here. | |
1610 | */ | |
1611 | if (s->ctor) | |
1612 | kmemcheck_mark_uninitialized_pages(page, pages); | |
1613 | else | |
1614 | kmemcheck_mark_unallocated_pages(page, pages); | |
5a896d9e VN |
1615 | } |
1616 | ||
834f3d11 | 1617 | page->objects = oo_objects(oo); |
81819f0f | 1618 | |
1f458cbf | 1619 | order = compound_order(page); |
1b4f59e3 | 1620 | page->slab_cache = s; |
c03f94cc | 1621 | __SetPageSlab(page); |
2f064f34 | 1622 | if (page_is_pfmemalloc(page)) |
072bb0aa | 1623 | SetPageSlabPfmemalloc(page); |
81819f0f CL |
1624 | |
1625 | start = page_address(page); | |
81819f0f CL |
1626 | |
1627 | if (unlikely(s->flags & SLAB_POISON)) | |
1f458cbf | 1628 | memset(start, POISON_INUSE, PAGE_SIZE << order); |
81819f0f | 1629 | |
0316bec2 AR |
1630 | kasan_poison_slab(page); |
1631 | ||
210e7a43 TG |
1632 | shuffle = shuffle_freelist(s, page); |
1633 | ||
1634 | if (!shuffle) { | |
1635 | for_each_object_idx(p, idx, s, start, page->objects) { | |
1636 | setup_object(s, page, p); | |
1637 | if (likely(idx < page->objects)) | |
1638 | set_freepointer(s, p, p + s->size); | |
1639 | else | |
1640 | set_freepointer(s, p, NULL); | |
1641 | } | |
1642 | page->freelist = fixup_red_left(s, start); | |
81819f0f | 1643 | } |
81819f0f | 1644 | |
e6e82ea1 | 1645 | page->inuse = page->objects; |
8cb0a506 | 1646 | page->frozen = 1; |
588f8ba9 | 1647 | |
81819f0f | 1648 | out: |
d0164adc | 1649 | if (gfpflags_allow_blocking(flags)) |
588f8ba9 TG |
1650 | local_irq_disable(); |
1651 | if (!page) | |
1652 | return NULL; | |
1653 | ||
7779f212 | 1654 | mod_lruvec_page_state(page, |
588f8ba9 TG |
1655 | (s->flags & SLAB_RECLAIM_ACCOUNT) ? |
1656 | NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE, | |
1657 | 1 << oo_order(oo)); | |
1658 | ||
1659 | inc_slabs_node(s, page_to_nid(page), page->objects); | |
1660 | ||
81819f0f CL |
1661 | return page; |
1662 | } | |
1663 | ||
588f8ba9 TG |
1664 | static struct page *new_slab(struct kmem_cache *s, gfp_t flags, int node) |
1665 | { | |
1666 | if (unlikely(flags & GFP_SLAB_BUG_MASK)) { | |
bacdcb34 | 1667 | gfp_t invalid_mask = flags & GFP_SLAB_BUG_MASK; |
72baeef0 MH |
1668 | flags &= ~GFP_SLAB_BUG_MASK; |
1669 | pr_warn("Unexpected gfp: %#x (%pGg). Fixing up to gfp: %#x (%pGg). Fix your code!\n", | |
1670 | invalid_mask, &invalid_mask, flags, &flags); | |
65b9de75 | 1671 | dump_stack(); |
588f8ba9 TG |
1672 | } |
1673 | ||
1674 | return allocate_slab(s, | |
1675 | flags & (GFP_RECLAIM_MASK | GFP_CONSTRAINT_MASK), node); | |
1676 | } | |
1677 | ||
81819f0f CL |
1678 | static void __free_slab(struct kmem_cache *s, struct page *page) |
1679 | { | |
834f3d11 CL |
1680 | int order = compound_order(page); |
1681 | int pages = 1 << order; | |
81819f0f | 1682 | |
becfda68 | 1683 | if (s->flags & SLAB_CONSISTENCY_CHECKS) { |
81819f0f CL |
1684 | void *p; |
1685 | ||
1686 | slab_pad_check(s, page); | |
224a88be CL |
1687 | for_each_object(p, s, page_address(page), |
1688 | page->objects) | |
f7cb1933 | 1689 | check_object(s, page, p, SLUB_RED_INACTIVE); |
81819f0f CL |
1690 | } |
1691 | ||
b1eeab67 | 1692 | kmemcheck_free_shadow(page, compound_order(page)); |
5a896d9e | 1693 | |
7779f212 | 1694 | mod_lruvec_page_state(page, |
81819f0f CL |
1695 | (s->flags & SLAB_RECLAIM_ACCOUNT) ? |
1696 | NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE, | |
06428780 | 1697 | -pages); |
81819f0f | 1698 | |
072bb0aa | 1699 | __ClearPageSlabPfmemalloc(page); |
49bd5221 | 1700 | __ClearPageSlab(page); |
1f458cbf | 1701 | |
22b751c3 | 1702 | page_mapcount_reset(page); |
1eb5ac64 NP |
1703 | if (current->reclaim_state) |
1704 | current->reclaim_state->reclaimed_slab += pages; | |
27ee57c9 VD |
1705 | memcg_uncharge_slab(page, order, s); |
1706 | __free_pages(page, order); | |
81819f0f CL |
1707 | } |
1708 | ||
da9a638c LJ |
1709 | #define need_reserve_slab_rcu \ |
1710 | (sizeof(((struct page *)NULL)->lru) < sizeof(struct rcu_head)) | |
1711 | ||
81819f0f CL |
1712 | static void rcu_free_slab(struct rcu_head *h) |
1713 | { | |
1714 | struct page *page; | |
1715 | ||
da9a638c LJ |
1716 | if (need_reserve_slab_rcu) |
1717 | page = virt_to_head_page(h); | |
1718 | else | |
1719 | page = container_of((struct list_head *)h, struct page, lru); | |
1720 | ||
1b4f59e3 | 1721 | __free_slab(page->slab_cache, page); |
81819f0f CL |
1722 | } |
1723 | ||
1724 | static void free_slab(struct kmem_cache *s, struct page *page) | |
1725 | { | |
5f0d5a3a | 1726 | if (unlikely(s->flags & SLAB_TYPESAFE_BY_RCU)) { |
da9a638c LJ |
1727 | struct rcu_head *head; |
1728 | ||
1729 | if (need_reserve_slab_rcu) { | |
1730 | int order = compound_order(page); | |
1731 | int offset = (PAGE_SIZE << order) - s->reserved; | |
1732 | ||
1733 | VM_BUG_ON(s->reserved != sizeof(*head)); | |
1734 | head = page_address(page) + offset; | |
1735 | } else { | |
bc4f610d | 1736 | head = &page->rcu_head; |
da9a638c | 1737 | } |
81819f0f CL |
1738 | |
1739 | call_rcu(head, rcu_free_slab); | |
1740 | } else | |
1741 | __free_slab(s, page); | |
1742 | } | |
1743 | ||
1744 | static void discard_slab(struct kmem_cache *s, struct page *page) | |
1745 | { | |
205ab99d | 1746 | dec_slabs_node(s, page_to_nid(page), page->objects); |
81819f0f CL |
1747 | free_slab(s, page); |
1748 | } | |
1749 | ||
1750 | /* | |
5cc6eee8 | 1751 | * Management of partially allocated slabs. |
81819f0f | 1752 | */ |
1e4dd946 SR |
1753 | static inline void |
1754 | __add_partial(struct kmem_cache_node *n, struct page *page, int tail) | |
81819f0f | 1755 | { |
e95eed57 | 1756 | n->nr_partial++; |
136333d1 | 1757 | if (tail == DEACTIVATE_TO_TAIL) |
7c2e132c CL |
1758 | list_add_tail(&page->lru, &n->partial); |
1759 | else | |
1760 | list_add(&page->lru, &n->partial); | |
81819f0f CL |
1761 | } |
1762 | ||
1e4dd946 SR |
1763 | static inline void add_partial(struct kmem_cache_node *n, |
1764 | struct page *page, int tail) | |
62e346a8 | 1765 | { |
c65c1877 | 1766 | lockdep_assert_held(&n->list_lock); |
1e4dd946 SR |
1767 | __add_partial(n, page, tail); |
1768 | } | |
c65c1877 | 1769 | |
1e4dd946 SR |
1770 | static inline void remove_partial(struct kmem_cache_node *n, |
1771 | struct page *page) | |
1772 | { | |
1773 | lockdep_assert_held(&n->list_lock); | |
52b4b950 DS |
1774 | list_del(&page->lru); |
1775 | n->nr_partial--; | |
1e4dd946 SR |
1776 | } |
1777 | ||
81819f0f | 1778 | /* |
7ced3719 CL |
1779 | * Remove slab from the partial list, freeze it and |
1780 | * return the pointer to the freelist. | |
81819f0f | 1781 | * |
497b66f2 | 1782 | * Returns a list of objects or NULL if it fails. |
81819f0f | 1783 | */ |
497b66f2 | 1784 | static inline void *acquire_slab(struct kmem_cache *s, |
acd19fd1 | 1785 | struct kmem_cache_node *n, struct page *page, |
633b0764 | 1786 | int mode, int *objects) |
81819f0f | 1787 | { |
2cfb7455 CL |
1788 | void *freelist; |
1789 | unsigned long counters; | |
1790 | struct page new; | |
1791 | ||
c65c1877 PZ |
1792 | lockdep_assert_held(&n->list_lock); |
1793 | ||
2cfb7455 CL |
1794 | /* |
1795 | * Zap the freelist and set the frozen bit. | |
1796 | * The old freelist is the list of objects for the | |
1797 | * per cpu allocation list. | |
1798 | */ | |
7ced3719 CL |
1799 | freelist = page->freelist; |
1800 | counters = page->counters; | |
1801 | new.counters = counters; | |
633b0764 | 1802 | *objects = new.objects - new.inuse; |
23910c50 | 1803 | if (mode) { |
7ced3719 | 1804 | new.inuse = page->objects; |
23910c50 PE |
1805 | new.freelist = NULL; |
1806 | } else { | |
1807 | new.freelist = freelist; | |
1808 | } | |
2cfb7455 | 1809 | |
a0132ac0 | 1810 | VM_BUG_ON(new.frozen); |
7ced3719 | 1811 | new.frozen = 1; |
2cfb7455 | 1812 | |
7ced3719 | 1813 | if (!__cmpxchg_double_slab(s, page, |
2cfb7455 | 1814 | freelist, counters, |
02d7633f | 1815 | new.freelist, new.counters, |
7ced3719 | 1816 | "acquire_slab")) |
7ced3719 | 1817 | return NULL; |
2cfb7455 CL |
1818 | |
1819 | remove_partial(n, page); | |
7ced3719 | 1820 | WARN_ON(!freelist); |
49e22585 | 1821 | return freelist; |
81819f0f CL |
1822 | } |
1823 | ||
633b0764 | 1824 | static void put_cpu_partial(struct kmem_cache *s, struct page *page, int drain); |
8ba00bb6 | 1825 | static inline bool pfmemalloc_match(struct page *page, gfp_t gfpflags); |
49e22585 | 1826 | |
81819f0f | 1827 | /* |
672bba3a | 1828 | * Try to allocate a partial slab from a specific node. |
81819f0f | 1829 | */ |
8ba00bb6 JK |
1830 | static void *get_partial_node(struct kmem_cache *s, struct kmem_cache_node *n, |
1831 | struct kmem_cache_cpu *c, gfp_t flags) | |
81819f0f | 1832 | { |
49e22585 CL |
1833 | struct page *page, *page2; |
1834 | void *object = NULL; | |
633b0764 JK |
1835 | int available = 0; |
1836 | int objects; | |
81819f0f CL |
1837 | |
1838 | /* | |
1839 | * Racy check. If we mistakenly see no partial slabs then we | |
1840 | * just allocate an empty slab. If we mistakenly try to get a | |
672bba3a CL |
1841 | * partial slab and there is none available then get_partials() |
1842 | * will return NULL. | |
81819f0f CL |
1843 | */ |
1844 | if (!n || !n->nr_partial) | |
1845 | return NULL; | |
1846 | ||
1847 | spin_lock(&n->list_lock); | |
49e22585 | 1848 | list_for_each_entry_safe(page, page2, &n->partial, lru) { |
8ba00bb6 | 1849 | void *t; |
49e22585 | 1850 | |
8ba00bb6 JK |
1851 | if (!pfmemalloc_match(page, flags)) |
1852 | continue; | |
1853 | ||
633b0764 | 1854 | t = acquire_slab(s, n, page, object == NULL, &objects); |
49e22585 CL |
1855 | if (!t) |
1856 | break; | |
1857 | ||
633b0764 | 1858 | available += objects; |
12d79634 | 1859 | if (!object) { |
49e22585 | 1860 | c->page = page; |
49e22585 | 1861 | stat(s, ALLOC_FROM_PARTIAL); |
49e22585 | 1862 | object = t; |
49e22585 | 1863 | } else { |
633b0764 | 1864 | put_cpu_partial(s, page, 0); |
8028dcea | 1865 | stat(s, CPU_PARTIAL_NODE); |
49e22585 | 1866 | } |
345c905d | 1867 | if (!kmem_cache_has_cpu_partial(s) |
e6d0e1dc | 1868 | || available > slub_cpu_partial(s) / 2) |
49e22585 CL |
1869 | break; |
1870 | ||
497b66f2 | 1871 | } |
81819f0f | 1872 | spin_unlock(&n->list_lock); |
497b66f2 | 1873 | return object; |
81819f0f CL |
1874 | } |
1875 | ||
1876 | /* | |
672bba3a | 1877 | * Get a page from somewhere. Search in increasing NUMA distances. |
81819f0f | 1878 | */ |
de3ec035 | 1879 | static void *get_any_partial(struct kmem_cache *s, gfp_t flags, |
acd19fd1 | 1880 | struct kmem_cache_cpu *c) |
81819f0f CL |
1881 | { |
1882 | #ifdef CONFIG_NUMA | |
1883 | struct zonelist *zonelist; | |
dd1a239f | 1884 | struct zoneref *z; |
54a6eb5c MG |
1885 | struct zone *zone; |
1886 | enum zone_type high_zoneidx = gfp_zone(flags); | |
497b66f2 | 1887 | void *object; |
cc9a6c87 | 1888 | unsigned int cpuset_mems_cookie; |
81819f0f CL |
1889 | |
1890 | /* | |
672bba3a CL |
1891 | * The defrag ratio allows a configuration of the tradeoffs between |
1892 | * inter node defragmentation and node local allocations. A lower | |
1893 | * defrag_ratio increases the tendency to do local allocations | |
1894 | * instead of attempting to obtain partial slabs from other nodes. | |
81819f0f | 1895 | * |
672bba3a CL |
1896 | * If the defrag_ratio is set to 0 then kmalloc() always |
1897 | * returns node local objects. If the ratio is higher then kmalloc() | |
1898 | * may return off node objects because partial slabs are obtained | |
1899 | * from other nodes and filled up. | |
81819f0f | 1900 | * |
43efd3ea LP |
1901 | * If /sys/kernel/slab/xx/remote_node_defrag_ratio is set to 100 |
1902 | * (which makes defrag_ratio = 1000) then every (well almost) | |
1903 | * allocation will first attempt to defrag slab caches on other nodes. | |
1904 | * This means scanning over all nodes to look for partial slabs which | |
1905 | * may be expensive if we do it every time we are trying to find a slab | |
672bba3a | 1906 | * with available objects. |
81819f0f | 1907 | */ |
9824601e CL |
1908 | if (!s->remote_node_defrag_ratio || |
1909 | get_cycles() % 1024 > s->remote_node_defrag_ratio) | |
81819f0f CL |
1910 | return NULL; |
1911 | ||
cc9a6c87 | 1912 | do { |
d26914d1 | 1913 | cpuset_mems_cookie = read_mems_allowed_begin(); |
2a389610 | 1914 | zonelist = node_zonelist(mempolicy_slab_node(), flags); |
cc9a6c87 MG |
1915 | for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) { |
1916 | struct kmem_cache_node *n; | |
1917 | ||
1918 | n = get_node(s, zone_to_nid(zone)); | |
1919 | ||
dee2f8aa | 1920 | if (n && cpuset_zone_allowed(zone, flags) && |
cc9a6c87 | 1921 | n->nr_partial > s->min_partial) { |
8ba00bb6 | 1922 | object = get_partial_node(s, n, c, flags); |
cc9a6c87 MG |
1923 | if (object) { |
1924 | /* | |
d26914d1 MG |
1925 | * Don't check read_mems_allowed_retry() |
1926 | * here - if mems_allowed was updated in | |
1927 | * parallel, that was a harmless race | |
1928 | * between allocation and the cpuset | |
1929 | * update | |
cc9a6c87 | 1930 | */ |
cc9a6c87 MG |
1931 | return object; |
1932 | } | |
c0ff7453 | 1933 | } |
81819f0f | 1934 | } |
d26914d1 | 1935 | } while (read_mems_allowed_retry(cpuset_mems_cookie)); |
81819f0f CL |
1936 | #endif |
1937 | return NULL; | |
1938 | } | |
1939 | ||
1940 | /* | |
1941 | * Get a partial page, lock it and return it. | |
1942 | */ | |
497b66f2 | 1943 | static void *get_partial(struct kmem_cache *s, gfp_t flags, int node, |
acd19fd1 | 1944 | struct kmem_cache_cpu *c) |
81819f0f | 1945 | { |
497b66f2 | 1946 | void *object; |
a561ce00 JK |
1947 | int searchnode = node; |
1948 | ||
1949 | if (node == NUMA_NO_NODE) | |
1950 | searchnode = numa_mem_id(); | |
1951 | else if (!node_present_pages(node)) | |
1952 | searchnode = node_to_mem_node(node); | |
81819f0f | 1953 | |
8ba00bb6 | 1954 | object = get_partial_node(s, get_node(s, searchnode), c, flags); |
497b66f2 CL |
1955 | if (object || node != NUMA_NO_NODE) |
1956 | return object; | |
81819f0f | 1957 | |
acd19fd1 | 1958 | return get_any_partial(s, flags, c); |
81819f0f CL |
1959 | } |
1960 | ||
8a5ec0ba CL |
1961 | #ifdef CONFIG_PREEMPT |
1962 | /* | |
1963 | * Calculate the next globally unique transaction for disambiguiation | |
1964 | * during cmpxchg. The transactions start with the cpu number and are then | |
1965 | * incremented by CONFIG_NR_CPUS. | |
1966 | */ | |
1967 | #define TID_STEP roundup_pow_of_two(CONFIG_NR_CPUS) | |
1968 | #else | |
1969 | /* | |
1970 | * No preemption supported therefore also no need to check for | |
1971 | * different cpus. | |
1972 | */ | |
1973 | #define TID_STEP 1 | |
1974 | #endif | |
1975 | ||
1976 | static inline unsigned long next_tid(unsigned long tid) | |
1977 | { | |
1978 | return tid + TID_STEP; | |
1979 | } | |
1980 | ||
1981 | static inline unsigned int tid_to_cpu(unsigned long tid) | |
1982 | { | |
1983 | return tid % TID_STEP; | |
1984 | } | |
1985 | ||
1986 | static inline unsigned long tid_to_event(unsigned long tid) | |
1987 | { | |
1988 | return tid / TID_STEP; | |
1989 | } | |
1990 | ||
1991 | static inline unsigned int init_tid(int cpu) | |
1992 | { | |
1993 | return cpu; | |
1994 | } | |
1995 | ||
1996 | static inline void note_cmpxchg_failure(const char *n, | |
1997 | const struct kmem_cache *s, unsigned long tid) | |
1998 | { | |
1999 | #ifdef SLUB_DEBUG_CMPXCHG | |
2000 | unsigned long actual_tid = __this_cpu_read(s->cpu_slab->tid); | |
2001 | ||
f9f58285 | 2002 | pr_info("%s %s: cmpxchg redo ", n, s->name); |
8a5ec0ba CL |
2003 | |
2004 | #ifdef CONFIG_PREEMPT | |
2005 | if (tid_to_cpu(tid) != tid_to_cpu(actual_tid)) | |
f9f58285 | 2006 | pr_warn("due to cpu change %d -> %d\n", |
8a5ec0ba CL |
2007 | tid_to_cpu(tid), tid_to_cpu(actual_tid)); |
2008 | else | |
2009 | #endif | |
2010 | if (tid_to_event(tid) != tid_to_event(actual_tid)) | |
f9f58285 | 2011 | pr_warn("due to cpu running other code. Event %ld->%ld\n", |
8a5ec0ba CL |
2012 | tid_to_event(tid), tid_to_event(actual_tid)); |
2013 | else | |
f9f58285 | 2014 | pr_warn("for unknown reason: actual=%lx was=%lx target=%lx\n", |
8a5ec0ba CL |
2015 | actual_tid, tid, next_tid(tid)); |
2016 | #endif | |
4fdccdfb | 2017 | stat(s, CMPXCHG_DOUBLE_CPU_FAIL); |
8a5ec0ba CL |
2018 | } |
2019 | ||
788e1aad | 2020 | static void init_kmem_cache_cpus(struct kmem_cache *s) |
8a5ec0ba | 2021 | { |
8a5ec0ba CL |
2022 | int cpu; |
2023 | ||
2024 | for_each_possible_cpu(cpu) | |
2025 | per_cpu_ptr(s->cpu_slab, cpu)->tid = init_tid(cpu); | |
8a5ec0ba | 2026 | } |
2cfb7455 | 2027 | |
81819f0f CL |
2028 | /* |
2029 | * Remove the cpu slab | |
2030 | */ | |
d0e0ac97 | 2031 | static void deactivate_slab(struct kmem_cache *s, struct page *page, |
d4ff6d35 | 2032 | void *freelist, struct kmem_cache_cpu *c) |
81819f0f | 2033 | { |
2cfb7455 | 2034 | enum slab_modes { M_NONE, M_PARTIAL, M_FULL, M_FREE }; |
2cfb7455 CL |
2035 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); |
2036 | int lock = 0; | |
2037 | enum slab_modes l = M_NONE, m = M_NONE; | |
2cfb7455 | 2038 | void *nextfree; |
136333d1 | 2039 | int tail = DEACTIVATE_TO_HEAD; |
2cfb7455 CL |
2040 | struct page new; |
2041 | struct page old; | |
2042 | ||
2043 | if (page->freelist) { | |
84e554e6 | 2044 | stat(s, DEACTIVATE_REMOTE_FREES); |
136333d1 | 2045 | tail = DEACTIVATE_TO_TAIL; |
2cfb7455 CL |
2046 | } |
2047 | ||
894b8788 | 2048 | /* |
2cfb7455 CL |
2049 | * Stage one: Free all available per cpu objects back |
2050 | * to the page freelist while it is still frozen. Leave the | |
2051 | * last one. | |
2052 | * | |
2053 | * There is no need to take the list->lock because the page | |
2054 | * is still frozen. | |
2055 | */ | |
2056 | while (freelist && (nextfree = get_freepointer(s, freelist))) { | |
2057 | void *prior; | |
2058 | unsigned long counters; | |
2059 | ||
2060 | do { | |
2061 | prior = page->freelist; | |
2062 | counters = page->counters; | |
2063 | set_freepointer(s, freelist, prior); | |
2064 | new.counters = counters; | |
2065 | new.inuse--; | |
a0132ac0 | 2066 | VM_BUG_ON(!new.frozen); |
2cfb7455 | 2067 | |
1d07171c | 2068 | } while (!__cmpxchg_double_slab(s, page, |
2cfb7455 CL |
2069 | prior, counters, |
2070 | freelist, new.counters, | |
2071 | "drain percpu freelist")); | |
2072 | ||
2073 | freelist = nextfree; | |
2074 | } | |
2075 | ||
894b8788 | 2076 | /* |
2cfb7455 CL |
2077 | * Stage two: Ensure that the page is unfrozen while the |
2078 | * list presence reflects the actual number of objects | |
2079 | * during unfreeze. | |
2080 | * | |
2081 | * We setup the list membership and then perform a cmpxchg | |
2082 | * with the count. If there is a mismatch then the page | |
2083 | * is not unfrozen but the page is on the wrong list. | |
2084 | * | |
2085 | * Then we restart the process which may have to remove | |
2086 | * the page from the list that we just put it on again | |
2087 | * because the number of objects in the slab may have | |
2088 | * changed. | |
894b8788 | 2089 | */ |
2cfb7455 | 2090 | redo: |
894b8788 | 2091 | |
2cfb7455 CL |
2092 | old.freelist = page->freelist; |
2093 | old.counters = page->counters; | |
a0132ac0 | 2094 | VM_BUG_ON(!old.frozen); |
7c2e132c | 2095 | |
2cfb7455 CL |
2096 | /* Determine target state of the slab */ |
2097 | new.counters = old.counters; | |
2098 | if (freelist) { | |
2099 | new.inuse--; | |
2100 | set_freepointer(s, freelist, old.freelist); | |
2101 | new.freelist = freelist; | |
2102 | } else | |
2103 | new.freelist = old.freelist; | |
2104 | ||
2105 | new.frozen = 0; | |
2106 | ||
8a5b20ae | 2107 | if (!new.inuse && n->nr_partial >= s->min_partial) |
2cfb7455 CL |
2108 | m = M_FREE; |
2109 | else if (new.freelist) { | |
2110 | m = M_PARTIAL; | |
2111 | if (!lock) { | |
2112 | lock = 1; | |
2113 | /* | |
2114 | * Taking the spinlock removes the possiblity | |
2115 | * that acquire_slab() will see a slab page that | |
2116 | * is frozen | |
2117 | */ | |
2118 | spin_lock(&n->list_lock); | |
2119 | } | |
2120 | } else { | |
2121 | m = M_FULL; | |
2122 | if (kmem_cache_debug(s) && !lock) { | |
2123 | lock = 1; | |
2124 | /* | |
2125 | * This also ensures that the scanning of full | |
2126 | * slabs from diagnostic functions will not see | |
2127 | * any frozen slabs. | |
2128 | */ | |
2129 | spin_lock(&n->list_lock); | |
2130 | } | |
2131 | } | |
2132 | ||
2133 | if (l != m) { | |
2134 | ||
2135 | if (l == M_PARTIAL) | |
2136 | ||
2137 | remove_partial(n, page); | |
2138 | ||
2139 | else if (l == M_FULL) | |
894b8788 | 2140 | |
c65c1877 | 2141 | remove_full(s, n, page); |
2cfb7455 CL |
2142 | |
2143 | if (m == M_PARTIAL) { | |
2144 | ||
2145 | add_partial(n, page, tail); | |
136333d1 | 2146 | stat(s, tail); |
2cfb7455 CL |
2147 | |
2148 | } else if (m == M_FULL) { | |
894b8788 | 2149 | |
2cfb7455 CL |
2150 | stat(s, DEACTIVATE_FULL); |
2151 | add_full(s, n, page); | |
2152 | ||
2153 | } | |
2154 | } | |
2155 | ||
2156 | l = m; | |
1d07171c | 2157 | if (!__cmpxchg_double_slab(s, page, |
2cfb7455 CL |
2158 | old.freelist, old.counters, |
2159 | new.freelist, new.counters, | |
2160 | "unfreezing slab")) | |
2161 | goto redo; | |
2162 | ||
2cfb7455 CL |
2163 | if (lock) |
2164 | spin_unlock(&n->list_lock); | |
2165 | ||
2166 | if (m == M_FREE) { | |
2167 | stat(s, DEACTIVATE_EMPTY); | |
2168 | discard_slab(s, page); | |
2169 | stat(s, FREE_SLAB); | |
894b8788 | 2170 | } |
d4ff6d35 WY |
2171 | |
2172 | c->page = NULL; | |
2173 | c->freelist = NULL; | |
81819f0f CL |
2174 | } |
2175 | ||
d24ac77f JK |
2176 | /* |
2177 | * Unfreeze all the cpu partial slabs. | |
2178 | * | |
59a09917 CL |
2179 | * This function must be called with interrupts disabled |
2180 | * for the cpu using c (or some other guarantee must be there | |
2181 | * to guarantee no concurrent accesses). | |
d24ac77f | 2182 | */ |
59a09917 CL |
2183 | static void unfreeze_partials(struct kmem_cache *s, |
2184 | struct kmem_cache_cpu *c) | |
49e22585 | 2185 | { |
345c905d | 2186 | #ifdef CONFIG_SLUB_CPU_PARTIAL |
43d77867 | 2187 | struct kmem_cache_node *n = NULL, *n2 = NULL; |
9ada1934 | 2188 | struct page *page, *discard_page = NULL; |
49e22585 CL |
2189 | |
2190 | while ((page = c->partial)) { | |
49e22585 CL |
2191 | struct page new; |
2192 | struct page old; | |
2193 | ||
2194 | c->partial = page->next; | |
43d77867 JK |
2195 | |
2196 | n2 = get_node(s, page_to_nid(page)); | |
2197 | if (n != n2) { | |
2198 | if (n) | |
2199 | spin_unlock(&n->list_lock); | |
2200 | ||
2201 | n = n2; | |
2202 | spin_lock(&n->list_lock); | |
2203 | } | |
49e22585 CL |
2204 | |
2205 | do { | |
2206 | ||
2207 | old.freelist = page->freelist; | |
2208 | old.counters = page->counters; | |
a0132ac0 | 2209 | VM_BUG_ON(!old.frozen); |
49e22585 CL |
2210 | |
2211 | new.counters = old.counters; | |
2212 | new.freelist = old.freelist; | |
2213 | ||
2214 | new.frozen = 0; | |
2215 | ||
d24ac77f | 2216 | } while (!__cmpxchg_double_slab(s, page, |
49e22585 CL |
2217 | old.freelist, old.counters, |
2218 | new.freelist, new.counters, | |
2219 | "unfreezing slab")); | |
2220 | ||
8a5b20ae | 2221 | if (unlikely(!new.inuse && n->nr_partial >= s->min_partial)) { |
9ada1934 SL |
2222 | page->next = discard_page; |
2223 | discard_page = page; | |
43d77867 JK |
2224 | } else { |
2225 | add_partial(n, page, DEACTIVATE_TO_TAIL); | |
2226 | stat(s, FREE_ADD_PARTIAL); | |
49e22585 CL |
2227 | } |
2228 | } | |
2229 | ||
2230 | if (n) | |
2231 | spin_unlock(&n->list_lock); | |
9ada1934 SL |
2232 | |
2233 | while (discard_page) { | |
2234 | page = discard_page; | |
2235 | discard_page = discard_page->next; | |
2236 | ||
2237 | stat(s, DEACTIVATE_EMPTY); | |
2238 | discard_slab(s, page); | |
2239 | stat(s, FREE_SLAB); | |
2240 | } | |
345c905d | 2241 | #endif |
49e22585 CL |
2242 | } |
2243 | ||
2244 | /* | |
2245 | * Put a page that was just frozen (in __slab_free) into a partial page | |
2246 | * slot if available. This is done without interrupts disabled and without | |
2247 | * preemption disabled. The cmpxchg is racy and may put the partial page | |
2248 | * onto a random cpus partial slot. | |
2249 | * | |
2250 | * If we did not find a slot then simply move all the partials to the | |
2251 | * per node partial list. | |
2252 | */ | |
633b0764 | 2253 | static void put_cpu_partial(struct kmem_cache *s, struct page *page, int drain) |
49e22585 | 2254 | { |
345c905d | 2255 | #ifdef CONFIG_SLUB_CPU_PARTIAL |
49e22585 CL |
2256 | struct page *oldpage; |
2257 | int pages; | |
2258 | int pobjects; | |
2259 | ||
d6e0b7fa | 2260 | preempt_disable(); |
49e22585 CL |
2261 | do { |
2262 | pages = 0; | |
2263 | pobjects = 0; | |
2264 | oldpage = this_cpu_read(s->cpu_slab->partial); | |
2265 | ||
2266 | if (oldpage) { | |
2267 | pobjects = oldpage->pobjects; | |
2268 | pages = oldpage->pages; | |
2269 | if (drain && pobjects > s->cpu_partial) { | |
2270 | unsigned long flags; | |
2271 | /* | |
2272 | * partial array is full. Move the existing | |
2273 | * set to the per node partial list. | |
2274 | */ | |
2275 | local_irq_save(flags); | |
59a09917 | 2276 | unfreeze_partials(s, this_cpu_ptr(s->cpu_slab)); |
49e22585 | 2277 | local_irq_restore(flags); |
e24fc410 | 2278 | oldpage = NULL; |
49e22585 CL |
2279 | pobjects = 0; |
2280 | pages = 0; | |
8028dcea | 2281 | stat(s, CPU_PARTIAL_DRAIN); |
49e22585 CL |
2282 | } |
2283 | } | |
2284 | ||
2285 | pages++; | |
2286 | pobjects += page->objects - page->inuse; | |
2287 | ||
2288 | page->pages = pages; | |
2289 | page->pobjects = pobjects; | |
2290 | page->next = oldpage; | |
2291 | ||
d0e0ac97 CG |
2292 | } while (this_cpu_cmpxchg(s->cpu_slab->partial, oldpage, page) |
2293 | != oldpage); | |
d6e0b7fa VD |
2294 | if (unlikely(!s->cpu_partial)) { |
2295 | unsigned long flags; | |
2296 | ||
2297 | local_irq_save(flags); | |
2298 | unfreeze_partials(s, this_cpu_ptr(s->cpu_slab)); | |
2299 | local_irq_restore(flags); | |
2300 | } | |
2301 | preempt_enable(); | |
345c905d | 2302 | #endif |
49e22585 CL |
2303 | } |
2304 | ||
dfb4f096 | 2305 | static inline void flush_slab(struct kmem_cache *s, struct kmem_cache_cpu *c) |
81819f0f | 2306 | { |
84e554e6 | 2307 | stat(s, CPUSLAB_FLUSH); |
d4ff6d35 | 2308 | deactivate_slab(s, c->page, c->freelist, c); |
c17dda40 CL |
2309 | |
2310 | c->tid = next_tid(c->tid); | |
81819f0f CL |
2311 | } |
2312 | ||
2313 | /* | |
2314 | * Flush cpu slab. | |
6446faa2 | 2315 | * |
81819f0f CL |
2316 | * Called from IPI handler with interrupts disabled. |
2317 | */ | |
0c710013 | 2318 | static inline void __flush_cpu_slab(struct kmem_cache *s, int cpu) |
81819f0f | 2319 | { |
9dfc6e68 | 2320 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu); |
81819f0f | 2321 | |
49e22585 CL |
2322 | if (likely(c)) { |
2323 | if (c->page) | |
2324 | flush_slab(s, c); | |
2325 | ||
59a09917 | 2326 | unfreeze_partials(s, c); |
49e22585 | 2327 | } |
81819f0f CL |
2328 | } |
2329 | ||
2330 | static void flush_cpu_slab(void *d) | |
2331 | { | |
2332 | struct kmem_cache *s = d; | |
81819f0f | 2333 | |
dfb4f096 | 2334 | __flush_cpu_slab(s, smp_processor_id()); |
81819f0f CL |
2335 | } |
2336 | ||
a8364d55 GBY |
2337 | static bool has_cpu_slab(int cpu, void *info) |
2338 | { | |
2339 | struct kmem_cache *s = info; | |
2340 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu); | |
2341 | ||
a93cf07b | 2342 | return c->page || slub_percpu_partial(c); |
a8364d55 GBY |
2343 | } |
2344 | ||
81819f0f CL |
2345 | static void flush_all(struct kmem_cache *s) |
2346 | { | |
a8364d55 | 2347 | on_each_cpu_cond(has_cpu_slab, flush_cpu_slab, s, 1, GFP_ATOMIC); |
81819f0f CL |
2348 | } |
2349 | ||
a96a87bf SAS |
2350 | /* |
2351 | * Use the cpu notifier to insure that the cpu slabs are flushed when | |
2352 | * necessary. | |
2353 | */ | |
2354 | static int slub_cpu_dead(unsigned int cpu) | |
2355 | { | |
2356 | struct kmem_cache *s; | |
2357 | unsigned long flags; | |
2358 | ||
2359 | mutex_lock(&slab_mutex); | |
2360 | list_for_each_entry(s, &slab_caches, list) { | |
2361 | local_irq_save(flags); | |
2362 | __flush_cpu_slab(s, cpu); | |
2363 | local_irq_restore(flags); | |
2364 | } | |
2365 | mutex_unlock(&slab_mutex); | |
2366 | return 0; | |
2367 | } | |
2368 | ||
dfb4f096 CL |
2369 | /* |
2370 | * Check if the objects in a per cpu structure fit numa | |
2371 | * locality expectations. | |
2372 | */ | |
57d437d2 | 2373 | static inline int node_match(struct page *page, int node) |
dfb4f096 CL |
2374 | { |
2375 | #ifdef CONFIG_NUMA | |
4d7868e6 | 2376 | if (!page || (node != NUMA_NO_NODE && page_to_nid(page) != node)) |
dfb4f096 CL |
2377 | return 0; |
2378 | #endif | |
2379 | return 1; | |
2380 | } | |
2381 | ||
9a02d699 | 2382 | #ifdef CONFIG_SLUB_DEBUG |
781b2ba6 PE |
2383 | static int count_free(struct page *page) |
2384 | { | |
2385 | return page->objects - page->inuse; | |
2386 | } | |
2387 | ||
9a02d699 DR |
2388 | static inline unsigned long node_nr_objs(struct kmem_cache_node *n) |
2389 | { | |
2390 | return atomic_long_read(&n->total_objects); | |
2391 | } | |
2392 | #endif /* CONFIG_SLUB_DEBUG */ | |
2393 | ||
2394 | #if defined(CONFIG_SLUB_DEBUG) || defined(CONFIG_SYSFS) | |
781b2ba6 PE |
2395 | static unsigned long count_partial(struct kmem_cache_node *n, |
2396 | int (*get_count)(struct page *)) | |
2397 | { | |
2398 | unsigned long flags; | |
2399 | unsigned long x = 0; | |
2400 | struct page *page; | |
2401 | ||
2402 | spin_lock_irqsave(&n->list_lock, flags); | |
2403 | list_for_each_entry(page, &n->partial, lru) | |
2404 | x += get_count(page); | |
2405 | spin_unlock_irqrestore(&n->list_lock, flags); | |
2406 | return x; | |
2407 | } | |
9a02d699 | 2408 | #endif /* CONFIG_SLUB_DEBUG || CONFIG_SYSFS */ |
26c02cf0 | 2409 | |
781b2ba6 PE |
2410 | static noinline void |
2411 | slab_out_of_memory(struct kmem_cache *s, gfp_t gfpflags, int nid) | |
2412 | { | |
9a02d699 DR |
2413 | #ifdef CONFIG_SLUB_DEBUG |
2414 | static DEFINE_RATELIMIT_STATE(slub_oom_rs, DEFAULT_RATELIMIT_INTERVAL, | |
2415 | DEFAULT_RATELIMIT_BURST); | |
781b2ba6 | 2416 | int node; |
fa45dc25 | 2417 | struct kmem_cache_node *n; |
781b2ba6 | 2418 | |
9a02d699 DR |
2419 | if ((gfpflags & __GFP_NOWARN) || !__ratelimit(&slub_oom_rs)) |
2420 | return; | |
2421 | ||
5b3810e5 VB |
2422 | pr_warn("SLUB: Unable to allocate memory on node %d, gfp=%#x(%pGg)\n", |
2423 | nid, gfpflags, &gfpflags); | |
f9f58285 FF |
2424 | pr_warn(" cache: %s, object size: %d, buffer size: %d, default order: %d, min order: %d\n", |
2425 | s->name, s->object_size, s->size, oo_order(s->oo), | |
2426 | oo_order(s->min)); | |
781b2ba6 | 2427 | |
3b0efdfa | 2428 | if (oo_order(s->min) > get_order(s->object_size)) |
f9f58285 FF |
2429 | pr_warn(" %s debugging increased min order, use slub_debug=O to disable.\n", |
2430 | s->name); | |
fa5ec8a1 | 2431 | |
fa45dc25 | 2432 | for_each_kmem_cache_node(s, node, n) { |
781b2ba6 PE |
2433 | unsigned long nr_slabs; |
2434 | unsigned long nr_objs; | |
2435 | unsigned long nr_free; | |
2436 | ||
26c02cf0 AB |
2437 | nr_free = count_partial(n, count_free); |
2438 | nr_slabs = node_nr_slabs(n); | |
2439 | nr_objs = node_nr_objs(n); | |
781b2ba6 | 2440 | |
f9f58285 | 2441 | pr_warn(" node %d: slabs: %ld, objs: %ld, free: %ld\n", |
781b2ba6 PE |
2442 | node, nr_slabs, nr_objs, nr_free); |
2443 | } | |
9a02d699 | 2444 | #endif |
781b2ba6 PE |
2445 | } |
2446 | ||
497b66f2 CL |
2447 | static inline void *new_slab_objects(struct kmem_cache *s, gfp_t flags, |
2448 | int node, struct kmem_cache_cpu **pc) | |
2449 | { | |
6faa6833 | 2450 | void *freelist; |
188fd063 CL |
2451 | struct kmem_cache_cpu *c = *pc; |
2452 | struct page *page; | |
497b66f2 | 2453 | |
188fd063 | 2454 | freelist = get_partial(s, flags, node, c); |
497b66f2 | 2455 | |
188fd063 CL |
2456 | if (freelist) |
2457 | return freelist; | |
2458 | ||
2459 | page = new_slab(s, flags, node); | |
497b66f2 | 2460 | if (page) { |
7c8e0181 | 2461 | c = raw_cpu_ptr(s->cpu_slab); |
497b66f2 CL |
2462 | if (c->page) |
2463 | flush_slab(s, c); | |
2464 | ||
2465 | /* | |
2466 | * No other reference to the page yet so we can | |
2467 | * muck around with it freely without cmpxchg | |
2468 | */ | |
6faa6833 | 2469 | freelist = page->freelist; |
497b66f2 CL |
2470 | page->freelist = NULL; |
2471 | ||
2472 | stat(s, ALLOC_SLAB); | |
497b66f2 CL |
2473 | c->page = page; |
2474 | *pc = c; | |
2475 | } else | |
6faa6833 | 2476 | freelist = NULL; |
497b66f2 | 2477 | |
6faa6833 | 2478 | return freelist; |
497b66f2 CL |
2479 | } |
2480 | ||
072bb0aa MG |
2481 | static inline bool pfmemalloc_match(struct page *page, gfp_t gfpflags) |
2482 | { | |
2483 | if (unlikely(PageSlabPfmemalloc(page))) | |
2484 | return gfp_pfmemalloc_allowed(gfpflags); | |
2485 | ||
2486 | return true; | |
2487 | } | |
2488 | ||
213eeb9f | 2489 | /* |
d0e0ac97 CG |
2490 | * Check the page->freelist of a page and either transfer the freelist to the |
2491 | * per cpu freelist or deactivate the page. | |
213eeb9f CL |
2492 | * |
2493 | * The page is still frozen if the return value is not NULL. | |
2494 | * | |
2495 | * If this function returns NULL then the page has been unfrozen. | |
d24ac77f JK |
2496 | * |
2497 | * This function must be called with interrupt disabled. | |
213eeb9f CL |
2498 | */ |
2499 | static inline void *get_freelist(struct kmem_cache *s, struct page *page) | |
2500 | { | |
2501 | struct page new; | |
2502 | unsigned long counters; | |
2503 | void *freelist; | |
2504 | ||
2505 | do { | |
2506 | freelist = page->freelist; | |
2507 | counters = page->counters; | |
6faa6833 | 2508 | |
213eeb9f | 2509 | new.counters = counters; |
a0132ac0 | 2510 | VM_BUG_ON(!new.frozen); |
213eeb9f CL |
2511 | |
2512 | new.inuse = page->objects; | |
2513 | new.frozen = freelist != NULL; | |
2514 | ||
d24ac77f | 2515 | } while (!__cmpxchg_double_slab(s, page, |
213eeb9f CL |
2516 | freelist, counters, |
2517 | NULL, new.counters, | |
2518 | "get_freelist")); | |
2519 | ||
2520 | return freelist; | |
2521 | } | |
2522 | ||
81819f0f | 2523 | /* |
894b8788 CL |
2524 | * Slow path. The lockless freelist is empty or we need to perform |
2525 | * debugging duties. | |
2526 | * | |
894b8788 CL |
2527 | * Processing is still very fast if new objects have been freed to the |
2528 | * regular freelist. In that case we simply take over the regular freelist | |
2529 | * as the lockless freelist and zap the regular freelist. | |
81819f0f | 2530 | * |
894b8788 CL |
2531 | * If that is not working then we fall back to the partial lists. We take the |
2532 | * first element of the freelist as the object to allocate now and move the | |
2533 | * rest of the freelist to the lockless freelist. | |
81819f0f | 2534 | * |
894b8788 | 2535 | * And if we were unable to get a new slab from the partial slab lists then |
6446faa2 CL |
2536 | * we need to allocate a new slab. This is the slowest path since it involves |
2537 | * a call to the page allocator and the setup of a new slab. | |
a380a3c7 CL |
2538 | * |
2539 | * Version of __slab_alloc to use when we know that interrupts are | |
2540 | * already disabled (which is the case for bulk allocation). | |
81819f0f | 2541 | */ |
a380a3c7 | 2542 | static void *___slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node, |
ce71e27c | 2543 | unsigned long addr, struct kmem_cache_cpu *c) |
81819f0f | 2544 | { |
6faa6833 | 2545 | void *freelist; |
f6e7def7 | 2546 | struct page *page; |
81819f0f | 2547 | |
f6e7def7 CL |
2548 | page = c->page; |
2549 | if (!page) | |
81819f0f | 2550 | goto new_slab; |
49e22585 | 2551 | redo: |
6faa6833 | 2552 | |
57d437d2 | 2553 | if (unlikely(!node_match(page, node))) { |
a561ce00 JK |
2554 | int searchnode = node; |
2555 | ||
2556 | if (node != NUMA_NO_NODE && !node_present_pages(node)) | |
2557 | searchnode = node_to_mem_node(node); | |
2558 | ||
2559 | if (unlikely(!node_match(page, searchnode))) { | |
2560 | stat(s, ALLOC_NODE_MISMATCH); | |
d4ff6d35 | 2561 | deactivate_slab(s, page, c->freelist, c); |
a561ce00 JK |
2562 | goto new_slab; |
2563 | } | |
fc59c053 | 2564 | } |
6446faa2 | 2565 | |
072bb0aa MG |
2566 | /* |
2567 | * By rights, we should be searching for a slab page that was | |
2568 | * PFMEMALLOC but right now, we are losing the pfmemalloc | |
2569 | * information when the page leaves the per-cpu allocator | |
2570 | */ | |
2571 | if (unlikely(!pfmemalloc_match(page, gfpflags))) { | |
d4ff6d35 | 2572 | deactivate_slab(s, page, c->freelist, c); |
072bb0aa MG |
2573 | goto new_slab; |
2574 | } | |
2575 | ||
73736e03 | 2576 | /* must check again c->freelist in case of cpu migration or IRQ */ |
6faa6833 CL |
2577 | freelist = c->freelist; |
2578 | if (freelist) | |
73736e03 | 2579 | goto load_freelist; |
03e404af | 2580 | |
f6e7def7 | 2581 | freelist = get_freelist(s, page); |
6446faa2 | 2582 | |
6faa6833 | 2583 | if (!freelist) { |
03e404af CL |
2584 | c->page = NULL; |
2585 | stat(s, DEACTIVATE_BYPASS); | |
fc59c053 | 2586 | goto new_slab; |
03e404af | 2587 | } |
6446faa2 | 2588 | |
84e554e6 | 2589 | stat(s, ALLOC_REFILL); |
6446faa2 | 2590 | |
894b8788 | 2591 | load_freelist: |
507effea CL |
2592 | /* |
2593 | * freelist is pointing to the list of objects to be used. | |
2594 | * page is pointing to the page from which the objects are obtained. | |
2595 | * That page must be frozen for per cpu allocations to work. | |
2596 | */ | |
a0132ac0 | 2597 | VM_BUG_ON(!c->page->frozen); |
6faa6833 | 2598 | c->freelist = get_freepointer(s, freelist); |
8a5ec0ba | 2599 | c->tid = next_tid(c->tid); |
6faa6833 | 2600 | return freelist; |
81819f0f | 2601 | |
81819f0f | 2602 | new_slab: |
2cfb7455 | 2603 | |
a93cf07b WY |
2604 | if (slub_percpu_partial(c)) { |
2605 | page = c->page = slub_percpu_partial(c); | |
2606 | slub_set_percpu_partial(c, page); | |
49e22585 | 2607 | stat(s, CPU_PARTIAL_ALLOC); |
49e22585 | 2608 | goto redo; |
81819f0f CL |
2609 | } |
2610 | ||
188fd063 | 2611 | freelist = new_slab_objects(s, gfpflags, node, &c); |
01ad8a7b | 2612 | |
f4697436 | 2613 | if (unlikely(!freelist)) { |
9a02d699 | 2614 | slab_out_of_memory(s, gfpflags, node); |
f4697436 | 2615 | return NULL; |
81819f0f | 2616 | } |
2cfb7455 | 2617 | |
f6e7def7 | 2618 | page = c->page; |
5091b74a | 2619 | if (likely(!kmem_cache_debug(s) && pfmemalloc_match(page, gfpflags))) |
4b6f0750 | 2620 | goto load_freelist; |
2cfb7455 | 2621 | |
497b66f2 | 2622 | /* Only entered in the debug case */ |
d0e0ac97 CG |
2623 | if (kmem_cache_debug(s) && |
2624 | !alloc_debug_processing(s, page, freelist, addr)) | |
497b66f2 | 2625 | goto new_slab; /* Slab failed checks. Next slab needed */ |
894b8788 | 2626 | |
d4ff6d35 | 2627 | deactivate_slab(s, page, get_freepointer(s, freelist), c); |
6faa6833 | 2628 | return freelist; |
894b8788 CL |
2629 | } |
2630 | ||
a380a3c7 CL |
2631 | /* |
2632 | * Another one that disabled interrupt and compensates for possible | |
2633 | * cpu changes by refetching the per cpu area pointer. | |
2634 | */ | |
2635 | static void *__slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node, | |
2636 | unsigned long addr, struct kmem_cache_cpu *c) | |
2637 | { | |
2638 | void *p; | |
2639 | unsigned long flags; | |
2640 | ||
2641 | local_irq_save(flags); | |
2642 | #ifdef CONFIG_PREEMPT | |
2643 | /* | |
2644 | * We may have been preempted and rescheduled on a different | |
2645 | * cpu before disabling interrupts. Need to reload cpu area | |
2646 | * pointer. | |
2647 | */ | |
2648 | c = this_cpu_ptr(s->cpu_slab); | |
2649 | #endif | |
2650 | ||
2651 | p = ___slab_alloc(s, gfpflags, node, addr, c); | |
2652 | local_irq_restore(flags); | |
2653 | return p; | |
2654 | } | |
2655 | ||
894b8788 CL |
2656 | /* |
2657 | * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc) | |
2658 | * have the fastpath folded into their functions. So no function call | |
2659 | * overhead for requests that can be satisfied on the fastpath. | |
2660 | * | |
2661 | * The fastpath works by first checking if the lockless freelist can be used. | |
2662 | * If not then __slab_alloc is called for slow processing. | |
2663 | * | |
2664 | * Otherwise we can simply pick the next object from the lockless free list. | |
2665 | */ | |
2b847c3c | 2666 | static __always_inline void *slab_alloc_node(struct kmem_cache *s, |
ce71e27c | 2667 | gfp_t gfpflags, int node, unsigned long addr) |
894b8788 | 2668 | { |
03ec0ed5 | 2669 | void *object; |
dfb4f096 | 2670 | struct kmem_cache_cpu *c; |
57d437d2 | 2671 | struct page *page; |
8a5ec0ba | 2672 | unsigned long tid; |
1f84260c | 2673 | |
8135be5a VD |
2674 | s = slab_pre_alloc_hook(s, gfpflags); |
2675 | if (!s) | |
773ff60e | 2676 | return NULL; |
8a5ec0ba | 2677 | redo: |
8a5ec0ba CL |
2678 | /* |
2679 | * Must read kmem_cache cpu data via this cpu ptr. Preemption is | |
2680 | * enabled. We may switch back and forth between cpus while | |
2681 | * reading from one cpu area. That does not matter as long | |
2682 | * as we end up on the original cpu again when doing the cmpxchg. | |
7cccd80b | 2683 | * |
9aabf810 JK |
2684 | * We should guarantee that tid and kmem_cache are retrieved on |
2685 | * the same cpu. It could be different if CONFIG_PREEMPT so we need | |
2686 | * to check if it is matched or not. | |
8a5ec0ba | 2687 | */ |
9aabf810 JK |
2688 | do { |
2689 | tid = this_cpu_read(s->cpu_slab->tid); | |
2690 | c = raw_cpu_ptr(s->cpu_slab); | |
859b7a0e MR |
2691 | } while (IS_ENABLED(CONFIG_PREEMPT) && |
2692 | unlikely(tid != READ_ONCE(c->tid))); | |
9aabf810 JK |
2693 | |
2694 | /* | |
2695 | * Irqless object alloc/free algorithm used here depends on sequence | |
2696 | * of fetching cpu_slab's data. tid should be fetched before anything | |
2697 | * on c to guarantee that object and page associated with previous tid | |
2698 | * won't be used with current tid. If we fetch tid first, object and | |
2699 | * page could be one associated with next tid and our alloc/free | |
2700 | * request will be failed. In this case, we will retry. So, no problem. | |
2701 | */ | |
2702 | barrier(); | |
8a5ec0ba | 2703 | |
8a5ec0ba CL |
2704 | /* |
2705 | * The transaction ids are globally unique per cpu and per operation on | |
2706 | * a per cpu queue. Thus they can be guarantee that the cmpxchg_double | |
2707 | * occurs on the right processor and that there was no operation on the | |
2708 | * linked list in between. | |
2709 | */ | |
8a5ec0ba | 2710 | |
9dfc6e68 | 2711 | object = c->freelist; |
57d437d2 | 2712 | page = c->page; |
8eae1492 | 2713 | if (unlikely(!object || !node_match(page, node))) { |
dfb4f096 | 2714 | object = __slab_alloc(s, gfpflags, node, addr, c); |
8eae1492 DH |
2715 | stat(s, ALLOC_SLOWPATH); |
2716 | } else { | |
0ad9500e ED |
2717 | void *next_object = get_freepointer_safe(s, object); |
2718 | ||
8a5ec0ba | 2719 | /* |
25985edc | 2720 | * The cmpxchg will only match if there was no additional |
8a5ec0ba CL |
2721 | * operation and if we are on the right processor. |
2722 | * | |
d0e0ac97 CG |
2723 | * The cmpxchg does the following atomically (without lock |
2724 | * semantics!) | |
8a5ec0ba CL |
2725 | * 1. Relocate first pointer to the current per cpu area. |
2726 | * 2. Verify that tid and freelist have not been changed | |
2727 | * 3. If they were not changed replace tid and freelist | |
2728 | * | |
d0e0ac97 CG |
2729 | * Since this is without lock semantics the protection is only |
2730 | * against code executing on this cpu *not* from access by | |
2731 | * other cpus. | |
8a5ec0ba | 2732 | */ |
933393f5 | 2733 | if (unlikely(!this_cpu_cmpxchg_double( |
8a5ec0ba CL |
2734 | s->cpu_slab->freelist, s->cpu_slab->tid, |
2735 | object, tid, | |
0ad9500e | 2736 | next_object, next_tid(tid)))) { |
8a5ec0ba CL |
2737 | |
2738 | note_cmpxchg_failure("slab_alloc", s, tid); | |
2739 | goto redo; | |
2740 | } | |
0ad9500e | 2741 | prefetch_freepointer(s, next_object); |
84e554e6 | 2742 | stat(s, ALLOC_FASTPATH); |
894b8788 | 2743 | } |
8a5ec0ba | 2744 | |
74e2134f | 2745 | if (unlikely(gfpflags & __GFP_ZERO) && object) |
3b0efdfa | 2746 | memset(object, 0, s->object_size); |
d07dbea4 | 2747 | |
03ec0ed5 | 2748 | slab_post_alloc_hook(s, gfpflags, 1, &object); |
5a896d9e | 2749 | |
894b8788 | 2750 | return object; |
81819f0f CL |
2751 | } |
2752 | ||
2b847c3c EG |
2753 | static __always_inline void *slab_alloc(struct kmem_cache *s, |
2754 | gfp_t gfpflags, unsigned long addr) | |
2755 | { | |
2756 | return slab_alloc_node(s, gfpflags, NUMA_NO_NODE, addr); | |
2757 | } | |
2758 | ||
81819f0f CL |
2759 | void *kmem_cache_alloc(struct kmem_cache *s, gfp_t gfpflags) |
2760 | { | |
2b847c3c | 2761 | void *ret = slab_alloc(s, gfpflags, _RET_IP_); |
5b882be4 | 2762 | |
d0e0ac97 CG |
2763 | trace_kmem_cache_alloc(_RET_IP_, ret, s->object_size, |
2764 | s->size, gfpflags); | |
5b882be4 EGM |
2765 | |
2766 | return ret; | |
81819f0f CL |
2767 | } |
2768 | EXPORT_SYMBOL(kmem_cache_alloc); | |
2769 | ||
0f24f128 | 2770 | #ifdef CONFIG_TRACING |
4a92379b RK |
2771 | void *kmem_cache_alloc_trace(struct kmem_cache *s, gfp_t gfpflags, size_t size) |
2772 | { | |
2b847c3c | 2773 | void *ret = slab_alloc(s, gfpflags, _RET_IP_); |
4a92379b | 2774 | trace_kmalloc(_RET_IP_, ret, size, s->size, gfpflags); |
505f5dcb | 2775 | kasan_kmalloc(s, ret, size, gfpflags); |
4a92379b RK |
2776 | return ret; |
2777 | } | |
2778 | EXPORT_SYMBOL(kmem_cache_alloc_trace); | |
5b882be4 EGM |
2779 | #endif |
2780 | ||
81819f0f CL |
2781 | #ifdef CONFIG_NUMA |
2782 | void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t gfpflags, int node) | |
2783 | { | |
2b847c3c | 2784 | void *ret = slab_alloc_node(s, gfpflags, node, _RET_IP_); |
5b882be4 | 2785 | |
ca2b84cb | 2786 | trace_kmem_cache_alloc_node(_RET_IP_, ret, |
3b0efdfa | 2787 | s->object_size, s->size, gfpflags, node); |
5b882be4 EGM |
2788 | |
2789 | return ret; | |
81819f0f CL |
2790 | } |
2791 | EXPORT_SYMBOL(kmem_cache_alloc_node); | |
81819f0f | 2792 | |
0f24f128 | 2793 | #ifdef CONFIG_TRACING |
4a92379b | 2794 | void *kmem_cache_alloc_node_trace(struct kmem_cache *s, |
5b882be4 | 2795 | gfp_t gfpflags, |
4a92379b | 2796 | int node, size_t size) |
5b882be4 | 2797 | { |
2b847c3c | 2798 | void *ret = slab_alloc_node(s, gfpflags, node, _RET_IP_); |
4a92379b RK |
2799 | |
2800 | trace_kmalloc_node(_RET_IP_, ret, | |
2801 | size, s->size, gfpflags, node); | |
0316bec2 | 2802 | |
505f5dcb | 2803 | kasan_kmalloc(s, ret, size, gfpflags); |
4a92379b | 2804 | return ret; |
5b882be4 | 2805 | } |
4a92379b | 2806 | EXPORT_SYMBOL(kmem_cache_alloc_node_trace); |
5b882be4 | 2807 | #endif |
5d1f57e4 | 2808 | #endif |
5b882be4 | 2809 | |
81819f0f | 2810 | /* |
94e4d712 | 2811 | * Slow path handling. This may still be called frequently since objects |
894b8788 | 2812 | * have a longer lifetime than the cpu slabs in most processing loads. |
81819f0f | 2813 | * |
894b8788 CL |
2814 | * So we still attempt to reduce cache line usage. Just take the slab |
2815 | * lock and free the item. If there is no additional partial page | |
2816 | * handling required then we can return immediately. | |
81819f0f | 2817 | */ |
894b8788 | 2818 | static void __slab_free(struct kmem_cache *s, struct page *page, |
81084651 JDB |
2819 | void *head, void *tail, int cnt, |
2820 | unsigned long addr) | |
2821 | ||
81819f0f CL |
2822 | { |
2823 | void *prior; | |
2cfb7455 | 2824 | int was_frozen; |
2cfb7455 CL |
2825 | struct page new; |
2826 | unsigned long counters; | |
2827 | struct kmem_cache_node *n = NULL; | |
61728d1e | 2828 | unsigned long uninitialized_var(flags); |
81819f0f | 2829 | |
8a5ec0ba | 2830 | stat(s, FREE_SLOWPATH); |
81819f0f | 2831 | |
19c7ff9e | 2832 | if (kmem_cache_debug(s) && |
282acb43 | 2833 | !free_debug_processing(s, page, head, tail, cnt, addr)) |
80f08c19 | 2834 | return; |
6446faa2 | 2835 | |
2cfb7455 | 2836 | do { |
837d678d JK |
2837 | if (unlikely(n)) { |
2838 | spin_unlock_irqrestore(&n->list_lock, flags); | |
2839 | n = NULL; | |
2840 | } | |
2cfb7455 CL |
2841 | prior = page->freelist; |
2842 | counters = page->counters; | |
81084651 | 2843 | set_freepointer(s, tail, prior); |
2cfb7455 CL |
2844 | new.counters = counters; |
2845 | was_frozen = new.frozen; | |
81084651 | 2846 | new.inuse -= cnt; |
837d678d | 2847 | if ((!new.inuse || !prior) && !was_frozen) { |
49e22585 | 2848 | |
c65c1877 | 2849 | if (kmem_cache_has_cpu_partial(s) && !prior) { |
49e22585 CL |
2850 | |
2851 | /* | |
d0e0ac97 CG |
2852 | * Slab was on no list before and will be |
2853 | * partially empty | |
2854 | * We can defer the list move and instead | |
2855 | * freeze it. | |
49e22585 CL |
2856 | */ |
2857 | new.frozen = 1; | |
2858 | ||
c65c1877 | 2859 | } else { /* Needs to be taken off a list */ |
49e22585 | 2860 | |
b455def2 | 2861 | n = get_node(s, page_to_nid(page)); |
49e22585 CL |
2862 | /* |
2863 | * Speculatively acquire the list_lock. | |
2864 | * If the cmpxchg does not succeed then we may | |
2865 | * drop the list_lock without any processing. | |
2866 | * | |
2867 | * Otherwise the list_lock will synchronize with | |
2868 | * other processors updating the list of slabs. | |
2869 | */ | |
2870 | spin_lock_irqsave(&n->list_lock, flags); | |
2871 | ||
2872 | } | |
2cfb7455 | 2873 | } |
81819f0f | 2874 | |
2cfb7455 CL |
2875 | } while (!cmpxchg_double_slab(s, page, |
2876 | prior, counters, | |
81084651 | 2877 | head, new.counters, |
2cfb7455 | 2878 | "__slab_free")); |
81819f0f | 2879 | |
2cfb7455 | 2880 | if (likely(!n)) { |
49e22585 CL |
2881 | |
2882 | /* | |
2883 | * If we just froze the page then put it onto the | |
2884 | * per cpu partial list. | |
2885 | */ | |
8028dcea | 2886 | if (new.frozen && !was_frozen) { |
49e22585 | 2887 | put_cpu_partial(s, page, 1); |
8028dcea AS |
2888 | stat(s, CPU_PARTIAL_FREE); |
2889 | } | |
49e22585 | 2890 | /* |
2cfb7455 CL |
2891 | * The list lock was not taken therefore no list |
2892 | * activity can be necessary. | |
2893 | */ | |
b455def2 L |
2894 | if (was_frozen) |
2895 | stat(s, FREE_FROZEN); | |
2896 | return; | |
2897 | } | |
81819f0f | 2898 | |
8a5b20ae | 2899 | if (unlikely(!new.inuse && n->nr_partial >= s->min_partial)) |
837d678d JK |
2900 | goto slab_empty; |
2901 | ||
81819f0f | 2902 | /* |
837d678d JK |
2903 | * Objects left in the slab. If it was not on the partial list before |
2904 | * then add it. | |
81819f0f | 2905 | */ |
345c905d JK |
2906 | if (!kmem_cache_has_cpu_partial(s) && unlikely(!prior)) { |
2907 | if (kmem_cache_debug(s)) | |
c65c1877 | 2908 | remove_full(s, n, page); |
837d678d JK |
2909 | add_partial(n, page, DEACTIVATE_TO_TAIL); |
2910 | stat(s, FREE_ADD_PARTIAL); | |
8ff12cfc | 2911 | } |
80f08c19 | 2912 | spin_unlock_irqrestore(&n->list_lock, flags); |
81819f0f CL |
2913 | return; |
2914 | ||
2915 | slab_empty: | |
a973e9dd | 2916 | if (prior) { |
81819f0f | 2917 | /* |
6fbabb20 | 2918 | * Slab on the partial list. |
81819f0f | 2919 | */ |
5cc6eee8 | 2920 | remove_partial(n, page); |
84e554e6 | 2921 | stat(s, FREE_REMOVE_PARTIAL); |
c65c1877 | 2922 | } else { |
6fbabb20 | 2923 | /* Slab must be on the full list */ |
c65c1877 PZ |
2924 | remove_full(s, n, page); |
2925 | } | |
2cfb7455 | 2926 | |
80f08c19 | 2927 | spin_unlock_irqrestore(&n->list_lock, flags); |
84e554e6 | 2928 | stat(s, FREE_SLAB); |
81819f0f | 2929 | discard_slab(s, page); |
81819f0f CL |
2930 | } |
2931 | ||
894b8788 CL |
2932 | /* |
2933 | * Fastpath with forced inlining to produce a kfree and kmem_cache_free that | |
2934 | * can perform fastpath freeing without additional function calls. | |
2935 | * | |
2936 | * The fastpath is only possible if we are freeing to the current cpu slab | |
2937 | * of this processor. This typically the case if we have just allocated | |
2938 | * the item before. | |
2939 | * | |
2940 | * If fastpath is not possible then fall back to __slab_free where we deal | |
2941 | * with all sorts of special processing. | |
81084651 JDB |
2942 | * |
2943 | * Bulk free of a freelist with several objects (all pointing to the | |
2944 | * same page) possible by specifying head and tail ptr, plus objects | |
2945 | * count (cnt). Bulk free indicated by tail pointer being set. | |
894b8788 | 2946 | */ |
80a9201a AP |
2947 | static __always_inline void do_slab_free(struct kmem_cache *s, |
2948 | struct page *page, void *head, void *tail, | |
2949 | int cnt, unsigned long addr) | |
894b8788 | 2950 | { |
81084651 | 2951 | void *tail_obj = tail ? : head; |
dfb4f096 | 2952 | struct kmem_cache_cpu *c; |
8a5ec0ba | 2953 | unsigned long tid; |
8a5ec0ba CL |
2954 | redo: |
2955 | /* | |
2956 | * Determine the currently cpus per cpu slab. | |
2957 | * The cpu may change afterward. However that does not matter since | |
2958 | * data is retrieved via this pointer. If we are on the same cpu | |
2ae44005 | 2959 | * during the cmpxchg then the free will succeed. |
8a5ec0ba | 2960 | */ |
9aabf810 JK |
2961 | do { |
2962 | tid = this_cpu_read(s->cpu_slab->tid); | |
2963 | c = raw_cpu_ptr(s->cpu_slab); | |
859b7a0e MR |
2964 | } while (IS_ENABLED(CONFIG_PREEMPT) && |
2965 | unlikely(tid != READ_ONCE(c->tid))); | |
c016b0bd | 2966 | |
9aabf810 JK |
2967 | /* Same with comment on barrier() in slab_alloc_node() */ |
2968 | barrier(); | |
c016b0bd | 2969 | |
442b06bc | 2970 | if (likely(page == c->page)) { |
81084651 | 2971 | set_freepointer(s, tail_obj, c->freelist); |
8a5ec0ba | 2972 | |
933393f5 | 2973 | if (unlikely(!this_cpu_cmpxchg_double( |
8a5ec0ba CL |
2974 | s->cpu_slab->freelist, s->cpu_slab->tid, |
2975 | c->freelist, tid, | |
81084651 | 2976 | head, next_tid(tid)))) { |
8a5ec0ba CL |
2977 | |
2978 | note_cmpxchg_failure("slab_free", s, tid); | |
2979 | goto redo; | |
2980 | } | |
84e554e6 | 2981 | stat(s, FREE_FASTPATH); |
894b8788 | 2982 | } else |
81084651 | 2983 | __slab_free(s, page, head, tail_obj, cnt, addr); |
894b8788 | 2984 | |
894b8788 CL |
2985 | } |
2986 | ||
80a9201a AP |
2987 | static __always_inline void slab_free(struct kmem_cache *s, struct page *page, |
2988 | void *head, void *tail, int cnt, | |
2989 | unsigned long addr) | |
2990 | { | |
2991 | slab_free_freelist_hook(s, head, tail); | |
2992 | /* | |
2993 | * slab_free_freelist_hook() could have put the items into quarantine. | |
2994 | * If so, no need to free them. | |
2995 | */ | |
5f0d5a3a | 2996 | if (s->flags & SLAB_KASAN && !(s->flags & SLAB_TYPESAFE_BY_RCU)) |
80a9201a AP |
2997 | return; |
2998 | do_slab_free(s, page, head, tail, cnt, addr); | |
2999 | } | |
3000 | ||
3001 | #ifdef CONFIG_KASAN | |
3002 | void ___cache_free(struct kmem_cache *cache, void *x, unsigned long addr) | |
3003 | { | |
3004 | do_slab_free(cache, virt_to_head_page(x), x, NULL, 1, addr); | |
3005 | } | |
3006 | #endif | |
3007 | ||
81819f0f CL |
3008 | void kmem_cache_free(struct kmem_cache *s, void *x) |
3009 | { | |
b9ce5ef4 GC |
3010 | s = cache_from_obj(s, x); |
3011 | if (!s) | |
79576102 | 3012 | return; |
81084651 | 3013 | slab_free(s, virt_to_head_page(x), x, NULL, 1, _RET_IP_); |
ca2b84cb | 3014 | trace_kmem_cache_free(_RET_IP_, x); |
81819f0f CL |
3015 | } |
3016 | EXPORT_SYMBOL(kmem_cache_free); | |
3017 | ||
d0ecd894 | 3018 | struct detached_freelist { |
fbd02630 | 3019 | struct page *page; |
d0ecd894 JDB |
3020 | void *tail; |
3021 | void *freelist; | |
3022 | int cnt; | |
376bf125 | 3023 | struct kmem_cache *s; |
d0ecd894 | 3024 | }; |
fbd02630 | 3025 | |
d0ecd894 JDB |
3026 | /* |
3027 | * This function progressively scans the array with free objects (with | |
3028 | * a limited look ahead) and extract objects belonging to the same | |
3029 | * page. It builds a detached freelist directly within the given | |
3030 | * page/objects. This can happen without any need for | |
3031 | * synchronization, because the objects are owned by running process. | |
3032 | * The freelist is build up as a single linked list in the objects. | |
3033 | * The idea is, that this detached freelist can then be bulk | |
3034 | * transferred to the real freelist(s), but only requiring a single | |
3035 | * synchronization primitive. Look ahead in the array is limited due | |
3036 | * to performance reasons. | |
3037 | */ | |
376bf125 JDB |
3038 | static inline |
3039 | int build_detached_freelist(struct kmem_cache *s, size_t size, | |
3040 | void **p, struct detached_freelist *df) | |
d0ecd894 JDB |
3041 | { |
3042 | size_t first_skipped_index = 0; | |
3043 | int lookahead = 3; | |
3044 | void *object; | |
ca257195 | 3045 | struct page *page; |
fbd02630 | 3046 | |
d0ecd894 JDB |
3047 | /* Always re-init detached_freelist */ |
3048 | df->page = NULL; | |
fbd02630 | 3049 | |
d0ecd894 JDB |
3050 | do { |
3051 | object = p[--size]; | |
ca257195 | 3052 | /* Do we need !ZERO_OR_NULL_PTR(object) here? (for kfree) */ |
d0ecd894 | 3053 | } while (!object && size); |
3eed034d | 3054 | |
d0ecd894 JDB |
3055 | if (!object) |
3056 | return 0; | |
fbd02630 | 3057 | |
ca257195 JDB |
3058 | page = virt_to_head_page(object); |
3059 | if (!s) { | |
3060 | /* Handle kalloc'ed objects */ | |
3061 | if (unlikely(!PageSlab(page))) { | |
3062 | BUG_ON(!PageCompound(page)); | |
3063 | kfree_hook(object); | |
4949148a | 3064 | __free_pages(page, compound_order(page)); |
ca257195 JDB |
3065 | p[size] = NULL; /* mark object processed */ |
3066 | return size; | |
3067 | } | |
3068 | /* Derive kmem_cache from object */ | |
3069 | df->s = page->slab_cache; | |
3070 | } else { | |
3071 | df->s = cache_from_obj(s, object); /* Support for memcg */ | |
3072 | } | |
376bf125 | 3073 | |
d0ecd894 | 3074 | /* Start new detached freelist */ |
ca257195 | 3075 | df->page = page; |
376bf125 | 3076 | set_freepointer(df->s, object, NULL); |
d0ecd894 JDB |
3077 | df->tail = object; |
3078 | df->freelist = object; | |
3079 | p[size] = NULL; /* mark object processed */ | |
3080 | df->cnt = 1; | |
3081 | ||
3082 | while (size) { | |
3083 | object = p[--size]; | |
3084 | if (!object) | |
3085 | continue; /* Skip processed objects */ | |
3086 | ||
3087 | /* df->page is always set at this point */ | |
3088 | if (df->page == virt_to_head_page(object)) { | |
3089 | /* Opportunity build freelist */ | |
376bf125 | 3090 | set_freepointer(df->s, object, df->freelist); |
d0ecd894 JDB |
3091 | df->freelist = object; |
3092 | df->cnt++; | |
3093 | p[size] = NULL; /* mark object processed */ | |
3094 | ||
3095 | continue; | |
fbd02630 | 3096 | } |
d0ecd894 JDB |
3097 | |
3098 | /* Limit look ahead search */ | |
3099 | if (!--lookahead) | |
3100 | break; | |
3101 | ||
3102 | if (!first_skipped_index) | |
3103 | first_skipped_index = size + 1; | |
fbd02630 | 3104 | } |
d0ecd894 JDB |
3105 | |
3106 | return first_skipped_index; | |
3107 | } | |
3108 | ||
d0ecd894 | 3109 | /* Note that interrupts must be enabled when calling this function. */ |
376bf125 | 3110 | void kmem_cache_free_bulk(struct kmem_cache *s, size_t size, void **p) |
d0ecd894 JDB |
3111 | { |
3112 | if (WARN_ON(!size)) | |
3113 | return; | |
3114 | ||
3115 | do { | |
3116 | struct detached_freelist df; | |
3117 | ||
3118 | size = build_detached_freelist(s, size, p, &df); | |
84582c8a | 3119 | if (!df.page) |
d0ecd894 JDB |
3120 | continue; |
3121 | ||
376bf125 | 3122 | slab_free(df.s, df.page, df.freelist, df.tail, df.cnt,_RET_IP_); |
d0ecd894 | 3123 | } while (likely(size)); |
484748f0 CL |
3124 | } |
3125 | EXPORT_SYMBOL(kmem_cache_free_bulk); | |
3126 | ||
994eb764 | 3127 | /* Note that interrupts must be enabled when calling this function. */ |
865762a8 JDB |
3128 | int kmem_cache_alloc_bulk(struct kmem_cache *s, gfp_t flags, size_t size, |
3129 | void **p) | |
484748f0 | 3130 | { |
994eb764 JDB |
3131 | struct kmem_cache_cpu *c; |
3132 | int i; | |
3133 | ||
03ec0ed5 JDB |
3134 | /* memcg and kmem_cache debug support */ |
3135 | s = slab_pre_alloc_hook(s, flags); | |
3136 | if (unlikely(!s)) | |
3137 | return false; | |
994eb764 JDB |
3138 | /* |
3139 | * Drain objects in the per cpu slab, while disabling local | |
3140 | * IRQs, which protects against PREEMPT and interrupts | |
3141 | * handlers invoking normal fastpath. | |
3142 | */ | |
3143 | local_irq_disable(); | |
3144 | c = this_cpu_ptr(s->cpu_slab); | |
3145 | ||
3146 | for (i = 0; i < size; i++) { | |
3147 | void *object = c->freelist; | |
3148 | ||
ebe909e0 | 3149 | if (unlikely(!object)) { |
ebe909e0 JDB |
3150 | /* |
3151 | * Invoking slow path likely have side-effect | |
3152 | * of re-populating per CPU c->freelist | |
3153 | */ | |
87098373 | 3154 | p[i] = ___slab_alloc(s, flags, NUMA_NO_NODE, |
ebe909e0 | 3155 | _RET_IP_, c); |
87098373 CL |
3156 | if (unlikely(!p[i])) |
3157 | goto error; | |
3158 | ||
ebe909e0 JDB |
3159 | c = this_cpu_ptr(s->cpu_slab); |
3160 | continue; /* goto for-loop */ | |
3161 | } | |
994eb764 JDB |
3162 | c->freelist = get_freepointer(s, object); |
3163 | p[i] = object; | |
3164 | } | |
3165 | c->tid = next_tid(c->tid); | |
3166 | local_irq_enable(); | |
3167 | ||
3168 | /* Clear memory outside IRQ disabled fastpath loop */ | |
3169 | if (unlikely(flags & __GFP_ZERO)) { | |
3170 | int j; | |
3171 | ||
3172 | for (j = 0; j < i; j++) | |
3173 | memset(p[j], 0, s->object_size); | |
3174 | } | |
3175 | ||
03ec0ed5 JDB |
3176 | /* memcg and kmem_cache debug support */ |
3177 | slab_post_alloc_hook(s, flags, size, p); | |
865762a8 | 3178 | return i; |
87098373 | 3179 | error: |
87098373 | 3180 | local_irq_enable(); |
03ec0ed5 JDB |
3181 | slab_post_alloc_hook(s, flags, i, p); |
3182 | __kmem_cache_free_bulk(s, i, p); | |
865762a8 | 3183 | return 0; |
484748f0 CL |
3184 | } |
3185 | EXPORT_SYMBOL(kmem_cache_alloc_bulk); | |
3186 | ||
3187 | ||
81819f0f | 3188 | /* |
672bba3a CL |
3189 | * Object placement in a slab is made very easy because we always start at |
3190 | * offset 0. If we tune the size of the object to the alignment then we can | |
3191 | * get the required alignment by putting one properly sized object after | |
3192 | * another. | |
81819f0f CL |
3193 | * |
3194 | * Notice that the allocation order determines the sizes of the per cpu | |
3195 | * caches. Each processor has always one slab available for allocations. | |
3196 | * Increasing the allocation order reduces the number of times that slabs | |
672bba3a | 3197 | * must be moved on and off the partial lists and is therefore a factor in |
81819f0f | 3198 | * locking overhead. |
81819f0f CL |
3199 | */ |
3200 | ||
3201 | /* | |
3202 | * Mininum / Maximum order of slab pages. This influences locking overhead | |
3203 | * and slab fragmentation. A higher order reduces the number of partial slabs | |
3204 | * and increases the number of allocations possible without having to | |
3205 | * take the list_lock. | |
3206 | */ | |
3207 | static int slub_min_order; | |
114e9e89 | 3208 | static int slub_max_order = PAGE_ALLOC_COSTLY_ORDER; |
9b2cd506 | 3209 | static int slub_min_objects; |
81819f0f | 3210 | |
81819f0f CL |
3211 | /* |
3212 | * Calculate the order of allocation given an slab object size. | |
3213 | * | |
672bba3a CL |
3214 | * The order of allocation has significant impact on performance and other |
3215 | * system components. Generally order 0 allocations should be preferred since | |
3216 | * order 0 does not cause fragmentation in the page allocator. Larger objects | |
3217 | * be problematic to put into order 0 slabs because there may be too much | |
c124f5b5 | 3218 | * unused space left. We go to a higher order if more than 1/16th of the slab |
672bba3a CL |
3219 | * would be wasted. |
3220 | * | |
3221 | * In order to reach satisfactory performance we must ensure that a minimum | |
3222 | * number of objects is in one slab. Otherwise we may generate too much | |
3223 | * activity on the partial lists which requires taking the list_lock. This is | |
3224 | * less a concern for large slabs though which are rarely used. | |
81819f0f | 3225 | * |
672bba3a CL |
3226 | * slub_max_order specifies the order where we begin to stop considering the |
3227 | * number of objects in a slab as critical. If we reach slub_max_order then | |
3228 | * we try to keep the page order as low as possible. So we accept more waste | |
3229 | * of space in favor of a small page order. | |
81819f0f | 3230 | * |
672bba3a CL |
3231 | * Higher order allocations also allow the placement of more objects in a |
3232 | * slab and thereby reduce object handling overhead. If the user has | |
3233 | * requested a higher mininum order then we start with that one instead of | |
3234 | * the smallest order which will fit the object. | |
81819f0f | 3235 | */ |
5e6d444e | 3236 | static inline int slab_order(int size, int min_objects, |
ab9a0f19 | 3237 | int max_order, int fract_leftover, int reserved) |
81819f0f CL |
3238 | { |
3239 | int order; | |
3240 | int rem; | |
6300ea75 | 3241 | int min_order = slub_min_order; |
81819f0f | 3242 | |
ab9a0f19 | 3243 | if (order_objects(min_order, size, reserved) > MAX_OBJS_PER_PAGE) |
210b5c06 | 3244 | return get_order(size * MAX_OBJS_PER_PAGE) - 1; |
39b26464 | 3245 | |
9f835703 | 3246 | for (order = max(min_order, get_order(min_objects * size + reserved)); |
5e6d444e | 3247 | order <= max_order; order++) { |
81819f0f | 3248 | |
5e6d444e | 3249 | unsigned long slab_size = PAGE_SIZE << order; |
81819f0f | 3250 | |
ab9a0f19 | 3251 | rem = (slab_size - reserved) % size; |
81819f0f | 3252 | |
5e6d444e | 3253 | if (rem <= slab_size / fract_leftover) |
81819f0f | 3254 | break; |
81819f0f | 3255 | } |
672bba3a | 3256 | |
81819f0f CL |
3257 | return order; |
3258 | } | |
3259 | ||
ab9a0f19 | 3260 | static inline int calculate_order(int size, int reserved) |
5e6d444e CL |
3261 | { |
3262 | int order; | |
3263 | int min_objects; | |
3264 | int fraction; | |
e8120ff1 | 3265 | int max_objects; |
5e6d444e CL |
3266 | |
3267 | /* | |
3268 | * Attempt to find best configuration for a slab. This | |
3269 | * works by first attempting to generate a layout with | |
3270 | * the best configuration and backing off gradually. | |
3271 | * | |
422ff4d7 | 3272 | * First we increase the acceptable waste in a slab. Then |
5e6d444e CL |
3273 | * we reduce the minimum objects required in a slab. |
3274 | */ | |
3275 | min_objects = slub_min_objects; | |
9b2cd506 CL |
3276 | if (!min_objects) |
3277 | min_objects = 4 * (fls(nr_cpu_ids) + 1); | |
ab9a0f19 | 3278 | max_objects = order_objects(slub_max_order, size, reserved); |
e8120ff1 ZY |
3279 | min_objects = min(min_objects, max_objects); |
3280 | ||
5e6d444e | 3281 | while (min_objects > 1) { |
c124f5b5 | 3282 | fraction = 16; |
5e6d444e CL |
3283 | while (fraction >= 4) { |
3284 | order = slab_order(size, min_objects, | |
ab9a0f19 | 3285 | slub_max_order, fraction, reserved); |
5e6d444e CL |
3286 | if (order <= slub_max_order) |
3287 | return order; | |
3288 | fraction /= 2; | |
3289 | } | |
5086c389 | 3290 | min_objects--; |
5e6d444e CL |
3291 | } |
3292 | ||
3293 | /* | |
3294 | * We were unable to place multiple objects in a slab. Now | |
3295 | * lets see if we can place a single object there. | |
3296 | */ | |
ab9a0f19 | 3297 | order = slab_order(size, 1, slub_max_order, 1, reserved); |
5e6d444e CL |
3298 | if (order <= slub_max_order) |
3299 | return order; | |
3300 | ||
3301 | /* | |
3302 | * Doh this slab cannot be placed using slub_max_order. | |
3303 | */ | |
ab9a0f19 | 3304 | order = slab_order(size, 1, MAX_ORDER, 1, reserved); |
818cf590 | 3305 | if (order < MAX_ORDER) |
5e6d444e CL |
3306 | return order; |
3307 | return -ENOSYS; | |
3308 | } | |
3309 | ||
5595cffc | 3310 | static void |
4053497d | 3311 | init_kmem_cache_node(struct kmem_cache_node *n) |
81819f0f CL |
3312 | { |
3313 | n->nr_partial = 0; | |
81819f0f CL |
3314 | spin_lock_init(&n->list_lock); |
3315 | INIT_LIST_HEAD(&n->partial); | |
8ab1372f | 3316 | #ifdef CONFIG_SLUB_DEBUG |
0f389ec6 | 3317 | atomic_long_set(&n->nr_slabs, 0); |
02b71b70 | 3318 | atomic_long_set(&n->total_objects, 0); |
643b1138 | 3319 | INIT_LIST_HEAD(&n->full); |
8ab1372f | 3320 | #endif |
81819f0f CL |
3321 | } |
3322 | ||
55136592 | 3323 | static inline int alloc_kmem_cache_cpus(struct kmem_cache *s) |
4c93c355 | 3324 | { |
6c182dc0 | 3325 | BUILD_BUG_ON(PERCPU_DYNAMIC_EARLY_SIZE < |
95a05b42 | 3326 | KMALLOC_SHIFT_HIGH * sizeof(struct kmem_cache_cpu)); |
4c93c355 | 3327 | |
8a5ec0ba | 3328 | /* |
d4d84fef CM |
3329 | * Must align to double word boundary for the double cmpxchg |
3330 | * instructions to work; see __pcpu_double_call_return_bool(). | |
8a5ec0ba | 3331 | */ |
d4d84fef CM |
3332 | s->cpu_slab = __alloc_percpu(sizeof(struct kmem_cache_cpu), |
3333 | 2 * sizeof(void *)); | |
8a5ec0ba CL |
3334 | |
3335 | if (!s->cpu_slab) | |
3336 | return 0; | |
3337 | ||
3338 | init_kmem_cache_cpus(s); | |
4c93c355 | 3339 | |
8a5ec0ba | 3340 | return 1; |
4c93c355 | 3341 | } |
4c93c355 | 3342 | |
51df1142 CL |
3343 | static struct kmem_cache *kmem_cache_node; |
3344 | ||
81819f0f CL |
3345 | /* |
3346 | * No kmalloc_node yet so do it by hand. We know that this is the first | |
3347 | * slab on the node for this slabcache. There are no concurrent accesses | |
3348 | * possible. | |
3349 | * | |
721ae22a ZYW |
3350 | * Note that this function only works on the kmem_cache_node |
3351 | * when allocating for the kmem_cache_node. This is used for bootstrapping | |
4c93c355 | 3352 | * memory on a fresh node that has no slab structures yet. |
81819f0f | 3353 | */ |
55136592 | 3354 | static void early_kmem_cache_node_alloc(int node) |
81819f0f CL |
3355 | { |
3356 | struct page *page; | |
3357 | struct kmem_cache_node *n; | |
3358 | ||
51df1142 | 3359 | BUG_ON(kmem_cache_node->size < sizeof(struct kmem_cache_node)); |
81819f0f | 3360 | |
51df1142 | 3361 | page = new_slab(kmem_cache_node, GFP_NOWAIT, node); |
81819f0f CL |
3362 | |
3363 | BUG_ON(!page); | |
a2f92ee7 | 3364 | if (page_to_nid(page) != node) { |
f9f58285 FF |
3365 | pr_err("SLUB: Unable to allocate memory from node %d\n", node); |
3366 | pr_err("SLUB: Allocating a useless per node structure in order to be able to continue\n"); | |
a2f92ee7 CL |
3367 | } |
3368 | ||
81819f0f CL |
3369 | n = page->freelist; |
3370 | BUG_ON(!n); | |
51df1142 | 3371 | page->freelist = get_freepointer(kmem_cache_node, n); |
e6e82ea1 | 3372 | page->inuse = 1; |
8cb0a506 | 3373 | page->frozen = 0; |
51df1142 | 3374 | kmem_cache_node->node[node] = n; |
8ab1372f | 3375 | #ifdef CONFIG_SLUB_DEBUG |
f7cb1933 | 3376 | init_object(kmem_cache_node, n, SLUB_RED_ACTIVE); |
51df1142 | 3377 | init_tracking(kmem_cache_node, n); |
8ab1372f | 3378 | #endif |
505f5dcb AP |
3379 | kasan_kmalloc(kmem_cache_node, n, sizeof(struct kmem_cache_node), |
3380 | GFP_KERNEL); | |
4053497d | 3381 | init_kmem_cache_node(n); |
51df1142 | 3382 | inc_slabs_node(kmem_cache_node, node, page->objects); |
6446faa2 | 3383 | |
67b6c900 | 3384 | /* |
1e4dd946 SR |
3385 | * No locks need to be taken here as it has just been |
3386 | * initialized and there is no concurrent access. | |
67b6c900 | 3387 | */ |
1e4dd946 | 3388 | __add_partial(n, page, DEACTIVATE_TO_HEAD); |
81819f0f CL |
3389 | } |
3390 | ||
3391 | static void free_kmem_cache_nodes(struct kmem_cache *s) | |
3392 | { | |
3393 | int node; | |
fa45dc25 | 3394 | struct kmem_cache_node *n; |
81819f0f | 3395 | |
fa45dc25 | 3396 | for_each_kmem_cache_node(s, node, n) { |
81819f0f | 3397 | s->node[node] = NULL; |
ea37df54 | 3398 | kmem_cache_free(kmem_cache_node, n); |
81819f0f CL |
3399 | } |
3400 | } | |
3401 | ||
52b4b950 DS |
3402 | void __kmem_cache_release(struct kmem_cache *s) |
3403 | { | |
210e7a43 | 3404 | cache_random_seq_destroy(s); |
52b4b950 DS |
3405 | free_percpu(s->cpu_slab); |
3406 | free_kmem_cache_nodes(s); | |
3407 | } | |
3408 | ||
55136592 | 3409 | static int init_kmem_cache_nodes(struct kmem_cache *s) |
81819f0f CL |
3410 | { |
3411 | int node; | |
81819f0f | 3412 | |
f64dc58c | 3413 | for_each_node_state(node, N_NORMAL_MEMORY) { |
81819f0f CL |
3414 | struct kmem_cache_node *n; |
3415 | ||
73367bd8 | 3416 | if (slab_state == DOWN) { |
55136592 | 3417 | early_kmem_cache_node_alloc(node); |
73367bd8 AD |
3418 | continue; |
3419 | } | |
51df1142 | 3420 | n = kmem_cache_alloc_node(kmem_cache_node, |
55136592 | 3421 | GFP_KERNEL, node); |
81819f0f | 3422 | |
73367bd8 AD |
3423 | if (!n) { |
3424 | free_kmem_cache_nodes(s); | |
3425 | return 0; | |
81819f0f | 3426 | } |
73367bd8 | 3427 | |
4053497d | 3428 | init_kmem_cache_node(n); |
ea37df54 | 3429 | s->node[node] = n; |
81819f0f CL |
3430 | } |
3431 | return 1; | |
3432 | } | |
81819f0f | 3433 | |
c0bdb232 | 3434 | static void set_min_partial(struct kmem_cache *s, unsigned long min) |
3b89d7d8 DR |
3435 | { |
3436 | if (min < MIN_PARTIAL) | |
3437 | min = MIN_PARTIAL; | |
3438 | else if (min > MAX_PARTIAL) | |
3439 | min = MAX_PARTIAL; | |
3440 | s->min_partial = min; | |
3441 | } | |
3442 | ||
e6d0e1dc WY |
3443 | static void set_cpu_partial(struct kmem_cache *s) |
3444 | { | |
3445 | #ifdef CONFIG_SLUB_CPU_PARTIAL | |
3446 | /* | |
3447 | * cpu_partial determined the maximum number of objects kept in the | |
3448 | * per cpu partial lists of a processor. | |
3449 | * | |
3450 | * Per cpu partial lists mainly contain slabs that just have one | |
3451 | * object freed. If they are used for allocation then they can be | |
3452 | * filled up again with minimal effort. The slab will never hit the | |
3453 | * per node partial lists and therefore no locking will be required. | |
3454 | * | |
3455 | * This setting also determines | |
3456 | * | |
3457 | * A) The number of objects from per cpu partial slabs dumped to the | |
3458 | * per node list when we reach the limit. | |
3459 | * B) The number of objects in cpu partial slabs to extract from the | |
3460 | * per node list when we run out of per cpu objects. We only fetch | |
3461 | * 50% to keep some capacity around for frees. | |
3462 | */ | |
3463 | if (!kmem_cache_has_cpu_partial(s)) | |
3464 | s->cpu_partial = 0; | |
3465 | else if (s->size >= PAGE_SIZE) | |
3466 | s->cpu_partial = 2; | |
3467 | else if (s->size >= 1024) | |
3468 | s->cpu_partial = 6; | |
3469 | else if (s->size >= 256) | |
3470 | s->cpu_partial = 13; | |
3471 | else | |
3472 | s->cpu_partial = 30; | |
3473 | #endif | |
3474 | } | |
3475 | ||
81819f0f CL |
3476 | /* |
3477 | * calculate_sizes() determines the order and the distribution of data within | |
3478 | * a slab object. | |
3479 | */ | |
06b285dc | 3480 | static int calculate_sizes(struct kmem_cache *s, int forced_order) |
81819f0f | 3481 | { |
d50112ed | 3482 | slab_flags_t flags = s->flags; |
80a9201a | 3483 | size_t size = s->object_size; |
834f3d11 | 3484 | int order; |
81819f0f | 3485 | |
d8b42bf5 CL |
3486 | /* |
3487 | * Round up object size to the next word boundary. We can only | |
3488 | * place the free pointer at word boundaries and this determines | |
3489 | * the possible location of the free pointer. | |
3490 | */ | |
3491 | size = ALIGN(size, sizeof(void *)); | |
3492 | ||
3493 | #ifdef CONFIG_SLUB_DEBUG | |
81819f0f CL |
3494 | /* |
3495 | * Determine if we can poison the object itself. If the user of | |
3496 | * the slab may touch the object after free or before allocation | |
3497 | * then we should never poison the object itself. | |
3498 | */ | |
5f0d5a3a | 3499 | if ((flags & SLAB_POISON) && !(flags & SLAB_TYPESAFE_BY_RCU) && |
c59def9f | 3500 | !s->ctor) |
81819f0f CL |
3501 | s->flags |= __OBJECT_POISON; |
3502 | else | |
3503 | s->flags &= ~__OBJECT_POISON; | |
3504 | ||
81819f0f CL |
3505 | |
3506 | /* | |
672bba3a | 3507 | * If we are Redzoning then check if there is some space between the |
81819f0f | 3508 | * end of the object and the free pointer. If not then add an |
672bba3a | 3509 | * additional word to have some bytes to store Redzone information. |
81819f0f | 3510 | */ |
3b0efdfa | 3511 | if ((flags & SLAB_RED_ZONE) && size == s->object_size) |
81819f0f | 3512 | size += sizeof(void *); |
41ecc55b | 3513 | #endif |
81819f0f CL |
3514 | |
3515 | /* | |
672bba3a CL |
3516 | * With that we have determined the number of bytes in actual use |
3517 | * by the object. This is the potential offset to the free pointer. | |
81819f0f CL |
3518 | */ |
3519 | s->inuse = size; | |
3520 | ||
5f0d5a3a | 3521 | if (((flags & (SLAB_TYPESAFE_BY_RCU | SLAB_POISON)) || |
c59def9f | 3522 | s->ctor)) { |
81819f0f CL |
3523 | /* |
3524 | * Relocate free pointer after the object if it is not | |
3525 | * permitted to overwrite the first word of the object on | |
3526 | * kmem_cache_free. | |
3527 | * | |
3528 | * This is the case if we do RCU, have a constructor or | |
3529 | * destructor or are poisoning the objects. | |
3530 | */ | |
3531 | s->offset = size; | |
3532 | size += sizeof(void *); | |
3533 | } | |
3534 | ||
c12b3c62 | 3535 | #ifdef CONFIG_SLUB_DEBUG |
81819f0f CL |
3536 | if (flags & SLAB_STORE_USER) |
3537 | /* | |
3538 | * Need to store information about allocs and frees after | |
3539 | * the object. | |
3540 | */ | |
3541 | size += 2 * sizeof(struct track); | |
80a9201a | 3542 | #endif |
81819f0f | 3543 | |
80a9201a AP |
3544 | kasan_cache_create(s, &size, &s->flags); |
3545 | #ifdef CONFIG_SLUB_DEBUG | |
d86bd1be | 3546 | if (flags & SLAB_RED_ZONE) { |
81819f0f CL |
3547 | /* |
3548 | * Add some empty padding so that we can catch | |
3549 | * overwrites from earlier objects rather than let | |
3550 | * tracking information or the free pointer be | |
0211a9c8 | 3551 | * corrupted if a user writes before the start |
81819f0f CL |
3552 | * of the object. |
3553 | */ | |
3554 | size += sizeof(void *); | |
d86bd1be JK |
3555 | |
3556 | s->red_left_pad = sizeof(void *); | |
3557 | s->red_left_pad = ALIGN(s->red_left_pad, s->align); | |
3558 | size += s->red_left_pad; | |
3559 | } | |
41ecc55b | 3560 | #endif |
672bba3a | 3561 | |
81819f0f CL |
3562 | /* |
3563 | * SLUB stores one object immediately after another beginning from | |
3564 | * offset 0. In order to align the objects we have to simply size | |
3565 | * each object to conform to the alignment. | |
3566 | */ | |
45906855 | 3567 | size = ALIGN(size, s->align); |
81819f0f | 3568 | s->size = size; |
06b285dc CL |
3569 | if (forced_order >= 0) |
3570 | order = forced_order; | |
3571 | else | |
ab9a0f19 | 3572 | order = calculate_order(size, s->reserved); |
81819f0f | 3573 | |
834f3d11 | 3574 | if (order < 0) |
81819f0f CL |
3575 | return 0; |
3576 | ||
b7a49f0d | 3577 | s->allocflags = 0; |
834f3d11 | 3578 | if (order) |
b7a49f0d CL |
3579 | s->allocflags |= __GFP_COMP; |
3580 | ||
3581 | if (s->flags & SLAB_CACHE_DMA) | |
2c59dd65 | 3582 | s->allocflags |= GFP_DMA; |
b7a49f0d CL |
3583 | |
3584 | if (s->flags & SLAB_RECLAIM_ACCOUNT) | |
3585 | s->allocflags |= __GFP_RECLAIMABLE; | |
3586 | ||
81819f0f CL |
3587 | /* |
3588 | * Determine the number of objects per slab | |
3589 | */ | |
ab9a0f19 LJ |
3590 | s->oo = oo_make(order, size, s->reserved); |
3591 | s->min = oo_make(get_order(size), size, s->reserved); | |
205ab99d CL |
3592 | if (oo_objects(s->oo) > oo_objects(s->max)) |
3593 | s->max = s->oo; | |
81819f0f | 3594 | |
834f3d11 | 3595 | return !!oo_objects(s->oo); |
81819f0f CL |
3596 | } |
3597 | ||
d50112ed | 3598 | static int kmem_cache_open(struct kmem_cache *s, slab_flags_t flags) |
81819f0f | 3599 | { |
8a13a4cc | 3600 | s->flags = kmem_cache_flags(s->size, flags, s->name, s->ctor); |
ab9a0f19 | 3601 | s->reserved = 0; |
2482ddec KC |
3602 | #ifdef CONFIG_SLAB_FREELIST_HARDENED |
3603 | s->random = get_random_long(); | |
3604 | #endif | |
81819f0f | 3605 | |
5f0d5a3a | 3606 | if (need_reserve_slab_rcu && (s->flags & SLAB_TYPESAFE_BY_RCU)) |
da9a638c | 3607 | s->reserved = sizeof(struct rcu_head); |
81819f0f | 3608 | |
06b285dc | 3609 | if (!calculate_sizes(s, -1)) |
81819f0f | 3610 | goto error; |
3de47213 DR |
3611 | if (disable_higher_order_debug) { |
3612 | /* | |
3613 | * Disable debugging flags that store metadata if the min slab | |
3614 | * order increased. | |
3615 | */ | |
3b0efdfa | 3616 | if (get_order(s->size) > get_order(s->object_size)) { |
3de47213 DR |
3617 | s->flags &= ~DEBUG_METADATA_FLAGS; |
3618 | s->offset = 0; | |
3619 | if (!calculate_sizes(s, -1)) | |
3620 | goto error; | |
3621 | } | |
3622 | } | |
81819f0f | 3623 | |
2565409f HC |
3624 | #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \ |
3625 | defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE) | |
149daaf3 | 3626 | if (system_has_cmpxchg_double() && (s->flags & SLAB_NO_CMPXCHG) == 0) |
b789ef51 CL |
3627 | /* Enable fast mode */ |
3628 | s->flags |= __CMPXCHG_DOUBLE; | |
3629 | #endif | |
3630 | ||
3b89d7d8 DR |
3631 | /* |
3632 | * The larger the object size is, the more pages we want on the partial | |
3633 | * list to avoid pounding the page allocator excessively. | |
3634 | */ | |
49e22585 CL |
3635 | set_min_partial(s, ilog2(s->size) / 2); |
3636 | ||
e6d0e1dc | 3637 | set_cpu_partial(s); |
49e22585 | 3638 | |
81819f0f | 3639 | #ifdef CONFIG_NUMA |
e2cb96b7 | 3640 | s->remote_node_defrag_ratio = 1000; |
81819f0f | 3641 | #endif |
210e7a43 TG |
3642 | |
3643 | /* Initialize the pre-computed randomized freelist if slab is up */ | |
3644 | if (slab_state >= UP) { | |
3645 | if (init_cache_random_seq(s)) | |
3646 | goto error; | |
3647 | } | |
3648 | ||
55136592 | 3649 | if (!init_kmem_cache_nodes(s)) |
dfb4f096 | 3650 | goto error; |
81819f0f | 3651 | |
55136592 | 3652 | if (alloc_kmem_cache_cpus(s)) |
278b1bb1 | 3653 | return 0; |
ff12059e | 3654 | |
4c93c355 | 3655 | free_kmem_cache_nodes(s); |
81819f0f CL |
3656 | error: |
3657 | if (flags & SLAB_PANIC) | |
756a025f JP |
3658 | panic("Cannot create slab %s size=%lu realsize=%u order=%u offset=%u flags=%lx\n", |
3659 | s->name, (unsigned long)s->size, s->size, | |
4fd0b46e | 3660 | oo_order(s->oo), s->offset, (unsigned long)flags); |
278b1bb1 | 3661 | return -EINVAL; |
81819f0f | 3662 | } |
81819f0f | 3663 | |
33b12c38 CL |
3664 | static void list_slab_objects(struct kmem_cache *s, struct page *page, |
3665 | const char *text) | |
3666 | { | |
3667 | #ifdef CONFIG_SLUB_DEBUG | |
3668 | void *addr = page_address(page); | |
3669 | void *p; | |
a5dd5c11 NK |
3670 | unsigned long *map = kzalloc(BITS_TO_LONGS(page->objects) * |
3671 | sizeof(long), GFP_ATOMIC); | |
bbd7d57b ED |
3672 | if (!map) |
3673 | return; | |
945cf2b6 | 3674 | slab_err(s, page, text, s->name); |
33b12c38 | 3675 | slab_lock(page); |
33b12c38 | 3676 | |
5f80b13a | 3677 | get_map(s, page, map); |
33b12c38 CL |
3678 | for_each_object(p, s, addr, page->objects) { |
3679 | ||
3680 | if (!test_bit(slab_index(p, s, addr), map)) { | |
f9f58285 | 3681 | pr_err("INFO: Object 0x%p @offset=%tu\n", p, p - addr); |
33b12c38 CL |
3682 | print_tracking(s, p); |
3683 | } | |
3684 | } | |
3685 | slab_unlock(page); | |
bbd7d57b | 3686 | kfree(map); |
33b12c38 CL |
3687 | #endif |
3688 | } | |
3689 | ||
81819f0f | 3690 | /* |
599870b1 | 3691 | * Attempt to free all partial slabs on a node. |
52b4b950 DS |
3692 | * This is called from __kmem_cache_shutdown(). We must take list_lock |
3693 | * because sysfs file might still access partial list after the shutdowning. | |
81819f0f | 3694 | */ |
599870b1 | 3695 | static void free_partial(struct kmem_cache *s, struct kmem_cache_node *n) |
81819f0f | 3696 | { |
60398923 | 3697 | LIST_HEAD(discard); |
81819f0f CL |
3698 | struct page *page, *h; |
3699 | ||
52b4b950 DS |
3700 | BUG_ON(irqs_disabled()); |
3701 | spin_lock_irq(&n->list_lock); | |
33b12c38 | 3702 | list_for_each_entry_safe(page, h, &n->partial, lru) { |
81819f0f | 3703 | if (!page->inuse) { |
52b4b950 | 3704 | remove_partial(n, page); |
60398923 | 3705 | list_add(&page->lru, &discard); |
33b12c38 CL |
3706 | } else { |
3707 | list_slab_objects(s, page, | |
52b4b950 | 3708 | "Objects remaining in %s on __kmem_cache_shutdown()"); |
599870b1 | 3709 | } |
33b12c38 | 3710 | } |
52b4b950 | 3711 | spin_unlock_irq(&n->list_lock); |
60398923 CW |
3712 | |
3713 | list_for_each_entry_safe(page, h, &discard, lru) | |
3714 | discard_slab(s, page); | |
81819f0f CL |
3715 | } |
3716 | ||
3717 | /* | |
672bba3a | 3718 | * Release all resources used by a slab cache. |
81819f0f | 3719 | */ |
52b4b950 | 3720 | int __kmem_cache_shutdown(struct kmem_cache *s) |
81819f0f CL |
3721 | { |
3722 | int node; | |
fa45dc25 | 3723 | struct kmem_cache_node *n; |
81819f0f CL |
3724 | |
3725 | flush_all(s); | |
81819f0f | 3726 | /* Attempt to free all objects */ |
fa45dc25 | 3727 | for_each_kmem_cache_node(s, node, n) { |
599870b1 CL |
3728 | free_partial(s, n); |
3729 | if (n->nr_partial || slabs_node(s, node)) | |
81819f0f CL |
3730 | return 1; |
3731 | } | |
bf5eb3de | 3732 | sysfs_slab_remove(s); |
81819f0f CL |
3733 | return 0; |
3734 | } | |
3735 | ||
81819f0f CL |
3736 | /******************************************************************** |
3737 | * Kmalloc subsystem | |
3738 | *******************************************************************/ | |
3739 | ||
81819f0f CL |
3740 | static int __init setup_slub_min_order(char *str) |
3741 | { | |
06428780 | 3742 | get_option(&str, &slub_min_order); |
81819f0f CL |
3743 | |
3744 | return 1; | |
3745 | } | |
3746 | ||
3747 | __setup("slub_min_order=", setup_slub_min_order); | |
3748 | ||
3749 | static int __init setup_slub_max_order(char *str) | |
3750 | { | |
06428780 | 3751 | get_option(&str, &slub_max_order); |
818cf590 | 3752 | slub_max_order = min(slub_max_order, MAX_ORDER - 1); |
81819f0f CL |
3753 | |
3754 | return 1; | |
3755 | } | |
3756 | ||
3757 | __setup("slub_max_order=", setup_slub_max_order); | |
3758 | ||
3759 | static int __init setup_slub_min_objects(char *str) | |
3760 | { | |
06428780 | 3761 | get_option(&str, &slub_min_objects); |
81819f0f CL |
3762 | |
3763 | return 1; | |
3764 | } | |
3765 | ||
3766 | __setup("slub_min_objects=", setup_slub_min_objects); | |
3767 | ||
81819f0f CL |
3768 | void *__kmalloc(size_t size, gfp_t flags) |
3769 | { | |
aadb4bc4 | 3770 | struct kmem_cache *s; |
5b882be4 | 3771 | void *ret; |
81819f0f | 3772 | |
95a05b42 | 3773 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) |
eada35ef | 3774 | return kmalloc_large(size, flags); |
aadb4bc4 | 3775 | |
2c59dd65 | 3776 | s = kmalloc_slab(size, flags); |
aadb4bc4 CL |
3777 | |
3778 | if (unlikely(ZERO_OR_NULL_PTR(s))) | |
6cb8f913 CL |
3779 | return s; |
3780 | ||
2b847c3c | 3781 | ret = slab_alloc(s, flags, _RET_IP_); |
5b882be4 | 3782 | |
ca2b84cb | 3783 | trace_kmalloc(_RET_IP_, ret, size, s->size, flags); |
5b882be4 | 3784 | |
505f5dcb | 3785 | kasan_kmalloc(s, ret, size, flags); |
0316bec2 | 3786 | |
5b882be4 | 3787 | return ret; |
81819f0f CL |
3788 | } |
3789 | EXPORT_SYMBOL(__kmalloc); | |
3790 | ||
5d1f57e4 | 3791 | #ifdef CONFIG_NUMA |
f619cfe1 CL |
3792 | static void *kmalloc_large_node(size_t size, gfp_t flags, int node) |
3793 | { | |
b1eeab67 | 3794 | struct page *page; |
e4f7c0b4 | 3795 | void *ptr = NULL; |
f619cfe1 | 3796 | |
52383431 | 3797 | flags |= __GFP_COMP | __GFP_NOTRACK; |
4949148a | 3798 | page = alloc_pages_node(node, flags, get_order(size)); |
f619cfe1 | 3799 | if (page) |
e4f7c0b4 CM |
3800 | ptr = page_address(page); |
3801 | ||
d56791b3 | 3802 | kmalloc_large_node_hook(ptr, size, flags); |
e4f7c0b4 | 3803 | return ptr; |
f619cfe1 CL |
3804 | } |
3805 | ||
81819f0f CL |
3806 | void *__kmalloc_node(size_t size, gfp_t flags, int node) |
3807 | { | |
aadb4bc4 | 3808 | struct kmem_cache *s; |
5b882be4 | 3809 | void *ret; |
81819f0f | 3810 | |
95a05b42 | 3811 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) { |
5b882be4 EGM |
3812 | ret = kmalloc_large_node(size, flags, node); |
3813 | ||
ca2b84cb EGM |
3814 | trace_kmalloc_node(_RET_IP_, ret, |
3815 | size, PAGE_SIZE << get_order(size), | |
3816 | flags, node); | |
5b882be4 EGM |
3817 | |
3818 | return ret; | |
3819 | } | |
aadb4bc4 | 3820 | |
2c59dd65 | 3821 | s = kmalloc_slab(size, flags); |
aadb4bc4 CL |
3822 | |
3823 | if (unlikely(ZERO_OR_NULL_PTR(s))) | |
6cb8f913 CL |
3824 | return s; |
3825 | ||
2b847c3c | 3826 | ret = slab_alloc_node(s, flags, node, _RET_IP_); |
5b882be4 | 3827 | |
ca2b84cb | 3828 | trace_kmalloc_node(_RET_IP_, ret, size, s->size, flags, node); |
5b882be4 | 3829 | |
505f5dcb | 3830 | kasan_kmalloc(s, ret, size, flags); |
0316bec2 | 3831 | |
5b882be4 | 3832 | return ret; |
81819f0f CL |
3833 | } |
3834 | EXPORT_SYMBOL(__kmalloc_node); | |
3835 | #endif | |
3836 | ||
ed18adc1 KC |
3837 | #ifdef CONFIG_HARDENED_USERCOPY |
3838 | /* | |
3839 | * Rejects objects that are incorrectly sized. | |
3840 | * | |
3841 | * Returns NULL if check passes, otherwise const char * to name of cache | |
3842 | * to indicate an error. | |
3843 | */ | |
3844 | const char *__check_heap_object(const void *ptr, unsigned long n, | |
3845 | struct page *page) | |
3846 | { | |
3847 | struct kmem_cache *s; | |
3848 | unsigned long offset; | |
3849 | size_t object_size; | |
3850 | ||
3851 | /* Find object and usable object size. */ | |
3852 | s = page->slab_cache; | |
3853 | object_size = slab_ksize(s); | |
3854 | ||
3855 | /* Reject impossible pointers. */ | |
3856 | if (ptr < page_address(page)) | |
3857 | return s->name; | |
3858 | ||
3859 | /* Find offset within object. */ | |
3860 | offset = (ptr - page_address(page)) % s->size; | |
3861 | ||
3862 | /* Adjust for redzone and reject if within the redzone. */ | |
3863 | if (kmem_cache_debug(s) && s->flags & SLAB_RED_ZONE) { | |
3864 | if (offset < s->red_left_pad) | |
3865 | return s->name; | |
3866 | offset -= s->red_left_pad; | |
3867 | } | |
3868 | ||
3869 | /* Allow address range falling entirely within object size. */ | |
3870 | if (offset <= object_size && n <= object_size - offset) | |
3871 | return NULL; | |
3872 | ||
3873 | return s->name; | |
3874 | } | |
3875 | #endif /* CONFIG_HARDENED_USERCOPY */ | |
3876 | ||
0316bec2 | 3877 | static size_t __ksize(const void *object) |
81819f0f | 3878 | { |
272c1d21 | 3879 | struct page *page; |
81819f0f | 3880 | |
ef8b4520 | 3881 | if (unlikely(object == ZERO_SIZE_PTR)) |
272c1d21 CL |
3882 | return 0; |
3883 | ||
294a80a8 | 3884 | page = virt_to_head_page(object); |
294a80a8 | 3885 | |
76994412 PE |
3886 | if (unlikely(!PageSlab(page))) { |
3887 | WARN_ON(!PageCompound(page)); | |
294a80a8 | 3888 | return PAGE_SIZE << compound_order(page); |
76994412 | 3889 | } |
81819f0f | 3890 | |
1b4f59e3 | 3891 | return slab_ksize(page->slab_cache); |
81819f0f | 3892 | } |
0316bec2 AR |
3893 | |
3894 | size_t ksize(const void *object) | |
3895 | { | |
3896 | size_t size = __ksize(object); | |
3897 | /* We assume that ksize callers could use whole allocated area, | |
4ebb31a4 AP |
3898 | * so we need to unpoison this area. |
3899 | */ | |
3900 | kasan_unpoison_shadow(object, size); | |
0316bec2 AR |
3901 | return size; |
3902 | } | |
b1aabecd | 3903 | EXPORT_SYMBOL(ksize); |
81819f0f CL |
3904 | |
3905 | void kfree(const void *x) | |
3906 | { | |
81819f0f | 3907 | struct page *page; |
5bb983b0 | 3908 | void *object = (void *)x; |
81819f0f | 3909 | |
2121db74 PE |
3910 | trace_kfree(_RET_IP_, x); |
3911 | ||
2408c550 | 3912 | if (unlikely(ZERO_OR_NULL_PTR(x))) |
81819f0f CL |
3913 | return; |
3914 | ||
b49af68f | 3915 | page = virt_to_head_page(x); |
aadb4bc4 | 3916 | if (unlikely(!PageSlab(page))) { |
0937502a | 3917 | BUG_ON(!PageCompound(page)); |
d56791b3 | 3918 | kfree_hook(x); |
4949148a | 3919 | __free_pages(page, compound_order(page)); |
aadb4bc4 CL |
3920 | return; |
3921 | } | |
81084651 | 3922 | slab_free(page->slab_cache, page, object, NULL, 1, _RET_IP_); |
81819f0f CL |
3923 | } |
3924 | EXPORT_SYMBOL(kfree); | |
3925 | ||
832f37f5 VD |
3926 | #define SHRINK_PROMOTE_MAX 32 |
3927 | ||
2086d26a | 3928 | /* |
832f37f5 VD |
3929 | * kmem_cache_shrink discards empty slabs and promotes the slabs filled |
3930 | * up most to the head of the partial lists. New allocations will then | |
3931 | * fill those up and thus they can be removed from the partial lists. | |
672bba3a CL |
3932 | * |
3933 | * The slabs with the least items are placed last. This results in them | |
3934 | * being allocated from last increasing the chance that the last objects | |
3935 | * are freed in them. | |
2086d26a | 3936 | */ |
c9fc5864 | 3937 | int __kmem_cache_shrink(struct kmem_cache *s) |
2086d26a CL |
3938 | { |
3939 | int node; | |
3940 | int i; | |
3941 | struct kmem_cache_node *n; | |
3942 | struct page *page; | |
3943 | struct page *t; | |
832f37f5 VD |
3944 | struct list_head discard; |
3945 | struct list_head promote[SHRINK_PROMOTE_MAX]; | |
2086d26a | 3946 | unsigned long flags; |
ce3712d7 | 3947 | int ret = 0; |
2086d26a | 3948 | |
2086d26a | 3949 | flush_all(s); |
fa45dc25 | 3950 | for_each_kmem_cache_node(s, node, n) { |
832f37f5 VD |
3951 | INIT_LIST_HEAD(&discard); |
3952 | for (i = 0; i < SHRINK_PROMOTE_MAX; i++) | |
3953 | INIT_LIST_HEAD(promote + i); | |
2086d26a CL |
3954 | |
3955 | spin_lock_irqsave(&n->list_lock, flags); | |
3956 | ||
3957 | /* | |
832f37f5 | 3958 | * Build lists of slabs to discard or promote. |
2086d26a | 3959 | * |
672bba3a CL |
3960 | * Note that concurrent frees may occur while we hold the |
3961 | * list_lock. page->inuse here is the upper limit. | |
2086d26a CL |
3962 | */ |
3963 | list_for_each_entry_safe(page, t, &n->partial, lru) { | |
832f37f5 VD |
3964 | int free = page->objects - page->inuse; |
3965 | ||
3966 | /* Do not reread page->inuse */ | |
3967 | barrier(); | |
3968 | ||
3969 | /* We do not keep full slabs on the list */ | |
3970 | BUG_ON(free <= 0); | |
3971 | ||
3972 | if (free == page->objects) { | |
3973 | list_move(&page->lru, &discard); | |
69cb8e6b | 3974 | n->nr_partial--; |
832f37f5 VD |
3975 | } else if (free <= SHRINK_PROMOTE_MAX) |
3976 | list_move(&page->lru, promote + free - 1); | |
2086d26a CL |
3977 | } |
3978 | ||
2086d26a | 3979 | /* |
832f37f5 VD |
3980 | * Promote the slabs filled up most to the head of the |
3981 | * partial list. | |
2086d26a | 3982 | */ |
832f37f5 VD |
3983 | for (i = SHRINK_PROMOTE_MAX - 1; i >= 0; i--) |
3984 | list_splice(promote + i, &n->partial); | |
2086d26a | 3985 | |
2086d26a | 3986 | spin_unlock_irqrestore(&n->list_lock, flags); |
69cb8e6b CL |
3987 | |
3988 | /* Release empty slabs */ | |
832f37f5 | 3989 | list_for_each_entry_safe(page, t, &discard, lru) |
69cb8e6b | 3990 | discard_slab(s, page); |
ce3712d7 VD |
3991 | |
3992 | if (slabs_node(s, node)) | |
3993 | ret = 1; | |
2086d26a CL |
3994 | } |
3995 | ||
ce3712d7 | 3996 | return ret; |
2086d26a | 3997 | } |
2086d26a | 3998 | |
c9fc5864 | 3999 | #ifdef CONFIG_MEMCG |
01fb58bc TH |
4000 | static void kmemcg_cache_deact_after_rcu(struct kmem_cache *s) |
4001 | { | |
50862ce7 TH |
4002 | /* |
4003 | * Called with all the locks held after a sched RCU grace period. | |
4004 | * Even if @s becomes empty after shrinking, we can't know that @s | |
4005 | * doesn't have allocations already in-flight and thus can't | |
4006 | * destroy @s until the associated memcg is released. | |
4007 | * | |
4008 | * However, let's remove the sysfs files for empty caches here. | |
4009 | * Each cache has a lot of interface files which aren't | |
4010 | * particularly useful for empty draining caches; otherwise, we can | |
4011 | * easily end up with millions of unnecessary sysfs files on | |
4012 | * systems which have a lot of memory and transient cgroups. | |
4013 | */ | |
4014 | if (!__kmem_cache_shrink(s)) | |
4015 | sysfs_slab_remove(s); | |
01fb58bc TH |
4016 | } |
4017 | ||
c9fc5864 TH |
4018 | void __kmemcg_cache_deactivate(struct kmem_cache *s) |
4019 | { | |
4020 | /* | |
4021 | * Disable empty slabs caching. Used to avoid pinning offline | |
4022 | * memory cgroups by kmem pages that can be freed. | |
4023 | */ | |
e6d0e1dc | 4024 | slub_set_cpu_partial(s, 0); |
c9fc5864 TH |
4025 | s->min_partial = 0; |
4026 | ||
4027 | /* | |
4028 | * s->cpu_partial is checked locklessly (see put_cpu_partial), so | |
01fb58bc | 4029 | * we have to make sure the change is visible before shrinking. |
c9fc5864 | 4030 | */ |
01fb58bc | 4031 | slab_deactivate_memcg_cache_rcu_sched(s, kmemcg_cache_deact_after_rcu); |
c9fc5864 TH |
4032 | } |
4033 | #endif | |
4034 | ||
b9049e23 YG |
4035 | static int slab_mem_going_offline_callback(void *arg) |
4036 | { | |
4037 | struct kmem_cache *s; | |
4038 | ||
18004c5d | 4039 | mutex_lock(&slab_mutex); |
b9049e23 | 4040 | list_for_each_entry(s, &slab_caches, list) |
c9fc5864 | 4041 | __kmem_cache_shrink(s); |
18004c5d | 4042 | mutex_unlock(&slab_mutex); |
b9049e23 YG |
4043 | |
4044 | return 0; | |
4045 | } | |
4046 | ||
4047 | static void slab_mem_offline_callback(void *arg) | |
4048 | { | |
4049 | struct kmem_cache_node *n; | |
4050 | struct kmem_cache *s; | |
4051 | struct memory_notify *marg = arg; | |
4052 | int offline_node; | |
4053 | ||
b9d5ab25 | 4054 | offline_node = marg->status_change_nid_normal; |
b9049e23 YG |
4055 | |
4056 | /* | |
4057 | * If the node still has available memory. we need kmem_cache_node | |
4058 | * for it yet. | |
4059 | */ | |
4060 | if (offline_node < 0) | |
4061 | return; | |
4062 | ||
18004c5d | 4063 | mutex_lock(&slab_mutex); |
b9049e23 YG |
4064 | list_for_each_entry(s, &slab_caches, list) { |
4065 | n = get_node(s, offline_node); | |
4066 | if (n) { | |
4067 | /* | |
4068 | * if n->nr_slabs > 0, slabs still exist on the node | |
4069 | * that is going down. We were unable to free them, | |
c9404c9c | 4070 | * and offline_pages() function shouldn't call this |
b9049e23 YG |
4071 | * callback. So, we must fail. |
4072 | */ | |
0f389ec6 | 4073 | BUG_ON(slabs_node(s, offline_node)); |
b9049e23 YG |
4074 | |
4075 | s->node[offline_node] = NULL; | |
8de66a0c | 4076 | kmem_cache_free(kmem_cache_node, n); |
b9049e23 YG |
4077 | } |
4078 | } | |
18004c5d | 4079 | mutex_unlock(&slab_mutex); |
b9049e23 YG |
4080 | } |
4081 | ||
4082 | static int slab_mem_going_online_callback(void *arg) | |
4083 | { | |
4084 | struct kmem_cache_node *n; | |
4085 | struct kmem_cache *s; | |
4086 | struct memory_notify *marg = arg; | |
b9d5ab25 | 4087 | int nid = marg->status_change_nid_normal; |
b9049e23 YG |
4088 | int ret = 0; |
4089 | ||
4090 | /* | |
4091 | * If the node's memory is already available, then kmem_cache_node is | |
4092 | * already created. Nothing to do. | |
4093 | */ | |
4094 | if (nid < 0) | |
4095 | return 0; | |
4096 | ||
4097 | /* | |
0121c619 | 4098 | * We are bringing a node online. No memory is available yet. We must |
b9049e23 YG |
4099 | * allocate a kmem_cache_node structure in order to bring the node |
4100 | * online. | |
4101 | */ | |
18004c5d | 4102 | mutex_lock(&slab_mutex); |
b9049e23 YG |
4103 | list_for_each_entry(s, &slab_caches, list) { |
4104 | /* | |
4105 | * XXX: kmem_cache_alloc_node will fallback to other nodes | |
4106 | * since memory is not yet available from the node that | |
4107 | * is brought up. | |
4108 | */ | |
8de66a0c | 4109 | n = kmem_cache_alloc(kmem_cache_node, GFP_KERNEL); |
b9049e23 YG |
4110 | if (!n) { |
4111 | ret = -ENOMEM; | |
4112 | goto out; | |
4113 | } | |
4053497d | 4114 | init_kmem_cache_node(n); |
b9049e23 YG |
4115 | s->node[nid] = n; |
4116 | } | |
4117 | out: | |
18004c5d | 4118 | mutex_unlock(&slab_mutex); |
b9049e23 YG |
4119 | return ret; |
4120 | } | |
4121 | ||
4122 | static int slab_memory_callback(struct notifier_block *self, | |
4123 | unsigned long action, void *arg) | |
4124 | { | |
4125 | int ret = 0; | |
4126 | ||
4127 | switch (action) { | |
4128 | case MEM_GOING_ONLINE: | |
4129 | ret = slab_mem_going_online_callback(arg); | |
4130 | break; | |
4131 | case MEM_GOING_OFFLINE: | |
4132 | ret = slab_mem_going_offline_callback(arg); | |
4133 | break; | |
4134 | case MEM_OFFLINE: | |
4135 | case MEM_CANCEL_ONLINE: | |
4136 | slab_mem_offline_callback(arg); | |
4137 | break; | |
4138 | case MEM_ONLINE: | |
4139 | case MEM_CANCEL_OFFLINE: | |
4140 | break; | |
4141 | } | |
dc19f9db KH |
4142 | if (ret) |
4143 | ret = notifier_from_errno(ret); | |
4144 | else | |
4145 | ret = NOTIFY_OK; | |
b9049e23 YG |
4146 | return ret; |
4147 | } | |
4148 | ||
3ac38faa AM |
4149 | static struct notifier_block slab_memory_callback_nb = { |
4150 | .notifier_call = slab_memory_callback, | |
4151 | .priority = SLAB_CALLBACK_PRI, | |
4152 | }; | |
b9049e23 | 4153 | |
81819f0f CL |
4154 | /******************************************************************** |
4155 | * Basic setup of slabs | |
4156 | *******************************************************************/ | |
4157 | ||
51df1142 CL |
4158 | /* |
4159 | * Used for early kmem_cache structures that were allocated using | |
dffb4d60 CL |
4160 | * the page allocator. Allocate them properly then fix up the pointers |
4161 | * that may be pointing to the wrong kmem_cache structure. | |
51df1142 CL |
4162 | */ |
4163 | ||
dffb4d60 | 4164 | static struct kmem_cache * __init bootstrap(struct kmem_cache *static_cache) |
51df1142 CL |
4165 | { |
4166 | int node; | |
dffb4d60 | 4167 | struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT); |
fa45dc25 | 4168 | struct kmem_cache_node *n; |
51df1142 | 4169 | |
dffb4d60 | 4170 | memcpy(s, static_cache, kmem_cache->object_size); |
51df1142 | 4171 | |
7d557b3c GC |
4172 | /* |
4173 | * This runs very early, and only the boot processor is supposed to be | |
4174 | * up. Even if it weren't true, IRQs are not up so we couldn't fire | |
4175 | * IPIs around. | |
4176 | */ | |
4177 | __flush_cpu_slab(s, smp_processor_id()); | |
fa45dc25 | 4178 | for_each_kmem_cache_node(s, node, n) { |
51df1142 CL |
4179 | struct page *p; |
4180 | ||
fa45dc25 CL |
4181 | list_for_each_entry(p, &n->partial, lru) |
4182 | p->slab_cache = s; | |
51df1142 | 4183 | |
607bf324 | 4184 | #ifdef CONFIG_SLUB_DEBUG |
fa45dc25 CL |
4185 | list_for_each_entry(p, &n->full, lru) |
4186 | p->slab_cache = s; | |
51df1142 | 4187 | #endif |
51df1142 | 4188 | } |
f7ce3190 | 4189 | slab_init_memcg_params(s); |
dffb4d60 | 4190 | list_add(&s->list, &slab_caches); |
510ded33 | 4191 | memcg_link_cache(s); |
dffb4d60 | 4192 | return s; |
51df1142 CL |
4193 | } |
4194 | ||
81819f0f CL |
4195 | void __init kmem_cache_init(void) |
4196 | { | |
dffb4d60 CL |
4197 | static __initdata struct kmem_cache boot_kmem_cache, |
4198 | boot_kmem_cache_node; | |
51df1142 | 4199 | |
fc8d8620 SG |
4200 | if (debug_guardpage_minorder()) |
4201 | slub_max_order = 0; | |
4202 | ||
dffb4d60 CL |
4203 | kmem_cache_node = &boot_kmem_cache_node; |
4204 | kmem_cache = &boot_kmem_cache; | |
51df1142 | 4205 | |
dffb4d60 CL |
4206 | create_boot_cache(kmem_cache_node, "kmem_cache_node", |
4207 | sizeof(struct kmem_cache_node), SLAB_HWCACHE_ALIGN); | |
b9049e23 | 4208 | |
3ac38faa | 4209 | register_hotmemory_notifier(&slab_memory_callback_nb); |
81819f0f CL |
4210 | |
4211 | /* Able to allocate the per node structures */ | |
4212 | slab_state = PARTIAL; | |
4213 | ||
dffb4d60 CL |
4214 | create_boot_cache(kmem_cache, "kmem_cache", |
4215 | offsetof(struct kmem_cache, node) + | |
4216 | nr_node_ids * sizeof(struct kmem_cache_node *), | |
4217 | SLAB_HWCACHE_ALIGN); | |
8a13a4cc | 4218 | |
dffb4d60 | 4219 | kmem_cache = bootstrap(&boot_kmem_cache); |
81819f0f | 4220 | |
51df1142 CL |
4221 | /* |
4222 | * Allocate kmem_cache_node properly from the kmem_cache slab. | |
4223 | * kmem_cache_node is separately allocated so no need to | |
4224 | * update any list pointers. | |
4225 | */ | |
dffb4d60 | 4226 | kmem_cache_node = bootstrap(&boot_kmem_cache_node); |
51df1142 CL |
4227 | |
4228 | /* Now we can use the kmem_cache to allocate kmalloc slabs */ | |
34cc6990 | 4229 | setup_kmalloc_cache_index_table(); |
f97d5f63 | 4230 | create_kmalloc_caches(0); |
81819f0f | 4231 | |
210e7a43 TG |
4232 | /* Setup random freelists for each cache */ |
4233 | init_freelist_randomization(); | |
4234 | ||
a96a87bf SAS |
4235 | cpuhp_setup_state_nocalls(CPUHP_SLUB_DEAD, "slub:dead", NULL, |
4236 | slub_cpu_dead); | |
81819f0f | 4237 | |
9b130ad5 | 4238 | pr_info("SLUB: HWalign=%d, Order=%d-%d, MinObjects=%d, CPUs=%u, Nodes=%d\n", |
f97d5f63 | 4239 | cache_line_size(), |
81819f0f CL |
4240 | slub_min_order, slub_max_order, slub_min_objects, |
4241 | nr_cpu_ids, nr_node_ids); | |
4242 | } | |
4243 | ||
7e85ee0c PE |
4244 | void __init kmem_cache_init_late(void) |
4245 | { | |
7e85ee0c PE |
4246 | } |
4247 | ||
2633d7a0 | 4248 | struct kmem_cache * |
a44cb944 | 4249 | __kmem_cache_alias(const char *name, size_t size, size_t align, |
d50112ed | 4250 | slab_flags_t flags, void (*ctor)(void *)) |
81819f0f | 4251 | { |
426589f5 | 4252 | struct kmem_cache *s, *c; |
81819f0f | 4253 | |
a44cb944 | 4254 | s = find_mergeable(size, align, flags, name, ctor); |
81819f0f CL |
4255 | if (s) { |
4256 | s->refcount++; | |
84d0ddd6 | 4257 | |
81819f0f CL |
4258 | /* |
4259 | * Adjust the object sizes so that we clear | |
4260 | * the complete object on kzalloc. | |
4261 | */ | |
3b0efdfa | 4262 | s->object_size = max(s->object_size, (int)size); |
81819f0f | 4263 | s->inuse = max_t(int, s->inuse, ALIGN(size, sizeof(void *))); |
6446faa2 | 4264 | |
426589f5 | 4265 | for_each_memcg_cache(c, s) { |
84d0ddd6 VD |
4266 | c->object_size = s->object_size; |
4267 | c->inuse = max_t(int, c->inuse, | |
4268 | ALIGN(size, sizeof(void *))); | |
4269 | } | |
4270 | ||
7b8f3b66 | 4271 | if (sysfs_slab_alias(s, name)) { |
7b8f3b66 | 4272 | s->refcount--; |
cbb79694 | 4273 | s = NULL; |
7b8f3b66 | 4274 | } |
a0e1d1be | 4275 | } |
6446faa2 | 4276 | |
cbb79694 CL |
4277 | return s; |
4278 | } | |
84c1cf62 | 4279 | |
d50112ed | 4280 | int __kmem_cache_create(struct kmem_cache *s, slab_flags_t flags) |
cbb79694 | 4281 | { |
aac3a166 PE |
4282 | int err; |
4283 | ||
4284 | err = kmem_cache_open(s, flags); | |
4285 | if (err) | |
4286 | return err; | |
20cea968 | 4287 | |
45530c44 CL |
4288 | /* Mutex is not taken during early boot */ |
4289 | if (slab_state <= UP) | |
4290 | return 0; | |
4291 | ||
107dab5c | 4292 | memcg_propagate_slab_attrs(s); |
aac3a166 | 4293 | err = sysfs_slab_add(s); |
aac3a166 | 4294 | if (err) |
52b4b950 | 4295 | __kmem_cache_release(s); |
20cea968 | 4296 | |
aac3a166 | 4297 | return err; |
81819f0f | 4298 | } |
81819f0f | 4299 | |
ce71e27c | 4300 | void *__kmalloc_track_caller(size_t size, gfp_t gfpflags, unsigned long caller) |
81819f0f | 4301 | { |
aadb4bc4 | 4302 | struct kmem_cache *s; |
94b528d0 | 4303 | void *ret; |
aadb4bc4 | 4304 | |
95a05b42 | 4305 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) |
eada35ef PE |
4306 | return kmalloc_large(size, gfpflags); |
4307 | ||
2c59dd65 | 4308 | s = kmalloc_slab(size, gfpflags); |
81819f0f | 4309 | |
2408c550 | 4310 | if (unlikely(ZERO_OR_NULL_PTR(s))) |
6cb8f913 | 4311 | return s; |
81819f0f | 4312 | |
2b847c3c | 4313 | ret = slab_alloc(s, gfpflags, caller); |
94b528d0 | 4314 | |
25985edc | 4315 | /* Honor the call site pointer we received. */ |
ca2b84cb | 4316 | trace_kmalloc(caller, ret, size, s->size, gfpflags); |
94b528d0 EGM |
4317 | |
4318 | return ret; | |
81819f0f CL |
4319 | } |
4320 | ||
5d1f57e4 | 4321 | #ifdef CONFIG_NUMA |
81819f0f | 4322 | void *__kmalloc_node_track_caller(size_t size, gfp_t gfpflags, |
ce71e27c | 4323 | int node, unsigned long caller) |
81819f0f | 4324 | { |
aadb4bc4 | 4325 | struct kmem_cache *s; |
94b528d0 | 4326 | void *ret; |
aadb4bc4 | 4327 | |
95a05b42 | 4328 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) { |
d3e14aa3 XF |
4329 | ret = kmalloc_large_node(size, gfpflags, node); |
4330 | ||
4331 | trace_kmalloc_node(caller, ret, | |
4332 | size, PAGE_SIZE << get_order(size), | |
4333 | gfpflags, node); | |
4334 | ||
4335 | return ret; | |
4336 | } | |
eada35ef | 4337 | |
2c59dd65 | 4338 | s = kmalloc_slab(size, gfpflags); |
81819f0f | 4339 | |
2408c550 | 4340 | if (unlikely(ZERO_OR_NULL_PTR(s))) |
6cb8f913 | 4341 | return s; |
81819f0f | 4342 | |
2b847c3c | 4343 | ret = slab_alloc_node(s, gfpflags, node, caller); |
94b528d0 | 4344 | |
25985edc | 4345 | /* Honor the call site pointer we received. */ |
ca2b84cb | 4346 | trace_kmalloc_node(caller, ret, size, s->size, gfpflags, node); |
94b528d0 EGM |
4347 | |
4348 | return ret; | |
81819f0f | 4349 | } |
5d1f57e4 | 4350 | #endif |
81819f0f | 4351 | |
ab4d5ed5 | 4352 | #ifdef CONFIG_SYSFS |
205ab99d CL |
4353 | static int count_inuse(struct page *page) |
4354 | { | |
4355 | return page->inuse; | |
4356 | } | |
4357 | ||
4358 | static int count_total(struct page *page) | |
4359 | { | |
4360 | return page->objects; | |
4361 | } | |
ab4d5ed5 | 4362 | #endif |
205ab99d | 4363 | |
ab4d5ed5 | 4364 | #ifdef CONFIG_SLUB_DEBUG |
434e245d CL |
4365 | static int validate_slab(struct kmem_cache *s, struct page *page, |
4366 | unsigned long *map) | |
53e15af0 CL |
4367 | { |
4368 | void *p; | |
a973e9dd | 4369 | void *addr = page_address(page); |
53e15af0 CL |
4370 | |
4371 | if (!check_slab(s, page) || | |
4372 | !on_freelist(s, page, NULL)) | |
4373 | return 0; | |
4374 | ||
4375 | /* Now we know that a valid freelist exists */ | |
39b26464 | 4376 | bitmap_zero(map, page->objects); |
53e15af0 | 4377 | |
5f80b13a CL |
4378 | get_map(s, page, map); |
4379 | for_each_object(p, s, addr, page->objects) { | |
4380 | if (test_bit(slab_index(p, s, addr), map)) | |
4381 | if (!check_object(s, page, p, SLUB_RED_INACTIVE)) | |
4382 | return 0; | |
53e15af0 CL |
4383 | } |
4384 | ||
224a88be | 4385 | for_each_object(p, s, addr, page->objects) |
7656c72b | 4386 | if (!test_bit(slab_index(p, s, addr), map)) |
37d57443 | 4387 | if (!check_object(s, page, p, SLUB_RED_ACTIVE)) |
53e15af0 CL |
4388 | return 0; |
4389 | return 1; | |
4390 | } | |
4391 | ||
434e245d CL |
4392 | static void validate_slab_slab(struct kmem_cache *s, struct page *page, |
4393 | unsigned long *map) | |
53e15af0 | 4394 | { |
881db7fb CL |
4395 | slab_lock(page); |
4396 | validate_slab(s, page, map); | |
4397 | slab_unlock(page); | |
53e15af0 CL |
4398 | } |
4399 | ||
434e245d CL |
4400 | static int validate_slab_node(struct kmem_cache *s, |
4401 | struct kmem_cache_node *n, unsigned long *map) | |
53e15af0 CL |
4402 | { |
4403 | unsigned long count = 0; | |
4404 | struct page *page; | |
4405 | unsigned long flags; | |
4406 | ||
4407 | spin_lock_irqsave(&n->list_lock, flags); | |
4408 | ||
4409 | list_for_each_entry(page, &n->partial, lru) { | |
434e245d | 4410 | validate_slab_slab(s, page, map); |
53e15af0 CL |
4411 | count++; |
4412 | } | |
4413 | if (count != n->nr_partial) | |
f9f58285 FF |
4414 | pr_err("SLUB %s: %ld partial slabs counted but counter=%ld\n", |
4415 | s->name, count, n->nr_partial); | |
53e15af0 CL |
4416 | |
4417 | if (!(s->flags & SLAB_STORE_USER)) | |
4418 | goto out; | |
4419 | ||
4420 | list_for_each_entry(page, &n->full, lru) { | |
434e245d | 4421 | validate_slab_slab(s, page, map); |
53e15af0 CL |
4422 | count++; |
4423 | } | |
4424 | if (count != atomic_long_read(&n->nr_slabs)) | |
f9f58285 FF |
4425 | pr_err("SLUB: %s %ld slabs counted but counter=%ld\n", |
4426 | s->name, count, atomic_long_read(&n->nr_slabs)); | |
53e15af0 CL |
4427 | |
4428 | out: | |
4429 | spin_unlock_irqrestore(&n->list_lock, flags); | |
4430 | return count; | |
4431 | } | |
4432 | ||
434e245d | 4433 | static long validate_slab_cache(struct kmem_cache *s) |
53e15af0 CL |
4434 | { |
4435 | int node; | |
4436 | unsigned long count = 0; | |
205ab99d | 4437 | unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) * |
434e245d | 4438 | sizeof(unsigned long), GFP_KERNEL); |
fa45dc25 | 4439 | struct kmem_cache_node *n; |
434e245d CL |
4440 | |
4441 | if (!map) | |
4442 | return -ENOMEM; | |
53e15af0 CL |
4443 | |
4444 | flush_all(s); | |
fa45dc25 | 4445 | for_each_kmem_cache_node(s, node, n) |
434e245d | 4446 | count += validate_slab_node(s, n, map); |
434e245d | 4447 | kfree(map); |
53e15af0 CL |
4448 | return count; |
4449 | } | |
88a420e4 | 4450 | /* |
672bba3a | 4451 | * Generate lists of code addresses where slabcache objects are allocated |
88a420e4 CL |
4452 | * and freed. |
4453 | */ | |
4454 | ||
4455 | struct location { | |
4456 | unsigned long count; | |
ce71e27c | 4457 | unsigned long addr; |
45edfa58 CL |
4458 | long long sum_time; |
4459 | long min_time; | |
4460 | long max_time; | |
4461 | long min_pid; | |
4462 | long max_pid; | |
174596a0 | 4463 | DECLARE_BITMAP(cpus, NR_CPUS); |
45edfa58 | 4464 | nodemask_t nodes; |
88a420e4 CL |
4465 | }; |
4466 | ||
4467 | struct loc_track { | |
4468 | unsigned long max; | |
4469 | unsigned long count; | |
4470 | struct location *loc; | |
4471 | }; | |
4472 | ||
4473 | static void free_loc_track(struct loc_track *t) | |
4474 | { | |
4475 | if (t->max) | |
4476 | free_pages((unsigned long)t->loc, | |
4477 | get_order(sizeof(struct location) * t->max)); | |
4478 | } | |
4479 | ||
68dff6a9 | 4480 | static int alloc_loc_track(struct loc_track *t, unsigned long max, gfp_t flags) |
88a420e4 CL |
4481 | { |
4482 | struct location *l; | |
4483 | int order; | |
4484 | ||
88a420e4 CL |
4485 | order = get_order(sizeof(struct location) * max); |
4486 | ||
68dff6a9 | 4487 | l = (void *)__get_free_pages(flags, order); |
88a420e4 CL |
4488 | if (!l) |
4489 | return 0; | |
4490 | ||
4491 | if (t->count) { | |
4492 | memcpy(l, t->loc, sizeof(struct location) * t->count); | |
4493 | free_loc_track(t); | |
4494 | } | |
4495 | t->max = max; | |
4496 | t->loc = l; | |
4497 | return 1; | |
4498 | } | |
4499 | ||
4500 | static int add_location(struct loc_track *t, struct kmem_cache *s, | |
45edfa58 | 4501 | const struct track *track) |
88a420e4 CL |
4502 | { |
4503 | long start, end, pos; | |
4504 | struct location *l; | |
ce71e27c | 4505 | unsigned long caddr; |
45edfa58 | 4506 | unsigned long age = jiffies - track->when; |
88a420e4 CL |
4507 | |
4508 | start = -1; | |
4509 | end = t->count; | |
4510 | ||
4511 | for ( ; ; ) { | |
4512 | pos = start + (end - start + 1) / 2; | |
4513 | ||
4514 | /* | |
4515 | * There is nothing at "end". If we end up there | |
4516 | * we need to add something to before end. | |
4517 | */ | |
4518 | if (pos == end) | |
4519 | break; | |
4520 | ||
4521 | caddr = t->loc[pos].addr; | |
45edfa58 CL |
4522 | if (track->addr == caddr) { |
4523 | ||
4524 | l = &t->loc[pos]; | |
4525 | l->count++; | |
4526 | if (track->when) { | |
4527 | l->sum_time += age; | |
4528 | if (age < l->min_time) | |
4529 | l->min_time = age; | |
4530 | if (age > l->max_time) | |
4531 | l->max_time = age; | |
4532 | ||
4533 | if (track->pid < l->min_pid) | |
4534 | l->min_pid = track->pid; | |
4535 | if (track->pid > l->max_pid) | |
4536 | l->max_pid = track->pid; | |
4537 | ||
174596a0 RR |
4538 | cpumask_set_cpu(track->cpu, |
4539 | to_cpumask(l->cpus)); | |
45edfa58 CL |
4540 | } |
4541 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
88a420e4 CL |
4542 | return 1; |
4543 | } | |
4544 | ||
45edfa58 | 4545 | if (track->addr < caddr) |
88a420e4 CL |
4546 | end = pos; |
4547 | else | |
4548 | start = pos; | |
4549 | } | |
4550 | ||
4551 | /* | |
672bba3a | 4552 | * Not found. Insert new tracking element. |
88a420e4 | 4553 | */ |
68dff6a9 | 4554 | if (t->count >= t->max && !alloc_loc_track(t, 2 * t->max, GFP_ATOMIC)) |
88a420e4 CL |
4555 | return 0; |
4556 | ||
4557 | l = t->loc + pos; | |
4558 | if (pos < t->count) | |
4559 | memmove(l + 1, l, | |
4560 | (t->count - pos) * sizeof(struct location)); | |
4561 | t->count++; | |
4562 | l->count = 1; | |
45edfa58 CL |
4563 | l->addr = track->addr; |
4564 | l->sum_time = age; | |
4565 | l->min_time = age; | |
4566 | l->max_time = age; | |
4567 | l->min_pid = track->pid; | |
4568 | l->max_pid = track->pid; | |
174596a0 RR |
4569 | cpumask_clear(to_cpumask(l->cpus)); |
4570 | cpumask_set_cpu(track->cpu, to_cpumask(l->cpus)); | |
45edfa58 CL |
4571 | nodes_clear(l->nodes); |
4572 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
88a420e4 CL |
4573 | return 1; |
4574 | } | |
4575 | ||
4576 | static void process_slab(struct loc_track *t, struct kmem_cache *s, | |
bbd7d57b | 4577 | struct page *page, enum track_item alloc, |
a5dd5c11 | 4578 | unsigned long *map) |
88a420e4 | 4579 | { |
a973e9dd | 4580 | void *addr = page_address(page); |
88a420e4 CL |
4581 | void *p; |
4582 | ||
39b26464 | 4583 | bitmap_zero(map, page->objects); |
5f80b13a | 4584 | get_map(s, page, map); |
88a420e4 | 4585 | |
224a88be | 4586 | for_each_object(p, s, addr, page->objects) |
45edfa58 CL |
4587 | if (!test_bit(slab_index(p, s, addr), map)) |
4588 | add_location(t, s, get_track(s, p, alloc)); | |
88a420e4 CL |
4589 | } |
4590 | ||
4591 | static int list_locations(struct kmem_cache *s, char *buf, | |
4592 | enum track_item alloc) | |
4593 | { | |
e374d483 | 4594 | int len = 0; |
88a420e4 | 4595 | unsigned long i; |
68dff6a9 | 4596 | struct loc_track t = { 0, 0, NULL }; |
88a420e4 | 4597 | int node; |
bbd7d57b ED |
4598 | unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) * |
4599 | sizeof(unsigned long), GFP_KERNEL); | |
fa45dc25 | 4600 | struct kmem_cache_node *n; |
88a420e4 | 4601 | |
bbd7d57b | 4602 | if (!map || !alloc_loc_track(&t, PAGE_SIZE / sizeof(struct location), |
0ee931c4 | 4603 | GFP_KERNEL)) { |
bbd7d57b | 4604 | kfree(map); |
68dff6a9 | 4605 | return sprintf(buf, "Out of memory\n"); |
bbd7d57b | 4606 | } |
88a420e4 CL |
4607 | /* Push back cpu slabs */ |
4608 | flush_all(s); | |
4609 | ||
fa45dc25 | 4610 | for_each_kmem_cache_node(s, node, n) { |
88a420e4 CL |
4611 | unsigned long flags; |
4612 | struct page *page; | |
4613 | ||
9e86943b | 4614 | if (!atomic_long_read(&n->nr_slabs)) |
88a420e4 CL |
4615 | continue; |
4616 | ||
4617 | spin_lock_irqsave(&n->list_lock, flags); | |
4618 | list_for_each_entry(page, &n->partial, lru) | |
bbd7d57b | 4619 | process_slab(&t, s, page, alloc, map); |
88a420e4 | 4620 | list_for_each_entry(page, &n->full, lru) |
bbd7d57b | 4621 | process_slab(&t, s, page, alloc, map); |
88a420e4 CL |
4622 | spin_unlock_irqrestore(&n->list_lock, flags); |
4623 | } | |
4624 | ||
4625 | for (i = 0; i < t.count; i++) { | |
45edfa58 | 4626 | struct location *l = &t.loc[i]; |
88a420e4 | 4627 | |
9c246247 | 4628 | if (len > PAGE_SIZE - KSYM_SYMBOL_LEN - 100) |
88a420e4 | 4629 | break; |
e374d483 | 4630 | len += sprintf(buf + len, "%7ld ", l->count); |
45edfa58 CL |
4631 | |
4632 | if (l->addr) | |
62c70bce | 4633 | len += sprintf(buf + len, "%pS", (void *)l->addr); |
88a420e4 | 4634 | else |
e374d483 | 4635 | len += sprintf(buf + len, "<not-available>"); |
45edfa58 CL |
4636 | |
4637 | if (l->sum_time != l->min_time) { | |
e374d483 | 4638 | len += sprintf(buf + len, " age=%ld/%ld/%ld", |
f8bd2258 RZ |
4639 | l->min_time, |
4640 | (long)div_u64(l->sum_time, l->count), | |
4641 | l->max_time); | |
45edfa58 | 4642 | } else |
e374d483 | 4643 | len += sprintf(buf + len, " age=%ld", |
45edfa58 CL |
4644 | l->min_time); |
4645 | ||
4646 | if (l->min_pid != l->max_pid) | |
e374d483 | 4647 | len += sprintf(buf + len, " pid=%ld-%ld", |
45edfa58 CL |
4648 | l->min_pid, l->max_pid); |
4649 | else | |
e374d483 | 4650 | len += sprintf(buf + len, " pid=%ld", |
45edfa58 CL |
4651 | l->min_pid); |
4652 | ||
174596a0 RR |
4653 | if (num_online_cpus() > 1 && |
4654 | !cpumask_empty(to_cpumask(l->cpus)) && | |
5024c1d7 TH |
4655 | len < PAGE_SIZE - 60) |
4656 | len += scnprintf(buf + len, PAGE_SIZE - len - 50, | |
4657 | " cpus=%*pbl", | |
4658 | cpumask_pr_args(to_cpumask(l->cpus))); | |
45edfa58 | 4659 | |
62bc62a8 | 4660 | if (nr_online_nodes > 1 && !nodes_empty(l->nodes) && |
5024c1d7 TH |
4661 | len < PAGE_SIZE - 60) |
4662 | len += scnprintf(buf + len, PAGE_SIZE - len - 50, | |
4663 | " nodes=%*pbl", | |
4664 | nodemask_pr_args(&l->nodes)); | |
45edfa58 | 4665 | |
e374d483 | 4666 | len += sprintf(buf + len, "\n"); |
88a420e4 CL |
4667 | } |
4668 | ||
4669 | free_loc_track(&t); | |
bbd7d57b | 4670 | kfree(map); |
88a420e4 | 4671 | if (!t.count) |
e374d483 HH |
4672 | len += sprintf(buf, "No data\n"); |
4673 | return len; | |
88a420e4 | 4674 | } |
ab4d5ed5 | 4675 | #endif |
88a420e4 | 4676 | |
a5a84755 | 4677 | #ifdef SLUB_RESILIENCY_TEST |
c07b8183 | 4678 | static void __init resiliency_test(void) |
a5a84755 CL |
4679 | { |
4680 | u8 *p; | |
4681 | ||
95a05b42 | 4682 | BUILD_BUG_ON(KMALLOC_MIN_SIZE > 16 || KMALLOC_SHIFT_HIGH < 10); |
a5a84755 | 4683 | |
f9f58285 FF |
4684 | pr_err("SLUB resiliency testing\n"); |
4685 | pr_err("-----------------------\n"); | |
4686 | pr_err("A. Corruption after allocation\n"); | |
a5a84755 CL |
4687 | |
4688 | p = kzalloc(16, GFP_KERNEL); | |
4689 | p[16] = 0x12; | |
f9f58285 FF |
4690 | pr_err("\n1. kmalloc-16: Clobber Redzone/next pointer 0x12->0x%p\n\n", |
4691 | p + 16); | |
a5a84755 CL |
4692 | |
4693 | validate_slab_cache(kmalloc_caches[4]); | |
4694 | ||
4695 | /* Hmmm... The next two are dangerous */ | |
4696 | p = kzalloc(32, GFP_KERNEL); | |
4697 | p[32 + sizeof(void *)] = 0x34; | |
f9f58285 FF |
4698 | pr_err("\n2. kmalloc-32: Clobber next pointer/next slab 0x34 -> -0x%p\n", |
4699 | p); | |
4700 | pr_err("If allocated object is overwritten then not detectable\n\n"); | |
a5a84755 CL |
4701 | |
4702 | validate_slab_cache(kmalloc_caches[5]); | |
4703 | p = kzalloc(64, GFP_KERNEL); | |
4704 | p += 64 + (get_cycles() & 0xff) * sizeof(void *); | |
4705 | *p = 0x56; | |
f9f58285 FF |
4706 | pr_err("\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n", |
4707 | p); | |
4708 | pr_err("If allocated object is overwritten then not detectable\n\n"); | |
a5a84755 CL |
4709 | validate_slab_cache(kmalloc_caches[6]); |
4710 | ||
f9f58285 | 4711 | pr_err("\nB. Corruption after free\n"); |
a5a84755 CL |
4712 | p = kzalloc(128, GFP_KERNEL); |
4713 | kfree(p); | |
4714 | *p = 0x78; | |
f9f58285 | 4715 | pr_err("1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p); |
a5a84755 CL |
4716 | validate_slab_cache(kmalloc_caches[7]); |
4717 | ||
4718 | p = kzalloc(256, GFP_KERNEL); | |
4719 | kfree(p); | |
4720 | p[50] = 0x9a; | |
f9f58285 | 4721 | pr_err("\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n", p); |
a5a84755 CL |
4722 | validate_slab_cache(kmalloc_caches[8]); |
4723 | ||
4724 | p = kzalloc(512, GFP_KERNEL); | |
4725 | kfree(p); | |
4726 | p[512] = 0xab; | |
f9f58285 | 4727 | pr_err("\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p); |
a5a84755 CL |
4728 | validate_slab_cache(kmalloc_caches[9]); |
4729 | } | |
4730 | #else | |
4731 | #ifdef CONFIG_SYSFS | |
4732 | static void resiliency_test(void) {}; | |
4733 | #endif | |
4734 | #endif | |
4735 | ||
ab4d5ed5 | 4736 | #ifdef CONFIG_SYSFS |
81819f0f | 4737 | enum slab_stat_type { |
205ab99d CL |
4738 | SL_ALL, /* All slabs */ |
4739 | SL_PARTIAL, /* Only partially allocated slabs */ | |
4740 | SL_CPU, /* Only slabs used for cpu caches */ | |
4741 | SL_OBJECTS, /* Determine allocated objects not slabs */ | |
4742 | SL_TOTAL /* Determine object capacity not slabs */ | |
81819f0f CL |
4743 | }; |
4744 | ||
205ab99d | 4745 | #define SO_ALL (1 << SL_ALL) |
81819f0f CL |
4746 | #define SO_PARTIAL (1 << SL_PARTIAL) |
4747 | #define SO_CPU (1 << SL_CPU) | |
4748 | #define SO_OBJECTS (1 << SL_OBJECTS) | |
205ab99d | 4749 | #define SO_TOTAL (1 << SL_TOTAL) |
81819f0f | 4750 | |
1663f26d TH |
4751 | #ifdef CONFIG_MEMCG |
4752 | static bool memcg_sysfs_enabled = IS_ENABLED(CONFIG_SLUB_MEMCG_SYSFS_ON); | |
4753 | ||
4754 | static int __init setup_slub_memcg_sysfs(char *str) | |
4755 | { | |
4756 | int v; | |
4757 | ||
4758 | if (get_option(&str, &v) > 0) | |
4759 | memcg_sysfs_enabled = v; | |
4760 | ||
4761 | return 1; | |
4762 | } | |
4763 | ||
4764 | __setup("slub_memcg_sysfs=", setup_slub_memcg_sysfs); | |
4765 | #endif | |
4766 | ||
62e5c4b4 CG |
4767 | static ssize_t show_slab_objects(struct kmem_cache *s, |
4768 | char *buf, unsigned long flags) | |
81819f0f CL |
4769 | { |
4770 | unsigned long total = 0; | |
81819f0f CL |
4771 | int node; |
4772 | int x; | |
4773 | unsigned long *nodes; | |
81819f0f | 4774 | |
e35e1a97 | 4775 | nodes = kzalloc(sizeof(unsigned long) * nr_node_ids, GFP_KERNEL); |
62e5c4b4 CG |
4776 | if (!nodes) |
4777 | return -ENOMEM; | |
81819f0f | 4778 | |
205ab99d CL |
4779 | if (flags & SO_CPU) { |
4780 | int cpu; | |
81819f0f | 4781 | |
205ab99d | 4782 | for_each_possible_cpu(cpu) { |
d0e0ac97 CG |
4783 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, |
4784 | cpu); | |
ec3ab083 | 4785 | int node; |
49e22585 | 4786 | struct page *page; |
dfb4f096 | 4787 | |
4db0c3c2 | 4788 | page = READ_ONCE(c->page); |
ec3ab083 CL |
4789 | if (!page) |
4790 | continue; | |
205ab99d | 4791 | |
ec3ab083 CL |
4792 | node = page_to_nid(page); |
4793 | if (flags & SO_TOTAL) | |
4794 | x = page->objects; | |
4795 | else if (flags & SO_OBJECTS) | |
4796 | x = page->inuse; | |
4797 | else | |
4798 | x = 1; | |
49e22585 | 4799 | |
ec3ab083 CL |
4800 | total += x; |
4801 | nodes[node] += x; | |
4802 | ||
a93cf07b | 4803 | page = slub_percpu_partial_read_once(c); |
49e22585 | 4804 | if (page) { |
8afb1474 LZ |
4805 | node = page_to_nid(page); |
4806 | if (flags & SO_TOTAL) | |
4807 | WARN_ON_ONCE(1); | |
4808 | else if (flags & SO_OBJECTS) | |
4809 | WARN_ON_ONCE(1); | |
4810 | else | |
4811 | x = page->pages; | |
bc6697d8 ED |
4812 | total += x; |
4813 | nodes[node] += x; | |
49e22585 | 4814 | } |
81819f0f CL |
4815 | } |
4816 | } | |
4817 | ||
bfc8c901 | 4818 | get_online_mems(); |
ab4d5ed5 | 4819 | #ifdef CONFIG_SLUB_DEBUG |
205ab99d | 4820 | if (flags & SO_ALL) { |
fa45dc25 CL |
4821 | struct kmem_cache_node *n; |
4822 | ||
4823 | for_each_kmem_cache_node(s, node, n) { | |
205ab99d | 4824 | |
d0e0ac97 CG |
4825 | if (flags & SO_TOTAL) |
4826 | x = atomic_long_read(&n->total_objects); | |
4827 | else if (flags & SO_OBJECTS) | |
4828 | x = atomic_long_read(&n->total_objects) - | |
4829 | count_partial(n, count_free); | |
81819f0f | 4830 | else |
205ab99d | 4831 | x = atomic_long_read(&n->nr_slabs); |
81819f0f CL |
4832 | total += x; |
4833 | nodes[node] += x; | |
4834 | } | |
4835 | ||
ab4d5ed5 CL |
4836 | } else |
4837 | #endif | |
4838 | if (flags & SO_PARTIAL) { | |
fa45dc25 | 4839 | struct kmem_cache_node *n; |
81819f0f | 4840 | |
fa45dc25 | 4841 | for_each_kmem_cache_node(s, node, n) { |
205ab99d CL |
4842 | if (flags & SO_TOTAL) |
4843 | x = count_partial(n, count_total); | |
4844 | else if (flags & SO_OBJECTS) | |
4845 | x = count_partial(n, count_inuse); | |
81819f0f | 4846 | else |
205ab99d | 4847 | x = n->nr_partial; |
81819f0f CL |
4848 | total += x; |
4849 | nodes[node] += x; | |
4850 | } | |
4851 | } | |
81819f0f CL |
4852 | x = sprintf(buf, "%lu", total); |
4853 | #ifdef CONFIG_NUMA | |
fa45dc25 | 4854 | for (node = 0; node < nr_node_ids; node++) |
81819f0f CL |
4855 | if (nodes[node]) |
4856 | x += sprintf(buf + x, " N%d=%lu", | |
4857 | node, nodes[node]); | |
4858 | #endif | |
bfc8c901 | 4859 | put_online_mems(); |
81819f0f CL |
4860 | kfree(nodes); |
4861 | return x + sprintf(buf + x, "\n"); | |
4862 | } | |
4863 | ||
ab4d5ed5 | 4864 | #ifdef CONFIG_SLUB_DEBUG |
81819f0f CL |
4865 | static int any_slab_objects(struct kmem_cache *s) |
4866 | { | |
4867 | int node; | |
fa45dc25 | 4868 | struct kmem_cache_node *n; |
81819f0f | 4869 | |
fa45dc25 | 4870 | for_each_kmem_cache_node(s, node, n) |
4ea33e2d | 4871 | if (atomic_long_read(&n->total_objects)) |
81819f0f | 4872 | return 1; |
fa45dc25 | 4873 | |
81819f0f CL |
4874 | return 0; |
4875 | } | |
ab4d5ed5 | 4876 | #endif |
81819f0f CL |
4877 | |
4878 | #define to_slab_attr(n) container_of(n, struct slab_attribute, attr) | |
497888cf | 4879 | #define to_slab(n) container_of(n, struct kmem_cache, kobj) |
81819f0f CL |
4880 | |
4881 | struct slab_attribute { | |
4882 | struct attribute attr; | |
4883 | ssize_t (*show)(struct kmem_cache *s, char *buf); | |
4884 | ssize_t (*store)(struct kmem_cache *s, const char *x, size_t count); | |
4885 | }; | |
4886 | ||
4887 | #define SLAB_ATTR_RO(_name) \ | |
ab067e99 VK |
4888 | static struct slab_attribute _name##_attr = \ |
4889 | __ATTR(_name, 0400, _name##_show, NULL) | |
81819f0f CL |
4890 | |
4891 | #define SLAB_ATTR(_name) \ | |
4892 | static struct slab_attribute _name##_attr = \ | |
ab067e99 | 4893 | __ATTR(_name, 0600, _name##_show, _name##_store) |
81819f0f | 4894 | |
81819f0f CL |
4895 | static ssize_t slab_size_show(struct kmem_cache *s, char *buf) |
4896 | { | |
4897 | return sprintf(buf, "%d\n", s->size); | |
4898 | } | |
4899 | SLAB_ATTR_RO(slab_size); | |
4900 | ||
4901 | static ssize_t align_show(struct kmem_cache *s, char *buf) | |
4902 | { | |
4903 | return sprintf(buf, "%d\n", s->align); | |
4904 | } | |
4905 | SLAB_ATTR_RO(align); | |
4906 | ||
4907 | static ssize_t object_size_show(struct kmem_cache *s, char *buf) | |
4908 | { | |
3b0efdfa | 4909 | return sprintf(buf, "%d\n", s->object_size); |
81819f0f CL |
4910 | } |
4911 | SLAB_ATTR_RO(object_size); | |
4912 | ||
4913 | static ssize_t objs_per_slab_show(struct kmem_cache *s, char *buf) | |
4914 | { | |
834f3d11 | 4915 | return sprintf(buf, "%d\n", oo_objects(s->oo)); |
81819f0f CL |
4916 | } |
4917 | SLAB_ATTR_RO(objs_per_slab); | |
4918 | ||
06b285dc CL |
4919 | static ssize_t order_store(struct kmem_cache *s, |
4920 | const char *buf, size_t length) | |
4921 | { | |
0121c619 CL |
4922 | unsigned long order; |
4923 | int err; | |
4924 | ||
3dbb95f7 | 4925 | err = kstrtoul(buf, 10, &order); |
0121c619 CL |
4926 | if (err) |
4927 | return err; | |
06b285dc CL |
4928 | |
4929 | if (order > slub_max_order || order < slub_min_order) | |
4930 | return -EINVAL; | |
4931 | ||
4932 | calculate_sizes(s, order); | |
4933 | return length; | |
4934 | } | |
4935 | ||
81819f0f CL |
4936 | static ssize_t order_show(struct kmem_cache *s, char *buf) |
4937 | { | |
834f3d11 | 4938 | return sprintf(buf, "%d\n", oo_order(s->oo)); |
81819f0f | 4939 | } |
06b285dc | 4940 | SLAB_ATTR(order); |
81819f0f | 4941 | |
73d342b1 DR |
4942 | static ssize_t min_partial_show(struct kmem_cache *s, char *buf) |
4943 | { | |
4944 | return sprintf(buf, "%lu\n", s->min_partial); | |
4945 | } | |
4946 | ||
4947 | static ssize_t min_partial_store(struct kmem_cache *s, const char *buf, | |
4948 | size_t length) | |
4949 | { | |
4950 | unsigned long min; | |
4951 | int err; | |
4952 | ||
3dbb95f7 | 4953 | err = kstrtoul(buf, 10, &min); |
73d342b1 DR |
4954 | if (err) |
4955 | return err; | |
4956 | ||
c0bdb232 | 4957 | set_min_partial(s, min); |
73d342b1 DR |
4958 | return length; |
4959 | } | |
4960 | SLAB_ATTR(min_partial); | |
4961 | ||
49e22585 CL |
4962 | static ssize_t cpu_partial_show(struct kmem_cache *s, char *buf) |
4963 | { | |
e6d0e1dc | 4964 | return sprintf(buf, "%u\n", slub_cpu_partial(s)); |
49e22585 CL |
4965 | } |
4966 | ||
4967 | static ssize_t cpu_partial_store(struct kmem_cache *s, const char *buf, | |
4968 | size_t length) | |
4969 | { | |
4970 | unsigned long objects; | |
4971 | int err; | |
4972 | ||
3dbb95f7 | 4973 | err = kstrtoul(buf, 10, &objects); |
49e22585 CL |
4974 | if (err) |
4975 | return err; | |
345c905d | 4976 | if (objects && !kmem_cache_has_cpu_partial(s)) |
74ee4ef1 | 4977 | return -EINVAL; |
49e22585 | 4978 | |
e6d0e1dc | 4979 | slub_set_cpu_partial(s, objects); |
49e22585 CL |
4980 | flush_all(s); |
4981 | return length; | |
4982 | } | |
4983 | SLAB_ATTR(cpu_partial); | |
4984 | ||
81819f0f CL |
4985 | static ssize_t ctor_show(struct kmem_cache *s, char *buf) |
4986 | { | |
62c70bce JP |
4987 | if (!s->ctor) |
4988 | return 0; | |
4989 | return sprintf(buf, "%pS\n", s->ctor); | |
81819f0f CL |
4990 | } |
4991 | SLAB_ATTR_RO(ctor); | |
4992 | ||
81819f0f CL |
4993 | static ssize_t aliases_show(struct kmem_cache *s, char *buf) |
4994 | { | |
4307c14f | 4995 | return sprintf(buf, "%d\n", s->refcount < 0 ? 0 : s->refcount - 1); |
81819f0f CL |
4996 | } |
4997 | SLAB_ATTR_RO(aliases); | |
4998 | ||
81819f0f CL |
4999 | static ssize_t partial_show(struct kmem_cache *s, char *buf) |
5000 | { | |
d9acf4b7 | 5001 | return show_slab_objects(s, buf, SO_PARTIAL); |
81819f0f CL |
5002 | } |
5003 | SLAB_ATTR_RO(partial); | |
5004 | ||
5005 | static ssize_t cpu_slabs_show(struct kmem_cache *s, char *buf) | |
5006 | { | |
d9acf4b7 | 5007 | return show_slab_objects(s, buf, SO_CPU); |
81819f0f CL |
5008 | } |
5009 | SLAB_ATTR_RO(cpu_slabs); | |
5010 | ||
5011 | static ssize_t objects_show(struct kmem_cache *s, char *buf) | |
5012 | { | |
205ab99d | 5013 | return show_slab_objects(s, buf, SO_ALL|SO_OBJECTS); |
81819f0f CL |
5014 | } |
5015 | SLAB_ATTR_RO(objects); | |
5016 | ||
205ab99d CL |
5017 | static ssize_t objects_partial_show(struct kmem_cache *s, char *buf) |
5018 | { | |
5019 | return show_slab_objects(s, buf, SO_PARTIAL|SO_OBJECTS); | |
5020 | } | |
5021 | SLAB_ATTR_RO(objects_partial); | |
5022 | ||
49e22585 CL |
5023 | static ssize_t slabs_cpu_partial_show(struct kmem_cache *s, char *buf) |
5024 | { | |
5025 | int objects = 0; | |
5026 | int pages = 0; | |
5027 | int cpu; | |
5028 | int len; | |
5029 | ||
5030 | for_each_online_cpu(cpu) { | |
a93cf07b WY |
5031 | struct page *page; |
5032 | ||
5033 | page = slub_percpu_partial(per_cpu_ptr(s->cpu_slab, cpu)); | |
49e22585 CL |
5034 | |
5035 | if (page) { | |
5036 | pages += page->pages; | |
5037 | objects += page->pobjects; | |
5038 | } | |
5039 | } | |
5040 | ||
5041 | len = sprintf(buf, "%d(%d)", objects, pages); | |
5042 | ||
5043 | #ifdef CONFIG_SMP | |
5044 | for_each_online_cpu(cpu) { | |
a93cf07b WY |
5045 | struct page *page; |
5046 | ||
5047 | page = slub_percpu_partial(per_cpu_ptr(s->cpu_slab, cpu)); | |
49e22585 CL |
5048 | |
5049 | if (page && len < PAGE_SIZE - 20) | |
5050 | len += sprintf(buf + len, " C%d=%d(%d)", cpu, | |
5051 | page->pobjects, page->pages); | |
5052 | } | |
5053 | #endif | |
5054 | return len + sprintf(buf + len, "\n"); | |
5055 | } | |
5056 | SLAB_ATTR_RO(slabs_cpu_partial); | |
5057 | ||
a5a84755 CL |
5058 | static ssize_t reclaim_account_show(struct kmem_cache *s, char *buf) |
5059 | { | |
5060 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_RECLAIM_ACCOUNT)); | |
5061 | } | |
5062 | ||
5063 | static ssize_t reclaim_account_store(struct kmem_cache *s, | |
5064 | const char *buf, size_t length) | |
5065 | { | |
5066 | s->flags &= ~SLAB_RECLAIM_ACCOUNT; | |
5067 | if (buf[0] == '1') | |
5068 | s->flags |= SLAB_RECLAIM_ACCOUNT; | |
5069 | return length; | |
5070 | } | |
5071 | SLAB_ATTR(reclaim_account); | |
5072 | ||
5073 | static ssize_t hwcache_align_show(struct kmem_cache *s, char *buf) | |
5074 | { | |
5075 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_HWCACHE_ALIGN)); | |
5076 | } | |
5077 | SLAB_ATTR_RO(hwcache_align); | |
5078 | ||
5079 | #ifdef CONFIG_ZONE_DMA | |
5080 | static ssize_t cache_dma_show(struct kmem_cache *s, char *buf) | |
5081 | { | |
5082 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_CACHE_DMA)); | |
5083 | } | |
5084 | SLAB_ATTR_RO(cache_dma); | |
5085 | #endif | |
5086 | ||
5087 | static ssize_t destroy_by_rcu_show(struct kmem_cache *s, char *buf) | |
5088 | { | |
5f0d5a3a | 5089 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_TYPESAFE_BY_RCU)); |
a5a84755 CL |
5090 | } |
5091 | SLAB_ATTR_RO(destroy_by_rcu); | |
5092 | ||
ab9a0f19 LJ |
5093 | static ssize_t reserved_show(struct kmem_cache *s, char *buf) |
5094 | { | |
5095 | return sprintf(buf, "%d\n", s->reserved); | |
5096 | } | |
5097 | SLAB_ATTR_RO(reserved); | |
5098 | ||
ab4d5ed5 | 5099 | #ifdef CONFIG_SLUB_DEBUG |
a5a84755 CL |
5100 | static ssize_t slabs_show(struct kmem_cache *s, char *buf) |
5101 | { | |
5102 | return show_slab_objects(s, buf, SO_ALL); | |
5103 | } | |
5104 | SLAB_ATTR_RO(slabs); | |
5105 | ||
205ab99d CL |
5106 | static ssize_t total_objects_show(struct kmem_cache *s, char *buf) |
5107 | { | |
5108 | return show_slab_objects(s, buf, SO_ALL|SO_TOTAL); | |
5109 | } | |
5110 | SLAB_ATTR_RO(total_objects); | |
5111 | ||
81819f0f CL |
5112 | static ssize_t sanity_checks_show(struct kmem_cache *s, char *buf) |
5113 | { | |
becfda68 | 5114 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_CONSISTENCY_CHECKS)); |
81819f0f CL |
5115 | } |
5116 | ||
5117 | static ssize_t sanity_checks_store(struct kmem_cache *s, | |
5118 | const char *buf, size_t length) | |
5119 | { | |
becfda68 | 5120 | s->flags &= ~SLAB_CONSISTENCY_CHECKS; |
b789ef51 CL |
5121 | if (buf[0] == '1') { |
5122 | s->flags &= ~__CMPXCHG_DOUBLE; | |
becfda68 | 5123 | s->flags |= SLAB_CONSISTENCY_CHECKS; |
b789ef51 | 5124 | } |
81819f0f CL |
5125 | return length; |
5126 | } | |
5127 | SLAB_ATTR(sanity_checks); | |
5128 | ||
5129 | static ssize_t trace_show(struct kmem_cache *s, char *buf) | |
5130 | { | |
5131 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_TRACE)); | |
5132 | } | |
5133 | ||
5134 | static ssize_t trace_store(struct kmem_cache *s, const char *buf, | |
5135 | size_t length) | |
5136 | { | |
c9e16131 CL |
5137 | /* |
5138 | * Tracing a merged cache is going to give confusing results | |
5139 | * as well as cause other issues like converting a mergeable | |
5140 | * cache into an umergeable one. | |
5141 | */ | |
5142 | if (s->refcount > 1) | |
5143 | return -EINVAL; | |
5144 | ||
81819f0f | 5145 | s->flags &= ~SLAB_TRACE; |
b789ef51 CL |
5146 | if (buf[0] == '1') { |
5147 | s->flags &= ~__CMPXCHG_DOUBLE; | |
81819f0f | 5148 | s->flags |= SLAB_TRACE; |
b789ef51 | 5149 | } |
81819f0f CL |
5150 | return length; |
5151 | } | |
5152 | SLAB_ATTR(trace); | |
5153 | ||
81819f0f CL |
5154 | static ssize_t red_zone_show(struct kmem_cache *s, char *buf) |
5155 | { | |
5156 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_RED_ZONE)); | |
5157 | } | |
5158 | ||
5159 | static ssize_t red_zone_store(struct kmem_cache *s, | |
5160 | const char *buf, size_t length) | |
5161 | { | |
5162 | if (any_slab_objects(s)) | |
5163 | return -EBUSY; | |
5164 | ||
5165 | s->flags &= ~SLAB_RED_ZONE; | |
b789ef51 | 5166 | if (buf[0] == '1') { |
81819f0f | 5167 | s->flags |= SLAB_RED_ZONE; |
b789ef51 | 5168 | } |
06b285dc | 5169 | calculate_sizes(s, -1); |
81819f0f CL |
5170 | return length; |
5171 | } | |
5172 | SLAB_ATTR(red_zone); | |
5173 | ||
5174 | static ssize_t poison_show(struct kmem_cache *s, char *buf) | |
5175 | { | |
5176 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_POISON)); | |
5177 | } | |
5178 | ||
5179 | static ssize_t poison_store(struct kmem_cache *s, | |
5180 | const char *buf, size_t length) | |
5181 | { | |
5182 | if (any_slab_objects(s)) | |
5183 | return -EBUSY; | |
5184 | ||
5185 | s->flags &= ~SLAB_POISON; | |
b789ef51 | 5186 | if (buf[0] == '1') { |
81819f0f | 5187 | s->flags |= SLAB_POISON; |
b789ef51 | 5188 | } |
06b285dc | 5189 | calculate_sizes(s, -1); |
81819f0f CL |
5190 | return length; |
5191 | } | |
5192 | SLAB_ATTR(poison); | |
5193 | ||
5194 | static ssize_t store_user_show(struct kmem_cache *s, char *buf) | |
5195 | { | |
5196 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_STORE_USER)); | |
5197 | } | |
5198 | ||
5199 | static ssize_t store_user_store(struct kmem_cache *s, | |
5200 | const char *buf, size_t length) | |
5201 | { | |
5202 | if (any_slab_objects(s)) | |
5203 | return -EBUSY; | |
5204 | ||
5205 | s->flags &= ~SLAB_STORE_USER; | |
b789ef51 CL |
5206 | if (buf[0] == '1') { |
5207 | s->flags &= ~__CMPXCHG_DOUBLE; | |
81819f0f | 5208 | s->flags |= SLAB_STORE_USER; |
b789ef51 | 5209 | } |
06b285dc | 5210 | calculate_sizes(s, -1); |
81819f0f CL |
5211 | return length; |
5212 | } | |
5213 | SLAB_ATTR(store_user); | |
5214 | ||
53e15af0 CL |
5215 | static ssize_t validate_show(struct kmem_cache *s, char *buf) |
5216 | { | |
5217 | return 0; | |
5218 | } | |
5219 | ||
5220 | static ssize_t validate_store(struct kmem_cache *s, | |
5221 | const char *buf, size_t length) | |
5222 | { | |
434e245d CL |
5223 | int ret = -EINVAL; |
5224 | ||
5225 | if (buf[0] == '1') { | |
5226 | ret = validate_slab_cache(s); | |
5227 | if (ret >= 0) | |
5228 | ret = length; | |
5229 | } | |
5230 | return ret; | |
53e15af0 CL |
5231 | } |
5232 | SLAB_ATTR(validate); | |
a5a84755 CL |
5233 | |
5234 | static ssize_t alloc_calls_show(struct kmem_cache *s, char *buf) | |
5235 | { | |
5236 | if (!(s->flags & SLAB_STORE_USER)) | |
5237 | return -ENOSYS; | |
5238 | return list_locations(s, buf, TRACK_ALLOC); | |
5239 | } | |
5240 | SLAB_ATTR_RO(alloc_calls); | |
5241 | ||
5242 | static ssize_t free_calls_show(struct kmem_cache *s, char *buf) | |
5243 | { | |
5244 | if (!(s->flags & SLAB_STORE_USER)) | |
5245 | return -ENOSYS; | |
5246 | return list_locations(s, buf, TRACK_FREE); | |
5247 | } | |
5248 | SLAB_ATTR_RO(free_calls); | |
5249 | #endif /* CONFIG_SLUB_DEBUG */ | |
5250 | ||
5251 | #ifdef CONFIG_FAILSLAB | |
5252 | static ssize_t failslab_show(struct kmem_cache *s, char *buf) | |
5253 | { | |
5254 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_FAILSLAB)); | |
5255 | } | |
5256 | ||
5257 | static ssize_t failslab_store(struct kmem_cache *s, const char *buf, | |
5258 | size_t length) | |
5259 | { | |
c9e16131 CL |
5260 | if (s->refcount > 1) |
5261 | return -EINVAL; | |
5262 | ||
a5a84755 CL |
5263 | s->flags &= ~SLAB_FAILSLAB; |
5264 | if (buf[0] == '1') | |
5265 | s->flags |= SLAB_FAILSLAB; | |
5266 | return length; | |
5267 | } | |
5268 | SLAB_ATTR(failslab); | |
ab4d5ed5 | 5269 | #endif |
53e15af0 | 5270 | |
2086d26a CL |
5271 | static ssize_t shrink_show(struct kmem_cache *s, char *buf) |
5272 | { | |
5273 | return 0; | |
5274 | } | |
5275 | ||
5276 | static ssize_t shrink_store(struct kmem_cache *s, | |
5277 | const char *buf, size_t length) | |
5278 | { | |
832f37f5 VD |
5279 | if (buf[0] == '1') |
5280 | kmem_cache_shrink(s); | |
5281 | else | |
2086d26a CL |
5282 | return -EINVAL; |
5283 | return length; | |
5284 | } | |
5285 | SLAB_ATTR(shrink); | |
5286 | ||
81819f0f | 5287 | #ifdef CONFIG_NUMA |
9824601e | 5288 | static ssize_t remote_node_defrag_ratio_show(struct kmem_cache *s, char *buf) |
81819f0f | 5289 | { |
9824601e | 5290 | return sprintf(buf, "%d\n", s->remote_node_defrag_ratio / 10); |
81819f0f CL |
5291 | } |
5292 | ||
9824601e | 5293 | static ssize_t remote_node_defrag_ratio_store(struct kmem_cache *s, |
81819f0f CL |
5294 | const char *buf, size_t length) |
5295 | { | |
0121c619 CL |
5296 | unsigned long ratio; |
5297 | int err; | |
5298 | ||
3dbb95f7 | 5299 | err = kstrtoul(buf, 10, &ratio); |
0121c619 CL |
5300 | if (err) |
5301 | return err; | |
5302 | ||
e2cb96b7 | 5303 | if (ratio <= 100) |
0121c619 | 5304 | s->remote_node_defrag_ratio = ratio * 10; |
81819f0f | 5305 | |
81819f0f CL |
5306 | return length; |
5307 | } | |
9824601e | 5308 | SLAB_ATTR(remote_node_defrag_ratio); |
81819f0f CL |
5309 | #endif |
5310 | ||
8ff12cfc | 5311 | #ifdef CONFIG_SLUB_STATS |
8ff12cfc CL |
5312 | static int show_stat(struct kmem_cache *s, char *buf, enum stat_item si) |
5313 | { | |
5314 | unsigned long sum = 0; | |
5315 | int cpu; | |
5316 | int len; | |
5317 | int *data = kmalloc(nr_cpu_ids * sizeof(int), GFP_KERNEL); | |
5318 | ||
5319 | if (!data) | |
5320 | return -ENOMEM; | |
5321 | ||
5322 | for_each_online_cpu(cpu) { | |
9dfc6e68 | 5323 | unsigned x = per_cpu_ptr(s->cpu_slab, cpu)->stat[si]; |
8ff12cfc CL |
5324 | |
5325 | data[cpu] = x; | |
5326 | sum += x; | |
5327 | } | |
5328 | ||
5329 | len = sprintf(buf, "%lu", sum); | |
5330 | ||
50ef37b9 | 5331 | #ifdef CONFIG_SMP |
8ff12cfc CL |
5332 | for_each_online_cpu(cpu) { |
5333 | if (data[cpu] && len < PAGE_SIZE - 20) | |
50ef37b9 | 5334 | len += sprintf(buf + len, " C%d=%u", cpu, data[cpu]); |
8ff12cfc | 5335 | } |
50ef37b9 | 5336 | #endif |
8ff12cfc CL |
5337 | kfree(data); |
5338 | return len + sprintf(buf + len, "\n"); | |
5339 | } | |
5340 | ||
78eb00cc DR |
5341 | static void clear_stat(struct kmem_cache *s, enum stat_item si) |
5342 | { | |
5343 | int cpu; | |
5344 | ||
5345 | for_each_online_cpu(cpu) | |
9dfc6e68 | 5346 | per_cpu_ptr(s->cpu_slab, cpu)->stat[si] = 0; |
78eb00cc DR |
5347 | } |
5348 | ||
8ff12cfc CL |
5349 | #define STAT_ATTR(si, text) \ |
5350 | static ssize_t text##_show(struct kmem_cache *s, char *buf) \ | |
5351 | { \ | |
5352 | return show_stat(s, buf, si); \ | |
5353 | } \ | |
78eb00cc DR |
5354 | static ssize_t text##_store(struct kmem_cache *s, \ |
5355 | const char *buf, size_t length) \ | |
5356 | { \ | |
5357 | if (buf[0] != '0') \ | |
5358 | return -EINVAL; \ | |
5359 | clear_stat(s, si); \ | |
5360 | return length; \ | |
5361 | } \ | |
5362 | SLAB_ATTR(text); \ | |
8ff12cfc CL |
5363 | |
5364 | STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath); | |
5365 | STAT_ATTR(ALLOC_SLOWPATH, alloc_slowpath); | |
5366 | STAT_ATTR(FREE_FASTPATH, free_fastpath); | |
5367 | STAT_ATTR(FREE_SLOWPATH, free_slowpath); | |
5368 | STAT_ATTR(FREE_FROZEN, free_frozen); | |
5369 | STAT_ATTR(FREE_ADD_PARTIAL, free_add_partial); | |
5370 | STAT_ATTR(FREE_REMOVE_PARTIAL, free_remove_partial); | |
5371 | STAT_ATTR(ALLOC_FROM_PARTIAL, alloc_from_partial); | |
5372 | STAT_ATTR(ALLOC_SLAB, alloc_slab); | |
5373 | STAT_ATTR(ALLOC_REFILL, alloc_refill); | |
e36a2652 | 5374 | STAT_ATTR(ALLOC_NODE_MISMATCH, alloc_node_mismatch); |
8ff12cfc CL |
5375 | STAT_ATTR(FREE_SLAB, free_slab); |
5376 | STAT_ATTR(CPUSLAB_FLUSH, cpuslab_flush); | |
5377 | STAT_ATTR(DEACTIVATE_FULL, deactivate_full); | |
5378 | STAT_ATTR(DEACTIVATE_EMPTY, deactivate_empty); | |
5379 | STAT_ATTR(DEACTIVATE_TO_HEAD, deactivate_to_head); | |
5380 | STAT_ATTR(DEACTIVATE_TO_TAIL, deactivate_to_tail); | |
5381 | STAT_ATTR(DEACTIVATE_REMOTE_FREES, deactivate_remote_frees); | |
03e404af | 5382 | STAT_ATTR(DEACTIVATE_BYPASS, deactivate_bypass); |
65c3376a | 5383 | STAT_ATTR(ORDER_FALLBACK, order_fallback); |
b789ef51 CL |
5384 | STAT_ATTR(CMPXCHG_DOUBLE_CPU_FAIL, cmpxchg_double_cpu_fail); |
5385 | STAT_ATTR(CMPXCHG_DOUBLE_FAIL, cmpxchg_double_fail); | |
49e22585 CL |
5386 | STAT_ATTR(CPU_PARTIAL_ALLOC, cpu_partial_alloc); |
5387 | STAT_ATTR(CPU_PARTIAL_FREE, cpu_partial_free); | |
8028dcea AS |
5388 | STAT_ATTR(CPU_PARTIAL_NODE, cpu_partial_node); |
5389 | STAT_ATTR(CPU_PARTIAL_DRAIN, cpu_partial_drain); | |
8ff12cfc CL |
5390 | #endif |
5391 | ||
06428780 | 5392 | static struct attribute *slab_attrs[] = { |
81819f0f CL |
5393 | &slab_size_attr.attr, |
5394 | &object_size_attr.attr, | |
5395 | &objs_per_slab_attr.attr, | |
5396 | &order_attr.attr, | |
73d342b1 | 5397 | &min_partial_attr.attr, |
49e22585 | 5398 | &cpu_partial_attr.attr, |
81819f0f | 5399 | &objects_attr.attr, |
205ab99d | 5400 | &objects_partial_attr.attr, |
81819f0f CL |
5401 | &partial_attr.attr, |
5402 | &cpu_slabs_attr.attr, | |
5403 | &ctor_attr.attr, | |
81819f0f CL |
5404 | &aliases_attr.attr, |
5405 | &align_attr.attr, | |
81819f0f CL |
5406 | &hwcache_align_attr.attr, |
5407 | &reclaim_account_attr.attr, | |
5408 | &destroy_by_rcu_attr.attr, | |
a5a84755 | 5409 | &shrink_attr.attr, |
ab9a0f19 | 5410 | &reserved_attr.attr, |
49e22585 | 5411 | &slabs_cpu_partial_attr.attr, |
ab4d5ed5 | 5412 | #ifdef CONFIG_SLUB_DEBUG |
a5a84755 CL |
5413 | &total_objects_attr.attr, |
5414 | &slabs_attr.attr, | |
5415 | &sanity_checks_attr.attr, | |
5416 | &trace_attr.attr, | |
81819f0f CL |
5417 | &red_zone_attr.attr, |
5418 | &poison_attr.attr, | |
5419 | &store_user_attr.attr, | |
53e15af0 | 5420 | &validate_attr.attr, |
88a420e4 CL |
5421 | &alloc_calls_attr.attr, |
5422 | &free_calls_attr.attr, | |
ab4d5ed5 | 5423 | #endif |
81819f0f CL |
5424 | #ifdef CONFIG_ZONE_DMA |
5425 | &cache_dma_attr.attr, | |
5426 | #endif | |
5427 | #ifdef CONFIG_NUMA | |
9824601e | 5428 | &remote_node_defrag_ratio_attr.attr, |
8ff12cfc CL |
5429 | #endif |
5430 | #ifdef CONFIG_SLUB_STATS | |
5431 | &alloc_fastpath_attr.attr, | |
5432 | &alloc_slowpath_attr.attr, | |
5433 | &free_fastpath_attr.attr, | |
5434 | &free_slowpath_attr.attr, | |
5435 | &free_frozen_attr.attr, | |
5436 | &free_add_partial_attr.attr, | |
5437 | &free_remove_partial_attr.attr, | |
5438 | &alloc_from_partial_attr.attr, | |
5439 | &alloc_slab_attr.attr, | |
5440 | &alloc_refill_attr.attr, | |
e36a2652 | 5441 | &alloc_node_mismatch_attr.attr, |
8ff12cfc CL |
5442 | &free_slab_attr.attr, |
5443 | &cpuslab_flush_attr.attr, | |
5444 | &deactivate_full_attr.attr, | |
5445 | &deactivate_empty_attr.attr, | |
5446 | &deactivate_to_head_attr.attr, | |
5447 | &deactivate_to_tail_attr.attr, | |
5448 | &deactivate_remote_frees_attr.attr, | |
03e404af | 5449 | &deactivate_bypass_attr.attr, |
65c3376a | 5450 | &order_fallback_attr.attr, |
b789ef51 CL |
5451 | &cmpxchg_double_fail_attr.attr, |
5452 | &cmpxchg_double_cpu_fail_attr.attr, | |
49e22585 CL |
5453 | &cpu_partial_alloc_attr.attr, |
5454 | &cpu_partial_free_attr.attr, | |
8028dcea AS |
5455 | &cpu_partial_node_attr.attr, |
5456 | &cpu_partial_drain_attr.attr, | |
81819f0f | 5457 | #endif |
4c13dd3b DM |
5458 | #ifdef CONFIG_FAILSLAB |
5459 | &failslab_attr.attr, | |
5460 | #endif | |
5461 | ||
81819f0f CL |
5462 | NULL |
5463 | }; | |
5464 | ||
1fdaaa23 | 5465 | static const struct attribute_group slab_attr_group = { |
81819f0f CL |
5466 | .attrs = slab_attrs, |
5467 | }; | |
5468 | ||
5469 | static ssize_t slab_attr_show(struct kobject *kobj, | |
5470 | struct attribute *attr, | |
5471 | char *buf) | |
5472 | { | |
5473 | struct slab_attribute *attribute; | |
5474 | struct kmem_cache *s; | |
5475 | int err; | |
5476 | ||
5477 | attribute = to_slab_attr(attr); | |
5478 | s = to_slab(kobj); | |
5479 | ||
5480 | if (!attribute->show) | |
5481 | return -EIO; | |
5482 | ||
5483 | err = attribute->show(s, buf); | |
5484 | ||
5485 | return err; | |
5486 | } | |
5487 | ||
5488 | static ssize_t slab_attr_store(struct kobject *kobj, | |
5489 | struct attribute *attr, | |
5490 | const char *buf, size_t len) | |
5491 | { | |
5492 | struct slab_attribute *attribute; | |
5493 | struct kmem_cache *s; | |
5494 | int err; | |
5495 | ||
5496 | attribute = to_slab_attr(attr); | |
5497 | s = to_slab(kobj); | |
5498 | ||
5499 | if (!attribute->store) | |
5500 | return -EIO; | |
5501 | ||
5502 | err = attribute->store(s, buf, len); | |
127424c8 | 5503 | #ifdef CONFIG_MEMCG |
107dab5c | 5504 | if (slab_state >= FULL && err >= 0 && is_root_cache(s)) { |
426589f5 | 5505 | struct kmem_cache *c; |
81819f0f | 5506 | |
107dab5c GC |
5507 | mutex_lock(&slab_mutex); |
5508 | if (s->max_attr_size < len) | |
5509 | s->max_attr_size = len; | |
5510 | ||
ebe945c2 GC |
5511 | /* |
5512 | * This is a best effort propagation, so this function's return | |
5513 | * value will be determined by the parent cache only. This is | |
5514 | * basically because not all attributes will have a well | |
5515 | * defined semantics for rollbacks - most of the actions will | |
5516 | * have permanent effects. | |
5517 | * | |
5518 | * Returning the error value of any of the children that fail | |
5519 | * is not 100 % defined, in the sense that users seeing the | |
5520 | * error code won't be able to know anything about the state of | |
5521 | * the cache. | |
5522 | * | |
5523 | * Only returning the error code for the parent cache at least | |
5524 | * has well defined semantics. The cache being written to | |
5525 | * directly either failed or succeeded, in which case we loop | |
5526 | * through the descendants with best-effort propagation. | |
5527 | */ | |
426589f5 VD |
5528 | for_each_memcg_cache(c, s) |
5529 | attribute->store(c, buf, len); | |
107dab5c GC |
5530 | mutex_unlock(&slab_mutex); |
5531 | } | |
5532 | #endif | |
81819f0f CL |
5533 | return err; |
5534 | } | |
5535 | ||
107dab5c GC |
5536 | static void memcg_propagate_slab_attrs(struct kmem_cache *s) |
5537 | { | |
127424c8 | 5538 | #ifdef CONFIG_MEMCG |
107dab5c GC |
5539 | int i; |
5540 | char *buffer = NULL; | |
93030d83 | 5541 | struct kmem_cache *root_cache; |
107dab5c | 5542 | |
93030d83 | 5543 | if (is_root_cache(s)) |
107dab5c GC |
5544 | return; |
5545 | ||
f7ce3190 | 5546 | root_cache = s->memcg_params.root_cache; |
93030d83 | 5547 | |
107dab5c GC |
5548 | /* |
5549 | * This mean this cache had no attribute written. Therefore, no point | |
5550 | * in copying default values around | |
5551 | */ | |
93030d83 | 5552 | if (!root_cache->max_attr_size) |
107dab5c GC |
5553 | return; |
5554 | ||
5555 | for (i = 0; i < ARRAY_SIZE(slab_attrs); i++) { | |
5556 | char mbuf[64]; | |
5557 | char *buf; | |
5558 | struct slab_attribute *attr = to_slab_attr(slab_attrs[i]); | |
478fe303 | 5559 | ssize_t len; |
107dab5c GC |
5560 | |
5561 | if (!attr || !attr->store || !attr->show) | |
5562 | continue; | |
5563 | ||
5564 | /* | |
5565 | * It is really bad that we have to allocate here, so we will | |
5566 | * do it only as a fallback. If we actually allocate, though, | |
5567 | * we can just use the allocated buffer until the end. | |
5568 | * | |
5569 | * Most of the slub attributes will tend to be very small in | |
5570 | * size, but sysfs allows buffers up to a page, so they can | |
5571 | * theoretically happen. | |
5572 | */ | |
5573 | if (buffer) | |
5574 | buf = buffer; | |
93030d83 | 5575 | else if (root_cache->max_attr_size < ARRAY_SIZE(mbuf)) |
107dab5c GC |
5576 | buf = mbuf; |
5577 | else { | |
5578 | buffer = (char *) get_zeroed_page(GFP_KERNEL); | |
5579 | if (WARN_ON(!buffer)) | |
5580 | continue; | |
5581 | buf = buffer; | |
5582 | } | |
5583 | ||
478fe303 TG |
5584 | len = attr->show(root_cache, buf); |
5585 | if (len > 0) | |
5586 | attr->store(s, buf, len); | |
107dab5c GC |
5587 | } |
5588 | ||
5589 | if (buffer) | |
5590 | free_page((unsigned long)buffer); | |
5591 | #endif | |
5592 | } | |
5593 | ||
41a21285 CL |
5594 | static void kmem_cache_release(struct kobject *k) |
5595 | { | |
5596 | slab_kmem_cache_release(to_slab(k)); | |
5597 | } | |
5598 | ||
52cf25d0 | 5599 | static const struct sysfs_ops slab_sysfs_ops = { |
81819f0f CL |
5600 | .show = slab_attr_show, |
5601 | .store = slab_attr_store, | |
5602 | }; | |
5603 | ||
5604 | static struct kobj_type slab_ktype = { | |
5605 | .sysfs_ops = &slab_sysfs_ops, | |
41a21285 | 5606 | .release = kmem_cache_release, |
81819f0f CL |
5607 | }; |
5608 | ||
5609 | static int uevent_filter(struct kset *kset, struct kobject *kobj) | |
5610 | { | |
5611 | struct kobj_type *ktype = get_ktype(kobj); | |
5612 | ||
5613 | if (ktype == &slab_ktype) | |
5614 | return 1; | |
5615 | return 0; | |
5616 | } | |
5617 | ||
9cd43611 | 5618 | static const struct kset_uevent_ops slab_uevent_ops = { |
81819f0f CL |
5619 | .filter = uevent_filter, |
5620 | }; | |
5621 | ||
27c3a314 | 5622 | static struct kset *slab_kset; |
81819f0f | 5623 | |
9a41707b VD |
5624 | static inline struct kset *cache_kset(struct kmem_cache *s) |
5625 | { | |
127424c8 | 5626 | #ifdef CONFIG_MEMCG |
9a41707b | 5627 | if (!is_root_cache(s)) |
f7ce3190 | 5628 | return s->memcg_params.root_cache->memcg_kset; |
9a41707b VD |
5629 | #endif |
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'; | |
5656 | if (s->flags & SLAB_RECLAIM_ACCOUNT) | |
5657 | *p++ = 'a'; | |
becfda68 | 5658 | if (s->flags & SLAB_CONSISTENCY_CHECKS) |
81819f0f | 5659 | *p++ = 'F'; |
5a896d9e VN |
5660 | if (!(s->flags & SLAB_NOTRACK)) |
5661 | *p++ = 't'; | |
230e9fc2 VD |
5662 | if (s->flags & SLAB_ACCOUNT) |
5663 | *p++ = 'A'; | |
81819f0f CL |
5664 | if (p != name + 1) |
5665 | *p++ = '-'; | |
5666 | p += sprintf(p, "%07d", s->size); | |
2633d7a0 | 5667 | |
81819f0f CL |
5668 | BUG_ON(p > name + ID_STR_LENGTH - 1); |
5669 | return name; | |
5670 | } | |
5671 | ||
3b7b3140 TH |
5672 | static void sysfs_slab_remove_workfn(struct work_struct *work) |
5673 | { | |
5674 | struct kmem_cache *s = | |
5675 | container_of(work, struct kmem_cache, kobj_remove_work); | |
5676 | ||
5677 | if (!s->kobj.state_in_sysfs) | |
5678 | /* | |
5679 | * For a memcg cache, this may be called during | |
5680 | * deactivation and again on shutdown. Remove only once. | |
5681 | * A cache is never shut down before deactivation is | |
5682 | * complete, so no need to worry about synchronization. | |
5683 | */ | |
f6ba4880 | 5684 | goto out; |
3b7b3140 TH |
5685 | |
5686 | #ifdef CONFIG_MEMCG | |
5687 | kset_unregister(s->memcg_kset); | |
5688 | #endif | |
5689 | kobject_uevent(&s->kobj, KOBJ_REMOVE); | |
5690 | kobject_del(&s->kobj); | |
f6ba4880 | 5691 | out: |
3b7b3140 TH |
5692 | kobject_put(&s->kobj); |
5693 | } | |
5694 | ||
81819f0f CL |
5695 | static int sysfs_slab_add(struct kmem_cache *s) |
5696 | { | |
5697 | int err; | |
5698 | const char *name; | |
1663f26d | 5699 | struct kset *kset = cache_kset(s); |
45530c44 | 5700 | int unmergeable = slab_unmergeable(s); |
81819f0f | 5701 | |
3b7b3140 TH |
5702 | INIT_WORK(&s->kobj_remove_work, sysfs_slab_remove_workfn); |
5703 | ||
1663f26d TH |
5704 | if (!kset) { |
5705 | kobject_init(&s->kobj, &slab_ktype); | |
5706 | return 0; | |
5707 | } | |
5708 | ||
81819f0f CL |
5709 | if (unmergeable) { |
5710 | /* | |
5711 | * Slabcache can never be merged so we can use the name proper. | |
5712 | * This is typically the case for debug situations. In that | |
5713 | * case we can catch duplicate names easily. | |
5714 | */ | |
27c3a314 | 5715 | sysfs_remove_link(&slab_kset->kobj, s->name); |
81819f0f CL |
5716 | name = s->name; |
5717 | } else { | |
5718 | /* | |
5719 | * Create a unique name for the slab as a target | |
5720 | * for the symlinks. | |
5721 | */ | |
5722 | name = create_unique_id(s); | |
5723 | } | |
5724 | ||
1663f26d | 5725 | s->kobj.kset = kset; |
26e4f205 | 5726 | err = kobject_init_and_add(&s->kobj, &slab_ktype, NULL, "%s", name); |
54b6a731 | 5727 | if (err) |
80da026a | 5728 | goto out; |
81819f0f CL |
5729 | |
5730 | err = sysfs_create_group(&s->kobj, &slab_attr_group); | |
54b6a731 DJ |
5731 | if (err) |
5732 | goto out_del_kobj; | |
9a41707b | 5733 | |
127424c8 | 5734 | #ifdef CONFIG_MEMCG |
1663f26d | 5735 | if (is_root_cache(s) && memcg_sysfs_enabled) { |
9a41707b VD |
5736 | s->memcg_kset = kset_create_and_add("cgroup", NULL, &s->kobj); |
5737 | if (!s->memcg_kset) { | |
54b6a731 DJ |
5738 | err = -ENOMEM; |
5739 | goto out_del_kobj; | |
9a41707b VD |
5740 | } |
5741 | } | |
5742 | #endif | |
5743 | ||
81819f0f CL |
5744 | kobject_uevent(&s->kobj, KOBJ_ADD); |
5745 | if (!unmergeable) { | |
5746 | /* Setup first alias */ | |
5747 | sysfs_slab_alias(s, s->name); | |
81819f0f | 5748 | } |
54b6a731 DJ |
5749 | out: |
5750 | if (!unmergeable) | |
5751 | kfree(name); | |
5752 | return err; | |
5753 | out_del_kobj: | |
5754 | kobject_del(&s->kobj); | |
54b6a731 | 5755 | goto out; |
81819f0f CL |
5756 | } |
5757 | ||
bf5eb3de | 5758 | static void sysfs_slab_remove(struct kmem_cache *s) |
81819f0f | 5759 | { |
97d06609 | 5760 | if (slab_state < FULL) |
2bce6485 CL |
5761 | /* |
5762 | * Sysfs has not been setup yet so no need to remove the | |
5763 | * cache from sysfs. | |
5764 | */ | |
5765 | return; | |
5766 | ||
3b7b3140 TH |
5767 | kobject_get(&s->kobj); |
5768 | schedule_work(&s->kobj_remove_work); | |
bf5eb3de TH |
5769 | } |
5770 | ||
5771 | void sysfs_slab_release(struct kmem_cache *s) | |
5772 | { | |
5773 | if (slab_state >= FULL) | |
5774 | kobject_put(&s->kobj); | |
81819f0f CL |
5775 | } |
5776 | ||
5777 | /* | |
5778 | * Need to buffer aliases during bootup until sysfs becomes | |
9f6c708e | 5779 | * available lest we lose that information. |
81819f0f CL |
5780 | */ |
5781 | struct saved_alias { | |
5782 | struct kmem_cache *s; | |
5783 | const char *name; | |
5784 | struct saved_alias *next; | |
5785 | }; | |
5786 | ||
5af328a5 | 5787 | static struct saved_alias *alias_list; |
81819f0f CL |
5788 | |
5789 | static int sysfs_slab_alias(struct kmem_cache *s, const char *name) | |
5790 | { | |
5791 | struct saved_alias *al; | |
5792 | ||
97d06609 | 5793 | if (slab_state == FULL) { |
81819f0f CL |
5794 | /* |
5795 | * If we have a leftover link then remove it. | |
5796 | */ | |
27c3a314 GKH |
5797 | sysfs_remove_link(&slab_kset->kobj, name); |
5798 | return sysfs_create_link(&slab_kset->kobj, &s->kobj, name); | |
81819f0f CL |
5799 | } |
5800 | ||
5801 | al = kmalloc(sizeof(struct saved_alias), GFP_KERNEL); | |
5802 | if (!al) | |
5803 | return -ENOMEM; | |
5804 | ||
5805 | al->s = s; | |
5806 | al->name = name; | |
5807 | al->next = alias_list; | |
5808 | alias_list = al; | |
5809 | return 0; | |
5810 | } | |
5811 | ||
5812 | static int __init slab_sysfs_init(void) | |
5813 | { | |
5b95a4ac | 5814 | struct kmem_cache *s; |
81819f0f CL |
5815 | int err; |
5816 | ||
18004c5d | 5817 | mutex_lock(&slab_mutex); |
2bce6485 | 5818 | |
0ff21e46 | 5819 | slab_kset = kset_create_and_add("slab", &slab_uevent_ops, kernel_kobj); |
27c3a314 | 5820 | if (!slab_kset) { |
18004c5d | 5821 | mutex_unlock(&slab_mutex); |
f9f58285 | 5822 | pr_err("Cannot register slab subsystem.\n"); |
81819f0f CL |
5823 | return -ENOSYS; |
5824 | } | |
5825 | ||
97d06609 | 5826 | slab_state = FULL; |
26a7bd03 | 5827 | |
5b95a4ac | 5828 | list_for_each_entry(s, &slab_caches, list) { |
26a7bd03 | 5829 | err = sysfs_slab_add(s); |
5d540fb7 | 5830 | if (err) |
f9f58285 FF |
5831 | pr_err("SLUB: Unable to add boot slab %s to sysfs\n", |
5832 | s->name); | |
26a7bd03 | 5833 | } |
81819f0f CL |
5834 | |
5835 | while (alias_list) { | |
5836 | struct saved_alias *al = alias_list; | |
5837 | ||
5838 | alias_list = alias_list->next; | |
5839 | err = sysfs_slab_alias(al->s, al->name); | |
5d540fb7 | 5840 | if (err) |
f9f58285 FF |
5841 | pr_err("SLUB: Unable to add boot slab alias %s to sysfs\n", |
5842 | al->name); | |
81819f0f CL |
5843 | kfree(al); |
5844 | } | |
5845 | ||
18004c5d | 5846 | mutex_unlock(&slab_mutex); |
81819f0f CL |
5847 | resiliency_test(); |
5848 | return 0; | |
5849 | } | |
5850 | ||
5851 | __initcall(slab_sysfs_init); | |
ab4d5ed5 | 5852 | #endif /* CONFIG_SYSFS */ |
57ed3eda PE |
5853 | |
5854 | /* | |
5855 | * The /proc/slabinfo ABI | |
5856 | */ | |
5b365771 | 5857 | #ifdef CONFIG_SLUB_DEBUG |
0d7561c6 | 5858 | void get_slabinfo(struct kmem_cache *s, struct slabinfo *sinfo) |
57ed3eda | 5859 | { |
57ed3eda | 5860 | unsigned long nr_slabs = 0; |
205ab99d CL |
5861 | unsigned long nr_objs = 0; |
5862 | unsigned long nr_free = 0; | |
57ed3eda | 5863 | int node; |
fa45dc25 | 5864 | struct kmem_cache_node *n; |
57ed3eda | 5865 | |
fa45dc25 | 5866 | for_each_kmem_cache_node(s, node, n) { |
c17fd13e WL |
5867 | nr_slabs += node_nr_slabs(n); |
5868 | nr_objs += node_nr_objs(n); | |
205ab99d | 5869 | nr_free += count_partial(n, count_free); |
57ed3eda PE |
5870 | } |
5871 | ||
0d7561c6 GC |
5872 | sinfo->active_objs = nr_objs - nr_free; |
5873 | sinfo->num_objs = nr_objs; | |
5874 | sinfo->active_slabs = nr_slabs; | |
5875 | sinfo->num_slabs = nr_slabs; | |
5876 | sinfo->objects_per_slab = oo_objects(s->oo); | |
5877 | sinfo->cache_order = oo_order(s->oo); | |
57ed3eda PE |
5878 | } |
5879 | ||
0d7561c6 | 5880 | void slabinfo_show_stats(struct seq_file *m, struct kmem_cache *s) |
7b3c3a50 | 5881 | { |
7b3c3a50 AD |
5882 | } |
5883 | ||
b7454ad3 GC |
5884 | ssize_t slabinfo_write(struct file *file, const char __user *buffer, |
5885 | size_t count, loff_t *ppos) | |
7b3c3a50 | 5886 | { |
b7454ad3 | 5887 | return -EIO; |
7b3c3a50 | 5888 | } |
5b365771 | 5889 | #endif /* CONFIG_SLUB_DEBUG */ |