]>
Commit | Line | Data |
---|---|---|
81819f0f CL |
1 | /* |
2 | * SLUB: A slab allocator that limits cache line use instead of queuing | |
3 | * objects in per cpu and per node lists. | |
4 | * | |
881db7fb CL |
5 | * The allocator synchronizes using per slab locks or atomic operatios |
6 | * and only uses a centralized lock to manage a pool of partial slabs. | |
81819f0f | 7 | * |
cde53535 | 8 | * (C) 2007 SGI, Christoph Lameter |
881db7fb | 9 | * (C) 2011 Linux Foundation, Christoph Lameter |
81819f0f CL |
10 | */ |
11 | ||
12 | #include <linux/mm.h> | |
1eb5ac64 | 13 | #include <linux/swap.h> /* struct reclaim_state */ |
81819f0f CL |
14 | #include <linux/module.h> |
15 | #include <linux/bit_spinlock.h> | |
16 | #include <linux/interrupt.h> | |
17 | #include <linux/bitops.h> | |
18 | #include <linux/slab.h> | |
97d06609 | 19 | #include "slab.h" |
7b3c3a50 | 20 | #include <linux/proc_fs.h> |
3ac38faa | 21 | #include <linux/notifier.h> |
81819f0f | 22 | #include <linux/seq_file.h> |
a79316c6 | 23 | #include <linux/kasan.h> |
5a896d9e | 24 | #include <linux/kmemcheck.h> |
81819f0f CL |
25 | #include <linux/cpu.h> |
26 | #include <linux/cpuset.h> | |
27 | #include <linux/mempolicy.h> | |
28 | #include <linux/ctype.h> | |
3ac7fe5a | 29 | #include <linux/debugobjects.h> |
81819f0f | 30 | #include <linux/kallsyms.h> |
b9049e23 | 31 | #include <linux/memory.h> |
f8bd2258 | 32 | #include <linux/math64.h> |
773ff60e | 33 | #include <linux/fault-inject.h> |
bfa71457 | 34 | #include <linux/stacktrace.h> |
4de900b4 | 35 | #include <linux/prefetch.h> |
2633d7a0 | 36 | #include <linux/memcontrol.h> |
81819f0f | 37 | |
4a92379b RK |
38 | #include <trace/events/kmem.h> |
39 | ||
072bb0aa MG |
40 | #include "internal.h" |
41 | ||
81819f0f CL |
42 | /* |
43 | * Lock order: | |
18004c5d | 44 | * 1. slab_mutex (Global Mutex) |
881db7fb CL |
45 | * 2. node->list_lock |
46 | * 3. slab_lock(page) (Only on some arches and for debugging) | |
81819f0f | 47 | * |
18004c5d | 48 | * slab_mutex |
881db7fb | 49 | * |
18004c5d | 50 | * The role of the slab_mutex is to protect the list of all the slabs |
881db7fb CL |
51 | * and to synchronize major metadata changes to slab cache structures. |
52 | * | |
53 | * The slab_lock is only used for debugging and on arches that do not | |
54 | * have the ability to do a cmpxchg_double. It only protects the second | |
55 | * double word in the page struct. Meaning | |
56 | * A. page->freelist -> List of object free in a page | |
57 | * B. page->counters -> Counters of objects | |
58 | * C. page->frozen -> frozen state | |
59 | * | |
60 | * If a slab is frozen then it is exempt from list management. It is not | |
61 | * on any list. The processor that froze the slab is the one who can | |
62 | * perform list operations on the page. Other processors may put objects | |
63 | * onto the freelist but the processor that froze the slab is the only | |
64 | * one that can retrieve the objects from the page's freelist. | |
81819f0f CL |
65 | * |
66 | * The list_lock protects the partial and full list on each node and | |
67 | * the partial slab counter. If taken then no new slabs may be added or | |
68 | * removed from the lists nor make the number of partial slabs be modified. | |
69 | * (Note that the total number of slabs is an atomic value that may be | |
70 | * modified without taking the list lock). | |
71 | * | |
72 | * The list_lock is a centralized lock and thus we avoid taking it as | |
73 | * much as possible. As long as SLUB does not have to handle partial | |
74 | * slabs, operations can continue without any centralized lock. F.e. | |
75 | * allocating a long series of objects that fill up slabs does not require | |
76 | * the list lock. | |
81819f0f CL |
77 | * Interrupts are disabled during allocation and deallocation in order to |
78 | * make the slab allocator safe to use in the context of an irq. In addition | |
79 | * interrupts are disabled to ensure that the processor does not change | |
80 | * while handling per_cpu slabs, due to kernel preemption. | |
81 | * | |
82 | * SLUB assigns one slab for allocation to each processor. | |
83 | * Allocations only occur from these slabs called cpu slabs. | |
84 | * | |
672bba3a CL |
85 | * Slabs with free elements are kept on a partial list and during regular |
86 | * operations no list for full slabs is used. If an object in a full slab is | |
81819f0f | 87 | * freed then the slab will show up again on the partial lists. |
672bba3a CL |
88 | * We track full slabs for debugging purposes though because otherwise we |
89 | * cannot scan all objects. | |
81819f0f CL |
90 | * |
91 | * Slabs are freed when they become empty. Teardown and setup is | |
92 | * minimal so we rely on the page allocators per cpu caches for | |
93 | * fast frees and allocs. | |
94 | * | |
95 | * Overloading of page flags that are otherwise used for LRU management. | |
96 | * | |
4b6f0750 CL |
97 | * PageActive The slab is frozen and exempt from list processing. |
98 | * This means that the slab is dedicated to a purpose | |
99 | * such as satisfying allocations for a specific | |
100 | * processor. Objects may be freed in the slab while | |
101 | * it is frozen but slab_free will then skip the usual | |
102 | * list operations. It is up to the processor holding | |
103 | * the slab to integrate the slab into the slab lists | |
104 | * when the slab is no longer needed. | |
105 | * | |
106 | * One use of this flag is to mark slabs that are | |
107 | * used for allocations. Then such a slab becomes a cpu | |
108 | * slab. The cpu slab may be equipped with an additional | |
dfb4f096 | 109 | * freelist that allows lockless access to |
894b8788 CL |
110 | * free objects in addition to the regular freelist |
111 | * that requires the slab lock. | |
81819f0f CL |
112 | * |
113 | * PageError Slab requires special handling due to debug | |
114 | * options set. This moves slab handling out of | |
894b8788 | 115 | * the fast path and disables lockless freelists. |
81819f0f CL |
116 | */ |
117 | ||
af537b0a CL |
118 | static inline int kmem_cache_debug(struct kmem_cache *s) |
119 | { | |
5577bd8a | 120 | #ifdef CONFIG_SLUB_DEBUG |
af537b0a | 121 | return unlikely(s->flags & SLAB_DEBUG_FLAGS); |
5577bd8a | 122 | #else |
af537b0a | 123 | return 0; |
5577bd8a | 124 | #endif |
af537b0a | 125 | } |
5577bd8a | 126 | |
d86bd1be JK |
127 | static inline void *fixup_red_left(struct kmem_cache *s, void *p) |
128 | { | |
129 | if (kmem_cache_debug(s) && s->flags & SLAB_RED_ZONE) | |
130 | p += s->red_left_pad; | |
131 | ||
132 | return p; | |
133 | } | |
134 | ||
345c905d JK |
135 | static inline bool kmem_cache_has_cpu_partial(struct kmem_cache *s) |
136 | { | |
137 | #ifdef CONFIG_SLUB_CPU_PARTIAL | |
138 | return !kmem_cache_debug(s); | |
139 | #else | |
140 | return false; | |
141 | #endif | |
142 | } | |
143 | ||
81819f0f CL |
144 | /* |
145 | * Issues still to be resolved: | |
146 | * | |
81819f0f CL |
147 | * - Support PAGE_ALLOC_DEBUG. Should be easy to do. |
148 | * | |
81819f0f CL |
149 | * - Variable sizing of the per node arrays |
150 | */ | |
151 | ||
152 | /* Enable to test recovery from slab corruption on boot */ | |
153 | #undef SLUB_RESILIENCY_TEST | |
154 | ||
b789ef51 CL |
155 | /* Enable to log cmpxchg failures */ |
156 | #undef SLUB_DEBUG_CMPXCHG | |
157 | ||
2086d26a CL |
158 | /* |
159 | * Mininum number of partial slabs. These will be left on the partial | |
160 | * lists even if they are empty. kmem_cache_shrink may reclaim them. | |
161 | */ | |
76be8950 | 162 | #define MIN_PARTIAL 5 |
e95eed57 | 163 | |
2086d26a CL |
164 | /* |
165 | * Maximum number of desirable partial slabs. | |
166 | * The existence of more partial slabs makes kmem_cache_shrink | |
721ae22a | 167 | * sort the partial list by the number of objects in use. |
2086d26a CL |
168 | */ |
169 | #define MAX_PARTIAL 10 | |
170 | ||
becfda68 | 171 | #define DEBUG_DEFAULT_FLAGS (SLAB_CONSISTENCY_CHECKS | SLAB_RED_ZONE | \ |
81819f0f | 172 | SLAB_POISON | SLAB_STORE_USER) |
672bba3a | 173 | |
149daaf3 LA |
174 | /* |
175 | * These debug flags cannot use CMPXCHG because there might be consistency | |
176 | * issues when checking or reading debug information | |
177 | */ | |
178 | #define SLAB_NO_CMPXCHG (SLAB_CONSISTENCY_CHECKS | SLAB_STORE_USER | \ | |
179 | SLAB_TRACE) | |
180 | ||
181 | ||
fa5ec8a1 | 182 | /* |
3de47213 DR |
183 | * Debugging flags that require metadata to be stored in the slab. These get |
184 | * disabled when slub_debug=O is used and a cache's min order increases with | |
185 | * metadata. | |
fa5ec8a1 | 186 | */ |
3de47213 | 187 | #define DEBUG_METADATA_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER) |
fa5ec8a1 | 188 | |
210b5c06 CG |
189 | #define OO_SHIFT 16 |
190 | #define OO_MASK ((1 << OO_SHIFT) - 1) | |
50d5c41c | 191 | #define MAX_OBJS_PER_PAGE 32767 /* since page.objects is u15 */ |
210b5c06 | 192 | |
81819f0f | 193 | /* Internal SLUB flags */ |
f90ec390 | 194 | #define __OBJECT_POISON 0x80000000UL /* Poison object */ |
b789ef51 | 195 | #define __CMPXCHG_DOUBLE 0x40000000UL /* Use cmpxchg_double */ |
81819f0f | 196 | |
81819f0f CL |
197 | #ifdef CONFIG_SMP |
198 | static struct notifier_block slab_notifier; | |
199 | #endif | |
200 | ||
02cbc874 CL |
201 | /* |
202 | * Tracking user of a slab. | |
203 | */ | |
d6543e39 | 204 | #define TRACK_ADDRS_COUNT 16 |
02cbc874 | 205 | struct track { |
ce71e27c | 206 | unsigned long addr; /* Called from address */ |
d6543e39 BG |
207 | #ifdef CONFIG_STACKTRACE |
208 | unsigned long addrs[TRACK_ADDRS_COUNT]; /* Called from address */ | |
209 | #endif | |
02cbc874 CL |
210 | int cpu; /* Was running on cpu */ |
211 | int pid; /* Pid context */ | |
212 | unsigned long when; /* When did the operation occur */ | |
213 | }; | |
214 | ||
215 | enum track_item { TRACK_ALLOC, TRACK_FREE }; | |
216 | ||
ab4d5ed5 | 217 | #ifdef CONFIG_SYSFS |
81819f0f CL |
218 | static int sysfs_slab_add(struct kmem_cache *); |
219 | static int sysfs_slab_alias(struct kmem_cache *, const char *); | |
107dab5c | 220 | static void memcg_propagate_slab_attrs(struct kmem_cache *s); |
81819f0f | 221 | #else |
0c710013 CL |
222 | static inline int sysfs_slab_add(struct kmem_cache *s) { return 0; } |
223 | static inline int sysfs_slab_alias(struct kmem_cache *s, const char *p) | |
224 | { return 0; } | |
107dab5c | 225 | static inline void memcg_propagate_slab_attrs(struct kmem_cache *s) { } |
81819f0f CL |
226 | #endif |
227 | ||
4fdccdfb | 228 | static inline void stat(const struct kmem_cache *s, enum stat_item si) |
8ff12cfc CL |
229 | { |
230 | #ifdef CONFIG_SLUB_STATS | |
88da03a6 CL |
231 | /* |
232 | * The rmw is racy on a preemptible kernel but this is acceptable, so | |
233 | * avoid this_cpu_add()'s irq-disable overhead. | |
234 | */ | |
235 | raw_cpu_inc(s->cpu_slab->stat[si]); | |
8ff12cfc CL |
236 | #endif |
237 | } | |
238 | ||
81819f0f CL |
239 | /******************************************************************** |
240 | * Core slab cache functions | |
241 | *******************************************************************/ | |
242 | ||
7656c72b CL |
243 | static inline void *get_freepointer(struct kmem_cache *s, void *object) |
244 | { | |
245 | return *(void **)(object + s->offset); | |
246 | } | |
247 | ||
0ad9500e ED |
248 | static void prefetch_freepointer(const struct kmem_cache *s, void *object) |
249 | { | |
250 | prefetch(object + s->offset); | |
251 | } | |
252 | ||
1393d9a1 CL |
253 | static inline void *get_freepointer_safe(struct kmem_cache *s, void *object) |
254 | { | |
255 | void *p; | |
256 | ||
922d566c JK |
257 | if (!debug_pagealloc_enabled()) |
258 | return get_freepointer(s, object); | |
259 | ||
1393d9a1 | 260 | probe_kernel_read(&p, (void **)(object + s->offset), sizeof(p)); |
1393d9a1 CL |
261 | return p; |
262 | } | |
263 | ||
7656c72b CL |
264 | static inline void set_freepointer(struct kmem_cache *s, void *object, void *fp) |
265 | { | |
266 | *(void **)(object + s->offset) = fp; | |
267 | } | |
268 | ||
269 | /* Loop over all objects in a slab */ | |
224a88be | 270 | #define for_each_object(__p, __s, __addr, __objects) \ |
d86bd1be JK |
271 | for (__p = fixup_red_left(__s, __addr); \ |
272 | __p < (__addr) + (__objects) * (__s)->size; \ | |
273 | __p += (__s)->size) | |
7656c72b | 274 | |
54266640 | 275 | #define for_each_object_idx(__p, __idx, __s, __addr, __objects) \ |
d86bd1be JK |
276 | for (__p = fixup_red_left(__s, __addr), __idx = 1; \ |
277 | __idx <= __objects; \ | |
278 | __p += (__s)->size, __idx++) | |
54266640 | 279 | |
7656c72b CL |
280 | /* Determine object index from a given position */ |
281 | static inline int slab_index(void *p, struct kmem_cache *s, void *addr) | |
282 | { | |
283 | return (p - addr) / s->size; | |
284 | } | |
285 | ||
ab9a0f19 LJ |
286 | static inline int order_objects(int order, unsigned long size, int reserved) |
287 | { | |
288 | return ((PAGE_SIZE << order) - reserved) / size; | |
289 | } | |
290 | ||
834f3d11 | 291 | static inline struct kmem_cache_order_objects oo_make(int order, |
ab9a0f19 | 292 | unsigned long size, int reserved) |
834f3d11 CL |
293 | { |
294 | struct kmem_cache_order_objects x = { | |
ab9a0f19 | 295 | (order << OO_SHIFT) + order_objects(order, size, reserved) |
834f3d11 CL |
296 | }; |
297 | ||
298 | return x; | |
299 | } | |
300 | ||
301 | static inline int oo_order(struct kmem_cache_order_objects x) | |
302 | { | |
210b5c06 | 303 | return x.x >> OO_SHIFT; |
834f3d11 CL |
304 | } |
305 | ||
306 | static inline int oo_objects(struct kmem_cache_order_objects x) | |
307 | { | |
210b5c06 | 308 | return x.x & OO_MASK; |
834f3d11 CL |
309 | } |
310 | ||
881db7fb CL |
311 | /* |
312 | * Per slab locking using the pagelock | |
313 | */ | |
314 | static __always_inline void slab_lock(struct page *page) | |
315 | { | |
48c935ad | 316 | VM_BUG_ON_PAGE(PageTail(page), page); |
881db7fb CL |
317 | bit_spin_lock(PG_locked, &page->flags); |
318 | } | |
319 | ||
320 | static __always_inline void slab_unlock(struct page *page) | |
321 | { | |
48c935ad | 322 | VM_BUG_ON_PAGE(PageTail(page), page); |
881db7fb CL |
323 | __bit_spin_unlock(PG_locked, &page->flags); |
324 | } | |
325 | ||
a0320865 DH |
326 | static inline void set_page_slub_counters(struct page *page, unsigned long counters_new) |
327 | { | |
328 | struct page tmp; | |
329 | tmp.counters = counters_new; | |
330 | /* | |
331 | * page->counters can cover frozen/inuse/objects as well | |
0139aa7b JK |
332 | * as page->_refcount. If we assign to ->counters directly |
333 | * we run the risk of losing updates to page->_refcount, so | |
a0320865 DH |
334 | * be careful and only assign to the fields we need. |
335 | */ | |
336 | page->frozen = tmp.frozen; | |
337 | page->inuse = tmp.inuse; | |
338 | page->objects = tmp.objects; | |
339 | } | |
340 | ||
1d07171c CL |
341 | /* Interrupts must be disabled (for the fallback code to work right) */ |
342 | static inline bool __cmpxchg_double_slab(struct kmem_cache *s, struct page *page, | |
343 | void *freelist_old, unsigned long counters_old, | |
344 | void *freelist_new, unsigned long counters_new, | |
345 | const char *n) | |
346 | { | |
347 | VM_BUG_ON(!irqs_disabled()); | |
2565409f HC |
348 | #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \ |
349 | defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE) | |
1d07171c | 350 | if (s->flags & __CMPXCHG_DOUBLE) { |
cdcd6298 | 351 | if (cmpxchg_double(&page->freelist, &page->counters, |
0aa9a13d DC |
352 | freelist_old, counters_old, |
353 | freelist_new, counters_new)) | |
6f6528a1 | 354 | return true; |
1d07171c CL |
355 | } else |
356 | #endif | |
357 | { | |
358 | slab_lock(page); | |
d0e0ac97 CG |
359 | if (page->freelist == freelist_old && |
360 | page->counters == counters_old) { | |
1d07171c | 361 | page->freelist = freelist_new; |
a0320865 | 362 | set_page_slub_counters(page, counters_new); |
1d07171c | 363 | slab_unlock(page); |
6f6528a1 | 364 | return true; |
1d07171c CL |
365 | } |
366 | slab_unlock(page); | |
367 | } | |
368 | ||
369 | cpu_relax(); | |
370 | stat(s, CMPXCHG_DOUBLE_FAIL); | |
371 | ||
372 | #ifdef SLUB_DEBUG_CMPXCHG | |
f9f58285 | 373 | pr_info("%s %s: cmpxchg double redo ", n, s->name); |
1d07171c CL |
374 | #endif |
375 | ||
6f6528a1 | 376 | return false; |
1d07171c CL |
377 | } |
378 | ||
b789ef51 CL |
379 | static inline bool cmpxchg_double_slab(struct kmem_cache *s, struct page *page, |
380 | void *freelist_old, unsigned long counters_old, | |
381 | void *freelist_new, unsigned long counters_new, | |
382 | const char *n) | |
383 | { | |
2565409f HC |
384 | #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \ |
385 | defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE) | |
b789ef51 | 386 | if (s->flags & __CMPXCHG_DOUBLE) { |
cdcd6298 | 387 | if (cmpxchg_double(&page->freelist, &page->counters, |
0aa9a13d DC |
388 | freelist_old, counters_old, |
389 | freelist_new, counters_new)) | |
6f6528a1 | 390 | return true; |
b789ef51 CL |
391 | } else |
392 | #endif | |
393 | { | |
1d07171c CL |
394 | unsigned long flags; |
395 | ||
396 | local_irq_save(flags); | |
881db7fb | 397 | slab_lock(page); |
d0e0ac97 CG |
398 | if (page->freelist == freelist_old && |
399 | page->counters == counters_old) { | |
b789ef51 | 400 | page->freelist = freelist_new; |
a0320865 | 401 | set_page_slub_counters(page, counters_new); |
881db7fb | 402 | slab_unlock(page); |
1d07171c | 403 | local_irq_restore(flags); |
6f6528a1 | 404 | return true; |
b789ef51 | 405 | } |
881db7fb | 406 | slab_unlock(page); |
1d07171c | 407 | local_irq_restore(flags); |
b789ef51 CL |
408 | } |
409 | ||
410 | cpu_relax(); | |
411 | stat(s, CMPXCHG_DOUBLE_FAIL); | |
412 | ||
413 | #ifdef SLUB_DEBUG_CMPXCHG | |
f9f58285 | 414 | pr_info("%s %s: cmpxchg double redo ", n, s->name); |
b789ef51 CL |
415 | #endif |
416 | ||
6f6528a1 | 417 | return false; |
b789ef51 CL |
418 | } |
419 | ||
41ecc55b | 420 | #ifdef CONFIG_SLUB_DEBUG |
5f80b13a CL |
421 | /* |
422 | * Determine a map of object in use on a page. | |
423 | * | |
881db7fb | 424 | * Node listlock must be held to guarantee that the page does |
5f80b13a CL |
425 | * not vanish from under us. |
426 | */ | |
427 | static void get_map(struct kmem_cache *s, struct page *page, unsigned long *map) | |
428 | { | |
429 | void *p; | |
430 | void *addr = page_address(page); | |
431 | ||
432 | for (p = page->freelist; p; p = get_freepointer(s, p)) | |
433 | set_bit(slab_index(p, s, addr), map); | |
434 | } | |
435 | ||
d86bd1be JK |
436 | static inline int size_from_object(struct kmem_cache *s) |
437 | { | |
438 | if (s->flags & SLAB_RED_ZONE) | |
439 | return s->size - s->red_left_pad; | |
440 | ||
441 | return s->size; | |
442 | } | |
443 | ||
444 | static inline void *restore_red_left(struct kmem_cache *s, void *p) | |
445 | { | |
446 | if (s->flags & SLAB_RED_ZONE) | |
447 | p -= s->red_left_pad; | |
448 | ||
449 | return p; | |
450 | } | |
451 | ||
41ecc55b CL |
452 | /* |
453 | * Debug settings: | |
454 | */ | |
89d3c87e | 455 | #if defined(CONFIG_SLUB_DEBUG_ON) |
f0630fff | 456 | static int slub_debug = DEBUG_DEFAULT_FLAGS; |
89d3c87e AR |
457 | #elif defined(CONFIG_KASAN) |
458 | static int slub_debug = SLAB_STORE_USER; | |
f0630fff | 459 | #else |
41ecc55b | 460 | static int slub_debug; |
f0630fff | 461 | #endif |
41ecc55b CL |
462 | |
463 | static char *slub_debug_slabs; | |
fa5ec8a1 | 464 | static int disable_higher_order_debug; |
41ecc55b | 465 | |
a79316c6 AR |
466 | /* |
467 | * slub is about to manipulate internal object metadata. This memory lies | |
468 | * outside the range of the allocated object, so accessing it would normally | |
469 | * be reported by kasan as a bounds error. metadata_access_enable() is used | |
470 | * to tell kasan that these accesses are OK. | |
471 | */ | |
472 | static inline void metadata_access_enable(void) | |
473 | { | |
474 | kasan_disable_current(); | |
475 | } | |
476 | ||
477 | static inline void metadata_access_disable(void) | |
478 | { | |
479 | kasan_enable_current(); | |
480 | } | |
481 | ||
81819f0f CL |
482 | /* |
483 | * Object debugging | |
484 | */ | |
d86bd1be JK |
485 | |
486 | /* Verify that a pointer has an address that is valid within a slab page */ | |
487 | static inline int check_valid_pointer(struct kmem_cache *s, | |
488 | struct page *page, void *object) | |
489 | { | |
490 | void *base; | |
491 | ||
492 | if (!object) | |
493 | return 1; | |
494 | ||
495 | base = page_address(page); | |
496 | object = restore_red_left(s, object); | |
497 | if (object < base || object >= base + page->objects * s->size || | |
498 | (object - base) % s->size) { | |
499 | return 0; | |
500 | } | |
501 | ||
502 | return 1; | |
503 | } | |
504 | ||
81819f0f CL |
505 | static void print_section(char *text, u8 *addr, unsigned int length) |
506 | { | |
a79316c6 | 507 | metadata_access_enable(); |
ffc79d28 SAS |
508 | print_hex_dump(KERN_ERR, text, DUMP_PREFIX_ADDRESS, 16, 1, addr, |
509 | length, 1); | |
a79316c6 | 510 | metadata_access_disable(); |
81819f0f CL |
511 | } |
512 | ||
81819f0f CL |
513 | static struct track *get_track(struct kmem_cache *s, void *object, |
514 | enum track_item alloc) | |
515 | { | |
516 | struct track *p; | |
517 | ||
518 | if (s->offset) | |
519 | p = object + s->offset + sizeof(void *); | |
520 | else | |
521 | p = object + s->inuse; | |
522 | ||
523 | return p + alloc; | |
524 | } | |
525 | ||
526 | static void set_track(struct kmem_cache *s, void *object, | |
ce71e27c | 527 | enum track_item alloc, unsigned long addr) |
81819f0f | 528 | { |
1a00df4a | 529 | struct track *p = get_track(s, object, alloc); |
81819f0f | 530 | |
81819f0f | 531 | if (addr) { |
d6543e39 BG |
532 | #ifdef CONFIG_STACKTRACE |
533 | struct stack_trace trace; | |
534 | int i; | |
535 | ||
536 | trace.nr_entries = 0; | |
537 | trace.max_entries = TRACK_ADDRS_COUNT; | |
538 | trace.entries = p->addrs; | |
539 | trace.skip = 3; | |
a79316c6 | 540 | metadata_access_enable(); |
d6543e39 | 541 | save_stack_trace(&trace); |
a79316c6 | 542 | metadata_access_disable(); |
d6543e39 BG |
543 | |
544 | /* See rant in lockdep.c */ | |
545 | if (trace.nr_entries != 0 && | |
546 | trace.entries[trace.nr_entries - 1] == ULONG_MAX) | |
547 | trace.nr_entries--; | |
548 | ||
549 | for (i = trace.nr_entries; i < TRACK_ADDRS_COUNT; i++) | |
550 | p->addrs[i] = 0; | |
551 | #endif | |
81819f0f CL |
552 | p->addr = addr; |
553 | p->cpu = smp_processor_id(); | |
88e4ccf2 | 554 | p->pid = current->pid; |
81819f0f CL |
555 | p->when = jiffies; |
556 | } else | |
557 | memset(p, 0, sizeof(struct track)); | |
558 | } | |
559 | ||
81819f0f CL |
560 | static void init_tracking(struct kmem_cache *s, void *object) |
561 | { | |
24922684 CL |
562 | if (!(s->flags & SLAB_STORE_USER)) |
563 | return; | |
564 | ||
ce71e27c EGM |
565 | set_track(s, object, TRACK_FREE, 0UL); |
566 | set_track(s, object, TRACK_ALLOC, 0UL); | |
81819f0f CL |
567 | } |
568 | ||
569 | static void print_track(const char *s, struct track *t) | |
570 | { | |
571 | if (!t->addr) | |
572 | return; | |
573 | ||
f9f58285 FF |
574 | pr_err("INFO: %s in %pS age=%lu cpu=%u pid=%d\n", |
575 | s, (void *)t->addr, jiffies - t->when, t->cpu, t->pid); | |
d6543e39 BG |
576 | #ifdef CONFIG_STACKTRACE |
577 | { | |
578 | int i; | |
579 | for (i = 0; i < TRACK_ADDRS_COUNT; i++) | |
580 | if (t->addrs[i]) | |
f9f58285 | 581 | pr_err("\t%pS\n", (void *)t->addrs[i]); |
d6543e39 BG |
582 | else |
583 | break; | |
584 | } | |
585 | #endif | |
24922684 CL |
586 | } |
587 | ||
588 | static void print_tracking(struct kmem_cache *s, void *object) | |
589 | { | |
590 | if (!(s->flags & SLAB_STORE_USER)) | |
591 | return; | |
592 | ||
593 | print_track("Allocated", get_track(s, object, TRACK_ALLOC)); | |
594 | print_track("Freed", get_track(s, object, TRACK_FREE)); | |
595 | } | |
596 | ||
597 | static void print_page_info(struct page *page) | |
598 | { | |
f9f58285 | 599 | pr_err("INFO: Slab 0x%p objects=%u used=%u fp=0x%p flags=0x%04lx\n", |
d0e0ac97 | 600 | page, page->objects, page->inuse, page->freelist, page->flags); |
24922684 CL |
601 | |
602 | } | |
603 | ||
604 | static void slab_bug(struct kmem_cache *s, char *fmt, ...) | |
605 | { | |
ecc42fbe | 606 | struct va_format vaf; |
24922684 | 607 | va_list args; |
24922684 CL |
608 | |
609 | va_start(args, fmt); | |
ecc42fbe FF |
610 | vaf.fmt = fmt; |
611 | vaf.va = &args; | |
f9f58285 | 612 | pr_err("=============================================================================\n"); |
ecc42fbe | 613 | pr_err("BUG %s (%s): %pV\n", s->name, print_tainted(), &vaf); |
f9f58285 | 614 | pr_err("-----------------------------------------------------------------------------\n\n"); |
645df230 | 615 | |
373d4d09 | 616 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); |
ecc42fbe | 617 | va_end(args); |
81819f0f CL |
618 | } |
619 | ||
24922684 CL |
620 | static void slab_fix(struct kmem_cache *s, char *fmt, ...) |
621 | { | |
ecc42fbe | 622 | struct va_format vaf; |
24922684 | 623 | va_list args; |
24922684 CL |
624 | |
625 | va_start(args, fmt); | |
ecc42fbe FF |
626 | vaf.fmt = fmt; |
627 | vaf.va = &args; | |
628 | pr_err("FIX %s: %pV\n", s->name, &vaf); | |
24922684 | 629 | va_end(args); |
24922684 CL |
630 | } |
631 | ||
632 | static void print_trailer(struct kmem_cache *s, struct page *page, u8 *p) | |
81819f0f CL |
633 | { |
634 | unsigned int off; /* Offset of last byte */ | |
a973e9dd | 635 | u8 *addr = page_address(page); |
24922684 CL |
636 | |
637 | print_tracking(s, p); | |
638 | ||
639 | print_page_info(page); | |
640 | ||
f9f58285 FF |
641 | pr_err("INFO: Object 0x%p @offset=%tu fp=0x%p\n\n", |
642 | p, p - addr, get_freepointer(s, p)); | |
24922684 | 643 | |
d86bd1be JK |
644 | if (s->flags & SLAB_RED_ZONE) |
645 | print_section("Redzone ", p - s->red_left_pad, s->red_left_pad); | |
646 | else if (p > addr + 16) | |
ffc79d28 | 647 | print_section("Bytes b4 ", p - 16, 16); |
81819f0f | 648 | |
3b0efdfa | 649 | print_section("Object ", p, min_t(unsigned long, s->object_size, |
ffc79d28 | 650 | PAGE_SIZE)); |
81819f0f | 651 | if (s->flags & SLAB_RED_ZONE) |
3b0efdfa CL |
652 | print_section("Redzone ", p + s->object_size, |
653 | s->inuse - s->object_size); | |
81819f0f | 654 | |
81819f0f CL |
655 | if (s->offset) |
656 | off = s->offset + sizeof(void *); | |
657 | else | |
658 | off = s->inuse; | |
659 | ||
24922684 | 660 | if (s->flags & SLAB_STORE_USER) |
81819f0f | 661 | off += 2 * sizeof(struct track); |
81819f0f | 662 | |
d86bd1be | 663 | if (off != size_from_object(s)) |
81819f0f | 664 | /* Beginning of the filler is the free pointer */ |
d86bd1be | 665 | print_section("Padding ", p + off, size_from_object(s) - off); |
24922684 CL |
666 | |
667 | dump_stack(); | |
81819f0f CL |
668 | } |
669 | ||
75c66def | 670 | void object_err(struct kmem_cache *s, struct page *page, |
81819f0f CL |
671 | u8 *object, char *reason) |
672 | { | |
3dc50637 | 673 | slab_bug(s, "%s", reason); |
24922684 | 674 | print_trailer(s, page, object); |
81819f0f CL |
675 | } |
676 | ||
d0e0ac97 CG |
677 | static void slab_err(struct kmem_cache *s, struct page *page, |
678 | const char *fmt, ...) | |
81819f0f CL |
679 | { |
680 | va_list args; | |
681 | char buf[100]; | |
682 | ||
24922684 CL |
683 | va_start(args, fmt); |
684 | vsnprintf(buf, sizeof(buf), fmt, args); | |
81819f0f | 685 | va_end(args); |
3dc50637 | 686 | slab_bug(s, "%s", buf); |
24922684 | 687 | print_page_info(page); |
81819f0f CL |
688 | dump_stack(); |
689 | } | |
690 | ||
f7cb1933 | 691 | static void init_object(struct kmem_cache *s, void *object, u8 val) |
81819f0f CL |
692 | { |
693 | u8 *p = object; | |
694 | ||
d86bd1be JK |
695 | if (s->flags & SLAB_RED_ZONE) |
696 | memset(p - s->red_left_pad, val, s->red_left_pad); | |
697 | ||
81819f0f | 698 | if (s->flags & __OBJECT_POISON) { |
3b0efdfa CL |
699 | memset(p, POISON_FREE, s->object_size - 1); |
700 | p[s->object_size - 1] = POISON_END; | |
81819f0f CL |
701 | } |
702 | ||
703 | if (s->flags & SLAB_RED_ZONE) | |
3b0efdfa | 704 | memset(p + s->object_size, val, s->inuse - s->object_size); |
81819f0f CL |
705 | } |
706 | ||
24922684 CL |
707 | static void restore_bytes(struct kmem_cache *s, char *message, u8 data, |
708 | void *from, void *to) | |
709 | { | |
710 | slab_fix(s, "Restoring 0x%p-0x%p=0x%x\n", from, to - 1, data); | |
711 | memset(from, data, to - from); | |
712 | } | |
713 | ||
714 | static int check_bytes_and_report(struct kmem_cache *s, struct page *page, | |
715 | u8 *object, char *what, | |
06428780 | 716 | u8 *start, unsigned int value, unsigned int bytes) |
24922684 CL |
717 | { |
718 | u8 *fault; | |
719 | u8 *end; | |
720 | ||
a79316c6 | 721 | metadata_access_enable(); |
79824820 | 722 | fault = memchr_inv(start, value, bytes); |
a79316c6 | 723 | metadata_access_disable(); |
24922684 CL |
724 | if (!fault) |
725 | return 1; | |
726 | ||
727 | end = start + bytes; | |
728 | while (end > fault && end[-1] == value) | |
729 | end--; | |
730 | ||
731 | slab_bug(s, "%s overwritten", what); | |
f9f58285 | 732 | pr_err("INFO: 0x%p-0x%p. First byte 0x%x instead of 0x%x\n", |
24922684 CL |
733 | fault, end - 1, fault[0], value); |
734 | print_trailer(s, page, object); | |
735 | ||
736 | restore_bytes(s, what, value, fault, end); | |
737 | return 0; | |
81819f0f CL |
738 | } |
739 | ||
81819f0f CL |
740 | /* |
741 | * Object layout: | |
742 | * | |
743 | * object address | |
744 | * Bytes of the object to be managed. | |
745 | * If the freepointer may overlay the object then the free | |
746 | * pointer is the first word of the object. | |
672bba3a | 747 | * |
81819f0f CL |
748 | * Poisoning uses 0x6b (POISON_FREE) and the last byte is |
749 | * 0xa5 (POISON_END) | |
750 | * | |
3b0efdfa | 751 | * object + s->object_size |
81819f0f | 752 | * Padding to reach word boundary. This is also used for Redzoning. |
672bba3a | 753 | * Padding is extended by another word if Redzoning is enabled and |
3b0efdfa | 754 | * object_size == inuse. |
672bba3a | 755 | * |
81819f0f CL |
756 | * We fill with 0xbb (RED_INACTIVE) for inactive objects and with |
757 | * 0xcc (RED_ACTIVE) for objects in use. | |
758 | * | |
759 | * object + s->inuse | |
672bba3a CL |
760 | * Meta data starts here. |
761 | * | |
81819f0f CL |
762 | * A. Free pointer (if we cannot overwrite object on free) |
763 | * B. Tracking data for SLAB_STORE_USER | |
672bba3a | 764 | * C. Padding to reach required alignment boundary or at mininum |
6446faa2 | 765 | * one word if debugging is on to be able to detect writes |
672bba3a CL |
766 | * before the word boundary. |
767 | * | |
768 | * Padding is done using 0x5a (POISON_INUSE) | |
81819f0f CL |
769 | * |
770 | * object + s->size | |
672bba3a | 771 | * Nothing is used beyond s->size. |
81819f0f | 772 | * |
3b0efdfa | 773 | * If slabcaches are merged then the object_size and inuse boundaries are mostly |
672bba3a | 774 | * ignored. And therefore no slab options that rely on these boundaries |
81819f0f CL |
775 | * may be used with merged slabcaches. |
776 | */ | |
777 | ||
81819f0f CL |
778 | static int check_pad_bytes(struct kmem_cache *s, struct page *page, u8 *p) |
779 | { | |
780 | unsigned long off = s->inuse; /* The end of info */ | |
781 | ||
782 | if (s->offset) | |
783 | /* Freepointer is placed after the object. */ | |
784 | off += sizeof(void *); | |
785 | ||
786 | if (s->flags & SLAB_STORE_USER) | |
787 | /* We also have user information there */ | |
788 | off += 2 * sizeof(struct track); | |
789 | ||
d86bd1be | 790 | if (size_from_object(s) == off) |
81819f0f CL |
791 | return 1; |
792 | ||
24922684 | 793 | return check_bytes_and_report(s, page, p, "Object padding", |
d86bd1be | 794 | p + off, POISON_INUSE, size_from_object(s) - off); |
81819f0f CL |
795 | } |
796 | ||
39b26464 | 797 | /* Check the pad bytes at the end of a slab page */ |
81819f0f CL |
798 | static int slab_pad_check(struct kmem_cache *s, struct page *page) |
799 | { | |
24922684 CL |
800 | u8 *start; |
801 | u8 *fault; | |
802 | u8 *end; | |
803 | int length; | |
804 | int remainder; | |
81819f0f CL |
805 | |
806 | if (!(s->flags & SLAB_POISON)) | |
807 | return 1; | |
808 | ||
a973e9dd | 809 | start = page_address(page); |
ab9a0f19 | 810 | length = (PAGE_SIZE << compound_order(page)) - s->reserved; |
39b26464 CL |
811 | end = start + length; |
812 | remainder = length % s->size; | |
81819f0f CL |
813 | if (!remainder) |
814 | return 1; | |
815 | ||
a79316c6 | 816 | metadata_access_enable(); |
79824820 | 817 | fault = memchr_inv(end - remainder, POISON_INUSE, remainder); |
a79316c6 | 818 | metadata_access_disable(); |
24922684 CL |
819 | if (!fault) |
820 | return 1; | |
821 | while (end > fault && end[-1] == POISON_INUSE) | |
822 | end--; | |
823 | ||
824 | slab_err(s, page, "Padding overwritten. 0x%p-0x%p", fault, end - 1); | |
ffc79d28 | 825 | print_section("Padding ", end - remainder, remainder); |
24922684 | 826 | |
8a3d271d | 827 | restore_bytes(s, "slab padding", POISON_INUSE, end - remainder, end); |
24922684 | 828 | return 0; |
81819f0f CL |
829 | } |
830 | ||
831 | static int check_object(struct kmem_cache *s, struct page *page, | |
f7cb1933 | 832 | void *object, u8 val) |
81819f0f CL |
833 | { |
834 | u8 *p = object; | |
3b0efdfa | 835 | u8 *endobject = object + s->object_size; |
81819f0f CL |
836 | |
837 | if (s->flags & SLAB_RED_ZONE) { | |
d86bd1be JK |
838 | if (!check_bytes_and_report(s, page, object, "Redzone", |
839 | object - s->red_left_pad, val, s->red_left_pad)) | |
840 | return 0; | |
841 | ||
24922684 | 842 | if (!check_bytes_and_report(s, page, object, "Redzone", |
3b0efdfa | 843 | endobject, val, s->inuse - s->object_size)) |
81819f0f | 844 | return 0; |
81819f0f | 845 | } else { |
3b0efdfa | 846 | if ((s->flags & SLAB_POISON) && s->object_size < s->inuse) { |
3adbefee | 847 | check_bytes_and_report(s, page, p, "Alignment padding", |
d0e0ac97 CG |
848 | endobject, POISON_INUSE, |
849 | s->inuse - s->object_size); | |
3adbefee | 850 | } |
81819f0f CL |
851 | } |
852 | ||
853 | if (s->flags & SLAB_POISON) { | |
f7cb1933 | 854 | if (val != SLUB_RED_ACTIVE && (s->flags & __OBJECT_POISON) && |
24922684 | 855 | (!check_bytes_and_report(s, page, p, "Poison", p, |
3b0efdfa | 856 | POISON_FREE, s->object_size - 1) || |
24922684 | 857 | !check_bytes_and_report(s, page, p, "Poison", |
3b0efdfa | 858 | p + s->object_size - 1, POISON_END, 1))) |
81819f0f | 859 | return 0; |
81819f0f CL |
860 | /* |
861 | * check_pad_bytes cleans up on its own. | |
862 | */ | |
863 | check_pad_bytes(s, page, p); | |
864 | } | |
865 | ||
f7cb1933 | 866 | if (!s->offset && val == SLUB_RED_ACTIVE) |
81819f0f CL |
867 | /* |
868 | * Object and freepointer overlap. Cannot check | |
869 | * freepointer while object is allocated. | |
870 | */ | |
871 | return 1; | |
872 | ||
873 | /* Check free pointer validity */ | |
874 | if (!check_valid_pointer(s, page, get_freepointer(s, p))) { | |
875 | object_err(s, page, p, "Freepointer corrupt"); | |
876 | /* | |
9f6c708e | 877 | * No choice but to zap it and thus lose the remainder |
81819f0f | 878 | * of the free objects in this slab. May cause |
672bba3a | 879 | * another error because the object count is now wrong. |
81819f0f | 880 | */ |
a973e9dd | 881 | set_freepointer(s, p, NULL); |
81819f0f CL |
882 | return 0; |
883 | } | |
884 | return 1; | |
885 | } | |
886 | ||
887 | static int check_slab(struct kmem_cache *s, struct page *page) | |
888 | { | |
39b26464 CL |
889 | int maxobj; |
890 | ||
81819f0f CL |
891 | VM_BUG_ON(!irqs_disabled()); |
892 | ||
893 | if (!PageSlab(page)) { | |
24922684 | 894 | slab_err(s, page, "Not a valid slab page"); |
81819f0f CL |
895 | return 0; |
896 | } | |
39b26464 | 897 | |
ab9a0f19 | 898 | maxobj = order_objects(compound_order(page), s->size, s->reserved); |
39b26464 CL |
899 | if (page->objects > maxobj) { |
900 | slab_err(s, page, "objects %u > max %u", | |
f6edde9c | 901 | page->objects, maxobj); |
39b26464 CL |
902 | return 0; |
903 | } | |
904 | if (page->inuse > page->objects) { | |
24922684 | 905 | slab_err(s, page, "inuse %u > max %u", |
f6edde9c | 906 | page->inuse, page->objects); |
81819f0f CL |
907 | return 0; |
908 | } | |
909 | /* Slab_pad_check fixes things up after itself */ | |
910 | slab_pad_check(s, page); | |
911 | return 1; | |
912 | } | |
913 | ||
914 | /* | |
672bba3a CL |
915 | * Determine if a certain object on a page is on the freelist. Must hold the |
916 | * slab lock to guarantee that the chains are in a consistent state. | |
81819f0f CL |
917 | */ |
918 | static int on_freelist(struct kmem_cache *s, struct page *page, void *search) | |
919 | { | |
920 | int nr = 0; | |
881db7fb | 921 | void *fp; |
81819f0f | 922 | void *object = NULL; |
f6edde9c | 923 | int max_objects; |
81819f0f | 924 | |
881db7fb | 925 | fp = page->freelist; |
39b26464 | 926 | while (fp && nr <= page->objects) { |
81819f0f CL |
927 | if (fp == search) |
928 | return 1; | |
929 | if (!check_valid_pointer(s, page, fp)) { | |
930 | if (object) { | |
931 | object_err(s, page, object, | |
932 | "Freechain corrupt"); | |
a973e9dd | 933 | set_freepointer(s, object, NULL); |
81819f0f | 934 | } else { |
24922684 | 935 | slab_err(s, page, "Freepointer corrupt"); |
a973e9dd | 936 | page->freelist = NULL; |
39b26464 | 937 | page->inuse = page->objects; |
24922684 | 938 | slab_fix(s, "Freelist cleared"); |
81819f0f CL |
939 | return 0; |
940 | } | |
941 | break; | |
942 | } | |
943 | object = fp; | |
944 | fp = get_freepointer(s, object); | |
945 | nr++; | |
946 | } | |
947 | ||
ab9a0f19 | 948 | max_objects = order_objects(compound_order(page), s->size, s->reserved); |
210b5c06 CG |
949 | if (max_objects > MAX_OBJS_PER_PAGE) |
950 | max_objects = MAX_OBJS_PER_PAGE; | |
224a88be CL |
951 | |
952 | if (page->objects != max_objects) { | |
756a025f JP |
953 | slab_err(s, page, "Wrong number of objects. Found %d but should be %d", |
954 | page->objects, max_objects); | |
224a88be CL |
955 | page->objects = max_objects; |
956 | slab_fix(s, "Number of objects adjusted."); | |
957 | } | |
39b26464 | 958 | if (page->inuse != page->objects - nr) { |
756a025f JP |
959 | slab_err(s, page, "Wrong object count. Counter is %d but counted were %d", |
960 | page->inuse, page->objects - nr); | |
39b26464 | 961 | page->inuse = page->objects - nr; |
24922684 | 962 | slab_fix(s, "Object count adjusted."); |
81819f0f CL |
963 | } |
964 | return search == NULL; | |
965 | } | |
966 | ||
0121c619 CL |
967 | static void trace(struct kmem_cache *s, struct page *page, void *object, |
968 | int alloc) | |
3ec09742 CL |
969 | { |
970 | if (s->flags & SLAB_TRACE) { | |
f9f58285 | 971 | pr_info("TRACE %s %s 0x%p inuse=%d fp=0x%p\n", |
3ec09742 CL |
972 | s->name, |
973 | alloc ? "alloc" : "free", | |
974 | object, page->inuse, | |
975 | page->freelist); | |
976 | ||
977 | if (!alloc) | |
d0e0ac97 CG |
978 | print_section("Object ", (void *)object, |
979 | s->object_size); | |
3ec09742 CL |
980 | |
981 | dump_stack(); | |
982 | } | |
983 | } | |
984 | ||
643b1138 | 985 | /* |
672bba3a | 986 | * Tracking of fully allocated slabs for debugging purposes. |
643b1138 | 987 | */ |
5cc6eee8 CL |
988 | static void add_full(struct kmem_cache *s, |
989 | struct kmem_cache_node *n, struct page *page) | |
643b1138 | 990 | { |
5cc6eee8 CL |
991 | if (!(s->flags & SLAB_STORE_USER)) |
992 | return; | |
993 | ||
255d0884 | 994 | lockdep_assert_held(&n->list_lock); |
643b1138 | 995 | list_add(&page->lru, &n->full); |
643b1138 CL |
996 | } |
997 | ||
c65c1877 | 998 | static void remove_full(struct kmem_cache *s, struct kmem_cache_node *n, struct page *page) |
643b1138 | 999 | { |
643b1138 CL |
1000 | if (!(s->flags & SLAB_STORE_USER)) |
1001 | return; | |
1002 | ||
255d0884 | 1003 | lockdep_assert_held(&n->list_lock); |
643b1138 | 1004 | list_del(&page->lru); |
643b1138 CL |
1005 | } |
1006 | ||
0f389ec6 CL |
1007 | /* Tracking of the number of slabs for debugging purposes */ |
1008 | static inline unsigned long slabs_node(struct kmem_cache *s, int node) | |
1009 | { | |
1010 | struct kmem_cache_node *n = get_node(s, node); | |
1011 | ||
1012 | return atomic_long_read(&n->nr_slabs); | |
1013 | } | |
1014 | ||
26c02cf0 AB |
1015 | static inline unsigned long node_nr_slabs(struct kmem_cache_node *n) |
1016 | { | |
1017 | return atomic_long_read(&n->nr_slabs); | |
1018 | } | |
1019 | ||
205ab99d | 1020 | static inline void inc_slabs_node(struct kmem_cache *s, int node, int objects) |
0f389ec6 CL |
1021 | { |
1022 | struct kmem_cache_node *n = get_node(s, node); | |
1023 | ||
1024 | /* | |
1025 | * May be called early in order to allocate a slab for the | |
1026 | * kmem_cache_node structure. Solve the chicken-egg | |
1027 | * dilemma by deferring the increment of the count during | |
1028 | * bootstrap (see early_kmem_cache_node_alloc). | |
1029 | */ | |
338b2642 | 1030 | if (likely(n)) { |
0f389ec6 | 1031 | atomic_long_inc(&n->nr_slabs); |
205ab99d CL |
1032 | atomic_long_add(objects, &n->total_objects); |
1033 | } | |
0f389ec6 | 1034 | } |
205ab99d | 1035 | static inline void dec_slabs_node(struct kmem_cache *s, int node, int objects) |
0f389ec6 CL |
1036 | { |
1037 | struct kmem_cache_node *n = get_node(s, node); | |
1038 | ||
1039 | atomic_long_dec(&n->nr_slabs); | |
205ab99d | 1040 | atomic_long_sub(objects, &n->total_objects); |
0f389ec6 CL |
1041 | } |
1042 | ||
1043 | /* Object debug checks for alloc/free paths */ | |
3ec09742 CL |
1044 | static void setup_object_debug(struct kmem_cache *s, struct page *page, |
1045 | void *object) | |
1046 | { | |
1047 | if (!(s->flags & (SLAB_STORE_USER|SLAB_RED_ZONE|__OBJECT_POISON))) | |
1048 | return; | |
1049 | ||
f7cb1933 | 1050 | init_object(s, object, SLUB_RED_INACTIVE); |
3ec09742 CL |
1051 | init_tracking(s, object); |
1052 | } | |
1053 | ||
becfda68 | 1054 | static inline int alloc_consistency_checks(struct kmem_cache *s, |
d0e0ac97 | 1055 | struct page *page, |
ce71e27c | 1056 | void *object, unsigned long addr) |
81819f0f CL |
1057 | { |
1058 | if (!check_slab(s, page)) | |
becfda68 | 1059 | return 0; |
81819f0f | 1060 | |
81819f0f CL |
1061 | if (!check_valid_pointer(s, page, object)) { |
1062 | object_err(s, page, object, "Freelist Pointer check fails"); | |
becfda68 | 1063 | return 0; |
81819f0f CL |
1064 | } |
1065 | ||
f7cb1933 | 1066 | if (!check_object(s, page, object, SLUB_RED_INACTIVE)) |
becfda68 LA |
1067 | return 0; |
1068 | ||
1069 | return 1; | |
1070 | } | |
1071 | ||
1072 | static noinline int alloc_debug_processing(struct kmem_cache *s, | |
1073 | struct page *page, | |
1074 | void *object, unsigned long addr) | |
1075 | { | |
1076 | if (s->flags & SLAB_CONSISTENCY_CHECKS) { | |
1077 | if (!alloc_consistency_checks(s, page, object, addr)) | |
1078 | goto bad; | |
1079 | } | |
81819f0f | 1080 | |
3ec09742 CL |
1081 | /* Success perform special debug activities for allocs */ |
1082 | if (s->flags & SLAB_STORE_USER) | |
1083 | set_track(s, object, TRACK_ALLOC, addr); | |
1084 | trace(s, page, object, 1); | |
f7cb1933 | 1085 | init_object(s, object, SLUB_RED_ACTIVE); |
81819f0f | 1086 | return 1; |
3ec09742 | 1087 | |
81819f0f CL |
1088 | bad: |
1089 | if (PageSlab(page)) { | |
1090 | /* | |
1091 | * If this is a slab page then lets do the best we can | |
1092 | * to avoid issues in the future. Marking all objects | |
672bba3a | 1093 | * as used avoids touching the remaining objects. |
81819f0f | 1094 | */ |
24922684 | 1095 | slab_fix(s, "Marking all objects used"); |
39b26464 | 1096 | page->inuse = page->objects; |
a973e9dd | 1097 | page->freelist = NULL; |
81819f0f CL |
1098 | } |
1099 | return 0; | |
1100 | } | |
1101 | ||
becfda68 LA |
1102 | static inline int free_consistency_checks(struct kmem_cache *s, |
1103 | struct page *page, void *object, unsigned long addr) | |
81819f0f | 1104 | { |
81819f0f | 1105 | if (!check_valid_pointer(s, page, object)) { |
70d71228 | 1106 | slab_err(s, page, "Invalid object pointer 0x%p", object); |
becfda68 | 1107 | return 0; |
81819f0f CL |
1108 | } |
1109 | ||
1110 | if (on_freelist(s, page, object)) { | |
24922684 | 1111 | object_err(s, page, object, "Object already free"); |
becfda68 | 1112 | return 0; |
81819f0f CL |
1113 | } |
1114 | ||
f7cb1933 | 1115 | if (!check_object(s, page, object, SLUB_RED_ACTIVE)) |
becfda68 | 1116 | return 0; |
81819f0f | 1117 | |
1b4f59e3 | 1118 | if (unlikely(s != page->slab_cache)) { |
3adbefee | 1119 | if (!PageSlab(page)) { |
756a025f JP |
1120 | slab_err(s, page, "Attempt to free object(0x%p) outside of slab", |
1121 | object); | |
1b4f59e3 | 1122 | } else if (!page->slab_cache) { |
f9f58285 FF |
1123 | pr_err("SLUB <none>: no slab for object 0x%p.\n", |
1124 | object); | |
70d71228 | 1125 | dump_stack(); |
06428780 | 1126 | } else |
24922684 CL |
1127 | object_err(s, page, object, |
1128 | "page slab pointer corrupt."); | |
becfda68 LA |
1129 | return 0; |
1130 | } | |
1131 | return 1; | |
1132 | } | |
1133 | ||
1134 | /* Supports checking bulk free of a constructed freelist */ | |
1135 | static noinline int free_debug_processing( | |
1136 | struct kmem_cache *s, struct page *page, | |
1137 | void *head, void *tail, int bulk_cnt, | |
1138 | unsigned long addr) | |
1139 | { | |
1140 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); | |
1141 | void *object = head; | |
1142 | int cnt = 0; | |
1143 | unsigned long uninitialized_var(flags); | |
1144 | int ret = 0; | |
1145 | ||
1146 | spin_lock_irqsave(&n->list_lock, flags); | |
1147 | slab_lock(page); | |
1148 | ||
1149 | if (s->flags & SLAB_CONSISTENCY_CHECKS) { | |
1150 | if (!check_slab(s, page)) | |
1151 | goto out; | |
1152 | } | |
1153 | ||
1154 | next_object: | |
1155 | cnt++; | |
1156 | ||
1157 | if (s->flags & SLAB_CONSISTENCY_CHECKS) { | |
1158 | if (!free_consistency_checks(s, page, object, addr)) | |
1159 | goto out; | |
81819f0f | 1160 | } |
3ec09742 | 1161 | |
3ec09742 CL |
1162 | if (s->flags & SLAB_STORE_USER) |
1163 | set_track(s, object, TRACK_FREE, addr); | |
1164 | trace(s, page, object, 0); | |
81084651 | 1165 | /* Freepointer not overwritten by init_object(), SLAB_POISON moved it */ |
f7cb1933 | 1166 | init_object(s, object, SLUB_RED_INACTIVE); |
81084651 JDB |
1167 | |
1168 | /* Reached end of constructed freelist yet? */ | |
1169 | if (object != tail) { | |
1170 | object = get_freepointer(s, object); | |
1171 | goto next_object; | |
1172 | } | |
804aa132 LA |
1173 | ret = 1; |
1174 | ||
5c2e4bbb | 1175 | out: |
81084651 JDB |
1176 | if (cnt != bulk_cnt) |
1177 | slab_err(s, page, "Bulk freelist count(%d) invalid(%d)\n", | |
1178 | bulk_cnt, cnt); | |
1179 | ||
881db7fb | 1180 | slab_unlock(page); |
282acb43 | 1181 | spin_unlock_irqrestore(&n->list_lock, flags); |
804aa132 LA |
1182 | if (!ret) |
1183 | slab_fix(s, "Object at 0x%p not freed", object); | |
1184 | return ret; | |
81819f0f CL |
1185 | } |
1186 | ||
41ecc55b CL |
1187 | static int __init setup_slub_debug(char *str) |
1188 | { | |
f0630fff CL |
1189 | slub_debug = DEBUG_DEFAULT_FLAGS; |
1190 | if (*str++ != '=' || !*str) | |
1191 | /* | |
1192 | * No options specified. Switch on full debugging. | |
1193 | */ | |
1194 | goto out; | |
1195 | ||
1196 | if (*str == ',') | |
1197 | /* | |
1198 | * No options but restriction on slabs. This means full | |
1199 | * debugging for slabs matching a pattern. | |
1200 | */ | |
1201 | goto check_slabs; | |
1202 | ||
1203 | slub_debug = 0; | |
1204 | if (*str == '-') | |
1205 | /* | |
1206 | * Switch off all debugging measures. | |
1207 | */ | |
1208 | goto out; | |
1209 | ||
1210 | /* | |
1211 | * Determine which debug features should be switched on | |
1212 | */ | |
06428780 | 1213 | for (; *str && *str != ','; str++) { |
f0630fff CL |
1214 | switch (tolower(*str)) { |
1215 | case 'f': | |
becfda68 | 1216 | slub_debug |= SLAB_CONSISTENCY_CHECKS; |
f0630fff CL |
1217 | break; |
1218 | case 'z': | |
1219 | slub_debug |= SLAB_RED_ZONE; | |
1220 | break; | |
1221 | case 'p': | |
1222 | slub_debug |= SLAB_POISON; | |
1223 | break; | |
1224 | case 'u': | |
1225 | slub_debug |= SLAB_STORE_USER; | |
1226 | break; | |
1227 | case 't': | |
1228 | slub_debug |= SLAB_TRACE; | |
1229 | break; | |
4c13dd3b DM |
1230 | case 'a': |
1231 | slub_debug |= SLAB_FAILSLAB; | |
1232 | break; | |
08303a73 CA |
1233 | case 'o': |
1234 | /* | |
1235 | * Avoid enabling debugging on caches if its minimum | |
1236 | * order would increase as a result. | |
1237 | */ | |
1238 | disable_higher_order_debug = 1; | |
1239 | break; | |
f0630fff | 1240 | default: |
f9f58285 FF |
1241 | pr_err("slub_debug option '%c' unknown. skipped\n", |
1242 | *str); | |
f0630fff | 1243 | } |
41ecc55b CL |
1244 | } |
1245 | ||
f0630fff | 1246 | check_slabs: |
41ecc55b CL |
1247 | if (*str == ',') |
1248 | slub_debug_slabs = str + 1; | |
f0630fff | 1249 | out: |
41ecc55b CL |
1250 | return 1; |
1251 | } | |
1252 | ||
1253 | __setup("slub_debug", setup_slub_debug); | |
1254 | ||
423c929c | 1255 | unsigned long kmem_cache_flags(unsigned long object_size, |
ba0268a8 | 1256 | unsigned long flags, const char *name, |
51cc5068 | 1257 | void (*ctor)(void *)) |
41ecc55b CL |
1258 | { |
1259 | /* | |
e153362a | 1260 | * Enable debugging if selected on the kernel commandline. |
41ecc55b | 1261 | */ |
c6f58d9b CL |
1262 | if (slub_debug && (!slub_debug_slabs || (name && |
1263 | !strncmp(slub_debug_slabs, name, strlen(slub_debug_slabs))))) | |
3de47213 | 1264 | flags |= slub_debug; |
ba0268a8 CL |
1265 | |
1266 | return flags; | |
41ecc55b | 1267 | } |
b4a64718 | 1268 | #else /* !CONFIG_SLUB_DEBUG */ |
3ec09742 CL |
1269 | static inline void setup_object_debug(struct kmem_cache *s, |
1270 | struct page *page, void *object) {} | |
41ecc55b | 1271 | |
3ec09742 | 1272 | static inline int alloc_debug_processing(struct kmem_cache *s, |
ce71e27c | 1273 | struct page *page, void *object, unsigned long addr) { return 0; } |
41ecc55b | 1274 | |
282acb43 | 1275 | static inline int free_debug_processing( |
81084651 JDB |
1276 | struct kmem_cache *s, struct page *page, |
1277 | void *head, void *tail, int bulk_cnt, | |
282acb43 | 1278 | unsigned long addr) { return 0; } |
41ecc55b | 1279 | |
41ecc55b CL |
1280 | static inline int slab_pad_check(struct kmem_cache *s, struct page *page) |
1281 | { return 1; } | |
1282 | static inline int check_object(struct kmem_cache *s, struct page *page, | |
f7cb1933 | 1283 | void *object, u8 val) { return 1; } |
5cc6eee8 CL |
1284 | static inline void add_full(struct kmem_cache *s, struct kmem_cache_node *n, |
1285 | struct page *page) {} | |
c65c1877 PZ |
1286 | static inline void remove_full(struct kmem_cache *s, struct kmem_cache_node *n, |
1287 | struct page *page) {} | |
423c929c | 1288 | unsigned long kmem_cache_flags(unsigned long object_size, |
ba0268a8 | 1289 | unsigned long flags, const char *name, |
51cc5068 | 1290 | void (*ctor)(void *)) |
ba0268a8 CL |
1291 | { |
1292 | return flags; | |
1293 | } | |
41ecc55b | 1294 | #define slub_debug 0 |
0f389ec6 | 1295 | |
fdaa45e9 IM |
1296 | #define disable_higher_order_debug 0 |
1297 | ||
0f389ec6 CL |
1298 | static inline unsigned long slabs_node(struct kmem_cache *s, int node) |
1299 | { return 0; } | |
26c02cf0 AB |
1300 | static inline unsigned long node_nr_slabs(struct kmem_cache_node *n) |
1301 | { return 0; } | |
205ab99d CL |
1302 | static inline void inc_slabs_node(struct kmem_cache *s, int node, |
1303 | int objects) {} | |
1304 | static inline void dec_slabs_node(struct kmem_cache *s, int node, | |
1305 | int objects) {} | |
7d550c56 | 1306 | |
02e72cc6 AR |
1307 | #endif /* CONFIG_SLUB_DEBUG */ |
1308 | ||
1309 | /* | |
1310 | * Hooks for other subsystems that check memory allocations. In a typical | |
1311 | * production configuration these hooks all should produce no code at all. | |
1312 | */ | |
d56791b3 RB |
1313 | static inline void kmalloc_large_node_hook(void *ptr, size_t size, gfp_t flags) |
1314 | { | |
1315 | kmemleak_alloc(ptr, size, 1, flags); | |
505f5dcb | 1316 | kasan_kmalloc_large(ptr, size, flags); |
d56791b3 RB |
1317 | } |
1318 | ||
1319 | static inline void kfree_hook(const void *x) | |
1320 | { | |
1321 | kmemleak_free(x); | |
0316bec2 | 1322 | kasan_kfree_large(x); |
d56791b3 RB |
1323 | } |
1324 | ||
d56791b3 RB |
1325 | static inline void slab_free_hook(struct kmem_cache *s, void *x) |
1326 | { | |
1327 | kmemleak_free_recursive(x, s->flags); | |
7d550c56 | 1328 | |
02e72cc6 AR |
1329 | /* |
1330 | * Trouble is that we may no longer disable interrupts in the fast path | |
1331 | * So in order to make the debug calls that expect irqs to be | |
1332 | * disabled we need to disable interrupts temporarily. | |
1333 | */ | |
1334 | #if defined(CONFIG_KMEMCHECK) || defined(CONFIG_LOCKDEP) | |
1335 | { | |
1336 | unsigned long flags; | |
1337 | ||
1338 | local_irq_save(flags); | |
1339 | kmemcheck_slab_free(s, x, s->object_size); | |
1340 | debug_check_no_locks_freed(x, s->object_size); | |
1341 | local_irq_restore(flags); | |
1342 | } | |
1343 | #endif | |
1344 | if (!(s->flags & SLAB_DEBUG_OBJECTS)) | |
1345 | debug_check_no_obj_freed(x, s->object_size); | |
0316bec2 AR |
1346 | |
1347 | kasan_slab_free(s, x); | |
02e72cc6 | 1348 | } |
205ab99d | 1349 | |
81084651 JDB |
1350 | static inline void slab_free_freelist_hook(struct kmem_cache *s, |
1351 | void *head, void *tail) | |
1352 | { | |
1353 | /* | |
1354 | * Compiler cannot detect this function can be removed if slab_free_hook() | |
1355 | * evaluates to nothing. Thus, catch all relevant config debug options here. | |
1356 | */ | |
1357 | #if defined(CONFIG_KMEMCHECK) || \ | |
1358 | defined(CONFIG_LOCKDEP) || \ | |
1359 | defined(CONFIG_DEBUG_KMEMLEAK) || \ | |
1360 | defined(CONFIG_DEBUG_OBJECTS_FREE) || \ | |
1361 | defined(CONFIG_KASAN) | |
1362 | ||
1363 | void *object = head; | |
1364 | void *tail_obj = tail ? : head; | |
1365 | ||
1366 | do { | |
1367 | slab_free_hook(s, object); | |
1368 | } while ((object != tail_obj) && | |
1369 | (object = get_freepointer(s, object))); | |
1370 | #endif | |
1371 | } | |
1372 | ||
588f8ba9 TG |
1373 | static void setup_object(struct kmem_cache *s, struct page *page, |
1374 | void *object) | |
1375 | { | |
1376 | setup_object_debug(s, page, object); | |
1377 | if (unlikely(s->ctor)) { | |
1378 | kasan_unpoison_object_data(s, object); | |
1379 | s->ctor(object); | |
1380 | kasan_poison_object_data(s, object); | |
1381 | } | |
1382 | } | |
1383 | ||
81819f0f CL |
1384 | /* |
1385 | * Slab allocation and freeing | |
1386 | */ | |
5dfb4175 VD |
1387 | static inline struct page *alloc_slab_page(struct kmem_cache *s, |
1388 | gfp_t flags, int node, struct kmem_cache_order_objects oo) | |
65c3376a | 1389 | { |
5dfb4175 | 1390 | struct page *page; |
65c3376a CL |
1391 | int order = oo_order(oo); |
1392 | ||
b1eeab67 VN |
1393 | flags |= __GFP_NOTRACK; |
1394 | ||
2154a336 | 1395 | if (node == NUMA_NO_NODE) |
5dfb4175 | 1396 | page = alloc_pages(flags, order); |
65c3376a | 1397 | else |
96db800f | 1398 | page = __alloc_pages_node(node, flags, order); |
5dfb4175 | 1399 | |
f3ccb2c4 VD |
1400 | if (page && memcg_charge_slab(page, flags, order, s)) { |
1401 | __free_pages(page, order); | |
1402 | page = NULL; | |
1403 | } | |
5dfb4175 VD |
1404 | |
1405 | return page; | |
65c3376a CL |
1406 | } |
1407 | ||
210e7a43 TG |
1408 | #ifdef CONFIG_SLAB_FREELIST_RANDOM |
1409 | /* Pre-initialize the random sequence cache */ | |
1410 | static int init_cache_random_seq(struct kmem_cache *s) | |
1411 | { | |
1412 | int err; | |
1413 | unsigned long i, count = oo_objects(s->oo); | |
1414 | ||
1415 | err = cache_random_seq_create(s, count, GFP_KERNEL); | |
1416 | if (err) { | |
1417 | pr_err("SLUB: Unable to initialize free list for %s\n", | |
1418 | s->name); | |
1419 | return err; | |
1420 | } | |
1421 | ||
1422 | /* Transform to an offset on the set of pages */ | |
1423 | if (s->random_seq) { | |
1424 | for (i = 0; i < count; i++) | |
1425 | s->random_seq[i] *= s->size; | |
1426 | } | |
1427 | return 0; | |
1428 | } | |
1429 | ||
1430 | /* Initialize each random sequence freelist per cache */ | |
1431 | static void __init init_freelist_randomization(void) | |
1432 | { | |
1433 | struct kmem_cache *s; | |
1434 | ||
1435 | mutex_lock(&slab_mutex); | |
1436 | ||
1437 | list_for_each_entry(s, &slab_caches, list) | |
1438 | init_cache_random_seq(s); | |
1439 | ||
1440 | mutex_unlock(&slab_mutex); | |
1441 | } | |
1442 | ||
1443 | /* Get the next entry on the pre-computed freelist randomized */ | |
1444 | static void *next_freelist_entry(struct kmem_cache *s, struct page *page, | |
1445 | unsigned long *pos, void *start, | |
1446 | unsigned long page_limit, | |
1447 | unsigned long freelist_count) | |
1448 | { | |
1449 | unsigned int idx; | |
1450 | ||
1451 | /* | |
1452 | * If the target page allocation failed, the number of objects on the | |
1453 | * page might be smaller than the usual size defined by the cache. | |
1454 | */ | |
1455 | do { | |
1456 | idx = s->random_seq[*pos]; | |
1457 | *pos += 1; | |
1458 | if (*pos >= freelist_count) | |
1459 | *pos = 0; | |
1460 | } while (unlikely(idx >= page_limit)); | |
1461 | ||
1462 | return (char *)start + idx; | |
1463 | } | |
1464 | ||
1465 | /* Shuffle the single linked freelist based on a random pre-computed sequence */ | |
1466 | static bool shuffle_freelist(struct kmem_cache *s, struct page *page) | |
1467 | { | |
1468 | void *start; | |
1469 | void *cur; | |
1470 | void *next; | |
1471 | unsigned long idx, pos, page_limit, freelist_count; | |
1472 | ||
1473 | if (page->objects < 2 || !s->random_seq) | |
1474 | return false; | |
1475 | ||
1476 | freelist_count = oo_objects(s->oo); | |
1477 | pos = get_random_int() % freelist_count; | |
1478 | ||
1479 | page_limit = page->objects * s->size; | |
1480 | start = fixup_red_left(s, page_address(page)); | |
1481 | ||
1482 | /* First entry is used as the base of the freelist */ | |
1483 | cur = next_freelist_entry(s, page, &pos, start, page_limit, | |
1484 | freelist_count); | |
1485 | page->freelist = cur; | |
1486 | ||
1487 | for (idx = 1; idx < page->objects; idx++) { | |
1488 | setup_object(s, page, cur); | |
1489 | next = next_freelist_entry(s, page, &pos, start, page_limit, | |
1490 | freelist_count); | |
1491 | set_freepointer(s, cur, next); | |
1492 | cur = next; | |
1493 | } | |
1494 | setup_object(s, page, cur); | |
1495 | set_freepointer(s, cur, NULL); | |
1496 | ||
1497 | return true; | |
1498 | } | |
1499 | #else | |
1500 | static inline int init_cache_random_seq(struct kmem_cache *s) | |
1501 | { | |
1502 | return 0; | |
1503 | } | |
1504 | static inline void init_freelist_randomization(void) { } | |
1505 | static inline bool shuffle_freelist(struct kmem_cache *s, struct page *page) | |
1506 | { | |
1507 | return false; | |
1508 | } | |
1509 | #endif /* CONFIG_SLAB_FREELIST_RANDOM */ | |
1510 | ||
81819f0f CL |
1511 | static struct page *allocate_slab(struct kmem_cache *s, gfp_t flags, int node) |
1512 | { | |
06428780 | 1513 | struct page *page; |
834f3d11 | 1514 | struct kmem_cache_order_objects oo = s->oo; |
ba52270d | 1515 | gfp_t alloc_gfp; |
588f8ba9 TG |
1516 | void *start, *p; |
1517 | int idx, order; | |
210e7a43 | 1518 | bool shuffle; |
81819f0f | 1519 | |
7e0528da CL |
1520 | flags &= gfp_allowed_mask; |
1521 | ||
d0164adc | 1522 | if (gfpflags_allow_blocking(flags)) |
7e0528da CL |
1523 | local_irq_enable(); |
1524 | ||
b7a49f0d | 1525 | flags |= s->allocflags; |
e12ba74d | 1526 | |
ba52270d PE |
1527 | /* |
1528 | * Let the initial higher-order allocation fail under memory pressure | |
1529 | * so we fall-back to the minimum order allocation. | |
1530 | */ | |
1531 | alloc_gfp = (flags | __GFP_NOWARN | __GFP_NORETRY) & ~__GFP_NOFAIL; | |
d0164adc | 1532 | if ((alloc_gfp & __GFP_DIRECT_RECLAIM) && oo_order(oo) > oo_order(s->min)) |
444eb2a4 | 1533 | alloc_gfp = (alloc_gfp | __GFP_NOMEMALLOC) & ~(__GFP_RECLAIM|__GFP_NOFAIL); |
ba52270d | 1534 | |
5dfb4175 | 1535 | page = alloc_slab_page(s, alloc_gfp, node, oo); |
65c3376a CL |
1536 | if (unlikely(!page)) { |
1537 | oo = s->min; | |
80c3a998 | 1538 | alloc_gfp = flags; |
65c3376a CL |
1539 | /* |
1540 | * Allocation may have failed due to fragmentation. | |
1541 | * Try a lower order alloc if possible | |
1542 | */ | |
5dfb4175 | 1543 | page = alloc_slab_page(s, alloc_gfp, node, oo); |
588f8ba9 TG |
1544 | if (unlikely(!page)) |
1545 | goto out; | |
1546 | stat(s, ORDER_FALLBACK); | |
65c3376a | 1547 | } |
5a896d9e | 1548 | |
588f8ba9 TG |
1549 | if (kmemcheck_enabled && |
1550 | !(s->flags & (SLAB_NOTRACK | DEBUG_DEFAULT_FLAGS))) { | |
b1eeab67 VN |
1551 | int pages = 1 << oo_order(oo); |
1552 | ||
80c3a998 | 1553 | kmemcheck_alloc_shadow(page, oo_order(oo), alloc_gfp, node); |
b1eeab67 VN |
1554 | |
1555 | /* | |
1556 | * Objects from caches that have a constructor don't get | |
1557 | * cleared when they're allocated, so we need to do it here. | |
1558 | */ | |
1559 | if (s->ctor) | |
1560 | kmemcheck_mark_uninitialized_pages(page, pages); | |
1561 | else | |
1562 | kmemcheck_mark_unallocated_pages(page, pages); | |
5a896d9e VN |
1563 | } |
1564 | ||
834f3d11 | 1565 | page->objects = oo_objects(oo); |
81819f0f | 1566 | |
1f458cbf | 1567 | order = compound_order(page); |
1b4f59e3 | 1568 | page->slab_cache = s; |
c03f94cc | 1569 | __SetPageSlab(page); |
2f064f34 | 1570 | if (page_is_pfmemalloc(page)) |
072bb0aa | 1571 | SetPageSlabPfmemalloc(page); |
81819f0f CL |
1572 | |
1573 | start = page_address(page); | |
81819f0f CL |
1574 | |
1575 | if (unlikely(s->flags & SLAB_POISON)) | |
1f458cbf | 1576 | memset(start, POISON_INUSE, PAGE_SIZE << order); |
81819f0f | 1577 | |
0316bec2 AR |
1578 | kasan_poison_slab(page); |
1579 | ||
210e7a43 TG |
1580 | shuffle = shuffle_freelist(s, page); |
1581 | ||
1582 | if (!shuffle) { | |
1583 | for_each_object_idx(p, idx, s, start, page->objects) { | |
1584 | setup_object(s, page, p); | |
1585 | if (likely(idx < page->objects)) | |
1586 | set_freepointer(s, p, p + s->size); | |
1587 | else | |
1588 | set_freepointer(s, p, NULL); | |
1589 | } | |
1590 | page->freelist = fixup_red_left(s, start); | |
81819f0f | 1591 | } |
81819f0f | 1592 | |
e6e82ea1 | 1593 | page->inuse = page->objects; |
8cb0a506 | 1594 | page->frozen = 1; |
588f8ba9 | 1595 | |
81819f0f | 1596 | out: |
d0164adc | 1597 | if (gfpflags_allow_blocking(flags)) |
588f8ba9 TG |
1598 | local_irq_disable(); |
1599 | if (!page) | |
1600 | return NULL; | |
1601 | ||
1602 | mod_zone_page_state(page_zone(page), | |
1603 | (s->flags & SLAB_RECLAIM_ACCOUNT) ? | |
1604 | NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE, | |
1605 | 1 << oo_order(oo)); | |
1606 | ||
1607 | inc_slabs_node(s, page_to_nid(page), page->objects); | |
1608 | ||
81819f0f CL |
1609 | return page; |
1610 | } | |
1611 | ||
588f8ba9 TG |
1612 | static struct page *new_slab(struct kmem_cache *s, gfp_t flags, int node) |
1613 | { | |
1614 | if (unlikely(flags & GFP_SLAB_BUG_MASK)) { | |
bacdcb34 | 1615 | gfp_t invalid_mask = flags & GFP_SLAB_BUG_MASK; |
72baeef0 MH |
1616 | flags &= ~GFP_SLAB_BUG_MASK; |
1617 | pr_warn("Unexpected gfp: %#x (%pGg). Fixing up to gfp: %#x (%pGg). Fix your code!\n", | |
1618 | invalid_mask, &invalid_mask, flags, &flags); | |
588f8ba9 TG |
1619 | } |
1620 | ||
1621 | return allocate_slab(s, | |
1622 | flags & (GFP_RECLAIM_MASK | GFP_CONSTRAINT_MASK), node); | |
1623 | } | |
1624 | ||
81819f0f CL |
1625 | static void __free_slab(struct kmem_cache *s, struct page *page) |
1626 | { | |
834f3d11 CL |
1627 | int order = compound_order(page); |
1628 | int pages = 1 << order; | |
81819f0f | 1629 | |
becfda68 | 1630 | if (s->flags & SLAB_CONSISTENCY_CHECKS) { |
81819f0f CL |
1631 | void *p; |
1632 | ||
1633 | slab_pad_check(s, page); | |
224a88be CL |
1634 | for_each_object(p, s, page_address(page), |
1635 | page->objects) | |
f7cb1933 | 1636 | check_object(s, page, p, SLUB_RED_INACTIVE); |
81819f0f CL |
1637 | } |
1638 | ||
b1eeab67 | 1639 | kmemcheck_free_shadow(page, compound_order(page)); |
5a896d9e | 1640 | |
81819f0f CL |
1641 | mod_zone_page_state(page_zone(page), |
1642 | (s->flags & SLAB_RECLAIM_ACCOUNT) ? | |
1643 | NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE, | |
06428780 | 1644 | -pages); |
81819f0f | 1645 | |
072bb0aa | 1646 | __ClearPageSlabPfmemalloc(page); |
49bd5221 | 1647 | __ClearPageSlab(page); |
1f458cbf | 1648 | |
22b751c3 | 1649 | page_mapcount_reset(page); |
1eb5ac64 NP |
1650 | if (current->reclaim_state) |
1651 | current->reclaim_state->reclaimed_slab += pages; | |
27ee57c9 VD |
1652 | memcg_uncharge_slab(page, order, s); |
1653 | __free_pages(page, order); | |
81819f0f CL |
1654 | } |
1655 | ||
da9a638c LJ |
1656 | #define need_reserve_slab_rcu \ |
1657 | (sizeof(((struct page *)NULL)->lru) < sizeof(struct rcu_head)) | |
1658 | ||
81819f0f CL |
1659 | static void rcu_free_slab(struct rcu_head *h) |
1660 | { | |
1661 | struct page *page; | |
1662 | ||
da9a638c LJ |
1663 | if (need_reserve_slab_rcu) |
1664 | page = virt_to_head_page(h); | |
1665 | else | |
1666 | page = container_of((struct list_head *)h, struct page, lru); | |
1667 | ||
1b4f59e3 | 1668 | __free_slab(page->slab_cache, page); |
81819f0f CL |
1669 | } |
1670 | ||
1671 | static void free_slab(struct kmem_cache *s, struct page *page) | |
1672 | { | |
1673 | if (unlikely(s->flags & SLAB_DESTROY_BY_RCU)) { | |
da9a638c LJ |
1674 | struct rcu_head *head; |
1675 | ||
1676 | if (need_reserve_slab_rcu) { | |
1677 | int order = compound_order(page); | |
1678 | int offset = (PAGE_SIZE << order) - s->reserved; | |
1679 | ||
1680 | VM_BUG_ON(s->reserved != sizeof(*head)); | |
1681 | head = page_address(page) + offset; | |
1682 | } else { | |
bc4f610d | 1683 | head = &page->rcu_head; |
da9a638c | 1684 | } |
81819f0f CL |
1685 | |
1686 | call_rcu(head, rcu_free_slab); | |
1687 | } else | |
1688 | __free_slab(s, page); | |
1689 | } | |
1690 | ||
1691 | static void discard_slab(struct kmem_cache *s, struct page *page) | |
1692 | { | |
205ab99d | 1693 | dec_slabs_node(s, page_to_nid(page), page->objects); |
81819f0f CL |
1694 | free_slab(s, page); |
1695 | } | |
1696 | ||
1697 | /* | |
5cc6eee8 | 1698 | * Management of partially allocated slabs. |
81819f0f | 1699 | */ |
1e4dd946 SR |
1700 | static inline void |
1701 | __add_partial(struct kmem_cache_node *n, struct page *page, int tail) | |
81819f0f | 1702 | { |
e95eed57 | 1703 | n->nr_partial++; |
136333d1 | 1704 | if (tail == DEACTIVATE_TO_TAIL) |
7c2e132c CL |
1705 | list_add_tail(&page->lru, &n->partial); |
1706 | else | |
1707 | list_add(&page->lru, &n->partial); | |
81819f0f CL |
1708 | } |
1709 | ||
1e4dd946 SR |
1710 | static inline void add_partial(struct kmem_cache_node *n, |
1711 | struct page *page, int tail) | |
62e346a8 | 1712 | { |
c65c1877 | 1713 | lockdep_assert_held(&n->list_lock); |
1e4dd946 SR |
1714 | __add_partial(n, page, tail); |
1715 | } | |
c65c1877 | 1716 | |
1e4dd946 SR |
1717 | static inline void remove_partial(struct kmem_cache_node *n, |
1718 | struct page *page) | |
1719 | { | |
1720 | lockdep_assert_held(&n->list_lock); | |
52b4b950 DS |
1721 | list_del(&page->lru); |
1722 | n->nr_partial--; | |
1e4dd946 SR |
1723 | } |
1724 | ||
81819f0f | 1725 | /* |
7ced3719 CL |
1726 | * Remove slab from the partial list, freeze it and |
1727 | * return the pointer to the freelist. | |
81819f0f | 1728 | * |
497b66f2 | 1729 | * Returns a list of objects or NULL if it fails. |
81819f0f | 1730 | */ |
497b66f2 | 1731 | static inline void *acquire_slab(struct kmem_cache *s, |
acd19fd1 | 1732 | struct kmem_cache_node *n, struct page *page, |
633b0764 | 1733 | int mode, int *objects) |
81819f0f | 1734 | { |
2cfb7455 CL |
1735 | void *freelist; |
1736 | unsigned long counters; | |
1737 | struct page new; | |
1738 | ||
c65c1877 PZ |
1739 | lockdep_assert_held(&n->list_lock); |
1740 | ||
2cfb7455 CL |
1741 | /* |
1742 | * Zap the freelist and set the frozen bit. | |
1743 | * The old freelist is the list of objects for the | |
1744 | * per cpu allocation list. | |
1745 | */ | |
7ced3719 CL |
1746 | freelist = page->freelist; |
1747 | counters = page->counters; | |
1748 | new.counters = counters; | |
633b0764 | 1749 | *objects = new.objects - new.inuse; |
23910c50 | 1750 | if (mode) { |
7ced3719 | 1751 | new.inuse = page->objects; |
23910c50 PE |
1752 | new.freelist = NULL; |
1753 | } else { | |
1754 | new.freelist = freelist; | |
1755 | } | |
2cfb7455 | 1756 | |
a0132ac0 | 1757 | VM_BUG_ON(new.frozen); |
7ced3719 | 1758 | new.frozen = 1; |
2cfb7455 | 1759 | |
7ced3719 | 1760 | if (!__cmpxchg_double_slab(s, page, |
2cfb7455 | 1761 | freelist, counters, |
02d7633f | 1762 | new.freelist, new.counters, |
7ced3719 | 1763 | "acquire_slab")) |
7ced3719 | 1764 | return NULL; |
2cfb7455 CL |
1765 | |
1766 | remove_partial(n, page); | |
7ced3719 | 1767 | WARN_ON(!freelist); |
49e22585 | 1768 | return freelist; |
81819f0f CL |
1769 | } |
1770 | ||
633b0764 | 1771 | static void put_cpu_partial(struct kmem_cache *s, struct page *page, int drain); |
8ba00bb6 | 1772 | static inline bool pfmemalloc_match(struct page *page, gfp_t gfpflags); |
49e22585 | 1773 | |
81819f0f | 1774 | /* |
672bba3a | 1775 | * Try to allocate a partial slab from a specific node. |
81819f0f | 1776 | */ |
8ba00bb6 JK |
1777 | static void *get_partial_node(struct kmem_cache *s, struct kmem_cache_node *n, |
1778 | struct kmem_cache_cpu *c, gfp_t flags) | |
81819f0f | 1779 | { |
49e22585 CL |
1780 | struct page *page, *page2; |
1781 | void *object = NULL; | |
633b0764 JK |
1782 | int available = 0; |
1783 | int objects; | |
81819f0f CL |
1784 | |
1785 | /* | |
1786 | * Racy check. If we mistakenly see no partial slabs then we | |
1787 | * just allocate an empty slab. If we mistakenly try to get a | |
672bba3a CL |
1788 | * partial slab and there is none available then get_partials() |
1789 | * will return NULL. | |
81819f0f CL |
1790 | */ |
1791 | if (!n || !n->nr_partial) | |
1792 | return NULL; | |
1793 | ||
1794 | spin_lock(&n->list_lock); | |
49e22585 | 1795 | list_for_each_entry_safe(page, page2, &n->partial, lru) { |
8ba00bb6 | 1796 | void *t; |
49e22585 | 1797 | |
8ba00bb6 JK |
1798 | if (!pfmemalloc_match(page, flags)) |
1799 | continue; | |
1800 | ||
633b0764 | 1801 | t = acquire_slab(s, n, page, object == NULL, &objects); |
49e22585 CL |
1802 | if (!t) |
1803 | break; | |
1804 | ||
633b0764 | 1805 | available += objects; |
12d79634 | 1806 | if (!object) { |
49e22585 | 1807 | c->page = page; |
49e22585 | 1808 | stat(s, ALLOC_FROM_PARTIAL); |
49e22585 | 1809 | object = t; |
49e22585 | 1810 | } else { |
633b0764 | 1811 | put_cpu_partial(s, page, 0); |
8028dcea | 1812 | stat(s, CPU_PARTIAL_NODE); |
49e22585 | 1813 | } |
345c905d JK |
1814 | if (!kmem_cache_has_cpu_partial(s) |
1815 | || available > s->cpu_partial / 2) | |
49e22585 CL |
1816 | break; |
1817 | ||
497b66f2 | 1818 | } |
81819f0f | 1819 | spin_unlock(&n->list_lock); |
497b66f2 | 1820 | return object; |
81819f0f CL |
1821 | } |
1822 | ||
1823 | /* | |
672bba3a | 1824 | * Get a page from somewhere. Search in increasing NUMA distances. |
81819f0f | 1825 | */ |
de3ec035 | 1826 | static void *get_any_partial(struct kmem_cache *s, gfp_t flags, |
acd19fd1 | 1827 | struct kmem_cache_cpu *c) |
81819f0f CL |
1828 | { |
1829 | #ifdef CONFIG_NUMA | |
1830 | struct zonelist *zonelist; | |
dd1a239f | 1831 | struct zoneref *z; |
54a6eb5c MG |
1832 | struct zone *zone; |
1833 | enum zone_type high_zoneidx = gfp_zone(flags); | |
497b66f2 | 1834 | void *object; |
cc9a6c87 | 1835 | unsigned int cpuset_mems_cookie; |
81819f0f CL |
1836 | |
1837 | /* | |
672bba3a CL |
1838 | * The defrag ratio allows a configuration of the tradeoffs between |
1839 | * inter node defragmentation and node local allocations. A lower | |
1840 | * defrag_ratio increases the tendency to do local allocations | |
1841 | * instead of attempting to obtain partial slabs from other nodes. | |
81819f0f | 1842 | * |
672bba3a CL |
1843 | * If the defrag_ratio is set to 0 then kmalloc() always |
1844 | * returns node local objects. If the ratio is higher then kmalloc() | |
1845 | * may return off node objects because partial slabs are obtained | |
1846 | * from other nodes and filled up. | |
81819f0f | 1847 | * |
43efd3ea LP |
1848 | * If /sys/kernel/slab/xx/remote_node_defrag_ratio is set to 100 |
1849 | * (which makes defrag_ratio = 1000) then every (well almost) | |
1850 | * allocation will first attempt to defrag slab caches on other nodes. | |
1851 | * This means scanning over all nodes to look for partial slabs which | |
1852 | * may be expensive if we do it every time we are trying to find a slab | |
672bba3a | 1853 | * with available objects. |
81819f0f | 1854 | */ |
9824601e CL |
1855 | if (!s->remote_node_defrag_ratio || |
1856 | get_cycles() % 1024 > s->remote_node_defrag_ratio) | |
81819f0f CL |
1857 | return NULL; |
1858 | ||
cc9a6c87 | 1859 | do { |
d26914d1 | 1860 | cpuset_mems_cookie = read_mems_allowed_begin(); |
2a389610 | 1861 | zonelist = node_zonelist(mempolicy_slab_node(), flags); |
cc9a6c87 MG |
1862 | for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) { |
1863 | struct kmem_cache_node *n; | |
1864 | ||
1865 | n = get_node(s, zone_to_nid(zone)); | |
1866 | ||
dee2f8aa | 1867 | if (n && cpuset_zone_allowed(zone, flags) && |
cc9a6c87 | 1868 | n->nr_partial > s->min_partial) { |
8ba00bb6 | 1869 | object = get_partial_node(s, n, c, flags); |
cc9a6c87 MG |
1870 | if (object) { |
1871 | /* | |
d26914d1 MG |
1872 | * Don't check read_mems_allowed_retry() |
1873 | * here - if mems_allowed was updated in | |
1874 | * parallel, that was a harmless race | |
1875 | * between allocation and the cpuset | |
1876 | * update | |
cc9a6c87 | 1877 | */ |
cc9a6c87 MG |
1878 | return object; |
1879 | } | |
c0ff7453 | 1880 | } |
81819f0f | 1881 | } |
d26914d1 | 1882 | } while (read_mems_allowed_retry(cpuset_mems_cookie)); |
81819f0f CL |
1883 | #endif |
1884 | return NULL; | |
1885 | } | |
1886 | ||
1887 | /* | |
1888 | * Get a partial page, lock it and return it. | |
1889 | */ | |
497b66f2 | 1890 | static void *get_partial(struct kmem_cache *s, gfp_t flags, int node, |
acd19fd1 | 1891 | struct kmem_cache_cpu *c) |
81819f0f | 1892 | { |
497b66f2 | 1893 | void *object; |
a561ce00 JK |
1894 | int searchnode = node; |
1895 | ||
1896 | if (node == NUMA_NO_NODE) | |
1897 | searchnode = numa_mem_id(); | |
1898 | else if (!node_present_pages(node)) | |
1899 | searchnode = node_to_mem_node(node); | |
81819f0f | 1900 | |
8ba00bb6 | 1901 | object = get_partial_node(s, get_node(s, searchnode), c, flags); |
497b66f2 CL |
1902 | if (object || node != NUMA_NO_NODE) |
1903 | return object; | |
81819f0f | 1904 | |
acd19fd1 | 1905 | return get_any_partial(s, flags, c); |
81819f0f CL |
1906 | } |
1907 | ||
8a5ec0ba CL |
1908 | #ifdef CONFIG_PREEMPT |
1909 | /* | |
1910 | * Calculate the next globally unique transaction for disambiguiation | |
1911 | * during cmpxchg. The transactions start with the cpu number and are then | |
1912 | * incremented by CONFIG_NR_CPUS. | |
1913 | */ | |
1914 | #define TID_STEP roundup_pow_of_two(CONFIG_NR_CPUS) | |
1915 | #else | |
1916 | /* | |
1917 | * No preemption supported therefore also no need to check for | |
1918 | * different cpus. | |
1919 | */ | |
1920 | #define TID_STEP 1 | |
1921 | #endif | |
1922 | ||
1923 | static inline unsigned long next_tid(unsigned long tid) | |
1924 | { | |
1925 | return tid + TID_STEP; | |
1926 | } | |
1927 | ||
1928 | static inline unsigned int tid_to_cpu(unsigned long tid) | |
1929 | { | |
1930 | return tid % TID_STEP; | |
1931 | } | |
1932 | ||
1933 | static inline unsigned long tid_to_event(unsigned long tid) | |
1934 | { | |
1935 | return tid / TID_STEP; | |
1936 | } | |
1937 | ||
1938 | static inline unsigned int init_tid(int cpu) | |
1939 | { | |
1940 | return cpu; | |
1941 | } | |
1942 | ||
1943 | static inline void note_cmpxchg_failure(const char *n, | |
1944 | const struct kmem_cache *s, unsigned long tid) | |
1945 | { | |
1946 | #ifdef SLUB_DEBUG_CMPXCHG | |
1947 | unsigned long actual_tid = __this_cpu_read(s->cpu_slab->tid); | |
1948 | ||
f9f58285 | 1949 | pr_info("%s %s: cmpxchg redo ", n, s->name); |
8a5ec0ba CL |
1950 | |
1951 | #ifdef CONFIG_PREEMPT | |
1952 | if (tid_to_cpu(tid) != tid_to_cpu(actual_tid)) | |
f9f58285 | 1953 | pr_warn("due to cpu change %d -> %d\n", |
8a5ec0ba CL |
1954 | tid_to_cpu(tid), tid_to_cpu(actual_tid)); |
1955 | else | |
1956 | #endif | |
1957 | if (tid_to_event(tid) != tid_to_event(actual_tid)) | |
f9f58285 | 1958 | pr_warn("due to cpu running other code. Event %ld->%ld\n", |
8a5ec0ba CL |
1959 | tid_to_event(tid), tid_to_event(actual_tid)); |
1960 | else | |
f9f58285 | 1961 | pr_warn("for unknown reason: actual=%lx was=%lx target=%lx\n", |
8a5ec0ba CL |
1962 | actual_tid, tid, next_tid(tid)); |
1963 | #endif | |
4fdccdfb | 1964 | stat(s, CMPXCHG_DOUBLE_CPU_FAIL); |
8a5ec0ba CL |
1965 | } |
1966 | ||
788e1aad | 1967 | static void init_kmem_cache_cpus(struct kmem_cache *s) |
8a5ec0ba | 1968 | { |
8a5ec0ba CL |
1969 | int cpu; |
1970 | ||
1971 | for_each_possible_cpu(cpu) | |
1972 | per_cpu_ptr(s->cpu_slab, cpu)->tid = init_tid(cpu); | |
8a5ec0ba | 1973 | } |
2cfb7455 | 1974 | |
81819f0f CL |
1975 | /* |
1976 | * Remove the cpu slab | |
1977 | */ | |
d0e0ac97 CG |
1978 | static void deactivate_slab(struct kmem_cache *s, struct page *page, |
1979 | void *freelist) | |
81819f0f | 1980 | { |
2cfb7455 | 1981 | enum slab_modes { M_NONE, M_PARTIAL, M_FULL, M_FREE }; |
2cfb7455 CL |
1982 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); |
1983 | int lock = 0; | |
1984 | enum slab_modes l = M_NONE, m = M_NONE; | |
2cfb7455 | 1985 | void *nextfree; |
136333d1 | 1986 | int tail = DEACTIVATE_TO_HEAD; |
2cfb7455 CL |
1987 | struct page new; |
1988 | struct page old; | |
1989 | ||
1990 | if (page->freelist) { | |
84e554e6 | 1991 | stat(s, DEACTIVATE_REMOTE_FREES); |
136333d1 | 1992 | tail = DEACTIVATE_TO_TAIL; |
2cfb7455 CL |
1993 | } |
1994 | ||
894b8788 | 1995 | /* |
2cfb7455 CL |
1996 | * Stage one: Free all available per cpu objects back |
1997 | * to the page freelist while it is still frozen. Leave the | |
1998 | * last one. | |
1999 | * | |
2000 | * There is no need to take the list->lock because the page | |
2001 | * is still frozen. | |
2002 | */ | |
2003 | while (freelist && (nextfree = get_freepointer(s, freelist))) { | |
2004 | void *prior; | |
2005 | unsigned long counters; | |
2006 | ||
2007 | do { | |
2008 | prior = page->freelist; | |
2009 | counters = page->counters; | |
2010 | set_freepointer(s, freelist, prior); | |
2011 | new.counters = counters; | |
2012 | new.inuse--; | |
a0132ac0 | 2013 | VM_BUG_ON(!new.frozen); |
2cfb7455 | 2014 | |
1d07171c | 2015 | } while (!__cmpxchg_double_slab(s, page, |
2cfb7455 CL |
2016 | prior, counters, |
2017 | freelist, new.counters, | |
2018 | "drain percpu freelist")); | |
2019 | ||
2020 | freelist = nextfree; | |
2021 | } | |
2022 | ||
894b8788 | 2023 | /* |
2cfb7455 CL |
2024 | * Stage two: Ensure that the page is unfrozen while the |
2025 | * list presence reflects the actual number of objects | |
2026 | * during unfreeze. | |
2027 | * | |
2028 | * We setup the list membership and then perform a cmpxchg | |
2029 | * with the count. If there is a mismatch then the page | |
2030 | * is not unfrozen but the page is on the wrong list. | |
2031 | * | |
2032 | * Then we restart the process which may have to remove | |
2033 | * the page from the list that we just put it on again | |
2034 | * because the number of objects in the slab may have | |
2035 | * changed. | |
894b8788 | 2036 | */ |
2cfb7455 | 2037 | redo: |
894b8788 | 2038 | |
2cfb7455 CL |
2039 | old.freelist = page->freelist; |
2040 | old.counters = page->counters; | |
a0132ac0 | 2041 | VM_BUG_ON(!old.frozen); |
7c2e132c | 2042 | |
2cfb7455 CL |
2043 | /* Determine target state of the slab */ |
2044 | new.counters = old.counters; | |
2045 | if (freelist) { | |
2046 | new.inuse--; | |
2047 | set_freepointer(s, freelist, old.freelist); | |
2048 | new.freelist = freelist; | |
2049 | } else | |
2050 | new.freelist = old.freelist; | |
2051 | ||
2052 | new.frozen = 0; | |
2053 | ||
8a5b20ae | 2054 | if (!new.inuse && n->nr_partial >= s->min_partial) |
2cfb7455 CL |
2055 | m = M_FREE; |
2056 | else if (new.freelist) { | |
2057 | m = M_PARTIAL; | |
2058 | if (!lock) { | |
2059 | lock = 1; | |
2060 | /* | |
2061 | * Taking the spinlock removes the possiblity | |
2062 | * that acquire_slab() will see a slab page that | |
2063 | * is frozen | |
2064 | */ | |
2065 | spin_lock(&n->list_lock); | |
2066 | } | |
2067 | } else { | |
2068 | m = M_FULL; | |
2069 | if (kmem_cache_debug(s) && !lock) { | |
2070 | lock = 1; | |
2071 | /* | |
2072 | * This also ensures that the scanning of full | |
2073 | * slabs from diagnostic functions will not see | |
2074 | * any frozen slabs. | |
2075 | */ | |
2076 | spin_lock(&n->list_lock); | |
2077 | } | |
2078 | } | |
2079 | ||
2080 | if (l != m) { | |
2081 | ||
2082 | if (l == M_PARTIAL) | |
2083 | ||
2084 | remove_partial(n, page); | |
2085 | ||
2086 | else if (l == M_FULL) | |
894b8788 | 2087 | |
c65c1877 | 2088 | remove_full(s, n, page); |
2cfb7455 CL |
2089 | |
2090 | if (m == M_PARTIAL) { | |
2091 | ||
2092 | add_partial(n, page, tail); | |
136333d1 | 2093 | stat(s, tail); |
2cfb7455 CL |
2094 | |
2095 | } else if (m == M_FULL) { | |
894b8788 | 2096 | |
2cfb7455 CL |
2097 | stat(s, DEACTIVATE_FULL); |
2098 | add_full(s, n, page); | |
2099 | ||
2100 | } | |
2101 | } | |
2102 | ||
2103 | l = m; | |
1d07171c | 2104 | if (!__cmpxchg_double_slab(s, page, |
2cfb7455 CL |
2105 | old.freelist, old.counters, |
2106 | new.freelist, new.counters, | |
2107 | "unfreezing slab")) | |
2108 | goto redo; | |
2109 | ||
2cfb7455 CL |
2110 | if (lock) |
2111 | spin_unlock(&n->list_lock); | |
2112 | ||
2113 | if (m == M_FREE) { | |
2114 | stat(s, DEACTIVATE_EMPTY); | |
2115 | discard_slab(s, page); | |
2116 | stat(s, FREE_SLAB); | |
894b8788 | 2117 | } |
81819f0f CL |
2118 | } |
2119 | ||
d24ac77f JK |
2120 | /* |
2121 | * Unfreeze all the cpu partial slabs. | |
2122 | * | |
59a09917 CL |
2123 | * This function must be called with interrupts disabled |
2124 | * for the cpu using c (or some other guarantee must be there | |
2125 | * to guarantee no concurrent accesses). | |
d24ac77f | 2126 | */ |
59a09917 CL |
2127 | static void unfreeze_partials(struct kmem_cache *s, |
2128 | struct kmem_cache_cpu *c) | |
49e22585 | 2129 | { |
345c905d | 2130 | #ifdef CONFIG_SLUB_CPU_PARTIAL |
43d77867 | 2131 | struct kmem_cache_node *n = NULL, *n2 = NULL; |
9ada1934 | 2132 | struct page *page, *discard_page = NULL; |
49e22585 CL |
2133 | |
2134 | while ((page = c->partial)) { | |
49e22585 CL |
2135 | struct page new; |
2136 | struct page old; | |
2137 | ||
2138 | c->partial = page->next; | |
43d77867 JK |
2139 | |
2140 | n2 = get_node(s, page_to_nid(page)); | |
2141 | if (n != n2) { | |
2142 | if (n) | |
2143 | spin_unlock(&n->list_lock); | |
2144 | ||
2145 | n = n2; | |
2146 | spin_lock(&n->list_lock); | |
2147 | } | |
49e22585 CL |
2148 | |
2149 | do { | |
2150 | ||
2151 | old.freelist = page->freelist; | |
2152 | old.counters = page->counters; | |
a0132ac0 | 2153 | VM_BUG_ON(!old.frozen); |
49e22585 CL |
2154 | |
2155 | new.counters = old.counters; | |
2156 | new.freelist = old.freelist; | |
2157 | ||
2158 | new.frozen = 0; | |
2159 | ||
d24ac77f | 2160 | } while (!__cmpxchg_double_slab(s, page, |
49e22585 CL |
2161 | old.freelist, old.counters, |
2162 | new.freelist, new.counters, | |
2163 | "unfreezing slab")); | |
2164 | ||
8a5b20ae | 2165 | if (unlikely(!new.inuse && n->nr_partial >= s->min_partial)) { |
9ada1934 SL |
2166 | page->next = discard_page; |
2167 | discard_page = page; | |
43d77867 JK |
2168 | } else { |
2169 | add_partial(n, page, DEACTIVATE_TO_TAIL); | |
2170 | stat(s, FREE_ADD_PARTIAL); | |
49e22585 CL |
2171 | } |
2172 | } | |
2173 | ||
2174 | if (n) | |
2175 | spin_unlock(&n->list_lock); | |
9ada1934 SL |
2176 | |
2177 | while (discard_page) { | |
2178 | page = discard_page; | |
2179 | discard_page = discard_page->next; | |
2180 | ||
2181 | stat(s, DEACTIVATE_EMPTY); | |
2182 | discard_slab(s, page); | |
2183 | stat(s, FREE_SLAB); | |
2184 | } | |
345c905d | 2185 | #endif |
49e22585 CL |
2186 | } |
2187 | ||
2188 | /* | |
2189 | * Put a page that was just frozen (in __slab_free) into a partial page | |
2190 | * slot if available. This is done without interrupts disabled and without | |
2191 | * preemption disabled. The cmpxchg is racy and may put the partial page | |
2192 | * onto a random cpus partial slot. | |
2193 | * | |
2194 | * If we did not find a slot then simply move all the partials to the | |
2195 | * per node partial list. | |
2196 | */ | |
633b0764 | 2197 | static void put_cpu_partial(struct kmem_cache *s, struct page *page, int drain) |
49e22585 | 2198 | { |
345c905d | 2199 | #ifdef CONFIG_SLUB_CPU_PARTIAL |
49e22585 CL |
2200 | struct page *oldpage; |
2201 | int pages; | |
2202 | int pobjects; | |
2203 | ||
d6e0b7fa | 2204 | preempt_disable(); |
49e22585 CL |
2205 | do { |
2206 | pages = 0; | |
2207 | pobjects = 0; | |
2208 | oldpage = this_cpu_read(s->cpu_slab->partial); | |
2209 | ||
2210 | if (oldpage) { | |
2211 | pobjects = oldpage->pobjects; | |
2212 | pages = oldpage->pages; | |
2213 | if (drain && pobjects > s->cpu_partial) { | |
2214 | unsigned long flags; | |
2215 | /* | |
2216 | * partial array is full. Move the existing | |
2217 | * set to the per node partial list. | |
2218 | */ | |
2219 | local_irq_save(flags); | |
59a09917 | 2220 | unfreeze_partials(s, this_cpu_ptr(s->cpu_slab)); |
49e22585 | 2221 | local_irq_restore(flags); |
e24fc410 | 2222 | oldpage = NULL; |
49e22585 CL |
2223 | pobjects = 0; |
2224 | pages = 0; | |
8028dcea | 2225 | stat(s, CPU_PARTIAL_DRAIN); |
49e22585 CL |
2226 | } |
2227 | } | |
2228 | ||
2229 | pages++; | |
2230 | pobjects += page->objects - page->inuse; | |
2231 | ||
2232 | page->pages = pages; | |
2233 | page->pobjects = pobjects; | |
2234 | page->next = oldpage; | |
2235 | ||
d0e0ac97 CG |
2236 | } while (this_cpu_cmpxchg(s->cpu_slab->partial, oldpage, page) |
2237 | != oldpage); | |
d6e0b7fa VD |
2238 | if (unlikely(!s->cpu_partial)) { |
2239 | unsigned long flags; | |
2240 | ||
2241 | local_irq_save(flags); | |
2242 | unfreeze_partials(s, this_cpu_ptr(s->cpu_slab)); | |
2243 | local_irq_restore(flags); | |
2244 | } | |
2245 | preempt_enable(); | |
345c905d | 2246 | #endif |
49e22585 CL |
2247 | } |
2248 | ||
dfb4f096 | 2249 | static inline void flush_slab(struct kmem_cache *s, struct kmem_cache_cpu *c) |
81819f0f | 2250 | { |
84e554e6 | 2251 | stat(s, CPUSLAB_FLUSH); |
c17dda40 CL |
2252 | deactivate_slab(s, c->page, c->freelist); |
2253 | ||
2254 | c->tid = next_tid(c->tid); | |
2255 | c->page = NULL; | |
2256 | c->freelist = NULL; | |
81819f0f CL |
2257 | } |
2258 | ||
2259 | /* | |
2260 | * Flush cpu slab. | |
6446faa2 | 2261 | * |
81819f0f CL |
2262 | * Called from IPI handler with interrupts disabled. |
2263 | */ | |
0c710013 | 2264 | static inline void __flush_cpu_slab(struct kmem_cache *s, int cpu) |
81819f0f | 2265 | { |
9dfc6e68 | 2266 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu); |
81819f0f | 2267 | |
49e22585 CL |
2268 | if (likely(c)) { |
2269 | if (c->page) | |
2270 | flush_slab(s, c); | |
2271 | ||
59a09917 | 2272 | unfreeze_partials(s, c); |
49e22585 | 2273 | } |
81819f0f CL |
2274 | } |
2275 | ||
2276 | static void flush_cpu_slab(void *d) | |
2277 | { | |
2278 | struct kmem_cache *s = d; | |
81819f0f | 2279 | |
dfb4f096 | 2280 | __flush_cpu_slab(s, smp_processor_id()); |
81819f0f CL |
2281 | } |
2282 | ||
a8364d55 GBY |
2283 | static bool has_cpu_slab(int cpu, void *info) |
2284 | { | |
2285 | struct kmem_cache *s = info; | |
2286 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu); | |
2287 | ||
02e1a9cd | 2288 | return c->page || c->partial; |
a8364d55 GBY |
2289 | } |
2290 | ||
81819f0f CL |
2291 | static void flush_all(struct kmem_cache *s) |
2292 | { | |
a8364d55 | 2293 | on_each_cpu_cond(has_cpu_slab, flush_cpu_slab, s, 1, GFP_ATOMIC); |
81819f0f CL |
2294 | } |
2295 | ||
dfb4f096 CL |
2296 | /* |
2297 | * Check if the objects in a per cpu structure fit numa | |
2298 | * locality expectations. | |
2299 | */ | |
57d437d2 | 2300 | static inline int node_match(struct page *page, int node) |
dfb4f096 CL |
2301 | { |
2302 | #ifdef CONFIG_NUMA | |
4d7868e6 | 2303 | if (!page || (node != NUMA_NO_NODE && page_to_nid(page) != node)) |
dfb4f096 CL |
2304 | return 0; |
2305 | #endif | |
2306 | return 1; | |
2307 | } | |
2308 | ||
9a02d699 | 2309 | #ifdef CONFIG_SLUB_DEBUG |
781b2ba6 PE |
2310 | static int count_free(struct page *page) |
2311 | { | |
2312 | return page->objects - page->inuse; | |
2313 | } | |
2314 | ||
9a02d699 DR |
2315 | static inline unsigned long node_nr_objs(struct kmem_cache_node *n) |
2316 | { | |
2317 | return atomic_long_read(&n->total_objects); | |
2318 | } | |
2319 | #endif /* CONFIG_SLUB_DEBUG */ | |
2320 | ||
2321 | #if defined(CONFIG_SLUB_DEBUG) || defined(CONFIG_SYSFS) | |
781b2ba6 PE |
2322 | static unsigned long count_partial(struct kmem_cache_node *n, |
2323 | int (*get_count)(struct page *)) | |
2324 | { | |
2325 | unsigned long flags; | |
2326 | unsigned long x = 0; | |
2327 | struct page *page; | |
2328 | ||
2329 | spin_lock_irqsave(&n->list_lock, flags); | |
2330 | list_for_each_entry(page, &n->partial, lru) | |
2331 | x += get_count(page); | |
2332 | spin_unlock_irqrestore(&n->list_lock, flags); | |
2333 | return x; | |
2334 | } | |
9a02d699 | 2335 | #endif /* CONFIG_SLUB_DEBUG || CONFIG_SYSFS */ |
26c02cf0 | 2336 | |
781b2ba6 PE |
2337 | static noinline void |
2338 | slab_out_of_memory(struct kmem_cache *s, gfp_t gfpflags, int nid) | |
2339 | { | |
9a02d699 DR |
2340 | #ifdef CONFIG_SLUB_DEBUG |
2341 | static DEFINE_RATELIMIT_STATE(slub_oom_rs, DEFAULT_RATELIMIT_INTERVAL, | |
2342 | DEFAULT_RATELIMIT_BURST); | |
781b2ba6 | 2343 | int node; |
fa45dc25 | 2344 | struct kmem_cache_node *n; |
781b2ba6 | 2345 | |
9a02d699 DR |
2346 | if ((gfpflags & __GFP_NOWARN) || !__ratelimit(&slub_oom_rs)) |
2347 | return; | |
2348 | ||
5b3810e5 VB |
2349 | pr_warn("SLUB: Unable to allocate memory on node %d, gfp=%#x(%pGg)\n", |
2350 | nid, gfpflags, &gfpflags); | |
f9f58285 FF |
2351 | pr_warn(" cache: %s, object size: %d, buffer size: %d, default order: %d, min order: %d\n", |
2352 | s->name, s->object_size, s->size, oo_order(s->oo), | |
2353 | oo_order(s->min)); | |
781b2ba6 | 2354 | |
3b0efdfa | 2355 | if (oo_order(s->min) > get_order(s->object_size)) |
f9f58285 FF |
2356 | pr_warn(" %s debugging increased min order, use slub_debug=O to disable.\n", |
2357 | s->name); | |
fa5ec8a1 | 2358 | |
fa45dc25 | 2359 | for_each_kmem_cache_node(s, node, n) { |
781b2ba6 PE |
2360 | unsigned long nr_slabs; |
2361 | unsigned long nr_objs; | |
2362 | unsigned long nr_free; | |
2363 | ||
26c02cf0 AB |
2364 | nr_free = count_partial(n, count_free); |
2365 | nr_slabs = node_nr_slabs(n); | |
2366 | nr_objs = node_nr_objs(n); | |
781b2ba6 | 2367 | |
f9f58285 | 2368 | pr_warn(" node %d: slabs: %ld, objs: %ld, free: %ld\n", |
781b2ba6 PE |
2369 | node, nr_slabs, nr_objs, nr_free); |
2370 | } | |
9a02d699 | 2371 | #endif |
781b2ba6 PE |
2372 | } |
2373 | ||
497b66f2 CL |
2374 | static inline void *new_slab_objects(struct kmem_cache *s, gfp_t flags, |
2375 | int node, struct kmem_cache_cpu **pc) | |
2376 | { | |
6faa6833 | 2377 | void *freelist; |
188fd063 CL |
2378 | struct kmem_cache_cpu *c = *pc; |
2379 | struct page *page; | |
497b66f2 | 2380 | |
188fd063 | 2381 | freelist = get_partial(s, flags, node, c); |
497b66f2 | 2382 | |
188fd063 CL |
2383 | if (freelist) |
2384 | return freelist; | |
2385 | ||
2386 | page = new_slab(s, flags, node); | |
497b66f2 | 2387 | if (page) { |
7c8e0181 | 2388 | c = raw_cpu_ptr(s->cpu_slab); |
497b66f2 CL |
2389 | if (c->page) |
2390 | flush_slab(s, c); | |
2391 | ||
2392 | /* | |
2393 | * No other reference to the page yet so we can | |
2394 | * muck around with it freely without cmpxchg | |
2395 | */ | |
6faa6833 | 2396 | freelist = page->freelist; |
497b66f2 CL |
2397 | page->freelist = NULL; |
2398 | ||
2399 | stat(s, ALLOC_SLAB); | |
497b66f2 CL |
2400 | c->page = page; |
2401 | *pc = c; | |
2402 | } else | |
6faa6833 | 2403 | freelist = NULL; |
497b66f2 | 2404 | |
6faa6833 | 2405 | return freelist; |
497b66f2 CL |
2406 | } |
2407 | ||
072bb0aa MG |
2408 | static inline bool pfmemalloc_match(struct page *page, gfp_t gfpflags) |
2409 | { | |
2410 | if (unlikely(PageSlabPfmemalloc(page))) | |
2411 | return gfp_pfmemalloc_allowed(gfpflags); | |
2412 | ||
2413 | return true; | |
2414 | } | |
2415 | ||
213eeb9f | 2416 | /* |
d0e0ac97 CG |
2417 | * Check the page->freelist of a page and either transfer the freelist to the |
2418 | * per cpu freelist or deactivate the page. | |
213eeb9f CL |
2419 | * |
2420 | * The page is still frozen if the return value is not NULL. | |
2421 | * | |
2422 | * If this function returns NULL then the page has been unfrozen. | |
d24ac77f JK |
2423 | * |
2424 | * This function must be called with interrupt disabled. | |
213eeb9f CL |
2425 | */ |
2426 | static inline void *get_freelist(struct kmem_cache *s, struct page *page) | |
2427 | { | |
2428 | struct page new; | |
2429 | unsigned long counters; | |
2430 | void *freelist; | |
2431 | ||
2432 | do { | |
2433 | freelist = page->freelist; | |
2434 | counters = page->counters; | |
6faa6833 | 2435 | |
213eeb9f | 2436 | new.counters = counters; |
a0132ac0 | 2437 | VM_BUG_ON(!new.frozen); |
213eeb9f CL |
2438 | |
2439 | new.inuse = page->objects; | |
2440 | new.frozen = freelist != NULL; | |
2441 | ||
d24ac77f | 2442 | } while (!__cmpxchg_double_slab(s, page, |
213eeb9f CL |
2443 | freelist, counters, |
2444 | NULL, new.counters, | |
2445 | "get_freelist")); | |
2446 | ||
2447 | return freelist; | |
2448 | } | |
2449 | ||
81819f0f | 2450 | /* |
894b8788 CL |
2451 | * Slow path. The lockless freelist is empty or we need to perform |
2452 | * debugging duties. | |
2453 | * | |
894b8788 CL |
2454 | * Processing is still very fast if new objects have been freed to the |
2455 | * regular freelist. In that case we simply take over the regular freelist | |
2456 | * as the lockless freelist and zap the regular freelist. | |
81819f0f | 2457 | * |
894b8788 CL |
2458 | * If that is not working then we fall back to the partial lists. We take the |
2459 | * first element of the freelist as the object to allocate now and move the | |
2460 | * rest of the freelist to the lockless freelist. | |
81819f0f | 2461 | * |
894b8788 | 2462 | * And if we were unable to get a new slab from the partial slab lists then |
6446faa2 CL |
2463 | * we need to allocate a new slab. This is the slowest path since it involves |
2464 | * a call to the page allocator and the setup of a new slab. | |
a380a3c7 CL |
2465 | * |
2466 | * Version of __slab_alloc to use when we know that interrupts are | |
2467 | * already disabled (which is the case for bulk allocation). | |
81819f0f | 2468 | */ |
a380a3c7 | 2469 | static void *___slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node, |
ce71e27c | 2470 | unsigned long addr, struct kmem_cache_cpu *c) |
81819f0f | 2471 | { |
6faa6833 | 2472 | void *freelist; |
f6e7def7 | 2473 | struct page *page; |
81819f0f | 2474 | |
f6e7def7 CL |
2475 | page = c->page; |
2476 | if (!page) | |
81819f0f | 2477 | goto new_slab; |
49e22585 | 2478 | redo: |
6faa6833 | 2479 | |
57d437d2 | 2480 | if (unlikely(!node_match(page, node))) { |
a561ce00 JK |
2481 | int searchnode = node; |
2482 | ||
2483 | if (node != NUMA_NO_NODE && !node_present_pages(node)) | |
2484 | searchnode = node_to_mem_node(node); | |
2485 | ||
2486 | if (unlikely(!node_match(page, searchnode))) { | |
2487 | stat(s, ALLOC_NODE_MISMATCH); | |
2488 | deactivate_slab(s, page, c->freelist); | |
2489 | c->page = NULL; | |
2490 | c->freelist = NULL; | |
2491 | goto new_slab; | |
2492 | } | |
fc59c053 | 2493 | } |
6446faa2 | 2494 | |
072bb0aa MG |
2495 | /* |
2496 | * By rights, we should be searching for a slab page that was | |
2497 | * PFMEMALLOC but right now, we are losing the pfmemalloc | |
2498 | * information when the page leaves the per-cpu allocator | |
2499 | */ | |
2500 | if (unlikely(!pfmemalloc_match(page, gfpflags))) { | |
2501 | deactivate_slab(s, page, c->freelist); | |
2502 | c->page = NULL; | |
2503 | c->freelist = NULL; | |
2504 | goto new_slab; | |
2505 | } | |
2506 | ||
73736e03 | 2507 | /* must check again c->freelist in case of cpu migration or IRQ */ |
6faa6833 CL |
2508 | freelist = c->freelist; |
2509 | if (freelist) | |
73736e03 | 2510 | goto load_freelist; |
03e404af | 2511 | |
f6e7def7 | 2512 | freelist = get_freelist(s, page); |
6446faa2 | 2513 | |
6faa6833 | 2514 | if (!freelist) { |
03e404af CL |
2515 | c->page = NULL; |
2516 | stat(s, DEACTIVATE_BYPASS); | |
fc59c053 | 2517 | goto new_slab; |
03e404af | 2518 | } |
6446faa2 | 2519 | |
84e554e6 | 2520 | stat(s, ALLOC_REFILL); |
6446faa2 | 2521 | |
894b8788 | 2522 | load_freelist: |
507effea CL |
2523 | /* |
2524 | * freelist is pointing to the list of objects to be used. | |
2525 | * page is pointing to the page from which the objects are obtained. | |
2526 | * That page must be frozen for per cpu allocations to work. | |
2527 | */ | |
a0132ac0 | 2528 | VM_BUG_ON(!c->page->frozen); |
6faa6833 | 2529 | c->freelist = get_freepointer(s, freelist); |
8a5ec0ba | 2530 | c->tid = next_tid(c->tid); |
6faa6833 | 2531 | return freelist; |
81819f0f | 2532 | |
81819f0f | 2533 | new_slab: |
2cfb7455 | 2534 | |
49e22585 | 2535 | if (c->partial) { |
f6e7def7 CL |
2536 | page = c->page = c->partial; |
2537 | c->partial = page->next; | |
49e22585 CL |
2538 | stat(s, CPU_PARTIAL_ALLOC); |
2539 | c->freelist = NULL; | |
2540 | goto redo; | |
81819f0f CL |
2541 | } |
2542 | ||
188fd063 | 2543 | freelist = new_slab_objects(s, gfpflags, node, &c); |
01ad8a7b | 2544 | |
f4697436 | 2545 | if (unlikely(!freelist)) { |
9a02d699 | 2546 | slab_out_of_memory(s, gfpflags, node); |
f4697436 | 2547 | return NULL; |
81819f0f | 2548 | } |
2cfb7455 | 2549 | |
f6e7def7 | 2550 | page = c->page; |
5091b74a | 2551 | if (likely(!kmem_cache_debug(s) && pfmemalloc_match(page, gfpflags))) |
4b6f0750 | 2552 | goto load_freelist; |
2cfb7455 | 2553 | |
497b66f2 | 2554 | /* Only entered in the debug case */ |
d0e0ac97 CG |
2555 | if (kmem_cache_debug(s) && |
2556 | !alloc_debug_processing(s, page, freelist, addr)) | |
497b66f2 | 2557 | goto new_slab; /* Slab failed checks. Next slab needed */ |
894b8788 | 2558 | |
f6e7def7 | 2559 | deactivate_slab(s, page, get_freepointer(s, freelist)); |
c17dda40 CL |
2560 | c->page = NULL; |
2561 | c->freelist = NULL; | |
6faa6833 | 2562 | return freelist; |
894b8788 CL |
2563 | } |
2564 | ||
a380a3c7 CL |
2565 | /* |
2566 | * Another one that disabled interrupt and compensates for possible | |
2567 | * cpu changes by refetching the per cpu area pointer. | |
2568 | */ | |
2569 | static void *__slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node, | |
2570 | unsigned long addr, struct kmem_cache_cpu *c) | |
2571 | { | |
2572 | void *p; | |
2573 | unsigned long flags; | |
2574 | ||
2575 | local_irq_save(flags); | |
2576 | #ifdef CONFIG_PREEMPT | |
2577 | /* | |
2578 | * We may have been preempted and rescheduled on a different | |
2579 | * cpu before disabling interrupts. Need to reload cpu area | |
2580 | * pointer. | |
2581 | */ | |
2582 | c = this_cpu_ptr(s->cpu_slab); | |
2583 | #endif | |
2584 | ||
2585 | p = ___slab_alloc(s, gfpflags, node, addr, c); | |
2586 | local_irq_restore(flags); | |
2587 | return p; | |
2588 | } | |
2589 | ||
894b8788 CL |
2590 | /* |
2591 | * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc) | |
2592 | * have the fastpath folded into their functions. So no function call | |
2593 | * overhead for requests that can be satisfied on the fastpath. | |
2594 | * | |
2595 | * The fastpath works by first checking if the lockless freelist can be used. | |
2596 | * If not then __slab_alloc is called for slow processing. | |
2597 | * | |
2598 | * Otherwise we can simply pick the next object from the lockless free list. | |
2599 | */ | |
2b847c3c | 2600 | static __always_inline void *slab_alloc_node(struct kmem_cache *s, |
ce71e27c | 2601 | gfp_t gfpflags, int node, unsigned long addr) |
894b8788 | 2602 | { |
03ec0ed5 | 2603 | void *object; |
dfb4f096 | 2604 | struct kmem_cache_cpu *c; |
57d437d2 | 2605 | struct page *page; |
8a5ec0ba | 2606 | unsigned long tid; |
1f84260c | 2607 | |
8135be5a VD |
2608 | s = slab_pre_alloc_hook(s, gfpflags); |
2609 | if (!s) | |
773ff60e | 2610 | return NULL; |
8a5ec0ba | 2611 | redo: |
8a5ec0ba CL |
2612 | /* |
2613 | * Must read kmem_cache cpu data via this cpu ptr. Preemption is | |
2614 | * enabled. We may switch back and forth between cpus while | |
2615 | * reading from one cpu area. That does not matter as long | |
2616 | * as we end up on the original cpu again when doing the cmpxchg. | |
7cccd80b | 2617 | * |
9aabf810 JK |
2618 | * We should guarantee that tid and kmem_cache are retrieved on |
2619 | * the same cpu. It could be different if CONFIG_PREEMPT so we need | |
2620 | * to check if it is matched or not. | |
8a5ec0ba | 2621 | */ |
9aabf810 JK |
2622 | do { |
2623 | tid = this_cpu_read(s->cpu_slab->tid); | |
2624 | c = raw_cpu_ptr(s->cpu_slab); | |
859b7a0e MR |
2625 | } while (IS_ENABLED(CONFIG_PREEMPT) && |
2626 | unlikely(tid != READ_ONCE(c->tid))); | |
9aabf810 JK |
2627 | |
2628 | /* | |
2629 | * Irqless object alloc/free algorithm used here depends on sequence | |
2630 | * of fetching cpu_slab's data. tid should be fetched before anything | |
2631 | * on c to guarantee that object and page associated with previous tid | |
2632 | * won't be used with current tid. If we fetch tid first, object and | |
2633 | * page could be one associated with next tid and our alloc/free | |
2634 | * request will be failed. In this case, we will retry. So, no problem. | |
2635 | */ | |
2636 | barrier(); | |
8a5ec0ba | 2637 | |
8a5ec0ba CL |
2638 | /* |
2639 | * The transaction ids are globally unique per cpu and per operation on | |
2640 | * a per cpu queue. Thus they can be guarantee that the cmpxchg_double | |
2641 | * occurs on the right processor and that there was no operation on the | |
2642 | * linked list in between. | |
2643 | */ | |
8a5ec0ba | 2644 | |
9dfc6e68 | 2645 | object = c->freelist; |
57d437d2 | 2646 | page = c->page; |
8eae1492 | 2647 | if (unlikely(!object || !node_match(page, node))) { |
dfb4f096 | 2648 | object = __slab_alloc(s, gfpflags, node, addr, c); |
8eae1492 DH |
2649 | stat(s, ALLOC_SLOWPATH); |
2650 | } else { | |
0ad9500e ED |
2651 | void *next_object = get_freepointer_safe(s, object); |
2652 | ||
8a5ec0ba | 2653 | /* |
25985edc | 2654 | * The cmpxchg will only match if there was no additional |
8a5ec0ba CL |
2655 | * operation and if we are on the right processor. |
2656 | * | |
d0e0ac97 CG |
2657 | * The cmpxchg does the following atomically (without lock |
2658 | * semantics!) | |
8a5ec0ba CL |
2659 | * 1. Relocate first pointer to the current per cpu area. |
2660 | * 2. Verify that tid and freelist have not been changed | |
2661 | * 3. If they were not changed replace tid and freelist | |
2662 | * | |
d0e0ac97 CG |
2663 | * Since this is without lock semantics the protection is only |
2664 | * against code executing on this cpu *not* from access by | |
2665 | * other cpus. | |
8a5ec0ba | 2666 | */ |
933393f5 | 2667 | if (unlikely(!this_cpu_cmpxchg_double( |
8a5ec0ba CL |
2668 | s->cpu_slab->freelist, s->cpu_slab->tid, |
2669 | object, tid, | |
0ad9500e | 2670 | next_object, next_tid(tid)))) { |
8a5ec0ba CL |
2671 | |
2672 | note_cmpxchg_failure("slab_alloc", s, tid); | |
2673 | goto redo; | |
2674 | } | |
0ad9500e | 2675 | prefetch_freepointer(s, next_object); |
84e554e6 | 2676 | stat(s, ALLOC_FASTPATH); |
894b8788 | 2677 | } |
8a5ec0ba | 2678 | |
74e2134f | 2679 | if (unlikely(gfpflags & __GFP_ZERO) && object) |
3b0efdfa | 2680 | memset(object, 0, s->object_size); |
d07dbea4 | 2681 | |
03ec0ed5 | 2682 | slab_post_alloc_hook(s, gfpflags, 1, &object); |
5a896d9e | 2683 | |
894b8788 | 2684 | return object; |
81819f0f CL |
2685 | } |
2686 | ||
2b847c3c EG |
2687 | static __always_inline void *slab_alloc(struct kmem_cache *s, |
2688 | gfp_t gfpflags, unsigned long addr) | |
2689 | { | |
2690 | return slab_alloc_node(s, gfpflags, NUMA_NO_NODE, addr); | |
2691 | } | |
2692 | ||
81819f0f CL |
2693 | void *kmem_cache_alloc(struct kmem_cache *s, gfp_t gfpflags) |
2694 | { | |
2b847c3c | 2695 | void *ret = slab_alloc(s, gfpflags, _RET_IP_); |
5b882be4 | 2696 | |
d0e0ac97 CG |
2697 | trace_kmem_cache_alloc(_RET_IP_, ret, s->object_size, |
2698 | s->size, gfpflags); | |
5b882be4 EGM |
2699 | |
2700 | return ret; | |
81819f0f CL |
2701 | } |
2702 | EXPORT_SYMBOL(kmem_cache_alloc); | |
2703 | ||
0f24f128 | 2704 | #ifdef CONFIG_TRACING |
4a92379b RK |
2705 | void *kmem_cache_alloc_trace(struct kmem_cache *s, gfp_t gfpflags, size_t size) |
2706 | { | |
2b847c3c | 2707 | void *ret = slab_alloc(s, gfpflags, _RET_IP_); |
4a92379b | 2708 | trace_kmalloc(_RET_IP_, ret, size, s->size, gfpflags); |
505f5dcb | 2709 | kasan_kmalloc(s, ret, size, gfpflags); |
4a92379b RK |
2710 | return ret; |
2711 | } | |
2712 | EXPORT_SYMBOL(kmem_cache_alloc_trace); | |
5b882be4 EGM |
2713 | #endif |
2714 | ||
81819f0f CL |
2715 | #ifdef CONFIG_NUMA |
2716 | void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t gfpflags, int node) | |
2717 | { | |
2b847c3c | 2718 | void *ret = slab_alloc_node(s, gfpflags, node, _RET_IP_); |
5b882be4 | 2719 | |
ca2b84cb | 2720 | trace_kmem_cache_alloc_node(_RET_IP_, ret, |
3b0efdfa | 2721 | s->object_size, s->size, gfpflags, node); |
5b882be4 EGM |
2722 | |
2723 | return ret; | |
81819f0f CL |
2724 | } |
2725 | EXPORT_SYMBOL(kmem_cache_alloc_node); | |
81819f0f | 2726 | |
0f24f128 | 2727 | #ifdef CONFIG_TRACING |
4a92379b | 2728 | void *kmem_cache_alloc_node_trace(struct kmem_cache *s, |
5b882be4 | 2729 | gfp_t gfpflags, |
4a92379b | 2730 | int node, size_t size) |
5b882be4 | 2731 | { |
2b847c3c | 2732 | void *ret = slab_alloc_node(s, gfpflags, node, _RET_IP_); |
4a92379b RK |
2733 | |
2734 | trace_kmalloc_node(_RET_IP_, ret, | |
2735 | size, s->size, gfpflags, node); | |
0316bec2 | 2736 | |
505f5dcb | 2737 | kasan_kmalloc(s, ret, size, gfpflags); |
4a92379b | 2738 | return ret; |
5b882be4 | 2739 | } |
4a92379b | 2740 | EXPORT_SYMBOL(kmem_cache_alloc_node_trace); |
5b882be4 | 2741 | #endif |
5d1f57e4 | 2742 | #endif |
5b882be4 | 2743 | |
81819f0f | 2744 | /* |
94e4d712 | 2745 | * Slow path handling. This may still be called frequently since objects |
894b8788 | 2746 | * have a longer lifetime than the cpu slabs in most processing loads. |
81819f0f | 2747 | * |
894b8788 CL |
2748 | * So we still attempt to reduce cache line usage. Just take the slab |
2749 | * lock and free the item. If there is no additional partial page | |
2750 | * handling required then we can return immediately. | |
81819f0f | 2751 | */ |
894b8788 | 2752 | static void __slab_free(struct kmem_cache *s, struct page *page, |
81084651 JDB |
2753 | void *head, void *tail, int cnt, |
2754 | unsigned long addr) | |
2755 | ||
81819f0f CL |
2756 | { |
2757 | void *prior; | |
2cfb7455 | 2758 | int was_frozen; |
2cfb7455 CL |
2759 | struct page new; |
2760 | unsigned long counters; | |
2761 | struct kmem_cache_node *n = NULL; | |
61728d1e | 2762 | unsigned long uninitialized_var(flags); |
81819f0f | 2763 | |
8a5ec0ba | 2764 | stat(s, FREE_SLOWPATH); |
81819f0f | 2765 | |
19c7ff9e | 2766 | if (kmem_cache_debug(s) && |
282acb43 | 2767 | !free_debug_processing(s, page, head, tail, cnt, addr)) |
80f08c19 | 2768 | return; |
6446faa2 | 2769 | |
2cfb7455 | 2770 | do { |
837d678d JK |
2771 | if (unlikely(n)) { |
2772 | spin_unlock_irqrestore(&n->list_lock, flags); | |
2773 | n = NULL; | |
2774 | } | |
2cfb7455 CL |
2775 | prior = page->freelist; |
2776 | counters = page->counters; | |
81084651 | 2777 | set_freepointer(s, tail, prior); |
2cfb7455 CL |
2778 | new.counters = counters; |
2779 | was_frozen = new.frozen; | |
81084651 | 2780 | new.inuse -= cnt; |
837d678d | 2781 | if ((!new.inuse || !prior) && !was_frozen) { |
49e22585 | 2782 | |
c65c1877 | 2783 | if (kmem_cache_has_cpu_partial(s) && !prior) { |
49e22585 CL |
2784 | |
2785 | /* | |
d0e0ac97 CG |
2786 | * Slab was on no list before and will be |
2787 | * partially empty | |
2788 | * We can defer the list move and instead | |
2789 | * freeze it. | |
49e22585 CL |
2790 | */ |
2791 | new.frozen = 1; | |
2792 | ||
c65c1877 | 2793 | } else { /* Needs to be taken off a list */ |
49e22585 | 2794 | |
b455def2 | 2795 | n = get_node(s, page_to_nid(page)); |
49e22585 CL |
2796 | /* |
2797 | * Speculatively acquire the list_lock. | |
2798 | * If the cmpxchg does not succeed then we may | |
2799 | * drop the list_lock without any processing. | |
2800 | * | |
2801 | * Otherwise the list_lock will synchronize with | |
2802 | * other processors updating the list of slabs. | |
2803 | */ | |
2804 | spin_lock_irqsave(&n->list_lock, flags); | |
2805 | ||
2806 | } | |
2cfb7455 | 2807 | } |
81819f0f | 2808 | |
2cfb7455 CL |
2809 | } while (!cmpxchg_double_slab(s, page, |
2810 | prior, counters, | |
81084651 | 2811 | head, new.counters, |
2cfb7455 | 2812 | "__slab_free")); |
81819f0f | 2813 | |
2cfb7455 | 2814 | if (likely(!n)) { |
49e22585 CL |
2815 | |
2816 | /* | |
2817 | * If we just froze the page then put it onto the | |
2818 | * per cpu partial list. | |
2819 | */ | |
8028dcea | 2820 | if (new.frozen && !was_frozen) { |
49e22585 | 2821 | put_cpu_partial(s, page, 1); |
8028dcea AS |
2822 | stat(s, CPU_PARTIAL_FREE); |
2823 | } | |
49e22585 | 2824 | /* |
2cfb7455 CL |
2825 | * The list lock was not taken therefore no list |
2826 | * activity can be necessary. | |
2827 | */ | |
b455def2 L |
2828 | if (was_frozen) |
2829 | stat(s, FREE_FROZEN); | |
2830 | return; | |
2831 | } | |
81819f0f | 2832 | |
8a5b20ae | 2833 | if (unlikely(!new.inuse && n->nr_partial >= s->min_partial)) |
837d678d JK |
2834 | goto slab_empty; |
2835 | ||
81819f0f | 2836 | /* |
837d678d JK |
2837 | * Objects left in the slab. If it was not on the partial list before |
2838 | * then add it. | |
81819f0f | 2839 | */ |
345c905d JK |
2840 | if (!kmem_cache_has_cpu_partial(s) && unlikely(!prior)) { |
2841 | if (kmem_cache_debug(s)) | |
c65c1877 | 2842 | remove_full(s, n, page); |
837d678d JK |
2843 | add_partial(n, page, DEACTIVATE_TO_TAIL); |
2844 | stat(s, FREE_ADD_PARTIAL); | |
8ff12cfc | 2845 | } |
80f08c19 | 2846 | spin_unlock_irqrestore(&n->list_lock, flags); |
81819f0f CL |
2847 | return; |
2848 | ||
2849 | slab_empty: | |
a973e9dd | 2850 | if (prior) { |
81819f0f | 2851 | /* |
6fbabb20 | 2852 | * Slab on the partial list. |
81819f0f | 2853 | */ |
5cc6eee8 | 2854 | remove_partial(n, page); |
84e554e6 | 2855 | stat(s, FREE_REMOVE_PARTIAL); |
c65c1877 | 2856 | } else { |
6fbabb20 | 2857 | /* Slab must be on the full list */ |
c65c1877 PZ |
2858 | remove_full(s, n, page); |
2859 | } | |
2cfb7455 | 2860 | |
80f08c19 | 2861 | spin_unlock_irqrestore(&n->list_lock, flags); |
84e554e6 | 2862 | stat(s, FREE_SLAB); |
81819f0f | 2863 | discard_slab(s, page); |
81819f0f CL |
2864 | } |
2865 | ||
894b8788 CL |
2866 | /* |
2867 | * Fastpath with forced inlining to produce a kfree and kmem_cache_free that | |
2868 | * can perform fastpath freeing without additional function calls. | |
2869 | * | |
2870 | * The fastpath is only possible if we are freeing to the current cpu slab | |
2871 | * of this processor. This typically the case if we have just allocated | |
2872 | * the item before. | |
2873 | * | |
2874 | * If fastpath is not possible then fall back to __slab_free where we deal | |
2875 | * with all sorts of special processing. | |
81084651 JDB |
2876 | * |
2877 | * Bulk free of a freelist with several objects (all pointing to the | |
2878 | * same page) possible by specifying head and tail ptr, plus objects | |
2879 | * count (cnt). Bulk free indicated by tail pointer being set. | |
894b8788 | 2880 | */ |
81084651 JDB |
2881 | static __always_inline void slab_free(struct kmem_cache *s, struct page *page, |
2882 | void *head, void *tail, int cnt, | |
2883 | unsigned long addr) | |
894b8788 | 2884 | { |
81084651 | 2885 | void *tail_obj = tail ? : head; |
dfb4f096 | 2886 | struct kmem_cache_cpu *c; |
8a5ec0ba | 2887 | unsigned long tid; |
1f84260c | 2888 | |
81084651 | 2889 | slab_free_freelist_hook(s, head, tail); |
c016b0bd | 2890 | |
8a5ec0ba CL |
2891 | redo: |
2892 | /* | |
2893 | * Determine the currently cpus per cpu slab. | |
2894 | * The cpu may change afterward. However that does not matter since | |
2895 | * data is retrieved via this pointer. If we are on the same cpu | |
2ae44005 | 2896 | * during the cmpxchg then the free will succeed. |
8a5ec0ba | 2897 | */ |
9aabf810 JK |
2898 | do { |
2899 | tid = this_cpu_read(s->cpu_slab->tid); | |
2900 | c = raw_cpu_ptr(s->cpu_slab); | |
859b7a0e MR |
2901 | } while (IS_ENABLED(CONFIG_PREEMPT) && |
2902 | unlikely(tid != READ_ONCE(c->tid))); | |
c016b0bd | 2903 | |
9aabf810 JK |
2904 | /* Same with comment on barrier() in slab_alloc_node() */ |
2905 | barrier(); | |
c016b0bd | 2906 | |
442b06bc | 2907 | if (likely(page == c->page)) { |
81084651 | 2908 | set_freepointer(s, tail_obj, c->freelist); |
8a5ec0ba | 2909 | |
933393f5 | 2910 | if (unlikely(!this_cpu_cmpxchg_double( |
8a5ec0ba CL |
2911 | s->cpu_slab->freelist, s->cpu_slab->tid, |
2912 | c->freelist, tid, | |
81084651 | 2913 | head, next_tid(tid)))) { |
8a5ec0ba CL |
2914 | |
2915 | note_cmpxchg_failure("slab_free", s, tid); | |
2916 | goto redo; | |
2917 | } | |
84e554e6 | 2918 | stat(s, FREE_FASTPATH); |
894b8788 | 2919 | } else |
81084651 | 2920 | __slab_free(s, page, head, tail_obj, cnt, addr); |
894b8788 | 2921 | |
894b8788 CL |
2922 | } |
2923 | ||
81819f0f CL |
2924 | void kmem_cache_free(struct kmem_cache *s, void *x) |
2925 | { | |
b9ce5ef4 GC |
2926 | s = cache_from_obj(s, x); |
2927 | if (!s) | |
79576102 | 2928 | return; |
81084651 | 2929 | slab_free(s, virt_to_head_page(x), x, NULL, 1, _RET_IP_); |
ca2b84cb | 2930 | trace_kmem_cache_free(_RET_IP_, x); |
81819f0f CL |
2931 | } |
2932 | EXPORT_SYMBOL(kmem_cache_free); | |
2933 | ||
d0ecd894 | 2934 | struct detached_freelist { |
fbd02630 | 2935 | struct page *page; |
d0ecd894 JDB |
2936 | void *tail; |
2937 | void *freelist; | |
2938 | int cnt; | |
376bf125 | 2939 | struct kmem_cache *s; |
d0ecd894 | 2940 | }; |
fbd02630 | 2941 | |
d0ecd894 JDB |
2942 | /* |
2943 | * This function progressively scans the array with free objects (with | |
2944 | * a limited look ahead) and extract objects belonging to the same | |
2945 | * page. It builds a detached freelist directly within the given | |
2946 | * page/objects. This can happen without any need for | |
2947 | * synchronization, because the objects are owned by running process. | |
2948 | * The freelist is build up as a single linked list in the objects. | |
2949 | * The idea is, that this detached freelist can then be bulk | |
2950 | * transferred to the real freelist(s), but only requiring a single | |
2951 | * synchronization primitive. Look ahead in the array is limited due | |
2952 | * to performance reasons. | |
2953 | */ | |
376bf125 JDB |
2954 | static inline |
2955 | int build_detached_freelist(struct kmem_cache *s, size_t size, | |
2956 | void **p, struct detached_freelist *df) | |
d0ecd894 JDB |
2957 | { |
2958 | size_t first_skipped_index = 0; | |
2959 | int lookahead = 3; | |
2960 | void *object; | |
ca257195 | 2961 | struct page *page; |
fbd02630 | 2962 | |
d0ecd894 JDB |
2963 | /* Always re-init detached_freelist */ |
2964 | df->page = NULL; | |
fbd02630 | 2965 | |
d0ecd894 JDB |
2966 | do { |
2967 | object = p[--size]; | |
ca257195 | 2968 | /* Do we need !ZERO_OR_NULL_PTR(object) here? (for kfree) */ |
d0ecd894 | 2969 | } while (!object && size); |
3eed034d | 2970 | |
d0ecd894 JDB |
2971 | if (!object) |
2972 | return 0; | |
fbd02630 | 2973 | |
ca257195 JDB |
2974 | page = virt_to_head_page(object); |
2975 | if (!s) { | |
2976 | /* Handle kalloc'ed objects */ | |
2977 | if (unlikely(!PageSlab(page))) { | |
2978 | BUG_ON(!PageCompound(page)); | |
2979 | kfree_hook(object); | |
4949148a | 2980 | __free_pages(page, compound_order(page)); |
ca257195 JDB |
2981 | p[size] = NULL; /* mark object processed */ |
2982 | return size; | |
2983 | } | |
2984 | /* Derive kmem_cache from object */ | |
2985 | df->s = page->slab_cache; | |
2986 | } else { | |
2987 | df->s = cache_from_obj(s, object); /* Support for memcg */ | |
2988 | } | |
376bf125 | 2989 | |
d0ecd894 | 2990 | /* Start new detached freelist */ |
ca257195 | 2991 | df->page = page; |
376bf125 | 2992 | set_freepointer(df->s, object, NULL); |
d0ecd894 JDB |
2993 | df->tail = object; |
2994 | df->freelist = object; | |
2995 | p[size] = NULL; /* mark object processed */ | |
2996 | df->cnt = 1; | |
2997 | ||
2998 | while (size) { | |
2999 | object = p[--size]; | |
3000 | if (!object) | |
3001 | continue; /* Skip processed objects */ | |
3002 | ||
3003 | /* df->page is always set at this point */ | |
3004 | if (df->page == virt_to_head_page(object)) { | |
3005 | /* Opportunity build freelist */ | |
376bf125 | 3006 | set_freepointer(df->s, object, df->freelist); |
d0ecd894 JDB |
3007 | df->freelist = object; |
3008 | df->cnt++; | |
3009 | p[size] = NULL; /* mark object processed */ | |
3010 | ||
3011 | continue; | |
fbd02630 | 3012 | } |
d0ecd894 JDB |
3013 | |
3014 | /* Limit look ahead search */ | |
3015 | if (!--lookahead) | |
3016 | break; | |
3017 | ||
3018 | if (!first_skipped_index) | |
3019 | first_skipped_index = size + 1; | |
fbd02630 | 3020 | } |
d0ecd894 JDB |
3021 | |
3022 | return first_skipped_index; | |
3023 | } | |
3024 | ||
d0ecd894 | 3025 | /* Note that interrupts must be enabled when calling this function. */ |
376bf125 | 3026 | void kmem_cache_free_bulk(struct kmem_cache *s, size_t size, void **p) |
d0ecd894 JDB |
3027 | { |
3028 | if (WARN_ON(!size)) | |
3029 | return; | |
3030 | ||
3031 | do { | |
3032 | struct detached_freelist df; | |
3033 | ||
3034 | size = build_detached_freelist(s, size, p, &df); | |
3035 | if (unlikely(!df.page)) | |
3036 | continue; | |
3037 | ||
376bf125 | 3038 | slab_free(df.s, df.page, df.freelist, df.tail, df.cnt,_RET_IP_); |
d0ecd894 | 3039 | } while (likely(size)); |
484748f0 CL |
3040 | } |
3041 | EXPORT_SYMBOL(kmem_cache_free_bulk); | |
3042 | ||
994eb764 | 3043 | /* Note that interrupts must be enabled when calling this function. */ |
865762a8 JDB |
3044 | int kmem_cache_alloc_bulk(struct kmem_cache *s, gfp_t flags, size_t size, |
3045 | void **p) | |
484748f0 | 3046 | { |
994eb764 JDB |
3047 | struct kmem_cache_cpu *c; |
3048 | int i; | |
3049 | ||
03ec0ed5 JDB |
3050 | /* memcg and kmem_cache debug support */ |
3051 | s = slab_pre_alloc_hook(s, flags); | |
3052 | if (unlikely(!s)) | |
3053 | return false; | |
994eb764 JDB |
3054 | /* |
3055 | * Drain objects in the per cpu slab, while disabling local | |
3056 | * IRQs, which protects against PREEMPT and interrupts | |
3057 | * handlers invoking normal fastpath. | |
3058 | */ | |
3059 | local_irq_disable(); | |
3060 | c = this_cpu_ptr(s->cpu_slab); | |
3061 | ||
3062 | for (i = 0; i < size; i++) { | |
3063 | void *object = c->freelist; | |
3064 | ||
ebe909e0 | 3065 | if (unlikely(!object)) { |
ebe909e0 JDB |
3066 | /* |
3067 | * Invoking slow path likely have side-effect | |
3068 | * of re-populating per CPU c->freelist | |
3069 | */ | |
87098373 | 3070 | p[i] = ___slab_alloc(s, flags, NUMA_NO_NODE, |
ebe909e0 | 3071 | _RET_IP_, c); |
87098373 CL |
3072 | if (unlikely(!p[i])) |
3073 | goto error; | |
3074 | ||
ebe909e0 JDB |
3075 | c = this_cpu_ptr(s->cpu_slab); |
3076 | continue; /* goto for-loop */ | |
3077 | } | |
994eb764 JDB |
3078 | c->freelist = get_freepointer(s, object); |
3079 | p[i] = object; | |
3080 | } | |
3081 | c->tid = next_tid(c->tid); | |
3082 | local_irq_enable(); | |
3083 | ||
3084 | /* Clear memory outside IRQ disabled fastpath loop */ | |
3085 | if (unlikely(flags & __GFP_ZERO)) { | |
3086 | int j; | |
3087 | ||
3088 | for (j = 0; j < i; j++) | |
3089 | memset(p[j], 0, s->object_size); | |
3090 | } | |
3091 | ||
03ec0ed5 JDB |
3092 | /* memcg and kmem_cache debug support */ |
3093 | slab_post_alloc_hook(s, flags, size, p); | |
865762a8 | 3094 | return i; |
87098373 | 3095 | error: |
87098373 | 3096 | local_irq_enable(); |
03ec0ed5 JDB |
3097 | slab_post_alloc_hook(s, flags, i, p); |
3098 | __kmem_cache_free_bulk(s, i, p); | |
865762a8 | 3099 | return 0; |
484748f0 CL |
3100 | } |
3101 | EXPORT_SYMBOL(kmem_cache_alloc_bulk); | |
3102 | ||
3103 | ||
81819f0f | 3104 | /* |
672bba3a CL |
3105 | * Object placement in a slab is made very easy because we always start at |
3106 | * offset 0. If we tune the size of the object to the alignment then we can | |
3107 | * get the required alignment by putting one properly sized object after | |
3108 | * another. | |
81819f0f CL |
3109 | * |
3110 | * Notice that the allocation order determines the sizes of the per cpu | |
3111 | * caches. Each processor has always one slab available for allocations. | |
3112 | * Increasing the allocation order reduces the number of times that slabs | |
672bba3a | 3113 | * must be moved on and off the partial lists and is therefore a factor in |
81819f0f | 3114 | * locking overhead. |
81819f0f CL |
3115 | */ |
3116 | ||
3117 | /* | |
3118 | * Mininum / Maximum order of slab pages. This influences locking overhead | |
3119 | * and slab fragmentation. A higher order reduces the number of partial slabs | |
3120 | * and increases the number of allocations possible without having to | |
3121 | * take the list_lock. | |
3122 | */ | |
3123 | static int slub_min_order; | |
114e9e89 | 3124 | static int slub_max_order = PAGE_ALLOC_COSTLY_ORDER; |
9b2cd506 | 3125 | static int slub_min_objects; |
81819f0f | 3126 | |
81819f0f CL |
3127 | /* |
3128 | * Calculate the order of allocation given an slab object size. | |
3129 | * | |
672bba3a CL |
3130 | * The order of allocation has significant impact on performance and other |
3131 | * system components. Generally order 0 allocations should be preferred since | |
3132 | * order 0 does not cause fragmentation in the page allocator. Larger objects | |
3133 | * be problematic to put into order 0 slabs because there may be too much | |
c124f5b5 | 3134 | * unused space left. We go to a higher order if more than 1/16th of the slab |
672bba3a CL |
3135 | * would be wasted. |
3136 | * | |
3137 | * In order to reach satisfactory performance we must ensure that a minimum | |
3138 | * number of objects is in one slab. Otherwise we may generate too much | |
3139 | * activity on the partial lists which requires taking the list_lock. This is | |
3140 | * less a concern for large slabs though which are rarely used. | |
81819f0f | 3141 | * |
672bba3a CL |
3142 | * slub_max_order specifies the order where we begin to stop considering the |
3143 | * number of objects in a slab as critical. If we reach slub_max_order then | |
3144 | * we try to keep the page order as low as possible. So we accept more waste | |
3145 | * of space in favor of a small page order. | |
81819f0f | 3146 | * |
672bba3a CL |
3147 | * Higher order allocations also allow the placement of more objects in a |
3148 | * slab and thereby reduce object handling overhead. If the user has | |
3149 | * requested a higher mininum order then we start with that one instead of | |
3150 | * the smallest order which will fit the object. | |
81819f0f | 3151 | */ |
5e6d444e | 3152 | static inline int slab_order(int size, int min_objects, |
ab9a0f19 | 3153 | int max_order, int fract_leftover, int reserved) |
81819f0f CL |
3154 | { |
3155 | int order; | |
3156 | int rem; | |
6300ea75 | 3157 | int min_order = slub_min_order; |
81819f0f | 3158 | |
ab9a0f19 | 3159 | if (order_objects(min_order, size, reserved) > MAX_OBJS_PER_PAGE) |
210b5c06 | 3160 | return get_order(size * MAX_OBJS_PER_PAGE) - 1; |
39b26464 | 3161 | |
9f835703 | 3162 | for (order = max(min_order, get_order(min_objects * size + reserved)); |
5e6d444e | 3163 | order <= max_order; order++) { |
81819f0f | 3164 | |
5e6d444e | 3165 | unsigned long slab_size = PAGE_SIZE << order; |
81819f0f | 3166 | |
ab9a0f19 | 3167 | rem = (slab_size - reserved) % size; |
81819f0f | 3168 | |
5e6d444e | 3169 | if (rem <= slab_size / fract_leftover) |
81819f0f | 3170 | break; |
81819f0f | 3171 | } |
672bba3a | 3172 | |
81819f0f CL |
3173 | return order; |
3174 | } | |
3175 | ||
ab9a0f19 | 3176 | static inline int calculate_order(int size, int reserved) |
5e6d444e CL |
3177 | { |
3178 | int order; | |
3179 | int min_objects; | |
3180 | int fraction; | |
e8120ff1 | 3181 | int max_objects; |
5e6d444e CL |
3182 | |
3183 | /* | |
3184 | * Attempt to find best configuration for a slab. This | |
3185 | * works by first attempting to generate a layout with | |
3186 | * the best configuration and backing off gradually. | |
3187 | * | |
422ff4d7 | 3188 | * First we increase the acceptable waste in a slab. Then |
5e6d444e CL |
3189 | * we reduce the minimum objects required in a slab. |
3190 | */ | |
3191 | min_objects = slub_min_objects; | |
9b2cd506 CL |
3192 | if (!min_objects) |
3193 | min_objects = 4 * (fls(nr_cpu_ids) + 1); | |
ab9a0f19 | 3194 | max_objects = order_objects(slub_max_order, size, reserved); |
e8120ff1 ZY |
3195 | min_objects = min(min_objects, max_objects); |
3196 | ||
5e6d444e | 3197 | while (min_objects > 1) { |
c124f5b5 | 3198 | fraction = 16; |
5e6d444e CL |
3199 | while (fraction >= 4) { |
3200 | order = slab_order(size, min_objects, | |
ab9a0f19 | 3201 | slub_max_order, fraction, reserved); |
5e6d444e CL |
3202 | if (order <= slub_max_order) |
3203 | return order; | |
3204 | fraction /= 2; | |
3205 | } | |
5086c389 | 3206 | min_objects--; |
5e6d444e CL |
3207 | } |
3208 | ||
3209 | /* | |
3210 | * We were unable to place multiple objects in a slab. Now | |
3211 | * lets see if we can place a single object there. | |
3212 | */ | |
ab9a0f19 | 3213 | order = slab_order(size, 1, slub_max_order, 1, reserved); |
5e6d444e CL |
3214 | if (order <= slub_max_order) |
3215 | return order; | |
3216 | ||
3217 | /* | |
3218 | * Doh this slab cannot be placed using slub_max_order. | |
3219 | */ | |
ab9a0f19 | 3220 | order = slab_order(size, 1, MAX_ORDER, 1, reserved); |
818cf590 | 3221 | if (order < MAX_ORDER) |
5e6d444e CL |
3222 | return order; |
3223 | return -ENOSYS; | |
3224 | } | |
3225 | ||
5595cffc | 3226 | static void |
4053497d | 3227 | init_kmem_cache_node(struct kmem_cache_node *n) |
81819f0f CL |
3228 | { |
3229 | n->nr_partial = 0; | |
81819f0f CL |
3230 | spin_lock_init(&n->list_lock); |
3231 | INIT_LIST_HEAD(&n->partial); | |
8ab1372f | 3232 | #ifdef CONFIG_SLUB_DEBUG |
0f389ec6 | 3233 | atomic_long_set(&n->nr_slabs, 0); |
02b71b70 | 3234 | atomic_long_set(&n->total_objects, 0); |
643b1138 | 3235 | INIT_LIST_HEAD(&n->full); |
8ab1372f | 3236 | #endif |
81819f0f CL |
3237 | } |
3238 | ||
55136592 | 3239 | static inline int alloc_kmem_cache_cpus(struct kmem_cache *s) |
4c93c355 | 3240 | { |
6c182dc0 | 3241 | BUILD_BUG_ON(PERCPU_DYNAMIC_EARLY_SIZE < |
95a05b42 | 3242 | KMALLOC_SHIFT_HIGH * sizeof(struct kmem_cache_cpu)); |
4c93c355 | 3243 | |
8a5ec0ba | 3244 | /* |
d4d84fef CM |
3245 | * Must align to double word boundary for the double cmpxchg |
3246 | * instructions to work; see __pcpu_double_call_return_bool(). | |
8a5ec0ba | 3247 | */ |
d4d84fef CM |
3248 | s->cpu_slab = __alloc_percpu(sizeof(struct kmem_cache_cpu), |
3249 | 2 * sizeof(void *)); | |
8a5ec0ba CL |
3250 | |
3251 | if (!s->cpu_slab) | |
3252 | return 0; | |
3253 | ||
3254 | init_kmem_cache_cpus(s); | |
4c93c355 | 3255 | |
8a5ec0ba | 3256 | return 1; |
4c93c355 | 3257 | } |
4c93c355 | 3258 | |
51df1142 CL |
3259 | static struct kmem_cache *kmem_cache_node; |
3260 | ||
81819f0f CL |
3261 | /* |
3262 | * No kmalloc_node yet so do it by hand. We know that this is the first | |
3263 | * slab on the node for this slabcache. There are no concurrent accesses | |
3264 | * possible. | |
3265 | * | |
721ae22a ZYW |
3266 | * Note that this function only works on the kmem_cache_node |
3267 | * when allocating for the kmem_cache_node. This is used for bootstrapping | |
4c93c355 | 3268 | * memory on a fresh node that has no slab structures yet. |
81819f0f | 3269 | */ |
55136592 | 3270 | static void early_kmem_cache_node_alloc(int node) |
81819f0f CL |
3271 | { |
3272 | struct page *page; | |
3273 | struct kmem_cache_node *n; | |
3274 | ||
51df1142 | 3275 | BUG_ON(kmem_cache_node->size < sizeof(struct kmem_cache_node)); |
81819f0f | 3276 | |
51df1142 | 3277 | page = new_slab(kmem_cache_node, GFP_NOWAIT, node); |
81819f0f CL |
3278 | |
3279 | BUG_ON(!page); | |
a2f92ee7 | 3280 | if (page_to_nid(page) != node) { |
f9f58285 FF |
3281 | pr_err("SLUB: Unable to allocate memory from node %d\n", node); |
3282 | pr_err("SLUB: Allocating a useless per node structure in order to be able to continue\n"); | |
a2f92ee7 CL |
3283 | } |
3284 | ||
81819f0f CL |
3285 | n = page->freelist; |
3286 | BUG_ON(!n); | |
51df1142 | 3287 | page->freelist = get_freepointer(kmem_cache_node, n); |
e6e82ea1 | 3288 | page->inuse = 1; |
8cb0a506 | 3289 | page->frozen = 0; |
51df1142 | 3290 | kmem_cache_node->node[node] = n; |
8ab1372f | 3291 | #ifdef CONFIG_SLUB_DEBUG |
f7cb1933 | 3292 | init_object(kmem_cache_node, n, SLUB_RED_ACTIVE); |
51df1142 | 3293 | init_tracking(kmem_cache_node, n); |
8ab1372f | 3294 | #endif |
505f5dcb AP |
3295 | kasan_kmalloc(kmem_cache_node, n, sizeof(struct kmem_cache_node), |
3296 | GFP_KERNEL); | |
4053497d | 3297 | init_kmem_cache_node(n); |
51df1142 | 3298 | inc_slabs_node(kmem_cache_node, node, page->objects); |
6446faa2 | 3299 | |
67b6c900 | 3300 | /* |
1e4dd946 SR |
3301 | * No locks need to be taken here as it has just been |
3302 | * initialized and there is no concurrent access. | |
67b6c900 | 3303 | */ |
1e4dd946 | 3304 | __add_partial(n, page, DEACTIVATE_TO_HEAD); |
81819f0f CL |
3305 | } |
3306 | ||
3307 | static void free_kmem_cache_nodes(struct kmem_cache *s) | |
3308 | { | |
3309 | int node; | |
fa45dc25 | 3310 | struct kmem_cache_node *n; |
81819f0f | 3311 | |
fa45dc25 CL |
3312 | for_each_kmem_cache_node(s, node, n) { |
3313 | kmem_cache_free(kmem_cache_node, n); | |
81819f0f CL |
3314 | s->node[node] = NULL; |
3315 | } | |
3316 | } | |
3317 | ||
52b4b950 DS |
3318 | void __kmem_cache_release(struct kmem_cache *s) |
3319 | { | |
210e7a43 | 3320 | cache_random_seq_destroy(s); |
52b4b950 DS |
3321 | free_percpu(s->cpu_slab); |
3322 | free_kmem_cache_nodes(s); | |
3323 | } | |
3324 | ||
55136592 | 3325 | static int init_kmem_cache_nodes(struct kmem_cache *s) |
81819f0f CL |
3326 | { |
3327 | int node; | |
81819f0f | 3328 | |
f64dc58c | 3329 | for_each_node_state(node, N_NORMAL_MEMORY) { |
81819f0f CL |
3330 | struct kmem_cache_node *n; |
3331 | ||
73367bd8 | 3332 | if (slab_state == DOWN) { |
55136592 | 3333 | early_kmem_cache_node_alloc(node); |
73367bd8 AD |
3334 | continue; |
3335 | } | |
51df1142 | 3336 | n = kmem_cache_alloc_node(kmem_cache_node, |
55136592 | 3337 | GFP_KERNEL, node); |
81819f0f | 3338 | |
73367bd8 AD |
3339 | if (!n) { |
3340 | free_kmem_cache_nodes(s); | |
3341 | return 0; | |
81819f0f | 3342 | } |
73367bd8 | 3343 | |
81819f0f | 3344 | s->node[node] = n; |
4053497d | 3345 | init_kmem_cache_node(n); |
81819f0f CL |
3346 | } |
3347 | return 1; | |
3348 | } | |
81819f0f | 3349 | |
c0bdb232 | 3350 | static void set_min_partial(struct kmem_cache *s, unsigned long min) |
3b89d7d8 DR |
3351 | { |
3352 | if (min < MIN_PARTIAL) | |
3353 | min = MIN_PARTIAL; | |
3354 | else if (min > MAX_PARTIAL) | |
3355 | min = MAX_PARTIAL; | |
3356 | s->min_partial = min; | |
3357 | } | |
3358 | ||
81819f0f CL |
3359 | /* |
3360 | * calculate_sizes() determines the order and the distribution of data within | |
3361 | * a slab object. | |
3362 | */ | |
06b285dc | 3363 | static int calculate_sizes(struct kmem_cache *s, int forced_order) |
81819f0f CL |
3364 | { |
3365 | unsigned long flags = s->flags; | |
3b0efdfa | 3366 | unsigned long size = s->object_size; |
834f3d11 | 3367 | int order; |
81819f0f | 3368 | |
d8b42bf5 CL |
3369 | /* |
3370 | * Round up object size to the next word boundary. We can only | |
3371 | * place the free pointer at word boundaries and this determines | |
3372 | * the possible location of the free pointer. | |
3373 | */ | |
3374 | size = ALIGN(size, sizeof(void *)); | |
3375 | ||
3376 | #ifdef CONFIG_SLUB_DEBUG | |
81819f0f CL |
3377 | /* |
3378 | * Determine if we can poison the object itself. If the user of | |
3379 | * the slab may touch the object after free or before allocation | |
3380 | * then we should never poison the object itself. | |
3381 | */ | |
3382 | if ((flags & SLAB_POISON) && !(flags & SLAB_DESTROY_BY_RCU) && | |
c59def9f | 3383 | !s->ctor) |
81819f0f CL |
3384 | s->flags |= __OBJECT_POISON; |
3385 | else | |
3386 | s->flags &= ~__OBJECT_POISON; | |
3387 | ||
81819f0f CL |
3388 | |
3389 | /* | |
672bba3a | 3390 | * If we are Redzoning then check if there is some space between the |
81819f0f | 3391 | * end of the object and the free pointer. If not then add an |
672bba3a | 3392 | * additional word to have some bytes to store Redzone information. |
81819f0f | 3393 | */ |
3b0efdfa | 3394 | if ((flags & SLAB_RED_ZONE) && size == s->object_size) |
81819f0f | 3395 | size += sizeof(void *); |
41ecc55b | 3396 | #endif |
81819f0f CL |
3397 | |
3398 | /* | |
672bba3a CL |
3399 | * With that we have determined the number of bytes in actual use |
3400 | * by the object. This is the potential offset to the free pointer. | |
81819f0f CL |
3401 | */ |
3402 | s->inuse = size; | |
3403 | ||
3404 | if (((flags & (SLAB_DESTROY_BY_RCU | SLAB_POISON)) || | |
c59def9f | 3405 | s->ctor)) { |
81819f0f CL |
3406 | /* |
3407 | * Relocate free pointer after the object if it is not | |
3408 | * permitted to overwrite the first word of the object on | |
3409 | * kmem_cache_free. | |
3410 | * | |
3411 | * This is the case if we do RCU, have a constructor or | |
3412 | * destructor or are poisoning the objects. | |
3413 | */ | |
3414 | s->offset = size; | |
3415 | size += sizeof(void *); | |
3416 | } | |
3417 | ||
c12b3c62 | 3418 | #ifdef CONFIG_SLUB_DEBUG |
81819f0f CL |
3419 | if (flags & SLAB_STORE_USER) |
3420 | /* | |
3421 | * Need to store information about allocs and frees after | |
3422 | * the object. | |
3423 | */ | |
3424 | size += 2 * sizeof(struct track); | |
3425 | ||
d86bd1be | 3426 | if (flags & SLAB_RED_ZONE) { |
81819f0f CL |
3427 | /* |
3428 | * Add some empty padding so that we can catch | |
3429 | * overwrites from earlier objects rather than let | |
3430 | * tracking information or the free pointer be | |
0211a9c8 | 3431 | * corrupted if a user writes before the start |
81819f0f CL |
3432 | * of the object. |
3433 | */ | |
3434 | size += sizeof(void *); | |
d86bd1be JK |
3435 | |
3436 | s->red_left_pad = sizeof(void *); | |
3437 | s->red_left_pad = ALIGN(s->red_left_pad, s->align); | |
3438 | size += s->red_left_pad; | |
3439 | } | |
41ecc55b | 3440 | #endif |
672bba3a | 3441 | |
81819f0f CL |
3442 | /* |
3443 | * SLUB stores one object immediately after another beginning from | |
3444 | * offset 0. In order to align the objects we have to simply size | |
3445 | * each object to conform to the alignment. | |
3446 | */ | |
45906855 | 3447 | size = ALIGN(size, s->align); |
81819f0f | 3448 | s->size = size; |
06b285dc CL |
3449 | if (forced_order >= 0) |
3450 | order = forced_order; | |
3451 | else | |
ab9a0f19 | 3452 | order = calculate_order(size, s->reserved); |
81819f0f | 3453 | |
834f3d11 | 3454 | if (order < 0) |
81819f0f CL |
3455 | return 0; |
3456 | ||
b7a49f0d | 3457 | s->allocflags = 0; |
834f3d11 | 3458 | if (order) |
b7a49f0d CL |
3459 | s->allocflags |= __GFP_COMP; |
3460 | ||
3461 | if (s->flags & SLAB_CACHE_DMA) | |
2c59dd65 | 3462 | s->allocflags |= GFP_DMA; |
b7a49f0d CL |
3463 | |
3464 | if (s->flags & SLAB_RECLAIM_ACCOUNT) | |
3465 | s->allocflags |= __GFP_RECLAIMABLE; | |
3466 | ||
81819f0f CL |
3467 | /* |
3468 | * Determine the number of objects per slab | |
3469 | */ | |
ab9a0f19 LJ |
3470 | s->oo = oo_make(order, size, s->reserved); |
3471 | s->min = oo_make(get_order(size), size, s->reserved); | |
205ab99d CL |
3472 | if (oo_objects(s->oo) > oo_objects(s->max)) |
3473 | s->max = s->oo; | |
81819f0f | 3474 | |
834f3d11 | 3475 | return !!oo_objects(s->oo); |
81819f0f CL |
3476 | } |
3477 | ||
8a13a4cc | 3478 | static int kmem_cache_open(struct kmem_cache *s, unsigned long flags) |
81819f0f | 3479 | { |
8a13a4cc | 3480 | s->flags = kmem_cache_flags(s->size, flags, s->name, s->ctor); |
ab9a0f19 | 3481 | s->reserved = 0; |
81819f0f | 3482 | |
da9a638c LJ |
3483 | if (need_reserve_slab_rcu && (s->flags & SLAB_DESTROY_BY_RCU)) |
3484 | s->reserved = sizeof(struct rcu_head); | |
81819f0f | 3485 | |
06b285dc | 3486 | if (!calculate_sizes(s, -1)) |
81819f0f | 3487 | goto error; |
3de47213 DR |
3488 | if (disable_higher_order_debug) { |
3489 | /* | |
3490 | * Disable debugging flags that store metadata if the min slab | |
3491 | * order increased. | |
3492 | */ | |
3b0efdfa | 3493 | if (get_order(s->size) > get_order(s->object_size)) { |
3de47213 DR |
3494 | s->flags &= ~DEBUG_METADATA_FLAGS; |
3495 | s->offset = 0; | |
3496 | if (!calculate_sizes(s, -1)) | |
3497 | goto error; | |
3498 | } | |
3499 | } | |
81819f0f | 3500 | |
2565409f HC |
3501 | #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \ |
3502 | defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE) | |
149daaf3 | 3503 | if (system_has_cmpxchg_double() && (s->flags & SLAB_NO_CMPXCHG) == 0) |
b789ef51 CL |
3504 | /* Enable fast mode */ |
3505 | s->flags |= __CMPXCHG_DOUBLE; | |
3506 | #endif | |
3507 | ||
3b89d7d8 DR |
3508 | /* |
3509 | * The larger the object size is, the more pages we want on the partial | |
3510 | * list to avoid pounding the page allocator excessively. | |
3511 | */ | |
49e22585 CL |
3512 | set_min_partial(s, ilog2(s->size) / 2); |
3513 | ||
3514 | /* | |
3515 | * cpu_partial determined the maximum number of objects kept in the | |
3516 | * per cpu partial lists of a processor. | |
3517 | * | |
3518 | * Per cpu partial lists mainly contain slabs that just have one | |
3519 | * object freed. If they are used for allocation then they can be | |
3520 | * filled up again with minimal effort. The slab will never hit the | |
3521 | * per node partial lists and therefore no locking will be required. | |
3522 | * | |
3523 | * This setting also determines | |
3524 | * | |
3525 | * A) The number of objects from per cpu partial slabs dumped to the | |
3526 | * per node list when we reach the limit. | |
9f264904 | 3527 | * B) The number of objects in cpu partial slabs to extract from the |
d0e0ac97 CG |
3528 | * per node list when we run out of per cpu objects. We only fetch |
3529 | * 50% to keep some capacity around for frees. | |
49e22585 | 3530 | */ |
345c905d | 3531 | if (!kmem_cache_has_cpu_partial(s)) |
8f1e33da CL |
3532 | s->cpu_partial = 0; |
3533 | else if (s->size >= PAGE_SIZE) | |
49e22585 CL |
3534 | s->cpu_partial = 2; |
3535 | else if (s->size >= 1024) | |
3536 | s->cpu_partial = 6; | |
3537 | else if (s->size >= 256) | |
3538 | s->cpu_partial = 13; | |
3539 | else | |
3540 | s->cpu_partial = 30; | |
3541 | ||
81819f0f | 3542 | #ifdef CONFIG_NUMA |
e2cb96b7 | 3543 | s->remote_node_defrag_ratio = 1000; |
81819f0f | 3544 | #endif |
210e7a43 TG |
3545 | |
3546 | /* Initialize the pre-computed randomized freelist if slab is up */ | |
3547 | if (slab_state >= UP) { | |
3548 | if (init_cache_random_seq(s)) | |
3549 | goto error; | |
3550 | } | |
3551 | ||
55136592 | 3552 | if (!init_kmem_cache_nodes(s)) |
dfb4f096 | 3553 | goto error; |
81819f0f | 3554 | |
55136592 | 3555 | if (alloc_kmem_cache_cpus(s)) |
278b1bb1 | 3556 | return 0; |
ff12059e | 3557 | |
4c93c355 | 3558 | free_kmem_cache_nodes(s); |
81819f0f CL |
3559 | error: |
3560 | if (flags & SLAB_PANIC) | |
756a025f JP |
3561 | panic("Cannot create slab %s size=%lu realsize=%u order=%u offset=%u flags=%lx\n", |
3562 | s->name, (unsigned long)s->size, s->size, | |
3563 | oo_order(s->oo), s->offset, flags); | |
278b1bb1 | 3564 | return -EINVAL; |
81819f0f | 3565 | } |
81819f0f | 3566 | |
33b12c38 CL |
3567 | static void list_slab_objects(struct kmem_cache *s, struct page *page, |
3568 | const char *text) | |
3569 | { | |
3570 | #ifdef CONFIG_SLUB_DEBUG | |
3571 | void *addr = page_address(page); | |
3572 | void *p; | |
a5dd5c11 NK |
3573 | unsigned long *map = kzalloc(BITS_TO_LONGS(page->objects) * |
3574 | sizeof(long), GFP_ATOMIC); | |
bbd7d57b ED |
3575 | if (!map) |
3576 | return; | |
945cf2b6 | 3577 | slab_err(s, page, text, s->name); |
33b12c38 | 3578 | slab_lock(page); |
33b12c38 | 3579 | |
5f80b13a | 3580 | get_map(s, page, map); |
33b12c38 CL |
3581 | for_each_object(p, s, addr, page->objects) { |
3582 | ||
3583 | if (!test_bit(slab_index(p, s, addr), map)) { | |
f9f58285 | 3584 | pr_err("INFO: Object 0x%p @offset=%tu\n", p, p - addr); |
33b12c38 CL |
3585 | print_tracking(s, p); |
3586 | } | |
3587 | } | |
3588 | slab_unlock(page); | |
bbd7d57b | 3589 | kfree(map); |
33b12c38 CL |
3590 | #endif |
3591 | } | |
3592 | ||
81819f0f | 3593 | /* |
599870b1 | 3594 | * Attempt to free all partial slabs on a node. |
52b4b950 DS |
3595 | * This is called from __kmem_cache_shutdown(). We must take list_lock |
3596 | * because sysfs file might still access partial list after the shutdowning. | |
81819f0f | 3597 | */ |
599870b1 | 3598 | static void free_partial(struct kmem_cache *s, struct kmem_cache_node *n) |
81819f0f | 3599 | { |
81819f0f CL |
3600 | struct page *page, *h; |
3601 | ||
52b4b950 DS |
3602 | BUG_ON(irqs_disabled()); |
3603 | spin_lock_irq(&n->list_lock); | |
33b12c38 | 3604 | list_for_each_entry_safe(page, h, &n->partial, lru) { |
81819f0f | 3605 | if (!page->inuse) { |
52b4b950 | 3606 | remove_partial(n, page); |
81819f0f | 3607 | discard_slab(s, page); |
33b12c38 CL |
3608 | } else { |
3609 | list_slab_objects(s, page, | |
52b4b950 | 3610 | "Objects remaining in %s on __kmem_cache_shutdown()"); |
599870b1 | 3611 | } |
33b12c38 | 3612 | } |
52b4b950 | 3613 | spin_unlock_irq(&n->list_lock); |
81819f0f CL |
3614 | } |
3615 | ||
3616 | /* | |
672bba3a | 3617 | * Release all resources used by a slab cache. |
81819f0f | 3618 | */ |
52b4b950 | 3619 | int __kmem_cache_shutdown(struct kmem_cache *s) |
81819f0f CL |
3620 | { |
3621 | int node; | |
fa45dc25 | 3622 | struct kmem_cache_node *n; |
81819f0f CL |
3623 | |
3624 | flush_all(s); | |
81819f0f | 3625 | /* Attempt to free all objects */ |
fa45dc25 | 3626 | for_each_kmem_cache_node(s, node, n) { |
599870b1 CL |
3627 | free_partial(s, n); |
3628 | if (n->nr_partial || slabs_node(s, node)) | |
81819f0f CL |
3629 | return 1; |
3630 | } | |
81819f0f CL |
3631 | return 0; |
3632 | } | |
3633 | ||
81819f0f CL |
3634 | /******************************************************************** |
3635 | * Kmalloc subsystem | |
3636 | *******************************************************************/ | |
3637 | ||
81819f0f CL |
3638 | static int __init setup_slub_min_order(char *str) |
3639 | { | |
06428780 | 3640 | get_option(&str, &slub_min_order); |
81819f0f CL |
3641 | |
3642 | return 1; | |
3643 | } | |
3644 | ||
3645 | __setup("slub_min_order=", setup_slub_min_order); | |
3646 | ||
3647 | static int __init setup_slub_max_order(char *str) | |
3648 | { | |
06428780 | 3649 | get_option(&str, &slub_max_order); |
818cf590 | 3650 | slub_max_order = min(slub_max_order, MAX_ORDER - 1); |
81819f0f CL |
3651 | |
3652 | return 1; | |
3653 | } | |
3654 | ||
3655 | __setup("slub_max_order=", setup_slub_max_order); | |
3656 | ||
3657 | static int __init setup_slub_min_objects(char *str) | |
3658 | { | |
06428780 | 3659 | get_option(&str, &slub_min_objects); |
81819f0f CL |
3660 | |
3661 | return 1; | |
3662 | } | |
3663 | ||
3664 | __setup("slub_min_objects=", setup_slub_min_objects); | |
3665 | ||
81819f0f CL |
3666 | void *__kmalloc(size_t size, gfp_t flags) |
3667 | { | |
aadb4bc4 | 3668 | struct kmem_cache *s; |
5b882be4 | 3669 | void *ret; |
81819f0f | 3670 | |
95a05b42 | 3671 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) |
eada35ef | 3672 | return kmalloc_large(size, flags); |
aadb4bc4 | 3673 | |
2c59dd65 | 3674 | s = kmalloc_slab(size, flags); |
aadb4bc4 CL |
3675 | |
3676 | if (unlikely(ZERO_OR_NULL_PTR(s))) | |
6cb8f913 CL |
3677 | return s; |
3678 | ||
2b847c3c | 3679 | ret = slab_alloc(s, flags, _RET_IP_); |
5b882be4 | 3680 | |
ca2b84cb | 3681 | trace_kmalloc(_RET_IP_, ret, size, s->size, flags); |
5b882be4 | 3682 | |
505f5dcb | 3683 | kasan_kmalloc(s, ret, size, flags); |
0316bec2 | 3684 | |
5b882be4 | 3685 | return ret; |
81819f0f CL |
3686 | } |
3687 | EXPORT_SYMBOL(__kmalloc); | |
3688 | ||
5d1f57e4 | 3689 | #ifdef CONFIG_NUMA |
f619cfe1 CL |
3690 | static void *kmalloc_large_node(size_t size, gfp_t flags, int node) |
3691 | { | |
b1eeab67 | 3692 | struct page *page; |
e4f7c0b4 | 3693 | void *ptr = NULL; |
f619cfe1 | 3694 | |
52383431 | 3695 | flags |= __GFP_COMP | __GFP_NOTRACK; |
4949148a | 3696 | page = alloc_pages_node(node, flags, get_order(size)); |
f619cfe1 | 3697 | if (page) |
e4f7c0b4 CM |
3698 | ptr = page_address(page); |
3699 | ||
d56791b3 | 3700 | kmalloc_large_node_hook(ptr, size, flags); |
e4f7c0b4 | 3701 | return ptr; |
f619cfe1 CL |
3702 | } |
3703 | ||
81819f0f CL |
3704 | void *__kmalloc_node(size_t size, gfp_t flags, int node) |
3705 | { | |
aadb4bc4 | 3706 | struct kmem_cache *s; |
5b882be4 | 3707 | void *ret; |
81819f0f | 3708 | |
95a05b42 | 3709 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) { |
5b882be4 EGM |
3710 | ret = kmalloc_large_node(size, flags, node); |
3711 | ||
ca2b84cb EGM |
3712 | trace_kmalloc_node(_RET_IP_, ret, |
3713 | size, PAGE_SIZE << get_order(size), | |
3714 | flags, node); | |
5b882be4 EGM |
3715 | |
3716 | return ret; | |
3717 | } | |
aadb4bc4 | 3718 | |
2c59dd65 | 3719 | s = kmalloc_slab(size, flags); |
aadb4bc4 CL |
3720 | |
3721 | if (unlikely(ZERO_OR_NULL_PTR(s))) | |
6cb8f913 CL |
3722 | return s; |
3723 | ||
2b847c3c | 3724 | ret = slab_alloc_node(s, flags, node, _RET_IP_); |
5b882be4 | 3725 | |
ca2b84cb | 3726 | trace_kmalloc_node(_RET_IP_, ret, size, s->size, flags, node); |
5b882be4 | 3727 | |
505f5dcb | 3728 | kasan_kmalloc(s, ret, size, flags); |
0316bec2 | 3729 | |
5b882be4 | 3730 | return ret; |
81819f0f CL |
3731 | } |
3732 | EXPORT_SYMBOL(__kmalloc_node); | |
3733 | #endif | |
3734 | ||
0316bec2 | 3735 | static size_t __ksize(const void *object) |
81819f0f | 3736 | { |
272c1d21 | 3737 | struct page *page; |
81819f0f | 3738 | |
ef8b4520 | 3739 | if (unlikely(object == ZERO_SIZE_PTR)) |
272c1d21 CL |
3740 | return 0; |
3741 | ||
294a80a8 | 3742 | page = virt_to_head_page(object); |
294a80a8 | 3743 | |
76994412 PE |
3744 | if (unlikely(!PageSlab(page))) { |
3745 | WARN_ON(!PageCompound(page)); | |
294a80a8 | 3746 | return PAGE_SIZE << compound_order(page); |
76994412 | 3747 | } |
81819f0f | 3748 | |
1b4f59e3 | 3749 | return slab_ksize(page->slab_cache); |
81819f0f | 3750 | } |
0316bec2 AR |
3751 | |
3752 | size_t ksize(const void *object) | |
3753 | { | |
3754 | size_t size = __ksize(object); | |
3755 | /* We assume that ksize callers could use whole allocated area, | |
4ebb31a4 AP |
3756 | * so we need to unpoison this area. |
3757 | */ | |
3758 | kasan_unpoison_shadow(object, size); | |
0316bec2 AR |
3759 | return size; |
3760 | } | |
b1aabecd | 3761 | EXPORT_SYMBOL(ksize); |
81819f0f CL |
3762 | |
3763 | void kfree(const void *x) | |
3764 | { | |
81819f0f | 3765 | struct page *page; |
5bb983b0 | 3766 | void *object = (void *)x; |
81819f0f | 3767 | |
2121db74 PE |
3768 | trace_kfree(_RET_IP_, x); |
3769 | ||
2408c550 | 3770 | if (unlikely(ZERO_OR_NULL_PTR(x))) |
81819f0f CL |
3771 | return; |
3772 | ||
b49af68f | 3773 | page = virt_to_head_page(x); |
aadb4bc4 | 3774 | if (unlikely(!PageSlab(page))) { |
0937502a | 3775 | BUG_ON(!PageCompound(page)); |
d56791b3 | 3776 | kfree_hook(x); |
4949148a | 3777 | __free_pages(page, compound_order(page)); |
aadb4bc4 CL |
3778 | return; |
3779 | } | |
81084651 | 3780 | slab_free(page->slab_cache, page, object, NULL, 1, _RET_IP_); |
81819f0f CL |
3781 | } |
3782 | EXPORT_SYMBOL(kfree); | |
3783 | ||
832f37f5 VD |
3784 | #define SHRINK_PROMOTE_MAX 32 |
3785 | ||
2086d26a | 3786 | /* |
832f37f5 VD |
3787 | * kmem_cache_shrink discards empty slabs and promotes the slabs filled |
3788 | * up most to the head of the partial lists. New allocations will then | |
3789 | * fill those up and thus they can be removed from the partial lists. | |
672bba3a CL |
3790 | * |
3791 | * The slabs with the least items are placed last. This results in them | |
3792 | * being allocated from last increasing the chance that the last objects | |
3793 | * are freed in them. | |
2086d26a | 3794 | */ |
d6e0b7fa | 3795 | int __kmem_cache_shrink(struct kmem_cache *s, bool deactivate) |
2086d26a CL |
3796 | { |
3797 | int node; | |
3798 | int i; | |
3799 | struct kmem_cache_node *n; | |
3800 | struct page *page; | |
3801 | struct page *t; | |
832f37f5 VD |
3802 | struct list_head discard; |
3803 | struct list_head promote[SHRINK_PROMOTE_MAX]; | |
2086d26a | 3804 | unsigned long flags; |
ce3712d7 | 3805 | int ret = 0; |
2086d26a | 3806 | |
d6e0b7fa VD |
3807 | if (deactivate) { |
3808 | /* | |
3809 | * Disable empty slabs caching. Used to avoid pinning offline | |
3810 | * memory cgroups by kmem pages that can be freed. | |
3811 | */ | |
3812 | s->cpu_partial = 0; | |
3813 | s->min_partial = 0; | |
3814 | ||
3815 | /* | |
3816 | * s->cpu_partial is checked locklessly (see put_cpu_partial), | |
3817 | * so we have to make sure the change is visible. | |
3818 | */ | |
81ae6d03 | 3819 | synchronize_sched(); |
d6e0b7fa VD |
3820 | } |
3821 | ||
2086d26a | 3822 | flush_all(s); |
fa45dc25 | 3823 | for_each_kmem_cache_node(s, node, n) { |
832f37f5 VD |
3824 | INIT_LIST_HEAD(&discard); |
3825 | for (i = 0; i < SHRINK_PROMOTE_MAX; i++) | |
3826 | INIT_LIST_HEAD(promote + i); | |
2086d26a CL |
3827 | |
3828 | spin_lock_irqsave(&n->list_lock, flags); | |
3829 | ||
3830 | /* | |
832f37f5 | 3831 | * Build lists of slabs to discard or promote. |
2086d26a | 3832 | * |
672bba3a CL |
3833 | * Note that concurrent frees may occur while we hold the |
3834 | * list_lock. page->inuse here is the upper limit. | |
2086d26a CL |
3835 | */ |
3836 | list_for_each_entry_safe(page, t, &n->partial, lru) { | |
832f37f5 VD |
3837 | int free = page->objects - page->inuse; |
3838 | ||
3839 | /* Do not reread page->inuse */ | |
3840 | barrier(); | |
3841 | ||
3842 | /* We do not keep full slabs on the list */ | |
3843 | BUG_ON(free <= 0); | |
3844 | ||
3845 | if (free == page->objects) { | |
3846 | list_move(&page->lru, &discard); | |
69cb8e6b | 3847 | n->nr_partial--; |
832f37f5 VD |
3848 | } else if (free <= SHRINK_PROMOTE_MAX) |
3849 | list_move(&page->lru, promote + free - 1); | |
2086d26a CL |
3850 | } |
3851 | ||
2086d26a | 3852 | /* |
832f37f5 VD |
3853 | * Promote the slabs filled up most to the head of the |
3854 | * partial list. | |
2086d26a | 3855 | */ |
832f37f5 VD |
3856 | for (i = SHRINK_PROMOTE_MAX - 1; i >= 0; i--) |
3857 | list_splice(promote + i, &n->partial); | |
2086d26a | 3858 | |
2086d26a | 3859 | spin_unlock_irqrestore(&n->list_lock, flags); |
69cb8e6b CL |
3860 | |
3861 | /* Release empty slabs */ | |
832f37f5 | 3862 | list_for_each_entry_safe(page, t, &discard, lru) |
69cb8e6b | 3863 | discard_slab(s, page); |
ce3712d7 VD |
3864 | |
3865 | if (slabs_node(s, node)) | |
3866 | ret = 1; | |
2086d26a CL |
3867 | } |
3868 | ||
ce3712d7 | 3869 | return ret; |
2086d26a | 3870 | } |
2086d26a | 3871 | |
b9049e23 YG |
3872 | static int slab_mem_going_offline_callback(void *arg) |
3873 | { | |
3874 | struct kmem_cache *s; | |
3875 | ||
18004c5d | 3876 | mutex_lock(&slab_mutex); |
b9049e23 | 3877 | list_for_each_entry(s, &slab_caches, list) |
d6e0b7fa | 3878 | __kmem_cache_shrink(s, false); |
18004c5d | 3879 | mutex_unlock(&slab_mutex); |
b9049e23 YG |
3880 | |
3881 | return 0; | |
3882 | } | |
3883 | ||
3884 | static void slab_mem_offline_callback(void *arg) | |
3885 | { | |
3886 | struct kmem_cache_node *n; | |
3887 | struct kmem_cache *s; | |
3888 | struct memory_notify *marg = arg; | |
3889 | int offline_node; | |
3890 | ||
b9d5ab25 | 3891 | offline_node = marg->status_change_nid_normal; |
b9049e23 YG |
3892 | |
3893 | /* | |
3894 | * If the node still has available memory. we need kmem_cache_node | |
3895 | * for it yet. | |
3896 | */ | |
3897 | if (offline_node < 0) | |
3898 | return; | |
3899 | ||
18004c5d | 3900 | mutex_lock(&slab_mutex); |
b9049e23 YG |
3901 | list_for_each_entry(s, &slab_caches, list) { |
3902 | n = get_node(s, offline_node); | |
3903 | if (n) { | |
3904 | /* | |
3905 | * if n->nr_slabs > 0, slabs still exist on the node | |
3906 | * that is going down. We were unable to free them, | |
c9404c9c | 3907 | * and offline_pages() function shouldn't call this |
b9049e23 YG |
3908 | * callback. So, we must fail. |
3909 | */ | |
0f389ec6 | 3910 | BUG_ON(slabs_node(s, offline_node)); |
b9049e23 YG |
3911 | |
3912 | s->node[offline_node] = NULL; | |
8de66a0c | 3913 | kmem_cache_free(kmem_cache_node, n); |
b9049e23 YG |
3914 | } |
3915 | } | |
18004c5d | 3916 | mutex_unlock(&slab_mutex); |
b9049e23 YG |
3917 | } |
3918 | ||
3919 | static int slab_mem_going_online_callback(void *arg) | |
3920 | { | |
3921 | struct kmem_cache_node *n; | |
3922 | struct kmem_cache *s; | |
3923 | struct memory_notify *marg = arg; | |
b9d5ab25 | 3924 | int nid = marg->status_change_nid_normal; |
b9049e23 YG |
3925 | int ret = 0; |
3926 | ||
3927 | /* | |
3928 | * If the node's memory is already available, then kmem_cache_node is | |
3929 | * already created. Nothing to do. | |
3930 | */ | |
3931 | if (nid < 0) | |
3932 | return 0; | |
3933 | ||
3934 | /* | |
0121c619 | 3935 | * We are bringing a node online. No memory is available yet. We must |
b9049e23 YG |
3936 | * allocate a kmem_cache_node structure in order to bring the node |
3937 | * online. | |
3938 | */ | |
18004c5d | 3939 | mutex_lock(&slab_mutex); |
b9049e23 YG |
3940 | list_for_each_entry(s, &slab_caches, list) { |
3941 | /* | |
3942 | * XXX: kmem_cache_alloc_node will fallback to other nodes | |
3943 | * since memory is not yet available from the node that | |
3944 | * is brought up. | |
3945 | */ | |
8de66a0c | 3946 | n = kmem_cache_alloc(kmem_cache_node, GFP_KERNEL); |
b9049e23 YG |
3947 | if (!n) { |
3948 | ret = -ENOMEM; | |
3949 | goto out; | |
3950 | } | |
4053497d | 3951 | init_kmem_cache_node(n); |
b9049e23 YG |
3952 | s->node[nid] = n; |
3953 | } | |
3954 | out: | |
18004c5d | 3955 | mutex_unlock(&slab_mutex); |
b9049e23 YG |
3956 | return ret; |
3957 | } | |
3958 | ||
3959 | static int slab_memory_callback(struct notifier_block *self, | |
3960 | unsigned long action, void *arg) | |
3961 | { | |
3962 | int ret = 0; | |
3963 | ||
3964 | switch (action) { | |
3965 | case MEM_GOING_ONLINE: | |
3966 | ret = slab_mem_going_online_callback(arg); | |
3967 | break; | |
3968 | case MEM_GOING_OFFLINE: | |
3969 | ret = slab_mem_going_offline_callback(arg); | |
3970 | break; | |
3971 | case MEM_OFFLINE: | |
3972 | case MEM_CANCEL_ONLINE: | |
3973 | slab_mem_offline_callback(arg); | |
3974 | break; | |
3975 | case MEM_ONLINE: | |
3976 | case MEM_CANCEL_OFFLINE: | |
3977 | break; | |
3978 | } | |
dc19f9db KH |
3979 | if (ret) |
3980 | ret = notifier_from_errno(ret); | |
3981 | else | |
3982 | ret = NOTIFY_OK; | |
b9049e23 YG |
3983 | return ret; |
3984 | } | |
3985 | ||
3ac38faa AM |
3986 | static struct notifier_block slab_memory_callback_nb = { |
3987 | .notifier_call = slab_memory_callback, | |
3988 | .priority = SLAB_CALLBACK_PRI, | |
3989 | }; | |
b9049e23 | 3990 | |
81819f0f CL |
3991 | /******************************************************************** |
3992 | * Basic setup of slabs | |
3993 | *******************************************************************/ | |
3994 | ||
51df1142 CL |
3995 | /* |
3996 | * Used for early kmem_cache structures that were allocated using | |
dffb4d60 CL |
3997 | * the page allocator. Allocate them properly then fix up the pointers |
3998 | * that may be pointing to the wrong kmem_cache structure. | |
51df1142 CL |
3999 | */ |
4000 | ||
dffb4d60 | 4001 | static struct kmem_cache * __init bootstrap(struct kmem_cache *static_cache) |
51df1142 CL |
4002 | { |
4003 | int node; | |
dffb4d60 | 4004 | struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT); |
fa45dc25 | 4005 | struct kmem_cache_node *n; |
51df1142 | 4006 | |
dffb4d60 | 4007 | memcpy(s, static_cache, kmem_cache->object_size); |
51df1142 | 4008 | |
7d557b3c GC |
4009 | /* |
4010 | * This runs very early, and only the boot processor is supposed to be | |
4011 | * up. Even if it weren't true, IRQs are not up so we couldn't fire | |
4012 | * IPIs around. | |
4013 | */ | |
4014 | __flush_cpu_slab(s, smp_processor_id()); | |
fa45dc25 | 4015 | for_each_kmem_cache_node(s, node, n) { |
51df1142 CL |
4016 | struct page *p; |
4017 | ||
fa45dc25 CL |
4018 | list_for_each_entry(p, &n->partial, lru) |
4019 | p->slab_cache = s; | |
51df1142 | 4020 | |
607bf324 | 4021 | #ifdef CONFIG_SLUB_DEBUG |
fa45dc25 CL |
4022 | list_for_each_entry(p, &n->full, lru) |
4023 | p->slab_cache = s; | |
51df1142 | 4024 | #endif |
51df1142 | 4025 | } |
f7ce3190 | 4026 | slab_init_memcg_params(s); |
dffb4d60 CL |
4027 | list_add(&s->list, &slab_caches); |
4028 | return s; | |
51df1142 CL |
4029 | } |
4030 | ||
81819f0f CL |
4031 | void __init kmem_cache_init(void) |
4032 | { | |
dffb4d60 CL |
4033 | static __initdata struct kmem_cache boot_kmem_cache, |
4034 | boot_kmem_cache_node; | |
51df1142 | 4035 | |
fc8d8620 SG |
4036 | if (debug_guardpage_minorder()) |
4037 | slub_max_order = 0; | |
4038 | ||
dffb4d60 CL |
4039 | kmem_cache_node = &boot_kmem_cache_node; |
4040 | kmem_cache = &boot_kmem_cache; | |
51df1142 | 4041 | |
dffb4d60 CL |
4042 | create_boot_cache(kmem_cache_node, "kmem_cache_node", |
4043 | sizeof(struct kmem_cache_node), SLAB_HWCACHE_ALIGN); | |
b9049e23 | 4044 | |
3ac38faa | 4045 | register_hotmemory_notifier(&slab_memory_callback_nb); |
81819f0f CL |
4046 | |
4047 | /* Able to allocate the per node structures */ | |
4048 | slab_state = PARTIAL; | |
4049 | ||
dffb4d60 CL |
4050 | create_boot_cache(kmem_cache, "kmem_cache", |
4051 | offsetof(struct kmem_cache, node) + | |
4052 | nr_node_ids * sizeof(struct kmem_cache_node *), | |
4053 | SLAB_HWCACHE_ALIGN); | |
8a13a4cc | 4054 | |
dffb4d60 | 4055 | kmem_cache = bootstrap(&boot_kmem_cache); |
81819f0f | 4056 | |
51df1142 CL |
4057 | /* |
4058 | * Allocate kmem_cache_node properly from the kmem_cache slab. | |
4059 | * kmem_cache_node is separately allocated so no need to | |
4060 | * update any list pointers. | |
4061 | */ | |
dffb4d60 | 4062 | kmem_cache_node = bootstrap(&boot_kmem_cache_node); |
51df1142 CL |
4063 | |
4064 | /* Now we can use the kmem_cache to allocate kmalloc slabs */ | |
34cc6990 | 4065 | setup_kmalloc_cache_index_table(); |
f97d5f63 | 4066 | create_kmalloc_caches(0); |
81819f0f | 4067 | |
210e7a43 TG |
4068 | /* Setup random freelists for each cache */ |
4069 | init_freelist_randomization(); | |
4070 | ||
81819f0f CL |
4071 | #ifdef CONFIG_SMP |
4072 | register_cpu_notifier(&slab_notifier); | |
9dfc6e68 | 4073 | #endif |
81819f0f | 4074 | |
f9f58285 | 4075 | pr_info("SLUB: HWalign=%d, Order=%d-%d, MinObjects=%d, CPUs=%d, Nodes=%d\n", |
f97d5f63 | 4076 | cache_line_size(), |
81819f0f CL |
4077 | slub_min_order, slub_max_order, slub_min_objects, |
4078 | nr_cpu_ids, nr_node_ids); | |
4079 | } | |
4080 | ||
7e85ee0c PE |
4081 | void __init kmem_cache_init_late(void) |
4082 | { | |
7e85ee0c PE |
4083 | } |
4084 | ||
2633d7a0 | 4085 | struct kmem_cache * |
a44cb944 VD |
4086 | __kmem_cache_alias(const char *name, size_t size, size_t align, |
4087 | unsigned long flags, void (*ctor)(void *)) | |
81819f0f | 4088 | { |
426589f5 | 4089 | struct kmem_cache *s, *c; |
81819f0f | 4090 | |
a44cb944 | 4091 | s = find_mergeable(size, align, flags, name, ctor); |
81819f0f CL |
4092 | if (s) { |
4093 | s->refcount++; | |
84d0ddd6 | 4094 | |
81819f0f CL |
4095 | /* |
4096 | * Adjust the object sizes so that we clear | |
4097 | * the complete object on kzalloc. | |
4098 | */ | |
3b0efdfa | 4099 | s->object_size = max(s->object_size, (int)size); |
81819f0f | 4100 | s->inuse = max_t(int, s->inuse, ALIGN(size, sizeof(void *))); |
6446faa2 | 4101 | |
426589f5 | 4102 | for_each_memcg_cache(c, s) { |
84d0ddd6 VD |
4103 | c->object_size = s->object_size; |
4104 | c->inuse = max_t(int, c->inuse, | |
4105 | ALIGN(size, sizeof(void *))); | |
4106 | } | |
4107 | ||
7b8f3b66 | 4108 | if (sysfs_slab_alias(s, name)) { |
7b8f3b66 | 4109 | s->refcount--; |
cbb79694 | 4110 | s = NULL; |
7b8f3b66 | 4111 | } |
a0e1d1be | 4112 | } |
6446faa2 | 4113 | |
cbb79694 CL |
4114 | return s; |
4115 | } | |
84c1cf62 | 4116 | |
8a13a4cc | 4117 | int __kmem_cache_create(struct kmem_cache *s, unsigned long flags) |
cbb79694 | 4118 | { |
aac3a166 PE |
4119 | int err; |
4120 | ||
4121 | err = kmem_cache_open(s, flags); | |
4122 | if (err) | |
4123 | return err; | |
20cea968 | 4124 | |
45530c44 CL |
4125 | /* Mutex is not taken during early boot */ |
4126 | if (slab_state <= UP) | |
4127 | return 0; | |
4128 | ||
107dab5c | 4129 | memcg_propagate_slab_attrs(s); |
aac3a166 | 4130 | err = sysfs_slab_add(s); |
aac3a166 | 4131 | if (err) |
52b4b950 | 4132 | __kmem_cache_release(s); |
20cea968 | 4133 | |
aac3a166 | 4134 | return err; |
81819f0f | 4135 | } |
81819f0f | 4136 | |
81819f0f | 4137 | #ifdef CONFIG_SMP |
81819f0f | 4138 | /* |
672bba3a CL |
4139 | * Use the cpu notifier to insure that the cpu slabs are flushed when |
4140 | * necessary. | |
81819f0f | 4141 | */ |
0db0628d | 4142 | static int slab_cpuup_callback(struct notifier_block *nfb, |
81819f0f CL |
4143 | unsigned long action, void *hcpu) |
4144 | { | |
4145 | long cpu = (long)hcpu; | |
5b95a4ac CL |
4146 | struct kmem_cache *s; |
4147 | unsigned long flags; | |
81819f0f CL |
4148 | |
4149 | switch (action) { | |
4150 | case CPU_UP_CANCELED: | |
8bb78442 | 4151 | case CPU_UP_CANCELED_FROZEN: |
81819f0f | 4152 | case CPU_DEAD: |
8bb78442 | 4153 | case CPU_DEAD_FROZEN: |
18004c5d | 4154 | mutex_lock(&slab_mutex); |
5b95a4ac CL |
4155 | list_for_each_entry(s, &slab_caches, list) { |
4156 | local_irq_save(flags); | |
4157 | __flush_cpu_slab(s, cpu); | |
4158 | local_irq_restore(flags); | |
4159 | } | |
18004c5d | 4160 | mutex_unlock(&slab_mutex); |
81819f0f CL |
4161 | break; |
4162 | default: | |
4163 | break; | |
4164 | } | |
4165 | return NOTIFY_OK; | |
4166 | } | |
4167 | ||
0db0628d | 4168 | static struct notifier_block slab_notifier = { |
3adbefee | 4169 | .notifier_call = slab_cpuup_callback |
06428780 | 4170 | }; |
81819f0f CL |
4171 | |
4172 | #endif | |
4173 | ||
ce71e27c | 4174 | void *__kmalloc_track_caller(size_t size, gfp_t gfpflags, unsigned long caller) |
81819f0f | 4175 | { |
aadb4bc4 | 4176 | struct kmem_cache *s; |
94b528d0 | 4177 | void *ret; |
aadb4bc4 | 4178 | |
95a05b42 | 4179 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) |
eada35ef PE |
4180 | return kmalloc_large(size, gfpflags); |
4181 | ||
2c59dd65 | 4182 | s = kmalloc_slab(size, gfpflags); |
81819f0f | 4183 | |
2408c550 | 4184 | if (unlikely(ZERO_OR_NULL_PTR(s))) |
6cb8f913 | 4185 | return s; |
81819f0f | 4186 | |
2b847c3c | 4187 | ret = slab_alloc(s, gfpflags, caller); |
94b528d0 | 4188 | |
25985edc | 4189 | /* Honor the call site pointer we received. */ |
ca2b84cb | 4190 | trace_kmalloc(caller, ret, size, s->size, gfpflags); |
94b528d0 EGM |
4191 | |
4192 | return ret; | |
81819f0f CL |
4193 | } |
4194 | ||
5d1f57e4 | 4195 | #ifdef CONFIG_NUMA |
81819f0f | 4196 | void *__kmalloc_node_track_caller(size_t size, gfp_t gfpflags, |
ce71e27c | 4197 | int node, unsigned long caller) |
81819f0f | 4198 | { |
aadb4bc4 | 4199 | struct kmem_cache *s; |
94b528d0 | 4200 | void *ret; |
aadb4bc4 | 4201 | |
95a05b42 | 4202 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) { |
d3e14aa3 XF |
4203 | ret = kmalloc_large_node(size, gfpflags, node); |
4204 | ||
4205 | trace_kmalloc_node(caller, ret, | |
4206 | size, PAGE_SIZE << get_order(size), | |
4207 | gfpflags, node); | |
4208 | ||
4209 | return ret; | |
4210 | } | |
eada35ef | 4211 | |
2c59dd65 | 4212 | s = kmalloc_slab(size, gfpflags); |
81819f0f | 4213 | |
2408c550 | 4214 | if (unlikely(ZERO_OR_NULL_PTR(s))) |
6cb8f913 | 4215 | return s; |
81819f0f | 4216 | |
2b847c3c | 4217 | ret = slab_alloc_node(s, gfpflags, node, caller); |
94b528d0 | 4218 | |
25985edc | 4219 | /* Honor the call site pointer we received. */ |
ca2b84cb | 4220 | trace_kmalloc_node(caller, ret, size, s->size, gfpflags, node); |
94b528d0 EGM |
4221 | |
4222 | return ret; | |
81819f0f | 4223 | } |
5d1f57e4 | 4224 | #endif |
81819f0f | 4225 | |
ab4d5ed5 | 4226 | #ifdef CONFIG_SYSFS |
205ab99d CL |
4227 | static int count_inuse(struct page *page) |
4228 | { | |
4229 | return page->inuse; | |
4230 | } | |
4231 | ||
4232 | static int count_total(struct page *page) | |
4233 | { | |
4234 | return page->objects; | |
4235 | } | |
ab4d5ed5 | 4236 | #endif |
205ab99d | 4237 | |
ab4d5ed5 | 4238 | #ifdef CONFIG_SLUB_DEBUG |
434e245d CL |
4239 | static int validate_slab(struct kmem_cache *s, struct page *page, |
4240 | unsigned long *map) | |
53e15af0 CL |
4241 | { |
4242 | void *p; | |
a973e9dd | 4243 | void *addr = page_address(page); |
53e15af0 CL |
4244 | |
4245 | if (!check_slab(s, page) || | |
4246 | !on_freelist(s, page, NULL)) | |
4247 | return 0; | |
4248 | ||
4249 | /* Now we know that a valid freelist exists */ | |
39b26464 | 4250 | bitmap_zero(map, page->objects); |
53e15af0 | 4251 | |
5f80b13a CL |
4252 | get_map(s, page, map); |
4253 | for_each_object(p, s, addr, page->objects) { | |
4254 | if (test_bit(slab_index(p, s, addr), map)) | |
4255 | if (!check_object(s, page, p, SLUB_RED_INACTIVE)) | |
4256 | return 0; | |
53e15af0 CL |
4257 | } |
4258 | ||
224a88be | 4259 | for_each_object(p, s, addr, page->objects) |
7656c72b | 4260 | if (!test_bit(slab_index(p, s, addr), map)) |
37d57443 | 4261 | if (!check_object(s, page, p, SLUB_RED_ACTIVE)) |
53e15af0 CL |
4262 | return 0; |
4263 | return 1; | |
4264 | } | |
4265 | ||
434e245d CL |
4266 | static void validate_slab_slab(struct kmem_cache *s, struct page *page, |
4267 | unsigned long *map) | |
53e15af0 | 4268 | { |
881db7fb CL |
4269 | slab_lock(page); |
4270 | validate_slab(s, page, map); | |
4271 | slab_unlock(page); | |
53e15af0 CL |
4272 | } |
4273 | ||
434e245d CL |
4274 | static int validate_slab_node(struct kmem_cache *s, |
4275 | struct kmem_cache_node *n, unsigned long *map) | |
53e15af0 CL |
4276 | { |
4277 | unsigned long count = 0; | |
4278 | struct page *page; | |
4279 | unsigned long flags; | |
4280 | ||
4281 | spin_lock_irqsave(&n->list_lock, flags); | |
4282 | ||
4283 | list_for_each_entry(page, &n->partial, lru) { | |
434e245d | 4284 | validate_slab_slab(s, page, map); |
53e15af0 CL |
4285 | count++; |
4286 | } | |
4287 | if (count != n->nr_partial) | |
f9f58285 FF |
4288 | pr_err("SLUB %s: %ld partial slabs counted but counter=%ld\n", |
4289 | s->name, count, n->nr_partial); | |
53e15af0 CL |
4290 | |
4291 | if (!(s->flags & SLAB_STORE_USER)) | |
4292 | goto out; | |
4293 | ||
4294 | list_for_each_entry(page, &n->full, lru) { | |
434e245d | 4295 | validate_slab_slab(s, page, map); |
53e15af0 CL |
4296 | count++; |
4297 | } | |
4298 | if (count != atomic_long_read(&n->nr_slabs)) | |
f9f58285 FF |
4299 | pr_err("SLUB: %s %ld slabs counted but counter=%ld\n", |
4300 | s->name, count, atomic_long_read(&n->nr_slabs)); | |
53e15af0 CL |
4301 | |
4302 | out: | |
4303 | spin_unlock_irqrestore(&n->list_lock, flags); | |
4304 | return count; | |
4305 | } | |
4306 | ||
434e245d | 4307 | static long validate_slab_cache(struct kmem_cache *s) |
53e15af0 CL |
4308 | { |
4309 | int node; | |
4310 | unsigned long count = 0; | |
205ab99d | 4311 | unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) * |
434e245d | 4312 | sizeof(unsigned long), GFP_KERNEL); |
fa45dc25 | 4313 | struct kmem_cache_node *n; |
434e245d CL |
4314 | |
4315 | if (!map) | |
4316 | return -ENOMEM; | |
53e15af0 CL |
4317 | |
4318 | flush_all(s); | |
fa45dc25 | 4319 | for_each_kmem_cache_node(s, node, n) |
434e245d | 4320 | count += validate_slab_node(s, n, map); |
434e245d | 4321 | kfree(map); |
53e15af0 CL |
4322 | return count; |
4323 | } | |
88a420e4 | 4324 | /* |
672bba3a | 4325 | * Generate lists of code addresses where slabcache objects are allocated |
88a420e4 CL |
4326 | * and freed. |
4327 | */ | |
4328 | ||
4329 | struct location { | |
4330 | unsigned long count; | |
ce71e27c | 4331 | unsigned long addr; |
45edfa58 CL |
4332 | long long sum_time; |
4333 | long min_time; | |
4334 | long max_time; | |
4335 | long min_pid; | |
4336 | long max_pid; | |
174596a0 | 4337 | DECLARE_BITMAP(cpus, NR_CPUS); |
45edfa58 | 4338 | nodemask_t nodes; |
88a420e4 CL |
4339 | }; |
4340 | ||
4341 | struct loc_track { | |
4342 | unsigned long max; | |
4343 | unsigned long count; | |
4344 | struct location *loc; | |
4345 | }; | |
4346 | ||
4347 | static void free_loc_track(struct loc_track *t) | |
4348 | { | |
4349 | if (t->max) | |
4350 | free_pages((unsigned long)t->loc, | |
4351 | get_order(sizeof(struct location) * t->max)); | |
4352 | } | |
4353 | ||
68dff6a9 | 4354 | static int alloc_loc_track(struct loc_track *t, unsigned long max, gfp_t flags) |
88a420e4 CL |
4355 | { |
4356 | struct location *l; | |
4357 | int order; | |
4358 | ||
88a420e4 CL |
4359 | order = get_order(sizeof(struct location) * max); |
4360 | ||
68dff6a9 | 4361 | l = (void *)__get_free_pages(flags, order); |
88a420e4 CL |
4362 | if (!l) |
4363 | return 0; | |
4364 | ||
4365 | if (t->count) { | |
4366 | memcpy(l, t->loc, sizeof(struct location) * t->count); | |
4367 | free_loc_track(t); | |
4368 | } | |
4369 | t->max = max; | |
4370 | t->loc = l; | |
4371 | return 1; | |
4372 | } | |
4373 | ||
4374 | static int add_location(struct loc_track *t, struct kmem_cache *s, | |
45edfa58 | 4375 | const struct track *track) |
88a420e4 CL |
4376 | { |
4377 | long start, end, pos; | |
4378 | struct location *l; | |
ce71e27c | 4379 | unsigned long caddr; |
45edfa58 | 4380 | unsigned long age = jiffies - track->when; |
88a420e4 CL |
4381 | |
4382 | start = -1; | |
4383 | end = t->count; | |
4384 | ||
4385 | for ( ; ; ) { | |
4386 | pos = start + (end - start + 1) / 2; | |
4387 | ||
4388 | /* | |
4389 | * There is nothing at "end". If we end up there | |
4390 | * we need to add something to before end. | |
4391 | */ | |
4392 | if (pos == end) | |
4393 | break; | |
4394 | ||
4395 | caddr = t->loc[pos].addr; | |
45edfa58 CL |
4396 | if (track->addr == caddr) { |
4397 | ||
4398 | l = &t->loc[pos]; | |
4399 | l->count++; | |
4400 | if (track->when) { | |
4401 | l->sum_time += age; | |
4402 | if (age < l->min_time) | |
4403 | l->min_time = age; | |
4404 | if (age > l->max_time) | |
4405 | l->max_time = age; | |
4406 | ||
4407 | if (track->pid < l->min_pid) | |
4408 | l->min_pid = track->pid; | |
4409 | if (track->pid > l->max_pid) | |
4410 | l->max_pid = track->pid; | |
4411 | ||
174596a0 RR |
4412 | cpumask_set_cpu(track->cpu, |
4413 | to_cpumask(l->cpus)); | |
45edfa58 CL |
4414 | } |
4415 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
88a420e4 CL |
4416 | return 1; |
4417 | } | |
4418 | ||
45edfa58 | 4419 | if (track->addr < caddr) |
88a420e4 CL |
4420 | end = pos; |
4421 | else | |
4422 | start = pos; | |
4423 | } | |
4424 | ||
4425 | /* | |
672bba3a | 4426 | * Not found. Insert new tracking element. |
88a420e4 | 4427 | */ |
68dff6a9 | 4428 | if (t->count >= t->max && !alloc_loc_track(t, 2 * t->max, GFP_ATOMIC)) |
88a420e4 CL |
4429 | return 0; |
4430 | ||
4431 | l = t->loc + pos; | |
4432 | if (pos < t->count) | |
4433 | memmove(l + 1, l, | |
4434 | (t->count - pos) * sizeof(struct location)); | |
4435 | t->count++; | |
4436 | l->count = 1; | |
45edfa58 CL |
4437 | l->addr = track->addr; |
4438 | l->sum_time = age; | |
4439 | l->min_time = age; | |
4440 | l->max_time = age; | |
4441 | l->min_pid = track->pid; | |
4442 | l->max_pid = track->pid; | |
174596a0 RR |
4443 | cpumask_clear(to_cpumask(l->cpus)); |
4444 | cpumask_set_cpu(track->cpu, to_cpumask(l->cpus)); | |
45edfa58 CL |
4445 | nodes_clear(l->nodes); |
4446 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
88a420e4 CL |
4447 | return 1; |
4448 | } | |
4449 | ||
4450 | static void process_slab(struct loc_track *t, struct kmem_cache *s, | |
bbd7d57b | 4451 | struct page *page, enum track_item alloc, |
a5dd5c11 | 4452 | unsigned long *map) |
88a420e4 | 4453 | { |
a973e9dd | 4454 | void *addr = page_address(page); |
88a420e4 CL |
4455 | void *p; |
4456 | ||
39b26464 | 4457 | bitmap_zero(map, page->objects); |
5f80b13a | 4458 | get_map(s, page, map); |
88a420e4 | 4459 | |
224a88be | 4460 | for_each_object(p, s, addr, page->objects) |
45edfa58 CL |
4461 | if (!test_bit(slab_index(p, s, addr), map)) |
4462 | add_location(t, s, get_track(s, p, alloc)); | |
88a420e4 CL |
4463 | } |
4464 | ||
4465 | static int list_locations(struct kmem_cache *s, char *buf, | |
4466 | enum track_item alloc) | |
4467 | { | |
e374d483 | 4468 | int len = 0; |
88a420e4 | 4469 | unsigned long i; |
68dff6a9 | 4470 | struct loc_track t = { 0, 0, NULL }; |
88a420e4 | 4471 | int node; |
bbd7d57b ED |
4472 | unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) * |
4473 | sizeof(unsigned long), GFP_KERNEL); | |
fa45dc25 | 4474 | struct kmem_cache_node *n; |
88a420e4 | 4475 | |
bbd7d57b ED |
4476 | if (!map || !alloc_loc_track(&t, PAGE_SIZE / sizeof(struct location), |
4477 | GFP_TEMPORARY)) { | |
4478 | kfree(map); | |
68dff6a9 | 4479 | return sprintf(buf, "Out of memory\n"); |
bbd7d57b | 4480 | } |
88a420e4 CL |
4481 | /* Push back cpu slabs */ |
4482 | flush_all(s); | |
4483 | ||
fa45dc25 | 4484 | for_each_kmem_cache_node(s, node, n) { |
88a420e4 CL |
4485 | unsigned long flags; |
4486 | struct page *page; | |
4487 | ||
9e86943b | 4488 | if (!atomic_long_read(&n->nr_slabs)) |
88a420e4 CL |
4489 | continue; |
4490 | ||
4491 | spin_lock_irqsave(&n->list_lock, flags); | |
4492 | list_for_each_entry(page, &n->partial, lru) | |
bbd7d57b | 4493 | process_slab(&t, s, page, alloc, map); |
88a420e4 | 4494 | list_for_each_entry(page, &n->full, lru) |
bbd7d57b | 4495 | process_slab(&t, s, page, alloc, map); |
88a420e4 CL |
4496 | spin_unlock_irqrestore(&n->list_lock, flags); |
4497 | } | |
4498 | ||
4499 | for (i = 0; i < t.count; i++) { | |
45edfa58 | 4500 | struct location *l = &t.loc[i]; |
88a420e4 | 4501 | |
9c246247 | 4502 | if (len > PAGE_SIZE - KSYM_SYMBOL_LEN - 100) |
88a420e4 | 4503 | break; |
e374d483 | 4504 | len += sprintf(buf + len, "%7ld ", l->count); |
45edfa58 CL |
4505 | |
4506 | if (l->addr) | |
62c70bce | 4507 | len += sprintf(buf + len, "%pS", (void *)l->addr); |
88a420e4 | 4508 | else |
e374d483 | 4509 | len += sprintf(buf + len, "<not-available>"); |
45edfa58 CL |
4510 | |
4511 | if (l->sum_time != l->min_time) { | |
e374d483 | 4512 | len += sprintf(buf + len, " age=%ld/%ld/%ld", |
f8bd2258 RZ |
4513 | l->min_time, |
4514 | (long)div_u64(l->sum_time, l->count), | |
4515 | l->max_time); | |
45edfa58 | 4516 | } else |
e374d483 | 4517 | len += sprintf(buf + len, " age=%ld", |
45edfa58 CL |
4518 | l->min_time); |
4519 | ||
4520 | if (l->min_pid != l->max_pid) | |
e374d483 | 4521 | len += sprintf(buf + len, " pid=%ld-%ld", |
45edfa58 CL |
4522 | l->min_pid, l->max_pid); |
4523 | else | |
e374d483 | 4524 | len += sprintf(buf + len, " pid=%ld", |
45edfa58 CL |
4525 | l->min_pid); |
4526 | ||
174596a0 RR |
4527 | if (num_online_cpus() > 1 && |
4528 | !cpumask_empty(to_cpumask(l->cpus)) && | |
5024c1d7 TH |
4529 | len < PAGE_SIZE - 60) |
4530 | len += scnprintf(buf + len, PAGE_SIZE - len - 50, | |
4531 | " cpus=%*pbl", | |
4532 | cpumask_pr_args(to_cpumask(l->cpus))); | |
45edfa58 | 4533 | |
62bc62a8 | 4534 | if (nr_online_nodes > 1 && !nodes_empty(l->nodes) && |
5024c1d7 TH |
4535 | len < PAGE_SIZE - 60) |
4536 | len += scnprintf(buf + len, PAGE_SIZE - len - 50, | |
4537 | " nodes=%*pbl", | |
4538 | nodemask_pr_args(&l->nodes)); | |
45edfa58 | 4539 | |
e374d483 | 4540 | len += sprintf(buf + len, "\n"); |
88a420e4 CL |
4541 | } |
4542 | ||
4543 | free_loc_track(&t); | |
bbd7d57b | 4544 | kfree(map); |
88a420e4 | 4545 | if (!t.count) |
e374d483 HH |
4546 | len += sprintf(buf, "No data\n"); |
4547 | return len; | |
88a420e4 | 4548 | } |
ab4d5ed5 | 4549 | #endif |
88a420e4 | 4550 | |
a5a84755 | 4551 | #ifdef SLUB_RESILIENCY_TEST |
c07b8183 | 4552 | static void __init resiliency_test(void) |
a5a84755 CL |
4553 | { |
4554 | u8 *p; | |
4555 | ||
95a05b42 | 4556 | BUILD_BUG_ON(KMALLOC_MIN_SIZE > 16 || KMALLOC_SHIFT_HIGH < 10); |
a5a84755 | 4557 | |
f9f58285 FF |
4558 | pr_err("SLUB resiliency testing\n"); |
4559 | pr_err("-----------------------\n"); | |
4560 | pr_err("A. Corruption after allocation\n"); | |
a5a84755 CL |
4561 | |
4562 | p = kzalloc(16, GFP_KERNEL); | |
4563 | p[16] = 0x12; | |
f9f58285 FF |
4564 | pr_err("\n1. kmalloc-16: Clobber Redzone/next pointer 0x12->0x%p\n\n", |
4565 | p + 16); | |
a5a84755 CL |
4566 | |
4567 | validate_slab_cache(kmalloc_caches[4]); | |
4568 | ||
4569 | /* Hmmm... The next two are dangerous */ | |
4570 | p = kzalloc(32, GFP_KERNEL); | |
4571 | p[32 + sizeof(void *)] = 0x34; | |
f9f58285 FF |
4572 | pr_err("\n2. kmalloc-32: Clobber next pointer/next slab 0x34 -> -0x%p\n", |
4573 | p); | |
4574 | pr_err("If allocated object is overwritten then not detectable\n\n"); | |
a5a84755 CL |
4575 | |
4576 | validate_slab_cache(kmalloc_caches[5]); | |
4577 | p = kzalloc(64, GFP_KERNEL); | |
4578 | p += 64 + (get_cycles() & 0xff) * sizeof(void *); | |
4579 | *p = 0x56; | |
f9f58285 FF |
4580 | pr_err("\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n", |
4581 | p); | |
4582 | pr_err("If allocated object is overwritten then not detectable\n\n"); | |
a5a84755 CL |
4583 | validate_slab_cache(kmalloc_caches[6]); |
4584 | ||
f9f58285 | 4585 | pr_err("\nB. Corruption after free\n"); |
a5a84755 CL |
4586 | p = kzalloc(128, GFP_KERNEL); |
4587 | kfree(p); | |
4588 | *p = 0x78; | |
f9f58285 | 4589 | pr_err("1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p); |
a5a84755 CL |
4590 | validate_slab_cache(kmalloc_caches[7]); |
4591 | ||
4592 | p = kzalloc(256, GFP_KERNEL); | |
4593 | kfree(p); | |
4594 | p[50] = 0x9a; | |
f9f58285 | 4595 | pr_err("\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n", p); |
a5a84755 CL |
4596 | validate_slab_cache(kmalloc_caches[8]); |
4597 | ||
4598 | p = kzalloc(512, GFP_KERNEL); | |
4599 | kfree(p); | |
4600 | p[512] = 0xab; | |
f9f58285 | 4601 | pr_err("\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p); |
a5a84755 CL |
4602 | validate_slab_cache(kmalloc_caches[9]); |
4603 | } | |
4604 | #else | |
4605 | #ifdef CONFIG_SYSFS | |
4606 | static void resiliency_test(void) {}; | |
4607 | #endif | |
4608 | #endif | |
4609 | ||
ab4d5ed5 | 4610 | #ifdef CONFIG_SYSFS |
81819f0f | 4611 | enum slab_stat_type { |
205ab99d CL |
4612 | SL_ALL, /* All slabs */ |
4613 | SL_PARTIAL, /* Only partially allocated slabs */ | |
4614 | SL_CPU, /* Only slabs used for cpu caches */ | |
4615 | SL_OBJECTS, /* Determine allocated objects not slabs */ | |
4616 | SL_TOTAL /* Determine object capacity not slabs */ | |
81819f0f CL |
4617 | }; |
4618 | ||
205ab99d | 4619 | #define SO_ALL (1 << SL_ALL) |
81819f0f CL |
4620 | #define SO_PARTIAL (1 << SL_PARTIAL) |
4621 | #define SO_CPU (1 << SL_CPU) | |
4622 | #define SO_OBJECTS (1 << SL_OBJECTS) | |
205ab99d | 4623 | #define SO_TOTAL (1 << SL_TOTAL) |
81819f0f | 4624 | |
62e5c4b4 CG |
4625 | static ssize_t show_slab_objects(struct kmem_cache *s, |
4626 | char *buf, unsigned long flags) | |
81819f0f CL |
4627 | { |
4628 | unsigned long total = 0; | |
81819f0f CL |
4629 | int node; |
4630 | int x; | |
4631 | unsigned long *nodes; | |
81819f0f | 4632 | |
e35e1a97 | 4633 | nodes = kzalloc(sizeof(unsigned long) * nr_node_ids, GFP_KERNEL); |
62e5c4b4 CG |
4634 | if (!nodes) |
4635 | return -ENOMEM; | |
81819f0f | 4636 | |
205ab99d CL |
4637 | if (flags & SO_CPU) { |
4638 | int cpu; | |
81819f0f | 4639 | |
205ab99d | 4640 | for_each_possible_cpu(cpu) { |
d0e0ac97 CG |
4641 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, |
4642 | cpu); | |
ec3ab083 | 4643 | int node; |
49e22585 | 4644 | struct page *page; |
dfb4f096 | 4645 | |
4db0c3c2 | 4646 | page = READ_ONCE(c->page); |
ec3ab083 CL |
4647 | if (!page) |
4648 | continue; | |
205ab99d | 4649 | |
ec3ab083 CL |
4650 | node = page_to_nid(page); |
4651 | if (flags & SO_TOTAL) | |
4652 | x = page->objects; | |
4653 | else if (flags & SO_OBJECTS) | |
4654 | x = page->inuse; | |
4655 | else | |
4656 | x = 1; | |
49e22585 | 4657 | |
ec3ab083 CL |
4658 | total += x; |
4659 | nodes[node] += x; | |
4660 | ||
4db0c3c2 | 4661 | page = READ_ONCE(c->partial); |
49e22585 | 4662 | if (page) { |
8afb1474 LZ |
4663 | node = page_to_nid(page); |
4664 | if (flags & SO_TOTAL) | |
4665 | WARN_ON_ONCE(1); | |
4666 | else if (flags & SO_OBJECTS) | |
4667 | WARN_ON_ONCE(1); | |
4668 | else | |
4669 | x = page->pages; | |
bc6697d8 ED |
4670 | total += x; |
4671 | nodes[node] += x; | |
49e22585 | 4672 | } |
81819f0f CL |
4673 | } |
4674 | } | |
4675 | ||
bfc8c901 | 4676 | get_online_mems(); |
ab4d5ed5 | 4677 | #ifdef CONFIG_SLUB_DEBUG |
205ab99d | 4678 | if (flags & SO_ALL) { |
fa45dc25 CL |
4679 | struct kmem_cache_node *n; |
4680 | ||
4681 | for_each_kmem_cache_node(s, node, n) { | |
205ab99d | 4682 | |
d0e0ac97 CG |
4683 | if (flags & SO_TOTAL) |
4684 | x = atomic_long_read(&n->total_objects); | |
4685 | else if (flags & SO_OBJECTS) | |
4686 | x = atomic_long_read(&n->total_objects) - | |
4687 | count_partial(n, count_free); | |
81819f0f | 4688 | else |
205ab99d | 4689 | x = atomic_long_read(&n->nr_slabs); |
81819f0f CL |
4690 | total += x; |
4691 | nodes[node] += x; | |
4692 | } | |
4693 | ||
ab4d5ed5 CL |
4694 | } else |
4695 | #endif | |
4696 | if (flags & SO_PARTIAL) { | |
fa45dc25 | 4697 | struct kmem_cache_node *n; |
81819f0f | 4698 | |
fa45dc25 | 4699 | for_each_kmem_cache_node(s, node, n) { |
205ab99d CL |
4700 | if (flags & SO_TOTAL) |
4701 | x = count_partial(n, count_total); | |
4702 | else if (flags & SO_OBJECTS) | |
4703 | x = count_partial(n, count_inuse); | |
81819f0f | 4704 | else |
205ab99d | 4705 | x = n->nr_partial; |
81819f0f CL |
4706 | total += x; |
4707 | nodes[node] += x; | |
4708 | } | |
4709 | } | |
81819f0f CL |
4710 | x = sprintf(buf, "%lu", total); |
4711 | #ifdef CONFIG_NUMA | |
fa45dc25 | 4712 | for (node = 0; node < nr_node_ids; node++) |
81819f0f CL |
4713 | if (nodes[node]) |
4714 | x += sprintf(buf + x, " N%d=%lu", | |
4715 | node, nodes[node]); | |
4716 | #endif | |
bfc8c901 | 4717 | put_online_mems(); |
81819f0f CL |
4718 | kfree(nodes); |
4719 | return x + sprintf(buf + x, "\n"); | |
4720 | } | |
4721 | ||
ab4d5ed5 | 4722 | #ifdef CONFIG_SLUB_DEBUG |
81819f0f CL |
4723 | static int any_slab_objects(struct kmem_cache *s) |
4724 | { | |
4725 | int node; | |
fa45dc25 | 4726 | struct kmem_cache_node *n; |
81819f0f | 4727 | |
fa45dc25 | 4728 | for_each_kmem_cache_node(s, node, n) |
4ea33e2d | 4729 | if (atomic_long_read(&n->total_objects)) |
81819f0f | 4730 | return 1; |
fa45dc25 | 4731 | |
81819f0f CL |
4732 | return 0; |
4733 | } | |
ab4d5ed5 | 4734 | #endif |
81819f0f CL |
4735 | |
4736 | #define to_slab_attr(n) container_of(n, struct slab_attribute, attr) | |
497888cf | 4737 | #define to_slab(n) container_of(n, struct kmem_cache, kobj) |
81819f0f CL |
4738 | |
4739 | struct slab_attribute { | |
4740 | struct attribute attr; | |
4741 | ssize_t (*show)(struct kmem_cache *s, char *buf); | |
4742 | ssize_t (*store)(struct kmem_cache *s, const char *x, size_t count); | |
4743 | }; | |
4744 | ||
4745 | #define SLAB_ATTR_RO(_name) \ | |
ab067e99 VK |
4746 | static struct slab_attribute _name##_attr = \ |
4747 | __ATTR(_name, 0400, _name##_show, NULL) | |
81819f0f CL |
4748 | |
4749 | #define SLAB_ATTR(_name) \ | |
4750 | static struct slab_attribute _name##_attr = \ | |
ab067e99 | 4751 | __ATTR(_name, 0600, _name##_show, _name##_store) |
81819f0f | 4752 | |
81819f0f CL |
4753 | static ssize_t slab_size_show(struct kmem_cache *s, char *buf) |
4754 | { | |
4755 | return sprintf(buf, "%d\n", s->size); | |
4756 | } | |
4757 | SLAB_ATTR_RO(slab_size); | |
4758 | ||
4759 | static ssize_t align_show(struct kmem_cache *s, char *buf) | |
4760 | { | |
4761 | return sprintf(buf, "%d\n", s->align); | |
4762 | } | |
4763 | SLAB_ATTR_RO(align); | |
4764 | ||
4765 | static ssize_t object_size_show(struct kmem_cache *s, char *buf) | |
4766 | { | |
3b0efdfa | 4767 | return sprintf(buf, "%d\n", s->object_size); |
81819f0f CL |
4768 | } |
4769 | SLAB_ATTR_RO(object_size); | |
4770 | ||
4771 | static ssize_t objs_per_slab_show(struct kmem_cache *s, char *buf) | |
4772 | { | |
834f3d11 | 4773 | return sprintf(buf, "%d\n", oo_objects(s->oo)); |
81819f0f CL |
4774 | } |
4775 | SLAB_ATTR_RO(objs_per_slab); | |
4776 | ||
06b285dc CL |
4777 | static ssize_t order_store(struct kmem_cache *s, |
4778 | const char *buf, size_t length) | |
4779 | { | |
0121c619 CL |
4780 | unsigned long order; |
4781 | int err; | |
4782 | ||
3dbb95f7 | 4783 | err = kstrtoul(buf, 10, &order); |
0121c619 CL |
4784 | if (err) |
4785 | return err; | |
06b285dc CL |
4786 | |
4787 | if (order > slub_max_order || order < slub_min_order) | |
4788 | return -EINVAL; | |
4789 | ||
4790 | calculate_sizes(s, order); | |
4791 | return length; | |
4792 | } | |
4793 | ||
81819f0f CL |
4794 | static ssize_t order_show(struct kmem_cache *s, char *buf) |
4795 | { | |
834f3d11 | 4796 | return sprintf(buf, "%d\n", oo_order(s->oo)); |
81819f0f | 4797 | } |
06b285dc | 4798 | SLAB_ATTR(order); |
81819f0f | 4799 | |
73d342b1 DR |
4800 | static ssize_t min_partial_show(struct kmem_cache *s, char *buf) |
4801 | { | |
4802 | return sprintf(buf, "%lu\n", s->min_partial); | |
4803 | } | |
4804 | ||
4805 | static ssize_t min_partial_store(struct kmem_cache *s, const char *buf, | |
4806 | size_t length) | |
4807 | { | |
4808 | unsigned long min; | |
4809 | int err; | |
4810 | ||
3dbb95f7 | 4811 | err = kstrtoul(buf, 10, &min); |
73d342b1 DR |
4812 | if (err) |
4813 | return err; | |
4814 | ||
c0bdb232 | 4815 | set_min_partial(s, min); |
73d342b1 DR |
4816 | return length; |
4817 | } | |
4818 | SLAB_ATTR(min_partial); | |
4819 | ||
49e22585 CL |
4820 | static ssize_t cpu_partial_show(struct kmem_cache *s, char *buf) |
4821 | { | |
4822 | return sprintf(buf, "%u\n", s->cpu_partial); | |
4823 | } | |
4824 | ||
4825 | static ssize_t cpu_partial_store(struct kmem_cache *s, const char *buf, | |
4826 | size_t length) | |
4827 | { | |
4828 | unsigned long objects; | |
4829 | int err; | |
4830 | ||
3dbb95f7 | 4831 | err = kstrtoul(buf, 10, &objects); |
49e22585 CL |
4832 | if (err) |
4833 | return err; | |
345c905d | 4834 | if (objects && !kmem_cache_has_cpu_partial(s)) |
74ee4ef1 | 4835 | return -EINVAL; |
49e22585 CL |
4836 | |
4837 | s->cpu_partial = objects; | |
4838 | flush_all(s); | |
4839 | return length; | |
4840 | } | |
4841 | SLAB_ATTR(cpu_partial); | |
4842 | ||
81819f0f CL |
4843 | static ssize_t ctor_show(struct kmem_cache *s, char *buf) |
4844 | { | |
62c70bce JP |
4845 | if (!s->ctor) |
4846 | return 0; | |
4847 | return sprintf(buf, "%pS\n", s->ctor); | |
81819f0f CL |
4848 | } |
4849 | SLAB_ATTR_RO(ctor); | |
4850 | ||
81819f0f CL |
4851 | static ssize_t aliases_show(struct kmem_cache *s, char *buf) |
4852 | { | |
4307c14f | 4853 | return sprintf(buf, "%d\n", s->refcount < 0 ? 0 : s->refcount - 1); |
81819f0f CL |
4854 | } |
4855 | SLAB_ATTR_RO(aliases); | |
4856 | ||
81819f0f CL |
4857 | static ssize_t partial_show(struct kmem_cache *s, char *buf) |
4858 | { | |
d9acf4b7 | 4859 | return show_slab_objects(s, buf, SO_PARTIAL); |
81819f0f CL |
4860 | } |
4861 | SLAB_ATTR_RO(partial); | |
4862 | ||
4863 | static ssize_t cpu_slabs_show(struct kmem_cache *s, char *buf) | |
4864 | { | |
d9acf4b7 | 4865 | return show_slab_objects(s, buf, SO_CPU); |
81819f0f CL |
4866 | } |
4867 | SLAB_ATTR_RO(cpu_slabs); | |
4868 | ||
4869 | static ssize_t objects_show(struct kmem_cache *s, char *buf) | |
4870 | { | |
205ab99d | 4871 | return show_slab_objects(s, buf, SO_ALL|SO_OBJECTS); |
81819f0f CL |
4872 | } |
4873 | SLAB_ATTR_RO(objects); | |
4874 | ||
205ab99d CL |
4875 | static ssize_t objects_partial_show(struct kmem_cache *s, char *buf) |
4876 | { | |
4877 | return show_slab_objects(s, buf, SO_PARTIAL|SO_OBJECTS); | |
4878 | } | |
4879 | SLAB_ATTR_RO(objects_partial); | |
4880 | ||
49e22585 CL |
4881 | static ssize_t slabs_cpu_partial_show(struct kmem_cache *s, char *buf) |
4882 | { | |
4883 | int objects = 0; | |
4884 | int pages = 0; | |
4885 | int cpu; | |
4886 | int len; | |
4887 | ||
4888 | for_each_online_cpu(cpu) { | |
4889 | struct page *page = per_cpu_ptr(s->cpu_slab, cpu)->partial; | |
4890 | ||
4891 | if (page) { | |
4892 | pages += page->pages; | |
4893 | objects += page->pobjects; | |
4894 | } | |
4895 | } | |
4896 | ||
4897 | len = sprintf(buf, "%d(%d)", objects, pages); | |
4898 | ||
4899 | #ifdef CONFIG_SMP | |
4900 | for_each_online_cpu(cpu) { | |
4901 | struct page *page = per_cpu_ptr(s->cpu_slab, cpu) ->partial; | |
4902 | ||
4903 | if (page && len < PAGE_SIZE - 20) | |
4904 | len += sprintf(buf + len, " C%d=%d(%d)", cpu, | |
4905 | page->pobjects, page->pages); | |
4906 | } | |
4907 | #endif | |
4908 | return len + sprintf(buf + len, "\n"); | |
4909 | } | |
4910 | SLAB_ATTR_RO(slabs_cpu_partial); | |
4911 | ||
a5a84755 CL |
4912 | static ssize_t reclaim_account_show(struct kmem_cache *s, char *buf) |
4913 | { | |
4914 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_RECLAIM_ACCOUNT)); | |
4915 | } | |
4916 | ||
4917 | static ssize_t reclaim_account_store(struct kmem_cache *s, | |
4918 | const char *buf, size_t length) | |
4919 | { | |
4920 | s->flags &= ~SLAB_RECLAIM_ACCOUNT; | |
4921 | if (buf[0] == '1') | |
4922 | s->flags |= SLAB_RECLAIM_ACCOUNT; | |
4923 | return length; | |
4924 | } | |
4925 | SLAB_ATTR(reclaim_account); | |
4926 | ||
4927 | static ssize_t hwcache_align_show(struct kmem_cache *s, char *buf) | |
4928 | { | |
4929 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_HWCACHE_ALIGN)); | |
4930 | } | |
4931 | SLAB_ATTR_RO(hwcache_align); | |
4932 | ||
4933 | #ifdef CONFIG_ZONE_DMA | |
4934 | static ssize_t cache_dma_show(struct kmem_cache *s, char *buf) | |
4935 | { | |
4936 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_CACHE_DMA)); | |
4937 | } | |
4938 | SLAB_ATTR_RO(cache_dma); | |
4939 | #endif | |
4940 | ||
4941 | static ssize_t destroy_by_rcu_show(struct kmem_cache *s, char *buf) | |
4942 | { | |
4943 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_DESTROY_BY_RCU)); | |
4944 | } | |
4945 | SLAB_ATTR_RO(destroy_by_rcu); | |
4946 | ||
ab9a0f19 LJ |
4947 | static ssize_t reserved_show(struct kmem_cache *s, char *buf) |
4948 | { | |
4949 | return sprintf(buf, "%d\n", s->reserved); | |
4950 | } | |
4951 | SLAB_ATTR_RO(reserved); | |
4952 | ||
ab4d5ed5 | 4953 | #ifdef CONFIG_SLUB_DEBUG |
a5a84755 CL |
4954 | static ssize_t slabs_show(struct kmem_cache *s, char *buf) |
4955 | { | |
4956 | return show_slab_objects(s, buf, SO_ALL); | |
4957 | } | |
4958 | SLAB_ATTR_RO(slabs); | |
4959 | ||
205ab99d CL |
4960 | static ssize_t total_objects_show(struct kmem_cache *s, char *buf) |
4961 | { | |
4962 | return show_slab_objects(s, buf, SO_ALL|SO_TOTAL); | |
4963 | } | |
4964 | SLAB_ATTR_RO(total_objects); | |
4965 | ||
81819f0f CL |
4966 | static ssize_t sanity_checks_show(struct kmem_cache *s, char *buf) |
4967 | { | |
becfda68 | 4968 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_CONSISTENCY_CHECKS)); |
81819f0f CL |
4969 | } |
4970 | ||
4971 | static ssize_t sanity_checks_store(struct kmem_cache *s, | |
4972 | const char *buf, size_t length) | |
4973 | { | |
becfda68 | 4974 | s->flags &= ~SLAB_CONSISTENCY_CHECKS; |
b789ef51 CL |
4975 | if (buf[0] == '1') { |
4976 | s->flags &= ~__CMPXCHG_DOUBLE; | |
becfda68 | 4977 | s->flags |= SLAB_CONSISTENCY_CHECKS; |
b789ef51 | 4978 | } |
81819f0f CL |
4979 | return length; |
4980 | } | |
4981 | SLAB_ATTR(sanity_checks); | |
4982 | ||
4983 | static ssize_t trace_show(struct kmem_cache *s, char *buf) | |
4984 | { | |
4985 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_TRACE)); | |
4986 | } | |
4987 | ||
4988 | static ssize_t trace_store(struct kmem_cache *s, const char *buf, | |
4989 | size_t length) | |
4990 | { | |
c9e16131 CL |
4991 | /* |
4992 | * Tracing a merged cache is going to give confusing results | |
4993 | * as well as cause other issues like converting a mergeable | |
4994 | * cache into an umergeable one. | |
4995 | */ | |
4996 | if (s->refcount > 1) | |
4997 | return -EINVAL; | |
4998 | ||
81819f0f | 4999 | s->flags &= ~SLAB_TRACE; |
b789ef51 CL |
5000 | if (buf[0] == '1') { |
5001 | s->flags &= ~__CMPXCHG_DOUBLE; | |
81819f0f | 5002 | s->flags |= SLAB_TRACE; |
b789ef51 | 5003 | } |
81819f0f CL |
5004 | return length; |
5005 | } | |
5006 | SLAB_ATTR(trace); | |
5007 | ||
81819f0f CL |
5008 | static ssize_t red_zone_show(struct kmem_cache *s, char *buf) |
5009 | { | |
5010 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_RED_ZONE)); | |
5011 | } | |
5012 | ||
5013 | static ssize_t red_zone_store(struct kmem_cache *s, | |
5014 | const char *buf, size_t length) | |
5015 | { | |
5016 | if (any_slab_objects(s)) | |
5017 | return -EBUSY; | |
5018 | ||
5019 | s->flags &= ~SLAB_RED_ZONE; | |
b789ef51 | 5020 | if (buf[0] == '1') { |
81819f0f | 5021 | s->flags |= SLAB_RED_ZONE; |
b789ef51 | 5022 | } |
06b285dc | 5023 | calculate_sizes(s, -1); |
81819f0f CL |
5024 | return length; |
5025 | } | |
5026 | SLAB_ATTR(red_zone); | |
5027 | ||
5028 | static ssize_t poison_show(struct kmem_cache *s, char *buf) | |
5029 | { | |
5030 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_POISON)); | |
5031 | } | |
5032 | ||
5033 | static ssize_t poison_store(struct kmem_cache *s, | |
5034 | const char *buf, size_t length) | |
5035 | { | |
5036 | if (any_slab_objects(s)) | |
5037 | return -EBUSY; | |
5038 | ||
5039 | s->flags &= ~SLAB_POISON; | |
b789ef51 | 5040 | if (buf[0] == '1') { |
81819f0f | 5041 | s->flags |= SLAB_POISON; |
b789ef51 | 5042 | } |
06b285dc | 5043 | calculate_sizes(s, -1); |
81819f0f CL |
5044 | return length; |
5045 | } | |
5046 | SLAB_ATTR(poison); | |
5047 | ||
5048 | static ssize_t store_user_show(struct kmem_cache *s, char *buf) | |
5049 | { | |
5050 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_STORE_USER)); | |
5051 | } | |
5052 | ||
5053 | static ssize_t store_user_store(struct kmem_cache *s, | |
5054 | const char *buf, size_t length) | |
5055 | { | |
5056 | if (any_slab_objects(s)) | |
5057 | return -EBUSY; | |
5058 | ||
5059 | s->flags &= ~SLAB_STORE_USER; | |
b789ef51 CL |
5060 | if (buf[0] == '1') { |
5061 | s->flags &= ~__CMPXCHG_DOUBLE; | |
81819f0f | 5062 | s->flags |= SLAB_STORE_USER; |
b789ef51 | 5063 | } |
06b285dc | 5064 | calculate_sizes(s, -1); |
81819f0f CL |
5065 | return length; |
5066 | } | |
5067 | SLAB_ATTR(store_user); | |
5068 | ||
53e15af0 CL |
5069 | static ssize_t validate_show(struct kmem_cache *s, char *buf) |
5070 | { | |
5071 | return 0; | |
5072 | } | |
5073 | ||
5074 | static ssize_t validate_store(struct kmem_cache *s, | |
5075 | const char *buf, size_t length) | |
5076 | { | |
434e245d CL |
5077 | int ret = -EINVAL; |
5078 | ||
5079 | if (buf[0] == '1') { | |
5080 | ret = validate_slab_cache(s); | |
5081 | if (ret >= 0) | |
5082 | ret = length; | |
5083 | } | |
5084 | return ret; | |
53e15af0 CL |
5085 | } |
5086 | SLAB_ATTR(validate); | |
a5a84755 CL |
5087 | |
5088 | static ssize_t alloc_calls_show(struct kmem_cache *s, char *buf) | |
5089 | { | |
5090 | if (!(s->flags & SLAB_STORE_USER)) | |
5091 | return -ENOSYS; | |
5092 | return list_locations(s, buf, TRACK_ALLOC); | |
5093 | } | |
5094 | SLAB_ATTR_RO(alloc_calls); | |
5095 | ||
5096 | static ssize_t free_calls_show(struct kmem_cache *s, char *buf) | |
5097 | { | |
5098 | if (!(s->flags & SLAB_STORE_USER)) | |
5099 | return -ENOSYS; | |
5100 | return list_locations(s, buf, TRACK_FREE); | |
5101 | } | |
5102 | SLAB_ATTR_RO(free_calls); | |
5103 | #endif /* CONFIG_SLUB_DEBUG */ | |
5104 | ||
5105 | #ifdef CONFIG_FAILSLAB | |
5106 | static ssize_t failslab_show(struct kmem_cache *s, char *buf) | |
5107 | { | |
5108 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_FAILSLAB)); | |
5109 | } | |
5110 | ||
5111 | static ssize_t failslab_store(struct kmem_cache *s, const char *buf, | |
5112 | size_t length) | |
5113 | { | |
c9e16131 CL |
5114 | if (s->refcount > 1) |
5115 | return -EINVAL; | |
5116 | ||
a5a84755 CL |
5117 | s->flags &= ~SLAB_FAILSLAB; |
5118 | if (buf[0] == '1') | |
5119 | s->flags |= SLAB_FAILSLAB; | |
5120 | return length; | |
5121 | } | |
5122 | SLAB_ATTR(failslab); | |
ab4d5ed5 | 5123 | #endif |
53e15af0 | 5124 | |
2086d26a CL |
5125 | static ssize_t shrink_show(struct kmem_cache *s, char *buf) |
5126 | { | |
5127 | return 0; | |
5128 | } | |
5129 | ||
5130 | static ssize_t shrink_store(struct kmem_cache *s, | |
5131 | const char *buf, size_t length) | |
5132 | { | |
832f37f5 VD |
5133 | if (buf[0] == '1') |
5134 | kmem_cache_shrink(s); | |
5135 | else | |
2086d26a CL |
5136 | return -EINVAL; |
5137 | return length; | |
5138 | } | |
5139 | SLAB_ATTR(shrink); | |
5140 | ||
81819f0f | 5141 | #ifdef CONFIG_NUMA |
9824601e | 5142 | static ssize_t remote_node_defrag_ratio_show(struct kmem_cache *s, char *buf) |
81819f0f | 5143 | { |
9824601e | 5144 | return sprintf(buf, "%d\n", s->remote_node_defrag_ratio / 10); |
81819f0f CL |
5145 | } |
5146 | ||
9824601e | 5147 | static ssize_t remote_node_defrag_ratio_store(struct kmem_cache *s, |
81819f0f CL |
5148 | const char *buf, size_t length) |
5149 | { | |
0121c619 CL |
5150 | unsigned long ratio; |
5151 | int err; | |
5152 | ||
3dbb95f7 | 5153 | err = kstrtoul(buf, 10, &ratio); |
0121c619 CL |
5154 | if (err) |
5155 | return err; | |
5156 | ||
e2cb96b7 | 5157 | if (ratio <= 100) |
0121c619 | 5158 | s->remote_node_defrag_ratio = ratio * 10; |
81819f0f | 5159 | |
81819f0f CL |
5160 | return length; |
5161 | } | |
9824601e | 5162 | SLAB_ATTR(remote_node_defrag_ratio); |
81819f0f CL |
5163 | #endif |
5164 | ||
8ff12cfc | 5165 | #ifdef CONFIG_SLUB_STATS |
8ff12cfc CL |
5166 | static int show_stat(struct kmem_cache *s, char *buf, enum stat_item si) |
5167 | { | |
5168 | unsigned long sum = 0; | |
5169 | int cpu; | |
5170 | int len; | |
5171 | int *data = kmalloc(nr_cpu_ids * sizeof(int), GFP_KERNEL); | |
5172 | ||
5173 | if (!data) | |
5174 | return -ENOMEM; | |
5175 | ||
5176 | for_each_online_cpu(cpu) { | |
9dfc6e68 | 5177 | unsigned x = per_cpu_ptr(s->cpu_slab, cpu)->stat[si]; |
8ff12cfc CL |
5178 | |
5179 | data[cpu] = x; | |
5180 | sum += x; | |
5181 | } | |
5182 | ||
5183 | len = sprintf(buf, "%lu", sum); | |
5184 | ||
50ef37b9 | 5185 | #ifdef CONFIG_SMP |
8ff12cfc CL |
5186 | for_each_online_cpu(cpu) { |
5187 | if (data[cpu] && len < PAGE_SIZE - 20) | |
50ef37b9 | 5188 | len += sprintf(buf + len, " C%d=%u", cpu, data[cpu]); |
8ff12cfc | 5189 | } |
50ef37b9 | 5190 | #endif |
8ff12cfc CL |
5191 | kfree(data); |
5192 | return len + sprintf(buf + len, "\n"); | |
5193 | } | |
5194 | ||
78eb00cc DR |
5195 | static void clear_stat(struct kmem_cache *s, enum stat_item si) |
5196 | { | |
5197 | int cpu; | |
5198 | ||
5199 | for_each_online_cpu(cpu) | |
9dfc6e68 | 5200 | per_cpu_ptr(s->cpu_slab, cpu)->stat[si] = 0; |
78eb00cc DR |
5201 | } |
5202 | ||
8ff12cfc CL |
5203 | #define STAT_ATTR(si, text) \ |
5204 | static ssize_t text##_show(struct kmem_cache *s, char *buf) \ | |
5205 | { \ | |
5206 | return show_stat(s, buf, si); \ | |
5207 | } \ | |
78eb00cc DR |
5208 | static ssize_t text##_store(struct kmem_cache *s, \ |
5209 | const char *buf, size_t length) \ | |
5210 | { \ | |
5211 | if (buf[0] != '0') \ | |
5212 | return -EINVAL; \ | |
5213 | clear_stat(s, si); \ | |
5214 | return length; \ | |
5215 | } \ | |
5216 | SLAB_ATTR(text); \ | |
8ff12cfc CL |
5217 | |
5218 | STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath); | |
5219 | STAT_ATTR(ALLOC_SLOWPATH, alloc_slowpath); | |
5220 | STAT_ATTR(FREE_FASTPATH, free_fastpath); | |
5221 | STAT_ATTR(FREE_SLOWPATH, free_slowpath); | |
5222 | STAT_ATTR(FREE_FROZEN, free_frozen); | |
5223 | STAT_ATTR(FREE_ADD_PARTIAL, free_add_partial); | |
5224 | STAT_ATTR(FREE_REMOVE_PARTIAL, free_remove_partial); | |
5225 | STAT_ATTR(ALLOC_FROM_PARTIAL, alloc_from_partial); | |
5226 | STAT_ATTR(ALLOC_SLAB, alloc_slab); | |
5227 | STAT_ATTR(ALLOC_REFILL, alloc_refill); | |
e36a2652 | 5228 | STAT_ATTR(ALLOC_NODE_MISMATCH, alloc_node_mismatch); |
8ff12cfc CL |
5229 | STAT_ATTR(FREE_SLAB, free_slab); |
5230 | STAT_ATTR(CPUSLAB_FLUSH, cpuslab_flush); | |
5231 | STAT_ATTR(DEACTIVATE_FULL, deactivate_full); | |
5232 | STAT_ATTR(DEACTIVATE_EMPTY, deactivate_empty); | |
5233 | STAT_ATTR(DEACTIVATE_TO_HEAD, deactivate_to_head); | |
5234 | STAT_ATTR(DEACTIVATE_TO_TAIL, deactivate_to_tail); | |
5235 | STAT_ATTR(DEACTIVATE_REMOTE_FREES, deactivate_remote_frees); | |
03e404af | 5236 | STAT_ATTR(DEACTIVATE_BYPASS, deactivate_bypass); |
65c3376a | 5237 | STAT_ATTR(ORDER_FALLBACK, order_fallback); |
b789ef51 CL |
5238 | STAT_ATTR(CMPXCHG_DOUBLE_CPU_FAIL, cmpxchg_double_cpu_fail); |
5239 | STAT_ATTR(CMPXCHG_DOUBLE_FAIL, cmpxchg_double_fail); | |
49e22585 CL |
5240 | STAT_ATTR(CPU_PARTIAL_ALLOC, cpu_partial_alloc); |
5241 | STAT_ATTR(CPU_PARTIAL_FREE, cpu_partial_free); | |
8028dcea AS |
5242 | STAT_ATTR(CPU_PARTIAL_NODE, cpu_partial_node); |
5243 | STAT_ATTR(CPU_PARTIAL_DRAIN, cpu_partial_drain); | |
8ff12cfc CL |
5244 | #endif |
5245 | ||
06428780 | 5246 | static struct attribute *slab_attrs[] = { |
81819f0f CL |
5247 | &slab_size_attr.attr, |
5248 | &object_size_attr.attr, | |
5249 | &objs_per_slab_attr.attr, | |
5250 | &order_attr.attr, | |
73d342b1 | 5251 | &min_partial_attr.attr, |
49e22585 | 5252 | &cpu_partial_attr.attr, |
81819f0f | 5253 | &objects_attr.attr, |
205ab99d | 5254 | &objects_partial_attr.attr, |
81819f0f CL |
5255 | &partial_attr.attr, |
5256 | &cpu_slabs_attr.attr, | |
5257 | &ctor_attr.attr, | |
81819f0f CL |
5258 | &aliases_attr.attr, |
5259 | &align_attr.attr, | |
81819f0f CL |
5260 | &hwcache_align_attr.attr, |
5261 | &reclaim_account_attr.attr, | |
5262 | &destroy_by_rcu_attr.attr, | |
a5a84755 | 5263 | &shrink_attr.attr, |
ab9a0f19 | 5264 | &reserved_attr.attr, |
49e22585 | 5265 | &slabs_cpu_partial_attr.attr, |
ab4d5ed5 | 5266 | #ifdef CONFIG_SLUB_DEBUG |
a5a84755 CL |
5267 | &total_objects_attr.attr, |
5268 | &slabs_attr.attr, | |
5269 | &sanity_checks_attr.attr, | |
5270 | &trace_attr.attr, | |
81819f0f CL |
5271 | &red_zone_attr.attr, |
5272 | &poison_attr.attr, | |
5273 | &store_user_attr.attr, | |
53e15af0 | 5274 | &validate_attr.attr, |
88a420e4 CL |
5275 | &alloc_calls_attr.attr, |
5276 | &free_calls_attr.attr, | |
ab4d5ed5 | 5277 | #endif |
81819f0f CL |
5278 | #ifdef CONFIG_ZONE_DMA |
5279 | &cache_dma_attr.attr, | |
5280 | #endif | |
5281 | #ifdef CONFIG_NUMA | |
9824601e | 5282 | &remote_node_defrag_ratio_attr.attr, |
8ff12cfc CL |
5283 | #endif |
5284 | #ifdef CONFIG_SLUB_STATS | |
5285 | &alloc_fastpath_attr.attr, | |
5286 | &alloc_slowpath_attr.attr, | |
5287 | &free_fastpath_attr.attr, | |
5288 | &free_slowpath_attr.attr, | |
5289 | &free_frozen_attr.attr, | |
5290 | &free_add_partial_attr.attr, | |
5291 | &free_remove_partial_attr.attr, | |
5292 | &alloc_from_partial_attr.attr, | |
5293 | &alloc_slab_attr.attr, | |
5294 | &alloc_refill_attr.attr, | |
e36a2652 | 5295 | &alloc_node_mismatch_attr.attr, |
8ff12cfc CL |
5296 | &free_slab_attr.attr, |
5297 | &cpuslab_flush_attr.attr, | |
5298 | &deactivate_full_attr.attr, | |
5299 | &deactivate_empty_attr.attr, | |
5300 | &deactivate_to_head_attr.attr, | |
5301 | &deactivate_to_tail_attr.attr, | |
5302 | &deactivate_remote_frees_attr.attr, | |
03e404af | 5303 | &deactivate_bypass_attr.attr, |
65c3376a | 5304 | &order_fallback_attr.attr, |
b789ef51 CL |
5305 | &cmpxchg_double_fail_attr.attr, |
5306 | &cmpxchg_double_cpu_fail_attr.attr, | |
49e22585 CL |
5307 | &cpu_partial_alloc_attr.attr, |
5308 | &cpu_partial_free_attr.attr, | |
8028dcea AS |
5309 | &cpu_partial_node_attr.attr, |
5310 | &cpu_partial_drain_attr.attr, | |
81819f0f | 5311 | #endif |
4c13dd3b DM |
5312 | #ifdef CONFIG_FAILSLAB |
5313 | &failslab_attr.attr, | |
5314 | #endif | |
5315 | ||
81819f0f CL |
5316 | NULL |
5317 | }; | |
5318 | ||
5319 | static struct attribute_group slab_attr_group = { | |
5320 | .attrs = slab_attrs, | |
5321 | }; | |
5322 | ||
5323 | static ssize_t slab_attr_show(struct kobject *kobj, | |
5324 | struct attribute *attr, | |
5325 | char *buf) | |
5326 | { | |
5327 | struct slab_attribute *attribute; | |
5328 | struct kmem_cache *s; | |
5329 | int err; | |
5330 | ||
5331 | attribute = to_slab_attr(attr); | |
5332 | s = to_slab(kobj); | |
5333 | ||
5334 | if (!attribute->show) | |
5335 | return -EIO; | |
5336 | ||
5337 | err = attribute->show(s, buf); | |
5338 | ||
5339 | return err; | |
5340 | } | |
5341 | ||
5342 | static ssize_t slab_attr_store(struct kobject *kobj, | |
5343 | struct attribute *attr, | |
5344 | const char *buf, size_t len) | |
5345 | { | |
5346 | struct slab_attribute *attribute; | |
5347 | struct kmem_cache *s; | |
5348 | int err; | |
5349 | ||
5350 | attribute = to_slab_attr(attr); | |
5351 | s = to_slab(kobj); | |
5352 | ||
5353 | if (!attribute->store) | |
5354 | return -EIO; | |
5355 | ||
5356 | err = attribute->store(s, buf, len); | |
127424c8 | 5357 | #ifdef CONFIG_MEMCG |
107dab5c | 5358 | if (slab_state >= FULL && err >= 0 && is_root_cache(s)) { |
426589f5 | 5359 | struct kmem_cache *c; |
81819f0f | 5360 | |
107dab5c GC |
5361 | mutex_lock(&slab_mutex); |
5362 | if (s->max_attr_size < len) | |
5363 | s->max_attr_size = len; | |
5364 | ||
ebe945c2 GC |
5365 | /* |
5366 | * This is a best effort propagation, so this function's return | |
5367 | * value will be determined by the parent cache only. This is | |
5368 | * basically because not all attributes will have a well | |
5369 | * defined semantics for rollbacks - most of the actions will | |
5370 | * have permanent effects. | |
5371 | * | |
5372 | * Returning the error value of any of the children that fail | |
5373 | * is not 100 % defined, in the sense that users seeing the | |
5374 | * error code won't be able to know anything about the state of | |
5375 | * the cache. | |
5376 | * | |
5377 | * Only returning the error code for the parent cache at least | |
5378 | * has well defined semantics. The cache being written to | |
5379 | * directly either failed or succeeded, in which case we loop | |
5380 | * through the descendants with best-effort propagation. | |
5381 | */ | |
426589f5 VD |
5382 | for_each_memcg_cache(c, s) |
5383 | attribute->store(c, buf, len); | |
107dab5c GC |
5384 | mutex_unlock(&slab_mutex); |
5385 | } | |
5386 | #endif | |
81819f0f CL |
5387 | return err; |
5388 | } | |
5389 | ||
107dab5c GC |
5390 | static void memcg_propagate_slab_attrs(struct kmem_cache *s) |
5391 | { | |
127424c8 | 5392 | #ifdef CONFIG_MEMCG |
107dab5c GC |
5393 | int i; |
5394 | char *buffer = NULL; | |
93030d83 | 5395 | struct kmem_cache *root_cache; |
107dab5c | 5396 | |
93030d83 | 5397 | if (is_root_cache(s)) |
107dab5c GC |
5398 | return; |
5399 | ||
f7ce3190 | 5400 | root_cache = s->memcg_params.root_cache; |
93030d83 | 5401 | |
107dab5c GC |
5402 | /* |
5403 | * This mean this cache had no attribute written. Therefore, no point | |
5404 | * in copying default values around | |
5405 | */ | |
93030d83 | 5406 | if (!root_cache->max_attr_size) |
107dab5c GC |
5407 | return; |
5408 | ||
5409 | for (i = 0; i < ARRAY_SIZE(slab_attrs); i++) { | |
5410 | char mbuf[64]; | |
5411 | char *buf; | |
5412 | struct slab_attribute *attr = to_slab_attr(slab_attrs[i]); | |
5413 | ||
5414 | if (!attr || !attr->store || !attr->show) | |
5415 | continue; | |
5416 | ||
5417 | /* | |
5418 | * It is really bad that we have to allocate here, so we will | |
5419 | * do it only as a fallback. If we actually allocate, though, | |
5420 | * we can just use the allocated buffer until the end. | |
5421 | * | |
5422 | * Most of the slub attributes will tend to be very small in | |
5423 | * size, but sysfs allows buffers up to a page, so they can | |
5424 | * theoretically happen. | |
5425 | */ | |
5426 | if (buffer) | |
5427 | buf = buffer; | |
93030d83 | 5428 | else if (root_cache->max_attr_size < ARRAY_SIZE(mbuf)) |
107dab5c GC |
5429 | buf = mbuf; |
5430 | else { | |
5431 | buffer = (char *) get_zeroed_page(GFP_KERNEL); | |
5432 | if (WARN_ON(!buffer)) | |
5433 | continue; | |
5434 | buf = buffer; | |
5435 | } | |
5436 | ||
93030d83 | 5437 | attr->show(root_cache, buf); |
107dab5c GC |
5438 | attr->store(s, buf, strlen(buf)); |
5439 | } | |
5440 | ||
5441 | if (buffer) | |
5442 | free_page((unsigned long)buffer); | |
5443 | #endif | |
5444 | } | |
5445 | ||
41a21285 CL |
5446 | static void kmem_cache_release(struct kobject *k) |
5447 | { | |
5448 | slab_kmem_cache_release(to_slab(k)); | |
5449 | } | |
5450 | ||
52cf25d0 | 5451 | static const struct sysfs_ops slab_sysfs_ops = { |
81819f0f CL |
5452 | .show = slab_attr_show, |
5453 | .store = slab_attr_store, | |
5454 | }; | |
5455 | ||
5456 | static struct kobj_type slab_ktype = { | |
5457 | .sysfs_ops = &slab_sysfs_ops, | |
41a21285 | 5458 | .release = kmem_cache_release, |
81819f0f CL |
5459 | }; |
5460 | ||
5461 | static int uevent_filter(struct kset *kset, struct kobject *kobj) | |
5462 | { | |
5463 | struct kobj_type *ktype = get_ktype(kobj); | |
5464 | ||
5465 | if (ktype == &slab_ktype) | |
5466 | return 1; | |
5467 | return 0; | |
5468 | } | |
5469 | ||
9cd43611 | 5470 | static const struct kset_uevent_ops slab_uevent_ops = { |
81819f0f CL |
5471 | .filter = uevent_filter, |
5472 | }; | |
5473 | ||
27c3a314 | 5474 | static struct kset *slab_kset; |
81819f0f | 5475 | |
9a41707b VD |
5476 | static inline struct kset *cache_kset(struct kmem_cache *s) |
5477 | { | |
127424c8 | 5478 | #ifdef CONFIG_MEMCG |
9a41707b | 5479 | if (!is_root_cache(s)) |
f7ce3190 | 5480 | return s->memcg_params.root_cache->memcg_kset; |
9a41707b VD |
5481 | #endif |
5482 | return slab_kset; | |
5483 | } | |
5484 | ||
81819f0f CL |
5485 | #define ID_STR_LENGTH 64 |
5486 | ||
5487 | /* Create a unique string id for a slab cache: | |
6446faa2 CL |
5488 | * |
5489 | * Format :[flags-]size | |
81819f0f CL |
5490 | */ |
5491 | static char *create_unique_id(struct kmem_cache *s) | |
5492 | { | |
5493 | char *name = kmalloc(ID_STR_LENGTH, GFP_KERNEL); | |
5494 | char *p = name; | |
5495 | ||
5496 | BUG_ON(!name); | |
5497 | ||
5498 | *p++ = ':'; | |
5499 | /* | |
5500 | * First flags affecting slabcache operations. We will only | |
5501 | * get here for aliasable slabs so we do not need to support | |
5502 | * too many flags. The flags here must cover all flags that | |
5503 | * are matched during merging to guarantee that the id is | |
5504 | * unique. | |
5505 | */ | |
5506 | if (s->flags & SLAB_CACHE_DMA) | |
5507 | *p++ = 'd'; | |
5508 | if (s->flags & SLAB_RECLAIM_ACCOUNT) | |
5509 | *p++ = 'a'; | |
becfda68 | 5510 | if (s->flags & SLAB_CONSISTENCY_CHECKS) |
81819f0f | 5511 | *p++ = 'F'; |
5a896d9e VN |
5512 | if (!(s->flags & SLAB_NOTRACK)) |
5513 | *p++ = 't'; | |
230e9fc2 VD |
5514 | if (s->flags & SLAB_ACCOUNT) |
5515 | *p++ = 'A'; | |
81819f0f CL |
5516 | if (p != name + 1) |
5517 | *p++ = '-'; | |
5518 | p += sprintf(p, "%07d", s->size); | |
2633d7a0 | 5519 | |
81819f0f CL |
5520 | BUG_ON(p > name + ID_STR_LENGTH - 1); |
5521 | return name; | |
5522 | } | |
5523 | ||
5524 | static int sysfs_slab_add(struct kmem_cache *s) | |
5525 | { | |
5526 | int err; | |
5527 | const char *name; | |
45530c44 | 5528 | int unmergeable = slab_unmergeable(s); |
81819f0f | 5529 | |
81819f0f CL |
5530 | if (unmergeable) { |
5531 | /* | |
5532 | * Slabcache can never be merged so we can use the name proper. | |
5533 | * This is typically the case for debug situations. In that | |
5534 | * case we can catch duplicate names easily. | |
5535 | */ | |
27c3a314 | 5536 | sysfs_remove_link(&slab_kset->kobj, s->name); |
81819f0f CL |
5537 | name = s->name; |
5538 | } else { | |
5539 | /* | |
5540 | * Create a unique name for the slab as a target | |
5541 | * for the symlinks. | |
5542 | */ | |
5543 | name = create_unique_id(s); | |
5544 | } | |
5545 | ||
9a41707b | 5546 | s->kobj.kset = cache_kset(s); |
26e4f205 | 5547 | err = kobject_init_and_add(&s->kobj, &slab_ktype, NULL, "%s", name); |
54b6a731 | 5548 | if (err) |
80da026a | 5549 | goto out; |
81819f0f CL |
5550 | |
5551 | err = sysfs_create_group(&s->kobj, &slab_attr_group); | |
54b6a731 DJ |
5552 | if (err) |
5553 | goto out_del_kobj; | |
9a41707b | 5554 | |
127424c8 | 5555 | #ifdef CONFIG_MEMCG |
9a41707b VD |
5556 | if (is_root_cache(s)) { |
5557 | s->memcg_kset = kset_create_and_add("cgroup", NULL, &s->kobj); | |
5558 | if (!s->memcg_kset) { | |
54b6a731 DJ |
5559 | err = -ENOMEM; |
5560 | goto out_del_kobj; | |
9a41707b VD |
5561 | } |
5562 | } | |
5563 | #endif | |
5564 | ||
81819f0f CL |
5565 | kobject_uevent(&s->kobj, KOBJ_ADD); |
5566 | if (!unmergeable) { | |
5567 | /* Setup first alias */ | |
5568 | sysfs_slab_alias(s, s->name); | |
81819f0f | 5569 | } |
54b6a731 DJ |
5570 | out: |
5571 | if (!unmergeable) | |
5572 | kfree(name); | |
5573 | return err; | |
5574 | out_del_kobj: | |
5575 | kobject_del(&s->kobj); | |
54b6a731 | 5576 | goto out; |
81819f0f CL |
5577 | } |
5578 | ||
41a21285 | 5579 | void sysfs_slab_remove(struct kmem_cache *s) |
81819f0f | 5580 | { |
97d06609 | 5581 | if (slab_state < FULL) |
2bce6485 CL |
5582 | /* |
5583 | * Sysfs has not been setup yet so no need to remove the | |
5584 | * cache from sysfs. | |
5585 | */ | |
5586 | return; | |
5587 | ||
127424c8 | 5588 | #ifdef CONFIG_MEMCG |
9a41707b VD |
5589 | kset_unregister(s->memcg_kset); |
5590 | #endif | |
81819f0f CL |
5591 | kobject_uevent(&s->kobj, KOBJ_REMOVE); |
5592 | kobject_del(&s->kobj); | |
151c602f | 5593 | kobject_put(&s->kobj); |
81819f0f CL |
5594 | } |
5595 | ||
5596 | /* | |
5597 | * Need to buffer aliases during bootup until sysfs becomes | |
9f6c708e | 5598 | * available lest we lose that information. |
81819f0f CL |
5599 | */ |
5600 | struct saved_alias { | |
5601 | struct kmem_cache *s; | |
5602 | const char *name; | |
5603 | struct saved_alias *next; | |
5604 | }; | |
5605 | ||
5af328a5 | 5606 | static struct saved_alias *alias_list; |
81819f0f CL |
5607 | |
5608 | static int sysfs_slab_alias(struct kmem_cache *s, const char *name) | |
5609 | { | |
5610 | struct saved_alias *al; | |
5611 | ||
97d06609 | 5612 | if (slab_state == FULL) { |
81819f0f CL |
5613 | /* |
5614 | * If we have a leftover link then remove it. | |
5615 | */ | |
27c3a314 GKH |
5616 | sysfs_remove_link(&slab_kset->kobj, name); |
5617 | return sysfs_create_link(&slab_kset->kobj, &s->kobj, name); | |
81819f0f CL |
5618 | } |
5619 | ||
5620 | al = kmalloc(sizeof(struct saved_alias), GFP_KERNEL); | |
5621 | if (!al) | |
5622 | return -ENOMEM; | |
5623 | ||
5624 | al->s = s; | |
5625 | al->name = name; | |
5626 | al->next = alias_list; | |
5627 | alias_list = al; | |
5628 | return 0; | |
5629 | } | |
5630 | ||
5631 | static int __init slab_sysfs_init(void) | |
5632 | { | |
5b95a4ac | 5633 | struct kmem_cache *s; |
81819f0f CL |
5634 | int err; |
5635 | ||
18004c5d | 5636 | mutex_lock(&slab_mutex); |
2bce6485 | 5637 | |
0ff21e46 | 5638 | slab_kset = kset_create_and_add("slab", &slab_uevent_ops, kernel_kobj); |
27c3a314 | 5639 | if (!slab_kset) { |
18004c5d | 5640 | mutex_unlock(&slab_mutex); |
f9f58285 | 5641 | pr_err("Cannot register slab subsystem.\n"); |
81819f0f CL |
5642 | return -ENOSYS; |
5643 | } | |
5644 | ||
97d06609 | 5645 | slab_state = FULL; |
26a7bd03 | 5646 | |
5b95a4ac | 5647 | list_for_each_entry(s, &slab_caches, list) { |
26a7bd03 | 5648 | err = sysfs_slab_add(s); |
5d540fb7 | 5649 | if (err) |
f9f58285 FF |
5650 | pr_err("SLUB: Unable to add boot slab %s to sysfs\n", |
5651 | s->name); | |
26a7bd03 | 5652 | } |
81819f0f CL |
5653 | |
5654 | while (alias_list) { | |
5655 | struct saved_alias *al = alias_list; | |
5656 | ||
5657 | alias_list = alias_list->next; | |
5658 | err = sysfs_slab_alias(al->s, al->name); | |
5d540fb7 | 5659 | if (err) |
f9f58285 FF |
5660 | pr_err("SLUB: Unable to add boot slab alias %s to sysfs\n", |
5661 | al->name); | |
81819f0f CL |
5662 | kfree(al); |
5663 | } | |
5664 | ||
18004c5d | 5665 | mutex_unlock(&slab_mutex); |
81819f0f CL |
5666 | resiliency_test(); |
5667 | return 0; | |
5668 | } | |
5669 | ||
5670 | __initcall(slab_sysfs_init); | |
ab4d5ed5 | 5671 | #endif /* CONFIG_SYSFS */ |
57ed3eda PE |
5672 | |
5673 | /* | |
5674 | * The /proc/slabinfo ABI | |
5675 | */ | |
158a9624 | 5676 | #ifdef CONFIG_SLABINFO |
0d7561c6 | 5677 | void get_slabinfo(struct kmem_cache *s, struct slabinfo *sinfo) |
57ed3eda | 5678 | { |
57ed3eda | 5679 | unsigned long nr_slabs = 0; |
205ab99d CL |
5680 | unsigned long nr_objs = 0; |
5681 | unsigned long nr_free = 0; | |
57ed3eda | 5682 | int node; |
fa45dc25 | 5683 | struct kmem_cache_node *n; |
57ed3eda | 5684 | |
fa45dc25 | 5685 | for_each_kmem_cache_node(s, node, n) { |
c17fd13e WL |
5686 | nr_slabs += node_nr_slabs(n); |
5687 | nr_objs += node_nr_objs(n); | |
205ab99d | 5688 | nr_free += count_partial(n, count_free); |
57ed3eda PE |
5689 | } |
5690 | ||
0d7561c6 GC |
5691 | sinfo->active_objs = nr_objs - nr_free; |
5692 | sinfo->num_objs = nr_objs; | |
5693 | sinfo->active_slabs = nr_slabs; | |
5694 | sinfo->num_slabs = nr_slabs; | |
5695 | sinfo->objects_per_slab = oo_objects(s->oo); | |
5696 | sinfo->cache_order = oo_order(s->oo); | |
57ed3eda PE |
5697 | } |
5698 | ||
0d7561c6 | 5699 | void slabinfo_show_stats(struct seq_file *m, struct kmem_cache *s) |
7b3c3a50 | 5700 | { |
7b3c3a50 AD |
5701 | } |
5702 | ||
b7454ad3 GC |
5703 | ssize_t slabinfo_write(struct file *file, const char __user *buffer, |
5704 | size_t count, loff_t *ppos) | |
7b3c3a50 | 5705 | { |
b7454ad3 | 5706 | return -EIO; |
7b3c3a50 | 5707 | } |
158a9624 | 5708 | #endif /* CONFIG_SLABINFO */ |