]>
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 | |
345c905d JK |
127 | static inline bool kmem_cache_has_cpu_partial(struct kmem_cache *s) |
128 | { | |
129 | #ifdef CONFIG_SLUB_CPU_PARTIAL | |
130 | return !kmem_cache_debug(s); | |
131 | #else | |
132 | return false; | |
133 | #endif | |
134 | } | |
135 | ||
81819f0f CL |
136 | /* |
137 | * Issues still to be resolved: | |
138 | * | |
81819f0f CL |
139 | * - Support PAGE_ALLOC_DEBUG. Should be easy to do. |
140 | * | |
81819f0f CL |
141 | * - Variable sizing of the per node arrays |
142 | */ | |
143 | ||
144 | /* Enable to test recovery from slab corruption on boot */ | |
145 | #undef SLUB_RESILIENCY_TEST | |
146 | ||
b789ef51 CL |
147 | /* Enable to log cmpxchg failures */ |
148 | #undef SLUB_DEBUG_CMPXCHG | |
149 | ||
2086d26a CL |
150 | /* |
151 | * Mininum number of partial slabs. These will be left on the partial | |
152 | * lists even if they are empty. kmem_cache_shrink may reclaim them. | |
153 | */ | |
76be8950 | 154 | #define MIN_PARTIAL 5 |
e95eed57 | 155 | |
2086d26a CL |
156 | /* |
157 | * Maximum number of desirable partial slabs. | |
158 | * The existence of more partial slabs makes kmem_cache_shrink | |
721ae22a | 159 | * sort the partial list by the number of objects in use. |
2086d26a CL |
160 | */ |
161 | #define MAX_PARTIAL 10 | |
162 | ||
becfda68 | 163 | #define DEBUG_DEFAULT_FLAGS (SLAB_CONSISTENCY_CHECKS | SLAB_RED_ZONE | \ |
81819f0f | 164 | SLAB_POISON | SLAB_STORE_USER) |
672bba3a | 165 | |
149daaf3 LA |
166 | /* |
167 | * These debug flags cannot use CMPXCHG because there might be consistency | |
168 | * issues when checking or reading debug information | |
169 | */ | |
170 | #define SLAB_NO_CMPXCHG (SLAB_CONSISTENCY_CHECKS | SLAB_STORE_USER | \ | |
171 | SLAB_TRACE) | |
172 | ||
173 | ||
fa5ec8a1 | 174 | /* |
3de47213 DR |
175 | * Debugging flags that require metadata to be stored in the slab. These get |
176 | * disabled when slub_debug=O is used and a cache's min order increases with | |
177 | * metadata. | |
fa5ec8a1 | 178 | */ |
3de47213 | 179 | #define DEBUG_METADATA_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER) |
fa5ec8a1 | 180 | |
210b5c06 CG |
181 | #define OO_SHIFT 16 |
182 | #define OO_MASK ((1 << OO_SHIFT) - 1) | |
50d5c41c | 183 | #define MAX_OBJS_PER_PAGE 32767 /* since page.objects is u15 */ |
210b5c06 | 184 | |
81819f0f | 185 | /* Internal SLUB flags */ |
f90ec390 | 186 | #define __OBJECT_POISON 0x80000000UL /* Poison object */ |
b789ef51 | 187 | #define __CMPXCHG_DOUBLE 0x40000000UL /* Use cmpxchg_double */ |
81819f0f | 188 | |
81819f0f CL |
189 | #ifdef CONFIG_SMP |
190 | static struct notifier_block slab_notifier; | |
191 | #endif | |
192 | ||
02cbc874 CL |
193 | /* |
194 | * Tracking user of a slab. | |
195 | */ | |
d6543e39 | 196 | #define TRACK_ADDRS_COUNT 16 |
02cbc874 | 197 | struct track { |
ce71e27c | 198 | unsigned long addr; /* Called from address */ |
d6543e39 BG |
199 | #ifdef CONFIG_STACKTRACE |
200 | unsigned long addrs[TRACK_ADDRS_COUNT]; /* Called from address */ | |
201 | #endif | |
02cbc874 CL |
202 | int cpu; /* Was running on cpu */ |
203 | int pid; /* Pid context */ | |
204 | unsigned long when; /* When did the operation occur */ | |
205 | }; | |
206 | ||
207 | enum track_item { TRACK_ALLOC, TRACK_FREE }; | |
208 | ||
ab4d5ed5 | 209 | #ifdef CONFIG_SYSFS |
81819f0f CL |
210 | static int sysfs_slab_add(struct kmem_cache *); |
211 | static int sysfs_slab_alias(struct kmem_cache *, const char *); | |
107dab5c | 212 | static void memcg_propagate_slab_attrs(struct kmem_cache *s); |
81819f0f | 213 | #else |
0c710013 CL |
214 | static inline int sysfs_slab_add(struct kmem_cache *s) { return 0; } |
215 | static inline int sysfs_slab_alias(struct kmem_cache *s, const char *p) | |
216 | { return 0; } | |
107dab5c | 217 | static inline void memcg_propagate_slab_attrs(struct kmem_cache *s) { } |
81819f0f CL |
218 | #endif |
219 | ||
4fdccdfb | 220 | static inline void stat(const struct kmem_cache *s, enum stat_item si) |
8ff12cfc CL |
221 | { |
222 | #ifdef CONFIG_SLUB_STATS | |
88da03a6 CL |
223 | /* |
224 | * The rmw is racy on a preemptible kernel but this is acceptable, so | |
225 | * avoid this_cpu_add()'s irq-disable overhead. | |
226 | */ | |
227 | raw_cpu_inc(s->cpu_slab->stat[si]); | |
8ff12cfc CL |
228 | #endif |
229 | } | |
230 | ||
81819f0f CL |
231 | /******************************************************************** |
232 | * Core slab cache functions | |
233 | *******************************************************************/ | |
234 | ||
6446faa2 | 235 | /* Verify that a pointer has an address that is valid within a slab page */ |
02cbc874 CL |
236 | static inline int check_valid_pointer(struct kmem_cache *s, |
237 | struct page *page, const void *object) | |
238 | { | |
239 | void *base; | |
240 | ||
a973e9dd | 241 | if (!object) |
02cbc874 CL |
242 | return 1; |
243 | ||
a973e9dd | 244 | base = page_address(page); |
39b26464 | 245 | if (object < base || object >= base + page->objects * s->size || |
02cbc874 CL |
246 | (object - base) % s->size) { |
247 | return 0; | |
248 | } | |
249 | ||
250 | return 1; | |
251 | } | |
252 | ||
7656c72b CL |
253 | static inline void *get_freepointer(struct kmem_cache *s, void *object) |
254 | { | |
255 | return *(void **)(object + s->offset); | |
256 | } | |
257 | ||
0ad9500e ED |
258 | static void prefetch_freepointer(const struct kmem_cache *s, void *object) |
259 | { | |
260 | prefetch(object + s->offset); | |
261 | } | |
262 | ||
1393d9a1 CL |
263 | static inline void *get_freepointer_safe(struct kmem_cache *s, void *object) |
264 | { | |
265 | void *p; | |
266 | ||
267 | #ifdef CONFIG_DEBUG_PAGEALLOC | |
268 | probe_kernel_read(&p, (void **)(object + s->offset), sizeof(p)); | |
269 | #else | |
270 | p = get_freepointer(s, object); | |
271 | #endif | |
272 | return p; | |
273 | } | |
274 | ||
7656c72b CL |
275 | static inline void set_freepointer(struct kmem_cache *s, void *object, void *fp) |
276 | { | |
277 | *(void **)(object + s->offset) = fp; | |
278 | } | |
279 | ||
280 | /* Loop over all objects in a slab */ | |
224a88be CL |
281 | #define for_each_object(__p, __s, __addr, __objects) \ |
282 | for (__p = (__addr); __p < (__addr) + (__objects) * (__s)->size;\ | |
7656c72b CL |
283 | __p += (__s)->size) |
284 | ||
54266640 WY |
285 | #define for_each_object_idx(__p, __idx, __s, __addr, __objects) \ |
286 | for (__p = (__addr), __idx = 1; __idx <= __objects;\ | |
287 | __p += (__s)->size, __idx++) | |
288 | ||
7656c72b CL |
289 | /* Determine object index from a given position */ |
290 | static inline int slab_index(void *p, struct kmem_cache *s, void *addr) | |
291 | { | |
292 | return (p - addr) / s->size; | |
293 | } | |
294 | ||
ab9a0f19 LJ |
295 | static inline int order_objects(int order, unsigned long size, int reserved) |
296 | { | |
297 | return ((PAGE_SIZE << order) - reserved) / size; | |
298 | } | |
299 | ||
834f3d11 | 300 | static inline struct kmem_cache_order_objects oo_make(int order, |
ab9a0f19 | 301 | unsigned long size, int reserved) |
834f3d11 CL |
302 | { |
303 | struct kmem_cache_order_objects x = { | |
ab9a0f19 | 304 | (order << OO_SHIFT) + order_objects(order, size, reserved) |
834f3d11 CL |
305 | }; |
306 | ||
307 | return x; | |
308 | } | |
309 | ||
310 | static inline int oo_order(struct kmem_cache_order_objects x) | |
311 | { | |
210b5c06 | 312 | return x.x >> OO_SHIFT; |
834f3d11 CL |
313 | } |
314 | ||
315 | static inline int oo_objects(struct kmem_cache_order_objects x) | |
316 | { | |
210b5c06 | 317 | return x.x & OO_MASK; |
834f3d11 CL |
318 | } |
319 | ||
881db7fb CL |
320 | /* |
321 | * Per slab locking using the pagelock | |
322 | */ | |
323 | static __always_inline void slab_lock(struct page *page) | |
324 | { | |
48c935ad | 325 | VM_BUG_ON_PAGE(PageTail(page), page); |
881db7fb CL |
326 | bit_spin_lock(PG_locked, &page->flags); |
327 | } | |
328 | ||
329 | static __always_inline void slab_unlock(struct page *page) | |
330 | { | |
48c935ad | 331 | VM_BUG_ON_PAGE(PageTail(page), page); |
881db7fb CL |
332 | __bit_spin_unlock(PG_locked, &page->flags); |
333 | } | |
334 | ||
a0320865 DH |
335 | static inline void set_page_slub_counters(struct page *page, unsigned long counters_new) |
336 | { | |
337 | struct page tmp; | |
338 | tmp.counters = counters_new; | |
339 | /* | |
340 | * page->counters can cover frozen/inuse/objects as well | |
341 | * as page->_count. If we assign to ->counters directly | |
342 | * we run the risk of losing updates to page->_count, so | |
343 | * be careful and only assign to the fields we need. | |
344 | */ | |
345 | page->frozen = tmp.frozen; | |
346 | page->inuse = tmp.inuse; | |
347 | page->objects = tmp.objects; | |
348 | } | |
349 | ||
1d07171c CL |
350 | /* Interrupts must be disabled (for the fallback code to work right) */ |
351 | static inline bool __cmpxchg_double_slab(struct kmem_cache *s, struct page *page, | |
352 | void *freelist_old, unsigned long counters_old, | |
353 | void *freelist_new, unsigned long counters_new, | |
354 | const char *n) | |
355 | { | |
356 | VM_BUG_ON(!irqs_disabled()); | |
2565409f HC |
357 | #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \ |
358 | defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE) | |
1d07171c | 359 | if (s->flags & __CMPXCHG_DOUBLE) { |
cdcd6298 | 360 | if (cmpxchg_double(&page->freelist, &page->counters, |
0aa9a13d DC |
361 | freelist_old, counters_old, |
362 | freelist_new, counters_new)) | |
6f6528a1 | 363 | return true; |
1d07171c CL |
364 | } else |
365 | #endif | |
366 | { | |
367 | slab_lock(page); | |
d0e0ac97 CG |
368 | if (page->freelist == freelist_old && |
369 | page->counters == counters_old) { | |
1d07171c | 370 | page->freelist = freelist_new; |
a0320865 | 371 | set_page_slub_counters(page, counters_new); |
1d07171c | 372 | slab_unlock(page); |
6f6528a1 | 373 | return true; |
1d07171c CL |
374 | } |
375 | slab_unlock(page); | |
376 | } | |
377 | ||
378 | cpu_relax(); | |
379 | stat(s, CMPXCHG_DOUBLE_FAIL); | |
380 | ||
381 | #ifdef SLUB_DEBUG_CMPXCHG | |
f9f58285 | 382 | pr_info("%s %s: cmpxchg double redo ", n, s->name); |
1d07171c CL |
383 | #endif |
384 | ||
6f6528a1 | 385 | return false; |
1d07171c CL |
386 | } |
387 | ||
b789ef51 CL |
388 | static inline bool cmpxchg_double_slab(struct kmem_cache *s, struct page *page, |
389 | void *freelist_old, unsigned long counters_old, | |
390 | void *freelist_new, unsigned long counters_new, | |
391 | const char *n) | |
392 | { | |
2565409f HC |
393 | #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \ |
394 | defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE) | |
b789ef51 | 395 | if (s->flags & __CMPXCHG_DOUBLE) { |
cdcd6298 | 396 | if (cmpxchg_double(&page->freelist, &page->counters, |
0aa9a13d DC |
397 | freelist_old, counters_old, |
398 | freelist_new, counters_new)) | |
6f6528a1 | 399 | return true; |
b789ef51 CL |
400 | } else |
401 | #endif | |
402 | { | |
1d07171c CL |
403 | unsigned long flags; |
404 | ||
405 | local_irq_save(flags); | |
881db7fb | 406 | slab_lock(page); |
d0e0ac97 CG |
407 | if (page->freelist == freelist_old && |
408 | page->counters == counters_old) { | |
b789ef51 | 409 | page->freelist = freelist_new; |
a0320865 | 410 | set_page_slub_counters(page, counters_new); |
881db7fb | 411 | slab_unlock(page); |
1d07171c | 412 | local_irq_restore(flags); |
6f6528a1 | 413 | return true; |
b789ef51 | 414 | } |
881db7fb | 415 | slab_unlock(page); |
1d07171c | 416 | local_irq_restore(flags); |
b789ef51 CL |
417 | } |
418 | ||
419 | cpu_relax(); | |
420 | stat(s, CMPXCHG_DOUBLE_FAIL); | |
421 | ||
422 | #ifdef SLUB_DEBUG_CMPXCHG | |
f9f58285 | 423 | pr_info("%s %s: cmpxchg double redo ", n, s->name); |
b789ef51 CL |
424 | #endif |
425 | ||
6f6528a1 | 426 | return false; |
b789ef51 CL |
427 | } |
428 | ||
41ecc55b | 429 | #ifdef CONFIG_SLUB_DEBUG |
5f80b13a CL |
430 | /* |
431 | * Determine a map of object in use on a page. | |
432 | * | |
881db7fb | 433 | * Node listlock must be held to guarantee that the page does |
5f80b13a CL |
434 | * not vanish from under us. |
435 | */ | |
436 | static void get_map(struct kmem_cache *s, struct page *page, unsigned long *map) | |
437 | { | |
438 | void *p; | |
439 | void *addr = page_address(page); | |
440 | ||
441 | for (p = page->freelist; p; p = get_freepointer(s, p)) | |
442 | set_bit(slab_index(p, s, addr), map); | |
443 | } | |
444 | ||
41ecc55b CL |
445 | /* |
446 | * Debug settings: | |
447 | */ | |
89d3c87e | 448 | #if defined(CONFIG_SLUB_DEBUG_ON) |
f0630fff | 449 | static int slub_debug = DEBUG_DEFAULT_FLAGS; |
89d3c87e AR |
450 | #elif defined(CONFIG_KASAN) |
451 | static int slub_debug = SLAB_STORE_USER; | |
f0630fff | 452 | #else |
41ecc55b | 453 | static int slub_debug; |
f0630fff | 454 | #endif |
41ecc55b CL |
455 | |
456 | static char *slub_debug_slabs; | |
fa5ec8a1 | 457 | static int disable_higher_order_debug; |
41ecc55b | 458 | |
a79316c6 AR |
459 | /* |
460 | * slub is about to manipulate internal object metadata. This memory lies | |
461 | * outside the range of the allocated object, so accessing it would normally | |
462 | * be reported by kasan as a bounds error. metadata_access_enable() is used | |
463 | * to tell kasan that these accesses are OK. | |
464 | */ | |
465 | static inline void metadata_access_enable(void) | |
466 | { | |
467 | kasan_disable_current(); | |
468 | } | |
469 | ||
470 | static inline void metadata_access_disable(void) | |
471 | { | |
472 | kasan_enable_current(); | |
473 | } | |
474 | ||
81819f0f CL |
475 | /* |
476 | * Object debugging | |
477 | */ | |
478 | static void print_section(char *text, u8 *addr, unsigned int length) | |
479 | { | |
a79316c6 | 480 | metadata_access_enable(); |
ffc79d28 SAS |
481 | print_hex_dump(KERN_ERR, text, DUMP_PREFIX_ADDRESS, 16, 1, addr, |
482 | length, 1); | |
a79316c6 | 483 | metadata_access_disable(); |
81819f0f CL |
484 | } |
485 | ||
81819f0f CL |
486 | static struct track *get_track(struct kmem_cache *s, void *object, |
487 | enum track_item alloc) | |
488 | { | |
489 | struct track *p; | |
490 | ||
491 | if (s->offset) | |
492 | p = object + s->offset + sizeof(void *); | |
493 | else | |
494 | p = object + s->inuse; | |
495 | ||
496 | return p + alloc; | |
497 | } | |
498 | ||
499 | static void set_track(struct kmem_cache *s, void *object, | |
ce71e27c | 500 | enum track_item alloc, unsigned long addr) |
81819f0f | 501 | { |
1a00df4a | 502 | struct track *p = get_track(s, object, alloc); |
81819f0f | 503 | |
81819f0f | 504 | if (addr) { |
d6543e39 BG |
505 | #ifdef CONFIG_STACKTRACE |
506 | struct stack_trace trace; | |
507 | int i; | |
508 | ||
509 | trace.nr_entries = 0; | |
510 | trace.max_entries = TRACK_ADDRS_COUNT; | |
511 | trace.entries = p->addrs; | |
512 | trace.skip = 3; | |
a79316c6 | 513 | metadata_access_enable(); |
d6543e39 | 514 | save_stack_trace(&trace); |
a79316c6 | 515 | metadata_access_disable(); |
d6543e39 BG |
516 | |
517 | /* See rant in lockdep.c */ | |
518 | if (trace.nr_entries != 0 && | |
519 | trace.entries[trace.nr_entries - 1] == ULONG_MAX) | |
520 | trace.nr_entries--; | |
521 | ||
522 | for (i = trace.nr_entries; i < TRACK_ADDRS_COUNT; i++) | |
523 | p->addrs[i] = 0; | |
524 | #endif | |
81819f0f CL |
525 | p->addr = addr; |
526 | p->cpu = smp_processor_id(); | |
88e4ccf2 | 527 | p->pid = current->pid; |
81819f0f CL |
528 | p->when = jiffies; |
529 | } else | |
530 | memset(p, 0, sizeof(struct track)); | |
531 | } | |
532 | ||
81819f0f CL |
533 | static void init_tracking(struct kmem_cache *s, void *object) |
534 | { | |
24922684 CL |
535 | if (!(s->flags & SLAB_STORE_USER)) |
536 | return; | |
537 | ||
ce71e27c EGM |
538 | set_track(s, object, TRACK_FREE, 0UL); |
539 | set_track(s, object, TRACK_ALLOC, 0UL); | |
81819f0f CL |
540 | } |
541 | ||
542 | static void print_track(const char *s, struct track *t) | |
543 | { | |
544 | if (!t->addr) | |
545 | return; | |
546 | ||
f9f58285 FF |
547 | pr_err("INFO: %s in %pS age=%lu cpu=%u pid=%d\n", |
548 | s, (void *)t->addr, jiffies - t->when, t->cpu, t->pid); | |
d6543e39 BG |
549 | #ifdef CONFIG_STACKTRACE |
550 | { | |
551 | int i; | |
552 | for (i = 0; i < TRACK_ADDRS_COUNT; i++) | |
553 | if (t->addrs[i]) | |
f9f58285 | 554 | pr_err("\t%pS\n", (void *)t->addrs[i]); |
d6543e39 BG |
555 | else |
556 | break; | |
557 | } | |
558 | #endif | |
24922684 CL |
559 | } |
560 | ||
561 | static void print_tracking(struct kmem_cache *s, void *object) | |
562 | { | |
563 | if (!(s->flags & SLAB_STORE_USER)) | |
564 | return; | |
565 | ||
566 | print_track("Allocated", get_track(s, object, TRACK_ALLOC)); | |
567 | print_track("Freed", get_track(s, object, TRACK_FREE)); | |
568 | } | |
569 | ||
570 | static void print_page_info(struct page *page) | |
571 | { | |
f9f58285 | 572 | pr_err("INFO: Slab 0x%p objects=%u used=%u fp=0x%p flags=0x%04lx\n", |
d0e0ac97 | 573 | page, page->objects, page->inuse, page->freelist, page->flags); |
24922684 CL |
574 | |
575 | } | |
576 | ||
577 | static void slab_bug(struct kmem_cache *s, char *fmt, ...) | |
578 | { | |
ecc42fbe | 579 | struct va_format vaf; |
24922684 | 580 | va_list args; |
24922684 CL |
581 | |
582 | va_start(args, fmt); | |
ecc42fbe FF |
583 | vaf.fmt = fmt; |
584 | vaf.va = &args; | |
f9f58285 | 585 | pr_err("=============================================================================\n"); |
ecc42fbe | 586 | pr_err("BUG %s (%s): %pV\n", s->name, print_tainted(), &vaf); |
f9f58285 | 587 | pr_err("-----------------------------------------------------------------------------\n\n"); |
645df230 | 588 | |
373d4d09 | 589 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); |
ecc42fbe | 590 | va_end(args); |
81819f0f CL |
591 | } |
592 | ||
24922684 CL |
593 | static void slab_fix(struct kmem_cache *s, char *fmt, ...) |
594 | { | |
ecc42fbe | 595 | struct va_format vaf; |
24922684 | 596 | va_list args; |
24922684 CL |
597 | |
598 | va_start(args, fmt); | |
ecc42fbe FF |
599 | vaf.fmt = fmt; |
600 | vaf.va = &args; | |
601 | pr_err("FIX %s: %pV\n", s->name, &vaf); | |
24922684 | 602 | va_end(args); |
24922684 CL |
603 | } |
604 | ||
605 | static void print_trailer(struct kmem_cache *s, struct page *page, u8 *p) | |
81819f0f CL |
606 | { |
607 | unsigned int off; /* Offset of last byte */ | |
a973e9dd | 608 | u8 *addr = page_address(page); |
24922684 CL |
609 | |
610 | print_tracking(s, p); | |
611 | ||
612 | print_page_info(page); | |
613 | ||
f9f58285 FF |
614 | pr_err("INFO: Object 0x%p @offset=%tu fp=0x%p\n\n", |
615 | p, p - addr, get_freepointer(s, p)); | |
24922684 CL |
616 | |
617 | if (p > addr + 16) | |
ffc79d28 | 618 | print_section("Bytes b4 ", p - 16, 16); |
81819f0f | 619 | |
3b0efdfa | 620 | print_section("Object ", p, min_t(unsigned long, s->object_size, |
ffc79d28 | 621 | PAGE_SIZE)); |
81819f0f | 622 | if (s->flags & SLAB_RED_ZONE) |
3b0efdfa CL |
623 | print_section("Redzone ", p + s->object_size, |
624 | s->inuse - s->object_size); | |
81819f0f | 625 | |
81819f0f CL |
626 | if (s->offset) |
627 | off = s->offset + sizeof(void *); | |
628 | else | |
629 | off = s->inuse; | |
630 | ||
24922684 | 631 | if (s->flags & SLAB_STORE_USER) |
81819f0f | 632 | off += 2 * sizeof(struct track); |
81819f0f CL |
633 | |
634 | if (off != s->size) | |
635 | /* Beginning of the filler is the free pointer */ | |
ffc79d28 | 636 | print_section("Padding ", p + off, s->size - off); |
24922684 CL |
637 | |
638 | dump_stack(); | |
81819f0f CL |
639 | } |
640 | ||
75c66def | 641 | void object_err(struct kmem_cache *s, struct page *page, |
81819f0f CL |
642 | u8 *object, char *reason) |
643 | { | |
3dc50637 | 644 | slab_bug(s, "%s", reason); |
24922684 | 645 | print_trailer(s, page, object); |
81819f0f CL |
646 | } |
647 | ||
d0e0ac97 CG |
648 | static void slab_err(struct kmem_cache *s, struct page *page, |
649 | const char *fmt, ...) | |
81819f0f CL |
650 | { |
651 | va_list args; | |
652 | char buf[100]; | |
653 | ||
24922684 CL |
654 | va_start(args, fmt); |
655 | vsnprintf(buf, sizeof(buf), fmt, args); | |
81819f0f | 656 | va_end(args); |
3dc50637 | 657 | slab_bug(s, "%s", buf); |
24922684 | 658 | print_page_info(page); |
81819f0f CL |
659 | dump_stack(); |
660 | } | |
661 | ||
f7cb1933 | 662 | static void init_object(struct kmem_cache *s, void *object, u8 val) |
81819f0f CL |
663 | { |
664 | u8 *p = object; | |
665 | ||
666 | if (s->flags & __OBJECT_POISON) { | |
3b0efdfa CL |
667 | memset(p, POISON_FREE, s->object_size - 1); |
668 | p[s->object_size - 1] = POISON_END; | |
81819f0f CL |
669 | } |
670 | ||
671 | if (s->flags & SLAB_RED_ZONE) | |
3b0efdfa | 672 | memset(p + s->object_size, val, s->inuse - s->object_size); |
81819f0f CL |
673 | } |
674 | ||
24922684 CL |
675 | static void restore_bytes(struct kmem_cache *s, char *message, u8 data, |
676 | void *from, void *to) | |
677 | { | |
678 | slab_fix(s, "Restoring 0x%p-0x%p=0x%x\n", from, to - 1, data); | |
679 | memset(from, data, to - from); | |
680 | } | |
681 | ||
682 | static int check_bytes_and_report(struct kmem_cache *s, struct page *page, | |
683 | u8 *object, char *what, | |
06428780 | 684 | u8 *start, unsigned int value, unsigned int bytes) |
24922684 CL |
685 | { |
686 | u8 *fault; | |
687 | u8 *end; | |
688 | ||
a79316c6 | 689 | metadata_access_enable(); |
79824820 | 690 | fault = memchr_inv(start, value, bytes); |
a79316c6 | 691 | metadata_access_disable(); |
24922684 CL |
692 | if (!fault) |
693 | return 1; | |
694 | ||
695 | end = start + bytes; | |
696 | while (end > fault && end[-1] == value) | |
697 | end--; | |
698 | ||
699 | slab_bug(s, "%s overwritten", what); | |
f9f58285 | 700 | pr_err("INFO: 0x%p-0x%p. First byte 0x%x instead of 0x%x\n", |
24922684 CL |
701 | fault, end - 1, fault[0], value); |
702 | print_trailer(s, page, object); | |
703 | ||
704 | restore_bytes(s, what, value, fault, end); | |
705 | return 0; | |
81819f0f CL |
706 | } |
707 | ||
81819f0f CL |
708 | /* |
709 | * Object layout: | |
710 | * | |
711 | * object address | |
712 | * Bytes of the object to be managed. | |
713 | * If the freepointer may overlay the object then the free | |
714 | * pointer is the first word of the object. | |
672bba3a | 715 | * |
81819f0f CL |
716 | * Poisoning uses 0x6b (POISON_FREE) and the last byte is |
717 | * 0xa5 (POISON_END) | |
718 | * | |
3b0efdfa | 719 | * object + s->object_size |
81819f0f | 720 | * Padding to reach word boundary. This is also used for Redzoning. |
672bba3a | 721 | * Padding is extended by another word if Redzoning is enabled and |
3b0efdfa | 722 | * object_size == inuse. |
672bba3a | 723 | * |
81819f0f CL |
724 | * We fill with 0xbb (RED_INACTIVE) for inactive objects and with |
725 | * 0xcc (RED_ACTIVE) for objects in use. | |
726 | * | |
727 | * object + s->inuse | |
672bba3a CL |
728 | * Meta data starts here. |
729 | * | |
81819f0f CL |
730 | * A. Free pointer (if we cannot overwrite object on free) |
731 | * B. Tracking data for SLAB_STORE_USER | |
672bba3a | 732 | * C. Padding to reach required alignment boundary or at mininum |
6446faa2 | 733 | * one word if debugging is on to be able to detect writes |
672bba3a CL |
734 | * before the word boundary. |
735 | * | |
736 | * Padding is done using 0x5a (POISON_INUSE) | |
81819f0f CL |
737 | * |
738 | * object + s->size | |
672bba3a | 739 | * Nothing is used beyond s->size. |
81819f0f | 740 | * |
3b0efdfa | 741 | * If slabcaches are merged then the object_size and inuse boundaries are mostly |
672bba3a | 742 | * ignored. And therefore no slab options that rely on these boundaries |
81819f0f CL |
743 | * may be used with merged slabcaches. |
744 | */ | |
745 | ||
81819f0f CL |
746 | static int check_pad_bytes(struct kmem_cache *s, struct page *page, u8 *p) |
747 | { | |
748 | unsigned long off = s->inuse; /* The end of info */ | |
749 | ||
750 | if (s->offset) | |
751 | /* Freepointer is placed after the object. */ | |
752 | off += sizeof(void *); | |
753 | ||
754 | if (s->flags & SLAB_STORE_USER) | |
755 | /* We also have user information there */ | |
756 | off += 2 * sizeof(struct track); | |
757 | ||
758 | if (s->size == off) | |
759 | return 1; | |
760 | ||
24922684 CL |
761 | return check_bytes_and_report(s, page, p, "Object padding", |
762 | p + off, POISON_INUSE, s->size - off); | |
81819f0f CL |
763 | } |
764 | ||
39b26464 | 765 | /* Check the pad bytes at the end of a slab page */ |
81819f0f CL |
766 | static int slab_pad_check(struct kmem_cache *s, struct page *page) |
767 | { | |
24922684 CL |
768 | u8 *start; |
769 | u8 *fault; | |
770 | u8 *end; | |
771 | int length; | |
772 | int remainder; | |
81819f0f CL |
773 | |
774 | if (!(s->flags & SLAB_POISON)) | |
775 | return 1; | |
776 | ||
a973e9dd | 777 | start = page_address(page); |
ab9a0f19 | 778 | length = (PAGE_SIZE << compound_order(page)) - s->reserved; |
39b26464 CL |
779 | end = start + length; |
780 | remainder = length % s->size; | |
81819f0f CL |
781 | if (!remainder) |
782 | return 1; | |
783 | ||
a79316c6 | 784 | metadata_access_enable(); |
79824820 | 785 | fault = memchr_inv(end - remainder, POISON_INUSE, remainder); |
a79316c6 | 786 | metadata_access_disable(); |
24922684 CL |
787 | if (!fault) |
788 | return 1; | |
789 | while (end > fault && end[-1] == POISON_INUSE) | |
790 | end--; | |
791 | ||
792 | slab_err(s, page, "Padding overwritten. 0x%p-0x%p", fault, end - 1); | |
ffc79d28 | 793 | print_section("Padding ", end - remainder, remainder); |
24922684 | 794 | |
8a3d271d | 795 | restore_bytes(s, "slab padding", POISON_INUSE, end - remainder, end); |
24922684 | 796 | return 0; |
81819f0f CL |
797 | } |
798 | ||
799 | static int check_object(struct kmem_cache *s, struct page *page, | |
f7cb1933 | 800 | void *object, u8 val) |
81819f0f CL |
801 | { |
802 | u8 *p = object; | |
3b0efdfa | 803 | u8 *endobject = object + s->object_size; |
81819f0f CL |
804 | |
805 | if (s->flags & SLAB_RED_ZONE) { | |
24922684 | 806 | if (!check_bytes_and_report(s, page, object, "Redzone", |
3b0efdfa | 807 | endobject, val, s->inuse - s->object_size)) |
81819f0f | 808 | return 0; |
81819f0f | 809 | } else { |
3b0efdfa | 810 | if ((s->flags & SLAB_POISON) && s->object_size < s->inuse) { |
3adbefee | 811 | check_bytes_and_report(s, page, p, "Alignment padding", |
d0e0ac97 CG |
812 | endobject, POISON_INUSE, |
813 | s->inuse - s->object_size); | |
3adbefee | 814 | } |
81819f0f CL |
815 | } |
816 | ||
817 | if (s->flags & SLAB_POISON) { | |
f7cb1933 | 818 | if (val != SLUB_RED_ACTIVE && (s->flags & __OBJECT_POISON) && |
24922684 | 819 | (!check_bytes_and_report(s, page, p, "Poison", p, |
3b0efdfa | 820 | POISON_FREE, s->object_size - 1) || |
24922684 | 821 | !check_bytes_and_report(s, page, p, "Poison", |
3b0efdfa | 822 | p + s->object_size - 1, POISON_END, 1))) |
81819f0f | 823 | return 0; |
81819f0f CL |
824 | /* |
825 | * check_pad_bytes cleans up on its own. | |
826 | */ | |
827 | check_pad_bytes(s, page, p); | |
828 | } | |
829 | ||
f7cb1933 | 830 | if (!s->offset && val == SLUB_RED_ACTIVE) |
81819f0f CL |
831 | /* |
832 | * Object and freepointer overlap. Cannot check | |
833 | * freepointer while object is allocated. | |
834 | */ | |
835 | return 1; | |
836 | ||
837 | /* Check free pointer validity */ | |
838 | if (!check_valid_pointer(s, page, get_freepointer(s, p))) { | |
839 | object_err(s, page, p, "Freepointer corrupt"); | |
840 | /* | |
9f6c708e | 841 | * No choice but to zap it and thus lose the remainder |
81819f0f | 842 | * of the free objects in this slab. May cause |
672bba3a | 843 | * another error because the object count is now wrong. |
81819f0f | 844 | */ |
a973e9dd | 845 | set_freepointer(s, p, NULL); |
81819f0f CL |
846 | return 0; |
847 | } | |
848 | return 1; | |
849 | } | |
850 | ||
851 | static int check_slab(struct kmem_cache *s, struct page *page) | |
852 | { | |
39b26464 CL |
853 | int maxobj; |
854 | ||
81819f0f CL |
855 | VM_BUG_ON(!irqs_disabled()); |
856 | ||
857 | if (!PageSlab(page)) { | |
24922684 | 858 | slab_err(s, page, "Not a valid slab page"); |
81819f0f CL |
859 | return 0; |
860 | } | |
39b26464 | 861 | |
ab9a0f19 | 862 | maxobj = order_objects(compound_order(page), s->size, s->reserved); |
39b26464 CL |
863 | if (page->objects > maxobj) { |
864 | slab_err(s, page, "objects %u > max %u", | |
f6edde9c | 865 | page->objects, maxobj); |
39b26464 CL |
866 | return 0; |
867 | } | |
868 | if (page->inuse > page->objects) { | |
24922684 | 869 | slab_err(s, page, "inuse %u > max %u", |
f6edde9c | 870 | page->inuse, page->objects); |
81819f0f CL |
871 | return 0; |
872 | } | |
873 | /* Slab_pad_check fixes things up after itself */ | |
874 | slab_pad_check(s, page); | |
875 | return 1; | |
876 | } | |
877 | ||
878 | /* | |
672bba3a CL |
879 | * Determine if a certain object on a page is on the freelist. Must hold the |
880 | * slab lock to guarantee that the chains are in a consistent state. | |
81819f0f CL |
881 | */ |
882 | static int on_freelist(struct kmem_cache *s, struct page *page, void *search) | |
883 | { | |
884 | int nr = 0; | |
881db7fb | 885 | void *fp; |
81819f0f | 886 | void *object = NULL; |
f6edde9c | 887 | int max_objects; |
81819f0f | 888 | |
881db7fb | 889 | fp = page->freelist; |
39b26464 | 890 | while (fp && nr <= page->objects) { |
81819f0f CL |
891 | if (fp == search) |
892 | return 1; | |
893 | if (!check_valid_pointer(s, page, fp)) { | |
894 | if (object) { | |
895 | object_err(s, page, object, | |
896 | "Freechain corrupt"); | |
a973e9dd | 897 | set_freepointer(s, object, NULL); |
81819f0f | 898 | } else { |
24922684 | 899 | slab_err(s, page, "Freepointer corrupt"); |
a973e9dd | 900 | page->freelist = NULL; |
39b26464 | 901 | page->inuse = page->objects; |
24922684 | 902 | slab_fix(s, "Freelist cleared"); |
81819f0f CL |
903 | return 0; |
904 | } | |
905 | break; | |
906 | } | |
907 | object = fp; | |
908 | fp = get_freepointer(s, object); | |
909 | nr++; | |
910 | } | |
911 | ||
ab9a0f19 | 912 | max_objects = order_objects(compound_order(page), s->size, s->reserved); |
210b5c06 CG |
913 | if (max_objects > MAX_OBJS_PER_PAGE) |
914 | max_objects = MAX_OBJS_PER_PAGE; | |
224a88be CL |
915 | |
916 | if (page->objects != max_objects) { | |
917 | slab_err(s, page, "Wrong number of objects. Found %d but " | |
918 | "should be %d", page->objects, max_objects); | |
919 | page->objects = max_objects; | |
920 | slab_fix(s, "Number of objects adjusted."); | |
921 | } | |
39b26464 | 922 | if (page->inuse != page->objects - nr) { |
70d71228 | 923 | slab_err(s, page, "Wrong object count. Counter is %d but " |
39b26464 CL |
924 | "counted were %d", page->inuse, page->objects - nr); |
925 | page->inuse = page->objects - nr; | |
24922684 | 926 | slab_fix(s, "Object count adjusted."); |
81819f0f CL |
927 | } |
928 | return search == NULL; | |
929 | } | |
930 | ||
0121c619 CL |
931 | static void trace(struct kmem_cache *s, struct page *page, void *object, |
932 | int alloc) | |
3ec09742 CL |
933 | { |
934 | if (s->flags & SLAB_TRACE) { | |
f9f58285 | 935 | pr_info("TRACE %s %s 0x%p inuse=%d fp=0x%p\n", |
3ec09742 CL |
936 | s->name, |
937 | alloc ? "alloc" : "free", | |
938 | object, page->inuse, | |
939 | page->freelist); | |
940 | ||
941 | if (!alloc) | |
d0e0ac97 CG |
942 | print_section("Object ", (void *)object, |
943 | s->object_size); | |
3ec09742 CL |
944 | |
945 | dump_stack(); | |
946 | } | |
947 | } | |
948 | ||
643b1138 | 949 | /* |
672bba3a | 950 | * Tracking of fully allocated slabs for debugging purposes. |
643b1138 | 951 | */ |
5cc6eee8 CL |
952 | static void add_full(struct kmem_cache *s, |
953 | struct kmem_cache_node *n, struct page *page) | |
643b1138 | 954 | { |
5cc6eee8 CL |
955 | if (!(s->flags & SLAB_STORE_USER)) |
956 | return; | |
957 | ||
255d0884 | 958 | lockdep_assert_held(&n->list_lock); |
643b1138 | 959 | list_add(&page->lru, &n->full); |
643b1138 CL |
960 | } |
961 | ||
c65c1877 | 962 | static void remove_full(struct kmem_cache *s, struct kmem_cache_node *n, struct page *page) |
643b1138 | 963 | { |
643b1138 CL |
964 | if (!(s->flags & SLAB_STORE_USER)) |
965 | return; | |
966 | ||
255d0884 | 967 | lockdep_assert_held(&n->list_lock); |
643b1138 | 968 | list_del(&page->lru); |
643b1138 CL |
969 | } |
970 | ||
0f389ec6 CL |
971 | /* Tracking of the number of slabs for debugging purposes */ |
972 | static inline unsigned long slabs_node(struct kmem_cache *s, int node) | |
973 | { | |
974 | struct kmem_cache_node *n = get_node(s, node); | |
975 | ||
976 | return atomic_long_read(&n->nr_slabs); | |
977 | } | |
978 | ||
26c02cf0 AB |
979 | static inline unsigned long node_nr_slabs(struct kmem_cache_node *n) |
980 | { | |
981 | return atomic_long_read(&n->nr_slabs); | |
982 | } | |
983 | ||
205ab99d | 984 | static inline void inc_slabs_node(struct kmem_cache *s, int node, int objects) |
0f389ec6 CL |
985 | { |
986 | struct kmem_cache_node *n = get_node(s, node); | |
987 | ||
988 | /* | |
989 | * May be called early in order to allocate a slab for the | |
990 | * kmem_cache_node structure. Solve the chicken-egg | |
991 | * dilemma by deferring the increment of the count during | |
992 | * bootstrap (see early_kmem_cache_node_alloc). | |
993 | */ | |
338b2642 | 994 | if (likely(n)) { |
0f389ec6 | 995 | atomic_long_inc(&n->nr_slabs); |
205ab99d CL |
996 | atomic_long_add(objects, &n->total_objects); |
997 | } | |
0f389ec6 | 998 | } |
205ab99d | 999 | static inline void dec_slabs_node(struct kmem_cache *s, int node, int objects) |
0f389ec6 CL |
1000 | { |
1001 | struct kmem_cache_node *n = get_node(s, node); | |
1002 | ||
1003 | atomic_long_dec(&n->nr_slabs); | |
205ab99d | 1004 | atomic_long_sub(objects, &n->total_objects); |
0f389ec6 CL |
1005 | } |
1006 | ||
1007 | /* Object debug checks for alloc/free paths */ | |
3ec09742 CL |
1008 | static void setup_object_debug(struct kmem_cache *s, struct page *page, |
1009 | void *object) | |
1010 | { | |
1011 | if (!(s->flags & (SLAB_STORE_USER|SLAB_RED_ZONE|__OBJECT_POISON))) | |
1012 | return; | |
1013 | ||
f7cb1933 | 1014 | init_object(s, object, SLUB_RED_INACTIVE); |
3ec09742 CL |
1015 | init_tracking(s, object); |
1016 | } | |
1017 | ||
becfda68 | 1018 | static inline int alloc_consistency_checks(struct kmem_cache *s, |
d0e0ac97 | 1019 | struct page *page, |
ce71e27c | 1020 | void *object, unsigned long addr) |
81819f0f CL |
1021 | { |
1022 | if (!check_slab(s, page)) | |
becfda68 | 1023 | return 0; |
81819f0f | 1024 | |
81819f0f CL |
1025 | if (!check_valid_pointer(s, page, object)) { |
1026 | object_err(s, page, object, "Freelist Pointer check fails"); | |
becfda68 | 1027 | return 0; |
81819f0f CL |
1028 | } |
1029 | ||
f7cb1933 | 1030 | if (!check_object(s, page, object, SLUB_RED_INACTIVE)) |
becfda68 LA |
1031 | return 0; |
1032 | ||
1033 | return 1; | |
1034 | } | |
1035 | ||
1036 | static noinline int alloc_debug_processing(struct kmem_cache *s, | |
1037 | struct page *page, | |
1038 | void *object, unsigned long addr) | |
1039 | { | |
1040 | if (s->flags & SLAB_CONSISTENCY_CHECKS) { | |
1041 | if (!alloc_consistency_checks(s, page, object, addr)) | |
1042 | goto bad; | |
1043 | } | |
81819f0f | 1044 | |
3ec09742 CL |
1045 | /* Success perform special debug activities for allocs */ |
1046 | if (s->flags & SLAB_STORE_USER) | |
1047 | set_track(s, object, TRACK_ALLOC, addr); | |
1048 | trace(s, page, object, 1); | |
f7cb1933 | 1049 | init_object(s, object, SLUB_RED_ACTIVE); |
81819f0f | 1050 | return 1; |
3ec09742 | 1051 | |
81819f0f CL |
1052 | bad: |
1053 | if (PageSlab(page)) { | |
1054 | /* | |
1055 | * If this is a slab page then lets do the best we can | |
1056 | * to avoid issues in the future. Marking all objects | |
672bba3a | 1057 | * as used avoids touching the remaining objects. |
81819f0f | 1058 | */ |
24922684 | 1059 | slab_fix(s, "Marking all objects used"); |
39b26464 | 1060 | page->inuse = page->objects; |
a973e9dd | 1061 | page->freelist = NULL; |
81819f0f CL |
1062 | } |
1063 | return 0; | |
1064 | } | |
1065 | ||
becfda68 LA |
1066 | static inline int free_consistency_checks(struct kmem_cache *s, |
1067 | struct page *page, void *object, unsigned long addr) | |
81819f0f | 1068 | { |
81819f0f | 1069 | if (!check_valid_pointer(s, page, object)) { |
70d71228 | 1070 | slab_err(s, page, "Invalid object pointer 0x%p", object); |
becfda68 | 1071 | return 0; |
81819f0f CL |
1072 | } |
1073 | ||
1074 | if (on_freelist(s, page, object)) { | |
24922684 | 1075 | object_err(s, page, object, "Object already free"); |
becfda68 | 1076 | return 0; |
81819f0f CL |
1077 | } |
1078 | ||
f7cb1933 | 1079 | if (!check_object(s, page, object, SLUB_RED_ACTIVE)) |
becfda68 | 1080 | return 0; |
81819f0f | 1081 | |
1b4f59e3 | 1082 | if (unlikely(s != page->slab_cache)) { |
3adbefee | 1083 | if (!PageSlab(page)) { |
70d71228 CL |
1084 | slab_err(s, page, "Attempt to free object(0x%p) " |
1085 | "outside of slab", object); | |
1b4f59e3 | 1086 | } else if (!page->slab_cache) { |
f9f58285 FF |
1087 | pr_err("SLUB <none>: no slab for object 0x%p.\n", |
1088 | object); | |
70d71228 | 1089 | dump_stack(); |
06428780 | 1090 | } else |
24922684 CL |
1091 | object_err(s, page, object, |
1092 | "page slab pointer corrupt."); | |
becfda68 LA |
1093 | return 0; |
1094 | } | |
1095 | return 1; | |
1096 | } | |
1097 | ||
1098 | /* Supports checking bulk free of a constructed freelist */ | |
1099 | static noinline int free_debug_processing( | |
1100 | struct kmem_cache *s, struct page *page, | |
1101 | void *head, void *tail, int bulk_cnt, | |
1102 | unsigned long addr) | |
1103 | { | |
1104 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); | |
1105 | void *object = head; | |
1106 | int cnt = 0; | |
1107 | unsigned long uninitialized_var(flags); | |
1108 | int ret = 0; | |
1109 | ||
1110 | spin_lock_irqsave(&n->list_lock, flags); | |
1111 | slab_lock(page); | |
1112 | ||
1113 | if (s->flags & SLAB_CONSISTENCY_CHECKS) { | |
1114 | if (!check_slab(s, page)) | |
1115 | goto out; | |
1116 | } | |
1117 | ||
1118 | next_object: | |
1119 | cnt++; | |
1120 | ||
1121 | if (s->flags & SLAB_CONSISTENCY_CHECKS) { | |
1122 | if (!free_consistency_checks(s, page, object, addr)) | |
1123 | goto out; | |
81819f0f | 1124 | } |
3ec09742 | 1125 | |
3ec09742 CL |
1126 | if (s->flags & SLAB_STORE_USER) |
1127 | set_track(s, object, TRACK_FREE, addr); | |
1128 | trace(s, page, object, 0); | |
81084651 | 1129 | /* Freepointer not overwritten by init_object(), SLAB_POISON moved it */ |
f7cb1933 | 1130 | init_object(s, object, SLUB_RED_INACTIVE); |
81084651 JDB |
1131 | |
1132 | /* Reached end of constructed freelist yet? */ | |
1133 | if (object != tail) { | |
1134 | object = get_freepointer(s, object); | |
1135 | goto next_object; | |
1136 | } | |
804aa132 LA |
1137 | ret = 1; |
1138 | ||
5c2e4bbb | 1139 | out: |
81084651 JDB |
1140 | if (cnt != bulk_cnt) |
1141 | slab_err(s, page, "Bulk freelist count(%d) invalid(%d)\n", | |
1142 | bulk_cnt, cnt); | |
1143 | ||
881db7fb | 1144 | slab_unlock(page); |
282acb43 | 1145 | spin_unlock_irqrestore(&n->list_lock, flags); |
804aa132 LA |
1146 | if (!ret) |
1147 | slab_fix(s, "Object at 0x%p not freed", object); | |
1148 | return ret; | |
81819f0f CL |
1149 | } |
1150 | ||
41ecc55b CL |
1151 | static int __init setup_slub_debug(char *str) |
1152 | { | |
f0630fff CL |
1153 | slub_debug = DEBUG_DEFAULT_FLAGS; |
1154 | if (*str++ != '=' || !*str) | |
1155 | /* | |
1156 | * No options specified. Switch on full debugging. | |
1157 | */ | |
1158 | goto out; | |
1159 | ||
1160 | if (*str == ',') | |
1161 | /* | |
1162 | * No options but restriction on slabs. This means full | |
1163 | * debugging for slabs matching a pattern. | |
1164 | */ | |
1165 | goto check_slabs; | |
1166 | ||
1167 | slub_debug = 0; | |
1168 | if (*str == '-') | |
1169 | /* | |
1170 | * Switch off all debugging measures. | |
1171 | */ | |
1172 | goto out; | |
1173 | ||
1174 | /* | |
1175 | * Determine which debug features should be switched on | |
1176 | */ | |
06428780 | 1177 | for (; *str && *str != ','; str++) { |
f0630fff CL |
1178 | switch (tolower(*str)) { |
1179 | case 'f': | |
becfda68 | 1180 | slub_debug |= SLAB_CONSISTENCY_CHECKS; |
f0630fff CL |
1181 | break; |
1182 | case 'z': | |
1183 | slub_debug |= SLAB_RED_ZONE; | |
1184 | break; | |
1185 | case 'p': | |
1186 | slub_debug |= SLAB_POISON; | |
1187 | break; | |
1188 | case 'u': | |
1189 | slub_debug |= SLAB_STORE_USER; | |
1190 | break; | |
1191 | case 't': | |
1192 | slub_debug |= SLAB_TRACE; | |
1193 | break; | |
4c13dd3b DM |
1194 | case 'a': |
1195 | slub_debug |= SLAB_FAILSLAB; | |
1196 | break; | |
08303a73 CA |
1197 | case 'o': |
1198 | /* | |
1199 | * Avoid enabling debugging on caches if its minimum | |
1200 | * order would increase as a result. | |
1201 | */ | |
1202 | disable_higher_order_debug = 1; | |
1203 | break; | |
f0630fff | 1204 | default: |
f9f58285 FF |
1205 | pr_err("slub_debug option '%c' unknown. skipped\n", |
1206 | *str); | |
f0630fff | 1207 | } |
41ecc55b CL |
1208 | } |
1209 | ||
f0630fff | 1210 | check_slabs: |
41ecc55b CL |
1211 | if (*str == ',') |
1212 | slub_debug_slabs = str + 1; | |
f0630fff | 1213 | out: |
41ecc55b CL |
1214 | return 1; |
1215 | } | |
1216 | ||
1217 | __setup("slub_debug", setup_slub_debug); | |
1218 | ||
423c929c | 1219 | unsigned long kmem_cache_flags(unsigned long object_size, |
ba0268a8 | 1220 | unsigned long flags, const char *name, |
51cc5068 | 1221 | void (*ctor)(void *)) |
41ecc55b CL |
1222 | { |
1223 | /* | |
e153362a | 1224 | * Enable debugging if selected on the kernel commandline. |
41ecc55b | 1225 | */ |
c6f58d9b CL |
1226 | if (slub_debug && (!slub_debug_slabs || (name && |
1227 | !strncmp(slub_debug_slabs, name, strlen(slub_debug_slabs))))) | |
3de47213 | 1228 | flags |= slub_debug; |
ba0268a8 CL |
1229 | |
1230 | return flags; | |
41ecc55b | 1231 | } |
b4a64718 | 1232 | #else /* !CONFIG_SLUB_DEBUG */ |
3ec09742 CL |
1233 | static inline void setup_object_debug(struct kmem_cache *s, |
1234 | struct page *page, void *object) {} | |
41ecc55b | 1235 | |
3ec09742 | 1236 | static inline int alloc_debug_processing(struct kmem_cache *s, |
ce71e27c | 1237 | struct page *page, void *object, unsigned long addr) { return 0; } |
41ecc55b | 1238 | |
282acb43 | 1239 | static inline int free_debug_processing( |
81084651 JDB |
1240 | struct kmem_cache *s, struct page *page, |
1241 | void *head, void *tail, int bulk_cnt, | |
282acb43 | 1242 | unsigned long addr) { return 0; } |
41ecc55b | 1243 | |
41ecc55b CL |
1244 | static inline int slab_pad_check(struct kmem_cache *s, struct page *page) |
1245 | { return 1; } | |
1246 | static inline int check_object(struct kmem_cache *s, struct page *page, | |
f7cb1933 | 1247 | void *object, u8 val) { return 1; } |
5cc6eee8 CL |
1248 | static inline void add_full(struct kmem_cache *s, struct kmem_cache_node *n, |
1249 | struct page *page) {} | |
c65c1877 PZ |
1250 | static inline void remove_full(struct kmem_cache *s, struct kmem_cache_node *n, |
1251 | struct page *page) {} | |
423c929c | 1252 | unsigned long kmem_cache_flags(unsigned long object_size, |
ba0268a8 | 1253 | unsigned long flags, const char *name, |
51cc5068 | 1254 | void (*ctor)(void *)) |
ba0268a8 CL |
1255 | { |
1256 | return flags; | |
1257 | } | |
41ecc55b | 1258 | #define slub_debug 0 |
0f389ec6 | 1259 | |
fdaa45e9 IM |
1260 | #define disable_higher_order_debug 0 |
1261 | ||
0f389ec6 CL |
1262 | static inline unsigned long slabs_node(struct kmem_cache *s, int node) |
1263 | { return 0; } | |
26c02cf0 AB |
1264 | static inline unsigned long node_nr_slabs(struct kmem_cache_node *n) |
1265 | { return 0; } | |
205ab99d CL |
1266 | static inline void inc_slabs_node(struct kmem_cache *s, int node, |
1267 | int objects) {} | |
1268 | static inline void dec_slabs_node(struct kmem_cache *s, int node, | |
1269 | int objects) {} | |
7d550c56 | 1270 | |
02e72cc6 AR |
1271 | #endif /* CONFIG_SLUB_DEBUG */ |
1272 | ||
1273 | /* | |
1274 | * Hooks for other subsystems that check memory allocations. In a typical | |
1275 | * production configuration these hooks all should produce no code at all. | |
1276 | */ | |
d56791b3 RB |
1277 | static inline void kmalloc_large_node_hook(void *ptr, size_t size, gfp_t flags) |
1278 | { | |
1279 | kmemleak_alloc(ptr, size, 1, flags); | |
0316bec2 | 1280 | kasan_kmalloc_large(ptr, size); |
d56791b3 RB |
1281 | } |
1282 | ||
1283 | static inline void kfree_hook(const void *x) | |
1284 | { | |
1285 | kmemleak_free(x); | |
0316bec2 | 1286 | kasan_kfree_large(x); |
d56791b3 RB |
1287 | } |
1288 | ||
d56791b3 RB |
1289 | static inline void slab_free_hook(struct kmem_cache *s, void *x) |
1290 | { | |
1291 | kmemleak_free_recursive(x, s->flags); | |
7d550c56 | 1292 | |
02e72cc6 AR |
1293 | /* |
1294 | * Trouble is that we may no longer disable interrupts in the fast path | |
1295 | * So in order to make the debug calls that expect irqs to be | |
1296 | * disabled we need to disable interrupts temporarily. | |
1297 | */ | |
1298 | #if defined(CONFIG_KMEMCHECK) || defined(CONFIG_LOCKDEP) | |
1299 | { | |
1300 | unsigned long flags; | |
1301 | ||
1302 | local_irq_save(flags); | |
1303 | kmemcheck_slab_free(s, x, s->object_size); | |
1304 | debug_check_no_locks_freed(x, s->object_size); | |
1305 | local_irq_restore(flags); | |
1306 | } | |
1307 | #endif | |
1308 | if (!(s->flags & SLAB_DEBUG_OBJECTS)) | |
1309 | debug_check_no_obj_freed(x, s->object_size); | |
0316bec2 AR |
1310 | |
1311 | kasan_slab_free(s, x); | |
02e72cc6 | 1312 | } |
205ab99d | 1313 | |
81084651 JDB |
1314 | static inline void slab_free_freelist_hook(struct kmem_cache *s, |
1315 | void *head, void *tail) | |
1316 | { | |
1317 | /* | |
1318 | * Compiler cannot detect this function can be removed if slab_free_hook() | |
1319 | * evaluates to nothing. Thus, catch all relevant config debug options here. | |
1320 | */ | |
1321 | #if defined(CONFIG_KMEMCHECK) || \ | |
1322 | defined(CONFIG_LOCKDEP) || \ | |
1323 | defined(CONFIG_DEBUG_KMEMLEAK) || \ | |
1324 | defined(CONFIG_DEBUG_OBJECTS_FREE) || \ | |
1325 | defined(CONFIG_KASAN) | |
1326 | ||
1327 | void *object = head; | |
1328 | void *tail_obj = tail ? : head; | |
1329 | ||
1330 | do { | |
1331 | slab_free_hook(s, object); | |
1332 | } while ((object != tail_obj) && | |
1333 | (object = get_freepointer(s, object))); | |
1334 | #endif | |
1335 | } | |
1336 | ||
588f8ba9 TG |
1337 | static void setup_object(struct kmem_cache *s, struct page *page, |
1338 | void *object) | |
1339 | { | |
1340 | setup_object_debug(s, page, object); | |
1341 | if (unlikely(s->ctor)) { | |
1342 | kasan_unpoison_object_data(s, object); | |
1343 | s->ctor(object); | |
1344 | kasan_poison_object_data(s, object); | |
1345 | } | |
1346 | } | |
1347 | ||
81819f0f CL |
1348 | /* |
1349 | * Slab allocation and freeing | |
1350 | */ | |
5dfb4175 VD |
1351 | static inline struct page *alloc_slab_page(struct kmem_cache *s, |
1352 | gfp_t flags, int node, struct kmem_cache_order_objects oo) | |
65c3376a | 1353 | { |
5dfb4175 | 1354 | struct page *page; |
65c3376a CL |
1355 | int order = oo_order(oo); |
1356 | ||
b1eeab67 VN |
1357 | flags |= __GFP_NOTRACK; |
1358 | ||
2154a336 | 1359 | if (node == NUMA_NO_NODE) |
5dfb4175 | 1360 | page = alloc_pages(flags, order); |
65c3376a | 1361 | else |
96db800f | 1362 | page = __alloc_pages_node(node, flags, order); |
5dfb4175 | 1363 | |
f3ccb2c4 VD |
1364 | if (page && memcg_charge_slab(page, flags, order, s)) { |
1365 | __free_pages(page, order); | |
1366 | page = NULL; | |
1367 | } | |
5dfb4175 VD |
1368 | |
1369 | return page; | |
65c3376a CL |
1370 | } |
1371 | ||
81819f0f CL |
1372 | static struct page *allocate_slab(struct kmem_cache *s, gfp_t flags, int node) |
1373 | { | |
06428780 | 1374 | struct page *page; |
834f3d11 | 1375 | struct kmem_cache_order_objects oo = s->oo; |
ba52270d | 1376 | gfp_t alloc_gfp; |
588f8ba9 TG |
1377 | void *start, *p; |
1378 | int idx, order; | |
81819f0f | 1379 | |
7e0528da CL |
1380 | flags &= gfp_allowed_mask; |
1381 | ||
d0164adc | 1382 | if (gfpflags_allow_blocking(flags)) |
7e0528da CL |
1383 | local_irq_enable(); |
1384 | ||
b7a49f0d | 1385 | flags |= s->allocflags; |
e12ba74d | 1386 | |
ba52270d PE |
1387 | /* |
1388 | * Let the initial higher-order allocation fail under memory pressure | |
1389 | * so we fall-back to the minimum order allocation. | |
1390 | */ | |
1391 | alloc_gfp = (flags | __GFP_NOWARN | __GFP_NORETRY) & ~__GFP_NOFAIL; | |
d0164adc MG |
1392 | if ((alloc_gfp & __GFP_DIRECT_RECLAIM) && oo_order(oo) > oo_order(s->min)) |
1393 | alloc_gfp = (alloc_gfp | __GFP_NOMEMALLOC) & ~__GFP_DIRECT_RECLAIM; | |
ba52270d | 1394 | |
5dfb4175 | 1395 | page = alloc_slab_page(s, alloc_gfp, node, oo); |
65c3376a CL |
1396 | if (unlikely(!page)) { |
1397 | oo = s->min; | |
80c3a998 | 1398 | alloc_gfp = flags; |
65c3376a CL |
1399 | /* |
1400 | * Allocation may have failed due to fragmentation. | |
1401 | * Try a lower order alloc if possible | |
1402 | */ | |
5dfb4175 | 1403 | page = alloc_slab_page(s, alloc_gfp, node, oo); |
588f8ba9 TG |
1404 | if (unlikely(!page)) |
1405 | goto out; | |
1406 | stat(s, ORDER_FALLBACK); | |
65c3376a | 1407 | } |
5a896d9e | 1408 | |
588f8ba9 TG |
1409 | if (kmemcheck_enabled && |
1410 | !(s->flags & (SLAB_NOTRACK | DEBUG_DEFAULT_FLAGS))) { | |
b1eeab67 VN |
1411 | int pages = 1 << oo_order(oo); |
1412 | ||
80c3a998 | 1413 | kmemcheck_alloc_shadow(page, oo_order(oo), alloc_gfp, node); |
b1eeab67 VN |
1414 | |
1415 | /* | |
1416 | * Objects from caches that have a constructor don't get | |
1417 | * cleared when they're allocated, so we need to do it here. | |
1418 | */ | |
1419 | if (s->ctor) | |
1420 | kmemcheck_mark_uninitialized_pages(page, pages); | |
1421 | else | |
1422 | kmemcheck_mark_unallocated_pages(page, pages); | |
5a896d9e VN |
1423 | } |
1424 | ||
834f3d11 | 1425 | page->objects = oo_objects(oo); |
81819f0f | 1426 | |
1f458cbf | 1427 | order = compound_order(page); |
1b4f59e3 | 1428 | page->slab_cache = s; |
c03f94cc | 1429 | __SetPageSlab(page); |
2f064f34 | 1430 | if (page_is_pfmemalloc(page)) |
072bb0aa | 1431 | SetPageSlabPfmemalloc(page); |
81819f0f CL |
1432 | |
1433 | start = page_address(page); | |
81819f0f CL |
1434 | |
1435 | if (unlikely(s->flags & SLAB_POISON)) | |
1f458cbf | 1436 | memset(start, POISON_INUSE, PAGE_SIZE << order); |
81819f0f | 1437 | |
0316bec2 AR |
1438 | kasan_poison_slab(page); |
1439 | ||
54266640 WY |
1440 | for_each_object_idx(p, idx, s, start, page->objects) { |
1441 | setup_object(s, page, p); | |
1442 | if (likely(idx < page->objects)) | |
1443 | set_freepointer(s, p, p + s->size); | |
1444 | else | |
1445 | set_freepointer(s, p, NULL); | |
81819f0f | 1446 | } |
81819f0f CL |
1447 | |
1448 | page->freelist = start; | |
e6e82ea1 | 1449 | page->inuse = page->objects; |
8cb0a506 | 1450 | page->frozen = 1; |
588f8ba9 | 1451 | |
81819f0f | 1452 | out: |
d0164adc | 1453 | if (gfpflags_allow_blocking(flags)) |
588f8ba9 TG |
1454 | local_irq_disable(); |
1455 | if (!page) | |
1456 | return NULL; | |
1457 | ||
1458 | mod_zone_page_state(page_zone(page), | |
1459 | (s->flags & SLAB_RECLAIM_ACCOUNT) ? | |
1460 | NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE, | |
1461 | 1 << oo_order(oo)); | |
1462 | ||
1463 | inc_slabs_node(s, page_to_nid(page), page->objects); | |
1464 | ||
81819f0f CL |
1465 | return page; |
1466 | } | |
1467 | ||
588f8ba9 TG |
1468 | static struct page *new_slab(struct kmem_cache *s, gfp_t flags, int node) |
1469 | { | |
1470 | if (unlikely(flags & GFP_SLAB_BUG_MASK)) { | |
1471 | pr_emerg("gfp: %u\n", flags & GFP_SLAB_BUG_MASK); | |
1472 | BUG(); | |
1473 | } | |
1474 | ||
1475 | return allocate_slab(s, | |
1476 | flags & (GFP_RECLAIM_MASK | GFP_CONSTRAINT_MASK), node); | |
1477 | } | |
1478 | ||
81819f0f CL |
1479 | static void __free_slab(struct kmem_cache *s, struct page *page) |
1480 | { | |
834f3d11 CL |
1481 | int order = compound_order(page); |
1482 | int pages = 1 << order; | |
81819f0f | 1483 | |
becfda68 | 1484 | if (s->flags & SLAB_CONSISTENCY_CHECKS) { |
81819f0f CL |
1485 | void *p; |
1486 | ||
1487 | slab_pad_check(s, page); | |
224a88be CL |
1488 | for_each_object(p, s, page_address(page), |
1489 | page->objects) | |
f7cb1933 | 1490 | check_object(s, page, p, SLUB_RED_INACTIVE); |
81819f0f CL |
1491 | } |
1492 | ||
b1eeab67 | 1493 | kmemcheck_free_shadow(page, compound_order(page)); |
5a896d9e | 1494 | |
81819f0f CL |
1495 | mod_zone_page_state(page_zone(page), |
1496 | (s->flags & SLAB_RECLAIM_ACCOUNT) ? | |
1497 | NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE, | |
06428780 | 1498 | -pages); |
81819f0f | 1499 | |
072bb0aa | 1500 | __ClearPageSlabPfmemalloc(page); |
49bd5221 | 1501 | __ClearPageSlab(page); |
1f458cbf | 1502 | |
22b751c3 | 1503 | page_mapcount_reset(page); |
1eb5ac64 NP |
1504 | if (current->reclaim_state) |
1505 | current->reclaim_state->reclaimed_slab += pages; | |
f3ccb2c4 | 1506 | __free_kmem_pages(page, order); |
81819f0f CL |
1507 | } |
1508 | ||
da9a638c LJ |
1509 | #define need_reserve_slab_rcu \ |
1510 | (sizeof(((struct page *)NULL)->lru) < sizeof(struct rcu_head)) | |
1511 | ||
81819f0f CL |
1512 | static void rcu_free_slab(struct rcu_head *h) |
1513 | { | |
1514 | struct page *page; | |
1515 | ||
da9a638c LJ |
1516 | if (need_reserve_slab_rcu) |
1517 | page = virt_to_head_page(h); | |
1518 | else | |
1519 | page = container_of((struct list_head *)h, struct page, lru); | |
1520 | ||
1b4f59e3 | 1521 | __free_slab(page->slab_cache, page); |
81819f0f CL |
1522 | } |
1523 | ||
1524 | static void free_slab(struct kmem_cache *s, struct page *page) | |
1525 | { | |
1526 | if (unlikely(s->flags & SLAB_DESTROY_BY_RCU)) { | |
da9a638c LJ |
1527 | struct rcu_head *head; |
1528 | ||
1529 | if (need_reserve_slab_rcu) { | |
1530 | int order = compound_order(page); | |
1531 | int offset = (PAGE_SIZE << order) - s->reserved; | |
1532 | ||
1533 | VM_BUG_ON(s->reserved != sizeof(*head)); | |
1534 | head = page_address(page) + offset; | |
1535 | } else { | |
bc4f610d | 1536 | head = &page->rcu_head; |
da9a638c | 1537 | } |
81819f0f CL |
1538 | |
1539 | call_rcu(head, rcu_free_slab); | |
1540 | } else | |
1541 | __free_slab(s, page); | |
1542 | } | |
1543 | ||
1544 | static void discard_slab(struct kmem_cache *s, struct page *page) | |
1545 | { | |
205ab99d | 1546 | dec_slabs_node(s, page_to_nid(page), page->objects); |
81819f0f CL |
1547 | free_slab(s, page); |
1548 | } | |
1549 | ||
1550 | /* | |
5cc6eee8 | 1551 | * Management of partially allocated slabs. |
81819f0f | 1552 | */ |
1e4dd946 SR |
1553 | static inline void |
1554 | __add_partial(struct kmem_cache_node *n, struct page *page, int tail) | |
81819f0f | 1555 | { |
e95eed57 | 1556 | n->nr_partial++; |
136333d1 | 1557 | if (tail == DEACTIVATE_TO_TAIL) |
7c2e132c CL |
1558 | list_add_tail(&page->lru, &n->partial); |
1559 | else | |
1560 | list_add(&page->lru, &n->partial); | |
81819f0f CL |
1561 | } |
1562 | ||
1e4dd946 SR |
1563 | static inline void add_partial(struct kmem_cache_node *n, |
1564 | struct page *page, int tail) | |
62e346a8 | 1565 | { |
c65c1877 | 1566 | lockdep_assert_held(&n->list_lock); |
1e4dd946 SR |
1567 | __add_partial(n, page, tail); |
1568 | } | |
c65c1877 | 1569 | |
1e4dd946 SR |
1570 | static inline void remove_partial(struct kmem_cache_node *n, |
1571 | struct page *page) | |
1572 | { | |
1573 | lockdep_assert_held(&n->list_lock); | |
52b4b950 DS |
1574 | list_del(&page->lru); |
1575 | n->nr_partial--; | |
1e4dd946 SR |
1576 | } |
1577 | ||
81819f0f | 1578 | /* |
7ced3719 CL |
1579 | * Remove slab from the partial list, freeze it and |
1580 | * return the pointer to the freelist. | |
81819f0f | 1581 | * |
497b66f2 | 1582 | * Returns a list of objects or NULL if it fails. |
81819f0f | 1583 | */ |
497b66f2 | 1584 | static inline void *acquire_slab(struct kmem_cache *s, |
acd19fd1 | 1585 | struct kmem_cache_node *n, struct page *page, |
633b0764 | 1586 | int mode, int *objects) |
81819f0f | 1587 | { |
2cfb7455 CL |
1588 | void *freelist; |
1589 | unsigned long counters; | |
1590 | struct page new; | |
1591 | ||
c65c1877 PZ |
1592 | lockdep_assert_held(&n->list_lock); |
1593 | ||
2cfb7455 CL |
1594 | /* |
1595 | * Zap the freelist and set the frozen bit. | |
1596 | * The old freelist is the list of objects for the | |
1597 | * per cpu allocation list. | |
1598 | */ | |
7ced3719 CL |
1599 | freelist = page->freelist; |
1600 | counters = page->counters; | |
1601 | new.counters = counters; | |
633b0764 | 1602 | *objects = new.objects - new.inuse; |
23910c50 | 1603 | if (mode) { |
7ced3719 | 1604 | new.inuse = page->objects; |
23910c50 PE |
1605 | new.freelist = NULL; |
1606 | } else { | |
1607 | new.freelist = freelist; | |
1608 | } | |
2cfb7455 | 1609 | |
a0132ac0 | 1610 | VM_BUG_ON(new.frozen); |
7ced3719 | 1611 | new.frozen = 1; |
2cfb7455 | 1612 | |
7ced3719 | 1613 | if (!__cmpxchg_double_slab(s, page, |
2cfb7455 | 1614 | freelist, counters, |
02d7633f | 1615 | new.freelist, new.counters, |
7ced3719 | 1616 | "acquire_slab")) |
7ced3719 | 1617 | return NULL; |
2cfb7455 CL |
1618 | |
1619 | remove_partial(n, page); | |
7ced3719 | 1620 | WARN_ON(!freelist); |
49e22585 | 1621 | return freelist; |
81819f0f CL |
1622 | } |
1623 | ||
633b0764 | 1624 | static void put_cpu_partial(struct kmem_cache *s, struct page *page, int drain); |
8ba00bb6 | 1625 | static inline bool pfmemalloc_match(struct page *page, gfp_t gfpflags); |
49e22585 | 1626 | |
81819f0f | 1627 | /* |
672bba3a | 1628 | * Try to allocate a partial slab from a specific node. |
81819f0f | 1629 | */ |
8ba00bb6 JK |
1630 | static void *get_partial_node(struct kmem_cache *s, struct kmem_cache_node *n, |
1631 | struct kmem_cache_cpu *c, gfp_t flags) | |
81819f0f | 1632 | { |
49e22585 CL |
1633 | struct page *page, *page2; |
1634 | void *object = NULL; | |
633b0764 JK |
1635 | int available = 0; |
1636 | int objects; | |
81819f0f CL |
1637 | |
1638 | /* | |
1639 | * Racy check. If we mistakenly see no partial slabs then we | |
1640 | * just allocate an empty slab. If we mistakenly try to get a | |
672bba3a CL |
1641 | * partial slab and there is none available then get_partials() |
1642 | * will return NULL. | |
81819f0f CL |
1643 | */ |
1644 | if (!n || !n->nr_partial) | |
1645 | return NULL; | |
1646 | ||
1647 | spin_lock(&n->list_lock); | |
49e22585 | 1648 | list_for_each_entry_safe(page, page2, &n->partial, lru) { |
8ba00bb6 | 1649 | void *t; |
49e22585 | 1650 | |
8ba00bb6 JK |
1651 | if (!pfmemalloc_match(page, flags)) |
1652 | continue; | |
1653 | ||
633b0764 | 1654 | t = acquire_slab(s, n, page, object == NULL, &objects); |
49e22585 CL |
1655 | if (!t) |
1656 | break; | |
1657 | ||
633b0764 | 1658 | available += objects; |
12d79634 | 1659 | if (!object) { |
49e22585 | 1660 | c->page = page; |
49e22585 | 1661 | stat(s, ALLOC_FROM_PARTIAL); |
49e22585 | 1662 | object = t; |
49e22585 | 1663 | } else { |
633b0764 | 1664 | put_cpu_partial(s, page, 0); |
8028dcea | 1665 | stat(s, CPU_PARTIAL_NODE); |
49e22585 | 1666 | } |
345c905d JK |
1667 | if (!kmem_cache_has_cpu_partial(s) |
1668 | || available > s->cpu_partial / 2) | |
49e22585 CL |
1669 | break; |
1670 | ||
497b66f2 | 1671 | } |
81819f0f | 1672 | spin_unlock(&n->list_lock); |
497b66f2 | 1673 | return object; |
81819f0f CL |
1674 | } |
1675 | ||
1676 | /* | |
672bba3a | 1677 | * Get a page from somewhere. Search in increasing NUMA distances. |
81819f0f | 1678 | */ |
de3ec035 | 1679 | static void *get_any_partial(struct kmem_cache *s, gfp_t flags, |
acd19fd1 | 1680 | struct kmem_cache_cpu *c) |
81819f0f CL |
1681 | { |
1682 | #ifdef CONFIG_NUMA | |
1683 | struct zonelist *zonelist; | |
dd1a239f | 1684 | struct zoneref *z; |
54a6eb5c MG |
1685 | struct zone *zone; |
1686 | enum zone_type high_zoneidx = gfp_zone(flags); | |
497b66f2 | 1687 | void *object; |
cc9a6c87 | 1688 | unsigned int cpuset_mems_cookie; |
81819f0f CL |
1689 | |
1690 | /* | |
672bba3a CL |
1691 | * The defrag ratio allows a configuration of the tradeoffs between |
1692 | * inter node defragmentation and node local allocations. A lower | |
1693 | * defrag_ratio increases the tendency to do local allocations | |
1694 | * instead of attempting to obtain partial slabs from other nodes. | |
81819f0f | 1695 | * |
672bba3a CL |
1696 | * If the defrag_ratio is set to 0 then kmalloc() always |
1697 | * returns node local objects. If the ratio is higher then kmalloc() | |
1698 | * may return off node objects because partial slabs are obtained | |
1699 | * from other nodes and filled up. | |
81819f0f | 1700 | * |
6446faa2 | 1701 | * If /sys/kernel/slab/xx/defrag_ratio is set to 100 (which makes |
672bba3a CL |
1702 | * defrag_ratio = 1000) then every (well almost) allocation will |
1703 | * first attempt to defrag slab caches on other nodes. This means | |
1704 | * scanning over all nodes to look for partial slabs which may be | |
1705 | * expensive if we do it every time we are trying to find a slab | |
1706 | * with available objects. | |
81819f0f | 1707 | */ |
9824601e CL |
1708 | if (!s->remote_node_defrag_ratio || |
1709 | get_cycles() % 1024 > s->remote_node_defrag_ratio) | |
81819f0f CL |
1710 | return NULL; |
1711 | ||
cc9a6c87 | 1712 | do { |
d26914d1 | 1713 | cpuset_mems_cookie = read_mems_allowed_begin(); |
2a389610 | 1714 | zonelist = node_zonelist(mempolicy_slab_node(), flags); |
cc9a6c87 MG |
1715 | for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) { |
1716 | struct kmem_cache_node *n; | |
1717 | ||
1718 | n = get_node(s, zone_to_nid(zone)); | |
1719 | ||
dee2f8aa | 1720 | if (n && cpuset_zone_allowed(zone, flags) && |
cc9a6c87 | 1721 | n->nr_partial > s->min_partial) { |
8ba00bb6 | 1722 | object = get_partial_node(s, n, c, flags); |
cc9a6c87 MG |
1723 | if (object) { |
1724 | /* | |
d26914d1 MG |
1725 | * Don't check read_mems_allowed_retry() |
1726 | * here - if mems_allowed was updated in | |
1727 | * parallel, that was a harmless race | |
1728 | * between allocation and the cpuset | |
1729 | * update | |
cc9a6c87 | 1730 | */ |
cc9a6c87 MG |
1731 | return object; |
1732 | } | |
c0ff7453 | 1733 | } |
81819f0f | 1734 | } |
d26914d1 | 1735 | } while (read_mems_allowed_retry(cpuset_mems_cookie)); |
81819f0f CL |
1736 | #endif |
1737 | return NULL; | |
1738 | } | |
1739 | ||
1740 | /* | |
1741 | * Get a partial page, lock it and return it. | |
1742 | */ | |
497b66f2 | 1743 | static void *get_partial(struct kmem_cache *s, gfp_t flags, int node, |
acd19fd1 | 1744 | struct kmem_cache_cpu *c) |
81819f0f | 1745 | { |
497b66f2 | 1746 | void *object; |
a561ce00 JK |
1747 | int searchnode = node; |
1748 | ||
1749 | if (node == NUMA_NO_NODE) | |
1750 | searchnode = numa_mem_id(); | |
1751 | else if (!node_present_pages(node)) | |
1752 | searchnode = node_to_mem_node(node); | |
81819f0f | 1753 | |
8ba00bb6 | 1754 | object = get_partial_node(s, get_node(s, searchnode), c, flags); |
497b66f2 CL |
1755 | if (object || node != NUMA_NO_NODE) |
1756 | return object; | |
81819f0f | 1757 | |
acd19fd1 | 1758 | return get_any_partial(s, flags, c); |
81819f0f CL |
1759 | } |
1760 | ||
8a5ec0ba CL |
1761 | #ifdef CONFIG_PREEMPT |
1762 | /* | |
1763 | * Calculate the next globally unique transaction for disambiguiation | |
1764 | * during cmpxchg. The transactions start with the cpu number and are then | |
1765 | * incremented by CONFIG_NR_CPUS. | |
1766 | */ | |
1767 | #define TID_STEP roundup_pow_of_two(CONFIG_NR_CPUS) | |
1768 | #else | |
1769 | /* | |
1770 | * No preemption supported therefore also no need to check for | |
1771 | * different cpus. | |
1772 | */ | |
1773 | #define TID_STEP 1 | |
1774 | #endif | |
1775 | ||
1776 | static inline unsigned long next_tid(unsigned long tid) | |
1777 | { | |
1778 | return tid + TID_STEP; | |
1779 | } | |
1780 | ||
1781 | static inline unsigned int tid_to_cpu(unsigned long tid) | |
1782 | { | |
1783 | return tid % TID_STEP; | |
1784 | } | |
1785 | ||
1786 | static inline unsigned long tid_to_event(unsigned long tid) | |
1787 | { | |
1788 | return tid / TID_STEP; | |
1789 | } | |
1790 | ||
1791 | static inline unsigned int init_tid(int cpu) | |
1792 | { | |
1793 | return cpu; | |
1794 | } | |
1795 | ||
1796 | static inline void note_cmpxchg_failure(const char *n, | |
1797 | const struct kmem_cache *s, unsigned long tid) | |
1798 | { | |
1799 | #ifdef SLUB_DEBUG_CMPXCHG | |
1800 | unsigned long actual_tid = __this_cpu_read(s->cpu_slab->tid); | |
1801 | ||
f9f58285 | 1802 | pr_info("%s %s: cmpxchg redo ", n, s->name); |
8a5ec0ba CL |
1803 | |
1804 | #ifdef CONFIG_PREEMPT | |
1805 | if (tid_to_cpu(tid) != tid_to_cpu(actual_tid)) | |
f9f58285 | 1806 | pr_warn("due to cpu change %d -> %d\n", |
8a5ec0ba CL |
1807 | tid_to_cpu(tid), tid_to_cpu(actual_tid)); |
1808 | else | |
1809 | #endif | |
1810 | if (tid_to_event(tid) != tid_to_event(actual_tid)) | |
f9f58285 | 1811 | pr_warn("due to cpu running other code. Event %ld->%ld\n", |
8a5ec0ba CL |
1812 | tid_to_event(tid), tid_to_event(actual_tid)); |
1813 | else | |
f9f58285 | 1814 | pr_warn("for unknown reason: actual=%lx was=%lx target=%lx\n", |
8a5ec0ba CL |
1815 | actual_tid, tid, next_tid(tid)); |
1816 | #endif | |
4fdccdfb | 1817 | stat(s, CMPXCHG_DOUBLE_CPU_FAIL); |
8a5ec0ba CL |
1818 | } |
1819 | ||
788e1aad | 1820 | static void init_kmem_cache_cpus(struct kmem_cache *s) |
8a5ec0ba | 1821 | { |
8a5ec0ba CL |
1822 | int cpu; |
1823 | ||
1824 | for_each_possible_cpu(cpu) | |
1825 | per_cpu_ptr(s->cpu_slab, cpu)->tid = init_tid(cpu); | |
8a5ec0ba | 1826 | } |
2cfb7455 | 1827 | |
81819f0f CL |
1828 | /* |
1829 | * Remove the cpu slab | |
1830 | */ | |
d0e0ac97 CG |
1831 | static void deactivate_slab(struct kmem_cache *s, struct page *page, |
1832 | void *freelist) | |
81819f0f | 1833 | { |
2cfb7455 | 1834 | enum slab_modes { M_NONE, M_PARTIAL, M_FULL, M_FREE }; |
2cfb7455 CL |
1835 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); |
1836 | int lock = 0; | |
1837 | enum slab_modes l = M_NONE, m = M_NONE; | |
2cfb7455 | 1838 | void *nextfree; |
136333d1 | 1839 | int tail = DEACTIVATE_TO_HEAD; |
2cfb7455 CL |
1840 | struct page new; |
1841 | struct page old; | |
1842 | ||
1843 | if (page->freelist) { | |
84e554e6 | 1844 | stat(s, DEACTIVATE_REMOTE_FREES); |
136333d1 | 1845 | tail = DEACTIVATE_TO_TAIL; |
2cfb7455 CL |
1846 | } |
1847 | ||
894b8788 | 1848 | /* |
2cfb7455 CL |
1849 | * Stage one: Free all available per cpu objects back |
1850 | * to the page freelist while it is still frozen. Leave the | |
1851 | * last one. | |
1852 | * | |
1853 | * There is no need to take the list->lock because the page | |
1854 | * is still frozen. | |
1855 | */ | |
1856 | while (freelist && (nextfree = get_freepointer(s, freelist))) { | |
1857 | void *prior; | |
1858 | unsigned long counters; | |
1859 | ||
1860 | do { | |
1861 | prior = page->freelist; | |
1862 | counters = page->counters; | |
1863 | set_freepointer(s, freelist, prior); | |
1864 | new.counters = counters; | |
1865 | new.inuse--; | |
a0132ac0 | 1866 | VM_BUG_ON(!new.frozen); |
2cfb7455 | 1867 | |
1d07171c | 1868 | } while (!__cmpxchg_double_slab(s, page, |
2cfb7455 CL |
1869 | prior, counters, |
1870 | freelist, new.counters, | |
1871 | "drain percpu freelist")); | |
1872 | ||
1873 | freelist = nextfree; | |
1874 | } | |
1875 | ||
894b8788 | 1876 | /* |
2cfb7455 CL |
1877 | * Stage two: Ensure that the page is unfrozen while the |
1878 | * list presence reflects the actual number of objects | |
1879 | * during unfreeze. | |
1880 | * | |
1881 | * We setup the list membership and then perform a cmpxchg | |
1882 | * with the count. If there is a mismatch then the page | |
1883 | * is not unfrozen but the page is on the wrong list. | |
1884 | * | |
1885 | * Then we restart the process which may have to remove | |
1886 | * the page from the list that we just put it on again | |
1887 | * because the number of objects in the slab may have | |
1888 | * changed. | |
894b8788 | 1889 | */ |
2cfb7455 | 1890 | redo: |
894b8788 | 1891 | |
2cfb7455 CL |
1892 | old.freelist = page->freelist; |
1893 | old.counters = page->counters; | |
a0132ac0 | 1894 | VM_BUG_ON(!old.frozen); |
7c2e132c | 1895 | |
2cfb7455 CL |
1896 | /* Determine target state of the slab */ |
1897 | new.counters = old.counters; | |
1898 | if (freelist) { | |
1899 | new.inuse--; | |
1900 | set_freepointer(s, freelist, old.freelist); | |
1901 | new.freelist = freelist; | |
1902 | } else | |
1903 | new.freelist = old.freelist; | |
1904 | ||
1905 | new.frozen = 0; | |
1906 | ||
8a5b20ae | 1907 | if (!new.inuse && n->nr_partial >= s->min_partial) |
2cfb7455 CL |
1908 | m = M_FREE; |
1909 | else if (new.freelist) { | |
1910 | m = M_PARTIAL; | |
1911 | if (!lock) { | |
1912 | lock = 1; | |
1913 | /* | |
1914 | * Taking the spinlock removes the possiblity | |
1915 | * that acquire_slab() will see a slab page that | |
1916 | * is frozen | |
1917 | */ | |
1918 | spin_lock(&n->list_lock); | |
1919 | } | |
1920 | } else { | |
1921 | m = M_FULL; | |
1922 | if (kmem_cache_debug(s) && !lock) { | |
1923 | lock = 1; | |
1924 | /* | |
1925 | * This also ensures that the scanning of full | |
1926 | * slabs from diagnostic functions will not see | |
1927 | * any frozen slabs. | |
1928 | */ | |
1929 | spin_lock(&n->list_lock); | |
1930 | } | |
1931 | } | |
1932 | ||
1933 | if (l != m) { | |
1934 | ||
1935 | if (l == M_PARTIAL) | |
1936 | ||
1937 | remove_partial(n, page); | |
1938 | ||
1939 | else if (l == M_FULL) | |
894b8788 | 1940 | |
c65c1877 | 1941 | remove_full(s, n, page); |
2cfb7455 CL |
1942 | |
1943 | if (m == M_PARTIAL) { | |
1944 | ||
1945 | add_partial(n, page, tail); | |
136333d1 | 1946 | stat(s, tail); |
2cfb7455 CL |
1947 | |
1948 | } else if (m == M_FULL) { | |
894b8788 | 1949 | |
2cfb7455 CL |
1950 | stat(s, DEACTIVATE_FULL); |
1951 | add_full(s, n, page); | |
1952 | ||
1953 | } | |
1954 | } | |
1955 | ||
1956 | l = m; | |
1d07171c | 1957 | if (!__cmpxchg_double_slab(s, page, |
2cfb7455 CL |
1958 | old.freelist, old.counters, |
1959 | new.freelist, new.counters, | |
1960 | "unfreezing slab")) | |
1961 | goto redo; | |
1962 | ||
2cfb7455 CL |
1963 | if (lock) |
1964 | spin_unlock(&n->list_lock); | |
1965 | ||
1966 | if (m == M_FREE) { | |
1967 | stat(s, DEACTIVATE_EMPTY); | |
1968 | discard_slab(s, page); | |
1969 | stat(s, FREE_SLAB); | |
894b8788 | 1970 | } |
81819f0f CL |
1971 | } |
1972 | ||
d24ac77f JK |
1973 | /* |
1974 | * Unfreeze all the cpu partial slabs. | |
1975 | * | |
59a09917 CL |
1976 | * This function must be called with interrupts disabled |
1977 | * for the cpu using c (or some other guarantee must be there | |
1978 | * to guarantee no concurrent accesses). | |
d24ac77f | 1979 | */ |
59a09917 CL |
1980 | static void unfreeze_partials(struct kmem_cache *s, |
1981 | struct kmem_cache_cpu *c) | |
49e22585 | 1982 | { |
345c905d | 1983 | #ifdef CONFIG_SLUB_CPU_PARTIAL |
43d77867 | 1984 | struct kmem_cache_node *n = NULL, *n2 = NULL; |
9ada1934 | 1985 | struct page *page, *discard_page = NULL; |
49e22585 CL |
1986 | |
1987 | while ((page = c->partial)) { | |
49e22585 CL |
1988 | struct page new; |
1989 | struct page old; | |
1990 | ||
1991 | c->partial = page->next; | |
43d77867 JK |
1992 | |
1993 | n2 = get_node(s, page_to_nid(page)); | |
1994 | if (n != n2) { | |
1995 | if (n) | |
1996 | spin_unlock(&n->list_lock); | |
1997 | ||
1998 | n = n2; | |
1999 | spin_lock(&n->list_lock); | |
2000 | } | |
49e22585 CL |
2001 | |
2002 | do { | |
2003 | ||
2004 | old.freelist = page->freelist; | |
2005 | old.counters = page->counters; | |
a0132ac0 | 2006 | VM_BUG_ON(!old.frozen); |
49e22585 CL |
2007 | |
2008 | new.counters = old.counters; | |
2009 | new.freelist = old.freelist; | |
2010 | ||
2011 | new.frozen = 0; | |
2012 | ||
d24ac77f | 2013 | } while (!__cmpxchg_double_slab(s, page, |
49e22585 CL |
2014 | old.freelist, old.counters, |
2015 | new.freelist, new.counters, | |
2016 | "unfreezing slab")); | |
2017 | ||
8a5b20ae | 2018 | if (unlikely(!new.inuse && n->nr_partial >= s->min_partial)) { |
9ada1934 SL |
2019 | page->next = discard_page; |
2020 | discard_page = page; | |
43d77867 JK |
2021 | } else { |
2022 | add_partial(n, page, DEACTIVATE_TO_TAIL); | |
2023 | stat(s, FREE_ADD_PARTIAL); | |
49e22585 CL |
2024 | } |
2025 | } | |
2026 | ||
2027 | if (n) | |
2028 | spin_unlock(&n->list_lock); | |
9ada1934 SL |
2029 | |
2030 | while (discard_page) { | |
2031 | page = discard_page; | |
2032 | discard_page = discard_page->next; | |
2033 | ||
2034 | stat(s, DEACTIVATE_EMPTY); | |
2035 | discard_slab(s, page); | |
2036 | stat(s, FREE_SLAB); | |
2037 | } | |
345c905d | 2038 | #endif |
49e22585 CL |
2039 | } |
2040 | ||
2041 | /* | |
2042 | * Put a page that was just frozen (in __slab_free) into a partial page | |
2043 | * slot if available. This is done without interrupts disabled and without | |
2044 | * preemption disabled. The cmpxchg is racy and may put the partial page | |
2045 | * onto a random cpus partial slot. | |
2046 | * | |
2047 | * If we did not find a slot then simply move all the partials to the | |
2048 | * per node partial list. | |
2049 | */ | |
633b0764 | 2050 | static void put_cpu_partial(struct kmem_cache *s, struct page *page, int drain) |
49e22585 | 2051 | { |
345c905d | 2052 | #ifdef CONFIG_SLUB_CPU_PARTIAL |
49e22585 CL |
2053 | struct page *oldpage; |
2054 | int pages; | |
2055 | int pobjects; | |
2056 | ||
d6e0b7fa | 2057 | preempt_disable(); |
49e22585 CL |
2058 | do { |
2059 | pages = 0; | |
2060 | pobjects = 0; | |
2061 | oldpage = this_cpu_read(s->cpu_slab->partial); | |
2062 | ||
2063 | if (oldpage) { | |
2064 | pobjects = oldpage->pobjects; | |
2065 | pages = oldpage->pages; | |
2066 | if (drain && pobjects > s->cpu_partial) { | |
2067 | unsigned long flags; | |
2068 | /* | |
2069 | * partial array is full. Move the existing | |
2070 | * set to the per node partial list. | |
2071 | */ | |
2072 | local_irq_save(flags); | |
59a09917 | 2073 | unfreeze_partials(s, this_cpu_ptr(s->cpu_slab)); |
49e22585 | 2074 | local_irq_restore(flags); |
e24fc410 | 2075 | oldpage = NULL; |
49e22585 CL |
2076 | pobjects = 0; |
2077 | pages = 0; | |
8028dcea | 2078 | stat(s, CPU_PARTIAL_DRAIN); |
49e22585 CL |
2079 | } |
2080 | } | |
2081 | ||
2082 | pages++; | |
2083 | pobjects += page->objects - page->inuse; | |
2084 | ||
2085 | page->pages = pages; | |
2086 | page->pobjects = pobjects; | |
2087 | page->next = oldpage; | |
2088 | ||
d0e0ac97 CG |
2089 | } while (this_cpu_cmpxchg(s->cpu_slab->partial, oldpage, page) |
2090 | != oldpage); | |
d6e0b7fa VD |
2091 | if (unlikely(!s->cpu_partial)) { |
2092 | unsigned long flags; | |
2093 | ||
2094 | local_irq_save(flags); | |
2095 | unfreeze_partials(s, this_cpu_ptr(s->cpu_slab)); | |
2096 | local_irq_restore(flags); | |
2097 | } | |
2098 | preempt_enable(); | |
345c905d | 2099 | #endif |
49e22585 CL |
2100 | } |
2101 | ||
dfb4f096 | 2102 | static inline void flush_slab(struct kmem_cache *s, struct kmem_cache_cpu *c) |
81819f0f | 2103 | { |
84e554e6 | 2104 | stat(s, CPUSLAB_FLUSH); |
c17dda40 CL |
2105 | deactivate_slab(s, c->page, c->freelist); |
2106 | ||
2107 | c->tid = next_tid(c->tid); | |
2108 | c->page = NULL; | |
2109 | c->freelist = NULL; | |
81819f0f CL |
2110 | } |
2111 | ||
2112 | /* | |
2113 | * Flush cpu slab. | |
6446faa2 | 2114 | * |
81819f0f CL |
2115 | * Called from IPI handler with interrupts disabled. |
2116 | */ | |
0c710013 | 2117 | static inline void __flush_cpu_slab(struct kmem_cache *s, int cpu) |
81819f0f | 2118 | { |
9dfc6e68 | 2119 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu); |
81819f0f | 2120 | |
49e22585 CL |
2121 | if (likely(c)) { |
2122 | if (c->page) | |
2123 | flush_slab(s, c); | |
2124 | ||
59a09917 | 2125 | unfreeze_partials(s, c); |
49e22585 | 2126 | } |
81819f0f CL |
2127 | } |
2128 | ||
2129 | static void flush_cpu_slab(void *d) | |
2130 | { | |
2131 | struct kmem_cache *s = d; | |
81819f0f | 2132 | |
dfb4f096 | 2133 | __flush_cpu_slab(s, smp_processor_id()); |
81819f0f CL |
2134 | } |
2135 | ||
a8364d55 GBY |
2136 | static bool has_cpu_slab(int cpu, void *info) |
2137 | { | |
2138 | struct kmem_cache *s = info; | |
2139 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu); | |
2140 | ||
02e1a9cd | 2141 | return c->page || c->partial; |
a8364d55 GBY |
2142 | } |
2143 | ||
81819f0f CL |
2144 | static void flush_all(struct kmem_cache *s) |
2145 | { | |
a8364d55 | 2146 | on_each_cpu_cond(has_cpu_slab, flush_cpu_slab, s, 1, GFP_ATOMIC); |
81819f0f CL |
2147 | } |
2148 | ||
dfb4f096 CL |
2149 | /* |
2150 | * Check if the objects in a per cpu structure fit numa | |
2151 | * locality expectations. | |
2152 | */ | |
57d437d2 | 2153 | static inline int node_match(struct page *page, int node) |
dfb4f096 CL |
2154 | { |
2155 | #ifdef CONFIG_NUMA | |
4d7868e6 | 2156 | if (!page || (node != NUMA_NO_NODE && page_to_nid(page) != node)) |
dfb4f096 CL |
2157 | return 0; |
2158 | #endif | |
2159 | return 1; | |
2160 | } | |
2161 | ||
9a02d699 | 2162 | #ifdef CONFIG_SLUB_DEBUG |
781b2ba6 PE |
2163 | static int count_free(struct page *page) |
2164 | { | |
2165 | return page->objects - page->inuse; | |
2166 | } | |
2167 | ||
9a02d699 DR |
2168 | static inline unsigned long node_nr_objs(struct kmem_cache_node *n) |
2169 | { | |
2170 | return atomic_long_read(&n->total_objects); | |
2171 | } | |
2172 | #endif /* CONFIG_SLUB_DEBUG */ | |
2173 | ||
2174 | #if defined(CONFIG_SLUB_DEBUG) || defined(CONFIG_SYSFS) | |
781b2ba6 PE |
2175 | static unsigned long count_partial(struct kmem_cache_node *n, |
2176 | int (*get_count)(struct page *)) | |
2177 | { | |
2178 | unsigned long flags; | |
2179 | unsigned long x = 0; | |
2180 | struct page *page; | |
2181 | ||
2182 | spin_lock_irqsave(&n->list_lock, flags); | |
2183 | list_for_each_entry(page, &n->partial, lru) | |
2184 | x += get_count(page); | |
2185 | spin_unlock_irqrestore(&n->list_lock, flags); | |
2186 | return x; | |
2187 | } | |
9a02d699 | 2188 | #endif /* CONFIG_SLUB_DEBUG || CONFIG_SYSFS */ |
26c02cf0 | 2189 | |
781b2ba6 PE |
2190 | static noinline void |
2191 | slab_out_of_memory(struct kmem_cache *s, gfp_t gfpflags, int nid) | |
2192 | { | |
9a02d699 DR |
2193 | #ifdef CONFIG_SLUB_DEBUG |
2194 | static DEFINE_RATELIMIT_STATE(slub_oom_rs, DEFAULT_RATELIMIT_INTERVAL, | |
2195 | DEFAULT_RATELIMIT_BURST); | |
781b2ba6 | 2196 | int node; |
fa45dc25 | 2197 | struct kmem_cache_node *n; |
781b2ba6 | 2198 | |
9a02d699 DR |
2199 | if ((gfpflags & __GFP_NOWARN) || !__ratelimit(&slub_oom_rs)) |
2200 | return; | |
2201 | ||
f9f58285 | 2202 | pr_warn("SLUB: Unable to allocate memory on node %d (gfp=0x%x)\n", |
781b2ba6 | 2203 | nid, gfpflags); |
f9f58285 FF |
2204 | pr_warn(" cache: %s, object size: %d, buffer size: %d, default order: %d, min order: %d\n", |
2205 | s->name, s->object_size, s->size, oo_order(s->oo), | |
2206 | oo_order(s->min)); | |
781b2ba6 | 2207 | |
3b0efdfa | 2208 | if (oo_order(s->min) > get_order(s->object_size)) |
f9f58285 FF |
2209 | pr_warn(" %s debugging increased min order, use slub_debug=O to disable.\n", |
2210 | s->name); | |
fa5ec8a1 | 2211 | |
fa45dc25 | 2212 | for_each_kmem_cache_node(s, node, n) { |
781b2ba6 PE |
2213 | unsigned long nr_slabs; |
2214 | unsigned long nr_objs; | |
2215 | unsigned long nr_free; | |
2216 | ||
26c02cf0 AB |
2217 | nr_free = count_partial(n, count_free); |
2218 | nr_slabs = node_nr_slabs(n); | |
2219 | nr_objs = node_nr_objs(n); | |
781b2ba6 | 2220 | |
f9f58285 | 2221 | pr_warn(" node %d: slabs: %ld, objs: %ld, free: %ld\n", |
781b2ba6 PE |
2222 | node, nr_slabs, nr_objs, nr_free); |
2223 | } | |
9a02d699 | 2224 | #endif |
781b2ba6 PE |
2225 | } |
2226 | ||
497b66f2 CL |
2227 | static inline void *new_slab_objects(struct kmem_cache *s, gfp_t flags, |
2228 | int node, struct kmem_cache_cpu **pc) | |
2229 | { | |
6faa6833 | 2230 | void *freelist; |
188fd063 CL |
2231 | struct kmem_cache_cpu *c = *pc; |
2232 | struct page *page; | |
497b66f2 | 2233 | |
188fd063 | 2234 | freelist = get_partial(s, flags, node, c); |
497b66f2 | 2235 | |
188fd063 CL |
2236 | if (freelist) |
2237 | return freelist; | |
2238 | ||
2239 | page = new_slab(s, flags, node); | |
497b66f2 | 2240 | if (page) { |
7c8e0181 | 2241 | c = raw_cpu_ptr(s->cpu_slab); |
497b66f2 CL |
2242 | if (c->page) |
2243 | flush_slab(s, c); | |
2244 | ||
2245 | /* | |
2246 | * No other reference to the page yet so we can | |
2247 | * muck around with it freely without cmpxchg | |
2248 | */ | |
6faa6833 | 2249 | freelist = page->freelist; |
497b66f2 CL |
2250 | page->freelist = NULL; |
2251 | ||
2252 | stat(s, ALLOC_SLAB); | |
497b66f2 CL |
2253 | c->page = page; |
2254 | *pc = c; | |
2255 | } else | |
6faa6833 | 2256 | freelist = NULL; |
497b66f2 | 2257 | |
6faa6833 | 2258 | return freelist; |
497b66f2 CL |
2259 | } |
2260 | ||
072bb0aa MG |
2261 | static inline bool pfmemalloc_match(struct page *page, gfp_t gfpflags) |
2262 | { | |
2263 | if (unlikely(PageSlabPfmemalloc(page))) | |
2264 | return gfp_pfmemalloc_allowed(gfpflags); | |
2265 | ||
2266 | return true; | |
2267 | } | |
2268 | ||
213eeb9f | 2269 | /* |
d0e0ac97 CG |
2270 | * Check the page->freelist of a page and either transfer the freelist to the |
2271 | * per cpu freelist or deactivate the page. | |
213eeb9f CL |
2272 | * |
2273 | * The page is still frozen if the return value is not NULL. | |
2274 | * | |
2275 | * If this function returns NULL then the page has been unfrozen. | |
d24ac77f JK |
2276 | * |
2277 | * This function must be called with interrupt disabled. | |
213eeb9f CL |
2278 | */ |
2279 | static inline void *get_freelist(struct kmem_cache *s, struct page *page) | |
2280 | { | |
2281 | struct page new; | |
2282 | unsigned long counters; | |
2283 | void *freelist; | |
2284 | ||
2285 | do { | |
2286 | freelist = page->freelist; | |
2287 | counters = page->counters; | |
6faa6833 | 2288 | |
213eeb9f | 2289 | new.counters = counters; |
a0132ac0 | 2290 | VM_BUG_ON(!new.frozen); |
213eeb9f CL |
2291 | |
2292 | new.inuse = page->objects; | |
2293 | new.frozen = freelist != NULL; | |
2294 | ||
d24ac77f | 2295 | } while (!__cmpxchg_double_slab(s, page, |
213eeb9f CL |
2296 | freelist, counters, |
2297 | NULL, new.counters, | |
2298 | "get_freelist")); | |
2299 | ||
2300 | return freelist; | |
2301 | } | |
2302 | ||
81819f0f | 2303 | /* |
894b8788 CL |
2304 | * Slow path. The lockless freelist is empty or we need to perform |
2305 | * debugging duties. | |
2306 | * | |
894b8788 CL |
2307 | * Processing is still very fast if new objects have been freed to the |
2308 | * regular freelist. In that case we simply take over the regular freelist | |
2309 | * as the lockless freelist and zap the regular freelist. | |
81819f0f | 2310 | * |
894b8788 CL |
2311 | * If that is not working then we fall back to the partial lists. We take the |
2312 | * first element of the freelist as the object to allocate now and move the | |
2313 | * rest of the freelist to the lockless freelist. | |
81819f0f | 2314 | * |
894b8788 | 2315 | * And if we were unable to get a new slab from the partial slab lists then |
6446faa2 CL |
2316 | * we need to allocate a new slab. This is the slowest path since it involves |
2317 | * a call to the page allocator and the setup of a new slab. | |
a380a3c7 CL |
2318 | * |
2319 | * Version of __slab_alloc to use when we know that interrupts are | |
2320 | * already disabled (which is the case for bulk allocation). | |
81819f0f | 2321 | */ |
a380a3c7 | 2322 | static void *___slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node, |
ce71e27c | 2323 | unsigned long addr, struct kmem_cache_cpu *c) |
81819f0f | 2324 | { |
6faa6833 | 2325 | void *freelist; |
f6e7def7 | 2326 | struct page *page; |
81819f0f | 2327 | |
f6e7def7 CL |
2328 | page = c->page; |
2329 | if (!page) | |
81819f0f | 2330 | goto new_slab; |
49e22585 | 2331 | redo: |
6faa6833 | 2332 | |
57d437d2 | 2333 | if (unlikely(!node_match(page, node))) { |
a561ce00 JK |
2334 | int searchnode = node; |
2335 | ||
2336 | if (node != NUMA_NO_NODE && !node_present_pages(node)) | |
2337 | searchnode = node_to_mem_node(node); | |
2338 | ||
2339 | if (unlikely(!node_match(page, searchnode))) { | |
2340 | stat(s, ALLOC_NODE_MISMATCH); | |
2341 | deactivate_slab(s, page, c->freelist); | |
2342 | c->page = NULL; | |
2343 | c->freelist = NULL; | |
2344 | goto new_slab; | |
2345 | } | |
fc59c053 | 2346 | } |
6446faa2 | 2347 | |
072bb0aa MG |
2348 | /* |
2349 | * By rights, we should be searching for a slab page that was | |
2350 | * PFMEMALLOC but right now, we are losing the pfmemalloc | |
2351 | * information when the page leaves the per-cpu allocator | |
2352 | */ | |
2353 | if (unlikely(!pfmemalloc_match(page, gfpflags))) { | |
2354 | deactivate_slab(s, page, c->freelist); | |
2355 | c->page = NULL; | |
2356 | c->freelist = NULL; | |
2357 | goto new_slab; | |
2358 | } | |
2359 | ||
73736e03 | 2360 | /* must check again c->freelist in case of cpu migration or IRQ */ |
6faa6833 CL |
2361 | freelist = c->freelist; |
2362 | if (freelist) | |
73736e03 | 2363 | goto load_freelist; |
03e404af | 2364 | |
f6e7def7 | 2365 | freelist = get_freelist(s, page); |
6446faa2 | 2366 | |
6faa6833 | 2367 | if (!freelist) { |
03e404af CL |
2368 | c->page = NULL; |
2369 | stat(s, DEACTIVATE_BYPASS); | |
fc59c053 | 2370 | goto new_slab; |
03e404af | 2371 | } |
6446faa2 | 2372 | |
84e554e6 | 2373 | stat(s, ALLOC_REFILL); |
6446faa2 | 2374 | |
894b8788 | 2375 | load_freelist: |
507effea CL |
2376 | /* |
2377 | * freelist is pointing to the list of objects to be used. | |
2378 | * page is pointing to the page from which the objects are obtained. | |
2379 | * That page must be frozen for per cpu allocations to work. | |
2380 | */ | |
a0132ac0 | 2381 | VM_BUG_ON(!c->page->frozen); |
6faa6833 | 2382 | c->freelist = get_freepointer(s, freelist); |
8a5ec0ba | 2383 | c->tid = next_tid(c->tid); |
6faa6833 | 2384 | return freelist; |
81819f0f | 2385 | |
81819f0f | 2386 | new_slab: |
2cfb7455 | 2387 | |
49e22585 | 2388 | if (c->partial) { |
f6e7def7 CL |
2389 | page = c->page = c->partial; |
2390 | c->partial = page->next; | |
49e22585 CL |
2391 | stat(s, CPU_PARTIAL_ALLOC); |
2392 | c->freelist = NULL; | |
2393 | goto redo; | |
81819f0f CL |
2394 | } |
2395 | ||
188fd063 | 2396 | freelist = new_slab_objects(s, gfpflags, node, &c); |
01ad8a7b | 2397 | |
f4697436 | 2398 | if (unlikely(!freelist)) { |
9a02d699 | 2399 | slab_out_of_memory(s, gfpflags, node); |
f4697436 | 2400 | return NULL; |
81819f0f | 2401 | } |
2cfb7455 | 2402 | |
f6e7def7 | 2403 | page = c->page; |
5091b74a | 2404 | if (likely(!kmem_cache_debug(s) && pfmemalloc_match(page, gfpflags))) |
4b6f0750 | 2405 | goto load_freelist; |
2cfb7455 | 2406 | |
497b66f2 | 2407 | /* Only entered in the debug case */ |
d0e0ac97 CG |
2408 | if (kmem_cache_debug(s) && |
2409 | !alloc_debug_processing(s, page, freelist, addr)) | |
497b66f2 | 2410 | goto new_slab; /* Slab failed checks. Next slab needed */ |
894b8788 | 2411 | |
f6e7def7 | 2412 | deactivate_slab(s, page, get_freepointer(s, freelist)); |
c17dda40 CL |
2413 | c->page = NULL; |
2414 | c->freelist = NULL; | |
6faa6833 | 2415 | return freelist; |
894b8788 CL |
2416 | } |
2417 | ||
a380a3c7 CL |
2418 | /* |
2419 | * Another one that disabled interrupt and compensates for possible | |
2420 | * cpu changes by refetching the per cpu area pointer. | |
2421 | */ | |
2422 | static void *__slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node, | |
2423 | unsigned long addr, struct kmem_cache_cpu *c) | |
2424 | { | |
2425 | void *p; | |
2426 | unsigned long flags; | |
2427 | ||
2428 | local_irq_save(flags); | |
2429 | #ifdef CONFIG_PREEMPT | |
2430 | /* | |
2431 | * We may have been preempted and rescheduled on a different | |
2432 | * cpu before disabling interrupts. Need to reload cpu area | |
2433 | * pointer. | |
2434 | */ | |
2435 | c = this_cpu_ptr(s->cpu_slab); | |
2436 | #endif | |
2437 | ||
2438 | p = ___slab_alloc(s, gfpflags, node, addr, c); | |
2439 | local_irq_restore(flags); | |
2440 | return p; | |
2441 | } | |
2442 | ||
894b8788 CL |
2443 | /* |
2444 | * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc) | |
2445 | * have the fastpath folded into their functions. So no function call | |
2446 | * overhead for requests that can be satisfied on the fastpath. | |
2447 | * | |
2448 | * The fastpath works by first checking if the lockless freelist can be used. | |
2449 | * If not then __slab_alloc is called for slow processing. | |
2450 | * | |
2451 | * Otherwise we can simply pick the next object from the lockless free list. | |
2452 | */ | |
2b847c3c | 2453 | static __always_inline void *slab_alloc_node(struct kmem_cache *s, |
ce71e27c | 2454 | gfp_t gfpflags, int node, unsigned long addr) |
894b8788 | 2455 | { |
03ec0ed5 | 2456 | void *object; |
dfb4f096 | 2457 | struct kmem_cache_cpu *c; |
57d437d2 | 2458 | struct page *page; |
8a5ec0ba | 2459 | unsigned long tid; |
1f84260c | 2460 | |
8135be5a VD |
2461 | s = slab_pre_alloc_hook(s, gfpflags); |
2462 | if (!s) | |
773ff60e | 2463 | return NULL; |
8a5ec0ba | 2464 | redo: |
8a5ec0ba CL |
2465 | /* |
2466 | * Must read kmem_cache cpu data via this cpu ptr. Preemption is | |
2467 | * enabled. We may switch back and forth between cpus while | |
2468 | * reading from one cpu area. That does not matter as long | |
2469 | * as we end up on the original cpu again when doing the cmpxchg. | |
7cccd80b | 2470 | * |
9aabf810 JK |
2471 | * We should guarantee that tid and kmem_cache are retrieved on |
2472 | * the same cpu. It could be different if CONFIG_PREEMPT so we need | |
2473 | * to check if it is matched or not. | |
8a5ec0ba | 2474 | */ |
9aabf810 JK |
2475 | do { |
2476 | tid = this_cpu_read(s->cpu_slab->tid); | |
2477 | c = raw_cpu_ptr(s->cpu_slab); | |
859b7a0e MR |
2478 | } while (IS_ENABLED(CONFIG_PREEMPT) && |
2479 | unlikely(tid != READ_ONCE(c->tid))); | |
9aabf810 JK |
2480 | |
2481 | /* | |
2482 | * Irqless object alloc/free algorithm used here depends on sequence | |
2483 | * of fetching cpu_slab's data. tid should be fetched before anything | |
2484 | * on c to guarantee that object and page associated with previous tid | |
2485 | * won't be used with current tid. If we fetch tid first, object and | |
2486 | * page could be one associated with next tid and our alloc/free | |
2487 | * request will be failed. In this case, we will retry. So, no problem. | |
2488 | */ | |
2489 | barrier(); | |
8a5ec0ba | 2490 | |
8a5ec0ba CL |
2491 | /* |
2492 | * The transaction ids are globally unique per cpu and per operation on | |
2493 | * a per cpu queue. Thus they can be guarantee that the cmpxchg_double | |
2494 | * occurs on the right processor and that there was no operation on the | |
2495 | * linked list in between. | |
2496 | */ | |
8a5ec0ba | 2497 | |
9dfc6e68 | 2498 | object = c->freelist; |
57d437d2 | 2499 | page = c->page; |
8eae1492 | 2500 | if (unlikely(!object || !node_match(page, node))) { |
dfb4f096 | 2501 | object = __slab_alloc(s, gfpflags, node, addr, c); |
8eae1492 DH |
2502 | stat(s, ALLOC_SLOWPATH); |
2503 | } else { | |
0ad9500e ED |
2504 | void *next_object = get_freepointer_safe(s, object); |
2505 | ||
8a5ec0ba | 2506 | /* |
25985edc | 2507 | * The cmpxchg will only match if there was no additional |
8a5ec0ba CL |
2508 | * operation and if we are on the right processor. |
2509 | * | |
d0e0ac97 CG |
2510 | * The cmpxchg does the following atomically (without lock |
2511 | * semantics!) | |
8a5ec0ba CL |
2512 | * 1. Relocate first pointer to the current per cpu area. |
2513 | * 2. Verify that tid and freelist have not been changed | |
2514 | * 3. If they were not changed replace tid and freelist | |
2515 | * | |
d0e0ac97 CG |
2516 | * Since this is without lock semantics the protection is only |
2517 | * against code executing on this cpu *not* from access by | |
2518 | * other cpus. | |
8a5ec0ba | 2519 | */ |
933393f5 | 2520 | if (unlikely(!this_cpu_cmpxchg_double( |
8a5ec0ba CL |
2521 | s->cpu_slab->freelist, s->cpu_slab->tid, |
2522 | object, tid, | |
0ad9500e | 2523 | next_object, next_tid(tid)))) { |
8a5ec0ba CL |
2524 | |
2525 | note_cmpxchg_failure("slab_alloc", s, tid); | |
2526 | goto redo; | |
2527 | } | |
0ad9500e | 2528 | prefetch_freepointer(s, next_object); |
84e554e6 | 2529 | stat(s, ALLOC_FASTPATH); |
894b8788 | 2530 | } |
8a5ec0ba | 2531 | |
74e2134f | 2532 | if (unlikely(gfpflags & __GFP_ZERO) && object) |
3b0efdfa | 2533 | memset(object, 0, s->object_size); |
d07dbea4 | 2534 | |
03ec0ed5 | 2535 | slab_post_alloc_hook(s, gfpflags, 1, &object); |
5a896d9e | 2536 | |
894b8788 | 2537 | return object; |
81819f0f CL |
2538 | } |
2539 | ||
2b847c3c EG |
2540 | static __always_inline void *slab_alloc(struct kmem_cache *s, |
2541 | gfp_t gfpflags, unsigned long addr) | |
2542 | { | |
2543 | return slab_alloc_node(s, gfpflags, NUMA_NO_NODE, addr); | |
2544 | } | |
2545 | ||
81819f0f CL |
2546 | void *kmem_cache_alloc(struct kmem_cache *s, gfp_t gfpflags) |
2547 | { | |
2b847c3c | 2548 | void *ret = slab_alloc(s, gfpflags, _RET_IP_); |
5b882be4 | 2549 | |
d0e0ac97 CG |
2550 | trace_kmem_cache_alloc(_RET_IP_, ret, s->object_size, |
2551 | s->size, gfpflags); | |
5b882be4 EGM |
2552 | |
2553 | return ret; | |
81819f0f CL |
2554 | } |
2555 | EXPORT_SYMBOL(kmem_cache_alloc); | |
2556 | ||
0f24f128 | 2557 | #ifdef CONFIG_TRACING |
4a92379b RK |
2558 | void *kmem_cache_alloc_trace(struct kmem_cache *s, gfp_t gfpflags, size_t size) |
2559 | { | |
2b847c3c | 2560 | void *ret = slab_alloc(s, gfpflags, _RET_IP_); |
4a92379b | 2561 | trace_kmalloc(_RET_IP_, ret, size, s->size, gfpflags); |
0316bec2 | 2562 | kasan_kmalloc(s, ret, size); |
4a92379b RK |
2563 | return ret; |
2564 | } | |
2565 | EXPORT_SYMBOL(kmem_cache_alloc_trace); | |
5b882be4 EGM |
2566 | #endif |
2567 | ||
81819f0f CL |
2568 | #ifdef CONFIG_NUMA |
2569 | void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t gfpflags, int node) | |
2570 | { | |
2b847c3c | 2571 | void *ret = slab_alloc_node(s, gfpflags, node, _RET_IP_); |
5b882be4 | 2572 | |
ca2b84cb | 2573 | trace_kmem_cache_alloc_node(_RET_IP_, ret, |
3b0efdfa | 2574 | s->object_size, s->size, gfpflags, node); |
5b882be4 EGM |
2575 | |
2576 | return ret; | |
81819f0f CL |
2577 | } |
2578 | EXPORT_SYMBOL(kmem_cache_alloc_node); | |
81819f0f | 2579 | |
0f24f128 | 2580 | #ifdef CONFIG_TRACING |
4a92379b | 2581 | void *kmem_cache_alloc_node_trace(struct kmem_cache *s, |
5b882be4 | 2582 | gfp_t gfpflags, |
4a92379b | 2583 | int node, size_t size) |
5b882be4 | 2584 | { |
2b847c3c | 2585 | void *ret = slab_alloc_node(s, gfpflags, node, _RET_IP_); |
4a92379b RK |
2586 | |
2587 | trace_kmalloc_node(_RET_IP_, ret, | |
2588 | size, s->size, gfpflags, node); | |
0316bec2 AR |
2589 | |
2590 | kasan_kmalloc(s, ret, size); | |
4a92379b | 2591 | return ret; |
5b882be4 | 2592 | } |
4a92379b | 2593 | EXPORT_SYMBOL(kmem_cache_alloc_node_trace); |
5b882be4 | 2594 | #endif |
5d1f57e4 | 2595 | #endif |
5b882be4 | 2596 | |
81819f0f | 2597 | /* |
94e4d712 | 2598 | * Slow path handling. This may still be called frequently since objects |
894b8788 | 2599 | * have a longer lifetime than the cpu slabs in most processing loads. |
81819f0f | 2600 | * |
894b8788 CL |
2601 | * So we still attempt to reduce cache line usage. Just take the slab |
2602 | * lock and free the item. If there is no additional partial page | |
2603 | * handling required then we can return immediately. | |
81819f0f | 2604 | */ |
894b8788 | 2605 | static void __slab_free(struct kmem_cache *s, struct page *page, |
81084651 JDB |
2606 | void *head, void *tail, int cnt, |
2607 | unsigned long addr) | |
2608 | ||
81819f0f CL |
2609 | { |
2610 | void *prior; | |
2cfb7455 | 2611 | int was_frozen; |
2cfb7455 CL |
2612 | struct page new; |
2613 | unsigned long counters; | |
2614 | struct kmem_cache_node *n = NULL; | |
61728d1e | 2615 | unsigned long uninitialized_var(flags); |
81819f0f | 2616 | |
8a5ec0ba | 2617 | stat(s, FREE_SLOWPATH); |
81819f0f | 2618 | |
19c7ff9e | 2619 | if (kmem_cache_debug(s) && |
282acb43 | 2620 | !free_debug_processing(s, page, head, tail, cnt, addr)) |
80f08c19 | 2621 | return; |
6446faa2 | 2622 | |
2cfb7455 | 2623 | do { |
837d678d JK |
2624 | if (unlikely(n)) { |
2625 | spin_unlock_irqrestore(&n->list_lock, flags); | |
2626 | n = NULL; | |
2627 | } | |
2cfb7455 CL |
2628 | prior = page->freelist; |
2629 | counters = page->counters; | |
81084651 | 2630 | set_freepointer(s, tail, prior); |
2cfb7455 CL |
2631 | new.counters = counters; |
2632 | was_frozen = new.frozen; | |
81084651 | 2633 | new.inuse -= cnt; |
837d678d | 2634 | if ((!new.inuse || !prior) && !was_frozen) { |
49e22585 | 2635 | |
c65c1877 | 2636 | if (kmem_cache_has_cpu_partial(s) && !prior) { |
49e22585 CL |
2637 | |
2638 | /* | |
d0e0ac97 CG |
2639 | * Slab was on no list before and will be |
2640 | * partially empty | |
2641 | * We can defer the list move and instead | |
2642 | * freeze it. | |
49e22585 CL |
2643 | */ |
2644 | new.frozen = 1; | |
2645 | ||
c65c1877 | 2646 | } else { /* Needs to be taken off a list */ |
49e22585 | 2647 | |
b455def2 | 2648 | n = get_node(s, page_to_nid(page)); |
49e22585 CL |
2649 | /* |
2650 | * Speculatively acquire the list_lock. | |
2651 | * If the cmpxchg does not succeed then we may | |
2652 | * drop the list_lock without any processing. | |
2653 | * | |
2654 | * Otherwise the list_lock will synchronize with | |
2655 | * other processors updating the list of slabs. | |
2656 | */ | |
2657 | spin_lock_irqsave(&n->list_lock, flags); | |
2658 | ||
2659 | } | |
2cfb7455 | 2660 | } |
81819f0f | 2661 | |
2cfb7455 CL |
2662 | } while (!cmpxchg_double_slab(s, page, |
2663 | prior, counters, | |
81084651 | 2664 | head, new.counters, |
2cfb7455 | 2665 | "__slab_free")); |
81819f0f | 2666 | |
2cfb7455 | 2667 | if (likely(!n)) { |
49e22585 CL |
2668 | |
2669 | /* | |
2670 | * If we just froze the page then put it onto the | |
2671 | * per cpu partial list. | |
2672 | */ | |
8028dcea | 2673 | if (new.frozen && !was_frozen) { |
49e22585 | 2674 | put_cpu_partial(s, page, 1); |
8028dcea AS |
2675 | stat(s, CPU_PARTIAL_FREE); |
2676 | } | |
49e22585 | 2677 | /* |
2cfb7455 CL |
2678 | * The list lock was not taken therefore no list |
2679 | * activity can be necessary. | |
2680 | */ | |
b455def2 L |
2681 | if (was_frozen) |
2682 | stat(s, FREE_FROZEN); | |
2683 | return; | |
2684 | } | |
81819f0f | 2685 | |
8a5b20ae | 2686 | if (unlikely(!new.inuse && n->nr_partial >= s->min_partial)) |
837d678d JK |
2687 | goto slab_empty; |
2688 | ||
81819f0f | 2689 | /* |
837d678d JK |
2690 | * Objects left in the slab. If it was not on the partial list before |
2691 | * then add it. | |
81819f0f | 2692 | */ |
345c905d JK |
2693 | if (!kmem_cache_has_cpu_partial(s) && unlikely(!prior)) { |
2694 | if (kmem_cache_debug(s)) | |
c65c1877 | 2695 | remove_full(s, n, page); |
837d678d JK |
2696 | add_partial(n, page, DEACTIVATE_TO_TAIL); |
2697 | stat(s, FREE_ADD_PARTIAL); | |
8ff12cfc | 2698 | } |
80f08c19 | 2699 | spin_unlock_irqrestore(&n->list_lock, flags); |
81819f0f CL |
2700 | return; |
2701 | ||
2702 | slab_empty: | |
a973e9dd | 2703 | if (prior) { |
81819f0f | 2704 | /* |
6fbabb20 | 2705 | * Slab on the partial list. |
81819f0f | 2706 | */ |
5cc6eee8 | 2707 | remove_partial(n, page); |
84e554e6 | 2708 | stat(s, FREE_REMOVE_PARTIAL); |
c65c1877 | 2709 | } else { |
6fbabb20 | 2710 | /* Slab must be on the full list */ |
c65c1877 PZ |
2711 | remove_full(s, n, page); |
2712 | } | |
2cfb7455 | 2713 | |
80f08c19 | 2714 | spin_unlock_irqrestore(&n->list_lock, flags); |
84e554e6 | 2715 | stat(s, FREE_SLAB); |
81819f0f | 2716 | discard_slab(s, page); |
81819f0f CL |
2717 | } |
2718 | ||
894b8788 CL |
2719 | /* |
2720 | * Fastpath with forced inlining to produce a kfree and kmem_cache_free that | |
2721 | * can perform fastpath freeing without additional function calls. | |
2722 | * | |
2723 | * The fastpath is only possible if we are freeing to the current cpu slab | |
2724 | * of this processor. This typically the case if we have just allocated | |
2725 | * the item before. | |
2726 | * | |
2727 | * If fastpath is not possible then fall back to __slab_free where we deal | |
2728 | * with all sorts of special processing. | |
81084651 JDB |
2729 | * |
2730 | * Bulk free of a freelist with several objects (all pointing to the | |
2731 | * same page) possible by specifying head and tail ptr, plus objects | |
2732 | * count (cnt). Bulk free indicated by tail pointer being set. | |
894b8788 | 2733 | */ |
81084651 JDB |
2734 | static __always_inline void slab_free(struct kmem_cache *s, struct page *page, |
2735 | void *head, void *tail, int cnt, | |
2736 | unsigned long addr) | |
894b8788 | 2737 | { |
81084651 | 2738 | void *tail_obj = tail ? : head; |
dfb4f096 | 2739 | struct kmem_cache_cpu *c; |
8a5ec0ba | 2740 | unsigned long tid; |
1f84260c | 2741 | |
81084651 | 2742 | slab_free_freelist_hook(s, head, tail); |
c016b0bd | 2743 | |
8a5ec0ba CL |
2744 | redo: |
2745 | /* | |
2746 | * Determine the currently cpus per cpu slab. | |
2747 | * The cpu may change afterward. However that does not matter since | |
2748 | * data is retrieved via this pointer. If we are on the same cpu | |
2ae44005 | 2749 | * during the cmpxchg then the free will succeed. |
8a5ec0ba | 2750 | */ |
9aabf810 JK |
2751 | do { |
2752 | tid = this_cpu_read(s->cpu_slab->tid); | |
2753 | c = raw_cpu_ptr(s->cpu_slab); | |
859b7a0e MR |
2754 | } while (IS_ENABLED(CONFIG_PREEMPT) && |
2755 | unlikely(tid != READ_ONCE(c->tid))); | |
c016b0bd | 2756 | |
9aabf810 JK |
2757 | /* Same with comment on barrier() in slab_alloc_node() */ |
2758 | barrier(); | |
c016b0bd | 2759 | |
442b06bc | 2760 | if (likely(page == c->page)) { |
81084651 | 2761 | set_freepointer(s, tail_obj, c->freelist); |
8a5ec0ba | 2762 | |
933393f5 | 2763 | if (unlikely(!this_cpu_cmpxchg_double( |
8a5ec0ba CL |
2764 | s->cpu_slab->freelist, s->cpu_slab->tid, |
2765 | c->freelist, tid, | |
81084651 | 2766 | head, next_tid(tid)))) { |
8a5ec0ba CL |
2767 | |
2768 | note_cmpxchg_failure("slab_free", s, tid); | |
2769 | goto redo; | |
2770 | } | |
84e554e6 | 2771 | stat(s, FREE_FASTPATH); |
894b8788 | 2772 | } else |
81084651 | 2773 | __slab_free(s, page, head, tail_obj, cnt, addr); |
894b8788 | 2774 | |
894b8788 CL |
2775 | } |
2776 | ||
81819f0f CL |
2777 | void kmem_cache_free(struct kmem_cache *s, void *x) |
2778 | { | |
b9ce5ef4 GC |
2779 | s = cache_from_obj(s, x); |
2780 | if (!s) | |
79576102 | 2781 | return; |
81084651 | 2782 | slab_free(s, virt_to_head_page(x), x, NULL, 1, _RET_IP_); |
ca2b84cb | 2783 | trace_kmem_cache_free(_RET_IP_, x); |
81819f0f CL |
2784 | } |
2785 | EXPORT_SYMBOL(kmem_cache_free); | |
2786 | ||
d0ecd894 | 2787 | struct detached_freelist { |
fbd02630 | 2788 | struct page *page; |
d0ecd894 JDB |
2789 | void *tail; |
2790 | void *freelist; | |
2791 | int cnt; | |
376bf125 | 2792 | struct kmem_cache *s; |
d0ecd894 | 2793 | }; |
fbd02630 | 2794 | |
d0ecd894 JDB |
2795 | /* |
2796 | * This function progressively scans the array with free objects (with | |
2797 | * a limited look ahead) and extract objects belonging to the same | |
2798 | * page. It builds a detached freelist directly within the given | |
2799 | * page/objects. This can happen without any need for | |
2800 | * synchronization, because the objects are owned by running process. | |
2801 | * The freelist is build up as a single linked list in the objects. | |
2802 | * The idea is, that this detached freelist can then be bulk | |
2803 | * transferred to the real freelist(s), but only requiring a single | |
2804 | * synchronization primitive. Look ahead in the array is limited due | |
2805 | * to performance reasons. | |
2806 | */ | |
376bf125 JDB |
2807 | static inline |
2808 | int build_detached_freelist(struct kmem_cache *s, size_t size, | |
2809 | void **p, struct detached_freelist *df) | |
d0ecd894 JDB |
2810 | { |
2811 | size_t first_skipped_index = 0; | |
2812 | int lookahead = 3; | |
2813 | void *object; | |
ca257195 | 2814 | struct page *page; |
fbd02630 | 2815 | |
d0ecd894 JDB |
2816 | /* Always re-init detached_freelist */ |
2817 | df->page = NULL; | |
fbd02630 | 2818 | |
d0ecd894 JDB |
2819 | do { |
2820 | object = p[--size]; | |
ca257195 | 2821 | /* Do we need !ZERO_OR_NULL_PTR(object) here? (for kfree) */ |
d0ecd894 | 2822 | } while (!object && size); |
3eed034d | 2823 | |
d0ecd894 JDB |
2824 | if (!object) |
2825 | return 0; | |
fbd02630 | 2826 | |
ca257195 JDB |
2827 | page = virt_to_head_page(object); |
2828 | if (!s) { | |
2829 | /* Handle kalloc'ed objects */ | |
2830 | if (unlikely(!PageSlab(page))) { | |
2831 | BUG_ON(!PageCompound(page)); | |
2832 | kfree_hook(object); | |
2833 | __free_kmem_pages(page, compound_order(page)); | |
2834 | p[size] = NULL; /* mark object processed */ | |
2835 | return size; | |
2836 | } | |
2837 | /* Derive kmem_cache from object */ | |
2838 | df->s = page->slab_cache; | |
2839 | } else { | |
2840 | df->s = cache_from_obj(s, object); /* Support for memcg */ | |
2841 | } | |
376bf125 | 2842 | |
d0ecd894 | 2843 | /* Start new detached freelist */ |
ca257195 | 2844 | df->page = page; |
376bf125 | 2845 | set_freepointer(df->s, object, NULL); |
d0ecd894 JDB |
2846 | df->tail = object; |
2847 | df->freelist = object; | |
2848 | p[size] = NULL; /* mark object processed */ | |
2849 | df->cnt = 1; | |
2850 | ||
2851 | while (size) { | |
2852 | object = p[--size]; | |
2853 | if (!object) | |
2854 | continue; /* Skip processed objects */ | |
2855 | ||
2856 | /* df->page is always set at this point */ | |
2857 | if (df->page == virt_to_head_page(object)) { | |
2858 | /* Opportunity build freelist */ | |
376bf125 | 2859 | set_freepointer(df->s, object, df->freelist); |
d0ecd894 JDB |
2860 | df->freelist = object; |
2861 | df->cnt++; | |
2862 | p[size] = NULL; /* mark object processed */ | |
2863 | ||
2864 | continue; | |
fbd02630 | 2865 | } |
d0ecd894 JDB |
2866 | |
2867 | /* Limit look ahead search */ | |
2868 | if (!--lookahead) | |
2869 | break; | |
2870 | ||
2871 | if (!first_skipped_index) | |
2872 | first_skipped_index = size + 1; | |
fbd02630 | 2873 | } |
d0ecd894 JDB |
2874 | |
2875 | return first_skipped_index; | |
2876 | } | |
2877 | ||
d0ecd894 | 2878 | /* Note that interrupts must be enabled when calling this function. */ |
376bf125 | 2879 | void kmem_cache_free_bulk(struct kmem_cache *s, size_t size, void **p) |
d0ecd894 JDB |
2880 | { |
2881 | if (WARN_ON(!size)) | |
2882 | return; | |
2883 | ||
2884 | do { | |
2885 | struct detached_freelist df; | |
2886 | ||
2887 | size = build_detached_freelist(s, size, p, &df); | |
2888 | if (unlikely(!df.page)) | |
2889 | continue; | |
2890 | ||
376bf125 | 2891 | slab_free(df.s, df.page, df.freelist, df.tail, df.cnt,_RET_IP_); |
d0ecd894 | 2892 | } while (likely(size)); |
484748f0 CL |
2893 | } |
2894 | EXPORT_SYMBOL(kmem_cache_free_bulk); | |
2895 | ||
994eb764 | 2896 | /* Note that interrupts must be enabled when calling this function. */ |
865762a8 JDB |
2897 | int kmem_cache_alloc_bulk(struct kmem_cache *s, gfp_t flags, size_t size, |
2898 | void **p) | |
484748f0 | 2899 | { |
994eb764 JDB |
2900 | struct kmem_cache_cpu *c; |
2901 | int i; | |
2902 | ||
03ec0ed5 JDB |
2903 | /* memcg and kmem_cache debug support */ |
2904 | s = slab_pre_alloc_hook(s, flags); | |
2905 | if (unlikely(!s)) | |
2906 | return false; | |
994eb764 JDB |
2907 | /* |
2908 | * Drain objects in the per cpu slab, while disabling local | |
2909 | * IRQs, which protects against PREEMPT and interrupts | |
2910 | * handlers invoking normal fastpath. | |
2911 | */ | |
2912 | local_irq_disable(); | |
2913 | c = this_cpu_ptr(s->cpu_slab); | |
2914 | ||
2915 | for (i = 0; i < size; i++) { | |
2916 | void *object = c->freelist; | |
2917 | ||
ebe909e0 | 2918 | if (unlikely(!object)) { |
ebe909e0 JDB |
2919 | /* |
2920 | * Invoking slow path likely have side-effect | |
2921 | * of re-populating per CPU c->freelist | |
2922 | */ | |
87098373 | 2923 | p[i] = ___slab_alloc(s, flags, NUMA_NO_NODE, |
ebe909e0 | 2924 | _RET_IP_, c); |
87098373 CL |
2925 | if (unlikely(!p[i])) |
2926 | goto error; | |
2927 | ||
ebe909e0 JDB |
2928 | c = this_cpu_ptr(s->cpu_slab); |
2929 | continue; /* goto for-loop */ | |
2930 | } | |
994eb764 JDB |
2931 | c->freelist = get_freepointer(s, object); |
2932 | p[i] = object; | |
2933 | } | |
2934 | c->tid = next_tid(c->tid); | |
2935 | local_irq_enable(); | |
2936 | ||
2937 | /* Clear memory outside IRQ disabled fastpath loop */ | |
2938 | if (unlikely(flags & __GFP_ZERO)) { | |
2939 | int j; | |
2940 | ||
2941 | for (j = 0; j < i; j++) | |
2942 | memset(p[j], 0, s->object_size); | |
2943 | } | |
2944 | ||
03ec0ed5 JDB |
2945 | /* memcg and kmem_cache debug support */ |
2946 | slab_post_alloc_hook(s, flags, size, p); | |
865762a8 | 2947 | return i; |
87098373 | 2948 | error: |
87098373 | 2949 | local_irq_enable(); |
03ec0ed5 JDB |
2950 | slab_post_alloc_hook(s, flags, i, p); |
2951 | __kmem_cache_free_bulk(s, i, p); | |
865762a8 | 2952 | return 0; |
484748f0 CL |
2953 | } |
2954 | EXPORT_SYMBOL(kmem_cache_alloc_bulk); | |
2955 | ||
2956 | ||
81819f0f | 2957 | /* |
672bba3a CL |
2958 | * Object placement in a slab is made very easy because we always start at |
2959 | * offset 0. If we tune the size of the object to the alignment then we can | |
2960 | * get the required alignment by putting one properly sized object after | |
2961 | * another. | |
81819f0f CL |
2962 | * |
2963 | * Notice that the allocation order determines the sizes of the per cpu | |
2964 | * caches. Each processor has always one slab available for allocations. | |
2965 | * Increasing the allocation order reduces the number of times that slabs | |
672bba3a | 2966 | * must be moved on and off the partial lists and is therefore a factor in |
81819f0f | 2967 | * locking overhead. |
81819f0f CL |
2968 | */ |
2969 | ||
2970 | /* | |
2971 | * Mininum / Maximum order of slab pages. This influences locking overhead | |
2972 | * and slab fragmentation. A higher order reduces the number of partial slabs | |
2973 | * and increases the number of allocations possible without having to | |
2974 | * take the list_lock. | |
2975 | */ | |
2976 | static int slub_min_order; | |
114e9e89 | 2977 | static int slub_max_order = PAGE_ALLOC_COSTLY_ORDER; |
9b2cd506 | 2978 | static int slub_min_objects; |
81819f0f | 2979 | |
81819f0f CL |
2980 | /* |
2981 | * Calculate the order of allocation given an slab object size. | |
2982 | * | |
672bba3a CL |
2983 | * The order of allocation has significant impact on performance and other |
2984 | * system components. Generally order 0 allocations should be preferred since | |
2985 | * order 0 does not cause fragmentation in the page allocator. Larger objects | |
2986 | * be problematic to put into order 0 slabs because there may be too much | |
c124f5b5 | 2987 | * unused space left. We go to a higher order if more than 1/16th of the slab |
672bba3a CL |
2988 | * would be wasted. |
2989 | * | |
2990 | * In order to reach satisfactory performance we must ensure that a minimum | |
2991 | * number of objects is in one slab. Otherwise we may generate too much | |
2992 | * activity on the partial lists which requires taking the list_lock. This is | |
2993 | * less a concern for large slabs though which are rarely used. | |
81819f0f | 2994 | * |
672bba3a CL |
2995 | * slub_max_order specifies the order where we begin to stop considering the |
2996 | * number of objects in a slab as critical. If we reach slub_max_order then | |
2997 | * we try to keep the page order as low as possible. So we accept more waste | |
2998 | * of space in favor of a small page order. | |
81819f0f | 2999 | * |
672bba3a CL |
3000 | * Higher order allocations also allow the placement of more objects in a |
3001 | * slab and thereby reduce object handling overhead. If the user has | |
3002 | * requested a higher mininum order then we start with that one instead of | |
3003 | * the smallest order which will fit the object. | |
81819f0f | 3004 | */ |
5e6d444e | 3005 | static inline int slab_order(int size, int min_objects, |
ab9a0f19 | 3006 | int max_order, int fract_leftover, int reserved) |
81819f0f CL |
3007 | { |
3008 | int order; | |
3009 | int rem; | |
6300ea75 | 3010 | int min_order = slub_min_order; |
81819f0f | 3011 | |
ab9a0f19 | 3012 | if (order_objects(min_order, size, reserved) > MAX_OBJS_PER_PAGE) |
210b5c06 | 3013 | return get_order(size * MAX_OBJS_PER_PAGE) - 1; |
39b26464 | 3014 | |
9f835703 | 3015 | for (order = max(min_order, get_order(min_objects * size + reserved)); |
5e6d444e | 3016 | order <= max_order; order++) { |
81819f0f | 3017 | |
5e6d444e | 3018 | unsigned long slab_size = PAGE_SIZE << order; |
81819f0f | 3019 | |
ab9a0f19 | 3020 | rem = (slab_size - reserved) % size; |
81819f0f | 3021 | |
5e6d444e | 3022 | if (rem <= slab_size / fract_leftover) |
81819f0f | 3023 | break; |
81819f0f | 3024 | } |
672bba3a | 3025 | |
81819f0f CL |
3026 | return order; |
3027 | } | |
3028 | ||
ab9a0f19 | 3029 | static inline int calculate_order(int size, int reserved) |
5e6d444e CL |
3030 | { |
3031 | int order; | |
3032 | int min_objects; | |
3033 | int fraction; | |
e8120ff1 | 3034 | int max_objects; |
5e6d444e CL |
3035 | |
3036 | /* | |
3037 | * Attempt to find best configuration for a slab. This | |
3038 | * works by first attempting to generate a layout with | |
3039 | * the best configuration and backing off gradually. | |
3040 | * | |
422ff4d7 | 3041 | * First we increase the acceptable waste in a slab. Then |
5e6d444e CL |
3042 | * we reduce the minimum objects required in a slab. |
3043 | */ | |
3044 | min_objects = slub_min_objects; | |
9b2cd506 CL |
3045 | if (!min_objects) |
3046 | min_objects = 4 * (fls(nr_cpu_ids) + 1); | |
ab9a0f19 | 3047 | max_objects = order_objects(slub_max_order, size, reserved); |
e8120ff1 ZY |
3048 | min_objects = min(min_objects, max_objects); |
3049 | ||
5e6d444e | 3050 | while (min_objects > 1) { |
c124f5b5 | 3051 | fraction = 16; |
5e6d444e CL |
3052 | while (fraction >= 4) { |
3053 | order = slab_order(size, min_objects, | |
ab9a0f19 | 3054 | slub_max_order, fraction, reserved); |
5e6d444e CL |
3055 | if (order <= slub_max_order) |
3056 | return order; | |
3057 | fraction /= 2; | |
3058 | } | |
5086c389 | 3059 | min_objects--; |
5e6d444e CL |
3060 | } |
3061 | ||
3062 | /* | |
3063 | * We were unable to place multiple objects in a slab. Now | |
3064 | * lets see if we can place a single object there. | |
3065 | */ | |
ab9a0f19 | 3066 | order = slab_order(size, 1, slub_max_order, 1, reserved); |
5e6d444e CL |
3067 | if (order <= slub_max_order) |
3068 | return order; | |
3069 | ||
3070 | /* | |
3071 | * Doh this slab cannot be placed using slub_max_order. | |
3072 | */ | |
ab9a0f19 | 3073 | order = slab_order(size, 1, MAX_ORDER, 1, reserved); |
818cf590 | 3074 | if (order < MAX_ORDER) |
5e6d444e CL |
3075 | return order; |
3076 | return -ENOSYS; | |
3077 | } | |
3078 | ||
5595cffc | 3079 | static void |
4053497d | 3080 | init_kmem_cache_node(struct kmem_cache_node *n) |
81819f0f CL |
3081 | { |
3082 | n->nr_partial = 0; | |
81819f0f CL |
3083 | spin_lock_init(&n->list_lock); |
3084 | INIT_LIST_HEAD(&n->partial); | |
8ab1372f | 3085 | #ifdef CONFIG_SLUB_DEBUG |
0f389ec6 | 3086 | atomic_long_set(&n->nr_slabs, 0); |
02b71b70 | 3087 | atomic_long_set(&n->total_objects, 0); |
643b1138 | 3088 | INIT_LIST_HEAD(&n->full); |
8ab1372f | 3089 | #endif |
81819f0f CL |
3090 | } |
3091 | ||
55136592 | 3092 | static inline int alloc_kmem_cache_cpus(struct kmem_cache *s) |
4c93c355 | 3093 | { |
6c182dc0 | 3094 | BUILD_BUG_ON(PERCPU_DYNAMIC_EARLY_SIZE < |
95a05b42 | 3095 | KMALLOC_SHIFT_HIGH * sizeof(struct kmem_cache_cpu)); |
4c93c355 | 3096 | |
8a5ec0ba | 3097 | /* |
d4d84fef CM |
3098 | * Must align to double word boundary for the double cmpxchg |
3099 | * instructions to work; see __pcpu_double_call_return_bool(). | |
8a5ec0ba | 3100 | */ |
d4d84fef CM |
3101 | s->cpu_slab = __alloc_percpu(sizeof(struct kmem_cache_cpu), |
3102 | 2 * sizeof(void *)); | |
8a5ec0ba CL |
3103 | |
3104 | if (!s->cpu_slab) | |
3105 | return 0; | |
3106 | ||
3107 | init_kmem_cache_cpus(s); | |
4c93c355 | 3108 | |
8a5ec0ba | 3109 | return 1; |
4c93c355 | 3110 | } |
4c93c355 | 3111 | |
51df1142 CL |
3112 | static struct kmem_cache *kmem_cache_node; |
3113 | ||
81819f0f CL |
3114 | /* |
3115 | * No kmalloc_node yet so do it by hand. We know that this is the first | |
3116 | * slab on the node for this slabcache. There are no concurrent accesses | |
3117 | * possible. | |
3118 | * | |
721ae22a ZYW |
3119 | * Note that this function only works on the kmem_cache_node |
3120 | * when allocating for the kmem_cache_node. This is used for bootstrapping | |
4c93c355 | 3121 | * memory on a fresh node that has no slab structures yet. |
81819f0f | 3122 | */ |
55136592 | 3123 | static void early_kmem_cache_node_alloc(int node) |
81819f0f CL |
3124 | { |
3125 | struct page *page; | |
3126 | struct kmem_cache_node *n; | |
3127 | ||
51df1142 | 3128 | BUG_ON(kmem_cache_node->size < sizeof(struct kmem_cache_node)); |
81819f0f | 3129 | |
51df1142 | 3130 | page = new_slab(kmem_cache_node, GFP_NOWAIT, node); |
81819f0f CL |
3131 | |
3132 | BUG_ON(!page); | |
a2f92ee7 | 3133 | if (page_to_nid(page) != node) { |
f9f58285 FF |
3134 | pr_err("SLUB: Unable to allocate memory from node %d\n", node); |
3135 | pr_err("SLUB: Allocating a useless per node structure in order to be able to continue\n"); | |
a2f92ee7 CL |
3136 | } |
3137 | ||
81819f0f CL |
3138 | n = page->freelist; |
3139 | BUG_ON(!n); | |
51df1142 | 3140 | page->freelist = get_freepointer(kmem_cache_node, n); |
e6e82ea1 | 3141 | page->inuse = 1; |
8cb0a506 | 3142 | page->frozen = 0; |
51df1142 | 3143 | kmem_cache_node->node[node] = n; |
8ab1372f | 3144 | #ifdef CONFIG_SLUB_DEBUG |
f7cb1933 | 3145 | init_object(kmem_cache_node, n, SLUB_RED_ACTIVE); |
51df1142 | 3146 | init_tracking(kmem_cache_node, n); |
8ab1372f | 3147 | #endif |
0316bec2 | 3148 | kasan_kmalloc(kmem_cache_node, n, sizeof(struct kmem_cache_node)); |
4053497d | 3149 | init_kmem_cache_node(n); |
51df1142 | 3150 | inc_slabs_node(kmem_cache_node, node, page->objects); |
6446faa2 | 3151 | |
67b6c900 | 3152 | /* |
1e4dd946 SR |
3153 | * No locks need to be taken here as it has just been |
3154 | * initialized and there is no concurrent access. | |
67b6c900 | 3155 | */ |
1e4dd946 | 3156 | __add_partial(n, page, DEACTIVATE_TO_HEAD); |
81819f0f CL |
3157 | } |
3158 | ||
3159 | static void free_kmem_cache_nodes(struct kmem_cache *s) | |
3160 | { | |
3161 | int node; | |
fa45dc25 | 3162 | struct kmem_cache_node *n; |
81819f0f | 3163 | |
fa45dc25 CL |
3164 | for_each_kmem_cache_node(s, node, n) { |
3165 | kmem_cache_free(kmem_cache_node, n); | |
81819f0f CL |
3166 | s->node[node] = NULL; |
3167 | } | |
3168 | } | |
3169 | ||
52b4b950 DS |
3170 | void __kmem_cache_release(struct kmem_cache *s) |
3171 | { | |
3172 | free_percpu(s->cpu_slab); | |
3173 | free_kmem_cache_nodes(s); | |
3174 | } | |
3175 | ||
55136592 | 3176 | static int init_kmem_cache_nodes(struct kmem_cache *s) |
81819f0f CL |
3177 | { |
3178 | int node; | |
81819f0f | 3179 | |
f64dc58c | 3180 | for_each_node_state(node, N_NORMAL_MEMORY) { |
81819f0f CL |
3181 | struct kmem_cache_node *n; |
3182 | ||
73367bd8 | 3183 | if (slab_state == DOWN) { |
55136592 | 3184 | early_kmem_cache_node_alloc(node); |
73367bd8 AD |
3185 | continue; |
3186 | } | |
51df1142 | 3187 | n = kmem_cache_alloc_node(kmem_cache_node, |
55136592 | 3188 | GFP_KERNEL, node); |
81819f0f | 3189 | |
73367bd8 AD |
3190 | if (!n) { |
3191 | free_kmem_cache_nodes(s); | |
3192 | return 0; | |
81819f0f | 3193 | } |
73367bd8 | 3194 | |
81819f0f | 3195 | s->node[node] = n; |
4053497d | 3196 | init_kmem_cache_node(n); |
81819f0f CL |
3197 | } |
3198 | return 1; | |
3199 | } | |
81819f0f | 3200 | |
c0bdb232 | 3201 | static void set_min_partial(struct kmem_cache *s, unsigned long min) |
3b89d7d8 DR |
3202 | { |
3203 | if (min < MIN_PARTIAL) | |
3204 | min = MIN_PARTIAL; | |
3205 | else if (min > MAX_PARTIAL) | |
3206 | min = MAX_PARTIAL; | |
3207 | s->min_partial = min; | |
3208 | } | |
3209 | ||
81819f0f CL |
3210 | /* |
3211 | * calculate_sizes() determines the order and the distribution of data within | |
3212 | * a slab object. | |
3213 | */ | |
06b285dc | 3214 | static int calculate_sizes(struct kmem_cache *s, int forced_order) |
81819f0f CL |
3215 | { |
3216 | unsigned long flags = s->flags; | |
3b0efdfa | 3217 | unsigned long size = s->object_size; |
834f3d11 | 3218 | int order; |
81819f0f | 3219 | |
d8b42bf5 CL |
3220 | /* |
3221 | * Round up object size to the next word boundary. We can only | |
3222 | * place the free pointer at word boundaries and this determines | |
3223 | * the possible location of the free pointer. | |
3224 | */ | |
3225 | size = ALIGN(size, sizeof(void *)); | |
3226 | ||
3227 | #ifdef CONFIG_SLUB_DEBUG | |
81819f0f CL |
3228 | /* |
3229 | * Determine if we can poison the object itself. If the user of | |
3230 | * the slab may touch the object after free or before allocation | |
3231 | * then we should never poison the object itself. | |
3232 | */ | |
3233 | if ((flags & SLAB_POISON) && !(flags & SLAB_DESTROY_BY_RCU) && | |
c59def9f | 3234 | !s->ctor) |
81819f0f CL |
3235 | s->flags |= __OBJECT_POISON; |
3236 | else | |
3237 | s->flags &= ~__OBJECT_POISON; | |
3238 | ||
81819f0f CL |
3239 | |
3240 | /* | |
672bba3a | 3241 | * If we are Redzoning then check if there is some space between the |
81819f0f | 3242 | * end of the object and the free pointer. If not then add an |
672bba3a | 3243 | * additional word to have some bytes to store Redzone information. |
81819f0f | 3244 | */ |
3b0efdfa | 3245 | if ((flags & SLAB_RED_ZONE) && size == s->object_size) |
81819f0f | 3246 | size += sizeof(void *); |
41ecc55b | 3247 | #endif |
81819f0f CL |
3248 | |
3249 | /* | |
672bba3a CL |
3250 | * With that we have determined the number of bytes in actual use |
3251 | * by the object. This is the potential offset to the free pointer. | |
81819f0f CL |
3252 | */ |
3253 | s->inuse = size; | |
3254 | ||
3255 | if (((flags & (SLAB_DESTROY_BY_RCU | SLAB_POISON)) || | |
c59def9f | 3256 | s->ctor)) { |
81819f0f CL |
3257 | /* |
3258 | * Relocate free pointer after the object if it is not | |
3259 | * permitted to overwrite the first word of the object on | |
3260 | * kmem_cache_free. | |
3261 | * | |
3262 | * This is the case if we do RCU, have a constructor or | |
3263 | * destructor or are poisoning the objects. | |
3264 | */ | |
3265 | s->offset = size; | |
3266 | size += sizeof(void *); | |
3267 | } | |
3268 | ||
c12b3c62 | 3269 | #ifdef CONFIG_SLUB_DEBUG |
81819f0f CL |
3270 | if (flags & SLAB_STORE_USER) |
3271 | /* | |
3272 | * Need to store information about allocs and frees after | |
3273 | * the object. | |
3274 | */ | |
3275 | size += 2 * sizeof(struct track); | |
3276 | ||
be7b3fbc | 3277 | if (flags & SLAB_RED_ZONE) |
81819f0f CL |
3278 | /* |
3279 | * Add some empty padding so that we can catch | |
3280 | * overwrites from earlier objects rather than let | |
3281 | * tracking information or the free pointer be | |
0211a9c8 | 3282 | * corrupted if a user writes before the start |
81819f0f CL |
3283 | * of the object. |
3284 | */ | |
3285 | size += sizeof(void *); | |
41ecc55b | 3286 | #endif |
672bba3a | 3287 | |
81819f0f CL |
3288 | /* |
3289 | * SLUB stores one object immediately after another beginning from | |
3290 | * offset 0. In order to align the objects we have to simply size | |
3291 | * each object to conform to the alignment. | |
3292 | */ | |
45906855 | 3293 | size = ALIGN(size, s->align); |
81819f0f | 3294 | s->size = size; |
06b285dc CL |
3295 | if (forced_order >= 0) |
3296 | order = forced_order; | |
3297 | else | |
ab9a0f19 | 3298 | order = calculate_order(size, s->reserved); |
81819f0f | 3299 | |
834f3d11 | 3300 | if (order < 0) |
81819f0f CL |
3301 | return 0; |
3302 | ||
b7a49f0d | 3303 | s->allocflags = 0; |
834f3d11 | 3304 | if (order) |
b7a49f0d CL |
3305 | s->allocflags |= __GFP_COMP; |
3306 | ||
3307 | if (s->flags & SLAB_CACHE_DMA) | |
2c59dd65 | 3308 | s->allocflags |= GFP_DMA; |
b7a49f0d CL |
3309 | |
3310 | if (s->flags & SLAB_RECLAIM_ACCOUNT) | |
3311 | s->allocflags |= __GFP_RECLAIMABLE; | |
3312 | ||
81819f0f CL |
3313 | /* |
3314 | * Determine the number of objects per slab | |
3315 | */ | |
ab9a0f19 LJ |
3316 | s->oo = oo_make(order, size, s->reserved); |
3317 | s->min = oo_make(get_order(size), size, s->reserved); | |
205ab99d CL |
3318 | if (oo_objects(s->oo) > oo_objects(s->max)) |
3319 | s->max = s->oo; | |
81819f0f | 3320 | |
834f3d11 | 3321 | return !!oo_objects(s->oo); |
81819f0f CL |
3322 | } |
3323 | ||
8a13a4cc | 3324 | static int kmem_cache_open(struct kmem_cache *s, unsigned long flags) |
81819f0f | 3325 | { |
8a13a4cc | 3326 | s->flags = kmem_cache_flags(s->size, flags, s->name, s->ctor); |
ab9a0f19 | 3327 | s->reserved = 0; |
81819f0f | 3328 | |
da9a638c LJ |
3329 | if (need_reserve_slab_rcu && (s->flags & SLAB_DESTROY_BY_RCU)) |
3330 | s->reserved = sizeof(struct rcu_head); | |
81819f0f | 3331 | |
06b285dc | 3332 | if (!calculate_sizes(s, -1)) |
81819f0f | 3333 | goto error; |
3de47213 DR |
3334 | if (disable_higher_order_debug) { |
3335 | /* | |
3336 | * Disable debugging flags that store metadata if the min slab | |
3337 | * order increased. | |
3338 | */ | |
3b0efdfa | 3339 | if (get_order(s->size) > get_order(s->object_size)) { |
3de47213 DR |
3340 | s->flags &= ~DEBUG_METADATA_FLAGS; |
3341 | s->offset = 0; | |
3342 | if (!calculate_sizes(s, -1)) | |
3343 | goto error; | |
3344 | } | |
3345 | } | |
81819f0f | 3346 | |
2565409f HC |
3347 | #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \ |
3348 | defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE) | |
149daaf3 | 3349 | if (system_has_cmpxchg_double() && (s->flags & SLAB_NO_CMPXCHG) == 0) |
b789ef51 CL |
3350 | /* Enable fast mode */ |
3351 | s->flags |= __CMPXCHG_DOUBLE; | |
3352 | #endif | |
3353 | ||
3b89d7d8 DR |
3354 | /* |
3355 | * The larger the object size is, the more pages we want on the partial | |
3356 | * list to avoid pounding the page allocator excessively. | |
3357 | */ | |
49e22585 CL |
3358 | set_min_partial(s, ilog2(s->size) / 2); |
3359 | ||
3360 | /* | |
3361 | * cpu_partial determined the maximum number of objects kept in the | |
3362 | * per cpu partial lists of a processor. | |
3363 | * | |
3364 | * Per cpu partial lists mainly contain slabs that just have one | |
3365 | * object freed. If they are used for allocation then they can be | |
3366 | * filled up again with minimal effort. The slab will never hit the | |
3367 | * per node partial lists and therefore no locking will be required. | |
3368 | * | |
3369 | * This setting also determines | |
3370 | * | |
3371 | * A) The number of objects from per cpu partial slabs dumped to the | |
3372 | * per node list when we reach the limit. | |
9f264904 | 3373 | * B) The number of objects in cpu partial slabs to extract from the |
d0e0ac97 CG |
3374 | * per node list when we run out of per cpu objects. We only fetch |
3375 | * 50% to keep some capacity around for frees. | |
49e22585 | 3376 | */ |
345c905d | 3377 | if (!kmem_cache_has_cpu_partial(s)) |
8f1e33da CL |
3378 | s->cpu_partial = 0; |
3379 | else if (s->size >= PAGE_SIZE) | |
49e22585 CL |
3380 | s->cpu_partial = 2; |
3381 | else if (s->size >= 1024) | |
3382 | s->cpu_partial = 6; | |
3383 | else if (s->size >= 256) | |
3384 | s->cpu_partial = 13; | |
3385 | else | |
3386 | s->cpu_partial = 30; | |
3387 | ||
81819f0f | 3388 | #ifdef CONFIG_NUMA |
e2cb96b7 | 3389 | s->remote_node_defrag_ratio = 1000; |
81819f0f | 3390 | #endif |
55136592 | 3391 | if (!init_kmem_cache_nodes(s)) |
dfb4f096 | 3392 | goto error; |
81819f0f | 3393 | |
55136592 | 3394 | if (alloc_kmem_cache_cpus(s)) |
278b1bb1 | 3395 | return 0; |
ff12059e | 3396 | |
4c93c355 | 3397 | free_kmem_cache_nodes(s); |
81819f0f CL |
3398 | error: |
3399 | if (flags & SLAB_PANIC) | |
3400 | panic("Cannot create slab %s size=%lu realsize=%u " | |
3401 | "order=%u offset=%u flags=%lx\n", | |
d0e0ac97 CG |
3402 | s->name, (unsigned long)s->size, s->size, |
3403 | oo_order(s->oo), s->offset, flags); | |
278b1bb1 | 3404 | return -EINVAL; |
81819f0f | 3405 | } |
81819f0f | 3406 | |
33b12c38 CL |
3407 | static void list_slab_objects(struct kmem_cache *s, struct page *page, |
3408 | const char *text) | |
3409 | { | |
3410 | #ifdef CONFIG_SLUB_DEBUG | |
3411 | void *addr = page_address(page); | |
3412 | void *p; | |
a5dd5c11 NK |
3413 | unsigned long *map = kzalloc(BITS_TO_LONGS(page->objects) * |
3414 | sizeof(long), GFP_ATOMIC); | |
bbd7d57b ED |
3415 | if (!map) |
3416 | return; | |
945cf2b6 | 3417 | slab_err(s, page, text, s->name); |
33b12c38 | 3418 | slab_lock(page); |
33b12c38 | 3419 | |
5f80b13a | 3420 | get_map(s, page, map); |
33b12c38 CL |
3421 | for_each_object(p, s, addr, page->objects) { |
3422 | ||
3423 | if (!test_bit(slab_index(p, s, addr), map)) { | |
f9f58285 | 3424 | pr_err("INFO: Object 0x%p @offset=%tu\n", p, p - addr); |
33b12c38 CL |
3425 | print_tracking(s, p); |
3426 | } | |
3427 | } | |
3428 | slab_unlock(page); | |
bbd7d57b | 3429 | kfree(map); |
33b12c38 CL |
3430 | #endif |
3431 | } | |
3432 | ||
81819f0f | 3433 | /* |
599870b1 | 3434 | * Attempt to free all partial slabs on a node. |
52b4b950 DS |
3435 | * This is called from __kmem_cache_shutdown(). We must take list_lock |
3436 | * because sysfs file might still access partial list after the shutdowning. | |
81819f0f | 3437 | */ |
599870b1 | 3438 | static void free_partial(struct kmem_cache *s, struct kmem_cache_node *n) |
81819f0f | 3439 | { |
81819f0f CL |
3440 | struct page *page, *h; |
3441 | ||
52b4b950 DS |
3442 | BUG_ON(irqs_disabled()); |
3443 | spin_lock_irq(&n->list_lock); | |
33b12c38 | 3444 | list_for_each_entry_safe(page, h, &n->partial, lru) { |
81819f0f | 3445 | if (!page->inuse) { |
52b4b950 | 3446 | remove_partial(n, page); |
81819f0f | 3447 | discard_slab(s, page); |
33b12c38 CL |
3448 | } else { |
3449 | list_slab_objects(s, page, | |
52b4b950 | 3450 | "Objects remaining in %s on __kmem_cache_shutdown()"); |
599870b1 | 3451 | } |
33b12c38 | 3452 | } |
52b4b950 | 3453 | spin_unlock_irq(&n->list_lock); |
81819f0f CL |
3454 | } |
3455 | ||
3456 | /* | |
672bba3a | 3457 | * Release all resources used by a slab cache. |
81819f0f | 3458 | */ |
52b4b950 | 3459 | int __kmem_cache_shutdown(struct kmem_cache *s) |
81819f0f CL |
3460 | { |
3461 | int node; | |
fa45dc25 | 3462 | struct kmem_cache_node *n; |
81819f0f CL |
3463 | |
3464 | flush_all(s); | |
81819f0f | 3465 | /* Attempt to free all objects */ |
fa45dc25 | 3466 | for_each_kmem_cache_node(s, node, n) { |
599870b1 CL |
3467 | free_partial(s, n); |
3468 | if (n->nr_partial || slabs_node(s, node)) | |
81819f0f CL |
3469 | return 1; |
3470 | } | |
81819f0f CL |
3471 | return 0; |
3472 | } | |
3473 | ||
81819f0f CL |
3474 | /******************************************************************** |
3475 | * Kmalloc subsystem | |
3476 | *******************************************************************/ | |
3477 | ||
81819f0f CL |
3478 | static int __init setup_slub_min_order(char *str) |
3479 | { | |
06428780 | 3480 | get_option(&str, &slub_min_order); |
81819f0f CL |
3481 | |
3482 | return 1; | |
3483 | } | |
3484 | ||
3485 | __setup("slub_min_order=", setup_slub_min_order); | |
3486 | ||
3487 | static int __init setup_slub_max_order(char *str) | |
3488 | { | |
06428780 | 3489 | get_option(&str, &slub_max_order); |
818cf590 | 3490 | slub_max_order = min(slub_max_order, MAX_ORDER - 1); |
81819f0f CL |
3491 | |
3492 | return 1; | |
3493 | } | |
3494 | ||
3495 | __setup("slub_max_order=", setup_slub_max_order); | |
3496 | ||
3497 | static int __init setup_slub_min_objects(char *str) | |
3498 | { | |
06428780 | 3499 | get_option(&str, &slub_min_objects); |
81819f0f CL |
3500 | |
3501 | return 1; | |
3502 | } | |
3503 | ||
3504 | __setup("slub_min_objects=", setup_slub_min_objects); | |
3505 | ||
81819f0f CL |
3506 | void *__kmalloc(size_t size, gfp_t flags) |
3507 | { | |
aadb4bc4 | 3508 | struct kmem_cache *s; |
5b882be4 | 3509 | void *ret; |
81819f0f | 3510 | |
95a05b42 | 3511 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) |
eada35ef | 3512 | return kmalloc_large(size, flags); |
aadb4bc4 | 3513 | |
2c59dd65 | 3514 | s = kmalloc_slab(size, flags); |
aadb4bc4 CL |
3515 | |
3516 | if (unlikely(ZERO_OR_NULL_PTR(s))) | |
6cb8f913 CL |
3517 | return s; |
3518 | ||
2b847c3c | 3519 | ret = slab_alloc(s, flags, _RET_IP_); |
5b882be4 | 3520 | |
ca2b84cb | 3521 | trace_kmalloc(_RET_IP_, ret, size, s->size, flags); |
5b882be4 | 3522 | |
0316bec2 AR |
3523 | kasan_kmalloc(s, ret, size); |
3524 | ||
5b882be4 | 3525 | return ret; |
81819f0f CL |
3526 | } |
3527 | EXPORT_SYMBOL(__kmalloc); | |
3528 | ||
5d1f57e4 | 3529 | #ifdef CONFIG_NUMA |
f619cfe1 CL |
3530 | static void *kmalloc_large_node(size_t size, gfp_t flags, int node) |
3531 | { | |
b1eeab67 | 3532 | struct page *page; |
e4f7c0b4 | 3533 | void *ptr = NULL; |
f619cfe1 | 3534 | |
52383431 VD |
3535 | flags |= __GFP_COMP | __GFP_NOTRACK; |
3536 | page = alloc_kmem_pages_node(node, flags, get_order(size)); | |
f619cfe1 | 3537 | if (page) |
e4f7c0b4 CM |
3538 | ptr = page_address(page); |
3539 | ||
d56791b3 | 3540 | kmalloc_large_node_hook(ptr, size, flags); |
e4f7c0b4 | 3541 | return ptr; |
f619cfe1 CL |
3542 | } |
3543 | ||
81819f0f CL |
3544 | void *__kmalloc_node(size_t size, gfp_t flags, int node) |
3545 | { | |
aadb4bc4 | 3546 | struct kmem_cache *s; |
5b882be4 | 3547 | void *ret; |
81819f0f | 3548 | |
95a05b42 | 3549 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) { |
5b882be4 EGM |
3550 | ret = kmalloc_large_node(size, flags, node); |
3551 | ||
ca2b84cb EGM |
3552 | trace_kmalloc_node(_RET_IP_, ret, |
3553 | size, PAGE_SIZE << get_order(size), | |
3554 | flags, node); | |
5b882be4 EGM |
3555 | |
3556 | return ret; | |
3557 | } | |
aadb4bc4 | 3558 | |
2c59dd65 | 3559 | s = kmalloc_slab(size, flags); |
aadb4bc4 CL |
3560 | |
3561 | if (unlikely(ZERO_OR_NULL_PTR(s))) | |
6cb8f913 CL |
3562 | return s; |
3563 | ||
2b847c3c | 3564 | ret = slab_alloc_node(s, flags, node, _RET_IP_); |
5b882be4 | 3565 | |
ca2b84cb | 3566 | trace_kmalloc_node(_RET_IP_, ret, size, s->size, flags, node); |
5b882be4 | 3567 | |
0316bec2 AR |
3568 | kasan_kmalloc(s, ret, size); |
3569 | ||
5b882be4 | 3570 | return ret; |
81819f0f CL |
3571 | } |
3572 | EXPORT_SYMBOL(__kmalloc_node); | |
3573 | #endif | |
3574 | ||
0316bec2 | 3575 | static size_t __ksize(const void *object) |
81819f0f | 3576 | { |
272c1d21 | 3577 | struct page *page; |
81819f0f | 3578 | |
ef8b4520 | 3579 | if (unlikely(object == ZERO_SIZE_PTR)) |
272c1d21 CL |
3580 | return 0; |
3581 | ||
294a80a8 | 3582 | page = virt_to_head_page(object); |
294a80a8 | 3583 | |
76994412 PE |
3584 | if (unlikely(!PageSlab(page))) { |
3585 | WARN_ON(!PageCompound(page)); | |
294a80a8 | 3586 | return PAGE_SIZE << compound_order(page); |
76994412 | 3587 | } |
81819f0f | 3588 | |
1b4f59e3 | 3589 | return slab_ksize(page->slab_cache); |
81819f0f | 3590 | } |
0316bec2 AR |
3591 | |
3592 | size_t ksize(const void *object) | |
3593 | { | |
3594 | size_t size = __ksize(object); | |
3595 | /* We assume that ksize callers could use whole allocated area, | |
3596 | so we need unpoison this area. */ | |
3597 | kasan_krealloc(object, size); | |
3598 | return size; | |
3599 | } | |
b1aabecd | 3600 | EXPORT_SYMBOL(ksize); |
81819f0f CL |
3601 | |
3602 | void kfree(const void *x) | |
3603 | { | |
81819f0f | 3604 | struct page *page; |
5bb983b0 | 3605 | void *object = (void *)x; |
81819f0f | 3606 | |
2121db74 PE |
3607 | trace_kfree(_RET_IP_, x); |
3608 | ||
2408c550 | 3609 | if (unlikely(ZERO_OR_NULL_PTR(x))) |
81819f0f CL |
3610 | return; |
3611 | ||
b49af68f | 3612 | page = virt_to_head_page(x); |
aadb4bc4 | 3613 | if (unlikely(!PageSlab(page))) { |
0937502a | 3614 | BUG_ON(!PageCompound(page)); |
d56791b3 | 3615 | kfree_hook(x); |
52383431 | 3616 | __free_kmem_pages(page, compound_order(page)); |
aadb4bc4 CL |
3617 | return; |
3618 | } | |
81084651 | 3619 | slab_free(page->slab_cache, page, object, NULL, 1, _RET_IP_); |
81819f0f CL |
3620 | } |
3621 | EXPORT_SYMBOL(kfree); | |
3622 | ||
832f37f5 VD |
3623 | #define SHRINK_PROMOTE_MAX 32 |
3624 | ||
2086d26a | 3625 | /* |
832f37f5 VD |
3626 | * kmem_cache_shrink discards empty slabs and promotes the slabs filled |
3627 | * up most to the head of the partial lists. New allocations will then | |
3628 | * fill those up and thus they can be removed from the partial lists. | |
672bba3a CL |
3629 | * |
3630 | * The slabs with the least items are placed last. This results in them | |
3631 | * being allocated from last increasing the chance that the last objects | |
3632 | * are freed in them. | |
2086d26a | 3633 | */ |
d6e0b7fa | 3634 | int __kmem_cache_shrink(struct kmem_cache *s, bool deactivate) |
2086d26a CL |
3635 | { |
3636 | int node; | |
3637 | int i; | |
3638 | struct kmem_cache_node *n; | |
3639 | struct page *page; | |
3640 | struct page *t; | |
832f37f5 VD |
3641 | struct list_head discard; |
3642 | struct list_head promote[SHRINK_PROMOTE_MAX]; | |
2086d26a | 3643 | unsigned long flags; |
ce3712d7 | 3644 | int ret = 0; |
2086d26a | 3645 | |
d6e0b7fa VD |
3646 | if (deactivate) { |
3647 | /* | |
3648 | * Disable empty slabs caching. Used to avoid pinning offline | |
3649 | * memory cgroups by kmem pages that can be freed. | |
3650 | */ | |
3651 | s->cpu_partial = 0; | |
3652 | s->min_partial = 0; | |
3653 | ||
3654 | /* | |
3655 | * s->cpu_partial is checked locklessly (see put_cpu_partial), | |
3656 | * so we have to make sure the change is visible. | |
3657 | */ | |
3658 | kick_all_cpus_sync(); | |
3659 | } | |
3660 | ||
2086d26a | 3661 | flush_all(s); |
fa45dc25 | 3662 | for_each_kmem_cache_node(s, node, n) { |
832f37f5 VD |
3663 | INIT_LIST_HEAD(&discard); |
3664 | for (i = 0; i < SHRINK_PROMOTE_MAX; i++) | |
3665 | INIT_LIST_HEAD(promote + i); | |
2086d26a CL |
3666 | |
3667 | spin_lock_irqsave(&n->list_lock, flags); | |
3668 | ||
3669 | /* | |
832f37f5 | 3670 | * Build lists of slabs to discard or promote. |
2086d26a | 3671 | * |
672bba3a CL |
3672 | * Note that concurrent frees may occur while we hold the |
3673 | * list_lock. page->inuse here is the upper limit. | |
2086d26a CL |
3674 | */ |
3675 | list_for_each_entry_safe(page, t, &n->partial, lru) { | |
832f37f5 VD |
3676 | int free = page->objects - page->inuse; |
3677 | ||
3678 | /* Do not reread page->inuse */ | |
3679 | barrier(); | |
3680 | ||
3681 | /* We do not keep full slabs on the list */ | |
3682 | BUG_ON(free <= 0); | |
3683 | ||
3684 | if (free == page->objects) { | |
3685 | list_move(&page->lru, &discard); | |
69cb8e6b | 3686 | n->nr_partial--; |
832f37f5 VD |
3687 | } else if (free <= SHRINK_PROMOTE_MAX) |
3688 | list_move(&page->lru, promote + free - 1); | |
2086d26a CL |
3689 | } |
3690 | ||
2086d26a | 3691 | /* |
832f37f5 VD |
3692 | * Promote the slabs filled up most to the head of the |
3693 | * partial list. | |
2086d26a | 3694 | */ |
832f37f5 VD |
3695 | for (i = SHRINK_PROMOTE_MAX - 1; i >= 0; i--) |
3696 | list_splice(promote + i, &n->partial); | |
2086d26a | 3697 | |
2086d26a | 3698 | spin_unlock_irqrestore(&n->list_lock, flags); |
69cb8e6b CL |
3699 | |
3700 | /* Release empty slabs */ | |
832f37f5 | 3701 | list_for_each_entry_safe(page, t, &discard, lru) |
69cb8e6b | 3702 | discard_slab(s, page); |
ce3712d7 VD |
3703 | |
3704 | if (slabs_node(s, node)) | |
3705 | ret = 1; | |
2086d26a CL |
3706 | } |
3707 | ||
ce3712d7 | 3708 | return ret; |
2086d26a | 3709 | } |
2086d26a | 3710 | |
b9049e23 YG |
3711 | static int slab_mem_going_offline_callback(void *arg) |
3712 | { | |
3713 | struct kmem_cache *s; | |
3714 | ||
18004c5d | 3715 | mutex_lock(&slab_mutex); |
b9049e23 | 3716 | list_for_each_entry(s, &slab_caches, list) |
d6e0b7fa | 3717 | __kmem_cache_shrink(s, false); |
18004c5d | 3718 | mutex_unlock(&slab_mutex); |
b9049e23 YG |
3719 | |
3720 | return 0; | |
3721 | } | |
3722 | ||
3723 | static void slab_mem_offline_callback(void *arg) | |
3724 | { | |
3725 | struct kmem_cache_node *n; | |
3726 | struct kmem_cache *s; | |
3727 | struct memory_notify *marg = arg; | |
3728 | int offline_node; | |
3729 | ||
b9d5ab25 | 3730 | offline_node = marg->status_change_nid_normal; |
b9049e23 YG |
3731 | |
3732 | /* | |
3733 | * If the node still has available memory. we need kmem_cache_node | |
3734 | * for it yet. | |
3735 | */ | |
3736 | if (offline_node < 0) | |
3737 | return; | |
3738 | ||
18004c5d | 3739 | mutex_lock(&slab_mutex); |
b9049e23 YG |
3740 | list_for_each_entry(s, &slab_caches, list) { |
3741 | n = get_node(s, offline_node); | |
3742 | if (n) { | |
3743 | /* | |
3744 | * if n->nr_slabs > 0, slabs still exist on the node | |
3745 | * that is going down. We were unable to free them, | |
c9404c9c | 3746 | * and offline_pages() function shouldn't call this |
b9049e23 YG |
3747 | * callback. So, we must fail. |
3748 | */ | |
0f389ec6 | 3749 | BUG_ON(slabs_node(s, offline_node)); |
b9049e23 YG |
3750 | |
3751 | s->node[offline_node] = NULL; | |
8de66a0c | 3752 | kmem_cache_free(kmem_cache_node, n); |
b9049e23 YG |
3753 | } |
3754 | } | |
18004c5d | 3755 | mutex_unlock(&slab_mutex); |
b9049e23 YG |
3756 | } |
3757 | ||
3758 | static int slab_mem_going_online_callback(void *arg) | |
3759 | { | |
3760 | struct kmem_cache_node *n; | |
3761 | struct kmem_cache *s; | |
3762 | struct memory_notify *marg = arg; | |
b9d5ab25 | 3763 | int nid = marg->status_change_nid_normal; |
b9049e23 YG |
3764 | int ret = 0; |
3765 | ||
3766 | /* | |
3767 | * If the node's memory is already available, then kmem_cache_node is | |
3768 | * already created. Nothing to do. | |
3769 | */ | |
3770 | if (nid < 0) | |
3771 | return 0; | |
3772 | ||
3773 | /* | |
0121c619 | 3774 | * We are bringing a node online. No memory is available yet. We must |
b9049e23 YG |
3775 | * allocate a kmem_cache_node structure in order to bring the node |
3776 | * online. | |
3777 | */ | |
18004c5d | 3778 | mutex_lock(&slab_mutex); |
b9049e23 YG |
3779 | list_for_each_entry(s, &slab_caches, list) { |
3780 | /* | |
3781 | * XXX: kmem_cache_alloc_node will fallback to other nodes | |
3782 | * since memory is not yet available from the node that | |
3783 | * is brought up. | |
3784 | */ | |
8de66a0c | 3785 | n = kmem_cache_alloc(kmem_cache_node, GFP_KERNEL); |
b9049e23 YG |
3786 | if (!n) { |
3787 | ret = -ENOMEM; | |
3788 | goto out; | |
3789 | } | |
4053497d | 3790 | init_kmem_cache_node(n); |
b9049e23 YG |
3791 | s->node[nid] = n; |
3792 | } | |
3793 | out: | |
18004c5d | 3794 | mutex_unlock(&slab_mutex); |
b9049e23 YG |
3795 | return ret; |
3796 | } | |
3797 | ||
3798 | static int slab_memory_callback(struct notifier_block *self, | |
3799 | unsigned long action, void *arg) | |
3800 | { | |
3801 | int ret = 0; | |
3802 | ||
3803 | switch (action) { | |
3804 | case MEM_GOING_ONLINE: | |
3805 | ret = slab_mem_going_online_callback(arg); | |
3806 | break; | |
3807 | case MEM_GOING_OFFLINE: | |
3808 | ret = slab_mem_going_offline_callback(arg); | |
3809 | break; | |
3810 | case MEM_OFFLINE: | |
3811 | case MEM_CANCEL_ONLINE: | |
3812 | slab_mem_offline_callback(arg); | |
3813 | break; | |
3814 | case MEM_ONLINE: | |
3815 | case MEM_CANCEL_OFFLINE: | |
3816 | break; | |
3817 | } | |
dc19f9db KH |
3818 | if (ret) |
3819 | ret = notifier_from_errno(ret); | |
3820 | else | |
3821 | ret = NOTIFY_OK; | |
b9049e23 YG |
3822 | return ret; |
3823 | } | |
3824 | ||
3ac38faa AM |
3825 | static struct notifier_block slab_memory_callback_nb = { |
3826 | .notifier_call = slab_memory_callback, | |
3827 | .priority = SLAB_CALLBACK_PRI, | |
3828 | }; | |
b9049e23 | 3829 | |
81819f0f CL |
3830 | /******************************************************************** |
3831 | * Basic setup of slabs | |
3832 | *******************************************************************/ | |
3833 | ||
51df1142 CL |
3834 | /* |
3835 | * Used for early kmem_cache structures that were allocated using | |
dffb4d60 CL |
3836 | * the page allocator. Allocate them properly then fix up the pointers |
3837 | * that may be pointing to the wrong kmem_cache structure. | |
51df1142 CL |
3838 | */ |
3839 | ||
dffb4d60 | 3840 | static struct kmem_cache * __init bootstrap(struct kmem_cache *static_cache) |
51df1142 CL |
3841 | { |
3842 | int node; | |
dffb4d60 | 3843 | struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT); |
fa45dc25 | 3844 | struct kmem_cache_node *n; |
51df1142 | 3845 | |
dffb4d60 | 3846 | memcpy(s, static_cache, kmem_cache->object_size); |
51df1142 | 3847 | |
7d557b3c GC |
3848 | /* |
3849 | * This runs very early, and only the boot processor is supposed to be | |
3850 | * up. Even if it weren't true, IRQs are not up so we couldn't fire | |
3851 | * IPIs around. | |
3852 | */ | |
3853 | __flush_cpu_slab(s, smp_processor_id()); | |
fa45dc25 | 3854 | for_each_kmem_cache_node(s, node, n) { |
51df1142 CL |
3855 | struct page *p; |
3856 | ||
fa45dc25 CL |
3857 | list_for_each_entry(p, &n->partial, lru) |
3858 | p->slab_cache = s; | |
51df1142 | 3859 | |
607bf324 | 3860 | #ifdef CONFIG_SLUB_DEBUG |
fa45dc25 CL |
3861 | list_for_each_entry(p, &n->full, lru) |
3862 | p->slab_cache = s; | |
51df1142 | 3863 | #endif |
51df1142 | 3864 | } |
f7ce3190 | 3865 | slab_init_memcg_params(s); |
dffb4d60 CL |
3866 | list_add(&s->list, &slab_caches); |
3867 | return s; | |
51df1142 CL |
3868 | } |
3869 | ||
81819f0f CL |
3870 | void __init kmem_cache_init(void) |
3871 | { | |
dffb4d60 CL |
3872 | static __initdata struct kmem_cache boot_kmem_cache, |
3873 | boot_kmem_cache_node; | |
51df1142 | 3874 | |
fc8d8620 SG |
3875 | if (debug_guardpage_minorder()) |
3876 | slub_max_order = 0; | |
3877 | ||
dffb4d60 CL |
3878 | kmem_cache_node = &boot_kmem_cache_node; |
3879 | kmem_cache = &boot_kmem_cache; | |
51df1142 | 3880 | |
dffb4d60 CL |
3881 | create_boot_cache(kmem_cache_node, "kmem_cache_node", |
3882 | sizeof(struct kmem_cache_node), SLAB_HWCACHE_ALIGN); | |
b9049e23 | 3883 | |
3ac38faa | 3884 | register_hotmemory_notifier(&slab_memory_callback_nb); |
81819f0f CL |
3885 | |
3886 | /* Able to allocate the per node structures */ | |
3887 | slab_state = PARTIAL; | |
3888 | ||
dffb4d60 CL |
3889 | create_boot_cache(kmem_cache, "kmem_cache", |
3890 | offsetof(struct kmem_cache, node) + | |
3891 | nr_node_ids * sizeof(struct kmem_cache_node *), | |
3892 | SLAB_HWCACHE_ALIGN); | |
8a13a4cc | 3893 | |
dffb4d60 | 3894 | kmem_cache = bootstrap(&boot_kmem_cache); |
81819f0f | 3895 | |
51df1142 CL |
3896 | /* |
3897 | * Allocate kmem_cache_node properly from the kmem_cache slab. | |
3898 | * kmem_cache_node is separately allocated so no need to | |
3899 | * update any list pointers. | |
3900 | */ | |
dffb4d60 | 3901 | kmem_cache_node = bootstrap(&boot_kmem_cache_node); |
51df1142 CL |
3902 | |
3903 | /* Now we can use the kmem_cache to allocate kmalloc slabs */ | |
34cc6990 | 3904 | setup_kmalloc_cache_index_table(); |
f97d5f63 | 3905 | create_kmalloc_caches(0); |
81819f0f CL |
3906 | |
3907 | #ifdef CONFIG_SMP | |
3908 | register_cpu_notifier(&slab_notifier); | |
9dfc6e68 | 3909 | #endif |
81819f0f | 3910 | |
f9f58285 | 3911 | pr_info("SLUB: HWalign=%d, Order=%d-%d, MinObjects=%d, CPUs=%d, Nodes=%d\n", |
f97d5f63 | 3912 | cache_line_size(), |
81819f0f CL |
3913 | slub_min_order, slub_max_order, slub_min_objects, |
3914 | nr_cpu_ids, nr_node_ids); | |
3915 | } | |
3916 | ||
7e85ee0c PE |
3917 | void __init kmem_cache_init_late(void) |
3918 | { | |
7e85ee0c PE |
3919 | } |
3920 | ||
2633d7a0 | 3921 | struct kmem_cache * |
a44cb944 VD |
3922 | __kmem_cache_alias(const char *name, size_t size, size_t align, |
3923 | unsigned long flags, void (*ctor)(void *)) | |
81819f0f | 3924 | { |
426589f5 | 3925 | struct kmem_cache *s, *c; |
81819f0f | 3926 | |
a44cb944 | 3927 | s = find_mergeable(size, align, flags, name, ctor); |
81819f0f CL |
3928 | if (s) { |
3929 | s->refcount++; | |
84d0ddd6 | 3930 | |
81819f0f CL |
3931 | /* |
3932 | * Adjust the object sizes so that we clear | |
3933 | * the complete object on kzalloc. | |
3934 | */ | |
3b0efdfa | 3935 | s->object_size = max(s->object_size, (int)size); |
81819f0f | 3936 | s->inuse = max_t(int, s->inuse, ALIGN(size, sizeof(void *))); |
6446faa2 | 3937 | |
426589f5 | 3938 | for_each_memcg_cache(c, s) { |
84d0ddd6 VD |
3939 | c->object_size = s->object_size; |
3940 | c->inuse = max_t(int, c->inuse, | |
3941 | ALIGN(size, sizeof(void *))); | |
3942 | } | |
3943 | ||
7b8f3b66 | 3944 | if (sysfs_slab_alias(s, name)) { |
7b8f3b66 | 3945 | s->refcount--; |
cbb79694 | 3946 | s = NULL; |
7b8f3b66 | 3947 | } |
a0e1d1be | 3948 | } |
6446faa2 | 3949 | |
cbb79694 CL |
3950 | return s; |
3951 | } | |
84c1cf62 | 3952 | |
8a13a4cc | 3953 | int __kmem_cache_create(struct kmem_cache *s, unsigned long flags) |
cbb79694 | 3954 | { |
aac3a166 PE |
3955 | int err; |
3956 | ||
3957 | err = kmem_cache_open(s, flags); | |
3958 | if (err) | |
3959 | return err; | |
20cea968 | 3960 | |
45530c44 CL |
3961 | /* Mutex is not taken during early boot */ |
3962 | if (slab_state <= UP) | |
3963 | return 0; | |
3964 | ||
107dab5c | 3965 | memcg_propagate_slab_attrs(s); |
aac3a166 | 3966 | err = sysfs_slab_add(s); |
aac3a166 | 3967 | if (err) |
52b4b950 | 3968 | __kmem_cache_release(s); |
20cea968 | 3969 | |
aac3a166 | 3970 | return err; |
81819f0f | 3971 | } |
81819f0f | 3972 | |
81819f0f | 3973 | #ifdef CONFIG_SMP |
81819f0f | 3974 | /* |
672bba3a CL |
3975 | * Use the cpu notifier to insure that the cpu slabs are flushed when |
3976 | * necessary. | |
81819f0f | 3977 | */ |
0db0628d | 3978 | static int slab_cpuup_callback(struct notifier_block *nfb, |
81819f0f CL |
3979 | unsigned long action, void *hcpu) |
3980 | { | |
3981 | long cpu = (long)hcpu; | |
5b95a4ac CL |
3982 | struct kmem_cache *s; |
3983 | unsigned long flags; | |
81819f0f CL |
3984 | |
3985 | switch (action) { | |
3986 | case CPU_UP_CANCELED: | |
8bb78442 | 3987 | case CPU_UP_CANCELED_FROZEN: |
81819f0f | 3988 | case CPU_DEAD: |
8bb78442 | 3989 | case CPU_DEAD_FROZEN: |
18004c5d | 3990 | mutex_lock(&slab_mutex); |
5b95a4ac CL |
3991 | list_for_each_entry(s, &slab_caches, list) { |
3992 | local_irq_save(flags); | |
3993 | __flush_cpu_slab(s, cpu); | |
3994 | local_irq_restore(flags); | |
3995 | } | |
18004c5d | 3996 | mutex_unlock(&slab_mutex); |
81819f0f CL |
3997 | break; |
3998 | default: | |
3999 | break; | |
4000 | } | |
4001 | return NOTIFY_OK; | |
4002 | } | |
4003 | ||
0db0628d | 4004 | static struct notifier_block slab_notifier = { |
3adbefee | 4005 | .notifier_call = slab_cpuup_callback |
06428780 | 4006 | }; |
81819f0f CL |
4007 | |
4008 | #endif | |
4009 | ||
ce71e27c | 4010 | void *__kmalloc_track_caller(size_t size, gfp_t gfpflags, unsigned long caller) |
81819f0f | 4011 | { |
aadb4bc4 | 4012 | struct kmem_cache *s; |
94b528d0 | 4013 | void *ret; |
aadb4bc4 | 4014 | |
95a05b42 | 4015 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) |
eada35ef PE |
4016 | return kmalloc_large(size, gfpflags); |
4017 | ||
2c59dd65 | 4018 | s = kmalloc_slab(size, gfpflags); |
81819f0f | 4019 | |
2408c550 | 4020 | if (unlikely(ZERO_OR_NULL_PTR(s))) |
6cb8f913 | 4021 | return s; |
81819f0f | 4022 | |
2b847c3c | 4023 | ret = slab_alloc(s, gfpflags, caller); |
94b528d0 | 4024 | |
25985edc | 4025 | /* Honor the call site pointer we received. */ |
ca2b84cb | 4026 | trace_kmalloc(caller, ret, size, s->size, gfpflags); |
94b528d0 EGM |
4027 | |
4028 | return ret; | |
81819f0f CL |
4029 | } |
4030 | ||
5d1f57e4 | 4031 | #ifdef CONFIG_NUMA |
81819f0f | 4032 | void *__kmalloc_node_track_caller(size_t size, gfp_t gfpflags, |
ce71e27c | 4033 | int node, unsigned long caller) |
81819f0f | 4034 | { |
aadb4bc4 | 4035 | struct kmem_cache *s; |
94b528d0 | 4036 | void *ret; |
aadb4bc4 | 4037 | |
95a05b42 | 4038 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) { |
d3e14aa3 XF |
4039 | ret = kmalloc_large_node(size, gfpflags, node); |
4040 | ||
4041 | trace_kmalloc_node(caller, ret, | |
4042 | size, PAGE_SIZE << get_order(size), | |
4043 | gfpflags, node); | |
4044 | ||
4045 | return ret; | |
4046 | } | |
eada35ef | 4047 | |
2c59dd65 | 4048 | s = kmalloc_slab(size, gfpflags); |
81819f0f | 4049 | |
2408c550 | 4050 | if (unlikely(ZERO_OR_NULL_PTR(s))) |
6cb8f913 | 4051 | return s; |
81819f0f | 4052 | |
2b847c3c | 4053 | ret = slab_alloc_node(s, gfpflags, node, caller); |
94b528d0 | 4054 | |
25985edc | 4055 | /* Honor the call site pointer we received. */ |
ca2b84cb | 4056 | trace_kmalloc_node(caller, ret, size, s->size, gfpflags, node); |
94b528d0 EGM |
4057 | |
4058 | return ret; | |
81819f0f | 4059 | } |
5d1f57e4 | 4060 | #endif |
81819f0f | 4061 | |
ab4d5ed5 | 4062 | #ifdef CONFIG_SYSFS |
205ab99d CL |
4063 | static int count_inuse(struct page *page) |
4064 | { | |
4065 | return page->inuse; | |
4066 | } | |
4067 | ||
4068 | static int count_total(struct page *page) | |
4069 | { | |
4070 | return page->objects; | |
4071 | } | |
ab4d5ed5 | 4072 | #endif |
205ab99d | 4073 | |
ab4d5ed5 | 4074 | #ifdef CONFIG_SLUB_DEBUG |
434e245d CL |
4075 | static int validate_slab(struct kmem_cache *s, struct page *page, |
4076 | unsigned long *map) | |
53e15af0 CL |
4077 | { |
4078 | void *p; | |
a973e9dd | 4079 | void *addr = page_address(page); |
53e15af0 CL |
4080 | |
4081 | if (!check_slab(s, page) || | |
4082 | !on_freelist(s, page, NULL)) | |
4083 | return 0; | |
4084 | ||
4085 | /* Now we know that a valid freelist exists */ | |
39b26464 | 4086 | bitmap_zero(map, page->objects); |
53e15af0 | 4087 | |
5f80b13a CL |
4088 | get_map(s, page, map); |
4089 | for_each_object(p, s, addr, page->objects) { | |
4090 | if (test_bit(slab_index(p, s, addr), map)) | |
4091 | if (!check_object(s, page, p, SLUB_RED_INACTIVE)) | |
4092 | return 0; | |
53e15af0 CL |
4093 | } |
4094 | ||
224a88be | 4095 | for_each_object(p, s, addr, page->objects) |
7656c72b | 4096 | if (!test_bit(slab_index(p, s, addr), map)) |
37d57443 | 4097 | if (!check_object(s, page, p, SLUB_RED_ACTIVE)) |
53e15af0 CL |
4098 | return 0; |
4099 | return 1; | |
4100 | } | |
4101 | ||
434e245d CL |
4102 | static void validate_slab_slab(struct kmem_cache *s, struct page *page, |
4103 | unsigned long *map) | |
53e15af0 | 4104 | { |
881db7fb CL |
4105 | slab_lock(page); |
4106 | validate_slab(s, page, map); | |
4107 | slab_unlock(page); | |
53e15af0 CL |
4108 | } |
4109 | ||
434e245d CL |
4110 | static int validate_slab_node(struct kmem_cache *s, |
4111 | struct kmem_cache_node *n, unsigned long *map) | |
53e15af0 CL |
4112 | { |
4113 | unsigned long count = 0; | |
4114 | struct page *page; | |
4115 | unsigned long flags; | |
4116 | ||
4117 | spin_lock_irqsave(&n->list_lock, flags); | |
4118 | ||
4119 | list_for_each_entry(page, &n->partial, lru) { | |
434e245d | 4120 | validate_slab_slab(s, page, map); |
53e15af0 CL |
4121 | count++; |
4122 | } | |
4123 | if (count != n->nr_partial) | |
f9f58285 FF |
4124 | pr_err("SLUB %s: %ld partial slabs counted but counter=%ld\n", |
4125 | s->name, count, n->nr_partial); | |
53e15af0 CL |
4126 | |
4127 | if (!(s->flags & SLAB_STORE_USER)) | |
4128 | goto out; | |
4129 | ||
4130 | list_for_each_entry(page, &n->full, lru) { | |
434e245d | 4131 | validate_slab_slab(s, page, map); |
53e15af0 CL |
4132 | count++; |
4133 | } | |
4134 | if (count != atomic_long_read(&n->nr_slabs)) | |
f9f58285 FF |
4135 | pr_err("SLUB: %s %ld slabs counted but counter=%ld\n", |
4136 | s->name, count, atomic_long_read(&n->nr_slabs)); | |
53e15af0 CL |
4137 | |
4138 | out: | |
4139 | spin_unlock_irqrestore(&n->list_lock, flags); | |
4140 | return count; | |
4141 | } | |
4142 | ||
434e245d | 4143 | static long validate_slab_cache(struct kmem_cache *s) |
53e15af0 CL |
4144 | { |
4145 | int node; | |
4146 | unsigned long count = 0; | |
205ab99d | 4147 | unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) * |
434e245d | 4148 | sizeof(unsigned long), GFP_KERNEL); |
fa45dc25 | 4149 | struct kmem_cache_node *n; |
434e245d CL |
4150 | |
4151 | if (!map) | |
4152 | return -ENOMEM; | |
53e15af0 CL |
4153 | |
4154 | flush_all(s); | |
fa45dc25 | 4155 | for_each_kmem_cache_node(s, node, n) |
434e245d | 4156 | count += validate_slab_node(s, n, map); |
434e245d | 4157 | kfree(map); |
53e15af0 CL |
4158 | return count; |
4159 | } | |
88a420e4 | 4160 | /* |
672bba3a | 4161 | * Generate lists of code addresses where slabcache objects are allocated |
88a420e4 CL |
4162 | * and freed. |
4163 | */ | |
4164 | ||
4165 | struct location { | |
4166 | unsigned long count; | |
ce71e27c | 4167 | unsigned long addr; |
45edfa58 CL |
4168 | long long sum_time; |
4169 | long min_time; | |
4170 | long max_time; | |
4171 | long min_pid; | |
4172 | long max_pid; | |
174596a0 | 4173 | DECLARE_BITMAP(cpus, NR_CPUS); |
45edfa58 | 4174 | nodemask_t nodes; |
88a420e4 CL |
4175 | }; |
4176 | ||
4177 | struct loc_track { | |
4178 | unsigned long max; | |
4179 | unsigned long count; | |
4180 | struct location *loc; | |
4181 | }; | |
4182 | ||
4183 | static void free_loc_track(struct loc_track *t) | |
4184 | { | |
4185 | if (t->max) | |
4186 | free_pages((unsigned long)t->loc, | |
4187 | get_order(sizeof(struct location) * t->max)); | |
4188 | } | |
4189 | ||
68dff6a9 | 4190 | static int alloc_loc_track(struct loc_track *t, unsigned long max, gfp_t flags) |
88a420e4 CL |
4191 | { |
4192 | struct location *l; | |
4193 | int order; | |
4194 | ||
88a420e4 CL |
4195 | order = get_order(sizeof(struct location) * max); |
4196 | ||
68dff6a9 | 4197 | l = (void *)__get_free_pages(flags, order); |
88a420e4 CL |
4198 | if (!l) |
4199 | return 0; | |
4200 | ||
4201 | if (t->count) { | |
4202 | memcpy(l, t->loc, sizeof(struct location) * t->count); | |
4203 | free_loc_track(t); | |
4204 | } | |
4205 | t->max = max; | |
4206 | t->loc = l; | |
4207 | return 1; | |
4208 | } | |
4209 | ||
4210 | static int add_location(struct loc_track *t, struct kmem_cache *s, | |
45edfa58 | 4211 | const struct track *track) |
88a420e4 CL |
4212 | { |
4213 | long start, end, pos; | |
4214 | struct location *l; | |
ce71e27c | 4215 | unsigned long caddr; |
45edfa58 | 4216 | unsigned long age = jiffies - track->when; |
88a420e4 CL |
4217 | |
4218 | start = -1; | |
4219 | end = t->count; | |
4220 | ||
4221 | for ( ; ; ) { | |
4222 | pos = start + (end - start + 1) / 2; | |
4223 | ||
4224 | /* | |
4225 | * There is nothing at "end". If we end up there | |
4226 | * we need to add something to before end. | |
4227 | */ | |
4228 | if (pos == end) | |
4229 | break; | |
4230 | ||
4231 | caddr = t->loc[pos].addr; | |
45edfa58 CL |
4232 | if (track->addr == caddr) { |
4233 | ||
4234 | l = &t->loc[pos]; | |
4235 | l->count++; | |
4236 | if (track->when) { | |
4237 | l->sum_time += age; | |
4238 | if (age < l->min_time) | |
4239 | l->min_time = age; | |
4240 | if (age > l->max_time) | |
4241 | l->max_time = age; | |
4242 | ||
4243 | if (track->pid < l->min_pid) | |
4244 | l->min_pid = track->pid; | |
4245 | if (track->pid > l->max_pid) | |
4246 | l->max_pid = track->pid; | |
4247 | ||
174596a0 RR |
4248 | cpumask_set_cpu(track->cpu, |
4249 | to_cpumask(l->cpus)); | |
45edfa58 CL |
4250 | } |
4251 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
88a420e4 CL |
4252 | return 1; |
4253 | } | |
4254 | ||
45edfa58 | 4255 | if (track->addr < caddr) |
88a420e4 CL |
4256 | end = pos; |
4257 | else | |
4258 | start = pos; | |
4259 | } | |
4260 | ||
4261 | /* | |
672bba3a | 4262 | * Not found. Insert new tracking element. |
88a420e4 | 4263 | */ |
68dff6a9 | 4264 | if (t->count >= t->max && !alloc_loc_track(t, 2 * t->max, GFP_ATOMIC)) |
88a420e4 CL |
4265 | return 0; |
4266 | ||
4267 | l = t->loc + pos; | |
4268 | if (pos < t->count) | |
4269 | memmove(l + 1, l, | |
4270 | (t->count - pos) * sizeof(struct location)); | |
4271 | t->count++; | |
4272 | l->count = 1; | |
45edfa58 CL |
4273 | l->addr = track->addr; |
4274 | l->sum_time = age; | |
4275 | l->min_time = age; | |
4276 | l->max_time = age; | |
4277 | l->min_pid = track->pid; | |
4278 | l->max_pid = track->pid; | |
174596a0 RR |
4279 | cpumask_clear(to_cpumask(l->cpus)); |
4280 | cpumask_set_cpu(track->cpu, to_cpumask(l->cpus)); | |
45edfa58 CL |
4281 | nodes_clear(l->nodes); |
4282 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
88a420e4 CL |
4283 | return 1; |
4284 | } | |
4285 | ||
4286 | static void process_slab(struct loc_track *t, struct kmem_cache *s, | |
bbd7d57b | 4287 | struct page *page, enum track_item alloc, |
a5dd5c11 | 4288 | unsigned long *map) |
88a420e4 | 4289 | { |
a973e9dd | 4290 | void *addr = page_address(page); |
88a420e4 CL |
4291 | void *p; |
4292 | ||
39b26464 | 4293 | bitmap_zero(map, page->objects); |
5f80b13a | 4294 | get_map(s, page, map); |
88a420e4 | 4295 | |
224a88be | 4296 | for_each_object(p, s, addr, page->objects) |
45edfa58 CL |
4297 | if (!test_bit(slab_index(p, s, addr), map)) |
4298 | add_location(t, s, get_track(s, p, alloc)); | |
88a420e4 CL |
4299 | } |
4300 | ||
4301 | static int list_locations(struct kmem_cache *s, char *buf, | |
4302 | enum track_item alloc) | |
4303 | { | |
e374d483 | 4304 | int len = 0; |
88a420e4 | 4305 | unsigned long i; |
68dff6a9 | 4306 | struct loc_track t = { 0, 0, NULL }; |
88a420e4 | 4307 | int node; |
bbd7d57b ED |
4308 | unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) * |
4309 | sizeof(unsigned long), GFP_KERNEL); | |
fa45dc25 | 4310 | struct kmem_cache_node *n; |
88a420e4 | 4311 | |
bbd7d57b ED |
4312 | if (!map || !alloc_loc_track(&t, PAGE_SIZE / sizeof(struct location), |
4313 | GFP_TEMPORARY)) { | |
4314 | kfree(map); | |
68dff6a9 | 4315 | return sprintf(buf, "Out of memory\n"); |
bbd7d57b | 4316 | } |
88a420e4 CL |
4317 | /* Push back cpu slabs */ |
4318 | flush_all(s); | |
4319 | ||
fa45dc25 | 4320 | for_each_kmem_cache_node(s, node, n) { |
88a420e4 CL |
4321 | unsigned long flags; |
4322 | struct page *page; | |
4323 | ||
9e86943b | 4324 | if (!atomic_long_read(&n->nr_slabs)) |
88a420e4 CL |
4325 | continue; |
4326 | ||
4327 | spin_lock_irqsave(&n->list_lock, flags); | |
4328 | list_for_each_entry(page, &n->partial, lru) | |
bbd7d57b | 4329 | process_slab(&t, s, page, alloc, map); |
88a420e4 | 4330 | list_for_each_entry(page, &n->full, lru) |
bbd7d57b | 4331 | process_slab(&t, s, page, alloc, map); |
88a420e4 CL |
4332 | spin_unlock_irqrestore(&n->list_lock, flags); |
4333 | } | |
4334 | ||
4335 | for (i = 0; i < t.count; i++) { | |
45edfa58 | 4336 | struct location *l = &t.loc[i]; |
88a420e4 | 4337 | |
9c246247 | 4338 | if (len > PAGE_SIZE - KSYM_SYMBOL_LEN - 100) |
88a420e4 | 4339 | break; |
e374d483 | 4340 | len += sprintf(buf + len, "%7ld ", l->count); |
45edfa58 CL |
4341 | |
4342 | if (l->addr) | |
62c70bce | 4343 | len += sprintf(buf + len, "%pS", (void *)l->addr); |
88a420e4 | 4344 | else |
e374d483 | 4345 | len += sprintf(buf + len, "<not-available>"); |
45edfa58 CL |
4346 | |
4347 | if (l->sum_time != l->min_time) { | |
e374d483 | 4348 | len += sprintf(buf + len, " age=%ld/%ld/%ld", |
f8bd2258 RZ |
4349 | l->min_time, |
4350 | (long)div_u64(l->sum_time, l->count), | |
4351 | l->max_time); | |
45edfa58 | 4352 | } else |
e374d483 | 4353 | len += sprintf(buf + len, " age=%ld", |
45edfa58 CL |
4354 | l->min_time); |
4355 | ||
4356 | if (l->min_pid != l->max_pid) | |
e374d483 | 4357 | len += sprintf(buf + len, " pid=%ld-%ld", |
45edfa58 CL |
4358 | l->min_pid, l->max_pid); |
4359 | else | |
e374d483 | 4360 | len += sprintf(buf + len, " pid=%ld", |
45edfa58 CL |
4361 | l->min_pid); |
4362 | ||
174596a0 RR |
4363 | if (num_online_cpus() > 1 && |
4364 | !cpumask_empty(to_cpumask(l->cpus)) && | |
5024c1d7 TH |
4365 | len < PAGE_SIZE - 60) |
4366 | len += scnprintf(buf + len, PAGE_SIZE - len - 50, | |
4367 | " cpus=%*pbl", | |
4368 | cpumask_pr_args(to_cpumask(l->cpus))); | |
45edfa58 | 4369 | |
62bc62a8 | 4370 | if (nr_online_nodes > 1 && !nodes_empty(l->nodes) && |
5024c1d7 TH |
4371 | len < PAGE_SIZE - 60) |
4372 | len += scnprintf(buf + len, PAGE_SIZE - len - 50, | |
4373 | " nodes=%*pbl", | |
4374 | nodemask_pr_args(&l->nodes)); | |
45edfa58 | 4375 | |
e374d483 | 4376 | len += sprintf(buf + len, "\n"); |
88a420e4 CL |
4377 | } |
4378 | ||
4379 | free_loc_track(&t); | |
bbd7d57b | 4380 | kfree(map); |
88a420e4 | 4381 | if (!t.count) |
e374d483 HH |
4382 | len += sprintf(buf, "No data\n"); |
4383 | return len; | |
88a420e4 | 4384 | } |
ab4d5ed5 | 4385 | #endif |
88a420e4 | 4386 | |
a5a84755 | 4387 | #ifdef SLUB_RESILIENCY_TEST |
c07b8183 | 4388 | static void __init resiliency_test(void) |
a5a84755 CL |
4389 | { |
4390 | u8 *p; | |
4391 | ||
95a05b42 | 4392 | BUILD_BUG_ON(KMALLOC_MIN_SIZE > 16 || KMALLOC_SHIFT_HIGH < 10); |
a5a84755 | 4393 | |
f9f58285 FF |
4394 | pr_err("SLUB resiliency testing\n"); |
4395 | pr_err("-----------------------\n"); | |
4396 | pr_err("A. Corruption after allocation\n"); | |
a5a84755 CL |
4397 | |
4398 | p = kzalloc(16, GFP_KERNEL); | |
4399 | p[16] = 0x12; | |
f9f58285 FF |
4400 | pr_err("\n1. kmalloc-16: Clobber Redzone/next pointer 0x12->0x%p\n\n", |
4401 | p + 16); | |
a5a84755 CL |
4402 | |
4403 | validate_slab_cache(kmalloc_caches[4]); | |
4404 | ||
4405 | /* Hmmm... The next two are dangerous */ | |
4406 | p = kzalloc(32, GFP_KERNEL); | |
4407 | p[32 + sizeof(void *)] = 0x34; | |
f9f58285 FF |
4408 | pr_err("\n2. kmalloc-32: Clobber next pointer/next slab 0x34 -> -0x%p\n", |
4409 | p); | |
4410 | pr_err("If allocated object is overwritten then not detectable\n\n"); | |
a5a84755 CL |
4411 | |
4412 | validate_slab_cache(kmalloc_caches[5]); | |
4413 | p = kzalloc(64, GFP_KERNEL); | |
4414 | p += 64 + (get_cycles() & 0xff) * sizeof(void *); | |
4415 | *p = 0x56; | |
f9f58285 FF |
4416 | pr_err("\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n", |
4417 | p); | |
4418 | pr_err("If allocated object is overwritten then not detectable\n\n"); | |
a5a84755 CL |
4419 | validate_slab_cache(kmalloc_caches[6]); |
4420 | ||
f9f58285 | 4421 | pr_err("\nB. Corruption after free\n"); |
a5a84755 CL |
4422 | p = kzalloc(128, GFP_KERNEL); |
4423 | kfree(p); | |
4424 | *p = 0x78; | |
f9f58285 | 4425 | pr_err("1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p); |
a5a84755 CL |
4426 | validate_slab_cache(kmalloc_caches[7]); |
4427 | ||
4428 | p = kzalloc(256, GFP_KERNEL); | |
4429 | kfree(p); | |
4430 | p[50] = 0x9a; | |
f9f58285 | 4431 | pr_err("\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n", p); |
a5a84755 CL |
4432 | validate_slab_cache(kmalloc_caches[8]); |
4433 | ||
4434 | p = kzalloc(512, GFP_KERNEL); | |
4435 | kfree(p); | |
4436 | p[512] = 0xab; | |
f9f58285 | 4437 | pr_err("\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p); |
a5a84755 CL |
4438 | validate_slab_cache(kmalloc_caches[9]); |
4439 | } | |
4440 | #else | |
4441 | #ifdef CONFIG_SYSFS | |
4442 | static void resiliency_test(void) {}; | |
4443 | #endif | |
4444 | #endif | |
4445 | ||
ab4d5ed5 | 4446 | #ifdef CONFIG_SYSFS |
81819f0f | 4447 | enum slab_stat_type { |
205ab99d CL |
4448 | SL_ALL, /* All slabs */ |
4449 | SL_PARTIAL, /* Only partially allocated slabs */ | |
4450 | SL_CPU, /* Only slabs used for cpu caches */ | |
4451 | SL_OBJECTS, /* Determine allocated objects not slabs */ | |
4452 | SL_TOTAL /* Determine object capacity not slabs */ | |
81819f0f CL |
4453 | }; |
4454 | ||
205ab99d | 4455 | #define SO_ALL (1 << SL_ALL) |
81819f0f CL |
4456 | #define SO_PARTIAL (1 << SL_PARTIAL) |
4457 | #define SO_CPU (1 << SL_CPU) | |
4458 | #define SO_OBJECTS (1 << SL_OBJECTS) | |
205ab99d | 4459 | #define SO_TOTAL (1 << SL_TOTAL) |
81819f0f | 4460 | |
62e5c4b4 CG |
4461 | static ssize_t show_slab_objects(struct kmem_cache *s, |
4462 | char *buf, unsigned long flags) | |
81819f0f CL |
4463 | { |
4464 | unsigned long total = 0; | |
81819f0f CL |
4465 | int node; |
4466 | int x; | |
4467 | unsigned long *nodes; | |
81819f0f | 4468 | |
e35e1a97 | 4469 | nodes = kzalloc(sizeof(unsigned long) * nr_node_ids, GFP_KERNEL); |
62e5c4b4 CG |
4470 | if (!nodes) |
4471 | return -ENOMEM; | |
81819f0f | 4472 | |
205ab99d CL |
4473 | if (flags & SO_CPU) { |
4474 | int cpu; | |
81819f0f | 4475 | |
205ab99d | 4476 | for_each_possible_cpu(cpu) { |
d0e0ac97 CG |
4477 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, |
4478 | cpu); | |
ec3ab083 | 4479 | int node; |
49e22585 | 4480 | struct page *page; |
dfb4f096 | 4481 | |
4db0c3c2 | 4482 | page = READ_ONCE(c->page); |
ec3ab083 CL |
4483 | if (!page) |
4484 | continue; | |
205ab99d | 4485 | |
ec3ab083 CL |
4486 | node = page_to_nid(page); |
4487 | if (flags & SO_TOTAL) | |
4488 | x = page->objects; | |
4489 | else if (flags & SO_OBJECTS) | |
4490 | x = page->inuse; | |
4491 | else | |
4492 | x = 1; | |
49e22585 | 4493 | |
ec3ab083 CL |
4494 | total += x; |
4495 | nodes[node] += x; | |
4496 | ||
4db0c3c2 | 4497 | page = READ_ONCE(c->partial); |
49e22585 | 4498 | if (page) { |
8afb1474 LZ |
4499 | node = page_to_nid(page); |
4500 | if (flags & SO_TOTAL) | |
4501 | WARN_ON_ONCE(1); | |
4502 | else if (flags & SO_OBJECTS) | |
4503 | WARN_ON_ONCE(1); | |
4504 | else | |
4505 | x = page->pages; | |
bc6697d8 ED |
4506 | total += x; |
4507 | nodes[node] += x; | |
49e22585 | 4508 | } |
81819f0f CL |
4509 | } |
4510 | } | |
4511 | ||
bfc8c901 | 4512 | get_online_mems(); |
ab4d5ed5 | 4513 | #ifdef CONFIG_SLUB_DEBUG |
205ab99d | 4514 | if (flags & SO_ALL) { |
fa45dc25 CL |
4515 | struct kmem_cache_node *n; |
4516 | ||
4517 | for_each_kmem_cache_node(s, node, n) { | |
205ab99d | 4518 | |
d0e0ac97 CG |
4519 | if (flags & SO_TOTAL) |
4520 | x = atomic_long_read(&n->total_objects); | |
4521 | else if (flags & SO_OBJECTS) | |
4522 | x = atomic_long_read(&n->total_objects) - | |
4523 | count_partial(n, count_free); | |
81819f0f | 4524 | else |
205ab99d | 4525 | x = atomic_long_read(&n->nr_slabs); |
81819f0f CL |
4526 | total += x; |
4527 | nodes[node] += x; | |
4528 | } | |
4529 | ||
ab4d5ed5 CL |
4530 | } else |
4531 | #endif | |
4532 | if (flags & SO_PARTIAL) { | |
fa45dc25 | 4533 | struct kmem_cache_node *n; |
81819f0f | 4534 | |
fa45dc25 | 4535 | for_each_kmem_cache_node(s, node, n) { |
205ab99d CL |
4536 | if (flags & SO_TOTAL) |
4537 | x = count_partial(n, count_total); | |
4538 | else if (flags & SO_OBJECTS) | |
4539 | x = count_partial(n, count_inuse); | |
81819f0f | 4540 | else |
205ab99d | 4541 | x = n->nr_partial; |
81819f0f CL |
4542 | total += x; |
4543 | nodes[node] += x; | |
4544 | } | |
4545 | } | |
81819f0f CL |
4546 | x = sprintf(buf, "%lu", total); |
4547 | #ifdef CONFIG_NUMA | |
fa45dc25 | 4548 | for (node = 0; node < nr_node_ids; node++) |
81819f0f CL |
4549 | if (nodes[node]) |
4550 | x += sprintf(buf + x, " N%d=%lu", | |
4551 | node, nodes[node]); | |
4552 | #endif | |
bfc8c901 | 4553 | put_online_mems(); |
81819f0f CL |
4554 | kfree(nodes); |
4555 | return x + sprintf(buf + x, "\n"); | |
4556 | } | |
4557 | ||
ab4d5ed5 | 4558 | #ifdef CONFIG_SLUB_DEBUG |
81819f0f CL |
4559 | static int any_slab_objects(struct kmem_cache *s) |
4560 | { | |
4561 | int node; | |
fa45dc25 | 4562 | struct kmem_cache_node *n; |
81819f0f | 4563 | |
fa45dc25 | 4564 | for_each_kmem_cache_node(s, node, n) |
4ea33e2d | 4565 | if (atomic_long_read(&n->total_objects)) |
81819f0f | 4566 | return 1; |
fa45dc25 | 4567 | |
81819f0f CL |
4568 | return 0; |
4569 | } | |
ab4d5ed5 | 4570 | #endif |
81819f0f CL |
4571 | |
4572 | #define to_slab_attr(n) container_of(n, struct slab_attribute, attr) | |
497888cf | 4573 | #define to_slab(n) container_of(n, struct kmem_cache, kobj) |
81819f0f CL |
4574 | |
4575 | struct slab_attribute { | |
4576 | struct attribute attr; | |
4577 | ssize_t (*show)(struct kmem_cache *s, char *buf); | |
4578 | ssize_t (*store)(struct kmem_cache *s, const char *x, size_t count); | |
4579 | }; | |
4580 | ||
4581 | #define SLAB_ATTR_RO(_name) \ | |
ab067e99 VK |
4582 | static struct slab_attribute _name##_attr = \ |
4583 | __ATTR(_name, 0400, _name##_show, NULL) | |
81819f0f CL |
4584 | |
4585 | #define SLAB_ATTR(_name) \ | |
4586 | static struct slab_attribute _name##_attr = \ | |
ab067e99 | 4587 | __ATTR(_name, 0600, _name##_show, _name##_store) |
81819f0f | 4588 | |
81819f0f CL |
4589 | static ssize_t slab_size_show(struct kmem_cache *s, char *buf) |
4590 | { | |
4591 | return sprintf(buf, "%d\n", s->size); | |
4592 | } | |
4593 | SLAB_ATTR_RO(slab_size); | |
4594 | ||
4595 | static ssize_t align_show(struct kmem_cache *s, char *buf) | |
4596 | { | |
4597 | return sprintf(buf, "%d\n", s->align); | |
4598 | } | |
4599 | SLAB_ATTR_RO(align); | |
4600 | ||
4601 | static ssize_t object_size_show(struct kmem_cache *s, char *buf) | |
4602 | { | |
3b0efdfa | 4603 | return sprintf(buf, "%d\n", s->object_size); |
81819f0f CL |
4604 | } |
4605 | SLAB_ATTR_RO(object_size); | |
4606 | ||
4607 | static ssize_t objs_per_slab_show(struct kmem_cache *s, char *buf) | |
4608 | { | |
834f3d11 | 4609 | return sprintf(buf, "%d\n", oo_objects(s->oo)); |
81819f0f CL |
4610 | } |
4611 | SLAB_ATTR_RO(objs_per_slab); | |
4612 | ||
06b285dc CL |
4613 | static ssize_t order_store(struct kmem_cache *s, |
4614 | const char *buf, size_t length) | |
4615 | { | |
0121c619 CL |
4616 | unsigned long order; |
4617 | int err; | |
4618 | ||
3dbb95f7 | 4619 | err = kstrtoul(buf, 10, &order); |
0121c619 CL |
4620 | if (err) |
4621 | return err; | |
06b285dc CL |
4622 | |
4623 | if (order > slub_max_order || order < slub_min_order) | |
4624 | return -EINVAL; | |
4625 | ||
4626 | calculate_sizes(s, order); | |
4627 | return length; | |
4628 | } | |
4629 | ||
81819f0f CL |
4630 | static ssize_t order_show(struct kmem_cache *s, char *buf) |
4631 | { | |
834f3d11 | 4632 | return sprintf(buf, "%d\n", oo_order(s->oo)); |
81819f0f | 4633 | } |
06b285dc | 4634 | SLAB_ATTR(order); |
81819f0f | 4635 | |
73d342b1 DR |
4636 | static ssize_t min_partial_show(struct kmem_cache *s, char *buf) |
4637 | { | |
4638 | return sprintf(buf, "%lu\n", s->min_partial); | |
4639 | } | |
4640 | ||
4641 | static ssize_t min_partial_store(struct kmem_cache *s, const char *buf, | |
4642 | size_t length) | |
4643 | { | |
4644 | unsigned long min; | |
4645 | int err; | |
4646 | ||
3dbb95f7 | 4647 | err = kstrtoul(buf, 10, &min); |
73d342b1 DR |
4648 | if (err) |
4649 | return err; | |
4650 | ||
c0bdb232 | 4651 | set_min_partial(s, min); |
73d342b1 DR |
4652 | return length; |
4653 | } | |
4654 | SLAB_ATTR(min_partial); | |
4655 | ||
49e22585 CL |
4656 | static ssize_t cpu_partial_show(struct kmem_cache *s, char *buf) |
4657 | { | |
4658 | return sprintf(buf, "%u\n", s->cpu_partial); | |
4659 | } | |
4660 | ||
4661 | static ssize_t cpu_partial_store(struct kmem_cache *s, const char *buf, | |
4662 | size_t length) | |
4663 | { | |
4664 | unsigned long objects; | |
4665 | int err; | |
4666 | ||
3dbb95f7 | 4667 | err = kstrtoul(buf, 10, &objects); |
49e22585 CL |
4668 | if (err) |
4669 | return err; | |
345c905d | 4670 | if (objects && !kmem_cache_has_cpu_partial(s)) |
74ee4ef1 | 4671 | return -EINVAL; |
49e22585 CL |
4672 | |
4673 | s->cpu_partial = objects; | |
4674 | flush_all(s); | |
4675 | return length; | |
4676 | } | |
4677 | SLAB_ATTR(cpu_partial); | |
4678 | ||
81819f0f CL |
4679 | static ssize_t ctor_show(struct kmem_cache *s, char *buf) |
4680 | { | |
62c70bce JP |
4681 | if (!s->ctor) |
4682 | return 0; | |
4683 | return sprintf(buf, "%pS\n", s->ctor); | |
81819f0f CL |
4684 | } |
4685 | SLAB_ATTR_RO(ctor); | |
4686 | ||
81819f0f CL |
4687 | static ssize_t aliases_show(struct kmem_cache *s, char *buf) |
4688 | { | |
4307c14f | 4689 | return sprintf(buf, "%d\n", s->refcount < 0 ? 0 : s->refcount - 1); |
81819f0f CL |
4690 | } |
4691 | SLAB_ATTR_RO(aliases); | |
4692 | ||
81819f0f CL |
4693 | static ssize_t partial_show(struct kmem_cache *s, char *buf) |
4694 | { | |
d9acf4b7 | 4695 | return show_slab_objects(s, buf, SO_PARTIAL); |
81819f0f CL |
4696 | } |
4697 | SLAB_ATTR_RO(partial); | |
4698 | ||
4699 | static ssize_t cpu_slabs_show(struct kmem_cache *s, char *buf) | |
4700 | { | |
d9acf4b7 | 4701 | return show_slab_objects(s, buf, SO_CPU); |
81819f0f CL |
4702 | } |
4703 | SLAB_ATTR_RO(cpu_slabs); | |
4704 | ||
4705 | static ssize_t objects_show(struct kmem_cache *s, char *buf) | |
4706 | { | |
205ab99d | 4707 | return show_slab_objects(s, buf, SO_ALL|SO_OBJECTS); |
81819f0f CL |
4708 | } |
4709 | SLAB_ATTR_RO(objects); | |
4710 | ||
205ab99d CL |
4711 | static ssize_t objects_partial_show(struct kmem_cache *s, char *buf) |
4712 | { | |
4713 | return show_slab_objects(s, buf, SO_PARTIAL|SO_OBJECTS); | |
4714 | } | |
4715 | SLAB_ATTR_RO(objects_partial); | |
4716 | ||
49e22585 CL |
4717 | static ssize_t slabs_cpu_partial_show(struct kmem_cache *s, char *buf) |
4718 | { | |
4719 | int objects = 0; | |
4720 | int pages = 0; | |
4721 | int cpu; | |
4722 | int len; | |
4723 | ||
4724 | for_each_online_cpu(cpu) { | |
4725 | struct page *page = per_cpu_ptr(s->cpu_slab, cpu)->partial; | |
4726 | ||
4727 | if (page) { | |
4728 | pages += page->pages; | |
4729 | objects += page->pobjects; | |
4730 | } | |
4731 | } | |
4732 | ||
4733 | len = sprintf(buf, "%d(%d)", objects, pages); | |
4734 | ||
4735 | #ifdef CONFIG_SMP | |
4736 | for_each_online_cpu(cpu) { | |
4737 | struct page *page = per_cpu_ptr(s->cpu_slab, cpu) ->partial; | |
4738 | ||
4739 | if (page && len < PAGE_SIZE - 20) | |
4740 | len += sprintf(buf + len, " C%d=%d(%d)", cpu, | |
4741 | page->pobjects, page->pages); | |
4742 | } | |
4743 | #endif | |
4744 | return len + sprintf(buf + len, "\n"); | |
4745 | } | |
4746 | SLAB_ATTR_RO(slabs_cpu_partial); | |
4747 | ||
a5a84755 CL |
4748 | static ssize_t reclaim_account_show(struct kmem_cache *s, char *buf) |
4749 | { | |
4750 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_RECLAIM_ACCOUNT)); | |
4751 | } | |
4752 | ||
4753 | static ssize_t reclaim_account_store(struct kmem_cache *s, | |
4754 | const char *buf, size_t length) | |
4755 | { | |
4756 | s->flags &= ~SLAB_RECLAIM_ACCOUNT; | |
4757 | if (buf[0] == '1') | |
4758 | s->flags |= SLAB_RECLAIM_ACCOUNT; | |
4759 | return length; | |
4760 | } | |
4761 | SLAB_ATTR(reclaim_account); | |
4762 | ||
4763 | static ssize_t hwcache_align_show(struct kmem_cache *s, char *buf) | |
4764 | { | |
4765 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_HWCACHE_ALIGN)); | |
4766 | } | |
4767 | SLAB_ATTR_RO(hwcache_align); | |
4768 | ||
4769 | #ifdef CONFIG_ZONE_DMA | |
4770 | static ssize_t cache_dma_show(struct kmem_cache *s, char *buf) | |
4771 | { | |
4772 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_CACHE_DMA)); | |
4773 | } | |
4774 | SLAB_ATTR_RO(cache_dma); | |
4775 | #endif | |
4776 | ||
4777 | static ssize_t destroy_by_rcu_show(struct kmem_cache *s, char *buf) | |
4778 | { | |
4779 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_DESTROY_BY_RCU)); | |
4780 | } | |
4781 | SLAB_ATTR_RO(destroy_by_rcu); | |
4782 | ||
ab9a0f19 LJ |
4783 | static ssize_t reserved_show(struct kmem_cache *s, char *buf) |
4784 | { | |
4785 | return sprintf(buf, "%d\n", s->reserved); | |
4786 | } | |
4787 | SLAB_ATTR_RO(reserved); | |
4788 | ||
ab4d5ed5 | 4789 | #ifdef CONFIG_SLUB_DEBUG |
a5a84755 CL |
4790 | static ssize_t slabs_show(struct kmem_cache *s, char *buf) |
4791 | { | |
4792 | return show_slab_objects(s, buf, SO_ALL); | |
4793 | } | |
4794 | SLAB_ATTR_RO(slabs); | |
4795 | ||
205ab99d CL |
4796 | static ssize_t total_objects_show(struct kmem_cache *s, char *buf) |
4797 | { | |
4798 | return show_slab_objects(s, buf, SO_ALL|SO_TOTAL); | |
4799 | } | |
4800 | SLAB_ATTR_RO(total_objects); | |
4801 | ||
81819f0f CL |
4802 | static ssize_t sanity_checks_show(struct kmem_cache *s, char *buf) |
4803 | { | |
becfda68 | 4804 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_CONSISTENCY_CHECKS)); |
81819f0f CL |
4805 | } |
4806 | ||
4807 | static ssize_t sanity_checks_store(struct kmem_cache *s, | |
4808 | const char *buf, size_t length) | |
4809 | { | |
becfda68 | 4810 | s->flags &= ~SLAB_CONSISTENCY_CHECKS; |
b789ef51 CL |
4811 | if (buf[0] == '1') { |
4812 | s->flags &= ~__CMPXCHG_DOUBLE; | |
becfda68 | 4813 | s->flags |= SLAB_CONSISTENCY_CHECKS; |
b789ef51 | 4814 | } |
81819f0f CL |
4815 | return length; |
4816 | } | |
4817 | SLAB_ATTR(sanity_checks); | |
4818 | ||
4819 | static ssize_t trace_show(struct kmem_cache *s, char *buf) | |
4820 | { | |
4821 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_TRACE)); | |
4822 | } | |
4823 | ||
4824 | static ssize_t trace_store(struct kmem_cache *s, const char *buf, | |
4825 | size_t length) | |
4826 | { | |
c9e16131 CL |
4827 | /* |
4828 | * Tracing a merged cache is going to give confusing results | |
4829 | * as well as cause other issues like converting a mergeable | |
4830 | * cache into an umergeable one. | |
4831 | */ | |
4832 | if (s->refcount > 1) | |
4833 | return -EINVAL; | |
4834 | ||
81819f0f | 4835 | s->flags &= ~SLAB_TRACE; |
b789ef51 CL |
4836 | if (buf[0] == '1') { |
4837 | s->flags &= ~__CMPXCHG_DOUBLE; | |
81819f0f | 4838 | s->flags |= SLAB_TRACE; |
b789ef51 | 4839 | } |
81819f0f CL |
4840 | return length; |
4841 | } | |
4842 | SLAB_ATTR(trace); | |
4843 | ||
81819f0f CL |
4844 | static ssize_t red_zone_show(struct kmem_cache *s, char *buf) |
4845 | { | |
4846 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_RED_ZONE)); | |
4847 | } | |
4848 | ||
4849 | static ssize_t red_zone_store(struct kmem_cache *s, | |
4850 | const char *buf, size_t length) | |
4851 | { | |
4852 | if (any_slab_objects(s)) | |
4853 | return -EBUSY; | |
4854 | ||
4855 | s->flags &= ~SLAB_RED_ZONE; | |
b789ef51 | 4856 | if (buf[0] == '1') { |
81819f0f | 4857 | s->flags |= SLAB_RED_ZONE; |
b789ef51 | 4858 | } |
06b285dc | 4859 | calculate_sizes(s, -1); |
81819f0f CL |
4860 | return length; |
4861 | } | |
4862 | SLAB_ATTR(red_zone); | |
4863 | ||
4864 | static ssize_t poison_show(struct kmem_cache *s, char *buf) | |
4865 | { | |
4866 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_POISON)); | |
4867 | } | |
4868 | ||
4869 | static ssize_t poison_store(struct kmem_cache *s, | |
4870 | const char *buf, size_t length) | |
4871 | { | |
4872 | if (any_slab_objects(s)) | |
4873 | return -EBUSY; | |
4874 | ||
4875 | s->flags &= ~SLAB_POISON; | |
b789ef51 | 4876 | if (buf[0] == '1') { |
81819f0f | 4877 | s->flags |= SLAB_POISON; |
b789ef51 | 4878 | } |
06b285dc | 4879 | calculate_sizes(s, -1); |
81819f0f CL |
4880 | return length; |
4881 | } | |
4882 | SLAB_ATTR(poison); | |
4883 | ||
4884 | static ssize_t store_user_show(struct kmem_cache *s, char *buf) | |
4885 | { | |
4886 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_STORE_USER)); | |
4887 | } | |
4888 | ||
4889 | static ssize_t store_user_store(struct kmem_cache *s, | |
4890 | const char *buf, size_t length) | |
4891 | { | |
4892 | if (any_slab_objects(s)) | |
4893 | return -EBUSY; | |
4894 | ||
4895 | s->flags &= ~SLAB_STORE_USER; | |
b789ef51 CL |
4896 | if (buf[0] == '1') { |
4897 | s->flags &= ~__CMPXCHG_DOUBLE; | |
81819f0f | 4898 | s->flags |= SLAB_STORE_USER; |
b789ef51 | 4899 | } |
06b285dc | 4900 | calculate_sizes(s, -1); |
81819f0f CL |
4901 | return length; |
4902 | } | |
4903 | SLAB_ATTR(store_user); | |
4904 | ||
53e15af0 CL |
4905 | static ssize_t validate_show(struct kmem_cache *s, char *buf) |
4906 | { | |
4907 | return 0; | |
4908 | } | |
4909 | ||
4910 | static ssize_t validate_store(struct kmem_cache *s, | |
4911 | const char *buf, size_t length) | |
4912 | { | |
434e245d CL |
4913 | int ret = -EINVAL; |
4914 | ||
4915 | if (buf[0] == '1') { | |
4916 | ret = validate_slab_cache(s); | |
4917 | if (ret >= 0) | |
4918 | ret = length; | |
4919 | } | |
4920 | return ret; | |
53e15af0 CL |
4921 | } |
4922 | SLAB_ATTR(validate); | |
a5a84755 CL |
4923 | |
4924 | static ssize_t alloc_calls_show(struct kmem_cache *s, char *buf) | |
4925 | { | |
4926 | if (!(s->flags & SLAB_STORE_USER)) | |
4927 | return -ENOSYS; | |
4928 | return list_locations(s, buf, TRACK_ALLOC); | |
4929 | } | |
4930 | SLAB_ATTR_RO(alloc_calls); | |
4931 | ||
4932 | static ssize_t free_calls_show(struct kmem_cache *s, char *buf) | |
4933 | { | |
4934 | if (!(s->flags & SLAB_STORE_USER)) | |
4935 | return -ENOSYS; | |
4936 | return list_locations(s, buf, TRACK_FREE); | |
4937 | } | |
4938 | SLAB_ATTR_RO(free_calls); | |
4939 | #endif /* CONFIG_SLUB_DEBUG */ | |
4940 | ||
4941 | #ifdef CONFIG_FAILSLAB | |
4942 | static ssize_t failslab_show(struct kmem_cache *s, char *buf) | |
4943 | { | |
4944 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_FAILSLAB)); | |
4945 | } | |
4946 | ||
4947 | static ssize_t failslab_store(struct kmem_cache *s, const char *buf, | |
4948 | size_t length) | |
4949 | { | |
c9e16131 CL |
4950 | if (s->refcount > 1) |
4951 | return -EINVAL; | |
4952 | ||
a5a84755 CL |
4953 | s->flags &= ~SLAB_FAILSLAB; |
4954 | if (buf[0] == '1') | |
4955 | s->flags |= SLAB_FAILSLAB; | |
4956 | return length; | |
4957 | } | |
4958 | SLAB_ATTR(failslab); | |
ab4d5ed5 | 4959 | #endif |
53e15af0 | 4960 | |
2086d26a CL |
4961 | static ssize_t shrink_show(struct kmem_cache *s, char *buf) |
4962 | { | |
4963 | return 0; | |
4964 | } | |
4965 | ||
4966 | static ssize_t shrink_store(struct kmem_cache *s, | |
4967 | const char *buf, size_t length) | |
4968 | { | |
832f37f5 VD |
4969 | if (buf[0] == '1') |
4970 | kmem_cache_shrink(s); | |
4971 | else | |
2086d26a CL |
4972 | return -EINVAL; |
4973 | return length; | |
4974 | } | |
4975 | SLAB_ATTR(shrink); | |
4976 | ||
81819f0f | 4977 | #ifdef CONFIG_NUMA |
9824601e | 4978 | static ssize_t remote_node_defrag_ratio_show(struct kmem_cache *s, char *buf) |
81819f0f | 4979 | { |
9824601e | 4980 | return sprintf(buf, "%d\n", s->remote_node_defrag_ratio / 10); |
81819f0f CL |
4981 | } |
4982 | ||
9824601e | 4983 | static ssize_t remote_node_defrag_ratio_store(struct kmem_cache *s, |
81819f0f CL |
4984 | const char *buf, size_t length) |
4985 | { | |
0121c619 CL |
4986 | unsigned long ratio; |
4987 | int err; | |
4988 | ||
3dbb95f7 | 4989 | err = kstrtoul(buf, 10, &ratio); |
0121c619 CL |
4990 | if (err) |
4991 | return err; | |
4992 | ||
e2cb96b7 | 4993 | if (ratio <= 100) |
0121c619 | 4994 | s->remote_node_defrag_ratio = ratio * 10; |
81819f0f | 4995 | |
81819f0f CL |
4996 | return length; |
4997 | } | |
9824601e | 4998 | SLAB_ATTR(remote_node_defrag_ratio); |
81819f0f CL |
4999 | #endif |
5000 | ||
8ff12cfc | 5001 | #ifdef CONFIG_SLUB_STATS |
8ff12cfc CL |
5002 | static int show_stat(struct kmem_cache *s, char *buf, enum stat_item si) |
5003 | { | |
5004 | unsigned long sum = 0; | |
5005 | int cpu; | |
5006 | int len; | |
5007 | int *data = kmalloc(nr_cpu_ids * sizeof(int), GFP_KERNEL); | |
5008 | ||
5009 | if (!data) | |
5010 | return -ENOMEM; | |
5011 | ||
5012 | for_each_online_cpu(cpu) { | |
9dfc6e68 | 5013 | unsigned x = per_cpu_ptr(s->cpu_slab, cpu)->stat[si]; |
8ff12cfc CL |
5014 | |
5015 | data[cpu] = x; | |
5016 | sum += x; | |
5017 | } | |
5018 | ||
5019 | len = sprintf(buf, "%lu", sum); | |
5020 | ||
50ef37b9 | 5021 | #ifdef CONFIG_SMP |
8ff12cfc CL |
5022 | for_each_online_cpu(cpu) { |
5023 | if (data[cpu] && len < PAGE_SIZE - 20) | |
50ef37b9 | 5024 | len += sprintf(buf + len, " C%d=%u", cpu, data[cpu]); |
8ff12cfc | 5025 | } |
50ef37b9 | 5026 | #endif |
8ff12cfc CL |
5027 | kfree(data); |
5028 | return len + sprintf(buf + len, "\n"); | |
5029 | } | |
5030 | ||
78eb00cc DR |
5031 | static void clear_stat(struct kmem_cache *s, enum stat_item si) |
5032 | { | |
5033 | int cpu; | |
5034 | ||
5035 | for_each_online_cpu(cpu) | |
9dfc6e68 | 5036 | per_cpu_ptr(s->cpu_slab, cpu)->stat[si] = 0; |
78eb00cc DR |
5037 | } |
5038 | ||
8ff12cfc CL |
5039 | #define STAT_ATTR(si, text) \ |
5040 | static ssize_t text##_show(struct kmem_cache *s, char *buf) \ | |
5041 | { \ | |
5042 | return show_stat(s, buf, si); \ | |
5043 | } \ | |
78eb00cc DR |
5044 | static ssize_t text##_store(struct kmem_cache *s, \ |
5045 | const char *buf, size_t length) \ | |
5046 | { \ | |
5047 | if (buf[0] != '0') \ | |
5048 | return -EINVAL; \ | |
5049 | clear_stat(s, si); \ | |
5050 | return length; \ | |
5051 | } \ | |
5052 | SLAB_ATTR(text); \ | |
8ff12cfc CL |
5053 | |
5054 | STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath); | |
5055 | STAT_ATTR(ALLOC_SLOWPATH, alloc_slowpath); | |
5056 | STAT_ATTR(FREE_FASTPATH, free_fastpath); | |
5057 | STAT_ATTR(FREE_SLOWPATH, free_slowpath); | |
5058 | STAT_ATTR(FREE_FROZEN, free_frozen); | |
5059 | STAT_ATTR(FREE_ADD_PARTIAL, free_add_partial); | |
5060 | STAT_ATTR(FREE_REMOVE_PARTIAL, free_remove_partial); | |
5061 | STAT_ATTR(ALLOC_FROM_PARTIAL, alloc_from_partial); | |
5062 | STAT_ATTR(ALLOC_SLAB, alloc_slab); | |
5063 | STAT_ATTR(ALLOC_REFILL, alloc_refill); | |
e36a2652 | 5064 | STAT_ATTR(ALLOC_NODE_MISMATCH, alloc_node_mismatch); |
8ff12cfc CL |
5065 | STAT_ATTR(FREE_SLAB, free_slab); |
5066 | STAT_ATTR(CPUSLAB_FLUSH, cpuslab_flush); | |
5067 | STAT_ATTR(DEACTIVATE_FULL, deactivate_full); | |
5068 | STAT_ATTR(DEACTIVATE_EMPTY, deactivate_empty); | |
5069 | STAT_ATTR(DEACTIVATE_TO_HEAD, deactivate_to_head); | |
5070 | STAT_ATTR(DEACTIVATE_TO_TAIL, deactivate_to_tail); | |
5071 | STAT_ATTR(DEACTIVATE_REMOTE_FREES, deactivate_remote_frees); | |
03e404af | 5072 | STAT_ATTR(DEACTIVATE_BYPASS, deactivate_bypass); |
65c3376a | 5073 | STAT_ATTR(ORDER_FALLBACK, order_fallback); |
b789ef51 CL |
5074 | STAT_ATTR(CMPXCHG_DOUBLE_CPU_FAIL, cmpxchg_double_cpu_fail); |
5075 | STAT_ATTR(CMPXCHG_DOUBLE_FAIL, cmpxchg_double_fail); | |
49e22585 CL |
5076 | STAT_ATTR(CPU_PARTIAL_ALLOC, cpu_partial_alloc); |
5077 | STAT_ATTR(CPU_PARTIAL_FREE, cpu_partial_free); | |
8028dcea AS |
5078 | STAT_ATTR(CPU_PARTIAL_NODE, cpu_partial_node); |
5079 | STAT_ATTR(CPU_PARTIAL_DRAIN, cpu_partial_drain); | |
8ff12cfc CL |
5080 | #endif |
5081 | ||
06428780 | 5082 | static struct attribute *slab_attrs[] = { |
81819f0f CL |
5083 | &slab_size_attr.attr, |
5084 | &object_size_attr.attr, | |
5085 | &objs_per_slab_attr.attr, | |
5086 | &order_attr.attr, | |
73d342b1 | 5087 | &min_partial_attr.attr, |
49e22585 | 5088 | &cpu_partial_attr.attr, |
81819f0f | 5089 | &objects_attr.attr, |
205ab99d | 5090 | &objects_partial_attr.attr, |
81819f0f CL |
5091 | &partial_attr.attr, |
5092 | &cpu_slabs_attr.attr, | |
5093 | &ctor_attr.attr, | |
81819f0f CL |
5094 | &aliases_attr.attr, |
5095 | &align_attr.attr, | |
81819f0f CL |
5096 | &hwcache_align_attr.attr, |
5097 | &reclaim_account_attr.attr, | |
5098 | &destroy_by_rcu_attr.attr, | |
a5a84755 | 5099 | &shrink_attr.attr, |
ab9a0f19 | 5100 | &reserved_attr.attr, |
49e22585 | 5101 | &slabs_cpu_partial_attr.attr, |
ab4d5ed5 | 5102 | #ifdef CONFIG_SLUB_DEBUG |
a5a84755 CL |
5103 | &total_objects_attr.attr, |
5104 | &slabs_attr.attr, | |
5105 | &sanity_checks_attr.attr, | |
5106 | &trace_attr.attr, | |
81819f0f CL |
5107 | &red_zone_attr.attr, |
5108 | &poison_attr.attr, | |
5109 | &store_user_attr.attr, | |
53e15af0 | 5110 | &validate_attr.attr, |
88a420e4 CL |
5111 | &alloc_calls_attr.attr, |
5112 | &free_calls_attr.attr, | |
ab4d5ed5 | 5113 | #endif |
81819f0f CL |
5114 | #ifdef CONFIG_ZONE_DMA |
5115 | &cache_dma_attr.attr, | |
5116 | #endif | |
5117 | #ifdef CONFIG_NUMA | |
9824601e | 5118 | &remote_node_defrag_ratio_attr.attr, |
8ff12cfc CL |
5119 | #endif |
5120 | #ifdef CONFIG_SLUB_STATS | |
5121 | &alloc_fastpath_attr.attr, | |
5122 | &alloc_slowpath_attr.attr, | |
5123 | &free_fastpath_attr.attr, | |
5124 | &free_slowpath_attr.attr, | |
5125 | &free_frozen_attr.attr, | |
5126 | &free_add_partial_attr.attr, | |
5127 | &free_remove_partial_attr.attr, | |
5128 | &alloc_from_partial_attr.attr, | |
5129 | &alloc_slab_attr.attr, | |
5130 | &alloc_refill_attr.attr, | |
e36a2652 | 5131 | &alloc_node_mismatch_attr.attr, |
8ff12cfc CL |
5132 | &free_slab_attr.attr, |
5133 | &cpuslab_flush_attr.attr, | |
5134 | &deactivate_full_attr.attr, | |
5135 | &deactivate_empty_attr.attr, | |
5136 | &deactivate_to_head_attr.attr, | |
5137 | &deactivate_to_tail_attr.attr, | |
5138 | &deactivate_remote_frees_attr.attr, | |
03e404af | 5139 | &deactivate_bypass_attr.attr, |
65c3376a | 5140 | &order_fallback_attr.attr, |
b789ef51 CL |
5141 | &cmpxchg_double_fail_attr.attr, |
5142 | &cmpxchg_double_cpu_fail_attr.attr, | |
49e22585 CL |
5143 | &cpu_partial_alloc_attr.attr, |
5144 | &cpu_partial_free_attr.attr, | |
8028dcea AS |
5145 | &cpu_partial_node_attr.attr, |
5146 | &cpu_partial_drain_attr.attr, | |
81819f0f | 5147 | #endif |
4c13dd3b DM |
5148 | #ifdef CONFIG_FAILSLAB |
5149 | &failslab_attr.attr, | |
5150 | #endif | |
5151 | ||
81819f0f CL |
5152 | NULL |
5153 | }; | |
5154 | ||
5155 | static struct attribute_group slab_attr_group = { | |
5156 | .attrs = slab_attrs, | |
5157 | }; | |
5158 | ||
5159 | static ssize_t slab_attr_show(struct kobject *kobj, | |
5160 | struct attribute *attr, | |
5161 | char *buf) | |
5162 | { | |
5163 | struct slab_attribute *attribute; | |
5164 | struct kmem_cache *s; | |
5165 | int err; | |
5166 | ||
5167 | attribute = to_slab_attr(attr); | |
5168 | s = to_slab(kobj); | |
5169 | ||
5170 | if (!attribute->show) | |
5171 | return -EIO; | |
5172 | ||
5173 | err = attribute->show(s, buf); | |
5174 | ||
5175 | return err; | |
5176 | } | |
5177 | ||
5178 | static ssize_t slab_attr_store(struct kobject *kobj, | |
5179 | struct attribute *attr, | |
5180 | const char *buf, size_t len) | |
5181 | { | |
5182 | struct slab_attribute *attribute; | |
5183 | struct kmem_cache *s; | |
5184 | int err; | |
5185 | ||
5186 | attribute = to_slab_attr(attr); | |
5187 | s = to_slab(kobj); | |
5188 | ||
5189 | if (!attribute->store) | |
5190 | return -EIO; | |
5191 | ||
5192 | err = attribute->store(s, buf, len); | |
127424c8 | 5193 | #ifdef CONFIG_MEMCG |
107dab5c | 5194 | if (slab_state >= FULL && err >= 0 && is_root_cache(s)) { |
426589f5 | 5195 | struct kmem_cache *c; |
81819f0f | 5196 | |
107dab5c GC |
5197 | mutex_lock(&slab_mutex); |
5198 | if (s->max_attr_size < len) | |
5199 | s->max_attr_size = len; | |
5200 | ||
ebe945c2 GC |
5201 | /* |
5202 | * This is a best effort propagation, so this function's return | |
5203 | * value will be determined by the parent cache only. This is | |
5204 | * basically because not all attributes will have a well | |
5205 | * defined semantics for rollbacks - most of the actions will | |
5206 | * have permanent effects. | |
5207 | * | |
5208 | * Returning the error value of any of the children that fail | |
5209 | * is not 100 % defined, in the sense that users seeing the | |
5210 | * error code won't be able to know anything about the state of | |
5211 | * the cache. | |
5212 | * | |
5213 | * Only returning the error code for the parent cache at least | |
5214 | * has well defined semantics. The cache being written to | |
5215 | * directly either failed or succeeded, in which case we loop | |
5216 | * through the descendants with best-effort propagation. | |
5217 | */ | |
426589f5 VD |
5218 | for_each_memcg_cache(c, s) |
5219 | attribute->store(c, buf, len); | |
107dab5c GC |
5220 | mutex_unlock(&slab_mutex); |
5221 | } | |
5222 | #endif | |
81819f0f CL |
5223 | return err; |
5224 | } | |
5225 | ||
107dab5c GC |
5226 | static void memcg_propagate_slab_attrs(struct kmem_cache *s) |
5227 | { | |
127424c8 | 5228 | #ifdef CONFIG_MEMCG |
107dab5c GC |
5229 | int i; |
5230 | char *buffer = NULL; | |
93030d83 | 5231 | struct kmem_cache *root_cache; |
107dab5c | 5232 | |
93030d83 | 5233 | if (is_root_cache(s)) |
107dab5c GC |
5234 | return; |
5235 | ||
f7ce3190 | 5236 | root_cache = s->memcg_params.root_cache; |
93030d83 | 5237 | |
107dab5c GC |
5238 | /* |
5239 | * This mean this cache had no attribute written. Therefore, no point | |
5240 | * in copying default values around | |
5241 | */ | |
93030d83 | 5242 | if (!root_cache->max_attr_size) |
107dab5c GC |
5243 | return; |
5244 | ||
5245 | for (i = 0; i < ARRAY_SIZE(slab_attrs); i++) { | |
5246 | char mbuf[64]; | |
5247 | char *buf; | |
5248 | struct slab_attribute *attr = to_slab_attr(slab_attrs[i]); | |
5249 | ||
5250 | if (!attr || !attr->store || !attr->show) | |
5251 | continue; | |
5252 | ||
5253 | /* | |
5254 | * It is really bad that we have to allocate here, so we will | |
5255 | * do it only as a fallback. If we actually allocate, though, | |
5256 | * we can just use the allocated buffer until the end. | |
5257 | * | |
5258 | * Most of the slub attributes will tend to be very small in | |
5259 | * size, but sysfs allows buffers up to a page, so they can | |
5260 | * theoretically happen. | |
5261 | */ | |
5262 | if (buffer) | |
5263 | buf = buffer; | |
93030d83 | 5264 | else if (root_cache->max_attr_size < ARRAY_SIZE(mbuf)) |
107dab5c GC |
5265 | buf = mbuf; |
5266 | else { | |
5267 | buffer = (char *) get_zeroed_page(GFP_KERNEL); | |
5268 | if (WARN_ON(!buffer)) | |
5269 | continue; | |
5270 | buf = buffer; | |
5271 | } | |
5272 | ||
93030d83 | 5273 | attr->show(root_cache, buf); |
107dab5c GC |
5274 | attr->store(s, buf, strlen(buf)); |
5275 | } | |
5276 | ||
5277 | if (buffer) | |
5278 | free_page((unsigned long)buffer); | |
5279 | #endif | |
5280 | } | |
5281 | ||
41a21285 CL |
5282 | static void kmem_cache_release(struct kobject *k) |
5283 | { | |
5284 | slab_kmem_cache_release(to_slab(k)); | |
5285 | } | |
5286 | ||
52cf25d0 | 5287 | static const struct sysfs_ops slab_sysfs_ops = { |
81819f0f CL |
5288 | .show = slab_attr_show, |
5289 | .store = slab_attr_store, | |
5290 | }; | |
5291 | ||
5292 | static struct kobj_type slab_ktype = { | |
5293 | .sysfs_ops = &slab_sysfs_ops, | |
41a21285 | 5294 | .release = kmem_cache_release, |
81819f0f CL |
5295 | }; |
5296 | ||
5297 | static int uevent_filter(struct kset *kset, struct kobject *kobj) | |
5298 | { | |
5299 | struct kobj_type *ktype = get_ktype(kobj); | |
5300 | ||
5301 | if (ktype == &slab_ktype) | |
5302 | return 1; | |
5303 | return 0; | |
5304 | } | |
5305 | ||
9cd43611 | 5306 | static const struct kset_uevent_ops slab_uevent_ops = { |
81819f0f CL |
5307 | .filter = uevent_filter, |
5308 | }; | |
5309 | ||
27c3a314 | 5310 | static struct kset *slab_kset; |
81819f0f | 5311 | |
9a41707b VD |
5312 | static inline struct kset *cache_kset(struct kmem_cache *s) |
5313 | { | |
127424c8 | 5314 | #ifdef CONFIG_MEMCG |
9a41707b | 5315 | if (!is_root_cache(s)) |
f7ce3190 | 5316 | return s->memcg_params.root_cache->memcg_kset; |
9a41707b VD |
5317 | #endif |
5318 | return slab_kset; | |
5319 | } | |
5320 | ||
81819f0f CL |
5321 | #define ID_STR_LENGTH 64 |
5322 | ||
5323 | /* Create a unique string id for a slab cache: | |
6446faa2 CL |
5324 | * |
5325 | * Format :[flags-]size | |
81819f0f CL |
5326 | */ |
5327 | static char *create_unique_id(struct kmem_cache *s) | |
5328 | { | |
5329 | char *name = kmalloc(ID_STR_LENGTH, GFP_KERNEL); | |
5330 | char *p = name; | |
5331 | ||
5332 | BUG_ON(!name); | |
5333 | ||
5334 | *p++ = ':'; | |
5335 | /* | |
5336 | * First flags affecting slabcache operations. We will only | |
5337 | * get here for aliasable slabs so we do not need to support | |
5338 | * too many flags. The flags here must cover all flags that | |
5339 | * are matched during merging to guarantee that the id is | |
5340 | * unique. | |
5341 | */ | |
5342 | if (s->flags & SLAB_CACHE_DMA) | |
5343 | *p++ = 'd'; | |
5344 | if (s->flags & SLAB_RECLAIM_ACCOUNT) | |
5345 | *p++ = 'a'; | |
becfda68 | 5346 | if (s->flags & SLAB_CONSISTENCY_CHECKS) |
81819f0f | 5347 | *p++ = 'F'; |
5a896d9e VN |
5348 | if (!(s->flags & SLAB_NOTRACK)) |
5349 | *p++ = 't'; | |
230e9fc2 VD |
5350 | if (s->flags & SLAB_ACCOUNT) |
5351 | *p++ = 'A'; | |
81819f0f CL |
5352 | if (p != name + 1) |
5353 | *p++ = '-'; | |
5354 | p += sprintf(p, "%07d", s->size); | |
2633d7a0 | 5355 | |
81819f0f CL |
5356 | BUG_ON(p > name + ID_STR_LENGTH - 1); |
5357 | return name; | |
5358 | } | |
5359 | ||
5360 | static int sysfs_slab_add(struct kmem_cache *s) | |
5361 | { | |
5362 | int err; | |
5363 | const char *name; | |
45530c44 | 5364 | int unmergeable = slab_unmergeable(s); |
81819f0f | 5365 | |
81819f0f CL |
5366 | if (unmergeable) { |
5367 | /* | |
5368 | * Slabcache can never be merged so we can use the name proper. | |
5369 | * This is typically the case for debug situations. In that | |
5370 | * case we can catch duplicate names easily. | |
5371 | */ | |
27c3a314 | 5372 | sysfs_remove_link(&slab_kset->kobj, s->name); |
81819f0f CL |
5373 | name = s->name; |
5374 | } else { | |
5375 | /* | |
5376 | * Create a unique name for the slab as a target | |
5377 | * for the symlinks. | |
5378 | */ | |
5379 | name = create_unique_id(s); | |
5380 | } | |
5381 | ||
9a41707b | 5382 | s->kobj.kset = cache_kset(s); |
26e4f205 | 5383 | err = kobject_init_and_add(&s->kobj, &slab_ktype, NULL, "%s", name); |
54b6a731 | 5384 | if (err) |
80da026a | 5385 | goto out; |
81819f0f CL |
5386 | |
5387 | err = sysfs_create_group(&s->kobj, &slab_attr_group); | |
54b6a731 DJ |
5388 | if (err) |
5389 | goto out_del_kobj; | |
9a41707b | 5390 | |
127424c8 | 5391 | #ifdef CONFIG_MEMCG |
9a41707b VD |
5392 | if (is_root_cache(s)) { |
5393 | s->memcg_kset = kset_create_and_add("cgroup", NULL, &s->kobj); | |
5394 | if (!s->memcg_kset) { | |
54b6a731 DJ |
5395 | err = -ENOMEM; |
5396 | goto out_del_kobj; | |
9a41707b VD |
5397 | } |
5398 | } | |
5399 | #endif | |
5400 | ||
81819f0f CL |
5401 | kobject_uevent(&s->kobj, KOBJ_ADD); |
5402 | if (!unmergeable) { | |
5403 | /* Setup first alias */ | |
5404 | sysfs_slab_alias(s, s->name); | |
81819f0f | 5405 | } |
54b6a731 DJ |
5406 | out: |
5407 | if (!unmergeable) | |
5408 | kfree(name); | |
5409 | return err; | |
5410 | out_del_kobj: | |
5411 | kobject_del(&s->kobj); | |
54b6a731 | 5412 | goto out; |
81819f0f CL |
5413 | } |
5414 | ||
41a21285 | 5415 | void sysfs_slab_remove(struct kmem_cache *s) |
81819f0f | 5416 | { |
97d06609 | 5417 | if (slab_state < FULL) |
2bce6485 CL |
5418 | /* |
5419 | * Sysfs has not been setup yet so no need to remove the | |
5420 | * cache from sysfs. | |
5421 | */ | |
5422 | return; | |
5423 | ||
127424c8 | 5424 | #ifdef CONFIG_MEMCG |
9a41707b VD |
5425 | kset_unregister(s->memcg_kset); |
5426 | #endif | |
81819f0f CL |
5427 | kobject_uevent(&s->kobj, KOBJ_REMOVE); |
5428 | kobject_del(&s->kobj); | |
151c602f | 5429 | kobject_put(&s->kobj); |
81819f0f CL |
5430 | } |
5431 | ||
5432 | /* | |
5433 | * Need to buffer aliases during bootup until sysfs becomes | |
9f6c708e | 5434 | * available lest we lose that information. |
81819f0f CL |
5435 | */ |
5436 | struct saved_alias { | |
5437 | struct kmem_cache *s; | |
5438 | const char *name; | |
5439 | struct saved_alias *next; | |
5440 | }; | |
5441 | ||
5af328a5 | 5442 | static struct saved_alias *alias_list; |
81819f0f CL |
5443 | |
5444 | static int sysfs_slab_alias(struct kmem_cache *s, const char *name) | |
5445 | { | |
5446 | struct saved_alias *al; | |
5447 | ||
97d06609 | 5448 | if (slab_state == FULL) { |
81819f0f CL |
5449 | /* |
5450 | * If we have a leftover link then remove it. | |
5451 | */ | |
27c3a314 GKH |
5452 | sysfs_remove_link(&slab_kset->kobj, name); |
5453 | return sysfs_create_link(&slab_kset->kobj, &s->kobj, name); | |
81819f0f CL |
5454 | } |
5455 | ||
5456 | al = kmalloc(sizeof(struct saved_alias), GFP_KERNEL); | |
5457 | if (!al) | |
5458 | return -ENOMEM; | |
5459 | ||
5460 | al->s = s; | |
5461 | al->name = name; | |
5462 | al->next = alias_list; | |
5463 | alias_list = al; | |
5464 | return 0; | |
5465 | } | |
5466 | ||
5467 | static int __init slab_sysfs_init(void) | |
5468 | { | |
5b95a4ac | 5469 | struct kmem_cache *s; |
81819f0f CL |
5470 | int err; |
5471 | ||
18004c5d | 5472 | mutex_lock(&slab_mutex); |
2bce6485 | 5473 | |
0ff21e46 | 5474 | slab_kset = kset_create_and_add("slab", &slab_uevent_ops, kernel_kobj); |
27c3a314 | 5475 | if (!slab_kset) { |
18004c5d | 5476 | mutex_unlock(&slab_mutex); |
f9f58285 | 5477 | pr_err("Cannot register slab subsystem.\n"); |
81819f0f CL |
5478 | return -ENOSYS; |
5479 | } | |
5480 | ||
97d06609 | 5481 | slab_state = FULL; |
26a7bd03 | 5482 | |
5b95a4ac | 5483 | list_for_each_entry(s, &slab_caches, list) { |
26a7bd03 | 5484 | err = sysfs_slab_add(s); |
5d540fb7 | 5485 | if (err) |
f9f58285 FF |
5486 | pr_err("SLUB: Unable to add boot slab %s to sysfs\n", |
5487 | s->name); | |
26a7bd03 | 5488 | } |
81819f0f CL |
5489 | |
5490 | while (alias_list) { | |
5491 | struct saved_alias *al = alias_list; | |
5492 | ||
5493 | alias_list = alias_list->next; | |
5494 | err = sysfs_slab_alias(al->s, al->name); | |
5d540fb7 | 5495 | if (err) |
f9f58285 FF |
5496 | pr_err("SLUB: Unable to add boot slab alias %s to sysfs\n", |
5497 | al->name); | |
81819f0f CL |
5498 | kfree(al); |
5499 | } | |
5500 | ||
18004c5d | 5501 | mutex_unlock(&slab_mutex); |
81819f0f CL |
5502 | resiliency_test(); |
5503 | return 0; | |
5504 | } | |
5505 | ||
5506 | __initcall(slab_sysfs_init); | |
ab4d5ed5 | 5507 | #endif /* CONFIG_SYSFS */ |
57ed3eda PE |
5508 | |
5509 | /* | |
5510 | * The /proc/slabinfo ABI | |
5511 | */ | |
158a9624 | 5512 | #ifdef CONFIG_SLABINFO |
0d7561c6 | 5513 | void get_slabinfo(struct kmem_cache *s, struct slabinfo *sinfo) |
57ed3eda | 5514 | { |
57ed3eda | 5515 | unsigned long nr_slabs = 0; |
205ab99d CL |
5516 | unsigned long nr_objs = 0; |
5517 | unsigned long nr_free = 0; | |
57ed3eda | 5518 | int node; |
fa45dc25 | 5519 | struct kmem_cache_node *n; |
57ed3eda | 5520 | |
fa45dc25 | 5521 | for_each_kmem_cache_node(s, node, n) { |
c17fd13e WL |
5522 | nr_slabs += node_nr_slabs(n); |
5523 | nr_objs += node_nr_objs(n); | |
205ab99d | 5524 | nr_free += count_partial(n, count_free); |
57ed3eda PE |
5525 | } |
5526 | ||
0d7561c6 GC |
5527 | sinfo->active_objs = nr_objs - nr_free; |
5528 | sinfo->num_objs = nr_objs; | |
5529 | sinfo->active_slabs = nr_slabs; | |
5530 | sinfo->num_slabs = nr_slabs; | |
5531 | sinfo->objects_per_slab = oo_objects(s->oo); | |
5532 | sinfo->cache_order = oo_order(s->oo); | |
57ed3eda PE |
5533 | } |
5534 | ||
0d7561c6 | 5535 | void slabinfo_show_stats(struct seq_file *m, struct kmem_cache *s) |
7b3c3a50 | 5536 | { |
7b3c3a50 AD |
5537 | } |
5538 | ||
b7454ad3 GC |
5539 | ssize_t slabinfo_write(struct file *file, const char __user *buffer, |
5540 | size_t count, loff_t *ppos) | |
7b3c3a50 | 5541 | { |
b7454ad3 | 5542 | return -EIO; |
7b3c3a50 | 5543 | } |
158a9624 | 5544 | #endif /* CONFIG_SLABINFO */ |