]>
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 | |
117d54df | 127 | void *fixup_red_left(struct kmem_cache *s, void *p) |
d86bd1be JK |
128 | { |
129 | if (kmem_cache_debug(s) && s->flags & SLAB_RED_ZONE) | |
130 | p += s->red_left_pad; | |
131 | ||
132 | return p; | |
133 | } | |
134 | ||
345c905d JK |
135 | static inline bool kmem_cache_has_cpu_partial(struct kmem_cache *s) |
136 | { | |
137 | #ifdef CONFIG_SLUB_CPU_PARTIAL | |
138 | return !kmem_cache_debug(s); | |
139 | #else | |
140 | return false; | |
141 | #endif | |
142 | } | |
143 | ||
81819f0f CL |
144 | /* |
145 | * Issues still to be resolved: | |
146 | * | |
81819f0f CL |
147 | * - Support PAGE_ALLOC_DEBUG. Should be easy to do. |
148 | * | |
81819f0f CL |
149 | * - Variable sizing of the per node arrays |
150 | */ | |
151 | ||
152 | /* Enable to test recovery from slab corruption on boot */ | |
153 | #undef SLUB_RESILIENCY_TEST | |
154 | ||
b789ef51 CL |
155 | /* Enable to log cmpxchg failures */ |
156 | #undef SLUB_DEBUG_CMPXCHG | |
157 | ||
2086d26a CL |
158 | /* |
159 | * Mininum number of partial slabs. These will be left on the partial | |
160 | * lists even if they are empty. kmem_cache_shrink may reclaim them. | |
161 | */ | |
76be8950 | 162 | #define MIN_PARTIAL 5 |
e95eed57 | 163 | |
2086d26a CL |
164 | /* |
165 | * Maximum number of desirable partial slabs. | |
166 | * The existence of more partial slabs makes kmem_cache_shrink | |
721ae22a | 167 | * sort the partial list by the number of objects in use. |
2086d26a CL |
168 | */ |
169 | #define MAX_PARTIAL 10 | |
170 | ||
becfda68 | 171 | #define DEBUG_DEFAULT_FLAGS (SLAB_CONSISTENCY_CHECKS | SLAB_RED_ZONE | \ |
81819f0f | 172 | SLAB_POISON | SLAB_STORE_USER) |
672bba3a | 173 | |
149daaf3 LA |
174 | /* |
175 | * These debug flags cannot use CMPXCHG because there might be consistency | |
176 | * issues when checking or reading debug information | |
177 | */ | |
178 | #define SLAB_NO_CMPXCHG (SLAB_CONSISTENCY_CHECKS | SLAB_STORE_USER | \ | |
179 | SLAB_TRACE) | |
180 | ||
181 | ||
fa5ec8a1 | 182 | /* |
3de47213 DR |
183 | * Debugging flags that require metadata to be stored in the slab. These get |
184 | * disabled when slub_debug=O is used and a cache's min order increases with | |
185 | * metadata. | |
fa5ec8a1 | 186 | */ |
3de47213 | 187 | #define DEBUG_METADATA_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER) |
fa5ec8a1 | 188 | |
210b5c06 CG |
189 | #define OO_SHIFT 16 |
190 | #define OO_MASK ((1 << OO_SHIFT) - 1) | |
50d5c41c | 191 | #define MAX_OBJS_PER_PAGE 32767 /* since page.objects is u15 */ |
210b5c06 | 192 | |
81819f0f | 193 | /* Internal SLUB flags */ |
f90ec390 | 194 | #define __OBJECT_POISON 0x80000000UL /* Poison object */ |
b789ef51 | 195 | #define __CMPXCHG_DOUBLE 0x40000000UL /* Use cmpxchg_double */ |
81819f0f | 196 | |
02cbc874 CL |
197 | /* |
198 | * Tracking user of a slab. | |
199 | */ | |
d6543e39 | 200 | #define TRACK_ADDRS_COUNT 16 |
02cbc874 | 201 | struct track { |
ce71e27c | 202 | unsigned long addr; /* Called from address */ |
d6543e39 BG |
203 | #ifdef CONFIG_STACKTRACE |
204 | unsigned long addrs[TRACK_ADDRS_COUNT]; /* Called from address */ | |
205 | #endif | |
02cbc874 CL |
206 | int cpu; /* Was running on cpu */ |
207 | int pid; /* Pid context */ | |
208 | unsigned long when; /* When did the operation occur */ | |
209 | }; | |
210 | ||
211 | enum track_item { TRACK_ALLOC, TRACK_FREE }; | |
212 | ||
ab4d5ed5 | 213 | #ifdef CONFIG_SYSFS |
81819f0f CL |
214 | static int sysfs_slab_add(struct kmem_cache *); |
215 | static int sysfs_slab_alias(struct kmem_cache *, const char *); | |
107dab5c | 216 | static void memcg_propagate_slab_attrs(struct kmem_cache *s); |
bf5eb3de | 217 | static void sysfs_slab_remove(struct kmem_cache *s); |
81819f0f | 218 | #else |
0c710013 CL |
219 | static inline int sysfs_slab_add(struct kmem_cache *s) { return 0; } |
220 | static inline int sysfs_slab_alias(struct kmem_cache *s, const char *p) | |
221 | { return 0; } | |
107dab5c | 222 | static inline void memcg_propagate_slab_attrs(struct kmem_cache *s) { } |
bf5eb3de | 223 | static inline void sysfs_slab_remove(struct kmem_cache *s) { } |
81819f0f CL |
224 | #endif |
225 | ||
4fdccdfb | 226 | static inline void stat(const struct kmem_cache *s, enum stat_item si) |
8ff12cfc CL |
227 | { |
228 | #ifdef CONFIG_SLUB_STATS | |
88da03a6 CL |
229 | /* |
230 | * The rmw is racy on a preemptible kernel but this is acceptable, so | |
231 | * avoid this_cpu_add()'s irq-disable overhead. | |
232 | */ | |
233 | raw_cpu_inc(s->cpu_slab->stat[si]); | |
8ff12cfc CL |
234 | #endif |
235 | } | |
236 | ||
81819f0f CL |
237 | /******************************************************************** |
238 | * Core slab cache functions | |
239 | *******************************************************************/ | |
240 | ||
7656c72b CL |
241 | static inline void *get_freepointer(struct kmem_cache *s, void *object) |
242 | { | |
243 | return *(void **)(object + s->offset); | |
244 | } | |
245 | ||
0ad9500e ED |
246 | static void prefetch_freepointer(const struct kmem_cache *s, void *object) |
247 | { | |
248 | prefetch(object + s->offset); | |
249 | } | |
250 | ||
1393d9a1 CL |
251 | static inline void *get_freepointer_safe(struct kmem_cache *s, void *object) |
252 | { | |
253 | void *p; | |
254 | ||
922d566c JK |
255 | if (!debug_pagealloc_enabled()) |
256 | return get_freepointer(s, object); | |
257 | ||
1393d9a1 | 258 | probe_kernel_read(&p, (void **)(object + s->offset), sizeof(p)); |
1393d9a1 CL |
259 | return p; |
260 | } | |
261 | ||
7656c72b CL |
262 | static inline void set_freepointer(struct kmem_cache *s, void *object, void *fp) |
263 | { | |
264 | *(void **)(object + s->offset) = fp; | |
265 | } | |
266 | ||
267 | /* Loop over all objects in a slab */ | |
224a88be | 268 | #define for_each_object(__p, __s, __addr, __objects) \ |
d86bd1be JK |
269 | for (__p = fixup_red_left(__s, __addr); \ |
270 | __p < (__addr) + (__objects) * (__s)->size; \ | |
271 | __p += (__s)->size) | |
7656c72b | 272 | |
54266640 | 273 | #define for_each_object_idx(__p, __idx, __s, __addr, __objects) \ |
d86bd1be JK |
274 | for (__p = fixup_red_left(__s, __addr), __idx = 1; \ |
275 | __idx <= __objects; \ | |
276 | __p += (__s)->size, __idx++) | |
54266640 | 277 | |
7656c72b CL |
278 | /* Determine object index from a given position */ |
279 | static inline int slab_index(void *p, struct kmem_cache *s, void *addr) | |
280 | { | |
281 | return (p - addr) / s->size; | |
282 | } | |
283 | ||
ab9a0f19 LJ |
284 | static inline int order_objects(int order, unsigned long size, int reserved) |
285 | { | |
286 | return ((PAGE_SIZE << order) - reserved) / size; | |
287 | } | |
288 | ||
834f3d11 | 289 | static inline struct kmem_cache_order_objects oo_make(int order, |
ab9a0f19 | 290 | unsigned long size, int reserved) |
834f3d11 CL |
291 | { |
292 | struct kmem_cache_order_objects x = { | |
ab9a0f19 | 293 | (order << OO_SHIFT) + order_objects(order, size, reserved) |
834f3d11 CL |
294 | }; |
295 | ||
296 | return x; | |
297 | } | |
298 | ||
299 | static inline int oo_order(struct kmem_cache_order_objects x) | |
300 | { | |
210b5c06 | 301 | return x.x >> OO_SHIFT; |
834f3d11 CL |
302 | } |
303 | ||
304 | static inline int oo_objects(struct kmem_cache_order_objects x) | |
305 | { | |
210b5c06 | 306 | return x.x & OO_MASK; |
834f3d11 CL |
307 | } |
308 | ||
881db7fb CL |
309 | /* |
310 | * Per slab locking using the pagelock | |
311 | */ | |
312 | static __always_inline void slab_lock(struct page *page) | |
313 | { | |
48c935ad | 314 | VM_BUG_ON_PAGE(PageTail(page), page); |
881db7fb CL |
315 | bit_spin_lock(PG_locked, &page->flags); |
316 | } | |
317 | ||
318 | static __always_inline void slab_unlock(struct page *page) | |
319 | { | |
48c935ad | 320 | VM_BUG_ON_PAGE(PageTail(page), page); |
881db7fb CL |
321 | __bit_spin_unlock(PG_locked, &page->flags); |
322 | } | |
323 | ||
a0320865 DH |
324 | static inline void set_page_slub_counters(struct page *page, unsigned long counters_new) |
325 | { | |
326 | struct page tmp; | |
327 | tmp.counters = counters_new; | |
328 | /* | |
329 | * page->counters can cover frozen/inuse/objects as well | |
0139aa7b JK |
330 | * as page->_refcount. If we assign to ->counters directly |
331 | * we run the risk of losing updates to page->_refcount, so | |
a0320865 DH |
332 | * be careful and only assign to the fields we need. |
333 | */ | |
334 | page->frozen = tmp.frozen; | |
335 | page->inuse = tmp.inuse; | |
336 | page->objects = tmp.objects; | |
337 | } | |
338 | ||
1d07171c CL |
339 | /* Interrupts must be disabled (for the fallback code to work right) */ |
340 | static inline bool __cmpxchg_double_slab(struct kmem_cache *s, struct page *page, | |
341 | void *freelist_old, unsigned long counters_old, | |
342 | void *freelist_new, unsigned long counters_new, | |
343 | const char *n) | |
344 | { | |
345 | VM_BUG_ON(!irqs_disabled()); | |
2565409f HC |
346 | #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \ |
347 | defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE) | |
1d07171c | 348 | if (s->flags & __CMPXCHG_DOUBLE) { |
cdcd6298 | 349 | if (cmpxchg_double(&page->freelist, &page->counters, |
0aa9a13d DC |
350 | freelist_old, counters_old, |
351 | freelist_new, counters_new)) | |
6f6528a1 | 352 | return true; |
1d07171c CL |
353 | } else |
354 | #endif | |
355 | { | |
356 | slab_lock(page); | |
d0e0ac97 CG |
357 | if (page->freelist == freelist_old && |
358 | page->counters == counters_old) { | |
1d07171c | 359 | page->freelist = freelist_new; |
a0320865 | 360 | set_page_slub_counters(page, counters_new); |
1d07171c | 361 | slab_unlock(page); |
6f6528a1 | 362 | return true; |
1d07171c CL |
363 | } |
364 | slab_unlock(page); | |
365 | } | |
366 | ||
367 | cpu_relax(); | |
368 | stat(s, CMPXCHG_DOUBLE_FAIL); | |
369 | ||
370 | #ifdef SLUB_DEBUG_CMPXCHG | |
f9f58285 | 371 | pr_info("%s %s: cmpxchg double redo ", n, s->name); |
1d07171c CL |
372 | #endif |
373 | ||
6f6528a1 | 374 | return false; |
1d07171c CL |
375 | } |
376 | ||
b789ef51 CL |
377 | static inline bool cmpxchg_double_slab(struct kmem_cache *s, struct page *page, |
378 | void *freelist_old, unsigned long counters_old, | |
379 | void *freelist_new, unsigned long counters_new, | |
380 | const char *n) | |
381 | { | |
2565409f HC |
382 | #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \ |
383 | defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE) | |
b789ef51 | 384 | if (s->flags & __CMPXCHG_DOUBLE) { |
cdcd6298 | 385 | if (cmpxchg_double(&page->freelist, &page->counters, |
0aa9a13d DC |
386 | freelist_old, counters_old, |
387 | freelist_new, counters_new)) | |
6f6528a1 | 388 | return true; |
b789ef51 CL |
389 | } else |
390 | #endif | |
391 | { | |
1d07171c CL |
392 | unsigned long flags; |
393 | ||
394 | local_irq_save(flags); | |
881db7fb | 395 | slab_lock(page); |
d0e0ac97 CG |
396 | if (page->freelist == freelist_old && |
397 | page->counters == counters_old) { | |
b789ef51 | 398 | page->freelist = freelist_new; |
a0320865 | 399 | set_page_slub_counters(page, counters_new); |
881db7fb | 400 | slab_unlock(page); |
1d07171c | 401 | local_irq_restore(flags); |
6f6528a1 | 402 | return true; |
b789ef51 | 403 | } |
881db7fb | 404 | slab_unlock(page); |
1d07171c | 405 | local_irq_restore(flags); |
b789ef51 CL |
406 | } |
407 | ||
408 | cpu_relax(); | |
409 | stat(s, CMPXCHG_DOUBLE_FAIL); | |
410 | ||
411 | #ifdef SLUB_DEBUG_CMPXCHG | |
f9f58285 | 412 | pr_info("%s %s: cmpxchg double redo ", n, s->name); |
b789ef51 CL |
413 | #endif |
414 | ||
6f6528a1 | 415 | return false; |
b789ef51 CL |
416 | } |
417 | ||
41ecc55b | 418 | #ifdef CONFIG_SLUB_DEBUG |
5f80b13a CL |
419 | /* |
420 | * Determine a map of object in use on a page. | |
421 | * | |
881db7fb | 422 | * Node listlock must be held to guarantee that the page does |
5f80b13a CL |
423 | * not vanish from under us. |
424 | */ | |
425 | static void get_map(struct kmem_cache *s, struct page *page, unsigned long *map) | |
426 | { | |
427 | void *p; | |
428 | void *addr = page_address(page); | |
429 | ||
430 | for (p = page->freelist; p; p = get_freepointer(s, p)) | |
431 | set_bit(slab_index(p, s, addr), map); | |
432 | } | |
433 | ||
d86bd1be JK |
434 | static inline int size_from_object(struct kmem_cache *s) |
435 | { | |
436 | if (s->flags & SLAB_RED_ZONE) | |
437 | return s->size - s->red_left_pad; | |
438 | ||
439 | return s->size; | |
440 | } | |
441 | ||
442 | static inline void *restore_red_left(struct kmem_cache *s, void *p) | |
443 | { | |
444 | if (s->flags & SLAB_RED_ZONE) | |
445 | p -= s->red_left_pad; | |
446 | ||
447 | return p; | |
448 | } | |
449 | ||
41ecc55b CL |
450 | /* |
451 | * Debug settings: | |
452 | */ | |
89d3c87e | 453 | #if defined(CONFIG_SLUB_DEBUG_ON) |
f0630fff CL |
454 | static int slub_debug = DEBUG_DEFAULT_FLAGS; |
455 | #else | |
41ecc55b | 456 | static int slub_debug; |
f0630fff | 457 | #endif |
41ecc55b CL |
458 | |
459 | static char *slub_debug_slabs; | |
fa5ec8a1 | 460 | static int disable_higher_order_debug; |
41ecc55b | 461 | |
a79316c6 AR |
462 | /* |
463 | * slub is about to manipulate internal object metadata. This memory lies | |
464 | * outside the range of the allocated object, so accessing it would normally | |
465 | * be reported by kasan as a bounds error. metadata_access_enable() is used | |
466 | * to tell kasan that these accesses are OK. | |
467 | */ | |
468 | static inline void metadata_access_enable(void) | |
469 | { | |
470 | kasan_disable_current(); | |
471 | } | |
472 | ||
473 | static inline void metadata_access_disable(void) | |
474 | { | |
475 | kasan_enable_current(); | |
476 | } | |
477 | ||
81819f0f CL |
478 | /* |
479 | * Object debugging | |
480 | */ | |
d86bd1be JK |
481 | |
482 | /* Verify that a pointer has an address that is valid within a slab page */ | |
483 | static inline int check_valid_pointer(struct kmem_cache *s, | |
484 | struct page *page, void *object) | |
485 | { | |
486 | void *base; | |
487 | ||
488 | if (!object) | |
489 | return 1; | |
490 | ||
491 | base = page_address(page); | |
492 | object = restore_red_left(s, object); | |
493 | if (object < base || object >= base + page->objects * s->size || | |
494 | (object - base) % s->size) { | |
495 | return 0; | |
496 | } | |
497 | ||
498 | return 1; | |
499 | } | |
500 | ||
aa2efd5e DT |
501 | static void print_section(char *level, char *text, u8 *addr, |
502 | unsigned int length) | |
81819f0f | 503 | { |
a79316c6 | 504 | metadata_access_enable(); |
aa2efd5e | 505 | print_hex_dump(level, text, DUMP_PREFIX_ADDRESS, 16, 1, addr, |
ffc79d28 | 506 | length, 1); |
a79316c6 | 507 | metadata_access_disable(); |
81819f0f CL |
508 | } |
509 | ||
81819f0f CL |
510 | static struct track *get_track(struct kmem_cache *s, void *object, |
511 | enum track_item alloc) | |
512 | { | |
513 | struct track *p; | |
514 | ||
515 | if (s->offset) | |
516 | p = object + s->offset + sizeof(void *); | |
517 | else | |
518 | p = object + s->inuse; | |
519 | ||
520 | return p + alloc; | |
521 | } | |
522 | ||
523 | static void set_track(struct kmem_cache *s, void *object, | |
ce71e27c | 524 | enum track_item alloc, unsigned long addr) |
81819f0f | 525 | { |
1a00df4a | 526 | struct track *p = get_track(s, object, alloc); |
81819f0f | 527 | |
81819f0f | 528 | if (addr) { |
d6543e39 BG |
529 | #ifdef CONFIG_STACKTRACE |
530 | struct stack_trace trace; | |
531 | int i; | |
532 | ||
533 | trace.nr_entries = 0; | |
534 | trace.max_entries = TRACK_ADDRS_COUNT; | |
535 | trace.entries = p->addrs; | |
536 | trace.skip = 3; | |
a79316c6 | 537 | metadata_access_enable(); |
d6543e39 | 538 | save_stack_trace(&trace); |
a79316c6 | 539 | metadata_access_disable(); |
d6543e39 BG |
540 | |
541 | /* See rant in lockdep.c */ | |
542 | if (trace.nr_entries != 0 && | |
543 | trace.entries[trace.nr_entries - 1] == ULONG_MAX) | |
544 | trace.nr_entries--; | |
545 | ||
546 | for (i = trace.nr_entries; i < TRACK_ADDRS_COUNT; i++) | |
547 | p->addrs[i] = 0; | |
548 | #endif | |
81819f0f CL |
549 | p->addr = addr; |
550 | p->cpu = smp_processor_id(); | |
88e4ccf2 | 551 | p->pid = current->pid; |
81819f0f CL |
552 | p->when = jiffies; |
553 | } else | |
554 | memset(p, 0, sizeof(struct track)); | |
555 | } | |
556 | ||
81819f0f CL |
557 | static void init_tracking(struct kmem_cache *s, void *object) |
558 | { | |
24922684 CL |
559 | if (!(s->flags & SLAB_STORE_USER)) |
560 | return; | |
561 | ||
ce71e27c EGM |
562 | set_track(s, object, TRACK_FREE, 0UL); |
563 | set_track(s, object, TRACK_ALLOC, 0UL); | |
81819f0f CL |
564 | } |
565 | ||
566 | static void print_track(const char *s, struct track *t) | |
567 | { | |
568 | if (!t->addr) | |
569 | return; | |
570 | ||
f9f58285 FF |
571 | pr_err("INFO: %s in %pS age=%lu cpu=%u pid=%d\n", |
572 | s, (void *)t->addr, jiffies - t->when, t->cpu, t->pid); | |
d6543e39 BG |
573 | #ifdef CONFIG_STACKTRACE |
574 | { | |
575 | int i; | |
576 | for (i = 0; i < TRACK_ADDRS_COUNT; i++) | |
577 | if (t->addrs[i]) | |
f9f58285 | 578 | pr_err("\t%pS\n", (void *)t->addrs[i]); |
d6543e39 BG |
579 | else |
580 | break; | |
581 | } | |
582 | #endif | |
24922684 CL |
583 | } |
584 | ||
585 | static void print_tracking(struct kmem_cache *s, void *object) | |
586 | { | |
587 | if (!(s->flags & SLAB_STORE_USER)) | |
588 | return; | |
589 | ||
590 | print_track("Allocated", get_track(s, object, TRACK_ALLOC)); | |
591 | print_track("Freed", get_track(s, object, TRACK_FREE)); | |
592 | } | |
593 | ||
594 | static void print_page_info(struct page *page) | |
595 | { | |
f9f58285 | 596 | pr_err("INFO: Slab 0x%p objects=%u used=%u fp=0x%p flags=0x%04lx\n", |
d0e0ac97 | 597 | page, page->objects, page->inuse, page->freelist, page->flags); |
24922684 CL |
598 | |
599 | } | |
600 | ||
601 | static void slab_bug(struct kmem_cache *s, char *fmt, ...) | |
602 | { | |
ecc42fbe | 603 | struct va_format vaf; |
24922684 | 604 | va_list args; |
24922684 CL |
605 | |
606 | va_start(args, fmt); | |
ecc42fbe FF |
607 | vaf.fmt = fmt; |
608 | vaf.va = &args; | |
f9f58285 | 609 | pr_err("=============================================================================\n"); |
ecc42fbe | 610 | pr_err("BUG %s (%s): %pV\n", s->name, print_tainted(), &vaf); |
f9f58285 | 611 | pr_err("-----------------------------------------------------------------------------\n\n"); |
645df230 | 612 | |
373d4d09 | 613 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); |
ecc42fbe | 614 | va_end(args); |
81819f0f CL |
615 | } |
616 | ||
24922684 CL |
617 | static void slab_fix(struct kmem_cache *s, char *fmt, ...) |
618 | { | |
ecc42fbe | 619 | struct va_format vaf; |
24922684 | 620 | va_list args; |
24922684 CL |
621 | |
622 | va_start(args, fmt); | |
ecc42fbe FF |
623 | vaf.fmt = fmt; |
624 | vaf.va = &args; | |
625 | pr_err("FIX %s: %pV\n", s->name, &vaf); | |
24922684 | 626 | va_end(args); |
24922684 CL |
627 | } |
628 | ||
629 | static void print_trailer(struct kmem_cache *s, struct page *page, u8 *p) | |
81819f0f CL |
630 | { |
631 | unsigned int off; /* Offset of last byte */ | |
a973e9dd | 632 | u8 *addr = page_address(page); |
24922684 CL |
633 | |
634 | print_tracking(s, p); | |
635 | ||
636 | print_page_info(page); | |
637 | ||
f9f58285 FF |
638 | pr_err("INFO: Object 0x%p @offset=%tu fp=0x%p\n\n", |
639 | p, p - addr, get_freepointer(s, p)); | |
24922684 | 640 | |
d86bd1be | 641 | if (s->flags & SLAB_RED_ZONE) |
aa2efd5e DT |
642 | print_section(KERN_ERR, "Redzone ", p - s->red_left_pad, |
643 | s->red_left_pad); | |
d86bd1be | 644 | else if (p > addr + 16) |
aa2efd5e | 645 | print_section(KERN_ERR, "Bytes b4 ", p - 16, 16); |
81819f0f | 646 | |
aa2efd5e DT |
647 | print_section(KERN_ERR, "Object ", p, |
648 | min_t(unsigned long, s->object_size, PAGE_SIZE)); | |
81819f0f | 649 | if (s->flags & SLAB_RED_ZONE) |
aa2efd5e | 650 | print_section(KERN_ERR, "Redzone ", p + s->object_size, |
3b0efdfa | 651 | s->inuse - s->object_size); |
81819f0f | 652 | |
81819f0f CL |
653 | if (s->offset) |
654 | off = s->offset + sizeof(void *); | |
655 | else | |
656 | off = s->inuse; | |
657 | ||
24922684 | 658 | if (s->flags & SLAB_STORE_USER) |
81819f0f | 659 | off += 2 * sizeof(struct track); |
81819f0f | 660 | |
80a9201a AP |
661 | off += kasan_metadata_size(s); |
662 | ||
d86bd1be | 663 | if (off != size_from_object(s)) |
81819f0f | 664 | /* Beginning of the filler is the free pointer */ |
aa2efd5e DT |
665 | print_section(KERN_ERR, "Padding ", p + off, |
666 | size_from_object(s) - off); | |
24922684 CL |
667 | |
668 | dump_stack(); | |
81819f0f CL |
669 | } |
670 | ||
75c66def | 671 | void object_err(struct kmem_cache *s, struct page *page, |
81819f0f CL |
672 | u8 *object, char *reason) |
673 | { | |
3dc50637 | 674 | slab_bug(s, "%s", reason); |
24922684 | 675 | print_trailer(s, page, object); |
81819f0f CL |
676 | } |
677 | ||
d0e0ac97 CG |
678 | static void slab_err(struct kmem_cache *s, struct page *page, |
679 | const char *fmt, ...) | |
81819f0f CL |
680 | { |
681 | va_list args; | |
682 | char buf[100]; | |
683 | ||
24922684 CL |
684 | va_start(args, fmt); |
685 | vsnprintf(buf, sizeof(buf), fmt, args); | |
81819f0f | 686 | va_end(args); |
3dc50637 | 687 | slab_bug(s, "%s", buf); |
24922684 | 688 | print_page_info(page); |
81819f0f CL |
689 | dump_stack(); |
690 | } | |
691 | ||
f7cb1933 | 692 | static void init_object(struct kmem_cache *s, void *object, u8 val) |
81819f0f CL |
693 | { |
694 | u8 *p = object; | |
695 | ||
d86bd1be JK |
696 | if (s->flags & SLAB_RED_ZONE) |
697 | memset(p - s->red_left_pad, val, s->red_left_pad); | |
698 | ||
81819f0f | 699 | if (s->flags & __OBJECT_POISON) { |
3b0efdfa CL |
700 | memset(p, POISON_FREE, s->object_size - 1); |
701 | p[s->object_size - 1] = POISON_END; | |
81819f0f CL |
702 | } |
703 | ||
704 | if (s->flags & SLAB_RED_ZONE) | |
3b0efdfa | 705 | memset(p + s->object_size, val, s->inuse - s->object_size); |
81819f0f CL |
706 | } |
707 | ||
24922684 CL |
708 | static void restore_bytes(struct kmem_cache *s, char *message, u8 data, |
709 | void *from, void *to) | |
710 | { | |
711 | slab_fix(s, "Restoring 0x%p-0x%p=0x%x\n", from, to - 1, data); | |
712 | memset(from, data, to - from); | |
713 | } | |
714 | ||
715 | static int check_bytes_and_report(struct kmem_cache *s, struct page *page, | |
716 | u8 *object, char *what, | |
06428780 | 717 | u8 *start, unsigned int value, unsigned int bytes) |
24922684 CL |
718 | { |
719 | u8 *fault; | |
720 | u8 *end; | |
721 | ||
a79316c6 | 722 | metadata_access_enable(); |
79824820 | 723 | fault = memchr_inv(start, value, bytes); |
a79316c6 | 724 | metadata_access_disable(); |
24922684 CL |
725 | if (!fault) |
726 | return 1; | |
727 | ||
728 | end = start + bytes; | |
729 | while (end > fault && end[-1] == value) | |
730 | end--; | |
731 | ||
732 | slab_bug(s, "%s overwritten", what); | |
f9f58285 | 733 | pr_err("INFO: 0x%p-0x%p. First byte 0x%x instead of 0x%x\n", |
24922684 CL |
734 | fault, end - 1, fault[0], value); |
735 | print_trailer(s, page, object); | |
736 | ||
737 | restore_bytes(s, what, value, fault, end); | |
738 | return 0; | |
81819f0f CL |
739 | } |
740 | ||
81819f0f CL |
741 | /* |
742 | * Object layout: | |
743 | * | |
744 | * object address | |
745 | * Bytes of the object to be managed. | |
746 | * If the freepointer may overlay the object then the free | |
747 | * pointer is the first word of the object. | |
672bba3a | 748 | * |
81819f0f CL |
749 | * Poisoning uses 0x6b (POISON_FREE) and the last byte is |
750 | * 0xa5 (POISON_END) | |
751 | * | |
3b0efdfa | 752 | * object + s->object_size |
81819f0f | 753 | * Padding to reach word boundary. This is also used for Redzoning. |
672bba3a | 754 | * Padding is extended by another word if Redzoning is enabled and |
3b0efdfa | 755 | * object_size == inuse. |
672bba3a | 756 | * |
81819f0f CL |
757 | * We fill with 0xbb (RED_INACTIVE) for inactive objects and with |
758 | * 0xcc (RED_ACTIVE) for objects in use. | |
759 | * | |
760 | * object + s->inuse | |
672bba3a CL |
761 | * Meta data starts here. |
762 | * | |
81819f0f CL |
763 | * A. Free pointer (if we cannot overwrite object on free) |
764 | * B. Tracking data for SLAB_STORE_USER | |
672bba3a | 765 | * C. Padding to reach required alignment boundary or at mininum |
6446faa2 | 766 | * one word if debugging is on to be able to detect writes |
672bba3a CL |
767 | * before the word boundary. |
768 | * | |
769 | * Padding is done using 0x5a (POISON_INUSE) | |
81819f0f CL |
770 | * |
771 | * object + s->size | |
672bba3a | 772 | * Nothing is used beyond s->size. |
81819f0f | 773 | * |
3b0efdfa | 774 | * If slabcaches are merged then the object_size and inuse boundaries are mostly |
672bba3a | 775 | * ignored. And therefore no slab options that rely on these boundaries |
81819f0f CL |
776 | * may be used with merged slabcaches. |
777 | */ | |
778 | ||
81819f0f CL |
779 | static int check_pad_bytes(struct kmem_cache *s, struct page *page, u8 *p) |
780 | { | |
781 | unsigned long off = s->inuse; /* The end of info */ | |
782 | ||
783 | if (s->offset) | |
784 | /* Freepointer is placed after the object. */ | |
785 | off += sizeof(void *); | |
786 | ||
787 | if (s->flags & SLAB_STORE_USER) | |
788 | /* We also have user information there */ | |
789 | off += 2 * sizeof(struct track); | |
790 | ||
80a9201a AP |
791 | off += kasan_metadata_size(s); |
792 | ||
d86bd1be | 793 | if (size_from_object(s) == off) |
81819f0f CL |
794 | return 1; |
795 | ||
24922684 | 796 | return check_bytes_and_report(s, page, p, "Object padding", |
d86bd1be | 797 | p + off, POISON_INUSE, size_from_object(s) - off); |
81819f0f CL |
798 | } |
799 | ||
39b26464 | 800 | /* Check the pad bytes at the end of a slab page */ |
81819f0f CL |
801 | static int slab_pad_check(struct kmem_cache *s, struct page *page) |
802 | { | |
24922684 CL |
803 | u8 *start; |
804 | u8 *fault; | |
805 | u8 *end; | |
806 | int length; | |
807 | int remainder; | |
81819f0f CL |
808 | |
809 | if (!(s->flags & SLAB_POISON)) | |
810 | return 1; | |
811 | ||
a973e9dd | 812 | start = page_address(page); |
ab9a0f19 | 813 | length = (PAGE_SIZE << compound_order(page)) - s->reserved; |
39b26464 CL |
814 | end = start + length; |
815 | remainder = length % s->size; | |
81819f0f CL |
816 | if (!remainder) |
817 | return 1; | |
818 | ||
a79316c6 | 819 | metadata_access_enable(); |
79824820 | 820 | fault = memchr_inv(end - remainder, POISON_INUSE, remainder); |
a79316c6 | 821 | metadata_access_disable(); |
24922684 CL |
822 | if (!fault) |
823 | return 1; | |
824 | while (end > fault && end[-1] == POISON_INUSE) | |
825 | end--; | |
826 | ||
827 | slab_err(s, page, "Padding overwritten. 0x%p-0x%p", fault, end - 1); | |
aa2efd5e | 828 | print_section(KERN_ERR, "Padding ", end - remainder, remainder); |
24922684 | 829 | |
8a3d271d | 830 | restore_bytes(s, "slab padding", POISON_INUSE, end - remainder, end); |
24922684 | 831 | return 0; |
81819f0f CL |
832 | } |
833 | ||
834 | static int check_object(struct kmem_cache *s, struct page *page, | |
f7cb1933 | 835 | void *object, u8 val) |
81819f0f CL |
836 | { |
837 | u8 *p = object; | |
3b0efdfa | 838 | u8 *endobject = object + s->object_size; |
81819f0f CL |
839 | |
840 | if (s->flags & SLAB_RED_ZONE) { | |
d86bd1be JK |
841 | if (!check_bytes_and_report(s, page, object, "Redzone", |
842 | object - s->red_left_pad, val, s->red_left_pad)) | |
843 | return 0; | |
844 | ||
24922684 | 845 | if (!check_bytes_and_report(s, page, object, "Redzone", |
3b0efdfa | 846 | endobject, val, s->inuse - s->object_size)) |
81819f0f | 847 | return 0; |
81819f0f | 848 | } else { |
3b0efdfa | 849 | if ((s->flags & SLAB_POISON) && s->object_size < s->inuse) { |
3adbefee | 850 | check_bytes_and_report(s, page, p, "Alignment padding", |
d0e0ac97 CG |
851 | endobject, POISON_INUSE, |
852 | s->inuse - s->object_size); | |
3adbefee | 853 | } |
81819f0f CL |
854 | } |
855 | ||
856 | if (s->flags & SLAB_POISON) { | |
f7cb1933 | 857 | if (val != SLUB_RED_ACTIVE && (s->flags & __OBJECT_POISON) && |
24922684 | 858 | (!check_bytes_and_report(s, page, p, "Poison", p, |
3b0efdfa | 859 | POISON_FREE, s->object_size - 1) || |
24922684 | 860 | !check_bytes_and_report(s, page, p, "Poison", |
3b0efdfa | 861 | p + s->object_size - 1, POISON_END, 1))) |
81819f0f | 862 | return 0; |
81819f0f CL |
863 | /* |
864 | * check_pad_bytes cleans up on its own. | |
865 | */ | |
866 | check_pad_bytes(s, page, p); | |
867 | } | |
868 | ||
f7cb1933 | 869 | if (!s->offset && val == SLUB_RED_ACTIVE) |
81819f0f CL |
870 | /* |
871 | * Object and freepointer overlap. Cannot check | |
872 | * freepointer while object is allocated. | |
873 | */ | |
874 | return 1; | |
875 | ||
876 | /* Check free pointer validity */ | |
877 | if (!check_valid_pointer(s, page, get_freepointer(s, p))) { | |
878 | object_err(s, page, p, "Freepointer corrupt"); | |
879 | /* | |
9f6c708e | 880 | * No choice but to zap it and thus lose the remainder |
81819f0f | 881 | * of the free objects in this slab. May cause |
672bba3a | 882 | * another error because the object count is now wrong. |
81819f0f | 883 | */ |
a973e9dd | 884 | set_freepointer(s, p, NULL); |
81819f0f CL |
885 | return 0; |
886 | } | |
887 | return 1; | |
888 | } | |
889 | ||
890 | static int check_slab(struct kmem_cache *s, struct page *page) | |
891 | { | |
39b26464 CL |
892 | int maxobj; |
893 | ||
81819f0f CL |
894 | VM_BUG_ON(!irqs_disabled()); |
895 | ||
896 | if (!PageSlab(page)) { | |
24922684 | 897 | slab_err(s, page, "Not a valid slab page"); |
81819f0f CL |
898 | return 0; |
899 | } | |
39b26464 | 900 | |
ab9a0f19 | 901 | maxobj = order_objects(compound_order(page), s->size, s->reserved); |
39b26464 CL |
902 | if (page->objects > maxobj) { |
903 | slab_err(s, page, "objects %u > max %u", | |
f6edde9c | 904 | page->objects, maxobj); |
39b26464 CL |
905 | return 0; |
906 | } | |
907 | if (page->inuse > page->objects) { | |
24922684 | 908 | slab_err(s, page, "inuse %u > max %u", |
f6edde9c | 909 | page->inuse, page->objects); |
81819f0f CL |
910 | return 0; |
911 | } | |
912 | /* Slab_pad_check fixes things up after itself */ | |
913 | slab_pad_check(s, page); | |
914 | return 1; | |
915 | } | |
916 | ||
917 | /* | |
672bba3a CL |
918 | * Determine if a certain object on a page is on the freelist. Must hold the |
919 | * slab lock to guarantee that the chains are in a consistent state. | |
81819f0f CL |
920 | */ |
921 | static int on_freelist(struct kmem_cache *s, struct page *page, void *search) | |
922 | { | |
923 | int nr = 0; | |
881db7fb | 924 | void *fp; |
81819f0f | 925 | void *object = NULL; |
f6edde9c | 926 | int max_objects; |
81819f0f | 927 | |
881db7fb | 928 | fp = page->freelist; |
39b26464 | 929 | while (fp && nr <= page->objects) { |
81819f0f CL |
930 | if (fp == search) |
931 | return 1; | |
932 | if (!check_valid_pointer(s, page, fp)) { | |
933 | if (object) { | |
934 | object_err(s, page, object, | |
935 | "Freechain corrupt"); | |
a973e9dd | 936 | set_freepointer(s, object, NULL); |
81819f0f | 937 | } else { |
24922684 | 938 | slab_err(s, page, "Freepointer corrupt"); |
a973e9dd | 939 | page->freelist = NULL; |
39b26464 | 940 | page->inuse = page->objects; |
24922684 | 941 | slab_fix(s, "Freelist cleared"); |
81819f0f CL |
942 | return 0; |
943 | } | |
944 | break; | |
945 | } | |
946 | object = fp; | |
947 | fp = get_freepointer(s, object); | |
948 | nr++; | |
949 | } | |
950 | ||
ab9a0f19 | 951 | max_objects = order_objects(compound_order(page), s->size, s->reserved); |
210b5c06 CG |
952 | if (max_objects > MAX_OBJS_PER_PAGE) |
953 | max_objects = MAX_OBJS_PER_PAGE; | |
224a88be CL |
954 | |
955 | if (page->objects != max_objects) { | |
756a025f JP |
956 | slab_err(s, page, "Wrong number of objects. Found %d but should be %d", |
957 | page->objects, max_objects); | |
224a88be CL |
958 | page->objects = max_objects; |
959 | slab_fix(s, "Number of objects adjusted."); | |
960 | } | |
39b26464 | 961 | if (page->inuse != page->objects - nr) { |
756a025f JP |
962 | slab_err(s, page, "Wrong object count. Counter is %d but counted were %d", |
963 | page->inuse, page->objects - nr); | |
39b26464 | 964 | page->inuse = page->objects - nr; |
24922684 | 965 | slab_fix(s, "Object count adjusted."); |
81819f0f CL |
966 | } |
967 | return search == NULL; | |
968 | } | |
969 | ||
0121c619 CL |
970 | static void trace(struct kmem_cache *s, struct page *page, void *object, |
971 | int alloc) | |
3ec09742 CL |
972 | { |
973 | if (s->flags & SLAB_TRACE) { | |
f9f58285 | 974 | pr_info("TRACE %s %s 0x%p inuse=%d fp=0x%p\n", |
3ec09742 CL |
975 | s->name, |
976 | alloc ? "alloc" : "free", | |
977 | object, page->inuse, | |
978 | page->freelist); | |
979 | ||
980 | if (!alloc) | |
aa2efd5e | 981 | print_section(KERN_INFO, "Object ", (void *)object, |
d0e0ac97 | 982 | s->object_size); |
3ec09742 CL |
983 | |
984 | dump_stack(); | |
985 | } | |
986 | } | |
987 | ||
643b1138 | 988 | /* |
672bba3a | 989 | * Tracking of fully allocated slabs for debugging purposes. |
643b1138 | 990 | */ |
5cc6eee8 CL |
991 | static void add_full(struct kmem_cache *s, |
992 | struct kmem_cache_node *n, struct page *page) | |
643b1138 | 993 | { |
5cc6eee8 CL |
994 | if (!(s->flags & SLAB_STORE_USER)) |
995 | return; | |
996 | ||
255d0884 | 997 | lockdep_assert_held(&n->list_lock); |
643b1138 | 998 | list_add(&page->lru, &n->full); |
643b1138 CL |
999 | } |
1000 | ||
c65c1877 | 1001 | static void remove_full(struct kmem_cache *s, struct kmem_cache_node *n, struct page *page) |
643b1138 | 1002 | { |
643b1138 CL |
1003 | if (!(s->flags & SLAB_STORE_USER)) |
1004 | return; | |
1005 | ||
255d0884 | 1006 | lockdep_assert_held(&n->list_lock); |
643b1138 | 1007 | list_del(&page->lru); |
643b1138 CL |
1008 | } |
1009 | ||
0f389ec6 CL |
1010 | /* Tracking of the number of slabs for debugging purposes */ |
1011 | static inline unsigned long slabs_node(struct kmem_cache *s, int node) | |
1012 | { | |
1013 | struct kmem_cache_node *n = get_node(s, node); | |
1014 | ||
1015 | return atomic_long_read(&n->nr_slabs); | |
1016 | } | |
1017 | ||
26c02cf0 AB |
1018 | static inline unsigned long node_nr_slabs(struct kmem_cache_node *n) |
1019 | { | |
1020 | return atomic_long_read(&n->nr_slabs); | |
1021 | } | |
1022 | ||
205ab99d | 1023 | static inline void inc_slabs_node(struct kmem_cache *s, int node, int objects) |
0f389ec6 CL |
1024 | { |
1025 | struct kmem_cache_node *n = get_node(s, node); | |
1026 | ||
1027 | /* | |
1028 | * May be called early in order to allocate a slab for the | |
1029 | * kmem_cache_node structure. Solve the chicken-egg | |
1030 | * dilemma by deferring the increment of the count during | |
1031 | * bootstrap (see early_kmem_cache_node_alloc). | |
1032 | */ | |
338b2642 | 1033 | if (likely(n)) { |
0f389ec6 | 1034 | atomic_long_inc(&n->nr_slabs); |
205ab99d CL |
1035 | atomic_long_add(objects, &n->total_objects); |
1036 | } | |
0f389ec6 | 1037 | } |
205ab99d | 1038 | static inline void dec_slabs_node(struct kmem_cache *s, int node, int objects) |
0f389ec6 CL |
1039 | { |
1040 | struct kmem_cache_node *n = get_node(s, node); | |
1041 | ||
1042 | atomic_long_dec(&n->nr_slabs); | |
205ab99d | 1043 | atomic_long_sub(objects, &n->total_objects); |
0f389ec6 CL |
1044 | } |
1045 | ||
1046 | /* Object debug checks for alloc/free paths */ | |
3ec09742 CL |
1047 | static void setup_object_debug(struct kmem_cache *s, struct page *page, |
1048 | void *object) | |
1049 | { | |
1050 | if (!(s->flags & (SLAB_STORE_USER|SLAB_RED_ZONE|__OBJECT_POISON))) | |
1051 | return; | |
1052 | ||
f7cb1933 | 1053 | init_object(s, object, SLUB_RED_INACTIVE); |
3ec09742 CL |
1054 | init_tracking(s, object); |
1055 | } | |
1056 | ||
becfda68 | 1057 | static inline int alloc_consistency_checks(struct kmem_cache *s, |
d0e0ac97 | 1058 | struct page *page, |
ce71e27c | 1059 | void *object, unsigned long addr) |
81819f0f CL |
1060 | { |
1061 | if (!check_slab(s, page)) | |
becfda68 | 1062 | return 0; |
81819f0f | 1063 | |
81819f0f CL |
1064 | if (!check_valid_pointer(s, page, object)) { |
1065 | object_err(s, page, object, "Freelist Pointer check fails"); | |
becfda68 | 1066 | return 0; |
81819f0f CL |
1067 | } |
1068 | ||
f7cb1933 | 1069 | if (!check_object(s, page, object, SLUB_RED_INACTIVE)) |
becfda68 LA |
1070 | return 0; |
1071 | ||
1072 | return 1; | |
1073 | } | |
1074 | ||
1075 | static noinline int alloc_debug_processing(struct kmem_cache *s, | |
1076 | struct page *page, | |
1077 | void *object, unsigned long addr) | |
1078 | { | |
1079 | if (s->flags & SLAB_CONSISTENCY_CHECKS) { | |
1080 | if (!alloc_consistency_checks(s, page, object, addr)) | |
1081 | goto bad; | |
1082 | } | |
81819f0f | 1083 | |
3ec09742 CL |
1084 | /* Success perform special debug activities for allocs */ |
1085 | if (s->flags & SLAB_STORE_USER) | |
1086 | set_track(s, object, TRACK_ALLOC, addr); | |
1087 | trace(s, page, object, 1); | |
f7cb1933 | 1088 | init_object(s, object, SLUB_RED_ACTIVE); |
81819f0f | 1089 | return 1; |
3ec09742 | 1090 | |
81819f0f CL |
1091 | bad: |
1092 | if (PageSlab(page)) { | |
1093 | /* | |
1094 | * If this is a slab page then lets do the best we can | |
1095 | * to avoid issues in the future. Marking all objects | |
672bba3a | 1096 | * as used avoids touching the remaining objects. |
81819f0f | 1097 | */ |
24922684 | 1098 | slab_fix(s, "Marking all objects used"); |
39b26464 | 1099 | page->inuse = page->objects; |
a973e9dd | 1100 | page->freelist = NULL; |
81819f0f CL |
1101 | } |
1102 | return 0; | |
1103 | } | |
1104 | ||
becfda68 LA |
1105 | static inline int free_consistency_checks(struct kmem_cache *s, |
1106 | struct page *page, void *object, unsigned long addr) | |
81819f0f | 1107 | { |
81819f0f | 1108 | if (!check_valid_pointer(s, page, object)) { |
70d71228 | 1109 | slab_err(s, page, "Invalid object pointer 0x%p", object); |
becfda68 | 1110 | return 0; |
81819f0f CL |
1111 | } |
1112 | ||
1113 | if (on_freelist(s, page, object)) { | |
24922684 | 1114 | object_err(s, page, object, "Object already free"); |
becfda68 | 1115 | return 0; |
81819f0f CL |
1116 | } |
1117 | ||
f7cb1933 | 1118 | if (!check_object(s, page, object, SLUB_RED_ACTIVE)) |
becfda68 | 1119 | return 0; |
81819f0f | 1120 | |
1b4f59e3 | 1121 | if (unlikely(s != page->slab_cache)) { |
3adbefee | 1122 | if (!PageSlab(page)) { |
756a025f JP |
1123 | slab_err(s, page, "Attempt to free object(0x%p) outside of slab", |
1124 | object); | |
1b4f59e3 | 1125 | } else if (!page->slab_cache) { |
f9f58285 FF |
1126 | pr_err("SLUB <none>: no slab for object 0x%p.\n", |
1127 | object); | |
70d71228 | 1128 | dump_stack(); |
06428780 | 1129 | } else |
24922684 CL |
1130 | object_err(s, page, object, |
1131 | "page slab pointer corrupt."); | |
becfda68 LA |
1132 | return 0; |
1133 | } | |
1134 | return 1; | |
1135 | } | |
1136 | ||
1137 | /* Supports checking bulk free of a constructed freelist */ | |
1138 | static noinline int free_debug_processing( | |
1139 | struct kmem_cache *s, struct page *page, | |
1140 | void *head, void *tail, int bulk_cnt, | |
1141 | unsigned long addr) | |
1142 | { | |
1143 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); | |
1144 | void *object = head; | |
1145 | int cnt = 0; | |
1146 | unsigned long uninitialized_var(flags); | |
1147 | int ret = 0; | |
1148 | ||
1149 | spin_lock_irqsave(&n->list_lock, flags); | |
1150 | slab_lock(page); | |
1151 | ||
1152 | if (s->flags & SLAB_CONSISTENCY_CHECKS) { | |
1153 | if (!check_slab(s, page)) | |
1154 | goto out; | |
1155 | } | |
1156 | ||
1157 | next_object: | |
1158 | cnt++; | |
1159 | ||
1160 | if (s->flags & SLAB_CONSISTENCY_CHECKS) { | |
1161 | if (!free_consistency_checks(s, page, object, addr)) | |
1162 | goto out; | |
81819f0f | 1163 | } |
3ec09742 | 1164 | |
3ec09742 CL |
1165 | if (s->flags & SLAB_STORE_USER) |
1166 | set_track(s, object, TRACK_FREE, addr); | |
1167 | trace(s, page, object, 0); | |
81084651 | 1168 | /* Freepointer not overwritten by init_object(), SLAB_POISON moved it */ |
f7cb1933 | 1169 | init_object(s, object, SLUB_RED_INACTIVE); |
81084651 JDB |
1170 | |
1171 | /* Reached end of constructed freelist yet? */ | |
1172 | if (object != tail) { | |
1173 | object = get_freepointer(s, object); | |
1174 | goto next_object; | |
1175 | } | |
804aa132 LA |
1176 | ret = 1; |
1177 | ||
5c2e4bbb | 1178 | out: |
81084651 JDB |
1179 | if (cnt != bulk_cnt) |
1180 | slab_err(s, page, "Bulk freelist count(%d) invalid(%d)\n", | |
1181 | bulk_cnt, cnt); | |
1182 | ||
881db7fb | 1183 | slab_unlock(page); |
282acb43 | 1184 | spin_unlock_irqrestore(&n->list_lock, flags); |
804aa132 LA |
1185 | if (!ret) |
1186 | slab_fix(s, "Object at 0x%p not freed", object); | |
1187 | return ret; | |
81819f0f CL |
1188 | } |
1189 | ||
41ecc55b CL |
1190 | static int __init setup_slub_debug(char *str) |
1191 | { | |
f0630fff CL |
1192 | slub_debug = DEBUG_DEFAULT_FLAGS; |
1193 | if (*str++ != '=' || !*str) | |
1194 | /* | |
1195 | * No options specified. Switch on full debugging. | |
1196 | */ | |
1197 | goto out; | |
1198 | ||
1199 | if (*str == ',') | |
1200 | /* | |
1201 | * No options but restriction on slabs. This means full | |
1202 | * debugging for slabs matching a pattern. | |
1203 | */ | |
1204 | goto check_slabs; | |
1205 | ||
1206 | slub_debug = 0; | |
1207 | if (*str == '-') | |
1208 | /* | |
1209 | * Switch off all debugging measures. | |
1210 | */ | |
1211 | goto out; | |
1212 | ||
1213 | /* | |
1214 | * Determine which debug features should be switched on | |
1215 | */ | |
06428780 | 1216 | for (; *str && *str != ','; str++) { |
f0630fff CL |
1217 | switch (tolower(*str)) { |
1218 | case 'f': | |
becfda68 | 1219 | slub_debug |= SLAB_CONSISTENCY_CHECKS; |
f0630fff CL |
1220 | break; |
1221 | case 'z': | |
1222 | slub_debug |= SLAB_RED_ZONE; | |
1223 | break; | |
1224 | case 'p': | |
1225 | slub_debug |= SLAB_POISON; | |
1226 | break; | |
1227 | case 'u': | |
1228 | slub_debug |= SLAB_STORE_USER; | |
1229 | break; | |
1230 | case 't': | |
1231 | slub_debug |= SLAB_TRACE; | |
1232 | break; | |
4c13dd3b DM |
1233 | case 'a': |
1234 | slub_debug |= SLAB_FAILSLAB; | |
1235 | break; | |
08303a73 CA |
1236 | case 'o': |
1237 | /* | |
1238 | * Avoid enabling debugging on caches if its minimum | |
1239 | * order would increase as a result. | |
1240 | */ | |
1241 | disable_higher_order_debug = 1; | |
1242 | break; | |
f0630fff | 1243 | default: |
f9f58285 FF |
1244 | pr_err("slub_debug option '%c' unknown. skipped\n", |
1245 | *str); | |
f0630fff | 1246 | } |
41ecc55b CL |
1247 | } |
1248 | ||
f0630fff | 1249 | check_slabs: |
41ecc55b CL |
1250 | if (*str == ',') |
1251 | slub_debug_slabs = str + 1; | |
f0630fff | 1252 | out: |
41ecc55b CL |
1253 | return 1; |
1254 | } | |
1255 | ||
1256 | __setup("slub_debug", setup_slub_debug); | |
1257 | ||
423c929c | 1258 | unsigned long kmem_cache_flags(unsigned long object_size, |
ba0268a8 | 1259 | unsigned long flags, const char *name, |
51cc5068 | 1260 | void (*ctor)(void *)) |
41ecc55b CL |
1261 | { |
1262 | /* | |
e153362a | 1263 | * Enable debugging if selected on the kernel commandline. |
41ecc55b | 1264 | */ |
c6f58d9b CL |
1265 | if (slub_debug && (!slub_debug_slabs || (name && |
1266 | !strncmp(slub_debug_slabs, name, strlen(slub_debug_slabs))))) | |
3de47213 | 1267 | flags |= slub_debug; |
ba0268a8 CL |
1268 | |
1269 | return flags; | |
41ecc55b | 1270 | } |
b4a64718 | 1271 | #else /* !CONFIG_SLUB_DEBUG */ |
3ec09742 CL |
1272 | static inline void setup_object_debug(struct kmem_cache *s, |
1273 | struct page *page, void *object) {} | |
41ecc55b | 1274 | |
3ec09742 | 1275 | static inline int alloc_debug_processing(struct kmem_cache *s, |
ce71e27c | 1276 | struct page *page, void *object, unsigned long addr) { return 0; } |
41ecc55b | 1277 | |
282acb43 | 1278 | static inline int free_debug_processing( |
81084651 JDB |
1279 | struct kmem_cache *s, struct page *page, |
1280 | void *head, void *tail, int bulk_cnt, | |
282acb43 | 1281 | unsigned long addr) { return 0; } |
41ecc55b | 1282 | |
41ecc55b CL |
1283 | static inline int slab_pad_check(struct kmem_cache *s, struct page *page) |
1284 | { return 1; } | |
1285 | static inline int check_object(struct kmem_cache *s, struct page *page, | |
f7cb1933 | 1286 | void *object, u8 val) { return 1; } |
5cc6eee8 CL |
1287 | static inline void add_full(struct kmem_cache *s, struct kmem_cache_node *n, |
1288 | struct page *page) {} | |
c65c1877 PZ |
1289 | static inline void remove_full(struct kmem_cache *s, struct kmem_cache_node *n, |
1290 | struct page *page) {} | |
423c929c | 1291 | unsigned long kmem_cache_flags(unsigned long object_size, |
ba0268a8 | 1292 | unsigned long flags, const char *name, |
51cc5068 | 1293 | void (*ctor)(void *)) |
ba0268a8 CL |
1294 | { |
1295 | return flags; | |
1296 | } | |
41ecc55b | 1297 | #define slub_debug 0 |
0f389ec6 | 1298 | |
fdaa45e9 IM |
1299 | #define disable_higher_order_debug 0 |
1300 | ||
0f389ec6 CL |
1301 | static inline unsigned long slabs_node(struct kmem_cache *s, int node) |
1302 | { return 0; } | |
26c02cf0 AB |
1303 | static inline unsigned long node_nr_slabs(struct kmem_cache_node *n) |
1304 | { return 0; } | |
205ab99d CL |
1305 | static inline void inc_slabs_node(struct kmem_cache *s, int node, |
1306 | int objects) {} | |
1307 | static inline void dec_slabs_node(struct kmem_cache *s, int node, | |
1308 | int objects) {} | |
7d550c56 | 1309 | |
02e72cc6 AR |
1310 | #endif /* CONFIG_SLUB_DEBUG */ |
1311 | ||
1312 | /* | |
1313 | * Hooks for other subsystems that check memory allocations. In a typical | |
1314 | * production configuration these hooks all should produce no code at all. | |
1315 | */ | |
d56791b3 RB |
1316 | static inline void kmalloc_large_node_hook(void *ptr, size_t size, gfp_t flags) |
1317 | { | |
1318 | kmemleak_alloc(ptr, size, 1, flags); | |
505f5dcb | 1319 | kasan_kmalloc_large(ptr, size, flags); |
d56791b3 RB |
1320 | } |
1321 | ||
1322 | static inline void kfree_hook(const void *x) | |
1323 | { | |
1324 | kmemleak_free(x); | |
0316bec2 | 1325 | kasan_kfree_large(x); |
d56791b3 RB |
1326 | } |
1327 | ||
80a9201a | 1328 | static inline void *slab_free_hook(struct kmem_cache *s, void *x) |
d56791b3 | 1329 | { |
80a9201a AP |
1330 | void *freeptr; |
1331 | ||
d56791b3 | 1332 | kmemleak_free_recursive(x, s->flags); |
7d550c56 | 1333 | |
02e72cc6 AR |
1334 | /* |
1335 | * Trouble is that we may no longer disable interrupts in the fast path | |
1336 | * So in order to make the debug calls that expect irqs to be | |
1337 | * disabled we need to disable interrupts temporarily. | |
1338 | */ | |
1339 | #if defined(CONFIG_KMEMCHECK) || defined(CONFIG_LOCKDEP) | |
1340 | { | |
1341 | unsigned long flags; | |
1342 | ||
1343 | local_irq_save(flags); | |
1344 | kmemcheck_slab_free(s, x, s->object_size); | |
1345 | debug_check_no_locks_freed(x, s->object_size); | |
1346 | local_irq_restore(flags); | |
1347 | } | |
1348 | #endif | |
1349 | if (!(s->flags & SLAB_DEBUG_OBJECTS)) | |
1350 | debug_check_no_obj_freed(x, s->object_size); | |
0316bec2 | 1351 | |
80a9201a AP |
1352 | freeptr = get_freepointer(s, x); |
1353 | /* | |
1354 | * kasan_slab_free() may put x into memory quarantine, delaying its | |
1355 | * reuse. In this case the object's freelist pointer is changed. | |
1356 | */ | |
0316bec2 | 1357 | kasan_slab_free(s, x); |
80a9201a | 1358 | return freeptr; |
02e72cc6 | 1359 | } |
205ab99d | 1360 | |
81084651 JDB |
1361 | static inline void slab_free_freelist_hook(struct kmem_cache *s, |
1362 | void *head, void *tail) | |
1363 | { | |
1364 | /* | |
1365 | * Compiler cannot detect this function can be removed if slab_free_hook() | |
1366 | * evaluates to nothing. Thus, catch all relevant config debug options here. | |
1367 | */ | |
1368 | #if defined(CONFIG_KMEMCHECK) || \ | |
1369 | defined(CONFIG_LOCKDEP) || \ | |
1370 | defined(CONFIG_DEBUG_KMEMLEAK) || \ | |
1371 | defined(CONFIG_DEBUG_OBJECTS_FREE) || \ | |
1372 | defined(CONFIG_KASAN) | |
1373 | ||
1374 | void *object = head; | |
1375 | void *tail_obj = tail ? : head; | |
80a9201a | 1376 | void *freeptr; |
81084651 JDB |
1377 | |
1378 | do { | |
80a9201a AP |
1379 | freeptr = slab_free_hook(s, object); |
1380 | } while ((object != tail_obj) && (object = freeptr)); | |
81084651 JDB |
1381 | #endif |
1382 | } | |
1383 | ||
588f8ba9 TG |
1384 | static void setup_object(struct kmem_cache *s, struct page *page, |
1385 | void *object) | |
1386 | { | |
1387 | setup_object_debug(s, page, object); | |
b3cbd9bf | 1388 | kasan_init_slab_obj(s, object); |
588f8ba9 TG |
1389 | if (unlikely(s->ctor)) { |
1390 | kasan_unpoison_object_data(s, object); | |
1391 | s->ctor(object); | |
1392 | kasan_poison_object_data(s, object); | |
1393 | } | |
1394 | } | |
1395 | ||
81819f0f CL |
1396 | /* |
1397 | * Slab allocation and freeing | |
1398 | */ | |
5dfb4175 VD |
1399 | static inline struct page *alloc_slab_page(struct kmem_cache *s, |
1400 | gfp_t flags, int node, struct kmem_cache_order_objects oo) | |
65c3376a | 1401 | { |
5dfb4175 | 1402 | struct page *page; |
65c3376a CL |
1403 | int order = oo_order(oo); |
1404 | ||
b1eeab67 VN |
1405 | flags |= __GFP_NOTRACK; |
1406 | ||
2154a336 | 1407 | if (node == NUMA_NO_NODE) |
5dfb4175 | 1408 | page = alloc_pages(flags, order); |
65c3376a | 1409 | else |
96db800f | 1410 | page = __alloc_pages_node(node, flags, order); |
5dfb4175 | 1411 | |
f3ccb2c4 VD |
1412 | if (page && memcg_charge_slab(page, flags, order, s)) { |
1413 | __free_pages(page, order); | |
1414 | page = NULL; | |
1415 | } | |
5dfb4175 VD |
1416 | |
1417 | return page; | |
65c3376a CL |
1418 | } |
1419 | ||
210e7a43 TG |
1420 | #ifdef CONFIG_SLAB_FREELIST_RANDOM |
1421 | /* Pre-initialize the random sequence cache */ | |
1422 | static int init_cache_random_seq(struct kmem_cache *s) | |
1423 | { | |
1424 | int err; | |
1425 | unsigned long i, count = oo_objects(s->oo); | |
1426 | ||
a810007a SR |
1427 | /* Bailout if already initialised */ |
1428 | if (s->random_seq) | |
1429 | return 0; | |
1430 | ||
210e7a43 TG |
1431 | err = cache_random_seq_create(s, count, GFP_KERNEL); |
1432 | if (err) { | |
1433 | pr_err("SLUB: Unable to initialize free list for %s\n", | |
1434 | s->name); | |
1435 | return err; | |
1436 | } | |
1437 | ||
1438 | /* Transform to an offset on the set of pages */ | |
1439 | if (s->random_seq) { | |
1440 | for (i = 0; i < count; i++) | |
1441 | s->random_seq[i] *= s->size; | |
1442 | } | |
1443 | return 0; | |
1444 | } | |
1445 | ||
1446 | /* Initialize each random sequence freelist per cache */ | |
1447 | static void __init init_freelist_randomization(void) | |
1448 | { | |
1449 | struct kmem_cache *s; | |
1450 | ||
1451 | mutex_lock(&slab_mutex); | |
1452 | ||
1453 | list_for_each_entry(s, &slab_caches, list) | |
1454 | init_cache_random_seq(s); | |
1455 | ||
1456 | mutex_unlock(&slab_mutex); | |
1457 | } | |
1458 | ||
1459 | /* Get the next entry on the pre-computed freelist randomized */ | |
1460 | static void *next_freelist_entry(struct kmem_cache *s, struct page *page, | |
1461 | unsigned long *pos, void *start, | |
1462 | unsigned long page_limit, | |
1463 | unsigned long freelist_count) | |
1464 | { | |
1465 | unsigned int idx; | |
1466 | ||
1467 | /* | |
1468 | * If the target page allocation failed, the number of objects on the | |
1469 | * page might be smaller than the usual size defined by the cache. | |
1470 | */ | |
1471 | do { | |
1472 | idx = s->random_seq[*pos]; | |
1473 | *pos += 1; | |
1474 | if (*pos >= freelist_count) | |
1475 | *pos = 0; | |
1476 | } while (unlikely(idx >= page_limit)); | |
1477 | ||
1478 | return (char *)start + idx; | |
1479 | } | |
1480 | ||
1481 | /* Shuffle the single linked freelist based on a random pre-computed sequence */ | |
1482 | static bool shuffle_freelist(struct kmem_cache *s, struct page *page) | |
1483 | { | |
1484 | void *start; | |
1485 | void *cur; | |
1486 | void *next; | |
1487 | unsigned long idx, pos, page_limit, freelist_count; | |
1488 | ||
1489 | if (page->objects < 2 || !s->random_seq) | |
1490 | return false; | |
1491 | ||
1492 | freelist_count = oo_objects(s->oo); | |
1493 | pos = get_random_int() % freelist_count; | |
1494 | ||
1495 | page_limit = page->objects * s->size; | |
1496 | start = fixup_red_left(s, page_address(page)); | |
1497 | ||
1498 | /* First entry is used as the base of the freelist */ | |
1499 | cur = next_freelist_entry(s, page, &pos, start, page_limit, | |
1500 | freelist_count); | |
1501 | page->freelist = cur; | |
1502 | ||
1503 | for (idx = 1; idx < page->objects; idx++) { | |
1504 | setup_object(s, page, cur); | |
1505 | next = next_freelist_entry(s, page, &pos, start, page_limit, | |
1506 | freelist_count); | |
1507 | set_freepointer(s, cur, next); | |
1508 | cur = next; | |
1509 | } | |
1510 | setup_object(s, page, cur); | |
1511 | set_freepointer(s, cur, NULL); | |
1512 | ||
1513 | return true; | |
1514 | } | |
1515 | #else | |
1516 | static inline int init_cache_random_seq(struct kmem_cache *s) | |
1517 | { | |
1518 | return 0; | |
1519 | } | |
1520 | static inline void init_freelist_randomization(void) { } | |
1521 | static inline bool shuffle_freelist(struct kmem_cache *s, struct page *page) | |
1522 | { | |
1523 | return false; | |
1524 | } | |
1525 | #endif /* CONFIG_SLAB_FREELIST_RANDOM */ | |
1526 | ||
81819f0f CL |
1527 | static struct page *allocate_slab(struct kmem_cache *s, gfp_t flags, int node) |
1528 | { | |
06428780 | 1529 | struct page *page; |
834f3d11 | 1530 | struct kmem_cache_order_objects oo = s->oo; |
ba52270d | 1531 | gfp_t alloc_gfp; |
588f8ba9 TG |
1532 | void *start, *p; |
1533 | int idx, order; | |
210e7a43 | 1534 | bool shuffle; |
81819f0f | 1535 | |
7e0528da CL |
1536 | flags &= gfp_allowed_mask; |
1537 | ||
d0164adc | 1538 | if (gfpflags_allow_blocking(flags)) |
7e0528da CL |
1539 | local_irq_enable(); |
1540 | ||
b7a49f0d | 1541 | flags |= s->allocflags; |
e12ba74d | 1542 | |
ba52270d PE |
1543 | /* |
1544 | * Let the initial higher-order allocation fail under memory pressure | |
1545 | * so we fall-back to the minimum order allocation. | |
1546 | */ | |
1547 | alloc_gfp = (flags | __GFP_NOWARN | __GFP_NORETRY) & ~__GFP_NOFAIL; | |
d0164adc | 1548 | if ((alloc_gfp & __GFP_DIRECT_RECLAIM) && oo_order(oo) > oo_order(s->min)) |
444eb2a4 | 1549 | alloc_gfp = (alloc_gfp | __GFP_NOMEMALLOC) & ~(__GFP_RECLAIM|__GFP_NOFAIL); |
ba52270d | 1550 | |
5dfb4175 | 1551 | page = alloc_slab_page(s, alloc_gfp, node, oo); |
65c3376a CL |
1552 | if (unlikely(!page)) { |
1553 | oo = s->min; | |
80c3a998 | 1554 | alloc_gfp = flags; |
65c3376a CL |
1555 | /* |
1556 | * Allocation may have failed due to fragmentation. | |
1557 | * Try a lower order alloc if possible | |
1558 | */ | |
5dfb4175 | 1559 | page = alloc_slab_page(s, alloc_gfp, node, oo); |
588f8ba9 TG |
1560 | if (unlikely(!page)) |
1561 | goto out; | |
1562 | stat(s, ORDER_FALLBACK); | |
65c3376a | 1563 | } |
5a896d9e | 1564 | |
588f8ba9 TG |
1565 | if (kmemcheck_enabled && |
1566 | !(s->flags & (SLAB_NOTRACK | DEBUG_DEFAULT_FLAGS))) { | |
b1eeab67 VN |
1567 | int pages = 1 << oo_order(oo); |
1568 | ||
80c3a998 | 1569 | kmemcheck_alloc_shadow(page, oo_order(oo), alloc_gfp, node); |
b1eeab67 VN |
1570 | |
1571 | /* | |
1572 | * Objects from caches that have a constructor don't get | |
1573 | * cleared when they're allocated, so we need to do it here. | |
1574 | */ | |
1575 | if (s->ctor) | |
1576 | kmemcheck_mark_uninitialized_pages(page, pages); | |
1577 | else | |
1578 | kmemcheck_mark_unallocated_pages(page, pages); | |
5a896d9e VN |
1579 | } |
1580 | ||
834f3d11 | 1581 | page->objects = oo_objects(oo); |
81819f0f | 1582 | |
1f458cbf | 1583 | order = compound_order(page); |
1b4f59e3 | 1584 | page->slab_cache = s; |
c03f94cc | 1585 | __SetPageSlab(page); |
2f064f34 | 1586 | if (page_is_pfmemalloc(page)) |
072bb0aa | 1587 | SetPageSlabPfmemalloc(page); |
81819f0f CL |
1588 | |
1589 | start = page_address(page); | |
81819f0f CL |
1590 | |
1591 | if (unlikely(s->flags & SLAB_POISON)) | |
1f458cbf | 1592 | memset(start, POISON_INUSE, PAGE_SIZE << order); |
81819f0f | 1593 | |
0316bec2 AR |
1594 | kasan_poison_slab(page); |
1595 | ||
210e7a43 TG |
1596 | shuffle = shuffle_freelist(s, page); |
1597 | ||
1598 | if (!shuffle) { | |
1599 | for_each_object_idx(p, idx, s, start, page->objects) { | |
1600 | setup_object(s, page, p); | |
1601 | if (likely(idx < page->objects)) | |
1602 | set_freepointer(s, p, p + s->size); | |
1603 | else | |
1604 | set_freepointer(s, p, NULL); | |
1605 | } | |
1606 | page->freelist = fixup_red_left(s, start); | |
81819f0f | 1607 | } |
81819f0f | 1608 | |
e6e82ea1 | 1609 | page->inuse = page->objects; |
8cb0a506 | 1610 | page->frozen = 1; |
588f8ba9 | 1611 | |
81819f0f | 1612 | out: |
d0164adc | 1613 | if (gfpflags_allow_blocking(flags)) |
588f8ba9 TG |
1614 | local_irq_disable(); |
1615 | if (!page) | |
1616 | return NULL; | |
1617 | ||
1618 | mod_zone_page_state(page_zone(page), | |
1619 | (s->flags & SLAB_RECLAIM_ACCOUNT) ? | |
1620 | NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE, | |
1621 | 1 << oo_order(oo)); | |
1622 | ||
1623 | inc_slabs_node(s, page_to_nid(page), page->objects); | |
1624 | ||
81819f0f CL |
1625 | return page; |
1626 | } | |
1627 | ||
588f8ba9 TG |
1628 | static struct page *new_slab(struct kmem_cache *s, gfp_t flags, int node) |
1629 | { | |
1630 | if (unlikely(flags & GFP_SLAB_BUG_MASK)) { | |
bacdcb34 | 1631 | gfp_t invalid_mask = flags & GFP_SLAB_BUG_MASK; |
72baeef0 MH |
1632 | flags &= ~GFP_SLAB_BUG_MASK; |
1633 | pr_warn("Unexpected gfp: %#x (%pGg). Fixing up to gfp: %#x (%pGg). Fix your code!\n", | |
1634 | invalid_mask, &invalid_mask, flags, &flags); | |
65b9de75 | 1635 | dump_stack(); |
588f8ba9 TG |
1636 | } |
1637 | ||
1638 | return allocate_slab(s, | |
1639 | flags & (GFP_RECLAIM_MASK | GFP_CONSTRAINT_MASK), node); | |
1640 | } | |
1641 | ||
81819f0f CL |
1642 | static void __free_slab(struct kmem_cache *s, struct page *page) |
1643 | { | |
834f3d11 CL |
1644 | int order = compound_order(page); |
1645 | int pages = 1 << order; | |
81819f0f | 1646 | |
becfda68 | 1647 | if (s->flags & SLAB_CONSISTENCY_CHECKS) { |
81819f0f CL |
1648 | void *p; |
1649 | ||
1650 | slab_pad_check(s, page); | |
224a88be CL |
1651 | for_each_object(p, s, page_address(page), |
1652 | page->objects) | |
f7cb1933 | 1653 | check_object(s, page, p, SLUB_RED_INACTIVE); |
81819f0f CL |
1654 | } |
1655 | ||
b1eeab67 | 1656 | kmemcheck_free_shadow(page, compound_order(page)); |
5a896d9e | 1657 | |
81819f0f CL |
1658 | mod_zone_page_state(page_zone(page), |
1659 | (s->flags & SLAB_RECLAIM_ACCOUNT) ? | |
1660 | NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE, | |
06428780 | 1661 | -pages); |
81819f0f | 1662 | |
072bb0aa | 1663 | __ClearPageSlabPfmemalloc(page); |
49bd5221 | 1664 | __ClearPageSlab(page); |
1f458cbf | 1665 | |
22b751c3 | 1666 | page_mapcount_reset(page); |
1eb5ac64 NP |
1667 | if (current->reclaim_state) |
1668 | current->reclaim_state->reclaimed_slab += pages; | |
27ee57c9 VD |
1669 | memcg_uncharge_slab(page, order, s); |
1670 | __free_pages(page, order); | |
81819f0f CL |
1671 | } |
1672 | ||
da9a638c LJ |
1673 | #define need_reserve_slab_rcu \ |
1674 | (sizeof(((struct page *)NULL)->lru) < sizeof(struct rcu_head)) | |
1675 | ||
81819f0f CL |
1676 | static void rcu_free_slab(struct rcu_head *h) |
1677 | { | |
1678 | struct page *page; | |
1679 | ||
da9a638c LJ |
1680 | if (need_reserve_slab_rcu) |
1681 | page = virt_to_head_page(h); | |
1682 | else | |
1683 | page = container_of((struct list_head *)h, struct page, lru); | |
1684 | ||
1b4f59e3 | 1685 | __free_slab(page->slab_cache, page); |
81819f0f CL |
1686 | } |
1687 | ||
1688 | static void free_slab(struct kmem_cache *s, struct page *page) | |
1689 | { | |
5f0d5a3a | 1690 | if (unlikely(s->flags & SLAB_TYPESAFE_BY_RCU)) { |
da9a638c LJ |
1691 | struct rcu_head *head; |
1692 | ||
1693 | if (need_reserve_slab_rcu) { | |
1694 | int order = compound_order(page); | |
1695 | int offset = (PAGE_SIZE << order) - s->reserved; | |
1696 | ||
1697 | VM_BUG_ON(s->reserved != sizeof(*head)); | |
1698 | head = page_address(page) + offset; | |
1699 | } else { | |
bc4f610d | 1700 | head = &page->rcu_head; |
da9a638c | 1701 | } |
81819f0f CL |
1702 | |
1703 | call_rcu(head, rcu_free_slab); | |
1704 | } else | |
1705 | __free_slab(s, page); | |
1706 | } | |
1707 | ||
1708 | static void discard_slab(struct kmem_cache *s, struct page *page) | |
1709 | { | |
205ab99d | 1710 | dec_slabs_node(s, page_to_nid(page), page->objects); |
81819f0f CL |
1711 | free_slab(s, page); |
1712 | } | |
1713 | ||
1714 | /* | |
5cc6eee8 | 1715 | * Management of partially allocated slabs. |
81819f0f | 1716 | */ |
1e4dd946 SR |
1717 | static inline void |
1718 | __add_partial(struct kmem_cache_node *n, struct page *page, int tail) | |
81819f0f | 1719 | { |
e95eed57 | 1720 | n->nr_partial++; |
136333d1 | 1721 | if (tail == DEACTIVATE_TO_TAIL) |
7c2e132c CL |
1722 | list_add_tail(&page->lru, &n->partial); |
1723 | else | |
1724 | list_add(&page->lru, &n->partial); | |
81819f0f CL |
1725 | } |
1726 | ||
1e4dd946 SR |
1727 | static inline void add_partial(struct kmem_cache_node *n, |
1728 | struct page *page, int tail) | |
62e346a8 | 1729 | { |
c65c1877 | 1730 | lockdep_assert_held(&n->list_lock); |
1e4dd946 SR |
1731 | __add_partial(n, page, tail); |
1732 | } | |
c65c1877 | 1733 | |
1e4dd946 SR |
1734 | static inline void remove_partial(struct kmem_cache_node *n, |
1735 | struct page *page) | |
1736 | { | |
1737 | lockdep_assert_held(&n->list_lock); | |
52b4b950 DS |
1738 | list_del(&page->lru); |
1739 | n->nr_partial--; | |
1e4dd946 SR |
1740 | } |
1741 | ||
81819f0f | 1742 | /* |
7ced3719 CL |
1743 | * Remove slab from the partial list, freeze it and |
1744 | * return the pointer to the freelist. | |
81819f0f | 1745 | * |
497b66f2 | 1746 | * Returns a list of objects or NULL if it fails. |
81819f0f | 1747 | */ |
497b66f2 | 1748 | static inline void *acquire_slab(struct kmem_cache *s, |
acd19fd1 | 1749 | struct kmem_cache_node *n, struct page *page, |
633b0764 | 1750 | int mode, int *objects) |
81819f0f | 1751 | { |
2cfb7455 CL |
1752 | void *freelist; |
1753 | unsigned long counters; | |
1754 | struct page new; | |
1755 | ||
c65c1877 PZ |
1756 | lockdep_assert_held(&n->list_lock); |
1757 | ||
2cfb7455 CL |
1758 | /* |
1759 | * Zap the freelist and set the frozen bit. | |
1760 | * The old freelist is the list of objects for the | |
1761 | * per cpu allocation list. | |
1762 | */ | |
7ced3719 CL |
1763 | freelist = page->freelist; |
1764 | counters = page->counters; | |
1765 | new.counters = counters; | |
633b0764 | 1766 | *objects = new.objects - new.inuse; |
23910c50 | 1767 | if (mode) { |
7ced3719 | 1768 | new.inuse = page->objects; |
23910c50 PE |
1769 | new.freelist = NULL; |
1770 | } else { | |
1771 | new.freelist = freelist; | |
1772 | } | |
2cfb7455 | 1773 | |
a0132ac0 | 1774 | VM_BUG_ON(new.frozen); |
7ced3719 | 1775 | new.frozen = 1; |
2cfb7455 | 1776 | |
7ced3719 | 1777 | if (!__cmpxchg_double_slab(s, page, |
2cfb7455 | 1778 | freelist, counters, |
02d7633f | 1779 | new.freelist, new.counters, |
7ced3719 | 1780 | "acquire_slab")) |
7ced3719 | 1781 | return NULL; |
2cfb7455 CL |
1782 | |
1783 | remove_partial(n, page); | |
7ced3719 | 1784 | WARN_ON(!freelist); |
49e22585 | 1785 | return freelist; |
81819f0f CL |
1786 | } |
1787 | ||
633b0764 | 1788 | static void put_cpu_partial(struct kmem_cache *s, struct page *page, int drain); |
8ba00bb6 | 1789 | static inline bool pfmemalloc_match(struct page *page, gfp_t gfpflags); |
49e22585 | 1790 | |
81819f0f | 1791 | /* |
672bba3a | 1792 | * Try to allocate a partial slab from a specific node. |
81819f0f | 1793 | */ |
8ba00bb6 JK |
1794 | static void *get_partial_node(struct kmem_cache *s, struct kmem_cache_node *n, |
1795 | struct kmem_cache_cpu *c, gfp_t flags) | |
81819f0f | 1796 | { |
49e22585 CL |
1797 | struct page *page, *page2; |
1798 | void *object = NULL; | |
633b0764 JK |
1799 | int available = 0; |
1800 | int objects; | |
81819f0f CL |
1801 | |
1802 | /* | |
1803 | * Racy check. If we mistakenly see no partial slabs then we | |
1804 | * just allocate an empty slab. If we mistakenly try to get a | |
672bba3a CL |
1805 | * partial slab and there is none available then get_partials() |
1806 | * will return NULL. | |
81819f0f CL |
1807 | */ |
1808 | if (!n || !n->nr_partial) | |
1809 | return NULL; | |
1810 | ||
1811 | spin_lock(&n->list_lock); | |
49e22585 | 1812 | list_for_each_entry_safe(page, page2, &n->partial, lru) { |
8ba00bb6 | 1813 | void *t; |
49e22585 | 1814 | |
8ba00bb6 JK |
1815 | if (!pfmemalloc_match(page, flags)) |
1816 | continue; | |
1817 | ||
633b0764 | 1818 | t = acquire_slab(s, n, page, object == NULL, &objects); |
49e22585 CL |
1819 | if (!t) |
1820 | break; | |
1821 | ||
633b0764 | 1822 | available += objects; |
12d79634 | 1823 | if (!object) { |
49e22585 | 1824 | c->page = page; |
49e22585 | 1825 | stat(s, ALLOC_FROM_PARTIAL); |
49e22585 | 1826 | object = t; |
49e22585 | 1827 | } else { |
633b0764 | 1828 | put_cpu_partial(s, page, 0); |
8028dcea | 1829 | stat(s, CPU_PARTIAL_NODE); |
49e22585 | 1830 | } |
345c905d JK |
1831 | if (!kmem_cache_has_cpu_partial(s) |
1832 | || available > s->cpu_partial / 2) | |
49e22585 CL |
1833 | break; |
1834 | ||
497b66f2 | 1835 | } |
81819f0f | 1836 | spin_unlock(&n->list_lock); |
497b66f2 | 1837 | return object; |
81819f0f CL |
1838 | } |
1839 | ||
1840 | /* | |
672bba3a | 1841 | * Get a page from somewhere. Search in increasing NUMA distances. |
81819f0f | 1842 | */ |
de3ec035 | 1843 | static void *get_any_partial(struct kmem_cache *s, gfp_t flags, |
acd19fd1 | 1844 | struct kmem_cache_cpu *c) |
81819f0f CL |
1845 | { |
1846 | #ifdef CONFIG_NUMA | |
1847 | struct zonelist *zonelist; | |
dd1a239f | 1848 | struct zoneref *z; |
54a6eb5c MG |
1849 | struct zone *zone; |
1850 | enum zone_type high_zoneidx = gfp_zone(flags); | |
497b66f2 | 1851 | void *object; |
cc9a6c87 | 1852 | unsigned int cpuset_mems_cookie; |
81819f0f CL |
1853 | |
1854 | /* | |
672bba3a CL |
1855 | * The defrag ratio allows a configuration of the tradeoffs between |
1856 | * inter node defragmentation and node local allocations. A lower | |
1857 | * defrag_ratio increases the tendency to do local allocations | |
1858 | * instead of attempting to obtain partial slabs from other nodes. | |
81819f0f | 1859 | * |
672bba3a CL |
1860 | * If the defrag_ratio is set to 0 then kmalloc() always |
1861 | * returns node local objects. If the ratio is higher then kmalloc() | |
1862 | * may return off node objects because partial slabs are obtained | |
1863 | * from other nodes and filled up. | |
81819f0f | 1864 | * |
43efd3ea LP |
1865 | * If /sys/kernel/slab/xx/remote_node_defrag_ratio is set to 100 |
1866 | * (which makes defrag_ratio = 1000) then every (well almost) | |
1867 | * allocation will first attempt to defrag slab caches on other nodes. | |
1868 | * This means scanning over all nodes to look for partial slabs which | |
1869 | * may be expensive if we do it every time we are trying to find a slab | |
672bba3a | 1870 | * with available objects. |
81819f0f | 1871 | */ |
9824601e CL |
1872 | if (!s->remote_node_defrag_ratio || |
1873 | get_cycles() % 1024 > s->remote_node_defrag_ratio) | |
81819f0f CL |
1874 | return NULL; |
1875 | ||
cc9a6c87 | 1876 | do { |
d26914d1 | 1877 | cpuset_mems_cookie = read_mems_allowed_begin(); |
2a389610 | 1878 | zonelist = node_zonelist(mempolicy_slab_node(), flags); |
cc9a6c87 MG |
1879 | for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) { |
1880 | struct kmem_cache_node *n; | |
1881 | ||
1882 | n = get_node(s, zone_to_nid(zone)); | |
1883 | ||
dee2f8aa | 1884 | if (n && cpuset_zone_allowed(zone, flags) && |
cc9a6c87 | 1885 | n->nr_partial > s->min_partial) { |
8ba00bb6 | 1886 | object = get_partial_node(s, n, c, flags); |
cc9a6c87 MG |
1887 | if (object) { |
1888 | /* | |
d26914d1 MG |
1889 | * Don't check read_mems_allowed_retry() |
1890 | * here - if mems_allowed was updated in | |
1891 | * parallel, that was a harmless race | |
1892 | * between allocation and the cpuset | |
1893 | * update | |
cc9a6c87 | 1894 | */ |
cc9a6c87 MG |
1895 | return object; |
1896 | } | |
c0ff7453 | 1897 | } |
81819f0f | 1898 | } |
d26914d1 | 1899 | } while (read_mems_allowed_retry(cpuset_mems_cookie)); |
81819f0f CL |
1900 | #endif |
1901 | return NULL; | |
1902 | } | |
1903 | ||
1904 | /* | |
1905 | * Get a partial page, lock it and return it. | |
1906 | */ | |
497b66f2 | 1907 | static void *get_partial(struct kmem_cache *s, gfp_t flags, int node, |
acd19fd1 | 1908 | struct kmem_cache_cpu *c) |
81819f0f | 1909 | { |
497b66f2 | 1910 | void *object; |
a561ce00 JK |
1911 | int searchnode = node; |
1912 | ||
1913 | if (node == NUMA_NO_NODE) | |
1914 | searchnode = numa_mem_id(); | |
1915 | else if (!node_present_pages(node)) | |
1916 | searchnode = node_to_mem_node(node); | |
81819f0f | 1917 | |
8ba00bb6 | 1918 | object = get_partial_node(s, get_node(s, searchnode), c, flags); |
497b66f2 CL |
1919 | if (object || node != NUMA_NO_NODE) |
1920 | return object; | |
81819f0f | 1921 | |
acd19fd1 | 1922 | return get_any_partial(s, flags, c); |
81819f0f CL |
1923 | } |
1924 | ||
8a5ec0ba CL |
1925 | #ifdef CONFIG_PREEMPT |
1926 | /* | |
1927 | * Calculate the next globally unique transaction for disambiguiation | |
1928 | * during cmpxchg. The transactions start with the cpu number and are then | |
1929 | * incremented by CONFIG_NR_CPUS. | |
1930 | */ | |
1931 | #define TID_STEP roundup_pow_of_two(CONFIG_NR_CPUS) | |
1932 | #else | |
1933 | /* | |
1934 | * No preemption supported therefore also no need to check for | |
1935 | * different cpus. | |
1936 | */ | |
1937 | #define TID_STEP 1 | |
1938 | #endif | |
1939 | ||
1940 | static inline unsigned long next_tid(unsigned long tid) | |
1941 | { | |
1942 | return tid + TID_STEP; | |
1943 | } | |
1944 | ||
1945 | static inline unsigned int tid_to_cpu(unsigned long tid) | |
1946 | { | |
1947 | return tid % TID_STEP; | |
1948 | } | |
1949 | ||
1950 | static inline unsigned long tid_to_event(unsigned long tid) | |
1951 | { | |
1952 | return tid / TID_STEP; | |
1953 | } | |
1954 | ||
1955 | static inline unsigned int init_tid(int cpu) | |
1956 | { | |
1957 | return cpu; | |
1958 | } | |
1959 | ||
1960 | static inline void note_cmpxchg_failure(const char *n, | |
1961 | const struct kmem_cache *s, unsigned long tid) | |
1962 | { | |
1963 | #ifdef SLUB_DEBUG_CMPXCHG | |
1964 | unsigned long actual_tid = __this_cpu_read(s->cpu_slab->tid); | |
1965 | ||
f9f58285 | 1966 | pr_info("%s %s: cmpxchg redo ", n, s->name); |
8a5ec0ba CL |
1967 | |
1968 | #ifdef CONFIG_PREEMPT | |
1969 | if (tid_to_cpu(tid) != tid_to_cpu(actual_tid)) | |
f9f58285 | 1970 | pr_warn("due to cpu change %d -> %d\n", |
8a5ec0ba CL |
1971 | tid_to_cpu(tid), tid_to_cpu(actual_tid)); |
1972 | else | |
1973 | #endif | |
1974 | if (tid_to_event(tid) != tid_to_event(actual_tid)) | |
f9f58285 | 1975 | pr_warn("due to cpu running other code. Event %ld->%ld\n", |
8a5ec0ba CL |
1976 | tid_to_event(tid), tid_to_event(actual_tid)); |
1977 | else | |
f9f58285 | 1978 | pr_warn("for unknown reason: actual=%lx was=%lx target=%lx\n", |
8a5ec0ba CL |
1979 | actual_tid, tid, next_tid(tid)); |
1980 | #endif | |
4fdccdfb | 1981 | stat(s, CMPXCHG_DOUBLE_CPU_FAIL); |
8a5ec0ba CL |
1982 | } |
1983 | ||
788e1aad | 1984 | static void init_kmem_cache_cpus(struct kmem_cache *s) |
8a5ec0ba | 1985 | { |
8a5ec0ba CL |
1986 | int cpu; |
1987 | ||
1988 | for_each_possible_cpu(cpu) | |
1989 | per_cpu_ptr(s->cpu_slab, cpu)->tid = init_tid(cpu); | |
8a5ec0ba | 1990 | } |
2cfb7455 | 1991 | |
81819f0f CL |
1992 | /* |
1993 | * Remove the cpu slab | |
1994 | */ | |
d0e0ac97 CG |
1995 | static void deactivate_slab(struct kmem_cache *s, struct page *page, |
1996 | void *freelist) | |
81819f0f | 1997 | { |
2cfb7455 | 1998 | enum slab_modes { M_NONE, M_PARTIAL, M_FULL, M_FREE }; |
2cfb7455 CL |
1999 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); |
2000 | int lock = 0; | |
2001 | enum slab_modes l = M_NONE, m = M_NONE; | |
2cfb7455 | 2002 | void *nextfree; |
136333d1 | 2003 | int tail = DEACTIVATE_TO_HEAD; |
2cfb7455 CL |
2004 | struct page new; |
2005 | struct page old; | |
2006 | ||
2007 | if (page->freelist) { | |
84e554e6 | 2008 | stat(s, DEACTIVATE_REMOTE_FREES); |
136333d1 | 2009 | tail = DEACTIVATE_TO_TAIL; |
2cfb7455 CL |
2010 | } |
2011 | ||
894b8788 | 2012 | /* |
2cfb7455 CL |
2013 | * Stage one: Free all available per cpu objects back |
2014 | * to the page freelist while it is still frozen. Leave the | |
2015 | * last one. | |
2016 | * | |
2017 | * There is no need to take the list->lock because the page | |
2018 | * is still frozen. | |
2019 | */ | |
2020 | while (freelist && (nextfree = get_freepointer(s, freelist))) { | |
2021 | void *prior; | |
2022 | unsigned long counters; | |
2023 | ||
2024 | do { | |
2025 | prior = page->freelist; | |
2026 | counters = page->counters; | |
2027 | set_freepointer(s, freelist, prior); | |
2028 | new.counters = counters; | |
2029 | new.inuse--; | |
a0132ac0 | 2030 | VM_BUG_ON(!new.frozen); |
2cfb7455 | 2031 | |
1d07171c | 2032 | } while (!__cmpxchg_double_slab(s, page, |
2cfb7455 CL |
2033 | prior, counters, |
2034 | freelist, new.counters, | |
2035 | "drain percpu freelist")); | |
2036 | ||
2037 | freelist = nextfree; | |
2038 | } | |
2039 | ||
894b8788 | 2040 | /* |
2cfb7455 CL |
2041 | * Stage two: Ensure that the page is unfrozen while the |
2042 | * list presence reflects the actual number of objects | |
2043 | * during unfreeze. | |
2044 | * | |
2045 | * We setup the list membership and then perform a cmpxchg | |
2046 | * with the count. If there is a mismatch then the page | |
2047 | * is not unfrozen but the page is on the wrong list. | |
2048 | * | |
2049 | * Then we restart the process which may have to remove | |
2050 | * the page from the list that we just put it on again | |
2051 | * because the number of objects in the slab may have | |
2052 | * changed. | |
894b8788 | 2053 | */ |
2cfb7455 | 2054 | redo: |
894b8788 | 2055 | |
2cfb7455 CL |
2056 | old.freelist = page->freelist; |
2057 | old.counters = page->counters; | |
a0132ac0 | 2058 | VM_BUG_ON(!old.frozen); |
7c2e132c | 2059 | |
2cfb7455 CL |
2060 | /* Determine target state of the slab */ |
2061 | new.counters = old.counters; | |
2062 | if (freelist) { | |
2063 | new.inuse--; | |
2064 | set_freepointer(s, freelist, old.freelist); | |
2065 | new.freelist = freelist; | |
2066 | } else | |
2067 | new.freelist = old.freelist; | |
2068 | ||
2069 | new.frozen = 0; | |
2070 | ||
8a5b20ae | 2071 | if (!new.inuse && n->nr_partial >= s->min_partial) |
2cfb7455 CL |
2072 | m = M_FREE; |
2073 | else if (new.freelist) { | |
2074 | m = M_PARTIAL; | |
2075 | if (!lock) { | |
2076 | lock = 1; | |
2077 | /* | |
2078 | * Taking the spinlock removes the possiblity | |
2079 | * that acquire_slab() will see a slab page that | |
2080 | * is frozen | |
2081 | */ | |
2082 | spin_lock(&n->list_lock); | |
2083 | } | |
2084 | } else { | |
2085 | m = M_FULL; | |
2086 | if (kmem_cache_debug(s) && !lock) { | |
2087 | lock = 1; | |
2088 | /* | |
2089 | * This also ensures that the scanning of full | |
2090 | * slabs from diagnostic functions will not see | |
2091 | * any frozen slabs. | |
2092 | */ | |
2093 | spin_lock(&n->list_lock); | |
2094 | } | |
2095 | } | |
2096 | ||
2097 | if (l != m) { | |
2098 | ||
2099 | if (l == M_PARTIAL) | |
2100 | ||
2101 | remove_partial(n, page); | |
2102 | ||
2103 | else if (l == M_FULL) | |
894b8788 | 2104 | |
c65c1877 | 2105 | remove_full(s, n, page); |
2cfb7455 CL |
2106 | |
2107 | if (m == M_PARTIAL) { | |
2108 | ||
2109 | add_partial(n, page, tail); | |
136333d1 | 2110 | stat(s, tail); |
2cfb7455 CL |
2111 | |
2112 | } else if (m == M_FULL) { | |
894b8788 | 2113 | |
2cfb7455 CL |
2114 | stat(s, DEACTIVATE_FULL); |
2115 | add_full(s, n, page); | |
2116 | ||
2117 | } | |
2118 | } | |
2119 | ||
2120 | l = m; | |
1d07171c | 2121 | if (!__cmpxchg_double_slab(s, page, |
2cfb7455 CL |
2122 | old.freelist, old.counters, |
2123 | new.freelist, new.counters, | |
2124 | "unfreezing slab")) | |
2125 | goto redo; | |
2126 | ||
2cfb7455 CL |
2127 | if (lock) |
2128 | spin_unlock(&n->list_lock); | |
2129 | ||
2130 | if (m == M_FREE) { | |
2131 | stat(s, DEACTIVATE_EMPTY); | |
2132 | discard_slab(s, page); | |
2133 | stat(s, FREE_SLAB); | |
894b8788 | 2134 | } |
81819f0f CL |
2135 | } |
2136 | ||
d24ac77f JK |
2137 | /* |
2138 | * Unfreeze all the cpu partial slabs. | |
2139 | * | |
59a09917 CL |
2140 | * This function must be called with interrupts disabled |
2141 | * for the cpu using c (or some other guarantee must be there | |
2142 | * to guarantee no concurrent accesses). | |
d24ac77f | 2143 | */ |
59a09917 CL |
2144 | static void unfreeze_partials(struct kmem_cache *s, |
2145 | struct kmem_cache_cpu *c) | |
49e22585 | 2146 | { |
345c905d | 2147 | #ifdef CONFIG_SLUB_CPU_PARTIAL |
43d77867 | 2148 | struct kmem_cache_node *n = NULL, *n2 = NULL; |
9ada1934 | 2149 | struct page *page, *discard_page = NULL; |
49e22585 CL |
2150 | |
2151 | while ((page = c->partial)) { | |
49e22585 CL |
2152 | struct page new; |
2153 | struct page old; | |
2154 | ||
2155 | c->partial = page->next; | |
43d77867 JK |
2156 | |
2157 | n2 = get_node(s, page_to_nid(page)); | |
2158 | if (n != n2) { | |
2159 | if (n) | |
2160 | spin_unlock(&n->list_lock); | |
2161 | ||
2162 | n = n2; | |
2163 | spin_lock(&n->list_lock); | |
2164 | } | |
49e22585 CL |
2165 | |
2166 | do { | |
2167 | ||
2168 | old.freelist = page->freelist; | |
2169 | old.counters = page->counters; | |
a0132ac0 | 2170 | VM_BUG_ON(!old.frozen); |
49e22585 CL |
2171 | |
2172 | new.counters = old.counters; | |
2173 | new.freelist = old.freelist; | |
2174 | ||
2175 | new.frozen = 0; | |
2176 | ||
d24ac77f | 2177 | } while (!__cmpxchg_double_slab(s, page, |
49e22585 CL |
2178 | old.freelist, old.counters, |
2179 | new.freelist, new.counters, | |
2180 | "unfreezing slab")); | |
2181 | ||
8a5b20ae | 2182 | if (unlikely(!new.inuse && n->nr_partial >= s->min_partial)) { |
9ada1934 SL |
2183 | page->next = discard_page; |
2184 | discard_page = page; | |
43d77867 JK |
2185 | } else { |
2186 | add_partial(n, page, DEACTIVATE_TO_TAIL); | |
2187 | stat(s, FREE_ADD_PARTIAL); | |
49e22585 CL |
2188 | } |
2189 | } | |
2190 | ||
2191 | if (n) | |
2192 | spin_unlock(&n->list_lock); | |
9ada1934 SL |
2193 | |
2194 | while (discard_page) { | |
2195 | page = discard_page; | |
2196 | discard_page = discard_page->next; | |
2197 | ||
2198 | stat(s, DEACTIVATE_EMPTY); | |
2199 | discard_slab(s, page); | |
2200 | stat(s, FREE_SLAB); | |
2201 | } | |
345c905d | 2202 | #endif |
49e22585 CL |
2203 | } |
2204 | ||
2205 | /* | |
2206 | * Put a page that was just frozen (in __slab_free) into a partial page | |
2207 | * slot if available. This is done without interrupts disabled and without | |
2208 | * preemption disabled. The cmpxchg is racy and may put the partial page | |
2209 | * onto a random cpus partial slot. | |
2210 | * | |
2211 | * If we did not find a slot then simply move all the partials to the | |
2212 | * per node partial list. | |
2213 | */ | |
633b0764 | 2214 | static void put_cpu_partial(struct kmem_cache *s, struct page *page, int drain) |
49e22585 | 2215 | { |
345c905d | 2216 | #ifdef CONFIG_SLUB_CPU_PARTIAL |
49e22585 CL |
2217 | struct page *oldpage; |
2218 | int pages; | |
2219 | int pobjects; | |
2220 | ||
d6e0b7fa | 2221 | preempt_disable(); |
49e22585 CL |
2222 | do { |
2223 | pages = 0; | |
2224 | pobjects = 0; | |
2225 | oldpage = this_cpu_read(s->cpu_slab->partial); | |
2226 | ||
2227 | if (oldpage) { | |
2228 | pobjects = oldpage->pobjects; | |
2229 | pages = oldpage->pages; | |
2230 | if (drain && pobjects > s->cpu_partial) { | |
2231 | unsigned long flags; | |
2232 | /* | |
2233 | * partial array is full. Move the existing | |
2234 | * set to the per node partial list. | |
2235 | */ | |
2236 | local_irq_save(flags); | |
59a09917 | 2237 | unfreeze_partials(s, this_cpu_ptr(s->cpu_slab)); |
49e22585 | 2238 | local_irq_restore(flags); |
e24fc410 | 2239 | oldpage = NULL; |
49e22585 CL |
2240 | pobjects = 0; |
2241 | pages = 0; | |
8028dcea | 2242 | stat(s, CPU_PARTIAL_DRAIN); |
49e22585 CL |
2243 | } |
2244 | } | |
2245 | ||
2246 | pages++; | |
2247 | pobjects += page->objects - page->inuse; | |
2248 | ||
2249 | page->pages = pages; | |
2250 | page->pobjects = pobjects; | |
2251 | page->next = oldpage; | |
2252 | ||
d0e0ac97 CG |
2253 | } while (this_cpu_cmpxchg(s->cpu_slab->partial, oldpage, page) |
2254 | != oldpage); | |
d6e0b7fa VD |
2255 | if (unlikely(!s->cpu_partial)) { |
2256 | unsigned long flags; | |
2257 | ||
2258 | local_irq_save(flags); | |
2259 | unfreeze_partials(s, this_cpu_ptr(s->cpu_slab)); | |
2260 | local_irq_restore(flags); | |
2261 | } | |
2262 | preempt_enable(); | |
345c905d | 2263 | #endif |
49e22585 CL |
2264 | } |
2265 | ||
dfb4f096 | 2266 | static inline void flush_slab(struct kmem_cache *s, struct kmem_cache_cpu *c) |
81819f0f | 2267 | { |
84e554e6 | 2268 | stat(s, CPUSLAB_FLUSH); |
c17dda40 CL |
2269 | deactivate_slab(s, c->page, c->freelist); |
2270 | ||
2271 | c->tid = next_tid(c->tid); | |
2272 | c->page = NULL; | |
2273 | c->freelist = NULL; | |
81819f0f CL |
2274 | } |
2275 | ||
2276 | /* | |
2277 | * Flush cpu slab. | |
6446faa2 | 2278 | * |
81819f0f CL |
2279 | * Called from IPI handler with interrupts disabled. |
2280 | */ | |
0c710013 | 2281 | static inline void __flush_cpu_slab(struct kmem_cache *s, int cpu) |
81819f0f | 2282 | { |
9dfc6e68 | 2283 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu); |
81819f0f | 2284 | |
49e22585 CL |
2285 | if (likely(c)) { |
2286 | if (c->page) | |
2287 | flush_slab(s, c); | |
2288 | ||
59a09917 | 2289 | unfreeze_partials(s, c); |
49e22585 | 2290 | } |
81819f0f CL |
2291 | } |
2292 | ||
2293 | static void flush_cpu_slab(void *d) | |
2294 | { | |
2295 | struct kmem_cache *s = d; | |
81819f0f | 2296 | |
dfb4f096 | 2297 | __flush_cpu_slab(s, smp_processor_id()); |
81819f0f CL |
2298 | } |
2299 | ||
a8364d55 GBY |
2300 | static bool has_cpu_slab(int cpu, void *info) |
2301 | { | |
2302 | struct kmem_cache *s = info; | |
2303 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu); | |
2304 | ||
02e1a9cd | 2305 | return c->page || c->partial; |
a8364d55 GBY |
2306 | } |
2307 | ||
81819f0f CL |
2308 | static void flush_all(struct kmem_cache *s) |
2309 | { | |
a8364d55 | 2310 | on_each_cpu_cond(has_cpu_slab, flush_cpu_slab, s, 1, GFP_ATOMIC); |
81819f0f CL |
2311 | } |
2312 | ||
a96a87bf SAS |
2313 | /* |
2314 | * Use the cpu notifier to insure that the cpu slabs are flushed when | |
2315 | * necessary. | |
2316 | */ | |
2317 | static int slub_cpu_dead(unsigned int cpu) | |
2318 | { | |
2319 | struct kmem_cache *s; | |
2320 | unsigned long flags; | |
2321 | ||
2322 | mutex_lock(&slab_mutex); | |
2323 | list_for_each_entry(s, &slab_caches, list) { | |
2324 | local_irq_save(flags); | |
2325 | __flush_cpu_slab(s, cpu); | |
2326 | local_irq_restore(flags); | |
2327 | } | |
2328 | mutex_unlock(&slab_mutex); | |
2329 | return 0; | |
2330 | } | |
2331 | ||
dfb4f096 CL |
2332 | /* |
2333 | * Check if the objects in a per cpu structure fit numa | |
2334 | * locality expectations. | |
2335 | */ | |
57d437d2 | 2336 | static inline int node_match(struct page *page, int node) |
dfb4f096 CL |
2337 | { |
2338 | #ifdef CONFIG_NUMA | |
4d7868e6 | 2339 | if (!page || (node != NUMA_NO_NODE && page_to_nid(page) != node)) |
dfb4f096 CL |
2340 | return 0; |
2341 | #endif | |
2342 | return 1; | |
2343 | } | |
2344 | ||
9a02d699 | 2345 | #ifdef CONFIG_SLUB_DEBUG |
781b2ba6 PE |
2346 | static int count_free(struct page *page) |
2347 | { | |
2348 | return page->objects - page->inuse; | |
2349 | } | |
2350 | ||
9a02d699 DR |
2351 | static inline unsigned long node_nr_objs(struct kmem_cache_node *n) |
2352 | { | |
2353 | return atomic_long_read(&n->total_objects); | |
2354 | } | |
2355 | #endif /* CONFIG_SLUB_DEBUG */ | |
2356 | ||
2357 | #if defined(CONFIG_SLUB_DEBUG) || defined(CONFIG_SYSFS) | |
781b2ba6 PE |
2358 | static unsigned long count_partial(struct kmem_cache_node *n, |
2359 | int (*get_count)(struct page *)) | |
2360 | { | |
2361 | unsigned long flags; | |
2362 | unsigned long x = 0; | |
2363 | struct page *page; | |
2364 | ||
2365 | spin_lock_irqsave(&n->list_lock, flags); | |
2366 | list_for_each_entry(page, &n->partial, lru) | |
2367 | x += get_count(page); | |
2368 | spin_unlock_irqrestore(&n->list_lock, flags); | |
2369 | return x; | |
2370 | } | |
9a02d699 | 2371 | #endif /* CONFIG_SLUB_DEBUG || CONFIG_SYSFS */ |
26c02cf0 | 2372 | |
781b2ba6 PE |
2373 | static noinline void |
2374 | slab_out_of_memory(struct kmem_cache *s, gfp_t gfpflags, int nid) | |
2375 | { | |
9a02d699 DR |
2376 | #ifdef CONFIG_SLUB_DEBUG |
2377 | static DEFINE_RATELIMIT_STATE(slub_oom_rs, DEFAULT_RATELIMIT_INTERVAL, | |
2378 | DEFAULT_RATELIMIT_BURST); | |
781b2ba6 | 2379 | int node; |
fa45dc25 | 2380 | struct kmem_cache_node *n; |
781b2ba6 | 2381 | |
9a02d699 DR |
2382 | if ((gfpflags & __GFP_NOWARN) || !__ratelimit(&slub_oom_rs)) |
2383 | return; | |
2384 | ||
5b3810e5 VB |
2385 | pr_warn("SLUB: Unable to allocate memory on node %d, gfp=%#x(%pGg)\n", |
2386 | nid, gfpflags, &gfpflags); | |
f9f58285 FF |
2387 | pr_warn(" cache: %s, object size: %d, buffer size: %d, default order: %d, min order: %d\n", |
2388 | s->name, s->object_size, s->size, oo_order(s->oo), | |
2389 | oo_order(s->min)); | |
781b2ba6 | 2390 | |
3b0efdfa | 2391 | if (oo_order(s->min) > get_order(s->object_size)) |
f9f58285 FF |
2392 | pr_warn(" %s debugging increased min order, use slub_debug=O to disable.\n", |
2393 | s->name); | |
fa5ec8a1 | 2394 | |
fa45dc25 | 2395 | for_each_kmem_cache_node(s, node, n) { |
781b2ba6 PE |
2396 | unsigned long nr_slabs; |
2397 | unsigned long nr_objs; | |
2398 | unsigned long nr_free; | |
2399 | ||
26c02cf0 AB |
2400 | nr_free = count_partial(n, count_free); |
2401 | nr_slabs = node_nr_slabs(n); | |
2402 | nr_objs = node_nr_objs(n); | |
781b2ba6 | 2403 | |
f9f58285 | 2404 | pr_warn(" node %d: slabs: %ld, objs: %ld, free: %ld\n", |
781b2ba6 PE |
2405 | node, nr_slabs, nr_objs, nr_free); |
2406 | } | |
9a02d699 | 2407 | #endif |
781b2ba6 PE |
2408 | } |
2409 | ||
497b66f2 CL |
2410 | static inline void *new_slab_objects(struct kmem_cache *s, gfp_t flags, |
2411 | int node, struct kmem_cache_cpu **pc) | |
2412 | { | |
6faa6833 | 2413 | void *freelist; |
188fd063 CL |
2414 | struct kmem_cache_cpu *c = *pc; |
2415 | struct page *page; | |
497b66f2 | 2416 | |
188fd063 | 2417 | freelist = get_partial(s, flags, node, c); |
497b66f2 | 2418 | |
188fd063 CL |
2419 | if (freelist) |
2420 | return freelist; | |
2421 | ||
2422 | page = new_slab(s, flags, node); | |
497b66f2 | 2423 | if (page) { |
7c8e0181 | 2424 | c = raw_cpu_ptr(s->cpu_slab); |
497b66f2 CL |
2425 | if (c->page) |
2426 | flush_slab(s, c); | |
2427 | ||
2428 | /* | |
2429 | * No other reference to the page yet so we can | |
2430 | * muck around with it freely without cmpxchg | |
2431 | */ | |
6faa6833 | 2432 | freelist = page->freelist; |
497b66f2 CL |
2433 | page->freelist = NULL; |
2434 | ||
2435 | stat(s, ALLOC_SLAB); | |
497b66f2 CL |
2436 | c->page = page; |
2437 | *pc = c; | |
2438 | } else | |
6faa6833 | 2439 | freelist = NULL; |
497b66f2 | 2440 | |
6faa6833 | 2441 | return freelist; |
497b66f2 CL |
2442 | } |
2443 | ||
072bb0aa MG |
2444 | static inline bool pfmemalloc_match(struct page *page, gfp_t gfpflags) |
2445 | { | |
2446 | if (unlikely(PageSlabPfmemalloc(page))) | |
2447 | return gfp_pfmemalloc_allowed(gfpflags); | |
2448 | ||
2449 | return true; | |
2450 | } | |
2451 | ||
213eeb9f | 2452 | /* |
d0e0ac97 CG |
2453 | * Check the page->freelist of a page and either transfer the freelist to the |
2454 | * per cpu freelist or deactivate the page. | |
213eeb9f CL |
2455 | * |
2456 | * The page is still frozen if the return value is not NULL. | |
2457 | * | |
2458 | * If this function returns NULL then the page has been unfrozen. | |
d24ac77f JK |
2459 | * |
2460 | * This function must be called with interrupt disabled. | |
213eeb9f CL |
2461 | */ |
2462 | static inline void *get_freelist(struct kmem_cache *s, struct page *page) | |
2463 | { | |
2464 | struct page new; | |
2465 | unsigned long counters; | |
2466 | void *freelist; | |
2467 | ||
2468 | do { | |
2469 | freelist = page->freelist; | |
2470 | counters = page->counters; | |
6faa6833 | 2471 | |
213eeb9f | 2472 | new.counters = counters; |
a0132ac0 | 2473 | VM_BUG_ON(!new.frozen); |
213eeb9f CL |
2474 | |
2475 | new.inuse = page->objects; | |
2476 | new.frozen = freelist != NULL; | |
2477 | ||
d24ac77f | 2478 | } while (!__cmpxchg_double_slab(s, page, |
213eeb9f CL |
2479 | freelist, counters, |
2480 | NULL, new.counters, | |
2481 | "get_freelist")); | |
2482 | ||
2483 | return freelist; | |
2484 | } | |
2485 | ||
81819f0f | 2486 | /* |
894b8788 CL |
2487 | * Slow path. The lockless freelist is empty or we need to perform |
2488 | * debugging duties. | |
2489 | * | |
894b8788 CL |
2490 | * Processing is still very fast if new objects have been freed to the |
2491 | * regular freelist. In that case we simply take over the regular freelist | |
2492 | * as the lockless freelist and zap the regular freelist. | |
81819f0f | 2493 | * |
894b8788 CL |
2494 | * If that is not working then we fall back to the partial lists. We take the |
2495 | * first element of the freelist as the object to allocate now and move the | |
2496 | * rest of the freelist to the lockless freelist. | |
81819f0f | 2497 | * |
894b8788 | 2498 | * And if we were unable to get a new slab from the partial slab lists then |
6446faa2 CL |
2499 | * we need to allocate a new slab. This is the slowest path since it involves |
2500 | * a call to the page allocator and the setup of a new slab. | |
a380a3c7 CL |
2501 | * |
2502 | * Version of __slab_alloc to use when we know that interrupts are | |
2503 | * already disabled (which is the case for bulk allocation). | |
81819f0f | 2504 | */ |
a380a3c7 | 2505 | static void *___slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node, |
ce71e27c | 2506 | unsigned long addr, struct kmem_cache_cpu *c) |
81819f0f | 2507 | { |
6faa6833 | 2508 | void *freelist; |
f6e7def7 | 2509 | struct page *page; |
81819f0f | 2510 | |
f6e7def7 CL |
2511 | page = c->page; |
2512 | if (!page) | |
81819f0f | 2513 | goto new_slab; |
49e22585 | 2514 | redo: |
6faa6833 | 2515 | |
57d437d2 | 2516 | if (unlikely(!node_match(page, node))) { |
a561ce00 JK |
2517 | int searchnode = node; |
2518 | ||
2519 | if (node != NUMA_NO_NODE && !node_present_pages(node)) | |
2520 | searchnode = node_to_mem_node(node); | |
2521 | ||
2522 | if (unlikely(!node_match(page, searchnode))) { | |
2523 | stat(s, ALLOC_NODE_MISMATCH); | |
2524 | deactivate_slab(s, page, c->freelist); | |
2525 | c->page = NULL; | |
2526 | c->freelist = NULL; | |
2527 | goto new_slab; | |
2528 | } | |
fc59c053 | 2529 | } |
6446faa2 | 2530 | |
072bb0aa MG |
2531 | /* |
2532 | * By rights, we should be searching for a slab page that was | |
2533 | * PFMEMALLOC but right now, we are losing the pfmemalloc | |
2534 | * information when the page leaves the per-cpu allocator | |
2535 | */ | |
2536 | if (unlikely(!pfmemalloc_match(page, gfpflags))) { | |
2537 | deactivate_slab(s, page, c->freelist); | |
2538 | c->page = NULL; | |
2539 | c->freelist = NULL; | |
2540 | goto new_slab; | |
2541 | } | |
2542 | ||
73736e03 | 2543 | /* must check again c->freelist in case of cpu migration or IRQ */ |
6faa6833 CL |
2544 | freelist = c->freelist; |
2545 | if (freelist) | |
73736e03 | 2546 | goto load_freelist; |
03e404af | 2547 | |
f6e7def7 | 2548 | freelist = get_freelist(s, page); |
6446faa2 | 2549 | |
6faa6833 | 2550 | if (!freelist) { |
03e404af CL |
2551 | c->page = NULL; |
2552 | stat(s, DEACTIVATE_BYPASS); | |
fc59c053 | 2553 | goto new_slab; |
03e404af | 2554 | } |
6446faa2 | 2555 | |
84e554e6 | 2556 | stat(s, ALLOC_REFILL); |
6446faa2 | 2557 | |
894b8788 | 2558 | load_freelist: |
507effea CL |
2559 | /* |
2560 | * freelist is pointing to the list of objects to be used. | |
2561 | * page is pointing to the page from which the objects are obtained. | |
2562 | * That page must be frozen for per cpu allocations to work. | |
2563 | */ | |
a0132ac0 | 2564 | VM_BUG_ON(!c->page->frozen); |
6faa6833 | 2565 | c->freelist = get_freepointer(s, freelist); |
8a5ec0ba | 2566 | c->tid = next_tid(c->tid); |
6faa6833 | 2567 | return freelist; |
81819f0f | 2568 | |
81819f0f | 2569 | new_slab: |
2cfb7455 | 2570 | |
49e22585 | 2571 | if (c->partial) { |
f6e7def7 CL |
2572 | page = c->page = c->partial; |
2573 | c->partial = page->next; | |
49e22585 | 2574 | stat(s, CPU_PARTIAL_ALLOC); |
49e22585 | 2575 | goto redo; |
81819f0f CL |
2576 | } |
2577 | ||
188fd063 | 2578 | freelist = new_slab_objects(s, gfpflags, node, &c); |
01ad8a7b | 2579 | |
f4697436 | 2580 | if (unlikely(!freelist)) { |
9a02d699 | 2581 | slab_out_of_memory(s, gfpflags, node); |
f4697436 | 2582 | return NULL; |
81819f0f | 2583 | } |
2cfb7455 | 2584 | |
f6e7def7 | 2585 | page = c->page; |
5091b74a | 2586 | if (likely(!kmem_cache_debug(s) && pfmemalloc_match(page, gfpflags))) |
4b6f0750 | 2587 | goto load_freelist; |
2cfb7455 | 2588 | |
497b66f2 | 2589 | /* Only entered in the debug case */ |
d0e0ac97 CG |
2590 | if (kmem_cache_debug(s) && |
2591 | !alloc_debug_processing(s, page, freelist, addr)) | |
497b66f2 | 2592 | goto new_slab; /* Slab failed checks. Next slab needed */ |
894b8788 | 2593 | |
f6e7def7 | 2594 | deactivate_slab(s, page, get_freepointer(s, freelist)); |
c17dda40 CL |
2595 | c->page = NULL; |
2596 | c->freelist = NULL; | |
6faa6833 | 2597 | return freelist; |
894b8788 CL |
2598 | } |
2599 | ||
a380a3c7 CL |
2600 | /* |
2601 | * Another one that disabled interrupt and compensates for possible | |
2602 | * cpu changes by refetching the per cpu area pointer. | |
2603 | */ | |
2604 | static void *__slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node, | |
2605 | unsigned long addr, struct kmem_cache_cpu *c) | |
2606 | { | |
2607 | void *p; | |
2608 | unsigned long flags; | |
2609 | ||
2610 | local_irq_save(flags); | |
2611 | #ifdef CONFIG_PREEMPT | |
2612 | /* | |
2613 | * We may have been preempted and rescheduled on a different | |
2614 | * cpu before disabling interrupts. Need to reload cpu area | |
2615 | * pointer. | |
2616 | */ | |
2617 | c = this_cpu_ptr(s->cpu_slab); | |
2618 | #endif | |
2619 | ||
2620 | p = ___slab_alloc(s, gfpflags, node, addr, c); | |
2621 | local_irq_restore(flags); | |
2622 | return p; | |
2623 | } | |
2624 | ||
894b8788 CL |
2625 | /* |
2626 | * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc) | |
2627 | * have the fastpath folded into their functions. So no function call | |
2628 | * overhead for requests that can be satisfied on the fastpath. | |
2629 | * | |
2630 | * The fastpath works by first checking if the lockless freelist can be used. | |
2631 | * If not then __slab_alloc is called for slow processing. | |
2632 | * | |
2633 | * Otherwise we can simply pick the next object from the lockless free list. | |
2634 | */ | |
2b847c3c | 2635 | static __always_inline void *slab_alloc_node(struct kmem_cache *s, |
ce71e27c | 2636 | gfp_t gfpflags, int node, unsigned long addr) |
894b8788 | 2637 | { |
03ec0ed5 | 2638 | void *object; |
dfb4f096 | 2639 | struct kmem_cache_cpu *c; |
57d437d2 | 2640 | struct page *page; |
8a5ec0ba | 2641 | unsigned long tid; |
1f84260c | 2642 | |
8135be5a VD |
2643 | s = slab_pre_alloc_hook(s, gfpflags); |
2644 | if (!s) | |
773ff60e | 2645 | return NULL; |
8a5ec0ba | 2646 | redo: |
8a5ec0ba CL |
2647 | /* |
2648 | * Must read kmem_cache cpu data via this cpu ptr. Preemption is | |
2649 | * enabled. We may switch back and forth between cpus while | |
2650 | * reading from one cpu area. That does not matter as long | |
2651 | * as we end up on the original cpu again when doing the cmpxchg. | |
7cccd80b | 2652 | * |
9aabf810 JK |
2653 | * We should guarantee that tid and kmem_cache are retrieved on |
2654 | * the same cpu. It could be different if CONFIG_PREEMPT so we need | |
2655 | * to check if it is matched or not. | |
8a5ec0ba | 2656 | */ |
9aabf810 JK |
2657 | do { |
2658 | tid = this_cpu_read(s->cpu_slab->tid); | |
2659 | c = raw_cpu_ptr(s->cpu_slab); | |
859b7a0e MR |
2660 | } while (IS_ENABLED(CONFIG_PREEMPT) && |
2661 | unlikely(tid != READ_ONCE(c->tid))); | |
9aabf810 JK |
2662 | |
2663 | /* | |
2664 | * Irqless object alloc/free algorithm used here depends on sequence | |
2665 | * of fetching cpu_slab's data. tid should be fetched before anything | |
2666 | * on c to guarantee that object and page associated with previous tid | |
2667 | * won't be used with current tid. If we fetch tid first, object and | |
2668 | * page could be one associated with next tid and our alloc/free | |
2669 | * request will be failed. In this case, we will retry. So, no problem. | |
2670 | */ | |
2671 | barrier(); | |
8a5ec0ba | 2672 | |
8a5ec0ba CL |
2673 | /* |
2674 | * The transaction ids are globally unique per cpu and per operation on | |
2675 | * a per cpu queue. Thus they can be guarantee that the cmpxchg_double | |
2676 | * occurs on the right processor and that there was no operation on the | |
2677 | * linked list in between. | |
2678 | */ | |
8a5ec0ba | 2679 | |
9dfc6e68 | 2680 | object = c->freelist; |
57d437d2 | 2681 | page = c->page; |
8eae1492 | 2682 | if (unlikely(!object || !node_match(page, node))) { |
dfb4f096 | 2683 | object = __slab_alloc(s, gfpflags, node, addr, c); |
8eae1492 DH |
2684 | stat(s, ALLOC_SLOWPATH); |
2685 | } else { | |
0ad9500e ED |
2686 | void *next_object = get_freepointer_safe(s, object); |
2687 | ||
8a5ec0ba | 2688 | /* |
25985edc | 2689 | * The cmpxchg will only match if there was no additional |
8a5ec0ba CL |
2690 | * operation and if we are on the right processor. |
2691 | * | |
d0e0ac97 CG |
2692 | * The cmpxchg does the following atomically (without lock |
2693 | * semantics!) | |
8a5ec0ba CL |
2694 | * 1. Relocate first pointer to the current per cpu area. |
2695 | * 2. Verify that tid and freelist have not been changed | |
2696 | * 3. If they were not changed replace tid and freelist | |
2697 | * | |
d0e0ac97 CG |
2698 | * Since this is without lock semantics the protection is only |
2699 | * against code executing on this cpu *not* from access by | |
2700 | * other cpus. | |
8a5ec0ba | 2701 | */ |
933393f5 | 2702 | if (unlikely(!this_cpu_cmpxchg_double( |
8a5ec0ba CL |
2703 | s->cpu_slab->freelist, s->cpu_slab->tid, |
2704 | object, tid, | |
0ad9500e | 2705 | next_object, next_tid(tid)))) { |
8a5ec0ba CL |
2706 | |
2707 | note_cmpxchg_failure("slab_alloc", s, tid); | |
2708 | goto redo; | |
2709 | } | |
0ad9500e | 2710 | prefetch_freepointer(s, next_object); |
84e554e6 | 2711 | stat(s, ALLOC_FASTPATH); |
894b8788 | 2712 | } |
8a5ec0ba | 2713 | |
74e2134f | 2714 | if (unlikely(gfpflags & __GFP_ZERO) && object) |
3b0efdfa | 2715 | memset(object, 0, s->object_size); |
d07dbea4 | 2716 | |
03ec0ed5 | 2717 | slab_post_alloc_hook(s, gfpflags, 1, &object); |
5a896d9e | 2718 | |
894b8788 | 2719 | return object; |
81819f0f CL |
2720 | } |
2721 | ||
2b847c3c EG |
2722 | static __always_inline void *slab_alloc(struct kmem_cache *s, |
2723 | gfp_t gfpflags, unsigned long addr) | |
2724 | { | |
2725 | return slab_alloc_node(s, gfpflags, NUMA_NO_NODE, addr); | |
2726 | } | |
2727 | ||
81819f0f CL |
2728 | void *kmem_cache_alloc(struct kmem_cache *s, gfp_t gfpflags) |
2729 | { | |
2b847c3c | 2730 | void *ret = slab_alloc(s, gfpflags, _RET_IP_); |
5b882be4 | 2731 | |
d0e0ac97 CG |
2732 | trace_kmem_cache_alloc(_RET_IP_, ret, s->object_size, |
2733 | s->size, gfpflags); | |
5b882be4 EGM |
2734 | |
2735 | return ret; | |
81819f0f CL |
2736 | } |
2737 | EXPORT_SYMBOL(kmem_cache_alloc); | |
2738 | ||
0f24f128 | 2739 | #ifdef CONFIG_TRACING |
4a92379b RK |
2740 | void *kmem_cache_alloc_trace(struct kmem_cache *s, gfp_t gfpflags, size_t size) |
2741 | { | |
2b847c3c | 2742 | void *ret = slab_alloc(s, gfpflags, _RET_IP_); |
4a92379b | 2743 | trace_kmalloc(_RET_IP_, ret, size, s->size, gfpflags); |
505f5dcb | 2744 | kasan_kmalloc(s, ret, size, gfpflags); |
4a92379b RK |
2745 | return ret; |
2746 | } | |
2747 | EXPORT_SYMBOL(kmem_cache_alloc_trace); | |
5b882be4 EGM |
2748 | #endif |
2749 | ||
81819f0f CL |
2750 | #ifdef CONFIG_NUMA |
2751 | void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t gfpflags, int node) | |
2752 | { | |
2b847c3c | 2753 | void *ret = slab_alloc_node(s, gfpflags, node, _RET_IP_); |
5b882be4 | 2754 | |
ca2b84cb | 2755 | trace_kmem_cache_alloc_node(_RET_IP_, ret, |
3b0efdfa | 2756 | s->object_size, s->size, gfpflags, node); |
5b882be4 EGM |
2757 | |
2758 | return ret; | |
81819f0f CL |
2759 | } |
2760 | EXPORT_SYMBOL(kmem_cache_alloc_node); | |
81819f0f | 2761 | |
0f24f128 | 2762 | #ifdef CONFIG_TRACING |
4a92379b | 2763 | void *kmem_cache_alloc_node_trace(struct kmem_cache *s, |
5b882be4 | 2764 | gfp_t gfpflags, |
4a92379b | 2765 | int node, size_t size) |
5b882be4 | 2766 | { |
2b847c3c | 2767 | void *ret = slab_alloc_node(s, gfpflags, node, _RET_IP_); |
4a92379b RK |
2768 | |
2769 | trace_kmalloc_node(_RET_IP_, ret, | |
2770 | size, s->size, gfpflags, node); | |
0316bec2 | 2771 | |
505f5dcb | 2772 | kasan_kmalloc(s, ret, size, gfpflags); |
4a92379b | 2773 | return ret; |
5b882be4 | 2774 | } |
4a92379b | 2775 | EXPORT_SYMBOL(kmem_cache_alloc_node_trace); |
5b882be4 | 2776 | #endif |
5d1f57e4 | 2777 | #endif |
5b882be4 | 2778 | |
81819f0f | 2779 | /* |
94e4d712 | 2780 | * Slow path handling. This may still be called frequently since objects |
894b8788 | 2781 | * have a longer lifetime than the cpu slabs in most processing loads. |
81819f0f | 2782 | * |
894b8788 CL |
2783 | * So we still attempt to reduce cache line usage. Just take the slab |
2784 | * lock and free the item. If there is no additional partial page | |
2785 | * handling required then we can return immediately. | |
81819f0f | 2786 | */ |
894b8788 | 2787 | static void __slab_free(struct kmem_cache *s, struct page *page, |
81084651 JDB |
2788 | void *head, void *tail, int cnt, |
2789 | unsigned long addr) | |
2790 | ||
81819f0f CL |
2791 | { |
2792 | void *prior; | |
2cfb7455 | 2793 | int was_frozen; |
2cfb7455 CL |
2794 | struct page new; |
2795 | unsigned long counters; | |
2796 | struct kmem_cache_node *n = NULL; | |
61728d1e | 2797 | unsigned long uninitialized_var(flags); |
81819f0f | 2798 | |
8a5ec0ba | 2799 | stat(s, FREE_SLOWPATH); |
81819f0f | 2800 | |
19c7ff9e | 2801 | if (kmem_cache_debug(s) && |
282acb43 | 2802 | !free_debug_processing(s, page, head, tail, cnt, addr)) |
80f08c19 | 2803 | return; |
6446faa2 | 2804 | |
2cfb7455 | 2805 | do { |
837d678d JK |
2806 | if (unlikely(n)) { |
2807 | spin_unlock_irqrestore(&n->list_lock, flags); | |
2808 | n = NULL; | |
2809 | } | |
2cfb7455 CL |
2810 | prior = page->freelist; |
2811 | counters = page->counters; | |
81084651 | 2812 | set_freepointer(s, tail, prior); |
2cfb7455 CL |
2813 | new.counters = counters; |
2814 | was_frozen = new.frozen; | |
81084651 | 2815 | new.inuse -= cnt; |
837d678d | 2816 | if ((!new.inuse || !prior) && !was_frozen) { |
49e22585 | 2817 | |
c65c1877 | 2818 | if (kmem_cache_has_cpu_partial(s) && !prior) { |
49e22585 CL |
2819 | |
2820 | /* | |
d0e0ac97 CG |
2821 | * Slab was on no list before and will be |
2822 | * partially empty | |
2823 | * We can defer the list move and instead | |
2824 | * freeze it. | |
49e22585 CL |
2825 | */ |
2826 | new.frozen = 1; | |
2827 | ||
c65c1877 | 2828 | } else { /* Needs to be taken off a list */ |
49e22585 | 2829 | |
b455def2 | 2830 | n = get_node(s, page_to_nid(page)); |
49e22585 CL |
2831 | /* |
2832 | * Speculatively acquire the list_lock. | |
2833 | * If the cmpxchg does not succeed then we may | |
2834 | * drop the list_lock without any processing. | |
2835 | * | |
2836 | * Otherwise the list_lock will synchronize with | |
2837 | * other processors updating the list of slabs. | |
2838 | */ | |
2839 | spin_lock_irqsave(&n->list_lock, flags); | |
2840 | ||
2841 | } | |
2cfb7455 | 2842 | } |
81819f0f | 2843 | |
2cfb7455 CL |
2844 | } while (!cmpxchg_double_slab(s, page, |
2845 | prior, counters, | |
81084651 | 2846 | head, new.counters, |
2cfb7455 | 2847 | "__slab_free")); |
81819f0f | 2848 | |
2cfb7455 | 2849 | if (likely(!n)) { |
49e22585 CL |
2850 | |
2851 | /* | |
2852 | * If we just froze the page then put it onto the | |
2853 | * per cpu partial list. | |
2854 | */ | |
8028dcea | 2855 | if (new.frozen && !was_frozen) { |
49e22585 | 2856 | put_cpu_partial(s, page, 1); |
8028dcea AS |
2857 | stat(s, CPU_PARTIAL_FREE); |
2858 | } | |
49e22585 | 2859 | /* |
2cfb7455 CL |
2860 | * The list lock was not taken therefore no list |
2861 | * activity can be necessary. | |
2862 | */ | |
b455def2 L |
2863 | if (was_frozen) |
2864 | stat(s, FREE_FROZEN); | |
2865 | return; | |
2866 | } | |
81819f0f | 2867 | |
8a5b20ae | 2868 | if (unlikely(!new.inuse && n->nr_partial >= s->min_partial)) |
837d678d JK |
2869 | goto slab_empty; |
2870 | ||
81819f0f | 2871 | /* |
837d678d JK |
2872 | * Objects left in the slab. If it was not on the partial list before |
2873 | * then add it. | |
81819f0f | 2874 | */ |
345c905d JK |
2875 | if (!kmem_cache_has_cpu_partial(s) && unlikely(!prior)) { |
2876 | if (kmem_cache_debug(s)) | |
c65c1877 | 2877 | remove_full(s, n, page); |
837d678d JK |
2878 | add_partial(n, page, DEACTIVATE_TO_TAIL); |
2879 | stat(s, FREE_ADD_PARTIAL); | |
8ff12cfc | 2880 | } |
80f08c19 | 2881 | spin_unlock_irqrestore(&n->list_lock, flags); |
81819f0f CL |
2882 | return; |
2883 | ||
2884 | slab_empty: | |
a973e9dd | 2885 | if (prior) { |
81819f0f | 2886 | /* |
6fbabb20 | 2887 | * Slab on the partial list. |
81819f0f | 2888 | */ |
5cc6eee8 | 2889 | remove_partial(n, page); |
84e554e6 | 2890 | stat(s, FREE_REMOVE_PARTIAL); |
c65c1877 | 2891 | } else { |
6fbabb20 | 2892 | /* Slab must be on the full list */ |
c65c1877 PZ |
2893 | remove_full(s, n, page); |
2894 | } | |
2cfb7455 | 2895 | |
80f08c19 | 2896 | spin_unlock_irqrestore(&n->list_lock, flags); |
84e554e6 | 2897 | stat(s, FREE_SLAB); |
81819f0f | 2898 | discard_slab(s, page); |
81819f0f CL |
2899 | } |
2900 | ||
894b8788 CL |
2901 | /* |
2902 | * Fastpath with forced inlining to produce a kfree and kmem_cache_free that | |
2903 | * can perform fastpath freeing without additional function calls. | |
2904 | * | |
2905 | * The fastpath is only possible if we are freeing to the current cpu slab | |
2906 | * of this processor. This typically the case if we have just allocated | |
2907 | * the item before. | |
2908 | * | |
2909 | * If fastpath is not possible then fall back to __slab_free where we deal | |
2910 | * with all sorts of special processing. | |
81084651 JDB |
2911 | * |
2912 | * Bulk free of a freelist with several objects (all pointing to the | |
2913 | * same page) possible by specifying head and tail ptr, plus objects | |
2914 | * count (cnt). Bulk free indicated by tail pointer being set. | |
894b8788 | 2915 | */ |
80a9201a AP |
2916 | static __always_inline void do_slab_free(struct kmem_cache *s, |
2917 | struct page *page, void *head, void *tail, | |
2918 | int cnt, unsigned long addr) | |
894b8788 | 2919 | { |
81084651 | 2920 | void *tail_obj = tail ? : head; |
dfb4f096 | 2921 | struct kmem_cache_cpu *c; |
8a5ec0ba | 2922 | unsigned long tid; |
8a5ec0ba CL |
2923 | redo: |
2924 | /* | |
2925 | * Determine the currently cpus per cpu slab. | |
2926 | * The cpu may change afterward. However that does not matter since | |
2927 | * data is retrieved via this pointer. If we are on the same cpu | |
2ae44005 | 2928 | * during the cmpxchg then the free will succeed. |
8a5ec0ba | 2929 | */ |
9aabf810 JK |
2930 | do { |
2931 | tid = this_cpu_read(s->cpu_slab->tid); | |
2932 | c = raw_cpu_ptr(s->cpu_slab); | |
859b7a0e MR |
2933 | } while (IS_ENABLED(CONFIG_PREEMPT) && |
2934 | unlikely(tid != READ_ONCE(c->tid))); | |
c016b0bd | 2935 | |
9aabf810 JK |
2936 | /* Same with comment on barrier() in slab_alloc_node() */ |
2937 | barrier(); | |
c016b0bd | 2938 | |
442b06bc | 2939 | if (likely(page == c->page)) { |
81084651 | 2940 | set_freepointer(s, tail_obj, c->freelist); |
8a5ec0ba | 2941 | |
933393f5 | 2942 | if (unlikely(!this_cpu_cmpxchg_double( |
8a5ec0ba CL |
2943 | s->cpu_slab->freelist, s->cpu_slab->tid, |
2944 | c->freelist, tid, | |
81084651 | 2945 | head, next_tid(tid)))) { |
8a5ec0ba CL |
2946 | |
2947 | note_cmpxchg_failure("slab_free", s, tid); | |
2948 | goto redo; | |
2949 | } | |
84e554e6 | 2950 | stat(s, FREE_FASTPATH); |
894b8788 | 2951 | } else |
81084651 | 2952 | __slab_free(s, page, head, tail_obj, cnt, addr); |
894b8788 | 2953 | |
894b8788 CL |
2954 | } |
2955 | ||
80a9201a AP |
2956 | static __always_inline void slab_free(struct kmem_cache *s, struct page *page, |
2957 | void *head, void *tail, int cnt, | |
2958 | unsigned long addr) | |
2959 | { | |
2960 | slab_free_freelist_hook(s, head, tail); | |
2961 | /* | |
2962 | * slab_free_freelist_hook() could have put the items into quarantine. | |
2963 | * If so, no need to free them. | |
2964 | */ | |
5f0d5a3a | 2965 | if (s->flags & SLAB_KASAN && !(s->flags & SLAB_TYPESAFE_BY_RCU)) |
80a9201a AP |
2966 | return; |
2967 | do_slab_free(s, page, head, tail, cnt, addr); | |
2968 | } | |
2969 | ||
2970 | #ifdef CONFIG_KASAN | |
2971 | void ___cache_free(struct kmem_cache *cache, void *x, unsigned long addr) | |
2972 | { | |
2973 | do_slab_free(cache, virt_to_head_page(x), x, NULL, 1, addr); | |
2974 | } | |
2975 | #endif | |
2976 | ||
81819f0f CL |
2977 | void kmem_cache_free(struct kmem_cache *s, void *x) |
2978 | { | |
b9ce5ef4 GC |
2979 | s = cache_from_obj(s, x); |
2980 | if (!s) | |
79576102 | 2981 | return; |
81084651 | 2982 | slab_free(s, virt_to_head_page(x), x, NULL, 1, _RET_IP_); |
ca2b84cb | 2983 | trace_kmem_cache_free(_RET_IP_, x); |
81819f0f CL |
2984 | } |
2985 | EXPORT_SYMBOL(kmem_cache_free); | |
2986 | ||
d0ecd894 | 2987 | struct detached_freelist { |
fbd02630 | 2988 | struct page *page; |
d0ecd894 JDB |
2989 | void *tail; |
2990 | void *freelist; | |
2991 | int cnt; | |
376bf125 | 2992 | struct kmem_cache *s; |
d0ecd894 | 2993 | }; |
fbd02630 | 2994 | |
d0ecd894 JDB |
2995 | /* |
2996 | * This function progressively scans the array with free objects (with | |
2997 | * a limited look ahead) and extract objects belonging to the same | |
2998 | * page. It builds a detached freelist directly within the given | |
2999 | * page/objects. This can happen without any need for | |
3000 | * synchronization, because the objects are owned by running process. | |
3001 | * The freelist is build up as a single linked list in the objects. | |
3002 | * The idea is, that this detached freelist can then be bulk | |
3003 | * transferred to the real freelist(s), but only requiring a single | |
3004 | * synchronization primitive. Look ahead in the array is limited due | |
3005 | * to performance reasons. | |
3006 | */ | |
376bf125 JDB |
3007 | static inline |
3008 | int build_detached_freelist(struct kmem_cache *s, size_t size, | |
3009 | void **p, struct detached_freelist *df) | |
d0ecd894 JDB |
3010 | { |
3011 | size_t first_skipped_index = 0; | |
3012 | int lookahead = 3; | |
3013 | void *object; | |
ca257195 | 3014 | struct page *page; |
fbd02630 | 3015 | |
d0ecd894 JDB |
3016 | /* Always re-init detached_freelist */ |
3017 | df->page = NULL; | |
fbd02630 | 3018 | |
d0ecd894 JDB |
3019 | do { |
3020 | object = p[--size]; | |
ca257195 | 3021 | /* Do we need !ZERO_OR_NULL_PTR(object) here? (for kfree) */ |
d0ecd894 | 3022 | } while (!object && size); |
3eed034d | 3023 | |
d0ecd894 JDB |
3024 | if (!object) |
3025 | return 0; | |
fbd02630 | 3026 | |
ca257195 JDB |
3027 | page = virt_to_head_page(object); |
3028 | if (!s) { | |
3029 | /* Handle kalloc'ed objects */ | |
3030 | if (unlikely(!PageSlab(page))) { | |
3031 | BUG_ON(!PageCompound(page)); | |
3032 | kfree_hook(object); | |
4949148a | 3033 | __free_pages(page, compound_order(page)); |
ca257195 JDB |
3034 | p[size] = NULL; /* mark object processed */ |
3035 | return size; | |
3036 | } | |
3037 | /* Derive kmem_cache from object */ | |
3038 | df->s = page->slab_cache; | |
3039 | } else { | |
3040 | df->s = cache_from_obj(s, object); /* Support for memcg */ | |
3041 | } | |
376bf125 | 3042 | |
d0ecd894 | 3043 | /* Start new detached freelist */ |
ca257195 | 3044 | df->page = page; |
376bf125 | 3045 | set_freepointer(df->s, object, NULL); |
d0ecd894 JDB |
3046 | df->tail = object; |
3047 | df->freelist = object; | |
3048 | p[size] = NULL; /* mark object processed */ | |
3049 | df->cnt = 1; | |
3050 | ||
3051 | while (size) { | |
3052 | object = p[--size]; | |
3053 | if (!object) | |
3054 | continue; /* Skip processed objects */ | |
3055 | ||
3056 | /* df->page is always set at this point */ | |
3057 | if (df->page == virt_to_head_page(object)) { | |
3058 | /* Opportunity build freelist */ | |
376bf125 | 3059 | set_freepointer(df->s, object, df->freelist); |
d0ecd894 JDB |
3060 | df->freelist = object; |
3061 | df->cnt++; | |
3062 | p[size] = NULL; /* mark object processed */ | |
3063 | ||
3064 | continue; | |
fbd02630 | 3065 | } |
d0ecd894 JDB |
3066 | |
3067 | /* Limit look ahead search */ | |
3068 | if (!--lookahead) | |
3069 | break; | |
3070 | ||
3071 | if (!first_skipped_index) | |
3072 | first_skipped_index = size + 1; | |
fbd02630 | 3073 | } |
d0ecd894 JDB |
3074 | |
3075 | return first_skipped_index; | |
3076 | } | |
3077 | ||
d0ecd894 | 3078 | /* Note that interrupts must be enabled when calling this function. */ |
376bf125 | 3079 | void kmem_cache_free_bulk(struct kmem_cache *s, size_t size, void **p) |
d0ecd894 JDB |
3080 | { |
3081 | if (WARN_ON(!size)) | |
3082 | return; | |
3083 | ||
3084 | do { | |
3085 | struct detached_freelist df; | |
3086 | ||
3087 | size = build_detached_freelist(s, size, p, &df); | |
84582c8a | 3088 | if (!df.page) |
d0ecd894 JDB |
3089 | continue; |
3090 | ||
376bf125 | 3091 | slab_free(df.s, df.page, df.freelist, df.tail, df.cnt,_RET_IP_); |
d0ecd894 | 3092 | } while (likely(size)); |
484748f0 CL |
3093 | } |
3094 | EXPORT_SYMBOL(kmem_cache_free_bulk); | |
3095 | ||
994eb764 | 3096 | /* Note that interrupts must be enabled when calling this function. */ |
865762a8 JDB |
3097 | int kmem_cache_alloc_bulk(struct kmem_cache *s, gfp_t flags, size_t size, |
3098 | void **p) | |
484748f0 | 3099 | { |
994eb764 JDB |
3100 | struct kmem_cache_cpu *c; |
3101 | int i; | |
3102 | ||
03ec0ed5 JDB |
3103 | /* memcg and kmem_cache debug support */ |
3104 | s = slab_pre_alloc_hook(s, flags); | |
3105 | if (unlikely(!s)) | |
3106 | return false; | |
994eb764 JDB |
3107 | /* |
3108 | * Drain objects in the per cpu slab, while disabling local | |
3109 | * IRQs, which protects against PREEMPT and interrupts | |
3110 | * handlers invoking normal fastpath. | |
3111 | */ | |
3112 | local_irq_disable(); | |
3113 | c = this_cpu_ptr(s->cpu_slab); | |
3114 | ||
3115 | for (i = 0; i < size; i++) { | |
3116 | void *object = c->freelist; | |
3117 | ||
ebe909e0 | 3118 | if (unlikely(!object)) { |
ebe909e0 JDB |
3119 | /* |
3120 | * Invoking slow path likely have side-effect | |
3121 | * of re-populating per CPU c->freelist | |
3122 | */ | |
87098373 | 3123 | p[i] = ___slab_alloc(s, flags, NUMA_NO_NODE, |
ebe909e0 | 3124 | _RET_IP_, c); |
87098373 CL |
3125 | if (unlikely(!p[i])) |
3126 | goto error; | |
3127 | ||
ebe909e0 JDB |
3128 | c = this_cpu_ptr(s->cpu_slab); |
3129 | continue; /* goto for-loop */ | |
3130 | } | |
994eb764 JDB |
3131 | c->freelist = get_freepointer(s, object); |
3132 | p[i] = object; | |
3133 | } | |
3134 | c->tid = next_tid(c->tid); | |
3135 | local_irq_enable(); | |
3136 | ||
3137 | /* Clear memory outside IRQ disabled fastpath loop */ | |
3138 | if (unlikely(flags & __GFP_ZERO)) { | |
3139 | int j; | |
3140 | ||
3141 | for (j = 0; j < i; j++) | |
3142 | memset(p[j], 0, s->object_size); | |
3143 | } | |
3144 | ||
03ec0ed5 JDB |
3145 | /* memcg and kmem_cache debug support */ |
3146 | slab_post_alloc_hook(s, flags, size, p); | |
865762a8 | 3147 | return i; |
87098373 | 3148 | error: |
87098373 | 3149 | local_irq_enable(); |
03ec0ed5 JDB |
3150 | slab_post_alloc_hook(s, flags, i, p); |
3151 | __kmem_cache_free_bulk(s, i, p); | |
865762a8 | 3152 | return 0; |
484748f0 CL |
3153 | } |
3154 | EXPORT_SYMBOL(kmem_cache_alloc_bulk); | |
3155 | ||
3156 | ||
81819f0f | 3157 | /* |
672bba3a CL |
3158 | * Object placement in a slab is made very easy because we always start at |
3159 | * offset 0. If we tune the size of the object to the alignment then we can | |
3160 | * get the required alignment by putting one properly sized object after | |
3161 | * another. | |
81819f0f CL |
3162 | * |
3163 | * Notice that the allocation order determines the sizes of the per cpu | |
3164 | * caches. Each processor has always one slab available for allocations. | |
3165 | * Increasing the allocation order reduces the number of times that slabs | |
672bba3a | 3166 | * must be moved on and off the partial lists and is therefore a factor in |
81819f0f | 3167 | * locking overhead. |
81819f0f CL |
3168 | */ |
3169 | ||
3170 | /* | |
3171 | * Mininum / Maximum order of slab pages. This influences locking overhead | |
3172 | * and slab fragmentation. A higher order reduces the number of partial slabs | |
3173 | * and increases the number of allocations possible without having to | |
3174 | * take the list_lock. | |
3175 | */ | |
3176 | static int slub_min_order; | |
114e9e89 | 3177 | static int slub_max_order = PAGE_ALLOC_COSTLY_ORDER; |
9b2cd506 | 3178 | static int slub_min_objects; |
81819f0f | 3179 | |
81819f0f CL |
3180 | /* |
3181 | * Calculate the order of allocation given an slab object size. | |
3182 | * | |
672bba3a CL |
3183 | * The order of allocation has significant impact on performance and other |
3184 | * system components. Generally order 0 allocations should be preferred since | |
3185 | * order 0 does not cause fragmentation in the page allocator. Larger objects | |
3186 | * be problematic to put into order 0 slabs because there may be too much | |
c124f5b5 | 3187 | * unused space left. We go to a higher order if more than 1/16th of the slab |
672bba3a CL |
3188 | * would be wasted. |
3189 | * | |
3190 | * In order to reach satisfactory performance we must ensure that a minimum | |
3191 | * number of objects is in one slab. Otherwise we may generate too much | |
3192 | * activity on the partial lists which requires taking the list_lock. This is | |
3193 | * less a concern for large slabs though which are rarely used. | |
81819f0f | 3194 | * |
672bba3a CL |
3195 | * slub_max_order specifies the order where we begin to stop considering the |
3196 | * number of objects in a slab as critical. If we reach slub_max_order then | |
3197 | * we try to keep the page order as low as possible. So we accept more waste | |
3198 | * of space in favor of a small page order. | |
81819f0f | 3199 | * |
672bba3a CL |
3200 | * Higher order allocations also allow the placement of more objects in a |
3201 | * slab and thereby reduce object handling overhead. If the user has | |
3202 | * requested a higher mininum order then we start with that one instead of | |
3203 | * the smallest order which will fit the object. | |
81819f0f | 3204 | */ |
5e6d444e | 3205 | static inline int slab_order(int size, int min_objects, |
ab9a0f19 | 3206 | int max_order, int fract_leftover, int reserved) |
81819f0f CL |
3207 | { |
3208 | int order; | |
3209 | int rem; | |
6300ea75 | 3210 | int min_order = slub_min_order; |
81819f0f | 3211 | |
ab9a0f19 | 3212 | if (order_objects(min_order, size, reserved) > MAX_OBJS_PER_PAGE) |
210b5c06 | 3213 | return get_order(size * MAX_OBJS_PER_PAGE) - 1; |
39b26464 | 3214 | |
9f835703 | 3215 | for (order = max(min_order, get_order(min_objects * size + reserved)); |
5e6d444e | 3216 | order <= max_order; order++) { |
81819f0f | 3217 | |
5e6d444e | 3218 | unsigned long slab_size = PAGE_SIZE << order; |
81819f0f | 3219 | |
ab9a0f19 | 3220 | rem = (slab_size - reserved) % size; |
81819f0f | 3221 | |
5e6d444e | 3222 | if (rem <= slab_size / fract_leftover) |
81819f0f | 3223 | break; |
81819f0f | 3224 | } |
672bba3a | 3225 | |
81819f0f CL |
3226 | return order; |
3227 | } | |
3228 | ||
ab9a0f19 | 3229 | static inline int calculate_order(int size, int reserved) |
5e6d444e CL |
3230 | { |
3231 | int order; | |
3232 | int min_objects; | |
3233 | int fraction; | |
e8120ff1 | 3234 | int max_objects; |
5e6d444e CL |
3235 | |
3236 | /* | |
3237 | * Attempt to find best configuration for a slab. This | |
3238 | * works by first attempting to generate a layout with | |
3239 | * the best configuration and backing off gradually. | |
3240 | * | |
422ff4d7 | 3241 | * First we increase the acceptable waste in a slab. Then |
5e6d444e CL |
3242 | * we reduce the minimum objects required in a slab. |
3243 | */ | |
3244 | min_objects = slub_min_objects; | |
9b2cd506 CL |
3245 | if (!min_objects) |
3246 | min_objects = 4 * (fls(nr_cpu_ids) + 1); | |
ab9a0f19 | 3247 | max_objects = order_objects(slub_max_order, size, reserved); |
e8120ff1 ZY |
3248 | min_objects = min(min_objects, max_objects); |
3249 | ||
5e6d444e | 3250 | while (min_objects > 1) { |
c124f5b5 | 3251 | fraction = 16; |
5e6d444e CL |
3252 | while (fraction >= 4) { |
3253 | order = slab_order(size, min_objects, | |
ab9a0f19 | 3254 | slub_max_order, fraction, reserved); |
5e6d444e CL |
3255 | if (order <= slub_max_order) |
3256 | return order; | |
3257 | fraction /= 2; | |
3258 | } | |
5086c389 | 3259 | min_objects--; |
5e6d444e CL |
3260 | } |
3261 | ||
3262 | /* | |
3263 | * We were unable to place multiple objects in a slab. Now | |
3264 | * lets see if we can place a single object there. | |
3265 | */ | |
ab9a0f19 | 3266 | order = slab_order(size, 1, slub_max_order, 1, reserved); |
5e6d444e CL |
3267 | if (order <= slub_max_order) |
3268 | return order; | |
3269 | ||
3270 | /* | |
3271 | * Doh this slab cannot be placed using slub_max_order. | |
3272 | */ | |
ab9a0f19 | 3273 | order = slab_order(size, 1, MAX_ORDER, 1, reserved); |
818cf590 | 3274 | if (order < MAX_ORDER) |
5e6d444e CL |
3275 | return order; |
3276 | return -ENOSYS; | |
3277 | } | |
3278 | ||
5595cffc | 3279 | static void |
4053497d | 3280 | init_kmem_cache_node(struct kmem_cache_node *n) |
81819f0f CL |
3281 | { |
3282 | n->nr_partial = 0; | |
81819f0f CL |
3283 | spin_lock_init(&n->list_lock); |
3284 | INIT_LIST_HEAD(&n->partial); | |
8ab1372f | 3285 | #ifdef CONFIG_SLUB_DEBUG |
0f389ec6 | 3286 | atomic_long_set(&n->nr_slabs, 0); |
02b71b70 | 3287 | atomic_long_set(&n->total_objects, 0); |
643b1138 | 3288 | INIT_LIST_HEAD(&n->full); |
8ab1372f | 3289 | #endif |
81819f0f CL |
3290 | } |
3291 | ||
55136592 | 3292 | static inline int alloc_kmem_cache_cpus(struct kmem_cache *s) |
4c93c355 | 3293 | { |
6c182dc0 | 3294 | BUILD_BUG_ON(PERCPU_DYNAMIC_EARLY_SIZE < |
95a05b42 | 3295 | KMALLOC_SHIFT_HIGH * sizeof(struct kmem_cache_cpu)); |
4c93c355 | 3296 | |
8a5ec0ba | 3297 | /* |
d4d84fef CM |
3298 | * Must align to double word boundary for the double cmpxchg |
3299 | * instructions to work; see __pcpu_double_call_return_bool(). | |
8a5ec0ba | 3300 | */ |
d4d84fef CM |
3301 | s->cpu_slab = __alloc_percpu(sizeof(struct kmem_cache_cpu), |
3302 | 2 * sizeof(void *)); | |
8a5ec0ba CL |
3303 | |
3304 | if (!s->cpu_slab) | |
3305 | return 0; | |
3306 | ||
3307 | init_kmem_cache_cpus(s); | |
4c93c355 | 3308 | |
8a5ec0ba | 3309 | return 1; |
4c93c355 | 3310 | } |
4c93c355 | 3311 | |
51df1142 CL |
3312 | static struct kmem_cache *kmem_cache_node; |
3313 | ||
81819f0f CL |
3314 | /* |
3315 | * No kmalloc_node yet so do it by hand. We know that this is the first | |
3316 | * slab on the node for this slabcache. There are no concurrent accesses | |
3317 | * possible. | |
3318 | * | |
721ae22a ZYW |
3319 | * Note that this function only works on the kmem_cache_node |
3320 | * when allocating for the kmem_cache_node. This is used for bootstrapping | |
4c93c355 | 3321 | * memory on a fresh node that has no slab structures yet. |
81819f0f | 3322 | */ |
55136592 | 3323 | static void early_kmem_cache_node_alloc(int node) |
81819f0f CL |
3324 | { |
3325 | struct page *page; | |
3326 | struct kmem_cache_node *n; | |
3327 | ||
51df1142 | 3328 | BUG_ON(kmem_cache_node->size < sizeof(struct kmem_cache_node)); |
81819f0f | 3329 | |
51df1142 | 3330 | page = new_slab(kmem_cache_node, GFP_NOWAIT, node); |
81819f0f CL |
3331 | |
3332 | BUG_ON(!page); | |
a2f92ee7 | 3333 | if (page_to_nid(page) != node) { |
f9f58285 FF |
3334 | pr_err("SLUB: Unable to allocate memory from node %d\n", node); |
3335 | pr_err("SLUB: Allocating a useless per node structure in order to be able to continue\n"); | |
a2f92ee7 CL |
3336 | } |
3337 | ||
81819f0f CL |
3338 | n = page->freelist; |
3339 | BUG_ON(!n); | |
51df1142 | 3340 | page->freelist = get_freepointer(kmem_cache_node, n); |
e6e82ea1 | 3341 | page->inuse = 1; |
8cb0a506 | 3342 | page->frozen = 0; |
51df1142 | 3343 | kmem_cache_node->node[node] = n; |
8ab1372f | 3344 | #ifdef CONFIG_SLUB_DEBUG |
f7cb1933 | 3345 | init_object(kmem_cache_node, n, SLUB_RED_ACTIVE); |
51df1142 | 3346 | init_tracking(kmem_cache_node, n); |
8ab1372f | 3347 | #endif |
505f5dcb AP |
3348 | kasan_kmalloc(kmem_cache_node, n, sizeof(struct kmem_cache_node), |
3349 | GFP_KERNEL); | |
4053497d | 3350 | init_kmem_cache_node(n); |
51df1142 | 3351 | inc_slabs_node(kmem_cache_node, node, page->objects); |
6446faa2 | 3352 | |
67b6c900 | 3353 | /* |
1e4dd946 SR |
3354 | * No locks need to be taken here as it has just been |
3355 | * initialized and there is no concurrent access. | |
67b6c900 | 3356 | */ |
1e4dd946 | 3357 | __add_partial(n, page, DEACTIVATE_TO_HEAD); |
81819f0f CL |
3358 | } |
3359 | ||
3360 | static void free_kmem_cache_nodes(struct kmem_cache *s) | |
3361 | { | |
3362 | int node; | |
fa45dc25 | 3363 | struct kmem_cache_node *n; |
81819f0f | 3364 | |
fa45dc25 CL |
3365 | for_each_kmem_cache_node(s, node, n) { |
3366 | kmem_cache_free(kmem_cache_node, n); | |
81819f0f CL |
3367 | s->node[node] = NULL; |
3368 | } | |
3369 | } | |
3370 | ||
52b4b950 DS |
3371 | void __kmem_cache_release(struct kmem_cache *s) |
3372 | { | |
210e7a43 | 3373 | cache_random_seq_destroy(s); |
52b4b950 DS |
3374 | free_percpu(s->cpu_slab); |
3375 | free_kmem_cache_nodes(s); | |
3376 | } | |
3377 | ||
55136592 | 3378 | static int init_kmem_cache_nodes(struct kmem_cache *s) |
81819f0f CL |
3379 | { |
3380 | int node; | |
81819f0f | 3381 | |
f64dc58c | 3382 | for_each_node_state(node, N_NORMAL_MEMORY) { |
81819f0f CL |
3383 | struct kmem_cache_node *n; |
3384 | ||
73367bd8 | 3385 | if (slab_state == DOWN) { |
55136592 | 3386 | early_kmem_cache_node_alloc(node); |
73367bd8 AD |
3387 | continue; |
3388 | } | |
51df1142 | 3389 | n = kmem_cache_alloc_node(kmem_cache_node, |
55136592 | 3390 | GFP_KERNEL, node); |
81819f0f | 3391 | |
73367bd8 AD |
3392 | if (!n) { |
3393 | free_kmem_cache_nodes(s); | |
3394 | return 0; | |
81819f0f | 3395 | } |
73367bd8 | 3396 | |
81819f0f | 3397 | s->node[node] = n; |
4053497d | 3398 | init_kmem_cache_node(n); |
81819f0f CL |
3399 | } |
3400 | return 1; | |
3401 | } | |
81819f0f | 3402 | |
c0bdb232 | 3403 | static void set_min_partial(struct kmem_cache *s, unsigned long min) |
3b89d7d8 DR |
3404 | { |
3405 | if (min < MIN_PARTIAL) | |
3406 | min = MIN_PARTIAL; | |
3407 | else if (min > MAX_PARTIAL) | |
3408 | min = MAX_PARTIAL; | |
3409 | s->min_partial = min; | |
3410 | } | |
3411 | ||
81819f0f CL |
3412 | /* |
3413 | * calculate_sizes() determines the order and the distribution of data within | |
3414 | * a slab object. | |
3415 | */ | |
06b285dc | 3416 | static int calculate_sizes(struct kmem_cache *s, int forced_order) |
81819f0f CL |
3417 | { |
3418 | unsigned long flags = s->flags; | |
80a9201a | 3419 | size_t size = s->object_size; |
834f3d11 | 3420 | int order; |
81819f0f | 3421 | |
d8b42bf5 CL |
3422 | /* |
3423 | * Round up object size to the next word boundary. We can only | |
3424 | * place the free pointer at word boundaries and this determines | |
3425 | * the possible location of the free pointer. | |
3426 | */ | |
3427 | size = ALIGN(size, sizeof(void *)); | |
3428 | ||
3429 | #ifdef CONFIG_SLUB_DEBUG | |
81819f0f CL |
3430 | /* |
3431 | * Determine if we can poison the object itself. If the user of | |
3432 | * the slab may touch the object after free or before allocation | |
3433 | * then we should never poison the object itself. | |
3434 | */ | |
5f0d5a3a | 3435 | if ((flags & SLAB_POISON) && !(flags & SLAB_TYPESAFE_BY_RCU) && |
c59def9f | 3436 | !s->ctor) |
81819f0f CL |
3437 | s->flags |= __OBJECT_POISON; |
3438 | else | |
3439 | s->flags &= ~__OBJECT_POISON; | |
3440 | ||
81819f0f CL |
3441 | |
3442 | /* | |
672bba3a | 3443 | * If we are Redzoning then check if there is some space between the |
81819f0f | 3444 | * end of the object and the free pointer. If not then add an |
672bba3a | 3445 | * additional word to have some bytes to store Redzone information. |
81819f0f | 3446 | */ |
3b0efdfa | 3447 | if ((flags & SLAB_RED_ZONE) && size == s->object_size) |
81819f0f | 3448 | size += sizeof(void *); |
41ecc55b | 3449 | #endif |
81819f0f CL |
3450 | |
3451 | /* | |
672bba3a CL |
3452 | * With that we have determined the number of bytes in actual use |
3453 | * by the object. This is the potential offset to the free pointer. | |
81819f0f CL |
3454 | */ |
3455 | s->inuse = size; | |
3456 | ||
5f0d5a3a | 3457 | if (((flags & (SLAB_TYPESAFE_BY_RCU | SLAB_POISON)) || |
c59def9f | 3458 | s->ctor)) { |
81819f0f CL |
3459 | /* |
3460 | * Relocate free pointer after the object if it is not | |
3461 | * permitted to overwrite the first word of the object on | |
3462 | * kmem_cache_free. | |
3463 | * | |
3464 | * This is the case if we do RCU, have a constructor or | |
3465 | * destructor or are poisoning the objects. | |
3466 | */ | |
3467 | s->offset = size; | |
3468 | size += sizeof(void *); | |
3469 | } | |
3470 | ||
c12b3c62 | 3471 | #ifdef CONFIG_SLUB_DEBUG |
81819f0f CL |
3472 | if (flags & SLAB_STORE_USER) |
3473 | /* | |
3474 | * Need to store information about allocs and frees after | |
3475 | * the object. | |
3476 | */ | |
3477 | size += 2 * sizeof(struct track); | |
80a9201a | 3478 | #endif |
81819f0f | 3479 | |
80a9201a AP |
3480 | kasan_cache_create(s, &size, &s->flags); |
3481 | #ifdef CONFIG_SLUB_DEBUG | |
d86bd1be | 3482 | if (flags & SLAB_RED_ZONE) { |
81819f0f CL |
3483 | /* |
3484 | * Add some empty padding so that we can catch | |
3485 | * overwrites from earlier objects rather than let | |
3486 | * tracking information or the free pointer be | |
0211a9c8 | 3487 | * corrupted if a user writes before the start |
81819f0f CL |
3488 | * of the object. |
3489 | */ | |
3490 | size += sizeof(void *); | |
d86bd1be JK |
3491 | |
3492 | s->red_left_pad = sizeof(void *); | |
3493 | s->red_left_pad = ALIGN(s->red_left_pad, s->align); | |
3494 | size += s->red_left_pad; | |
3495 | } | |
41ecc55b | 3496 | #endif |
672bba3a | 3497 | |
81819f0f CL |
3498 | /* |
3499 | * SLUB stores one object immediately after another beginning from | |
3500 | * offset 0. In order to align the objects we have to simply size | |
3501 | * each object to conform to the alignment. | |
3502 | */ | |
45906855 | 3503 | size = ALIGN(size, s->align); |
81819f0f | 3504 | s->size = size; |
06b285dc CL |
3505 | if (forced_order >= 0) |
3506 | order = forced_order; | |
3507 | else | |
ab9a0f19 | 3508 | order = calculate_order(size, s->reserved); |
81819f0f | 3509 | |
834f3d11 | 3510 | if (order < 0) |
81819f0f CL |
3511 | return 0; |
3512 | ||
b7a49f0d | 3513 | s->allocflags = 0; |
834f3d11 | 3514 | if (order) |
b7a49f0d CL |
3515 | s->allocflags |= __GFP_COMP; |
3516 | ||
3517 | if (s->flags & SLAB_CACHE_DMA) | |
2c59dd65 | 3518 | s->allocflags |= GFP_DMA; |
b7a49f0d CL |
3519 | |
3520 | if (s->flags & SLAB_RECLAIM_ACCOUNT) | |
3521 | s->allocflags |= __GFP_RECLAIMABLE; | |
3522 | ||
81819f0f CL |
3523 | /* |
3524 | * Determine the number of objects per slab | |
3525 | */ | |
ab9a0f19 LJ |
3526 | s->oo = oo_make(order, size, s->reserved); |
3527 | s->min = oo_make(get_order(size), size, s->reserved); | |
205ab99d CL |
3528 | if (oo_objects(s->oo) > oo_objects(s->max)) |
3529 | s->max = s->oo; | |
81819f0f | 3530 | |
834f3d11 | 3531 | return !!oo_objects(s->oo); |
81819f0f CL |
3532 | } |
3533 | ||
8a13a4cc | 3534 | static int kmem_cache_open(struct kmem_cache *s, unsigned long flags) |
81819f0f | 3535 | { |
8a13a4cc | 3536 | s->flags = kmem_cache_flags(s->size, flags, s->name, s->ctor); |
ab9a0f19 | 3537 | s->reserved = 0; |
81819f0f | 3538 | |
5f0d5a3a | 3539 | if (need_reserve_slab_rcu && (s->flags & SLAB_TYPESAFE_BY_RCU)) |
da9a638c | 3540 | s->reserved = sizeof(struct rcu_head); |
81819f0f | 3541 | |
06b285dc | 3542 | if (!calculate_sizes(s, -1)) |
81819f0f | 3543 | goto error; |
3de47213 DR |
3544 | if (disable_higher_order_debug) { |
3545 | /* | |
3546 | * Disable debugging flags that store metadata if the min slab | |
3547 | * order increased. | |
3548 | */ | |
3b0efdfa | 3549 | if (get_order(s->size) > get_order(s->object_size)) { |
3de47213 DR |
3550 | s->flags &= ~DEBUG_METADATA_FLAGS; |
3551 | s->offset = 0; | |
3552 | if (!calculate_sizes(s, -1)) | |
3553 | goto error; | |
3554 | } | |
3555 | } | |
81819f0f | 3556 | |
2565409f HC |
3557 | #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \ |
3558 | defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE) | |
149daaf3 | 3559 | if (system_has_cmpxchg_double() && (s->flags & SLAB_NO_CMPXCHG) == 0) |
b789ef51 CL |
3560 | /* Enable fast mode */ |
3561 | s->flags |= __CMPXCHG_DOUBLE; | |
3562 | #endif | |
3563 | ||
3b89d7d8 DR |
3564 | /* |
3565 | * The larger the object size is, the more pages we want on the partial | |
3566 | * list to avoid pounding the page allocator excessively. | |
3567 | */ | |
49e22585 CL |
3568 | set_min_partial(s, ilog2(s->size) / 2); |
3569 | ||
3570 | /* | |
3571 | * cpu_partial determined the maximum number of objects kept in the | |
3572 | * per cpu partial lists of a processor. | |
3573 | * | |
3574 | * Per cpu partial lists mainly contain slabs that just have one | |
3575 | * object freed. If they are used for allocation then they can be | |
3576 | * filled up again with minimal effort. The slab will never hit the | |
3577 | * per node partial lists and therefore no locking will be required. | |
3578 | * | |
3579 | * This setting also determines | |
3580 | * | |
3581 | * A) The number of objects from per cpu partial slabs dumped to the | |
3582 | * per node list when we reach the limit. | |
9f264904 | 3583 | * B) The number of objects in cpu partial slabs to extract from the |
d0e0ac97 CG |
3584 | * per node list when we run out of per cpu objects. We only fetch |
3585 | * 50% to keep some capacity around for frees. | |
49e22585 | 3586 | */ |
345c905d | 3587 | if (!kmem_cache_has_cpu_partial(s)) |
8f1e33da CL |
3588 | s->cpu_partial = 0; |
3589 | else if (s->size >= PAGE_SIZE) | |
49e22585 CL |
3590 | s->cpu_partial = 2; |
3591 | else if (s->size >= 1024) | |
3592 | s->cpu_partial = 6; | |
3593 | else if (s->size >= 256) | |
3594 | s->cpu_partial = 13; | |
3595 | else | |
3596 | s->cpu_partial = 30; | |
3597 | ||
81819f0f | 3598 | #ifdef CONFIG_NUMA |
e2cb96b7 | 3599 | s->remote_node_defrag_ratio = 1000; |
81819f0f | 3600 | #endif |
210e7a43 TG |
3601 | |
3602 | /* Initialize the pre-computed randomized freelist if slab is up */ | |
3603 | if (slab_state >= UP) { | |
3604 | if (init_cache_random_seq(s)) | |
3605 | goto error; | |
3606 | } | |
3607 | ||
55136592 | 3608 | if (!init_kmem_cache_nodes(s)) |
dfb4f096 | 3609 | goto error; |
81819f0f | 3610 | |
55136592 | 3611 | if (alloc_kmem_cache_cpus(s)) |
278b1bb1 | 3612 | return 0; |
ff12059e | 3613 | |
4c93c355 | 3614 | free_kmem_cache_nodes(s); |
81819f0f CL |
3615 | error: |
3616 | if (flags & SLAB_PANIC) | |
756a025f JP |
3617 | panic("Cannot create slab %s size=%lu realsize=%u order=%u offset=%u flags=%lx\n", |
3618 | s->name, (unsigned long)s->size, s->size, | |
3619 | oo_order(s->oo), s->offset, flags); | |
278b1bb1 | 3620 | return -EINVAL; |
81819f0f | 3621 | } |
81819f0f | 3622 | |
33b12c38 CL |
3623 | static void list_slab_objects(struct kmem_cache *s, struct page *page, |
3624 | const char *text) | |
3625 | { | |
3626 | #ifdef CONFIG_SLUB_DEBUG | |
3627 | void *addr = page_address(page); | |
3628 | void *p; | |
a5dd5c11 NK |
3629 | unsigned long *map = kzalloc(BITS_TO_LONGS(page->objects) * |
3630 | sizeof(long), GFP_ATOMIC); | |
bbd7d57b ED |
3631 | if (!map) |
3632 | return; | |
945cf2b6 | 3633 | slab_err(s, page, text, s->name); |
33b12c38 | 3634 | slab_lock(page); |
33b12c38 | 3635 | |
5f80b13a | 3636 | get_map(s, page, map); |
33b12c38 CL |
3637 | for_each_object(p, s, addr, page->objects) { |
3638 | ||
3639 | if (!test_bit(slab_index(p, s, addr), map)) { | |
f9f58285 | 3640 | pr_err("INFO: Object 0x%p @offset=%tu\n", p, p - addr); |
33b12c38 CL |
3641 | print_tracking(s, p); |
3642 | } | |
3643 | } | |
3644 | slab_unlock(page); | |
bbd7d57b | 3645 | kfree(map); |
33b12c38 CL |
3646 | #endif |
3647 | } | |
3648 | ||
81819f0f | 3649 | /* |
599870b1 | 3650 | * Attempt to free all partial slabs on a node. |
52b4b950 DS |
3651 | * This is called from __kmem_cache_shutdown(). We must take list_lock |
3652 | * because sysfs file might still access partial list after the shutdowning. | |
81819f0f | 3653 | */ |
599870b1 | 3654 | static void free_partial(struct kmem_cache *s, struct kmem_cache_node *n) |
81819f0f | 3655 | { |
60398923 | 3656 | LIST_HEAD(discard); |
81819f0f CL |
3657 | struct page *page, *h; |
3658 | ||
52b4b950 DS |
3659 | BUG_ON(irqs_disabled()); |
3660 | spin_lock_irq(&n->list_lock); | |
33b12c38 | 3661 | list_for_each_entry_safe(page, h, &n->partial, lru) { |
81819f0f | 3662 | if (!page->inuse) { |
52b4b950 | 3663 | remove_partial(n, page); |
60398923 | 3664 | list_add(&page->lru, &discard); |
33b12c38 CL |
3665 | } else { |
3666 | list_slab_objects(s, page, | |
52b4b950 | 3667 | "Objects remaining in %s on __kmem_cache_shutdown()"); |
599870b1 | 3668 | } |
33b12c38 | 3669 | } |
52b4b950 | 3670 | spin_unlock_irq(&n->list_lock); |
60398923 CW |
3671 | |
3672 | list_for_each_entry_safe(page, h, &discard, lru) | |
3673 | discard_slab(s, page); | |
81819f0f CL |
3674 | } |
3675 | ||
3676 | /* | |
672bba3a | 3677 | * Release all resources used by a slab cache. |
81819f0f | 3678 | */ |
52b4b950 | 3679 | int __kmem_cache_shutdown(struct kmem_cache *s) |
81819f0f CL |
3680 | { |
3681 | int node; | |
fa45dc25 | 3682 | struct kmem_cache_node *n; |
81819f0f CL |
3683 | |
3684 | flush_all(s); | |
81819f0f | 3685 | /* Attempt to free all objects */ |
fa45dc25 | 3686 | for_each_kmem_cache_node(s, node, n) { |
599870b1 CL |
3687 | free_partial(s, n); |
3688 | if (n->nr_partial || slabs_node(s, node)) | |
81819f0f CL |
3689 | return 1; |
3690 | } | |
bf5eb3de | 3691 | sysfs_slab_remove(s); |
81819f0f CL |
3692 | return 0; |
3693 | } | |
3694 | ||
81819f0f CL |
3695 | /******************************************************************** |
3696 | * Kmalloc subsystem | |
3697 | *******************************************************************/ | |
3698 | ||
81819f0f CL |
3699 | static int __init setup_slub_min_order(char *str) |
3700 | { | |
06428780 | 3701 | get_option(&str, &slub_min_order); |
81819f0f CL |
3702 | |
3703 | return 1; | |
3704 | } | |
3705 | ||
3706 | __setup("slub_min_order=", setup_slub_min_order); | |
3707 | ||
3708 | static int __init setup_slub_max_order(char *str) | |
3709 | { | |
06428780 | 3710 | get_option(&str, &slub_max_order); |
818cf590 | 3711 | slub_max_order = min(slub_max_order, MAX_ORDER - 1); |
81819f0f CL |
3712 | |
3713 | return 1; | |
3714 | } | |
3715 | ||
3716 | __setup("slub_max_order=", setup_slub_max_order); | |
3717 | ||
3718 | static int __init setup_slub_min_objects(char *str) | |
3719 | { | |
06428780 | 3720 | get_option(&str, &slub_min_objects); |
81819f0f CL |
3721 | |
3722 | return 1; | |
3723 | } | |
3724 | ||
3725 | __setup("slub_min_objects=", setup_slub_min_objects); | |
3726 | ||
81819f0f CL |
3727 | void *__kmalloc(size_t size, gfp_t flags) |
3728 | { | |
aadb4bc4 | 3729 | struct kmem_cache *s; |
5b882be4 | 3730 | void *ret; |
81819f0f | 3731 | |
95a05b42 | 3732 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) |
eada35ef | 3733 | return kmalloc_large(size, flags); |
aadb4bc4 | 3734 | |
2c59dd65 | 3735 | s = kmalloc_slab(size, flags); |
aadb4bc4 CL |
3736 | |
3737 | if (unlikely(ZERO_OR_NULL_PTR(s))) | |
6cb8f913 CL |
3738 | return s; |
3739 | ||
2b847c3c | 3740 | ret = slab_alloc(s, flags, _RET_IP_); |
5b882be4 | 3741 | |
ca2b84cb | 3742 | trace_kmalloc(_RET_IP_, ret, size, s->size, flags); |
5b882be4 | 3743 | |
505f5dcb | 3744 | kasan_kmalloc(s, ret, size, flags); |
0316bec2 | 3745 | |
5b882be4 | 3746 | return ret; |
81819f0f CL |
3747 | } |
3748 | EXPORT_SYMBOL(__kmalloc); | |
3749 | ||
5d1f57e4 | 3750 | #ifdef CONFIG_NUMA |
f619cfe1 CL |
3751 | static void *kmalloc_large_node(size_t size, gfp_t flags, int node) |
3752 | { | |
b1eeab67 | 3753 | struct page *page; |
e4f7c0b4 | 3754 | void *ptr = NULL; |
f619cfe1 | 3755 | |
52383431 | 3756 | flags |= __GFP_COMP | __GFP_NOTRACK; |
4949148a | 3757 | page = alloc_pages_node(node, flags, get_order(size)); |
f619cfe1 | 3758 | if (page) |
e4f7c0b4 CM |
3759 | ptr = page_address(page); |
3760 | ||
d56791b3 | 3761 | kmalloc_large_node_hook(ptr, size, flags); |
e4f7c0b4 | 3762 | return ptr; |
f619cfe1 CL |
3763 | } |
3764 | ||
81819f0f CL |
3765 | void *__kmalloc_node(size_t size, gfp_t flags, int node) |
3766 | { | |
aadb4bc4 | 3767 | struct kmem_cache *s; |
5b882be4 | 3768 | void *ret; |
81819f0f | 3769 | |
95a05b42 | 3770 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) { |
5b882be4 EGM |
3771 | ret = kmalloc_large_node(size, flags, node); |
3772 | ||
ca2b84cb EGM |
3773 | trace_kmalloc_node(_RET_IP_, ret, |
3774 | size, PAGE_SIZE << get_order(size), | |
3775 | flags, node); | |
5b882be4 EGM |
3776 | |
3777 | return ret; | |
3778 | } | |
aadb4bc4 | 3779 | |
2c59dd65 | 3780 | s = kmalloc_slab(size, flags); |
aadb4bc4 CL |
3781 | |
3782 | if (unlikely(ZERO_OR_NULL_PTR(s))) | |
6cb8f913 CL |
3783 | return s; |
3784 | ||
2b847c3c | 3785 | ret = slab_alloc_node(s, flags, node, _RET_IP_); |
5b882be4 | 3786 | |
ca2b84cb | 3787 | trace_kmalloc_node(_RET_IP_, ret, size, s->size, flags, node); |
5b882be4 | 3788 | |
505f5dcb | 3789 | kasan_kmalloc(s, ret, size, flags); |
0316bec2 | 3790 | |
5b882be4 | 3791 | return ret; |
81819f0f CL |
3792 | } |
3793 | EXPORT_SYMBOL(__kmalloc_node); | |
3794 | #endif | |
3795 | ||
ed18adc1 KC |
3796 | #ifdef CONFIG_HARDENED_USERCOPY |
3797 | /* | |
3798 | * Rejects objects that are incorrectly sized. | |
3799 | * | |
3800 | * Returns NULL if check passes, otherwise const char * to name of cache | |
3801 | * to indicate an error. | |
3802 | */ | |
3803 | const char *__check_heap_object(const void *ptr, unsigned long n, | |
3804 | struct page *page) | |
3805 | { | |
3806 | struct kmem_cache *s; | |
3807 | unsigned long offset; | |
3808 | size_t object_size; | |
3809 | ||
3810 | /* Find object and usable object size. */ | |
3811 | s = page->slab_cache; | |
3812 | object_size = slab_ksize(s); | |
3813 | ||
3814 | /* Reject impossible pointers. */ | |
3815 | if (ptr < page_address(page)) | |
3816 | return s->name; | |
3817 | ||
3818 | /* Find offset within object. */ | |
3819 | offset = (ptr - page_address(page)) % s->size; | |
3820 | ||
3821 | /* Adjust for redzone and reject if within the redzone. */ | |
3822 | if (kmem_cache_debug(s) && s->flags & SLAB_RED_ZONE) { | |
3823 | if (offset < s->red_left_pad) | |
3824 | return s->name; | |
3825 | offset -= s->red_left_pad; | |
3826 | } | |
3827 | ||
3828 | /* Allow address range falling entirely within object size. */ | |
3829 | if (offset <= object_size && n <= object_size - offset) | |
3830 | return NULL; | |
3831 | ||
3832 | return s->name; | |
3833 | } | |
3834 | #endif /* CONFIG_HARDENED_USERCOPY */ | |
3835 | ||
0316bec2 | 3836 | static size_t __ksize(const void *object) |
81819f0f | 3837 | { |
272c1d21 | 3838 | struct page *page; |
81819f0f | 3839 | |
ef8b4520 | 3840 | if (unlikely(object == ZERO_SIZE_PTR)) |
272c1d21 CL |
3841 | return 0; |
3842 | ||
294a80a8 | 3843 | page = virt_to_head_page(object); |
294a80a8 | 3844 | |
76994412 PE |
3845 | if (unlikely(!PageSlab(page))) { |
3846 | WARN_ON(!PageCompound(page)); | |
294a80a8 | 3847 | return PAGE_SIZE << compound_order(page); |
76994412 | 3848 | } |
81819f0f | 3849 | |
1b4f59e3 | 3850 | return slab_ksize(page->slab_cache); |
81819f0f | 3851 | } |
0316bec2 AR |
3852 | |
3853 | size_t ksize(const void *object) | |
3854 | { | |
3855 | size_t size = __ksize(object); | |
3856 | /* We assume that ksize callers could use whole allocated area, | |
4ebb31a4 AP |
3857 | * so we need to unpoison this area. |
3858 | */ | |
3859 | kasan_unpoison_shadow(object, size); | |
0316bec2 AR |
3860 | return size; |
3861 | } | |
b1aabecd | 3862 | EXPORT_SYMBOL(ksize); |
81819f0f CL |
3863 | |
3864 | void kfree(const void *x) | |
3865 | { | |
81819f0f | 3866 | struct page *page; |
5bb983b0 | 3867 | void *object = (void *)x; |
81819f0f | 3868 | |
2121db74 PE |
3869 | trace_kfree(_RET_IP_, x); |
3870 | ||
2408c550 | 3871 | if (unlikely(ZERO_OR_NULL_PTR(x))) |
81819f0f CL |
3872 | return; |
3873 | ||
b49af68f | 3874 | page = virt_to_head_page(x); |
aadb4bc4 | 3875 | if (unlikely(!PageSlab(page))) { |
0937502a | 3876 | BUG_ON(!PageCompound(page)); |
d56791b3 | 3877 | kfree_hook(x); |
4949148a | 3878 | __free_pages(page, compound_order(page)); |
aadb4bc4 CL |
3879 | return; |
3880 | } | |
81084651 | 3881 | slab_free(page->slab_cache, page, object, NULL, 1, _RET_IP_); |
81819f0f CL |
3882 | } |
3883 | EXPORT_SYMBOL(kfree); | |
3884 | ||
832f37f5 VD |
3885 | #define SHRINK_PROMOTE_MAX 32 |
3886 | ||
2086d26a | 3887 | /* |
832f37f5 VD |
3888 | * kmem_cache_shrink discards empty slabs and promotes the slabs filled |
3889 | * up most to the head of the partial lists. New allocations will then | |
3890 | * fill those up and thus they can be removed from the partial lists. | |
672bba3a CL |
3891 | * |
3892 | * The slabs with the least items are placed last. This results in them | |
3893 | * being allocated from last increasing the chance that the last objects | |
3894 | * are freed in them. | |
2086d26a | 3895 | */ |
c9fc5864 | 3896 | int __kmem_cache_shrink(struct kmem_cache *s) |
2086d26a CL |
3897 | { |
3898 | int node; | |
3899 | int i; | |
3900 | struct kmem_cache_node *n; | |
3901 | struct page *page; | |
3902 | struct page *t; | |
832f37f5 VD |
3903 | struct list_head discard; |
3904 | struct list_head promote[SHRINK_PROMOTE_MAX]; | |
2086d26a | 3905 | unsigned long flags; |
ce3712d7 | 3906 | int ret = 0; |
2086d26a | 3907 | |
2086d26a | 3908 | flush_all(s); |
fa45dc25 | 3909 | for_each_kmem_cache_node(s, node, n) { |
832f37f5 VD |
3910 | INIT_LIST_HEAD(&discard); |
3911 | for (i = 0; i < SHRINK_PROMOTE_MAX; i++) | |
3912 | INIT_LIST_HEAD(promote + i); | |
2086d26a CL |
3913 | |
3914 | spin_lock_irqsave(&n->list_lock, flags); | |
3915 | ||
3916 | /* | |
832f37f5 | 3917 | * Build lists of slabs to discard or promote. |
2086d26a | 3918 | * |
672bba3a CL |
3919 | * Note that concurrent frees may occur while we hold the |
3920 | * list_lock. page->inuse here is the upper limit. | |
2086d26a CL |
3921 | */ |
3922 | list_for_each_entry_safe(page, t, &n->partial, lru) { | |
832f37f5 VD |
3923 | int free = page->objects - page->inuse; |
3924 | ||
3925 | /* Do not reread page->inuse */ | |
3926 | barrier(); | |
3927 | ||
3928 | /* We do not keep full slabs on the list */ | |
3929 | BUG_ON(free <= 0); | |
3930 | ||
3931 | if (free == page->objects) { | |
3932 | list_move(&page->lru, &discard); | |
69cb8e6b | 3933 | n->nr_partial--; |
832f37f5 VD |
3934 | } else if (free <= SHRINK_PROMOTE_MAX) |
3935 | list_move(&page->lru, promote + free - 1); | |
2086d26a CL |
3936 | } |
3937 | ||
2086d26a | 3938 | /* |
832f37f5 VD |
3939 | * Promote the slabs filled up most to the head of the |
3940 | * partial list. | |
2086d26a | 3941 | */ |
832f37f5 VD |
3942 | for (i = SHRINK_PROMOTE_MAX - 1; i >= 0; i--) |
3943 | list_splice(promote + i, &n->partial); | |
2086d26a | 3944 | |
2086d26a | 3945 | spin_unlock_irqrestore(&n->list_lock, flags); |
69cb8e6b CL |
3946 | |
3947 | /* Release empty slabs */ | |
832f37f5 | 3948 | list_for_each_entry_safe(page, t, &discard, lru) |
69cb8e6b | 3949 | discard_slab(s, page); |
ce3712d7 VD |
3950 | |
3951 | if (slabs_node(s, node)) | |
3952 | ret = 1; | |
2086d26a CL |
3953 | } |
3954 | ||
ce3712d7 | 3955 | return ret; |
2086d26a | 3956 | } |
2086d26a | 3957 | |
c9fc5864 | 3958 | #ifdef CONFIG_MEMCG |
01fb58bc TH |
3959 | static void kmemcg_cache_deact_after_rcu(struct kmem_cache *s) |
3960 | { | |
50862ce7 TH |
3961 | /* |
3962 | * Called with all the locks held after a sched RCU grace period. | |
3963 | * Even if @s becomes empty after shrinking, we can't know that @s | |
3964 | * doesn't have allocations already in-flight and thus can't | |
3965 | * destroy @s until the associated memcg is released. | |
3966 | * | |
3967 | * However, let's remove the sysfs files for empty caches here. | |
3968 | * Each cache has a lot of interface files which aren't | |
3969 | * particularly useful for empty draining caches; otherwise, we can | |
3970 | * easily end up with millions of unnecessary sysfs files on | |
3971 | * systems which have a lot of memory and transient cgroups. | |
3972 | */ | |
3973 | if (!__kmem_cache_shrink(s)) | |
3974 | sysfs_slab_remove(s); | |
01fb58bc TH |
3975 | } |
3976 | ||
c9fc5864 TH |
3977 | void __kmemcg_cache_deactivate(struct kmem_cache *s) |
3978 | { | |
3979 | /* | |
3980 | * Disable empty slabs caching. Used to avoid pinning offline | |
3981 | * memory cgroups by kmem pages that can be freed. | |
3982 | */ | |
3983 | s->cpu_partial = 0; | |
3984 | s->min_partial = 0; | |
3985 | ||
3986 | /* | |
3987 | * s->cpu_partial is checked locklessly (see put_cpu_partial), so | |
01fb58bc | 3988 | * we have to make sure the change is visible before shrinking. |
c9fc5864 | 3989 | */ |
01fb58bc | 3990 | slab_deactivate_memcg_cache_rcu_sched(s, kmemcg_cache_deact_after_rcu); |
c9fc5864 TH |
3991 | } |
3992 | #endif | |
3993 | ||
b9049e23 YG |
3994 | static int slab_mem_going_offline_callback(void *arg) |
3995 | { | |
3996 | struct kmem_cache *s; | |
3997 | ||
18004c5d | 3998 | mutex_lock(&slab_mutex); |
b9049e23 | 3999 | list_for_each_entry(s, &slab_caches, list) |
c9fc5864 | 4000 | __kmem_cache_shrink(s); |
18004c5d | 4001 | mutex_unlock(&slab_mutex); |
b9049e23 YG |
4002 | |
4003 | return 0; | |
4004 | } | |
4005 | ||
4006 | static void slab_mem_offline_callback(void *arg) | |
4007 | { | |
4008 | struct kmem_cache_node *n; | |
4009 | struct kmem_cache *s; | |
4010 | struct memory_notify *marg = arg; | |
4011 | int offline_node; | |
4012 | ||
b9d5ab25 | 4013 | offline_node = marg->status_change_nid_normal; |
b9049e23 YG |
4014 | |
4015 | /* | |
4016 | * If the node still has available memory. we need kmem_cache_node | |
4017 | * for it yet. | |
4018 | */ | |
4019 | if (offline_node < 0) | |
4020 | return; | |
4021 | ||
18004c5d | 4022 | mutex_lock(&slab_mutex); |
b9049e23 YG |
4023 | list_for_each_entry(s, &slab_caches, list) { |
4024 | n = get_node(s, offline_node); | |
4025 | if (n) { | |
4026 | /* | |
4027 | * if n->nr_slabs > 0, slabs still exist on the node | |
4028 | * that is going down. We were unable to free them, | |
c9404c9c | 4029 | * and offline_pages() function shouldn't call this |
b9049e23 YG |
4030 | * callback. So, we must fail. |
4031 | */ | |
0f389ec6 | 4032 | BUG_ON(slabs_node(s, offline_node)); |
b9049e23 YG |
4033 | |
4034 | s->node[offline_node] = NULL; | |
8de66a0c | 4035 | kmem_cache_free(kmem_cache_node, n); |
b9049e23 YG |
4036 | } |
4037 | } | |
18004c5d | 4038 | mutex_unlock(&slab_mutex); |
b9049e23 YG |
4039 | } |
4040 | ||
4041 | static int slab_mem_going_online_callback(void *arg) | |
4042 | { | |
4043 | struct kmem_cache_node *n; | |
4044 | struct kmem_cache *s; | |
4045 | struct memory_notify *marg = arg; | |
b9d5ab25 | 4046 | int nid = marg->status_change_nid_normal; |
b9049e23 YG |
4047 | int ret = 0; |
4048 | ||
4049 | /* | |
4050 | * If the node's memory is already available, then kmem_cache_node is | |
4051 | * already created. Nothing to do. | |
4052 | */ | |
4053 | if (nid < 0) | |
4054 | return 0; | |
4055 | ||
4056 | /* | |
0121c619 | 4057 | * We are bringing a node online. No memory is available yet. We must |
b9049e23 YG |
4058 | * allocate a kmem_cache_node structure in order to bring the node |
4059 | * online. | |
4060 | */ | |
18004c5d | 4061 | mutex_lock(&slab_mutex); |
b9049e23 YG |
4062 | list_for_each_entry(s, &slab_caches, list) { |
4063 | /* | |
4064 | * XXX: kmem_cache_alloc_node will fallback to other nodes | |
4065 | * since memory is not yet available from the node that | |
4066 | * is brought up. | |
4067 | */ | |
8de66a0c | 4068 | n = kmem_cache_alloc(kmem_cache_node, GFP_KERNEL); |
b9049e23 YG |
4069 | if (!n) { |
4070 | ret = -ENOMEM; | |
4071 | goto out; | |
4072 | } | |
4053497d | 4073 | init_kmem_cache_node(n); |
b9049e23 YG |
4074 | s->node[nid] = n; |
4075 | } | |
4076 | out: | |
18004c5d | 4077 | mutex_unlock(&slab_mutex); |
b9049e23 YG |
4078 | return ret; |
4079 | } | |
4080 | ||
4081 | static int slab_memory_callback(struct notifier_block *self, | |
4082 | unsigned long action, void *arg) | |
4083 | { | |
4084 | int ret = 0; | |
4085 | ||
4086 | switch (action) { | |
4087 | case MEM_GOING_ONLINE: | |
4088 | ret = slab_mem_going_online_callback(arg); | |
4089 | break; | |
4090 | case MEM_GOING_OFFLINE: | |
4091 | ret = slab_mem_going_offline_callback(arg); | |
4092 | break; | |
4093 | case MEM_OFFLINE: | |
4094 | case MEM_CANCEL_ONLINE: | |
4095 | slab_mem_offline_callback(arg); | |
4096 | break; | |
4097 | case MEM_ONLINE: | |
4098 | case MEM_CANCEL_OFFLINE: | |
4099 | break; | |
4100 | } | |
dc19f9db KH |
4101 | if (ret) |
4102 | ret = notifier_from_errno(ret); | |
4103 | else | |
4104 | ret = NOTIFY_OK; | |
b9049e23 YG |
4105 | return ret; |
4106 | } | |
4107 | ||
3ac38faa AM |
4108 | static struct notifier_block slab_memory_callback_nb = { |
4109 | .notifier_call = slab_memory_callback, | |
4110 | .priority = SLAB_CALLBACK_PRI, | |
4111 | }; | |
b9049e23 | 4112 | |
81819f0f CL |
4113 | /******************************************************************** |
4114 | * Basic setup of slabs | |
4115 | *******************************************************************/ | |
4116 | ||
51df1142 CL |
4117 | /* |
4118 | * Used for early kmem_cache structures that were allocated using | |
dffb4d60 CL |
4119 | * the page allocator. Allocate them properly then fix up the pointers |
4120 | * that may be pointing to the wrong kmem_cache structure. | |
51df1142 CL |
4121 | */ |
4122 | ||
dffb4d60 | 4123 | static struct kmem_cache * __init bootstrap(struct kmem_cache *static_cache) |
51df1142 CL |
4124 | { |
4125 | int node; | |
dffb4d60 | 4126 | struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT); |
fa45dc25 | 4127 | struct kmem_cache_node *n; |
51df1142 | 4128 | |
dffb4d60 | 4129 | memcpy(s, static_cache, kmem_cache->object_size); |
51df1142 | 4130 | |
7d557b3c GC |
4131 | /* |
4132 | * This runs very early, and only the boot processor is supposed to be | |
4133 | * up. Even if it weren't true, IRQs are not up so we couldn't fire | |
4134 | * IPIs around. | |
4135 | */ | |
4136 | __flush_cpu_slab(s, smp_processor_id()); | |
fa45dc25 | 4137 | for_each_kmem_cache_node(s, node, n) { |
51df1142 CL |
4138 | struct page *p; |
4139 | ||
fa45dc25 CL |
4140 | list_for_each_entry(p, &n->partial, lru) |
4141 | p->slab_cache = s; | |
51df1142 | 4142 | |
607bf324 | 4143 | #ifdef CONFIG_SLUB_DEBUG |
fa45dc25 CL |
4144 | list_for_each_entry(p, &n->full, lru) |
4145 | p->slab_cache = s; | |
51df1142 | 4146 | #endif |
51df1142 | 4147 | } |
f7ce3190 | 4148 | slab_init_memcg_params(s); |
dffb4d60 | 4149 | list_add(&s->list, &slab_caches); |
510ded33 | 4150 | memcg_link_cache(s); |
dffb4d60 | 4151 | return s; |
51df1142 CL |
4152 | } |
4153 | ||
81819f0f CL |
4154 | void __init kmem_cache_init(void) |
4155 | { | |
dffb4d60 CL |
4156 | static __initdata struct kmem_cache boot_kmem_cache, |
4157 | boot_kmem_cache_node; | |
51df1142 | 4158 | |
fc8d8620 SG |
4159 | if (debug_guardpage_minorder()) |
4160 | slub_max_order = 0; | |
4161 | ||
dffb4d60 CL |
4162 | kmem_cache_node = &boot_kmem_cache_node; |
4163 | kmem_cache = &boot_kmem_cache; | |
51df1142 | 4164 | |
dffb4d60 CL |
4165 | create_boot_cache(kmem_cache_node, "kmem_cache_node", |
4166 | sizeof(struct kmem_cache_node), SLAB_HWCACHE_ALIGN); | |
b9049e23 | 4167 | |
3ac38faa | 4168 | register_hotmemory_notifier(&slab_memory_callback_nb); |
81819f0f CL |
4169 | |
4170 | /* Able to allocate the per node structures */ | |
4171 | slab_state = PARTIAL; | |
4172 | ||
dffb4d60 CL |
4173 | create_boot_cache(kmem_cache, "kmem_cache", |
4174 | offsetof(struct kmem_cache, node) + | |
4175 | nr_node_ids * sizeof(struct kmem_cache_node *), | |
4176 | SLAB_HWCACHE_ALIGN); | |
8a13a4cc | 4177 | |
dffb4d60 | 4178 | kmem_cache = bootstrap(&boot_kmem_cache); |
81819f0f | 4179 | |
51df1142 CL |
4180 | /* |
4181 | * Allocate kmem_cache_node properly from the kmem_cache slab. | |
4182 | * kmem_cache_node is separately allocated so no need to | |
4183 | * update any list pointers. | |
4184 | */ | |
dffb4d60 | 4185 | kmem_cache_node = bootstrap(&boot_kmem_cache_node); |
51df1142 CL |
4186 | |
4187 | /* Now we can use the kmem_cache to allocate kmalloc slabs */ | |
34cc6990 | 4188 | setup_kmalloc_cache_index_table(); |
f97d5f63 | 4189 | create_kmalloc_caches(0); |
81819f0f | 4190 | |
210e7a43 TG |
4191 | /* Setup random freelists for each cache */ |
4192 | init_freelist_randomization(); | |
4193 | ||
a96a87bf SAS |
4194 | cpuhp_setup_state_nocalls(CPUHP_SLUB_DEAD, "slub:dead", NULL, |
4195 | slub_cpu_dead); | |
81819f0f | 4196 | |
f9f58285 | 4197 | pr_info("SLUB: HWalign=%d, Order=%d-%d, MinObjects=%d, CPUs=%d, Nodes=%d\n", |
f97d5f63 | 4198 | cache_line_size(), |
81819f0f CL |
4199 | slub_min_order, slub_max_order, slub_min_objects, |
4200 | nr_cpu_ids, nr_node_ids); | |
4201 | } | |
4202 | ||
7e85ee0c PE |
4203 | void __init kmem_cache_init_late(void) |
4204 | { | |
7e85ee0c PE |
4205 | } |
4206 | ||
2633d7a0 | 4207 | struct kmem_cache * |
a44cb944 VD |
4208 | __kmem_cache_alias(const char *name, size_t size, size_t align, |
4209 | unsigned long flags, void (*ctor)(void *)) | |
81819f0f | 4210 | { |
426589f5 | 4211 | struct kmem_cache *s, *c; |
81819f0f | 4212 | |
a44cb944 | 4213 | s = find_mergeable(size, align, flags, name, ctor); |
81819f0f CL |
4214 | if (s) { |
4215 | s->refcount++; | |
84d0ddd6 | 4216 | |
81819f0f CL |
4217 | /* |
4218 | * Adjust the object sizes so that we clear | |
4219 | * the complete object on kzalloc. | |
4220 | */ | |
3b0efdfa | 4221 | s->object_size = max(s->object_size, (int)size); |
81819f0f | 4222 | s->inuse = max_t(int, s->inuse, ALIGN(size, sizeof(void *))); |
6446faa2 | 4223 | |
426589f5 | 4224 | for_each_memcg_cache(c, s) { |
84d0ddd6 VD |
4225 | c->object_size = s->object_size; |
4226 | c->inuse = max_t(int, c->inuse, | |
4227 | ALIGN(size, sizeof(void *))); | |
4228 | } | |
4229 | ||
7b8f3b66 | 4230 | if (sysfs_slab_alias(s, name)) { |
7b8f3b66 | 4231 | s->refcount--; |
cbb79694 | 4232 | s = NULL; |
7b8f3b66 | 4233 | } |
a0e1d1be | 4234 | } |
6446faa2 | 4235 | |
cbb79694 CL |
4236 | return s; |
4237 | } | |
84c1cf62 | 4238 | |
8a13a4cc | 4239 | int __kmem_cache_create(struct kmem_cache *s, unsigned long flags) |
cbb79694 | 4240 | { |
aac3a166 PE |
4241 | int err; |
4242 | ||
4243 | err = kmem_cache_open(s, flags); | |
4244 | if (err) | |
4245 | return err; | |
20cea968 | 4246 | |
45530c44 CL |
4247 | /* Mutex is not taken during early boot */ |
4248 | if (slab_state <= UP) | |
4249 | return 0; | |
4250 | ||
107dab5c | 4251 | memcg_propagate_slab_attrs(s); |
aac3a166 | 4252 | err = sysfs_slab_add(s); |
aac3a166 | 4253 | if (err) |
52b4b950 | 4254 | __kmem_cache_release(s); |
20cea968 | 4255 | |
aac3a166 | 4256 | return err; |
81819f0f | 4257 | } |
81819f0f | 4258 | |
ce71e27c | 4259 | void *__kmalloc_track_caller(size_t size, gfp_t gfpflags, unsigned long caller) |
81819f0f | 4260 | { |
aadb4bc4 | 4261 | struct kmem_cache *s; |
94b528d0 | 4262 | void *ret; |
aadb4bc4 | 4263 | |
95a05b42 | 4264 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) |
eada35ef PE |
4265 | return kmalloc_large(size, gfpflags); |
4266 | ||
2c59dd65 | 4267 | s = kmalloc_slab(size, gfpflags); |
81819f0f | 4268 | |
2408c550 | 4269 | if (unlikely(ZERO_OR_NULL_PTR(s))) |
6cb8f913 | 4270 | return s; |
81819f0f | 4271 | |
2b847c3c | 4272 | ret = slab_alloc(s, gfpflags, caller); |
94b528d0 | 4273 | |
25985edc | 4274 | /* Honor the call site pointer we received. */ |
ca2b84cb | 4275 | trace_kmalloc(caller, ret, size, s->size, gfpflags); |
94b528d0 EGM |
4276 | |
4277 | return ret; | |
81819f0f CL |
4278 | } |
4279 | ||
5d1f57e4 | 4280 | #ifdef CONFIG_NUMA |
81819f0f | 4281 | void *__kmalloc_node_track_caller(size_t size, gfp_t gfpflags, |
ce71e27c | 4282 | int node, unsigned long caller) |
81819f0f | 4283 | { |
aadb4bc4 | 4284 | struct kmem_cache *s; |
94b528d0 | 4285 | void *ret; |
aadb4bc4 | 4286 | |
95a05b42 | 4287 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) { |
d3e14aa3 XF |
4288 | ret = kmalloc_large_node(size, gfpflags, node); |
4289 | ||
4290 | trace_kmalloc_node(caller, ret, | |
4291 | size, PAGE_SIZE << get_order(size), | |
4292 | gfpflags, node); | |
4293 | ||
4294 | return ret; | |
4295 | } | |
eada35ef | 4296 | |
2c59dd65 | 4297 | s = kmalloc_slab(size, gfpflags); |
81819f0f | 4298 | |
2408c550 | 4299 | if (unlikely(ZERO_OR_NULL_PTR(s))) |
6cb8f913 | 4300 | return s; |
81819f0f | 4301 | |
2b847c3c | 4302 | ret = slab_alloc_node(s, gfpflags, node, caller); |
94b528d0 | 4303 | |
25985edc | 4304 | /* Honor the call site pointer we received. */ |
ca2b84cb | 4305 | trace_kmalloc_node(caller, ret, size, s->size, gfpflags, node); |
94b528d0 EGM |
4306 | |
4307 | return ret; | |
81819f0f | 4308 | } |
5d1f57e4 | 4309 | #endif |
81819f0f | 4310 | |
ab4d5ed5 | 4311 | #ifdef CONFIG_SYSFS |
205ab99d CL |
4312 | static int count_inuse(struct page *page) |
4313 | { | |
4314 | return page->inuse; | |
4315 | } | |
4316 | ||
4317 | static int count_total(struct page *page) | |
4318 | { | |
4319 | return page->objects; | |
4320 | } | |
ab4d5ed5 | 4321 | #endif |
205ab99d | 4322 | |
ab4d5ed5 | 4323 | #ifdef CONFIG_SLUB_DEBUG |
434e245d CL |
4324 | static int validate_slab(struct kmem_cache *s, struct page *page, |
4325 | unsigned long *map) | |
53e15af0 CL |
4326 | { |
4327 | void *p; | |
a973e9dd | 4328 | void *addr = page_address(page); |
53e15af0 CL |
4329 | |
4330 | if (!check_slab(s, page) || | |
4331 | !on_freelist(s, page, NULL)) | |
4332 | return 0; | |
4333 | ||
4334 | /* Now we know that a valid freelist exists */ | |
39b26464 | 4335 | bitmap_zero(map, page->objects); |
53e15af0 | 4336 | |
5f80b13a CL |
4337 | get_map(s, page, map); |
4338 | for_each_object(p, s, addr, page->objects) { | |
4339 | if (test_bit(slab_index(p, s, addr), map)) | |
4340 | if (!check_object(s, page, p, SLUB_RED_INACTIVE)) | |
4341 | return 0; | |
53e15af0 CL |
4342 | } |
4343 | ||
224a88be | 4344 | for_each_object(p, s, addr, page->objects) |
7656c72b | 4345 | if (!test_bit(slab_index(p, s, addr), map)) |
37d57443 | 4346 | if (!check_object(s, page, p, SLUB_RED_ACTIVE)) |
53e15af0 CL |
4347 | return 0; |
4348 | return 1; | |
4349 | } | |
4350 | ||
434e245d CL |
4351 | static void validate_slab_slab(struct kmem_cache *s, struct page *page, |
4352 | unsigned long *map) | |
53e15af0 | 4353 | { |
881db7fb CL |
4354 | slab_lock(page); |
4355 | validate_slab(s, page, map); | |
4356 | slab_unlock(page); | |
53e15af0 CL |
4357 | } |
4358 | ||
434e245d CL |
4359 | static int validate_slab_node(struct kmem_cache *s, |
4360 | struct kmem_cache_node *n, unsigned long *map) | |
53e15af0 CL |
4361 | { |
4362 | unsigned long count = 0; | |
4363 | struct page *page; | |
4364 | unsigned long flags; | |
4365 | ||
4366 | spin_lock_irqsave(&n->list_lock, flags); | |
4367 | ||
4368 | list_for_each_entry(page, &n->partial, lru) { | |
434e245d | 4369 | validate_slab_slab(s, page, map); |
53e15af0 CL |
4370 | count++; |
4371 | } | |
4372 | if (count != n->nr_partial) | |
f9f58285 FF |
4373 | pr_err("SLUB %s: %ld partial slabs counted but counter=%ld\n", |
4374 | s->name, count, n->nr_partial); | |
53e15af0 CL |
4375 | |
4376 | if (!(s->flags & SLAB_STORE_USER)) | |
4377 | goto out; | |
4378 | ||
4379 | list_for_each_entry(page, &n->full, lru) { | |
434e245d | 4380 | validate_slab_slab(s, page, map); |
53e15af0 CL |
4381 | count++; |
4382 | } | |
4383 | if (count != atomic_long_read(&n->nr_slabs)) | |
f9f58285 FF |
4384 | pr_err("SLUB: %s %ld slabs counted but counter=%ld\n", |
4385 | s->name, count, atomic_long_read(&n->nr_slabs)); | |
53e15af0 CL |
4386 | |
4387 | out: | |
4388 | spin_unlock_irqrestore(&n->list_lock, flags); | |
4389 | return count; | |
4390 | } | |
4391 | ||
434e245d | 4392 | static long validate_slab_cache(struct kmem_cache *s) |
53e15af0 CL |
4393 | { |
4394 | int node; | |
4395 | unsigned long count = 0; | |
205ab99d | 4396 | unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) * |
434e245d | 4397 | sizeof(unsigned long), GFP_KERNEL); |
fa45dc25 | 4398 | struct kmem_cache_node *n; |
434e245d CL |
4399 | |
4400 | if (!map) | |
4401 | return -ENOMEM; | |
53e15af0 CL |
4402 | |
4403 | flush_all(s); | |
fa45dc25 | 4404 | for_each_kmem_cache_node(s, node, n) |
434e245d | 4405 | count += validate_slab_node(s, n, map); |
434e245d | 4406 | kfree(map); |
53e15af0 CL |
4407 | return count; |
4408 | } | |
88a420e4 | 4409 | /* |
672bba3a | 4410 | * Generate lists of code addresses where slabcache objects are allocated |
88a420e4 CL |
4411 | * and freed. |
4412 | */ | |
4413 | ||
4414 | struct location { | |
4415 | unsigned long count; | |
ce71e27c | 4416 | unsigned long addr; |
45edfa58 CL |
4417 | long long sum_time; |
4418 | long min_time; | |
4419 | long max_time; | |
4420 | long min_pid; | |
4421 | long max_pid; | |
174596a0 | 4422 | DECLARE_BITMAP(cpus, NR_CPUS); |
45edfa58 | 4423 | nodemask_t nodes; |
88a420e4 CL |
4424 | }; |
4425 | ||
4426 | struct loc_track { | |
4427 | unsigned long max; | |
4428 | unsigned long count; | |
4429 | struct location *loc; | |
4430 | }; | |
4431 | ||
4432 | static void free_loc_track(struct loc_track *t) | |
4433 | { | |
4434 | if (t->max) | |
4435 | free_pages((unsigned long)t->loc, | |
4436 | get_order(sizeof(struct location) * t->max)); | |
4437 | } | |
4438 | ||
68dff6a9 | 4439 | static int alloc_loc_track(struct loc_track *t, unsigned long max, gfp_t flags) |
88a420e4 CL |
4440 | { |
4441 | struct location *l; | |
4442 | int order; | |
4443 | ||
88a420e4 CL |
4444 | order = get_order(sizeof(struct location) * max); |
4445 | ||
68dff6a9 | 4446 | l = (void *)__get_free_pages(flags, order); |
88a420e4 CL |
4447 | if (!l) |
4448 | return 0; | |
4449 | ||
4450 | if (t->count) { | |
4451 | memcpy(l, t->loc, sizeof(struct location) * t->count); | |
4452 | free_loc_track(t); | |
4453 | } | |
4454 | t->max = max; | |
4455 | t->loc = l; | |
4456 | return 1; | |
4457 | } | |
4458 | ||
4459 | static int add_location(struct loc_track *t, struct kmem_cache *s, | |
45edfa58 | 4460 | const struct track *track) |
88a420e4 CL |
4461 | { |
4462 | long start, end, pos; | |
4463 | struct location *l; | |
ce71e27c | 4464 | unsigned long caddr; |
45edfa58 | 4465 | unsigned long age = jiffies - track->when; |
88a420e4 CL |
4466 | |
4467 | start = -1; | |
4468 | end = t->count; | |
4469 | ||
4470 | for ( ; ; ) { | |
4471 | pos = start + (end - start + 1) / 2; | |
4472 | ||
4473 | /* | |
4474 | * There is nothing at "end". If we end up there | |
4475 | * we need to add something to before end. | |
4476 | */ | |
4477 | if (pos == end) | |
4478 | break; | |
4479 | ||
4480 | caddr = t->loc[pos].addr; | |
45edfa58 CL |
4481 | if (track->addr == caddr) { |
4482 | ||
4483 | l = &t->loc[pos]; | |
4484 | l->count++; | |
4485 | if (track->when) { | |
4486 | l->sum_time += age; | |
4487 | if (age < l->min_time) | |
4488 | l->min_time = age; | |
4489 | if (age > l->max_time) | |
4490 | l->max_time = age; | |
4491 | ||
4492 | if (track->pid < l->min_pid) | |
4493 | l->min_pid = track->pid; | |
4494 | if (track->pid > l->max_pid) | |
4495 | l->max_pid = track->pid; | |
4496 | ||
174596a0 RR |
4497 | cpumask_set_cpu(track->cpu, |
4498 | to_cpumask(l->cpus)); | |
45edfa58 CL |
4499 | } |
4500 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
88a420e4 CL |
4501 | return 1; |
4502 | } | |
4503 | ||
45edfa58 | 4504 | if (track->addr < caddr) |
88a420e4 CL |
4505 | end = pos; |
4506 | else | |
4507 | start = pos; | |
4508 | } | |
4509 | ||
4510 | /* | |
672bba3a | 4511 | * Not found. Insert new tracking element. |
88a420e4 | 4512 | */ |
68dff6a9 | 4513 | if (t->count >= t->max && !alloc_loc_track(t, 2 * t->max, GFP_ATOMIC)) |
88a420e4 CL |
4514 | return 0; |
4515 | ||
4516 | l = t->loc + pos; | |
4517 | if (pos < t->count) | |
4518 | memmove(l + 1, l, | |
4519 | (t->count - pos) * sizeof(struct location)); | |
4520 | t->count++; | |
4521 | l->count = 1; | |
45edfa58 CL |
4522 | l->addr = track->addr; |
4523 | l->sum_time = age; | |
4524 | l->min_time = age; | |
4525 | l->max_time = age; | |
4526 | l->min_pid = track->pid; | |
4527 | l->max_pid = track->pid; | |
174596a0 RR |
4528 | cpumask_clear(to_cpumask(l->cpus)); |
4529 | cpumask_set_cpu(track->cpu, to_cpumask(l->cpus)); | |
45edfa58 CL |
4530 | nodes_clear(l->nodes); |
4531 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
88a420e4 CL |
4532 | return 1; |
4533 | } | |
4534 | ||
4535 | static void process_slab(struct loc_track *t, struct kmem_cache *s, | |
bbd7d57b | 4536 | struct page *page, enum track_item alloc, |
a5dd5c11 | 4537 | unsigned long *map) |
88a420e4 | 4538 | { |
a973e9dd | 4539 | void *addr = page_address(page); |
88a420e4 CL |
4540 | void *p; |
4541 | ||
39b26464 | 4542 | bitmap_zero(map, page->objects); |
5f80b13a | 4543 | get_map(s, page, map); |
88a420e4 | 4544 | |
224a88be | 4545 | for_each_object(p, s, addr, page->objects) |
45edfa58 CL |
4546 | if (!test_bit(slab_index(p, s, addr), map)) |
4547 | add_location(t, s, get_track(s, p, alloc)); | |
88a420e4 CL |
4548 | } |
4549 | ||
4550 | static int list_locations(struct kmem_cache *s, char *buf, | |
4551 | enum track_item alloc) | |
4552 | { | |
e374d483 | 4553 | int len = 0; |
88a420e4 | 4554 | unsigned long i; |
68dff6a9 | 4555 | struct loc_track t = { 0, 0, NULL }; |
88a420e4 | 4556 | int node; |
bbd7d57b ED |
4557 | unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) * |
4558 | sizeof(unsigned long), GFP_KERNEL); | |
fa45dc25 | 4559 | struct kmem_cache_node *n; |
88a420e4 | 4560 | |
bbd7d57b ED |
4561 | if (!map || !alloc_loc_track(&t, PAGE_SIZE / sizeof(struct location), |
4562 | GFP_TEMPORARY)) { | |
4563 | kfree(map); | |
68dff6a9 | 4564 | return sprintf(buf, "Out of memory\n"); |
bbd7d57b | 4565 | } |
88a420e4 CL |
4566 | /* Push back cpu slabs */ |
4567 | flush_all(s); | |
4568 | ||
fa45dc25 | 4569 | for_each_kmem_cache_node(s, node, n) { |
88a420e4 CL |
4570 | unsigned long flags; |
4571 | struct page *page; | |
4572 | ||
9e86943b | 4573 | if (!atomic_long_read(&n->nr_slabs)) |
88a420e4 CL |
4574 | continue; |
4575 | ||
4576 | spin_lock_irqsave(&n->list_lock, flags); | |
4577 | list_for_each_entry(page, &n->partial, lru) | |
bbd7d57b | 4578 | process_slab(&t, s, page, alloc, map); |
88a420e4 | 4579 | list_for_each_entry(page, &n->full, lru) |
bbd7d57b | 4580 | process_slab(&t, s, page, alloc, map); |
88a420e4 CL |
4581 | spin_unlock_irqrestore(&n->list_lock, flags); |
4582 | } | |
4583 | ||
4584 | for (i = 0; i < t.count; i++) { | |
45edfa58 | 4585 | struct location *l = &t.loc[i]; |
88a420e4 | 4586 | |
9c246247 | 4587 | if (len > PAGE_SIZE - KSYM_SYMBOL_LEN - 100) |
88a420e4 | 4588 | break; |
e374d483 | 4589 | len += sprintf(buf + len, "%7ld ", l->count); |
45edfa58 CL |
4590 | |
4591 | if (l->addr) | |
62c70bce | 4592 | len += sprintf(buf + len, "%pS", (void *)l->addr); |
88a420e4 | 4593 | else |
e374d483 | 4594 | len += sprintf(buf + len, "<not-available>"); |
45edfa58 CL |
4595 | |
4596 | if (l->sum_time != l->min_time) { | |
e374d483 | 4597 | len += sprintf(buf + len, " age=%ld/%ld/%ld", |
f8bd2258 RZ |
4598 | l->min_time, |
4599 | (long)div_u64(l->sum_time, l->count), | |
4600 | l->max_time); | |
45edfa58 | 4601 | } else |
e374d483 | 4602 | len += sprintf(buf + len, " age=%ld", |
45edfa58 CL |
4603 | l->min_time); |
4604 | ||
4605 | if (l->min_pid != l->max_pid) | |
e374d483 | 4606 | len += sprintf(buf + len, " pid=%ld-%ld", |
45edfa58 CL |
4607 | l->min_pid, l->max_pid); |
4608 | else | |
e374d483 | 4609 | len += sprintf(buf + len, " pid=%ld", |
45edfa58 CL |
4610 | l->min_pid); |
4611 | ||
174596a0 RR |
4612 | if (num_online_cpus() > 1 && |
4613 | !cpumask_empty(to_cpumask(l->cpus)) && | |
5024c1d7 TH |
4614 | len < PAGE_SIZE - 60) |
4615 | len += scnprintf(buf + len, PAGE_SIZE - len - 50, | |
4616 | " cpus=%*pbl", | |
4617 | cpumask_pr_args(to_cpumask(l->cpus))); | |
45edfa58 | 4618 | |
62bc62a8 | 4619 | if (nr_online_nodes > 1 && !nodes_empty(l->nodes) && |
5024c1d7 TH |
4620 | len < PAGE_SIZE - 60) |
4621 | len += scnprintf(buf + len, PAGE_SIZE - len - 50, | |
4622 | " nodes=%*pbl", | |
4623 | nodemask_pr_args(&l->nodes)); | |
45edfa58 | 4624 | |
e374d483 | 4625 | len += sprintf(buf + len, "\n"); |
88a420e4 CL |
4626 | } |
4627 | ||
4628 | free_loc_track(&t); | |
bbd7d57b | 4629 | kfree(map); |
88a420e4 | 4630 | if (!t.count) |
e374d483 HH |
4631 | len += sprintf(buf, "No data\n"); |
4632 | return len; | |
88a420e4 | 4633 | } |
ab4d5ed5 | 4634 | #endif |
88a420e4 | 4635 | |
a5a84755 | 4636 | #ifdef SLUB_RESILIENCY_TEST |
c07b8183 | 4637 | static void __init resiliency_test(void) |
a5a84755 CL |
4638 | { |
4639 | u8 *p; | |
4640 | ||
95a05b42 | 4641 | BUILD_BUG_ON(KMALLOC_MIN_SIZE > 16 || KMALLOC_SHIFT_HIGH < 10); |
a5a84755 | 4642 | |
f9f58285 FF |
4643 | pr_err("SLUB resiliency testing\n"); |
4644 | pr_err("-----------------------\n"); | |
4645 | pr_err("A. Corruption after allocation\n"); | |
a5a84755 CL |
4646 | |
4647 | p = kzalloc(16, GFP_KERNEL); | |
4648 | p[16] = 0x12; | |
f9f58285 FF |
4649 | pr_err("\n1. kmalloc-16: Clobber Redzone/next pointer 0x12->0x%p\n\n", |
4650 | p + 16); | |
a5a84755 CL |
4651 | |
4652 | validate_slab_cache(kmalloc_caches[4]); | |
4653 | ||
4654 | /* Hmmm... The next two are dangerous */ | |
4655 | p = kzalloc(32, GFP_KERNEL); | |
4656 | p[32 + sizeof(void *)] = 0x34; | |
f9f58285 FF |
4657 | pr_err("\n2. kmalloc-32: Clobber next pointer/next slab 0x34 -> -0x%p\n", |
4658 | p); | |
4659 | pr_err("If allocated object is overwritten then not detectable\n\n"); | |
a5a84755 CL |
4660 | |
4661 | validate_slab_cache(kmalloc_caches[5]); | |
4662 | p = kzalloc(64, GFP_KERNEL); | |
4663 | p += 64 + (get_cycles() & 0xff) * sizeof(void *); | |
4664 | *p = 0x56; | |
f9f58285 FF |
4665 | pr_err("\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n", |
4666 | p); | |
4667 | pr_err("If allocated object is overwritten then not detectable\n\n"); | |
a5a84755 CL |
4668 | validate_slab_cache(kmalloc_caches[6]); |
4669 | ||
f9f58285 | 4670 | pr_err("\nB. Corruption after free\n"); |
a5a84755 CL |
4671 | p = kzalloc(128, GFP_KERNEL); |
4672 | kfree(p); | |
4673 | *p = 0x78; | |
f9f58285 | 4674 | pr_err("1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p); |
a5a84755 CL |
4675 | validate_slab_cache(kmalloc_caches[7]); |
4676 | ||
4677 | p = kzalloc(256, GFP_KERNEL); | |
4678 | kfree(p); | |
4679 | p[50] = 0x9a; | |
f9f58285 | 4680 | pr_err("\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n", p); |
a5a84755 CL |
4681 | validate_slab_cache(kmalloc_caches[8]); |
4682 | ||
4683 | p = kzalloc(512, GFP_KERNEL); | |
4684 | kfree(p); | |
4685 | p[512] = 0xab; | |
f9f58285 | 4686 | pr_err("\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p); |
a5a84755 CL |
4687 | validate_slab_cache(kmalloc_caches[9]); |
4688 | } | |
4689 | #else | |
4690 | #ifdef CONFIG_SYSFS | |
4691 | static void resiliency_test(void) {}; | |
4692 | #endif | |
4693 | #endif | |
4694 | ||
ab4d5ed5 | 4695 | #ifdef CONFIG_SYSFS |
81819f0f | 4696 | enum slab_stat_type { |
205ab99d CL |
4697 | SL_ALL, /* All slabs */ |
4698 | SL_PARTIAL, /* Only partially allocated slabs */ | |
4699 | SL_CPU, /* Only slabs used for cpu caches */ | |
4700 | SL_OBJECTS, /* Determine allocated objects not slabs */ | |
4701 | SL_TOTAL /* Determine object capacity not slabs */ | |
81819f0f CL |
4702 | }; |
4703 | ||
205ab99d | 4704 | #define SO_ALL (1 << SL_ALL) |
81819f0f CL |
4705 | #define SO_PARTIAL (1 << SL_PARTIAL) |
4706 | #define SO_CPU (1 << SL_CPU) | |
4707 | #define SO_OBJECTS (1 << SL_OBJECTS) | |
205ab99d | 4708 | #define SO_TOTAL (1 << SL_TOTAL) |
81819f0f | 4709 | |
1663f26d TH |
4710 | #ifdef CONFIG_MEMCG |
4711 | static bool memcg_sysfs_enabled = IS_ENABLED(CONFIG_SLUB_MEMCG_SYSFS_ON); | |
4712 | ||
4713 | static int __init setup_slub_memcg_sysfs(char *str) | |
4714 | { | |
4715 | int v; | |
4716 | ||
4717 | if (get_option(&str, &v) > 0) | |
4718 | memcg_sysfs_enabled = v; | |
4719 | ||
4720 | return 1; | |
4721 | } | |
4722 | ||
4723 | __setup("slub_memcg_sysfs=", setup_slub_memcg_sysfs); | |
4724 | #endif | |
4725 | ||
62e5c4b4 CG |
4726 | static ssize_t show_slab_objects(struct kmem_cache *s, |
4727 | char *buf, unsigned long flags) | |
81819f0f CL |
4728 | { |
4729 | unsigned long total = 0; | |
81819f0f CL |
4730 | int node; |
4731 | int x; | |
4732 | unsigned long *nodes; | |
81819f0f | 4733 | |
e35e1a97 | 4734 | nodes = kzalloc(sizeof(unsigned long) * nr_node_ids, GFP_KERNEL); |
62e5c4b4 CG |
4735 | if (!nodes) |
4736 | return -ENOMEM; | |
81819f0f | 4737 | |
205ab99d CL |
4738 | if (flags & SO_CPU) { |
4739 | int cpu; | |
81819f0f | 4740 | |
205ab99d | 4741 | for_each_possible_cpu(cpu) { |
d0e0ac97 CG |
4742 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, |
4743 | cpu); | |
ec3ab083 | 4744 | int node; |
49e22585 | 4745 | struct page *page; |
dfb4f096 | 4746 | |
4db0c3c2 | 4747 | page = READ_ONCE(c->page); |
ec3ab083 CL |
4748 | if (!page) |
4749 | continue; | |
205ab99d | 4750 | |
ec3ab083 CL |
4751 | node = page_to_nid(page); |
4752 | if (flags & SO_TOTAL) | |
4753 | x = page->objects; | |
4754 | else if (flags & SO_OBJECTS) | |
4755 | x = page->inuse; | |
4756 | else | |
4757 | x = 1; | |
49e22585 | 4758 | |
ec3ab083 CL |
4759 | total += x; |
4760 | nodes[node] += x; | |
4761 | ||
4db0c3c2 | 4762 | page = READ_ONCE(c->partial); |
49e22585 | 4763 | if (page) { |
8afb1474 LZ |
4764 | node = page_to_nid(page); |
4765 | if (flags & SO_TOTAL) | |
4766 | WARN_ON_ONCE(1); | |
4767 | else if (flags & SO_OBJECTS) | |
4768 | WARN_ON_ONCE(1); | |
4769 | else | |
4770 | x = page->pages; | |
bc6697d8 ED |
4771 | total += x; |
4772 | nodes[node] += x; | |
49e22585 | 4773 | } |
81819f0f CL |
4774 | } |
4775 | } | |
4776 | ||
bfc8c901 | 4777 | get_online_mems(); |
ab4d5ed5 | 4778 | #ifdef CONFIG_SLUB_DEBUG |
205ab99d | 4779 | if (flags & SO_ALL) { |
fa45dc25 CL |
4780 | struct kmem_cache_node *n; |
4781 | ||
4782 | for_each_kmem_cache_node(s, node, n) { | |
205ab99d | 4783 | |
d0e0ac97 CG |
4784 | if (flags & SO_TOTAL) |
4785 | x = atomic_long_read(&n->total_objects); | |
4786 | else if (flags & SO_OBJECTS) | |
4787 | x = atomic_long_read(&n->total_objects) - | |
4788 | count_partial(n, count_free); | |
81819f0f | 4789 | else |
205ab99d | 4790 | x = atomic_long_read(&n->nr_slabs); |
81819f0f CL |
4791 | total += x; |
4792 | nodes[node] += x; | |
4793 | } | |
4794 | ||
ab4d5ed5 CL |
4795 | } else |
4796 | #endif | |
4797 | if (flags & SO_PARTIAL) { | |
fa45dc25 | 4798 | struct kmem_cache_node *n; |
81819f0f | 4799 | |
fa45dc25 | 4800 | for_each_kmem_cache_node(s, node, n) { |
205ab99d CL |
4801 | if (flags & SO_TOTAL) |
4802 | x = count_partial(n, count_total); | |
4803 | else if (flags & SO_OBJECTS) | |
4804 | x = count_partial(n, count_inuse); | |
81819f0f | 4805 | else |
205ab99d | 4806 | x = n->nr_partial; |
81819f0f CL |
4807 | total += x; |
4808 | nodes[node] += x; | |
4809 | } | |
4810 | } | |
81819f0f CL |
4811 | x = sprintf(buf, "%lu", total); |
4812 | #ifdef CONFIG_NUMA | |
fa45dc25 | 4813 | for (node = 0; node < nr_node_ids; node++) |
81819f0f CL |
4814 | if (nodes[node]) |
4815 | x += sprintf(buf + x, " N%d=%lu", | |
4816 | node, nodes[node]); | |
4817 | #endif | |
bfc8c901 | 4818 | put_online_mems(); |
81819f0f CL |
4819 | kfree(nodes); |
4820 | return x + sprintf(buf + x, "\n"); | |
4821 | } | |
4822 | ||
ab4d5ed5 | 4823 | #ifdef CONFIG_SLUB_DEBUG |
81819f0f CL |
4824 | static int any_slab_objects(struct kmem_cache *s) |
4825 | { | |
4826 | int node; | |
fa45dc25 | 4827 | struct kmem_cache_node *n; |
81819f0f | 4828 | |
fa45dc25 | 4829 | for_each_kmem_cache_node(s, node, n) |
4ea33e2d | 4830 | if (atomic_long_read(&n->total_objects)) |
81819f0f | 4831 | return 1; |
fa45dc25 | 4832 | |
81819f0f CL |
4833 | return 0; |
4834 | } | |
ab4d5ed5 | 4835 | #endif |
81819f0f CL |
4836 | |
4837 | #define to_slab_attr(n) container_of(n, struct slab_attribute, attr) | |
497888cf | 4838 | #define to_slab(n) container_of(n, struct kmem_cache, kobj) |
81819f0f CL |
4839 | |
4840 | struct slab_attribute { | |
4841 | struct attribute attr; | |
4842 | ssize_t (*show)(struct kmem_cache *s, char *buf); | |
4843 | ssize_t (*store)(struct kmem_cache *s, const char *x, size_t count); | |
4844 | }; | |
4845 | ||
4846 | #define SLAB_ATTR_RO(_name) \ | |
ab067e99 VK |
4847 | static struct slab_attribute _name##_attr = \ |
4848 | __ATTR(_name, 0400, _name##_show, NULL) | |
81819f0f CL |
4849 | |
4850 | #define SLAB_ATTR(_name) \ | |
4851 | static struct slab_attribute _name##_attr = \ | |
ab067e99 | 4852 | __ATTR(_name, 0600, _name##_show, _name##_store) |
81819f0f | 4853 | |
81819f0f CL |
4854 | static ssize_t slab_size_show(struct kmem_cache *s, char *buf) |
4855 | { | |
4856 | return sprintf(buf, "%d\n", s->size); | |
4857 | } | |
4858 | SLAB_ATTR_RO(slab_size); | |
4859 | ||
4860 | static ssize_t align_show(struct kmem_cache *s, char *buf) | |
4861 | { | |
4862 | return sprintf(buf, "%d\n", s->align); | |
4863 | } | |
4864 | SLAB_ATTR_RO(align); | |
4865 | ||
4866 | static ssize_t object_size_show(struct kmem_cache *s, char *buf) | |
4867 | { | |
3b0efdfa | 4868 | return sprintf(buf, "%d\n", s->object_size); |
81819f0f CL |
4869 | } |
4870 | SLAB_ATTR_RO(object_size); | |
4871 | ||
4872 | static ssize_t objs_per_slab_show(struct kmem_cache *s, char *buf) | |
4873 | { | |
834f3d11 | 4874 | return sprintf(buf, "%d\n", oo_objects(s->oo)); |
81819f0f CL |
4875 | } |
4876 | SLAB_ATTR_RO(objs_per_slab); | |
4877 | ||
06b285dc CL |
4878 | static ssize_t order_store(struct kmem_cache *s, |
4879 | const char *buf, size_t length) | |
4880 | { | |
0121c619 CL |
4881 | unsigned long order; |
4882 | int err; | |
4883 | ||
3dbb95f7 | 4884 | err = kstrtoul(buf, 10, &order); |
0121c619 CL |
4885 | if (err) |
4886 | return err; | |
06b285dc CL |
4887 | |
4888 | if (order > slub_max_order || order < slub_min_order) | |
4889 | return -EINVAL; | |
4890 | ||
4891 | calculate_sizes(s, order); | |
4892 | return length; | |
4893 | } | |
4894 | ||
81819f0f CL |
4895 | static ssize_t order_show(struct kmem_cache *s, char *buf) |
4896 | { | |
834f3d11 | 4897 | return sprintf(buf, "%d\n", oo_order(s->oo)); |
81819f0f | 4898 | } |
06b285dc | 4899 | SLAB_ATTR(order); |
81819f0f | 4900 | |
73d342b1 DR |
4901 | static ssize_t min_partial_show(struct kmem_cache *s, char *buf) |
4902 | { | |
4903 | return sprintf(buf, "%lu\n", s->min_partial); | |
4904 | } | |
4905 | ||
4906 | static ssize_t min_partial_store(struct kmem_cache *s, const char *buf, | |
4907 | size_t length) | |
4908 | { | |
4909 | unsigned long min; | |
4910 | int err; | |
4911 | ||
3dbb95f7 | 4912 | err = kstrtoul(buf, 10, &min); |
73d342b1 DR |
4913 | if (err) |
4914 | return err; | |
4915 | ||
c0bdb232 | 4916 | set_min_partial(s, min); |
73d342b1 DR |
4917 | return length; |
4918 | } | |
4919 | SLAB_ATTR(min_partial); | |
4920 | ||
49e22585 CL |
4921 | static ssize_t cpu_partial_show(struct kmem_cache *s, char *buf) |
4922 | { | |
4923 | return sprintf(buf, "%u\n", s->cpu_partial); | |
4924 | } | |
4925 | ||
4926 | static ssize_t cpu_partial_store(struct kmem_cache *s, const char *buf, | |
4927 | size_t length) | |
4928 | { | |
4929 | unsigned long objects; | |
4930 | int err; | |
4931 | ||
3dbb95f7 | 4932 | err = kstrtoul(buf, 10, &objects); |
49e22585 CL |
4933 | if (err) |
4934 | return err; | |
345c905d | 4935 | if (objects && !kmem_cache_has_cpu_partial(s)) |
74ee4ef1 | 4936 | return -EINVAL; |
49e22585 CL |
4937 | |
4938 | s->cpu_partial = objects; | |
4939 | flush_all(s); | |
4940 | return length; | |
4941 | } | |
4942 | SLAB_ATTR(cpu_partial); | |
4943 | ||
81819f0f CL |
4944 | static ssize_t ctor_show(struct kmem_cache *s, char *buf) |
4945 | { | |
62c70bce JP |
4946 | if (!s->ctor) |
4947 | return 0; | |
4948 | return sprintf(buf, "%pS\n", s->ctor); | |
81819f0f CL |
4949 | } |
4950 | SLAB_ATTR_RO(ctor); | |
4951 | ||
81819f0f CL |
4952 | static ssize_t aliases_show(struct kmem_cache *s, char *buf) |
4953 | { | |
4307c14f | 4954 | return sprintf(buf, "%d\n", s->refcount < 0 ? 0 : s->refcount - 1); |
81819f0f CL |
4955 | } |
4956 | SLAB_ATTR_RO(aliases); | |
4957 | ||
81819f0f CL |
4958 | static ssize_t partial_show(struct kmem_cache *s, char *buf) |
4959 | { | |
d9acf4b7 | 4960 | return show_slab_objects(s, buf, SO_PARTIAL); |
81819f0f CL |
4961 | } |
4962 | SLAB_ATTR_RO(partial); | |
4963 | ||
4964 | static ssize_t cpu_slabs_show(struct kmem_cache *s, char *buf) | |
4965 | { | |
d9acf4b7 | 4966 | return show_slab_objects(s, buf, SO_CPU); |
81819f0f CL |
4967 | } |
4968 | SLAB_ATTR_RO(cpu_slabs); | |
4969 | ||
4970 | static ssize_t objects_show(struct kmem_cache *s, char *buf) | |
4971 | { | |
205ab99d | 4972 | return show_slab_objects(s, buf, SO_ALL|SO_OBJECTS); |
81819f0f CL |
4973 | } |
4974 | SLAB_ATTR_RO(objects); | |
4975 | ||
205ab99d CL |
4976 | static ssize_t objects_partial_show(struct kmem_cache *s, char *buf) |
4977 | { | |
4978 | return show_slab_objects(s, buf, SO_PARTIAL|SO_OBJECTS); | |
4979 | } | |
4980 | SLAB_ATTR_RO(objects_partial); | |
4981 | ||
49e22585 CL |
4982 | static ssize_t slabs_cpu_partial_show(struct kmem_cache *s, char *buf) |
4983 | { | |
4984 | int objects = 0; | |
4985 | int pages = 0; | |
4986 | int cpu; | |
4987 | int len; | |
4988 | ||
4989 | for_each_online_cpu(cpu) { | |
4990 | struct page *page = per_cpu_ptr(s->cpu_slab, cpu)->partial; | |
4991 | ||
4992 | if (page) { | |
4993 | pages += page->pages; | |
4994 | objects += page->pobjects; | |
4995 | } | |
4996 | } | |
4997 | ||
4998 | len = sprintf(buf, "%d(%d)", objects, pages); | |
4999 | ||
5000 | #ifdef CONFIG_SMP | |
5001 | for_each_online_cpu(cpu) { | |
5002 | struct page *page = per_cpu_ptr(s->cpu_slab, cpu) ->partial; | |
5003 | ||
5004 | if (page && len < PAGE_SIZE - 20) | |
5005 | len += sprintf(buf + len, " C%d=%d(%d)", cpu, | |
5006 | page->pobjects, page->pages); | |
5007 | } | |
5008 | #endif | |
5009 | return len + sprintf(buf + len, "\n"); | |
5010 | } | |
5011 | SLAB_ATTR_RO(slabs_cpu_partial); | |
5012 | ||
a5a84755 CL |
5013 | static ssize_t reclaim_account_show(struct kmem_cache *s, char *buf) |
5014 | { | |
5015 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_RECLAIM_ACCOUNT)); | |
5016 | } | |
5017 | ||
5018 | static ssize_t reclaim_account_store(struct kmem_cache *s, | |
5019 | const char *buf, size_t length) | |
5020 | { | |
5021 | s->flags &= ~SLAB_RECLAIM_ACCOUNT; | |
5022 | if (buf[0] == '1') | |
5023 | s->flags |= SLAB_RECLAIM_ACCOUNT; | |
5024 | return length; | |
5025 | } | |
5026 | SLAB_ATTR(reclaim_account); | |
5027 | ||
5028 | static ssize_t hwcache_align_show(struct kmem_cache *s, char *buf) | |
5029 | { | |
5030 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_HWCACHE_ALIGN)); | |
5031 | } | |
5032 | SLAB_ATTR_RO(hwcache_align); | |
5033 | ||
5034 | #ifdef CONFIG_ZONE_DMA | |
5035 | static ssize_t cache_dma_show(struct kmem_cache *s, char *buf) | |
5036 | { | |
5037 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_CACHE_DMA)); | |
5038 | } | |
5039 | SLAB_ATTR_RO(cache_dma); | |
5040 | #endif | |
5041 | ||
5042 | static ssize_t destroy_by_rcu_show(struct kmem_cache *s, char *buf) | |
5043 | { | |
5f0d5a3a | 5044 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_TYPESAFE_BY_RCU)); |
a5a84755 CL |
5045 | } |
5046 | SLAB_ATTR_RO(destroy_by_rcu); | |
5047 | ||
ab9a0f19 LJ |
5048 | static ssize_t reserved_show(struct kmem_cache *s, char *buf) |
5049 | { | |
5050 | return sprintf(buf, "%d\n", s->reserved); | |
5051 | } | |
5052 | SLAB_ATTR_RO(reserved); | |
5053 | ||
ab4d5ed5 | 5054 | #ifdef CONFIG_SLUB_DEBUG |
a5a84755 CL |
5055 | static ssize_t slabs_show(struct kmem_cache *s, char *buf) |
5056 | { | |
5057 | return show_slab_objects(s, buf, SO_ALL); | |
5058 | } | |
5059 | SLAB_ATTR_RO(slabs); | |
5060 | ||
205ab99d CL |
5061 | static ssize_t total_objects_show(struct kmem_cache *s, char *buf) |
5062 | { | |
5063 | return show_slab_objects(s, buf, SO_ALL|SO_TOTAL); | |
5064 | } | |
5065 | SLAB_ATTR_RO(total_objects); | |
5066 | ||
81819f0f CL |
5067 | static ssize_t sanity_checks_show(struct kmem_cache *s, char *buf) |
5068 | { | |
becfda68 | 5069 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_CONSISTENCY_CHECKS)); |
81819f0f CL |
5070 | } |
5071 | ||
5072 | static ssize_t sanity_checks_store(struct kmem_cache *s, | |
5073 | const char *buf, size_t length) | |
5074 | { | |
becfda68 | 5075 | s->flags &= ~SLAB_CONSISTENCY_CHECKS; |
b789ef51 CL |
5076 | if (buf[0] == '1') { |
5077 | s->flags &= ~__CMPXCHG_DOUBLE; | |
becfda68 | 5078 | s->flags |= SLAB_CONSISTENCY_CHECKS; |
b789ef51 | 5079 | } |
81819f0f CL |
5080 | return length; |
5081 | } | |
5082 | SLAB_ATTR(sanity_checks); | |
5083 | ||
5084 | static ssize_t trace_show(struct kmem_cache *s, char *buf) | |
5085 | { | |
5086 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_TRACE)); | |
5087 | } | |
5088 | ||
5089 | static ssize_t trace_store(struct kmem_cache *s, const char *buf, | |
5090 | size_t length) | |
5091 | { | |
c9e16131 CL |
5092 | /* |
5093 | * Tracing a merged cache is going to give confusing results | |
5094 | * as well as cause other issues like converting a mergeable | |
5095 | * cache into an umergeable one. | |
5096 | */ | |
5097 | if (s->refcount > 1) | |
5098 | return -EINVAL; | |
5099 | ||
81819f0f | 5100 | s->flags &= ~SLAB_TRACE; |
b789ef51 CL |
5101 | if (buf[0] == '1') { |
5102 | s->flags &= ~__CMPXCHG_DOUBLE; | |
81819f0f | 5103 | s->flags |= SLAB_TRACE; |
b789ef51 | 5104 | } |
81819f0f CL |
5105 | return length; |
5106 | } | |
5107 | SLAB_ATTR(trace); | |
5108 | ||
81819f0f CL |
5109 | static ssize_t red_zone_show(struct kmem_cache *s, char *buf) |
5110 | { | |
5111 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_RED_ZONE)); | |
5112 | } | |
5113 | ||
5114 | static ssize_t red_zone_store(struct kmem_cache *s, | |
5115 | const char *buf, size_t length) | |
5116 | { | |
5117 | if (any_slab_objects(s)) | |
5118 | return -EBUSY; | |
5119 | ||
5120 | s->flags &= ~SLAB_RED_ZONE; | |
b789ef51 | 5121 | if (buf[0] == '1') { |
81819f0f | 5122 | s->flags |= SLAB_RED_ZONE; |
b789ef51 | 5123 | } |
06b285dc | 5124 | calculate_sizes(s, -1); |
81819f0f CL |
5125 | return length; |
5126 | } | |
5127 | SLAB_ATTR(red_zone); | |
5128 | ||
5129 | static ssize_t poison_show(struct kmem_cache *s, char *buf) | |
5130 | { | |
5131 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_POISON)); | |
5132 | } | |
5133 | ||
5134 | static ssize_t poison_store(struct kmem_cache *s, | |
5135 | const char *buf, size_t length) | |
5136 | { | |
5137 | if (any_slab_objects(s)) | |
5138 | return -EBUSY; | |
5139 | ||
5140 | s->flags &= ~SLAB_POISON; | |
b789ef51 | 5141 | if (buf[0] == '1') { |
81819f0f | 5142 | s->flags |= SLAB_POISON; |
b789ef51 | 5143 | } |
06b285dc | 5144 | calculate_sizes(s, -1); |
81819f0f CL |
5145 | return length; |
5146 | } | |
5147 | SLAB_ATTR(poison); | |
5148 | ||
5149 | static ssize_t store_user_show(struct kmem_cache *s, char *buf) | |
5150 | { | |
5151 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_STORE_USER)); | |
5152 | } | |
5153 | ||
5154 | static ssize_t store_user_store(struct kmem_cache *s, | |
5155 | const char *buf, size_t length) | |
5156 | { | |
5157 | if (any_slab_objects(s)) | |
5158 | return -EBUSY; | |
5159 | ||
5160 | s->flags &= ~SLAB_STORE_USER; | |
b789ef51 CL |
5161 | if (buf[0] == '1') { |
5162 | s->flags &= ~__CMPXCHG_DOUBLE; | |
81819f0f | 5163 | s->flags |= SLAB_STORE_USER; |
b789ef51 | 5164 | } |
06b285dc | 5165 | calculate_sizes(s, -1); |
81819f0f CL |
5166 | return length; |
5167 | } | |
5168 | SLAB_ATTR(store_user); | |
5169 | ||
53e15af0 CL |
5170 | static ssize_t validate_show(struct kmem_cache *s, char *buf) |
5171 | { | |
5172 | return 0; | |
5173 | } | |
5174 | ||
5175 | static ssize_t validate_store(struct kmem_cache *s, | |
5176 | const char *buf, size_t length) | |
5177 | { | |
434e245d CL |
5178 | int ret = -EINVAL; |
5179 | ||
5180 | if (buf[0] == '1') { | |
5181 | ret = validate_slab_cache(s); | |
5182 | if (ret >= 0) | |
5183 | ret = length; | |
5184 | } | |
5185 | return ret; | |
53e15af0 CL |
5186 | } |
5187 | SLAB_ATTR(validate); | |
a5a84755 CL |
5188 | |
5189 | static ssize_t alloc_calls_show(struct kmem_cache *s, char *buf) | |
5190 | { | |
5191 | if (!(s->flags & SLAB_STORE_USER)) | |
5192 | return -ENOSYS; | |
5193 | return list_locations(s, buf, TRACK_ALLOC); | |
5194 | } | |
5195 | SLAB_ATTR_RO(alloc_calls); | |
5196 | ||
5197 | static ssize_t free_calls_show(struct kmem_cache *s, char *buf) | |
5198 | { | |
5199 | if (!(s->flags & SLAB_STORE_USER)) | |
5200 | return -ENOSYS; | |
5201 | return list_locations(s, buf, TRACK_FREE); | |
5202 | } | |
5203 | SLAB_ATTR_RO(free_calls); | |
5204 | #endif /* CONFIG_SLUB_DEBUG */ | |
5205 | ||
5206 | #ifdef CONFIG_FAILSLAB | |
5207 | static ssize_t failslab_show(struct kmem_cache *s, char *buf) | |
5208 | { | |
5209 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_FAILSLAB)); | |
5210 | } | |
5211 | ||
5212 | static ssize_t failslab_store(struct kmem_cache *s, const char *buf, | |
5213 | size_t length) | |
5214 | { | |
c9e16131 CL |
5215 | if (s->refcount > 1) |
5216 | return -EINVAL; | |
5217 | ||
a5a84755 CL |
5218 | s->flags &= ~SLAB_FAILSLAB; |
5219 | if (buf[0] == '1') | |
5220 | s->flags |= SLAB_FAILSLAB; | |
5221 | return length; | |
5222 | } | |
5223 | SLAB_ATTR(failslab); | |
ab4d5ed5 | 5224 | #endif |
53e15af0 | 5225 | |
2086d26a CL |
5226 | static ssize_t shrink_show(struct kmem_cache *s, char *buf) |
5227 | { | |
5228 | return 0; | |
5229 | } | |
5230 | ||
5231 | static ssize_t shrink_store(struct kmem_cache *s, | |
5232 | const char *buf, size_t length) | |
5233 | { | |
832f37f5 VD |
5234 | if (buf[0] == '1') |
5235 | kmem_cache_shrink(s); | |
5236 | else | |
2086d26a CL |
5237 | return -EINVAL; |
5238 | return length; | |
5239 | } | |
5240 | SLAB_ATTR(shrink); | |
5241 | ||
81819f0f | 5242 | #ifdef CONFIG_NUMA |
9824601e | 5243 | static ssize_t remote_node_defrag_ratio_show(struct kmem_cache *s, char *buf) |
81819f0f | 5244 | { |
9824601e | 5245 | return sprintf(buf, "%d\n", s->remote_node_defrag_ratio / 10); |
81819f0f CL |
5246 | } |
5247 | ||
9824601e | 5248 | static ssize_t remote_node_defrag_ratio_store(struct kmem_cache *s, |
81819f0f CL |
5249 | const char *buf, size_t length) |
5250 | { | |
0121c619 CL |
5251 | unsigned long ratio; |
5252 | int err; | |
5253 | ||
3dbb95f7 | 5254 | err = kstrtoul(buf, 10, &ratio); |
0121c619 CL |
5255 | if (err) |
5256 | return err; | |
5257 | ||
e2cb96b7 | 5258 | if (ratio <= 100) |
0121c619 | 5259 | s->remote_node_defrag_ratio = ratio * 10; |
81819f0f | 5260 | |
81819f0f CL |
5261 | return length; |
5262 | } | |
9824601e | 5263 | SLAB_ATTR(remote_node_defrag_ratio); |
81819f0f CL |
5264 | #endif |
5265 | ||
8ff12cfc | 5266 | #ifdef CONFIG_SLUB_STATS |
8ff12cfc CL |
5267 | static int show_stat(struct kmem_cache *s, char *buf, enum stat_item si) |
5268 | { | |
5269 | unsigned long sum = 0; | |
5270 | int cpu; | |
5271 | int len; | |
5272 | int *data = kmalloc(nr_cpu_ids * sizeof(int), GFP_KERNEL); | |
5273 | ||
5274 | if (!data) | |
5275 | return -ENOMEM; | |
5276 | ||
5277 | for_each_online_cpu(cpu) { | |
9dfc6e68 | 5278 | unsigned x = per_cpu_ptr(s->cpu_slab, cpu)->stat[si]; |
8ff12cfc CL |
5279 | |
5280 | data[cpu] = x; | |
5281 | sum += x; | |
5282 | } | |
5283 | ||
5284 | len = sprintf(buf, "%lu", sum); | |
5285 | ||
50ef37b9 | 5286 | #ifdef CONFIG_SMP |
8ff12cfc CL |
5287 | for_each_online_cpu(cpu) { |
5288 | if (data[cpu] && len < PAGE_SIZE - 20) | |
50ef37b9 | 5289 | len += sprintf(buf + len, " C%d=%u", cpu, data[cpu]); |
8ff12cfc | 5290 | } |
50ef37b9 | 5291 | #endif |
8ff12cfc CL |
5292 | kfree(data); |
5293 | return len + sprintf(buf + len, "\n"); | |
5294 | } | |
5295 | ||
78eb00cc DR |
5296 | static void clear_stat(struct kmem_cache *s, enum stat_item si) |
5297 | { | |
5298 | int cpu; | |
5299 | ||
5300 | for_each_online_cpu(cpu) | |
9dfc6e68 | 5301 | per_cpu_ptr(s->cpu_slab, cpu)->stat[si] = 0; |
78eb00cc DR |
5302 | } |
5303 | ||
8ff12cfc CL |
5304 | #define STAT_ATTR(si, text) \ |
5305 | static ssize_t text##_show(struct kmem_cache *s, char *buf) \ | |
5306 | { \ | |
5307 | return show_stat(s, buf, si); \ | |
5308 | } \ | |
78eb00cc DR |
5309 | static ssize_t text##_store(struct kmem_cache *s, \ |
5310 | const char *buf, size_t length) \ | |
5311 | { \ | |
5312 | if (buf[0] != '0') \ | |
5313 | return -EINVAL; \ | |
5314 | clear_stat(s, si); \ | |
5315 | return length; \ | |
5316 | } \ | |
5317 | SLAB_ATTR(text); \ | |
8ff12cfc CL |
5318 | |
5319 | STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath); | |
5320 | STAT_ATTR(ALLOC_SLOWPATH, alloc_slowpath); | |
5321 | STAT_ATTR(FREE_FASTPATH, free_fastpath); | |
5322 | STAT_ATTR(FREE_SLOWPATH, free_slowpath); | |
5323 | STAT_ATTR(FREE_FROZEN, free_frozen); | |
5324 | STAT_ATTR(FREE_ADD_PARTIAL, free_add_partial); | |
5325 | STAT_ATTR(FREE_REMOVE_PARTIAL, free_remove_partial); | |
5326 | STAT_ATTR(ALLOC_FROM_PARTIAL, alloc_from_partial); | |
5327 | STAT_ATTR(ALLOC_SLAB, alloc_slab); | |
5328 | STAT_ATTR(ALLOC_REFILL, alloc_refill); | |
e36a2652 | 5329 | STAT_ATTR(ALLOC_NODE_MISMATCH, alloc_node_mismatch); |
8ff12cfc CL |
5330 | STAT_ATTR(FREE_SLAB, free_slab); |
5331 | STAT_ATTR(CPUSLAB_FLUSH, cpuslab_flush); | |
5332 | STAT_ATTR(DEACTIVATE_FULL, deactivate_full); | |
5333 | STAT_ATTR(DEACTIVATE_EMPTY, deactivate_empty); | |
5334 | STAT_ATTR(DEACTIVATE_TO_HEAD, deactivate_to_head); | |
5335 | STAT_ATTR(DEACTIVATE_TO_TAIL, deactivate_to_tail); | |
5336 | STAT_ATTR(DEACTIVATE_REMOTE_FREES, deactivate_remote_frees); | |
03e404af | 5337 | STAT_ATTR(DEACTIVATE_BYPASS, deactivate_bypass); |
65c3376a | 5338 | STAT_ATTR(ORDER_FALLBACK, order_fallback); |
b789ef51 CL |
5339 | STAT_ATTR(CMPXCHG_DOUBLE_CPU_FAIL, cmpxchg_double_cpu_fail); |
5340 | STAT_ATTR(CMPXCHG_DOUBLE_FAIL, cmpxchg_double_fail); | |
49e22585 CL |
5341 | STAT_ATTR(CPU_PARTIAL_ALLOC, cpu_partial_alloc); |
5342 | STAT_ATTR(CPU_PARTIAL_FREE, cpu_partial_free); | |
8028dcea AS |
5343 | STAT_ATTR(CPU_PARTIAL_NODE, cpu_partial_node); |
5344 | STAT_ATTR(CPU_PARTIAL_DRAIN, cpu_partial_drain); | |
8ff12cfc CL |
5345 | #endif |
5346 | ||
06428780 | 5347 | static struct attribute *slab_attrs[] = { |
81819f0f CL |
5348 | &slab_size_attr.attr, |
5349 | &object_size_attr.attr, | |
5350 | &objs_per_slab_attr.attr, | |
5351 | &order_attr.attr, | |
73d342b1 | 5352 | &min_partial_attr.attr, |
49e22585 | 5353 | &cpu_partial_attr.attr, |
81819f0f | 5354 | &objects_attr.attr, |
205ab99d | 5355 | &objects_partial_attr.attr, |
81819f0f CL |
5356 | &partial_attr.attr, |
5357 | &cpu_slabs_attr.attr, | |
5358 | &ctor_attr.attr, | |
81819f0f CL |
5359 | &aliases_attr.attr, |
5360 | &align_attr.attr, | |
81819f0f CL |
5361 | &hwcache_align_attr.attr, |
5362 | &reclaim_account_attr.attr, | |
5363 | &destroy_by_rcu_attr.attr, | |
a5a84755 | 5364 | &shrink_attr.attr, |
ab9a0f19 | 5365 | &reserved_attr.attr, |
49e22585 | 5366 | &slabs_cpu_partial_attr.attr, |
ab4d5ed5 | 5367 | #ifdef CONFIG_SLUB_DEBUG |
a5a84755 CL |
5368 | &total_objects_attr.attr, |
5369 | &slabs_attr.attr, | |
5370 | &sanity_checks_attr.attr, | |
5371 | &trace_attr.attr, | |
81819f0f CL |
5372 | &red_zone_attr.attr, |
5373 | &poison_attr.attr, | |
5374 | &store_user_attr.attr, | |
53e15af0 | 5375 | &validate_attr.attr, |
88a420e4 CL |
5376 | &alloc_calls_attr.attr, |
5377 | &free_calls_attr.attr, | |
ab4d5ed5 | 5378 | #endif |
81819f0f CL |
5379 | #ifdef CONFIG_ZONE_DMA |
5380 | &cache_dma_attr.attr, | |
5381 | #endif | |
5382 | #ifdef CONFIG_NUMA | |
9824601e | 5383 | &remote_node_defrag_ratio_attr.attr, |
8ff12cfc CL |
5384 | #endif |
5385 | #ifdef CONFIG_SLUB_STATS | |
5386 | &alloc_fastpath_attr.attr, | |
5387 | &alloc_slowpath_attr.attr, | |
5388 | &free_fastpath_attr.attr, | |
5389 | &free_slowpath_attr.attr, | |
5390 | &free_frozen_attr.attr, | |
5391 | &free_add_partial_attr.attr, | |
5392 | &free_remove_partial_attr.attr, | |
5393 | &alloc_from_partial_attr.attr, | |
5394 | &alloc_slab_attr.attr, | |
5395 | &alloc_refill_attr.attr, | |
e36a2652 | 5396 | &alloc_node_mismatch_attr.attr, |
8ff12cfc CL |
5397 | &free_slab_attr.attr, |
5398 | &cpuslab_flush_attr.attr, | |
5399 | &deactivate_full_attr.attr, | |
5400 | &deactivate_empty_attr.attr, | |
5401 | &deactivate_to_head_attr.attr, | |
5402 | &deactivate_to_tail_attr.attr, | |
5403 | &deactivate_remote_frees_attr.attr, | |
03e404af | 5404 | &deactivate_bypass_attr.attr, |
65c3376a | 5405 | &order_fallback_attr.attr, |
b789ef51 CL |
5406 | &cmpxchg_double_fail_attr.attr, |
5407 | &cmpxchg_double_cpu_fail_attr.attr, | |
49e22585 CL |
5408 | &cpu_partial_alloc_attr.attr, |
5409 | &cpu_partial_free_attr.attr, | |
8028dcea AS |
5410 | &cpu_partial_node_attr.attr, |
5411 | &cpu_partial_drain_attr.attr, | |
81819f0f | 5412 | #endif |
4c13dd3b DM |
5413 | #ifdef CONFIG_FAILSLAB |
5414 | &failslab_attr.attr, | |
5415 | #endif | |
5416 | ||
81819f0f CL |
5417 | NULL |
5418 | }; | |
5419 | ||
5420 | static struct attribute_group slab_attr_group = { | |
5421 | .attrs = slab_attrs, | |
5422 | }; | |
5423 | ||
5424 | static ssize_t slab_attr_show(struct kobject *kobj, | |
5425 | struct attribute *attr, | |
5426 | char *buf) | |
5427 | { | |
5428 | struct slab_attribute *attribute; | |
5429 | struct kmem_cache *s; | |
5430 | int err; | |
5431 | ||
5432 | attribute = to_slab_attr(attr); | |
5433 | s = to_slab(kobj); | |
5434 | ||
5435 | if (!attribute->show) | |
5436 | return -EIO; | |
5437 | ||
5438 | err = attribute->show(s, buf); | |
5439 | ||
5440 | return err; | |
5441 | } | |
5442 | ||
5443 | static ssize_t slab_attr_store(struct kobject *kobj, | |
5444 | struct attribute *attr, | |
5445 | const char *buf, size_t len) | |
5446 | { | |
5447 | struct slab_attribute *attribute; | |
5448 | struct kmem_cache *s; | |
5449 | int err; | |
5450 | ||
5451 | attribute = to_slab_attr(attr); | |
5452 | s = to_slab(kobj); | |
5453 | ||
5454 | if (!attribute->store) | |
5455 | return -EIO; | |
5456 | ||
5457 | err = attribute->store(s, buf, len); | |
127424c8 | 5458 | #ifdef CONFIG_MEMCG |
107dab5c | 5459 | if (slab_state >= FULL && err >= 0 && is_root_cache(s)) { |
426589f5 | 5460 | struct kmem_cache *c; |
81819f0f | 5461 | |
107dab5c GC |
5462 | mutex_lock(&slab_mutex); |
5463 | if (s->max_attr_size < len) | |
5464 | s->max_attr_size = len; | |
5465 | ||
ebe945c2 GC |
5466 | /* |
5467 | * This is a best effort propagation, so this function's return | |
5468 | * value will be determined by the parent cache only. This is | |
5469 | * basically because not all attributes will have a well | |
5470 | * defined semantics for rollbacks - most of the actions will | |
5471 | * have permanent effects. | |
5472 | * | |
5473 | * Returning the error value of any of the children that fail | |
5474 | * is not 100 % defined, in the sense that users seeing the | |
5475 | * error code won't be able to know anything about the state of | |
5476 | * the cache. | |
5477 | * | |
5478 | * Only returning the error code for the parent cache at least | |
5479 | * has well defined semantics. The cache being written to | |
5480 | * directly either failed or succeeded, in which case we loop | |
5481 | * through the descendants with best-effort propagation. | |
5482 | */ | |
426589f5 VD |
5483 | for_each_memcg_cache(c, s) |
5484 | attribute->store(c, buf, len); | |
107dab5c GC |
5485 | mutex_unlock(&slab_mutex); |
5486 | } | |
5487 | #endif | |
81819f0f CL |
5488 | return err; |
5489 | } | |
5490 | ||
107dab5c GC |
5491 | static void memcg_propagate_slab_attrs(struct kmem_cache *s) |
5492 | { | |
127424c8 | 5493 | #ifdef CONFIG_MEMCG |
107dab5c GC |
5494 | int i; |
5495 | char *buffer = NULL; | |
93030d83 | 5496 | struct kmem_cache *root_cache; |
107dab5c | 5497 | |
93030d83 | 5498 | if (is_root_cache(s)) |
107dab5c GC |
5499 | return; |
5500 | ||
f7ce3190 | 5501 | root_cache = s->memcg_params.root_cache; |
93030d83 | 5502 | |
107dab5c GC |
5503 | /* |
5504 | * This mean this cache had no attribute written. Therefore, no point | |
5505 | * in copying default values around | |
5506 | */ | |
93030d83 | 5507 | if (!root_cache->max_attr_size) |
107dab5c GC |
5508 | return; |
5509 | ||
5510 | for (i = 0; i < ARRAY_SIZE(slab_attrs); i++) { | |
5511 | char mbuf[64]; | |
5512 | char *buf; | |
5513 | struct slab_attribute *attr = to_slab_attr(slab_attrs[i]); | |
478fe303 | 5514 | ssize_t len; |
107dab5c GC |
5515 | |
5516 | if (!attr || !attr->store || !attr->show) | |
5517 | continue; | |
5518 | ||
5519 | /* | |
5520 | * It is really bad that we have to allocate here, so we will | |
5521 | * do it only as a fallback. If we actually allocate, though, | |
5522 | * we can just use the allocated buffer until the end. | |
5523 | * | |
5524 | * Most of the slub attributes will tend to be very small in | |
5525 | * size, but sysfs allows buffers up to a page, so they can | |
5526 | * theoretically happen. | |
5527 | */ | |
5528 | if (buffer) | |
5529 | buf = buffer; | |
93030d83 | 5530 | else if (root_cache->max_attr_size < ARRAY_SIZE(mbuf)) |
107dab5c GC |
5531 | buf = mbuf; |
5532 | else { | |
5533 | buffer = (char *) get_zeroed_page(GFP_KERNEL); | |
5534 | if (WARN_ON(!buffer)) | |
5535 | continue; | |
5536 | buf = buffer; | |
5537 | } | |
5538 | ||
478fe303 TG |
5539 | len = attr->show(root_cache, buf); |
5540 | if (len > 0) | |
5541 | attr->store(s, buf, len); | |
107dab5c GC |
5542 | } |
5543 | ||
5544 | if (buffer) | |
5545 | free_page((unsigned long)buffer); | |
5546 | #endif | |
5547 | } | |
5548 | ||
41a21285 CL |
5549 | static void kmem_cache_release(struct kobject *k) |
5550 | { | |
5551 | slab_kmem_cache_release(to_slab(k)); | |
5552 | } | |
5553 | ||
52cf25d0 | 5554 | static const struct sysfs_ops slab_sysfs_ops = { |
81819f0f CL |
5555 | .show = slab_attr_show, |
5556 | .store = slab_attr_store, | |
5557 | }; | |
5558 | ||
5559 | static struct kobj_type slab_ktype = { | |
5560 | .sysfs_ops = &slab_sysfs_ops, | |
41a21285 | 5561 | .release = kmem_cache_release, |
81819f0f CL |
5562 | }; |
5563 | ||
5564 | static int uevent_filter(struct kset *kset, struct kobject *kobj) | |
5565 | { | |
5566 | struct kobj_type *ktype = get_ktype(kobj); | |
5567 | ||
5568 | if (ktype == &slab_ktype) | |
5569 | return 1; | |
5570 | return 0; | |
5571 | } | |
5572 | ||
9cd43611 | 5573 | static const struct kset_uevent_ops slab_uevent_ops = { |
81819f0f CL |
5574 | .filter = uevent_filter, |
5575 | }; | |
5576 | ||
27c3a314 | 5577 | static struct kset *slab_kset; |
81819f0f | 5578 | |
9a41707b VD |
5579 | static inline struct kset *cache_kset(struct kmem_cache *s) |
5580 | { | |
127424c8 | 5581 | #ifdef CONFIG_MEMCG |
9a41707b | 5582 | if (!is_root_cache(s)) |
f7ce3190 | 5583 | return s->memcg_params.root_cache->memcg_kset; |
9a41707b VD |
5584 | #endif |
5585 | return slab_kset; | |
5586 | } | |
5587 | ||
81819f0f CL |
5588 | #define ID_STR_LENGTH 64 |
5589 | ||
5590 | /* Create a unique string id for a slab cache: | |
6446faa2 CL |
5591 | * |
5592 | * Format :[flags-]size | |
81819f0f CL |
5593 | */ |
5594 | static char *create_unique_id(struct kmem_cache *s) | |
5595 | { | |
5596 | char *name = kmalloc(ID_STR_LENGTH, GFP_KERNEL); | |
5597 | char *p = name; | |
5598 | ||
5599 | BUG_ON(!name); | |
5600 | ||
5601 | *p++ = ':'; | |
5602 | /* | |
5603 | * First flags affecting slabcache operations. We will only | |
5604 | * get here for aliasable slabs so we do not need to support | |
5605 | * too many flags. The flags here must cover all flags that | |
5606 | * are matched during merging to guarantee that the id is | |
5607 | * unique. | |
5608 | */ | |
5609 | if (s->flags & SLAB_CACHE_DMA) | |
5610 | *p++ = 'd'; | |
5611 | if (s->flags & SLAB_RECLAIM_ACCOUNT) | |
5612 | *p++ = 'a'; | |
becfda68 | 5613 | if (s->flags & SLAB_CONSISTENCY_CHECKS) |
81819f0f | 5614 | *p++ = 'F'; |
5a896d9e VN |
5615 | if (!(s->flags & SLAB_NOTRACK)) |
5616 | *p++ = 't'; | |
230e9fc2 VD |
5617 | if (s->flags & SLAB_ACCOUNT) |
5618 | *p++ = 'A'; | |
81819f0f CL |
5619 | if (p != name + 1) |
5620 | *p++ = '-'; | |
5621 | p += sprintf(p, "%07d", s->size); | |
2633d7a0 | 5622 | |
81819f0f CL |
5623 | BUG_ON(p > name + ID_STR_LENGTH - 1); |
5624 | return name; | |
5625 | } | |
5626 | ||
3b7b3140 TH |
5627 | static void sysfs_slab_remove_workfn(struct work_struct *work) |
5628 | { | |
5629 | struct kmem_cache *s = | |
5630 | container_of(work, struct kmem_cache, kobj_remove_work); | |
5631 | ||
5632 | if (!s->kobj.state_in_sysfs) | |
5633 | /* | |
5634 | * For a memcg cache, this may be called during | |
5635 | * deactivation and again on shutdown. Remove only once. | |
5636 | * A cache is never shut down before deactivation is | |
5637 | * complete, so no need to worry about synchronization. | |
5638 | */ | |
5639 | return; | |
5640 | ||
5641 | #ifdef CONFIG_MEMCG | |
5642 | kset_unregister(s->memcg_kset); | |
5643 | #endif | |
5644 | kobject_uevent(&s->kobj, KOBJ_REMOVE); | |
5645 | kobject_del(&s->kobj); | |
5646 | kobject_put(&s->kobj); | |
5647 | } | |
5648 | ||
81819f0f CL |
5649 | static int sysfs_slab_add(struct kmem_cache *s) |
5650 | { | |
5651 | int err; | |
5652 | const char *name; | |
1663f26d | 5653 | struct kset *kset = cache_kset(s); |
45530c44 | 5654 | int unmergeable = slab_unmergeable(s); |
81819f0f | 5655 | |
3b7b3140 TH |
5656 | INIT_WORK(&s->kobj_remove_work, sysfs_slab_remove_workfn); |
5657 | ||
1663f26d TH |
5658 | if (!kset) { |
5659 | kobject_init(&s->kobj, &slab_ktype); | |
5660 | return 0; | |
5661 | } | |
5662 | ||
81819f0f CL |
5663 | if (unmergeable) { |
5664 | /* | |
5665 | * Slabcache can never be merged so we can use the name proper. | |
5666 | * This is typically the case for debug situations. In that | |
5667 | * case we can catch duplicate names easily. | |
5668 | */ | |
27c3a314 | 5669 | sysfs_remove_link(&slab_kset->kobj, s->name); |
81819f0f CL |
5670 | name = s->name; |
5671 | } else { | |
5672 | /* | |
5673 | * Create a unique name for the slab as a target | |
5674 | * for the symlinks. | |
5675 | */ | |
5676 | name = create_unique_id(s); | |
5677 | } | |
5678 | ||
1663f26d | 5679 | s->kobj.kset = kset; |
26e4f205 | 5680 | err = kobject_init_and_add(&s->kobj, &slab_ktype, NULL, "%s", name); |
54b6a731 | 5681 | if (err) |
80da026a | 5682 | goto out; |
81819f0f CL |
5683 | |
5684 | err = sysfs_create_group(&s->kobj, &slab_attr_group); | |
54b6a731 DJ |
5685 | if (err) |
5686 | goto out_del_kobj; | |
9a41707b | 5687 | |
127424c8 | 5688 | #ifdef CONFIG_MEMCG |
1663f26d | 5689 | if (is_root_cache(s) && memcg_sysfs_enabled) { |
9a41707b VD |
5690 | s->memcg_kset = kset_create_and_add("cgroup", NULL, &s->kobj); |
5691 | if (!s->memcg_kset) { | |
54b6a731 DJ |
5692 | err = -ENOMEM; |
5693 | goto out_del_kobj; | |
9a41707b VD |
5694 | } |
5695 | } | |
5696 | #endif | |
5697 | ||
81819f0f CL |
5698 | kobject_uevent(&s->kobj, KOBJ_ADD); |
5699 | if (!unmergeable) { | |
5700 | /* Setup first alias */ | |
5701 | sysfs_slab_alias(s, s->name); | |
81819f0f | 5702 | } |
54b6a731 DJ |
5703 | out: |
5704 | if (!unmergeable) | |
5705 | kfree(name); | |
5706 | return err; | |
5707 | out_del_kobj: | |
5708 | kobject_del(&s->kobj); | |
54b6a731 | 5709 | goto out; |
81819f0f CL |
5710 | } |
5711 | ||
bf5eb3de | 5712 | static void sysfs_slab_remove(struct kmem_cache *s) |
81819f0f | 5713 | { |
97d06609 | 5714 | if (slab_state < FULL) |
2bce6485 CL |
5715 | /* |
5716 | * Sysfs has not been setup yet so no need to remove the | |
5717 | * cache from sysfs. | |
5718 | */ | |
5719 | return; | |
5720 | ||
3b7b3140 TH |
5721 | kobject_get(&s->kobj); |
5722 | schedule_work(&s->kobj_remove_work); | |
bf5eb3de TH |
5723 | } |
5724 | ||
5725 | void sysfs_slab_release(struct kmem_cache *s) | |
5726 | { | |
5727 | if (slab_state >= FULL) | |
5728 | kobject_put(&s->kobj); | |
81819f0f CL |
5729 | } |
5730 | ||
5731 | /* | |
5732 | * Need to buffer aliases during bootup until sysfs becomes | |
9f6c708e | 5733 | * available lest we lose that information. |
81819f0f CL |
5734 | */ |
5735 | struct saved_alias { | |
5736 | struct kmem_cache *s; | |
5737 | const char *name; | |
5738 | struct saved_alias *next; | |
5739 | }; | |
5740 | ||
5af328a5 | 5741 | static struct saved_alias *alias_list; |
81819f0f CL |
5742 | |
5743 | static int sysfs_slab_alias(struct kmem_cache *s, const char *name) | |
5744 | { | |
5745 | struct saved_alias *al; | |
5746 | ||
97d06609 | 5747 | if (slab_state == FULL) { |
81819f0f CL |
5748 | /* |
5749 | * If we have a leftover link then remove it. | |
5750 | */ | |
27c3a314 GKH |
5751 | sysfs_remove_link(&slab_kset->kobj, name); |
5752 | return sysfs_create_link(&slab_kset->kobj, &s->kobj, name); | |
81819f0f CL |
5753 | } |
5754 | ||
5755 | al = kmalloc(sizeof(struct saved_alias), GFP_KERNEL); | |
5756 | if (!al) | |
5757 | return -ENOMEM; | |
5758 | ||
5759 | al->s = s; | |
5760 | al->name = name; | |
5761 | al->next = alias_list; | |
5762 | alias_list = al; | |
5763 | return 0; | |
5764 | } | |
5765 | ||
5766 | static int __init slab_sysfs_init(void) | |
5767 | { | |
5b95a4ac | 5768 | struct kmem_cache *s; |
81819f0f CL |
5769 | int err; |
5770 | ||
18004c5d | 5771 | mutex_lock(&slab_mutex); |
2bce6485 | 5772 | |
0ff21e46 | 5773 | slab_kset = kset_create_and_add("slab", &slab_uevent_ops, kernel_kobj); |
27c3a314 | 5774 | if (!slab_kset) { |
18004c5d | 5775 | mutex_unlock(&slab_mutex); |
f9f58285 | 5776 | pr_err("Cannot register slab subsystem.\n"); |
81819f0f CL |
5777 | return -ENOSYS; |
5778 | } | |
5779 | ||
97d06609 | 5780 | slab_state = FULL; |
26a7bd03 | 5781 | |
5b95a4ac | 5782 | list_for_each_entry(s, &slab_caches, list) { |
26a7bd03 | 5783 | err = sysfs_slab_add(s); |
5d540fb7 | 5784 | if (err) |
f9f58285 FF |
5785 | pr_err("SLUB: Unable to add boot slab %s to sysfs\n", |
5786 | s->name); | |
26a7bd03 | 5787 | } |
81819f0f CL |
5788 | |
5789 | while (alias_list) { | |
5790 | struct saved_alias *al = alias_list; | |
5791 | ||
5792 | alias_list = alias_list->next; | |
5793 | err = sysfs_slab_alias(al->s, al->name); | |
5d540fb7 | 5794 | if (err) |
f9f58285 FF |
5795 | pr_err("SLUB: Unable to add boot slab alias %s to sysfs\n", |
5796 | al->name); | |
81819f0f CL |
5797 | kfree(al); |
5798 | } | |
5799 | ||
18004c5d | 5800 | mutex_unlock(&slab_mutex); |
81819f0f CL |
5801 | resiliency_test(); |
5802 | return 0; | |
5803 | } | |
5804 | ||
5805 | __initcall(slab_sysfs_init); | |
ab4d5ed5 | 5806 | #endif /* CONFIG_SYSFS */ |
57ed3eda PE |
5807 | |
5808 | /* | |
5809 | * The /proc/slabinfo ABI | |
5810 | */ | |
158a9624 | 5811 | #ifdef CONFIG_SLABINFO |
0d7561c6 | 5812 | void get_slabinfo(struct kmem_cache *s, struct slabinfo *sinfo) |
57ed3eda | 5813 | { |
57ed3eda | 5814 | unsigned long nr_slabs = 0; |
205ab99d CL |
5815 | unsigned long nr_objs = 0; |
5816 | unsigned long nr_free = 0; | |
57ed3eda | 5817 | int node; |
fa45dc25 | 5818 | struct kmem_cache_node *n; |
57ed3eda | 5819 | |
fa45dc25 | 5820 | for_each_kmem_cache_node(s, node, n) { |
c17fd13e WL |
5821 | nr_slabs += node_nr_slabs(n); |
5822 | nr_objs += node_nr_objs(n); | |
205ab99d | 5823 | nr_free += count_partial(n, count_free); |
57ed3eda PE |
5824 | } |
5825 | ||
0d7561c6 GC |
5826 | sinfo->active_objs = nr_objs - nr_free; |
5827 | sinfo->num_objs = nr_objs; | |
5828 | sinfo->active_slabs = nr_slabs; | |
5829 | sinfo->num_slabs = nr_slabs; | |
5830 | sinfo->objects_per_slab = oo_objects(s->oo); | |
5831 | sinfo->cache_order = oo_order(s->oo); | |
57ed3eda PE |
5832 | } |
5833 | ||
0d7561c6 | 5834 | void slabinfo_show_stats(struct seq_file *m, struct kmem_cache *s) |
7b3c3a50 | 5835 | { |
7b3c3a50 AD |
5836 | } |
5837 | ||
b7454ad3 GC |
5838 | ssize_t slabinfo_write(struct file *file, const char __user *buffer, |
5839 | size_t count, loff_t *ppos) | |
7b3c3a50 | 5840 | { |
b7454ad3 | 5841 | return -EIO; |
7b3c3a50 | 5842 | } |
158a9624 | 5843 | #endif /* CONFIG_SLABINFO */ |