]>
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); |
81819f0f | 217 | #else |
0c710013 CL |
218 | static inline int sysfs_slab_add(struct kmem_cache *s) { return 0; } |
219 | static inline int sysfs_slab_alias(struct kmem_cache *s, const char *p) | |
220 | { return 0; } | |
107dab5c | 221 | static inline void memcg_propagate_slab_attrs(struct kmem_cache *s) { } |
81819f0f CL |
222 | #endif |
223 | ||
4fdccdfb | 224 | static inline void stat(const struct kmem_cache *s, enum stat_item si) |
8ff12cfc CL |
225 | { |
226 | #ifdef CONFIG_SLUB_STATS | |
88da03a6 CL |
227 | /* |
228 | * The rmw is racy on a preemptible kernel but this is acceptable, so | |
229 | * avoid this_cpu_add()'s irq-disable overhead. | |
230 | */ | |
231 | raw_cpu_inc(s->cpu_slab->stat[si]); | |
8ff12cfc CL |
232 | #endif |
233 | } | |
234 | ||
81819f0f CL |
235 | /******************************************************************** |
236 | * Core slab cache functions | |
237 | *******************************************************************/ | |
238 | ||
7656c72b CL |
239 | static inline void *get_freepointer(struct kmem_cache *s, void *object) |
240 | { | |
241 | return *(void **)(object + s->offset); | |
242 | } | |
243 | ||
0ad9500e ED |
244 | static void prefetch_freepointer(const struct kmem_cache *s, void *object) |
245 | { | |
246 | prefetch(object + s->offset); | |
247 | } | |
248 | ||
1393d9a1 CL |
249 | static inline void *get_freepointer_safe(struct kmem_cache *s, void *object) |
250 | { | |
251 | void *p; | |
252 | ||
922d566c JK |
253 | if (!debug_pagealloc_enabled()) |
254 | return get_freepointer(s, object); | |
255 | ||
1393d9a1 | 256 | probe_kernel_read(&p, (void **)(object + s->offset), sizeof(p)); |
1393d9a1 CL |
257 | return p; |
258 | } | |
259 | ||
7656c72b CL |
260 | static inline void set_freepointer(struct kmem_cache *s, void *object, void *fp) |
261 | { | |
262 | *(void **)(object + s->offset) = fp; | |
263 | } | |
264 | ||
265 | /* Loop over all objects in a slab */ | |
224a88be | 266 | #define for_each_object(__p, __s, __addr, __objects) \ |
d86bd1be JK |
267 | for (__p = fixup_red_left(__s, __addr); \ |
268 | __p < (__addr) + (__objects) * (__s)->size; \ | |
269 | __p += (__s)->size) | |
7656c72b | 270 | |
54266640 | 271 | #define for_each_object_idx(__p, __idx, __s, __addr, __objects) \ |
d86bd1be JK |
272 | for (__p = fixup_red_left(__s, __addr), __idx = 1; \ |
273 | __idx <= __objects; \ | |
274 | __p += (__s)->size, __idx++) | |
54266640 | 275 | |
7656c72b CL |
276 | /* Determine object index from a given position */ |
277 | static inline int slab_index(void *p, struct kmem_cache *s, void *addr) | |
278 | { | |
279 | return (p - addr) / s->size; | |
280 | } | |
281 | ||
ab9a0f19 LJ |
282 | static inline int order_objects(int order, unsigned long size, int reserved) |
283 | { | |
284 | return ((PAGE_SIZE << order) - reserved) / size; | |
285 | } | |
286 | ||
834f3d11 | 287 | static inline struct kmem_cache_order_objects oo_make(int order, |
ab9a0f19 | 288 | unsigned long size, int reserved) |
834f3d11 CL |
289 | { |
290 | struct kmem_cache_order_objects x = { | |
ab9a0f19 | 291 | (order << OO_SHIFT) + order_objects(order, size, reserved) |
834f3d11 CL |
292 | }; |
293 | ||
294 | return x; | |
295 | } | |
296 | ||
297 | static inline int oo_order(struct kmem_cache_order_objects x) | |
298 | { | |
210b5c06 | 299 | return x.x >> OO_SHIFT; |
834f3d11 CL |
300 | } |
301 | ||
302 | static inline int oo_objects(struct kmem_cache_order_objects x) | |
303 | { | |
210b5c06 | 304 | return x.x & OO_MASK; |
834f3d11 CL |
305 | } |
306 | ||
881db7fb CL |
307 | /* |
308 | * Per slab locking using the pagelock | |
309 | */ | |
310 | static __always_inline void slab_lock(struct page *page) | |
311 | { | |
48c935ad | 312 | VM_BUG_ON_PAGE(PageTail(page), page); |
881db7fb CL |
313 | bit_spin_lock(PG_locked, &page->flags); |
314 | } | |
315 | ||
316 | static __always_inline void slab_unlock(struct page *page) | |
317 | { | |
48c935ad | 318 | VM_BUG_ON_PAGE(PageTail(page), page); |
881db7fb CL |
319 | __bit_spin_unlock(PG_locked, &page->flags); |
320 | } | |
321 | ||
a0320865 DH |
322 | static inline void set_page_slub_counters(struct page *page, unsigned long counters_new) |
323 | { | |
324 | struct page tmp; | |
325 | tmp.counters = counters_new; | |
326 | /* | |
327 | * page->counters can cover frozen/inuse/objects as well | |
0139aa7b JK |
328 | * as page->_refcount. If we assign to ->counters directly |
329 | * we run the risk of losing updates to page->_refcount, so | |
a0320865 DH |
330 | * be careful and only assign to the fields we need. |
331 | */ | |
332 | page->frozen = tmp.frozen; | |
333 | page->inuse = tmp.inuse; | |
334 | page->objects = tmp.objects; | |
335 | } | |
336 | ||
1d07171c CL |
337 | /* Interrupts must be disabled (for the fallback code to work right) */ |
338 | static inline bool __cmpxchg_double_slab(struct kmem_cache *s, struct page *page, | |
339 | void *freelist_old, unsigned long counters_old, | |
340 | void *freelist_new, unsigned long counters_new, | |
341 | const char *n) | |
342 | { | |
343 | VM_BUG_ON(!irqs_disabled()); | |
2565409f HC |
344 | #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \ |
345 | defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE) | |
1d07171c | 346 | if (s->flags & __CMPXCHG_DOUBLE) { |
cdcd6298 | 347 | if (cmpxchg_double(&page->freelist, &page->counters, |
0aa9a13d DC |
348 | freelist_old, counters_old, |
349 | freelist_new, counters_new)) | |
6f6528a1 | 350 | return true; |
1d07171c CL |
351 | } else |
352 | #endif | |
353 | { | |
354 | slab_lock(page); | |
d0e0ac97 CG |
355 | if (page->freelist == freelist_old && |
356 | page->counters == counters_old) { | |
1d07171c | 357 | page->freelist = freelist_new; |
a0320865 | 358 | set_page_slub_counters(page, counters_new); |
1d07171c | 359 | slab_unlock(page); |
6f6528a1 | 360 | return true; |
1d07171c CL |
361 | } |
362 | slab_unlock(page); | |
363 | } | |
364 | ||
365 | cpu_relax(); | |
366 | stat(s, CMPXCHG_DOUBLE_FAIL); | |
367 | ||
368 | #ifdef SLUB_DEBUG_CMPXCHG | |
f9f58285 | 369 | pr_info("%s %s: cmpxchg double redo ", n, s->name); |
1d07171c CL |
370 | #endif |
371 | ||
6f6528a1 | 372 | return false; |
1d07171c CL |
373 | } |
374 | ||
b789ef51 CL |
375 | static inline bool cmpxchg_double_slab(struct kmem_cache *s, struct page *page, |
376 | void *freelist_old, unsigned long counters_old, | |
377 | void *freelist_new, unsigned long counters_new, | |
378 | const char *n) | |
379 | { | |
2565409f HC |
380 | #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \ |
381 | defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE) | |
b789ef51 | 382 | if (s->flags & __CMPXCHG_DOUBLE) { |
cdcd6298 | 383 | if (cmpxchg_double(&page->freelist, &page->counters, |
0aa9a13d DC |
384 | freelist_old, counters_old, |
385 | freelist_new, counters_new)) | |
6f6528a1 | 386 | return true; |
b789ef51 CL |
387 | } else |
388 | #endif | |
389 | { | |
1d07171c CL |
390 | unsigned long flags; |
391 | ||
392 | local_irq_save(flags); | |
881db7fb | 393 | slab_lock(page); |
d0e0ac97 CG |
394 | if (page->freelist == freelist_old && |
395 | page->counters == counters_old) { | |
b789ef51 | 396 | page->freelist = freelist_new; |
a0320865 | 397 | set_page_slub_counters(page, counters_new); |
881db7fb | 398 | slab_unlock(page); |
1d07171c | 399 | local_irq_restore(flags); |
6f6528a1 | 400 | return true; |
b789ef51 | 401 | } |
881db7fb | 402 | slab_unlock(page); |
1d07171c | 403 | local_irq_restore(flags); |
b789ef51 CL |
404 | } |
405 | ||
406 | cpu_relax(); | |
407 | stat(s, CMPXCHG_DOUBLE_FAIL); | |
408 | ||
409 | #ifdef SLUB_DEBUG_CMPXCHG | |
f9f58285 | 410 | pr_info("%s %s: cmpxchg double redo ", n, s->name); |
b789ef51 CL |
411 | #endif |
412 | ||
6f6528a1 | 413 | return false; |
b789ef51 CL |
414 | } |
415 | ||
41ecc55b | 416 | #ifdef CONFIG_SLUB_DEBUG |
5f80b13a CL |
417 | /* |
418 | * Determine a map of object in use on a page. | |
419 | * | |
881db7fb | 420 | * Node listlock must be held to guarantee that the page does |
5f80b13a CL |
421 | * not vanish from under us. |
422 | */ | |
423 | static void get_map(struct kmem_cache *s, struct page *page, unsigned long *map) | |
424 | { | |
425 | void *p; | |
426 | void *addr = page_address(page); | |
427 | ||
428 | for (p = page->freelist; p; p = get_freepointer(s, p)) | |
429 | set_bit(slab_index(p, s, addr), map); | |
430 | } | |
431 | ||
d86bd1be JK |
432 | static inline int size_from_object(struct kmem_cache *s) |
433 | { | |
434 | if (s->flags & SLAB_RED_ZONE) | |
435 | return s->size - s->red_left_pad; | |
436 | ||
437 | return s->size; | |
438 | } | |
439 | ||
440 | static inline void *restore_red_left(struct kmem_cache *s, void *p) | |
441 | { | |
442 | if (s->flags & SLAB_RED_ZONE) | |
443 | p -= s->red_left_pad; | |
444 | ||
445 | return p; | |
446 | } | |
447 | ||
41ecc55b CL |
448 | /* |
449 | * Debug settings: | |
450 | */ | |
89d3c87e | 451 | #if defined(CONFIG_SLUB_DEBUG_ON) |
f0630fff CL |
452 | static int slub_debug = DEBUG_DEFAULT_FLAGS; |
453 | #else | |
41ecc55b | 454 | static int slub_debug; |
f0630fff | 455 | #endif |
41ecc55b CL |
456 | |
457 | static char *slub_debug_slabs; | |
fa5ec8a1 | 458 | static int disable_higher_order_debug; |
41ecc55b | 459 | |
a79316c6 AR |
460 | /* |
461 | * slub is about to manipulate internal object metadata. This memory lies | |
462 | * outside the range of the allocated object, so accessing it would normally | |
463 | * be reported by kasan as a bounds error. metadata_access_enable() is used | |
464 | * to tell kasan that these accesses are OK. | |
465 | */ | |
466 | static inline void metadata_access_enable(void) | |
467 | { | |
468 | kasan_disable_current(); | |
469 | } | |
470 | ||
471 | static inline void metadata_access_disable(void) | |
472 | { | |
473 | kasan_enable_current(); | |
474 | } | |
475 | ||
81819f0f CL |
476 | /* |
477 | * Object debugging | |
478 | */ | |
d86bd1be JK |
479 | |
480 | /* Verify that a pointer has an address that is valid within a slab page */ | |
481 | static inline int check_valid_pointer(struct kmem_cache *s, | |
482 | struct page *page, void *object) | |
483 | { | |
484 | void *base; | |
485 | ||
486 | if (!object) | |
487 | return 1; | |
488 | ||
489 | base = page_address(page); | |
490 | object = restore_red_left(s, object); | |
491 | if (object < base || object >= base + page->objects * s->size || | |
492 | (object - base) % s->size) { | |
493 | return 0; | |
494 | } | |
495 | ||
496 | return 1; | |
497 | } | |
498 | ||
aa2efd5e DT |
499 | static void print_section(char *level, char *text, u8 *addr, |
500 | unsigned int length) | |
81819f0f | 501 | { |
a79316c6 | 502 | metadata_access_enable(); |
aa2efd5e | 503 | print_hex_dump(level, text, DUMP_PREFIX_ADDRESS, 16, 1, addr, |
ffc79d28 | 504 | length, 1); |
a79316c6 | 505 | metadata_access_disable(); |
81819f0f CL |
506 | } |
507 | ||
81819f0f CL |
508 | static struct track *get_track(struct kmem_cache *s, void *object, |
509 | enum track_item alloc) | |
510 | { | |
511 | struct track *p; | |
512 | ||
513 | if (s->offset) | |
514 | p = object + s->offset + sizeof(void *); | |
515 | else | |
516 | p = object + s->inuse; | |
517 | ||
518 | return p + alloc; | |
519 | } | |
520 | ||
521 | static void set_track(struct kmem_cache *s, void *object, | |
ce71e27c | 522 | enum track_item alloc, unsigned long addr) |
81819f0f | 523 | { |
1a00df4a | 524 | struct track *p = get_track(s, object, alloc); |
81819f0f | 525 | |
81819f0f | 526 | if (addr) { |
d6543e39 BG |
527 | #ifdef CONFIG_STACKTRACE |
528 | struct stack_trace trace; | |
529 | int i; | |
530 | ||
531 | trace.nr_entries = 0; | |
532 | trace.max_entries = TRACK_ADDRS_COUNT; | |
533 | trace.entries = p->addrs; | |
534 | trace.skip = 3; | |
a79316c6 | 535 | metadata_access_enable(); |
d6543e39 | 536 | save_stack_trace(&trace); |
a79316c6 | 537 | metadata_access_disable(); |
d6543e39 BG |
538 | |
539 | /* See rant in lockdep.c */ | |
540 | if (trace.nr_entries != 0 && | |
541 | trace.entries[trace.nr_entries - 1] == ULONG_MAX) | |
542 | trace.nr_entries--; | |
543 | ||
544 | for (i = trace.nr_entries; i < TRACK_ADDRS_COUNT; i++) | |
545 | p->addrs[i] = 0; | |
546 | #endif | |
81819f0f CL |
547 | p->addr = addr; |
548 | p->cpu = smp_processor_id(); | |
88e4ccf2 | 549 | p->pid = current->pid; |
81819f0f CL |
550 | p->when = jiffies; |
551 | } else | |
552 | memset(p, 0, sizeof(struct track)); | |
553 | } | |
554 | ||
81819f0f CL |
555 | static void init_tracking(struct kmem_cache *s, void *object) |
556 | { | |
24922684 CL |
557 | if (!(s->flags & SLAB_STORE_USER)) |
558 | return; | |
559 | ||
ce71e27c EGM |
560 | set_track(s, object, TRACK_FREE, 0UL); |
561 | set_track(s, object, TRACK_ALLOC, 0UL); | |
81819f0f CL |
562 | } |
563 | ||
564 | static void print_track(const char *s, struct track *t) | |
565 | { | |
566 | if (!t->addr) | |
567 | return; | |
568 | ||
f9f58285 FF |
569 | pr_err("INFO: %s in %pS age=%lu cpu=%u pid=%d\n", |
570 | s, (void *)t->addr, jiffies - t->when, t->cpu, t->pid); | |
d6543e39 BG |
571 | #ifdef CONFIG_STACKTRACE |
572 | { | |
573 | int i; | |
574 | for (i = 0; i < TRACK_ADDRS_COUNT; i++) | |
575 | if (t->addrs[i]) | |
f9f58285 | 576 | pr_err("\t%pS\n", (void *)t->addrs[i]); |
d6543e39 BG |
577 | else |
578 | break; | |
579 | } | |
580 | #endif | |
24922684 CL |
581 | } |
582 | ||
583 | static void print_tracking(struct kmem_cache *s, void *object) | |
584 | { | |
585 | if (!(s->flags & SLAB_STORE_USER)) | |
586 | return; | |
587 | ||
588 | print_track("Allocated", get_track(s, object, TRACK_ALLOC)); | |
589 | print_track("Freed", get_track(s, object, TRACK_FREE)); | |
590 | } | |
591 | ||
592 | static void print_page_info(struct page *page) | |
593 | { | |
f9f58285 | 594 | pr_err("INFO: Slab 0x%p objects=%u used=%u fp=0x%p flags=0x%04lx\n", |
d0e0ac97 | 595 | page, page->objects, page->inuse, page->freelist, page->flags); |
24922684 CL |
596 | |
597 | } | |
598 | ||
599 | static void slab_bug(struct kmem_cache *s, char *fmt, ...) | |
600 | { | |
ecc42fbe | 601 | struct va_format vaf; |
24922684 | 602 | va_list args; |
24922684 CL |
603 | |
604 | va_start(args, fmt); | |
ecc42fbe FF |
605 | vaf.fmt = fmt; |
606 | vaf.va = &args; | |
f9f58285 | 607 | pr_err("=============================================================================\n"); |
ecc42fbe | 608 | pr_err("BUG %s (%s): %pV\n", s->name, print_tainted(), &vaf); |
f9f58285 | 609 | pr_err("-----------------------------------------------------------------------------\n\n"); |
645df230 | 610 | |
373d4d09 | 611 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); |
ecc42fbe | 612 | va_end(args); |
81819f0f CL |
613 | } |
614 | ||
24922684 CL |
615 | static void slab_fix(struct kmem_cache *s, char *fmt, ...) |
616 | { | |
ecc42fbe | 617 | struct va_format vaf; |
24922684 | 618 | va_list args; |
24922684 CL |
619 | |
620 | va_start(args, fmt); | |
ecc42fbe FF |
621 | vaf.fmt = fmt; |
622 | vaf.va = &args; | |
623 | pr_err("FIX %s: %pV\n", s->name, &vaf); | |
24922684 | 624 | va_end(args); |
24922684 CL |
625 | } |
626 | ||
627 | static void print_trailer(struct kmem_cache *s, struct page *page, u8 *p) | |
81819f0f CL |
628 | { |
629 | unsigned int off; /* Offset of last byte */ | |
a973e9dd | 630 | u8 *addr = page_address(page); |
24922684 CL |
631 | |
632 | print_tracking(s, p); | |
633 | ||
634 | print_page_info(page); | |
635 | ||
f9f58285 FF |
636 | pr_err("INFO: Object 0x%p @offset=%tu fp=0x%p\n\n", |
637 | p, p - addr, get_freepointer(s, p)); | |
24922684 | 638 | |
d86bd1be | 639 | if (s->flags & SLAB_RED_ZONE) |
aa2efd5e DT |
640 | print_section(KERN_ERR, "Redzone ", p - s->red_left_pad, |
641 | s->red_left_pad); | |
d86bd1be | 642 | else if (p > addr + 16) |
aa2efd5e | 643 | print_section(KERN_ERR, "Bytes b4 ", p - 16, 16); |
81819f0f | 644 | |
aa2efd5e DT |
645 | print_section(KERN_ERR, "Object ", p, |
646 | min_t(unsigned long, s->object_size, PAGE_SIZE)); | |
81819f0f | 647 | if (s->flags & SLAB_RED_ZONE) |
aa2efd5e | 648 | print_section(KERN_ERR, "Redzone ", p + s->object_size, |
3b0efdfa | 649 | s->inuse - s->object_size); |
81819f0f | 650 | |
81819f0f CL |
651 | if (s->offset) |
652 | off = s->offset + sizeof(void *); | |
653 | else | |
654 | off = s->inuse; | |
655 | ||
24922684 | 656 | if (s->flags & SLAB_STORE_USER) |
81819f0f | 657 | off += 2 * sizeof(struct track); |
81819f0f | 658 | |
80a9201a AP |
659 | off += kasan_metadata_size(s); |
660 | ||
d86bd1be | 661 | if (off != size_from_object(s)) |
81819f0f | 662 | /* Beginning of the filler is the free pointer */ |
aa2efd5e DT |
663 | print_section(KERN_ERR, "Padding ", p + off, |
664 | size_from_object(s) - off); | |
24922684 CL |
665 | |
666 | dump_stack(); | |
81819f0f CL |
667 | } |
668 | ||
75c66def | 669 | void object_err(struct kmem_cache *s, struct page *page, |
81819f0f CL |
670 | u8 *object, char *reason) |
671 | { | |
3dc50637 | 672 | slab_bug(s, "%s", reason); |
24922684 | 673 | print_trailer(s, page, object); |
81819f0f CL |
674 | } |
675 | ||
d0e0ac97 CG |
676 | static void slab_err(struct kmem_cache *s, struct page *page, |
677 | const char *fmt, ...) | |
81819f0f CL |
678 | { |
679 | va_list args; | |
680 | char buf[100]; | |
681 | ||
24922684 CL |
682 | va_start(args, fmt); |
683 | vsnprintf(buf, sizeof(buf), fmt, args); | |
81819f0f | 684 | va_end(args); |
3dc50637 | 685 | slab_bug(s, "%s", buf); |
24922684 | 686 | print_page_info(page); |
81819f0f CL |
687 | dump_stack(); |
688 | } | |
689 | ||
f7cb1933 | 690 | static void init_object(struct kmem_cache *s, void *object, u8 val) |
81819f0f CL |
691 | { |
692 | u8 *p = object; | |
693 | ||
d86bd1be JK |
694 | if (s->flags & SLAB_RED_ZONE) |
695 | memset(p - s->red_left_pad, val, s->red_left_pad); | |
696 | ||
81819f0f | 697 | if (s->flags & __OBJECT_POISON) { |
3b0efdfa CL |
698 | memset(p, POISON_FREE, s->object_size - 1); |
699 | p[s->object_size - 1] = POISON_END; | |
81819f0f CL |
700 | } |
701 | ||
702 | if (s->flags & SLAB_RED_ZONE) | |
3b0efdfa | 703 | memset(p + s->object_size, val, s->inuse - s->object_size); |
81819f0f CL |
704 | } |
705 | ||
24922684 CL |
706 | static void restore_bytes(struct kmem_cache *s, char *message, u8 data, |
707 | void *from, void *to) | |
708 | { | |
709 | slab_fix(s, "Restoring 0x%p-0x%p=0x%x\n", from, to - 1, data); | |
710 | memset(from, data, to - from); | |
711 | } | |
712 | ||
713 | static int check_bytes_and_report(struct kmem_cache *s, struct page *page, | |
714 | u8 *object, char *what, | |
06428780 | 715 | u8 *start, unsigned int value, unsigned int bytes) |
24922684 CL |
716 | { |
717 | u8 *fault; | |
718 | u8 *end; | |
719 | ||
a79316c6 | 720 | metadata_access_enable(); |
79824820 | 721 | fault = memchr_inv(start, value, bytes); |
a79316c6 | 722 | metadata_access_disable(); |
24922684 CL |
723 | if (!fault) |
724 | return 1; | |
725 | ||
726 | end = start + bytes; | |
727 | while (end > fault && end[-1] == value) | |
728 | end--; | |
729 | ||
730 | slab_bug(s, "%s overwritten", what); | |
f9f58285 | 731 | pr_err("INFO: 0x%p-0x%p. First byte 0x%x instead of 0x%x\n", |
24922684 CL |
732 | fault, end - 1, fault[0], value); |
733 | print_trailer(s, page, object); | |
734 | ||
735 | restore_bytes(s, what, value, fault, end); | |
736 | return 0; | |
81819f0f CL |
737 | } |
738 | ||
81819f0f CL |
739 | /* |
740 | * Object layout: | |
741 | * | |
742 | * object address | |
743 | * Bytes of the object to be managed. | |
744 | * If the freepointer may overlay the object then the free | |
745 | * pointer is the first word of the object. | |
672bba3a | 746 | * |
81819f0f CL |
747 | * Poisoning uses 0x6b (POISON_FREE) and the last byte is |
748 | * 0xa5 (POISON_END) | |
749 | * | |
3b0efdfa | 750 | * object + s->object_size |
81819f0f | 751 | * Padding to reach word boundary. This is also used for Redzoning. |
672bba3a | 752 | * Padding is extended by another word if Redzoning is enabled and |
3b0efdfa | 753 | * object_size == inuse. |
672bba3a | 754 | * |
81819f0f CL |
755 | * We fill with 0xbb (RED_INACTIVE) for inactive objects and with |
756 | * 0xcc (RED_ACTIVE) for objects in use. | |
757 | * | |
758 | * object + s->inuse | |
672bba3a CL |
759 | * Meta data starts here. |
760 | * | |
81819f0f CL |
761 | * A. Free pointer (if we cannot overwrite object on free) |
762 | * B. Tracking data for SLAB_STORE_USER | |
672bba3a | 763 | * C. Padding to reach required alignment boundary or at mininum |
6446faa2 | 764 | * one word if debugging is on to be able to detect writes |
672bba3a CL |
765 | * before the word boundary. |
766 | * | |
767 | * Padding is done using 0x5a (POISON_INUSE) | |
81819f0f CL |
768 | * |
769 | * object + s->size | |
672bba3a | 770 | * Nothing is used beyond s->size. |
81819f0f | 771 | * |
3b0efdfa | 772 | * If slabcaches are merged then the object_size and inuse boundaries are mostly |
672bba3a | 773 | * ignored. And therefore no slab options that rely on these boundaries |
81819f0f CL |
774 | * may be used with merged slabcaches. |
775 | */ | |
776 | ||
81819f0f CL |
777 | static int check_pad_bytes(struct kmem_cache *s, struct page *page, u8 *p) |
778 | { | |
779 | unsigned long off = s->inuse; /* The end of info */ | |
780 | ||
781 | if (s->offset) | |
782 | /* Freepointer is placed after the object. */ | |
783 | off += sizeof(void *); | |
784 | ||
785 | if (s->flags & SLAB_STORE_USER) | |
786 | /* We also have user information there */ | |
787 | off += 2 * sizeof(struct track); | |
788 | ||
80a9201a AP |
789 | off += kasan_metadata_size(s); |
790 | ||
d86bd1be | 791 | if (size_from_object(s) == off) |
81819f0f CL |
792 | return 1; |
793 | ||
24922684 | 794 | return check_bytes_and_report(s, page, p, "Object padding", |
d86bd1be | 795 | p + off, POISON_INUSE, size_from_object(s) - off); |
81819f0f CL |
796 | } |
797 | ||
39b26464 | 798 | /* Check the pad bytes at the end of a slab page */ |
81819f0f CL |
799 | static int slab_pad_check(struct kmem_cache *s, struct page *page) |
800 | { | |
24922684 CL |
801 | u8 *start; |
802 | u8 *fault; | |
803 | u8 *end; | |
804 | int length; | |
805 | int remainder; | |
81819f0f CL |
806 | |
807 | if (!(s->flags & SLAB_POISON)) | |
808 | return 1; | |
809 | ||
a973e9dd | 810 | start = page_address(page); |
ab9a0f19 | 811 | length = (PAGE_SIZE << compound_order(page)) - s->reserved; |
39b26464 CL |
812 | end = start + length; |
813 | remainder = length % s->size; | |
81819f0f CL |
814 | if (!remainder) |
815 | return 1; | |
816 | ||
a79316c6 | 817 | metadata_access_enable(); |
79824820 | 818 | fault = memchr_inv(end - remainder, POISON_INUSE, remainder); |
a79316c6 | 819 | metadata_access_disable(); |
24922684 CL |
820 | if (!fault) |
821 | return 1; | |
822 | while (end > fault && end[-1] == POISON_INUSE) | |
823 | end--; | |
824 | ||
825 | slab_err(s, page, "Padding overwritten. 0x%p-0x%p", fault, end - 1); | |
aa2efd5e | 826 | print_section(KERN_ERR, "Padding ", end - remainder, remainder); |
24922684 | 827 | |
8a3d271d | 828 | restore_bytes(s, "slab padding", POISON_INUSE, end - remainder, end); |
24922684 | 829 | return 0; |
81819f0f CL |
830 | } |
831 | ||
832 | static int check_object(struct kmem_cache *s, struct page *page, | |
f7cb1933 | 833 | void *object, u8 val) |
81819f0f CL |
834 | { |
835 | u8 *p = object; | |
3b0efdfa | 836 | u8 *endobject = object + s->object_size; |
81819f0f CL |
837 | |
838 | if (s->flags & SLAB_RED_ZONE) { | |
d86bd1be JK |
839 | if (!check_bytes_and_report(s, page, object, "Redzone", |
840 | object - s->red_left_pad, val, s->red_left_pad)) | |
841 | return 0; | |
842 | ||
24922684 | 843 | if (!check_bytes_and_report(s, page, object, "Redzone", |
3b0efdfa | 844 | endobject, val, s->inuse - s->object_size)) |
81819f0f | 845 | return 0; |
81819f0f | 846 | } else { |
3b0efdfa | 847 | if ((s->flags & SLAB_POISON) && s->object_size < s->inuse) { |
3adbefee | 848 | check_bytes_and_report(s, page, p, "Alignment padding", |
d0e0ac97 CG |
849 | endobject, POISON_INUSE, |
850 | s->inuse - s->object_size); | |
3adbefee | 851 | } |
81819f0f CL |
852 | } |
853 | ||
854 | if (s->flags & SLAB_POISON) { | |
f7cb1933 | 855 | if (val != SLUB_RED_ACTIVE && (s->flags & __OBJECT_POISON) && |
24922684 | 856 | (!check_bytes_and_report(s, page, p, "Poison", p, |
3b0efdfa | 857 | POISON_FREE, s->object_size - 1) || |
24922684 | 858 | !check_bytes_and_report(s, page, p, "Poison", |
3b0efdfa | 859 | p + s->object_size - 1, POISON_END, 1))) |
81819f0f | 860 | return 0; |
81819f0f CL |
861 | /* |
862 | * check_pad_bytes cleans up on its own. | |
863 | */ | |
864 | check_pad_bytes(s, page, p); | |
865 | } | |
866 | ||
f7cb1933 | 867 | if (!s->offset && val == SLUB_RED_ACTIVE) |
81819f0f CL |
868 | /* |
869 | * Object and freepointer overlap. Cannot check | |
870 | * freepointer while object is allocated. | |
871 | */ | |
872 | return 1; | |
873 | ||
874 | /* Check free pointer validity */ | |
875 | if (!check_valid_pointer(s, page, get_freepointer(s, p))) { | |
876 | object_err(s, page, p, "Freepointer corrupt"); | |
877 | /* | |
9f6c708e | 878 | * No choice but to zap it and thus lose the remainder |
81819f0f | 879 | * of the free objects in this slab. May cause |
672bba3a | 880 | * another error because the object count is now wrong. |
81819f0f | 881 | */ |
a973e9dd | 882 | set_freepointer(s, p, NULL); |
81819f0f CL |
883 | return 0; |
884 | } | |
885 | return 1; | |
886 | } | |
887 | ||
888 | static int check_slab(struct kmem_cache *s, struct page *page) | |
889 | { | |
39b26464 CL |
890 | int maxobj; |
891 | ||
81819f0f CL |
892 | VM_BUG_ON(!irqs_disabled()); |
893 | ||
894 | if (!PageSlab(page)) { | |
24922684 | 895 | slab_err(s, page, "Not a valid slab page"); |
81819f0f CL |
896 | return 0; |
897 | } | |
39b26464 | 898 | |
ab9a0f19 | 899 | maxobj = order_objects(compound_order(page), s->size, s->reserved); |
39b26464 CL |
900 | if (page->objects > maxobj) { |
901 | slab_err(s, page, "objects %u > max %u", | |
f6edde9c | 902 | page->objects, maxobj); |
39b26464 CL |
903 | return 0; |
904 | } | |
905 | if (page->inuse > page->objects) { | |
24922684 | 906 | slab_err(s, page, "inuse %u > max %u", |
f6edde9c | 907 | page->inuse, page->objects); |
81819f0f CL |
908 | return 0; |
909 | } | |
910 | /* Slab_pad_check fixes things up after itself */ | |
911 | slab_pad_check(s, page); | |
912 | return 1; | |
913 | } | |
914 | ||
915 | /* | |
672bba3a CL |
916 | * Determine if a certain object on a page is on the freelist. Must hold the |
917 | * slab lock to guarantee that the chains are in a consistent state. | |
81819f0f CL |
918 | */ |
919 | static int on_freelist(struct kmem_cache *s, struct page *page, void *search) | |
920 | { | |
921 | int nr = 0; | |
881db7fb | 922 | void *fp; |
81819f0f | 923 | void *object = NULL; |
f6edde9c | 924 | int max_objects; |
81819f0f | 925 | |
881db7fb | 926 | fp = page->freelist; |
39b26464 | 927 | while (fp && nr <= page->objects) { |
81819f0f CL |
928 | if (fp == search) |
929 | return 1; | |
930 | if (!check_valid_pointer(s, page, fp)) { | |
931 | if (object) { | |
932 | object_err(s, page, object, | |
933 | "Freechain corrupt"); | |
a973e9dd | 934 | set_freepointer(s, object, NULL); |
81819f0f | 935 | } else { |
24922684 | 936 | slab_err(s, page, "Freepointer corrupt"); |
a973e9dd | 937 | page->freelist = NULL; |
39b26464 | 938 | page->inuse = page->objects; |
24922684 | 939 | slab_fix(s, "Freelist cleared"); |
81819f0f CL |
940 | return 0; |
941 | } | |
942 | break; | |
943 | } | |
944 | object = fp; | |
945 | fp = get_freepointer(s, object); | |
946 | nr++; | |
947 | } | |
948 | ||
ab9a0f19 | 949 | max_objects = order_objects(compound_order(page), s->size, s->reserved); |
210b5c06 CG |
950 | if (max_objects > MAX_OBJS_PER_PAGE) |
951 | max_objects = MAX_OBJS_PER_PAGE; | |
224a88be CL |
952 | |
953 | if (page->objects != max_objects) { | |
756a025f JP |
954 | slab_err(s, page, "Wrong number of objects. Found %d but should be %d", |
955 | page->objects, max_objects); | |
224a88be CL |
956 | page->objects = max_objects; |
957 | slab_fix(s, "Number of objects adjusted."); | |
958 | } | |
39b26464 | 959 | if (page->inuse != page->objects - nr) { |
756a025f JP |
960 | slab_err(s, page, "Wrong object count. Counter is %d but counted were %d", |
961 | page->inuse, page->objects - nr); | |
39b26464 | 962 | page->inuse = page->objects - nr; |
24922684 | 963 | slab_fix(s, "Object count adjusted."); |
81819f0f CL |
964 | } |
965 | return search == NULL; | |
966 | } | |
967 | ||
0121c619 CL |
968 | static void trace(struct kmem_cache *s, struct page *page, void *object, |
969 | int alloc) | |
3ec09742 CL |
970 | { |
971 | if (s->flags & SLAB_TRACE) { | |
f9f58285 | 972 | pr_info("TRACE %s %s 0x%p inuse=%d fp=0x%p\n", |
3ec09742 CL |
973 | s->name, |
974 | alloc ? "alloc" : "free", | |
975 | object, page->inuse, | |
976 | page->freelist); | |
977 | ||
978 | if (!alloc) | |
aa2efd5e | 979 | print_section(KERN_INFO, "Object ", (void *)object, |
d0e0ac97 | 980 | s->object_size); |
3ec09742 CL |
981 | |
982 | dump_stack(); | |
983 | } | |
984 | } | |
985 | ||
643b1138 | 986 | /* |
672bba3a | 987 | * Tracking of fully allocated slabs for debugging purposes. |
643b1138 | 988 | */ |
5cc6eee8 CL |
989 | static void add_full(struct kmem_cache *s, |
990 | struct kmem_cache_node *n, struct page *page) | |
643b1138 | 991 | { |
5cc6eee8 CL |
992 | if (!(s->flags & SLAB_STORE_USER)) |
993 | return; | |
994 | ||
255d0884 | 995 | lockdep_assert_held(&n->list_lock); |
643b1138 | 996 | list_add(&page->lru, &n->full); |
643b1138 CL |
997 | } |
998 | ||
c65c1877 | 999 | static void remove_full(struct kmem_cache *s, struct kmem_cache_node *n, struct page *page) |
643b1138 | 1000 | { |
643b1138 CL |
1001 | if (!(s->flags & SLAB_STORE_USER)) |
1002 | return; | |
1003 | ||
255d0884 | 1004 | lockdep_assert_held(&n->list_lock); |
643b1138 | 1005 | list_del(&page->lru); |
643b1138 CL |
1006 | } |
1007 | ||
0f389ec6 CL |
1008 | /* Tracking of the number of slabs for debugging purposes */ |
1009 | static inline unsigned long slabs_node(struct kmem_cache *s, int node) | |
1010 | { | |
1011 | struct kmem_cache_node *n = get_node(s, node); | |
1012 | ||
1013 | return atomic_long_read(&n->nr_slabs); | |
1014 | } | |
1015 | ||
26c02cf0 AB |
1016 | static inline unsigned long node_nr_slabs(struct kmem_cache_node *n) |
1017 | { | |
1018 | return atomic_long_read(&n->nr_slabs); | |
1019 | } | |
1020 | ||
205ab99d | 1021 | static inline void inc_slabs_node(struct kmem_cache *s, int node, int objects) |
0f389ec6 CL |
1022 | { |
1023 | struct kmem_cache_node *n = get_node(s, node); | |
1024 | ||
1025 | /* | |
1026 | * May be called early in order to allocate a slab for the | |
1027 | * kmem_cache_node structure. Solve the chicken-egg | |
1028 | * dilemma by deferring the increment of the count during | |
1029 | * bootstrap (see early_kmem_cache_node_alloc). | |
1030 | */ | |
338b2642 | 1031 | if (likely(n)) { |
0f389ec6 | 1032 | atomic_long_inc(&n->nr_slabs); |
205ab99d CL |
1033 | atomic_long_add(objects, &n->total_objects); |
1034 | } | |
0f389ec6 | 1035 | } |
205ab99d | 1036 | static inline void dec_slabs_node(struct kmem_cache *s, int node, int objects) |
0f389ec6 CL |
1037 | { |
1038 | struct kmem_cache_node *n = get_node(s, node); | |
1039 | ||
1040 | atomic_long_dec(&n->nr_slabs); | |
205ab99d | 1041 | atomic_long_sub(objects, &n->total_objects); |
0f389ec6 CL |
1042 | } |
1043 | ||
1044 | /* Object debug checks for alloc/free paths */ | |
3ec09742 CL |
1045 | static void setup_object_debug(struct kmem_cache *s, struct page *page, |
1046 | void *object) | |
1047 | { | |
1048 | if (!(s->flags & (SLAB_STORE_USER|SLAB_RED_ZONE|__OBJECT_POISON))) | |
1049 | return; | |
1050 | ||
f7cb1933 | 1051 | init_object(s, object, SLUB_RED_INACTIVE); |
3ec09742 CL |
1052 | init_tracking(s, object); |
1053 | } | |
1054 | ||
becfda68 | 1055 | static inline int alloc_consistency_checks(struct kmem_cache *s, |
d0e0ac97 | 1056 | struct page *page, |
ce71e27c | 1057 | void *object, unsigned long addr) |
81819f0f CL |
1058 | { |
1059 | if (!check_slab(s, page)) | |
becfda68 | 1060 | return 0; |
81819f0f | 1061 | |
81819f0f CL |
1062 | if (!check_valid_pointer(s, page, object)) { |
1063 | object_err(s, page, object, "Freelist Pointer check fails"); | |
becfda68 | 1064 | return 0; |
81819f0f CL |
1065 | } |
1066 | ||
f7cb1933 | 1067 | if (!check_object(s, page, object, SLUB_RED_INACTIVE)) |
becfda68 LA |
1068 | return 0; |
1069 | ||
1070 | return 1; | |
1071 | } | |
1072 | ||
1073 | static noinline int alloc_debug_processing(struct kmem_cache *s, | |
1074 | struct page *page, | |
1075 | void *object, unsigned long addr) | |
1076 | { | |
1077 | if (s->flags & SLAB_CONSISTENCY_CHECKS) { | |
1078 | if (!alloc_consistency_checks(s, page, object, addr)) | |
1079 | goto bad; | |
1080 | } | |
81819f0f | 1081 | |
3ec09742 CL |
1082 | /* Success perform special debug activities for allocs */ |
1083 | if (s->flags & SLAB_STORE_USER) | |
1084 | set_track(s, object, TRACK_ALLOC, addr); | |
1085 | trace(s, page, object, 1); | |
f7cb1933 | 1086 | init_object(s, object, SLUB_RED_ACTIVE); |
81819f0f | 1087 | return 1; |
3ec09742 | 1088 | |
81819f0f CL |
1089 | bad: |
1090 | if (PageSlab(page)) { | |
1091 | /* | |
1092 | * If this is a slab page then lets do the best we can | |
1093 | * to avoid issues in the future. Marking all objects | |
672bba3a | 1094 | * as used avoids touching the remaining objects. |
81819f0f | 1095 | */ |
24922684 | 1096 | slab_fix(s, "Marking all objects used"); |
39b26464 | 1097 | page->inuse = page->objects; |
a973e9dd | 1098 | page->freelist = NULL; |
81819f0f CL |
1099 | } |
1100 | return 0; | |
1101 | } | |
1102 | ||
becfda68 LA |
1103 | static inline int free_consistency_checks(struct kmem_cache *s, |
1104 | struct page *page, void *object, unsigned long addr) | |
81819f0f | 1105 | { |
81819f0f | 1106 | if (!check_valid_pointer(s, page, object)) { |
70d71228 | 1107 | slab_err(s, page, "Invalid object pointer 0x%p", object); |
becfda68 | 1108 | return 0; |
81819f0f CL |
1109 | } |
1110 | ||
1111 | if (on_freelist(s, page, object)) { | |
24922684 | 1112 | object_err(s, page, object, "Object already free"); |
becfda68 | 1113 | return 0; |
81819f0f CL |
1114 | } |
1115 | ||
f7cb1933 | 1116 | if (!check_object(s, page, object, SLUB_RED_ACTIVE)) |
becfda68 | 1117 | return 0; |
81819f0f | 1118 | |
1b4f59e3 | 1119 | if (unlikely(s != page->slab_cache)) { |
3adbefee | 1120 | if (!PageSlab(page)) { |
756a025f JP |
1121 | slab_err(s, page, "Attempt to free object(0x%p) outside of slab", |
1122 | object); | |
1b4f59e3 | 1123 | } else if (!page->slab_cache) { |
f9f58285 FF |
1124 | pr_err("SLUB <none>: no slab for object 0x%p.\n", |
1125 | object); | |
70d71228 | 1126 | dump_stack(); |
06428780 | 1127 | } else |
24922684 CL |
1128 | object_err(s, page, object, |
1129 | "page slab pointer corrupt."); | |
becfda68 LA |
1130 | return 0; |
1131 | } | |
1132 | return 1; | |
1133 | } | |
1134 | ||
1135 | /* Supports checking bulk free of a constructed freelist */ | |
1136 | static noinline int free_debug_processing( | |
1137 | struct kmem_cache *s, struct page *page, | |
1138 | void *head, void *tail, int bulk_cnt, | |
1139 | unsigned long addr) | |
1140 | { | |
1141 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); | |
1142 | void *object = head; | |
1143 | int cnt = 0; | |
1144 | unsigned long uninitialized_var(flags); | |
1145 | int ret = 0; | |
1146 | ||
1147 | spin_lock_irqsave(&n->list_lock, flags); | |
1148 | slab_lock(page); | |
1149 | ||
1150 | if (s->flags & SLAB_CONSISTENCY_CHECKS) { | |
1151 | if (!check_slab(s, page)) | |
1152 | goto out; | |
1153 | } | |
1154 | ||
1155 | next_object: | |
1156 | cnt++; | |
1157 | ||
1158 | if (s->flags & SLAB_CONSISTENCY_CHECKS) { | |
1159 | if (!free_consistency_checks(s, page, object, addr)) | |
1160 | goto out; | |
81819f0f | 1161 | } |
3ec09742 | 1162 | |
3ec09742 CL |
1163 | if (s->flags & SLAB_STORE_USER) |
1164 | set_track(s, object, TRACK_FREE, addr); | |
1165 | trace(s, page, object, 0); | |
81084651 | 1166 | /* Freepointer not overwritten by init_object(), SLAB_POISON moved it */ |
f7cb1933 | 1167 | init_object(s, object, SLUB_RED_INACTIVE); |
81084651 JDB |
1168 | |
1169 | /* Reached end of constructed freelist yet? */ | |
1170 | if (object != tail) { | |
1171 | object = get_freepointer(s, object); | |
1172 | goto next_object; | |
1173 | } | |
804aa132 LA |
1174 | ret = 1; |
1175 | ||
5c2e4bbb | 1176 | out: |
81084651 JDB |
1177 | if (cnt != bulk_cnt) |
1178 | slab_err(s, page, "Bulk freelist count(%d) invalid(%d)\n", | |
1179 | bulk_cnt, cnt); | |
1180 | ||
881db7fb | 1181 | slab_unlock(page); |
282acb43 | 1182 | spin_unlock_irqrestore(&n->list_lock, flags); |
804aa132 LA |
1183 | if (!ret) |
1184 | slab_fix(s, "Object at 0x%p not freed", object); | |
1185 | return ret; | |
81819f0f CL |
1186 | } |
1187 | ||
41ecc55b CL |
1188 | static int __init setup_slub_debug(char *str) |
1189 | { | |
f0630fff CL |
1190 | slub_debug = DEBUG_DEFAULT_FLAGS; |
1191 | if (*str++ != '=' || !*str) | |
1192 | /* | |
1193 | * No options specified. Switch on full debugging. | |
1194 | */ | |
1195 | goto out; | |
1196 | ||
1197 | if (*str == ',') | |
1198 | /* | |
1199 | * No options but restriction on slabs. This means full | |
1200 | * debugging for slabs matching a pattern. | |
1201 | */ | |
1202 | goto check_slabs; | |
1203 | ||
1204 | slub_debug = 0; | |
1205 | if (*str == '-') | |
1206 | /* | |
1207 | * Switch off all debugging measures. | |
1208 | */ | |
1209 | goto out; | |
1210 | ||
1211 | /* | |
1212 | * Determine which debug features should be switched on | |
1213 | */ | |
06428780 | 1214 | for (; *str && *str != ','; str++) { |
f0630fff CL |
1215 | switch (tolower(*str)) { |
1216 | case 'f': | |
becfda68 | 1217 | slub_debug |= SLAB_CONSISTENCY_CHECKS; |
f0630fff CL |
1218 | break; |
1219 | case 'z': | |
1220 | slub_debug |= SLAB_RED_ZONE; | |
1221 | break; | |
1222 | case 'p': | |
1223 | slub_debug |= SLAB_POISON; | |
1224 | break; | |
1225 | case 'u': | |
1226 | slub_debug |= SLAB_STORE_USER; | |
1227 | break; | |
1228 | case 't': | |
1229 | slub_debug |= SLAB_TRACE; | |
1230 | break; | |
4c13dd3b DM |
1231 | case 'a': |
1232 | slub_debug |= SLAB_FAILSLAB; | |
1233 | break; | |
08303a73 CA |
1234 | case 'o': |
1235 | /* | |
1236 | * Avoid enabling debugging on caches if its minimum | |
1237 | * order would increase as a result. | |
1238 | */ | |
1239 | disable_higher_order_debug = 1; | |
1240 | break; | |
f0630fff | 1241 | default: |
f9f58285 FF |
1242 | pr_err("slub_debug option '%c' unknown. skipped\n", |
1243 | *str); | |
f0630fff | 1244 | } |
41ecc55b CL |
1245 | } |
1246 | ||
f0630fff | 1247 | check_slabs: |
41ecc55b CL |
1248 | if (*str == ',') |
1249 | slub_debug_slabs = str + 1; | |
f0630fff | 1250 | out: |
41ecc55b CL |
1251 | return 1; |
1252 | } | |
1253 | ||
1254 | __setup("slub_debug", setup_slub_debug); | |
1255 | ||
423c929c | 1256 | unsigned long kmem_cache_flags(unsigned long object_size, |
ba0268a8 | 1257 | unsigned long flags, const char *name, |
51cc5068 | 1258 | void (*ctor)(void *)) |
41ecc55b CL |
1259 | { |
1260 | /* | |
e153362a | 1261 | * Enable debugging if selected on the kernel commandline. |
41ecc55b | 1262 | */ |
c6f58d9b CL |
1263 | if (slub_debug && (!slub_debug_slabs || (name && |
1264 | !strncmp(slub_debug_slabs, name, strlen(slub_debug_slabs))))) | |
3de47213 | 1265 | flags |= slub_debug; |
ba0268a8 CL |
1266 | |
1267 | return flags; | |
41ecc55b | 1268 | } |
b4a64718 | 1269 | #else /* !CONFIG_SLUB_DEBUG */ |
3ec09742 CL |
1270 | static inline void setup_object_debug(struct kmem_cache *s, |
1271 | struct page *page, void *object) {} | |
41ecc55b | 1272 | |
3ec09742 | 1273 | static inline int alloc_debug_processing(struct kmem_cache *s, |
ce71e27c | 1274 | struct page *page, void *object, unsigned long addr) { return 0; } |
41ecc55b | 1275 | |
282acb43 | 1276 | static inline int free_debug_processing( |
81084651 JDB |
1277 | struct kmem_cache *s, struct page *page, |
1278 | void *head, void *tail, int bulk_cnt, | |
282acb43 | 1279 | unsigned long addr) { return 0; } |
41ecc55b | 1280 | |
41ecc55b CL |
1281 | static inline int slab_pad_check(struct kmem_cache *s, struct page *page) |
1282 | { return 1; } | |
1283 | static inline int check_object(struct kmem_cache *s, struct page *page, | |
f7cb1933 | 1284 | void *object, u8 val) { return 1; } |
5cc6eee8 CL |
1285 | static inline void add_full(struct kmem_cache *s, struct kmem_cache_node *n, |
1286 | struct page *page) {} | |
c65c1877 PZ |
1287 | static inline void remove_full(struct kmem_cache *s, struct kmem_cache_node *n, |
1288 | struct page *page) {} | |
423c929c | 1289 | unsigned long kmem_cache_flags(unsigned long object_size, |
ba0268a8 | 1290 | unsigned long flags, const char *name, |
51cc5068 | 1291 | void (*ctor)(void *)) |
ba0268a8 CL |
1292 | { |
1293 | return flags; | |
1294 | } | |
41ecc55b | 1295 | #define slub_debug 0 |
0f389ec6 | 1296 | |
fdaa45e9 IM |
1297 | #define disable_higher_order_debug 0 |
1298 | ||
0f389ec6 CL |
1299 | static inline unsigned long slabs_node(struct kmem_cache *s, int node) |
1300 | { return 0; } | |
26c02cf0 AB |
1301 | static inline unsigned long node_nr_slabs(struct kmem_cache_node *n) |
1302 | { return 0; } | |
205ab99d CL |
1303 | static inline void inc_slabs_node(struct kmem_cache *s, int node, |
1304 | int objects) {} | |
1305 | static inline void dec_slabs_node(struct kmem_cache *s, int node, | |
1306 | int objects) {} | |
7d550c56 | 1307 | |
02e72cc6 AR |
1308 | #endif /* CONFIG_SLUB_DEBUG */ |
1309 | ||
1310 | /* | |
1311 | * Hooks for other subsystems that check memory allocations. In a typical | |
1312 | * production configuration these hooks all should produce no code at all. | |
1313 | */ | |
d56791b3 RB |
1314 | static inline void kmalloc_large_node_hook(void *ptr, size_t size, gfp_t flags) |
1315 | { | |
1316 | kmemleak_alloc(ptr, size, 1, flags); | |
505f5dcb | 1317 | kasan_kmalloc_large(ptr, size, flags); |
d56791b3 RB |
1318 | } |
1319 | ||
1320 | static inline void kfree_hook(const void *x) | |
1321 | { | |
1322 | kmemleak_free(x); | |
0316bec2 | 1323 | kasan_kfree_large(x); |
d56791b3 RB |
1324 | } |
1325 | ||
80a9201a | 1326 | static inline void *slab_free_hook(struct kmem_cache *s, void *x) |
d56791b3 | 1327 | { |
80a9201a AP |
1328 | void *freeptr; |
1329 | ||
d56791b3 | 1330 | kmemleak_free_recursive(x, s->flags); |
7d550c56 | 1331 | |
02e72cc6 AR |
1332 | /* |
1333 | * Trouble is that we may no longer disable interrupts in the fast path | |
1334 | * So in order to make the debug calls that expect irqs to be | |
1335 | * disabled we need to disable interrupts temporarily. | |
1336 | */ | |
1337 | #if defined(CONFIG_KMEMCHECK) || defined(CONFIG_LOCKDEP) | |
1338 | { | |
1339 | unsigned long flags; | |
1340 | ||
1341 | local_irq_save(flags); | |
1342 | kmemcheck_slab_free(s, x, s->object_size); | |
1343 | debug_check_no_locks_freed(x, s->object_size); | |
1344 | local_irq_restore(flags); | |
1345 | } | |
1346 | #endif | |
1347 | if (!(s->flags & SLAB_DEBUG_OBJECTS)) | |
1348 | debug_check_no_obj_freed(x, s->object_size); | |
0316bec2 | 1349 | |
80a9201a AP |
1350 | freeptr = get_freepointer(s, x); |
1351 | /* | |
1352 | * kasan_slab_free() may put x into memory quarantine, delaying its | |
1353 | * reuse. In this case the object's freelist pointer is changed. | |
1354 | */ | |
0316bec2 | 1355 | kasan_slab_free(s, x); |
80a9201a | 1356 | return freeptr; |
02e72cc6 | 1357 | } |
205ab99d | 1358 | |
81084651 JDB |
1359 | static inline void slab_free_freelist_hook(struct kmem_cache *s, |
1360 | void *head, void *tail) | |
1361 | { | |
1362 | /* | |
1363 | * Compiler cannot detect this function can be removed if slab_free_hook() | |
1364 | * evaluates to nothing. Thus, catch all relevant config debug options here. | |
1365 | */ | |
1366 | #if defined(CONFIG_KMEMCHECK) || \ | |
1367 | defined(CONFIG_LOCKDEP) || \ | |
1368 | defined(CONFIG_DEBUG_KMEMLEAK) || \ | |
1369 | defined(CONFIG_DEBUG_OBJECTS_FREE) || \ | |
1370 | defined(CONFIG_KASAN) | |
1371 | ||
1372 | void *object = head; | |
1373 | void *tail_obj = tail ? : head; | |
80a9201a | 1374 | void *freeptr; |
81084651 JDB |
1375 | |
1376 | do { | |
80a9201a AP |
1377 | freeptr = slab_free_hook(s, object); |
1378 | } while ((object != tail_obj) && (object = freeptr)); | |
81084651 JDB |
1379 | #endif |
1380 | } | |
1381 | ||
588f8ba9 TG |
1382 | static void setup_object(struct kmem_cache *s, struct page *page, |
1383 | void *object) | |
1384 | { | |
1385 | setup_object_debug(s, page, object); | |
b3cbd9bf | 1386 | kasan_init_slab_obj(s, object); |
588f8ba9 TG |
1387 | if (unlikely(s->ctor)) { |
1388 | kasan_unpoison_object_data(s, object); | |
1389 | s->ctor(object); | |
1390 | kasan_poison_object_data(s, object); | |
1391 | } | |
1392 | } | |
1393 | ||
81819f0f CL |
1394 | /* |
1395 | * Slab allocation and freeing | |
1396 | */ | |
5dfb4175 VD |
1397 | static inline struct page *alloc_slab_page(struct kmem_cache *s, |
1398 | gfp_t flags, int node, struct kmem_cache_order_objects oo) | |
65c3376a | 1399 | { |
5dfb4175 | 1400 | struct page *page; |
65c3376a CL |
1401 | int order = oo_order(oo); |
1402 | ||
b1eeab67 VN |
1403 | flags |= __GFP_NOTRACK; |
1404 | ||
2154a336 | 1405 | if (node == NUMA_NO_NODE) |
5dfb4175 | 1406 | page = alloc_pages(flags, order); |
65c3376a | 1407 | else |
96db800f | 1408 | page = __alloc_pages_node(node, flags, order); |
5dfb4175 | 1409 | |
f3ccb2c4 VD |
1410 | if (page && memcg_charge_slab(page, flags, order, s)) { |
1411 | __free_pages(page, order); | |
1412 | page = NULL; | |
1413 | } | |
5dfb4175 VD |
1414 | |
1415 | return page; | |
65c3376a CL |
1416 | } |
1417 | ||
210e7a43 TG |
1418 | #ifdef CONFIG_SLAB_FREELIST_RANDOM |
1419 | /* Pre-initialize the random sequence cache */ | |
1420 | static int init_cache_random_seq(struct kmem_cache *s) | |
1421 | { | |
1422 | int err; | |
1423 | unsigned long i, count = oo_objects(s->oo); | |
1424 | ||
a810007a SR |
1425 | /* Bailout if already initialised */ |
1426 | if (s->random_seq) | |
1427 | return 0; | |
1428 | ||
210e7a43 TG |
1429 | err = cache_random_seq_create(s, count, GFP_KERNEL); |
1430 | if (err) { | |
1431 | pr_err("SLUB: Unable to initialize free list for %s\n", | |
1432 | s->name); | |
1433 | return err; | |
1434 | } | |
1435 | ||
1436 | /* Transform to an offset on the set of pages */ | |
1437 | if (s->random_seq) { | |
1438 | for (i = 0; i < count; i++) | |
1439 | s->random_seq[i] *= s->size; | |
1440 | } | |
1441 | return 0; | |
1442 | } | |
1443 | ||
1444 | /* Initialize each random sequence freelist per cache */ | |
1445 | static void __init init_freelist_randomization(void) | |
1446 | { | |
1447 | struct kmem_cache *s; | |
1448 | ||
1449 | mutex_lock(&slab_mutex); | |
1450 | ||
1451 | list_for_each_entry(s, &slab_caches, list) | |
1452 | init_cache_random_seq(s); | |
1453 | ||
1454 | mutex_unlock(&slab_mutex); | |
1455 | } | |
1456 | ||
1457 | /* Get the next entry on the pre-computed freelist randomized */ | |
1458 | static void *next_freelist_entry(struct kmem_cache *s, struct page *page, | |
1459 | unsigned long *pos, void *start, | |
1460 | unsigned long page_limit, | |
1461 | unsigned long freelist_count) | |
1462 | { | |
1463 | unsigned int idx; | |
1464 | ||
1465 | /* | |
1466 | * If the target page allocation failed, the number of objects on the | |
1467 | * page might be smaller than the usual size defined by the cache. | |
1468 | */ | |
1469 | do { | |
1470 | idx = s->random_seq[*pos]; | |
1471 | *pos += 1; | |
1472 | if (*pos >= freelist_count) | |
1473 | *pos = 0; | |
1474 | } while (unlikely(idx >= page_limit)); | |
1475 | ||
1476 | return (char *)start + idx; | |
1477 | } | |
1478 | ||
1479 | /* Shuffle the single linked freelist based on a random pre-computed sequence */ | |
1480 | static bool shuffle_freelist(struct kmem_cache *s, struct page *page) | |
1481 | { | |
1482 | void *start; | |
1483 | void *cur; | |
1484 | void *next; | |
1485 | unsigned long idx, pos, page_limit, freelist_count; | |
1486 | ||
1487 | if (page->objects < 2 || !s->random_seq) | |
1488 | return false; | |
1489 | ||
1490 | freelist_count = oo_objects(s->oo); | |
1491 | pos = get_random_int() % freelist_count; | |
1492 | ||
1493 | page_limit = page->objects * s->size; | |
1494 | start = fixup_red_left(s, page_address(page)); | |
1495 | ||
1496 | /* First entry is used as the base of the freelist */ | |
1497 | cur = next_freelist_entry(s, page, &pos, start, page_limit, | |
1498 | freelist_count); | |
1499 | page->freelist = cur; | |
1500 | ||
1501 | for (idx = 1; idx < page->objects; idx++) { | |
1502 | setup_object(s, page, cur); | |
1503 | next = next_freelist_entry(s, page, &pos, start, page_limit, | |
1504 | freelist_count); | |
1505 | set_freepointer(s, cur, next); | |
1506 | cur = next; | |
1507 | } | |
1508 | setup_object(s, page, cur); | |
1509 | set_freepointer(s, cur, NULL); | |
1510 | ||
1511 | return true; | |
1512 | } | |
1513 | #else | |
1514 | static inline int init_cache_random_seq(struct kmem_cache *s) | |
1515 | { | |
1516 | return 0; | |
1517 | } | |
1518 | static inline void init_freelist_randomization(void) { } | |
1519 | static inline bool shuffle_freelist(struct kmem_cache *s, struct page *page) | |
1520 | { | |
1521 | return false; | |
1522 | } | |
1523 | #endif /* CONFIG_SLAB_FREELIST_RANDOM */ | |
1524 | ||
81819f0f CL |
1525 | static struct page *allocate_slab(struct kmem_cache *s, gfp_t flags, int node) |
1526 | { | |
06428780 | 1527 | struct page *page; |
834f3d11 | 1528 | struct kmem_cache_order_objects oo = s->oo; |
ba52270d | 1529 | gfp_t alloc_gfp; |
588f8ba9 TG |
1530 | void *start, *p; |
1531 | int idx, order; | |
210e7a43 | 1532 | bool shuffle; |
81819f0f | 1533 | |
7e0528da CL |
1534 | flags &= gfp_allowed_mask; |
1535 | ||
d0164adc | 1536 | if (gfpflags_allow_blocking(flags)) |
7e0528da CL |
1537 | local_irq_enable(); |
1538 | ||
b7a49f0d | 1539 | flags |= s->allocflags; |
e12ba74d | 1540 | |
ba52270d PE |
1541 | /* |
1542 | * Let the initial higher-order allocation fail under memory pressure | |
1543 | * so we fall-back to the minimum order allocation. | |
1544 | */ | |
1545 | alloc_gfp = (flags | __GFP_NOWARN | __GFP_NORETRY) & ~__GFP_NOFAIL; | |
d0164adc | 1546 | if ((alloc_gfp & __GFP_DIRECT_RECLAIM) && oo_order(oo) > oo_order(s->min)) |
444eb2a4 | 1547 | alloc_gfp = (alloc_gfp | __GFP_NOMEMALLOC) & ~(__GFP_RECLAIM|__GFP_NOFAIL); |
ba52270d | 1548 | |
5dfb4175 | 1549 | page = alloc_slab_page(s, alloc_gfp, node, oo); |
65c3376a CL |
1550 | if (unlikely(!page)) { |
1551 | oo = s->min; | |
80c3a998 | 1552 | alloc_gfp = flags; |
65c3376a CL |
1553 | /* |
1554 | * Allocation may have failed due to fragmentation. | |
1555 | * Try a lower order alloc if possible | |
1556 | */ | |
5dfb4175 | 1557 | page = alloc_slab_page(s, alloc_gfp, node, oo); |
588f8ba9 TG |
1558 | if (unlikely(!page)) |
1559 | goto out; | |
1560 | stat(s, ORDER_FALLBACK); | |
65c3376a | 1561 | } |
5a896d9e | 1562 | |
588f8ba9 TG |
1563 | if (kmemcheck_enabled && |
1564 | !(s->flags & (SLAB_NOTRACK | DEBUG_DEFAULT_FLAGS))) { | |
b1eeab67 VN |
1565 | int pages = 1 << oo_order(oo); |
1566 | ||
80c3a998 | 1567 | kmemcheck_alloc_shadow(page, oo_order(oo), alloc_gfp, node); |
b1eeab67 VN |
1568 | |
1569 | /* | |
1570 | * Objects from caches that have a constructor don't get | |
1571 | * cleared when they're allocated, so we need to do it here. | |
1572 | */ | |
1573 | if (s->ctor) | |
1574 | kmemcheck_mark_uninitialized_pages(page, pages); | |
1575 | else | |
1576 | kmemcheck_mark_unallocated_pages(page, pages); | |
5a896d9e VN |
1577 | } |
1578 | ||
834f3d11 | 1579 | page->objects = oo_objects(oo); |
81819f0f | 1580 | |
1f458cbf | 1581 | order = compound_order(page); |
1b4f59e3 | 1582 | page->slab_cache = s; |
c03f94cc | 1583 | __SetPageSlab(page); |
2f064f34 | 1584 | if (page_is_pfmemalloc(page)) |
072bb0aa | 1585 | SetPageSlabPfmemalloc(page); |
81819f0f CL |
1586 | |
1587 | start = page_address(page); | |
81819f0f CL |
1588 | |
1589 | if (unlikely(s->flags & SLAB_POISON)) | |
1f458cbf | 1590 | memset(start, POISON_INUSE, PAGE_SIZE << order); |
81819f0f | 1591 | |
0316bec2 AR |
1592 | kasan_poison_slab(page); |
1593 | ||
210e7a43 TG |
1594 | shuffle = shuffle_freelist(s, page); |
1595 | ||
1596 | if (!shuffle) { | |
1597 | for_each_object_idx(p, idx, s, start, page->objects) { | |
1598 | setup_object(s, page, p); | |
1599 | if (likely(idx < page->objects)) | |
1600 | set_freepointer(s, p, p + s->size); | |
1601 | else | |
1602 | set_freepointer(s, p, NULL); | |
1603 | } | |
1604 | page->freelist = fixup_red_left(s, start); | |
81819f0f | 1605 | } |
81819f0f | 1606 | |
e6e82ea1 | 1607 | page->inuse = page->objects; |
8cb0a506 | 1608 | page->frozen = 1; |
588f8ba9 | 1609 | |
81819f0f | 1610 | out: |
d0164adc | 1611 | if (gfpflags_allow_blocking(flags)) |
588f8ba9 TG |
1612 | local_irq_disable(); |
1613 | if (!page) | |
1614 | return NULL; | |
1615 | ||
1616 | mod_zone_page_state(page_zone(page), | |
1617 | (s->flags & SLAB_RECLAIM_ACCOUNT) ? | |
1618 | NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE, | |
1619 | 1 << oo_order(oo)); | |
1620 | ||
1621 | inc_slabs_node(s, page_to_nid(page), page->objects); | |
1622 | ||
81819f0f CL |
1623 | return page; |
1624 | } | |
1625 | ||
588f8ba9 TG |
1626 | static struct page *new_slab(struct kmem_cache *s, gfp_t flags, int node) |
1627 | { | |
1628 | if (unlikely(flags & GFP_SLAB_BUG_MASK)) { | |
bacdcb34 | 1629 | gfp_t invalid_mask = flags & GFP_SLAB_BUG_MASK; |
72baeef0 MH |
1630 | flags &= ~GFP_SLAB_BUG_MASK; |
1631 | pr_warn("Unexpected gfp: %#x (%pGg). Fixing up to gfp: %#x (%pGg). Fix your code!\n", | |
1632 | invalid_mask, &invalid_mask, flags, &flags); | |
65b9de75 | 1633 | dump_stack(); |
588f8ba9 TG |
1634 | } |
1635 | ||
1636 | return allocate_slab(s, | |
1637 | flags & (GFP_RECLAIM_MASK | GFP_CONSTRAINT_MASK), node); | |
1638 | } | |
1639 | ||
81819f0f CL |
1640 | static void __free_slab(struct kmem_cache *s, struct page *page) |
1641 | { | |
834f3d11 CL |
1642 | int order = compound_order(page); |
1643 | int pages = 1 << order; | |
81819f0f | 1644 | |
becfda68 | 1645 | if (s->flags & SLAB_CONSISTENCY_CHECKS) { |
81819f0f CL |
1646 | void *p; |
1647 | ||
1648 | slab_pad_check(s, page); | |
224a88be CL |
1649 | for_each_object(p, s, page_address(page), |
1650 | page->objects) | |
f7cb1933 | 1651 | check_object(s, page, p, SLUB_RED_INACTIVE); |
81819f0f CL |
1652 | } |
1653 | ||
b1eeab67 | 1654 | kmemcheck_free_shadow(page, compound_order(page)); |
5a896d9e | 1655 | |
81819f0f CL |
1656 | mod_zone_page_state(page_zone(page), |
1657 | (s->flags & SLAB_RECLAIM_ACCOUNT) ? | |
1658 | NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE, | |
06428780 | 1659 | -pages); |
81819f0f | 1660 | |
072bb0aa | 1661 | __ClearPageSlabPfmemalloc(page); |
49bd5221 | 1662 | __ClearPageSlab(page); |
1f458cbf | 1663 | |
22b751c3 | 1664 | page_mapcount_reset(page); |
1eb5ac64 NP |
1665 | if (current->reclaim_state) |
1666 | current->reclaim_state->reclaimed_slab += pages; | |
27ee57c9 VD |
1667 | memcg_uncharge_slab(page, order, s); |
1668 | __free_pages(page, order); | |
81819f0f CL |
1669 | } |
1670 | ||
da9a638c LJ |
1671 | #define need_reserve_slab_rcu \ |
1672 | (sizeof(((struct page *)NULL)->lru) < sizeof(struct rcu_head)) | |
1673 | ||
81819f0f CL |
1674 | static void rcu_free_slab(struct rcu_head *h) |
1675 | { | |
1676 | struct page *page; | |
1677 | ||
da9a638c LJ |
1678 | if (need_reserve_slab_rcu) |
1679 | page = virt_to_head_page(h); | |
1680 | else | |
1681 | page = container_of((struct list_head *)h, struct page, lru); | |
1682 | ||
1b4f59e3 | 1683 | __free_slab(page->slab_cache, page); |
81819f0f CL |
1684 | } |
1685 | ||
1686 | static void free_slab(struct kmem_cache *s, struct page *page) | |
1687 | { | |
1688 | if (unlikely(s->flags & SLAB_DESTROY_BY_RCU)) { | |
da9a638c LJ |
1689 | struct rcu_head *head; |
1690 | ||
1691 | if (need_reserve_slab_rcu) { | |
1692 | int order = compound_order(page); | |
1693 | int offset = (PAGE_SIZE << order) - s->reserved; | |
1694 | ||
1695 | VM_BUG_ON(s->reserved != sizeof(*head)); | |
1696 | head = page_address(page) + offset; | |
1697 | } else { | |
bc4f610d | 1698 | head = &page->rcu_head; |
da9a638c | 1699 | } |
81819f0f CL |
1700 | |
1701 | call_rcu(head, rcu_free_slab); | |
1702 | } else | |
1703 | __free_slab(s, page); | |
1704 | } | |
1705 | ||
1706 | static void discard_slab(struct kmem_cache *s, struct page *page) | |
1707 | { | |
205ab99d | 1708 | dec_slabs_node(s, page_to_nid(page), page->objects); |
81819f0f CL |
1709 | free_slab(s, page); |
1710 | } | |
1711 | ||
1712 | /* | |
5cc6eee8 | 1713 | * Management of partially allocated slabs. |
81819f0f | 1714 | */ |
1e4dd946 SR |
1715 | static inline void |
1716 | __add_partial(struct kmem_cache_node *n, struct page *page, int tail) | |
81819f0f | 1717 | { |
e95eed57 | 1718 | n->nr_partial++; |
136333d1 | 1719 | if (tail == DEACTIVATE_TO_TAIL) |
7c2e132c CL |
1720 | list_add_tail(&page->lru, &n->partial); |
1721 | else | |
1722 | list_add(&page->lru, &n->partial); | |
81819f0f CL |
1723 | } |
1724 | ||
1e4dd946 SR |
1725 | static inline void add_partial(struct kmem_cache_node *n, |
1726 | struct page *page, int tail) | |
62e346a8 | 1727 | { |
c65c1877 | 1728 | lockdep_assert_held(&n->list_lock); |
1e4dd946 SR |
1729 | __add_partial(n, page, tail); |
1730 | } | |
c65c1877 | 1731 | |
1e4dd946 SR |
1732 | static inline void remove_partial(struct kmem_cache_node *n, |
1733 | struct page *page) | |
1734 | { | |
1735 | lockdep_assert_held(&n->list_lock); | |
52b4b950 DS |
1736 | list_del(&page->lru); |
1737 | n->nr_partial--; | |
1e4dd946 SR |
1738 | } |
1739 | ||
81819f0f | 1740 | /* |
7ced3719 CL |
1741 | * Remove slab from the partial list, freeze it and |
1742 | * return the pointer to the freelist. | |
81819f0f | 1743 | * |
497b66f2 | 1744 | * Returns a list of objects or NULL if it fails. |
81819f0f | 1745 | */ |
497b66f2 | 1746 | static inline void *acquire_slab(struct kmem_cache *s, |
acd19fd1 | 1747 | struct kmem_cache_node *n, struct page *page, |
633b0764 | 1748 | int mode, int *objects) |
81819f0f | 1749 | { |
2cfb7455 CL |
1750 | void *freelist; |
1751 | unsigned long counters; | |
1752 | struct page new; | |
1753 | ||
c65c1877 PZ |
1754 | lockdep_assert_held(&n->list_lock); |
1755 | ||
2cfb7455 CL |
1756 | /* |
1757 | * Zap the freelist and set the frozen bit. | |
1758 | * The old freelist is the list of objects for the | |
1759 | * per cpu allocation list. | |
1760 | */ | |
7ced3719 CL |
1761 | freelist = page->freelist; |
1762 | counters = page->counters; | |
1763 | new.counters = counters; | |
633b0764 | 1764 | *objects = new.objects - new.inuse; |
23910c50 | 1765 | if (mode) { |
7ced3719 | 1766 | new.inuse = page->objects; |
23910c50 PE |
1767 | new.freelist = NULL; |
1768 | } else { | |
1769 | new.freelist = freelist; | |
1770 | } | |
2cfb7455 | 1771 | |
a0132ac0 | 1772 | VM_BUG_ON(new.frozen); |
7ced3719 | 1773 | new.frozen = 1; |
2cfb7455 | 1774 | |
7ced3719 | 1775 | if (!__cmpxchg_double_slab(s, page, |
2cfb7455 | 1776 | freelist, counters, |
02d7633f | 1777 | new.freelist, new.counters, |
7ced3719 | 1778 | "acquire_slab")) |
7ced3719 | 1779 | return NULL; |
2cfb7455 CL |
1780 | |
1781 | remove_partial(n, page); | |
7ced3719 | 1782 | WARN_ON(!freelist); |
49e22585 | 1783 | return freelist; |
81819f0f CL |
1784 | } |
1785 | ||
633b0764 | 1786 | static void put_cpu_partial(struct kmem_cache *s, struct page *page, int drain); |
8ba00bb6 | 1787 | static inline bool pfmemalloc_match(struct page *page, gfp_t gfpflags); |
49e22585 | 1788 | |
81819f0f | 1789 | /* |
672bba3a | 1790 | * Try to allocate a partial slab from a specific node. |
81819f0f | 1791 | */ |
8ba00bb6 JK |
1792 | static void *get_partial_node(struct kmem_cache *s, struct kmem_cache_node *n, |
1793 | struct kmem_cache_cpu *c, gfp_t flags) | |
81819f0f | 1794 | { |
49e22585 CL |
1795 | struct page *page, *page2; |
1796 | void *object = NULL; | |
633b0764 JK |
1797 | int available = 0; |
1798 | int objects; | |
81819f0f CL |
1799 | |
1800 | /* | |
1801 | * Racy check. If we mistakenly see no partial slabs then we | |
1802 | * just allocate an empty slab. If we mistakenly try to get a | |
672bba3a CL |
1803 | * partial slab and there is none available then get_partials() |
1804 | * will return NULL. | |
81819f0f CL |
1805 | */ |
1806 | if (!n || !n->nr_partial) | |
1807 | return NULL; | |
1808 | ||
1809 | spin_lock(&n->list_lock); | |
49e22585 | 1810 | list_for_each_entry_safe(page, page2, &n->partial, lru) { |
8ba00bb6 | 1811 | void *t; |
49e22585 | 1812 | |
8ba00bb6 JK |
1813 | if (!pfmemalloc_match(page, flags)) |
1814 | continue; | |
1815 | ||
633b0764 | 1816 | t = acquire_slab(s, n, page, object == NULL, &objects); |
49e22585 CL |
1817 | if (!t) |
1818 | break; | |
1819 | ||
633b0764 | 1820 | available += objects; |
12d79634 | 1821 | if (!object) { |
49e22585 | 1822 | c->page = page; |
49e22585 | 1823 | stat(s, ALLOC_FROM_PARTIAL); |
49e22585 | 1824 | object = t; |
49e22585 | 1825 | } else { |
633b0764 | 1826 | put_cpu_partial(s, page, 0); |
8028dcea | 1827 | stat(s, CPU_PARTIAL_NODE); |
49e22585 | 1828 | } |
345c905d JK |
1829 | if (!kmem_cache_has_cpu_partial(s) |
1830 | || available > s->cpu_partial / 2) | |
49e22585 CL |
1831 | break; |
1832 | ||
497b66f2 | 1833 | } |
81819f0f | 1834 | spin_unlock(&n->list_lock); |
497b66f2 | 1835 | return object; |
81819f0f CL |
1836 | } |
1837 | ||
1838 | /* | |
672bba3a | 1839 | * Get a page from somewhere. Search in increasing NUMA distances. |
81819f0f | 1840 | */ |
de3ec035 | 1841 | static void *get_any_partial(struct kmem_cache *s, gfp_t flags, |
acd19fd1 | 1842 | struct kmem_cache_cpu *c) |
81819f0f CL |
1843 | { |
1844 | #ifdef CONFIG_NUMA | |
1845 | struct zonelist *zonelist; | |
dd1a239f | 1846 | struct zoneref *z; |
54a6eb5c MG |
1847 | struct zone *zone; |
1848 | enum zone_type high_zoneidx = gfp_zone(flags); | |
497b66f2 | 1849 | void *object; |
cc9a6c87 | 1850 | unsigned int cpuset_mems_cookie; |
81819f0f CL |
1851 | |
1852 | /* | |
672bba3a CL |
1853 | * The defrag ratio allows a configuration of the tradeoffs between |
1854 | * inter node defragmentation and node local allocations. A lower | |
1855 | * defrag_ratio increases the tendency to do local allocations | |
1856 | * instead of attempting to obtain partial slabs from other nodes. | |
81819f0f | 1857 | * |
672bba3a CL |
1858 | * If the defrag_ratio is set to 0 then kmalloc() always |
1859 | * returns node local objects. If the ratio is higher then kmalloc() | |
1860 | * may return off node objects because partial slabs are obtained | |
1861 | * from other nodes and filled up. | |
81819f0f | 1862 | * |
43efd3ea LP |
1863 | * If /sys/kernel/slab/xx/remote_node_defrag_ratio is set to 100 |
1864 | * (which makes defrag_ratio = 1000) then every (well almost) | |
1865 | * allocation will first attempt to defrag slab caches on other nodes. | |
1866 | * This means scanning over all nodes to look for partial slabs which | |
1867 | * may be expensive if we do it every time we are trying to find a slab | |
672bba3a | 1868 | * with available objects. |
81819f0f | 1869 | */ |
9824601e CL |
1870 | if (!s->remote_node_defrag_ratio || |
1871 | get_cycles() % 1024 > s->remote_node_defrag_ratio) | |
81819f0f CL |
1872 | return NULL; |
1873 | ||
cc9a6c87 | 1874 | do { |
d26914d1 | 1875 | cpuset_mems_cookie = read_mems_allowed_begin(); |
2a389610 | 1876 | zonelist = node_zonelist(mempolicy_slab_node(), flags); |
cc9a6c87 MG |
1877 | for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) { |
1878 | struct kmem_cache_node *n; | |
1879 | ||
1880 | n = get_node(s, zone_to_nid(zone)); | |
1881 | ||
dee2f8aa | 1882 | if (n && cpuset_zone_allowed(zone, flags) && |
cc9a6c87 | 1883 | n->nr_partial > s->min_partial) { |
8ba00bb6 | 1884 | object = get_partial_node(s, n, c, flags); |
cc9a6c87 MG |
1885 | if (object) { |
1886 | /* | |
d26914d1 MG |
1887 | * Don't check read_mems_allowed_retry() |
1888 | * here - if mems_allowed was updated in | |
1889 | * parallel, that was a harmless race | |
1890 | * between allocation and the cpuset | |
1891 | * update | |
cc9a6c87 | 1892 | */ |
cc9a6c87 MG |
1893 | return object; |
1894 | } | |
c0ff7453 | 1895 | } |
81819f0f | 1896 | } |
d26914d1 | 1897 | } while (read_mems_allowed_retry(cpuset_mems_cookie)); |
81819f0f CL |
1898 | #endif |
1899 | return NULL; | |
1900 | } | |
1901 | ||
1902 | /* | |
1903 | * Get a partial page, lock it and return it. | |
1904 | */ | |
497b66f2 | 1905 | static void *get_partial(struct kmem_cache *s, gfp_t flags, int node, |
acd19fd1 | 1906 | struct kmem_cache_cpu *c) |
81819f0f | 1907 | { |
497b66f2 | 1908 | void *object; |
a561ce00 JK |
1909 | int searchnode = node; |
1910 | ||
1911 | if (node == NUMA_NO_NODE) | |
1912 | searchnode = numa_mem_id(); | |
1913 | else if (!node_present_pages(node)) | |
1914 | searchnode = node_to_mem_node(node); | |
81819f0f | 1915 | |
8ba00bb6 | 1916 | object = get_partial_node(s, get_node(s, searchnode), c, flags); |
497b66f2 CL |
1917 | if (object || node != NUMA_NO_NODE) |
1918 | return object; | |
81819f0f | 1919 | |
acd19fd1 | 1920 | return get_any_partial(s, flags, c); |
81819f0f CL |
1921 | } |
1922 | ||
8a5ec0ba CL |
1923 | #ifdef CONFIG_PREEMPT |
1924 | /* | |
1925 | * Calculate the next globally unique transaction for disambiguiation | |
1926 | * during cmpxchg. The transactions start with the cpu number and are then | |
1927 | * incremented by CONFIG_NR_CPUS. | |
1928 | */ | |
1929 | #define TID_STEP roundup_pow_of_two(CONFIG_NR_CPUS) | |
1930 | #else | |
1931 | /* | |
1932 | * No preemption supported therefore also no need to check for | |
1933 | * different cpus. | |
1934 | */ | |
1935 | #define TID_STEP 1 | |
1936 | #endif | |
1937 | ||
1938 | static inline unsigned long next_tid(unsigned long tid) | |
1939 | { | |
1940 | return tid + TID_STEP; | |
1941 | } | |
1942 | ||
1943 | static inline unsigned int tid_to_cpu(unsigned long tid) | |
1944 | { | |
1945 | return tid % TID_STEP; | |
1946 | } | |
1947 | ||
1948 | static inline unsigned long tid_to_event(unsigned long tid) | |
1949 | { | |
1950 | return tid / TID_STEP; | |
1951 | } | |
1952 | ||
1953 | static inline unsigned int init_tid(int cpu) | |
1954 | { | |
1955 | return cpu; | |
1956 | } | |
1957 | ||
1958 | static inline void note_cmpxchg_failure(const char *n, | |
1959 | const struct kmem_cache *s, unsigned long tid) | |
1960 | { | |
1961 | #ifdef SLUB_DEBUG_CMPXCHG | |
1962 | unsigned long actual_tid = __this_cpu_read(s->cpu_slab->tid); | |
1963 | ||
f9f58285 | 1964 | pr_info("%s %s: cmpxchg redo ", n, s->name); |
8a5ec0ba CL |
1965 | |
1966 | #ifdef CONFIG_PREEMPT | |
1967 | if (tid_to_cpu(tid) != tid_to_cpu(actual_tid)) | |
f9f58285 | 1968 | pr_warn("due to cpu change %d -> %d\n", |
8a5ec0ba CL |
1969 | tid_to_cpu(tid), tid_to_cpu(actual_tid)); |
1970 | else | |
1971 | #endif | |
1972 | if (tid_to_event(tid) != tid_to_event(actual_tid)) | |
f9f58285 | 1973 | pr_warn("due to cpu running other code. Event %ld->%ld\n", |
8a5ec0ba CL |
1974 | tid_to_event(tid), tid_to_event(actual_tid)); |
1975 | else | |
f9f58285 | 1976 | pr_warn("for unknown reason: actual=%lx was=%lx target=%lx\n", |
8a5ec0ba CL |
1977 | actual_tid, tid, next_tid(tid)); |
1978 | #endif | |
4fdccdfb | 1979 | stat(s, CMPXCHG_DOUBLE_CPU_FAIL); |
8a5ec0ba CL |
1980 | } |
1981 | ||
788e1aad | 1982 | static void init_kmem_cache_cpus(struct kmem_cache *s) |
8a5ec0ba | 1983 | { |
8a5ec0ba CL |
1984 | int cpu; |
1985 | ||
1986 | for_each_possible_cpu(cpu) | |
1987 | per_cpu_ptr(s->cpu_slab, cpu)->tid = init_tid(cpu); | |
8a5ec0ba | 1988 | } |
2cfb7455 | 1989 | |
81819f0f CL |
1990 | /* |
1991 | * Remove the cpu slab | |
1992 | */ | |
d0e0ac97 CG |
1993 | static void deactivate_slab(struct kmem_cache *s, struct page *page, |
1994 | void *freelist) | |
81819f0f | 1995 | { |
2cfb7455 | 1996 | enum slab_modes { M_NONE, M_PARTIAL, M_FULL, M_FREE }; |
2cfb7455 CL |
1997 | struct kmem_cache_node *n = get_node(s, page_to_nid(page)); |
1998 | int lock = 0; | |
1999 | enum slab_modes l = M_NONE, m = M_NONE; | |
2cfb7455 | 2000 | void *nextfree; |
136333d1 | 2001 | int tail = DEACTIVATE_TO_HEAD; |
2cfb7455 CL |
2002 | struct page new; |
2003 | struct page old; | |
2004 | ||
2005 | if (page->freelist) { | |
84e554e6 | 2006 | stat(s, DEACTIVATE_REMOTE_FREES); |
136333d1 | 2007 | tail = DEACTIVATE_TO_TAIL; |
2cfb7455 CL |
2008 | } |
2009 | ||
894b8788 | 2010 | /* |
2cfb7455 CL |
2011 | * Stage one: Free all available per cpu objects back |
2012 | * to the page freelist while it is still frozen. Leave the | |
2013 | * last one. | |
2014 | * | |
2015 | * There is no need to take the list->lock because the page | |
2016 | * is still frozen. | |
2017 | */ | |
2018 | while (freelist && (nextfree = get_freepointer(s, freelist))) { | |
2019 | void *prior; | |
2020 | unsigned long counters; | |
2021 | ||
2022 | do { | |
2023 | prior = page->freelist; | |
2024 | counters = page->counters; | |
2025 | set_freepointer(s, freelist, prior); | |
2026 | new.counters = counters; | |
2027 | new.inuse--; | |
a0132ac0 | 2028 | VM_BUG_ON(!new.frozen); |
2cfb7455 | 2029 | |
1d07171c | 2030 | } while (!__cmpxchg_double_slab(s, page, |
2cfb7455 CL |
2031 | prior, counters, |
2032 | freelist, new.counters, | |
2033 | "drain percpu freelist")); | |
2034 | ||
2035 | freelist = nextfree; | |
2036 | } | |
2037 | ||
894b8788 | 2038 | /* |
2cfb7455 CL |
2039 | * Stage two: Ensure that the page is unfrozen while the |
2040 | * list presence reflects the actual number of objects | |
2041 | * during unfreeze. | |
2042 | * | |
2043 | * We setup the list membership and then perform a cmpxchg | |
2044 | * with the count. If there is a mismatch then the page | |
2045 | * is not unfrozen but the page is on the wrong list. | |
2046 | * | |
2047 | * Then we restart the process which may have to remove | |
2048 | * the page from the list that we just put it on again | |
2049 | * because the number of objects in the slab may have | |
2050 | * changed. | |
894b8788 | 2051 | */ |
2cfb7455 | 2052 | redo: |
894b8788 | 2053 | |
2cfb7455 CL |
2054 | old.freelist = page->freelist; |
2055 | old.counters = page->counters; | |
a0132ac0 | 2056 | VM_BUG_ON(!old.frozen); |
7c2e132c | 2057 | |
2cfb7455 CL |
2058 | /* Determine target state of the slab */ |
2059 | new.counters = old.counters; | |
2060 | if (freelist) { | |
2061 | new.inuse--; | |
2062 | set_freepointer(s, freelist, old.freelist); | |
2063 | new.freelist = freelist; | |
2064 | } else | |
2065 | new.freelist = old.freelist; | |
2066 | ||
2067 | new.frozen = 0; | |
2068 | ||
8a5b20ae | 2069 | if (!new.inuse && n->nr_partial >= s->min_partial) |
2cfb7455 CL |
2070 | m = M_FREE; |
2071 | else if (new.freelist) { | |
2072 | m = M_PARTIAL; | |
2073 | if (!lock) { | |
2074 | lock = 1; | |
2075 | /* | |
2076 | * Taking the spinlock removes the possiblity | |
2077 | * that acquire_slab() will see a slab page that | |
2078 | * is frozen | |
2079 | */ | |
2080 | spin_lock(&n->list_lock); | |
2081 | } | |
2082 | } else { | |
2083 | m = M_FULL; | |
2084 | if (kmem_cache_debug(s) && !lock) { | |
2085 | lock = 1; | |
2086 | /* | |
2087 | * This also ensures that the scanning of full | |
2088 | * slabs from diagnostic functions will not see | |
2089 | * any frozen slabs. | |
2090 | */ | |
2091 | spin_lock(&n->list_lock); | |
2092 | } | |
2093 | } | |
2094 | ||
2095 | if (l != m) { | |
2096 | ||
2097 | if (l == M_PARTIAL) | |
2098 | ||
2099 | remove_partial(n, page); | |
2100 | ||
2101 | else if (l == M_FULL) | |
894b8788 | 2102 | |
c65c1877 | 2103 | remove_full(s, n, page); |
2cfb7455 CL |
2104 | |
2105 | if (m == M_PARTIAL) { | |
2106 | ||
2107 | add_partial(n, page, tail); | |
136333d1 | 2108 | stat(s, tail); |
2cfb7455 CL |
2109 | |
2110 | } else if (m == M_FULL) { | |
894b8788 | 2111 | |
2cfb7455 CL |
2112 | stat(s, DEACTIVATE_FULL); |
2113 | add_full(s, n, page); | |
2114 | ||
2115 | } | |
2116 | } | |
2117 | ||
2118 | l = m; | |
1d07171c | 2119 | if (!__cmpxchg_double_slab(s, page, |
2cfb7455 CL |
2120 | old.freelist, old.counters, |
2121 | new.freelist, new.counters, | |
2122 | "unfreezing slab")) | |
2123 | goto redo; | |
2124 | ||
2cfb7455 CL |
2125 | if (lock) |
2126 | spin_unlock(&n->list_lock); | |
2127 | ||
2128 | if (m == M_FREE) { | |
2129 | stat(s, DEACTIVATE_EMPTY); | |
2130 | discard_slab(s, page); | |
2131 | stat(s, FREE_SLAB); | |
894b8788 | 2132 | } |
81819f0f CL |
2133 | } |
2134 | ||
d24ac77f JK |
2135 | /* |
2136 | * Unfreeze all the cpu partial slabs. | |
2137 | * | |
59a09917 CL |
2138 | * This function must be called with interrupts disabled |
2139 | * for the cpu using c (or some other guarantee must be there | |
2140 | * to guarantee no concurrent accesses). | |
d24ac77f | 2141 | */ |
59a09917 CL |
2142 | static void unfreeze_partials(struct kmem_cache *s, |
2143 | struct kmem_cache_cpu *c) | |
49e22585 | 2144 | { |
345c905d | 2145 | #ifdef CONFIG_SLUB_CPU_PARTIAL |
43d77867 | 2146 | struct kmem_cache_node *n = NULL, *n2 = NULL; |
9ada1934 | 2147 | struct page *page, *discard_page = NULL; |
49e22585 CL |
2148 | |
2149 | while ((page = c->partial)) { | |
49e22585 CL |
2150 | struct page new; |
2151 | struct page old; | |
2152 | ||
2153 | c->partial = page->next; | |
43d77867 JK |
2154 | |
2155 | n2 = get_node(s, page_to_nid(page)); | |
2156 | if (n != n2) { | |
2157 | if (n) | |
2158 | spin_unlock(&n->list_lock); | |
2159 | ||
2160 | n = n2; | |
2161 | spin_lock(&n->list_lock); | |
2162 | } | |
49e22585 CL |
2163 | |
2164 | do { | |
2165 | ||
2166 | old.freelist = page->freelist; | |
2167 | old.counters = page->counters; | |
a0132ac0 | 2168 | VM_BUG_ON(!old.frozen); |
49e22585 CL |
2169 | |
2170 | new.counters = old.counters; | |
2171 | new.freelist = old.freelist; | |
2172 | ||
2173 | new.frozen = 0; | |
2174 | ||
d24ac77f | 2175 | } while (!__cmpxchg_double_slab(s, page, |
49e22585 CL |
2176 | old.freelist, old.counters, |
2177 | new.freelist, new.counters, | |
2178 | "unfreezing slab")); | |
2179 | ||
8a5b20ae | 2180 | if (unlikely(!new.inuse && n->nr_partial >= s->min_partial)) { |
9ada1934 SL |
2181 | page->next = discard_page; |
2182 | discard_page = page; | |
43d77867 JK |
2183 | } else { |
2184 | add_partial(n, page, DEACTIVATE_TO_TAIL); | |
2185 | stat(s, FREE_ADD_PARTIAL); | |
49e22585 CL |
2186 | } |
2187 | } | |
2188 | ||
2189 | if (n) | |
2190 | spin_unlock(&n->list_lock); | |
9ada1934 SL |
2191 | |
2192 | while (discard_page) { | |
2193 | page = discard_page; | |
2194 | discard_page = discard_page->next; | |
2195 | ||
2196 | stat(s, DEACTIVATE_EMPTY); | |
2197 | discard_slab(s, page); | |
2198 | stat(s, FREE_SLAB); | |
2199 | } | |
345c905d | 2200 | #endif |
49e22585 CL |
2201 | } |
2202 | ||
2203 | /* | |
2204 | * Put a page that was just frozen (in __slab_free) into a partial page | |
2205 | * slot if available. This is done without interrupts disabled and without | |
2206 | * preemption disabled. The cmpxchg is racy and may put the partial page | |
2207 | * onto a random cpus partial slot. | |
2208 | * | |
2209 | * If we did not find a slot then simply move all the partials to the | |
2210 | * per node partial list. | |
2211 | */ | |
633b0764 | 2212 | static void put_cpu_partial(struct kmem_cache *s, struct page *page, int drain) |
49e22585 | 2213 | { |
345c905d | 2214 | #ifdef CONFIG_SLUB_CPU_PARTIAL |
49e22585 CL |
2215 | struct page *oldpage; |
2216 | int pages; | |
2217 | int pobjects; | |
2218 | ||
d6e0b7fa | 2219 | preempt_disable(); |
49e22585 CL |
2220 | do { |
2221 | pages = 0; | |
2222 | pobjects = 0; | |
2223 | oldpage = this_cpu_read(s->cpu_slab->partial); | |
2224 | ||
2225 | if (oldpage) { | |
2226 | pobjects = oldpage->pobjects; | |
2227 | pages = oldpage->pages; | |
2228 | if (drain && pobjects > s->cpu_partial) { | |
2229 | unsigned long flags; | |
2230 | /* | |
2231 | * partial array is full. Move the existing | |
2232 | * set to the per node partial list. | |
2233 | */ | |
2234 | local_irq_save(flags); | |
59a09917 | 2235 | unfreeze_partials(s, this_cpu_ptr(s->cpu_slab)); |
49e22585 | 2236 | local_irq_restore(flags); |
e24fc410 | 2237 | oldpage = NULL; |
49e22585 CL |
2238 | pobjects = 0; |
2239 | pages = 0; | |
8028dcea | 2240 | stat(s, CPU_PARTIAL_DRAIN); |
49e22585 CL |
2241 | } |
2242 | } | |
2243 | ||
2244 | pages++; | |
2245 | pobjects += page->objects - page->inuse; | |
2246 | ||
2247 | page->pages = pages; | |
2248 | page->pobjects = pobjects; | |
2249 | page->next = oldpage; | |
2250 | ||
d0e0ac97 CG |
2251 | } while (this_cpu_cmpxchg(s->cpu_slab->partial, oldpage, page) |
2252 | != oldpage); | |
d6e0b7fa VD |
2253 | if (unlikely(!s->cpu_partial)) { |
2254 | unsigned long flags; | |
2255 | ||
2256 | local_irq_save(flags); | |
2257 | unfreeze_partials(s, this_cpu_ptr(s->cpu_slab)); | |
2258 | local_irq_restore(flags); | |
2259 | } | |
2260 | preempt_enable(); | |
345c905d | 2261 | #endif |
49e22585 CL |
2262 | } |
2263 | ||
dfb4f096 | 2264 | static inline void flush_slab(struct kmem_cache *s, struct kmem_cache_cpu *c) |
81819f0f | 2265 | { |
84e554e6 | 2266 | stat(s, CPUSLAB_FLUSH); |
c17dda40 CL |
2267 | deactivate_slab(s, c->page, c->freelist); |
2268 | ||
2269 | c->tid = next_tid(c->tid); | |
2270 | c->page = NULL; | |
2271 | c->freelist = NULL; | |
81819f0f CL |
2272 | } |
2273 | ||
2274 | /* | |
2275 | * Flush cpu slab. | |
6446faa2 | 2276 | * |
81819f0f CL |
2277 | * Called from IPI handler with interrupts disabled. |
2278 | */ | |
0c710013 | 2279 | static inline void __flush_cpu_slab(struct kmem_cache *s, int cpu) |
81819f0f | 2280 | { |
9dfc6e68 | 2281 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu); |
81819f0f | 2282 | |
49e22585 CL |
2283 | if (likely(c)) { |
2284 | if (c->page) | |
2285 | flush_slab(s, c); | |
2286 | ||
59a09917 | 2287 | unfreeze_partials(s, c); |
49e22585 | 2288 | } |
81819f0f CL |
2289 | } |
2290 | ||
2291 | static void flush_cpu_slab(void *d) | |
2292 | { | |
2293 | struct kmem_cache *s = d; | |
81819f0f | 2294 | |
dfb4f096 | 2295 | __flush_cpu_slab(s, smp_processor_id()); |
81819f0f CL |
2296 | } |
2297 | ||
a8364d55 GBY |
2298 | static bool has_cpu_slab(int cpu, void *info) |
2299 | { | |
2300 | struct kmem_cache *s = info; | |
2301 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu); | |
2302 | ||
02e1a9cd | 2303 | return c->page || c->partial; |
a8364d55 GBY |
2304 | } |
2305 | ||
81819f0f CL |
2306 | static void flush_all(struct kmem_cache *s) |
2307 | { | |
a8364d55 | 2308 | on_each_cpu_cond(has_cpu_slab, flush_cpu_slab, s, 1, GFP_ATOMIC); |
81819f0f CL |
2309 | } |
2310 | ||
a96a87bf SAS |
2311 | /* |
2312 | * Use the cpu notifier to insure that the cpu slabs are flushed when | |
2313 | * necessary. | |
2314 | */ | |
2315 | static int slub_cpu_dead(unsigned int cpu) | |
2316 | { | |
2317 | struct kmem_cache *s; | |
2318 | unsigned long flags; | |
2319 | ||
2320 | mutex_lock(&slab_mutex); | |
2321 | list_for_each_entry(s, &slab_caches, list) { | |
2322 | local_irq_save(flags); | |
2323 | __flush_cpu_slab(s, cpu); | |
2324 | local_irq_restore(flags); | |
2325 | } | |
2326 | mutex_unlock(&slab_mutex); | |
2327 | return 0; | |
2328 | } | |
2329 | ||
dfb4f096 CL |
2330 | /* |
2331 | * Check if the objects in a per cpu structure fit numa | |
2332 | * locality expectations. | |
2333 | */ | |
57d437d2 | 2334 | static inline int node_match(struct page *page, int node) |
dfb4f096 CL |
2335 | { |
2336 | #ifdef CONFIG_NUMA | |
4d7868e6 | 2337 | if (!page || (node != NUMA_NO_NODE && page_to_nid(page) != node)) |
dfb4f096 CL |
2338 | return 0; |
2339 | #endif | |
2340 | return 1; | |
2341 | } | |
2342 | ||
9a02d699 | 2343 | #ifdef CONFIG_SLUB_DEBUG |
781b2ba6 PE |
2344 | static int count_free(struct page *page) |
2345 | { | |
2346 | return page->objects - page->inuse; | |
2347 | } | |
2348 | ||
9a02d699 DR |
2349 | static inline unsigned long node_nr_objs(struct kmem_cache_node *n) |
2350 | { | |
2351 | return atomic_long_read(&n->total_objects); | |
2352 | } | |
2353 | #endif /* CONFIG_SLUB_DEBUG */ | |
2354 | ||
2355 | #if defined(CONFIG_SLUB_DEBUG) || defined(CONFIG_SYSFS) | |
781b2ba6 PE |
2356 | static unsigned long count_partial(struct kmem_cache_node *n, |
2357 | int (*get_count)(struct page *)) | |
2358 | { | |
2359 | unsigned long flags; | |
2360 | unsigned long x = 0; | |
2361 | struct page *page; | |
2362 | ||
2363 | spin_lock_irqsave(&n->list_lock, flags); | |
2364 | list_for_each_entry(page, &n->partial, lru) | |
2365 | x += get_count(page); | |
2366 | spin_unlock_irqrestore(&n->list_lock, flags); | |
2367 | return x; | |
2368 | } | |
9a02d699 | 2369 | #endif /* CONFIG_SLUB_DEBUG || CONFIG_SYSFS */ |
26c02cf0 | 2370 | |
781b2ba6 PE |
2371 | static noinline void |
2372 | slab_out_of_memory(struct kmem_cache *s, gfp_t gfpflags, int nid) | |
2373 | { | |
9a02d699 DR |
2374 | #ifdef CONFIG_SLUB_DEBUG |
2375 | static DEFINE_RATELIMIT_STATE(slub_oom_rs, DEFAULT_RATELIMIT_INTERVAL, | |
2376 | DEFAULT_RATELIMIT_BURST); | |
781b2ba6 | 2377 | int node; |
fa45dc25 | 2378 | struct kmem_cache_node *n; |
781b2ba6 | 2379 | |
9a02d699 DR |
2380 | if ((gfpflags & __GFP_NOWARN) || !__ratelimit(&slub_oom_rs)) |
2381 | return; | |
2382 | ||
5b3810e5 VB |
2383 | pr_warn("SLUB: Unable to allocate memory on node %d, gfp=%#x(%pGg)\n", |
2384 | nid, gfpflags, &gfpflags); | |
f9f58285 FF |
2385 | pr_warn(" cache: %s, object size: %d, buffer size: %d, default order: %d, min order: %d\n", |
2386 | s->name, s->object_size, s->size, oo_order(s->oo), | |
2387 | oo_order(s->min)); | |
781b2ba6 | 2388 | |
3b0efdfa | 2389 | if (oo_order(s->min) > get_order(s->object_size)) |
f9f58285 FF |
2390 | pr_warn(" %s debugging increased min order, use slub_debug=O to disable.\n", |
2391 | s->name); | |
fa5ec8a1 | 2392 | |
fa45dc25 | 2393 | for_each_kmem_cache_node(s, node, n) { |
781b2ba6 PE |
2394 | unsigned long nr_slabs; |
2395 | unsigned long nr_objs; | |
2396 | unsigned long nr_free; | |
2397 | ||
26c02cf0 AB |
2398 | nr_free = count_partial(n, count_free); |
2399 | nr_slabs = node_nr_slabs(n); | |
2400 | nr_objs = node_nr_objs(n); | |
781b2ba6 | 2401 | |
f9f58285 | 2402 | pr_warn(" node %d: slabs: %ld, objs: %ld, free: %ld\n", |
781b2ba6 PE |
2403 | node, nr_slabs, nr_objs, nr_free); |
2404 | } | |
9a02d699 | 2405 | #endif |
781b2ba6 PE |
2406 | } |
2407 | ||
497b66f2 CL |
2408 | static inline void *new_slab_objects(struct kmem_cache *s, gfp_t flags, |
2409 | int node, struct kmem_cache_cpu **pc) | |
2410 | { | |
6faa6833 | 2411 | void *freelist; |
188fd063 CL |
2412 | struct kmem_cache_cpu *c = *pc; |
2413 | struct page *page; | |
497b66f2 | 2414 | |
188fd063 | 2415 | freelist = get_partial(s, flags, node, c); |
497b66f2 | 2416 | |
188fd063 CL |
2417 | if (freelist) |
2418 | return freelist; | |
2419 | ||
2420 | page = new_slab(s, flags, node); | |
497b66f2 | 2421 | if (page) { |
7c8e0181 | 2422 | c = raw_cpu_ptr(s->cpu_slab); |
497b66f2 CL |
2423 | if (c->page) |
2424 | flush_slab(s, c); | |
2425 | ||
2426 | /* | |
2427 | * No other reference to the page yet so we can | |
2428 | * muck around with it freely without cmpxchg | |
2429 | */ | |
6faa6833 | 2430 | freelist = page->freelist; |
497b66f2 CL |
2431 | page->freelist = NULL; |
2432 | ||
2433 | stat(s, ALLOC_SLAB); | |
497b66f2 CL |
2434 | c->page = page; |
2435 | *pc = c; | |
2436 | } else | |
6faa6833 | 2437 | freelist = NULL; |
497b66f2 | 2438 | |
6faa6833 | 2439 | return freelist; |
497b66f2 CL |
2440 | } |
2441 | ||
072bb0aa MG |
2442 | static inline bool pfmemalloc_match(struct page *page, gfp_t gfpflags) |
2443 | { | |
2444 | if (unlikely(PageSlabPfmemalloc(page))) | |
2445 | return gfp_pfmemalloc_allowed(gfpflags); | |
2446 | ||
2447 | return true; | |
2448 | } | |
2449 | ||
213eeb9f | 2450 | /* |
d0e0ac97 CG |
2451 | * Check the page->freelist of a page and either transfer the freelist to the |
2452 | * per cpu freelist or deactivate the page. | |
213eeb9f CL |
2453 | * |
2454 | * The page is still frozen if the return value is not NULL. | |
2455 | * | |
2456 | * If this function returns NULL then the page has been unfrozen. | |
d24ac77f JK |
2457 | * |
2458 | * This function must be called with interrupt disabled. | |
213eeb9f CL |
2459 | */ |
2460 | static inline void *get_freelist(struct kmem_cache *s, struct page *page) | |
2461 | { | |
2462 | struct page new; | |
2463 | unsigned long counters; | |
2464 | void *freelist; | |
2465 | ||
2466 | do { | |
2467 | freelist = page->freelist; | |
2468 | counters = page->counters; | |
6faa6833 | 2469 | |
213eeb9f | 2470 | new.counters = counters; |
a0132ac0 | 2471 | VM_BUG_ON(!new.frozen); |
213eeb9f CL |
2472 | |
2473 | new.inuse = page->objects; | |
2474 | new.frozen = freelist != NULL; | |
2475 | ||
d24ac77f | 2476 | } while (!__cmpxchg_double_slab(s, page, |
213eeb9f CL |
2477 | freelist, counters, |
2478 | NULL, new.counters, | |
2479 | "get_freelist")); | |
2480 | ||
2481 | return freelist; | |
2482 | } | |
2483 | ||
81819f0f | 2484 | /* |
894b8788 CL |
2485 | * Slow path. The lockless freelist is empty or we need to perform |
2486 | * debugging duties. | |
2487 | * | |
894b8788 CL |
2488 | * Processing is still very fast if new objects have been freed to the |
2489 | * regular freelist. In that case we simply take over the regular freelist | |
2490 | * as the lockless freelist and zap the regular freelist. | |
81819f0f | 2491 | * |
894b8788 CL |
2492 | * If that is not working then we fall back to the partial lists. We take the |
2493 | * first element of the freelist as the object to allocate now and move the | |
2494 | * rest of the freelist to the lockless freelist. | |
81819f0f | 2495 | * |
894b8788 | 2496 | * And if we were unable to get a new slab from the partial slab lists then |
6446faa2 CL |
2497 | * we need to allocate a new slab. This is the slowest path since it involves |
2498 | * a call to the page allocator and the setup of a new slab. | |
a380a3c7 CL |
2499 | * |
2500 | * Version of __slab_alloc to use when we know that interrupts are | |
2501 | * already disabled (which is the case for bulk allocation). | |
81819f0f | 2502 | */ |
a380a3c7 | 2503 | static void *___slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node, |
ce71e27c | 2504 | unsigned long addr, struct kmem_cache_cpu *c) |
81819f0f | 2505 | { |
6faa6833 | 2506 | void *freelist; |
f6e7def7 | 2507 | struct page *page; |
81819f0f | 2508 | |
f6e7def7 CL |
2509 | page = c->page; |
2510 | if (!page) | |
81819f0f | 2511 | goto new_slab; |
49e22585 | 2512 | redo: |
6faa6833 | 2513 | |
57d437d2 | 2514 | if (unlikely(!node_match(page, node))) { |
a561ce00 JK |
2515 | int searchnode = node; |
2516 | ||
2517 | if (node != NUMA_NO_NODE && !node_present_pages(node)) | |
2518 | searchnode = node_to_mem_node(node); | |
2519 | ||
2520 | if (unlikely(!node_match(page, searchnode))) { | |
2521 | stat(s, ALLOC_NODE_MISMATCH); | |
2522 | deactivate_slab(s, page, c->freelist); | |
2523 | c->page = NULL; | |
2524 | c->freelist = NULL; | |
2525 | goto new_slab; | |
2526 | } | |
fc59c053 | 2527 | } |
6446faa2 | 2528 | |
072bb0aa MG |
2529 | /* |
2530 | * By rights, we should be searching for a slab page that was | |
2531 | * PFMEMALLOC but right now, we are losing the pfmemalloc | |
2532 | * information when the page leaves the per-cpu allocator | |
2533 | */ | |
2534 | if (unlikely(!pfmemalloc_match(page, gfpflags))) { | |
2535 | deactivate_slab(s, page, c->freelist); | |
2536 | c->page = NULL; | |
2537 | c->freelist = NULL; | |
2538 | goto new_slab; | |
2539 | } | |
2540 | ||
73736e03 | 2541 | /* must check again c->freelist in case of cpu migration or IRQ */ |
6faa6833 CL |
2542 | freelist = c->freelist; |
2543 | if (freelist) | |
73736e03 | 2544 | goto load_freelist; |
03e404af | 2545 | |
f6e7def7 | 2546 | freelist = get_freelist(s, page); |
6446faa2 | 2547 | |
6faa6833 | 2548 | if (!freelist) { |
03e404af CL |
2549 | c->page = NULL; |
2550 | stat(s, DEACTIVATE_BYPASS); | |
fc59c053 | 2551 | goto new_slab; |
03e404af | 2552 | } |
6446faa2 | 2553 | |
84e554e6 | 2554 | stat(s, ALLOC_REFILL); |
6446faa2 | 2555 | |
894b8788 | 2556 | load_freelist: |
507effea CL |
2557 | /* |
2558 | * freelist is pointing to the list of objects to be used. | |
2559 | * page is pointing to the page from which the objects are obtained. | |
2560 | * That page must be frozen for per cpu allocations to work. | |
2561 | */ | |
a0132ac0 | 2562 | VM_BUG_ON(!c->page->frozen); |
6faa6833 | 2563 | c->freelist = get_freepointer(s, freelist); |
8a5ec0ba | 2564 | c->tid = next_tid(c->tid); |
6faa6833 | 2565 | return freelist; |
81819f0f | 2566 | |
81819f0f | 2567 | new_slab: |
2cfb7455 | 2568 | |
49e22585 | 2569 | if (c->partial) { |
f6e7def7 CL |
2570 | page = c->page = c->partial; |
2571 | c->partial = page->next; | |
49e22585 CL |
2572 | stat(s, CPU_PARTIAL_ALLOC); |
2573 | c->freelist = NULL; | |
2574 | goto redo; | |
81819f0f CL |
2575 | } |
2576 | ||
188fd063 | 2577 | freelist = new_slab_objects(s, gfpflags, node, &c); |
01ad8a7b | 2578 | |
f4697436 | 2579 | if (unlikely(!freelist)) { |
9a02d699 | 2580 | slab_out_of_memory(s, gfpflags, node); |
f4697436 | 2581 | return NULL; |
81819f0f | 2582 | } |
2cfb7455 | 2583 | |
f6e7def7 | 2584 | page = c->page; |
5091b74a | 2585 | if (likely(!kmem_cache_debug(s) && pfmemalloc_match(page, gfpflags))) |
4b6f0750 | 2586 | goto load_freelist; |
2cfb7455 | 2587 | |
497b66f2 | 2588 | /* Only entered in the debug case */ |
d0e0ac97 CG |
2589 | if (kmem_cache_debug(s) && |
2590 | !alloc_debug_processing(s, page, freelist, addr)) | |
497b66f2 | 2591 | goto new_slab; /* Slab failed checks. Next slab needed */ |
894b8788 | 2592 | |
f6e7def7 | 2593 | deactivate_slab(s, page, get_freepointer(s, freelist)); |
c17dda40 CL |
2594 | c->page = NULL; |
2595 | c->freelist = NULL; | |
6faa6833 | 2596 | return freelist; |
894b8788 CL |
2597 | } |
2598 | ||
a380a3c7 CL |
2599 | /* |
2600 | * Another one that disabled interrupt and compensates for possible | |
2601 | * cpu changes by refetching the per cpu area pointer. | |
2602 | */ | |
2603 | static void *__slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node, | |
2604 | unsigned long addr, struct kmem_cache_cpu *c) | |
2605 | { | |
2606 | void *p; | |
2607 | unsigned long flags; | |
2608 | ||
2609 | local_irq_save(flags); | |
2610 | #ifdef CONFIG_PREEMPT | |
2611 | /* | |
2612 | * We may have been preempted and rescheduled on a different | |
2613 | * cpu before disabling interrupts. Need to reload cpu area | |
2614 | * pointer. | |
2615 | */ | |
2616 | c = this_cpu_ptr(s->cpu_slab); | |
2617 | #endif | |
2618 | ||
2619 | p = ___slab_alloc(s, gfpflags, node, addr, c); | |
2620 | local_irq_restore(flags); | |
2621 | return p; | |
2622 | } | |
2623 | ||
894b8788 CL |
2624 | /* |
2625 | * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc) | |
2626 | * have the fastpath folded into their functions. So no function call | |
2627 | * overhead for requests that can be satisfied on the fastpath. | |
2628 | * | |
2629 | * The fastpath works by first checking if the lockless freelist can be used. | |
2630 | * If not then __slab_alloc is called for slow processing. | |
2631 | * | |
2632 | * Otherwise we can simply pick the next object from the lockless free list. | |
2633 | */ | |
2b847c3c | 2634 | static __always_inline void *slab_alloc_node(struct kmem_cache *s, |
ce71e27c | 2635 | gfp_t gfpflags, int node, unsigned long addr) |
894b8788 | 2636 | { |
03ec0ed5 | 2637 | void *object; |
dfb4f096 | 2638 | struct kmem_cache_cpu *c; |
57d437d2 | 2639 | struct page *page; |
8a5ec0ba | 2640 | unsigned long tid; |
1f84260c | 2641 | |
8135be5a VD |
2642 | s = slab_pre_alloc_hook(s, gfpflags); |
2643 | if (!s) | |
773ff60e | 2644 | return NULL; |
8a5ec0ba | 2645 | redo: |
8a5ec0ba CL |
2646 | /* |
2647 | * Must read kmem_cache cpu data via this cpu ptr. Preemption is | |
2648 | * enabled. We may switch back and forth between cpus while | |
2649 | * reading from one cpu area. That does not matter as long | |
2650 | * as we end up on the original cpu again when doing the cmpxchg. | |
7cccd80b | 2651 | * |
9aabf810 JK |
2652 | * We should guarantee that tid and kmem_cache are retrieved on |
2653 | * the same cpu. It could be different if CONFIG_PREEMPT so we need | |
2654 | * to check if it is matched or not. | |
8a5ec0ba | 2655 | */ |
9aabf810 JK |
2656 | do { |
2657 | tid = this_cpu_read(s->cpu_slab->tid); | |
2658 | c = raw_cpu_ptr(s->cpu_slab); | |
859b7a0e MR |
2659 | } while (IS_ENABLED(CONFIG_PREEMPT) && |
2660 | unlikely(tid != READ_ONCE(c->tid))); | |
9aabf810 JK |
2661 | |
2662 | /* | |
2663 | * Irqless object alloc/free algorithm used here depends on sequence | |
2664 | * of fetching cpu_slab's data. tid should be fetched before anything | |
2665 | * on c to guarantee that object and page associated with previous tid | |
2666 | * won't be used with current tid. If we fetch tid first, object and | |
2667 | * page could be one associated with next tid and our alloc/free | |
2668 | * request will be failed. In this case, we will retry. So, no problem. | |
2669 | */ | |
2670 | barrier(); | |
8a5ec0ba | 2671 | |
8a5ec0ba CL |
2672 | /* |
2673 | * The transaction ids are globally unique per cpu and per operation on | |
2674 | * a per cpu queue. Thus they can be guarantee that the cmpxchg_double | |
2675 | * occurs on the right processor and that there was no operation on the | |
2676 | * linked list in between. | |
2677 | */ | |
8a5ec0ba | 2678 | |
9dfc6e68 | 2679 | object = c->freelist; |
57d437d2 | 2680 | page = c->page; |
8eae1492 | 2681 | if (unlikely(!object || !node_match(page, node))) { |
dfb4f096 | 2682 | object = __slab_alloc(s, gfpflags, node, addr, c); |
8eae1492 DH |
2683 | stat(s, ALLOC_SLOWPATH); |
2684 | } else { | |
0ad9500e ED |
2685 | void *next_object = get_freepointer_safe(s, object); |
2686 | ||
8a5ec0ba | 2687 | /* |
25985edc | 2688 | * The cmpxchg will only match if there was no additional |
8a5ec0ba CL |
2689 | * operation and if we are on the right processor. |
2690 | * | |
d0e0ac97 CG |
2691 | * The cmpxchg does the following atomically (without lock |
2692 | * semantics!) | |
8a5ec0ba CL |
2693 | * 1. Relocate first pointer to the current per cpu area. |
2694 | * 2. Verify that tid and freelist have not been changed | |
2695 | * 3. If they were not changed replace tid and freelist | |
2696 | * | |
d0e0ac97 CG |
2697 | * Since this is without lock semantics the protection is only |
2698 | * against code executing on this cpu *not* from access by | |
2699 | * other cpus. | |
8a5ec0ba | 2700 | */ |
933393f5 | 2701 | if (unlikely(!this_cpu_cmpxchg_double( |
8a5ec0ba CL |
2702 | s->cpu_slab->freelist, s->cpu_slab->tid, |
2703 | object, tid, | |
0ad9500e | 2704 | next_object, next_tid(tid)))) { |
8a5ec0ba CL |
2705 | |
2706 | note_cmpxchg_failure("slab_alloc", s, tid); | |
2707 | goto redo; | |
2708 | } | |
0ad9500e | 2709 | prefetch_freepointer(s, next_object); |
84e554e6 | 2710 | stat(s, ALLOC_FASTPATH); |
894b8788 | 2711 | } |
8a5ec0ba | 2712 | |
74e2134f | 2713 | if (unlikely(gfpflags & __GFP_ZERO) && object) |
3b0efdfa | 2714 | memset(object, 0, s->object_size); |
d07dbea4 | 2715 | |
03ec0ed5 | 2716 | slab_post_alloc_hook(s, gfpflags, 1, &object); |
5a896d9e | 2717 | |
894b8788 | 2718 | return object; |
81819f0f CL |
2719 | } |
2720 | ||
2b847c3c EG |
2721 | static __always_inline void *slab_alloc(struct kmem_cache *s, |
2722 | gfp_t gfpflags, unsigned long addr) | |
2723 | { | |
2724 | return slab_alloc_node(s, gfpflags, NUMA_NO_NODE, addr); | |
2725 | } | |
2726 | ||
81819f0f CL |
2727 | void *kmem_cache_alloc(struct kmem_cache *s, gfp_t gfpflags) |
2728 | { | |
2b847c3c | 2729 | void *ret = slab_alloc(s, gfpflags, _RET_IP_); |
5b882be4 | 2730 | |
d0e0ac97 CG |
2731 | trace_kmem_cache_alloc(_RET_IP_, ret, s->object_size, |
2732 | s->size, gfpflags); | |
5b882be4 EGM |
2733 | |
2734 | return ret; | |
81819f0f CL |
2735 | } |
2736 | EXPORT_SYMBOL(kmem_cache_alloc); | |
2737 | ||
0f24f128 | 2738 | #ifdef CONFIG_TRACING |
4a92379b RK |
2739 | void *kmem_cache_alloc_trace(struct kmem_cache *s, gfp_t gfpflags, size_t size) |
2740 | { | |
2b847c3c | 2741 | void *ret = slab_alloc(s, gfpflags, _RET_IP_); |
4a92379b | 2742 | trace_kmalloc(_RET_IP_, ret, size, s->size, gfpflags); |
505f5dcb | 2743 | kasan_kmalloc(s, ret, size, gfpflags); |
4a92379b RK |
2744 | return ret; |
2745 | } | |
2746 | EXPORT_SYMBOL(kmem_cache_alloc_trace); | |
5b882be4 EGM |
2747 | #endif |
2748 | ||
81819f0f CL |
2749 | #ifdef CONFIG_NUMA |
2750 | void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t gfpflags, int node) | |
2751 | { | |
2b847c3c | 2752 | void *ret = slab_alloc_node(s, gfpflags, node, _RET_IP_); |
5b882be4 | 2753 | |
ca2b84cb | 2754 | trace_kmem_cache_alloc_node(_RET_IP_, ret, |
3b0efdfa | 2755 | s->object_size, s->size, gfpflags, node); |
5b882be4 EGM |
2756 | |
2757 | return ret; | |
81819f0f CL |
2758 | } |
2759 | EXPORT_SYMBOL(kmem_cache_alloc_node); | |
81819f0f | 2760 | |
0f24f128 | 2761 | #ifdef CONFIG_TRACING |
4a92379b | 2762 | void *kmem_cache_alloc_node_trace(struct kmem_cache *s, |
5b882be4 | 2763 | gfp_t gfpflags, |
4a92379b | 2764 | int node, size_t size) |
5b882be4 | 2765 | { |
2b847c3c | 2766 | void *ret = slab_alloc_node(s, gfpflags, node, _RET_IP_); |
4a92379b RK |
2767 | |
2768 | trace_kmalloc_node(_RET_IP_, ret, | |
2769 | size, s->size, gfpflags, node); | |
0316bec2 | 2770 | |
505f5dcb | 2771 | kasan_kmalloc(s, ret, size, gfpflags); |
4a92379b | 2772 | return ret; |
5b882be4 | 2773 | } |
4a92379b | 2774 | EXPORT_SYMBOL(kmem_cache_alloc_node_trace); |
5b882be4 | 2775 | #endif |
5d1f57e4 | 2776 | #endif |
5b882be4 | 2777 | |
81819f0f | 2778 | /* |
94e4d712 | 2779 | * Slow path handling. This may still be called frequently since objects |
894b8788 | 2780 | * have a longer lifetime than the cpu slabs in most processing loads. |
81819f0f | 2781 | * |
894b8788 CL |
2782 | * So we still attempt to reduce cache line usage. Just take the slab |
2783 | * lock and free the item. If there is no additional partial page | |
2784 | * handling required then we can return immediately. | |
81819f0f | 2785 | */ |
894b8788 | 2786 | static void __slab_free(struct kmem_cache *s, struct page *page, |
81084651 JDB |
2787 | void *head, void *tail, int cnt, |
2788 | unsigned long addr) | |
2789 | ||
81819f0f CL |
2790 | { |
2791 | void *prior; | |
2cfb7455 | 2792 | int was_frozen; |
2cfb7455 CL |
2793 | struct page new; |
2794 | unsigned long counters; | |
2795 | struct kmem_cache_node *n = NULL; | |
61728d1e | 2796 | unsigned long uninitialized_var(flags); |
81819f0f | 2797 | |
8a5ec0ba | 2798 | stat(s, FREE_SLOWPATH); |
81819f0f | 2799 | |
19c7ff9e | 2800 | if (kmem_cache_debug(s) && |
282acb43 | 2801 | !free_debug_processing(s, page, head, tail, cnt, addr)) |
80f08c19 | 2802 | return; |
6446faa2 | 2803 | |
2cfb7455 | 2804 | do { |
837d678d JK |
2805 | if (unlikely(n)) { |
2806 | spin_unlock_irqrestore(&n->list_lock, flags); | |
2807 | n = NULL; | |
2808 | } | |
2cfb7455 CL |
2809 | prior = page->freelist; |
2810 | counters = page->counters; | |
81084651 | 2811 | set_freepointer(s, tail, prior); |
2cfb7455 CL |
2812 | new.counters = counters; |
2813 | was_frozen = new.frozen; | |
81084651 | 2814 | new.inuse -= cnt; |
837d678d | 2815 | if ((!new.inuse || !prior) && !was_frozen) { |
49e22585 | 2816 | |
c65c1877 | 2817 | if (kmem_cache_has_cpu_partial(s) && !prior) { |
49e22585 CL |
2818 | |
2819 | /* | |
d0e0ac97 CG |
2820 | * Slab was on no list before and will be |
2821 | * partially empty | |
2822 | * We can defer the list move and instead | |
2823 | * freeze it. | |
49e22585 CL |
2824 | */ |
2825 | new.frozen = 1; | |
2826 | ||
c65c1877 | 2827 | } else { /* Needs to be taken off a list */ |
49e22585 | 2828 | |
b455def2 | 2829 | n = get_node(s, page_to_nid(page)); |
49e22585 CL |
2830 | /* |
2831 | * Speculatively acquire the list_lock. | |
2832 | * If the cmpxchg does not succeed then we may | |
2833 | * drop the list_lock without any processing. | |
2834 | * | |
2835 | * Otherwise the list_lock will synchronize with | |
2836 | * other processors updating the list of slabs. | |
2837 | */ | |
2838 | spin_lock_irqsave(&n->list_lock, flags); | |
2839 | ||
2840 | } | |
2cfb7455 | 2841 | } |
81819f0f | 2842 | |
2cfb7455 CL |
2843 | } while (!cmpxchg_double_slab(s, page, |
2844 | prior, counters, | |
81084651 | 2845 | head, new.counters, |
2cfb7455 | 2846 | "__slab_free")); |
81819f0f | 2847 | |
2cfb7455 | 2848 | if (likely(!n)) { |
49e22585 CL |
2849 | |
2850 | /* | |
2851 | * If we just froze the page then put it onto the | |
2852 | * per cpu partial list. | |
2853 | */ | |
8028dcea | 2854 | if (new.frozen && !was_frozen) { |
49e22585 | 2855 | put_cpu_partial(s, page, 1); |
8028dcea AS |
2856 | stat(s, CPU_PARTIAL_FREE); |
2857 | } | |
49e22585 | 2858 | /* |
2cfb7455 CL |
2859 | * The list lock was not taken therefore no list |
2860 | * activity can be necessary. | |
2861 | */ | |
b455def2 L |
2862 | if (was_frozen) |
2863 | stat(s, FREE_FROZEN); | |
2864 | return; | |
2865 | } | |
81819f0f | 2866 | |
8a5b20ae | 2867 | if (unlikely(!new.inuse && n->nr_partial >= s->min_partial)) |
837d678d JK |
2868 | goto slab_empty; |
2869 | ||
81819f0f | 2870 | /* |
837d678d JK |
2871 | * Objects left in the slab. If it was not on the partial list before |
2872 | * then add it. | |
81819f0f | 2873 | */ |
345c905d JK |
2874 | if (!kmem_cache_has_cpu_partial(s) && unlikely(!prior)) { |
2875 | if (kmem_cache_debug(s)) | |
c65c1877 | 2876 | remove_full(s, n, page); |
837d678d JK |
2877 | add_partial(n, page, DEACTIVATE_TO_TAIL); |
2878 | stat(s, FREE_ADD_PARTIAL); | |
8ff12cfc | 2879 | } |
80f08c19 | 2880 | spin_unlock_irqrestore(&n->list_lock, flags); |
81819f0f CL |
2881 | return; |
2882 | ||
2883 | slab_empty: | |
a973e9dd | 2884 | if (prior) { |
81819f0f | 2885 | /* |
6fbabb20 | 2886 | * Slab on the partial list. |
81819f0f | 2887 | */ |
5cc6eee8 | 2888 | remove_partial(n, page); |
84e554e6 | 2889 | stat(s, FREE_REMOVE_PARTIAL); |
c65c1877 | 2890 | } else { |
6fbabb20 | 2891 | /* Slab must be on the full list */ |
c65c1877 PZ |
2892 | remove_full(s, n, page); |
2893 | } | |
2cfb7455 | 2894 | |
80f08c19 | 2895 | spin_unlock_irqrestore(&n->list_lock, flags); |
84e554e6 | 2896 | stat(s, FREE_SLAB); |
81819f0f | 2897 | discard_slab(s, page); |
81819f0f CL |
2898 | } |
2899 | ||
894b8788 CL |
2900 | /* |
2901 | * Fastpath with forced inlining to produce a kfree and kmem_cache_free that | |
2902 | * can perform fastpath freeing without additional function calls. | |
2903 | * | |
2904 | * The fastpath is only possible if we are freeing to the current cpu slab | |
2905 | * of this processor. This typically the case if we have just allocated | |
2906 | * the item before. | |
2907 | * | |
2908 | * If fastpath is not possible then fall back to __slab_free where we deal | |
2909 | * with all sorts of special processing. | |
81084651 JDB |
2910 | * |
2911 | * Bulk free of a freelist with several objects (all pointing to the | |
2912 | * same page) possible by specifying head and tail ptr, plus objects | |
2913 | * count (cnt). Bulk free indicated by tail pointer being set. | |
894b8788 | 2914 | */ |
80a9201a AP |
2915 | static __always_inline void do_slab_free(struct kmem_cache *s, |
2916 | struct page *page, void *head, void *tail, | |
2917 | int cnt, unsigned long addr) | |
894b8788 | 2918 | { |
81084651 | 2919 | void *tail_obj = tail ? : head; |
dfb4f096 | 2920 | struct kmem_cache_cpu *c; |
8a5ec0ba | 2921 | unsigned long tid; |
8a5ec0ba CL |
2922 | redo: |
2923 | /* | |
2924 | * Determine the currently cpus per cpu slab. | |
2925 | * The cpu may change afterward. However that does not matter since | |
2926 | * data is retrieved via this pointer. If we are on the same cpu | |
2ae44005 | 2927 | * during the cmpxchg then the free will succeed. |
8a5ec0ba | 2928 | */ |
9aabf810 JK |
2929 | do { |
2930 | tid = this_cpu_read(s->cpu_slab->tid); | |
2931 | c = raw_cpu_ptr(s->cpu_slab); | |
859b7a0e MR |
2932 | } while (IS_ENABLED(CONFIG_PREEMPT) && |
2933 | unlikely(tid != READ_ONCE(c->tid))); | |
c016b0bd | 2934 | |
9aabf810 JK |
2935 | /* Same with comment on barrier() in slab_alloc_node() */ |
2936 | barrier(); | |
c016b0bd | 2937 | |
442b06bc | 2938 | if (likely(page == c->page)) { |
81084651 | 2939 | set_freepointer(s, tail_obj, c->freelist); |
8a5ec0ba | 2940 | |
933393f5 | 2941 | if (unlikely(!this_cpu_cmpxchg_double( |
8a5ec0ba CL |
2942 | s->cpu_slab->freelist, s->cpu_slab->tid, |
2943 | c->freelist, tid, | |
81084651 | 2944 | head, next_tid(tid)))) { |
8a5ec0ba CL |
2945 | |
2946 | note_cmpxchg_failure("slab_free", s, tid); | |
2947 | goto redo; | |
2948 | } | |
84e554e6 | 2949 | stat(s, FREE_FASTPATH); |
894b8788 | 2950 | } else |
81084651 | 2951 | __slab_free(s, page, head, tail_obj, cnt, addr); |
894b8788 | 2952 | |
894b8788 CL |
2953 | } |
2954 | ||
80a9201a AP |
2955 | static __always_inline void slab_free(struct kmem_cache *s, struct page *page, |
2956 | void *head, void *tail, int cnt, | |
2957 | unsigned long addr) | |
2958 | { | |
2959 | slab_free_freelist_hook(s, head, tail); | |
2960 | /* | |
2961 | * slab_free_freelist_hook() could have put the items into quarantine. | |
2962 | * If so, no need to free them. | |
2963 | */ | |
2964 | if (s->flags & SLAB_KASAN && !(s->flags & SLAB_DESTROY_BY_RCU)) | |
2965 | return; | |
2966 | do_slab_free(s, page, head, tail, cnt, addr); | |
2967 | } | |
2968 | ||
2969 | #ifdef CONFIG_KASAN | |
2970 | void ___cache_free(struct kmem_cache *cache, void *x, unsigned long addr) | |
2971 | { | |
2972 | do_slab_free(cache, virt_to_head_page(x), x, NULL, 1, addr); | |
2973 | } | |
2974 | #endif | |
2975 | ||
81819f0f CL |
2976 | void kmem_cache_free(struct kmem_cache *s, void *x) |
2977 | { | |
b9ce5ef4 GC |
2978 | s = cache_from_obj(s, x); |
2979 | if (!s) | |
79576102 | 2980 | return; |
81084651 | 2981 | slab_free(s, virt_to_head_page(x), x, NULL, 1, _RET_IP_); |
ca2b84cb | 2982 | trace_kmem_cache_free(_RET_IP_, x); |
81819f0f CL |
2983 | } |
2984 | EXPORT_SYMBOL(kmem_cache_free); | |
2985 | ||
d0ecd894 | 2986 | struct detached_freelist { |
fbd02630 | 2987 | struct page *page; |
d0ecd894 JDB |
2988 | void *tail; |
2989 | void *freelist; | |
2990 | int cnt; | |
376bf125 | 2991 | struct kmem_cache *s; |
d0ecd894 | 2992 | }; |
fbd02630 | 2993 | |
d0ecd894 JDB |
2994 | /* |
2995 | * This function progressively scans the array with free objects (with | |
2996 | * a limited look ahead) and extract objects belonging to the same | |
2997 | * page. It builds a detached freelist directly within the given | |
2998 | * page/objects. This can happen without any need for | |
2999 | * synchronization, because the objects are owned by running process. | |
3000 | * The freelist is build up as a single linked list in the objects. | |
3001 | * The idea is, that this detached freelist can then be bulk | |
3002 | * transferred to the real freelist(s), but only requiring a single | |
3003 | * synchronization primitive. Look ahead in the array is limited due | |
3004 | * to performance reasons. | |
3005 | */ | |
376bf125 JDB |
3006 | static inline |
3007 | int build_detached_freelist(struct kmem_cache *s, size_t size, | |
3008 | void **p, struct detached_freelist *df) | |
d0ecd894 JDB |
3009 | { |
3010 | size_t first_skipped_index = 0; | |
3011 | int lookahead = 3; | |
3012 | void *object; | |
ca257195 | 3013 | struct page *page; |
fbd02630 | 3014 | |
d0ecd894 JDB |
3015 | /* Always re-init detached_freelist */ |
3016 | df->page = NULL; | |
fbd02630 | 3017 | |
d0ecd894 JDB |
3018 | do { |
3019 | object = p[--size]; | |
ca257195 | 3020 | /* Do we need !ZERO_OR_NULL_PTR(object) here? (for kfree) */ |
d0ecd894 | 3021 | } while (!object && size); |
3eed034d | 3022 | |
d0ecd894 JDB |
3023 | if (!object) |
3024 | return 0; | |
fbd02630 | 3025 | |
ca257195 JDB |
3026 | page = virt_to_head_page(object); |
3027 | if (!s) { | |
3028 | /* Handle kalloc'ed objects */ | |
3029 | if (unlikely(!PageSlab(page))) { | |
3030 | BUG_ON(!PageCompound(page)); | |
3031 | kfree_hook(object); | |
4949148a | 3032 | __free_pages(page, compound_order(page)); |
ca257195 JDB |
3033 | p[size] = NULL; /* mark object processed */ |
3034 | return size; | |
3035 | } | |
3036 | /* Derive kmem_cache from object */ | |
3037 | df->s = page->slab_cache; | |
3038 | } else { | |
3039 | df->s = cache_from_obj(s, object); /* Support for memcg */ | |
3040 | } | |
376bf125 | 3041 | |
d0ecd894 | 3042 | /* Start new detached freelist */ |
ca257195 | 3043 | df->page = page; |
376bf125 | 3044 | set_freepointer(df->s, object, NULL); |
d0ecd894 JDB |
3045 | df->tail = object; |
3046 | df->freelist = object; | |
3047 | p[size] = NULL; /* mark object processed */ | |
3048 | df->cnt = 1; | |
3049 | ||
3050 | while (size) { | |
3051 | object = p[--size]; | |
3052 | if (!object) | |
3053 | continue; /* Skip processed objects */ | |
3054 | ||
3055 | /* df->page is always set at this point */ | |
3056 | if (df->page == virt_to_head_page(object)) { | |
3057 | /* Opportunity build freelist */ | |
376bf125 | 3058 | set_freepointer(df->s, object, df->freelist); |
d0ecd894 JDB |
3059 | df->freelist = object; |
3060 | df->cnt++; | |
3061 | p[size] = NULL; /* mark object processed */ | |
3062 | ||
3063 | continue; | |
fbd02630 | 3064 | } |
d0ecd894 JDB |
3065 | |
3066 | /* Limit look ahead search */ | |
3067 | if (!--lookahead) | |
3068 | break; | |
3069 | ||
3070 | if (!first_skipped_index) | |
3071 | first_skipped_index = size + 1; | |
fbd02630 | 3072 | } |
d0ecd894 JDB |
3073 | |
3074 | return first_skipped_index; | |
3075 | } | |
3076 | ||
d0ecd894 | 3077 | /* Note that interrupts must be enabled when calling this function. */ |
376bf125 | 3078 | void kmem_cache_free_bulk(struct kmem_cache *s, size_t size, void **p) |
d0ecd894 JDB |
3079 | { |
3080 | if (WARN_ON(!size)) | |
3081 | return; | |
3082 | ||
3083 | do { | |
3084 | struct detached_freelist df; | |
3085 | ||
3086 | size = build_detached_freelist(s, size, p, &df); | |
84582c8a | 3087 | if (!df.page) |
d0ecd894 JDB |
3088 | continue; |
3089 | ||
376bf125 | 3090 | slab_free(df.s, df.page, df.freelist, df.tail, df.cnt,_RET_IP_); |
d0ecd894 | 3091 | } while (likely(size)); |
484748f0 CL |
3092 | } |
3093 | EXPORT_SYMBOL(kmem_cache_free_bulk); | |
3094 | ||
994eb764 | 3095 | /* Note that interrupts must be enabled when calling this function. */ |
865762a8 JDB |
3096 | int kmem_cache_alloc_bulk(struct kmem_cache *s, gfp_t flags, size_t size, |
3097 | void **p) | |
484748f0 | 3098 | { |
994eb764 JDB |
3099 | struct kmem_cache_cpu *c; |
3100 | int i; | |
3101 | ||
03ec0ed5 JDB |
3102 | /* memcg and kmem_cache debug support */ |
3103 | s = slab_pre_alloc_hook(s, flags); | |
3104 | if (unlikely(!s)) | |
3105 | return false; | |
994eb764 JDB |
3106 | /* |
3107 | * Drain objects in the per cpu slab, while disabling local | |
3108 | * IRQs, which protects against PREEMPT and interrupts | |
3109 | * handlers invoking normal fastpath. | |
3110 | */ | |
3111 | local_irq_disable(); | |
3112 | c = this_cpu_ptr(s->cpu_slab); | |
3113 | ||
3114 | for (i = 0; i < size; i++) { | |
3115 | void *object = c->freelist; | |
3116 | ||
ebe909e0 | 3117 | if (unlikely(!object)) { |
ebe909e0 JDB |
3118 | /* |
3119 | * Invoking slow path likely have side-effect | |
3120 | * of re-populating per CPU c->freelist | |
3121 | */ | |
87098373 | 3122 | p[i] = ___slab_alloc(s, flags, NUMA_NO_NODE, |
ebe909e0 | 3123 | _RET_IP_, c); |
87098373 CL |
3124 | if (unlikely(!p[i])) |
3125 | goto error; | |
3126 | ||
ebe909e0 JDB |
3127 | c = this_cpu_ptr(s->cpu_slab); |
3128 | continue; /* goto for-loop */ | |
3129 | } | |
994eb764 JDB |
3130 | c->freelist = get_freepointer(s, object); |
3131 | p[i] = object; | |
3132 | } | |
3133 | c->tid = next_tid(c->tid); | |
3134 | local_irq_enable(); | |
3135 | ||
3136 | /* Clear memory outside IRQ disabled fastpath loop */ | |
3137 | if (unlikely(flags & __GFP_ZERO)) { | |
3138 | int j; | |
3139 | ||
3140 | for (j = 0; j < i; j++) | |
3141 | memset(p[j], 0, s->object_size); | |
3142 | } | |
3143 | ||
03ec0ed5 JDB |
3144 | /* memcg and kmem_cache debug support */ |
3145 | slab_post_alloc_hook(s, flags, size, p); | |
865762a8 | 3146 | return i; |
87098373 | 3147 | error: |
87098373 | 3148 | local_irq_enable(); |
03ec0ed5 JDB |
3149 | slab_post_alloc_hook(s, flags, i, p); |
3150 | __kmem_cache_free_bulk(s, i, p); | |
865762a8 | 3151 | return 0; |
484748f0 CL |
3152 | } |
3153 | EXPORT_SYMBOL(kmem_cache_alloc_bulk); | |
3154 | ||
3155 | ||
81819f0f | 3156 | /* |
672bba3a CL |
3157 | * Object placement in a slab is made very easy because we always start at |
3158 | * offset 0. If we tune the size of the object to the alignment then we can | |
3159 | * get the required alignment by putting one properly sized object after | |
3160 | * another. | |
81819f0f CL |
3161 | * |
3162 | * Notice that the allocation order determines the sizes of the per cpu | |
3163 | * caches. Each processor has always one slab available for allocations. | |
3164 | * Increasing the allocation order reduces the number of times that slabs | |
672bba3a | 3165 | * must be moved on and off the partial lists and is therefore a factor in |
81819f0f | 3166 | * locking overhead. |
81819f0f CL |
3167 | */ |
3168 | ||
3169 | /* | |
3170 | * Mininum / Maximum order of slab pages. This influences locking overhead | |
3171 | * and slab fragmentation. A higher order reduces the number of partial slabs | |
3172 | * and increases the number of allocations possible without having to | |
3173 | * take the list_lock. | |
3174 | */ | |
3175 | static int slub_min_order; | |
114e9e89 | 3176 | static int slub_max_order = PAGE_ALLOC_COSTLY_ORDER; |
9b2cd506 | 3177 | static int slub_min_objects; |
81819f0f | 3178 | |
81819f0f CL |
3179 | /* |
3180 | * Calculate the order of allocation given an slab object size. | |
3181 | * | |
672bba3a CL |
3182 | * The order of allocation has significant impact on performance and other |
3183 | * system components. Generally order 0 allocations should be preferred since | |
3184 | * order 0 does not cause fragmentation in the page allocator. Larger objects | |
3185 | * be problematic to put into order 0 slabs because there may be too much | |
c124f5b5 | 3186 | * unused space left. We go to a higher order if more than 1/16th of the slab |
672bba3a CL |
3187 | * would be wasted. |
3188 | * | |
3189 | * In order to reach satisfactory performance we must ensure that a minimum | |
3190 | * number of objects is in one slab. Otherwise we may generate too much | |
3191 | * activity on the partial lists which requires taking the list_lock. This is | |
3192 | * less a concern for large slabs though which are rarely used. | |
81819f0f | 3193 | * |
672bba3a CL |
3194 | * slub_max_order specifies the order where we begin to stop considering the |
3195 | * number of objects in a slab as critical. If we reach slub_max_order then | |
3196 | * we try to keep the page order as low as possible. So we accept more waste | |
3197 | * of space in favor of a small page order. | |
81819f0f | 3198 | * |
672bba3a CL |
3199 | * Higher order allocations also allow the placement of more objects in a |
3200 | * slab and thereby reduce object handling overhead. If the user has | |
3201 | * requested a higher mininum order then we start with that one instead of | |
3202 | * the smallest order which will fit the object. | |
81819f0f | 3203 | */ |
5e6d444e | 3204 | static inline int slab_order(int size, int min_objects, |
ab9a0f19 | 3205 | int max_order, int fract_leftover, int reserved) |
81819f0f CL |
3206 | { |
3207 | int order; | |
3208 | int rem; | |
6300ea75 | 3209 | int min_order = slub_min_order; |
81819f0f | 3210 | |
ab9a0f19 | 3211 | if (order_objects(min_order, size, reserved) > MAX_OBJS_PER_PAGE) |
210b5c06 | 3212 | return get_order(size * MAX_OBJS_PER_PAGE) - 1; |
39b26464 | 3213 | |
9f835703 | 3214 | for (order = max(min_order, get_order(min_objects * size + reserved)); |
5e6d444e | 3215 | order <= max_order; order++) { |
81819f0f | 3216 | |
5e6d444e | 3217 | unsigned long slab_size = PAGE_SIZE << order; |
81819f0f | 3218 | |
ab9a0f19 | 3219 | rem = (slab_size - reserved) % size; |
81819f0f | 3220 | |
5e6d444e | 3221 | if (rem <= slab_size / fract_leftover) |
81819f0f | 3222 | break; |
81819f0f | 3223 | } |
672bba3a | 3224 | |
81819f0f CL |
3225 | return order; |
3226 | } | |
3227 | ||
ab9a0f19 | 3228 | static inline int calculate_order(int size, int reserved) |
5e6d444e CL |
3229 | { |
3230 | int order; | |
3231 | int min_objects; | |
3232 | int fraction; | |
e8120ff1 | 3233 | int max_objects; |
5e6d444e CL |
3234 | |
3235 | /* | |
3236 | * Attempt to find best configuration for a slab. This | |
3237 | * works by first attempting to generate a layout with | |
3238 | * the best configuration and backing off gradually. | |
3239 | * | |
422ff4d7 | 3240 | * First we increase the acceptable waste in a slab. Then |
5e6d444e CL |
3241 | * we reduce the minimum objects required in a slab. |
3242 | */ | |
3243 | min_objects = slub_min_objects; | |
9b2cd506 CL |
3244 | if (!min_objects) |
3245 | min_objects = 4 * (fls(nr_cpu_ids) + 1); | |
ab9a0f19 | 3246 | max_objects = order_objects(slub_max_order, size, reserved); |
e8120ff1 ZY |
3247 | min_objects = min(min_objects, max_objects); |
3248 | ||
5e6d444e | 3249 | while (min_objects > 1) { |
c124f5b5 | 3250 | fraction = 16; |
5e6d444e CL |
3251 | while (fraction >= 4) { |
3252 | order = slab_order(size, min_objects, | |
ab9a0f19 | 3253 | slub_max_order, fraction, reserved); |
5e6d444e CL |
3254 | if (order <= slub_max_order) |
3255 | return order; | |
3256 | fraction /= 2; | |
3257 | } | |
5086c389 | 3258 | min_objects--; |
5e6d444e CL |
3259 | } |
3260 | ||
3261 | /* | |
3262 | * We were unable to place multiple objects in a slab. Now | |
3263 | * lets see if we can place a single object there. | |
3264 | */ | |
ab9a0f19 | 3265 | order = slab_order(size, 1, slub_max_order, 1, reserved); |
5e6d444e CL |
3266 | if (order <= slub_max_order) |
3267 | return order; | |
3268 | ||
3269 | /* | |
3270 | * Doh this slab cannot be placed using slub_max_order. | |
3271 | */ | |
ab9a0f19 | 3272 | order = slab_order(size, 1, MAX_ORDER, 1, reserved); |
818cf590 | 3273 | if (order < MAX_ORDER) |
5e6d444e CL |
3274 | return order; |
3275 | return -ENOSYS; | |
3276 | } | |
3277 | ||
5595cffc | 3278 | static void |
4053497d | 3279 | init_kmem_cache_node(struct kmem_cache_node *n) |
81819f0f CL |
3280 | { |
3281 | n->nr_partial = 0; | |
81819f0f CL |
3282 | spin_lock_init(&n->list_lock); |
3283 | INIT_LIST_HEAD(&n->partial); | |
8ab1372f | 3284 | #ifdef CONFIG_SLUB_DEBUG |
0f389ec6 | 3285 | atomic_long_set(&n->nr_slabs, 0); |
02b71b70 | 3286 | atomic_long_set(&n->total_objects, 0); |
643b1138 | 3287 | INIT_LIST_HEAD(&n->full); |
8ab1372f | 3288 | #endif |
81819f0f CL |
3289 | } |
3290 | ||
55136592 | 3291 | static inline int alloc_kmem_cache_cpus(struct kmem_cache *s) |
4c93c355 | 3292 | { |
6c182dc0 | 3293 | BUILD_BUG_ON(PERCPU_DYNAMIC_EARLY_SIZE < |
95a05b42 | 3294 | KMALLOC_SHIFT_HIGH * sizeof(struct kmem_cache_cpu)); |
4c93c355 | 3295 | |
8a5ec0ba | 3296 | /* |
d4d84fef CM |
3297 | * Must align to double word boundary for the double cmpxchg |
3298 | * instructions to work; see __pcpu_double_call_return_bool(). | |
8a5ec0ba | 3299 | */ |
d4d84fef CM |
3300 | s->cpu_slab = __alloc_percpu(sizeof(struct kmem_cache_cpu), |
3301 | 2 * sizeof(void *)); | |
8a5ec0ba CL |
3302 | |
3303 | if (!s->cpu_slab) | |
3304 | return 0; | |
3305 | ||
3306 | init_kmem_cache_cpus(s); | |
4c93c355 | 3307 | |
8a5ec0ba | 3308 | return 1; |
4c93c355 | 3309 | } |
4c93c355 | 3310 | |
51df1142 CL |
3311 | static struct kmem_cache *kmem_cache_node; |
3312 | ||
81819f0f CL |
3313 | /* |
3314 | * No kmalloc_node yet so do it by hand. We know that this is the first | |
3315 | * slab on the node for this slabcache. There are no concurrent accesses | |
3316 | * possible. | |
3317 | * | |
721ae22a ZYW |
3318 | * Note that this function only works on the kmem_cache_node |
3319 | * when allocating for the kmem_cache_node. This is used for bootstrapping | |
4c93c355 | 3320 | * memory on a fresh node that has no slab structures yet. |
81819f0f | 3321 | */ |
55136592 | 3322 | static void early_kmem_cache_node_alloc(int node) |
81819f0f CL |
3323 | { |
3324 | struct page *page; | |
3325 | struct kmem_cache_node *n; | |
3326 | ||
51df1142 | 3327 | BUG_ON(kmem_cache_node->size < sizeof(struct kmem_cache_node)); |
81819f0f | 3328 | |
51df1142 | 3329 | page = new_slab(kmem_cache_node, GFP_NOWAIT, node); |
81819f0f CL |
3330 | |
3331 | BUG_ON(!page); | |
a2f92ee7 | 3332 | if (page_to_nid(page) != node) { |
f9f58285 FF |
3333 | pr_err("SLUB: Unable to allocate memory from node %d\n", node); |
3334 | pr_err("SLUB: Allocating a useless per node structure in order to be able to continue\n"); | |
a2f92ee7 CL |
3335 | } |
3336 | ||
81819f0f CL |
3337 | n = page->freelist; |
3338 | BUG_ON(!n); | |
51df1142 | 3339 | page->freelist = get_freepointer(kmem_cache_node, n); |
e6e82ea1 | 3340 | page->inuse = 1; |
8cb0a506 | 3341 | page->frozen = 0; |
51df1142 | 3342 | kmem_cache_node->node[node] = n; |
8ab1372f | 3343 | #ifdef CONFIG_SLUB_DEBUG |
f7cb1933 | 3344 | init_object(kmem_cache_node, n, SLUB_RED_ACTIVE); |
51df1142 | 3345 | init_tracking(kmem_cache_node, n); |
8ab1372f | 3346 | #endif |
505f5dcb AP |
3347 | kasan_kmalloc(kmem_cache_node, n, sizeof(struct kmem_cache_node), |
3348 | GFP_KERNEL); | |
4053497d | 3349 | init_kmem_cache_node(n); |
51df1142 | 3350 | inc_slabs_node(kmem_cache_node, node, page->objects); |
6446faa2 | 3351 | |
67b6c900 | 3352 | /* |
1e4dd946 SR |
3353 | * No locks need to be taken here as it has just been |
3354 | * initialized and there is no concurrent access. | |
67b6c900 | 3355 | */ |
1e4dd946 | 3356 | __add_partial(n, page, DEACTIVATE_TO_HEAD); |
81819f0f CL |
3357 | } |
3358 | ||
3359 | static void free_kmem_cache_nodes(struct kmem_cache *s) | |
3360 | { | |
3361 | int node; | |
fa45dc25 | 3362 | struct kmem_cache_node *n; |
81819f0f | 3363 | |
fa45dc25 CL |
3364 | for_each_kmem_cache_node(s, node, n) { |
3365 | kmem_cache_free(kmem_cache_node, n); | |
81819f0f CL |
3366 | s->node[node] = NULL; |
3367 | } | |
3368 | } | |
3369 | ||
52b4b950 DS |
3370 | void __kmem_cache_release(struct kmem_cache *s) |
3371 | { | |
210e7a43 | 3372 | cache_random_seq_destroy(s); |
52b4b950 DS |
3373 | free_percpu(s->cpu_slab); |
3374 | free_kmem_cache_nodes(s); | |
3375 | } | |
3376 | ||
55136592 | 3377 | static int init_kmem_cache_nodes(struct kmem_cache *s) |
81819f0f CL |
3378 | { |
3379 | int node; | |
81819f0f | 3380 | |
f64dc58c | 3381 | for_each_node_state(node, N_NORMAL_MEMORY) { |
81819f0f CL |
3382 | struct kmem_cache_node *n; |
3383 | ||
73367bd8 | 3384 | if (slab_state == DOWN) { |
55136592 | 3385 | early_kmem_cache_node_alloc(node); |
73367bd8 AD |
3386 | continue; |
3387 | } | |
51df1142 | 3388 | n = kmem_cache_alloc_node(kmem_cache_node, |
55136592 | 3389 | GFP_KERNEL, node); |
81819f0f | 3390 | |
73367bd8 AD |
3391 | if (!n) { |
3392 | free_kmem_cache_nodes(s); | |
3393 | return 0; | |
81819f0f | 3394 | } |
73367bd8 | 3395 | |
81819f0f | 3396 | s->node[node] = n; |
4053497d | 3397 | init_kmem_cache_node(n); |
81819f0f CL |
3398 | } |
3399 | return 1; | |
3400 | } | |
81819f0f | 3401 | |
c0bdb232 | 3402 | static void set_min_partial(struct kmem_cache *s, unsigned long min) |
3b89d7d8 DR |
3403 | { |
3404 | if (min < MIN_PARTIAL) | |
3405 | min = MIN_PARTIAL; | |
3406 | else if (min > MAX_PARTIAL) | |
3407 | min = MAX_PARTIAL; | |
3408 | s->min_partial = min; | |
3409 | } | |
3410 | ||
81819f0f CL |
3411 | /* |
3412 | * calculate_sizes() determines the order and the distribution of data within | |
3413 | * a slab object. | |
3414 | */ | |
06b285dc | 3415 | static int calculate_sizes(struct kmem_cache *s, int forced_order) |
81819f0f CL |
3416 | { |
3417 | unsigned long flags = s->flags; | |
80a9201a | 3418 | size_t size = s->object_size; |
834f3d11 | 3419 | int order; |
81819f0f | 3420 | |
d8b42bf5 CL |
3421 | /* |
3422 | * Round up object size to the next word boundary. We can only | |
3423 | * place the free pointer at word boundaries and this determines | |
3424 | * the possible location of the free pointer. | |
3425 | */ | |
3426 | size = ALIGN(size, sizeof(void *)); | |
3427 | ||
3428 | #ifdef CONFIG_SLUB_DEBUG | |
81819f0f CL |
3429 | /* |
3430 | * Determine if we can poison the object itself. If the user of | |
3431 | * the slab may touch the object after free or before allocation | |
3432 | * then we should never poison the object itself. | |
3433 | */ | |
3434 | if ((flags & SLAB_POISON) && !(flags & SLAB_DESTROY_BY_RCU) && | |
c59def9f | 3435 | !s->ctor) |
81819f0f CL |
3436 | s->flags |= __OBJECT_POISON; |
3437 | else | |
3438 | s->flags &= ~__OBJECT_POISON; | |
3439 | ||
81819f0f CL |
3440 | |
3441 | /* | |
672bba3a | 3442 | * If we are Redzoning then check if there is some space between the |
81819f0f | 3443 | * end of the object and the free pointer. If not then add an |
672bba3a | 3444 | * additional word to have some bytes to store Redzone information. |
81819f0f | 3445 | */ |
3b0efdfa | 3446 | if ((flags & SLAB_RED_ZONE) && size == s->object_size) |
81819f0f | 3447 | size += sizeof(void *); |
41ecc55b | 3448 | #endif |
81819f0f CL |
3449 | |
3450 | /* | |
672bba3a CL |
3451 | * With that we have determined the number of bytes in actual use |
3452 | * by the object. This is the potential offset to the free pointer. | |
81819f0f CL |
3453 | */ |
3454 | s->inuse = size; | |
3455 | ||
3456 | if (((flags & (SLAB_DESTROY_BY_RCU | SLAB_POISON)) || | |
c59def9f | 3457 | s->ctor)) { |
81819f0f CL |
3458 | /* |
3459 | * Relocate free pointer after the object if it is not | |
3460 | * permitted to overwrite the first word of the object on | |
3461 | * kmem_cache_free. | |
3462 | * | |
3463 | * This is the case if we do RCU, have a constructor or | |
3464 | * destructor or are poisoning the objects. | |
3465 | */ | |
3466 | s->offset = size; | |
3467 | size += sizeof(void *); | |
3468 | } | |
3469 | ||
c12b3c62 | 3470 | #ifdef CONFIG_SLUB_DEBUG |
81819f0f CL |
3471 | if (flags & SLAB_STORE_USER) |
3472 | /* | |
3473 | * Need to store information about allocs and frees after | |
3474 | * the object. | |
3475 | */ | |
3476 | size += 2 * sizeof(struct track); | |
80a9201a | 3477 | #endif |
81819f0f | 3478 | |
80a9201a AP |
3479 | kasan_cache_create(s, &size, &s->flags); |
3480 | #ifdef CONFIG_SLUB_DEBUG | |
d86bd1be | 3481 | if (flags & SLAB_RED_ZONE) { |
81819f0f CL |
3482 | /* |
3483 | * Add some empty padding so that we can catch | |
3484 | * overwrites from earlier objects rather than let | |
3485 | * tracking information or the free pointer be | |
0211a9c8 | 3486 | * corrupted if a user writes before the start |
81819f0f CL |
3487 | * of the object. |
3488 | */ | |
3489 | size += sizeof(void *); | |
d86bd1be JK |
3490 | |
3491 | s->red_left_pad = sizeof(void *); | |
3492 | s->red_left_pad = ALIGN(s->red_left_pad, s->align); | |
3493 | size += s->red_left_pad; | |
3494 | } | |
41ecc55b | 3495 | #endif |
672bba3a | 3496 | |
81819f0f CL |
3497 | /* |
3498 | * SLUB stores one object immediately after another beginning from | |
3499 | * offset 0. In order to align the objects we have to simply size | |
3500 | * each object to conform to the alignment. | |
3501 | */ | |
45906855 | 3502 | size = ALIGN(size, s->align); |
81819f0f | 3503 | s->size = size; |
06b285dc CL |
3504 | if (forced_order >= 0) |
3505 | order = forced_order; | |
3506 | else | |
ab9a0f19 | 3507 | order = calculate_order(size, s->reserved); |
81819f0f | 3508 | |
834f3d11 | 3509 | if (order < 0) |
81819f0f CL |
3510 | return 0; |
3511 | ||
b7a49f0d | 3512 | s->allocflags = 0; |
834f3d11 | 3513 | if (order) |
b7a49f0d CL |
3514 | s->allocflags |= __GFP_COMP; |
3515 | ||
3516 | if (s->flags & SLAB_CACHE_DMA) | |
2c59dd65 | 3517 | s->allocflags |= GFP_DMA; |
b7a49f0d CL |
3518 | |
3519 | if (s->flags & SLAB_RECLAIM_ACCOUNT) | |
3520 | s->allocflags |= __GFP_RECLAIMABLE; | |
3521 | ||
81819f0f CL |
3522 | /* |
3523 | * Determine the number of objects per slab | |
3524 | */ | |
ab9a0f19 LJ |
3525 | s->oo = oo_make(order, size, s->reserved); |
3526 | s->min = oo_make(get_order(size), size, s->reserved); | |
205ab99d CL |
3527 | if (oo_objects(s->oo) > oo_objects(s->max)) |
3528 | s->max = s->oo; | |
81819f0f | 3529 | |
834f3d11 | 3530 | return !!oo_objects(s->oo); |
81819f0f CL |
3531 | } |
3532 | ||
8a13a4cc | 3533 | static int kmem_cache_open(struct kmem_cache *s, unsigned long flags) |
81819f0f | 3534 | { |
8a13a4cc | 3535 | s->flags = kmem_cache_flags(s->size, flags, s->name, s->ctor); |
ab9a0f19 | 3536 | s->reserved = 0; |
81819f0f | 3537 | |
da9a638c LJ |
3538 | if (need_reserve_slab_rcu && (s->flags & SLAB_DESTROY_BY_RCU)) |
3539 | s->reserved = sizeof(struct rcu_head); | |
81819f0f | 3540 | |
06b285dc | 3541 | if (!calculate_sizes(s, -1)) |
81819f0f | 3542 | goto error; |
3de47213 DR |
3543 | if (disable_higher_order_debug) { |
3544 | /* | |
3545 | * Disable debugging flags that store metadata if the min slab | |
3546 | * order increased. | |
3547 | */ | |
3b0efdfa | 3548 | if (get_order(s->size) > get_order(s->object_size)) { |
3de47213 DR |
3549 | s->flags &= ~DEBUG_METADATA_FLAGS; |
3550 | s->offset = 0; | |
3551 | if (!calculate_sizes(s, -1)) | |
3552 | goto error; | |
3553 | } | |
3554 | } | |
81819f0f | 3555 | |
2565409f HC |
3556 | #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \ |
3557 | defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE) | |
149daaf3 | 3558 | if (system_has_cmpxchg_double() && (s->flags & SLAB_NO_CMPXCHG) == 0) |
b789ef51 CL |
3559 | /* Enable fast mode */ |
3560 | s->flags |= __CMPXCHG_DOUBLE; | |
3561 | #endif | |
3562 | ||
3b89d7d8 DR |
3563 | /* |
3564 | * The larger the object size is, the more pages we want on the partial | |
3565 | * list to avoid pounding the page allocator excessively. | |
3566 | */ | |
49e22585 CL |
3567 | set_min_partial(s, ilog2(s->size) / 2); |
3568 | ||
3569 | /* | |
3570 | * cpu_partial determined the maximum number of objects kept in the | |
3571 | * per cpu partial lists of a processor. | |
3572 | * | |
3573 | * Per cpu partial lists mainly contain slabs that just have one | |
3574 | * object freed. If they are used for allocation then they can be | |
3575 | * filled up again with minimal effort. The slab will never hit the | |
3576 | * per node partial lists and therefore no locking will be required. | |
3577 | * | |
3578 | * This setting also determines | |
3579 | * | |
3580 | * A) The number of objects from per cpu partial slabs dumped to the | |
3581 | * per node list when we reach the limit. | |
9f264904 | 3582 | * B) The number of objects in cpu partial slabs to extract from the |
d0e0ac97 CG |
3583 | * per node list when we run out of per cpu objects. We only fetch |
3584 | * 50% to keep some capacity around for frees. | |
49e22585 | 3585 | */ |
345c905d | 3586 | if (!kmem_cache_has_cpu_partial(s)) |
8f1e33da CL |
3587 | s->cpu_partial = 0; |
3588 | else if (s->size >= PAGE_SIZE) | |
49e22585 CL |
3589 | s->cpu_partial = 2; |
3590 | else if (s->size >= 1024) | |
3591 | s->cpu_partial = 6; | |
3592 | else if (s->size >= 256) | |
3593 | s->cpu_partial = 13; | |
3594 | else | |
3595 | s->cpu_partial = 30; | |
3596 | ||
81819f0f | 3597 | #ifdef CONFIG_NUMA |
e2cb96b7 | 3598 | s->remote_node_defrag_ratio = 1000; |
81819f0f | 3599 | #endif |
210e7a43 TG |
3600 | |
3601 | /* Initialize the pre-computed randomized freelist if slab is up */ | |
3602 | if (slab_state >= UP) { | |
3603 | if (init_cache_random_seq(s)) | |
3604 | goto error; | |
3605 | } | |
3606 | ||
55136592 | 3607 | if (!init_kmem_cache_nodes(s)) |
dfb4f096 | 3608 | goto error; |
81819f0f | 3609 | |
55136592 | 3610 | if (alloc_kmem_cache_cpus(s)) |
278b1bb1 | 3611 | return 0; |
ff12059e | 3612 | |
4c93c355 | 3613 | free_kmem_cache_nodes(s); |
81819f0f CL |
3614 | error: |
3615 | if (flags & SLAB_PANIC) | |
756a025f JP |
3616 | panic("Cannot create slab %s size=%lu realsize=%u order=%u offset=%u flags=%lx\n", |
3617 | s->name, (unsigned long)s->size, s->size, | |
3618 | oo_order(s->oo), s->offset, flags); | |
278b1bb1 | 3619 | return -EINVAL; |
81819f0f | 3620 | } |
81819f0f | 3621 | |
33b12c38 CL |
3622 | static void list_slab_objects(struct kmem_cache *s, struct page *page, |
3623 | const char *text) | |
3624 | { | |
3625 | #ifdef CONFIG_SLUB_DEBUG | |
3626 | void *addr = page_address(page); | |
3627 | void *p; | |
a5dd5c11 NK |
3628 | unsigned long *map = kzalloc(BITS_TO_LONGS(page->objects) * |
3629 | sizeof(long), GFP_ATOMIC); | |
bbd7d57b ED |
3630 | if (!map) |
3631 | return; | |
945cf2b6 | 3632 | slab_err(s, page, text, s->name); |
33b12c38 | 3633 | slab_lock(page); |
33b12c38 | 3634 | |
5f80b13a | 3635 | get_map(s, page, map); |
33b12c38 CL |
3636 | for_each_object(p, s, addr, page->objects) { |
3637 | ||
3638 | if (!test_bit(slab_index(p, s, addr), map)) { | |
f9f58285 | 3639 | pr_err("INFO: Object 0x%p @offset=%tu\n", p, p - addr); |
33b12c38 CL |
3640 | print_tracking(s, p); |
3641 | } | |
3642 | } | |
3643 | slab_unlock(page); | |
bbd7d57b | 3644 | kfree(map); |
33b12c38 CL |
3645 | #endif |
3646 | } | |
3647 | ||
81819f0f | 3648 | /* |
599870b1 | 3649 | * Attempt to free all partial slabs on a node. |
52b4b950 DS |
3650 | * This is called from __kmem_cache_shutdown(). We must take list_lock |
3651 | * because sysfs file might still access partial list after the shutdowning. | |
81819f0f | 3652 | */ |
599870b1 | 3653 | static void free_partial(struct kmem_cache *s, struct kmem_cache_node *n) |
81819f0f | 3654 | { |
60398923 | 3655 | LIST_HEAD(discard); |
81819f0f CL |
3656 | struct page *page, *h; |
3657 | ||
52b4b950 DS |
3658 | BUG_ON(irqs_disabled()); |
3659 | spin_lock_irq(&n->list_lock); | |
33b12c38 | 3660 | list_for_each_entry_safe(page, h, &n->partial, lru) { |
81819f0f | 3661 | if (!page->inuse) { |
52b4b950 | 3662 | remove_partial(n, page); |
60398923 | 3663 | list_add(&page->lru, &discard); |
33b12c38 CL |
3664 | } else { |
3665 | list_slab_objects(s, page, | |
52b4b950 | 3666 | "Objects remaining in %s on __kmem_cache_shutdown()"); |
599870b1 | 3667 | } |
33b12c38 | 3668 | } |
52b4b950 | 3669 | spin_unlock_irq(&n->list_lock); |
60398923 CW |
3670 | |
3671 | list_for_each_entry_safe(page, h, &discard, lru) | |
3672 | discard_slab(s, page); | |
81819f0f CL |
3673 | } |
3674 | ||
3675 | /* | |
672bba3a | 3676 | * Release all resources used by a slab cache. |
81819f0f | 3677 | */ |
52b4b950 | 3678 | int __kmem_cache_shutdown(struct kmem_cache *s) |
81819f0f CL |
3679 | { |
3680 | int node; | |
fa45dc25 | 3681 | struct kmem_cache_node *n; |
81819f0f CL |
3682 | |
3683 | flush_all(s); | |
81819f0f | 3684 | /* Attempt to free all objects */ |
fa45dc25 | 3685 | for_each_kmem_cache_node(s, node, n) { |
599870b1 CL |
3686 | free_partial(s, n); |
3687 | if (n->nr_partial || slabs_node(s, node)) | |
81819f0f CL |
3688 | return 1; |
3689 | } | |
81819f0f CL |
3690 | return 0; |
3691 | } | |
3692 | ||
81819f0f CL |
3693 | /******************************************************************** |
3694 | * Kmalloc subsystem | |
3695 | *******************************************************************/ | |
3696 | ||
81819f0f CL |
3697 | static int __init setup_slub_min_order(char *str) |
3698 | { | |
06428780 | 3699 | get_option(&str, &slub_min_order); |
81819f0f CL |
3700 | |
3701 | return 1; | |
3702 | } | |
3703 | ||
3704 | __setup("slub_min_order=", setup_slub_min_order); | |
3705 | ||
3706 | static int __init setup_slub_max_order(char *str) | |
3707 | { | |
06428780 | 3708 | get_option(&str, &slub_max_order); |
818cf590 | 3709 | slub_max_order = min(slub_max_order, MAX_ORDER - 1); |
81819f0f CL |
3710 | |
3711 | return 1; | |
3712 | } | |
3713 | ||
3714 | __setup("slub_max_order=", setup_slub_max_order); | |
3715 | ||
3716 | static int __init setup_slub_min_objects(char *str) | |
3717 | { | |
06428780 | 3718 | get_option(&str, &slub_min_objects); |
81819f0f CL |
3719 | |
3720 | return 1; | |
3721 | } | |
3722 | ||
3723 | __setup("slub_min_objects=", setup_slub_min_objects); | |
3724 | ||
81819f0f CL |
3725 | void *__kmalloc(size_t size, gfp_t flags) |
3726 | { | |
aadb4bc4 | 3727 | struct kmem_cache *s; |
5b882be4 | 3728 | void *ret; |
81819f0f | 3729 | |
95a05b42 | 3730 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) |
eada35ef | 3731 | return kmalloc_large(size, flags); |
aadb4bc4 | 3732 | |
2c59dd65 | 3733 | s = kmalloc_slab(size, flags); |
aadb4bc4 CL |
3734 | |
3735 | if (unlikely(ZERO_OR_NULL_PTR(s))) | |
6cb8f913 CL |
3736 | return s; |
3737 | ||
2b847c3c | 3738 | ret = slab_alloc(s, flags, _RET_IP_); |
5b882be4 | 3739 | |
ca2b84cb | 3740 | trace_kmalloc(_RET_IP_, ret, size, s->size, flags); |
5b882be4 | 3741 | |
505f5dcb | 3742 | kasan_kmalloc(s, ret, size, flags); |
0316bec2 | 3743 | |
5b882be4 | 3744 | return ret; |
81819f0f CL |
3745 | } |
3746 | EXPORT_SYMBOL(__kmalloc); | |
3747 | ||
5d1f57e4 | 3748 | #ifdef CONFIG_NUMA |
f619cfe1 CL |
3749 | static void *kmalloc_large_node(size_t size, gfp_t flags, int node) |
3750 | { | |
b1eeab67 | 3751 | struct page *page; |
e4f7c0b4 | 3752 | void *ptr = NULL; |
f619cfe1 | 3753 | |
52383431 | 3754 | flags |= __GFP_COMP | __GFP_NOTRACK; |
4949148a | 3755 | page = alloc_pages_node(node, flags, get_order(size)); |
f619cfe1 | 3756 | if (page) |
e4f7c0b4 CM |
3757 | ptr = page_address(page); |
3758 | ||
d56791b3 | 3759 | kmalloc_large_node_hook(ptr, size, flags); |
e4f7c0b4 | 3760 | return ptr; |
f619cfe1 CL |
3761 | } |
3762 | ||
81819f0f CL |
3763 | void *__kmalloc_node(size_t size, gfp_t flags, int node) |
3764 | { | |
aadb4bc4 | 3765 | struct kmem_cache *s; |
5b882be4 | 3766 | void *ret; |
81819f0f | 3767 | |
95a05b42 | 3768 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) { |
5b882be4 EGM |
3769 | ret = kmalloc_large_node(size, flags, node); |
3770 | ||
ca2b84cb EGM |
3771 | trace_kmalloc_node(_RET_IP_, ret, |
3772 | size, PAGE_SIZE << get_order(size), | |
3773 | flags, node); | |
5b882be4 EGM |
3774 | |
3775 | return ret; | |
3776 | } | |
aadb4bc4 | 3777 | |
2c59dd65 | 3778 | s = kmalloc_slab(size, flags); |
aadb4bc4 CL |
3779 | |
3780 | if (unlikely(ZERO_OR_NULL_PTR(s))) | |
6cb8f913 CL |
3781 | return s; |
3782 | ||
2b847c3c | 3783 | ret = slab_alloc_node(s, flags, node, _RET_IP_); |
5b882be4 | 3784 | |
ca2b84cb | 3785 | trace_kmalloc_node(_RET_IP_, ret, size, s->size, flags, node); |
5b882be4 | 3786 | |
505f5dcb | 3787 | kasan_kmalloc(s, ret, size, flags); |
0316bec2 | 3788 | |
5b882be4 | 3789 | return ret; |
81819f0f CL |
3790 | } |
3791 | EXPORT_SYMBOL(__kmalloc_node); | |
3792 | #endif | |
3793 | ||
ed18adc1 KC |
3794 | #ifdef CONFIG_HARDENED_USERCOPY |
3795 | /* | |
3796 | * Rejects objects that are incorrectly sized. | |
3797 | * | |
3798 | * Returns NULL if check passes, otherwise const char * to name of cache | |
3799 | * to indicate an error. | |
3800 | */ | |
3801 | const char *__check_heap_object(const void *ptr, unsigned long n, | |
3802 | struct page *page) | |
3803 | { | |
3804 | struct kmem_cache *s; | |
3805 | unsigned long offset; | |
3806 | size_t object_size; | |
3807 | ||
3808 | /* Find object and usable object size. */ | |
3809 | s = page->slab_cache; | |
3810 | object_size = slab_ksize(s); | |
3811 | ||
3812 | /* Reject impossible pointers. */ | |
3813 | if (ptr < page_address(page)) | |
3814 | return s->name; | |
3815 | ||
3816 | /* Find offset within object. */ | |
3817 | offset = (ptr - page_address(page)) % s->size; | |
3818 | ||
3819 | /* Adjust for redzone and reject if within the redzone. */ | |
3820 | if (kmem_cache_debug(s) && s->flags & SLAB_RED_ZONE) { | |
3821 | if (offset < s->red_left_pad) | |
3822 | return s->name; | |
3823 | offset -= s->red_left_pad; | |
3824 | } | |
3825 | ||
3826 | /* Allow address range falling entirely within object size. */ | |
3827 | if (offset <= object_size && n <= object_size - offset) | |
3828 | return NULL; | |
3829 | ||
3830 | return s->name; | |
3831 | } | |
3832 | #endif /* CONFIG_HARDENED_USERCOPY */ | |
3833 | ||
0316bec2 | 3834 | static size_t __ksize(const void *object) |
81819f0f | 3835 | { |
272c1d21 | 3836 | struct page *page; |
81819f0f | 3837 | |
ef8b4520 | 3838 | if (unlikely(object == ZERO_SIZE_PTR)) |
272c1d21 CL |
3839 | return 0; |
3840 | ||
294a80a8 | 3841 | page = virt_to_head_page(object); |
294a80a8 | 3842 | |
76994412 PE |
3843 | if (unlikely(!PageSlab(page))) { |
3844 | WARN_ON(!PageCompound(page)); | |
294a80a8 | 3845 | return PAGE_SIZE << compound_order(page); |
76994412 | 3846 | } |
81819f0f | 3847 | |
1b4f59e3 | 3848 | return slab_ksize(page->slab_cache); |
81819f0f | 3849 | } |
0316bec2 AR |
3850 | |
3851 | size_t ksize(const void *object) | |
3852 | { | |
3853 | size_t size = __ksize(object); | |
3854 | /* We assume that ksize callers could use whole allocated area, | |
4ebb31a4 AP |
3855 | * so we need to unpoison this area. |
3856 | */ | |
3857 | kasan_unpoison_shadow(object, size); | |
0316bec2 AR |
3858 | return size; |
3859 | } | |
b1aabecd | 3860 | EXPORT_SYMBOL(ksize); |
81819f0f CL |
3861 | |
3862 | void kfree(const void *x) | |
3863 | { | |
81819f0f | 3864 | struct page *page; |
5bb983b0 | 3865 | void *object = (void *)x; |
81819f0f | 3866 | |
2121db74 PE |
3867 | trace_kfree(_RET_IP_, x); |
3868 | ||
2408c550 | 3869 | if (unlikely(ZERO_OR_NULL_PTR(x))) |
81819f0f CL |
3870 | return; |
3871 | ||
b49af68f | 3872 | page = virt_to_head_page(x); |
aadb4bc4 | 3873 | if (unlikely(!PageSlab(page))) { |
0937502a | 3874 | BUG_ON(!PageCompound(page)); |
d56791b3 | 3875 | kfree_hook(x); |
4949148a | 3876 | __free_pages(page, compound_order(page)); |
aadb4bc4 CL |
3877 | return; |
3878 | } | |
81084651 | 3879 | slab_free(page->slab_cache, page, object, NULL, 1, _RET_IP_); |
81819f0f CL |
3880 | } |
3881 | EXPORT_SYMBOL(kfree); | |
3882 | ||
832f37f5 VD |
3883 | #define SHRINK_PROMOTE_MAX 32 |
3884 | ||
2086d26a | 3885 | /* |
832f37f5 VD |
3886 | * kmem_cache_shrink discards empty slabs and promotes the slabs filled |
3887 | * up most to the head of the partial lists. New allocations will then | |
3888 | * fill those up and thus they can be removed from the partial lists. | |
672bba3a CL |
3889 | * |
3890 | * The slabs with the least items are placed last. This results in them | |
3891 | * being allocated from last increasing the chance that the last objects | |
3892 | * are freed in them. | |
2086d26a | 3893 | */ |
290b6a58 | 3894 | int __kmem_cache_shrink(struct kmem_cache *s, bool deactivate) |
2086d26a CL |
3895 | { |
3896 | int node; | |
3897 | int i; | |
3898 | struct kmem_cache_node *n; | |
3899 | struct page *page; | |
3900 | struct page *t; | |
832f37f5 VD |
3901 | struct list_head discard; |
3902 | struct list_head promote[SHRINK_PROMOTE_MAX]; | |
2086d26a | 3903 | unsigned long flags; |
ce3712d7 | 3904 | int ret = 0; |
2086d26a | 3905 | |
290b6a58 TH |
3906 | if (deactivate) { |
3907 | /* | |
3908 | * Disable empty slabs caching. Used to avoid pinning offline | |
3909 | * memory cgroups by kmem pages that can be freed. | |
3910 | */ | |
3911 | s->cpu_partial = 0; | |
3912 | s->min_partial = 0; | |
3913 | ||
3914 | /* | |
3915 | * s->cpu_partial is checked locklessly (see put_cpu_partial), | |
3916 | * so we have to make sure the change is visible. | |
3917 | */ | |
3918 | synchronize_sched(); | |
3919 | } | |
3920 | ||
2086d26a | 3921 | flush_all(s); |
fa45dc25 | 3922 | for_each_kmem_cache_node(s, node, n) { |
832f37f5 VD |
3923 | INIT_LIST_HEAD(&discard); |
3924 | for (i = 0; i < SHRINK_PROMOTE_MAX; i++) | |
3925 | INIT_LIST_HEAD(promote + i); | |
2086d26a CL |
3926 | |
3927 | spin_lock_irqsave(&n->list_lock, flags); | |
3928 | ||
3929 | /* | |
832f37f5 | 3930 | * Build lists of slabs to discard or promote. |
2086d26a | 3931 | * |
672bba3a CL |
3932 | * Note that concurrent frees may occur while we hold the |
3933 | * list_lock. page->inuse here is the upper limit. | |
2086d26a CL |
3934 | */ |
3935 | list_for_each_entry_safe(page, t, &n->partial, lru) { | |
832f37f5 VD |
3936 | int free = page->objects - page->inuse; |
3937 | ||
3938 | /* Do not reread page->inuse */ | |
3939 | barrier(); | |
3940 | ||
3941 | /* We do not keep full slabs on the list */ | |
3942 | BUG_ON(free <= 0); | |
3943 | ||
3944 | if (free == page->objects) { | |
3945 | list_move(&page->lru, &discard); | |
69cb8e6b | 3946 | n->nr_partial--; |
832f37f5 VD |
3947 | } else if (free <= SHRINK_PROMOTE_MAX) |
3948 | list_move(&page->lru, promote + free - 1); | |
2086d26a CL |
3949 | } |
3950 | ||
2086d26a | 3951 | /* |
832f37f5 VD |
3952 | * Promote the slabs filled up most to the head of the |
3953 | * partial list. | |
2086d26a | 3954 | */ |
832f37f5 VD |
3955 | for (i = SHRINK_PROMOTE_MAX - 1; i >= 0; i--) |
3956 | list_splice(promote + i, &n->partial); | |
2086d26a | 3957 | |
2086d26a | 3958 | spin_unlock_irqrestore(&n->list_lock, flags); |
69cb8e6b CL |
3959 | |
3960 | /* Release empty slabs */ | |
832f37f5 | 3961 | list_for_each_entry_safe(page, t, &discard, lru) |
69cb8e6b | 3962 | discard_slab(s, page); |
ce3712d7 VD |
3963 | |
3964 | if (slabs_node(s, node)) | |
3965 | ret = 1; | |
2086d26a CL |
3966 | } |
3967 | ||
ce3712d7 | 3968 | return ret; |
2086d26a | 3969 | } |
2086d26a | 3970 | |
b9049e23 YG |
3971 | static int slab_mem_going_offline_callback(void *arg) |
3972 | { | |
3973 | struct kmem_cache *s; | |
3974 | ||
18004c5d | 3975 | mutex_lock(&slab_mutex); |
b9049e23 | 3976 | list_for_each_entry(s, &slab_caches, list) |
290b6a58 | 3977 | __kmem_cache_shrink(s, false); |
18004c5d | 3978 | mutex_unlock(&slab_mutex); |
b9049e23 YG |
3979 | |
3980 | return 0; | |
3981 | } | |
3982 | ||
3983 | static void slab_mem_offline_callback(void *arg) | |
3984 | { | |
3985 | struct kmem_cache_node *n; | |
3986 | struct kmem_cache *s; | |
3987 | struct memory_notify *marg = arg; | |
3988 | int offline_node; | |
3989 | ||
b9d5ab25 | 3990 | offline_node = marg->status_change_nid_normal; |
b9049e23 YG |
3991 | |
3992 | /* | |
3993 | * If the node still has available memory. we need kmem_cache_node | |
3994 | * for it yet. | |
3995 | */ | |
3996 | if (offline_node < 0) | |
3997 | return; | |
3998 | ||
18004c5d | 3999 | mutex_lock(&slab_mutex); |
b9049e23 YG |
4000 | list_for_each_entry(s, &slab_caches, list) { |
4001 | n = get_node(s, offline_node); | |
4002 | if (n) { | |
4003 | /* | |
4004 | * if n->nr_slabs > 0, slabs still exist on the node | |
4005 | * that is going down. We were unable to free them, | |
c9404c9c | 4006 | * and offline_pages() function shouldn't call this |
b9049e23 YG |
4007 | * callback. So, we must fail. |
4008 | */ | |
0f389ec6 | 4009 | BUG_ON(slabs_node(s, offline_node)); |
b9049e23 YG |
4010 | |
4011 | s->node[offline_node] = NULL; | |
8de66a0c | 4012 | kmem_cache_free(kmem_cache_node, n); |
b9049e23 YG |
4013 | } |
4014 | } | |
18004c5d | 4015 | mutex_unlock(&slab_mutex); |
b9049e23 YG |
4016 | } |
4017 | ||
4018 | static int slab_mem_going_online_callback(void *arg) | |
4019 | { | |
4020 | struct kmem_cache_node *n; | |
4021 | struct kmem_cache *s; | |
4022 | struct memory_notify *marg = arg; | |
b9d5ab25 | 4023 | int nid = marg->status_change_nid_normal; |
b9049e23 YG |
4024 | int ret = 0; |
4025 | ||
4026 | /* | |
4027 | * If the node's memory is already available, then kmem_cache_node is | |
4028 | * already created. Nothing to do. | |
4029 | */ | |
4030 | if (nid < 0) | |
4031 | return 0; | |
4032 | ||
4033 | /* | |
0121c619 | 4034 | * We are bringing a node online. No memory is available yet. We must |
b9049e23 YG |
4035 | * allocate a kmem_cache_node structure in order to bring the node |
4036 | * online. | |
4037 | */ | |
18004c5d | 4038 | mutex_lock(&slab_mutex); |
b9049e23 YG |
4039 | list_for_each_entry(s, &slab_caches, list) { |
4040 | /* | |
4041 | * XXX: kmem_cache_alloc_node will fallback to other nodes | |
4042 | * since memory is not yet available from the node that | |
4043 | * is brought up. | |
4044 | */ | |
8de66a0c | 4045 | n = kmem_cache_alloc(kmem_cache_node, GFP_KERNEL); |
b9049e23 YG |
4046 | if (!n) { |
4047 | ret = -ENOMEM; | |
4048 | goto out; | |
4049 | } | |
4053497d | 4050 | init_kmem_cache_node(n); |
b9049e23 YG |
4051 | s->node[nid] = n; |
4052 | } | |
4053 | out: | |
18004c5d | 4054 | mutex_unlock(&slab_mutex); |
b9049e23 YG |
4055 | return ret; |
4056 | } | |
4057 | ||
4058 | static int slab_memory_callback(struct notifier_block *self, | |
4059 | unsigned long action, void *arg) | |
4060 | { | |
4061 | int ret = 0; | |
4062 | ||
4063 | switch (action) { | |
4064 | case MEM_GOING_ONLINE: | |
4065 | ret = slab_mem_going_online_callback(arg); | |
4066 | break; | |
4067 | case MEM_GOING_OFFLINE: | |
4068 | ret = slab_mem_going_offline_callback(arg); | |
4069 | break; | |
4070 | case MEM_OFFLINE: | |
4071 | case MEM_CANCEL_ONLINE: | |
4072 | slab_mem_offline_callback(arg); | |
4073 | break; | |
4074 | case MEM_ONLINE: | |
4075 | case MEM_CANCEL_OFFLINE: | |
4076 | break; | |
4077 | } | |
dc19f9db KH |
4078 | if (ret) |
4079 | ret = notifier_from_errno(ret); | |
4080 | else | |
4081 | ret = NOTIFY_OK; | |
b9049e23 YG |
4082 | return ret; |
4083 | } | |
4084 | ||
3ac38faa AM |
4085 | static struct notifier_block slab_memory_callback_nb = { |
4086 | .notifier_call = slab_memory_callback, | |
4087 | .priority = SLAB_CALLBACK_PRI, | |
4088 | }; | |
b9049e23 | 4089 | |
81819f0f CL |
4090 | /******************************************************************** |
4091 | * Basic setup of slabs | |
4092 | *******************************************************************/ | |
4093 | ||
51df1142 CL |
4094 | /* |
4095 | * Used for early kmem_cache structures that were allocated using | |
dffb4d60 CL |
4096 | * the page allocator. Allocate them properly then fix up the pointers |
4097 | * that may be pointing to the wrong kmem_cache structure. | |
51df1142 CL |
4098 | */ |
4099 | ||
dffb4d60 | 4100 | static struct kmem_cache * __init bootstrap(struct kmem_cache *static_cache) |
51df1142 CL |
4101 | { |
4102 | int node; | |
dffb4d60 | 4103 | struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT); |
fa45dc25 | 4104 | struct kmem_cache_node *n; |
51df1142 | 4105 | |
dffb4d60 | 4106 | memcpy(s, static_cache, kmem_cache->object_size); |
51df1142 | 4107 | |
7d557b3c GC |
4108 | /* |
4109 | * This runs very early, and only the boot processor is supposed to be | |
4110 | * up. Even if it weren't true, IRQs are not up so we couldn't fire | |
4111 | * IPIs around. | |
4112 | */ | |
4113 | __flush_cpu_slab(s, smp_processor_id()); | |
fa45dc25 | 4114 | for_each_kmem_cache_node(s, node, n) { |
51df1142 CL |
4115 | struct page *p; |
4116 | ||
fa45dc25 CL |
4117 | list_for_each_entry(p, &n->partial, lru) |
4118 | p->slab_cache = s; | |
51df1142 | 4119 | |
607bf324 | 4120 | #ifdef CONFIG_SLUB_DEBUG |
fa45dc25 CL |
4121 | list_for_each_entry(p, &n->full, lru) |
4122 | p->slab_cache = s; | |
51df1142 | 4123 | #endif |
51df1142 | 4124 | } |
f7ce3190 | 4125 | slab_init_memcg_params(s); |
dffb4d60 CL |
4126 | list_add(&s->list, &slab_caches); |
4127 | return s; | |
51df1142 CL |
4128 | } |
4129 | ||
81819f0f CL |
4130 | void __init kmem_cache_init(void) |
4131 | { | |
dffb4d60 CL |
4132 | static __initdata struct kmem_cache boot_kmem_cache, |
4133 | boot_kmem_cache_node; | |
51df1142 | 4134 | |
fc8d8620 SG |
4135 | if (debug_guardpage_minorder()) |
4136 | slub_max_order = 0; | |
4137 | ||
dffb4d60 CL |
4138 | kmem_cache_node = &boot_kmem_cache_node; |
4139 | kmem_cache = &boot_kmem_cache; | |
51df1142 | 4140 | |
dffb4d60 CL |
4141 | create_boot_cache(kmem_cache_node, "kmem_cache_node", |
4142 | sizeof(struct kmem_cache_node), SLAB_HWCACHE_ALIGN); | |
b9049e23 | 4143 | |
3ac38faa | 4144 | register_hotmemory_notifier(&slab_memory_callback_nb); |
81819f0f CL |
4145 | |
4146 | /* Able to allocate the per node structures */ | |
4147 | slab_state = PARTIAL; | |
4148 | ||
dffb4d60 CL |
4149 | create_boot_cache(kmem_cache, "kmem_cache", |
4150 | offsetof(struct kmem_cache, node) + | |
4151 | nr_node_ids * sizeof(struct kmem_cache_node *), | |
4152 | SLAB_HWCACHE_ALIGN); | |
8a13a4cc | 4153 | |
dffb4d60 | 4154 | kmem_cache = bootstrap(&boot_kmem_cache); |
81819f0f | 4155 | |
51df1142 CL |
4156 | /* |
4157 | * Allocate kmem_cache_node properly from the kmem_cache slab. | |
4158 | * kmem_cache_node is separately allocated so no need to | |
4159 | * update any list pointers. | |
4160 | */ | |
dffb4d60 | 4161 | kmem_cache_node = bootstrap(&boot_kmem_cache_node); |
51df1142 CL |
4162 | |
4163 | /* Now we can use the kmem_cache to allocate kmalloc slabs */ | |
34cc6990 | 4164 | setup_kmalloc_cache_index_table(); |
f97d5f63 | 4165 | create_kmalloc_caches(0); |
81819f0f | 4166 | |
210e7a43 TG |
4167 | /* Setup random freelists for each cache */ |
4168 | init_freelist_randomization(); | |
4169 | ||
a96a87bf SAS |
4170 | cpuhp_setup_state_nocalls(CPUHP_SLUB_DEAD, "slub:dead", NULL, |
4171 | slub_cpu_dead); | |
81819f0f | 4172 | |
f9f58285 | 4173 | pr_info("SLUB: HWalign=%d, Order=%d-%d, MinObjects=%d, CPUs=%d, Nodes=%d\n", |
f97d5f63 | 4174 | cache_line_size(), |
81819f0f CL |
4175 | slub_min_order, slub_max_order, slub_min_objects, |
4176 | nr_cpu_ids, nr_node_ids); | |
4177 | } | |
4178 | ||
7e85ee0c PE |
4179 | void __init kmem_cache_init_late(void) |
4180 | { | |
7e85ee0c PE |
4181 | } |
4182 | ||
2633d7a0 | 4183 | struct kmem_cache * |
a44cb944 VD |
4184 | __kmem_cache_alias(const char *name, size_t size, size_t align, |
4185 | unsigned long flags, void (*ctor)(void *)) | |
81819f0f | 4186 | { |
426589f5 | 4187 | struct kmem_cache *s, *c; |
81819f0f | 4188 | |
a44cb944 | 4189 | s = find_mergeable(size, align, flags, name, ctor); |
81819f0f CL |
4190 | if (s) { |
4191 | s->refcount++; | |
84d0ddd6 | 4192 | |
81819f0f CL |
4193 | /* |
4194 | * Adjust the object sizes so that we clear | |
4195 | * the complete object on kzalloc. | |
4196 | */ | |
3b0efdfa | 4197 | s->object_size = max(s->object_size, (int)size); |
81819f0f | 4198 | s->inuse = max_t(int, s->inuse, ALIGN(size, sizeof(void *))); |
6446faa2 | 4199 | |
426589f5 | 4200 | for_each_memcg_cache(c, s) { |
84d0ddd6 VD |
4201 | c->object_size = s->object_size; |
4202 | c->inuse = max_t(int, c->inuse, | |
4203 | ALIGN(size, sizeof(void *))); | |
4204 | } | |
4205 | ||
7b8f3b66 | 4206 | if (sysfs_slab_alias(s, name)) { |
7b8f3b66 | 4207 | s->refcount--; |
cbb79694 | 4208 | s = NULL; |
7b8f3b66 | 4209 | } |
a0e1d1be | 4210 | } |
6446faa2 | 4211 | |
cbb79694 CL |
4212 | return s; |
4213 | } | |
84c1cf62 | 4214 | |
8a13a4cc | 4215 | int __kmem_cache_create(struct kmem_cache *s, unsigned long flags) |
cbb79694 | 4216 | { |
aac3a166 PE |
4217 | int err; |
4218 | ||
4219 | err = kmem_cache_open(s, flags); | |
4220 | if (err) | |
4221 | return err; | |
20cea968 | 4222 | |
45530c44 CL |
4223 | /* Mutex is not taken during early boot */ |
4224 | if (slab_state <= UP) | |
4225 | return 0; | |
4226 | ||
107dab5c | 4227 | memcg_propagate_slab_attrs(s); |
aac3a166 | 4228 | err = sysfs_slab_add(s); |
aac3a166 | 4229 | if (err) |
52b4b950 | 4230 | __kmem_cache_release(s); |
20cea968 | 4231 | |
aac3a166 | 4232 | return err; |
81819f0f | 4233 | } |
81819f0f | 4234 | |
ce71e27c | 4235 | void *__kmalloc_track_caller(size_t size, gfp_t gfpflags, unsigned long caller) |
81819f0f | 4236 | { |
aadb4bc4 | 4237 | struct kmem_cache *s; |
94b528d0 | 4238 | void *ret; |
aadb4bc4 | 4239 | |
95a05b42 | 4240 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) |
eada35ef PE |
4241 | return kmalloc_large(size, gfpflags); |
4242 | ||
2c59dd65 | 4243 | s = kmalloc_slab(size, gfpflags); |
81819f0f | 4244 | |
2408c550 | 4245 | if (unlikely(ZERO_OR_NULL_PTR(s))) |
6cb8f913 | 4246 | return s; |
81819f0f | 4247 | |
2b847c3c | 4248 | ret = slab_alloc(s, gfpflags, caller); |
94b528d0 | 4249 | |
25985edc | 4250 | /* Honor the call site pointer we received. */ |
ca2b84cb | 4251 | trace_kmalloc(caller, ret, size, s->size, gfpflags); |
94b528d0 EGM |
4252 | |
4253 | return ret; | |
81819f0f CL |
4254 | } |
4255 | ||
5d1f57e4 | 4256 | #ifdef CONFIG_NUMA |
81819f0f | 4257 | void *__kmalloc_node_track_caller(size_t size, gfp_t gfpflags, |
ce71e27c | 4258 | int node, unsigned long caller) |
81819f0f | 4259 | { |
aadb4bc4 | 4260 | struct kmem_cache *s; |
94b528d0 | 4261 | void *ret; |
aadb4bc4 | 4262 | |
95a05b42 | 4263 | if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) { |
d3e14aa3 XF |
4264 | ret = kmalloc_large_node(size, gfpflags, node); |
4265 | ||
4266 | trace_kmalloc_node(caller, ret, | |
4267 | size, PAGE_SIZE << get_order(size), | |
4268 | gfpflags, node); | |
4269 | ||
4270 | return ret; | |
4271 | } | |
eada35ef | 4272 | |
2c59dd65 | 4273 | s = kmalloc_slab(size, gfpflags); |
81819f0f | 4274 | |
2408c550 | 4275 | if (unlikely(ZERO_OR_NULL_PTR(s))) |
6cb8f913 | 4276 | return s; |
81819f0f | 4277 | |
2b847c3c | 4278 | ret = slab_alloc_node(s, gfpflags, node, caller); |
94b528d0 | 4279 | |
25985edc | 4280 | /* Honor the call site pointer we received. */ |
ca2b84cb | 4281 | trace_kmalloc_node(caller, ret, size, s->size, gfpflags, node); |
94b528d0 EGM |
4282 | |
4283 | return ret; | |
81819f0f | 4284 | } |
5d1f57e4 | 4285 | #endif |
81819f0f | 4286 | |
ab4d5ed5 | 4287 | #ifdef CONFIG_SYSFS |
205ab99d CL |
4288 | static int count_inuse(struct page *page) |
4289 | { | |
4290 | return page->inuse; | |
4291 | } | |
4292 | ||
4293 | static int count_total(struct page *page) | |
4294 | { | |
4295 | return page->objects; | |
4296 | } | |
ab4d5ed5 | 4297 | #endif |
205ab99d | 4298 | |
ab4d5ed5 | 4299 | #ifdef CONFIG_SLUB_DEBUG |
434e245d CL |
4300 | static int validate_slab(struct kmem_cache *s, struct page *page, |
4301 | unsigned long *map) | |
53e15af0 CL |
4302 | { |
4303 | void *p; | |
a973e9dd | 4304 | void *addr = page_address(page); |
53e15af0 CL |
4305 | |
4306 | if (!check_slab(s, page) || | |
4307 | !on_freelist(s, page, NULL)) | |
4308 | return 0; | |
4309 | ||
4310 | /* Now we know that a valid freelist exists */ | |
39b26464 | 4311 | bitmap_zero(map, page->objects); |
53e15af0 | 4312 | |
5f80b13a CL |
4313 | get_map(s, page, map); |
4314 | for_each_object(p, s, addr, page->objects) { | |
4315 | if (test_bit(slab_index(p, s, addr), map)) | |
4316 | if (!check_object(s, page, p, SLUB_RED_INACTIVE)) | |
4317 | return 0; | |
53e15af0 CL |
4318 | } |
4319 | ||
224a88be | 4320 | for_each_object(p, s, addr, page->objects) |
7656c72b | 4321 | if (!test_bit(slab_index(p, s, addr), map)) |
37d57443 | 4322 | if (!check_object(s, page, p, SLUB_RED_ACTIVE)) |
53e15af0 CL |
4323 | return 0; |
4324 | return 1; | |
4325 | } | |
4326 | ||
434e245d CL |
4327 | static void validate_slab_slab(struct kmem_cache *s, struct page *page, |
4328 | unsigned long *map) | |
53e15af0 | 4329 | { |
881db7fb CL |
4330 | slab_lock(page); |
4331 | validate_slab(s, page, map); | |
4332 | slab_unlock(page); | |
53e15af0 CL |
4333 | } |
4334 | ||
434e245d CL |
4335 | static int validate_slab_node(struct kmem_cache *s, |
4336 | struct kmem_cache_node *n, unsigned long *map) | |
53e15af0 CL |
4337 | { |
4338 | unsigned long count = 0; | |
4339 | struct page *page; | |
4340 | unsigned long flags; | |
4341 | ||
4342 | spin_lock_irqsave(&n->list_lock, flags); | |
4343 | ||
4344 | list_for_each_entry(page, &n->partial, lru) { | |
434e245d | 4345 | validate_slab_slab(s, page, map); |
53e15af0 CL |
4346 | count++; |
4347 | } | |
4348 | if (count != n->nr_partial) | |
f9f58285 FF |
4349 | pr_err("SLUB %s: %ld partial slabs counted but counter=%ld\n", |
4350 | s->name, count, n->nr_partial); | |
53e15af0 CL |
4351 | |
4352 | if (!(s->flags & SLAB_STORE_USER)) | |
4353 | goto out; | |
4354 | ||
4355 | list_for_each_entry(page, &n->full, lru) { | |
434e245d | 4356 | validate_slab_slab(s, page, map); |
53e15af0 CL |
4357 | count++; |
4358 | } | |
4359 | if (count != atomic_long_read(&n->nr_slabs)) | |
f9f58285 FF |
4360 | pr_err("SLUB: %s %ld slabs counted but counter=%ld\n", |
4361 | s->name, count, atomic_long_read(&n->nr_slabs)); | |
53e15af0 CL |
4362 | |
4363 | out: | |
4364 | spin_unlock_irqrestore(&n->list_lock, flags); | |
4365 | return count; | |
4366 | } | |
4367 | ||
434e245d | 4368 | static long validate_slab_cache(struct kmem_cache *s) |
53e15af0 CL |
4369 | { |
4370 | int node; | |
4371 | unsigned long count = 0; | |
205ab99d | 4372 | unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) * |
434e245d | 4373 | sizeof(unsigned long), GFP_KERNEL); |
fa45dc25 | 4374 | struct kmem_cache_node *n; |
434e245d CL |
4375 | |
4376 | if (!map) | |
4377 | return -ENOMEM; | |
53e15af0 CL |
4378 | |
4379 | flush_all(s); | |
fa45dc25 | 4380 | for_each_kmem_cache_node(s, node, n) |
434e245d | 4381 | count += validate_slab_node(s, n, map); |
434e245d | 4382 | kfree(map); |
53e15af0 CL |
4383 | return count; |
4384 | } | |
88a420e4 | 4385 | /* |
672bba3a | 4386 | * Generate lists of code addresses where slabcache objects are allocated |
88a420e4 CL |
4387 | * and freed. |
4388 | */ | |
4389 | ||
4390 | struct location { | |
4391 | unsigned long count; | |
ce71e27c | 4392 | unsigned long addr; |
45edfa58 CL |
4393 | long long sum_time; |
4394 | long min_time; | |
4395 | long max_time; | |
4396 | long min_pid; | |
4397 | long max_pid; | |
174596a0 | 4398 | DECLARE_BITMAP(cpus, NR_CPUS); |
45edfa58 | 4399 | nodemask_t nodes; |
88a420e4 CL |
4400 | }; |
4401 | ||
4402 | struct loc_track { | |
4403 | unsigned long max; | |
4404 | unsigned long count; | |
4405 | struct location *loc; | |
4406 | }; | |
4407 | ||
4408 | static void free_loc_track(struct loc_track *t) | |
4409 | { | |
4410 | if (t->max) | |
4411 | free_pages((unsigned long)t->loc, | |
4412 | get_order(sizeof(struct location) * t->max)); | |
4413 | } | |
4414 | ||
68dff6a9 | 4415 | static int alloc_loc_track(struct loc_track *t, unsigned long max, gfp_t flags) |
88a420e4 CL |
4416 | { |
4417 | struct location *l; | |
4418 | int order; | |
4419 | ||
88a420e4 CL |
4420 | order = get_order(sizeof(struct location) * max); |
4421 | ||
68dff6a9 | 4422 | l = (void *)__get_free_pages(flags, order); |
88a420e4 CL |
4423 | if (!l) |
4424 | return 0; | |
4425 | ||
4426 | if (t->count) { | |
4427 | memcpy(l, t->loc, sizeof(struct location) * t->count); | |
4428 | free_loc_track(t); | |
4429 | } | |
4430 | t->max = max; | |
4431 | t->loc = l; | |
4432 | return 1; | |
4433 | } | |
4434 | ||
4435 | static int add_location(struct loc_track *t, struct kmem_cache *s, | |
45edfa58 | 4436 | const struct track *track) |
88a420e4 CL |
4437 | { |
4438 | long start, end, pos; | |
4439 | struct location *l; | |
ce71e27c | 4440 | unsigned long caddr; |
45edfa58 | 4441 | unsigned long age = jiffies - track->when; |
88a420e4 CL |
4442 | |
4443 | start = -1; | |
4444 | end = t->count; | |
4445 | ||
4446 | for ( ; ; ) { | |
4447 | pos = start + (end - start + 1) / 2; | |
4448 | ||
4449 | /* | |
4450 | * There is nothing at "end". If we end up there | |
4451 | * we need to add something to before end. | |
4452 | */ | |
4453 | if (pos == end) | |
4454 | break; | |
4455 | ||
4456 | caddr = t->loc[pos].addr; | |
45edfa58 CL |
4457 | if (track->addr == caddr) { |
4458 | ||
4459 | l = &t->loc[pos]; | |
4460 | l->count++; | |
4461 | if (track->when) { | |
4462 | l->sum_time += age; | |
4463 | if (age < l->min_time) | |
4464 | l->min_time = age; | |
4465 | if (age > l->max_time) | |
4466 | l->max_time = age; | |
4467 | ||
4468 | if (track->pid < l->min_pid) | |
4469 | l->min_pid = track->pid; | |
4470 | if (track->pid > l->max_pid) | |
4471 | l->max_pid = track->pid; | |
4472 | ||
174596a0 RR |
4473 | cpumask_set_cpu(track->cpu, |
4474 | to_cpumask(l->cpus)); | |
45edfa58 CL |
4475 | } |
4476 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
88a420e4 CL |
4477 | return 1; |
4478 | } | |
4479 | ||
45edfa58 | 4480 | if (track->addr < caddr) |
88a420e4 CL |
4481 | end = pos; |
4482 | else | |
4483 | start = pos; | |
4484 | } | |
4485 | ||
4486 | /* | |
672bba3a | 4487 | * Not found. Insert new tracking element. |
88a420e4 | 4488 | */ |
68dff6a9 | 4489 | if (t->count >= t->max && !alloc_loc_track(t, 2 * t->max, GFP_ATOMIC)) |
88a420e4 CL |
4490 | return 0; |
4491 | ||
4492 | l = t->loc + pos; | |
4493 | if (pos < t->count) | |
4494 | memmove(l + 1, l, | |
4495 | (t->count - pos) * sizeof(struct location)); | |
4496 | t->count++; | |
4497 | l->count = 1; | |
45edfa58 CL |
4498 | l->addr = track->addr; |
4499 | l->sum_time = age; | |
4500 | l->min_time = age; | |
4501 | l->max_time = age; | |
4502 | l->min_pid = track->pid; | |
4503 | l->max_pid = track->pid; | |
174596a0 RR |
4504 | cpumask_clear(to_cpumask(l->cpus)); |
4505 | cpumask_set_cpu(track->cpu, to_cpumask(l->cpus)); | |
45edfa58 CL |
4506 | nodes_clear(l->nodes); |
4507 | node_set(page_to_nid(virt_to_page(track)), l->nodes); | |
88a420e4 CL |
4508 | return 1; |
4509 | } | |
4510 | ||
4511 | static void process_slab(struct loc_track *t, struct kmem_cache *s, | |
bbd7d57b | 4512 | struct page *page, enum track_item alloc, |
a5dd5c11 | 4513 | unsigned long *map) |
88a420e4 | 4514 | { |
a973e9dd | 4515 | void *addr = page_address(page); |
88a420e4 CL |
4516 | void *p; |
4517 | ||
39b26464 | 4518 | bitmap_zero(map, page->objects); |
5f80b13a | 4519 | get_map(s, page, map); |
88a420e4 | 4520 | |
224a88be | 4521 | for_each_object(p, s, addr, page->objects) |
45edfa58 CL |
4522 | if (!test_bit(slab_index(p, s, addr), map)) |
4523 | add_location(t, s, get_track(s, p, alloc)); | |
88a420e4 CL |
4524 | } |
4525 | ||
4526 | static int list_locations(struct kmem_cache *s, char *buf, | |
4527 | enum track_item alloc) | |
4528 | { | |
e374d483 | 4529 | int len = 0; |
88a420e4 | 4530 | unsigned long i; |
68dff6a9 | 4531 | struct loc_track t = { 0, 0, NULL }; |
88a420e4 | 4532 | int node; |
bbd7d57b ED |
4533 | unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) * |
4534 | sizeof(unsigned long), GFP_KERNEL); | |
fa45dc25 | 4535 | struct kmem_cache_node *n; |
88a420e4 | 4536 | |
bbd7d57b ED |
4537 | if (!map || !alloc_loc_track(&t, PAGE_SIZE / sizeof(struct location), |
4538 | GFP_TEMPORARY)) { | |
4539 | kfree(map); | |
68dff6a9 | 4540 | return sprintf(buf, "Out of memory\n"); |
bbd7d57b | 4541 | } |
88a420e4 CL |
4542 | /* Push back cpu slabs */ |
4543 | flush_all(s); | |
4544 | ||
fa45dc25 | 4545 | for_each_kmem_cache_node(s, node, n) { |
88a420e4 CL |
4546 | unsigned long flags; |
4547 | struct page *page; | |
4548 | ||
9e86943b | 4549 | if (!atomic_long_read(&n->nr_slabs)) |
88a420e4 CL |
4550 | continue; |
4551 | ||
4552 | spin_lock_irqsave(&n->list_lock, flags); | |
4553 | list_for_each_entry(page, &n->partial, lru) | |
bbd7d57b | 4554 | process_slab(&t, s, page, alloc, map); |
88a420e4 | 4555 | list_for_each_entry(page, &n->full, lru) |
bbd7d57b | 4556 | process_slab(&t, s, page, alloc, map); |
88a420e4 CL |
4557 | spin_unlock_irqrestore(&n->list_lock, flags); |
4558 | } | |
4559 | ||
4560 | for (i = 0; i < t.count; i++) { | |
45edfa58 | 4561 | struct location *l = &t.loc[i]; |
88a420e4 | 4562 | |
9c246247 | 4563 | if (len > PAGE_SIZE - KSYM_SYMBOL_LEN - 100) |
88a420e4 | 4564 | break; |
e374d483 | 4565 | len += sprintf(buf + len, "%7ld ", l->count); |
45edfa58 CL |
4566 | |
4567 | if (l->addr) | |
62c70bce | 4568 | len += sprintf(buf + len, "%pS", (void *)l->addr); |
88a420e4 | 4569 | else |
e374d483 | 4570 | len += sprintf(buf + len, "<not-available>"); |
45edfa58 CL |
4571 | |
4572 | if (l->sum_time != l->min_time) { | |
e374d483 | 4573 | len += sprintf(buf + len, " age=%ld/%ld/%ld", |
f8bd2258 RZ |
4574 | l->min_time, |
4575 | (long)div_u64(l->sum_time, l->count), | |
4576 | l->max_time); | |
45edfa58 | 4577 | } else |
e374d483 | 4578 | len += sprintf(buf + len, " age=%ld", |
45edfa58 CL |
4579 | l->min_time); |
4580 | ||
4581 | if (l->min_pid != l->max_pid) | |
e374d483 | 4582 | len += sprintf(buf + len, " pid=%ld-%ld", |
45edfa58 CL |
4583 | l->min_pid, l->max_pid); |
4584 | else | |
e374d483 | 4585 | len += sprintf(buf + len, " pid=%ld", |
45edfa58 CL |
4586 | l->min_pid); |
4587 | ||
174596a0 RR |
4588 | if (num_online_cpus() > 1 && |
4589 | !cpumask_empty(to_cpumask(l->cpus)) && | |
5024c1d7 TH |
4590 | len < PAGE_SIZE - 60) |
4591 | len += scnprintf(buf + len, PAGE_SIZE - len - 50, | |
4592 | " cpus=%*pbl", | |
4593 | cpumask_pr_args(to_cpumask(l->cpus))); | |
45edfa58 | 4594 | |
62bc62a8 | 4595 | if (nr_online_nodes > 1 && !nodes_empty(l->nodes) && |
5024c1d7 TH |
4596 | len < PAGE_SIZE - 60) |
4597 | len += scnprintf(buf + len, PAGE_SIZE - len - 50, | |
4598 | " nodes=%*pbl", | |
4599 | nodemask_pr_args(&l->nodes)); | |
45edfa58 | 4600 | |
e374d483 | 4601 | len += sprintf(buf + len, "\n"); |
88a420e4 CL |
4602 | } |
4603 | ||
4604 | free_loc_track(&t); | |
bbd7d57b | 4605 | kfree(map); |
88a420e4 | 4606 | if (!t.count) |
e374d483 HH |
4607 | len += sprintf(buf, "No data\n"); |
4608 | return len; | |
88a420e4 | 4609 | } |
ab4d5ed5 | 4610 | #endif |
88a420e4 | 4611 | |
a5a84755 | 4612 | #ifdef SLUB_RESILIENCY_TEST |
c07b8183 | 4613 | static void __init resiliency_test(void) |
a5a84755 CL |
4614 | { |
4615 | u8 *p; | |
4616 | ||
95a05b42 | 4617 | BUILD_BUG_ON(KMALLOC_MIN_SIZE > 16 || KMALLOC_SHIFT_HIGH < 10); |
a5a84755 | 4618 | |
f9f58285 FF |
4619 | pr_err("SLUB resiliency testing\n"); |
4620 | pr_err("-----------------------\n"); | |
4621 | pr_err("A. Corruption after allocation\n"); | |
a5a84755 CL |
4622 | |
4623 | p = kzalloc(16, GFP_KERNEL); | |
4624 | p[16] = 0x12; | |
f9f58285 FF |
4625 | pr_err("\n1. kmalloc-16: Clobber Redzone/next pointer 0x12->0x%p\n\n", |
4626 | p + 16); | |
a5a84755 CL |
4627 | |
4628 | validate_slab_cache(kmalloc_caches[4]); | |
4629 | ||
4630 | /* Hmmm... The next two are dangerous */ | |
4631 | p = kzalloc(32, GFP_KERNEL); | |
4632 | p[32 + sizeof(void *)] = 0x34; | |
f9f58285 FF |
4633 | pr_err("\n2. kmalloc-32: Clobber next pointer/next slab 0x34 -> -0x%p\n", |
4634 | p); | |
4635 | pr_err("If allocated object is overwritten then not detectable\n\n"); | |
a5a84755 CL |
4636 | |
4637 | validate_slab_cache(kmalloc_caches[5]); | |
4638 | p = kzalloc(64, GFP_KERNEL); | |
4639 | p += 64 + (get_cycles() & 0xff) * sizeof(void *); | |
4640 | *p = 0x56; | |
f9f58285 FF |
4641 | pr_err("\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n", |
4642 | p); | |
4643 | pr_err("If allocated object is overwritten then not detectable\n\n"); | |
a5a84755 CL |
4644 | validate_slab_cache(kmalloc_caches[6]); |
4645 | ||
f9f58285 | 4646 | pr_err("\nB. Corruption after free\n"); |
a5a84755 CL |
4647 | p = kzalloc(128, GFP_KERNEL); |
4648 | kfree(p); | |
4649 | *p = 0x78; | |
f9f58285 | 4650 | pr_err("1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p); |
a5a84755 CL |
4651 | validate_slab_cache(kmalloc_caches[7]); |
4652 | ||
4653 | p = kzalloc(256, GFP_KERNEL); | |
4654 | kfree(p); | |
4655 | p[50] = 0x9a; | |
f9f58285 | 4656 | pr_err("\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n", p); |
a5a84755 CL |
4657 | validate_slab_cache(kmalloc_caches[8]); |
4658 | ||
4659 | p = kzalloc(512, GFP_KERNEL); | |
4660 | kfree(p); | |
4661 | p[512] = 0xab; | |
f9f58285 | 4662 | pr_err("\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p); |
a5a84755 CL |
4663 | validate_slab_cache(kmalloc_caches[9]); |
4664 | } | |
4665 | #else | |
4666 | #ifdef CONFIG_SYSFS | |
4667 | static void resiliency_test(void) {}; | |
4668 | #endif | |
4669 | #endif | |
4670 | ||
ab4d5ed5 | 4671 | #ifdef CONFIG_SYSFS |
81819f0f | 4672 | enum slab_stat_type { |
205ab99d CL |
4673 | SL_ALL, /* All slabs */ |
4674 | SL_PARTIAL, /* Only partially allocated slabs */ | |
4675 | SL_CPU, /* Only slabs used for cpu caches */ | |
4676 | SL_OBJECTS, /* Determine allocated objects not slabs */ | |
4677 | SL_TOTAL /* Determine object capacity not slabs */ | |
81819f0f CL |
4678 | }; |
4679 | ||
205ab99d | 4680 | #define SO_ALL (1 << SL_ALL) |
81819f0f CL |
4681 | #define SO_PARTIAL (1 << SL_PARTIAL) |
4682 | #define SO_CPU (1 << SL_CPU) | |
4683 | #define SO_OBJECTS (1 << SL_OBJECTS) | |
205ab99d | 4684 | #define SO_TOTAL (1 << SL_TOTAL) |
81819f0f | 4685 | |
62e5c4b4 CG |
4686 | static ssize_t show_slab_objects(struct kmem_cache *s, |
4687 | char *buf, unsigned long flags) | |
81819f0f CL |
4688 | { |
4689 | unsigned long total = 0; | |
81819f0f CL |
4690 | int node; |
4691 | int x; | |
4692 | unsigned long *nodes; | |
81819f0f | 4693 | |
e35e1a97 | 4694 | nodes = kzalloc(sizeof(unsigned long) * nr_node_ids, GFP_KERNEL); |
62e5c4b4 CG |
4695 | if (!nodes) |
4696 | return -ENOMEM; | |
81819f0f | 4697 | |
205ab99d CL |
4698 | if (flags & SO_CPU) { |
4699 | int cpu; | |
81819f0f | 4700 | |
205ab99d | 4701 | for_each_possible_cpu(cpu) { |
d0e0ac97 CG |
4702 | struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, |
4703 | cpu); | |
ec3ab083 | 4704 | int node; |
49e22585 | 4705 | struct page *page; |
dfb4f096 | 4706 | |
4db0c3c2 | 4707 | page = READ_ONCE(c->page); |
ec3ab083 CL |
4708 | if (!page) |
4709 | continue; | |
205ab99d | 4710 | |
ec3ab083 CL |
4711 | node = page_to_nid(page); |
4712 | if (flags & SO_TOTAL) | |
4713 | x = page->objects; | |
4714 | else if (flags & SO_OBJECTS) | |
4715 | x = page->inuse; | |
4716 | else | |
4717 | x = 1; | |
49e22585 | 4718 | |
ec3ab083 CL |
4719 | total += x; |
4720 | nodes[node] += x; | |
4721 | ||
4db0c3c2 | 4722 | page = READ_ONCE(c->partial); |
49e22585 | 4723 | if (page) { |
8afb1474 LZ |
4724 | node = page_to_nid(page); |
4725 | if (flags & SO_TOTAL) | |
4726 | WARN_ON_ONCE(1); | |
4727 | else if (flags & SO_OBJECTS) | |
4728 | WARN_ON_ONCE(1); | |
4729 | else | |
4730 | x = page->pages; | |
bc6697d8 ED |
4731 | total += x; |
4732 | nodes[node] += x; | |
49e22585 | 4733 | } |
81819f0f CL |
4734 | } |
4735 | } | |
4736 | ||
bfc8c901 | 4737 | get_online_mems(); |
ab4d5ed5 | 4738 | #ifdef CONFIG_SLUB_DEBUG |
205ab99d | 4739 | if (flags & SO_ALL) { |
fa45dc25 CL |
4740 | struct kmem_cache_node *n; |
4741 | ||
4742 | for_each_kmem_cache_node(s, node, n) { | |
205ab99d | 4743 | |
d0e0ac97 CG |
4744 | if (flags & SO_TOTAL) |
4745 | x = atomic_long_read(&n->total_objects); | |
4746 | else if (flags & SO_OBJECTS) | |
4747 | x = atomic_long_read(&n->total_objects) - | |
4748 | count_partial(n, count_free); | |
81819f0f | 4749 | else |
205ab99d | 4750 | x = atomic_long_read(&n->nr_slabs); |
81819f0f CL |
4751 | total += x; |
4752 | nodes[node] += x; | |
4753 | } | |
4754 | ||
ab4d5ed5 CL |
4755 | } else |
4756 | #endif | |
4757 | if (flags & SO_PARTIAL) { | |
fa45dc25 | 4758 | struct kmem_cache_node *n; |
81819f0f | 4759 | |
fa45dc25 | 4760 | for_each_kmem_cache_node(s, node, n) { |
205ab99d CL |
4761 | if (flags & SO_TOTAL) |
4762 | x = count_partial(n, count_total); | |
4763 | else if (flags & SO_OBJECTS) | |
4764 | x = count_partial(n, count_inuse); | |
81819f0f | 4765 | else |
205ab99d | 4766 | x = n->nr_partial; |
81819f0f CL |
4767 | total += x; |
4768 | nodes[node] += x; | |
4769 | } | |
4770 | } | |
81819f0f CL |
4771 | x = sprintf(buf, "%lu", total); |
4772 | #ifdef CONFIG_NUMA | |
fa45dc25 | 4773 | for (node = 0; node < nr_node_ids; node++) |
81819f0f CL |
4774 | if (nodes[node]) |
4775 | x += sprintf(buf + x, " N%d=%lu", | |
4776 | node, nodes[node]); | |
4777 | #endif | |
bfc8c901 | 4778 | put_online_mems(); |
81819f0f CL |
4779 | kfree(nodes); |
4780 | return x + sprintf(buf + x, "\n"); | |
4781 | } | |
4782 | ||
ab4d5ed5 | 4783 | #ifdef CONFIG_SLUB_DEBUG |
81819f0f CL |
4784 | static int any_slab_objects(struct kmem_cache *s) |
4785 | { | |
4786 | int node; | |
fa45dc25 | 4787 | struct kmem_cache_node *n; |
81819f0f | 4788 | |
fa45dc25 | 4789 | for_each_kmem_cache_node(s, node, n) |
4ea33e2d | 4790 | if (atomic_long_read(&n->total_objects)) |
81819f0f | 4791 | return 1; |
fa45dc25 | 4792 | |
81819f0f CL |
4793 | return 0; |
4794 | } | |
ab4d5ed5 | 4795 | #endif |
81819f0f CL |
4796 | |
4797 | #define to_slab_attr(n) container_of(n, struct slab_attribute, attr) | |
497888cf | 4798 | #define to_slab(n) container_of(n, struct kmem_cache, kobj) |
81819f0f CL |
4799 | |
4800 | struct slab_attribute { | |
4801 | struct attribute attr; | |
4802 | ssize_t (*show)(struct kmem_cache *s, char *buf); | |
4803 | ssize_t (*store)(struct kmem_cache *s, const char *x, size_t count); | |
4804 | }; | |
4805 | ||
4806 | #define SLAB_ATTR_RO(_name) \ | |
ab067e99 VK |
4807 | static struct slab_attribute _name##_attr = \ |
4808 | __ATTR(_name, 0400, _name##_show, NULL) | |
81819f0f CL |
4809 | |
4810 | #define SLAB_ATTR(_name) \ | |
4811 | static struct slab_attribute _name##_attr = \ | |
ab067e99 | 4812 | __ATTR(_name, 0600, _name##_show, _name##_store) |
81819f0f | 4813 | |
81819f0f CL |
4814 | static ssize_t slab_size_show(struct kmem_cache *s, char *buf) |
4815 | { | |
4816 | return sprintf(buf, "%d\n", s->size); | |
4817 | } | |
4818 | SLAB_ATTR_RO(slab_size); | |
4819 | ||
4820 | static ssize_t align_show(struct kmem_cache *s, char *buf) | |
4821 | { | |
4822 | return sprintf(buf, "%d\n", s->align); | |
4823 | } | |
4824 | SLAB_ATTR_RO(align); | |
4825 | ||
4826 | static ssize_t object_size_show(struct kmem_cache *s, char *buf) | |
4827 | { | |
3b0efdfa | 4828 | return sprintf(buf, "%d\n", s->object_size); |
81819f0f CL |
4829 | } |
4830 | SLAB_ATTR_RO(object_size); | |
4831 | ||
4832 | static ssize_t objs_per_slab_show(struct kmem_cache *s, char *buf) | |
4833 | { | |
834f3d11 | 4834 | return sprintf(buf, "%d\n", oo_objects(s->oo)); |
81819f0f CL |
4835 | } |
4836 | SLAB_ATTR_RO(objs_per_slab); | |
4837 | ||
06b285dc CL |
4838 | static ssize_t order_store(struct kmem_cache *s, |
4839 | const char *buf, size_t length) | |
4840 | { | |
0121c619 CL |
4841 | unsigned long order; |
4842 | int err; | |
4843 | ||
3dbb95f7 | 4844 | err = kstrtoul(buf, 10, &order); |
0121c619 CL |
4845 | if (err) |
4846 | return err; | |
06b285dc CL |
4847 | |
4848 | if (order > slub_max_order || order < slub_min_order) | |
4849 | return -EINVAL; | |
4850 | ||
4851 | calculate_sizes(s, order); | |
4852 | return length; | |
4853 | } | |
4854 | ||
81819f0f CL |
4855 | static ssize_t order_show(struct kmem_cache *s, char *buf) |
4856 | { | |
834f3d11 | 4857 | return sprintf(buf, "%d\n", oo_order(s->oo)); |
81819f0f | 4858 | } |
06b285dc | 4859 | SLAB_ATTR(order); |
81819f0f | 4860 | |
73d342b1 DR |
4861 | static ssize_t min_partial_show(struct kmem_cache *s, char *buf) |
4862 | { | |
4863 | return sprintf(buf, "%lu\n", s->min_partial); | |
4864 | } | |
4865 | ||
4866 | static ssize_t min_partial_store(struct kmem_cache *s, const char *buf, | |
4867 | size_t length) | |
4868 | { | |
4869 | unsigned long min; | |
4870 | int err; | |
4871 | ||
3dbb95f7 | 4872 | err = kstrtoul(buf, 10, &min); |
73d342b1 DR |
4873 | if (err) |
4874 | return err; | |
4875 | ||
c0bdb232 | 4876 | set_min_partial(s, min); |
73d342b1 DR |
4877 | return length; |
4878 | } | |
4879 | SLAB_ATTR(min_partial); | |
4880 | ||
49e22585 CL |
4881 | static ssize_t cpu_partial_show(struct kmem_cache *s, char *buf) |
4882 | { | |
4883 | return sprintf(buf, "%u\n", s->cpu_partial); | |
4884 | } | |
4885 | ||
4886 | static ssize_t cpu_partial_store(struct kmem_cache *s, const char *buf, | |
4887 | size_t length) | |
4888 | { | |
4889 | unsigned long objects; | |
4890 | int err; | |
4891 | ||
3dbb95f7 | 4892 | err = kstrtoul(buf, 10, &objects); |
49e22585 CL |
4893 | if (err) |
4894 | return err; | |
345c905d | 4895 | if (objects && !kmem_cache_has_cpu_partial(s)) |
74ee4ef1 | 4896 | return -EINVAL; |
49e22585 CL |
4897 | |
4898 | s->cpu_partial = objects; | |
4899 | flush_all(s); | |
4900 | return length; | |
4901 | } | |
4902 | SLAB_ATTR(cpu_partial); | |
4903 | ||
81819f0f CL |
4904 | static ssize_t ctor_show(struct kmem_cache *s, char *buf) |
4905 | { | |
62c70bce JP |
4906 | if (!s->ctor) |
4907 | return 0; | |
4908 | return sprintf(buf, "%pS\n", s->ctor); | |
81819f0f CL |
4909 | } |
4910 | SLAB_ATTR_RO(ctor); | |
4911 | ||
81819f0f CL |
4912 | static ssize_t aliases_show(struct kmem_cache *s, char *buf) |
4913 | { | |
4307c14f | 4914 | return sprintf(buf, "%d\n", s->refcount < 0 ? 0 : s->refcount - 1); |
81819f0f CL |
4915 | } |
4916 | SLAB_ATTR_RO(aliases); | |
4917 | ||
81819f0f CL |
4918 | static ssize_t partial_show(struct kmem_cache *s, char *buf) |
4919 | { | |
d9acf4b7 | 4920 | return show_slab_objects(s, buf, SO_PARTIAL); |
81819f0f CL |
4921 | } |
4922 | SLAB_ATTR_RO(partial); | |
4923 | ||
4924 | static ssize_t cpu_slabs_show(struct kmem_cache *s, char *buf) | |
4925 | { | |
d9acf4b7 | 4926 | return show_slab_objects(s, buf, SO_CPU); |
81819f0f CL |
4927 | } |
4928 | SLAB_ATTR_RO(cpu_slabs); | |
4929 | ||
4930 | static ssize_t objects_show(struct kmem_cache *s, char *buf) | |
4931 | { | |
205ab99d | 4932 | return show_slab_objects(s, buf, SO_ALL|SO_OBJECTS); |
81819f0f CL |
4933 | } |
4934 | SLAB_ATTR_RO(objects); | |
4935 | ||
205ab99d CL |
4936 | static ssize_t objects_partial_show(struct kmem_cache *s, char *buf) |
4937 | { | |
4938 | return show_slab_objects(s, buf, SO_PARTIAL|SO_OBJECTS); | |
4939 | } | |
4940 | SLAB_ATTR_RO(objects_partial); | |
4941 | ||
49e22585 CL |
4942 | static ssize_t slabs_cpu_partial_show(struct kmem_cache *s, char *buf) |
4943 | { | |
4944 | int objects = 0; | |
4945 | int pages = 0; | |
4946 | int cpu; | |
4947 | int len; | |
4948 | ||
4949 | for_each_online_cpu(cpu) { | |
4950 | struct page *page = per_cpu_ptr(s->cpu_slab, cpu)->partial; | |
4951 | ||
4952 | if (page) { | |
4953 | pages += page->pages; | |
4954 | objects += page->pobjects; | |
4955 | } | |
4956 | } | |
4957 | ||
4958 | len = sprintf(buf, "%d(%d)", objects, pages); | |
4959 | ||
4960 | #ifdef CONFIG_SMP | |
4961 | for_each_online_cpu(cpu) { | |
4962 | struct page *page = per_cpu_ptr(s->cpu_slab, cpu) ->partial; | |
4963 | ||
4964 | if (page && len < PAGE_SIZE - 20) | |
4965 | len += sprintf(buf + len, " C%d=%d(%d)", cpu, | |
4966 | page->pobjects, page->pages); | |
4967 | } | |
4968 | #endif | |
4969 | return len + sprintf(buf + len, "\n"); | |
4970 | } | |
4971 | SLAB_ATTR_RO(slabs_cpu_partial); | |
4972 | ||
a5a84755 CL |
4973 | static ssize_t reclaim_account_show(struct kmem_cache *s, char *buf) |
4974 | { | |
4975 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_RECLAIM_ACCOUNT)); | |
4976 | } | |
4977 | ||
4978 | static ssize_t reclaim_account_store(struct kmem_cache *s, | |
4979 | const char *buf, size_t length) | |
4980 | { | |
4981 | s->flags &= ~SLAB_RECLAIM_ACCOUNT; | |
4982 | if (buf[0] == '1') | |
4983 | s->flags |= SLAB_RECLAIM_ACCOUNT; | |
4984 | return length; | |
4985 | } | |
4986 | SLAB_ATTR(reclaim_account); | |
4987 | ||
4988 | static ssize_t hwcache_align_show(struct kmem_cache *s, char *buf) | |
4989 | { | |
4990 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_HWCACHE_ALIGN)); | |
4991 | } | |
4992 | SLAB_ATTR_RO(hwcache_align); | |
4993 | ||
4994 | #ifdef CONFIG_ZONE_DMA | |
4995 | static ssize_t cache_dma_show(struct kmem_cache *s, char *buf) | |
4996 | { | |
4997 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_CACHE_DMA)); | |
4998 | } | |
4999 | SLAB_ATTR_RO(cache_dma); | |
5000 | #endif | |
5001 | ||
5002 | static ssize_t destroy_by_rcu_show(struct kmem_cache *s, char *buf) | |
5003 | { | |
5004 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_DESTROY_BY_RCU)); | |
5005 | } | |
5006 | SLAB_ATTR_RO(destroy_by_rcu); | |
5007 | ||
ab9a0f19 LJ |
5008 | static ssize_t reserved_show(struct kmem_cache *s, char *buf) |
5009 | { | |
5010 | return sprintf(buf, "%d\n", s->reserved); | |
5011 | } | |
5012 | SLAB_ATTR_RO(reserved); | |
5013 | ||
ab4d5ed5 | 5014 | #ifdef CONFIG_SLUB_DEBUG |
a5a84755 CL |
5015 | static ssize_t slabs_show(struct kmem_cache *s, char *buf) |
5016 | { | |
5017 | return show_slab_objects(s, buf, SO_ALL); | |
5018 | } | |
5019 | SLAB_ATTR_RO(slabs); | |
5020 | ||
205ab99d CL |
5021 | static ssize_t total_objects_show(struct kmem_cache *s, char *buf) |
5022 | { | |
5023 | return show_slab_objects(s, buf, SO_ALL|SO_TOTAL); | |
5024 | } | |
5025 | SLAB_ATTR_RO(total_objects); | |
5026 | ||
81819f0f CL |
5027 | static ssize_t sanity_checks_show(struct kmem_cache *s, char *buf) |
5028 | { | |
becfda68 | 5029 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_CONSISTENCY_CHECKS)); |
81819f0f CL |
5030 | } |
5031 | ||
5032 | static ssize_t sanity_checks_store(struct kmem_cache *s, | |
5033 | const char *buf, size_t length) | |
5034 | { | |
becfda68 | 5035 | s->flags &= ~SLAB_CONSISTENCY_CHECKS; |
b789ef51 CL |
5036 | if (buf[0] == '1') { |
5037 | s->flags &= ~__CMPXCHG_DOUBLE; | |
becfda68 | 5038 | s->flags |= SLAB_CONSISTENCY_CHECKS; |
b789ef51 | 5039 | } |
81819f0f CL |
5040 | return length; |
5041 | } | |
5042 | SLAB_ATTR(sanity_checks); | |
5043 | ||
5044 | static ssize_t trace_show(struct kmem_cache *s, char *buf) | |
5045 | { | |
5046 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_TRACE)); | |
5047 | } | |
5048 | ||
5049 | static ssize_t trace_store(struct kmem_cache *s, const char *buf, | |
5050 | size_t length) | |
5051 | { | |
c9e16131 CL |
5052 | /* |
5053 | * Tracing a merged cache is going to give confusing results | |
5054 | * as well as cause other issues like converting a mergeable | |
5055 | * cache into an umergeable one. | |
5056 | */ | |
5057 | if (s->refcount > 1) | |
5058 | return -EINVAL; | |
5059 | ||
81819f0f | 5060 | s->flags &= ~SLAB_TRACE; |
b789ef51 CL |
5061 | if (buf[0] == '1') { |
5062 | s->flags &= ~__CMPXCHG_DOUBLE; | |
81819f0f | 5063 | s->flags |= SLAB_TRACE; |
b789ef51 | 5064 | } |
81819f0f CL |
5065 | return length; |
5066 | } | |
5067 | SLAB_ATTR(trace); | |
5068 | ||
81819f0f CL |
5069 | static ssize_t red_zone_show(struct kmem_cache *s, char *buf) |
5070 | { | |
5071 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_RED_ZONE)); | |
5072 | } | |
5073 | ||
5074 | static ssize_t red_zone_store(struct kmem_cache *s, | |
5075 | const char *buf, size_t length) | |
5076 | { | |
5077 | if (any_slab_objects(s)) | |
5078 | return -EBUSY; | |
5079 | ||
5080 | s->flags &= ~SLAB_RED_ZONE; | |
b789ef51 | 5081 | if (buf[0] == '1') { |
81819f0f | 5082 | s->flags |= SLAB_RED_ZONE; |
b789ef51 | 5083 | } |
06b285dc | 5084 | calculate_sizes(s, -1); |
81819f0f CL |
5085 | return length; |
5086 | } | |
5087 | SLAB_ATTR(red_zone); | |
5088 | ||
5089 | static ssize_t poison_show(struct kmem_cache *s, char *buf) | |
5090 | { | |
5091 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_POISON)); | |
5092 | } | |
5093 | ||
5094 | static ssize_t poison_store(struct kmem_cache *s, | |
5095 | const char *buf, size_t length) | |
5096 | { | |
5097 | if (any_slab_objects(s)) | |
5098 | return -EBUSY; | |
5099 | ||
5100 | s->flags &= ~SLAB_POISON; | |
b789ef51 | 5101 | if (buf[0] == '1') { |
81819f0f | 5102 | s->flags |= SLAB_POISON; |
b789ef51 | 5103 | } |
06b285dc | 5104 | calculate_sizes(s, -1); |
81819f0f CL |
5105 | return length; |
5106 | } | |
5107 | SLAB_ATTR(poison); | |
5108 | ||
5109 | static ssize_t store_user_show(struct kmem_cache *s, char *buf) | |
5110 | { | |
5111 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_STORE_USER)); | |
5112 | } | |
5113 | ||
5114 | static ssize_t store_user_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_STORE_USER; | |
b789ef51 CL |
5121 | if (buf[0] == '1') { |
5122 | s->flags &= ~__CMPXCHG_DOUBLE; | |
81819f0f | 5123 | s->flags |= SLAB_STORE_USER; |
b789ef51 | 5124 | } |
06b285dc | 5125 | calculate_sizes(s, -1); |
81819f0f CL |
5126 | return length; |
5127 | } | |
5128 | SLAB_ATTR(store_user); | |
5129 | ||
53e15af0 CL |
5130 | static ssize_t validate_show(struct kmem_cache *s, char *buf) |
5131 | { | |
5132 | return 0; | |
5133 | } | |
5134 | ||
5135 | static ssize_t validate_store(struct kmem_cache *s, | |
5136 | const char *buf, size_t length) | |
5137 | { | |
434e245d CL |
5138 | int ret = -EINVAL; |
5139 | ||
5140 | if (buf[0] == '1') { | |
5141 | ret = validate_slab_cache(s); | |
5142 | if (ret >= 0) | |
5143 | ret = length; | |
5144 | } | |
5145 | return ret; | |
53e15af0 CL |
5146 | } |
5147 | SLAB_ATTR(validate); | |
a5a84755 CL |
5148 | |
5149 | static ssize_t alloc_calls_show(struct kmem_cache *s, char *buf) | |
5150 | { | |
5151 | if (!(s->flags & SLAB_STORE_USER)) | |
5152 | return -ENOSYS; | |
5153 | return list_locations(s, buf, TRACK_ALLOC); | |
5154 | } | |
5155 | SLAB_ATTR_RO(alloc_calls); | |
5156 | ||
5157 | static ssize_t free_calls_show(struct kmem_cache *s, char *buf) | |
5158 | { | |
5159 | if (!(s->flags & SLAB_STORE_USER)) | |
5160 | return -ENOSYS; | |
5161 | return list_locations(s, buf, TRACK_FREE); | |
5162 | } | |
5163 | SLAB_ATTR_RO(free_calls); | |
5164 | #endif /* CONFIG_SLUB_DEBUG */ | |
5165 | ||
5166 | #ifdef CONFIG_FAILSLAB | |
5167 | static ssize_t failslab_show(struct kmem_cache *s, char *buf) | |
5168 | { | |
5169 | return sprintf(buf, "%d\n", !!(s->flags & SLAB_FAILSLAB)); | |
5170 | } | |
5171 | ||
5172 | static ssize_t failslab_store(struct kmem_cache *s, const char *buf, | |
5173 | size_t length) | |
5174 | { | |
c9e16131 CL |
5175 | if (s->refcount > 1) |
5176 | return -EINVAL; | |
5177 | ||
a5a84755 CL |
5178 | s->flags &= ~SLAB_FAILSLAB; |
5179 | if (buf[0] == '1') | |
5180 | s->flags |= SLAB_FAILSLAB; | |
5181 | return length; | |
5182 | } | |
5183 | SLAB_ATTR(failslab); | |
ab4d5ed5 | 5184 | #endif |
53e15af0 | 5185 | |
2086d26a CL |
5186 | static ssize_t shrink_show(struct kmem_cache *s, char *buf) |
5187 | { | |
5188 | return 0; | |
5189 | } | |
5190 | ||
5191 | static ssize_t shrink_store(struct kmem_cache *s, | |
5192 | const char *buf, size_t length) | |
5193 | { | |
832f37f5 VD |
5194 | if (buf[0] == '1') |
5195 | kmem_cache_shrink(s); | |
5196 | else | |
2086d26a CL |
5197 | return -EINVAL; |
5198 | return length; | |
5199 | } | |
5200 | SLAB_ATTR(shrink); | |
5201 | ||
81819f0f | 5202 | #ifdef CONFIG_NUMA |
9824601e | 5203 | static ssize_t remote_node_defrag_ratio_show(struct kmem_cache *s, char *buf) |
81819f0f | 5204 | { |
9824601e | 5205 | return sprintf(buf, "%d\n", s->remote_node_defrag_ratio / 10); |
81819f0f CL |
5206 | } |
5207 | ||
9824601e | 5208 | static ssize_t remote_node_defrag_ratio_store(struct kmem_cache *s, |
81819f0f CL |
5209 | const char *buf, size_t length) |
5210 | { | |
0121c619 CL |
5211 | unsigned long ratio; |
5212 | int err; | |
5213 | ||
3dbb95f7 | 5214 | err = kstrtoul(buf, 10, &ratio); |
0121c619 CL |
5215 | if (err) |
5216 | return err; | |
5217 | ||
e2cb96b7 | 5218 | if (ratio <= 100) |
0121c619 | 5219 | s->remote_node_defrag_ratio = ratio * 10; |
81819f0f | 5220 | |
81819f0f CL |
5221 | return length; |
5222 | } | |
9824601e | 5223 | SLAB_ATTR(remote_node_defrag_ratio); |
81819f0f CL |
5224 | #endif |
5225 | ||
8ff12cfc | 5226 | #ifdef CONFIG_SLUB_STATS |
8ff12cfc CL |
5227 | static int show_stat(struct kmem_cache *s, char *buf, enum stat_item si) |
5228 | { | |
5229 | unsigned long sum = 0; | |
5230 | int cpu; | |
5231 | int len; | |
5232 | int *data = kmalloc(nr_cpu_ids * sizeof(int), GFP_KERNEL); | |
5233 | ||
5234 | if (!data) | |
5235 | return -ENOMEM; | |
5236 | ||
5237 | for_each_online_cpu(cpu) { | |
9dfc6e68 | 5238 | unsigned x = per_cpu_ptr(s->cpu_slab, cpu)->stat[si]; |
8ff12cfc CL |
5239 | |
5240 | data[cpu] = x; | |
5241 | sum += x; | |
5242 | } | |
5243 | ||
5244 | len = sprintf(buf, "%lu", sum); | |
5245 | ||
50ef37b9 | 5246 | #ifdef CONFIG_SMP |
8ff12cfc CL |
5247 | for_each_online_cpu(cpu) { |
5248 | if (data[cpu] && len < PAGE_SIZE - 20) | |
50ef37b9 | 5249 | len += sprintf(buf + len, " C%d=%u", cpu, data[cpu]); |
8ff12cfc | 5250 | } |
50ef37b9 | 5251 | #endif |
8ff12cfc CL |
5252 | kfree(data); |
5253 | return len + sprintf(buf + len, "\n"); | |
5254 | } | |
5255 | ||
78eb00cc DR |
5256 | static void clear_stat(struct kmem_cache *s, enum stat_item si) |
5257 | { | |
5258 | int cpu; | |
5259 | ||
5260 | for_each_online_cpu(cpu) | |
9dfc6e68 | 5261 | per_cpu_ptr(s->cpu_slab, cpu)->stat[si] = 0; |
78eb00cc DR |
5262 | } |
5263 | ||
8ff12cfc CL |
5264 | #define STAT_ATTR(si, text) \ |
5265 | static ssize_t text##_show(struct kmem_cache *s, char *buf) \ | |
5266 | { \ | |
5267 | return show_stat(s, buf, si); \ | |
5268 | } \ | |
78eb00cc DR |
5269 | static ssize_t text##_store(struct kmem_cache *s, \ |
5270 | const char *buf, size_t length) \ | |
5271 | { \ | |
5272 | if (buf[0] != '0') \ | |
5273 | return -EINVAL; \ | |
5274 | clear_stat(s, si); \ | |
5275 | return length; \ | |
5276 | } \ | |
5277 | SLAB_ATTR(text); \ | |
8ff12cfc CL |
5278 | |
5279 | STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath); | |
5280 | STAT_ATTR(ALLOC_SLOWPATH, alloc_slowpath); | |
5281 | STAT_ATTR(FREE_FASTPATH, free_fastpath); | |
5282 | STAT_ATTR(FREE_SLOWPATH, free_slowpath); | |
5283 | STAT_ATTR(FREE_FROZEN, free_frozen); | |
5284 | STAT_ATTR(FREE_ADD_PARTIAL, free_add_partial); | |
5285 | STAT_ATTR(FREE_REMOVE_PARTIAL, free_remove_partial); | |
5286 | STAT_ATTR(ALLOC_FROM_PARTIAL, alloc_from_partial); | |
5287 | STAT_ATTR(ALLOC_SLAB, alloc_slab); | |
5288 | STAT_ATTR(ALLOC_REFILL, alloc_refill); | |
e36a2652 | 5289 | STAT_ATTR(ALLOC_NODE_MISMATCH, alloc_node_mismatch); |
8ff12cfc CL |
5290 | STAT_ATTR(FREE_SLAB, free_slab); |
5291 | STAT_ATTR(CPUSLAB_FLUSH, cpuslab_flush); | |
5292 | STAT_ATTR(DEACTIVATE_FULL, deactivate_full); | |
5293 | STAT_ATTR(DEACTIVATE_EMPTY, deactivate_empty); | |
5294 | STAT_ATTR(DEACTIVATE_TO_HEAD, deactivate_to_head); | |
5295 | STAT_ATTR(DEACTIVATE_TO_TAIL, deactivate_to_tail); | |
5296 | STAT_ATTR(DEACTIVATE_REMOTE_FREES, deactivate_remote_frees); | |
03e404af | 5297 | STAT_ATTR(DEACTIVATE_BYPASS, deactivate_bypass); |
65c3376a | 5298 | STAT_ATTR(ORDER_FALLBACK, order_fallback); |
b789ef51 CL |
5299 | STAT_ATTR(CMPXCHG_DOUBLE_CPU_FAIL, cmpxchg_double_cpu_fail); |
5300 | STAT_ATTR(CMPXCHG_DOUBLE_FAIL, cmpxchg_double_fail); | |
49e22585 CL |
5301 | STAT_ATTR(CPU_PARTIAL_ALLOC, cpu_partial_alloc); |
5302 | STAT_ATTR(CPU_PARTIAL_FREE, cpu_partial_free); | |
8028dcea AS |
5303 | STAT_ATTR(CPU_PARTIAL_NODE, cpu_partial_node); |
5304 | STAT_ATTR(CPU_PARTIAL_DRAIN, cpu_partial_drain); | |
8ff12cfc CL |
5305 | #endif |
5306 | ||
06428780 | 5307 | static struct attribute *slab_attrs[] = { |
81819f0f CL |
5308 | &slab_size_attr.attr, |
5309 | &object_size_attr.attr, | |
5310 | &objs_per_slab_attr.attr, | |
5311 | &order_attr.attr, | |
73d342b1 | 5312 | &min_partial_attr.attr, |
49e22585 | 5313 | &cpu_partial_attr.attr, |
81819f0f | 5314 | &objects_attr.attr, |
205ab99d | 5315 | &objects_partial_attr.attr, |
81819f0f CL |
5316 | &partial_attr.attr, |
5317 | &cpu_slabs_attr.attr, | |
5318 | &ctor_attr.attr, | |
81819f0f CL |
5319 | &aliases_attr.attr, |
5320 | &align_attr.attr, | |
81819f0f CL |
5321 | &hwcache_align_attr.attr, |
5322 | &reclaim_account_attr.attr, | |
5323 | &destroy_by_rcu_attr.attr, | |
a5a84755 | 5324 | &shrink_attr.attr, |
ab9a0f19 | 5325 | &reserved_attr.attr, |
49e22585 | 5326 | &slabs_cpu_partial_attr.attr, |
ab4d5ed5 | 5327 | #ifdef CONFIG_SLUB_DEBUG |
a5a84755 CL |
5328 | &total_objects_attr.attr, |
5329 | &slabs_attr.attr, | |
5330 | &sanity_checks_attr.attr, | |
5331 | &trace_attr.attr, | |
81819f0f CL |
5332 | &red_zone_attr.attr, |
5333 | &poison_attr.attr, | |
5334 | &store_user_attr.attr, | |
53e15af0 | 5335 | &validate_attr.attr, |
88a420e4 CL |
5336 | &alloc_calls_attr.attr, |
5337 | &free_calls_attr.attr, | |
ab4d5ed5 | 5338 | #endif |
81819f0f CL |
5339 | #ifdef CONFIG_ZONE_DMA |
5340 | &cache_dma_attr.attr, | |
5341 | #endif | |
5342 | #ifdef CONFIG_NUMA | |
9824601e | 5343 | &remote_node_defrag_ratio_attr.attr, |
8ff12cfc CL |
5344 | #endif |
5345 | #ifdef CONFIG_SLUB_STATS | |
5346 | &alloc_fastpath_attr.attr, | |
5347 | &alloc_slowpath_attr.attr, | |
5348 | &free_fastpath_attr.attr, | |
5349 | &free_slowpath_attr.attr, | |
5350 | &free_frozen_attr.attr, | |
5351 | &free_add_partial_attr.attr, | |
5352 | &free_remove_partial_attr.attr, | |
5353 | &alloc_from_partial_attr.attr, | |
5354 | &alloc_slab_attr.attr, | |
5355 | &alloc_refill_attr.attr, | |
e36a2652 | 5356 | &alloc_node_mismatch_attr.attr, |
8ff12cfc CL |
5357 | &free_slab_attr.attr, |
5358 | &cpuslab_flush_attr.attr, | |
5359 | &deactivate_full_attr.attr, | |
5360 | &deactivate_empty_attr.attr, | |
5361 | &deactivate_to_head_attr.attr, | |
5362 | &deactivate_to_tail_attr.attr, | |
5363 | &deactivate_remote_frees_attr.attr, | |
03e404af | 5364 | &deactivate_bypass_attr.attr, |
65c3376a | 5365 | &order_fallback_attr.attr, |
b789ef51 CL |
5366 | &cmpxchg_double_fail_attr.attr, |
5367 | &cmpxchg_double_cpu_fail_attr.attr, | |
49e22585 CL |
5368 | &cpu_partial_alloc_attr.attr, |
5369 | &cpu_partial_free_attr.attr, | |
8028dcea AS |
5370 | &cpu_partial_node_attr.attr, |
5371 | &cpu_partial_drain_attr.attr, | |
81819f0f | 5372 | #endif |
4c13dd3b DM |
5373 | #ifdef CONFIG_FAILSLAB |
5374 | &failslab_attr.attr, | |
5375 | #endif | |
5376 | ||
81819f0f CL |
5377 | NULL |
5378 | }; | |
5379 | ||
5380 | static struct attribute_group slab_attr_group = { | |
5381 | .attrs = slab_attrs, | |
5382 | }; | |
5383 | ||
5384 | static ssize_t slab_attr_show(struct kobject *kobj, | |
5385 | struct attribute *attr, | |
5386 | char *buf) | |
5387 | { | |
5388 | struct slab_attribute *attribute; | |
5389 | struct kmem_cache *s; | |
5390 | int err; | |
5391 | ||
5392 | attribute = to_slab_attr(attr); | |
5393 | s = to_slab(kobj); | |
5394 | ||
5395 | if (!attribute->show) | |
5396 | return -EIO; | |
5397 | ||
5398 | err = attribute->show(s, buf); | |
5399 | ||
5400 | return err; | |
5401 | } | |
5402 | ||
5403 | static ssize_t slab_attr_store(struct kobject *kobj, | |
5404 | struct attribute *attr, | |
5405 | const char *buf, size_t len) | |
5406 | { | |
5407 | struct slab_attribute *attribute; | |
5408 | struct kmem_cache *s; | |
5409 | int err; | |
5410 | ||
5411 | attribute = to_slab_attr(attr); | |
5412 | s = to_slab(kobj); | |
5413 | ||
5414 | if (!attribute->store) | |
5415 | return -EIO; | |
5416 | ||
5417 | err = attribute->store(s, buf, len); | |
127424c8 | 5418 | #ifdef CONFIG_MEMCG |
107dab5c | 5419 | if (slab_state >= FULL && err >= 0 && is_root_cache(s)) { |
426589f5 | 5420 | struct kmem_cache *c; |
81819f0f | 5421 | |
107dab5c GC |
5422 | mutex_lock(&slab_mutex); |
5423 | if (s->max_attr_size < len) | |
5424 | s->max_attr_size = len; | |
5425 | ||
ebe945c2 GC |
5426 | /* |
5427 | * This is a best effort propagation, so this function's return | |
5428 | * value will be determined by the parent cache only. This is | |
5429 | * basically because not all attributes will have a well | |
5430 | * defined semantics for rollbacks - most of the actions will | |
5431 | * have permanent effects. | |
5432 | * | |
5433 | * Returning the error value of any of the children that fail | |
5434 | * is not 100 % defined, in the sense that users seeing the | |
5435 | * error code won't be able to know anything about the state of | |
5436 | * the cache. | |
5437 | * | |
5438 | * Only returning the error code for the parent cache at least | |
5439 | * has well defined semantics. The cache being written to | |
5440 | * directly either failed or succeeded, in which case we loop | |
5441 | * through the descendants with best-effort propagation. | |
5442 | */ | |
426589f5 VD |
5443 | for_each_memcg_cache(c, s) |
5444 | attribute->store(c, buf, len); | |
107dab5c GC |
5445 | mutex_unlock(&slab_mutex); |
5446 | } | |
5447 | #endif | |
81819f0f CL |
5448 | return err; |
5449 | } | |
5450 | ||
107dab5c GC |
5451 | static void memcg_propagate_slab_attrs(struct kmem_cache *s) |
5452 | { | |
127424c8 | 5453 | #ifdef CONFIG_MEMCG |
107dab5c GC |
5454 | int i; |
5455 | char *buffer = NULL; | |
93030d83 | 5456 | struct kmem_cache *root_cache; |
107dab5c | 5457 | |
93030d83 | 5458 | if (is_root_cache(s)) |
107dab5c GC |
5459 | return; |
5460 | ||
f7ce3190 | 5461 | root_cache = s->memcg_params.root_cache; |
93030d83 | 5462 | |
107dab5c GC |
5463 | /* |
5464 | * This mean this cache had no attribute written. Therefore, no point | |
5465 | * in copying default values around | |
5466 | */ | |
93030d83 | 5467 | if (!root_cache->max_attr_size) |
107dab5c GC |
5468 | return; |
5469 | ||
5470 | for (i = 0; i < ARRAY_SIZE(slab_attrs); i++) { | |
5471 | char mbuf[64]; | |
5472 | char *buf; | |
5473 | struct slab_attribute *attr = to_slab_attr(slab_attrs[i]); | |
5474 | ||
5475 | if (!attr || !attr->store || !attr->show) | |
5476 | continue; | |
5477 | ||
5478 | /* | |
5479 | * It is really bad that we have to allocate here, so we will | |
5480 | * do it only as a fallback. If we actually allocate, though, | |
5481 | * we can just use the allocated buffer until the end. | |
5482 | * | |
5483 | * Most of the slub attributes will tend to be very small in | |
5484 | * size, but sysfs allows buffers up to a page, so they can | |
5485 | * theoretically happen. | |
5486 | */ | |
5487 | if (buffer) | |
5488 | buf = buffer; | |
93030d83 | 5489 | else if (root_cache->max_attr_size < ARRAY_SIZE(mbuf)) |
107dab5c GC |
5490 | buf = mbuf; |
5491 | else { | |
5492 | buffer = (char *) get_zeroed_page(GFP_KERNEL); | |
5493 | if (WARN_ON(!buffer)) | |
5494 | continue; | |
5495 | buf = buffer; | |
5496 | } | |
5497 | ||
93030d83 | 5498 | attr->show(root_cache, buf); |
107dab5c GC |
5499 | attr->store(s, buf, strlen(buf)); |
5500 | } | |
5501 | ||
5502 | if (buffer) | |
5503 | free_page((unsigned long)buffer); | |
5504 | #endif | |
5505 | } | |
5506 | ||
41a21285 CL |
5507 | static void kmem_cache_release(struct kobject *k) |
5508 | { | |
5509 | slab_kmem_cache_release(to_slab(k)); | |
5510 | } | |
5511 | ||
52cf25d0 | 5512 | static const struct sysfs_ops slab_sysfs_ops = { |
81819f0f CL |
5513 | .show = slab_attr_show, |
5514 | .store = slab_attr_store, | |
5515 | }; | |
5516 | ||
5517 | static struct kobj_type slab_ktype = { | |
5518 | .sysfs_ops = &slab_sysfs_ops, | |
41a21285 | 5519 | .release = kmem_cache_release, |
81819f0f CL |
5520 | }; |
5521 | ||
5522 | static int uevent_filter(struct kset *kset, struct kobject *kobj) | |
5523 | { | |
5524 | struct kobj_type *ktype = get_ktype(kobj); | |
5525 | ||
5526 | if (ktype == &slab_ktype) | |
5527 | return 1; | |
5528 | return 0; | |
5529 | } | |
5530 | ||
9cd43611 | 5531 | static const struct kset_uevent_ops slab_uevent_ops = { |
81819f0f CL |
5532 | .filter = uevent_filter, |
5533 | }; | |
5534 | ||
27c3a314 | 5535 | static struct kset *slab_kset; |
81819f0f | 5536 | |
9a41707b VD |
5537 | static inline struct kset *cache_kset(struct kmem_cache *s) |
5538 | { | |
127424c8 | 5539 | #ifdef CONFIG_MEMCG |
9a41707b | 5540 | if (!is_root_cache(s)) |
f7ce3190 | 5541 | return s->memcg_params.root_cache->memcg_kset; |
9a41707b VD |
5542 | #endif |
5543 | return slab_kset; | |
5544 | } | |
5545 | ||
81819f0f CL |
5546 | #define ID_STR_LENGTH 64 |
5547 | ||
5548 | /* Create a unique string id for a slab cache: | |
6446faa2 CL |
5549 | * |
5550 | * Format :[flags-]size | |
81819f0f CL |
5551 | */ |
5552 | static char *create_unique_id(struct kmem_cache *s) | |
5553 | { | |
5554 | char *name = kmalloc(ID_STR_LENGTH, GFP_KERNEL); | |
5555 | char *p = name; | |
5556 | ||
5557 | BUG_ON(!name); | |
5558 | ||
5559 | *p++ = ':'; | |
5560 | /* | |
5561 | * First flags affecting slabcache operations. We will only | |
5562 | * get here for aliasable slabs so we do not need to support | |
5563 | * too many flags. The flags here must cover all flags that | |
5564 | * are matched during merging to guarantee that the id is | |
5565 | * unique. | |
5566 | */ | |
5567 | if (s->flags & SLAB_CACHE_DMA) | |
5568 | *p++ = 'd'; | |
5569 | if (s->flags & SLAB_RECLAIM_ACCOUNT) | |
5570 | *p++ = 'a'; | |
becfda68 | 5571 | if (s->flags & SLAB_CONSISTENCY_CHECKS) |
81819f0f | 5572 | *p++ = 'F'; |
5a896d9e VN |
5573 | if (!(s->flags & SLAB_NOTRACK)) |
5574 | *p++ = 't'; | |
230e9fc2 VD |
5575 | if (s->flags & SLAB_ACCOUNT) |
5576 | *p++ = 'A'; | |
81819f0f CL |
5577 | if (p != name + 1) |
5578 | *p++ = '-'; | |
5579 | p += sprintf(p, "%07d", s->size); | |
2633d7a0 | 5580 | |
81819f0f CL |
5581 | BUG_ON(p > name + ID_STR_LENGTH - 1); |
5582 | return name; | |
5583 | } | |
5584 | ||
5585 | static int sysfs_slab_add(struct kmem_cache *s) | |
5586 | { | |
5587 | int err; | |
5588 | const char *name; | |
45530c44 | 5589 | int unmergeable = slab_unmergeable(s); |
81819f0f | 5590 | |
81819f0f CL |
5591 | if (unmergeable) { |
5592 | /* | |
5593 | * Slabcache can never be merged so we can use the name proper. | |
5594 | * This is typically the case for debug situations. In that | |
5595 | * case we can catch duplicate names easily. | |
5596 | */ | |
27c3a314 | 5597 | sysfs_remove_link(&slab_kset->kobj, s->name); |
81819f0f CL |
5598 | name = s->name; |
5599 | } else { | |
5600 | /* | |
5601 | * Create a unique name for the slab as a target | |
5602 | * for the symlinks. | |
5603 | */ | |
5604 | name = create_unique_id(s); | |
5605 | } | |
5606 | ||
9a41707b | 5607 | s->kobj.kset = cache_kset(s); |
26e4f205 | 5608 | err = kobject_init_and_add(&s->kobj, &slab_ktype, NULL, "%s", name); |
54b6a731 | 5609 | if (err) |
80da026a | 5610 | goto out; |
81819f0f CL |
5611 | |
5612 | err = sysfs_create_group(&s->kobj, &slab_attr_group); | |
54b6a731 DJ |
5613 | if (err) |
5614 | goto out_del_kobj; | |
9a41707b | 5615 | |
127424c8 | 5616 | #ifdef CONFIG_MEMCG |
9a41707b VD |
5617 | if (is_root_cache(s)) { |
5618 | s->memcg_kset = kset_create_and_add("cgroup", NULL, &s->kobj); | |
5619 | if (!s->memcg_kset) { | |
54b6a731 DJ |
5620 | err = -ENOMEM; |
5621 | goto out_del_kobj; | |
9a41707b VD |
5622 | } |
5623 | } | |
5624 | #endif | |
5625 | ||
81819f0f CL |
5626 | kobject_uevent(&s->kobj, KOBJ_ADD); |
5627 | if (!unmergeable) { | |
5628 | /* Setup first alias */ | |
5629 | sysfs_slab_alias(s, s->name); | |
81819f0f | 5630 | } |
54b6a731 DJ |
5631 | out: |
5632 | if (!unmergeable) | |
5633 | kfree(name); | |
5634 | return err; | |
5635 | out_del_kobj: | |
5636 | kobject_del(&s->kobj); | |
54b6a731 | 5637 | goto out; |
81819f0f CL |
5638 | } |
5639 | ||
41a21285 | 5640 | void sysfs_slab_remove(struct kmem_cache *s) |
81819f0f | 5641 | { |
97d06609 | 5642 | if (slab_state < FULL) |
2bce6485 CL |
5643 | /* |
5644 | * Sysfs has not been setup yet so no need to remove the | |
5645 | * cache from sysfs. | |
5646 | */ | |
5647 | return; | |
5648 | ||
127424c8 | 5649 | #ifdef CONFIG_MEMCG |
9a41707b VD |
5650 | kset_unregister(s->memcg_kset); |
5651 | #endif | |
81819f0f CL |
5652 | kobject_uevent(&s->kobj, KOBJ_REMOVE); |
5653 | kobject_del(&s->kobj); | |
151c602f | 5654 | kobject_put(&s->kobj); |
81819f0f CL |
5655 | } |
5656 | ||
5657 | /* | |
5658 | * Need to buffer aliases during bootup until sysfs becomes | |
9f6c708e | 5659 | * available lest we lose that information. |
81819f0f CL |
5660 | */ |
5661 | struct saved_alias { | |
5662 | struct kmem_cache *s; | |
5663 | const char *name; | |
5664 | struct saved_alias *next; | |
5665 | }; | |
5666 | ||
5af328a5 | 5667 | static struct saved_alias *alias_list; |
81819f0f CL |
5668 | |
5669 | static int sysfs_slab_alias(struct kmem_cache *s, const char *name) | |
5670 | { | |
5671 | struct saved_alias *al; | |
5672 | ||
97d06609 | 5673 | if (slab_state == FULL) { |
81819f0f CL |
5674 | /* |
5675 | * If we have a leftover link then remove it. | |
5676 | */ | |
27c3a314 GKH |
5677 | sysfs_remove_link(&slab_kset->kobj, name); |
5678 | return sysfs_create_link(&slab_kset->kobj, &s->kobj, name); | |
81819f0f CL |
5679 | } |
5680 | ||
5681 | al = kmalloc(sizeof(struct saved_alias), GFP_KERNEL); | |
5682 | if (!al) | |
5683 | return -ENOMEM; | |
5684 | ||
5685 | al->s = s; | |
5686 | al->name = name; | |
5687 | al->next = alias_list; | |
5688 | alias_list = al; | |
5689 | return 0; | |
5690 | } | |
5691 | ||
5692 | static int __init slab_sysfs_init(void) | |
5693 | { | |
5b95a4ac | 5694 | struct kmem_cache *s; |
81819f0f CL |
5695 | int err; |
5696 | ||
18004c5d | 5697 | mutex_lock(&slab_mutex); |
2bce6485 | 5698 | |
0ff21e46 | 5699 | slab_kset = kset_create_and_add("slab", &slab_uevent_ops, kernel_kobj); |
27c3a314 | 5700 | if (!slab_kset) { |
18004c5d | 5701 | mutex_unlock(&slab_mutex); |
f9f58285 | 5702 | pr_err("Cannot register slab subsystem.\n"); |
81819f0f CL |
5703 | return -ENOSYS; |
5704 | } | |
5705 | ||
97d06609 | 5706 | slab_state = FULL; |
26a7bd03 | 5707 | |
5b95a4ac | 5708 | list_for_each_entry(s, &slab_caches, list) { |
26a7bd03 | 5709 | err = sysfs_slab_add(s); |
5d540fb7 | 5710 | if (err) |
f9f58285 FF |
5711 | pr_err("SLUB: Unable to add boot slab %s to sysfs\n", |
5712 | s->name); | |
26a7bd03 | 5713 | } |
81819f0f CL |
5714 | |
5715 | while (alias_list) { | |
5716 | struct saved_alias *al = alias_list; | |
5717 | ||
5718 | alias_list = alias_list->next; | |
5719 | err = sysfs_slab_alias(al->s, al->name); | |
5d540fb7 | 5720 | if (err) |
f9f58285 FF |
5721 | pr_err("SLUB: Unable to add boot slab alias %s to sysfs\n", |
5722 | al->name); | |
81819f0f CL |
5723 | kfree(al); |
5724 | } | |
5725 | ||
18004c5d | 5726 | mutex_unlock(&slab_mutex); |
81819f0f CL |
5727 | resiliency_test(); |
5728 | return 0; | |
5729 | } | |
5730 | ||
5731 | __initcall(slab_sysfs_init); | |
ab4d5ed5 | 5732 | #endif /* CONFIG_SYSFS */ |
57ed3eda PE |
5733 | |
5734 | /* | |
5735 | * The /proc/slabinfo ABI | |
5736 | */ | |
158a9624 | 5737 | #ifdef CONFIG_SLABINFO |
0d7561c6 | 5738 | void get_slabinfo(struct kmem_cache *s, struct slabinfo *sinfo) |
57ed3eda | 5739 | { |
57ed3eda | 5740 | unsigned long nr_slabs = 0; |
205ab99d CL |
5741 | unsigned long nr_objs = 0; |
5742 | unsigned long nr_free = 0; | |
57ed3eda | 5743 | int node; |
fa45dc25 | 5744 | struct kmem_cache_node *n; |
57ed3eda | 5745 | |
fa45dc25 | 5746 | for_each_kmem_cache_node(s, node, n) { |
c17fd13e WL |
5747 | nr_slabs += node_nr_slabs(n); |
5748 | nr_objs += node_nr_objs(n); | |
205ab99d | 5749 | nr_free += count_partial(n, count_free); |
57ed3eda PE |
5750 | } |
5751 | ||
0d7561c6 GC |
5752 | sinfo->active_objs = nr_objs - nr_free; |
5753 | sinfo->num_objs = nr_objs; | |
5754 | sinfo->active_slabs = nr_slabs; | |
5755 | sinfo->num_slabs = nr_slabs; | |
5756 | sinfo->objects_per_slab = oo_objects(s->oo); | |
5757 | sinfo->cache_order = oo_order(s->oo); | |
57ed3eda PE |
5758 | } |
5759 | ||
0d7561c6 | 5760 | void slabinfo_show_stats(struct seq_file *m, struct kmem_cache *s) |
7b3c3a50 | 5761 | { |
7b3c3a50 AD |
5762 | } |
5763 | ||
b7454ad3 GC |
5764 | ssize_t slabinfo_write(struct file *file, const char __user *buffer, |
5765 | size_t count, loff_t *ppos) | |
7b3c3a50 | 5766 | { |
b7454ad3 | 5767 | return -EIO; |
7b3c3a50 | 5768 | } |
158a9624 | 5769 | #endif /* CONFIG_SLABINFO */ |