blk-mq: really fix plug list flushing for nomerge queues
[linux-2.6-block.git] / mm / slab.h
CommitLineData
97d06609
CL
1#ifndef MM_SLAB_H
2#define MM_SLAB_H
3/*
4 * Internal slab definitions
5 */
6
07f361b2
JK
7#ifdef CONFIG_SLOB
8/*
9 * Common fields provided in kmem_cache by all slab allocators
10 * This struct is either used directly by the allocator (SLOB)
11 * or the allocator must include definitions for all fields
12 * provided in kmem_cache_common in their definition of kmem_cache.
13 *
14 * Once we can do anonymous structs (C11 standard) we could put a
15 * anonymous struct definition in these allocators so that the
16 * separate allocations in the kmem_cache structure of SLAB and
17 * SLUB is no longer needed.
18 */
19struct kmem_cache {
20 unsigned int object_size;/* The original size of the object */
21 unsigned int size; /* The aligned/padded/added on size */
22 unsigned int align; /* Alignment as calculated */
23 unsigned long flags; /* Active flags on the slab */
24 const char *name; /* Slab name for sysfs */
25 int refcount; /* Use counter */
26 void (*ctor)(void *); /* Called on object slot creation */
27 struct list_head list; /* List of all slab caches on the system */
28};
29
30#endif /* CONFIG_SLOB */
31
32#ifdef CONFIG_SLAB
33#include <linux/slab_def.h>
34#endif
35
36#ifdef CONFIG_SLUB
37#include <linux/slub_def.h>
38#endif
39
40#include <linux/memcontrol.h>
11c7aec2
JDB
41#include <linux/fault-inject.h>
42#include <linux/kmemcheck.h>
43#include <linux/kasan.h>
44#include <linux/kmemleak.h>
07f361b2 45
97d06609
CL
46/*
47 * State of the slab allocator.
48 *
49 * This is used to describe the states of the allocator during bootup.
50 * Allocators use this to gradually bootstrap themselves. Most allocators
51 * have the problem that the structures used for managing slab caches are
52 * allocated from slab caches themselves.
53 */
54enum slab_state {
55 DOWN, /* No slab functionality yet */
56 PARTIAL, /* SLUB: kmem_cache_node available */
ce8eb6c4 57 PARTIAL_NODE, /* SLAB: kmalloc size for node struct available */
97d06609
CL
58 UP, /* Slab caches usable but not all extras yet */
59 FULL /* Everything is working */
60};
61
62extern enum slab_state slab_state;
63
18004c5d
CL
64/* The slab cache mutex protects the management structures during changes */
65extern struct mutex slab_mutex;
9b030cb8
CL
66
67/* The list of all slab caches on the system */
18004c5d
CL
68extern struct list_head slab_caches;
69
9b030cb8
CL
70/* The slab cache that manages slab cache information */
71extern struct kmem_cache *kmem_cache;
72
45906855
CL
73unsigned long calculate_alignment(unsigned long flags,
74 unsigned long align, unsigned long size);
75
f97d5f63
CL
76#ifndef CONFIG_SLOB
77/* Kmalloc array related functions */
34cc6990 78void setup_kmalloc_cache_index_table(void);
f97d5f63 79void create_kmalloc_caches(unsigned long);
2c59dd65
CL
80
81/* Find the kmalloc slab corresponding for a certain size */
82struct kmem_cache *kmalloc_slab(size_t, gfp_t);
f97d5f63
CL
83#endif
84
85
9b030cb8 86/* Functions provided by the slab allocators */
8a13a4cc 87extern int __kmem_cache_create(struct kmem_cache *, unsigned long flags);
97d06609 88
45530c44
CL
89extern struct kmem_cache *create_kmalloc_cache(const char *name, size_t size,
90 unsigned long flags);
91extern void create_boot_cache(struct kmem_cache *, const char *name,
92 size_t size, unsigned long flags);
93
423c929c
JK
94int slab_unmergeable(struct kmem_cache *s);
95struct kmem_cache *find_mergeable(size_t size, size_t align,
96 unsigned long flags, const char *name, void (*ctor)(void *));
12220dea 97#ifndef CONFIG_SLOB
2633d7a0 98struct kmem_cache *
a44cb944
VD
99__kmem_cache_alias(const char *name, size_t size, size_t align,
100 unsigned long flags, void (*ctor)(void *));
423c929c
JK
101
102unsigned long kmem_cache_flags(unsigned long object_size,
103 unsigned long flags, const char *name,
104 void (*ctor)(void *));
cbb79694 105#else
2633d7a0 106static inline struct kmem_cache *
a44cb944
VD
107__kmem_cache_alias(const char *name, size_t size, size_t align,
108 unsigned long flags, void (*ctor)(void *))
cbb79694 109{ return NULL; }
423c929c
JK
110
111static inline unsigned long kmem_cache_flags(unsigned long object_size,
112 unsigned long flags, const char *name,
113 void (*ctor)(void *))
114{
115 return flags;
116}
cbb79694
CL
117#endif
118
119
d8843922
GC
120/* Legal flag mask for kmem_cache_create(), for various configurations */
121#define SLAB_CORE_FLAGS (SLAB_HWCACHE_ALIGN | SLAB_CACHE_DMA | SLAB_PANIC | \
122 SLAB_DESTROY_BY_RCU | SLAB_DEBUG_OBJECTS )
123
124#if defined(CONFIG_DEBUG_SLAB)
125#define SLAB_DEBUG_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER)
126#elif defined(CONFIG_SLUB_DEBUG)
127#define SLAB_DEBUG_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
becfda68 128 SLAB_TRACE | SLAB_CONSISTENCY_CHECKS)
d8843922
GC
129#else
130#define SLAB_DEBUG_FLAGS (0)
131#endif
132
133#if defined(CONFIG_SLAB)
134#define SLAB_CACHE_FLAGS (SLAB_MEM_SPREAD | SLAB_NOLEAKTRACE | \
230e9fc2
VD
135 SLAB_RECLAIM_ACCOUNT | SLAB_TEMPORARY | \
136 SLAB_NOTRACK | SLAB_ACCOUNT)
d8843922
GC
137#elif defined(CONFIG_SLUB)
138#define SLAB_CACHE_FLAGS (SLAB_NOLEAKTRACE | SLAB_RECLAIM_ACCOUNT | \
230e9fc2 139 SLAB_TEMPORARY | SLAB_NOTRACK | SLAB_ACCOUNT)
d8843922
GC
140#else
141#define SLAB_CACHE_FLAGS (0)
142#endif
143
144#define CACHE_CREATE_MASK (SLAB_CORE_FLAGS | SLAB_DEBUG_FLAGS | SLAB_CACHE_FLAGS)
145
945cf2b6 146int __kmem_cache_shutdown(struct kmem_cache *);
52b4b950 147void __kmem_cache_release(struct kmem_cache *);
d6e0b7fa 148int __kmem_cache_shrink(struct kmem_cache *, bool);
41a21285 149void slab_kmem_cache_release(struct kmem_cache *);
945cf2b6 150
b7454ad3
GC
151struct seq_file;
152struct file;
b7454ad3 153
0d7561c6
GC
154struct slabinfo {
155 unsigned long active_objs;
156 unsigned long num_objs;
157 unsigned long active_slabs;
158 unsigned long num_slabs;
159 unsigned long shared_avail;
160 unsigned int limit;
161 unsigned int batchcount;
162 unsigned int shared;
163 unsigned int objects_per_slab;
164 unsigned int cache_order;
165};
166
167void get_slabinfo(struct kmem_cache *s, struct slabinfo *sinfo);
168void slabinfo_show_stats(struct seq_file *m, struct kmem_cache *s);
b7454ad3
GC
169ssize_t slabinfo_write(struct file *file, const char __user *buffer,
170 size_t count, loff_t *ppos);
ba6c496e 171
484748f0
CL
172/*
173 * Generic implementation of bulk operations
174 * These are useful for situations in which the allocator cannot
9f706d68 175 * perform optimizations. In that case segments of the object listed
484748f0
CL
176 * may be allocated or freed using these operations.
177 */
178void __kmem_cache_free_bulk(struct kmem_cache *, size_t, void **);
865762a8 179int __kmem_cache_alloc_bulk(struct kmem_cache *, gfp_t, size_t, void **);
484748f0 180
127424c8 181#if defined(CONFIG_MEMCG) && !defined(CONFIG_SLOB)
426589f5
VD
182/*
183 * Iterate over all memcg caches of the given root cache. The caller must hold
184 * slab_mutex.
185 */
186#define for_each_memcg_cache(iter, root) \
187 list_for_each_entry(iter, &(root)->memcg_params.list, \
188 memcg_params.list)
189
ba6c496e
GC
190static inline bool is_root_cache(struct kmem_cache *s)
191{
f7ce3190 192 return s->memcg_params.is_root_cache;
ba6c496e 193}
2633d7a0 194
b9ce5ef4 195static inline bool slab_equal_or_root(struct kmem_cache *s,
f7ce3190 196 struct kmem_cache *p)
b9ce5ef4 197{
f7ce3190 198 return p == s || p == s->memcg_params.root_cache;
b9ce5ef4 199}
749c5415
GC
200
201/*
202 * We use suffixes to the name in memcg because we can't have caches
203 * created in the system with the same name. But when we print them
204 * locally, better refer to them with the base name
205 */
206static inline const char *cache_name(struct kmem_cache *s)
207{
208 if (!is_root_cache(s))
f7ce3190 209 s = s->memcg_params.root_cache;
749c5415
GC
210 return s->name;
211}
212
f8570263
VD
213/*
214 * Note, we protect with RCU only the memcg_caches array, not per-memcg caches.
f7ce3190
VD
215 * That said the caller must assure the memcg's cache won't go away by either
216 * taking a css reference to the owner cgroup, or holding the slab_mutex.
f8570263 217 */
2ade4de8
QH
218static inline struct kmem_cache *
219cache_from_memcg_idx(struct kmem_cache *s, int idx)
749c5415 220{
959c8963 221 struct kmem_cache *cachep;
f7ce3190 222 struct memcg_cache_array *arr;
f8570263
VD
223
224 rcu_read_lock();
f7ce3190 225 arr = rcu_dereference(s->memcg_params.memcg_caches);
959c8963
VD
226
227 /*
228 * Make sure we will access the up-to-date value. The code updating
229 * memcg_caches issues a write barrier to match this (see
f7ce3190 230 * memcg_create_kmem_cache()).
959c8963 231 */
f7ce3190 232 cachep = lockless_dereference(arr->entries[idx]);
8df0c2dc
PK
233 rcu_read_unlock();
234
959c8963 235 return cachep;
749c5415 236}
943a451a
GC
237
238static inline struct kmem_cache *memcg_root_cache(struct kmem_cache *s)
239{
240 if (is_root_cache(s))
241 return s;
f7ce3190 242 return s->memcg_params.root_cache;
943a451a 243}
5dfb4175 244
f3ccb2c4
VD
245static __always_inline int memcg_charge_slab(struct page *page,
246 gfp_t gfp, int order,
247 struct kmem_cache *s)
5dfb4175 248{
27ee57c9
VD
249 int ret;
250
5dfb4175
VD
251 if (!memcg_kmem_enabled())
252 return 0;
253 if (is_root_cache(s))
254 return 0;
27ee57c9
VD
255
256 ret = __memcg_kmem_charge_memcg(page, gfp, order,
257 s->memcg_params.memcg);
258 if (ret)
259 return ret;
260
261 memcg_kmem_update_page_stat(page,
262 (s->flags & SLAB_RECLAIM_ACCOUNT) ?
263 MEMCG_SLAB_RECLAIMABLE : MEMCG_SLAB_UNRECLAIMABLE,
264 1 << order);
265 return 0;
266}
267
268static __always_inline void memcg_uncharge_slab(struct page *page, int order,
269 struct kmem_cache *s)
270{
271 memcg_kmem_update_page_stat(page,
272 (s->flags & SLAB_RECLAIM_ACCOUNT) ?
273 MEMCG_SLAB_RECLAIMABLE : MEMCG_SLAB_UNRECLAIMABLE,
274 -(1 << order));
275 memcg_kmem_uncharge(page, order);
5dfb4175 276}
f7ce3190
VD
277
278extern void slab_init_memcg_params(struct kmem_cache *);
279
127424c8 280#else /* CONFIG_MEMCG && !CONFIG_SLOB */
f7ce3190 281
426589f5
VD
282#define for_each_memcg_cache(iter, root) \
283 for ((void)(iter), (void)(root); 0; )
426589f5 284
ba6c496e
GC
285static inline bool is_root_cache(struct kmem_cache *s)
286{
287 return true;
288}
289
b9ce5ef4
GC
290static inline bool slab_equal_or_root(struct kmem_cache *s,
291 struct kmem_cache *p)
292{
293 return true;
294}
749c5415
GC
295
296static inline const char *cache_name(struct kmem_cache *s)
297{
298 return s->name;
299}
300
2ade4de8
QH
301static inline struct kmem_cache *
302cache_from_memcg_idx(struct kmem_cache *s, int idx)
749c5415
GC
303{
304 return NULL;
305}
943a451a
GC
306
307static inline struct kmem_cache *memcg_root_cache(struct kmem_cache *s)
308{
309 return s;
310}
5dfb4175 311
f3ccb2c4
VD
312static inline int memcg_charge_slab(struct page *page, gfp_t gfp, int order,
313 struct kmem_cache *s)
5dfb4175
VD
314{
315 return 0;
316}
317
27ee57c9
VD
318static inline void memcg_uncharge_slab(struct page *page, int order,
319 struct kmem_cache *s)
320{
321}
322
f7ce3190
VD
323static inline void slab_init_memcg_params(struct kmem_cache *s)
324{
325}
127424c8 326#endif /* CONFIG_MEMCG && !CONFIG_SLOB */
b9ce5ef4
GC
327
328static inline struct kmem_cache *cache_from_obj(struct kmem_cache *s, void *x)
329{
330 struct kmem_cache *cachep;
331 struct page *page;
332
333 /*
334 * When kmemcg is not being used, both assignments should return the
335 * same value. but we don't want to pay the assignment price in that
336 * case. If it is not compiled in, the compiler should be smart enough
337 * to not do even the assignment. In that case, slab_equal_or_root
338 * will also be a constant.
339 */
becfda68
LA
340 if (!memcg_kmem_enabled() &&
341 !unlikely(s->flags & SLAB_CONSISTENCY_CHECKS))
b9ce5ef4
GC
342 return s;
343
344 page = virt_to_head_page(x);
345 cachep = page->slab_cache;
346 if (slab_equal_or_root(cachep, s))
347 return cachep;
348
349 pr_err("%s: Wrong slab cache. %s but object is from %s\n",
2d16e0fd 350 __func__, s->name, cachep->name);
b9ce5ef4
GC
351 WARN_ON_ONCE(1);
352 return s;
353}
ca34956b 354
11c7aec2
JDB
355static inline size_t slab_ksize(const struct kmem_cache *s)
356{
357#ifndef CONFIG_SLUB
358 return s->object_size;
359
360#else /* CONFIG_SLUB */
361# ifdef CONFIG_SLUB_DEBUG
362 /*
363 * Debugging requires use of the padding between object
364 * and whatever may come after it.
365 */
366 if (s->flags & (SLAB_RED_ZONE | SLAB_POISON))
367 return s->object_size;
368# endif
369 /*
370 * If we have the need to store the freelist pointer
371 * back there or track user information then we can
372 * only use the space before that information.
373 */
374 if (s->flags & (SLAB_DESTROY_BY_RCU | SLAB_STORE_USER))
375 return s->inuse;
376 /*
377 * Else we can use all the padding etc for the allocation
378 */
379 return s->size;
380#endif
381}
382
383static inline struct kmem_cache *slab_pre_alloc_hook(struct kmem_cache *s,
384 gfp_t flags)
385{
386 flags &= gfp_allowed_mask;
387 lockdep_trace_alloc(flags);
388 might_sleep_if(gfpflags_allow_blocking(flags));
389
fab9963a 390 if (should_failslab(s, flags))
11c7aec2
JDB
391 return NULL;
392
393 return memcg_kmem_get_cache(s, flags);
394}
395
396static inline void slab_post_alloc_hook(struct kmem_cache *s, gfp_t flags,
397 size_t size, void **p)
398{
399 size_t i;
400
401 flags &= gfp_allowed_mask;
402 for (i = 0; i < size; i++) {
403 void *object = p[i];
404
405 kmemcheck_slab_alloc(s, flags, object, slab_ksize(s));
406 kmemleak_alloc_recursive(object, s->object_size, 1,
407 s->flags, flags);
505f5dcb 408 kasan_slab_alloc(s, object, flags);
11c7aec2
JDB
409 }
410 memcg_kmem_put_cache(s);
411}
412
44c5356f 413#ifndef CONFIG_SLOB
ca34956b
CL
414/*
415 * The slab lists for all objects.
416 */
417struct kmem_cache_node {
418 spinlock_t list_lock;
419
420#ifdef CONFIG_SLAB
421 struct list_head slabs_partial; /* partial list first, better asm code */
422 struct list_head slabs_full;
423 struct list_head slabs_free;
424 unsigned long free_objects;
425 unsigned int free_limit;
426 unsigned int colour_next; /* Per-node cache coloring */
427 struct array_cache *shared; /* shared per node */
c8522a3a 428 struct alien_cache **alien; /* on other nodes */
ca34956b
CL
429 unsigned long next_reap; /* updated without locking */
430 int free_touched; /* updated without locking */
431#endif
432
433#ifdef CONFIG_SLUB
434 unsigned long nr_partial;
435 struct list_head partial;
436#ifdef CONFIG_SLUB_DEBUG
437 atomic_long_t nr_slabs;
438 atomic_long_t total_objects;
439 struct list_head full;
440#endif
441#endif
442
443};
e25839f6 444
44c5356f
CL
445static inline struct kmem_cache_node *get_node(struct kmem_cache *s, int node)
446{
447 return s->node[node];
448}
449
450/*
451 * Iterator over all nodes. The body will be executed for each node that has
452 * a kmem_cache_node structure allocated (which is true for all online nodes)
453 */
454#define for_each_kmem_cache_node(__s, __node, __n) \
9163582c
MP
455 for (__node = 0; __node < nr_node_ids; __node++) \
456 if ((__n = get_node(__s, __node)))
44c5356f
CL
457
458#endif
459
1df3b26f 460void *slab_start(struct seq_file *m, loff_t *pos);
276a2439
WL
461void *slab_next(struct seq_file *m, void *p, loff_t *pos);
462void slab_stop(struct seq_file *m, void *p);
b047501c 463int memcg_slab_show(struct seq_file *m, void *p);
5240ab40 464
55834c59
AP
465void ___cache_free(struct kmem_cache *cache, void *x, unsigned long addr);
466
5240ab40 467#endif /* MM_SLAB_H */