blob: 73004808537ea841e05c85e0b68312ea645eb114 [file] [log] [blame]
Christoph Lameter81819f02007-05-06 14:49:36 -07001/*
2 * SLUB: A slab allocator that limits cache line use instead of queuing
3 * objects in per cpu and per node lists.
4 *
Christoph Lameter881db7f2011-06-01 12:25:53 -05005 * The allocator synchronizes using per slab locks or atomic operatios
6 * and only uses a centralized lock to manage a pool of partial slabs.
Christoph Lameter81819f02007-05-06 14:49:36 -07007 *
Christoph Lametercde53532008-07-04 09:59:22 -07008 * (C) 2007 SGI, Christoph Lameter
Christoph Lameter881db7f2011-06-01 12:25:53 -05009 * (C) 2011 Linux Foundation, Christoph Lameter
Christoph Lameter81819f02007-05-06 14:49:36 -070010 */
11
12#include <linux/mm.h>
Nick Piggin1eb5ac62009-05-05 19:13:44 +100013#include <linux/swap.h> /* struct reclaim_state */
Christoph Lameter81819f02007-05-06 14:49:36 -070014#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>
Christoph Lameter97d06602012-07-06 15:25:11 -050019#include "slab.h"
Alexey Dobriyan7b3c3a52008-10-06 02:42:17 +040020#include <linux/proc_fs.h>
Andrew Morton3ac38fa2013-04-29 15:08:06 -070021#include <linux/notifier.h>
Christoph Lameter81819f02007-05-06 14:49:36 -070022#include <linux/seq_file.h>
Vegard Nossum5a896d92008-04-04 00:54:48 +020023#include <linux/kmemcheck.h>
Christoph Lameter81819f02007-05-06 14:49:36 -070024#include <linux/cpu.h>
25#include <linux/cpuset.h>
26#include <linux/mempolicy.h>
27#include <linux/ctype.h>
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -070028#include <linux/debugobjects.h>
Christoph Lameter81819f02007-05-06 14:49:36 -070029#include <linux/kallsyms.h>
Yasunori Gotob9049e22007-10-21 16:41:37 -070030#include <linux/memory.h>
Roman Zippelf8bd2252008-05-01 04:34:31 -070031#include <linux/math64.h>
Akinobu Mita773ff602008-12-23 19:37:01 +090032#include <linux/fault-inject.h>
Pekka Enbergbfa71452011-07-07 22:47:01 +030033#include <linux/stacktrace.h>
Christoph Lameter4de900b2012-01-30 15:53:51 -060034#include <linux/prefetch.h>
Glauber Costa2633d7a2012-12-18 14:22:34 -080035#include <linux/memcontrol.h>
Christoph Lameter81819f02007-05-06 14:49:36 -070036
Richard Kennedy4a923792010-10-21 10:29:19 +010037#include <trace/events/kmem.h>
38
Mel Gorman072bb0a2012-07-31 16:43:58 -070039#include "internal.h"
40
Christoph Lameter81819f02007-05-06 14:49:36 -070041/*
42 * Lock order:
Christoph Lameter18004c52012-07-06 15:25:12 -050043 * 1. slab_mutex (Global Mutex)
Christoph Lameter881db7f2011-06-01 12:25:53 -050044 * 2. node->list_lock
45 * 3. slab_lock(page) (Only on some arches and for debugging)
Christoph Lameter81819f02007-05-06 14:49:36 -070046 *
Christoph Lameter18004c52012-07-06 15:25:12 -050047 * slab_mutex
Christoph Lameter881db7f2011-06-01 12:25:53 -050048 *
Christoph Lameter18004c52012-07-06 15:25:12 -050049 * The role of the slab_mutex is to protect the list of all the slabs
Christoph Lameter881db7f2011-06-01 12:25:53 -050050 * and to synchronize major metadata changes to slab cache structures.
51 *
52 * The slab_lock is only used for debugging and on arches that do not
53 * have the ability to do a cmpxchg_double. It only protects the second
54 * double word in the page struct. Meaning
55 * A. page->freelist -> List of object free in a page
56 * B. page->counters -> Counters of objects
57 * C. page->frozen -> frozen state
58 *
59 * If a slab is frozen then it is exempt from list management. It is not
60 * on any list. The processor that froze the slab is the one who can
61 * perform list operations on the page. Other processors may put objects
62 * onto the freelist but the processor that froze the slab is the only
63 * one that can retrieve the objects from the page's freelist.
Christoph Lameter81819f02007-05-06 14:49:36 -070064 *
65 * The list_lock protects the partial and full list on each node and
66 * the partial slab counter. If taken then no new slabs may be added or
67 * removed from the lists nor make the number of partial slabs be modified.
68 * (Note that the total number of slabs is an atomic value that may be
69 * modified without taking the list lock).
70 *
71 * The list_lock is a centralized lock and thus we avoid taking it as
72 * much as possible. As long as SLUB does not have to handle partial
73 * slabs, operations can continue without any centralized lock. F.e.
74 * allocating a long series of objects that fill up slabs does not require
75 * the list lock.
Christoph Lameter81819f02007-05-06 14:49:36 -070076 * Interrupts are disabled during allocation and deallocation in order to
77 * make the slab allocator safe to use in the context of an irq. In addition
78 * interrupts are disabled to ensure that the processor does not change
79 * while handling per_cpu slabs, due to kernel preemption.
80 *
81 * SLUB assigns one slab for allocation to each processor.
82 * Allocations only occur from these slabs called cpu slabs.
83 *
Christoph Lameter672bba32007-05-09 02:32:39 -070084 * Slabs with free elements are kept on a partial list and during regular
85 * operations no list for full slabs is used. If an object in a full slab is
Christoph Lameter81819f02007-05-06 14:49:36 -070086 * freed then the slab will show up again on the partial lists.
Christoph Lameter672bba32007-05-09 02:32:39 -070087 * We track full slabs for debugging purposes though because otherwise we
88 * cannot scan all objects.
Christoph Lameter81819f02007-05-06 14:49:36 -070089 *
90 * Slabs are freed when they become empty. Teardown and setup is
91 * minimal so we rely on the page allocators per cpu caches for
92 * fast frees and allocs.
93 *
94 * Overloading of page flags that are otherwise used for LRU management.
95 *
Christoph Lameter4b6f0752007-05-16 22:10:53 -070096 * PageActive The slab is frozen and exempt from list processing.
97 * This means that the slab is dedicated to a purpose
98 * such as satisfying allocations for a specific
99 * processor. Objects may be freed in the slab while
100 * it is frozen but slab_free will then skip the usual
101 * list operations. It is up to the processor holding
102 * the slab to integrate the slab into the slab lists
103 * when the slab is no longer needed.
104 *
105 * One use of this flag is to mark slabs that are
106 * used for allocations. Then such a slab becomes a cpu
107 * slab. The cpu slab may be equipped with an additional
Christoph Lameterdfb4f092007-10-16 01:26:05 -0700108 * freelist that allows lockless access to
Christoph Lameter894b8782007-05-10 03:15:16 -0700109 * free objects in addition to the regular freelist
110 * that requires the slab lock.
Christoph Lameter81819f02007-05-06 14:49:36 -0700111 *
112 * PageError Slab requires special handling due to debug
113 * options set. This moves slab handling out of
Christoph Lameter894b8782007-05-10 03:15:16 -0700114 * the fast path and disables lockless freelists.
Christoph Lameter81819f02007-05-06 14:49:36 -0700115 */
116
Christoph Lameteraf537b02010-07-09 14:07:14 -0500117static inline int kmem_cache_debug(struct kmem_cache *s)
118{
Christoph Lameter5577bd82007-05-16 22:10:56 -0700119#ifdef CONFIG_SLUB_DEBUG
Christoph Lameteraf537b02010-07-09 14:07:14 -0500120 return unlikely(s->flags & SLAB_DEBUG_FLAGS);
Christoph Lameter5577bd82007-05-16 22:10:56 -0700121#else
Christoph Lameteraf537b02010-07-09 14:07:14 -0500122 return 0;
Christoph Lameter5577bd82007-05-16 22:10:56 -0700123#endif
Christoph Lameteraf537b02010-07-09 14:07:14 -0500124}
Christoph Lameter5577bd82007-05-16 22:10:56 -0700125
Joonsoo Kim345c9052013-06-19 14:05:52 +0900126static inline bool kmem_cache_has_cpu_partial(struct kmem_cache *s)
127{
128#ifdef CONFIG_SLUB_CPU_PARTIAL
129 return !kmem_cache_debug(s);
130#else
131 return false;
132#endif
133}
134
Christoph Lameter81819f02007-05-06 14:49:36 -0700135/*
136 * Issues still to be resolved:
137 *
Christoph Lameter81819f02007-05-06 14:49:36 -0700138 * - Support PAGE_ALLOC_DEBUG. Should be easy to do.
139 *
Christoph Lameter81819f02007-05-06 14:49:36 -0700140 * - Variable sizing of the per node arrays
141 */
142
143/* Enable to test recovery from slab corruption on boot */
144#undef SLUB_RESILIENCY_TEST
145
Christoph Lameterb789ef52011-06-01 12:25:49 -0500146/* Enable to log cmpxchg failures */
147#undef SLUB_DEBUG_CMPXCHG
148
Christoph Lameter81819f02007-05-06 14:49:36 -0700149/*
Christoph Lameter2086d262007-05-06 14:49:46 -0700150 * Mininum number of partial slabs. These will be left on the partial
151 * lists even if they are empty. kmem_cache_shrink may reclaim them.
152 */
Christoph Lameter76be8952007-12-21 14:37:37 -0800153#define MIN_PARTIAL 5
Christoph Lametere95eed52007-05-06 14:49:44 -0700154
Christoph Lameter2086d262007-05-06 14:49:46 -0700155/*
156 * Maximum number of desirable partial slabs.
157 * The existence of more partial slabs makes kmem_cache_shrink
Zhi Yong Wu721ae222013-11-08 20:47:37 +0800158 * sort the partial list by the number of objects in use.
Christoph Lameter2086d262007-05-06 14:49:46 -0700159 */
160#define MAX_PARTIAL 10
161
Christoph Lameter81819f02007-05-06 14:49:36 -0700162#define DEBUG_DEFAULT_FLAGS (SLAB_DEBUG_FREE | SLAB_RED_ZONE | \
163 SLAB_POISON | SLAB_STORE_USER)
Christoph Lameter672bba32007-05-09 02:32:39 -0700164
Christoph Lameter81819f02007-05-06 14:49:36 -0700165/*
David Rientjes3de47212009-07-27 18:30:35 -0700166 * Debugging flags that require metadata to be stored in the slab. These get
167 * disabled when slub_debug=O is used and a cache's min order increases with
168 * metadata.
David Rientjesfa5ec8a2009-07-07 00:14:14 -0700169 */
David Rientjes3de47212009-07-27 18:30:35 -0700170#define DEBUG_METADATA_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER)
David Rientjesfa5ec8a2009-07-07 00:14:14 -0700171
172/*
Christoph Lameter81819f02007-05-06 14:49:36 -0700173 * Set of flags that will prevent slab merging
174 */
175#define SLUB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
Dmitry Monakhov4c13dd32010-02-26 09:36:12 +0300176 SLAB_TRACE | SLAB_DESTROY_BY_RCU | SLAB_NOLEAKTRACE | \
177 SLAB_FAILSLAB)
Christoph Lameter81819f02007-05-06 14:49:36 -0700178
179#define SLUB_MERGE_SAME (SLAB_DEBUG_FREE | SLAB_RECLAIM_ACCOUNT | \
Vegard Nossum5a896d92008-04-04 00:54:48 +0200180 SLAB_CACHE_DMA | SLAB_NOTRACK)
Christoph Lameter81819f02007-05-06 14:49:36 -0700181
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +0400182#define OO_SHIFT 16
183#define OO_MASK ((1 << OO_SHIFT) - 1)
Christoph Lameter50d5c412011-06-01 12:25:45 -0500184#define MAX_OBJS_PER_PAGE 32767 /* since page.objects is u15 */
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +0400185
Christoph Lameter81819f02007-05-06 14:49:36 -0700186/* Internal SLUB flags */
Christoph Lameterf90ec392010-07-09 14:07:11 -0500187#define __OBJECT_POISON 0x80000000UL /* Poison object */
Christoph Lameterb789ef52011-06-01 12:25:49 -0500188#define __CMPXCHG_DOUBLE 0x40000000UL /* Use cmpxchg_double */
Christoph Lameter81819f02007-05-06 14:49:36 -0700189
Christoph Lameter81819f02007-05-06 14:49:36 -0700190#ifdef CONFIG_SMP
191static struct notifier_block slab_notifier;
192#endif
193
Christoph Lameter02cbc872007-05-09 02:32:43 -0700194/*
195 * Tracking user of a slab.
196 */
Ben Greeard6543e32011-07-07 11:36:36 -0700197#define TRACK_ADDRS_COUNT 16
Christoph Lameter02cbc872007-05-09 02:32:43 -0700198struct track {
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300199 unsigned long addr; /* Called from address */
Ben Greeard6543e32011-07-07 11:36:36 -0700200#ifdef CONFIG_STACKTRACE
201 unsigned long addrs[TRACK_ADDRS_COUNT]; /* Called from address */
202#endif
Christoph Lameter02cbc872007-05-09 02:32:43 -0700203 int cpu; /* Was running on cpu */
204 int pid; /* Pid context */
205 unsigned long when; /* When did the operation occur */
206};
207
208enum track_item { TRACK_ALLOC, TRACK_FREE };
209
Christoph Lameterab4d5ed2010-10-05 13:57:26 -0500210#ifdef CONFIG_SYSFS
Christoph Lameter81819f02007-05-06 14:49:36 -0700211static int sysfs_slab_add(struct kmem_cache *);
212static int sysfs_slab_alias(struct kmem_cache *, const char *);
Glauber Costa107dab52012-12-18 14:23:05 -0800213static void memcg_propagate_slab_attrs(struct kmem_cache *s);
Christoph Lameter81819f02007-05-06 14:49:36 -0700214#else
Christoph Lameter0c710012007-07-17 04:03:24 -0700215static inline int sysfs_slab_add(struct kmem_cache *s) { return 0; }
216static inline int sysfs_slab_alias(struct kmem_cache *s, const char *p)
217 { return 0; }
Glauber Costa107dab52012-12-18 14:23:05 -0800218static inline void memcg_propagate_slab_attrs(struct kmem_cache *s) { }
Christoph Lameter81819f02007-05-06 14:49:36 -0700219#endif
220
Christoph Lameter4fdccdf2011-03-22 13:35:00 -0500221static inline void stat(const struct kmem_cache *s, enum stat_item si)
Christoph Lameter8ff12cf2008-02-07 17:47:41 -0800222{
223#ifdef CONFIG_SLUB_STATS
Christoph Lameter88da03a2014-04-07 15:39:42 -0700224 /*
225 * The rmw is racy on a preemptible kernel but this is acceptable, so
226 * avoid this_cpu_add()'s irq-disable overhead.
227 */
228 raw_cpu_inc(s->cpu_slab->stat[si]);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -0800229#endif
230}
231
Christoph Lameter81819f02007-05-06 14:49:36 -0700232/********************************************************************
233 * Core slab cache functions
234 *******************************************************************/
235
Christoph Lameter81819f02007-05-06 14:49:36 -0700236static inline struct kmem_cache_node *get_node(struct kmem_cache *s, int node)
237{
Christoph Lameter81819f02007-05-06 14:49:36 -0700238 return s->node[node];
Christoph Lameter81819f02007-05-06 14:49:36 -0700239}
240
Christoph Lameter6446faa2008-02-15 23:45:26 -0800241/* Verify that a pointer has an address that is valid within a slab page */
Christoph Lameter02cbc872007-05-09 02:32:43 -0700242static inline int check_valid_pointer(struct kmem_cache *s,
243 struct page *page, const void *object)
244{
245 void *base;
246
Christoph Lametera973e9d2008-03-01 13:40:44 -0800247 if (!object)
Christoph Lameter02cbc872007-05-09 02:32:43 -0700248 return 1;
249
Christoph Lametera973e9d2008-03-01 13:40:44 -0800250 base = page_address(page);
Christoph Lameter39b26462008-04-14 19:11:30 +0300251 if (object < base || object >= base + page->objects * s->size ||
Christoph Lameter02cbc872007-05-09 02:32:43 -0700252 (object - base) % s->size) {
253 return 0;
254 }
255
256 return 1;
257}
258
Christoph Lameter7656c722007-05-09 02:32:40 -0700259static inline void *get_freepointer(struct kmem_cache *s, void *object)
260{
261 return *(void **)(object + s->offset);
262}
263
Eric Dumazet0ad95002011-12-16 16:25:34 +0100264static void prefetch_freepointer(const struct kmem_cache *s, void *object)
265{
266 prefetch(object + s->offset);
267}
268
Christoph Lameter1393d9a2011-05-16 15:26:08 -0500269static inline void *get_freepointer_safe(struct kmem_cache *s, void *object)
270{
271 void *p;
272
273#ifdef CONFIG_DEBUG_PAGEALLOC
274 probe_kernel_read(&p, (void **)(object + s->offset), sizeof(p));
275#else
276 p = get_freepointer(s, object);
277#endif
278 return p;
279}
280
Christoph Lameter7656c722007-05-09 02:32:40 -0700281static inline void set_freepointer(struct kmem_cache *s, void *object, void *fp)
282{
283 *(void **)(object + s->offset) = fp;
284}
285
286/* Loop over all objects in a slab */
Christoph Lameter224a88b2008-04-14 19:11:31 +0300287#define for_each_object(__p, __s, __addr, __objects) \
288 for (__p = (__addr); __p < (__addr) + (__objects) * (__s)->size;\
Christoph Lameter7656c722007-05-09 02:32:40 -0700289 __p += (__s)->size)
290
Christoph Lameter7656c722007-05-09 02:32:40 -0700291/* Determine object index from a given position */
292static inline int slab_index(void *p, struct kmem_cache *s, void *addr)
293{
294 return (p - addr) / s->size;
295}
296
Mariusz Kozlowskid71f6062011-02-26 20:10:26 +0100297static inline size_t slab_ksize(const struct kmem_cache *s)
298{
299#ifdef CONFIG_SLUB_DEBUG
300 /*
301 * Debugging requires use of the padding between object
302 * and whatever may come after it.
303 */
304 if (s->flags & (SLAB_RED_ZONE | SLAB_POISON))
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500305 return s->object_size;
Mariusz Kozlowskid71f6062011-02-26 20:10:26 +0100306
307#endif
308 /*
309 * If we have the need to store the freelist pointer
310 * back there or track user information then we can
311 * only use the space before that information.
312 */
313 if (s->flags & (SLAB_DESTROY_BY_RCU | SLAB_STORE_USER))
314 return s->inuse;
315 /*
316 * Else we can use all the padding etc for the allocation
317 */
318 return s->size;
319}
320
Lai Jiangshanab9a0f12011-03-10 15:21:48 +0800321static inline int order_objects(int order, unsigned long size, int reserved)
322{
323 return ((PAGE_SIZE << order) - reserved) / size;
324}
325
Christoph Lameter834f3d12008-04-14 19:11:31 +0300326static inline struct kmem_cache_order_objects oo_make(int order,
Lai Jiangshanab9a0f12011-03-10 15:21:48 +0800327 unsigned long size, int reserved)
Christoph Lameter834f3d12008-04-14 19:11:31 +0300328{
329 struct kmem_cache_order_objects x = {
Lai Jiangshanab9a0f12011-03-10 15:21:48 +0800330 (order << OO_SHIFT) + order_objects(order, size, reserved)
Christoph Lameter834f3d12008-04-14 19:11:31 +0300331 };
332
333 return x;
334}
335
336static inline int oo_order(struct kmem_cache_order_objects x)
337{
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +0400338 return x.x >> OO_SHIFT;
Christoph Lameter834f3d12008-04-14 19:11:31 +0300339}
340
341static inline int oo_objects(struct kmem_cache_order_objects x)
342{
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +0400343 return x.x & OO_MASK;
Christoph Lameter834f3d12008-04-14 19:11:31 +0300344}
345
Christoph Lameter881db7f2011-06-01 12:25:53 -0500346/*
347 * Per slab locking using the pagelock
348 */
349static __always_inline void slab_lock(struct page *page)
350{
351 bit_spin_lock(PG_locked, &page->flags);
352}
353
354static __always_inline void slab_unlock(struct page *page)
355{
356 __bit_spin_unlock(PG_locked, &page->flags);
357}
358
Dave Hansena0320862014-01-30 15:46:09 -0800359static inline void set_page_slub_counters(struct page *page, unsigned long counters_new)
360{
361 struct page tmp;
362 tmp.counters = counters_new;
363 /*
364 * page->counters can cover frozen/inuse/objects as well
365 * as page->_count. If we assign to ->counters directly
366 * we run the risk of losing updates to page->_count, so
367 * be careful and only assign to the fields we need.
368 */
369 page->frozen = tmp.frozen;
370 page->inuse = tmp.inuse;
371 page->objects = tmp.objects;
372}
373
Christoph Lameter1d071712011-07-14 12:49:12 -0500374/* Interrupts must be disabled (for the fallback code to work right) */
375static inline bool __cmpxchg_double_slab(struct kmem_cache *s, struct page *page,
Christoph Lameterb789ef52011-06-01 12:25:49 -0500376 void *freelist_old, unsigned long counters_old,
377 void *freelist_new, unsigned long counters_new,
378 const char *n)
379{
Christoph Lameter1d071712011-07-14 12:49:12 -0500380 VM_BUG_ON(!irqs_disabled());
Heiko Carstens25654092012-01-12 17:17:33 -0800381#if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \
382 defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
Christoph Lameterb789ef52011-06-01 12:25:49 -0500383 if (s->flags & __CMPXCHG_DOUBLE) {
Jan Beulichcdcd6292012-01-02 17:02:18 +0000384 if (cmpxchg_double(&page->freelist, &page->counters,
Christoph Lameterb789ef52011-06-01 12:25:49 -0500385 freelist_old, counters_old,
386 freelist_new, counters_new))
387 return 1;
388 } else
389#endif
390 {
Christoph Lameter881db7f2011-06-01 12:25:53 -0500391 slab_lock(page);
Chen Gangd0e0ac92013-07-15 09:05:29 +0800392 if (page->freelist == freelist_old &&
393 page->counters == counters_old) {
Christoph Lameterb789ef52011-06-01 12:25:49 -0500394 page->freelist = freelist_new;
Dave Hansena0320862014-01-30 15:46:09 -0800395 set_page_slub_counters(page, counters_new);
Christoph Lameter881db7f2011-06-01 12:25:53 -0500396 slab_unlock(page);
Christoph Lameterb789ef52011-06-01 12:25:49 -0500397 return 1;
398 }
Christoph Lameter881db7f2011-06-01 12:25:53 -0500399 slab_unlock(page);
Christoph Lameterb789ef52011-06-01 12:25:49 -0500400 }
401
402 cpu_relax();
403 stat(s, CMPXCHG_DOUBLE_FAIL);
404
405#ifdef SLUB_DEBUG_CMPXCHG
Fabian Frederickf9f58282014-06-04 16:06:34 -0700406 pr_info("%s %s: cmpxchg double redo ", n, s->name);
Christoph Lameterb789ef52011-06-01 12:25:49 -0500407#endif
408
409 return 0;
410}
411
Christoph Lameter1d071712011-07-14 12:49:12 -0500412static inline bool cmpxchg_double_slab(struct kmem_cache *s, struct page *page,
413 void *freelist_old, unsigned long counters_old,
414 void *freelist_new, unsigned long counters_new,
415 const char *n)
416{
Heiko Carstens25654092012-01-12 17:17:33 -0800417#if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \
418 defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
Christoph Lameter1d071712011-07-14 12:49:12 -0500419 if (s->flags & __CMPXCHG_DOUBLE) {
Jan Beulichcdcd6292012-01-02 17:02:18 +0000420 if (cmpxchg_double(&page->freelist, &page->counters,
Christoph Lameter1d071712011-07-14 12:49:12 -0500421 freelist_old, counters_old,
422 freelist_new, counters_new))
423 return 1;
424 } else
425#endif
426 {
427 unsigned long flags;
428
429 local_irq_save(flags);
430 slab_lock(page);
Chen Gangd0e0ac92013-07-15 09:05:29 +0800431 if (page->freelist == freelist_old &&
432 page->counters == counters_old) {
Christoph Lameter1d071712011-07-14 12:49:12 -0500433 page->freelist = freelist_new;
Dave Hansena0320862014-01-30 15:46:09 -0800434 set_page_slub_counters(page, counters_new);
Christoph Lameter1d071712011-07-14 12:49:12 -0500435 slab_unlock(page);
436 local_irq_restore(flags);
437 return 1;
438 }
439 slab_unlock(page);
440 local_irq_restore(flags);
441 }
442
443 cpu_relax();
444 stat(s, CMPXCHG_DOUBLE_FAIL);
445
446#ifdef SLUB_DEBUG_CMPXCHG
Fabian Frederickf9f58282014-06-04 16:06:34 -0700447 pr_info("%s %s: cmpxchg double redo ", n, s->name);
Christoph Lameter1d071712011-07-14 12:49:12 -0500448#endif
449
450 return 0;
451}
452
Christoph Lameter41ecc552007-05-09 02:32:44 -0700453#ifdef CONFIG_SLUB_DEBUG
454/*
Christoph Lameter5f80b132011-04-15 14:48:13 -0500455 * Determine a map of object in use on a page.
456 *
Christoph Lameter881db7f2011-06-01 12:25:53 -0500457 * Node listlock must be held to guarantee that the page does
Christoph Lameter5f80b132011-04-15 14:48:13 -0500458 * not vanish from under us.
459 */
460static void get_map(struct kmem_cache *s, struct page *page, unsigned long *map)
461{
462 void *p;
463 void *addr = page_address(page);
464
465 for (p = page->freelist; p; p = get_freepointer(s, p))
466 set_bit(slab_index(p, s, addr), map);
467}
468
Christoph Lameter41ecc552007-05-09 02:32:44 -0700469/*
470 * Debug settings:
471 */
Christoph Lameterf0630ff2007-07-15 23:38:14 -0700472#ifdef CONFIG_SLUB_DEBUG_ON
473static int slub_debug = DEBUG_DEFAULT_FLAGS;
474#else
Christoph Lameter41ecc552007-05-09 02:32:44 -0700475static int slub_debug;
Christoph Lameterf0630ff2007-07-15 23:38:14 -0700476#endif
Christoph Lameter41ecc552007-05-09 02:32:44 -0700477
478static char *slub_debug_slabs;
David Rientjesfa5ec8a2009-07-07 00:14:14 -0700479static int disable_higher_order_debug;
Christoph Lameter41ecc552007-05-09 02:32:44 -0700480
Christoph Lameter7656c722007-05-09 02:32:40 -0700481/*
Christoph Lameter81819f02007-05-06 14:49:36 -0700482 * Object debugging
483 */
484static void print_section(char *text, u8 *addr, unsigned int length)
485{
Sebastian Andrzej Siewiorffc79d22011-07-29 14:10:20 +0200486 print_hex_dump(KERN_ERR, text, DUMP_PREFIX_ADDRESS, 16, 1, addr,
487 length, 1);
Christoph Lameter81819f02007-05-06 14:49:36 -0700488}
489
Christoph Lameter81819f02007-05-06 14:49:36 -0700490static struct track *get_track(struct kmem_cache *s, void *object,
491 enum track_item alloc)
492{
493 struct track *p;
494
495 if (s->offset)
496 p = object + s->offset + sizeof(void *);
497 else
498 p = object + s->inuse;
499
500 return p + alloc;
501}
502
503static void set_track(struct kmem_cache *s, void *object,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300504 enum track_item alloc, unsigned long addr)
Christoph Lameter81819f02007-05-06 14:49:36 -0700505{
Akinobu Mita1a00df42009-03-07 00:36:21 +0900506 struct track *p = get_track(s, object, alloc);
Christoph Lameter81819f02007-05-06 14:49:36 -0700507
Christoph Lameter81819f02007-05-06 14:49:36 -0700508 if (addr) {
Ben Greeard6543e32011-07-07 11:36:36 -0700509#ifdef CONFIG_STACKTRACE
510 struct stack_trace trace;
511 int i;
512
513 trace.nr_entries = 0;
514 trace.max_entries = TRACK_ADDRS_COUNT;
515 trace.entries = p->addrs;
516 trace.skip = 3;
517 save_stack_trace(&trace);
518
519 /* See rant in lockdep.c */
520 if (trace.nr_entries != 0 &&
521 trace.entries[trace.nr_entries - 1] == ULONG_MAX)
522 trace.nr_entries--;
523
524 for (i = trace.nr_entries; i < TRACK_ADDRS_COUNT; i++)
525 p->addrs[i] = 0;
526#endif
Christoph Lameter81819f02007-05-06 14:49:36 -0700527 p->addr = addr;
528 p->cpu = smp_processor_id();
Alexey Dobriyan88e4ccf2008-06-23 02:58:37 +0400529 p->pid = current->pid;
Christoph Lameter81819f02007-05-06 14:49:36 -0700530 p->when = jiffies;
531 } else
532 memset(p, 0, sizeof(struct track));
533}
534
Christoph Lameter81819f02007-05-06 14:49:36 -0700535static void init_tracking(struct kmem_cache *s, void *object)
536{
Christoph Lameter24922682007-07-17 04:03:18 -0700537 if (!(s->flags & SLAB_STORE_USER))
538 return;
539
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300540 set_track(s, object, TRACK_FREE, 0UL);
541 set_track(s, object, TRACK_ALLOC, 0UL);
Christoph Lameter81819f02007-05-06 14:49:36 -0700542}
543
544static void print_track(const char *s, struct track *t)
545{
546 if (!t->addr)
547 return;
548
Fabian Frederickf9f58282014-06-04 16:06:34 -0700549 pr_err("INFO: %s in %pS age=%lu cpu=%u pid=%d\n",
550 s, (void *)t->addr, jiffies - t->when, t->cpu, t->pid);
Ben Greeard6543e32011-07-07 11:36:36 -0700551#ifdef CONFIG_STACKTRACE
552 {
553 int i;
554 for (i = 0; i < TRACK_ADDRS_COUNT; i++)
555 if (t->addrs[i])
Fabian Frederickf9f58282014-06-04 16:06:34 -0700556 pr_err("\t%pS\n", (void *)t->addrs[i]);
Ben Greeard6543e32011-07-07 11:36:36 -0700557 else
558 break;
559 }
560#endif
Christoph Lameter81819f02007-05-06 14:49:36 -0700561}
562
Christoph Lameter24922682007-07-17 04:03:18 -0700563static void print_tracking(struct kmem_cache *s, void *object)
564{
565 if (!(s->flags & SLAB_STORE_USER))
566 return;
567
568 print_track("Allocated", get_track(s, object, TRACK_ALLOC));
569 print_track("Freed", get_track(s, object, TRACK_FREE));
570}
571
572static void print_page_info(struct page *page)
573{
Fabian Frederickf9f58282014-06-04 16:06:34 -0700574 pr_err("INFO: Slab 0x%p objects=%u used=%u fp=0x%p flags=0x%04lx\n",
Chen Gangd0e0ac92013-07-15 09:05:29 +0800575 page, page->objects, page->inuse, page->freelist, page->flags);
Christoph Lameter24922682007-07-17 04:03:18 -0700576
577}
578
579static void slab_bug(struct kmem_cache *s, char *fmt, ...)
580{
Fabian Frederickecc42fb2014-06-04 16:06:35 -0700581 struct va_format vaf;
Christoph Lameter24922682007-07-17 04:03:18 -0700582 va_list args;
Christoph Lameter24922682007-07-17 04:03:18 -0700583
584 va_start(args, fmt);
Fabian Frederickecc42fb2014-06-04 16:06:35 -0700585 vaf.fmt = fmt;
586 vaf.va = &args;
Fabian Frederickf9f58282014-06-04 16:06:34 -0700587 pr_err("=============================================================================\n");
Fabian Frederickecc42fb2014-06-04 16:06:35 -0700588 pr_err("BUG %s (%s): %pV\n", s->name, print_tainted(), &vaf);
Fabian Frederickf9f58282014-06-04 16:06:34 -0700589 pr_err("-----------------------------------------------------------------------------\n\n");
Dave Jones645df232012-09-18 15:54:12 -0400590
Rusty Russell373d4d02013-01-21 17:17:39 +1030591 add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
Fabian Frederickecc42fb2014-06-04 16:06:35 -0700592 va_end(args);
Christoph Lameter24922682007-07-17 04:03:18 -0700593}
594
595static void slab_fix(struct kmem_cache *s, char *fmt, ...)
596{
Fabian Frederickecc42fb2014-06-04 16:06:35 -0700597 struct va_format vaf;
Christoph Lameter24922682007-07-17 04:03:18 -0700598 va_list args;
Christoph Lameter24922682007-07-17 04:03:18 -0700599
600 va_start(args, fmt);
Fabian Frederickecc42fb2014-06-04 16:06:35 -0700601 vaf.fmt = fmt;
602 vaf.va = &args;
603 pr_err("FIX %s: %pV\n", s->name, &vaf);
Christoph Lameter24922682007-07-17 04:03:18 -0700604 va_end(args);
Christoph Lameter24922682007-07-17 04:03:18 -0700605}
606
607static void print_trailer(struct kmem_cache *s, struct page *page, u8 *p)
Christoph Lameter81819f02007-05-06 14:49:36 -0700608{
609 unsigned int off; /* Offset of last byte */
Christoph Lametera973e9d2008-03-01 13:40:44 -0800610 u8 *addr = page_address(page);
Christoph Lameter24922682007-07-17 04:03:18 -0700611
612 print_tracking(s, p);
613
614 print_page_info(page);
615
Fabian Frederickf9f58282014-06-04 16:06:34 -0700616 pr_err("INFO: Object 0x%p @offset=%tu fp=0x%p\n\n",
617 p, p - addr, get_freepointer(s, p));
Christoph Lameter24922682007-07-17 04:03:18 -0700618
619 if (p > addr + 16)
Sebastian Andrzej Siewiorffc79d22011-07-29 14:10:20 +0200620 print_section("Bytes b4 ", p - 16, 16);
Christoph Lameter24922682007-07-17 04:03:18 -0700621
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500622 print_section("Object ", p, min_t(unsigned long, s->object_size,
Sebastian Andrzej Siewiorffc79d22011-07-29 14:10:20 +0200623 PAGE_SIZE));
Christoph Lameter81819f02007-05-06 14:49:36 -0700624 if (s->flags & SLAB_RED_ZONE)
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500625 print_section("Redzone ", p + s->object_size,
626 s->inuse - s->object_size);
Christoph Lameter81819f02007-05-06 14:49:36 -0700627
Christoph Lameter81819f02007-05-06 14:49:36 -0700628 if (s->offset)
629 off = s->offset + sizeof(void *);
630 else
631 off = s->inuse;
632
Christoph Lameter24922682007-07-17 04:03:18 -0700633 if (s->flags & SLAB_STORE_USER)
Christoph Lameter81819f02007-05-06 14:49:36 -0700634 off += 2 * sizeof(struct track);
Christoph Lameter81819f02007-05-06 14:49:36 -0700635
636 if (off != s->size)
637 /* Beginning of the filler is the free pointer */
Sebastian Andrzej Siewiorffc79d22011-07-29 14:10:20 +0200638 print_section("Padding ", p + off, s->size - off);
Christoph Lameter24922682007-07-17 04:03:18 -0700639
640 dump_stack();
Christoph Lameter81819f02007-05-06 14:49:36 -0700641}
642
643static void object_err(struct kmem_cache *s, struct page *page,
644 u8 *object, char *reason)
645{
Christoph Lameter3dc50632008-04-23 12:28:01 -0700646 slab_bug(s, "%s", reason);
Christoph Lameter24922682007-07-17 04:03:18 -0700647 print_trailer(s, page, object);
Christoph Lameter81819f02007-05-06 14:49:36 -0700648}
649
Chen Gangd0e0ac92013-07-15 09:05:29 +0800650static void slab_err(struct kmem_cache *s, struct page *page,
651 const char *fmt, ...)
Christoph Lameter81819f02007-05-06 14:49:36 -0700652{
653 va_list args;
654 char buf[100];
655
Christoph Lameter24922682007-07-17 04:03:18 -0700656 va_start(args, fmt);
657 vsnprintf(buf, sizeof(buf), fmt, args);
Christoph Lameter81819f02007-05-06 14:49:36 -0700658 va_end(args);
Christoph Lameter3dc50632008-04-23 12:28:01 -0700659 slab_bug(s, "%s", buf);
Christoph Lameter24922682007-07-17 04:03:18 -0700660 print_page_info(page);
Christoph Lameter81819f02007-05-06 14:49:36 -0700661 dump_stack();
662}
663
Christoph Lameterf7cb1932010-09-29 07:15:01 -0500664static void init_object(struct kmem_cache *s, void *object, u8 val)
Christoph Lameter81819f02007-05-06 14:49:36 -0700665{
666 u8 *p = object;
667
668 if (s->flags & __OBJECT_POISON) {
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500669 memset(p, POISON_FREE, s->object_size - 1);
670 p[s->object_size - 1] = POISON_END;
Christoph Lameter81819f02007-05-06 14:49:36 -0700671 }
672
673 if (s->flags & SLAB_RED_ZONE)
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500674 memset(p + s->object_size, val, s->inuse - s->object_size);
Christoph Lameter81819f02007-05-06 14:49:36 -0700675}
676
Christoph Lameter24922682007-07-17 04:03:18 -0700677static void restore_bytes(struct kmem_cache *s, char *message, u8 data,
678 void *from, void *to)
679{
680 slab_fix(s, "Restoring 0x%p-0x%p=0x%x\n", from, to - 1, data);
681 memset(from, data, to - from);
682}
683
684static int check_bytes_and_report(struct kmem_cache *s, struct page *page,
685 u8 *object, char *what,
Pekka Enberg06428782008-01-07 23:20:27 -0800686 u8 *start, unsigned int value, unsigned int bytes)
Christoph Lameter24922682007-07-17 04:03:18 -0700687{
688 u8 *fault;
689 u8 *end;
690
Akinobu Mita798248202011-10-31 17:08:07 -0700691 fault = memchr_inv(start, value, bytes);
Christoph Lameter24922682007-07-17 04:03:18 -0700692 if (!fault)
693 return 1;
694
695 end = start + bytes;
696 while (end > fault && end[-1] == value)
697 end--;
698
699 slab_bug(s, "%s overwritten", what);
Fabian Frederickf9f58282014-06-04 16:06:34 -0700700 pr_err("INFO: 0x%p-0x%p. First byte 0x%x instead of 0x%x\n",
Christoph Lameter24922682007-07-17 04:03:18 -0700701 fault, end - 1, fault[0], value);
702 print_trailer(s, page, object);
703
704 restore_bytes(s, what, value, fault, end);
705 return 0;
Christoph Lameter81819f02007-05-06 14:49:36 -0700706}
707
Christoph Lameter81819f02007-05-06 14:49:36 -0700708/*
709 * Object layout:
710 *
711 * object address
712 * Bytes of the object to be managed.
713 * If the freepointer may overlay the object then the free
714 * pointer is the first word of the object.
Christoph Lameter672bba32007-05-09 02:32:39 -0700715 *
Christoph Lameter81819f02007-05-06 14:49:36 -0700716 * Poisoning uses 0x6b (POISON_FREE) and the last byte is
717 * 0xa5 (POISON_END)
718 *
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500719 * object + s->object_size
Christoph Lameter81819f02007-05-06 14:49:36 -0700720 * Padding to reach word boundary. This is also used for Redzoning.
Christoph Lameter672bba32007-05-09 02:32:39 -0700721 * Padding is extended by another word if Redzoning is enabled and
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500722 * object_size == inuse.
Christoph Lameter672bba32007-05-09 02:32:39 -0700723 *
Christoph Lameter81819f02007-05-06 14:49:36 -0700724 * We fill with 0xbb (RED_INACTIVE) for inactive objects and with
725 * 0xcc (RED_ACTIVE) for objects in use.
726 *
727 * object + s->inuse
Christoph Lameter672bba32007-05-09 02:32:39 -0700728 * Meta data starts here.
729 *
Christoph Lameter81819f02007-05-06 14:49:36 -0700730 * A. Free pointer (if we cannot overwrite object on free)
731 * B. Tracking data for SLAB_STORE_USER
Christoph Lameter672bba32007-05-09 02:32:39 -0700732 * C. Padding to reach required alignment boundary or at mininum
Christoph Lameter6446faa2008-02-15 23:45:26 -0800733 * one word if debugging is on to be able to detect writes
Christoph Lameter672bba32007-05-09 02:32:39 -0700734 * before the word boundary.
735 *
736 * Padding is done using 0x5a (POISON_INUSE)
Christoph Lameter81819f02007-05-06 14:49:36 -0700737 *
738 * object + s->size
Christoph Lameter672bba32007-05-09 02:32:39 -0700739 * Nothing is used beyond s->size.
Christoph Lameter81819f02007-05-06 14:49:36 -0700740 *
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500741 * If slabcaches are merged then the object_size and inuse boundaries are mostly
Christoph Lameter672bba32007-05-09 02:32:39 -0700742 * ignored. And therefore no slab options that rely on these boundaries
Christoph Lameter81819f02007-05-06 14:49:36 -0700743 * may be used with merged slabcaches.
744 */
745
Christoph Lameter81819f02007-05-06 14:49:36 -0700746static int check_pad_bytes(struct kmem_cache *s, struct page *page, u8 *p)
747{
748 unsigned long off = s->inuse; /* The end of info */
749
750 if (s->offset)
751 /* Freepointer is placed after the object. */
752 off += sizeof(void *);
753
754 if (s->flags & SLAB_STORE_USER)
755 /* We also have user information there */
756 off += 2 * sizeof(struct track);
757
758 if (s->size == off)
759 return 1;
760
Christoph Lameter24922682007-07-17 04:03:18 -0700761 return check_bytes_and_report(s, page, p, "Object padding",
762 p + off, POISON_INUSE, s->size - off);
Christoph Lameter81819f02007-05-06 14:49:36 -0700763}
764
Christoph Lameter39b26462008-04-14 19:11:30 +0300765/* Check the pad bytes at the end of a slab page */
Christoph Lameter81819f02007-05-06 14:49:36 -0700766static int slab_pad_check(struct kmem_cache *s, struct page *page)
767{
Christoph Lameter24922682007-07-17 04:03:18 -0700768 u8 *start;
769 u8 *fault;
770 u8 *end;
771 int length;
772 int remainder;
Christoph Lameter81819f02007-05-06 14:49:36 -0700773
774 if (!(s->flags & SLAB_POISON))
775 return 1;
776
Christoph Lametera973e9d2008-03-01 13:40:44 -0800777 start = page_address(page);
Lai Jiangshanab9a0f12011-03-10 15:21:48 +0800778 length = (PAGE_SIZE << compound_order(page)) - s->reserved;
Christoph Lameter39b26462008-04-14 19:11:30 +0300779 end = start + length;
780 remainder = length % s->size;
Christoph Lameter81819f02007-05-06 14:49:36 -0700781 if (!remainder)
782 return 1;
783
Akinobu Mita798248202011-10-31 17:08:07 -0700784 fault = memchr_inv(end - remainder, POISON_INUSE, remainder);
Christoph Lameter24922682007-07-17 04:03:18 -0700785 if (!fault)
786 return 1;
787 while (end > fault && end[-1] == POISON_INUSE)
788 end--;
789
790 slab_err(s, page, "Padding overwritten. 0x%p-0x%p", fault, end - 1);
Sebastian Andrzej Siewiorffc79d22011-07-29 14:10:20 +0200791 print_section("Padding ", end - remainder, remainder);
Christoph Lameter24922682007-07-17 04:03:18 -0700792
Eric Dumazet8a3d2712009-09-03 16:08:06 +0200793 restore_bytes(s, "slab padding", POISON_INUSE, end - remainder, end);
Christoph Lameter24922682007-07-17 04:03:18 -0700794 return 0;
Christoph Lameter81819f02007-05-06 14:49:36 -0700795}
796
797static int check_object(struct kmem_cache *s, struct page *page,
Christoph Lameterf7cb1932010-09-29 07:15:01 -0500798 void *object, u8 val)
Christoph Lameter81819f02007-05-06 14:49:36 -0700799{
800 u8 *p = object;
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500801 u8 *endobject = object + s->object_size;
Christoph Lameter81819f02007-05-06 14:49:36 -0700802
803 if (s->flags & SLAB_RED_ZONE) {
Christoph Lameter24922682007-07-17 04:03:18 -0700804 if (!check_bytes_and_report(s, page, object, "Redzone",
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500805 endobject, val, s->inuse - s->object_size))
Christoph Lameter81819f02007-05-06 14:49:36 -0700806 return 0;
Christoph Lameter81819f02007-05-06 14:49:36 -0700807 } else {
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500808 if ((s->flags & SLAB_POISON) && s->object_size < s->inuse) {
Ingo Molnar3adbefe2008-02-05 17:57:39 -0800809 check_bytes_and_report(s, page, p, "Alignment padding",
Chen Gangd0e0ac92013-07-15 09:05:29 +0800810 endobject, POISON_INUSE,
811 s->inuse - s->object_size);
Ingo Molnar3adbefe2008-02-05 17:57:39 -0800812 }
Christoph Lameter81819f02007-05-06 14:49:36 -0700813 }
814
815 if (s->flags & SLAB_POISON) {
Christoph Lameterf7cb1932010-09-29 07:15:01 -0500816 if (val != SLUB_RED_ACTIVE && (s->flags & __OBJECT_POISON) &&
Christoph Lameter24922682007-07-17 04:03:18 -0700817 (!check_bytes_and_report(s, page, p, "Poison", p,
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500818 POISON_FREE, s->object_size - 1) ||
Christoph Lameter24922682007-07-17 04:03:18 -0700819 !check_bytes_and_report(s, page, p, "Poison",
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500820 p + s->object_size - 1, POISON_END, 1)))
Christoph Lameter81819f02007-05-06 14:49:36 -0700821 return 0;
Christoph Lameter81819f02007-05-06 14:49:36 -0700822 /*
823 * check_pad_bytes cleans up on its own.
824 */
825 check_pad_bytes(s, page, p);
826 }
827
Christoph Lameterf7cb1932010-09-29 07:15:01 -0500828 if (!s->offset && val == SLUB_RED_ACTIVE)
Christoph Lameter81819f02007-05-06 14:49:36 -0700829 /*
830 * Object and freepointer overlap. Cannot check
831 * freepointer while object is allocated.
832 */
833 return 1;
834
835 /* Check free pointer validity */
836 if (!check_valid_pointer(s, page, get_freepointer(s, p))) {
837 object_err(s, page, p, "Freepointer corrupt");
838 /*
Nick Andrew9f6c708e2008-12-05 14:08:08 +1100839 * No choice but to zap it and thus lose the remainder
Christoph Lameter81819f02007-05-06 14:49:36 -0700840 * of the free objects in this slab. May cause
Christoph Lameter672bba32007-05-09 02:32:39 -0700841 * another error because the object count is now wrong.
Christoph Lameter81819f02007-05-06 14:49:36 -0700842 */
Christoph Lametera973e9d2008-03-01 13:40:44 -0800843 set_freepointer(s, p, NULL);
Christoph Lameter81819f02007-05-06 14:49:36 -0700844 return 0;
845 }
846 return 1;
847}
848
849static int check_slab(struct kmem_cache *s, struct page *page)
850{
Christoph Lameter39b26462008-04-14 19:11:30 +0300851 int maxobj;
852
Christoph Lameter81819f02007-05-06 14:49:36 -0700853 VM_BUG_ON(!irqs_disabled());
854
855 if (!PageSlab(page)) {
Christoph Lameter24922682007-07-17 04:03:18 -0700856 slab_err(s, page, "Not a valid slab page");
Christoph Lameter81819f02007-05-06 14:49:36 -0700857 return 0;
858 }
Christoph Lameter39b26462008-04-14 19:11:30 +0300859
Lai Jiangshanab9a0f12011-03-10 15:21:48 +0800860 maxobj = order_objects(compound_order(page), s->size, s->reserved);
Christoph Lameter39b26462008-04-14 19:11:30 +0300861 if (page->objects > maxobj) {
862 slab_err(s, page, "objects %u > max %u",
863 s->name, page->objects, maxobj);
864 return 0;
865 }
866 if (page->inuse > page->objects) {
Christoph Lameter24922682007-07-17 04:03:18 -0700867 slab_err(s, page, "inuse %u > max %u",
Christoph Lameter39b26462008-04-14 19:11:30 +0300868 s->name, page->inuse, page->objects);
Christoph Lameter81819f02007-05-06 14:49:36 -0700869 return 0;
870 }
871 /* Slab_pad_check fixes things up after itself */
872 slab_pad_check(s, page);
873 return 1;
874}
875
876/*
Christoph Lameter672bba32007-05-09 02:32:39 -0700877 * Determine if a certain object on a page is on the freelist. Must hold the
878 * slab lock to guarantee that the chains are in a consistent state.
Christoph Lameter81819f02007-05-06 14:49:36 -0700879 */
880static int on_freelist(struct kmem_cache *s, struct page *page, void *search)
881{
882 int nr = 0;
Christoph Lameter881db7f2011-06-01 12:25:53 -0500883 void *fp;
Christoph Lameter81819f02007-05-06 14:49:36 -0700884 void *object = NULL;
Christoph Lameter224a88b2008-04-14 19:11:31 +0300885 unsigned long max_objects;
Christoph Lameter81819f02007-05-06 14:49:36 -0700886
Christoph Lameter881db7f2011-06-01 12:25:53 -0500887 fp = page->freelist;
Christoph Lameter39b26462008-04-14 19:11:30 +0300888 while (fp && nr <= page->objects) {
Christoph Lameter81819f02007-05-06 14:49:36 -0700889 if (fp == search)
890 return 1;
891 if (!check_valid_pointer(s, page, fp)) {
892 if (object) {
893 object_err(s, page, object,
894 "Freechain corrupt");
Christoph Lametera973e9d2008-03-01 13:40:44 -0800895 set_freepointer(s, object, NULL);
Christoph Lameter81819f02007-05-06 14:49:36 -0700896 } else {
Christoph Lameter24922682007-07-17 04:03:18 -0700897 slab_err(s, page, "Freepointer corrupt");
Christoph Lametera973e9d2008-03-01 13:40:44 -0800898 page->freelist = NULL;
Christoph Lameter39b26462008-04-14 19:11:30 +0300899 page->inuse = page->objects;
Christoph Lameter24922682007-07-17 04:03:18 -0700900 slab_fix(s, "Freelist cleared");
Christoph Lameter81819f02007-05-06 14:49:36 -0700901 return 0;
902 }
903 break;
904 }
905 object = fp;
906 fp = get_freepointer(s, object);
907 nr++;
908 }
909
Lai Jiangshanab9a0f12011-03-10 15:21:48 +0800910 max_objects = order_objects(compound_order(page), s->size, s->reserved);
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +0400911 if (max_objects > MAX_OBJS_PER_PAGE)
912 max_objects = MAX_OBJS_PER_PAGE;
Christoph Lameter224a88b2008-04-14 19:11:31 +0300913
914 if (page->objects != max_objects) {
915 slab_err(s, page, "Wrong number of objects. Found %d but "
916 "should be %d", page->objects, max_objects);
917 page->objects = max_objects;
918 slab_fix(s, "Number of objects adjusted.");
919 }
Christoph Lameter39b26462008-04-14 19:11:30 +0300920 if (page->inuse != page->objects - nr) {
Christoph Lameter70d71222007-05-06 14:49:47 -0700921 slab_err(s, page, "Wrong object count. Counter is %d but "
Christoph Lameter39b26462008-04-14 19:11:30 +0300922 "counted were %d", page->inuse, page->objects - nr);
923 page->inuse = page->objects - nr;
Christoph Lameter24922682007-07-17 04:03:18 -0700924 slab_fix(s, "Object count adjusted.");
Christoph Lameter81819f02007-05-06 14:49:36 -0700925 }
926 return search == NULL;
927}
928
Christoph Lameter0121c6192008-04-29 16:11:12 -0700929static void trace(struct kmem_cache *s, struct page *page, void *object,
930 int alloc)
Christoph Lameter3ec09742007-05-16 22:11:00 -0700931{
932 if (s->flags & SLAB_TRACE) {
Fabian Frederickf9f58282014-06-04 16:06:34 -0700933 pr_info("TRACE %s %s 0x%p inuse=%d fp=0x%p\n",
Christoph Lameter3ec09742007-05-16 22:11:00 -0700934 s->name,
935 alloc ? "alloc" : "free",
936 object, page->inuse,
937 page->freelist);
938
939 if (!alloc)
Chen Gangd0e0ac92013-07-15 09:05:29 +0800940 print_section("Object ", (void *)object,
941 s->object_size);
Christoph Lameter3ec09742007-05-16 22:11:00 -0700942
943 dump_stack();
944 }
945}
946
Christoph Lameter643b1132007-05-06 14:49:42 -0700947/*
Christoph Lameterc016b0b2010-08-20 12:37:16 -0500948 * Hooks for other subsystems that check memory allocations. In a typical
949 * production configuration these hooks all should produce no code at all.
950 */
Roman Bobnievd56791b2013-10-08 15:58:57 -0700951static inline void kmalloc_large_node_hook(void *ptr, size_t size, gfp_t flags)
952{
953 kmemleak_alloc(ptr, size, 1, flags);
954}
955
956static inline void kfree_hook(const void *x)
957{
958 kmemleak_free(x);
959}
960
Christoph Lameterc016b0b2010-08-20 12:37:16 -0500961static inline int slab_pre_alloc_hook(struct kmem_cache *s, gfp_t flags)
962{
Christoph Lameterc1d50832010-08-20 12:37:17 -0500963 flags &= gfp_allowed_mask;
Christoph Lameterc016b0b2010-08-20 12:37:16 -0500964 lockdep_trace_alloc(flags);
965 might_sleep_if(flags & __GFP_WAIT);
966
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500967 return should_failslab(s->object_size, flags, s->flags);
Christoph Lameterc016b0b2010-08-20 12:37:16 -0500968}
969
Chen Gangd0e0ac92013-07-15 09:05:29 +0800970static inline void slab_post_alloc_hook(struct kmem_cache *s,
971 gfp_t flags, void *object)
Christoph Lameterc016b0b2010-08-20 12:37:16 -0500972{
Christoph Lameterc1d50832010-08-20 12:37:17 -0500973 flags &= gfp_allowed_mask;
Eric Dumazetb3d41882011-02-14 18:35:22 +0100974 kmemcheck_slab_alloc(s, flags, object, slab_ksize(s));
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500975 kmemleak_alloc_recursive(object, s->object_size, 1, s->flags, flags);
Christoph Lameterc016b0b2010-08-20 12:37:16 -0500976}
977
978static inline void slab_free_hook(struct kmem_cache *s, void *x)
979{
980 kmemleak_free_recursive(x, s->flags);
Christoph Lameterc016b0b2010-08-20 12:37:16 -0500981
Christoph Lameterd3f661d2011-02-25 11:38:52 -0600982 /*
Xie XiuQid1756172013-10-18 09:12:43 +0800983 * Trouble is that we may no longer disable interrupts in the fast path
Christoph Lameterd3f661d2011-02-25 11:38:52 -0600984 * So in order to make the debug calls that expect irqs to be
985 * disabled we need to disable interrupts temporarily.
986 */
987#if defined(CONFIG_KMEMCHECK) || defined(CONFIG_LOCKDEP)
988 {
989 unsigned long flags;
990
991 local_irq_save(flags);
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500992 kmemcheck_slab_free(s, x, s->object_size);
993 debug_check_no_locks_freed(x, s->object_size);
Christoph Lameterd3f661d2011-02-25 11:38:52 -0600994 local_irq_restore(flags);
995 }
996#endif
Thomas Gleixnerf9b615d2011-03-24 21:26:46 +0200997 if (!(s->flags & SLAB_DEBUG_OBJECTS))
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500998 debug_check_no_obj_freed(x, s->object_size);
Christoph Lameterc016b0b2010-08-20 12:37:16 -0500999}
1000
1001/*
Christoph Lameter672bba32007-05-09 02:32:39 -07001002 * Tracking of fully allocated slabs for debugging purposes.
Christoph Lameter643b1132007-05-06 14:49:42 -07001003 */
Christoph Lameter5cc6eee2011-06-01 12:25:50 -05001004static void add_full(struct kmem_cache *s,
1005 struct kmem_cache_node *n, struct page *page)
Christoph Lameter643b1132007-05-06 14:49:42 -07001006{
Christoph Lameter643b1132007-05-06 14:49:42 -07001007 if (!(s->flags & SLAB_STORE_USER))
1008 return;
1009
David Rientjes255d0882014-02-10 14:25:39 -08001010 lockdep_assert_held(&n->list_lock);
Christoph Lameter5cc6eee2011-06-01 12:25:50 -05001011 list_add(&page->lru, &n->full);
1012}
Christoph Lameter643b1132007-05-06 14:49:42 -07001013
Peter Zijlstrac65c1872014-01-10 13:23:49 +01001014static void remove_full(struct kmem_cache *s, struct kmem_cache_node *n, struct page *page)
Christoph Lameter5cc6eee2011-06-01 12:25:50 -05001015{
1016 if (!(s->flags & SLAB_STORE_USER))
1017 return;
1018
David Rientjes255d0882014-02-10 14:25:39 -08001019 lockdep_assert_held(&n->list_lock);
Christoph Lameter643b1132007-05-06 14:49:42 -07001020 list_del(&page->lru);
Christoph Lameter643b1132007-05-06 14:49:42 -07001021}
1022
Christoph Lameter0f389ec2008-04-14 18:53:02 +03001023/* Tracking of the number of slabs for debugging purposes */
1024static inline unsigned long slabs_node(struct kmem_cache *s, int node)
1025{
1026 struct kmem_cache_node *n = get_node(s, node);
1027
1028 return atomic_long_read(&n->nr_slabs);
1029}
1030
Alexander Beregalov26c02cf2009-06-11 14:08:48 +04001031static inline unsigned long node_nr_slabs(struct kmem_cache_node *n)
1032{
1033 return atomic_long_read(&n->nr_slabs);
1034}
1035
Christoph Lameter205ab992008-04-14 19:11:40 +03001036static inline void inc_slabs_node(struct kmem_cache *s, int node, int objects)
Christoph Lameter0f389ec2008-04-14 18:53:02 +03001037{
1038 struct kmem_cache_node *n = get_node(s, node);
1039
1040 /*
1041 * May be called early in order to allocate a slab for the
1042 * kmem_cache_node structure. Solve the chicken-egg
1043 * dilemma by deferring the increment of the count during
1044 * bootstrap (see early_kmem_cache_node_alloc).
1045 */
Joonsoo Kim338b2642013-01-21 17:01:27 +09001046 if (likely(n)) {
Christoph Lameter0f389ec2008-04-14 18:53:02 +03001047 atomic_long_inc(&n->nr_slabs);
Christoph Lameter205ab992008-04-14 19:11:40 +03001048 atomic_long_add(objects, &n->total_objects);
1049 }
Christoph Lameter0f389ec2008-04-14 18:53:02 +03001050}
Christoph Lameter205ab992008-04-14 19:11:40 +03001051static inline void dec_slabs_node(struct kmem_cache *s, int node, int objects)
Christoph Lameter0f389ec2008-04-14 18:53:02 +03001052{
1053 struct kmem_cache_node *n = get_node(s, node);
1054
1055 atomic_long_dec(&n->nr_slabs);
Christoph Lameter205ab992008-04-14 19:11:40 +03001056 atomic_long_sub(objects, &n->total_objects);
Christoph Lameter0f389ec2008-04-14 18:53:02 +03001057}
1058
1059/* Object debug checks for alloc/free paths */
Christoph Lameter3ec09742007-05-16 22:11:00 -07001060static void setup_object_debug(struct kmem_cache *s, struct page *page,
1061 void *object)
1062{
1063 if (!(s->flags & (SLAB_STORE_USER|SLAB_RED_ZONE|__OBJECT_POISON)))
1064 return;
1065
Christoph Lameterf7cb1932010-09-29 07:15:01 -05001066 init_object(s, object, SLUB_RED_INACTIVE);
Christoph Lameter3ec09742007-05-16 22:11:00 -07001067 init_tracking(s, object);
1068}
1069
Chen Gangd0e0ac92013-07-15 09:05:29 +08001070static noinline int alloc_debug_processing(struct kmem_cache *s,
1071 struct page *page,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03001072 void *object, unsigned long addr)
Christoph Lameter81819f02007-05-06 14:49:36 -07001073{
1074 if (!check_slab(s, page))
1075 goto bad;
1076
Christoph Lameter81819f02007-05-06 14:49:36 -07001077 if (!check_valid_pointer(s, page, object)) {
1078 object_err(s, page, object, "Freelist Pointer check fails");
Christoph Lameter70d71222007-05-06 14:49:47 -07001079 goto bad;
Christoph Lameter81819f02007-05-06 14:49:36 -07001080 }
1081
Christoph Lameterf7cb1932010-09-29 07:15:01 -05001082 if (!check_object(s, page, object, SLUB_RED_INACTIVE))
Christoph Lameter81819f02007-05-06 14:49:36 -07001083 goto bad;
Christoph Lameter81819f02007-05-06 14:49:36 -07001084
Christoph Lameter3ec09742007-05-16 22:11:00 -07001085 /* Success perform special debug activities for allocs */
1086 if (s->flags & SLAB_STORE_USER)
1087 set_track(s, object, TRACK_ALLOC, addr);
1088 trace(s, page, object, 1);
Christoph Lameterf7cb1932010-09-29 07:15:01 -05001089 init_object(s, object, SLUB_RED_ACTIVE);
Christoph Lameter81819f02007-05-06 14:49:36 -07001090 return 1;
Christoph Lameter3ec09742007-05-16 22:11:00 -07001091
Christoph Lameter81819f02007-05-06 14:49:36 -07001092bad:
1093 if (PageSlab(page)) {
1094 /*
1095 * If this is a slab page then lets do the best we can
1096 * to avoid issues in the future. Marking all objects
Christoph Lameter672bba32007-05-09 02:32:39 -07001097 * as used avoids touching the remaining objects.
Christoph Lameter81819f02007-05-06 14:49:36 -07001098 */
Christoph Lameter24922682007-07-17 04:03:18 -07001099 slab_fix(s, "Marking all objects used");
Christoph Lameter39b26462008-04-14 19:11:30 +03001100 page->inuse = page->objects;
Christoph Lametera973e9d2008-03-01 13:40:44 -08001101 page->freelist = NULL;
Christoph Lameter81819f02007-05-06 14:49:36 -07001102 }
1103 return 0;
1104}
1105
Christoph Lameter19c7ff92012-05-30 12:54:46 -05001106static noinline struct kmem_cache_node *free_debug_processing(
1107 struct kmem_cache *s, struct page *page, void *object,
1108 unsigned long addr, unsigned long *flags)
Christoph Lameter81819f02007-05-06 14:49:36 -07001109{
Christoph Lameter19c7ff92012-05-30 12:54:46 -05001110 struct kmem_cache_node *n = get_node(s, page_to_nid(page));
Christoph Lameter5c2e4bb2011-06-01 12:25:54 -05001111
Christoph Lameter19c7ff92012-05-30 12:54:46 -05001112 spin_lock_irqsave(&n->list_lock, *flags);
Christoph Lameter881db7f2011-06-01 12:25:53 -05001113 slab_lock(page);
1114
Christoph Lameter81819f02007-05-06 14:49:36 -07001115 if (!check_slab(s, page))
1116 goto fail;
1117
1118 if (!check_valid_pointer(s, page, object)) {
Christoph Lameter70d71222007-05-06 14:49:47 -07001119 slab_err(s, page, "Invalid object pointer 0x%p", object);
Christoph Lameter81819f02007-05-06 14:49:36 -07001120 goto fail;
1121 }
1122
1123 if (on_freelist(s, page, object)) {
Christoph Lameter24922682007-07-17 04:03:18 -07001124 object_err(s, page, object, "Object already free");
Christoph Lameter81819f02007-05-06 14:49:36 -07001125 goto fail;
1126 }
1127
Christoph Lameterf7cb1932010-09-29 07:15:01 -05001128 if (!check_object(s, page, object, SLUB_RED_ACTIVE))
Christoph Lameter5c2e4bb2011-06-01 12:25:54 -05001129 goto out;
Christoph Lameter81819f02007-05-06 14:49:36 -07001130
Glauber Costa1b4f59e32012-10-22 18:05:36 +04001131 if (unlikely(s != page->slab_cache)) {
Ingo Molnar3adbefe2008-02-05 17:57:39 -08001132 if (!PageSlab(page)) {
Christoph Lameter70d71222007-05-06 14:49:47 -07001133 slab_err(s, page, "Attempt to free object(0x%p) "
1134 "outside of slab", object);
Glauber Costa1b4f59e32012-10-22 18:05:36 +04001135 } else if (!page->slab_cache) {
Fabian Frederickf9f58282014-06-04 16:06:34 -07001136 pr_err("SLUB <none>: no slab for object 0x%p.\n",
1137 object);
Christoph Lameter70d71222007-05-06 14:49:47 -07001138 dump_stack();
Pekka Enberg06428782008-01-07 23:20:27 -08001139 } else
Christoph Lameter24922682007-07-17 04:03:18 -07001140 object_err(s, page, object,
1141 "page slab pointer corrupt.");
Christoph Lameter81819f02007-05-06 14:49:36 -07001142 goto fail;
1143 }
Christoph Lameter3ec09742007-05-16 22:11:00 -07001144
Christoph Lameter3ec09742007-05-16 22:11:00 -07001145 if (s->flags & SLAB_STORE_USER)
1146 set_track(s, object, TRACK_FREE, addr);
1147 trace(s, page, object, 0);
Christoph Lameterf7cb1932010-09-29 07:15:01 -05001148 init_object(s, object, SLUB_RED_INACTIVE);
Christoph Lameter5c2e4bb2011-06-01 12:25:54 -05001149out:
Christoph Lameter881db7f2011-06-01 12:25:53 -05001150 slab_unlock(page);
Christoph Lameter19c7ff92012-05-30 12:54:46 -05001151 /*
1152 * Keep node_lock to preserve integrity
1153 * until the object is actually freed
1154 */
1155 return n;
Christoph Lameter3ec09742007-05-16 22:11:00 -07001156
Christoph Lameter81819f02007-05-06 14:49:36 -07001157fail:
Christoph Lameter19c7ff92012-05-30 12:54:46 -05001158 slab_unlock(page);
1159 spin_unlock_irqrestore(&n->list_lock, *flags);
Christoph Lameter24922682007-07-17 04:03:18 -07001160 slab_fix(s, "Object at 0x%p not freed", object);
Christoph Lameter19c7ff92012-05-30 12:54:46 -05001161 return NULL;
Christoph Lameter81819f02007-05-06 14:49:36 -07001162}
1163
Christoph Lameter41ecc552007-05-09 02:32:44 -07001164static int __init setup_slub_debug(char *str)
1165{
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001166 slub_debug = DEBUG_DEFAULT_FLAGS;
1167 if (*str++ != '=' || !*str)
1168 /*
1169 * No options specified. Switch on full debugging.
1170 */
1171 goto out;
Christoph Lameter41ecc552007-05-09 02:32:44 -07001172
1173 if (*str == ',')
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001174 /*
1175 * No options but restriction on slabs. This means full
1176 * debugging for slabs matching a pattern.
1177 */
1178 goto check_slabs;
1179
David Rientjesfa5ec8a2009-07-07 00:14:14 -07001180 if (tolower(*str) == 'o') {
1181 /*
1182 * Avoid enabling debugging on caches if its minimum order
1183 * would increase as a result.
1184 */
1185 disable_higher_order_debug = 1;
1186 goto out;
1187 }
1188
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001189 slub_debug = 0;
1190 if (*str == '-')
1191 /*
1192 * Switch off all debugging measures.
1193 */
1194 goto out;
1195
1196 /*
1197 * Determine which debug features should be switched on
1198 */
Pekka Enberg06428782008-01-07 23:20:27 -08001199 for (; *str && *str != ','; str++) {
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001200 switch (tolower(*str)) {
1201 case 'f':
1202 slub_debug |= SLAB_DEBUG_FREE;
1203 break;
1204 case 'z':
1205 slub_debug |= SLAB_RED_ZONE;
1206 break;
1207 case 'p':
1208 slub_debug |= SLAB_POISON;
1209 break;
1210 case 'u':
1211 slub_debug |= SLAB_STORE_USER;
1212 break;
1213 case 't':
1214 slub_debug |= SLAB_TRACE;
1215 break;
Dmitry Monakhov4c13dd32010-02-26 09:36:12 +03001216 case 'a':
1217 slub_debug |= SLAB_FAILSLAB;
1218 break;
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001219 default:
Fabian Frederickf9f58282014-06-04 16:06:34 -07001220 pr_err("slub_debug option '%c' unknown. skipped\n",
1221 *str);
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001222 }
1223 }
1224
1225check_slabs:
1226 if (*str == ',')
Christoph Lameter41ecc552007-05-09 02:32:44 -07001227 slub_debug_slabs = str + 1;
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001228out:
Christoph Lameter41ecc552007-05-09 02:32:44 -07001229 return 1;
1230}
1231
1232__setup("slub_debug", setup_slub_debug);
1233
Christoph Lameter3b0efdf2012-06-13 10:24:57 -05001234static unsigned long kmem_cache_flags(unsigned long object_size,
Christoph Lameterba0268a2007-09-11 15:24:11 -07001235 unsigned long flags, const char *name,
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07001236 void (*ctor)(void *))
Christoph Lameter41ecc552007-05-09 02:32:44 -07001237{
1238 /*
Christoph Lametere1533622008-02-15 23:45:24 -08001239 * Enable debugging if selected on the kernel commandline.
Christoph Lameter41ecc552007-05-09 02:32:44 -07001240 */
Christoph Lameterc6f58d92013-11-07 16:29:15 +00001241 if (slub_debug && (!slub_debug_slabs || (name &&
1242 !strncmp(slub_debug_slabs, name, strlen(slub_debug_slabs)))))
David Rientjes3de47212009-07-27 18:30:35 -07001243 flags |= slub_debug;
Christoph Lameterba0268a2007-09-11 15:24:11 -07001244
1245 return flags;
Christoph Lameter41ecc552007-05-09 02:32:44 -07001246}
1247#else
Christoph Lameter3ec09742007-05-16 22:11:00 -07001248static inline void setup_object_debug(struct kmem_cache *s,
1249 struct page *page, void *object) {}
Christoph Lameter41ecc552007-05-09 02:32:44 -07001250
Christoph Lameter3ec09742007-05-16 22:11:00 -07001251static inline int alloc_debug_processing(struct kmem_cache *s,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03001252 struct page *page, void *object, unsigned long addr) { return 0; }
Christoph Lameter41ecc552007-05-09 02:32:44 -07001253
Christoph Lameter19c7ff92012-05-30 12:54:46 -05001254static inline struct kmem_cache_node *free_debug_processing(
1255 struct kmem_cache *s, struct page *page, void *object,
1256 unsigned long addr, unsigned long *flags) { return NULL; }
Christoph Lameter41ecc552007-05-09 02:32:44 -07001257
Christoph Lameter41ecc552007-05-09 02:32:44 -07001258static inline int slab_pad_check(struct kmem_cache *s, struct page *page)
1259 { return 1; }
1260static inline int check_object(struct kmem_cache *s, struct page *page,
Christoph Lameterf7cb1932010-09-29 07:15:01 -05001261 void *object, u8 val) { return 1; }
Christoph Lameter5cc6eee2011-06-01 12:25:50 -05001262static inline void add_full(struct kmem_cache *s, struct kmem_cache_node *n,
1263 struct page *page) {}
Peter Zijlstrac65c1872014-01-10 13:23:49 +01001264static inline void remove_full(struct kmem_cache *s, struct kmem_cache_node *n,
1265 struct page *page) {}
Christoph Lameter3b0efdf2012-06-13 10:24:57 -05001266static inline unsigned long kmem_cache_flags(unsigned long object_size,
Christoph Lameterba0268a2007-09-11 15:24:11 -07001267 unsigned long flags, const char *name,
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07001268 void (*ctor)(void *))
Christoph Lameterba0268a2007-09-11 15:24:11 -07001269{
1270 return flags;
1271}
Christoph Lameter41ecc552007-05-09 02:32:44 -07001272#define slub_debug 0
Christoph Lameter0f389ec2008-04-14 18:53:02 +03001273
Ingo Molnarfdaa45e2009-09-15 11:00:26 +02001274#define disable_higher_order_debug 0
1275
Christoph Lameter0f389ec2008-04-14 18:53:02 +03001276static inline unsigned long slabs_node(struct kmem_cache *s, int node)
1277 { return 0; }
Alexander Beregalov26c02cf2009-06-11 14:08:48 +04001278static inline unsigned long node_nr_slabs(struct kmem_cache_node *n)
1279 { return 0; }
Christoph Lameter205ab992008-04-14 19:11:40 +03001280static inline void inc_slabs_node(struct kmem_cache *s, int node,
1281 int objects) {}
1282static inline void dec_slabs_node(struct kmem_cache *s, int node,
1283 int objects) {}
Christoph Lameter7d550c52010-08-25 14:07:16 -05001284
Roman Bobnievd56791b2013-10-08 15:58:57 -07001285static inline void kmalloc_large_node_hook(void *ptr, size_t size, gfp_t flags)
1286{
1287 kmemleak_alloc(ptr, size, 1, flags);
1288}
1289
1290static inline void kfree_hook(const void *x)
1291{
1292 kmemleak_free(x);
1293}
1294
Christoph Lameter7d550c52010-08-25 14:07:16 -05001295static inline int slab_pre_alloc_hook(struct kmem_cache *s, gfp_t flags)
1296 { return 0; }
1297
1298static inline void slab_post_alloc_hook(struct kmem_cache *s, gfp_t flags,
Roman Bobnievd56791b2013-10-08 15:58:57 -07001299 void *object)
1300{
1301 kmemleak_alloc_recursive(object, s->object_size, 1, s->flags,
1302 flags & gfp_allowed_mask);
1303}
Christoph Lameter7d550c52010-08-25 14:07:16 -05001304
Roman Bobnievd56791b2013-10-08 15:58:57 -07001305static inline void slab_free_hook(struct kmem_cache *s, void *x)
1306{
1307 kmemleak_free_recursive(x, s->flags);
1308}
Christoph Lameter7d550c52010-08-25 14:07:16 -05001309
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05001310#endif /* CONFIG_SLUB_DEBUG */
Christoph Lameter205ab992008-04-14 19:11:40 +03001311
Christoph Lameter81819f02007-05-06 14:49:36 -07001312/*
1313 * Slab allocation and freeing
1314 */
Vladimir Davydov5dfb4172014-06-04 16:06:38 -07001315static inline struct page *alloc_slab_page(struct kmem_cache *s,
1316 gfp_t flags, int node, struct kmem_cache_order_objects oo)
Christoph Lameter65c33762008-04-14 19:11:40 +03001317{
Vladimir Davydov5dfb4172014-06-04 16:06:38 -07001318 struct page *page;
Christoph Lameter65c33762008-04-14 19:11:40 +03001319 int order = oo_order(oo);
1320
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001321 flags |= __GFP_NOTRACK;
1322
Vladimir Davydov5dfb4172014-06-04 16:06:38 -07001323 if (memcg_charge_slab(s, flags, order))
1324 return NULL;
1325
Christoph Lameter2154a332010-07-09 14:07:10 -05001326 if (node == NUMA_NO_NODE)
Vladimir Davydov5dfb4172014-06-04 16:06:38 -07001327 page = alloc_pages(flags, order);
Christoph Lameter65c33762008-04-14 19:11:40 +03001328 else
Vladimir Davydov5dfb4172014-06-04 16:06:38 -07001329 page = alloc_pages_exact_node(node, flags, order);
1330
1331 if (!page)
1332 memcg_uncharge_slab(s, order);
1333
1334 return page;
Christoph Lameter65c33762008-04-14 19:11:40 +03001335}
1336
Christoph Lameter81819f02007-05-06 14:49:36 -07001337static struct page *allocate_slab(struct kmem_cache *s, gfp_t flags, int node)
1338{
Pekka Enberg06428782008-01-07 23:20:27 -08001339 struct page *page;
Christoph Lameter834f3d12008-04-14 19:11:31 +03001340 struct kmem_cache_order_objects oo = s->oo;
Pekka Enbergba522702009-06-24 21:59:51 +03001341 gfp_t alloc_gfp;
Christoph Lameter81819f02007-05-06 14:49:36 -07001342
Christoph Lameter7e0528d2011-06-01 12:25:44 -05001343 flags &= gfp_allowed_mask;
1344
1345 if (flags & __GFP_WAIT)
1346 local_irq_enable();
1347
Christoph Lameterb7a49f02008-02-14 14:21:32 -08001348 flags |= s->allocflags;
Mel Gormane12ba742007-10-16 01:25:52 -07001349
Pekka Enbergba522702009-06-24 21:59:51 +03001350 /*
1351 * Let the initial higher-order allocation fail under memory pressure
1352 * so we fall-back to the minimum order allocation.
1353 */
1354 alloc_gfp = (flags | __GFP_NOWARN | __GFP_NORETRY) & ~__GFP_NOFAIL;
1355
Vladimir Davydov5dfb4172014-06-04 16:06:38 -07001356 page = alloc_slab_page(s, alloc_gfp, node, oo);
Christoph Lameter65c33762008-04-14 19:11:40 +03001357 if (unlikely(!page)) {
1358 oo = s->min;
Joonsoo Kim80c3a992014-03-12 17:26:20 +09001359 alloc_gfp = flags;
Christoph Lameter65c33762008-04-14 19:11:40 +03001360 /*
1361 * Allocation may have failed due to fragmentation.
1362 * Try a lower order alloc if possible
1363 */
Vladimir Davydov5dfb4172014-06-04 16:06:38 -07001364 page = alloc_slab_page(s, alloc_gfp, node, oo);
Christoph Lameter81819f02007-05-06 14:49:36 -07001365
Christoph Lameter7e0528d2011-06-01 12:25:44 -05001366 if (page)
1367 stat(s, ORDER_FALLBACK);
Christoph Lameter65c33762008-04-14 19:11:40 +03001368 }
Vegard Nossum5a896d92008-04-04 00:54:48 +02001369
David Rientjes737b7192012-07-09 14:00:38 -07001370 if (kmemcheck_enabled && page
Amerigo Wang5086c389c2009-08-19 21:44:13 +03001371 && !(s->flags & (SLAB_NOTRACK | DEBUG_DEFAULT_FLAGS))) {
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001372 int pages = 1 << oo_order(oo);
1373
Joonsoo Kim80c3a992014-03-12 17:26:20 +09001374 kmemcheck_alloc_shadow(page, oo_order(oo), alloc_gfp, node);
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001375
1376 /*
1377 * Objects from caches that have a constructor don't get
1378 * cleared when they're allocated, so we need to do it here.
1379 */
1380 if (s->ctor)
1381 kmemcheck_mark_uninitialized_pages(page, pages);
1382 else
1383 kmemcheck_mark_unallocated_pages(page, pages);
Vegard Nossum5a896d92008-04-04 00:54:48 +02001384 }
1385
David Rientjes737b7192012-07-09 14:00:38 -07001386 if (flags & __GFP_WAIT)
1387 local_irq_disable();
1388 if (!page)
1389 return NULL;
1390
Christoph Lameter834f3d12008-04-14 19:11:31 +03001391 page->objects = oo_objects(oo);
Christoph Lameter81819f02007-05-06 14:49:36 -07001392 mod_zone_page_state(page_zone(page),
1393 (s->flags & SLAB_RECLAIM_ACCOUNT) ?
1394 NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE,
Christoph Lameter65c33762008-04-14 19:11:40 +03001395 1 << oo_order(oo));
Christoph Lameter81819f02007-05-06 14:49:36 -07001396
1397 return page;
1398}
1399
1400static void setup_object(struct kmem_cache *s, struct page *page,
1401 void *object)
1402{
Christoph Lameter3ec09742007-05-16 22:11:00 -07001403 setup_object_debug(s, page, object);
Christoph Lameter4f104932007-05-06 14:50:17 -07001404 if (unlikely(s->ctor))
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07001405 s->ctor(object);
Christoph Lameter81819f02007-05-06 14:49:36 -07001406}
1407
1408static struct page *new_slab(struct kmem_cache *s, gfp_t flags, int node)
1409{
1410 struct page *page;
Christoph Lameter81819f02007-05-06 14:49:36 -07001411 void *start;
Christoph Lameter81819f02007-05-06 14:49:36 -07001412 void *last;
1413 void *p;
Glauber Costa1f458cb2012-12-18 14:22:50 -08001414 int order;
Christoph Lameter81819f02007-05-06 14:49:36 -07001415
Christoph Lameter6cb06222007-10-16 01:25:41 -07001416 BUG_ON(flags & GFP_SLAB_BUG_MASK);
Christoph Lameter81819f02007-05-06 14:49:36 -07001417
Christoph Lameter6cb06222007-10-16 01:25:41 -07001418 page = allocate_slab(s,
1419 flags & (GFP_RECLAIM_MASK | GFP_CONSTRAINT_MASK), node);
Christoph Lameter81819f02007-05-06 14:49:36 -07001420 if (!page)
1421 goto out;
1422
Glauber Costa1f458cb2012-12-18 14:22:50 -08001423 order = compound_order(page);
Christoph Lameter205ab992008-04-14 19:11:40 +03001424 inc_slabs_node(s, page_to_nid(page), page->objects);
Glauber Costa1b4f59e32012-10-22 18:05:36 +04001425 page->slab_cache = s;
Joonsoo Kimc03f94c2012-05-18 00:47:47 +09001426 __SetPageSlab(page);
Mel Gorman072bb0a2012-07-31 16:43:58 -07001427 if (page->pfmemalloc)
1428 SetPageSlabPfmemalloc(page);
Christoph Lameter81819f02007-05-06 14:49:36 -07001429
1430 start = page_address(page);
Christoph Lameter81819f02007-05-06 14:49:36 -07001431
1432 if (unlikely(s->flags & SLAB_POISON))
Glauber Costa1f458cb2012-12-18 14:22:50 -08001433 memset(start, POISON_INUSE, PAGE_SIZE << order);
Christoph Lameter81819f02007-05-06 14:49:36 -07001434
1435 last = start;
Christoph Lameter224a88b2008-04-14 19:11:31 +03001436 for_each_object(p, s, start, page->objects) {
Christoph Lameter81819f02007-05-06 14:49:36 -07001437 setup_object(s, page, last);
1438 set_freepointer(s, last, p);
1439 last = p;
1440 }
1441 setup_object(s, page, last);
Christoph Lametera973e9d2008-03-01 13:40:44 -08001442 set_freepointer(s, last, NULL);
Christoph Lameter81819f02007-05-06 14:49:36 -07001443
1444 page->freelist = start;
Christoph Lametere6e82ea2011-08-09 16:12:24 -05001445 page->inuse = page->objects;
Christoph Lameter8cb0a502011-06-01 12:25:46 -05001446 page->frozen = 1;
Christoph Lameter81819f02007-05-06 14:49:36 -07001447out:
Christoph Lameter81819f02007-05-06 14:49:36 -07001448 return page;
1449}
1450
1451static void __free_slab(struct kmem_cache *s, struct page *page)
1452{
Christoph Lameter834f3d12008-04-14 19:11:31 +03001453 int order = compound_order(page);
1454 int pages = 1 << order;
Christoph Lameter81819f02007-05-06 14:49:36 -07001455
Christoph Lameteraf537b02010-07-09 14:07:14 -05001456 if (kmem_cache_debug(s)) {
Christoph Lameter81819f02007-05-06 14:49:36 -07001457 void *p;
1458
1459 slab_pad_check(s, page);
Christoph Lameter224a88b2008-04-14 19:11:31 +03001460 for_each_object(p, s, page_address(page),
1461 page->objects)
Christoph Lameterf7cb1932010-09-29 07:15:01 -05001462 check_object(s, page, p, SLUB_RED_INACTIVE);
Christoph Lameter81819f02007-05-06 14:49:36 -07001463 }
1464
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001465 kmemcheck_free_shadow(page, compound_order(page));
Vegard Nossum5a896d92008-04-04 00:54:48 +02001466
Christoph Lameter81819f02007-05-06 14:49:36 -07001467 mod_zone_page_state(page_zone(page),
1468 (s->flags & SLAB_RECLAIM_ACCOUNT) ?
1469 NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE,
Pekka Enberg06428782008-01-07 23:20:27 -08001470 -pages);
Christoph Lameter81819f02007-05-06 14:49:36 -07001471
Mel Gorman072bb0a2012-07-31 16:43:58 -07001472 __ClearPageSlabPfmemalloc(page);
Christoph Lameter49bd5222008-04-14 18:52:18 +03001473 __ClearPageSlab(page);
Glauber Costa1f458cb2012-12-18 14:22:50 -08001474
Mel Gorman22b751c2013-02-22 16:34:59 -08001475 page_mapcount_reset(page);
Nick Piggin1eb5ac62009-05-05 19:13:44 +10001476 if (current->reclaim_state)
1477 current->reclaim_state->reclaimed_slab += pages;
Vladimir Davydov5dfb4172014-06-04 16:06:38 -07001478 __free_pages(page, order);
1479 memcg_uncharge_slab(s, order);
Christoph Lameter81819f02007-05-06 14:49:36 -07001480}
1481
Lai Jiangshanda9a6382011-03-10 15:22:00 +08001482#define need_reserve_slab_rcu \
1483 (sizeof(((struct page *)NULL)->lru) < sizeof(struct rcu_head))
1484
Christoph Lameter81819f02007-05-06 14:49:36 -07001485static void rcu_free_slab(struct rcu_head *h)
1486{
1487 struct page *page;
1488
Lai Jiangshanda9a6382011-03-10 15:22:00 +08001489 if (need_reserve_slab_rcu)
1490 page = virt_to_head_page(h);
1491 else
1492 page = container_of((struct list_head *)h, struct page, lru);
1493
Glauber Costa1b4f59e32012-10-22 18:05:36 +04001494 __free_slab(page->slab_cache, page);
Christoph Lameter81819f02007-05-06 14:49:36 -07001495}
1496
1497static void free_slab(struct kmem_cache *s, struct page *page)
1498{
1499 if (unlikely(s->flags & SLAB_DESTROY_BY_RCU)) {
Lai Jiangshanda9a6382011-03-10 15:22:00 +08001500 struct rcu_head *head;
1501
1502 if (need_reserve_slab_rcu) {
1503 int order = compound_order(page);
1504 int offset = (PAGE_SIZE << order) - s->reserved;
1505
1506 VM_BUG_ON(s->reserved != sizeof(*head));
1507 head = page_address(page) + offset;
1508 } else {
1509 /*
1510 * RCU free overloads the RCU head over the LRU
1511 */
1512 head = (void *)&page->lru;
1513 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001514
1515 call_rcu(head, rcu_free_slab);
1516 } else
1517 __free_slab(s, page);
1518}
1519
1520static void discard_slab(struct kmem_cache *s, struct page *page)
1521{
Christoph Lameter205ab992008-04-14 19:11:40 +03001522 dec_slabs_node(s, page_to_nid(page), page->objects);
Christoph Lameter81819f02007-05-06 14:49:36 -07001523 free_slab(s, page);
1524}
1525
1526/*
Christoph Lameter5cc6eee2011-06-01 12:25:50 -05001527 * Management of partially allocated slabs.
Christoph Lameter81819f02007-05-06 14:49:36 -07001528 */
Steven Rostedt1e4dd942014-02-10 14:25:46 -08001529static inline void
1530__add_partial(struct kmem_cache_node *n, struct page *page, int tail)
Christoph Lameter81819f02007-05-06 14:49:36 -07001531{
Christoph Lametere95eed52007-05-06 14:49:44 -07001532 n->nr_partial++;
Shaohua Li136333d2011-08-24 08:57:52 +08001533 if (tail == DEACTIVATE_TO_TAIL)
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001534 list_add_tail(&page->lru, &n->partial);
1535 else
1536 list_add(&page->lru, &n->partial);
Christoph Lameter81819f02007-05-06 14:49:36 -07001537}
1538
Steven Rostedt1e4dd942014-02-10 14:25:46 -08001539static inline void add_partial(struct kmem_cache_node *n,
1540 struct page *page, int tail)
1541{
1542 lockdep_assert_held(&n->list_lock);
1543 __add_partial(n, page, tail);
1544}
1545
1546static inline void
1547__remove_partial(struct kmem_cache_node *n, struct page *page)
1548{
1549 list_del(&page->lru);
1550 n->nr_partial--;
1551}
1552
Christoph Lameter5cc6eee2011-06-01 12:25:50 -05001553static inline void remove_partial(struct kmem_cache_node *n,
Christoph Lameter62e346a2010-09-28 08:10:28 -05001554 struct page *page)
1555{
Peter Zijlstrac65c1872014-01-10 13:23:49 +01001556 lockdep_assert_held(&n->list_lock);
Steven Rostedt1e4dd942014-02-10 14:25:46 -08001557 __remove_partial(n, page);
Christoph Lameter62e346a2010-09-28 08:10:28 -05001558}
1559
Christoph Lameter81819f02007-05-06 14:49:36 -07001560/*
Christoph Lameter7ced3712012-05-09 10:09:53 -05001561 * Remove slab from the partial list, freeze it and
1562 * return the pointer to the freelist.
Christoph Lameter81819f02007-05-06 14:49:36 -07001563 *
Christoph Lameter497b66f2011-08-09 16:12:26 -05001564 * Returns a list of objects or NULL if it fails.
Christoph Lameter81819f02007-05-06 14:49:36 -07001565 */
Christoph Lameter497b66f2011-08-09 16:12:26 -05001566static inline void *acquire_slab(struct kmem_cache *s,
Christoph Lameteracd19fd2011-08-09 16:12:25 -05001567 struct kmem_cache_node *n, struct page *page,
Joonsoo Kim633b0762013-01-21 17:01:25 +09001568 int mode, int *objects)
Christoph Lameter81819f02007-05-06 14:49:36 -07001569{
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001570 void *freelist;
1571 unsigned long counters;
1572 struct page new;
1573
Peter Zijlstrac65c1872014-01-10 13:23:49 +01001574 lockdep_assert_held(&n->list_lock);
1575
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001576 /*
1577 * Zap the freelist and set the frozen bit.
1578 * The old freelist is the list of objects for the
1579 * per cpu allocation list.
1580 */
Christoph Lameter7ced3712012-05-09 10:09:53 -05001581 freelist = page->freelist;
1582 counters = page->counters;
1583 new.counters = counters;
Joonsoo Kim633b0762013-01-21 17:01:25 +09001584 *objects = new.objects - new.inuse;
Pekka Enberg23910c52012-06-04 10:14:58 +03001585 if (mode) {
Christoph Lameter7ced3712012-05-09 10:09:53 -05001586 new.inuse = page->objects;
Pekka Enberg23910c52012-06-04 10:14:58 +03001587 new.freelist = NULL;
1588 } else {
1589 new.freelist = freelist;
1590 }
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001591
Dave Hansena0132ac2014-01-29 14:05:50 -08001592 VM_BUG_ON(new.frozen);
Christoph Lameter7ced3712012-05-09 10:09:53 -05001593 new.frozen = 1;
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001594
Christoph Lameter7ced3712012-05-09 10:09:53 -05001595 if (!__cmpxchg_double_slab(s, page,
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001596 freelist, counters,
Joonsoo Kim02d76332012-05-17 00:13:02 +09001597 new.freelist, new.counters,
Christoph Lameter7ced3712012-05-09 10:09:53 -05001598 "acquire_slab"))
Christoph Lameter7ced3712012-05-09 10:09:53 -05001599 return NULL;
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001600
1601 remove_partial(n, page);
Christoph Lameter7ced3712012-05-09 10:09:53 -05001602 WARN_ON(!freelist);
Christoph Lameter49e22582011-08-09 16:12:27 -05001603 return freelist;
Christoph Lameter81819f02007-05-06 14:49:36 -07001604}
1605
Joonsoo Kim633b0762013-01-21 17:01:25 +09001606static void put_cpu_partial(struct kmem_cache *s, struct page *page, int drain);
Joonsoo Kim8ba00bb2012-09-17 14:09:09 -07001607static inline bool pfmemalloc_match(struct page *page, gfp_t gfpflags);
Christoph Lameter49e22582011-08-09 16:12:27 -05001608
Christoph Lameter81819f02007-05-06 14:49:36 -07001609/*
Christoph Lameter672bba32007-05-09 02:32:39 -07001610 * Try to allocate a partial slab from a specific node.
Christoph Lameter81819f02007-05-06 14:49:36 -07001611 */
Joonsoo Kim8ba00bb2012-09-17 14:09:09 -07001612static void *get_partial_node(struct kmem_cache *s, struct kmem_cache_node *n,
1613 struct kmem_cache_cpu *c, gfp_t flags)
Christoph Lameter81819f02007-05-06 14:49:36 -07001614{
Christoph Lameter49e22582011-08-09 16:12:27 -05001615 struct page *page, *page2;
1616 void *object = NULL;
Joonsoo Kim633b0762013-01-21 17:01:25 +09001617 int available = 0;
1618 int objects;
Christoph Lameter81819f02007-05-06 14:49:36 -07001619
1620 /*
1621 * Racy check. If we mistakenly see no partial slabs then we
1622 * just allocate an empty slab. If we mistakenly try to get a
Christoph Lameter672bba32007-05-09 02:32:39 -07001623 * partial slab and there is none available then get_partials()
1624 * will return NULL.
Christoph Lameter81819f02007-05-06 14:49:36 -07001625 */
1626 if (!n || !n->nr_partial)
1627 return NULL;
1628
1629 spin_lock(&n->list_lock);
Christoph Lameter49e22582011-08-09 16:12:27 -05001630 list_for_each_entry_safe(page, page2, &n->partial, lru) {
Joonsoo Kim8ba00bb2012-09-17 14:09:09 -07001631 void *t;
Christoph Lameter49e22582011-08-09 16:12:27 -05001632
Joonsoo Kim8ba00bb2012-09-17 14:09:09 -07001633 if (!pfmemalloc_match(page, flags))
1634 continue;
1635
Joonsoo Kim633b0762013-01-21 17:01:25 +09001636 t = acquire_slab(s, n, page, object == NULL, &objects);
Christoph Lameter49e22582011-08-09 16:12:27 -05001637 if (!t)
1638 break;
1639
Joonsoo Kim633b0762013-01-21 17:01:25 +09001640 available += objects;
Alex,Shi12d79632011-09-07 10:26:36 +08001641 if (!object) {
Christoph Lameter49e22582011-08-09 16:12:27 -05001642 c->page = page;
Christoph Lameter49e22582011-08-09 16:12:27 -05001643 stat(s, ALLOC_FROM_PARTIAL);
Christoph Lameter49e22582011-08-09 16:12:27 -05001644 object = t;
Christoph Lameter49e22582011-08-09 16:12:27 -05001645 } else {
Joonsoo Kim633b0762013-01-21 17:01:25 +09001646 put_cpu_partial(s, page, 0);
Alex Shi8028dce2012-02-03 23:34:56 +08001647 stat(s, CPU_PARTIAL_NODE);
Christoph Lameter49e22582011-08-09 16:12:27 -05001648 }
Joonsoo Kim345c9052013-06-19 14:05:52 +09001649 if (!kmem_cache_has_cpu_partial(s)
1650 || available > s->cpu_partial / 2)
Christoph Lameter49e22582011-08-09 16:12:27 -05001651 break;
1652
Christoph Lameter497b66f2011-08-09 16:12:26 -05001653 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001654 spin_unlock(&n->list_lock);
Christoph Lameter497b66f2011-08-09 16:12:26 -05001655 return object;
Christoph Lameter81819f02007-05-06 14:49:36 -07001656}
1657
1658/*
Christoph Lameter672bba32007-05-09 02:32:39 -07001659 * Get a page from somewhere. Search in increasing NUMA distances.
Christoph Lameter81819f02007-05-06 14:49:36 -07001660 */
Joonsoo Kimde3ec032012-01-27 00:12:23 -08001661static void *get_any_partial(struct kmem_cache *s, gfp_t flags,
Christoph Lameteracd19fd2011-08-09 16:12:25 -05001662 struct kmem_cache_cpu *c)
Christoph Lameter81819f02007-05-06 14:49:36 -07001663{
1664#ifdef CONFIG_NUMA
1665 struct zonelist *zonelist;
Mel Gormandd1a2392008-04-28 02:12:17 -07001666 struct zoneref *z;
Mel Gorman54a6eb52008-04-28 02:12:16 -07001667 struct zone *zone;
1668 enum zone_type high_zoneidx = gfp_zone(flags);
Christoph Lameter497b66f2011-08-09 16:12:26 -05001669 void *object;
Mel Gormancc9a6c82012-03-21 16:34:11 -07001670 unsigned int cpuset_mems_cookie;
Christoph Lameter81819f02007-05-06 14:49:36 -07001671
1672 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07001673 * The defrag ratio allows a configuration of the tradeoffs between
1674 * inter node defragmentation and node local allocations. A lower
1675 * defrag_ratio increases the tendency to do local allocations
1676 * instead of attempting to obtain partial slabs from other nodes.
Christoph Lameter81819f02007-05-06 14:49:36 -07001677 *
Christoph Lameter672bba32007-05-09 02:32:39 -07001678 * If the defrag_ratio is set to 0 then kmalloc() always
1679 * returns node local objects. If the ratio is higher then kmalloc()
1680 * may return off node objects because partial slabs are obtained
1681 * from other nodes and filled up.
Christoph Lameter81819f02007-05-06 14:49:36 -07001682 *
Christoph Lameter6446faa2008-02-15 23:45:26 -08001683 * If /sys/kernel/slab/xx/defrag_ratio is set to 100 (which makes
Christoph Lameter672bba32007-05-09 02:32:39 -07001684 * defrag_ratio = 1000) then every (well almost) allocation will
1685 * first attempt to defrag slab caches on other nodes. This means
1686 * scanning over all nodes to look for partial slabs which may be
1687 * expensive if we do it every time we are trying to find a slab
1688 * with available objects.
Christoph Lameter81819f02007-05-06 14:49:36 -07001689 */
Christoph Lameter98246012008-01-07 23:20:26 -08001690 if (!s->remote_node_defrag_ratio ||
1691 get_cycles() % 1024 > s->remote_node_defrag_ratio)
Christoph Lameter81819f02007-05-06 14:49:36 -07001692 return NULL;
1693
Mel Gormancc9a6c82012-03-21 16:34:11 -07001694 do {
Mel Gormand26914d2014-04-03 14:47:24 -07001695 cpuset_mems_cookie = read_mems_allowed_begin();
David Rientjes2a389612014-04-07 15:37:29 -07001696 zonelist = node_zonelist(mempolicy_slab_node(), flags);
Mel Gormancc9a6c82012-03-21 16:34:11 -07001697 for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) {
1698 struct kmem_cache_node *n;
Christoph Lameter81819f02007-05-06 14:49:36 -07001699
Mel Gormancc9a6c82012-03-21 16:34:11 -07001700 n = get_node(s, zone_to_nid(zone));
Christoph Lameter81819f02007-05-06 14:49:36 -07001701
Mel Gormancc9a6c82012-03-21 16:34:11 -07001702 if (n && cpuset_zone_allowed_hardwall(zone, flags) &&
1703 n->nr_partial > s->min_partial) {
Joonsoo Kim8ba00bb2012-09-17 14:09:09 -07001704 object = get_partial_node(s, n, c, flags);
Mel Gormancc9a6c82012-03-21 16:34:11 -07001705 if (object) {
1706 /*
Mel Gormand26914d2014-04-03 14:47:24 -07001707 * Don't check read_mems_allowed_retry()
1708 * here - if mems_allowed was updated in
1709 * parallel, that was a harmless race
1710 * between allocation and the cpuset
1711 * update
Mel Gormancc9a6c82012-03-21 16:34:11 -07001712 */
Mel Gormancc9a6c82012-03-21 16:34:11 -07001713 return object;
1714 }
Miao Xiec0ff7452010-05-24 14:32:08 -07001715 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001716 }
Mel Gormand26914d2014-04-03 14:47:24 -07001717 } while (read_mems_allowed_retry(cpuset_mems_cookie));
Christoph Lameter81819f02007-05-06 14:49:36 -07001718#endif
1719 return NULL;
1720}
1721
1722/*
1723 * Get a partial page, lock it and return it.
1724 */
Christoph Lameter497b66f2011-08-09 16:12:26 -05001725static void *get_partial(struct kmem_cache *s, gfp_t flags, int node,
Christoph Lameteracd19fd2011-08-09 16:12:25 -05001726 struct kmem_cache_cpu *c)
Christoph Lameter81819f02007-05-06 14:49:36 -07001727{
Christoph Lameter497b66f2011-08-09 16:12:26 -05001728 void *object;
Joonsoo Kim844e4d62014-06-06 14:35:36 -07001729 int searchnode = (node == NUMA_NO_NODE) ? numa_mem_id() : node;
Christoph Lameter81819f02007-05-06 14:49:36 -07001730
Joonsoo Kim8ba00bb2012-09-17 14:09:09 -07001731 object = get_partial_node(s, get_node(s, searchnode), c, flags);
Christoph Lameter497b66f2011-08-09 16:12:26 -05001732 if (object || node != NUMA_NO_NODE)
1733 return object;
Christoph Lameter81819f02007-05-06 14:49:36 -07001734
Christoph Lameteracd19fd2011-08-09 16:12:25 -05001735 return get_any_partial(s, flags, c);
Christoph Lameter81819f02007-05-06 14:49:36 -07001736}
1737
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001738#ifdef CONFIG_PREEMPT
1739/*
1740 * Calculate the next globally unique transaction for disambiguiation
1741 * during cmpxchg. The transactions start with the cpu number and are then
1742 * incremented by CONFIG_NR_CPUS.
1743 */
1744#define TID_STEP roundup_pow_of_two(CONFIG_NR_CPUS)
1745#else
1746/*
1747 * No preemption supported therefore also no need to check for
1748 * different cpus.
1749 */
1750#define TID_STEP 1
1751#endif
1752
1753static inline unsigned long next_tid(unsigned long tid)
1754{
1755 return tid + TID_STEP;
1756}
1757
1758static inline unsigned int tid_to_cpu(unsigned long tid)
1759{
1760 return tid % TID_STEP;
1761}
1762
1763static inline unsigned long tid_to_event(unsigned long tid)
1764{
1765 return tid / TID_STEP;
1766}
1767
1768static inline unsigned int init_tid(int cpu)
1769{
1770 return cpu;
1771}
1772
1773static inline void note_cmpxchg_failure(const char *n,
1774 const struct kmem_cache *s, unsigned long tid)
1775{
1776#ifdef SLUB_DEBUG_CMPXCHG
1777 unsigned long actual_tid = __this_cpu_read(s->cpu_slab->tid);
1778
Fabian Frederickf9f58282014-06-04 16:06:34 -07001779 pr_info("%s %s: cmpxchg redo ", n, s->name);
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001780
1781#ifdef CONFIG_PREEMPT
1782 if (tid_to_cpu(tid) != tid_to_cpu(actual_tid))
Fabian Frederickf9f58282014-06-04 16:06:34 -07001783 pr_warn("due to cpu change %d -> %d\n",
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001784 tid_to_cpu(tid), tid_to_cpu(actual_tid));
1785 else
1786#endif
1787 if (tid_to_event(tid) != tid_to_event(actual_tid))
Fabian Frederickf9f58282014-06-04 16:06:34 -07001788 pr_warn("due to cpu running other code. Event %ld->%ld\n",
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001789 tid_to_event(tid), tid_to_event(actual_tid));
1790 else
Fabian Frederickf9f58282014-06-04 16:06:34 -07001791 pr_warn("for unknown reason: actual=%lx was=%lx target=%lx\n",
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001792 actual_tid, tid, next_tid(tid));
1793#endif
Christoph Lameter4fdccdf2011-03-22 13:35:00 -05001794 stat(s, CMPXCHG_DOUBLE_CPU_FAIL);
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001795}
1796
Fengguang Wu788e1aa2012-09-28 16:34:05 +08001797static void init_kmem_cache_cpus(struct kmem_cache *s)
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001798{
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001799 int cpu;
1800
1801 for_each_possible_cpu(cpu)
1802 per_cpu_ptr(s->cpu_slab, cpu)->tid = init_tid(cpu);
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001803}
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001804
1805/*
1806 * Remove the cpu slab
1807 */
Chen Gangd0e0ac92013-07-15 09:05:29 +08001808static void deactivate_slab(struct kmem_cache *s, struct page *page,
1809 void *freelist)
Christoph Lameter81819f02007-05-06 14:49:36 -07001810{
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001811 enum slab_modes { M_NONE, M_PARTIAL, M_FULL, M_FREE };
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001812 struct kmem_cache_node *n = get_node(s, page_to_nid(page));
1813 int lock = 0;
1814 enum slab_modes l = M_NONE, m = M_NONE;
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001815 void *nextfree;
Shaohua Li136333d2011-08-24 08:57:52 +08001816 int tail = DEACTIVATE_TO_HEAD;
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001817 struct page new;
1818 struct page old;
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001819
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001820 if (page->freelist) {
Christoph Lameter84e554e62009-12-18 16:26:23 -06001821 stat(s, DEACTIVATE_REMOTE_FREES);
Shaohua Li136333d2011-08-24 08:57:52 +08001822 tail = DEACTIVATE_TO_TAIL;
Christoph Lameter894b8782007-05-10 03:15:16 -07001823 }
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001824
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001825 /*
1826 * Stage one: Free all available per cpu objects back
1827 * to the page freelist while it is still frozen. Leave the
1828 * last one.
1829 *
1830 * There is no need to take the list->lock because the page
1831 * is still frozen.
1832 */
1833 while (freelist && (nextfree = get_freepointer(s, freelist))) {
1834 void *prior;
1835 unsigned long counters;
1836
1837 do {
1838 prior = page->freelist;
1839 counters = page->counters;
1840 set_freepointer(s, freelist, prior);
1841 new.counters = counters;
1842 new.inuse--;
Dave Hansena0132ac2014-01-29 14:05:50 -08001843 VM_BUG_ON(!new.frozen);
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001844
Christoph Lameter1d071712011-07-14 12:49:12 -05001845 } while (!__cmpxchg_double_slab(s, page,
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001846 prior, counters,
1847 freelist, new.counters,
1848 "drain percpu freelist"));
1849
1850 freelist = nextfree;
1851 }
1852
1853 /*
1854 * Stage two: Ensure that the page is unfrozen while the
1855 * list presence reflects the actual number of objects
1856 * during unfreeze.
1857 *
1858 * We setup the list membership and then perform a cmpxchg
1859 * with the count. If there is a mismatch then the page
1860 * is not unfrozen but the page is on the wrong list.
1861 *
1862 * Then we restart the process which may have to remove
1863 * the page from the list that we just put it on again
1864 * because the number of objects in the slab may have
1865 * changed.
1866 */
1867redo:
1868
1869 old.freelist = page->freelist;
1870 old.counters = page->counters;
Dave Hansena0132ac2014-01-29 14:05:50 -08001871 VM_BUG_ON(!old.frozen);
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001872
1873 /* Determine target state of the slab */
1874 new.counters = old.counters;
1875 if (freelist) {
1876 new.inuse--;
1877 set_freepointer(s, freelist, old.freelist);
1878 new.freelist = freelist;
1879 } else
1880 new.freelist = old.freelist;
1881
1882 new.frozen = 0;
1883
Joonsoo Kim8a5b20a2014-07-02 15:22:35 -07001884 if (!new.inuse && n->nr_partial >= s->min_partial)
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001885 m = M_FREE;
1886 else if (new.freelist) {
1887 m = M_PARTIAL;
1888 if (!lock) {
1889 lock = 1;
1890 /*
1891 * Taking the spinlock removes the possiblity
1892 * that acquire_slab() will see a slab page that
1893 * is frozen
1894 */
1895 spin_lock(&n->list_lock);
1896 }
1897 } else {
1898 m = M_FULL;
1899 if (kmem_cache_debug(s) && !lock) {
1900 lock = 1;
1901 /*
1902 * This also ensures that the scanning of full
1903 * slabs from diagnostic functions will not see
1904 * any frozen slabs.
1905 */
1906 spin_lock(&n->list_lock);
1907 }
1908 }
1909
1910 if (l != m) {
1911
1912 if (l == M_PARTIAL)
1913
1914 remove_partial(n, page);
1915
1916 else if (l == M_FULL)
1917
Peter Zijlstrac65c1872014-01-10 13:23:49 +01001918 remove_full(s, n, page);
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001919
1920 if (m == M_PARTIAL) {
1921
1922 add_partial(n, page, tail);
Shaohua Li136333d2011-08-24 08:57:52 +08001923 stat(s, tail);
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001924
1925 } else if (m == M_FULL) {
1926
1927 stat(s, DEACTIVATE_FULL);
1928 add_full(s, n, page);
1929
1930 }
1931 }
1932
1933 l = m;
Christoph Lameter1d071712011-07-14 12:49:12 -05001934 if (!__cmpxchg_double_slab(s, page,
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001935 old.freelist, old.counters,
1936 new.freelist, new.counters,
1937 "unfreezing slab"))
1938 goto redo;
1939
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001940 if (lock)
1941 spin_unlock(&n->list_lock);
1942
1943 if (m == M_FREE) {
1944 stat(s, DEACTIVATE_EMPTY);
1945 discard_slab(s, page);
1946 stat(s, FREE_SLAB);
1947 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001948}
1949
Joonsoo Kimd24ac772012-05-18 22:01:17 +09001950/*
1951 * Unfreeze all the cpu partial slabs.
1952 *
Christoph Lameter59a09912012-11-28 16:23:00 +00001953 * This function must be called with interrupts disabled
1954 * for the cpu using c (or some other guarantee must be there
1955 * to guarantee no concurrent accesses).
Joonsoo Kimd24ac772012-05-18 22:01:17 +09001956 */
Christoph Lameter59a09912012-11-28 16:23:00 +00001957static void unfreeze_partials(struct kmem_cache *s,
1958 struct kmem_cache_cpu *c)
Christoph Lameter49e22582011-08-09 16:12:27 -05001959{
Joonsoo Kim345c9052013-06-19 14:05:52 +09001960#ifdef CONFIG_SLUB_CPU_PARTIAL
Joonsoo Kim43d77862012-06-09 02:23:16 +09001961 struct kmem_cache_node *n = NULL, *n2 = NULL;
Shaohua Li9ada1932011-11-14 13:34:13 +08001962 struct page *page, *discard_page = NULL;
Christoph Lameter49e22582011-08-09 16:12:27 -05001963
1964 while ((page = c->partial)) {
Christoph Lameter49e22582011-08-09 16:12:27 -05001965 struct page new;
1966 struct page old;
1967
1968 c->partial = page->next;
Joonsoo Kim43d77862012-06-09 02:23:16 +09001969
1970 n2 = get_node(s, page_to_nid(page));
1971 if (n != n2) {
1972 if (n)
1973 spin_unlock(&n->list_lock);
1974
1975 n = n2;
1976 spin_lock(&n->list_lock);
1977 }
Christoph Lameter49e22582011-08-09 16:12:27 -05001978
1979 do {
1980
1981 old.freelist = page->freelist;
1982 old.counters = page->counters;
Dave Hansena0132ac2014-01-29 14:05:50 -08001983 VM_BUG_ON(!old.frozen);
Christoph Lameter49e22582011-08-09 16:12:27 -05001984
1985 new.counters = old.counters;
1986 new.freelist = old.freelist;
1987
1988 new.frozen = 0;
1989
Joonsoo Kimd24ac772012-05-18 22:01:17 +09001990 } while (!__cmpxchg_double_slab(s, page,
Christoph Lameter49e22582011-08-09 16:12:27 -05001991 old.freelist, old.counters,
1992 new.freelist, new.counters,
1993 "unfreezing slab"));
1994
Joonsoo Kim8a5b20a2014-07-02 15:22:35 -07001995 if (unlikely(!new.inuse && n->nr_partial >= s->min_partial)) {
Shaohua Li9ada1932011-11-14 13:34:13 +08001996 page->next = discard_page;
1997 discard_page = page;
Joonsoo Kim43d77862012-06-09 02:23:16 +09001998 } else {
1999 add_partial(n, page, DEACTIVATE_TO_TAIL);
2000 stat(s, FREE_ADD_PARTIAL);
Christoph Lameter49e22582011-08-09 16:12:27 -05002001 }
2002 }
2003
2004 if (n)
2005 spin_unlock(&n->list_lock);
Shaohua Li9ada1932011-11-14 13:34:13 +08002006
2007 while (discard_page) {
2008 page = discard_page;
2009 discard_page = discard_page->next;
2010
2011 stat(s, DEACTIVATE_EMPTY);
2012 discard_slab(s, page);
2013 stat(s, FREE_SLAB);
2014 }
Joonsoo Kim345c9052013-06-19 14:05:52 +09002015#endif
Christoph Lameter49e22582011-08-09 16:12:27 -05002016}
2017
2018/*
2019 * Put a page that was just frozen (in __slab_free) into a partial page
2020 * slot if available. This is done without interrupts disabled and without
2021 * preemption disabled. The cmpxchg is racy and may put the partial page
2022 * onto a random cpus partial slot.
2023 *
2024 * If we did not find a slot then simply move all the partials to the
2025 * per node partial list.
2026 */
Joonsoo Kim633b0762013-01-21 17:01:25 +09002027static void put_cpu_partial(struct kmem_cache *s, struct page *page, int drain)
Christoph Lameter49e22582011-08-09 16:12:27 -05002028{
Joonsoo Kim345c9052013-06-19 14:05:52 +09002029#ifdef CONFIG_SLUB_CPU_PARTIAL
Christoph Lameter49e22582011-08-09 16:12:27 -05002030 struct page *oldpage;
2031 int pages;
2032 int pobjects;
2033
2034 do {
2035 pages = 0;
2036 pobjects = 0;
2037 oldpage = this_cpu_read(s->cpu_slab->partial);
2038
2039 if (oldpage) {
2040 pobjects = oldpage->pobjects;
2041 pages = oldpage->pages;
2042 if (drain && pobjects > s->cpu_partial) {
2043 unsigned long flags;
2044 /*
2045 * partial array is full. Move the existing
2046 * set to the per node partial list.
2047 */
2048 local_irq_save(flags);
Christoph Lameter59a09912012-11-28 16:23:00 +00002049 unfreeze_partials(s, this_cpu_ptr(s->cpu_slab));
Christoph Lameter49e22582011-08-09 16:12:27 -05002050 local_irq_restore(flags);
Joonsoo Kime24fc412012-06-23 03:22:38 +09002051 oldpage = NULL;
Christoph Lameter49e22582011-08-09 16:12:27 -05002052 pobjects = 0;
2053 pages = 0;
Alex Shi8028dce2012-02-03 23:34:56 +08002054 stat(s, CPU_PARTIAL_DRAIN);
Christoph Lameter49e22582011-08-09 16:12:27 -05002055 }
2056 }
2057
2058 pages++;
2059 pobjects += page->objects - page->inuse;
2060
2061 page->pages = pages;
2062 page->pobjects = pobjects;
2063 page->next = oldpage;
2064
Chen Gangd0e0ac92013-07-15 09:05:29 +08002065 } while (this_cpu_cmpxchg(s->cpu_slab->partial, oldpage, page)
2066 != oldpage);
Joonsoo Kim345c9052013-06-19 14:05:52 +09002067#endif
Christoph Lameter49e22582011-08-09 16:12:27 -05002068}
2069
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002070static inline void flush_slab(struct kmem_cache *s, struct kmem_cache_cpu *c)
Christoph Lameter81819f02007-05-06 14:49:36 -07002071{
Christoph Lameter84e554e62009-12-18 16:26:23 -06002072 stat(s, CPUSLAB_FLUSH);
Christoph Lameterc17dda42012-05-09 10:09:57 -05002073 deactivate_slab(s, c->page, c->freelist);
2074
2075 c->tid = next_tid(c->tid);
2076 c->page = NULL;
2077 c->freelist = NULL;
Christoph Lameter81819f02007-05-06 14:49:36 -07002078}
2079
2080/*
2081 * Flush cpu slab.
Christoph Lameter6446faa2008-02-15 23:45:26 -08002082 *
Christoph Lameter81819f02007-05-06 14:49:36 -07002083 * Called from IPI handler with interrupts disabled.
2084 */
Christoph Lameter0c710012007-07-17 04:03:24 -07002085static inline void __flush_cpu_slab(struct kmem_cache *s, int cpu)
Christoph Lameter81819f02007-05-06 14:49:36 -07002086{
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06002087 struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu);
Christoph Lameter81819f02007-05-06 14:49:36 -07002088
Christoph Lameter49e22582011-08-09 16:12:27 -05002089 if (likely(c)) {
2090 if (c->page)
2091 flush_slab(s, c);
2092
Christoph Lameter59a09912012-11-28 16:23:00 +00002093 unfreeze_partials(s, c);
Christoph Lameter49e22582011-08-09 16:12:27 -05002094 }
Christoph Lameter81819f02007-05-06 14:49:36 -07002095}
2096
2097static void flush_cpu_slab(void *d)
2098{
2099 struct kmem_cache *s = d;
Christoph Lameter81819f02007-05-06 14:49:36 -07002100
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002101 __flush_cpu_slab(s, smp_processor_id());
Christoph Lameter81819f02007-05-06 14:49:36 -07002102}
2103
Gilad Ben-Yossefa8364d52012-03-28 14:42:44 -07002104static bool has_cpu_slab(int cpu, void *info)
2105{
2106 struct kmem_cache *s = info;
2107 struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu);
2108
majianpeng02e1a9c2012-05-17 17:03:26 -07002109 return c->page || c->partial;
Gilad Ben-Yossefa8364d52012-03-28 14:42:44 -07002110}
2111
Christoph Lameter81819f02007-05-06 14:49:36 -07002112static void flush_all(struct kmem_cache *s)
2113{
Gilad Ben-Yossefa8364d52012-03-28 14:42:44 -07002114 on_each_cpu_cond(has_cpu_slab, flush_cpu_slab, s, 1, GFP_ATOMIC);
Christoph Lameter81819f02007-05-06 14:49:36 -07002115}
2116
2117/*
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002118 * Check if the objects in a per cpu structure fit numa
2119 * locality expectations.
2120 */
Christoph Lameter57d437d2012-05-09 10:09:59 -05002121static inline int node_match(struct page *page, int node)
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002122{
2123#ifdef CONFIG_NUMA
Christoph Lameter4d7868e2013-01-23 21:45:47 +00002124 if (!page || (node != NUMA_NO_NODE && page_to_nid(page) != node))
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002125 return 0;
2126#endif
2127 return 1;
2128}
2129
David Rientjes9a02d692014-06-04 16:06:36 -07002130#ifdef CONFIG_SLUB_DEBUG
Pekka Enberg781b2ba2009-06-10 18:50:32 +03002131static int count_free(struct page *page)
2132{
2133 return page->objects - page->inuse;
2134}
2135
David Rientjes9a02d692014-06-04 16:06:36 -07002136static inline unsigned long node_nr_objs(struct kmem_cache_node *n)
2137{
2138 return atomic_long_read(&n->total_objects);
2139}
2140#endif /* CONFIG_SLUB_DEBUG */
2141
2142#if defined(CONFIG_SLUB_DEBUG) || defined(CONFIG_SYSFS)
Pekka Enberg781b2ba2009-06-10 18:50:32 +03002143static unsigned long count_partial(struct kmem_cache_node *n,
2144 int (*get_count)(struct page *))
2145{
2146 unsigned long flags;
2147 unsigned long x = 0;
2148 struct page *page;
2149
2150 spin_lock_irqsave(&n->list_lock, flags);
2151 list_for_each_entry(page, &n->partial, lru)
2152 x += get_count(page);
2153 spin_unlock_irqrestore(&n->list_lock, flags);
2154 return x;
2155}
David Rientjes9a02d692014-06-04 16:06:36 -07002156#endif /* CONFIG_SLUB_DEBUG || CONFIG_SYSFS */
Alexander Beregalov26c02cf2009-06-11 14:08:48 +04002157
Pekka Enberg781b2ba2009-06-10 18:50:32 +03002158static noinline void
2159slab_out_of_memory(struct kmem_cache *s, gfp_t gfpflags, int nid)
2160{
David Rientjes9a02d692014-06-04 16:06:36 -07002161#ifdef CONFIG_SLUB_DEBUG
2162 static DEFINE_RATELIMIT_STATE(slub_oom_rs, DEFAULT_RATELIMIT_INTERVAL,
2163 DEFAULT_RATELIMIT_BURST);
Pekka Enberg781b2ba2009-06-10 18:50:32 +03002164 int node;
2165
David Rientjes9a02d692014-06-04 16:06:36 -07002166 if ((gfpflags & __GFP_NOWARN) || !__ratelimit(&slub_oom_rs))
2167 return;
2168
Fabian Frederickf9f58282014-06-04 16:06:34 -07002169 pr_warn("SLUB: Unable to allocate memory on node %d (gfp=0x%x)\n",
Pekka Enberg781b2ba2009-06-10 18:50:32 +03002170 nid, gfpflags);
Fabian Frederickf9f58282014-06-04 16:06:34 -07002171 pr_warn(" cache: %s, object size: %d, buffer size: %d, default order: %d, min order: %d\n",
2172 s->name, s->object_size, s->size, oo_order(s->oo),
2173 oo_order(s->min));
Pekka Enberg781b2ba2009-06-10 18:50:32 +03002174
Christoph Lameter3b0efdf2012-06-13 10:24:57 -05002175 if (oo_order(s->min) > get_order(s->object_size))
Fabian Frederickf9f58282014-06-04 16:06:34 -07002176 pr_warn(" %s debugging increased min order, use slub_debug=O to disable.\n",
2177 s->name);
David Rientjesfa5ec8a2009-07-07 00:14:14 -07002178
Pekka Enberg781b2ba2009-06-10 18:50:32 +03002179 for_each_online_node(node) {
2180 struct kmem_cache_node *n = get_node(s, node);
2181 unsigned long nr_slabs;
2182 unsigned long nr_objs;
2183 unsigned long nr_free;
2184
2185 if (!n)
2186 continue;
2187
Alexander Beregalov26c02cf2009-06-11 14:08:48 +04002188 nr_free = count_partial(n, count_free);
2189 nr_slabs = node_nr_slabs(n);
2190 nr_objs = node_nr_objs(n);
Pekka Enberg781b2ba2009-06-10 18:50:32 +03002191
Fabian Frederickf9f58282014-06-04 16:06:34 -07002192 pr_warn(" node %d: slabs: %ld, objs: %ld, free: %ld\n",
Pekka Enberg781b2ba2009-06-10 18:50:32 +03002193 node, nr_slabs, nr_objs, nr_free);
2194 }
David Rientjes9a02d692014-06-04 16:06:36 -07002195#endif
Pekka Enberg781b2ba2009-06-10 18:50:32 +03002196}
2197
Christoph Lameter497b66f2011-08-09 16:12:26 -05002198static inline void *new_slab_objects(struct kmem_cache *s, gfp_t flags,
2199 int node, struct kmem_cache_cpu **pc)
2200{
Christoph Lameter6faa6832012-05-09 10:09:51 -05002201 void *freelist;
Christoph Lameter188fd062012-05-09 10:09:55 -05002202 struct kmem_cache_cpu *c = *pc;
2203 struct page *page;
Christoph Lameter497b66f2011-08-09 16:12:26 -05002204
Christoph Lameter188fd062012-05-09 10:09:55 -05002205 freelist = get_partial(s, flags, node, c);
2206
2207 if (freelist)
2208 return freelist;
2209
2210 page = new_slab(s, flags, node);
Christoph Lameter497b66f2011-08-09 16:12:26 -05002211 if (page) {
Christoph Lameter7c8e0182014-06-04 16:07:56 -07002212 c = raw_cpu_ptr(s->cpu_slab);
Christoph Lameter497b66f2011-08-09 16:12:26 -05002213 if (c->page)
2214 flush_slab(s, c);
2215
2216 /*
2217 * No other reference to the page yet so we can
2218 * muck around with it freely without cmpxchg
2219 */
Christoph Lameter6faa6832012-05-09 10:09:51 -05002220 freelist = page->freelist;
Christoph Lameter497b66f2011-08-09 16:12:26 -05002221 page->freelist = NULL;
2222
2223 stat(s, ALLOC_SLAB);
Christoph Lameter497b66f2011-08-09 16:12:26 -05002224 c->page = page;
2225 *pc = c;
2226 } else
Christoph Lameter6faa6832012-05-09 10:09:51 -05002227 freelist = NULL;
Christoph Lameter497b66f2011-08-09 16:12:26 -05002228
Christoph Lameter6faa6832012-05-09 10:09:51 -05002229 return freelist;
Christoph Lameter497b66f2011-08-09 16:12:26 -05002230}
2231
Mel Gorman072bb0a2012-07-31 16:43:58 -07002232static inline bool pfmemalloc_match(struct page *page, gfp_t gfpflags)
2233{
2234 if (unlikely(PageSlabPfmemalloc(page)))
2235 return gfp_pfmemalloc_allowed(gfpflags);
2236
2237 return true;
2238}
2239
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002240/*
Chen Gangd0e0ac92013-07-15 09:05:29 +08002241 * Check the page->freelist of a page and either transfer the freelist to the
2242 * per cpu freelist or deactivate the page.
Christoph Lameter213eeb92011-11-11 14:07:14 -06002243 *
2244 * The page is still frozen if the return value is not NULL.
2245 *
2246 * If this function returns NULL then the page has been unfrozen.
Joonsoo Kimd24ac772012-05-18 22:01:17 +09002247 *
2248 * This function must be called with interrupt disabled.
Christoph Lameter213eeb92011-11-11 14:07:14 -06002249 */
2250static inline void *get_freelist(struct kmem_cache *s, struct page *page)
2251{
2252 struct page new;
2253 unsigned long counters;
2254 void *freelist;
2255
2256 do {
2257 freelist = page->freelist;
2258 counters = page->counters;
Christoph Lameter6faa6832012-05-09 10:09:51 -05002259
Christoph Lameter213eeb92011-11-11 14:07:14 -06002260 new.counters = counters;
Dave Hansena0132ac2014-01-29 14:05:50 -08002261 VM_BUG_ON(!new.frozen);
Christoph Lameter213eeb92011-11-11 14:07:14 -06002262
2263 new.inuse = page->objects;
2264 new.frozen = freelist != NULL;
2265
Joonsoo Kimd24ac772012-05-18 22:01:17 +09002266 } while (!__cmpxchg_double_slab(s, page,
Christoph Lameter213eeb92011-11-11 14:07:14 -06002267 freelist, counters,
2268 NULL, new.counters,
2269 "get_freelist"));
2270
2271 return freelist;
2272}
2273
2274/*
Christoph Lameter894b8782007-05-10 03:15:16 -07002275 * Slow path. The lockless freelist is empty or we need to perform
2276 * debugging duties.
Christoph Lameter81819f02007-05-06 14:49:36 -07002277 *
Christoph Lameter894b8782007-05-10 03:15:16 -07002278 * Processing is still very fast if new objects have been freed to the
2279 * regular freelist. In that case we simply take over the regular freelist
2280 * as the lockless freelist and zap the regular freelist.
Christoph Lameter81819f02007-05-06 14:49:36 -07002281 *
Christoph Lameter894b8782007-05-10 03:15:16 -07002282 * If that is not working then we fall back to the partial lists. We take the
2283 * first element of the freelist as the object to allocate now and move the
2284 * rest of the freelist to the lockless freelist.
2285 *
2286 * And if we were unable to get a new slab from the partial slab lists then
Christoph Lameter6446faa2008-02-15 23:45:26 -08002287 * we need to allocate a new slab. This is the slowest path since it involves
2288 * a call to the page allocator and the setup of a new slab.
Christoph Lameter81819f02007-05-06 14:49:36 -07002289 */
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03002290static void *__slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node,
2291 unsigned long addr, struct kmem_cache_cpu *c)
Christoph Lameter81819f02007-05-06 14:49:36 -07002292{
Christoph Lameter6faa6832012-05-09 10:09:51 -05002293 void *freelist;
Christoph Lameterf6e7def2012-05-09 10:09:58 -05002294 struct page *page;
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002295 unsigned long flags;
2296
2297 local_irq_save(flags);
2298#ifdef CONFIG_PREEMPT
2299 /*
2300 * We may have been preempted and rescheduled on a different
2301 * cpu before disabling interrupts. Need to reload cpu area
2302 * pointer.
2303 */
2304 c = this_cpu_ptr(s->cpu_slab);
2305#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07002306
Christoph Lameterf6e7def2012-05-09 10:09:58 -05002307 page = c->page;
2308 if (!page)
Christoph Lameter81819f02007-05-06 14:49:36 -07002309 goto new_slab;
Christoph Lameter49e22582011-08-09 16:12:27 -05002310redo:
Christoph Lameter6faa6832012-05-09 10:09:51 -05002311
Christoph Lameter57d437d2012-05-09 10:09:59 -05002312 if (unlikely(!node_match(page, node))) {
Christoph Lametere36a2652011-06-01 12:25:57 -05002313 stat(s, ALLOC_NODE_MISMATCH);
Christoph Lameterf6e7def2012-05-09 10:09:58 -05002314 deactivate_slab(s, page, c->freelist);
Christoph Lameterc17dda42012-05-09 10:09:57 -05002315 c->page = NULL;
2316 c->freelist = NULL;
Christoph Lameterfc59c052011-06-01 12:25:56 -05002317 goto new_slab;
2318 }
Christoph Lameter6446faa2008-02-15 23:45:26 -08002319
Mel Gorman072bb0a2012-07-31 16:43:58 -07002320 /*
2321 * By rights, we should be searching for a slab page that was
2322 * PFMEMALLOC but right now, we are losing the pfmemalloc
2323 * information when the page leaves the per-cpu allocator
2324 */
2325 if (unlikely(!pfmemalloc_match(page, gfpflags))) {
2326 deactivate_slab(s, page, c->freelist);
2327 c->page = NULL;
2328 c->freelist = NULL;
2329 goto new_slab;
2330 }
2331
Eric Dumazet73736e02011-12-13 04:57:06 +01002332 /* must check again c->freelist in case of cpu migration or IRQ */
Christoph Lameter6faa6832012-05-09 10:09:51 -05002333 freelist = c->freelist;
2334 if (freelist)
Eric Dumazet73736e02011-12-13 04:57:06 +01002335 goto load_freelist;
2336
Christoph Lameterf6e7def2012-05-09 10:09:58 -05002337 freelist = get_freelist(s, page);
Christoph Lameter6446faa2008-02-15 23:45:26 -08002338
Christoph Lameter6faa6832012-05-09 10:09:51 -05002339 if (!freelist) {
Christoph Lameter03e404a2011-06-01 12:25:58 -05002340 c->page = NULL;
2341 stat(s, DEACTIVATE_BYPASS);
Christoph Lameterfc59c052011-06-01 12:25:56 -05002342 goto new_slab;
Christoph Lameter03e404a2011-06-01 12:25:58 -05002343 }
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002344
Christoph Lameter81819f02007-05-06 14:49:36 -07002345 stat(s, ALLOC_REFILL);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08002346
Christoph Lameter894b8782007-05-10 03:15:16 -07002347load_freelist:
Christoph Lameter507effe2012-05-09 10:09:52 -05002348 /*
2349 * freelist is pointing to the list of objects to be used.
2350 * page is pointing to the page from which the objects are obtained.
2351 * That page must be frozen for per cpu allocations to work.
2352 */
Dave Hansena0132ac2014-01-29 14:05:50 -08002353 VM_BUG_ON(!c->page->frozen);
Christoph Lameter6faa6832012-05-09 10:09:51 -05002354 c->freelist = get_freepointer(s, freelist);
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002355 c->tid = next_tid(c->tid);
2356 local_irq_restore(flags);
Christoph Lameter6faa6832012-05-09 10:09:51 -05002357 return freelist;
Christoph Lameter81819f02007-05-06 14:49:36 -07002358
Christoph Lameter81819f02007-05-06 14:49:36 -07002359new_slab:
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002360
Christoph Lameter49e22582011-08-09 16:12:27 -05002361 if (c->partial) {
Christoph Lameterf6e7def2012-05-09 10:09:58 -05002362 page = c->page = c->partial;
2363 c->partial = page->next;
Christoph Lameter49e22582011-08-09 16:12:27 -05002364 stat(s, CPU_PARTIAL_ALLOC);
2365 c->freelist = NULL;
2366 goto redo;
Christoph Lameter81819f02007-05-06 14:49:36 -07002367 }
2368
Christoph Lameter188fd062012-05-09 10:09:55 -05002369 freelist = new_slab_objects(s, gfpflags, node, &c);
Christoph Lameterb811c202007-10-16 23:25:51 -07002370
Christoph Lameterf46974362012-05-09 10:09:54 -05002371 if (unlikely(!freelist)) {
David Rientjes9a02d692014-06-04 16:06:36 -07002372 slab_out_of_memory(s, gfpflags, node);
Christoph Lameterf46974362012-05-09 10:09:54 -05002373 local_irq_restore(flags);
2374 return NULL;
Christoph Lameter81819f02007-05-06 14:49:36 -07002375 }
Christoph Lameter894b8782007-05-10 03:15:16 -07002376
Christoph Lameterf6e7def2012-05-09 10:09:58 -05002377 page = c->page;
Christoph Lameter5091b742012-07-31 16:44:00 -07002378 if (likely(!kmem_cache_debug(s) && pfmemalloc_match(page, gfpflags)))
Christoph Lameter81819f02007-05-06 14:49:36 -07002379 goto load_freelist;
Christoph Lameter894b8782007-05-10 03:15:16 -07002380
Christoph Lameter497b66f2011-08-09 16:12:26 -05002381 /* Only entered in the debug case */
Chen Gangd0e0ac92013-07-15 09:05:29 +08002382 if (kmem_cache_debug(s) &&
2383 !alloc_debug_processing(s, page, freelist, addr))
Christoph Lameter497b66f2011-08-09 16:12:26 -05002384 goto new_slab; /* Slab failed checks. Next slab needed */
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002385
Christoph Lameterf6e7def2012-05-09 10:09:58 -05002386 deactivate_slab(s, page, get_freepointer(s, freelist));
Christoph Lameterc17dda42012-05-09 10:09:57 -05002387 c->page = NULL;
2388 c->freelist = NULL;
Christoph Lametera71ae472011-05-25 09:47:43 -05002389 local_irq_restore(flags);
Christoph Lameter6faa6832012-05-09 10:09:51 -05002390 return freelist;
Christoph Lameter894b8782007-05-10 03:15:16 -07002391}
2392
2393/*
2394 * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc)
2395 * have the fastpath folded into their functions. So no function call
2396 * overhead for requests that can be satisfied on the fastpath.
2397 *
2398 * The fastpath works by first checking if the lockless freelist can be used.
2399 * If not then __slab_alloc is called for slow processing.
2400 *
2401 * Otherwise we can simply pick the next object from the lockless free list.
2402 */
Ezequiel Garcia2b847c32012-09-08 17:47:58 -03002403static __always_inline void *slab_alloc_node(struct kmem_cache *s,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03002404 gfp_t gfpflags, int node, unsigned long addr)
Christoph Lameter894b8782007-05-10 03:15:16 -07002405{
Christoph Lameter894b8782007-05-10 03:15:16 -07002406 void **object;
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002407 struct kmem_cache_cpu *c;
Christoph Lameter57d437d2012-05-09 10:09:59 -05002408 struct page *page;
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002409 unsigned long tid;
Christoph Lameter1f842602008-01-07 23:20:30 -08002410
Christoph Lameterc016b0b2010-08-20 12:37:16 -05002411 if (slab_pre_alloc_hook(s, gfpflags))
Akinobu Mita773ff602008-12-23 19:37:01 +09002412 return NULL;
2413
Glauber Costad79923f2012-12-18 14:22:48 -08002414 s = memcg_kmem_get_cache(s, gfpflags);
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002415redo:
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002416 /*
2417 * Must read kmem_cache cpu data via this cpu ptr. Preemption is
2418 * enabled. We may switch back and forth between cpus while
2419 * reading from one cpu area. That does not matter as long
2420 * as we end up on the original cpu again when doing the cmpxchg.
Christoph Lameter7cccd802013-01-23 21:45:48 +00002421 *
2422 * Preemption is disabled for the retrieval of the tid because that
2423 * must occur from the current processor. We cannot allow rescheduling
2424 * on a different processor between the determination of the pointer
2425 * and the retrieval of the tid.
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002426 */
Christoph Lameter7cccd802013-01-23 21:45:48 +00002427 preempt_disable();
Christoph Lameter7c8e0182014-06-04 16:07:56 -07002428 c = this_cpu_ptr(s->cpu_slab);
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002429
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002430 /*
2431 * The transaction ids are globally unique per cpu and per operation on
2432 * a per cpu queue. Thus they can be guarantee that the cmpxchg_double
2433 * occurs on the right processor and that there was no operation on the
2434 * linked list in between.
2435 */
2436 tid = c->tid;
Christoph Lameter7cccd802013-01-23 21:45:48 +00002437 preempt_enable();
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002438
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06002439 object = c->freelist;
Christoph Lameter57d437d2012-05-09 10:09:59 -05002440 page = c->page;
Dave Hansen8eae1492014-06-04 16:06:37 -07002441 if (unlikely(!object || !node_match(page, node))) {
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002442 object = __slab_alloc(s, gfpflags, node, addr, c);
Dave Hansen8eae1492014-06-04 16:06:37 -07002443 stat(s, ALLOC_SLOWPATH);
2444 } else {
Eric Dumazet0ad95002011-12-16 16:25:34 +01002445 void *next_object = get_freepointer_safe(s, object);
2446
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002447 /*
Lucas De Marchi25985ed2011-03-30 22:57:33 -03002448 * The cmpxchg will only match if there was no additional
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002449 * operation and if we are on the right processor.
2450 *
Chen Gangd0e0ac92013-07-15 09:05:29 +08002451 * The cmpxchg does the following atomically (without lock
2452 * semantics!)
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002453 * 1. Relocate first pointer to the current per cpu area.
2454 * 2. Verify that tid and freelist have not been changed
2455 * 3. If they were not changed replace tid and freelist
2456 *
Chen Gangd0e0ac92013-07-15 09:05:29 +08002457 * Since this is without lock semantics the protection is only
2458 * against code executing on this cpu *not* from access by
2459 * other cpus.
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002460 */
Christoph Lameter933393f2011-12-22 11:58:51 -06002461 if (unlikely(!this_cpu_cmpxchg_double(
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002462 s->cpu_slab->freelist, s->cpu_slab->tid,
2463 object, tid,
Eric Dumazet0ad95002011-12-16 16:25:34 +01002464 next_object, next_tid(tid)))) {
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002465
2466 note_cmpxchg_failure("slab_alloc", s, tid);
2467 goto redo;
2468 }
Eric Dumazet0ad95002011-12-16 16:25:34 +01002469 prefetch_freepointer(s, next_object);
Christoph Lameter84e554e62009-12-18 16:26:23 -06002470 stat(s, ALLOC_FASTPATH);
Christoph Lameter894b8782007-05-10 03:15:16 -07002471 }
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002472
Pekka Enberg74e21342009-11-25 20:14:48 +02002473 if (unlikely(gfpflags & __GFP_ZERO) && object)
Christoph Lameter3b0efdf2012-06-13 10:24:57 -05002474 memset(object, 0, s->object_size);
Christoph Lameterd07dbea2007-07-17 04:03:23 -07002475
Christoph Lameterc016b0b2010-08-20 12:37:16 -05002476 slab_post_alloc_hook(s, gfpflags, object);
Vegard Nossum5a896d92008-04-04 00:54:48 +02002477
Christoph Lameter894b8782007-05-10 03:15:16 -07002478 return object;
Christoph Lameter81819f02007-05-06 14:49:36 -07002479}
2480
Ezequiel Garcia2b847c32012-09-08 17:47:58 -03002481static __always_inline void *slab_alloc(struct kmem_cache *s,
2482 gfp_t gfpflags, unsigned long addr)
2483{
2484 return slab_alloc_node(s, gfpflags, NUMA_NO_NODE, addr);
2485}
2486
Christoph Lameter81819f02007-05-06 14:49:36 -07002487void *kmem_cache_alloc(struct kmem_cache *s, gfp_t gfpflags)
2488{
Ezequiel Garcia2b847c32012-09-08 17:47:58 -03002489 void *ret = slab_alloc(s, gfpflags, _RET_IP_);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002490
Chen Gangd0e0ac92013-07-15 09:05:29 +08002491 trace_kmem_cache_alloc(_RET_IP_, ret, s->object_size,
2492 s->size, gfpflags);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002493
2494 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07002495}
2496EXPORT_SYMBOL(kmem_cache_alloc);
2497
Li Zefan0f24f122009-12-11 15:45:30 +08002498#ifdef CONFIG_TRACING
Richard Kennedy4a923792010-10-21 10:29:19 +01002499void *kmem_cache_alloc_trace(struct kmem_cache *s, gfp_t gfpflags, size_t size)
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002500{
Ezequiel Garcia2b847c32012-09-08 17:47:58 -03002501 void *ret = slab_alloc(s, gfpflags, _RET_IP_);
Richard Kennedy4a923792010-10-21 10:29:19 +01002502 trace_kmalloc(_RET_IP_, ret, size, s->size, gfpflags);
2503 return ret;
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002504}
Richard Kennedy4a923792010-10-21 10:29:19 +01002505EXPORT_SYMBOL(kmem_cache_alloc_trace);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002506#endif
2507
Christoph Lameter81819f02007-05-06 14:49:36 -07002508#ifdef CONFIG_NUMA
2509void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t gfpflags, int node)
2510{
Ezequiel Garcia2b847c32012-09-08 17:47:58 -03002511 void *ret = slab_alloc_node(s, gfpflags, node, _RET_IP_);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002512
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02002513 trace_kmem_cache_alloc_node(_RET_IP_, ret,
Christoph Lameter3b0efdf2012-06-13 10:24:57 -05002514 s->object_size, s->size, gfpflags, node);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002515
2516 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07002517}
2518EXPORT_SYMBOL(kmem_cache_alloc_node);
Christoph Lameter81819f02007-05-06 14:49:36 -07002519
Li Zefan0f24f122009-12-11 15:45:30 +08002520#ifdef CONFIG_TRACING
Richard Kennedy4a923792010-10-21 10:29:19 +01002521void *kmem_cache_alloc_node_trace(struct kmem_cache *s,
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002522 gfp_t gfpflags,
Richard Kennedy4a923792010-10-21 10:29:19 +01002523 int node, size_t size)
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002524{
Ezequiel Garcia2b847c32012-09-08 17:47:58 -03002525 void *ret = slab_alloc_node(s, gfpflags, node, _RET_IP_);
Richard Kennedy4a923792010-10-21 10:29:19 +01002526
2527 trace_kmalloc_node(_RET_IP_, ret,
2528 size, s->size, gfpflags, node);
2529 return ret;
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002530}
Richard Kennedy4a923792010-10-21 10:29:19 +01002531EXPORT_SYMBOL(kmem_cache_alloc_node_trace);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002532#endif
Namhyung Kim5d1f57e2010-09-29 21:02:15 +09002533#endif
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002534
Christoph Lameter81819f02007-05-06 14:49:36 -07002535/*
Christoph Lameter894b8782007-05-10 03:15:16 -07002536 * Slow patch handling. This may still be called frequently since objects
2537 * have a longer lifetime than the cpu slabs in most processing loads.
Christoph Lameter81819f02007-05-06 14:49:36 -07002538 *
Christoph Lameter894b8782007-05-10 03:15:16 -07002539 * So we still attempt to reduce cache line usage. Just take the slab
2540 * lock and free the item. If there is no additional partial page
2541 * handling required then we can return immediately.
Christoph Lameter81819f02007-05-06 14:49:36 -07002542 */
Christoph Lameter894b8782007-05-10 03:15:16 -07002543static void __slab_free(struct kmem_cache *s, struct page *page,
Christoph Lameterff120592009-12-18 16:26:22 -06002544 void *x, unsigned long addr)
Christoph Lameter81819f02007-05-06 14:49:36 -07002545{
2546 void *prior;
2547 void **object = (void *)x;
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002548 int was_frozen;
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002549 struct page new;
2550 unsigned long counters;
2551 struct kmem_cache_node *n = NULL;
Christoph Lameter61728d12011-06-01 12:25:51 -05002552 unsigned long uninitialized_var(flags);
Christoph Lameter81819f02007-05-06 14:49:36 -07002553
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002554 stat(s, FREE_SLOWPATH);
Christoph Lameter81819f02007-05-06 14:49:36 -07002555
Christoph Lameter19c7ff92012-05-30 12:54:46 -05002556 if (kmem_cache_debug(s) &&
2557 !(n = free_debug_processing(s, page, x, addr, &flags)))
Christoph Lameter80f08c12011-06-01 12:25:55 -05002558 return;
Christoph Lameter6446faa2008-02-15 23:45:26 -08002559
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002560 do {
Joonsoo Kim837d6782012-08-16 00:02:40 +09002561 if (unlikely(n)) {
2562 spin_unlock_irqrestore(&n->list_lock, flags);
2563 n = NULL;
2564 }
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002565 prior = page->freelist;
2566 counters = page->counters;
2567 set_freepointer(s, object, prior);
2568 new.counters = counters;
2569 was_frozen = new.frozen;
2570 new.inuse--;
Joonsoo Kim837d6782012-08-16 00:02:40 +09002571 if ((!new.inuse || !prior) && !was_frozen) {
Christoph Lameter49e22582011-08-09 16:12:27 -05002572
Peter Zijlstrac65c1872014-01-10 13:23:49 +01002573 if (kmem_cache_has_cpu_partial(s) && !prior) {
Christoph Lameter49e22582011-08-09 16:12:27 -05002574
2575 /*
Chen Gangd0e0ac92013-07-15 09:05:29 +08002576 * Slab was on no list before and will be
2577 * partially empty
2578 * We can defer the list move and instead
2579 * freeze it.
Christoph Lameter49e22582011-08-09 16:12:27 -05002580 */
2581 new.frozen = 1;
2582
Peter Zijlstrac65c1872014-01-10 13:23:49 +01002583 } else { /* Needs to be taken off a list */
Christoph Lameter49e22582011-08-09 16:12:27 -05002584
2585 n = get_node(s, page_to_nid(page));
2586 /*
2587 * Speculatively acquire the list_lock.
2588 * If the cmpxchg does not succeed then we may
2589 * drop the list_lock without any processing.
2590 *
2591 * Otherwise the list_lock will synchronize with
2592 * other processors updating the list of slabs.
2593 */
2594 spin_lock_irqsave(&n->list_lock, flags);
2595
2596 }
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002597 }
Christoph Lameter81819f02007-05-06 14:49:36 -07002598
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002599 } while (!cmpxchg_double_slab(s, page,
2600 prior, counters,
2601 object, new.counters,
2602 "__slab_free"));
Christoph Lameter81819f02007-05-06 14:49:36 -07002603
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002604 if (likely(!n)) {
Christoph Lameter49e22582011-08-09 16:12:27 -05002605
2606 /*
2607 * If we just froze the page then put it onto the
2608 * per cpu partial list.
2609 */
Alex Shi8028dce2012-02-03 23:34:56 +08002610 if (new.frozen && !was_frozen) {
Christoph Lameter49e22582011-08-09 16:12:27 -05002611 put_cpu_partial(s, page, 1);
Alex Shi8028dce2012-02-03 23:34:56 +08002612 stat(s, CPU_PARTIAL_FREE);
2613 }
Christoph Lameter49e22582011-08-09 16:12:27 -05002614 /*
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002615 * The list lock was not taken therefore no list
2616 * activity can be necessary.
2617 */
2618 if (was_frozen)
2619 stat(s, FREE_FROZEN);
Christoph Lameter80f08c12011-06-01 12:25:55 -05002620 return;
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002621 }
Christoph Lameter81819f02007-05-06 14:49:36 -07002622
Joonsoo Kim8a5b20a2014-07-02 15:22:35 -07002623 if (unlikely(!new.inuse && n->nr_partial >= s->min_partial))
Joonsoo Kim837d6782012-08-16 00:02:40 +09002624 goto slab_empty;
Christoph Lameter81819f02007-05-06 14:49:36 -07002625
Joonsoo Kim837d6782012-08-16 00:02:40 +09002626 /*
2627 * Objects left in the slab. If it was not on the partial list before
2628 * then add it.
2629 */
Joonsoo Kim345c9052013-06-19 14:05:52 +09002630 if (!kmem_cache_has_cpu_partial(s) && unlikely(!prior)) {
2631 if (kmem_cache_debug(s))
Peter Zijlstrac65c1872014-01-10 13:23:49 +01002632 remove_full(s, n, page);
Joonsoo Kim837d6782012-08-16 00:02:40 +09002633 add_partial(n, page, DEACTIVATE_TO_TAIL);
2634 stat(s, FREE_ADD_PARTIAL);
Christoph Lameter81819f02007-05-06 14:49:36 -07002635 }
Christoph Lameter80f08c12011-06-01 12:25:55 -05002636 spin_unlock_irqrestore(&n->list_lock, flags);
Christoph Lameter81819f02007-05-06 14:49:36 -07002637 return;
2638
2639slab_empty:
Christoph Lametera973e9d2008-03-01 13:40:44 -08002640 if (prior) {
Christoph Lameter81819f02007-05-06 14:49:36 -07002641 /*
Christoph Lameter6fbabb22011-08-08 11:16:56 -05002642 * Slab on the partial list.
Christoph Lameter81819f02007-05-06 14:49:36 -07002643 */
Christoph Lameter5cc6eee2011-06-01 12:25:50 -05002644 remove_partial(n, page);
Christoph Lameter84e554e62009-12-18 16:26:23 -06002645 stat(s, FREE_REMOVE_PARTIAL);
Peter Zijlstrac65c1872014-01-10 13:23:49 +01002646 } else {
Christoph Lameter6fbabb22011-08-08 11:16:56 -05002647 /* Slab must be on the full list */
Peter Zijlstrac65c1872014-01-10 13:23:49 +01002648 remove_full(s, n, page);
2649 }
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002650
Christoph Lameter80f08c12011-06-01 12:25:55 -05002651 spin_unlock_irqrestore(&n->list_lock, flags);
Christoph Lameter84e554e62009-12-18 16:26:23 -06002652 stat(s, FREE_SLAB);
Christoph Lameter81819f02007-05-06 14:49:36 -07002653 discard_slab(s, page);
Christoph Lameter81819f02007-05-06 14:49:36 -07002654}
2655
Christoph Lameter894b8782007-05-10 03:15:16 -07002656/*
2657 * Fastpath with forced inlining to produce a kfree and kmem_cache_free that
2658 * can perform fastpath freeing without additional function calls.
2659 *
2660 * The fastpath is only possible if we are freeing to the current cpu slab
2661 * of this processor. This typically the case if we have just allocated
2662 * the item before.
2663 *
2664 * If fastpath is not possible then fall back to __slab_free where we deal
2665 * with all sorts of special processing.
2666 */
Pekka Enberg06428782008-01-07 23:20:27 -08002667static __always_inline void slab_free(struct kmem_cache *s,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03002668 struct page *page, void *x, unsigned long addr)
Christoph Lameter894b8782007-05-10 03:15:16 -07002669{
2670 void **object = (void *)x;
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002671 struct kmem_cache_cpu *c;
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002672 unsigned long tid;
Christoph Lameter1f842602008-01-07 23:20:30 -08002673
Christoph Lameterc016b0b2010-08-20 12:37:16 -05002674 slab_free_hook(s, x);
2675
Christoph Lametera24c5a02011-03-15 12:45:21 -05002676redo:
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002677 /*
2678 * Determine the currently cpus per cpu slab.
2679 * The cpu may change afterward. However that does not matter since
2680 * data is retrieved via this pointer. If we are on the same cpu
2681 * during the cmpxchg then the free will succedd.
2682 */
Christoph Lameter7cccd802013-01-23 21:45:48 +00002683 preempt_disable();
Christoph Lameter7c8e0182014-06-04 16:07:56 -07002684 c = this_cpu_ptr(s->cpu_slab);
Christoph Lameterc016b0b2010-08-20 12:37:16 -05002685
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002686 tid = c->tid;
Christoph Lameter7cccd802013-01-23 21:45:48 +00002687 preempt_enable();
Christoph Lameterc016b0b2010-08-20 12:37:16 -05002688
Christoph Lameter442b06b2011-05-17 16:29:31 -05002689 if (likely(page == c->page)) {
Christoph Lameterff120592009-12-18 16:26:22 -06002690 set_freepointer(s, object, c->freelist);
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002691
Christoph Lameter933393f2011-12-22 11:58:51 -06002692 if (unlikely(!this_cpu_cmpxchg_double(
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002693 s->cpu_slab->freelist, s->cpu_slab->tid,
2694 c->freelist, tid,
2695 object, next_tid(tid)))) {
2696
2697 note_cmpxchg_failure("slab_free", s, tid);
2698 goto redo;
2699 }
Christoph Lameter84e554e62009-12-18 16:26:23 -06002700 stat(s, FREE_FASTPATH);
Christoph Lameter894b8782007-05-10 03:15:16 -07002701 } else
Christoph Lameterff120592009-12-18 16:26:22 -06002702 __slab_free(s, page, x, addr);
Christoph Lameter894b8782007-05-10 03:15:16 -07002703
Christoph Lameter894b8782007-05-10 03:15:16 -07002704}
2705
Christoph Lameter81819f02007-05-06 14:49:36 -07002706void kmem_cache_free(struct kmem_cache *s, void *x)
2707{
Glauber Costab9ce5ef2012-12-18 14:22:46 -08002708 s = cache_from_obj(s, x);
2709 if (!s)
Christoph Lameter79576102012-09-04 23:06:14 +00002710 return;
Glauber Costab9ce5ef2012-12-18 14:22:46 -08002711 slab_free(s, virt_to_head_page(x), x, _RET_IP_);
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02002712 trace_kmem_cache_free(_RET_IP_, x);
Christoph Lameter81819f02007-05-06 14:49:36 -07002713}
2714EXPORT_SYMBOL(kmem_cache_free);
2715
Christoph Lameter81819f02007-05-06 14:49:36 -07002716/*
Christoph Lameter672bba32007-05-09 02:32:39 -07002717 * Object placement in a slab is made very easy because we always start at
2718 * offset 0. If we tune the size of the object to the alignment then we can
2719 * get the required alignment by putting one properly sized object after
2720 * another.
Christoph Lameter81819f02007-05-06 14:49:36 -07002721 *
2722 * Notice that the allocation order determines the sizes of the per cpu
2723 * caches. Each processor has always one slab available for allocations.
2724 * Increasing the allocation order reduces the number of times that slabs
Christoph Lameter672bba32007-05-09 02:32:39 -07002725 * must be moved on and off the partial lists and is therefore a factor in
Christoph Lameter81819f02007-05-06 14:49:36 -07002726 * locking overhead.
Christoph Lameter81819f02007-05-06 14:49:36 -07002727 */
2728
2729/*
2730 * Mininum / Maximum order of slab pages. This influences locking overhead
2731 * and slab fragmentation. A higher order reduces the number of partial slabs
2732 * and increases the number of allocations possible without having to
2733 * take the list_lock.
2734 */
2735static int slub_min_order;
Christoph Lameter114e9e82008-04-14 19:11:41 +03002736static int slub_max_order = PAGE_ALLOC_COSTLY_ORDER;
Christoph Lameter9b2cd502008-04-14 19:11:41 +03002737static int slub_min_objects;
Christoph Lameter81819f02007-05-06 14:49:36 -07002738
2739/*
2740 * Merge control. If this is set then no merging of slab caches will occur.
Christoph Lameter672bba32007-05-09 02:32:39 -07002741 * (Could be removed. This was introduced to pacify the merge skeptics.)
Christoph Lameter81819f02007-05-06 14:49:36 -07002742 */
2743static int slub_nomerge;
2744
2745/*
Christoph Lameter81819f02007-05-06 14:49:36 -07002746 * Calculate the order of allocation given an slab object size.
2747 *
Christoph Lameter672bba32007-05-09 02:32:39 -07002748 * The order of allocation has significant impact on performance and other
2749 * system components. Generally order 0 allocations should be preferred since
2750 * order 0 does not cause fragmentation in the page allocator. Larger objects
2751 * be problematic to put into order 0 slabs because there may be too much
Christoph Lameterc124f5b2008-04-14 19:13:29 +03002752 * unused space left. We go to a higher order if more than 1/16th of the slab
Christoph Lameter672bba32007-05-09 02:32:39 -07002753 * would be wasted.
Christoph Lameter81819f02007-05-06 14:49:36 -07002754 *
Christoph Lameter672bba32007-05-09 02:32:39 -07002755 * In order to reach satisfactory performance we must ensure that a minimum
2756 * number of objects is in one slab. Otherwise we may generate too much
2757 * activity on the partial lists which requires taking the list_lock. This is
2758 * less a concern for large slabs though which are rarely used.
Christoph Lameter81819f02007-05-06 14:49:36 -07002759 *
Christoph Lameter672bba32007-05-09 02:32:39 -07002760 * slub_max_order specifies the order where we begin to stop considering the
2761 * number of objects in a slab as critical. If we reach slub_max_order then
2762 * we try to keep the page order as low as possible. So we accept more waste
2763 * of space in favor of a small page order.
2764 *
2765 * Higher order allocations also allow the placement of more objects in a
2766 * slab and thereby reduce object handling overhead. If the user has
2767 * requested a higher mininum order then we start with that one instead of
2768 * the smallest order which will fit the object.
Christoph Lameter81819f02007-05-06 14:49:36 -07002769 */
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002770static inline int slab_order(int size, int min_objects,
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002771 int max_order, int fract_leftover, int reserved)
Christoph Lameter81819f02007-05-06 14:49:36 -07002772{
2773 int order;
2774 int rem;
Christoph Lameter6300ea72007-07-17 04:03:20 -07002775 int min_order = slub_min_order;
Christoph Lameter81819f02007-05-06 14:49:36 -07002776
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002777 if (order_objects(min_order, size, reserved) > MAX_OBJS_PER_PAGE)
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +04002778 return get_order(size * MAX_OBJS_PER_PAGE) - 1;
Christoph Lameter39b26462008-04-14 19:11:30 +03002779
Christoph Lameter6300ea72007-07-17 04:03:20 -07002780 for (order = max(min_order,
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002781 fls(min_objects * size - 1) - PAGE_SHIFT);
2782 order <= max_order; order++) {
2783
Christoph Lameter81819f02007-05-06 14:49:36 -07002784 unsigned long slab_size = PAGE_SIZE << order;
2785
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002786 if (slab_size < min_objects * size + reserved)
Christoph Lameter81819f02007-05-06 14:49:36 -07002787 continue;
2788
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002789 rem = (slab_size - reserved) % size;
Christoph Lameter81819f02007-05-06 14:49:36 -07002790
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002791 if (rem <= slab_size / fract_leftover)
Christoph Lameter81819f02007-05-06 14:49:36 -07002792 break;
2793
2794 }
Christoph Lameter672bba32007-05-09 02:32:39 -07002795
Christoph Lameter81819f02007-05-06 14:49:36 -07002796 return order;
2797}
2798
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002799static inline int calculate_order(int size, int reserved)
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002800{
2801 int order;
2802 int min_objects;
2803 int fraction;
Zhang Yanmine8120ff2009-02-12 18:00:17 +02002804 int max_objects;
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002805
2806 /*
2807 * Attempt to find best configuration for a slab. This
2808 * works by first attempting to generate a layout with
2809 * the best configuration and backing off gradually.
2810 *
2811 * First we reduce the acceptable waste in a slab. Then
2812 * we reduce the minimum objects required in a slab.
2813 */
2814 min_objects = slub_min_objects;
Christoph Lameter9b2cd502008-04-14 19:11:41 +03002815 if (!min_objects)
2816 min_objects = 4 * (fls(nr_cpu_ids) + 1);
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002817 max_objects = order_objects(slub_max_order, size, reserved);
Zhang Yanmine8120ff2009-02-12 18:00:17 +02002818 min_objects = min(min_objects, max_objects);
2819
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002820 while (min_objects > 1) {
Christoph Lameterc124f5b2008-04-14 19:13:29 +03002821 fraction = 16;
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002822 while (fraction >= 4) {
2823 order = slab_order(size, min_objects,
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002824 slub_max_order, fraction, reserved);
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002825 if (order <= slub_max_order)
2826 return order;
2827 fraction /= 2;
2828 }
Amerigo Wang5086c389c2009-08-19 21:44:13 +03002829 min_objects--;
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002830 }
2831
2832 /*
2833 * We were unable to place multiple objects in a slab. Now
2834 * lets see if we can place a single object there.
2835 */
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002836 order = slab_order(size, 1, slub_max_order, 1, reserved);
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002837 if (order <= slub_max_order)
2838 return order;
2839
2840 /*
2841 * Doh this slab cannot be placed using slub_max_order.
2842 */
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002843 order = slab_order(size, 1, MAX_ORDER, 1, reserved);
David Rientjes818cf592009-04-23 09:58:22 +03002844 if (order < MAX_ORDER)
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002845 return order;
2846 return -ENOSYS;
2847}
2848
Pekka Enberg5595cff2008-08-05 09:28:47 +03002849static void
Joonsoo Kim40534972012-05-11 00:50:47 +09002850init_kmem_cache_node(struct kmem_cache_node *n)
Christoph Lameter81819f02007-05-06 14:49:36 -07002851{
2852 n->nr_partial = 0;
Christoph Lameter81819f02007-05-06 14:49:36 -07002853 spin_lock_init(&n->list_lock);
2854 INIT_LIST_HEAD(&n->partial);
Christoph Lameter8ab13722007-07-17 04:03:32 -07002855#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter0f389ec2008-04-14 18:53:02 +03002856 atomic_long_set(&n->nr_slabs, 0);
Salman Qazi02b71b72008-09-11 12:25:41 -07002857 atomic_long_set(&n->total_objects, 0);
Christoph Lameter643b1132007-05-06 14:49:42 -07002858 INIT_LIST_HEAD(&n->full);
Christoph Lameter8ab13722007-07-17 04:03:32 -07002859#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07002860}
2861
Christoph Lameter55136592010-08-20 12:37:13 -05002862static inline int alloc_kmem_cache_cpus(struct kmem_cache *s)
Christoph Lameter4c93c3552007-10-16 01:26:08 -07002863{
Christoph Lameter6c182dc2010-08-20 12:37:14 -05002864 BUILD_BUG_ON(PERCPU_DYNAMIC_EARLY_SIZE <
Christoph Lameter95a05b42013-01-10 19:14:19 +00002865 KMALLOC_SHIFT_HIGH * sizeof(struct kmem_cache_cpu));
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06002866
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002867 /*
Chris Metcalfd4d84fe2011-06-02 10:19:41 -04002868 * Must align to double word boundary for the double cmpxchg
2869 * instructions to work; see __pcpu_double_call_return_bool().
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002870 */
Chris Metcalfd4d84fe2011-06-02 10:19:41 -04002871 s->cpu_slab = __alloc_percpu(sizeof(struct kmem_cache_cpu),
2872 2 * sizeof(void *));
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06002873
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002874 if (!s->cpu_slab)
2875 return 0;
2876
2877 init_kmem_cache_cpus(s);
2878
2879 return 1;
Christoph Lameter4c93c3552007-10-16 01:26:08 -07002880}
Christoph Lameter4c93c3552007-10-16 01:26:08 -07002881
Christoph Lameter51df1142010-08-20 12:37:15 -05002882static struct kmem_cache *kmem_cache_node;
2883
Christoph Lameter81819f02007-05-06 14:49:36 -07002884/*
2885 * No kmalloc_node yet so do it by hand. We know that this is the first
2886 * slab on the node for this slabcache. There are no concurrent accesses
2887 * possible.
2888 *
Zhi Yong Wu721ae222013-11-08 20:47:37 +08002889 * Note that this function only works on the kmem_cache_node
2890 * when allocating for the kmem_cache_node. This is used for bootstrapping
Christoph Lameter4c93c3552007-10-16 01:26:08 -07002891 * memory on a fresh node that has no slab structures yet.
Christoph Lameter81819f02007-05-06 14:49:36 -07002892 */
Christoph Lameter55136592010-08-20 12:37:13 -05002893static void early_kmem_cache_node_alloc(int node)
Christoph Lameter81819f02007-05-06 14:49:36 -07002894{
2895 struct page *page;
2896 struct kmem_cache_node *n;
2897
Christoph Lameter51df1142010-08-20 12:37:15 -05002898 BUG_ON(kmem_cache_node->size < sizeof(struct kmem_cache_node));
Christoph Lameter81819f02007-05-06 14:49:36 -07002899
Christoph Lameter51df1142010-08-20 12:37:15 -05002900 page = new_slab(kmem_cache_node, GFP_NOWAIT, node);
Christoph Lameter81819f02007-05-06 14:49:36 -07002901
2902 BUG_ON(!page);
Christoph Lametera2f92ee2007-08-22 14:01:57 -07002903 if (page_to_nid(page) != node) {
Fabian Frederickf9f58282014-06-04 16:06:34 -07002904 pr_err("SLUB: Unable to allocate memory from node %d\n", node);
2905 pr_err("SLUB: Allocating a useless per node structure in order to be able to continue\n");
Christoph Lametera2f92ee2007-08-22 14:01:57 -07002906 }
2907
Christoph Lameter81819f02007-05-06 14:49:36 -07002908 n = page->freelist;
2909 BUG_ON(!n);
Christoph Lameter51df1142010-08-20 12:37:15 -05002910 page->freelist = get_freepointer(kmem_cache_node, n);
Christoph Lametere6e82ea2011-08-09 16:12:24 -05002911 page->inuse = 1;
Christoph Lameter8cb0a502011-06-01 12:25:46 -05002912 page->frozen = 0;
Christoph Lameter51df1142010-08-20 12:37:15 -05002913 kmem_cache_node->node[node] = n;
Christoph Lameter8ab13722007-07-17 04:03:32 -07002914#ifdef CONFIG_SLUB_DEBUG
Christoph Lameterf7cb1932010-09-29 07:15:01 -05002915 init_object(kmem_cache_node, n, SLUB_RED_ACTIVE);
Christoph Lameter51df1142010-08-20 12:37:15 -05002916 init_tracking(kmem_cache_node, n);
Christoph Lameter8ab13722007-07-17 04:03:32 -07002917#endif
Joonsoo Kim40534972012-05-11 00:50:47 +09002918 init_kmem_cache_node(n);
Christoph Lameter51df1142010-08-20 12:37:15 -05002919 inc_slabs_node(kmem_cache_node, node, page->objects);
Christoph Lameter6446faa2008-02-15 23:45:26 -08002920
Dave Hansen67b6c902014-01-24 07:20:23 -08002921 /*
Steven Rostedt1e4dd942014-02-10 14:25:46 -08002922 * No locks need to be taken here as it has just been
2923 * initialized and there is no concurrent access.
Dave Hansen67b6c902014-01-24 07:20:23 -08002924 */
Steven Rostedt1e4dd942014-02-10 14:25:46 -08002925 __add_partial(n, page, DEACTIVATE_TO_HEAD);
Christoph Lameter81819f02007-05-06 14:49:36 -07002926}
2927
2928static void free_kmem_cache_nodes(struct kmem_cache *s)
2929{
2930 int node;
2931
Christoph Lameterf64dc582007-10-16 01:25:33 -07002932 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter81819f02007-05-06 14:49:36 -07002933 struct kmem_cache_node *n = s->node[node];
Christoph Lameter51df1142010-08-20 12:37:15 -05002934
Alexander Duyck73367bd2010-05-21 14:41:35 -07002935 if (n)
Christoph Lameter51df1142010-08-20 12:37:15 -05002936 kmem_cache_free(kmem_cache_node, n);
2937
Christoph Lameter81819f02007-05-06 14:49:36 -07002938 s->node[node] = NULL;
2939 }
2940}
2941
Christoph Lameter55136592010-08-20 12:37:13 -05002942static int init_kmem_cache_nodes(struct kmem_cache *s)
Christoph Lameter81819f02007-05-06 14:49:36 -07002943{
2944 int node;
Christoph Lameter81819f02007-05-06 14:49:36 -07002945
Christoph Lameterf64dc582007-10-16 01:25:33 -07002946 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter81819f02007-05-06 14:49:36 -07002947 struct kmem_cache_node *n;
2948
Alexander Duyck73367bd2010-05-21 14:41:35 -07002949 if (slab_state == DOWN) {
Christoph Lameter55136592010-08-20 12:37:13 -05002950 early_kmem_cache_node_alloc(node);
Alexander Duyck73367bd2010-05-21 14:41:35 -07002951 continue;
Christoph Lameter81819f02007-05-06 14:49:36 -07002952 }
Christoph Lameter51df1142010-08-20 12:37:15 -05002953 n = kmem_cache_alloc_node(kmem_cache_node,
Christoph Lameter55136592010-08-20 12:37:13 -05002954 GFP_KERNEL, node);
Alexander Duyck73367bd2010-05-21 14:41:35 -07002955
2956 if (!n) {
2957 free_kmem_cache_nodes(s);
2958 return 0;
2959 }
2960
Christoph Lameter81819f02007-05-06 14:49:36 -07002961 s->node[node] = n;
Joonsoo Kim40534972012-05-11 00:50:47 +09002962 init_kmem_cache_node(n);
Christoph Lameter81819f02007-05-06 14:49:36 -07002963 }
2964 return 1;
2965}
Christoph Lameter81819f02007-05-06 14:49:36 -07002966
David Rientjesc0bdb232009-02-25 09:16:35 +02002967static void set_min_partial(struct kmem_cache *s, unsigned long min)
David Rientjes3b89d7d2009-02-22 17:40:07 -08002968{
2969 if (min < MIN_PARTIAL)
2970 min = MIN_PARTIAL;
2971 else if (min > MAX_PARTIAL)
2972 min = MAX_PARTIAL;
2973 s->min_partial = min;
2974}
2975
Christoph Lameter81819f02007-05-06 14:49:36 -07002976/*
2977 * calculate_sizes() determines the order and the distribution of data within
2978 * a slab object.
2979 */
Christoph Lameter06b285d2008-04-14 19:11:41 +03002980static int calculate_sizes(struct kmem_cache *s, int forced_order)
Christoph Lameter81819f02007-05-06 14:49:36 -07002981{
2982 unsigned long flags = s->flags;
Christoph Lameter3b0efdf2012-06-13 10:24:57 -05002983 unsigned long size = s->object_size;
Christoph Lameter834f3d12008-04-14 19:11:31 +03002984 int order;
Christoph Lameter81819f02007-05-06 14:49:36 -07002985
2986 /*
Christoph Lameterd8b42bf2008-02-15 23:45:25 -08002987 * Round up object size to the next word boundary. We can only
2988 * place the free pointer at word boundaries and this determines
2989 * the possible location of the free pointer.
2990 */
2991 size = ALIGN(size, sizeof(void *));
2992
2993#ifdef CONFIG_SLUB_DEBUG
2994 /*
Christoph Lameter81819f02007-05-06 14:49:36 -07002995 * Determine if we can poison the object itself. If the user of
2996 * the slab may touch the object after free or before allocation
2997 * then we should never poison the object itself.
2998 */
2999 if ((flags & SLAB_POISON) && !(flags & SLAB_DESTROY_BY_RCU) &&
Christoph Lameterc59def92007-05-16 22:10:50 -07003000 !s->ctor)
Christoph Lameter81819f02007-05-06 14:49:36 -07003001 s->flags |= __OBJECT_POISON;
3002 else
3003 s->flags &= ~__OBJECT_POISON;
3004
Christoph Lameter81819f02007-05-06 14:49:36 -07003005
3006 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07003007 * If we are Redzoning then check if there is some space between the
Christoph Lameter81819f02007-05-06 14:49:36 -07003008 * end of the object and the free pointer. If not then add an
Christoph Lameter672bba32007-05-09 02:32:39 -07003009 * additional word to have some bytes to store Redzone information.
Christoph Lameter81819f02007-05-06 14:49:36 -07003010 */
Christoph Lameter3b0efdf2012-06-13 10:24:57 -05003011 if ((flags & SLAB_RED_ZONE) && size == s->object_size)
Christoph Lameter81819f02007-05-06 14:49:36 -07003012 size += sizeof(void *);
Christoph Lameter41ecc552007-05-09 02:32:44 -07003013#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07003014
3015 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07003016 * With that we have determined the number of bytes in actual use
3017 * by the object. This is the potential offset to the free pointer.
Christoph Lameter81819f02007-05-06 14:49:36 -07003018 */
3019 s->inuse = size;
3020
3021 if (((flags & (SLAB_DESTROY_BY_RCU | SLAB_POISON)) ||
Christoph Lameterc59def92007-05-16 22:10:50 -07003022 s->ctor)) {
Christoph Lameter81819f02007-05-06 14:49:36 -07003023 /*
3024 * Relocate free pointer after the object if it is not
3025 * permitted to overwrite the first word of the object on
3026 * kmem_cache_free.
3027 *
3028 * This is the case if we do RCU, have a constructor or
3029 * destructor or are poisoning the objects.
3030 */
3031 s->offset = size;
3032 size += sizeof(void *);
3033 }
3034
Christoph Lameterc12b3c62007-05-23 13:57:31 -07003035#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter81819f02007-05-06 14:49:36 -07003036 if (flags & SLAB_STORE_USER)
3037 /*
3038 * Need to store information about allocs and frees after
3039 * the object.
3040 */
3041 size += 2 * sizeof(struct track);
3042
Christoph Lameterbe7b3fb2007-05-09 02:32:36 -07003043 if (flags & SLAB_RED_ZONE)
Christoph Lameter81819f02007-05-06 14:49:36 -07003044 /*
3045 * Add some empty padding so that we can catch
3046 * overwrites from earlier objects rather than let
3047 * tracking information or the free pointer be
Frederik Schwarzer0211a9c2008-12-29 22:14:56 +01003048 * corrupted if a user writes before the start
Christoph Lameter81819f02007-05-06 14:49:36 -07003049 * of the object.
3050 */
3051 size += sizeof(void *);
Christoph Lameter41ecc552007-05-09 02:32:44 -07003052#endif
Christoph Lameter672bba32007-05-09 02:32:39 -07003053
Christoph Lameter81819f02007-05-06 14:49:36 -07003054 /*
Christoph Lameter81819f02007-05-06 14:49:36 -07003055 * SLUB stores one object immediately after another beginning from
3056 * offset 0. In order to align the objects we have to simply size
3057 * each object to conform to the alignment.
3058 */
Christoph Lameter45906852012-11-28 16:23:16 +00003059 size = ALIGN(size, s->align);
Christoph Lameter81819f02007-05-06 14:49:36 -07003060 s->size = size;
Christoph Lameter06b285d2008-04-14 19:11:41 +03003061 if (forced_order >= 0)
3062 order = forced_order;
3063 else
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08003064 order = calculate_order(size, s->reserved);
Christoph Lameter81819f02007-05-06 14:49:36 -07003065
Christoph Lameter834f3d12008-04-14 19:11:31 +03003066 if (order < 0)
Christoph Lameter81819f02007-05-06 14:49:36 -07003067 return 0;
3068
Christoph Lameterb7a49f02008-02-14 14:21:32 -08003069 s->allocflags = 0;
Christoph Lameter834f3d12008-04-14 19:11:31 +03003070 if (order)
Christoph Lameterb7a49f02008-02-14 14:21:32 -08003071 s->allocflags |= __GFP_COMP;
3072
3073 if (s->flags & SLAB_CACHE_DMA)
Christoph Lameter2c59dd62013-01-10 19:14:19 +00003074 s->allocflags |= GFP_DMA;
Christoph Lameterb7a49f02008-02-14 14:21:32 -08003075
3076 if (s->flags & SLAB_RECLAIM_ACCOUNT)
3077 s->allocflags |= __GFP_RECLAIMABLE;
3078
Christoph Lameter81819f02007-05-06 14:49:36 -07003079 /*
3080 * Determine the number of objects per slab
3081 */
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08003082 s->oo = oo_make(order, size, s->reserved);
3083 s->min = oo_make(get_order(size), size, s->reserved);
Christoph Lameter205ab992008-04-14 19:11:40 +03003084 if (oo_objects(s->oo) > oo_objects(s->max))
3085 s->max = s->oo;
Christoph Lameter81819f02007-05-06 14:49:36 -07003086
Christoph Lameter834f3d12008-04-14 19:11:31 +03003087 return !!oo_objects(s->oo);
Christoph Lameter81819f02007-05-06 14:49:36 -07003088}
3089
Christoph Lameter8a13a4c2012-09-04 23:18:33 +00003090static int kmem_cache_open(struct kmem_cache *s, unsigned long flags)
Christoph Lameter81819f02007-05-06 14:49:36 -07003091{
Christoph Lameter8a13a4c2012-09-04 23:18:33 +00003092 s->flags = kmem_cache_flags(s->size, flags, s->name, s->ctor);
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08003093 s->reserved = 0;
Christoph Lameter81819f02007-05-06 14:49:36 -07003094
Lai Jiangshanda9a6382011-03-10 15:22:00 +08003095 if (need_reserve_slab_rcu && (s->flags & SLAB_DESTROY_BY_RCU))
3096 s->reserved = sizeof(struct rcu_head);
Christoph Lameter81819f02007-05-06 14:49:36 -07003097
Christoph Lameter06b285d2008-04-14 19:11:41 +03003098 if (!calculate_sizes(s, -1))
Christoph Lameter81819f02007-05-06 14:49:36 -07003099 goto error;
David Rientjes3de47212009-07-27 18:30:35 -07003100 if (disable_higher_order_debug) {
3101 /*
3102 * Disable debugging flags that store metadata if the min slab
3103 * order increased.
3104 */
Christoph Lameter3b0efdf2012-06-13 10:24:57 -05003105 if (get_order(s->size) > get_order(s->object_size)) {
David Rientjes3de47212009-07-27 18:30:35 -07003106 s->flags &= ~DEBUG_METADATA_FLAGS;
3107 s->offset = 0;
3108 if (!calculate_sizes(s, -1))
3109 goto error;
3110 }
3111 }
Christoph Lameter81819f02007-05-06 14:49:36 -07003112
Heiko Carstens25654092012-01-12 17:17:33 -08003113#if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \
3114 defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
Christoph Lameterb789ef52011-06-01 12:25:49 -05003115 if (system_has_cmpxchg_double() && (s->flags & SLAB_DEBUG_FLAGS) == 0)
3116 /* Enable fast mode */
3117 s->flags |= __CMPXCHG_DOUBLE;
3118#endif
3119
David Rientjes3b89d7d2009-02-22 17:40:07 -08003120 /*
3121 * The larger the object size is, the more pages we want on the partial
3122 * list to avoid pounding the page allocator excessively.
3123 */
Christoph Lameter49e22582011-08-09 16:12:27 -05003124 set_min_partial(s, ilog2(s->size) / 2);
3125
3126 /*
3127 * cpu_partial determined the maximum number of objects kept in the
3128 * per cpu partial lists of a processor.
3129 *
3130 * Per cpu partial lists mainly contain slabs that just have one
3131 * object freed. If they are used for allocation then they can be
3132 * filled up again with minimal effort. The slab will never hit the
3133 * per node partial lists and therefore no locking will be required.
3134 *
3135 * This setting also determines
3136 *
3137 * A) The number of objects from per cpu partial slabs dumped to the
3138 * per node list when we reach the limit.
Alex Shi9f264902011-09-01 11:32:18 +08003139 * B) The number of objects in cpu partial slabs to extract from the
Chen Gangd0e0ac92013-07-15 09:05:29 +08003140 * per node list when we run out of per cpu objects. We only fetch
3141 * 50% to keep some capacity around for frees.
Christoph Lameter49e22582011-08-09 16:12:27 -05003142 */
Joonsoo Kim345c9052013-06-19 14:05:52 +09003143 if (!kmem_cache_has_cpu_partial(s))
Christoph Lameter8f1e33d2011-11-23 09:24:27 -06003144 s->cpu_partial = 0;
3145 else if (s->size >= PAGE_SIZE)
Christoph Lameter49e22582011-08-09 16:12:27 -05003146 s->cpu_partial = 2;
3147 else if (s->size >= 1024)
3148 s->cpu_partial = 6;
3149 else if (s->size >= 256)
3150 s->cpu_partial = 13;
3151 else
3152 s->cpu_partial = 30;
3153
Christoph Lameter81819f02007-05-06 14:49:36 -07003154#ifdef CONFIG_NUMA
Christoph Lametere2cb96b2008-08-19 08:51:22 -05003155 s->remote_node_defrag_ratio = 1000;
Christoph Lameter81819f02007-05-06 14:49:36 -07003156#endif
Christoph Lameter55136592010-08-20 12:37:13 -05003157 if (!init_kmem_cache_nodes(s))
Christoph Lameterdfb4f092007-10-16 01:26:05 -07003158 goto error;
Christoph Lameter81819f02007-05-06 14:49:36 -07003159
Christoph Lameter55136592010-08-20 12:37:13 -05003160 if (alloc_kmem_cache_cpus(s))
Christoph Lameter278b1bb2012-09-05 00:20:34 +00003161 return 0;
Christoph Lameterff120592009-12-18 16:26:22 -06003162
Christoph Lameter4c93c3552007-10-16 01:26:08 -07003163 free_kmem_cache_nodes(s);
Christoph Lameter81819f02007-05-06 14:49:36 -07003164error:
3165 if (flags & SLAB_PANIC)
3166 panic("Cannot create slab %s size=%lu realsize=%u "
3167 "order=%u offset=%u flags=%lx\n",
Chen Gangd0e0ac92013-07-15 09:05:29 +08003168 s->name, (unsigned long)s->size, s->size,
3169 oo_order(s->oo), s->offset, flags);
Christoph Lameter278b1bb2012-09-05 00:20:34 +00003170 return -EINVAL;
Christoph Lameter81819f02007-05-06 14:49:36 -07003171}
Christoph Lameter81819f02007-05-06 14:49:36 -07003172
Christoph Lameter33b12c32008-04-25 12:22:43 -07003173static void list_slab_objects(struct kmem_cache *s, struct page *page,
3174 const char *text)
Christoph Lameter81819f02007-05-06 14:49:36 -07003175{
Christoph Lameter33b12c32008-04-25 12:22:43 -07003176#ifdef CONFIG_SLUB_DEBUG
3177 void *addr = page_address(page);
3178 void *p;
Namhyung Kima5dd5c12010-09-29 21:02:13 +09003179 unsigned long *map = kzalloc(BITS_TO_LONGS(page->objects) *
3180 sizeof(long), GFP_ATOMIC);
Eric Dumazetbbd7d572010-03-24 22:25:47 +01003181 if (!map)
3182 return;
Christoph Lameter945cf2b2012-09-04 23:18:33 +00003183 slab_err(s, page, text, s->name);
Christoph Lameter33b12c32008-04-25 12:22:43 -07003184 slab_lock(page);
Christoph Lameter33b12c32008-04-25 12:22:43 -07003185
Christoph Lameter5f80b132011-04-15 14:48:13 -05003186 get_map(s, page, map);
Christoph Lameter33b12c32008-04-25 12:22:43 -07003187 for_each_object(p, s, addr, page->objects) {
3188
3189 if (!test_bit(slab_index(p, s, addr), map)) {
Fabian Frederickf9f58282014-06-04 16:06:34 -07003190 pr_err("INFO: Object 0x%p @offset=%tu\n", p, p - addr);
Christoph Lameter33b12c32008-04-25 12:22:43 -07003191 print_tracking(s, p);
3192 }
3193 }
3194 slab_unlock(page);
Eric Dumazetbbd7d572010-03-24 22:25:47 +01003195 kfree(map);
Christoph Lameter33b12c32008-04-25 12:22:43 -07003196#endif
3197}
3198
Christoph Lameter81819f02007-05-06 14:49:36 -07003199/*
Christoph Lameter599870b2008-04-23 12:36:52 -07003200 * Attempt to free all partial slabs on a node.
Christoph Lameter69cb8e62011-08-09 16:12:22 -05003201 * This is called from kmem_cache_close(). We must be the last thread
3202 * using the cache and therefore we do not need to lock anymore.
Christoph Lameter81819f02007-05-06 14:49:36 -07003203 */
Christoph Lameter599870b2008-04-23 12:36:52 -07003204static void free_partial(struct kmem_cache *s, struct kmem_cache_node *n)
Christoph Lameter81819f02007-05-06 14:49:36 -07003205{
Christoph Lameter81819f02007-05-06 14:49:36 -07003206 struct page *page, *h;
3207
Christoph Lameter33b12c32008-04-25 12:22:43 -07003208 list_for_each_entry_safe(page, h, &n->partial, lru) {
Christoph Lameter81819f02007-05-06 14:49:36 -07003209 if (!page->inuse) {
Steven Rostedt1e4dd942014-02-10 14:25:46 -08003210 __remove_partial(n, page);
Christoph Lameter81819f02007-05-06 14:49:36 -07003211 discard_slab(s, page);
Christoph Lameter33b12c32008-04-25 12:22:43 -07003212 } else {
3213 list_slab_objects(s, page,
Christoph Lameter945cf2b2012-09-04 23:18:33 +00003214 "Objects remaining in %s on kmem_cache_close()");
Christoph Lameter599870b2008-04-23 12:36:52 -07003215 }
Christoph Lameter33b12c32008-04-25 12:22:43 -07003216 }
Christoph Lameter81819f02007-05-06 14:49:36 -07003217}
3218
3219/*
Christoph Lameter672bba32007-05-09 02:32:39 -07003220 * Release all resources used by a slab cache.
Christoph Lameter81819f02007-05-06 14:49:36 -07003221 */
Christoph Lameter0c710012007-07-17 04:03:24 -07003222static inline int kmem_cache_close(struct kmem_cache *s)
Christoph Lameter81819f02007-05-06 14:49:36 -07003223{
3224 int node;
3225
3226 flush_all(s);
Christoph Lameter81819f02007-05-06 14:49:36 -07003227 /* Attempt to free all objects */
Christoph Lameterf64dc582007-10-16 01:25:33 -07003228 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter81819f02007-05-06 14:49:36 -07003229 struct kmem_cache_node *n = get_node(s, node);
3230
Christoph Lameter599870b2008-04-23 12:36:52 -07003231 free_partial(s, n);
3232 if (n->nr_partial || slabs_node(s, node))
Christoph Lameter81819f02007-05-06 14:49:36 -07003233 return 1;
3234 }
Christoph Lameter945cf2b2012-09-04 23:18:33 +00003235 free_percpu(s->cpu_slab);
Christoph Lameter81819f02007-05-06 14:49:36 -07003236 free_kmem_cache_nodes(s);
3237 return 0;
3238}
3239
Christoph Lameter945cf2b2012-09-04 23:18:33 +00003240int __kmem_cache_shutdown(struct kmem_cache *s)
Christoph Lameter81819f02007-05-06 14:49:36 -07003241{
Christoph Lameter41a21282014-05-06 12:50:08 -07003242 return kmem_cache_close(s);
Christoph Lameter81819f02007-05-06 14:49:36 -07003243}
Christoph Lameter81819f02007-05-06 14:49:36 -07003244
3245/********************************************************************
3246 * Kmalloc subsystem
3247 *******************************************************************/
3248
Christoph Lameter81819f02007-05-06 14:49:36 -07003249static int __init setup_slub_min_order(char *str)
3250{
Pekka Enberg06428782008-01-07 23:20:27 -08003251 get_option(&str, &slub_min_order);
Christoph Lameter81819f02007-05-06 14:49:36 -07003252
3253 return 1;
3254}
3255
3256__setup("slub_min_order=", setup_slub_min_order);
3257
3258static int __init setup_slub_max_order(char *str)
3259{
Pekka Enberg06428782008-01-07 23:20:27 -08003260 get_option(&str, &slub_max_order);
David Rientjes818cf592009-04-23 09:58:22 +03003261 slub_max_order = min(slub_max_order, MAX_ORDER - 1);
Christoph Lameter81819f02007-05-06 14:49:36 -07003262
3263 return 1;
3264}
3265
3266__setup("slub_max_order=", setup_slub_max_order);
3267
3268static int __init setup_slub_min_objects(char *str)
3269{
Pekka Enberg06428782008-01-07 23:20:27 -08003270 get_option(&str, &slub_min_objects);
Christoph Lameter81819f02007-05-06 14:49:36 -07003271
3272 return 1;
3273}
3274
3275__setup("slub_min_objects=", setup_slub_min_objects);
3276
3277static int __init setup_slub_nomerge(char *str)
3278{
3279 slub_nomerge = 1;
3280 return 1;
3281}
3282
3283__setup("slub_nomerge", setup_slub_nomerge);
3284
Christoph Lameter81819f02007-05-06 14:49:36 -07003285void *__kmalloc(size_t size, gfp_t flags)
3286{
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003287 struct kmem_cache *s;
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03003288 void *ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07003289
Christoph Lameter95a05b42013-01-10 19:14:19 +00003290 if (unlikely(size > KMALLOC_MAX_CACHE_SIZE))
Pekka Enbergeada35e2008-02-11 22:47:46 +02003291 return kmalloc_large(size, flags);
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003292
Christoph Lameter2c59dd62013-01-10 19:14:19 +00003293 s = kmalloc_slab(size, flags);
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003294
3295 if (unlikely(ZERO_OR_NULL_PTR(s)))
Christoph Lameter6cb8f912007-07-17 04:03:22 -07003296 return s;
3297
Ezequiel Garcia2b847c32012-09-08 17:47:58 -03003298 ret = slab_alloc(s, flags, _RET_IP_);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03003299
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02003300 trace_kmalloc(_RET_IP_, ret, size, s->size, flags);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03003301
3302 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07003303}
3304EXPORT_SYMBOL(__kmalloc);
3305
Namhyung Kim5d1f57e2010-09-29 21:02:15 +09003306#ifdef CONFIG_NUMA
Christoph Lameterf619cfe2008-03-01 13:56:40 -08003307static void *kmalloc_large_node(size_t size, gfp_t flags, int node)
3308{
Vegard Nossumb1eeab62008-11-25 16:55:53 +01003309 struct page *page;
Catalin Marinase4f7c0b42009-07-07 10:32:59 +01003310 void *ptr = NULL;
Christoph Lameterf619cfe2008-03-01 13:56:40 -08003311
Vladimir Davydov52383432014-06-04 16:06:39 -07003312 flags |= __GFP_COMP | __GFP_NOTRACK;
3313 page = alloc_kmem_pages_node(node, flags, get_order(size));
Christoph Lameterf619cfe2008-03-01 13:56:40 -08003314 if (page)
Catalin Marinase4f7c0b42009-07-07 10:32:59 +01003315 ptr = page_address(page);
3316
Roman Bobnievd56791b2013-10-08 15:58:57 -07003317 kmalloc_large_node_hook(ptr, size, flags);
Catalin Marinase4f7c0b42009-07-07 10:32:59 +01003318 return ptr;
Christoph Lameterf619cfe2008-03-01 13:56:40 -08003319}
3320
Christoph Lameter81819f02007-05-06 14:49:36 -07003321void *__kmalloc_node(size_t size, gfp_t flags, int node)
3322{
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003323 struct kmem_cache *s;
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03003324 void *ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07003325
Christoph Lameter95a05b42013-01-10 19:14:19 +00003326 if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) {
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03003327 ret = kmalloc_large_node(size, flags, node);
3328
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02003329 trace_kmalloc_node(_RET_IP_, ret,
3330 size, PAGE_SIZE << get_order(size),
3331 flags, node);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03003332
3333 return ret;
3334 }
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003335
Christoph Lameter2c59dd62013-01-10 19:14:19 +00003336 s = kmalloc_slab(size, flags);
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003337
3338 if (unlikely(ZERO_OR_NULL_PTR(s)))
Christoph Lameter6cb8f912007-07-17 04:03:22 -07003339 return s;
3340
Ezequiel Garcia2b847c32012-09-08 17:47:58 -03003341 ret = slab_alloc_node(s, flags, node, _RET_IP_);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03003342
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02003343 trace_kmalloc_node(_RET_IP_, ret, size, s->size, flags, node);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03003344
3345 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07003346}
3347EXPORT_SYMBOL(__kmalloc_node);
3348#endif
3349
3350size_t ksize(const void *object)
3351{
Christoph Lameter272c1d22007-06-08 13:46:49 -07003352 struct page *page;
Christoph Lameter81819f02007-05-06 14:49:36 -07003353
Christoph Lameteref8b4522007-10-16 01:24:46 -07003354 if (unlikely(object == ZERO_SIZE_PTR))
Christoph Lameter272c1d22007-06-08 13:46:49 -07003355 return 0;
3356
Vegard Nossum294a80a2007-12-04 23:45:30 -08003357 page = virt_to_head_page(object);
Vegard Nossum294a80a2007-12-04 23:45:30 -08003358
Pekka Enberg76994412008-05-22 19:22:25 +03003359 if (unlikely(!PageSlab(page))) {
3360 WARN_ON(!PageCompound(page));
Vegard Nossum294a80a2007-12-04 23:45:30 -08003361 return PAGE_SIZE << compound_order(page);
Pekka Enberg76994412008-05-22 19:22:25 +03003362 }
Christoph Lameter81819f02007-05-06 14:49:36 -07003363
Glauber Costa1b4f59e32012-10-22 18:05:36 +04003364 return slab_ksize(page->slab_cache);
Christoph Lameter81819f02007-05-06 14:49:36 -07003365}
Kirill A. Shutemovb1aabec2009-02-10 15:21:44 +02003366EXPORT_SYMBOL(ksize);
Christoph Lameter81819f02007-05-06 14:49:36 -07003367
3368void kfree(const void *x)
3369{
Christoph Lameter81819f02007-05-06 14:49:36 -07003370 struct page *page;
Christoph Lameter5bb983b2008-02-07 17:47:41 -08003371 void *object = (void *)x;
Christoph Lameter81819f02007-05-06 14:49:36 -07003372
Pekka Enberg2121db72009-03-25 11:05:57 +02003373 trace_kfree(_RET_IP_, x);
3374
Satyam Sharma2408c552007-10-16 01:24:44 -07003375 if (unlikely(ZERO_OR_NULL_PTR(x)))
Christoph Lameter81819f02007-05-06 14:49:36 -07003376 return;
3377
Christoph Lameterb49af682007-05-06 14:49:41 -07003378 page = virt_to_head_page(x);
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003379 if (unlikely(!PageSlab(page))) {
Christoph Lameter09375022008-05-28 10:32:22 -07003380 BUG_ON(!PageCompound(page));
Roman Bobnievd56791b2013-10-08 15:58:57 -07003381 kfree_hook(x);
Vladimir Davydov52383432014-06-04 16:06:39 -07003382 __free_kmem_pages(page, compound_order(page));
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003383 return;
3384 }
Glauber Costa1b4f59e32012-10-22 18:05:36 +04003385 slab_free(page->slab_cache, page, object, _RET_IP_);
Christoph Lameter81819f02007-05-06 14:49:36 -07003386}
3387EXPORT_SYMBOL(kfree);
3388
Christoph Lameter2086d262007-05-06 14:49:46 -07003389/*
Christoph Lameter672bba32007-05-09 02:32:39 -07003390 * kmem_cache_shrink removes empty slabs from the partial lists and sorts
3391 * the remaining slabs by the number of items in use. The slabs with the
3392 * most items in use come first. New allocations will then fill those up
3393 * and thus they can be removed from the partial lists.
3394 *
3395 * The slabs with the least items are placed last. This results in them
3396 * being allocated from last increasing the chance that the last objects
3397 * are freed in them.
Christoph Lameter2086d262007-05-06 14:49:46 -07003398 */
Vladimir Davydov03afc0e2014-06-04 16:07:20 -07003399int __kmem_cache_shrink(struct kmem_cache *s)
Christoph Lameter2086d262007-05-06 14:49:46 -07003400{
3401 int node;
3402 int i;
3403 struct kmem_cache_node *n;
3404 struct page *page;
3405 struct page *t;
Christoph Lameter205ab992008-04-14 19:11:40 +03003406 int objects = oo_objects(s->max);
Christoph Lameter2086d262007-05-06 14:49:46 -07003407 struct list_head *slabs_by_inuse =
Christoph Lameter834f3d12008-04-14 19:11:31 +03003408 kmalloc(sizeof(struct list_head) * objects, GFP_KERNEL);
Christoph Lameter2086d262007-05-06 14:49:46 -07003409 unsigned long flags;
3410
3411 if (!slabs_by_inuse)
3412 return -ENOMEM;
3413
3414 flush_all(s);
Christoph Lameterf64dc582007-10-16 01:25:33 -07003415 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter2086d262007-05-06 14:49:46 -07003416 n = get_node(s, node);
3417
3418 if (!n->nr_partial)
3419 continue;
3420
Christoph Lameter834f3d12008-04-14 19:11:31 +03003421 for (i = 0; i < objects; i++)
Christoph Lameter2086d262007-05-06 14:49:46 -07003422 INIT_LIST_HEAD(slabs_by_inuse + i);
3423
3424 spin_lock_irqsave(&n->list_lock, flags);
3425
3426 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07003427 * Build lists indexed by the items in use in each slab.
Christoph Lameter2086d262007-05-06 14:49:46 -07003428 *
Christoph Lameter672bba32007-05-09 02:32:39 -07003429 * Note that concurrent frees may occur while we hold the
3430 * list_lock. page->inuse here is the upper limit.
Christoph Lameter2086d262007-05-06 14:49:46 -07003431 */
3432 list_for_each_entry_safe(page, t, &n->partial, lru) {
Christoph Lameter69cb8e62011-08-09 16:12:22 -05003433 list_move(&page->lru, slabs_by_inuse + page->inuse);
3434 if (!page->inuse)
3435 n->nr_partial--;
Christoph Lameter2086d262007-05-06 14:49:46 -07003436 }
3437
Christoph Lameter2086d262007-05-06 14:49:46 -07003438 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07003439 * Rebuild the partial list with the slabs filled up most
3440 * first and the least used slabs at the end.
Christoph Lameter2086d262007-05-06 14:49:46 -07003441 */
Christoph Lameter69cb8e62011-08-09 16:12:22 -05003442 for (i = objects - 1; i > 0; i--)
Christoph Lameter2086d262007-05-06 14:49:46 -07003443 list_splice(slabs_by_inuse + i, n->partial.prev);
3444
Christoph Lameter2086d262007-05-06 14:49:46 -07003445 spin_unlock_irqrestore(&n->list_lock, flags);
Christoph Lameter69cb8e62011-08-09 16:12:22 -05003446
3447 /* Release empty slabs */
3448 list_for_each_entry_safe(page, t, slabs_by_inuse, lru)
3449 discard_slab(s, page);
Christoph Lameter2086d262007-05-06 14:49:46 -07003450 }
3451
3452 kfree(slabs_by_inuse);
3453 return 0;
3454}
Christoph Lameter2086d262007-05-06 14:49:46 -07003455
Yasunori Gotob9049e22007-10-21 16:41:37 -07003456static int slab_mem_going_offline_callback(void *arg)
3457{
3458 struct kmem_cache *s;
3459
Christoph Lameter18004c52012-07-06 15:25:12 -05003460 mutex_lock(&slab_mutex);
Yasunori Gotob9049e22007-10-21 16:41:37 -07003461 list_for_each_entry(s, &slab_caches, list)
Vladimir Davydov03afc0e2014-06-04 16:07:20 -07003462 __kmem_cache_shrink(s);
Christoph Lameter18004c52012-07-06 15:25:12 -05003463 mutex_unlock(&slab_mutex);
Yasunori Gotob9049e22007-10-21 16:41:37 -07003464
3465 return 0;
3466}
3467
3468static void slab_mem_offline_callback(void *arg)
3469{
3470 struct kmem_cache_node *n;
3471 struct kmem_cache *s;
3472 struct memory_notify *marg = arg;
3473 int offline_node;
3474
Lai Jiangshanb9d5ab22012-12-11 16:01:05 -08003475 offline_node = marg->status_change_nid_normal;
Yasunori Gotob9049e22007-10-21 16:41:37 -07003476
3477 /*
3478 * If the node still has available memory. we need kmem_cache_node
3479 * for it yet.
3480 */
3481 if (offline_node < 0)
3482 return;
3483
Christoph Lameter18004c52012-07-06 15:25:12 -05003484 mutex_lock(&slab_mutex);
Yasunori Gotob9049e22007-10-21 16:41:37 -07003485 list_for_each_entry(s, &slab_caches, list) {
3486 n = get_node(s, offline_node);
3487 if (n) {
3488 /*
3489 * if n->nr_slabs > 0, slabs still exist on the node
3490 * that is going down. We were unable to free them,
Adam Buchbinderc9404c92009-12-18 15:40:42 -05003491 * and offline_pages() function shouldn't call this
Yasunori Gotob9049e22007-10-21 16:41:37 -07003492 * callback. So, we must fail.
3493 */
Christoph Lameter0f389ec2008-04-14 18:53:02 +03003494 BUG_ON(slabs_node(s, offline_node));
Yasunori Gotob9049e22007-10-21 16:41:37 -07003495
3496 s->node[offline_node] = NULL;
Christoph Lameter8de66a02010-08-25 14:51:14 -05003497 kmem_cache_free(kmem_cache_node, n);
Yasunori Gotob9049e22007-10-21 16:41:37 -07003498 }
3499 }
Christoph Lameter18004c52012-07-06 15:25:12 -05003500 mutex_unlock(&slab_mutex);
Yasunori Gotob9049e22007-10-21 16:41:37 -07003501}
3502
3503static int slab_mem_going_online_callback(void *arg)
3504{
3505 struct kmem_cache_node *n;
3506 struct kmem_cache *s;
3507 struct memory_notify *marg = arg;
Lai Jiangshanb9d5ab22012-12-11 16:01:05 -08003508 int nid = marg->status_change_nid_normal;
Yasunori Gotob9049e22007-10-21 16:41:37 -07003509 int ret = 0;
3510
3511 /*
3512 * If the node's memory is already available, then kmem_cache_node is
3513 * already created. Nothing to do.
3514 */
3515 if (nid < 0)
3516 return 0;
3517
3518 /*
Christoph Lameter0121c6192008-04-29 16:11:12 -07003519 * We are bringing a node online. No memory is available yet. We must
Yasunori Gotob9049e22007-10-21 16:41:37 -07003520 * allocate a kmem_cache_node structure in order to bring the node
3521 * online.
3522 */
Christoph Lameter18004c52012-07-06 15:25:12 -05003523 mutex_lock(&slab_mutex);
Yasunori Gotob9049e22007-10-21 16:41:37 -07003524 list_for_each_entry(s, &slab_caches, list) {
3525 /*
3526 * XXX: kmem_cache_alloc_node will fallback to other nodes
3527 * since memory is not yet available from the node that
3528 * is brought up.
3529 */
Christoph Lameter8de66a02010-08-25 14:51:14 -05003530 n = kmem_cache_alloc(kmem_cache_node, GFP_KERNEL);
Yasunori Gotob9049e22007-10-21 16:41:37 -07003531 if (!n) {
3532 ret = -ENOMEM;
3533 goto out;
3534 }
Joonsoo Kim40534972012-05-11 00:50:47 +09003535 init_kmem_cache_node(n);
Yasunori Gotob9049e22007-10-21 16:41:37 -07003536 s->node[nid] = n;
3537 }
3538out:
Christoph Lameter18004c52012-07-06 15:25:12 -05003539 mutex_unlock(&slab_mutex);
Yasunori Gotob9049e22007-10-21 16:41:37 -07003540 return ret;
3541}
3542
3543static int slab_memory_callback(struct notifier_block *self,
3544 unsigned long action, void *arg)
3545{
3546 int ret = 0;
3547
3548 switch (action) {
3549 case MEM_GOING_ONLINE:
3550 ret = slab_mem_going_online_callback(arg);
3551 break;
3552 case MEM_GOING_OFFLINE:
3553 ret = slab_mem_going_offline_callback(arg);
3554 break;
3555 case MEM_OFFLINE:
3556 case MEM_CANCEL_ONLINE:
3557 slab_mem_offline_callback(arg);
3558 break;
3559 case MEM_ONLINE:
3560 case MEM_CANCEL_OFFLINE:
3561 break;
3562 }
KAMEZAWA Hiroyukidc19f9d2008-12-01 13:13:48 -08003563 if (ret)
3564 ret = notifier_from_errno(ret);
3565 else
3566 ret = NOTIFY_OK;
Yasunori Gotob9049e22007-10-21 16:41:37 -07003567 return ret;
3568}
3569
Andrew Morton3ac38fa2013-04-29 15:08:06 -07003570static struct notifier_block slab_memory_callback_nb = {
3571 .notifier_call = slab_memory_callback,
3572 .priority = SLAB_CALLBACK_PRI,
3573};
Yasunori Gotob9049e22007-10-21 16:41:37 -07003574
Christoph Lameter81819f02007-05-06 14:49:36 -07003575/********************************************************************
3576 * Basic setup of slabs
3577 *******************************************************************/
3578
Christoph Lameter51df1142010-08-20 12:37:15 -05003579/*
3580 * Used for early kmem_cache structures that were allocated using
Christoph Lameterdffb4d62012-11-28 16:23:07 +00003581 * the page allocator. Allocate them properly then fix up the pointers
3582 * that may be pointing to the wrong kmem_cache structure.
Christoph Lameter51df1142010-08-20 12:37:15 -05003583 */
3584
Christoph Lameterdffb4d62012-11-28 16:23:07 +00003585static struct kmem_cache * __init bootstrap(struct kmem_cache *static_cache)
Christoph Lameter51df1142010-08-20 12:37:15 -05003586{
3587 int node;
Christoph Lameterdffb4d62012-11-28 16:23:07 +00003588 struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT);
Christoph Lameter51df1142010-08-20 12:37:15 -05003589
Christoph Lameterdffb4d62012-11-28 16:23:07 +00003590 memcpy(s, static_cache, kmem_cache->object_size);
Christoph Lameter51df1142010-08-20 12:37:15 -05003591
Glauber Costa7d557b32013-02-22 20:20:00 +04003592 /*
3593 * This runs very early, and only the boot processor is supposed to be
3594 * up. Even if it weren't true, IRQs are not up so we couldn't fire
3595 * IPIs around.
3596 */
3597 __flush_cpu_slab(s, smp_processor_id());
Christoph Lameter51df1142010-08-20 12:37:15 -05003598 for_each_node_state(node, N_NORMAL_MEMORY) {
3599 struct kmem_cache_node *n = get_node(s, node);
3600 struct page *p;
3601
3602 if (n) {
3603 list_for_each_entry(p, &n->partial, lru)
Glauber Costa1b4f59e32012-10-22 18:05:36 +04003604 p->slab_cache = s;
Christoph Lameter51df1142010-08-20 12:37:15 -05003605
Li Zefan607bf322011-04-12 15:22:26 +08003606#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter51df1142010-08-20 12:37:15 -05003607 list_for_each_entry(p, &n->full, lru)
Glauber Costa1b4f59e32012-10-22 18:05:36 +04003608 p->slab_cache = s;
Christoph Lameter51df1142010-08-20 12:37:15 -05003609#endif
3610 }
3611 }
Christoph Lameterdffb4d62012-11-28 16:23:07 +00003612 list_add(&s->list, &slab_caches);
3613 return s;
Christoph Lameter51df1142010-08-20 12:37:15 -05003614}
3615
Christoph Lameter81819f02007-05-06 14:49:36 -07003616void __init kmem_cache_init(void)
3617{
Christoph Lameterdffb4d62012-11-28 16:23:07 +00003618 static __initdata struct kmem_cache boot_kmem_cache,
3619 boot_kmem_cache_node;
Christoph Lameter51df1142010-08-20 12:37:15 -05003620
Stanislaw Gruszkafc8d8622012-01-10 15:07:32 -08003621 if (debug_guardpage_minorder())
3622 slub_max_order = 0;
3623
Christoph Lameterdffb4d62012-11-28 16:23:07 +00003624 kmem_cache_node = &boot_kmem_cache_node;
3625 kmem_cache = &boot_kmem_cache;
Christoph Lameter51df1142010-08-20 12:37:15 -05003626
Christoph Lameterdffb4d62012-11-28 16:23:07 +00003627 create_boot_cache(kmem_cache_node, "kmem_cache_node",
3628 sizeof(struct kmem_cache_node), SLAB_HWCACHE_ALIGN);
Yasunori Gotob9049e22007-10-21 16:41:37 -07003629
Andrew Morton3ac38fa2013-04-29 15:08:06 -07003630 register_hotmemory_notifier(&slab_memory_callback_nb);
Christoph Lameter81819f02007-05-06 14:49:36 -07003631
3632 /* Able to allocate the per node structures */
3633 slab_state = PARTIAL;
3634
Christoph Lameterdffb4d62012-11-28 16:23:07 +00003635 create_boot_cache(kmem_cache, "kmem_cache",
3636 offsetof(struct kmem_cache, node) +
3637 nr_node_ids * sizeof(struct kmem_cache_node *),
3638 SLAB_HWCACHE_ALIGN);
Christoph Lameter8a13a4c2012-09-04 23:18:33 +00003639
Christoph Lameterdffb4d62012-11-28 16:23:07 +00003640 kmem_cache = bootstrap(&boot_kmem_cache);
Christoph Lameter81819f02007-05-06 14:49:36 -07003641
Christoph Lameter51df1142010-08-20 12:37:15 -05003642 /*
3643 * Allocate kmem_cache_node properly from the kmem_cache slab.
3644 * kmem_cache_node is separately allocated so no need to
3645 * update any list pointers.
3646 */
Christoph Lameterdffb4d62012-11-28 16:23:07 +00003647 kmem_cache_node = bootstrap(&boot_kmem_cache_node);
Christoph Lameter51df1142010-08-20 12:37:15 -05003648
3649 /* Now we can use the kmem_cache to allocate kmalloc slabs */
Christoph Lameterf97d5f62013-01-10 19:12:17 +00003650 create_kmalloc_caches(0);
Christoph Lameter81819f02007-05-06 14:49:36 -07003651
3652#ifdef CONFIG_SMP
3653 register_cpu_notifier(&slab_notifier);
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06003654#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07003655
Fabian Frederickf9f58282014-06-04 16:06:34 -07003656 pr_info("SLUB: HWalign=%d, Order=%d-%d, MinObjects=%d, CPUs=%d, Nodes=%d\n",
Christoph Lameterf97d5f62013-01-10 19:12:17 +00003657 cache_line_size(),
Christoph Lameter81819f02007-05-06 14:49:36 -07003658 slub_min_order, slub_max_order, slub_min_objects,
3659 nr_cpu_ids, nr_node_ids);
3660}
3661
Pekka Enberg7e85ee02009-06-12 14:03:06 +03003662void __init kmem_cache_init_late(void)
3663{
Pekka Enberg7e85ee02009-06-12 14:03:06 +03003664}
3665
Christoph Lameter81819f02007-05-06 14:49:36 -07003666/*
3667 * Find a mergeable slab cache
3668 */
3669static int slab_unmergeable(struct kmem_cache *s)
3670{
3671 if (slub_nomerge || (s->flags & SLUB_NEVER_MERGE))
3672 return 1;
3673
Vladimir Davydova44cb942014-04-07 15:39:23 -07003674 if (!is_root_cache(s))
3675 return 1;
3676
Christoph Lameterc59def92007-05-16 22:10:50 -07003677 if (s->ctor)
Christoph Lameter81819f02007-05-06 14:49:36 -07003678 return 1;
3679
Christoph Lameter8ffa6872007-05-31 00:40:51 -07003680 /*
3681 * We may have set a slab to be unmergeable during bootstrap.
3682 */
3683 if (s->refcount < 0)
3684 return 1;
3685
Christoph Lameter81819f02007-05-06 14:49:36 -07003686 return 0;
3687}
3688
Vladimir Davydova44cb942014-04-07 15:39:23 -07003689static struct kmem_cache *find_mergeable(size_t size, size_t align,
3690 unsigned long flags, const char *name, void (*ctor)(void *))
Christoph Lameter81819f02007-05-06 14:49:36 -07003691{
Christoph Lameter5b95a4ac2007-07-17 04:03:19 -07003692 struct kmem_cache *s;
Christoph Lameter81819f02007-05-06 14:49:36 -07003693
3694 if (slub_nomerge || (flags & SLUB_NEVER_MERGE))
3695 return NULL;
3696
Christoph Lameterc59def92007-05-16 22:10:50 -07003697 if (ctor)
Christoph Lameter81819f02007-05-06 14:49:36 -07003698 return NULL;
3699
3700 size = ALIGN(size, sizeof(void *));
3701 align = calculate_alignment(flags, align, size);
3702 size = ALIGN(size, align);
Christoph Lameterba0268a2007-09-11 15:24:11 -07003703 flags = kmem_cache_flags(size, flags, name, NULL);
Christoph Lameter81819f02007-05-06 14:49:36 -07003704
Christoph Lameter5b95a4ac2007-07-17 04:03:19 -07003705 list_for_each_entry(s, &slab_caches, list) {
Christoph Lameter81819f02007-05-06 14:49:36 -07003706 if (slab_unmergeable(s))
3707 continue;
3708
3709 if (size > s->size)
3710 continue;
3711
Christoph Lameterba0268a2007-09-11 15:24:11 -07003712 if ((flags & SLUB_MERGE_SAME) != (s->flags & SLUB_MERGE_SAME))
Vladimir Davydova44cb942014-04-07 15:39:23 -07003713 continue;
Christoph Lameter81819f02007-05-06 14:49:36 -07003714 /*
3715 * Check if alignment is compatible.
3716 * Courtesy of Adrian Drzewiecki
3717 */
Pekka Enberg06428782008-01-07 23:20:27 -08003718 if ((s->size & ~(align - 1)) != s->size)
Christoph Lameter81819f02007-05-06 14:49:36 -07003719 continue;
3720
3721 if (s->size - size >= sizeof(void *))
3722 continue;
3723
3724 return s;
3725 }
3726 return NULL;
3727}
3728
Glauber Costa2633d7a2012-12-18 14:22:34 -08003729struct kmem_cache *
Vladimir Davydova44cb942014-04-07 15:39:23 -07003730__kmem_cache_alias(const char *name, size_t size, size_t align,
3731 unsigned long flags, void (*ctor)(void *))
Christoph Lameter81819f02007-05-06 14:49:36 -07003732{
3733 struct kmem_cache *s;
3734
Vladimir Davydova44cb942014-04-07 15:39:23 -07003735 s = find_mergeable(size, align, flags, name, ctor);
Christoph Lameter81819f02007-05-06 14:49:36 -07003736 if (s) {
Vladimir Davydov84d0ddd2014-04-07 15:39:29 -07003737 int i;
3738 struct kmem_cache *c;
3739
Christoph Lameter81819f02007-05-06 14:49:36 -07003740 s->refcount++;
Vladimir Davydov84d0ddd2014-04-07 15:39:29 -07003741
Christoph Lameter81819f02007-05-06 14:49:36 -07003742 /*
3743 * Adjust the object sizes so that we clear
3744 * the complete object on kzalloc.
3745 */
Christoph Lameter3b0efdf2012-06-13 10:24:57 -05003746 s->object_size = max(s->object_size, (int)size);
Christoph Lameter81819f02007-05-06 14:49:36 -07003747 s->inuse = max_t(int, s->inuse, ALIGN(size, sizeof(void *)));
Christoph Lameter6446faa2008-02-15 23:45:26 -08003748
Vladimir Davydov84d0ddd2014-04-07 15:39:29 -07003749 for_each_memcg_cache_index(i) {
3750 c = cache_from_memcg_idx(s, i);
3751 if (!c)
3752 continue;
3753 c->object_size = s->object_size;
3754 c->inuse = max_t(int, c->inuse,
3755 ALIGN(size, sizeof(void *)));
3756 }
3757
David Rientjes7b8f3b62008-12-17 22:09:46 -08003758 if (sysfs_slab_alias(s, name)) {
David Rientjes7b8f3b62008-12-17 22:09:46 -08003759 s->refcount--;
Christoph Lametercbb79692012-09-05 00:18:32 +00003760 s = NULL;
David Rientjes7b8f3b62008-12-17 22:09:46 -08003761 }
Christoph Lametera0e1d1b2007-07-17 04:03:31 -07003762 }
Christoph Lameter6446faa2008-02-15 23:45:26 -08003763
Christoph Lametercbb79692012-09-05 00:18:32 +00003764 return s;
3765}
Pekka Enberg84c1cf62010-09-14 23:21:12 +03003766
Christoph Lameter8a13a4c2012-09-04 23:18:33 +00003767int __kmem_cache_create(struct kmem_cache *s, unsigned long flags)
Christoph Lametercbb79692012-09-05 00:18:32 +00003768{
Pekka Enbergaac3a162012-09-05 12:07:44 +03003769 int err;
Christoph Lameter20cea962012-07-06 15:25:13 -05003770
Pekka Enbergaac3a162012-09-05 12:07:44 +03003771 err = kmem_cache_open(s, flags);
3772 if (err)
3773 return err;
Christoph Lameter20cea962012-07-06 15:25:13 -05003774
Christoph Lameter45530c42012-11-28 16:23:07 +00003775 /* Mutex is not taken during early boot */
3776 if (slab_state <= UP)
3777 return 0;
3778
Glauber Costa107dab52012-12-18 14:23:05 -08003779 memcg_propagate_slab_attrs(s);
Pekka Enbergaac3a162012-09-05 12:07:44 +03003780 err = sysfs_slab_add(s);
Pekka Enbergaac3a162012-09-05 12:07:44 +03003781 if (err)
3782 kmem_cache_close(s);
3783
3784 return err;
Christoph Lameter81819f02007-05-06 14:49:36 -07003785}
Christoph Lameter81819f02007-05-06 14:49:36 -07003786
Christoph Lameter81819f02007-05-06 14:49:36 -07003787#ifdef CONFIG_SMP
Christoph Lameter27390bc2007-06-01 00:47:09 -07003788/*
Christoph Lameter672bba32007-05-09 02:32:39 -07003789 * Use the cpu notifier to insure that the cpu slabs are flushed when
3790 * necessary.
Christoph Lameter81819f02007-05-06 14:49:36 -07003791 */
Paul Gortmaker0db06282013-06-19 14:53:51 -04003792static int slab_cpuup_callback(struct notifier_block *nfb,
Christoph Lameter81819f02007-05-06 14:49:36 -07003793 unsigned long action, void *hcpu)
3794{
3795 long cpu = (long)hcpu;
Christoph Lameter5b95a4ac2007-07-17 04:03:19 -07003796 struct kmem_cache *s;
3797 unsigned long flags;
Christoph Lameter81819f02007-05-06 14:49:36 -07003798
3799 switch (action) {
3800 case CPU_UP_CANCELED:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07003801 case CPU_UP_CANCELED_FROZEN:
Christoph Lameter81819f02007-05-06 14:49:36 -07003802 case CPU_DEAD:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07003803 case CPU_DEAD_FROZEN:
Christoph Lameter18004c52012-07-06 15:25:12 -05003804 mutex_lock(&slab_mutex);
Christoph Lameter5b95a4ac2007-07-17 04:03:19 -07003805 list_for_each_entry(s, &slab_caches, list) {
3806 local_irq_save(flags);
3807 __flush_cpu_slab(s, cpu);
3808 local_irq_restore(flags);
3809 }
Christoph Lameter18004c52012-07-06 15:25:12 -05003810 mutex_unlock(&slab_mutex);
Christoph Lameter81819f02007-05-06 14:49:36 -07003811 break;
3812 default:
3813 break;
3814 }
3815 return NOTIFY_OK;
3816}
3817
Paul Gortmaker0db06282013-06-19 14:53:51 -04003818static struct notifier_block slab_notifier = {
Ingo Molnar3adbefe2008-02-05 17:57:39 -08003819 .notifier_call = slab_cpuup_callback
Pekka Enberg06428782008-01-07 23:20:27 -08003820};
Christoph Lameter81819f02007-05-06 14:49:36 -07003821
3822#endif
3823
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03003824void *__kmalloc_track_caller(size_t size, gfp_t gfpflags, unsigned long caller)
Christoph Lameter81819f02007-05-06 14:49:36 -07003825{
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003826 struct kmem_cache *s;
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03003827 void *ret;
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003828
Christoph Lameter95a05b42013-01-10 19:14:19 +00003829 if (unlikely(size > KMALLOC_MAX_CACHE_SIZE))
Pekka Enbergeada35e2008-02-11 22:47:46 +02003830 return kmalloc_large(size, gfpflags);
3831
Christoph Lameter2c59dd62013-01-10 19:14:19 +00003832 s = kmalloc_slab(size, gfpflags);
Christoph Lameter81819f02007-05-06 14:49:36 -07003833
Satyam Sharma2408c552007-10-16 01:24:44 -07003834 if (unlikely(ZERO_OR_NULL_PTR(s)))
Christoph Lameter6cb8f912007-07-17 04:03:22 -07003835 return s;
Christoph Lameter81819f02007-05-06 14:49:36 -07003836
Ezequiel Garcia2b847c32012-09-08 17:47:58 -03003837 ret = slab_alloc(s, gfpflags, caller);
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03003838
Lucas De Marchi25985ed2011-03-30 22:57:33 -03003839 /* Honor the call site pointer we received. */
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02003840 trace_kmalloc(caller, ret, size, s->size, gfpflags);
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03003841
3842 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07003843}
3844
Namhyung Kim5d1f57e2010-09-29 21:02:15 +09003845#ifdef CONFIG_NUMA
Christoph Lameter81819f02007-05-06 14:49:36 -07003846void *__kmalloc_node_track_caller(size_t size, gfp_t gfpflags,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03003847 int node, unsigned long caller)
Christoph Lameter81819f02007-05-06 14:49:36 -07003848{
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003849 struct kmem_cache *s;
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03003850 void *ret;
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003851
Christoph Lameter95a05b42013-01-10 19:14:19 +00003852 if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) {
Xiaotian Fengd3e14aa2010-04-08 17:26:44 +08003853 ret = kmalloc_large_node(size, gfpflags, node);
3854
3855 trace_kmalloc_node(caller, ret,
3856 size, PAGE_SIZE << get_order(size),
3857 gfpflags, node);
3858
3859 return ret;
3860 }
Pekka Enbergeada35e2008-02-11 22:47:46 +02003861
Christoph Lameter2c59dd62013-01-10 19:14:19 +00003862 s = kmalloc_slab(size, gfpflags);
Christoph Lameter81819f02007-05-06 14:49:36 -07003863
Satyam Sharma2408c552007-10-16 01:24:44 -07003864 if (unlikely(ZERO_OR_NULL_PTR(s)))
Christoph Lameter6cb8f912007-07-17 04:03:22 -07003865 return s;
Christoph Lameter81819f02007-05-06 14:49:36 -07003866
Ezequiel Garcia2b847c32012-09-08 17:47:58 -03003867 ret = slab_alloc_node(s, gfpflags, node, caller);
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03003868
Lucas De Marchi25985ed2011-03-30 22:57:33 -03003869 /* Honor the call site pointer we received. */
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02003870 trace_kmalloc_node(caller, ret, size, s->size, gfpflags, node);
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03003871
3872 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07003873}
Namhyung Kim5d1f57e2010-09-29 21:02:15 +09003874#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07003875
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05003876#ifdef CONFIG_SYSFS
Christoph Lameter205ab992008-04-14 19:11:40 +03003877static int count_inuse(struct page *page)
3878{
3879 return page->inuse;
3880}
3881
3882static int count_total(struct page *page)
3883{
3884 return page->objects;
3885}
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05003886#endif
Christoph Lameter205ab992008-04-14 19:11:40 +03003887
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05003888#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter434e2452007-07-17 04:03:30 -07003889static int validate_slab(struct kmem_cache *s, struct page *page,
3890 unsigned long *map)
Christoph Lameter53e15af2007-05-06 14:49:43 -07003891{
3892 void *p;
Christoph Lametera973e9d2008-03-01 13:40:44 -08003893 void *addr = page_address(page);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003894
3895 if (!check_slab(s, page) ||
3896 !on_freelist(s, page, NULL))
3897 return 0;
3898
3899 /* Now we know that a valid freelist exists */
Christoph Lameter39b26462008-04-14 19:11:30 +03003900 bitmap_zero(map, page->objects);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003901
Christoph Lameter5f80b132011-04-15 14:48:13 -05003902 get_map(s, page, map);
3903 for_each_object(p, s, addr, page->objects) {
3904 if (test_bit(slab_index(p, s, addr), map))
3905 if (!check_object(s, page, p, SLUB_RED_INACTIVE))
3906 return 0;
Christoph Lameter53e15af2007-05-06 14:49:43 -07003907 }
3908
Christoph Lameter224a88b2008-04-14 19:11:31 +03003909 for_each_object(p, s, addr, page->objects)
Christoph Lameter7656c722007-05-09 02:32:40 -07003910 if (!test_bit(slab_index(p, s, addr), map))
Tero Roponen37d57442010-12-01 20:04:20 +02003911 if (!check_object(s, page, p, SLUB_RED_ACTIVE))
Christoph Lameter53e15af2007-05-06 14:49:43 -07003912 return 0;
3913 return 1;
3914}
3915
Christoph Lameter434e2452007-07-17 04:03:30 -07003916static void validate_slab_slab(struct kmem_cache *s, struct page *page,
3917 unsigned long *map)
Christoph Lameter53e15af2007-05-06 14:49:43 -07003918{
Christoph Lameter881db7f2011-06-01 12:25:53 -05003919 slab_lock(page);
3920 validate_slab(s, page, map);
3921 slab_unlock(page);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003922}
3923
Christoph Lameter434e2452007-07-17 04:03:30 -07003924static int validate_slab_node(struct kmem_cache *s,
3925 struct kmem_cache_node *n, unsigned long *map)
Christoph Lameter53e15af2007-05-06 14:49:43 -07003926{
3927 unsigned long count = 0;
3928 struct page *page;
3929 unsigned long flags;
3930
3931 spin_lock_irqsave(&n->list_lock, flags);
3932
3933 list_for_each_entry(page, &n->partial, lru) {
Christoph Lameter434e2452007-07-17 04:03:30 -07003934 validate_slab_slab(s, page, map);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003935 count++;
3936 }
3937 if (count != n->nr_partial)
Fabian Frederickf9f58282014-06-04 16:06:34 -07003938 pr_err("SLUB %s: %ld partial slabs counted but counter=%ld\n",
3939 s->name, count, n->nr_partial);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003940
3941 if (!(s->flags & SLAB_STORE_USER))
3942 goto out;
3943
3944 list_for_each_entry(page, &n->full, lru) {
Christoph Lameter434e2452007-07-17 04:03:30 -07003945 validate_slab_slab(s, page, map);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003946 count++;
3947 }
3948 if (count != atomic_long_read(&n->nr_slabs))
Fabian Frederickf9f58282014-06-04 16:06:34 -07003949 pr_err("SLUB: %s %ld slabs counted but counter=%ld\n",
3950 s->name, count, atomic_long_read(&n->nr_slabs));
Christoph Lameter53e15af2007-05-06 14:49:43 -07003951
3952out:
3953 spin_unlock_irqrestore(&n->list_lock, flags);
3954 return count;
3955}
3956
Christoph Lameter434e2452007-07-17 04:03:30 -07003957static long validate_slab_cache(struct kmem_cache *s)
Christoph Lameter53e15af2007-05-06 14:49:43 -07003958{
3959 int node;
3960 unsigned long count = 0;
Christoph Lameter205ab992008-04-14 19:11:40 +03003961 unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) *
Christoph Lameter434e2452007-07-17 04:03:30 -07003962 sizeof(unsigned long), GFP_KERNEL);
3963
3964 if (!map)
3965 return -ENOMEM;
Christoph Lameter53e15af2007-05-06 14:49:43 -07003966
3967 flush_all(s);
Christoph Lameterf64dc582007-10-16 01:25:33 -07003968 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter53e15af2007-05-06 14:49:43 -07003969 struct kmem_cache_node *n = get_node(s, node);
3970
Christoph Lameter434e2452007-07-17 04:03:30 -07003971 count += validate_slab_node(s, n, map);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003972 }
Christoph Lameter434e2452007-07-17 04:03:30 -07003973 kfree(map);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003974 return count;
3975}
Christoph Lameter88a420e2007-05-06 14:49:45 -07003976/*
Christoph Lameter672bba32007-05-09 02:32:39 -07003977 * Generate lists of code addresses where slabcache objects are allocated
Christoph Lameter88a420e2007-05-06 14:49:45 -07003978 * and freed.
3979 */
3980
3981struct location {
3982 unsigned long count;
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03003983 unsigned long addr;
Christoph Lameter45edfa52007-05-09 02:32:45 -07003984 long long sum_time;
3985 long min_time;
3986 long max_time;
3987 long min_pid;
3988 long max_pid;
Rusty Russell174596a2009-01-01 10:12:29 +10303989 DECLARE_BITMAP(cpus, NR_CPUS);
Christoph Lameter45edfa52007-05-09 02:32:45 -07003990 nodemask_t nodes;
Christoph Lameter88a420e2007-05-06 14:49:45 -07003991};
3992
3993struct loc_track {
3994 unsigned long max;
3995 unsigned long count;
3996 struct location *loc;
3997};
3998
3999static void free_loc_track(struct loc_track *t)
4000{
4001 if (t->max)
4002 free_pages((unsigned long)t->loc,
4003 get_order(sizeof(struct location) * t->max));
4004}
4005
Christoph Lameter68dff6a2007-07-17 04:03:20 -07004006static int alloc_loc_track(struct loc_track *t, unsigned long max, gfp_t flags)
Christoph Lameter88a420e2007-05-06 14:49:45 -07004007{
4008 struct location *l;
4009 int order;
4010
Christoph Lameter88a420e2007-05-06 14:49:45 -07004011 order = get_order(sizeof(struct location) * max);
4012
Christoph Lameter68dff6a2007-07-17 04:03:20 -07004013 l = (void *)__get_free_pages(flags, order);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004014 if (!l)
4015 return 0;
4016
4017 if (t->count) {
4018 memcpy(l, t->loc, sizeof(struct location) * t->count);
4019 free_loc_track(t);
4020 }
4021 t->max = max;
4022 t->loc = l;
4023 return 1;
4024}
4025
4026static int add_location(struct loc_track *t, struct kmem_cache *s,
Christoph Lameter45edfa52007-05-09 02:32:45 -07004027 const struct track *track)
Christoph Lameter88a420e2007-05-06 14:49:45 -07004028{
4029 long start, end, pos;
4030 struct location *l;
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03004031 unsigned long caddr;
Christoph Lameter45edfa52007-05-09 02:32:45 -07004032 unsigned long age = jiffies - track->when;
Christoph Lameter88a420e2007-05-06 14:49:45 -07004033
4034 start = -1;
4035 end = t->count;
4036
4037 for ( ; ; ) {
4038 pos = start + (end - start + 1) / 2;
4039
4040 /*
4041 * There is nothing at "end". If we end up there
4042 * we need to add something to before end.
4043 */
4044 if (pos == end)
4045 break;
4046
4047 caddr = t->loc[pos].addr;
Christoph Lameter45edfa52007-05-09 02:32:45 -07004048 if (track->addr == caddr) {
4049
4050 l = &t->loc[pos];
4051 l->count++;
4052 if (track->when) {
4053 l->sum_time += age;
4054 if (age < l->min_time)
4055 l->min_time = age;
4056 if (age > l->max_time)
4057 l->max_time = age;
4058
4059 if (track->pid < l->min_pid)
4060 l->min_pid = track->pid;
4061 if (track->pid > l->max_pid)
4062 l->max_pid = track->pid;
4063
Rusty Russell174596a2009-01-01 10:12:29 +10304064 cpumask_set_cpu(track->cpu,
4065 to_cpumask(l->cpus));
Christoph Lameter45edfa52007-05-09 02:32:45 -07004066 }
4067 node_set(page_to_nid(virt_to_page(track)), l->nodes);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004068 return 1;
4069 }
4070
Christoph Lameter45edfa52007-05-09 02:32:45 -07004071 if (track->addr < caddr)
Christoph Lameter88a420e2007-05-06 14:49:45 -07004072 end = pos;
4073 else
4074 start = pos;
4075 }
4076
4077 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07004078 * Not found. Insert new tracking element.
Christoph Lameter88a420e2007-05-06 14:49:45 -07004079 */
Christoph Lameter68dff6a2007-07-17 04:03:20 -07004080 if (t->count >= t->max && !alloc_loc_track(t, 2 * t->max, GFP_ATOMIC))
Christoph Lameter88a420e2007-05-06 14:49:45 -07004081 return 0;
4082
4083 l = t->loc + pos;
4084 if (pos < t->count)
4085 memmove(l + 1, l,
4086 (t->count - pos) * sizeof(struct location));
4087 t->count++;
4088 l->count = 1;
Christoph Lameter45edfa52007-05-09 02:32:45 -07004089 l->addr = track->addr;
4090 l->sum_time = age;
4091 l->min_time = age;
4092 l->max_time = age;
4093 l->min_pid = track->pid;
4094 l->max_pid = track->pid;
Rusty Russell174596a2009-01-01 10:12:29 +10304095 cpumask_clear(to_cpumask(l->cpus));
4096 cpumask_set_cpu(track->cpu, to_cpumask(l->cpus));
Christoph Lameter45edfa52007-05-09 02:32:45 -07004097 nodes_clear(l->nodes);
4098 node_set(page_to_nid(virt_to_page(track)), l->nodes);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004099 return 1;
4100}
4101
4102static void process_slab(struct loc_track *t, struct kmem_cache *s,
Eric Dumazetbbd7d572010-03-24 22:25:47 +01004103 struct page *page, enum track_item alloc,
Namhyung Kima5dd5c12010-09-29 21:02:13 +09004104 unsigned long *map)
Christoph Lameter88a420e2007-05-06 14:49:45 -07004105{
Christoph Lametera973e9d2008-03-01 13:40:44 -08004106 void *addr = page_address(page);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004107 void *p;
4108
Christoph Lameter39b26462008-04-14 19:11:30 +03004109 bitmap_zero(map, page->objects);
Christoph Lameter5f80b132011-04-15 14:48:13 -05004110 get_map(s, page, map);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004111
Christoph Lameter224a88b2008-04-14 19:11:31 +03004112 for_each_object(p, s, addr, page->objects)
Christoph Lameter45edfa52007-05-09 02:32:45 -07004113 if (!test_bit(slab_index(p, s, addr), map))
4114 add_location(t, s, get_track(s, p, alloc));
Christoph Lameter88a420e2007-05-06 14:49:45 -07004115}
4116
4117static int list_locations(struct kmem_cache *s, char *buf,
4118 enum track_item alloc)
4119{
Harvey Harrisone374d482008-01-31 15:20:50 -08004120 int len = 0;
Christoph Lameter88a420e2007-05-06 14:49:45 -07004121 unsigned long i;
Christoph Lameter68dff6a2007-07-17 04:03:20 -07004122 struct loc_track t = { 0, 0, NULL };
Christoph Lameter88a420e2007-05-06 14:49:45 -07004123 int node;
Eric Dumazetbbd7d572010-03-24 22:25:47 +01004124 unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) *
4125 sizeof(unsigned long), GFP_KERNEL);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004126
Eric Dumazetbbd7d572010-03-24 22:25:47 +01004127 if (!map || !alloc_loc_track(&t, PAGE_SIZE / sizeof(struct location),
4128 GFP_TEMPORARY)) {
4129 kfree(map);
Christoph Lameter68dff6a2007-07-17 04:03:20 -07004130 return sprintf(buf, "Out of memory\n");
Eric Dumazetbbd7d572010-03-24 22:25:47 +01004131 }
Christoph Lameter88a420e2007-05-06 14:49:45 -07004132 /* Push back cpu slabs */
4133 flush_all(s);
4134
Christoph Lameterf64dc582007-10-16 01:25:33 -07004135 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter88a420e2007-05-06 14:49:45 -07004136 struct kmem_cache_node *n = get_node(s, node);
4137 unsigned long flags;
4138 struct page *page;
4139
Christoph Lameter9e869432007-08-22 14:01:56 -07004140 if (!atomic_long_read(&n->nr_slabs))
Christoph Lameter88a420e2007-05-06 14:49:45 -07004141 continue;
4142
4143 spin_lock_irqsave(&n->list_lock, flags);
4144 list_for_each_entry(page, &n->partial, lru)
Eric Dumazetbbd7d572010-03-24 22:25:47 +01004145 process_slab(&t, s, page, alloc, map);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004146 list_for_each_entry(page, &n->full, lru)
Eric Dumazetbbd7d572010-03-24 22:25:47 +01004147 process_slab(&t, s, page, alloc, map);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004148 spin_unlock_irqrestore(&n->list_lock, flags);
4149 }
4150
4151 for (i = 0; i < t.count; i++) {
Christoph Lameter45edfa52007-05-09 02:32:45 -07004152 struct location *l = &t.loc[i];
Christoph Lameter88a420e2007-05-06 14:49:45 -07004153
Hugh Dickins9c246242008-12-09 13:14:27 -08004154 if (len > PAGE_SIZE - KSYM_SYMBOL_LEN - 100)
Christoph Lameter88a420e2007-05-06 14:49:45 -07004155 break;
Harvey Harrisone374d482008-01-31 15:20:50 -08004156 len += sprintf(buf + len, "%7ld ", l->count);
Christoph Lameter45edfa52007-05-09 02:32:45 -07004157
4158 if (l->addr)
Joe Perches62c70bc2011-01-13 15:45:52 -08004159 len += sprintf(buf + len, "%pS", (void *)l->addr);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004160 else
Harvey Harrisone374d482008-01-31 15:20:50 -08004161 len += sprintf(buf + len, "<not-available>");
Christoph Lameter45edfa52007-05-09 02:32:45 -07004162
4163 if (l->sum_time != l->min_time) {
Harvey Harrisone374d482008-01-31 15:20:50 -08004164 len += sprintf(buf + len, " age=%ld/%ld/%ld",
Roman Zippelf8bd2252008-05-01 04:34:31 -07004165 l->min_time,
4166 (long)div_u64(l->sum_time, l->count),
4167 l->max_time);
Christoph Lameter45edfa52007-05-09 02:32:45 -07004168 } else
Harvey Harrisone374d482008-01-31 15:20:50 -08004169 len += sprintf(buf + len, " age=%ld",
Christoph Lameter45edfa52007-05-09 02:32:45 -07004170 l->min_time);
4171
4172 if (l->min_pid != l->max_pid)
Harvey Harrisone374d482008-01-31 15:20:50 -08004173 len += sprintf(buf + len, " pid=%ld-%ld",
Christoph Lameter45edfa52007-05-09 02:32:45 -07004174 l->min_pid, l->max_pid);
4175 else
Harvey Harrisone374d482008-01-31 15:20:50 -08004176 len += sprintf(buf + len, " pid=%ld",
Christoph Lameter45edfa52007-05-09 02:32:45 -07004177 l->min_pid);
4178
Rusty Russell174596a2009-01-01 10:12:29 +10304179 if (num_online_cpus() > 1 &&
4180 !cpumask_empty(to_cpumask(l->cpus)) &&
Harvey Harrisone374d482008-01-31 15:20:50 -08004181 len < PAGE_SIZE - 60) {
4182 len += sprintf(buf + len, " cpus=");
Chen Gangd0e0ac92013-07-15 09:05:29 +08004183 len += cpulist_scnprintf(buf + len,
4184 PAGE_SIZE - len - 50,
Rusty Russell174596a2009-01-01 10:12:29 +10304185 to_cpumask(l->cpus));
Christoph Lameter45edfa52007-05-09 02:32:45 -07004186 }
4187
Christoph Lameter62bc62a2009-06-16 15:32:15 -07004188 if (nr_online_nodes > 1 && !nodes_empty(l->nodes) &&
Harvey Harrisone374d482008-01-31 15:20:50 -08004189 len < PAGE_SIZE - 60) {
4190 len += sprintf(buf + len, " nodes=");
Chen Gangd0e0ac92013-07-15 09:05:29 +08004191 len += nodelist_scnprintf(buf + len,
4192 PAGE_SIZE - len - 50,
4193 l->nodes);
Christoph Lameter45edfa52007-05-09 02:32:45 -07004194 }
4195
Harvey Harrisone374d482008-01-31 15:20:50 -08004196 len += sprintf(buf + len, "\n");
Christoph Lameter88a420e2007-05-06 14:49:45 -07004197 }
4198
4199 free_loc_track(&t);
Eric Dumazetbbd7d572010-03-24 22:25:47 +01004200 kfree(map);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004201 if (!t.count)
Harvey Harrisone374d482008-01-31 15:20:50 -08004202 len += sprintf(buf, "No data\n");
4203 return len;
Christoph Lameter88a420e2007-05-06 14:49:45 -07004204}
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004205#endif
Christoph Lameter88a420e2007-05-06 14:49:45 -07004206
Christoph Lametera5a84752010-10-05 13:57:27 -05004207#ifdef SLUB_RESILIENCY_TEST
4208static void resiliency_test(void)
4209{
4210 u8 *p;
4211
Christoph Lameter95a05b42013-01-10 19:14:19 +00004212 BUILD_BUG_ON(KMALLOC_MIN_SIZE > 16 || KMALLOC_SHIFT_HIGH < 10);
Christoph Lametera5a84752010-10-05 13:57:27 -05004213
Fabian Frederickf9f58282014-06-04 16:06:34 -07004214 pr_err("SLUB resiliency testing\n");
4215 pr_err("-----------------------\n");
4216 pr_err("A. Corruption after allocation\n");
Christoph Lametera5a84752010-10-05 13:57:27 -05004217
4218 p = kzalloc(16, GFP_KERNEL);
4219 p[16] = 0x12;
Fabian Frederickf9f58282014-06-04 16:06:34 -07004220 pr_err("\n1. kmalloc-16: Clobber Redzone/next pointer 0x12->0x%p\n\n",
4221 p + 16);
Christoph Lametera5a84752010-10-05 13:57:27 -05004222
4223 validate_slab_cache(kmalloc_caches[4]);
4224
4225 /* Hmmm... The next two are dangerous */
4226 p = kzalloc(32, GFP_KERNEL);
4227 p[32 + sizeof(void *)] = 0x34;
Fabian Frederickf9f58282014-06-04 16:06:34 -07004228 pr_err("\n2. kmalloc-32: Clobber next pointer/next slab 0x34 -> -0x%p\n",
4229 p);
4230 pr_err("If allocated object is overwritten then not detectable\n\n");
Christoph Lametera5a84752010-10-05 13:57:27 -05004231
4232 validate_slab_cache(kmalloc_caches[5]);
4233 p = kzalloc(64, GFP_KERNEL);
4234 p += 64 + (get_cycles() & 0xff) * sizeof(void *);
4235 *p = 0x56;
Fabian Frederickf9f58282014-06-04 16:06:34 -07004236 pr_err("\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n",
4237 p);
4238 pr_err("If allocated object is overwritten then not detectable\n\n");
Christoph Lametera5a84752010-10-05 13:57:27 -05004239 validate_slab_cache(kmalloc_caches[6]);
4240
Fabian Frederickf9f58282014-06-04 16:06:34 -07004241 pr_err("\nB. Corruption after free\n");
Christoph Lametera5a84752010-10-05 13:57:27 -05004242 p = kzalloc(128, GFP_KERNEL);
4243 kfree(p);
4244 *p = 0x78;
Fabian Frederickf9f58282014-06-04 16:06:34 -07004245 pr_err("1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p);
Christoph Lametera5a84752010-10-05 13:57:27 -05004246 validate_slab_cache(kmalloc_caches[7]);
4247
4248 p = kzalloc(256, GFP_KERNEL);
4249 kfree(p);
4250 p[50] = 0x9a;
Fabian Frederickf9f58282014-06-04 16:06:34 -07004251 pr_err("\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n", p);
Christoph Lametera5a84752010-10-05 13:57:27 -05004252 validate_slab_cache(kmalloc_caches[8]);
4253
4254 p = kzalloc(512, GFP_KERNEL);
4255 kfree(p);
4256 p[512] = 0xab;
Fabian Frederickf9f58282014-06-04 16:06:34 -07004257 pr_err("\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p);
Christoph Lametera5a84752010-10-05 13:57:27 -05004258 validate_slab_cache(kmalloc_caches[9]);
4259}
4260#else
4261#ifdef CONFIG_SYSFS
4262static void resiliency_test(void) {};
4263#endif
4264#endif
4265
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004266#ifdef CONFIG_SYSFS
Christoph Lameter81819f02007-05-06 14:49:36 -07004267enum slab_stat_type {
Christoph Lameter205ab992008-04-14 19:11:40 +03004268 SL_ALL, /* All slabs */
4269 SL_PARTIAL, /* Only partially allocated slabs */
4270 SL_CPU, /* Only slabs used for cpu caches */
4271 SL_OBJECTS, /* Determine allocated objects not slabs */
4272 SL_TOTAL /* Determine object capacity not slabs */
Christoph Lameter81819f02007-05-06 14:49:36 -07004273};
4274
Christoph Lameter205ab992008-04-14 19:11:40 +03004275#define SO_ALL (1 << SL_ALL)
Christoph Lameter81819f02007-05-06 14:49:36 -07004276#define SO_PARTIAL (1 << SL_PARTIAL)
4277#define SO_CPU (1 << SL_CPU)
4278#define SO_OBJECTS (1 << SL_OBJECTS)
Christoph Lameter205ab992008-04-14 19:11:40 +03004279#define SO_TOTAL (1 << SL_TOTAL)
Christoph Lameter81819f02007-05-06 14:49:36 -07004280
Cyrill Gorcunov62e5c4b2008-03-02 23:28:24 +03004281static ssize_t show_slab_objects(struct kmem_cache *s,
4282 char *buf, unsigned long flags)
Christoph Lameter81819f02007-05-06 14:49:36 -07004283{
4284 unsigned long total = 0;
Christoph Lameter81819f02007-05-06 14:49:36 -07004285 int node;
4286 int x;
4287 unsigned long *nodes;
Christoph Lameter81819f02007-05-06 14:49:36 -07004288
Chen Gange35e1a92013-07-12 08:23:48 +08004289 nodes = kzalloc(sizeof(unsigned long) * nr_node_ids, GFP_KERNEL);
Cyrill Gorcunov62e5c4b2008-03-02 23:28:24 +03004290 if (!nodes)
4291 return -ENOMEM;
Christoph Lameter81819f02007-05-06 14:49:36 -07004292
Christoph Lameter205ab992008-04-14 19:11:40 +03004293 if (flags & SO_CPU) {
4294 int cpu;
Christoph Lameter81819f02007-05-06 14:49:36 -07004295
Christoph Lameter205ab992008-04-14 19:11:40 +03004296 for_each_possible_cpu(cpu) {
Chen Gangd0e0ac92013-07-15 09:05:29 +08004297 struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab,
4298 cpu);
Christoph Lameterec3ab082012-05-09 10:09:56 -05004299 int node;
Christoph Lameter49e22582011-08-09 16:12:27 -05004300 struct page *page;
Christoph Lameterdfb4f092007-10-16 01:26:05 -07004301
Eric Dumazetbc6697d2011-11-22 16:02:02 +01004302 page = ACCESS_ONCE(c->page);
Christoph Lameterec3ab082012-05-09 10:09:56 -05004303 if (!page)
4304 continue;
Christoph Lameter205ab992008-04-14 19:11:40 +03004305
Christoph Lameterec3ab082012-05-09 10:09:56 -05004306 node = page_to_nid(page);
4307 if (flags & SO_TOTAL)
4308 x = page->objects;
4309 else if (flags & SO_OBJECTS)
4310 x = page->inuse;
4311 else
4312 x = 1;
Christoph Lameter49e22582011-08-09 16:12:27 -05004313
Christoph Lameterec3ab082012-05-09 10:09:56 -05004314 total += x;
4315 nodes[node] += x;
4316
4317 page = ACCESS_ONCE(c->partial);
Christoph Lameter49e22582011-08-09 16:12:27 -05004318 if (page) {
Li Zefan8afb1472013-09-10 11:43:37 +08004319 node = page_to_nid(page);
4320 if (flags & SO_TOTAL)
4321 WARN_ON_ONCE(1);
4322 else if (flags & SO_OBJECTS)
4323 WARN_ON_ONCE(1);
4324 else
4325 x = page->pages;
Eric Dumazetbc6697d2011-11-22 16:02:02 +01004326 total += x;
4327 nodes[node] += x;
Christoph Lameter49e22582011-08-09 16:12:27 -05004328 }
Christoph Lameter81819f02007-05-06 14:49:36 -07004329 }
4330 }
4331
Vladimir Davydovbfc8c902014-06-04 16:07:18 -07004332 get_online_mems();
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004333#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter205ab992008-04-14 19:11:40 +03004334 if (flags & SO_ALL) {
4335 for_each_node_state(node, N_NORMAL_MEMORY) {
4336 struct kmem_cache_node *n = get_node(s, node);
Christoph Lameter81819f02007-05-06 14:49:36 -07004337
Chen Gangd0e0ac92013-07-15 09:05:29 +08004338 if (flags & SO_TOTAL)
4339 x = atomic_long_read(&n->total_objects);
4340 else if (flags & SO_OBJECTS)
4341 x = atomic_long_read(&n->total_objects) -
4342 count_partial(n, count_free);
Christoph Lameter205ab992008-04-14 19:11:40 +03004343 else
4344 x = atomic_long_read(&n->nr_slabs);
4345 total += x;
4346 nodes[node] += x;
4347 }
4348
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004349 } else
4350#endif
4351 if (flags & SO_PARTIAL) {
Christoph Lameter205ab992008-04-14 19:11:40 +03004352 for_each_node_state(node, N_NORMAL_MEMORY) {
4353 struct kmem_cache_node *n = get_node(s, node);
4354
4355 if (flags & SO_TOTAL)
4356 x = count_partial(n, count_total);
4357 else if (flags & SO_OBJECTS)
4358 x = count_partial(n, count_inuse);
Christoph Lameter81819f02007-05-06 14:49:36 -07004359 else
4360 x = n->nr_partial;
4361 total += x;
4362 nodes[node] += x;
4363 }
Christoph Lameter81819f02007-05-06 14:49:36 -07004364 }
Christoph Lameter81819f02007-05-06 14:49:36 -07004365 x = sprintf(buf, "%lu", total);
4366#ifdef CONFIG_NUMA
Christoph Lameterf64dc582007-10-16 01:25:33 -07004367 for_each_node_state(node, N_NORMAL_MEMORY)
Christoph Lameter81819f02007-05-06 14:49:36 -07004368 if (nodes[node])
4369 x += sprintf(buf + x, " N%d=%lu",
4370 node, nodes[node]);
4371#endif
Vladimir Davydovbfc8c902014-06-04 16:07:18 -07004372 put_online_mems();
Christoph Lameter81819f02007-05-06 14:49:36 -07004373 kfree(nodes);
4374 return x + sprintf(buf + x, "\n");
4375}
4376
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004377#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter81819f02007-05-06 14:49:36 -07004378static int any_slab_objects(struct kmem_cache *s)
4379{
4380 int node;
Christoph Lameterdfb4f092007-10-16 01:26:05 -07004381
4382 for_each_online_node(node) {
Christoph Lameter81819f02007-05-06 14:49:36 -07004383 struct kmem_cache_node *n = get_node(s, node);
4384
Christoph Lameterdfb4f092007-10-16 01:26:05 -07004385 if (!n)
4386 continue;
4387
Benjamin Herrenschmidt4ea33e22008-05-06 20:42:39 -07004388 if (atomic_long_read(&n->total_objects))
Christoph Lameter81819f02007-05-06 14:49:36 -07004389 return 1;
4390 }
4391 return 0;
4392}
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004393#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07004394
4395#define to_slab_attr(n) container_of(n, struct slab_attribute, attr)
Phil Carmody497888c2011-07-14 15:07:13 +03004396#define to_slab(n) container_of(n, struct kmem_cache, kobj)
Christoph Lameter81819f02007-05-06 14:49:36 -07004397
4398struct slab_attribute {
4399 struct attribute attr;
4400 ssize_t (*show)(struct kmem_cache *s, char *buf);
4401 ssize_t (*store)(struct kmem_cache *s, const char *x, size_t count);
4402};
4403
4404#define SLAB_ATTR_RO(_name) \
Vasiliy Kulikovab067e92011-09-27 21:54:53 +04004405 static struct slab_attribute _name##_attr = \
4406 __ATTR(_name, 0400, _name##_show, NULL)
Christoph Lameter81819f02007-05-06 14:49:36 -07004407
4408#define SLAB_ATTR(_name) \
4409 static struct slab_attribute _name##_attr = \
Vasiliy Kulikovab067e92011-09-27 21:54:53 +04004410 __ATTR(_name, 0600, _name##_show, _name##_store)
Christoph Lameter81819f02007-05-06 14:49:36 -07004411
Christoph Lameter81819f02007-05-06 14:49:36 -07004412static ssize_t slab_size_show(struct kmem_cache *s, char *buf)
4413{
4414 return sprintf(buf, "%d\n", s->size);
4415}
4416SLAB_ATTR_RO(slab_size);
4417
4418static ssize_t align_show(struct kmem_cache *s, char *buf)
4419{
4420 return sprintf(buf, "%d\n", s->align);
4421}
4422SLAB_ATTR_RO(align);
4423
4424static ssize_t object_size_show(struct kmem_cache *s, char *buf)
4425{
Christoph Lameter3b0efdf2012-06-13 10:24:57 -05004426 return sprintf(buf, "%d\n", s->object_size);
Christoph Lameter81819f02007-05-06 14:49:36 -07004427}
4428SLAB_ATTR_RO(object_size);
4429
4430static ssize_t objs_per_slab_show(struct kmem_cache *s, char *buf)
4431{
Christoph Lameter834f3d12008-04-14 19:11:31 +03004432 return sprintf(buf, "%d\n", oo_objects(s->oo));
Christoph Lameter81819f02007-05-06 14:49:36 -07004433}
4434SLAB_ATTR_RO(objs_per_slab);
4435
Christoph Lameter06b285d2008-04-14 19:11:41 +03004436static ssize_t order_store(struct kmem_cache *s,
4437 const char *buf, size_t length)
4438{
Christoph Lameter0121c6192008-04-29 16:11:12 -07004439 unsigned long order;
4440 int err;
4441
Jingoo Han3dbb95f2013-09-11 14:20:25 -07004442 err = kstrtoul(buf, 10, &order);
Christoph Lameter0121c6192008-04-29 16:11:12 -07004443 if (err)
4444 return err;
Christoph Lameter06b285d2008-04-14 19:11:41 +03004445
4446 if (order > slub_max_order || order < slub_min_order)
4447 return -EINVAL;
4448
4449 calculate_sizes(s, order);
4450 return length;
4451}
4452
Christoph Lameter81819f02007-05-06 14:49:36 -07004453static ssize_t order_show(struct kmem_cache *s, char *buf)
4454{
Christoph Lameter834f3d12008-04-14 19:11:31 +03004455 return sprintf(buf, "%d\n", oo_order(s->oo));
Christoph Lameter81819f02007-05-06 14:49:36 -07004456}
Christoph Lameter06b285d2008-04-14 19:11:41 +03004457SLAB_ATTR(order);
Christoph Lameter81819f02007-05-06 14:49:36 -07004458
David Rientjes73d342b2009-02-22 17:40:09 -08004459static ssize_t min_partial_show(struct kmem_cache *s, char *buf)
4460{
4461 return sprintf(buf, "%lu\n", s->min_partial);
4462}
4463
4464static ssize_t min_partial_store(struct kmem_cache *s, const char *buf,
4465 size_t length)
4466{
4467 unsigned long min;
4468 int err;
4469
Jingoo Han3dbb95f2013-09-11 14:20:25 -07004470 err = kstrtoul(buf, 10, &min);
David Rientjes73d342b2009-02-22 17:40:09 -08004471 if (err)
4472 return err;
4473
David Rientjesc0bdb232009-02-25 09:16:35 +02004474 set_min_partial(s, min);
David Rientjes73d342b2009-02-22 17:40:09 -08004475 return length;
4476}
4477SLAB_ATTR(min_partial);
4478
Christoph Lameter49e22582011-08-09 16:12:27 -05004479static ssize_t cpu_partial_show(struct kmem_cache *s, char *buf)
4480{
4481 return sprintf(buf, "%u\n", s->cpu_partial);
4482}
4483
4484static ssize_t cpu_partial_store(struct kmem_cache *s, const char *buf,
4485 size_t length)
4486{
4487 unsigned long objects;
4488 int err;
4489
Jingoo Han3dbb95f2013-09-11 14:20:25 -07004490 err = kstrtoul(buf, 10, &objects);
Christoph Lameter49e22582011-08-09 16:12:27 -05004491 if (err)
4492 return err;
Joonsoo Kim345c9052013-06-19 14:05:52 +09004493 if (objects && !kmem_cache_has_cpu_partial(s))
David Rientjes74ee4ef2012-01-09 13:19:45 -08004494 return -EINVAL;
Christoph Lameter49e22582011-08-09 16:12:27 -05004495
4496 s->cpu_partial = objects;
4497 flush_all(s);
4498 return length;
4499}
4500SLAB_ATTR(cpu_partial);
4501
Christoph Lameter81819f02007-05-06 14:49:36 -07004502static ssize_t ctor_show(struct kmem_cache *s, char *buf)
4503{
Joe Perches62c70bc2011-01-13 15:45:52 -08004504 if (!s->ctor)
4505 return 0;
4506 return sprintf(buf, "%pS\n", s->ctor);
Christoph Lameter81819f02007-05-06 14:49:36 -07004507}
4508SLAB_ATTR_RO(ctor);
4509
Christoph Lameter81819f02007-05-06 14:49:36 -07004510static ssize_t aliases_show(struct kmem_cache *s, char *buf)
4511{
4512 return sprintf(buf, "%d\n", s->refcount - 1);
4513}
4514SLAB_ATTR_RO(aliases);
4515
Christoph Lameter81819f02007-05-06 14:49:36 -07004516static ssize_t partial_show(struct kmem_cache *s, char *buf)
4517{
Christoph Lameterd9acf4b2008-02-15 15:22:21 -08004518 return show_slab_objects(s, buf, SO_PARTIAL);
Christoph Lameter81819f02007-05-06 14:49:36 -07004519}
4520SLAB_ATTR_RO(partial);
4521
4522static ssize_t cpu_slabs_show(struct kmem_cache *s, char *buf)
4523{
Christoph Lameterd9acf4b2008-02-15 15:22:21 -08004524 return show_slab_objects(s, buf, SO_CPU);
Christoph Lameter81819f02007-05-06 14:49:36 -07004525}
4526SLAB_ATTR_RO(cpu_slabs);
4527
4528static ssize_t objects_show(struct kmem_cache *s, char *buf)
4529{
Christoph Lameter205ab992008-04-14 19:11:40 +03004530 return show_slab_objects(s, buf, SO_ALL|SO_OBJECTS);
Christoph Lameter81819f02007-05-06 14:49:36 -07004531}
4532SLAB_ATTR_RO(objects);
4533
Christoph Lameter205ab992008-04-14 19:11:40 +03004534static ssize_t objects_partial_show(struct kmem_cache *s, char *buf)
4535{
4536 return show_slab_objects(s, buf, SO_PARTIAL|SO_OBJECTS);
4537}
4538SLAB_ATTR_RO(objects_partial);
4539
Christoph Lameter49e22582011-08-09 16:12:27 -05004540static ssize_t slabs_cpu_partial_show(struct kmem_cache *s, char *buf)
4541{
4542 int objects = 0;
4543 int pages = 0;
4544 int cpu;
4545 int len;
4546
4547 for_each_online_cpu(cpu) {
4548 struct page *page = per_cpu_ptr(s->cpu_slab, cpu)->partial;
4549
4550 if (page) {
4551 pages += page->pages;
4552 objects += page->pobjects;
4553 }
4554 }
4555
4556 len = sprintf(buf, "%d(%d)", objects, pages);
4557
4558#ifdef CONFIG_SMP
4559 for_each_online_cpu(cpu) {
4560 struct page *page = per_cpu_ptr(s->cpu_slab, cpu) ->partial;
4561
4562 if (page && len < PAGE_SIZE - 20)
4563 len += sprintf(buf + len, " C%d=%d(%d)", cpu,
4564 page->pobjects, page->pages);
4565 }
4566#endif
4567 return len + sprintf(buf + len, "\n");
4568}
4569SLAB_ATTR_RO(slabs_cpu_partial);
4570
Christoph Lameter81819f02007-05-06 14:49:36 -07004571static ssize_t reclaim_account_show(struct kmem_cache *s, char *buf)
4572{
4573 return sprintf(buf, "%d\n", !!(s->flags & SLAB_RECLAIM_ACCOUNT));
4574}
4575
4576static ssize_t reclaim_account_store(struct kmem_cache *s,
4577 const char *buf, size_t length)
4578{
4579 s->flags &= ~SLAB_RECLAIM_ACCOUNT;
4580 if (buf[0] == '1')
4581 s->flags |= SLAB_RECLAIM_ACCOUNT;
4582 return length;
4583}
4584SLAB_ATTR(reclaim_account);
4585
4586static ssize_t hwcache_align_show(struct kmem_cache *s, char *buf)
4587{
Christoph Lameter5af60832007-05-06 14:49:56 -07004588 return sprintf(buf, "%d\n", !!(s->flags & SLAB_HWCACHE_ALIGN));
Christoph Lameter81819f02007-05-06 14:49:36 -07004589}
4590SLAB_ATTR_RO(hwcache_align);
4591
4592#ifdef CONFIG_ZONE_DMA
4593static ssize_t cache_dma_show(struct kmem_cache *s, char *buf)
4594{
4595 return sprintf(buf, "%d\n", !!(s->flags & SLAB_CACHE_DMA));
4596}
4597SLAB_ATTR_RO(cache_dma);
4598#endif
4599
4600static ssize_t destroy_by_rcu_show(struct kmem_cache *s, char *buf)
4601{
4602 return sprintf(buf, "%d\n", !!(s->flags & SLAB_DESTROY_BY_RCU));
4603}
4604SLAB_ATTR_RO(destroy_by_rcu);
4605
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08004606static ssize_t reserved_show(struct kmem_cache *s, char *buf)
4607{
4608 return sprintf(buf, "%d\n", s->reserved);
4609}
4610SLAB_ATTR_RO(reserved);
4611
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004612#ifdef CONFIG_SLUB_DEBUG
Christoph Lametera5a84752010-10-05 13:57:27 -05004613static ssize_t slabs_show(struct kmem_cache *s, char *buf)
4614{
4615 return show_slab_objects(s, buf, SO_ALL);
4616}
4617SLAB_ATTR_RO(slabs);
4618
4619static ssize_t total_objects_show(struct kmem_cache *s, char *buf)
4620{
4621 return show_slab_objects(s, buf, SO_ALL|SO_TOTAL);
4622}
4623SLAB_ATTR_RO(total_objects);
4624
4625static ssize_t sanity_checks_show(struct kmem_cache *s, char *buf)
4626{
4627 return sprintf(buf, "%d\n", !!(s->flags & SLAB_DEBUG_FREE));
4628}
4629
4630static ssize_t sanity_checks_store(struct kmem_cache *s,
4631 const char *buf, size_t length)
4632{
4633 s->flags &= ~SLAB_DEBUG_FREE;
Christoph Lameterb789ef52011-06-01 12:25:49 -05004634 if (buf[0] == '1') {
4635 s->flags &= ~__CMPXCHG_DOUBLE;
Christoph Lametera5a84752010-10-05 13:57:27 -05004636 s->flags |= SLAB_DEBUG_FREE;
Christoph Lameterb789ef52011-06-01 12:25:49 -05004637 }
Christoph Lametera5a84752010-10-05 13:57:27 -05004638 return length;
4639}
4640SLAB_ATTR(sanity_checks);
4641
4642static ssize_t trace_show(struct kmem_cache *s, char *buf)
4643{
4644 return sprintf(buf, "%d\n", !!(s->flags & SLAB_TRACE));
4645}
4646
4647static ssize_t trace_store(struct kmem_cache *s, const char *buf,
4648 size_t length)
4649{
4650 s->flags &= ~SLAB_TRACE;
Christoph Lameterb789ef52011-06-01 12:25:49 -05004651 if (buf[0] == '1') {
4652 s->flags &= ~__CMPXCHG_DOUBLE;
Christoph Lametera5a84752010-10-05 13:57:27 -05004653 s->flags |= SLAB_TRACE;
Christoph Lameterb789ef52011-06-01 12:25:49 -05004654 }
Christoph Lametera5a84752010-10-05 13:57:27 -05004655 return length;
4656}
4657SLAB_ATTR(trace);
4658
Christoph Lameter81819f02007-05-06 14:49:36 -07004659static ssize_t red_zone_show(struct kmem_cache *s, char *buf)
4660{
4661 return sprintf(buf, "%d\n", !!(s->flags & SLAB_RED_ZONE));
4662}
4663
4664static ssize_t red_zone_store(struct kmem_cache *s,
4665 const char *buf, size_t length)
4666{
4667 if (any_slab_objects(s))
4668 return -EBUSY;
4669
4670 s->flags &= ~SLAB_RED_ZONE;
Christoph Lameterb789ef52011-06-01 12:25:49 -05004671 if (buf[0] == '1') {
4672 s->flags &= ~__CMPXCHG_DOUBLE;
Christoph Lameter81819f02007-05-06 14:49:36 -07004673 s->flags |= SLAB_RED_ZONE;
Christoph Lameterb789ef52011-06-01 12:25:49 -05004674 }
Christoph Lameter06b285d2008-04-14 19:11:41 +03004675 calculate_sizes(s, -1);
Christoph Lameter81819f02007-05-06 14:49:36 -07004676 return length;
4677}
4678SLAB_ATTR(red_zone);
4679
4680static ssize_t poison_show(struct kmem_cache *s, char *buf)
4681{
4682 return sprintf(buf, "%d\n", !!(s->flags & SLAB_POISON));
4683}
4684
4685static ssize_t poison_store(struct kmem_cache *s,
4686 const char *buf, size_t length)
4687{
4688 if (any_slab_objects(s))
4689 return -EBUSY;
4690
4691 s->flags &= ~SLAB_POISON;
Christoph Lameterb789ef52011-06-01 12:25:49 -05004692 if (buf[0] == '1') {
4693 s->flags &= ~__CMPXCHG_DOUBLE;
Christoph Lameter81819f02007-05-06 14:49:36 -07004694 s->flags |= SLAB_POISON;
Christoph Lameterb789ef52011-06-01 12:25:49 -05004695 }
Christoph Lameter06b285d2008-04-14 19:11:41 +03004696 calculate_sizes(s, -1);
Christoph Lameter81819f02007-05-06 14:49:36 -07004697 return length;
4698}
4699SLAB_ATTR(poison);
4700
4701static ssize_t store_user_show(struct kmem_cache *s, char *buf)
4702{
4703 return sprintf(buf, "%d\n", !!(s->flags & SLAB_STORE_USER));
4704}
4705
4706static ssize_t store_user_store(struct kmem_cache *s,
4707 const char *buf, size_t length)
4708{
4709 if (any_slab_objects(s))
4710 return -EBUSY;
4711
4712 s->flags &= ~SLAB_STORE_USER;
Christoph Lameterb789ef52011-06-01 12:25:49 -05004713 if (buf[0] == '1') {
4714 s->flags &= ~__CMPXCHG_DOUBLE;
Christoph Lameter81819f02007-05-06 14:49:36 -07004715 s->flags |= SLAB_STORE_USER;
Christoph Lameterb789ef52011-06-01 12:25:49 -05004716 }
Christoph Lameter06b285d2008-04-14 19:11:41 +03004717 calculate_sizes(s, -1);
Christoph Lameter81819f02007-05-06 14:49:36 -07004718 return length;
4719}
4720SLAB_ATTR(store_user);
4721
Christoph Lameter53e15af2007-05-06 14:49:43 -07004722static ssize_t validate_show(struct kmem_cache *s, char *buf)
4723{
4724 return 0;
4725}
4726
4727static ssize_t validate_store(struct kmem_cache *s,
4728 const char *buf, size_t length)
4729{
Christoph Lameter434e2452007-07-17 04:03:30 -07004730 int ret = -EINVAL;
4731
4732 if (buf[0] == '1') {
4733 ret = validate_slab_cache(s);
4734 if (ret >= 0)
4735 ret = length;
4736 }
4737 return ret;
Christoph Lameter53e15af2007-05-06 14:49:43 -07004738}
4739SLAB_ATTR(validate);
Christoph Lametera5a84752010-10-05 13:57:27 -05004740
4741static ssize_t alloc_calls_show(struct kmem_cache *s, char *buf)
4742{
4743 if (!(s->flags & SLAB_STORE_USER))
4744 return -ENOSYS;
4745 return list_locations(s, buf, TRACK_ALLOC);
4746}
4747SLAB_ATTR_RO(alloc_calls);
4748
4749static ssize_t free_calls_show(struct kmem_cache *s, char *buf)
4750{
4751 if (!(s->flags & SLAB_STORE_USER))
4752 return -ENOSYS;
4753 return list_locations(s, buf, TRACK_FREE);
4754}
4755SLAB_ATTR_RO(free_calls);
4756#endif /* CONFIG_SLUB_DEBUG */
4757
4758#ifdef CONFIG_FAILSLAB
4759static ssize_t failslab_show(struct kmem_cache *s, char *buf)
4760{
4761 return sprintf(buf, "%d\n", !!(s->flags & SLAB_FAILSLAB));
4762}
4763
4764static ssize_t failslab_store(struct kmem_cache *s, const char *buf,
4765 size_t length)
4766{
4767 s->flags &= ~SLAB_FAILSLAB;
4768 if (buf[0] == '1')
4769 s->flags |= SLAB_FAILSLAB;
4770 return length;
4771}
4772SLAB_ATTR(failslab);
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004773#endif
Christoph Lameter53e15af2007-05-06 14:49:43 -07004774
Christoph Lameter2086d262007-05-06 14:49:46 -07004775static ssize_t shrink_show(struct kmem_cache *s, char *buf)
4776{
4777 return 0;
4778}
4779
4780static ssize_t shrink_store(struct kmem_cache *s,
4781 const char *buf, size_t length)
4782{
4783 if (buf[0] == '1') {
4784 int rc = kmem_cache_shrink(s);
4785
4786 if (rc)
4787 return rc;
4788 } else
4789 return -EINVAL;
4790 return length;
4791}
4792SLAB_ATTR(shrink);
4793
Christoph Lameter81819f02007-05-06 14:49:36 -07004794#ifdef CONFIG_NUMA
Christoph Lameter98246012008-01-07 23:20:26 -08004795static ssize_t remote_node_defrag_ratio_show(struct kmem_cache *s, char *buf)
Christoph Lameter81819f02007-05-06 14:49:36 -07004796{
Christoph Lameter98246012008-01-07 23:20:26 -08004797 return sprintf(buf, "%d\n", s->remote_node_defrag_ratio / 10);
Christoph Lameter81819f02007-05-06 14:49:36 -07004798}
4799
Christoph Lameter98246012008-01-07 23:20:26 -08004800static ssize_t remote_node_defrag_ratio_store(struct kmem_cache *s,
Christoph Lameter81819f02007-05-06 14:49:36 -07004801 const char *buf, size_t length)
4802{
Christoph Lameter0121c6192008-04-29 16:11:12 -07004803 unsigned long ratio;
4804 int err;
Christoph Lameter81819f02007-05-06 14:49:36 -07004805
Jingoo Han3dbb95f2013-09-11 14:20:25 -07004806 err = kstrtoul(buf, 10, &ratio);
Christoph Lameter0121c6192008-04-29 16:11:12 -07004807 if (err)
4808 return err;
4809
Christoph Lametere2cb96b2008-08-19 08:51:22 -05004810 if (ratio <= 100)
Christoph Lameter0121c6192008-04-29 16:11:12 -07004811 s->remote_node_defrag_ratio = ratio * 10;
4812
Christoph Lameter81819f02007-05-06 14:49:36 -07004813 return length;
4814}
Christoph Lameter98246012008-01-07 23:20:26 -08004815SLAB_ATTR(remote_node_defrag_ratio);
Christoph Lameter81819f02007-05-06 14:49:36 -07004816#endif
4817
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004818#ifdef CONFIG_SLUB_STATS
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004819static int show_stat(struct kmem_cache *s, char *buf, enum stat_item si)
4820{
4821 unsigned long sum = 0;
4822 int cpu;
4823 int len;
4824 int *data = kmalloc(nr_cpu_ids * sizeof(int), GFP_KERNEL);
4825
4826 if (!data)
4827 return -ENOMEM;
4828
4829 for_each_online_cpu(cpu) {
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06004830 unsigned x = per_cpu_ptr(s->cpu_slab, cpu)->stat[si];
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004831
4832 data[cpu] = x;
4833 sum += x;
4834 }
4835
4836 len = sprintf(buf, "%lu", sum);
4837
Christoph Lameter50ef37b2008-04-14 18:52:05 +03004838#ifdef CONFIG_SMP
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004839 for_each_online_cpu(cpu) {
4840 if (data[cpu] && len < PAGE_SIZE - 20)
Christoph Lameter50ef37b2008-04-14 18:52:05 +03004841 len += sprintf(buf + len, " C%d=%u", cpu, data[cpu]);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004842 }
Christoph Lameter50ef37b2008-04-14 18:52:05 +03004843#endif
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004844 kfree(data);
4845 return len + sprintf(buf + len, "\n");
4846}
4847
David Rientjes78eb00c2009-10-15 02:20:22 -07004848static void clear_stat(struct kmem_cache *s, enum stat_item si)
4849{
4850 int cpu;
4851
4852 for_each_online_cpu(cpu)
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06004853 per_cpu_ptr(s->cpu_slab, cpu)->stat[si] = 0;
David Rientjes78eb00c2009-10-15 02:20:22 -07004854}
4855
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004856#define STAT_ATTR(si, text) \
4857static ssize_t text##_show(struct kmem_cache *s, char *buf) \
4858{ \
4859 return show_stat(s, buf, si); \
4860} \
David Rientjes78eb00c2009-10-15 02:20:22 -07004861static ssize_t text##_store(struct kmem_cache *s, \
4862 const char *buf, size_t length) \
4863{ \
4864 if (buf[0] != '0') \
4865 return -EINVAL; \
4866 clear_stat(s, si); \
4867 return length; \
4868} \
4869SLAB_ATTR(text); \
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004870
4871STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath);
4872STAT_ATTR(ALLOC_SLOWPATH, alloc_slowpath);
4873STAT_ATTR(FREE_FASTPATH, free_fastpath);
4874STAT_ATTR(FREE_SLOWPATH, free_slowpath);
4875STAT_ATTR(FREE_FROZEN, free_frozen);
4876STAT_ATTR(FREE_ADD_PARTIAL, free_add_partial);
4877STAT_ATTR(FREE_REMOVE_PARTIAL, free_remove_partial);
4878STAT_ATTR(ALLOC_FROM_PARTIAL, alloc_from_partial);
4879STAT_ATTR(ALLOC_SLAB, alloc_slab);
4880STAT_ATTR(ALLOC_REFILL, alloc_refill);
Christoph Lametere36a2652011-06-01 12:25:57 -05004881STAT_ATTR(ALLOC_NODE_MISMATCH, alloc_node_mismatch);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004882STAT_ATTR(FREE_SLAB, free_slab);
4883STAT_ATTR(CPUSLAB_FLUSH, cpuslab_flush);
4884STAT_ATTR(DEACTIVATE_FULL, deactivate_full);
4885STAT_ATTR(DEACTIVATE_EMPTY, deactivate_empty);
4886STAT_ATTR(DEACTIVATE_TO_HEAD, deactivate_to_head);
4887STAT_ATTR(DEACTIVATE_TO_TAIL, deactivate_to_tail);
4888STAT_ATTR(DEACTIVATE_REMOTE_FREES, deactivate_remote_frees);
Christoph Lameter03e404a2011-06-01 12:25:58 -05004889STAT_ATTR(DEACTIVATE_BYPASS, deactivate_bypass);
Christoph Lameter65c33762008-04-14 19:11:40 +03004890STAT_ATTR(ORDER_FALLBACK, order_fallback);
Christoph Lameterb789ef52011-06-01 12:25:49 -05004891STAT_ATTR(CMPXCHG_DOUBLE_CPU_FAIL, cmpxchg_double_cpu_fail);
4892STAT_ATTR(CMPXCHG_DOUBLE_FAIL, cmpxchg_double_fail);
Christoph Lameter49e22582011-08-09 16:12:27 -05004893STAT_ATTR(CPU_PARTIAL_ALLOC, cpu_partial_alloc);
4894STAT_ATTR(CPU_PARTIAL_FREE, cpu_partial_free);
Alex Shi8028dce2012-02-03 23:34:56 +08004895STAT_ATTR(CPU_PARTIAL_NODE, cpu_partial_node);
4896STAT_ATTR(CPU_PARTIAL_DRAIN, cpu_partial_drain);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004897#endif
4898
Pekka Enberg06428782008-01-07 23:20:27 -08004899static struct attribute *slab_attrs[] = {
Christoph Lameter81819f02007-05-06 14:49:36 -07004900 &slab_size_attr.attr,
4901 &object_size_attr.attr,
4902 &objs_per_slab_attr.attr,
4903 &order_attr.attr,
David Rientjes73d342b2009-02-22 17:40:09 -08004904 &min_partial_attr.attr,
Christoph Lameter49e22582011-08-09 16:12:27 -05004905 &cpu_partial_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07004906 &objects_attr.attr,
Christoph Lameter205ab992008-04-14 19:11:40 +03004907 &objects_partial_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07004908 &partial_attr.attr,
4909 &cpu_slabs_attr.attr,
4910 &ctor_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07004911 &aliases_attr.attr,
4912 &align_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07004913 &hwcache_align_attr.attr,
4914 &reclaim_account_attr.attr,
4915 &destroy_by_rcu_attr.attr,
Christoph Lametera5a84752010-10-05 13:57:27 -05004916 &shrink_attr.attr,
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08004917 &reserved_attr.attr,
Christoph Lameter49e22582011-08-09 16:12:27 -05004918 &slabs_cpu_partial_attr.attr,
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004919#ifdef CONFIG_SLUB_DEBUG
Christoph Lametera5a84752010-10-05 13:57:27 -05004920 &total_objects_attr.attr,
4921 &slabs_attr.attr,
4922 &sanity_checks_attr.attr,
4923 &trace_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07004924 &red_zone_attr.attr,
4925 &poison_attr.attr,
4926 &store_user_attr.attr,
Christoph Lameter53e15af2007-05-06 14:49:43 -07004927 &validate_attr.attr,
Christoph Lameter88a420e2007-05-06 14:49:45 -07004928 &alloc_calls_attr.attr,
4929 &free_calls_attr.attr,
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004930#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07004931#ifdef CONFIG_ZONE_DMA
4932 &cache_dma_attr.attr,
4933#endif
4934#ifdef CONFIG_NUMA
Christoph Lameter98246012008-01-07 23:20:26 -08004935 &remote_node_defrag_ratio_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07004936#endif
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004937#ifdef CONFIG_SLUB_STATS
4938 &alloc_fastpath_attr.attr,
4939 &alloc_slowpath_attr.attr,
4940 &free_fastpath_attr.attr,
4941 &free_slowpath_attr.attr,
4942 &free_frozen_attr.attr,
4943 &free_add_partial_attr.attr,
4944 &free_remove_partial_attr.attr,
4945 &alloc_from_partial_attr.attr,
4946 &alloc_slab_attr.attr,
4947 &alloc_refill_attr.attr,
Christoph Lametere36a2652011-06-01 12:25:57 -05004948 &alloc_node_mismatch_attr.attr,
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004949 &free_slab_attr.attr,
4950 &cpuslab_flush_attr.attr,
4951 &deactivate_full_attr.attr,
4952 &deactivate_empty_attr.attr,
4953 &deactivate_to_head_attr.attr,
4954 &deactivate_to_tail_attr.attr,
4955 &deactivate_remote_frees_attr.attr,
Christoph Lameter03e404a2011-06-01 12:25:58 -05004956 &deactivate_bypass_attr.attr,
Christoph Lameter65c33762008-04-14 19:11:40 +03004957 &order_fallback_attr.attr,
Christoph Lameterb789ef52011-06-01 12:25:49 -05004958 &cmpxchg_double_fail_attr.attr,
4959 &cmpxchg_double_cpu_fail_attr.attr,
Christoph Lameter49e22582011-08-09 16:12:27 -05004960 &cpu_partial_alloc_attr.attr,
4961 &cpu_partial_free_attr.attr,
Alex Shi8028dce2012-02-03 23:34:56 +08004962 &cpu_partial_node_attr.attr,
4963 &cpu_partial_drain_attr.attr,
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004964#endif
Dmitry Monakhov4c13dd32010-02-26 09:36:12 +03004965#ifdef CONFIG_FAILSLAB
4966 &failslab_attr.attr,
4967#endif
4968
Christoph Lameter81819f02007-05-06 14:49:36 -07004969 NULL
4970};
4971
4972static struct attribute_group slab_attr_group = {
4973 .attrs = slab_attrs,
4974};
4975
4976static ssize_t slab_attr_show(struct kobject *kobj,
4977 struct attribute *attr,
4978 char *buf)
4979{
4980 struct slab_attribute *attribute;
4981 struct kmem_cache *s;
4982 int err;
4983
4984 attribute = to_slab_attr(attr);
4985 s = to_slab(kobj);
4986
4987 if (!attribute->show)
4988 return -EIO;
4989
4990 err = attribute->show(s, buf);
4991
4992 return err;
4993}
4994
4995static ssize_t slab_attr_store(struct kobject *kobj,
4996 struct attribute *attr,
4997 const char *buf, size_t len)
4998{
4999 struct slab_attribute *attribute;
5000 struct kmem_cache *s;
5001 int err;
5002
5003 attribute = to_slab_attr(attr);
5004 s = to_slab(kobj);
5005
5006 if (!attribute->store)
5007 return -EIO;
5008
5009 err = attribute->store(s, buf, len);
Glauber Costa107dab52012-12-18 14:23:05 -08005010#ifdef CONFIG_MEMCG_KMEM
5011 if (slab_state >= FULL && err >= 0 && is_root_cache(s)) {
5012 int i;
Christoph Lameter81819f02007-05-06 14:49:36 -07005013
Glauber Costa107dab52012-12-18 14:23:05 -08005014 mutex_lock(&slab_mutex);
5015 if (s->max_attr_size < len)
5016 s->max_attr_size = len;
5017
Glauber Costaebe945c2012-12-18 14:23:10 -08005018 /*
5019 * This is a best effort propagation, so this function's return
5020 * value will be determined by the parent cache only. This is
5021 * basically because not all attributes will have a well
5022 * defined semantics for rollbacks - most of the actions will
5023 * have permanent effects.
5024 *
5025 * Returning the error value of any of the children that fail
5026 * is not 100 % defined, in the sense that users seeing the
5027 * error code won't be able to know anything about the state of
5028 * the cache.
5029 *
5030 * Only returning the error code for the parent cache at least
5031 * has well defined semantics. The cache being written to
5032 * directly either failed or succeeded, in which case we loop
5033 * through the descendants with best-effort propagation.
5034 */
Glauber Costa107dab52012-12-18 14:23:05 -08005035 for_each_memcg_cache_index(i) {
Qiang Huang2ade4de2013-11-12 15:08:23 -08005036 struct kmem_cache *c = cache_from_memcg_idx(s, i);
Glauber Costa107dab52012-12-18 14:23:05 -08005037 if (c)
5038 attribute->store(c, buf, len);
5039 }
5040 mutex_unlock(&slab_mutex);
5041 }
5042#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07005043 return err;
5044}
5045
Glauber Costa107dab52012-12-18 14:23:05 -08005046static void memcg_propagate_slab_attrs(struct kmem_cache *s)
5047{
5048#ifdef CONFIG_MEMCG_KMEM
5049 int i;
5050 char *buffer = NULL;
Vladimir Davydov93030d82014-05-06 12:49:59 -07005051 struct kmem_cache *root_cache;
Glauber Costa107dab52012-12-18 14:23:05 -08005052
Vladimir Davydov93030d82014-05-06 12:49:59 -07005053 if (is_root_cache(s))
Glauber Costa107dab52012-12-18 14:23:05 -08005054 return;
5055
Vladimir Davydov93030d82014-05-06 12:49:59 -07005056 root_cache = s->memcg_params->root_cache;
5057
Glauber Costa107dab52012-12-18 14:23:05 -08005058 /*
5059 * This mean this cache had no attribute written. Therefore, no point
5060 * in copying default values around
5061 */
Vladimir Davydov93030d82014-05-06 12:49:59 -07005062 if (!root_cache->max_attr_size)
Glauber Costa107dab52012-12-18 14:23:05 -08005063 return;
5064
5065 for (i = 0; i < ARRAY_SIZE(slab_attrs); i++) {
5066 char mbuf[64];
5067 char *buf;
5068 struct slab_attribute *attr = to_slab_attr(slab_attrs[i]);
5069
5070 if (!attr || !attr->store || !attr->show)
5071 continue;
5072
5073 /*
5074 * It is really bad that we have to allocate here, so we will
5075 * do it only as a fallback. If we actually allocate, though,
5076 * we can just use the allocated buffer until the end.
5077 *
5078 * Most of the slub attributes will tend to be very small in
5079 * size, but sysfs allows buffers up to a page, so they can
5080 * theoretically happen.
5081 */
5082 if (buffer)
5083 buf = buffer;
Vladimir Davydov93030d82014-05-06 12:49:59 -07005084 else if (root_cache->max_attr_size < ARRAY_SIZE(mbuf))
Glauber Costa107dab52012-12-18 14:23:05 -08005085 buf = mbuf;
5086 else {
5087 buffer = (char *) get_zeroed_page(GFP_KERNEL);
5088 if (WARN_ON(!buffer))
5089 continue;
5090 buf = buffer;
5091 }
5092
Vladimir Davydov93030d82014-05-06 12:49:59 -07005093 attr->show(root_cache, buf);
Glauber Costa107dab52012-12-18 14:23:05 -08005094 attr->store(s, buf, strlen(buf));
5095 }
5096
5097 if (buffer)
5098 free_page((unsigned long)buffer);
5099#endif
5100}
5101
Christoph Lameter41a21282014-05-06 12:50:08 -07005102static void kmem_cache_release(struct kobject *k)
5103{
5104 slab_kmem_cache_release(to_slab(k));
5105}
5106
Emese Revfy52cf25d2010-01-19 02:58:23 +01005107static const struct sysfs_ops slab_sysfs_ops = {
Christoph Lameter81819f02007-05-06 14:49:36 -07005108 .show = slab_attr_show,
5109 .store = slab_attr_store,
5110};
5111
5112static struct kobj_type slab_ktype = {
5113 .sysfs_ops = &slab_sysfs_ops,
Christoph Lameter41a21282014-05-06 12:50:08 -07005114 .release = kmem_cache_release,
Christoph Lameter81819f02007-05-06 14:49:36 -07005115};
5116
5117static int uevent_filter(struct kset *kset, struct kobject *kobj)
5118{
5119 struct kobj_type *ktype = get_ktype(kobj);
5120
5121 if (ktype == &slab_ktype)
5122 return 1;
5123 return 0;
5124}
5125
Emese Revfy9cd43612009-12-31 14:52:51 +01005126static const struct kset_uevent_ops slab_uevent_ops = {
Christoph Lameter81819f02007-05-06 14:49:36 -07005127 .filter = uevent_filter,
5128};
5129
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06005130static struct kset *slab_kset;
Christoph Lameter81819f02007-05-06 14:49:36 -07005131
Vladimir Davydov9a417072014-04-07 15:39:31 -07005132static inline struct kset *cache_kset(struct kmem_cache *s)
5133{
5134#ifdef CONFIG_MEMCG_KMEM
5135 if (!is_root_cache(s))
5136 return s->memcg_params->root_cache->memcg_kset;
5137#endif
5138 return slab_kset;
5139}
5140
Christoph Lameter81819f02007-05-06 14:49:36 -07005141#define ID_STR_LENGTH 64
5142
5143/* Create a unique string id for a slab cache:
Christoph Lameter6446faa2008-02-15 23:45:26 -08005144 *
5145 * Format :[flags-]size
Christoph Lameter81819f02007-05-06 14:49:36 -07005146 */
5147static char *create_unique_id(struct kmem_cache *s)
5148{
5149 char *name = kmalloc(ID_STR_LENGTH, GFP_KERNEL);
5150 char *p = name;
5151
5152 BUG_ON(!name);
5153
5154 *p++ = ':';
5155 /*
5156 * First flags affecting slabcache operations. We will only
5157 * get here for aliasable slabs so we do not need to support
5158 * too many flags. The flags here must cover all flags that
5159 * are matched during merging to guarantee that the id is
5160 * unique.
5161 */
5162 if (s->flags & SLAB_CACHE_DMA)
5163 *p++ = 'd';
5164 if (s->flags & SLAB_RECLAIM_ACCOUNT)
5165 *p++ = 'a';
5166 if (s->flags & SLAB_DEBUG_FREE)
5167 *p++ = 'F';
Vegard Nossum5a896d92008-04-04 00:54:48 +02005168 if (!(s->flags & SLAB_NOTRACK))
5169 *p++ = 't';
Christoph Lameter81819f02007-05-06 14:49:36 -07005170 if (p != name + 1)
5171 *p++ = '-';
5172 p += sprintf(p, "%07d", s->size);
Glauber Costa2633d7a2012-12-18 14:22:34 -08005173
5174#ifdef CONFIG_MEMCG_KMEM
5175 if (!is_root_cache(s))
Chen Gangd0e0ac92013-07-15 09:05:29 +08005176 p += sprintf(p, "-%08d",
5177 memcg_cache_id(s->memcg_params->memcg));
Glauber Costa2633d7a2012-12-18 14:22:34 -08005178#endif
5179
Christoph Lameter81819f02007-05-06 14:49:36 -07005180 BUG_ON(p > name + ID_STR_LENGTH - 1);
5181 return name;
5182}
5183
5184static int sysfs_slab_add(struct kmem_cache *s)
5185{
5186 int err;
5187 const char *name;
Christoph Lameter45530c42012-11-28 16:23:07 +00005188 int unmergeable = slab_unmergeable(s);
Christoph Lameter81819f02007-05-06 14:49:36 -07005189
Christoph Lameter81819f02007-05-06 14:49:36 -07005190 if (unmergeable) {
5191 /*
5192 * Slabcache can never be merged so we can use the name proper.
5193 * This is typically the case for debug situations. In that
5194 * case we can catch duplicate names easily.
5195 */
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06005196 sysfs_remove_link(&slab_kset->kobj, s->name);
Christoph Lameter81819f02007-05-06 14:49:36 -07005197 name = s->name;
5198 } else {
5199 /*
5200 * Create a unique name for the slab as a target
5201 * for the symlinks.
5202 */
5203 name = create_unique_id(s);
5204 }
5205
Vladimir Davydov9a417072014-04-07 15:39:31 -07005206 s->kobj.kset = cache_kset(s);
Tetsuo Handa26e4f202014-01-04 16:32:31 +09005207 err = kobject_init_and_add(&s->kobj, &slab_ktype, NULL, "%s", name);
Dave Jones54b6a732014-04-07 15:39:32 -07005208 if (err)
5209 goto out_put_kobj;
Christoph Lameter81819f02007-05-06 14:49:36 -07005210
5211 err = sysfs_create_group(&s->kobj, &slab_attr_group);
Dave Jones54b6a732014-04-07 15:39:32 -07005212 if (err)
5213 goto out_del_kobj;
Vladimir Davydov9a417072014-04-07 15:39:31 -07005214
5215#ifdef CONFIG_MEMCG_KMEM
5216 if (is_root_cache(s)) {
5217 s->memcg_kset = kset_create_and_add("cgroup", NULL, &s->kobj);
5218 if (!s->memcg_kset) {
Dave Jones54b6a732014-04-07 15:39:32 -07005219 err = -ENOMEM;
5220 goto out_del_kobj;
Vladimir Davydov9a417072014-04-07 15:39:31 -07005221 }
5222 }
5223#endif
5224
Christoph Lameter81819f02007-05-06 14:49:36 -07005225 kobject_uevent(&s->kobj, KOBJ_ADD);
5226 if (!unmergeable) {
5227 /* Setup first alias */
5228 sysfs_slab_alias(s, s->name);
Christoph Lameter81819f02007-05-06 14:49:36 -07005229 }
Dave Jones54b6a732014-04-07 15:39:32 -07005230out:
5231 if (!unmergeable)
5232 kfree(name);
5233 return err;
5234out_del_kobj:
5235 kobject_del(&s->kobj);
5236out_put_kobj:
5237 kobject_put(&s->kobj);
5238 goto out;
Christoph Lameter81819f02007-05-06 14:49:36 -07005239}
5240
Christoph Lameter41a21282014-05-06 12:50:08 -07005241void sysfs_slab_remove(struct kmem_cache *s)
Christoph Lameter81819f02007-05-06 14:49:36 -07005242{
Christoph Lameter97d06602012-07-06 15:25:11 -05005243 if (slab_state < FULL)
Christoph Lameter2bce6482010-07-19 11:39:11 -05005244 /*
5245 * Sysfs has not been setup yet so no need to remove the
5246 * cache from sysfs.
5247 */
5248 return;
5249
Vladimir Davydov9a417072014-04-07 15:39:31 -07005250#ifdef CONFIG_MEMCG_KMEM
5251 kset_unregister(s->memcg_kset);
5252#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07005253 kobject_uevent(&s->kobj, KOBJ_REMOVE);
5254 kobject_del(&s->kobj);
Christoph Lameter151c6022008-01-07 22:29:05 -08005255 kobject_put(&s->kobj);
Christoph Lameter81819f02007-05-06 14:49:36 -07005256}
5257
5258/*
5259 * Need to buffer aliases during bootup until sysfs becomes
Nick Andrew9f6c708e2008-12-05 14:08:08 +11005260 * available lest we lose that information.
Christoph Lameter81819f02007-05-06 14:49:36 -07005261 */
5262struct saved_alias {
5263 struct kmem_cache *s;
5264 const char *name;
5265 struct saved_alias *next;
5266};
5267
Adrian Bunk5af328a2007-07-17 04:03:27 -07005268static struct saved_alias *alias_list;
Christoph Lameter81819f02007-05-06 14:49:36 -07005269
5270static int sysfs_slab_alias(struct kmem_cache *s, const char *name)
5271{
5272 struct saved_alias *al;
5273
Christoph Lameter97d06602012-07-06 15:25:11 -05005274 if (slab_state == FULL) {
Christoph Lameter81819f02007-05-06 14:49:36 -07005275 /*
5276 * If we have a leftover link then remove it.
5277 */
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06005278 sysfs_remove_link(&slab_kset->kobj, name);
5279 return sysfs_create_link(&slab_kset->kobj, &s->kobj, name);
Christoph Lameter81819f02007-05-06 14:49:36 -07005280 }
5281
5282 al = kmalloc(sizeof(struct saved_alias), GFP_KERNEL);
5283 if (!al)
5284 return -ENOMEM;
5285
5286 al->s = s;
5287 al->name = name;
5288 al->next = alias_list;
5289 alias_list = al;
5290 return 0;
5291}
5292
5293static int __init slab_sysfs_init(void)
5294{
Christoph Lameter5b95a4ac2007-07-17 04:03:19 -07005295 struct kmem_cache *s;
Christoph Lameter81819f02007-05-06 14:49:36 -07005296 int err;
5297
Christoph Lameter18004c52012-07-06 15:25:12 -05005298 mutex_lock(&slab_mutex);
Christoph Lameter2bce6482010-07-19 11:39:11 -05005299
Greg Kroah-Hartman0ff21e42007-11-06 10:36:58 -08005300 slab_kset = kset_create_and_add("slab", &slab_uevent_ops, kernel_kobj);
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06005301 if (!slab_kset) {
Christoph Lameter18004c52012-07-06 15:25:12 -05005302 mutex_unlock(&slab_mutex);
Fabian Frederickf9f58282014-06-04 16:06:34 -07005303 pr_err("Cannot register slab subsystem.\n");
Christoph Lameter81819f02007-05-06 14:49:36 -07005304 return -ENOSYS;
5305 }
5306
Christoph Lameter97d06602012-07-06 15:25:11 -05005307 slab_state = FULL;
Christoph Lameter26a7bd02007-05-09 02:32:39 -07005308
Christoph Lameter5b95a4ac2007-07-17 04:03:19 -07005309 list_for_each_entry(s, &slab_caches, list) {
Christoph Lameter26a7bd02007-05-09 02:32:39 -07005310 err = sysfs_slab_add(s);
Christoph Lameter5d540fb2007-08-30 23:56:26 -07005311 if (err)
Fabian Frederickf9f58282014-06-04 16:06:34 -07005312 pr_err("SLUB: Unable to add boot slab %s to sysfs\n",
5313 s->name);
Christoph Lameter26a7bd02007-05-09 02:32:39 -07005314 }
Christoph Lameter81819f02007-05-06 14:49:36 -07005315
5316 while (alias_list) {
5317 struct saved_alias *al = alias_list;
5318
5319 alias_list = alias_list->next;
5320 err = sysfs_slab_alias(al->s, al->name);
Christoph Lameter5d540fb2007-08-30 23:56:26 -07005321 if (err)
Fabian Frederickf9f58282014-06-04 16:06:34 -07005322 pr_err("SLUB: Unable to add boot slab alias %s to sysfs\n",
5323 al->name);
Christoph Lameter81819f02007-05-06 14:49:36 -07005324 kfree(al);
5325 }
5326
Christoph Lameter18004c52012-07-06 15:25:12 -05005327 mutex_unlock(&slab_mutex);
Christoph Lameter81819f02007-05-06 14:49:36 -07005328 resiliency_test();
5329 return 0;
5330}
5331
5332__initcall(slab_sysfs_init);
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05005333#endif /* CONFIG_SYSFS */
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01005334
5335/*
5336 * The /proc/slabinfo ABI
5337 */
Linus Torvalds158a9622008-01-02 13:04:48 -08005338#ifdef CONFIG_SLABINFO
Glauber Costa0d7561c2012-10-19 18:20:27 +04005339void get_slabinfo(struct kmem_cache *s, struct slabinfo *sinfo)
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01005340{
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01005341 unsigned long nr_slabs = 0;
Christoph Lameter205ab992008-04-14 19:11:40 +03005342 unsigned long nr_objs = 0;
5343 unsigned long nr_free = 0;
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01005344 int node;
5345
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01005346 for_each_online_node(node) {
5347 struct kmem_cache_node *n = get_node(s, node);
5348
5349 if (!n)
5350 continue;
5351
Wanpeng Lic17fd132013-07-04 08:33:26 +08005352 nr_slabs += node_nr_slabs(n);
5353 nr_objs += node_nr_objs(n);
Christoph Lameter205ab992008-04-14 19:11:40 +03005354 nr_free += count_partial(n, count_free);
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01005355 }
5356
Glauber Costa0d7561c2012-10-19 18:20:27 +04005357 sinfo->active_objs = nr_objs - nr_free;
5358 sinfo->num_objs = nr_objs;
5359 sinfo->active_slabs = nr_slabs;
5360 sinfo->num_slabs = nr_slabs;
5361 sinfo->objects_per_slab = oo_objects(s->oo);
5362 sinfo->cache_order = oo_order(s->oo);
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01005363}
5364
Glauber Costa0d7561c2012-10-19 18:20:27 +04005365void slabinfo_show_stats(struct seq_file *m, struct kmem_cache *s)
Alexey Dobriyan7b3c3a52008-10-06 02:42:17 +04005366{
Alexey Dobriyan7b3c3a52008-10-06 02:42:17 +04005367}
5368
Glauber Costab7454ad2012-10-19 18:20:25 +04005369ssize_t slabinfo_write(struct file *file, const char __user *buffer,
5370 size_t count, loff_t *ppos)
Alexey Dobriyan7b3c3a52008-10-06 02:42:17 +04005371{
Glauber Costab7454ad2012-10-19 18:20:25 +04005372 return -EIO;
Alexey Dobriyan7b3c3a52008-10-06 02:42:17 +04005373}
Linus Torvalds158a9622008-01-02 13:04:48 -08005374#endif /* CONFIG_SLABINFO */