blob: 0fdd6c1e1f82cf84dc909872e5b6ee3368dee7e4 [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>
Andrey Ryabinina79316c2015-02-13 14:39:38 -080023#include <linux/kasan.h>
Vegard Nossum5a896d92008-04-04 00:54:48 +020024#include <linux/kmemcheck.h>
Christoph Lameter81819f02007-05-06 14:49:36 -070025#include <linux/cpu.h>
26#include <linux/cpuset.h>
27#include <linux/mempolicy.h>
28#include <linux/ctype.h>
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -070029#include <linux/debugobjects.h>
Christoph Lameter81819f02007-05-06 14:49:36 -070030#include <linux/kallsyms.h>
Yasunori Gotob9049e22007-10-21 16:41:37 -070031#include <linux/memory.h>
Roman Zippelf8bd2252008-05-01 04:34:31 -070032#include <linux/math64.h>
Akinobu Mita773ff602008-12-23 19:37:01 +090033#include <linux/fault-inject.h>
Pekka Enbergbfa71452011-07-07 22:47:01 +030034#include <linux/stacktrace.h>
Christoph Lameter4de900b2012-01-30 15:53:51 -060035#include <linux/prefetch.h>
Glauber Costa2633d7a2012-12-18 14:22:34 -080036#include <linux/memcontrol.h>
Christoph Lameter81819f02007-05-06 14:49:36 -070037
Richard Kennedy4a923792010-10-21 10:29:19 +010038#include <trace/events/kmem.h>
39
Mel Gorman072bb0a2012-07-31 16:43:58 -070040#include "internal.h"
41
Christoph Lameter81819f02007-05-06 14:49:36 -070042/*
43 * Lock order:
Christoph Lameter18004c52012-07-06 15:25:12 -050044 * 1. slab_mutex (Global Mutex)
Christoph Lameter881db7f2011-06-01 12:25:53 -050045 * 2. node->list_lock
46 * 3. slab_lock(page) (Only on some arches and for debugging)
Christoph Lameter81819f02007-05-06 14:49:36 -070047 *
Christoph Lameter18004c52012-07-06 15:25:12 -050048 * slab_mutex
Christoph Lameter881db7f2011-06-01 12:25:53 -050049 *
Christoph Lameter18004c52012-07-06 15:25:12 -050050 * The role of the slab_mutex is to protect the list of all the slabs
Christoph Lameter881db7f2011-06-01 12:25:53 -050051 * and to synchronize major metadata changes to slab cache structures.
52 *
53 * The slab_lock is only used for debugging and on arches that do not
54 * have the ability to do a cmpxchg_double. It only protects the second
55 * double word in the page struct. Meaning
56 * A. page->freelist -> List of object free in a page
57 * B. page->counters -> Counters of objects
58 * C. page->frozen -> frozen state
59 *
60 * If a slab is frozen then it is exempt from list management. It is not
61 * on any list. The processor that froze the slab is the one who can
62 * perform list operations on the page. Other processors may put objects
63 * onto the freelist but the processor that froze the slab is the only
64 * one that can retrieve the objects from the page's freelist.
Christoph Lameter81819f02007-05-06 14:49:36 -070065 *
66 * The list_lock protects the partial and full list on each node and
67 * the partial slab counter. If taken then no new slabs may be added or
68 * removed from the lists nor make the number of partial slabs be modified.
69 * (Note that the total number of slabs is an atomic value that may be
70 * modified without taking the list lock).
71 *
72 * The list_lock is a centralized lock and thus we avoid taking it as
73 * much as possible. As long as SLUB does not have to handle partial
74 * slabs, operations can continue without any centralized lock. F.e.
75 * allocating a long series of objects that fill up slabs does not require
76 * the list lock.
Christoph Lameter81819f02007-05-06 14:49:36 -070077 * Interrupts are disabled during allocation and deallocation in order to
78 * make the slab allocator safe to use in the context of an irq. In addition
79 * interrupts are disabled to ensure that the processor does not change
80 * while handling per_cpu slabs, due to kernel preemption.
81 *
82 * SLUB assigns one slab for allocation to each processor.
83 * Allocations only occur from these slabs called cpu slabs.
84 *
Christoph Lameter672bba32007-05-09 02:32:39 -070085 * Slabs with free elements are kept on a partial list and during regular
86 * operations no list for full slabs is used. If an object in a full slab is
Christoph Lameter81819f02007-05-06 14:49:36 -070087 * freed then the slab will show up again on the partial lists.
Christoph Lameter672bba32007-05-09 02:32:39 -070088 * We track full slabs for debugging purposes though because otherwise we
89 * cannot scan all objects.
Christoph Lameter81819f02007-05-06 14:49:36 -070090 *
91 * Slabs are freed when they become empty. Teardown and setup is
92 * minimal so we rely on the page allocators per cpu caches for
93 * fast frees and allocs.
94 *
95 * Overloading of page flags that are otherwise used for LRU management.
96 *
Christoph Lameter4b6f0752007-05-16 22:10:53 -070097 * PageActive The slab is frozen and exempt from list processing.
98 * This means that the slab is dedicated to a purpose
99 * such as satisfying allocations for a specific
100 * processor. Objects may be freed in the slab while
101 * it is frozen but slab_free will then skip the usual
102 * list operations. It is up to the processor holding
103 * the slab to integrate the slab into the slab lists
104 * when the slab is no longer needed.
105 *
106 * One use of this flag is to mark slabs that are
107 * used for allocations. Then such a slab becomes a cpu
108 * slab. The cpu slab may be equipped with an additional
Christoph Lameterdfb4f092007-10-16 01:26:05 -0700109 * freelist that allows lockless access to
Christoph Lameter894b8782007-05-10 03:15:16 -0700110 * free objects in addition to the regular freelist
111 * that requires the slab lock.
Christoph Lameter81819f02007-05-06 14:49:36 -0700112 *
113 * PageError Slab requires special handling due to debug
114 * options set. This moves slab handling out of
Christoph Lameter894b8782007-05-10 03:15:16 -0700115 * the fast path and disables lockless freelists.
Christoph Lameter81819f02007-05-06 14:49:36 -0700116 */
117
Christoph Lameteraf537b02010-07-09 14:07:14 -0500118static inline int kmem_cache_debug(struct kmem_cache *s)
119{
Christoph Lameter5577bd82007-05-16 22:10:56 -0700120#ifdef CONFIG_SLUB_DEBUG
Christoph Lameteraf537b02010-07-09 14:07:14 -0500121 return unlikely(s->flags & SLAB_DEBUG_FLAGS);
Christoph Lameter5577bd82007-05-16 22:10:56 -0700122#else
Christoph Lameteraf537b02010-07-09 14:07:14 -0500123 return 0;
Christoph Lameter5577bd82007-05-16 22:10:56 -0700124#endif
Christoph Lameteraf537b02010-07-09 14:07:14 -0500125}
Christoph Lameter5577bd82007-05-16 22:10:56 -0700126
Joonsoo Kim345c9052013-06-19 14:05:52 +0900127static inline bool kmem_cache_has_cpu_partial(struct kmem_cache *s)
128{
129#ifdef CONFIG_SLUB_CPU_PARTIAL
130 return !kmem_cache_debug(s);
131#else
132 return false;
133#endif
134}
135
Christoph Lameter81819f02007-05-06 14:49:36 -0700136/*
137 * Issues still to be resolved:
138 *
Christoph Lameter81819f02007-05-06 14:49:36 -0700139 * - Support PAGE_ALLOC_DEBUG. Should be easy to do.
140 *
Christoph Lameter81819f02007-05-06 14:49:36 -0700141 * - Variable sizing of the per node arrays
142 */
143
144/* Enable to test recovery from slab corruption on boot */
145#undef SLUB_RESILIENCY_TEST
146
Christoph Lameterb789ef52011-06-01 12:25:49 -0500147/* Enable to log cmpxchg failures */
148#undef SLUB_DEBUG_CMPXCHG
149
Christoph Lameter81819f02007-05-06 14:49:36 -0700150/*
Christoph Lameter2086d262007-05-06 14:49:46 -0700151 * Mininum number of partial slabs. These will be left on the partial
152 * lists even if they are empty. kmem_cache_shrink may reclaim them.
153 */
Christoph Lameter76be8952007-12-21 14:37:37 -0800154#define MIN_PARTIAL 5
Christoph Lametere95eed52007-05-06 14:49:44 -0700155
Christoph Lameter2086d262007-05-06 14:49:46 -0700156/*
157 * Maximum number of desirable partial slabs.
158 * The existence of more partial slabs makes kmem_cache_shrink
Zhi Yong Wu721ae222013-11-08 20:47:37 +0800159 * sort the partial list by the number of objects in use.
Christoph Lameter2086d262007-05-06 14:49:46 -0700160 */
161#define MAX_PARTIAL 10
162
Christoph Lameter81819f02007-05-06 14:49:36 -0700163#define DEBUG_DEFAULT_FLAGS (SLAB_DEBUG_FREE | SLAB_RED_ZONE | \
164 SLAB_POISON | SLAB_STORE_USER)
Christoph Lameter672bba32007-05-09 02:32:39 -0700165
Christoph Lameter81819f02007-05-06 14:49:36 -0700166/*
David Rientjes3de47212009-07-27 18:30:35 -0700167 * Debugging flags that require metadata to be stored in the slab. These get
168 * disabled when slub_debug=O is used and a cache's min order increases with
169 * metadata.
David Rientjesfa5ec8a2009-07-07 00:14:14 -0700170 */
David Rientjes3de47212009-07-27 18:30:35 -0700171#define DEBUG_METADATA_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER)
David Rientjesfa5ec8a2009-07-07 00:14:14 -0700172
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +0400173#define OO_SHIFT 16
174#define OO_MASK ((1 << OO_SHIFT) - 1)
Christoph Lameter50d5c412011-06-01 12:25:45 -0500175#define MAX_OBJS_PER_PAGE 32767 /* since page.objects is u15 */
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +0400176
Christoph Lameter81819f02007-05-06 14:49:36 -0700177/* Internal SLUB flags */
Christoph Lameterf90ec392010-07-09 14:07:11 -0500178#define __OBJECT_POISON 0x80000000UL /* Poison object */
Christoph Lameterb789ef52011-06-01 12:25:49 -0500179#define __CMPXCHG_DOUBLE 0x40000000UL /* Use cmpxchg_double */
Christoph Lameter81819f02007-05-06 14:49:36 -0700180
Christoph Lameter81819f02007-05-06 14:49:36 -0700181#ifdef CONFIG_SMP
182static struct notifier_block slab_notifier;
183#endif
184
Christoph Lameter02cbc872007-05-09 02:32:43 -0700185/*
186 * Tracking user of a slab.
187 */
Ben Greeard6543e32011-07-07 11:36:36 -0700188#define TRACK_ADDRS_COUNT 16
Christoph Lameter02cbc872007-05-09 02:32:43 -0700189struct track {
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300190 unsigned long addr; /* Called from address */
Ben Greeard6543e32011-07-07 11:36:36 -0700191#ifdef CONFIG_STACKTRACE
192 unsigned long addrs[TRACK_ADDRS_COUNT]; /* Called from address */
193#endif
Christoph Lameter02cbc872007-05-09 02:32:43 -0700194 int cpu; /* Was running on cpu */
195 int pid; /* Pid context */
196 unsigned long when; /* When did the operation occur */
197};
198
199enum track_item { TRACK_ALLOC, TRACK_FREE };
200
Christoph Lameterab4d5ed2010-10-05 13:57:26 -0500201#ifdef CONFIG_SYSFS
Christoph Lameter81819f02007-05-06 14:49:36 -0700202static int sysfs_slab_add(struct kmem_cache *);
203static int sysfs_slab_alias(struct kmem_cache *, const char *);
Glauber Costa107dab52012-12-18 14:23:05 -0800204static void memcg_propagate_slab_attrs(struct kmem_cache *s);
Christoph Lameter81819f02007-05-06 14:49:36 -0700205#else
Christoph Lameter0c710012007-07-17 04:03:24 -0700206static inline int sysfs_slab_add(struct kmem_cache *s) { return 0; }
207static inline int sysfs_slab_alias(struct kmem_cache *s, const char *p)
208 { return 0; }
Glauber Costa107dab52012-12-18 14:23:05 -0800209static inline void memcg_propagate_slab_attrs(struct kmem_cache *s) { }
Christoph Lameter81819f02007-05-06 14:49:36 -0700210#endif
211
Christoph Lameter4fdccdf2011-03-22 13:35:00 -0500212static inline void stat(const struct kmem_cache *s, enum stat_item si)
Christoph Lameter8ff12cf2008-02-07 17:47:41 -0800213{
214#ifdef CONFIG_SLUB_STATS
Christoph Lameter88da03a2014-04-07 15:39:42 -0700215 /*
216 * The rmw is racy on a preemptible kernel but this is acceptable, so
217 * avoid this_cpu_add()'s irq-disable overhead.
218 */
219 raw_cpu_inc(s->cpu_slab->stat[si]);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -0800220#endif
221}
222
Christoph Lameter81819f02007-05-06 14:49:36 -0700223/********************************************************************
224 * Core slab cache functions
225 *******************************************************************/
226
Christoph Lameter6446faa2008-02-15 23:45:26 -0800227/* Verify that a pointer has an address that is valid within a slab page */
Christoph Lameter02cbc872007-05-09 02:32:43 -0700228static inline int check_valid_pointer(struct kmem_cache *s,
229 struct page *page, const void *object)
230{
231 void *base;
232
Christoph Lametera973e9d2008-03-01 13:40:44 -0800233 if (!object)
Christoph Lameter02cbc872007-05-09 02:32:43 -0700234 return 1;
235
Christoph Lametera973e9d2008-03-01 13:40:44 -0800236 base = page_address(page);
Christoph Lameter39b26462008-04-14 19:11:30 +0300237 if (object < base || object >= base + page->objects * s->size ||
Christoph Lameter02cbc872007-05-09 02:32:43 -0700238 (object - base) % s->size) {
239 return 0;
240 }
241
242 return 1;
243}
244
Christoph Lameter7656c722007-05-09 02:32:40 -0700245static inline void *get_freepointer(struct kmem_cache *s, void *object)
246{
247 return *(void **)(object + s->offset);
248}
249
Eric Dumazet0ad95002011-12-16 16:25:34 +0100250static void prefetch_freepointer(const struct kmem_cache *s, void *object)
251{
252 prefetch(object + s->offset);
253}
254
Christoph Lameter1393d9a2011-05-16 15:26:08 -0500255static inline void *get_freepointer_safe(struct kmem_cache *s, void *object)
256{
257 void *p;
258
259#ifdef CONFIG_DEBUG_PAGEALLOC
260 probe_kernel_read(&p, (void **)(object + s->offset), sizeof(p));
261#else
262 p = get_freepointer(s, object);
263#endif
264 return p;
265}
266
Christoph Lameter7656c722007-05-09 02:32:40 -0700267static inline void set_freepointer(struct kmem_cache *s, void *object, void *fp)
268{
269 *(void **)(object + s->offset) = fp;
270}
271
272/* Loop over all objects in a slab */
Christoph Lameter224a88b2008-04-14 19:11:31 +0300273#define for_each_object(__p, __s, __addr, __objects) \
274 for (__p = (__addr); __p < (__addr) + (__objects) * (__s)->size;\
Christoph Lameter7656c722007-05-09 02:32:40 -0700275 __p += (__s)->size)
276
Wei Yang54266642014-08-06 16:04:42 -0700277#define for_each_object_idx(__p, __idx, __s, __addr, __objects) \
278 for (__p = (__addr), __idx = 1; __idx <= __objects;\
279 __p += (__s)->size, __idx++)
280
Christoph Lameter7656c722007-05-09 02:32:40 -0700281/* Determine object index from a given position */
282static inline int slab_index(void *p, struct kmem_cache *s, void *addr)
283{
284 return (p - addr) / s->size;
285}
286
Mariusz Kozlowskid71f6062011-02-26 20:10:26 +0100287static inline size_t slab_ksize(const struct kmem_cache *s)
288{
289#ifdef CONFIG_SLUB_DEBUG
290 /*
291 * Debugging requires use of the padding between object
292 * and whatever may come after it.
293 */
294 if (s->flags & (SLAB_RED_ZONE | SLAB_POISON))
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500295 return s->object_size;
Mariusz Kozlowskid71f6062011-02-26 20:10:26 +0100296
297#endif
298 /*
299 * If we have the need to store the freelist pointer
300 * back there or track user information then we can
301 * only use the space before that information.
302 */
303 if (s->flags & (SLAB_DESTROY_BY_RCU | SLAB_STORE_USER))
304 return s->inuse;
305 /*
306 * Else we can use all the padding etc for the allocation
307 */
308 return s->size;
309}
310
Lai Jiangshanab9a0f12011-03-10 15:21:48 +0800311static inline int order_objects(int order, unsigned long size, int reserved)
312{
313 return ((PAGE_SIZE << order) - reserved) / size;
314}
315
Christoph Lameter834f3d12008-04-14 19:11:31 +0300316static inline struct kmem_cache_order_objects oo_make(int order,
Lai Jiangshanab9a0f12011-03-10 15:21:48 +0800317 unsigned long size, int reserved)
Christoph Lameter834f3d12008-04-14 19:11:31 +0300318{
319 struct kmem_cache_order_objects x = {
Lai Jiangshanab9a0f12011-03-10 15:21:48 +0800320 (order << OO_SHIFT) + order_objects(order, size, reserved)
Christoph Lameter834f3d12008-04-14 19:11:31 +0300321 };
322
323 return x;
324}
325
326static inline int oo_order(struct kmem_cache_order_objects x)
327{
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +0400328 return x.x >> OO_SHIFT;
Christoph Lameter834f3d12008-04-14 19:11:31 +0300329}
330
331static inline int oo_objects(struct kmem_cache_order_objects x)
332{
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +0400333 return x.x & OO_MASK;
Christoph Lameter834f3d12008-04-14 19:11:31 +0300334}
335
Christoph Lameter881db7f2011-06-01 12:25:53 -0500336/*
337 * Per slab locking using the pagelock
338 */
339static __always_inline void slab_lock(struct page *page)
340{
341 bit_spin_lock(PG_locked, &page->flags);
342}
343
344static __always_inline void slab_unlock(struct page *page)
345{
346 __bit_spin_unlock(PG_locked, &page->flags);
347}
348
Dave Hansena0320862014-01-30 15:46:09 -0800349static inline void set_page_slub_counters(struct page *page, unsigned long counters_new)
350{
351 struct page tmp;
352 tmp.counters = counters_new;
353 /*
354 * page->counters can cover frozen/inuse/objects as well
355 * as page->_count. If we assign to ->counters directly
356 * we run the risk of losing updates to page->_count, so
357 * be careful and only assign to the fields we need.
358 */
359 page->frozen = tmp.frozen;
360 page->inuse = tmp.inuse;
361 page->objects = tmp.objects;
362}
363
Christoph Lameter1d071712011-07-14 12:49:12 -0500364/* Interrupts must be disabled (for the fallback code to work right) */
365static inline bool __cmpxchg_double_slab(struct kmem_cache *s, struct page *page,
Christoph Lameterb789ef52011-06-01 12:25:49 -0500366 void *freelist_old, unsigned long counters_old,
367 void *freelist_new, unsigned long counters_new,
368 const char *n)
369{
Christoph Lameter1d071712011-07-14 12:49:12 -0500370 VM_BUG_ON(!irqs_disabled());
Heiko Carstens25654092012-01-12 17:17:33 -0800371#if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \
372 defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
Christoph Lameterb789ef52011-06-01 12:25:49 -0500373 if (s->flags & __CMPXCHG_DOUBLE) {
Jan Beulichcdcd6292012-01-02 17:02:18 +0000374 if (cmpxchg_double(&page->freelist, &page->counters,
Dan Carpenter0aa9a132014-08-06 16:04:48 -0700375 freelist_old, counters_old,
376 freelist_new, counters_new))
Joe Perches6f6528a2015-04-14 15:44:31 -0700377 return true;
Christoph Lameterb789ef52011-06-01 12:25:49 -0500378 } else
379#endif
380 {
Christoph Lameter881db7f2011-06-01 12:25:53 -0500381 slab_lock(page);
Chen Gangd0e0ac92013-07-15 09:05:29 +0800382 if (page->freelist == freelist_old &&
383 page->counters == counters_old) {
Christoph Lameterb789ef52011-06-01 12:25:49 -0500384 page->freelist = freelist_new;
Dave Hansena0320862014-01-30 15:46:09 -0800385 set_page_slub_counters(page, counters_new);
Christoph Lameter881db7f2011-06-01 12:25:53 -0500386 slab_unlock(page);
Joe Perches6f6528a2015-04-14 15:44:31 -0700387 return true;
Christoph Lameterb789ef52011-06-01 12:25:49 -0500388 }
Christoph Lameter881db7f2011-06-01 12:25:53 -0500389 slab_unlock(page);
Christoph Lameterb789ef52011-06-01 12:25:49 -0500390 }
391
392 cpu_relax();
393 stat(s, CMPXCHG_DOUBLE_FAIL);
394
395#ifdef SLUB_DEBUG_CMPXCHG
Fabian Frederickf9f58282014-06-04 16:06:34 -0700396 pr_info("%s %s: cmpxchg double redo ", n, s->name);
Christoph Lameterb789ef52011-06-01 12:25:49 -0500397#endif
398
Joe Perches6f6528a2015-04-14 15:44:31 -0700399 return false;
Christoph Lameterb789ef52011-06-01 12:25:49 -0500400}
401
Christoph Lameter1d071712011-07-14 12:49:12 -0500402static inline bool cmpxchg_double_slab(struct kmem_cache *s, struct page *page,
403 void *freelist_old, unsigned long counters_old,
404 void *freelist_new, unsigned long counters_new,
405 const char *n)
406{
Heiko Carstens25654092012-01-12 17:17:33 -0800407#if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \
408 defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
Christoph Lameter1d071712011-07-14 12:49:12 -0500409 if (s->flags & __CMPXCHG_DOUBLE) {
Jan Beulichcdcd6292012-01-02 17:02:18 +0000410 if (cmpxchg_double(&page->freelist, &page->counters,
Dan Carpenter0aa9a132014-08-06 16:04:48 -0700411 freelist_old, counters_old,
412 freelist_new, counters_new))
Joe Perches6f6528a2015-04-14 15:44:31 -0700413 return true;
Christoph Lameter1d071712011-07-14 12:49:12 -0500414 } else
415#endif
416 {
417 unsigned long flags;
418
419 local_irq_save(flags);
420 slab_lock(page);
Chen Gangd0e0ac92013-07-15 09:05:29 +0800421 if (page->freelist == freelist_old &&
422 page->counters == counters_old) {
Christoph Lameter1d071712011-07-14 12:49:12 -0500423 page->freelist = freelist_new;
Dave Hansena0320862014-01-30 15:46:09 -0800424 set_page_slub_counters(page, counters_new);
Christoph Lameter1d071712011-07-14 12:49:12 -0500425 slab_unlock(page);
426 local_irq_restore(flags);
Joe Perches6f6528a2015-04-14 15:44:31 -0700427 return true;
Christoph Lameter1d071712011-07-14 12:49:12 -0500428 }
429 slab_unlock(page);
430 local_irq_restore(flags);
431 }
432
433 cpu_relax();
434 stat(s, CMPXCHG_DOUBLE_FAIL);
435
436#ifdef SLUB_DEBUG_CMPXCHG
Fabian Frederickf9f58282014-06-04 16:06:34 -0700437 pr_info("%s %s: cmpxchg double redo ", n, s->name);
Christoph Lameter1d071712011-07-14 12:49:12 -0500438#endif
439
Joe Perches6f6528a2015-04-14 15:44:31 -0700440 return false;
Christoph Lameter1d071712011-07-14 12:49:12 -0500441}
442
Christoph Lameter41ecc552007-05-09 02:32:44 -0700443#ifdef CONFIG_SLUB_DEBUG
444/*
Christoph Lameter5f80b132011-04-15 14:48:13 -0500445 * Determine a map of object in use on a page.
446 *
Christoph Lameter881db7f2011-06-01 12:25:53 -0500447 * Node listlock must be held to guarantee that the page does
Christoph Lameter5f80b132011-04-15 14:48:13 -0500448 * not vanish from under us.
449 */
450static void get_map(struct kmem_cache *s, struct page *page, unsigned long *map)
451{
452 void *p;
453 void *addr = page_address(page);
454
455 for (p = page->freelist; p; p = get_freepointer(s, p))
456 set_bit(slab_index(p, s, addr), map);
457}
458
Christoph Lameter41ecc552007-05-09 02:32:44 -0700459/*
460 * Debug settings:
461 */
Christoph Lameterf0630ff2007-07-15 23:38:14 -0700462#ifdef CONFIG_SLUB_DEBUG_ON
463static int slub_debug = DEBUG_DEFAULT_FLAGS;
464#else
Christoph Lameter41ecc552007-05-09 02:32:44 -0700465static int slub_debug;
Christoph Lameterf0630ff2007-07-15 23:38:14 -0700466#endif
Christoph Lameter41ecc552007-05-09 02:32:44 -0700467
468static char *slub_debug_slabs;
David Rientjesfa5ec8a2009-07-07 00:14:14 -0700469static int disable_higher_order_debug;
Christoph Lameter41ecc552007-05-09 02:32:44 -0700470
Christoph Lameter7656c722007-05-09 02:32:40 -0700471/*
Andrey Ryabinina79316c2015-02-13 14:39:38 -0800472 * slub is about to manipulate internal object metadata. This memory lies
473 * outside the range of the allocated object, so accessing it would normally
474 * be reported by kasan as a bounds error. metadata_access_enable() is used
475 * to tell kasan that these accesses are OK.
476 */
477static inline void metadata_access_enable(void)
478{
479 kasan_disable_current();
480}
481
482static inline void metadata_access_disable(void)
483{
484 kasan_enable_current();
485}
486
487/*
Christoph Lameter81819f02007-05-06 14:49:36 -0700488 * Object debugging
489 */
490static void print_section(char *text, u8 *addr, unsigned int length)
491{
Andrey Ryabinina79316c2015-02-13 14:39:38 -0800492 metadata_access_enable();
Sebastian Andrzej Siewiorffc79d22011-07-29 14:10:20 +0200493 print_hex_dump(KERN_ERR, text, DUMP_PREFIX_ADDRESS, 16, 1, addr,
494 length, 1);
Andrey Ryabinina79316c2015-02-13 14:39:38 -0800495 metadata_access_disable();
Christoph Lameter81819f02007-05-06 14:49:36 -0700496}
497
Christoph Lameter81819f02007-05-06 14:49:36 -0700498static struct track *get_track(struct kmem_cache *s, void *object,
499 enum track_item alloc)
500{
501 struct track *p;
502
503 if (s->offset)
504 p = object + s->offset + sizeof(void *);
505 else
506 p = object + s->inuse;
507
508 return p + alloc;
509}
510
511static void set_track(struct kmem_cache *s, void *object,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300512 enum track_item alloc, unsigned long addr)
Christoph Lameter81819f02007-05-06 14:49:36 -0700513{
Akinobu Mita1a00df42009-03-07 00:36:21 +0900514 struct track *p = get_track(s, object, alloc);
Christoph Lameter81819f02007-05-06 14:49:36 -0700515
Christoph Lameter81819f02007-05-06 14:49:36 -0700516 if (addr) {
Ben Greeard6543e32011-07-07 11:36:36 -0700517#ifdef CONFIG_STACKTRACE
518 struct stack_trace trace;
519 int i;
520
521 trace.nr_entries = 0;
522 trace.max_entries = TRACK_ADDRS_COUNT;
523 trace.entries = p->addrs;
524 trace.skip = 3;
Andrey Ryabinina79316c2015-02-13 14:39:38 -0800525 metadata_access_enable();
Ben Greeard6543e32011-07-07 11:36:36 -0700526 save_stack_trace(&trace);
Andrey Ryabinina79316c2015-02-13 14:39:38 -0800527 metadata_access_disable();
Ben Greeard6543e32011-07-07 11:36:36 -0700528
529 /* See rant in lockdep.c */
530 if (trace.nr_entries != 0 &&
531 trace.entries[trace.nr_entries - 1] == ULONG_MAX)
532 trace.nr_entries--;
533
534 for (i = trace.nr_entries; i < TRACK_ADDRS_COUNT; i++)
535 p->addrs[i] = 0;
536#endif
Christoph Lameter81819f02007-05-06 14:49:36 -0700537 p->addr = addr;
538 p->cpu = smp_processor_id();
Alexey Dobriyan88e4ccf2008-06-23 02:58:37 +0400539 p->pid = current->pid;
Christoph Lameter81819f02007-05-06 14:49:36 -0700540 p->when = jiffies;
541 } else
542 memset(p, 0, sizeof(struct track));
543}
544
Christoph Lameter81819f02007-05-06 14:49:36 -0700545static void init_tracking(struct kmem_cache *s, void *object)
546{
Christoph Lameter24922682007-07-17 04:03:18 -0700547 if (!(s->flags & SLAB_STORE_USER))
548 return;
549
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300550 set_track(s, object, TRACK_FREE, 0UL);
551 set_track(s, object, TRACK_ALLOC, 0UL);
Christoph Lameter81819f02007-05-06 14:49:36 -0700552}
553
554static void print_track(const char *s, struct track *t)
555{
556 if (!t->addr)
557 return;
558
Fabian Frederickf9f58282014-06-04 16:06:34 -0700559 pr_err("INFO: %s in %pS age=%lu cpu=%u pid=%d\n",
560 s, (void *)t->addr, jiffies - t->when, t->cpu, t->pid);
Ben Greeard6543e32011-07-07 11:36:36 -0700561#ifdef CONFIG_STACKTRACE
562 {
563 int i;
564 for (i = 0; i < TRACK_ADDRS_COUNT; i++)
565 if (t->addrs[i])
Fabian Frederickf9f58282014-06-04 16:06:34 -0700566 pr_err("\t%pS\n", (void *)t->addrs[i]);
Ben Greeard6543e32011-07-07 11:36:36 -0700567 else
568 break;
569 }
570#endif
Christoph Lameter81819f02007-05-06 14:49:36 -0700571}
572
Christoph Lameter24922682007-07-17 04:03:18 -0700573static void print_tracking(struct kmem_cache *s, void *object)
574{
575 if (!(s->flags & SLAB_STORE_USER))
576 return;
577
578 print_track("Allocated", get_track(s, object, TRACK_ALLOC));
579 print_track("Freed", get_track(s, object, TRACK_FREE));
580}
581
582static void print_page_info(struct page *page)
583{
Fabian Frederickf9f58282014-06-04 16:06:34 -0700584 pr_err("INFO: Slab 0x%p objects=%u used=%u fp=0x%p flags=0x%04lx\n",
Chen Gangd0e0ac92013-07-15 09:05:29 +0800585 page, page->objects, page->inuse, page->freelist, page->flags);
Christoph Lameter24922682007-07-17 04:03:18 -0700586
587}
588
589static void slab_bug(struct kmem_cache *s, char *fmt, ...)
590{
Fabian Frederickecc42fb2014-06-04 16:06:35 -0700591 struct va_format vaf;
Christoph Lameter24922682007-07-17 04:03:18 -0700592 va_list args;
Christoph Lameter24922682007-07-17 04:03:18 -0700593
594 va_start(args, fmt);
Fabian Frederickecc42fb2014-06-04 16:06:35 -0700595 vaf.fmt = fmt;
596 vaf.va = &args;
Fabian Frederickf9f58282014-06-04 16:06:34 -0700597 pr_err("=============================================================================\n");
Fabian Frederickecc42fb2014-06-04 16:06:35 -0700598 pr_err("BUG %s (%s): %pV\n", s->name, print_tainted(), &vaf);
Fabian Frederickf9f58282014-06-04 16:06:34 -0700599 pr_err("-----------------------------------------------------------------------------\n\n");
Dave Jones645df232012-09-18 15:54:12 -0400600
Rusty Russell373d4d02013-01-21 17:17:39 +1030601 add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
Fabian Frederickecc42fb2014-06-04 16:06:35 -0700602 va_end(args);
Christoph Lameter24922682007-07-17 04:03:18 -0700603}
604
605static void slab_fix(struct kmem_cache *s, char *fmt, ...)
606{
Fabian Frederickecc42fb2014-06-04 16:06:35 -0700607 struct va_format vaf;
Christoph Lameter24922682007-07-17 04:03:18 -0700608 va_list args;
Christoph Lameter24922682007-07-17 04:03:18 -0700609
610 va_start(args, fmt);
Fabian Frederickecc42fb2014-06-04 16:06:35 -0700611 vaf.fmt = fmt;
612 vaf.va = &args;
613 pr_err("FIX %s: %pV\n", s->name, &vaf);
Christoph Lameter24922682007-07-17 04:03:18 -0700614 va_end(args);
Christoph Lameter24922682007-07-17 04:03:18 -0700615}
616
617static void print_trailer(struct kmem_cache *s, struct page *page, u8 *p)
Christoph Lameter81819f02007-05-06 14:49:36 -0700618{
619 unsigned int off; /* Offset of last byte */
Christoph Lametera973e9d2008-03-01 13:40:44 -0800620 u8 *addr = page_address(page);
Christoph Lameter24922682007-07-17 04:03:18 -0700621
622 print_tracking(s, p);
623
624 print_page_info(page);
625
Fabian Frederickf9f58282014-06-04 16:06:34 -0700626 pr_err("INFO: Object 0x%p @offset=%tu fp=0x%p\n\n",
627 p, p - addr, get_freepointer(s, p));
Christoph Lameter24922682007-07-17 04:03:18 -0700628
629 if (p > addr + 16)
Sebastian Andrzej Siewiorffc79d22011-07-29 14:10:20 +0200630 print_section("Bytes b4 ", p - 16, 16);
Christoph Lameter24922682007-07-17 04:03:18 -0700631
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500632 print_section("Object ", p, min_t(unsigned long, s->object_size,
Sebastian Andrzej Siewiorffc79d22011-07-29 14:10:20 +0200633 PAGE_SIZE));
Christoph Lameter81819f02007-05-06 14:49:36 -0700634 if (s->flags & SLAB_RED_ZONE)
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500635 print_section("Redzone ", p + s->object_size,
636 s->inuse - s->object_size);
Christoph Lameter81819f02007-05-06 14:49:36 -0700637
Christoph Lameter81819f02007-05-06 14:49:36 -0700638 if (s->offset)
639 off = s->offset + sizeof(void *);
640 else
641 off = s->inuse;
642
Christoph Lameter24922682007-07-17 04:03:18 -0700643 if (s->flags & SLAB_STORE_USER)
Christoph Lameter81819f02007-05-06 14:49:36 -0700644 off += 2 * sizeof(struct track);
Christoph Lameter81819f02007-05-06 14:49:36 -0700645
646 if (off != s->size)
647 /* Beginning of the filler is the free pointer */
Sebastian Andrzej Siewiorffc79d22011-07-29 14:10:20 +0200648 print_section("Padding ", p + off, s->size - off);
Christoph Lameter24922682007-07-17 04:03:18 -0700649
650 dump_stack();
Christoph Lameter81819f02007-05-06 14:49:36 -0700651}
652
Andrey Ryabinin75c66de2015-02-13 14:39:35 -0800653void object_err(struct kmem_cache *s, struct page *page,
Christoph Lameter81819f02007-05-06 14:49:36 -0700654 u8 *object, char *reason)
655{
Christoph Lameter3dc50632008-04-23 12:28:01 -0700656 slab_bug(s, "%s", reason);
Christoph Lameter24922682007-07-17 04:03:18 -0700657 print_trailer(s, page, object);
Christoph Lameter81819f02007-05-06 14:49:36 -0700658}
659
Chen Gangd0e0ac92013-07-15 09:05:29 +0800660static void slab_err(struct kmem_cache *s, struct page *page,
661 const char *fmt, ...)
Christoph Lameter81819f02007-05-06 14:49:36 -0700662{
663 va_list args;
664 char buf[100];
665
Christoph Lameter24922682007-07-17 04:03:18 -0700666 va_start(args, fmt);
667 vsnprintf(buf, sizeof(buf), fmt, args);
Christoph Lameter81819f02007-05-06 14:49:36 -0700668 va_end(args);
Christoph Lameter3dc50632008-04-23 12:28:01 -0700669 slab_bug(s, "%s", buf);
Christoph Lameter24922682007-07-17 04:03:18 -0700670 print_page_info(page);
Christoph Lameter81819f02007-05-06 14:49:36 -0700671 dump_stack();
672}
673
Christoph Lameterf7cb1932010-09-29 07:15:01 -0500674static void init_object(struct kmem_cache *s, void *object, u8 val)
Christoph Lameter81819f02007-05-06 14:49:36 -0700675{
676 u8 *p = object;
677
678 if (s->flags & __OBJECT_POISON) {
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500679 memset(p, POISON_FREE, s->object_size - 1);
680 p[s->object_size - 1] = POISON_END;
Christoph Lameter81819f02007-05-06 14:49:36 -0700681 }
682
683 if (s->flags & SLAB_RED_ZONE)
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500684 memset(p + s->object_size, val, s->inuse - s->object_size);
Christoph Lameter81819f02007-05-06 14:49:36 -0700685}
686
Christoph Lameter24922682007-07-17 04:03:18 -0700687static void restore_bytes(struct kmem_cache *s, char *message, u8 data,
688 void *from, void *to)
689{
690 slab_fix(s, "Restoring 0x%p-0x%p=0x%x\n", from, to - 1, data);
691 memset(from, data, to - from);
692}
693
694static int check_bytes_and_report(struct kmem_cache *s, struct page *page,
695 u8 *object, char *what,
Pekka Enberg06428782008-01-07 23:20:27 -0800696 u8 *start, unsigned int value, unsigned int bytes)
Christoph Lameter24922682007-07-17 04:03:18 -0700697{
698 u8 *fault;
699 u8 *end;
700
Andrey Ryabinina79316c2015-02-13 14:39:38 -0800701 metadata_access_enable();
Akinobu Mita798248202011-10-31 17:08:07 -0700702 fault = memchr_inv(start, value, bytes);
Andrey Ryabinina79316c2015-02-13 14:39:38 -0800703 metadata_access_disable();
Christoph Lameter24922682007-07-17 04:03:18 -0700704 if (!fault)
705 return 1;
706
707 end = start + bytes;
708 while (end > fault && end[-1] == value)
709 end--;
710
711 slab_bug(s, "%s overwritten", what);
Fabian Frederickf9f58282014-06-04 16:06:34 -0700712 pr_err("INFO: 0x%p-0x%p. First byte 0x%x instead of 0x%x\n",
Christoph Lameter24922682007-07-17 04:03:18 -0700713 fault, end - 1, fault[0], value);
714 print_trailer(s, page, object);
715
716 restore_bytes(s, what, value, fault, end);
717 return 0;
Christoph Lameter81819f02007-05-06 14:49:36 -0700718}
719
Christoph Lameter81819f02007-05-06 14:49:36 -0700720/*
721 * Object layout:
722 *
723 * object address
724 * Bytes of the object to be managed.
725 * If the freepointer may overlay the object then the free
726 * pointer is the first word of the object.
Christoph Lameter672bba32007-05-09 02:32:39 -0700727 *
Christoph Lameter81819f02007-05-06 14:49:36 -0700728 * Poisoning uses 0x6b (POISON_FREE) and the last byte is
729 * 0xa5 (POISON_END)
730 *
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500731 * object + s->object_size
Christoph Lameter81819f02007-05-06 14:49:36 -0700732 * Padding to reach word boundary. This is also used for Redzoning.
Christoph Lameter672bba32007-05-09 02:32:39 -0700733 * Padding is extended by another word if Redzoning is enabled and
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500734 * object_size == inuse.
Christoph Lameter672bba32007-05-09 02:32:39 -0700735 *
Christoph Lameter81819f02007-05-06 14:49:36 -0700736 * We fill with 0xbb (RED_INACTIVE) for inactive objects and with
737 * 0xcc (RED_ACTIVE) for objects in use.
738 *
739 * object + s->inuse
Christoph Lameter672bba32007-05-09 02:32:39 -0700740 * Meta data starts here.
741 *
Christoph Lameter81819f02007-05-06 14:49:36 -0700742 * A. Free pointer (if we cannot overwrite object on free)
743 * B. Tracking data for SLAB_STORE_USER
Christoph Lameter672bba32007-05-09 02:32:39 -0700744 * C. Padding to reach required alignment boundary or at mininum
Christoph Lameter6446faa2008-02-15 23:45:26 -0800745 * one word if debugging is on to be able to detect writes
Christoph Lameter672bba32007-05-09 02:32:39 -0700746 * before the word boundary.
747 *
748 * Padding is done using 0x5a (POISON_INUSE)
Christoph Lameter81819f02007-05-06 14:49:36 -0700749 *
750 * object + s->size
Christoph Lameter672bba32007-05-09 02:32:39 -0700751 * Nothing is used beyond s->size.
Christoph Lameter81819f02007-05-06 14:49:36 -0700752 *
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500753 * If slabcaches are merged then the object_size and inuse boundaries are mostly
Christoph Lameter672bba32007-05-09 02:32:39 -0700754 * ignored. And therefore no slab options that rely on these boundaries
Christoph Lameter81819f02007-05-06 14:49:36 -0700755 * may be used with merged slabcaches.
756 */
757
Christoph Lameter81819f02007-05-06 14:49:36 -0700758static int check_pad_bytes(struct kmem_cache *s, struct page *page, u8 *p)
759{
760 unsigned long off = s->inuse; /* The end of info */
761
762 if (s->offset)
763 /* Freepointer is placed after the object. */
764 off += sizeof(void *);
765
766 if (s->flags & SLAB_STORE_USER)
767 /* We also have user information there */
768 off += 2 * sizeof(struct track);
769
770 if (s->size == off)
771 return 1;
772
Christoph Lameter24922682007-07-17 04:03:18 -0700773 return check_bytes_and_report(s, page, p, "Object padding",
774 p + off, POISON_INUSE, s->size - off);
Christoph Lameter81819f02007-05-06 14:49:36 -0700775}
776
Christoph Lameter39b26462008-04-14 19:11:30 +0300777/* Check the pad bytes at the end of a slab page */
Christoph Lameter81819f02007-05-06 14:49:36 -0700778static int slab_pad_check(struct kmem_cache *s, struct page *page)
779{
Christoph Lameter24922682007-07-17 04:03:18 -0700780 u8 *start;
781 u8 *fault;
782 u8 *end;
783 int length;
784 int remainder;
Christoph Lameter81819f02007-05-06 14:49:36 -0700785
786 if (!(s->flags & SLAB_POISON))
787 return 1;
788
Christoph Lametera973e9d2008-03-01 13:40:44 -0800789 start = page_address(page);
Lai Jiangshanab9a0f12011-03-10 15:21:48 +0800790 length = (PAGE_SIZE << compound_order(page)) - s->reserved;
Christoph Lameter39b26462008-04-14 19:11:30 +0300791 end = start + length;
792 remainder = length % s->size;
Christoph Lameter81819f02007-05-06 14:49:36 -0700793 if (!remainder)
794 return 1;
795
Andrey Ryabinina79316c2015-02-13 14:39:38 -0800796 metadata_access_enable();
Akinobu Mita798248202011-10-31 17:08:07 -0700797 fault = memchr_inv(end - remainder, POISON_INUSE, remainder);
Andrey Ryabinina79316c2015-02-13 14:39:38 -0800798 metadata_access_disable();
Christoph Lameter24922682007-07-17 04:03:18 -0700799 if (!fault)
800 return 1;
801 while (end > fault && end[-1] == POISON_INUSE)
802 end--;
803
804 slab_err(s, page, "Padding overwritten. 0x%p-0x%p", fault, end - 1);
Sebastian Andrzej Siewiorffc79d22011-07-29 14:10:20 +0200805 print_section("Padding ", end - remainder, remainder);
Christoph Lameter24922682007-07-17 04:03:18 -0700806
Eric Dumazet8a3d2712009-09-03 16:08:06 +0200807 restore_bytes(s, "slab padding", POISON_INUSE, end - remainder, end);
Christoph Lameter24922682007-07-17 04:03:18 -0700808 return 0;
Christoph Lameter81819f02007-05-06 14:49:36 -0700809}
810
811static int check_object(struct kmem_cache *s, struct page *page,
Christoph Lameterf7cb1932010-09-29 07:15:01 -0500812 void *object, u8 val)
Christoph Lameter81819f02007-05-06 14:49:36 -0700813{
814 u8 *p = object;
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500815 u8 *endobject = object + s->object_size;
Christoph Lameter81819f02007-05-06 14:49:36 -0700816
817 if (s->flags & SLAB_RED_ZONE) {
Christoph Lameter24922682007-07-17 04:03:18 -0700818 if (!check_bytes_and_report(s, page, object, "Redzone",
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500819 endobject, val, s->inuse - s->object_size))
Christoph Lameter81819f02007-05-06 14:49:36 -0700820 return 0;
Christoph Lameter81819f02007-05-06 14:49:36 -0700821 } else {
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500822 if ((s->flags & SLAB_POISON) && s->object_size < s->inuse) {
Ingo Molnar3adbefe2008-02-05 17:57:39 -0800823 check_bytes_and_report(s, page, p, "Alignment padding",
Chen Gangd0e0ac92013-07-15 09:05:29 +0800824 endobject, POISON_INUSE,
825 s->inuse - s->object_size);
Ingo Molnar3adbefe2008-02-05 17:57:39 -0800826 }
Christoph Lameter81819f02007-05-06 14:49:36 -0700827 }
828
829 if (s->flags & SLAB_POISON) {
Christoph Lameterf7cb1932010-09-29 07:15:01 -0500830 if (val != SLUB_RED_ACTIVE && (s->flags & __OBJECT_POISON) &&
Christoph Lameter24922682007-07-17 04:03:18 -0700831 (!check_bytes_and_report(s, page, p, "Poison", p,
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500832 POISON_FREE, s->object_size - 1) ||
Christoph Lameter24922682007-07-17 04:03:18 -0700833 !check_bytes_and_report(s, page, p, "Poison",
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500834 p + s->object_size - 1, POISON_END, 1)))
Christoph Lameter81819f02007-05-06 14:49:36 -0700835 return 0;
Christoph Lameter81819f02007-05-06 14:49:36 -0700836 /*
837 * check_pad_bytes cleans up on its own.
838 */
839 check_pad_bytes(s, page, p);
840 }
841
Christoph Lameterf7cb1932010-09-29 07:15:01 -0500842 if (!s->offset && val == SLUB_RED_ACTIVE)
Christoph Lameter81819f02007-05-06 14:49:36 -0700843 /*
844 * Object and freepointer overlap. Cannot check
845 * freepointer while object is allocated.
846 */
847 return 1;
848
849 /* Check free pointer validity */
850 if (!check_valid_pointer(s, page, get_freepointer(s, p))) {
851 object_err(s, page, p, "Freepointer corrupt");
852 /*
Nick Andrew9f6c708e2008-12-05 14:08:08 +1100853 * No choice but to zap it and thus lose the remainder
Christoph Lameter81819f02007-05-06 14:49:36 -0700854 * of the free objects in this slab. May cause
Christoph Lameter672bba32007-05-09 02:32:39 -0700855 * another error because the object count is now wrong.
Christoph Lameter81819f02007-05-06 14:49:36 -0700856 */
Christoph Lametera973e9d2008-03-01 13:40:44 -0800857 set_freepointer(s, p, NULL);
Christoph Lameter81819f02007-05-06 14:49:36 -0700858 return 0;
859 }
860 return 1;
861}
862
863static int check_slab(struct kmem_cache *s, struct page *page)
864{
Christoph Lameter39b26462008-04-14 19:11:30 +0300865 int maxobj;
866
Christoph Lameter81819f02007-05-06 14:49:36 -0700867 VM_BUG_ON(!irqs_disabled());
868
869 if (!PageSlab(page)) {
Christoph Lameter24922682007-07-17 04:03:18 -0700870 slab_err(s, page, "Not a valid slab page");
Christoph Lameter81819f02007-05-06 14:49:36 -0700871 return 0;
872 }
Christoph Lameter39b26462008-04-14 19:11:30 +0300873
Lai Jiangshanab9a0f12011-03-10 15:21:48 +0800874 maxobj = order_objects(compound_order(page), s->size, s->reserved);
Christoph Lameter39b26462008-04-14 19:11:30 +0300875 if (page->objects > maxobj) {
876 slab_err(s, page, "objects %u > max %u",
Andrey Ryabininf6edde92014-12-10 15:42:22 -0800877 page->objects, maxobj);
Christoph Lameter39b26462008-04-14 19:11:30 +0300878 return 0;
879 }
880 if (page->inuse > page->objects) {
Christoph Lameter24922682007-07-17 04:03:18 -0700881 slab_err(s, page, "inuse %u > max %u",
Andrey Ryabininf6edde92014-12-10 15:42:22 -0800882 page->inuse, page->objects);
Christoph Lameter81819f02007-05-06 14:49:36 -0700883 return 0;
884 }
885 /* Slab_pad_check fixes things up after itself */
886 slab_pad_check(s, page);
887 return 1;
888}
889
890/*
Christoph Lameter672bba32007-05-09 02:32:39 -0700891 * Determine if a certain object on a page is on the freelist. Must hold the
892 * slab lock to guarantee that the chains are in a consistent state.
Christoph Lameter81819f02007-05-06 14:49:36 -0700893 */
894static int on_freelist(struct kmem_cache *s, struct page *page, void *search)
895{
896 int nr = 0;
Christoph Lameter881db7f2011-06-01 12:25:53 -0500897 void *fp;
Christoph Lameter81819f02007-05-06 14:49:36 -0700898 void *object = NULL;
Andrey Ryabininf6edde92014-12-10 15:42:22 -0800899 int max_objects;
Christoph Lameter81819f02007-05-06 14:49:36 -0700900
Christoph Lameter881db7f2011-06-01 12:25:53 -0500901 fp = page->freelist;
Christoph Lameter39b26462008-04-14 19:11:30 +0300902 while (fp && nr <= page->objects) {
Christoph Lameter81819f02007-05-06 14:49:36 -0700903 if (fp == search)
904 return 1;
905 if (!check_valid_pointer(s, page, fp)) {
906 if (object) {
907 object_err(s, page, object,
908 "Freechain corrupt");
Christoph Lametera973e9d2008-03-01 13:40:44 -0800909 set_freepointer(s, object, NULL);
Christoph Lameter81819f02007-05-06 14:49:36 -0700910 } else {
Christoph Lameter24922682007-07-17 04:03:18 -0700911 slab_err(s, page, "Freepointer corrupt");
Christoph Lametera973e9d2008-03-01 13:40:44 -0800912 page->freelist = NULL;
Christoph Lameter39b26462008-04-14 19:11:30 +0300913 page->inuse = page->objects;
Christoph Lameter24922682007-07-17 04:03:18 -0700914 slab_fix(s, "Freelist cleared");
Christoph Lameter81819f02007-05-06 14:49:36 -0700915 return 0;
916 }
917 break;
918 }
919 object = fp;
920 fp = get_freepointer(s, object);
921 nr++;
922 }
923
Lai Jiangshanab9a0f12011-03-10 15:21:48 +0800924 max_objects = order_objects(compound_order(page), s->size, s->reserved);
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +0400925 if (max_objects > MAX_OBJS_PER_PAGE)
926 max_objects = MAX_OBJS_PER_PAGE;
Christoph Lameter224a88b2008-04-14 19:11:31 +0300927
928 if (page->objects != max_objects) {
929 slab_err(s, page, "Wrong number of objects. Found %d but "
930 "should be %d", page->objects, max_objects);
931 page->objects = max_objects;
932 slab_fix(s, "Number of objects adjusted.");
933 }
Christoph Lameter39b26462008-04-14 19:11:30 +0300934 if (page->inuse != page->objects - nr) {
Christoph Lameter70d71222007-05-06 14:49:47 -0700935 slab_err(s, page, "Wrong object count. Counter is %d but "
Christoph Lameter39b26462008-04-14 19:11:30 +0300936 "counted were %d", page->inuse, page->objects - nr);
937 page->inuse = page->objects - nr;
Christoph Lameter24922682007-07-17 04:03:18 -0700938 slab_fix(s, "Object count adjusted.");
Christoph Lameter81819f02007-05-06 14:49:36 -0700939 }
940 return search == NULL;
941}
942
Christoph Lameter0121c6192008-04-29 16:11:12 -0700943static void trace(struct kmem_cache *s, struct page *page, void *object,
944 int alloc)
Christoph Lameter3ec09742007-05-16 22:11:00 -0700945{
946 if (s->flags & SLAB_TRACE) {
Fabian Frederickf9f58282014-06-04 16:06:34 -0700947 pr_info("TRACE %s %s 0x%p inuse=%d fp=0x%p\n",
Christoph Lameter3ec09742007-05-16 22:11:00 -0700948 s->name,
949 alloc ? "alloc" : "free",
950 object, page->inuse,
951 page->freelist);
952
953 if (!alloc)
Chen Gangd0e0ac92013-07-15 09:05:29 +0800954 print_section("Object ", (void *)object,
955 s->object_size);
Christoph Lameter3ec09742007-05-16 22:11:00 -0700956
957 dump_stack();
958 }
959}
960
Christoph Lameter643b1132007-05-06 14:49:42 -0700961/*
Christoph Lameter672bba32007-05-09 02:32:39 -0700962 * Tracking of fully allocated slabs for debugging purposes.
Christoph Lameter643b1132007-05-06 14:49:42 -0700963 */
Christoph Lameter5cc6eee2011-06-01 12:25:50 -0500964static void add_full(struct kmem_cache *s,
965 struct kmem_cache_node *n, struct page *page)
Christoph Lameter643b1132007-05-06 14:49:42 -0700966{
Christoph Lameter643b1132007-05-06 14:49:42 -0700967 if (!(s->flags & SLAB_STORE_USER))
968 return;
969
David Rientjes255d0882014-02-10 14:25:39 -0800970 lockdep_assert_held(&n->list_lock);
Christoph Lameter5cc6eee2011-06-01 12:25:50 -0500971 list_add(&page->lru, &n->full);
972}
Christoph Lameter643b1132007-05-06 14:49:42 -0700973
Peter Zijlstrac65c1872014-01-10 13:23:49 +0100974static void remove_full(struct kmem_cache *s, struct kmem_cache_node *n, struct page *page)
Christoph Lameter5cc6eee2011-06-01 12:25:50 -0500975{
976 if (!(s->flags & SLAB_STORE_USER))
977 return;
978
David Rientjes255d0882014-02-10 14:25:39 -0800979 lockdep_assert_held(&n->list_lock);
Christoph Lameter643b1132007-05-06 14:49:42 -0700980 list_del(&page->lru);
Christoph Lameter643b1132007-05-06 14:49:42 -0700981}
982
Christoph Lameter0f389ec2008-04-14 18:53:02 +0300983/* Tracking of the number of slabs for debugging purposes */
984static inline unsigned long slabs_node(struct kmem_cache *s, int node)
985{
986 struct kmem_cache_node *n = get_node(s, node);
987
988 return atomic_long_read(&n->nr_slabs);
989}
990
Alexander Beregalov26c02cf2009-06-11 14:08:48 +0400991static inline unsigned long node_nr_slabs(struct kmem_cache_node *n)
992{
993 return atomic_long_read(&n->nr_slabs);
994}
995
Christoph Lameter205ab992008-04-14 19:11:40 +0300996static inline void inc_slabs_node(struct kmem_cache *s, int node, int objects)
Christoph Lameter0f389ec2008-04-14 18:53:02 +0300997{
998 struct kmem_cache_node *n = get_node(s, node);
999
1000 /*
1001 * May be called early in order to allocate a slab for the
1002 * kmem_cache_node structure. Solve the chicken-egg
1003 * dilemma by deferring the increment of the count during
1004 * bootstrap (see early_kmem_cache_node_alloc).
1005 */
Joonsoo Kim338b2642013-01-21 17:01:27 +09001006 if (likely(n)) {
Christoph Lameter0f389ec2008-04-14 18:53:02 +03001007 atomic_long_inc(&n->nr_slabs);
Christoph Lameter205ab992008-04-14 19:11:40 +03001008 atomic_long_add(objects, &n->total_objects);
1009 }
Christoph Lameter0f389ec2008-04-14 18:53:02 +03001010}
Christoph Lameter205ab992008-04-14 19:11:40 +03001011static inline void dec_slabs_node(struct kmem_cache *s, int node, int objects)
Christoph Lameter0f389ec2008-04-14 18:53:02 +03001012{
1013 struct kmem_cache_node *n = get_node(s, node);
1014
1015 atomic_long_dec(&n->nr_slabs);
Christoph Lameter205ab992008-04-14 19:11:40 +03001016 atomic_long_sub(objects, &n->total_objects);
Christoph Lameter0f389ec2008-04-14 18:53:02 +03001017}
1018
1019/* Object debug checks for alloc/free paths */
Christoph Lameter3ec09742007-05-16 22:11:00 -07001020static void setup_object_debug(struct kmem_cache *s, struct page *page,
1021 void *object)
1022{
1023 if (!(s->flags & (SLAB_STORE_USER|SLAB_RED_ZONE|__OBJECT_POISON)))
1024 return;
1025
Christoph Lameterf7cb1932010-09-29 07:15:01 -05001026 init_object(s, object, SLUB_RED_INACTIVE);
Christoph Lameter3ec09742007-05-16 22:11:00 -07001027 init_tracking(s, object);
1028}
1029
Chen Gangd0e0ac92013-07-15 09:05:29 +08001030static noinline int alloc_debug_processing(struct kmem_cache *s,
1031 struct page *page,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03001032 void *object, unsigned long addr)
Christoph Lameter81819f02007-05-06 14:49:36 -07001033{
1034 if (!check_slab(s, page))
1035 goto bad;
1036
Christoph Lameter81819f02007-05-06 14:49:36 -07001037 if (!check_valid_pointer(s, page, object)) {
1038 object_err(s, page, object, "Freelist Pointer check fails");
Christoph Lameter70d71222007-05-06 14:49:47 -07001039 goto bad;
Christoph Lameter81819f02007-05-06 14:49:36 -07001040 }
1041
Christoph Lameterf7cb1932010-09-29 07:15:01 -05001042 if (!check_object(s, page, object, SLUB_RED_INACTIVE))
Christoph Lameter81819f02007-05-06 14:49:36 -07001043 goto bad;
Christoph Lameter81819f02007-05-06 14:49:36 -07001044
Christoph Lameter3ec09742007-05-16 22:11:00 -07001045 /* Success perform special debug activities for allocs */
1046 if (s->flags & SLAB_STORE_USER)
1047 set_track(s, object, TRACK_ALLOC, addr);
1048 trace(s, page, object, 1);
Christoph Lameterf7cb1932010-09-29 07:15:01 -05001049 init_object(s, object, SLUB_RED_ACTIVE);
Christoph Lameter81819f02007-05-06 14:49:36 -07001050 return 1;
Christoph Lameter3ec09742007-05-16 22:11:00 -07001051
Christoph Lameter81819f02007-05-06 14:49:36 -07001052bad:
1053 if (PageSlab(page)) {
1054 /*
1055 * If this is a slab page then lets do the best we can
1056 * to avoid issues in the future. Marking all objects
Christoph Lameter672bba32007-05-09 02:32:39 -07001057 * as used avoids touching the remaining objects.
Christoph Lameter81819f02007-05-06 14:49:36 -07001058 */
Christoph Lameter24922682007-07-17 04:03:18 -07001059 slab_fix(s, "Marking all objects used");
Christoph Lameter39b26462008-04-14 19:11:30 +03001060 page->inuse = page->objects;
Christoph Lametera973e9d2008-03-01 13:40:44 -08001061 page->freelist = NULL;
Christoph Lameter81819f02007-05-06 14:49:36 -07001062 }
1063 return 0;
1064}
1065
Christoph Lameter19c7ff92012-05-30 12:54:46 -05001066static noinline struct kmem_cache_node *free_debug_processing(
1067 struct kmem_cache *s, struct page *page, void *object,
1068 unsigned long addr, unsigned long *flags)
Christoph Lameter81819f02007-05-06 14:49:36 -07001069{
Christoph Lameter19c7ff92012-05-30 12:54:46 -05001070 struct kmem_cache_node *n = get_node(s, page_to_nid(page));
Christoph Lameter5c2e4bb2011-06-01 12:25:54 -05001071
Christoph Lameter19c7ff92012-05-30 12:54:46 -05001072 spin_lock_irqsave(&n->list_lock, *flags);
Christoph Lameter881db7f2011-06-01 12:25:53 -05001073 slab_lock(page);
1074
Christoph Lameter81819f02007-05-06 14:49:36 -07001075 if (!check_slab(s, page))
1076 goto fail;
1077
1078 if (!check_valid_pointer(s, page, object)) {
Christoph Lameter70d71222007-05-06 14:49:47 -07001079 slab_err(s, page, "Invalid object pointer 0x%p", object);
Christoph Lameter81819f02007-05-06 14:49:36 -07001080 goto fail;
1081 }
1082
1083 if (on_freelist(s, page, object)) {
Christoph Lameter24922682007-07-17 04:03:18 -07001084 object_err(s, page, object, "Object already free");
Christoph Lameter81819f02007-05-06 14:49:36 -07001085 goto fail;
1086 }
1087
Christoph Lameterf7cb1932010-09-29 07:15:01 -05001088 if (!check_object(s, page, object, SLUB_RED_ACTIVE))
Christoph Lameter5c2e4bb2011-06-01 12:25:54 -05001089 goto out;
Christoph Lameter81819f02007-05-06 14:49:36 -07001090
Glauber Costa1b4f59e32012-10-22 18:05:36 +04001091 if (unlikely(s != page->slab_cache)) {
Ingo Molnar3adbefe2008-02-05 17:57:39 -08001092 if (!PageSlab(page)) {
Christoph Lameter70d71222007-05-06 14:49:47 -07001093 slab_err(s, page, "Attempt to free object(0x%p) "
1094 "outside of slab", object);
Glauber Costa1b4f59e32012-10-22 18:05:36 +04001095 } else if (!page->slab_cache) {
Fabian Frederickf9f58282014-06-04 16:06:34 -07001096 pr_err("SLUB <none>: no slab for object 0x%p.\n",
1097 object);
Christoph Lameter70d71222007-05-06 14:49:47 -07001098 dump_stack();
Pekka Enberg06428782008-01-07 23:20:27 -08001099 } else
Christoph Lameter24922682007-07-17 04:03:18 -07001100 object_err(s, page, object,
1101 "page slab pointer corrupt.");
Christoph Lameter81819f02007-05-06 14:49:36 -07001102 goto fail;
1103 }
Christoph Lameter3ec09742007-05-16 22:11:00 -07001104
Christoph Lameter3ec09742007-05-16 22:11:00 -07001105 if (s->flags & SLAB_STORE_USER)
1106 set_track(s, object, TRACK_FREE, addr);
1107 trace(s, page, object, 0);
Christoph Lameterf7cb1932010-09-29 07:15:01 -05001108 init_object(s, object, SLUB_RED_INACTIVE);
Christoph Lameter5c2e4bb2011-06-01 12:25:54 -05001109out:
Christoph Lameter881db7f2011-06-01 12:25:53 -05001110 slab_unlock(page);
Christoph Lameter19c7ff92012-05-30 12:54:46 -05001111 /*
1112 * Keep node_lock to preserve integrity
1113 * until the object is actually freed
1114 */
1115 return n;
Christoph Lameter3ec09742007-05-16 22:11:00 -07001116
Christoph Lameter81819f02007-05-06 14:49:36 -07001117fail:
Christoph Lameter19c7ff92012-05-30 12:54:46 -05001118 slab_unlock(page);
1119 spin_unlock_irqrestore(&n->list_lock, *flags);
Christoph Lameter24922682007-07-17 04:03:18 -07001120 slab_fix(s, "Object at 0x%p not freed", object);
Christoph Lameter19c7ff92012-05-30 12:54:46 -05001121 return NULL;
Christoph Lameter81819f02007-05-06 14:49:36 -07001122}
1123
Christoph Lameter41ecc552007-05-09 02:32:44 -07001124static int __init setup_slub_debug(char *str)
1125{
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001126 slub_debug = DEBUG_DEFAULT_FLAGS;
1127 if (*str++ != '=' || !*str)
1128 /*
1129 * No options specified. Switch on full debugging.
1130 */
1131 goto out;
Christoph Lameter41ecc552007-05-09 02:32:44 -07001132
1133 if (*str == ',')
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001134 /*
1135 * No options but restriction on slabs. This means full
1136 * debugging for slabs matching a pattern.
1137 */
1138 goto check_slabs;
1139
1140 slub_debug = 0;
1141 if (*str == '-')
1142 /*
1143 * Switch off all debugging measures.
1144 */
1145 goto out;
1146
1147 /*
1148 * Determine which debug features should be switched on
1149 */
Pekka Enberg06428782008-01-07 23:20:27 -08001150 for (; *str && *str != ','; str++) {
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001151 switch (tolower(*str)) {
1152 case 'f':
1153 slub_debug |= SLAB_DEBUG_FREE;
1154 break;
1155 case 'z':
1156 slub_debug |= SLAB_RED_ZONE;
1157 break;
1158 case 'p':
1159 slub_debug |= SLAB_POISON;
1160 break;
1161 case 'u':
1162 slub_debug |= SLAB_STORE_USER;
1163 break;
1164 case 't':
1165 slub_debug |= SLAB_TRACE;
1166 break;
Dmitry Monakhov4c13dd32010-02-26 09:36:12 +03001167 case 'a':
1168 slub_debug |= SLAB_FAILSLAB;
1169 break;
Chris J Arges08303a72015-04-14 15:44:25 -07001170 case 'o':
1171 /*
1172 * Avoid enabling debugging on caches if its minimum
1173 * order would increase as a result.
1174 */
1175 disable_higher_order_debug = 1;
1176 break;
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001177 default:
Fabian Frederickf9f58282014-06-04 16:06:34 -07001178 pr_err("slub_debug option '%c' unknown. skipped\n",
1179 *str);
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001180 }
1181 }
1182
1183check_slabs:
1184 if (*str == ',')
Christoph Lameter41ecc552007-05-09 02:32:44 -07001185 slub_debug_slabs = str + 1;
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001186out:
Christoph Lameter41ecc552007-05-09 02:32:44 -07001187 return 1;
1188}
1189
1190__setup("slub_debug", setup_slub_debug);
1191
Joonsoo Kim423c9292014-10-09 15:26:22 -07001192unsigned long kmem_cache_flags(unsigned long object_size,
Christoph Lameterba0268a2007-09-11 15:24:11 -07001193 unsigned long flags, const char *name,
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07001194 void (*ctor)(void *))
Christoph Lameter41ecc552007-05-09 02:32:44 -07001195{
1196 /*
Christoph Lametere1533622008-02-15 23:45:24 -08001197 * Enable debugging if selected on the kernel commandline.
Christoph Lameter41ecc552007-05-09 02:32:44 -07001198 */
Christoph Lameterc6f58d92013-11-07 16:29:15 +00001199 if (slub_debug && (!slub_debug_slabs || (name &&
1200 !strncmp(slub_debug_slabs, name, strlen(slub_debug_slabs)))))
David Rientjes3de47212009-07-27 18:30:35 -07001201 flags |= slub_debug;
Christoph Lameterba0268a2007-09-11 15:24:11 -07001202
1203 return flags;
Christoph Lameter41ecc552007-05-09 02:32:44 -07001204}
1205#else
Christoph Lameter3ec09742007-05-16 22:11:00 -07001206static inline void setup_object_debug(struct kmem_cache *s,
1207 struct page *page, void *object) {}
Christoph Lameter41ecc552007-05-09 02:32:44 -07001208
Christoph Lameter3ec09742007-05-16 22:11:00 -07001209static inline int alloc_debug_processing(struct kmem_cache *s,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03001210 struct page *page, void *object, unsigned long addr) { return 0; }
Christoph Lameter41ecc552007-05-09 02:32:44 -07001211
Christoph Lameter19c7ff92012-05-30 12:54:46 -05001212static inline struct kmem_cache_node *free_debug_processing(
1213 struct kmem_cache *s, struct page *page, void *object,
1214 unsigned long addr, unsigned long *flags) { return NULL; }
Christoph Lameter41ecc552007-05-09 02:32:44 -07001215
Christoph Lameter41ecc552007-05-09 02:32:44 -07001216static inline int slab_pad_check(struct kmem_cache *s, struct page *page)
1217 { return 1; }
1218static inline int check_object(struct kmem_cache *s, struct page *page,
Christoph Lameterf7cb1932010-09-29 07:15:01 -05001219 void *object, u8 val) { return 1; }
Christoph Lameter5cc6eee2011-06-01 12:25:50 -05001220static inline void add_full(struct kmem_cache *s, struct kmem_cache_node *n,
1221 struct page *page) {}
Peter Zijlstrac65c1872014-01-10 13:23:49 +01001222static inline void remove_full(struct kmem_cache *s, struct kmem_cache_node *n,
1223 struct page *page) {}
Joonsoo Kim423c9292014-10-09 15:26:22 -07001224unsigned long kmem_cache_flags(unsigned long object_size,
Christoph Lameterba0268a2007-09-11 15:24:11 -07001225 unsigned long flags, const char *name,
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07001226 void (*ctor)(void *))
Christoph Lameterba0268a2007-09-11 15:24:11 -07001227{
1228 return flags;
1229}
Christoph Lameter41ecc552007-05-09 02:32:44 -07001230#define slub_debug 0
Christoph Lameter0f389ec2008-04-14 18:53:02 +03001231
Ingo Molnarfdaa45e2009-09-15 11:00:26 +02001232#define disable_higher_order_debug 0
1233
Christoph Lameter0f389ec2008-04-14 18:53:02 +03001234static inline unsigned long slabs_node(struct kmem_cache *s, int node)
1235 { return 0; }
Alexander Beregalov26c02cf2009-06-11 14:08:48 +04001236static inline unsigned long node_nr_slabs(struct kmem_cache_node *n)
1237 { return 0; }
Christoph Lameter205ab992008-04-14 19:11:40 +03001238static inline void inc_slabs_node(struct kmem_cache *s, int node,
1239 int objects) {}
1240static inline void dec_slabs_node(struct kmem_cache *s, int node,
1241 int objects) {}
Christoph Lameter7d550c52010-08-25 14:07:16 -05001242
Andrey Ryabinin02e72cc2014-08-06 16:04:18 -07001243#endif /* CONFIG_SLUB_DEBUG */
1244
1245/*
1246 * Hooks for other subsystems that check memory allocations. In a typical
1247 * production configuration these hooks all should produce no code at all.
1248 */
Roman Bobnievd56791b2013-10-08 15:58:57 -07001249static inline void kmalloc_large_node_hook(void *ptr, size_t size, gfp_t flags)
1250{
1251 kmemleak_alloc(ptr, size, 1, flags);
Andrey Ryabinin0316bec2015-02-13 14:39:42 -08001252 kasan_kmalloc_large(ptr, size);
Roman Bobnievd56791b2013-10-08 15:58:57 -07001253}
1254
1255static inline void kfree_hook(const void *x)
1256{
1257 kmemleak_free(x);
Andrey Ryabinin0316bec2015-02-13 14:39:42 -08001258 kasan_kfree_large(x);
Roman Bobnievd56791b2013-10-08 15:58:57 -07001259}
1260
Vladimir Davydov8135be52014-12-12 16:56:38 -08001261static inline struct kmem_cache *slab_pre_alloc_hook(struct kmem_cache *s,
1262 gfp_t flags)
Roman Bobnievd56791b2013-10-08 15:58:57 -07001263{
Andrey Ryabinin02e72cc2014-08-06 16:04:18 -07001264 flags &= gfp_allowed_mask;
1265 lockdep_trace_alloc(flags);
1266 might_sleep_if(flags & __GFP_WAIT);
1267
Vladimir Davydov8135be52014-12-12 16:56:38 -08001268 if (should_failslab(s->object_size, flags, s->flags))
1269 return NULL;
1270
1271 return memcg_kmem_get_cache(s, flags);
Andrey Ryabinin02e72cc2014-08-06 16:04:18 -07001272}
1273
1274static inline void slab_post_alloc_hook(struct kmem_cache *s,
1275 gfp_t flags, void *object)
1276{
1277 flags &= gfp_allowed_mask;
1278 kmemcheck_slab_alloc(s, flags, object, slab_ksize(s));
1279 kmemleak_alloc_recursive(object, s->object_size, 1, s->flags, flags);
Vladimir Davydov8135be52014-12-12 16:56:38 -08001280 memcg_kmem_put_cache(s);
Andrey Ryabinin0316bec2015-02-13 14:39:42 -08001281 kasan_slab_alloc(s, object);
Roman Bobnievd56791b2013-10-08 15:58:57 -07001282}
Christoph Lameter7d550c52010-08-25 14:07:16 -05001283
Roman Bobnievd56791b2013-10-08 15:58:57 -07001284static inline void slab_free_hook(struct kmem_cache *s, void *x)
1285{
1286 kmemleak_free_recursive(x, s->flags);
Christoph Lameter7d550c52010-08-25 14:07:16 -05001287
Andrey Ryabinin02e72cc2014-08-06 16:04:18 -07001288 /*
1289 * Trouble is that we may no longer disable interrupts in the fast path
1290 * So in order to make the debug calls that expect irqs to be
1291 * disabled we need to disable interrupts temporarily.
1292 */
1293#if defined(CONFIG_KMEMCHECK) || defined(CONFIG_LOCKDEP)
1294 {
1295 unsigned long flags;
1296
1297 local_irq_save(flags);
1298 kmemcheck_slab_free(s, x, s->object_size);
1299 debug_check_no_locks_freed(x, s->object_size);
1300 local_irq_restore(flags);
1301 }
1302#endif
1303 if (!(s->flags & SLAB_DEBUG_OBJECTS))
1304 debug_check_no_obj_freed(x, s->object_size);
Andrey Ryabinin0316bec2015-02-13 14:39:42 -08001305
1306 kasan_slab_free(s, x);
Andrey Ryabinin02e72cc2014-08-06 16:04:18 -07001307}
Christoph Lameter205ab992008-04-14 19:11:40 +03001308
Christoph Lameter81819f02007-05-06 14:49:36 -07001309/*
1310 * Slab allocation and freeing
1311 */
Vladimir Davydov5dfb4172014-06-04 16:06:38 -07001312static inline struct page *alloc_slab_page(struct kmem_cache *s,
1313 gfp_t flags, int node, struct kmem_cache_order_objects oo)
Christoph Lameter65c33762008-04-14 19:11:40 +03001314{
Vladimir Davydov5dfb4172014-06-04 16:06:38 -07001315 struct page *page;
Christoph Lameter65c33762008-04-14 19:11:40 +03001316 int order = oo_order(oo);
1317
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001318 flags |= __GFP_NOTRACK;
1319
Vladimir Davydov5dfb4172014-06-04 16:06:38 -07001320 if (memcg_charge_slab(s, flags, order))
1321 return NULL;
1322
Christoph Lameter2154a332010-07-09 14:07:10 -05001323 if (node == NUMA_NO_NODE)
Vladimir Davydov5dfb4172014-06-04 16:06:38 -07001324 page = alloc_pages(flags, order);
Christoph Lameter65c33762008-04-14 19:11:40 +03001325 else
Vladimir Davydov5dfb4172014-06-04 16:06:38 -07001326 page = alloc_pages_exact_node(node, flags, order);
1327
1328 if (!page)
1329 memcg_uncharge_slab(s, order);
1330
1331 return page;
Christoph Lameter65c33762008-04-14 19:11:40 +03001332}
1333
Christoph Lameter81819f02007-05-06 14:49:36 -07001334static struct page *allocate_slab(struct kmem_cache *s, gfp_t flags, int node)
1335{
Pekka Enberg06428782008-01-07 23:20:27 -08001336 struct page *page;
Christoph Lameter834f3d12008-04-14 19:11:31 +03001337 struct kmem_cache_order_objects oo = s->oo;
Pekka Enbergba522702009-06-24 21:59:51 +03001338 gfp_t alloc_gfp;
Christoph Lameter81819f02007-05-06 14:49:36 -07001339
Christoph Lameter7e0528d2011-06-01 12:25:44 -05001340 flags &= gfp_allowed_mask;
1341
1342 if (flags & __GFP_WAIT)
1343 local_irq_enable();
1344
Christoph Lameterb7a49f02008-02-14 14:21:32 -08001345 flags |= s->allocflags;
Mel Gormane12ba742007-10-16 01:25:52 -07001346
Pekka Enbergba522702009-06-24 21:59:51 +03001347 /*
1348 * Let the initial higher-order allocation fail under memory pressure
1349 * so we fall-back to the minimum order allocation.
1350 */
1351 alloc_gfp = (flags | __GFP_NOWARN | __GFP_NORETRY) & ~__GFP_NOFAIL;
1352
Vladimir Davydov5dfb4172014-06-04 16:06:38 -07001353 page = alloc_slab_page(s, alloc_gfp, node, oo);
Christoph Lameter65c33762008-04-14 19:11:40 +03001354 if (unlikely(!page)) {
1355 oo = s->min;
Joonsoo Kim80c3a992014-03-12 17:26:20 +09001356 alloc_gfp = flags;
Christoph Lameter65c33762008-04-14 19:11:40 +03001357 /*
1358 * Allocation may have failed due to fragmentation.
1359 * Try a lower order alloc if possible
1360 */
Vladimir Davydov5dfb4172014-06-04 16:06:38 -07001361 page = alloc_slab_page(s, alloc_gfp, node, oo);
Christoph Lameter81819f02007-05-06 14:49:36 -07001362
Christoph Lameter7e0528d2011-06-01 12:25:44 -05001363 if (page)
1364 stat(s, ORDER_FALLBACK);
Christoph Lameter65c33762008-04-14 19:11:40 +03001365 }
Vegard Nossum5a896d92008-04-04 00:54:48 +02001366
David Rientjes737b7192012-07-09 14:00:38 -07001367 if (kmemcheck_enabled && page
Amerigo Wang5086c389c2009-08-19 21:44:13 +03001368 && !(s->flags & (SLAB_NOTRACK | DEBUG_DEFAULT_FLAGS))) {
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001369 int pages = 1 << oo_order(oo);
1370
Joonsoo Kim80c3a992014-03-12 17:26:20 +09001371 kmemcheck_alloc_shadow(page, oo_order(oo), alloc_gfp, node);
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001372
1373 /*
1374 * Objects from caches that have a constructor don't get
1375 * cleared when they're allocated, so we need to do it here.
1376 */
1377 if (s->ctor)
1378 kmemcheck_mark_uninitialized_pages(page, pages);
1379 else
1380 kmemcheck_mark_unallocated_pages(page, pages);
Vegard Nossum5a896d92008-04-04 00:54:48 +02001381 }
1382
David Rientjes737b7192012-07-09 14:00:38 -07001383 if (flags & __GFP_WAIT)
1384 local_irq_disable();
1385 if (!page)
1386 return NULL;
1387
Christoph Lameter834f3d12008-04-14 19:11:31 +03001388 page->objects = oo_objects(oo);
Christoph Lameter81819f02007-05-06 14:49:36 -07001389 mod_zone_page_state(page_zone(page),
1390 (s->flags & SLAB_RECLAIM_ACCOUNT) ?
1391 NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE,
Christoph Lameter65c33762008-04-14 19:11:40 +03001392 1 << oo_order(oo));
Christoph Lameter81819f02007-05-06 14:49:36 -07001393
1394 return page;
1395}
1396
1397static void setup_object(struct kmem_cache *s, struct page *page,
1398 void *object)
1399{
Christoph Lameter3ec09742007-05-16 22:11:00 -07001400 setup_object_debug(s, page, object);
Andrey Ryabinin0316bec2015-02-13 14:39:42 -08001401 if (unlikely(s->ctor)) {
1402 kasan_unpoison_object_data(s, object);
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07001403 s->ctor(object);
Andrey Ryabinin0316bec2015-02-13 14:39:42 -08001404 kasan_poison_object_data(s, object);
1405 }
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 *p;
Glauber Costa1f458cb2012-12-18 14:22:50 -08001413 int order;
Wei Yang54266642014-08-06 16:04:42 -07001414 int idx;
Christoph Lameter81819f02007-05-06 14:49:36 -07001415
Andrew Mortonc871ac42014-12-10 15:42:25 -08001416 if (unlikely(flags & GFP_SLAB_BUG_MASK)) {
1417 pr_emerg("gfp: %u\n", flags & GFP_SLAB_BUG_MASK);
1418 BUG();
1419 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001420
Christoph Lameter6cb06222007-10-16 01:25:41 -07001421 page = allocate_slab(s,
1422 flags & (GFP_RECLAIM_MASK | GFP_CONSTRAINT_MASK), node);
Christoph Lameter81819f02007-05-06 14:49:36 -07001423 if (!page)
1424 goto out;
1425
Glauber Costa1f458cb2012-12-18 14:22:50 -08001426 order = compound_order(page);
Christoph Lameter205ab992008-04-14 19:11:40 +03001427 inc_slabs_node(s, page_to_nid(page), page->objects);
Glauber Costa1b4f59e32012-10-22 18:05:36 +04001428 page->slab_cache = s;
Joonsoo Kimc03f94c2012-05-18 00:47:47 +09001429 __SetPageSlab(page);
Mel Gorman072bb0a2012-07-31 16:43:58 -07001430 if (page->pfmemalloc)
1431 SetPageSlabPfmemalloc(page);
Christoph Lameter81819f02007-05-06 14:49:36 -07001432
1433 start = page_address(page);
Christoph Lameter81819f02007-05-06 14:49:36 -07001434
1435 if (unlikely(s->flags & SLAB_POISON))
Glauber Costa1f458cb2012-12-18 14:22:50 -08001436 memset(start, POISON_INUSE, PAGE_SIZE << order);
Christoph Lameter81819f02007-05-06 14:49:36 -07001437
Andrey Ryabinin0316bec2015-02-13 14:39:42 -08001438 kasan_poison_slab(page);
1439
Wei Yang54266642014-08-06 16:04:42 -07001440 for_each_object_idx(p, idx, s, start, page->objects) {
1441 setup_object(s, page, p);
1442 if (likely(idx < page->objects))
1443 set_freepointer(s, p, p + s->size);
1444 else
1445 set_freepointer(s, p, NULL);
Christoph Lameter81819f02007-05-06 14:49:36 -07001446 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001447
1448 page->freelist = start;
Christoph Lametere6e82ea2011-08-09 16:12:24 -05001449 page->inuse = page->objects;
Christoph Lameter8cb0a502011-06-01 12:25:46 -05001450 page->frozen = 1;
Christoph Lameter81819f02007-05-06 14:49:36 -07001451out:
Christoph Lameter81819f02007-05-06 14:49:36 -07001452 return page;
1453}
1454
1455static void __free_slab(struct kmem_cache *s, struct page *page)
1456{
Christoph Lameter834f3d12008-04-14 19:11:31 +03001457 int order = compound_order(page);
1458 int pages = 1 << order;
Christoph Lameter81819f02007-05-06 14:49:36 -07001459
Christoph Lameteraf537b02010-07-09 14:07:14 -05001460 if (kmem_cache_debug(s)) {
Christoph Lameter81819f02007-05-06 14:49:36 -07001461 void *p;
1462
1463 slab_pad_check(s, page);
Christoph Lameter224a88b2008-04-14 19:11:31 +03001464 for_each_object(p, s, page_address(page),
1465 page->objects)
Christoph Lameterf7cb1932010-09-29 07:15:01 -05001466 check_object(s, page, p, SLUB_RED_INACTIVE);
Christoph Lameter81819f02007-05-06 14:49:36 -07001467 }
1468
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001469 kmemcheck_free_shadow(page, compound_order(page));
Vegard Nossum5a896d92008-04-04 00:54:48 +02001470
Christoph Lameter81819f02007-05-06 14:49:36 -07001471 mod_zone_page_state(page_zone(page),
1472 (s->flags & SLAB_RECLAIM_ACCOUNT) ?
1473 NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE,
Pekka Enberg06428782008-01-07 23:20:27 -08001474 -pages);
Christoph Lameter81819f02007-05-06 14:49:36 -07001475
Mel Gorman072bb0a2012-07-31 16:43:58 -07001476 __ClearPageSlabPfmemalloc(page);
Christoph Lameter49bd5222008-04-14 18:52:18 +03001477 __ClearPageSlab(page);
Glauber Costa1f458cb2012-12-18 14:22:50 -08001478
Mel Gorman22b751c2013-02-22 16:34:59 -08001479 page_mapcount_reset(page);
Nick Piggin1eb5ac62009-05-05 19:13:44 +10001480 if (current->reclaim_state)
1481 current->reclaim_state->reclaimed_slab += pages;
Vladimir Davydov5dfb4172014-06-04 16:06:38 -07001482 __free_pages(page, order);
1483 memcg_uncharge_slab(s, order);
Christoph Lameter81819f02007-05-06 14:49:36 -07001484}
1485
Lai Jiangshanda9a6382011-03-10 15:22:00 +08001486#define need_reserve_slab_rcu \
1487 (sizeof(((struct page *)NULL)->lru) < sizeof(struct rcu_head))
1488
Christoph Lameter81819f02007-05-06 14:49:36 -07001489static void rcu_free_slab(struct rcu_head *h)
1490{
1491 struct page *page;
1492
Lai Jiangshanda9a6382011-03-10 15:22:00 +08001493 if (need_reserve_slab_rcu)
1494 page = virt_to_head_page(h);
1495 else
1496 page = container_of((struct list_head *)h, struct page, lru);
1497
Glauber Costa1b4f59e32012-10-22 18:05:36 +04001498 __free_slab(page->slab_cache, page);
Christoph Lameter81819f02007-05-06 14:49:36 -07001499}
1500
1501static void free_slab(struct kmem_cache *s, struct page *page)
1502{
1503 if (unlikely(s->flags & SLAB_DESTROY_BY_RCU)) {
Lai Jiangshanda9a6382011-03-10 15:22:00 +08001504 struct rcu_head *head;
1505
1506 if (need_reserve_slab_rcu) {
1507 int order = compound_order(page);
1508 int offset = (PAGE_SIZE << order) - s->reserved;
1509
1510 VM_BUG_ON(s->reserved != sizeof(*head));
1511 head = page_address(page) + offset;
1512 } else {
1513 /*
1514 * RCU free overloads the RCU head over the LRU
1515 */
1516 head = (void *)&page->lru;
1517 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001518
1519 call_rcu(head, rcu_free_slab);
1520 } else
1521 __free_slab(s, page);
1522}
1523
1524static void discard_slab(struct kmem_cache *s, struct page *page)
1525{
Christoph Lameter205ab992008-04-14 19:11:40 +03001526 dec_slabs_node(s, page_to_nid(page), page->objects);
Christoph Lameter81819f02007-05-06 14:49:36 -07001527 free_slab(s, page);
1528}
1529
1530/*
Christoph Lameter5cc6eee2011-06-01 12:25:50 -05001531 * Management of partially allocated slabs.
Christoph Lameter81819f02007-05-06 14:49:36 -07001532 */
Steven Rostedt1e4dd942014-02-10 14:25:46 -08001533static inline void
1534__add_partial(struct kmem_cache_node *n, struct page *page, int tail)
Christoph Lameter81819f02007-05-06 14:49:36 -07001535{
Christoph Lametere95eed52007-05-06 14:49:44 -07001536 n->nr_partial++;
Shaohua Li136333d2011-08-24 08:57:52 +08001537 if (tail == DEACTIVATE_TO_TAIL)
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001538 list_add_tail(&page->lru, &n->partial);
1539 else
1540 list_add(&page->lru, &n->partial);
Christoph Lameter81819f02007-05-06 14:49:36 -07001541}
1542
Steven Rostedt1e4dd942014-02-10 14:25:46 -08001543static inline void add_partial(struct kmem_cache_node *n,
1544 struct page *page, int tail)
1545{
1546 lockdep_assert_held(&n->list_lock);
1547 __add_partial(n, page, tail);
1548}
1549
1550static inline void
1551__remove_partial(struct kmem_cache_node *n, struct page *page)
1552{
1553 list_del(&page->lru);
1554 n->nr_partial--;
1555}
1556
Christoph Lameter5cc6eee2011-06-01 12:25:50 -05001557static inline void remove_partial(struct kmem_cache_node *n,
Christoph Lameter62e346a2010-09-28 08:10:28 -05001558 struct page *page)
1559{
Peter Zijlstrac65c1872014-01-10 13:23:49 +01001560 lockdep_assert_held(&n->list_lock);
Steven Rostedt1e4dd942014-02-10 14:25:46 -08001561 __remove_partial(n, page);
Christoph Lameter62e346a2010-09-28 08:10:28 -05001562}
1563
Christoph Lameter81819f02007-05-06 14:49:36 -07001564/*
Christoph Lameter7ced3712012-05-09 10:09:53 -05001565 * Remove slab from the partial list, freeze it and
1566 * return the pointer to the freelist.
Christoph Lameter81819f02007-05-06 14:49:36 -07001567 *
Christoph Lameter497b66f2011-08-09 16:12:26 -05001568 * Returns a list of objects or NULL if it fails.
Christoph Lameter81819f02007-05-06 14:49:36 -07001569 */
Christoph Lameter497b66f2011-08-09 16:12:26 -05001570static inline void *acquire_slab(struct kmem_cache *s,
Christoph Lameteracd19fd2011-08-09 16:12:25 -05001571 struct kmem_cache_node *n, struct page *page,
Joonsoo Kim633b0762013-01-21 17:01:25 +09001572 int mode, int *objects)
Christoph Lameter81819f02007-05-06 14:49:36 -07001573{
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001574 void *freelist;
1575 unsigned long counters;
1576 struct page new;
1577
Peter Zijlstrac65c1872014-01-10 13:23:49 +01001578 lockdep_assert_held(&n->list_lock);
1579
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001580 /*
1581 * Zap the freelist and set the frozen bit.
1582 * The old freelist is the list of objects for the
1583 * per cpu allocation list.
1584 */
Christoph Lameter7ced3712012-05-09 10:09:53 -05001585 freelist = page->freelist;
1586 counters = page->counters;
1587 new.counters = counters;
Joonsoo Kim633b0762013-01-21 17:01:25 +09001588 *objects = new.objects - new.inuse;
Pekka Enberg23910c52012-06-04 10:14:58 +03001589 if (mode) {
Christoph Lameter7ced3712012-05-09 10:09:53 -05001590 new.inuse = page->objects;
Pekka Enberg23910c52012-06-04 10:14:58 +03001591 new.freelist = NULL;
1592 } else {
1593 new.freelist = freelist;
1594 }
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001595
Dave Hansena0132ac2014-01-29 14:05:50 -08001596 VM_BUG_ON(new.frozen);
Christoph Lameter7ced3712012-05-09 10:09:53 -05001597 new.frozen = 1;
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001598
Christoph Lameter7ced3712012-05-09 10:09:53 -05001599 if (!__cmpxchg_double_slab(s, page,
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001600 freelist, counters,
Joonsoo Kim02d76332012-05-17 00:13:02 +09001601 new.freelist, new.counters,
Christoph Lameter7ced3712012-05-09 10:09:53 -05001602 "acquire_slab"))
Christoph Lameter7ced3712012-05-09 10:09:53 -05001603 return NULL;
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001604
1605 remove_partial(n, page);
Christoph Lameter7ced3712012-05-09 10:09:53 -05001606 WARN_ON(!freelist);
Christoph Lameter49e22582011-08-09 16:12:27 -05001607 return freelist;
Christoph Lameter81819f02007-05-06 14:49:36 -07001608}
1609
Joonsoo Kim633b0762013-01-21 17:01:25 +09001610static void put_cpu_partial(struct kmem_cache *s, struct page *page, int drain);
Joonsoo Kim8ba00bb2012-09-17 14:09:09 -07001611static inline bool pfmemalloc_match(struct page *page, gfp_t gfpflags);
Christoph Lameter49e22582011-08-09 16:12:27 -05001612
Christoph Lameter81819f02007-05-06 14:49:36 -07001613/*
Christoph Lameter672bba32007-05-09 02:32:39 -07001614 * Try to allocate a partial slab from a specific node.
Christoph Lameter81819f02007-05-06 14:49:36 -07001615 */
Joonsoo Kim8ba00bb2012-09-17 14:09:09 -07001616static void *get_partial_node(struct kmem_cache *s, struct kmem_cache_node *n,
1617 struct kmem_cache_cpu *c, gfp_t flags)
Christoph Lameter81819f02007-05-06 14:49:36 -07001618{
Christoph Lameter49e22582011-08-09 16:12:27 -05001619 struct page *page, *page2;
1620 void *object = NULL;
Joonsoo Kim633b0762013-01-21 17:01:25 +09001621 int available = 0;
1622 int objects;
Christoph Lameter81819f02007-05-06 14:49:36 -07001623
1624 /*
1625 * Racy check. If we mistakenly see no partial slabs then we
1626 * just allocate an empty slab. If we mistakenly try to get a
Christoph Lameter672bba32007-05-09 02:32:39 -07001627 * partial slab and there is none available then get_partials()
1628 * will return NULL.
Christoph Lameter81819f02007-05-06 14:49:36 -07001629 */
1630 if (!n || !n->nr_partial)
1631 return NULL;
1632
1633 spin_lock(&n->list_lock);
Christoph Lameter49e22582011-08-09 16:12:27 -05001634 list_for_each_entry_safe(page, page2, &n->partial, lru) {
Joonsoo Kim8ba00bb2012-09-17 14:09:09 -07001635 void *t;
Christoph Lameter49e22582011-08-09 16:12:27 -05001636
Joonsoo Kim8ba00bb2012-09-17 14:09:09 -07001637 if (!pfmemalloc_match(page, flags))
1638 continue;
1639
Joonsoo Kim633b0762013-01-21 17:01:25 +09001640 t = acquire_slab(s, n, page, object == NULL, &objects);
Christoph Lameter49e22582011-08-09 16:12:27 -05001641 if (!t)
1642 break;
1643
Joonsoo Kim633b0762013-01-21 17:01:25 +09001644 available += objects;
Alex,Shi12d79632011-09-07 10:26:36 +08001645 if (!object) {
Christoph Lameter49e22582011-08-09 16:12:27 -05001646 c->page = page;
Christoph Lameter49e22582011-08-09 16:12:27 -05001647 stat(s, ALLOC_FROM_PARTIAL);
Christoph Lameter49e22582011-08-09 16:12:27 -05001648 object = t;
Christoph Lameter49e22582011-08-09 16:12:27 -05001649 } else {
Joonsoo Kim633b0762013-01-21 17:01:25 +09001650 put_cpu_partial(s, page, 0);
Alex Shi8028dce2012-02-03 23:34:56 +08001651 stat(s, CPU_PARTIAL_NODE);
Christoph Lameter49e22582011-08-09 16:12:27 -05001652 }
Joonsoo Kim345c9052013-06-19 14:05:52 +09001653 if (!kmem_cache_has_cpu_partial(s)
1654 || available > s->cpu_partial / 2)
Christoph Lameter49e22582011-08-09 16:12:27 -05001655 break;
1656
Christoph Lameter497b66f2011-08-09 16:12:26 -05001657 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001658 spin_unlock(&n->list_lock);
Christoph Lameter497b66f2011-08-09 16:12:26 -05001659 return object;
Christoph Lameter81819f02007-05-06 14:49:36 -07001660}
1661
1662/*
Christoph Lameter672bba32007-05-09 02:32:39 -07001663 * Get a page from somewhere. Search in increasing NUMA distances.
Christoph Lameter81819f02007-05-06 14:49:36 -07001664 */
Joonsoo Kimde3ec032012-01-27 00:12:23 -08001665static void *get_any_partial(struct kmem_cache *s, gfp_t flags,
Christoph Lameteracd19fd2011-08-09 16:12:25 -05001666 struct kmem_cache_cpu *c)
Christoph Lameter81819f02007-05-06 14:49:36 -07001667{
1668#ifdef CONFIG_NUMA
1669 struct zonelist *zonelist;
Mel Gormandd1a2392008-04-28 02:12:17 -07001670 struct zoneref *z;
Mel Gorman54a6eb52008-04-28 02:12:16 -07001671 struct zone *zone;
1672 enum zone_type high_zoneidx = gfp_zone(flags);
Christoph Lameter497b66f2011-08-09 16:12:26 -05001673 void *object;
Mel Gormancc9a6c82012-03-21 16:34:11 -07001674 unsigned int cpuset_mems_cookie;
Christoph Lameter81819f02007-05-06 14:49:36 -07001675
1676 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07001677 * The defrag ratio allows a configuration of the tradeoffs between
1678 * inter node defragmentation and node local allocations. A lower
1679 * defrag_ratio increases the tendency to do local allocations
1680 * instead of attempting to obtain partial slabs from other nodes.
Christoph Lameter81819f02007-05-06 14:49:36 -07001681 *
Christoph Lameter672bba32007-05-09 02:32:39 -07001682 * If the defrag_ratio is set to 0 then kmalloc() always
1683 * returns node local objects. If the ratio is higher then kmalloc()
1684 * may return off node objects because partial slabs are obtained
1685 * from other nodes and filled up.
Christoph Lameter81819f02007-05-06 14:49:36 -07001686 *
Christoph Lameter6446faa2008-02-15 23:45:26 -08001687 * If /sys/kernel/slab/xx/defrag_ratio is set to 100 (which makes
Christoph Lameter672bba32007-05-09 02:32:39 -07001688 * defrag_ratio = 1000) then every (well almost) allocation will
1689 * first attempt to defrag slab caches on other nodes. This means
1690 * scanning over all nodes to look for partial slabs which may be
1691 * expensive if we do it every time we are trying to find a slab
1692 * with available objects.
Christoph Lameter81819f02007-05-06 14:49:36 -07001693 */
Christoph Lameter98246012008-01-07 23:20:26 -08001694 if (!s->remote_node_defrag_ratio ||
1695 get_cycles() % 1024 > s->remote_node_defrag_ratio)
Christoph Lameter81819f02007-05-06 14:49:36 -07001696 return NULL;
1697
Mel Gormancc9a6c82012-03-21 16:34:11 -07001698 do {
Mel Gormand26914d2014-04-03 14:47:24 -07001699 cpuset_mems_cookie = read_mems_allowed_begin();
David Rientjes2a389612014-04-07 15:37:29 -07001700 zonelist = node_zonelist(mempolicy_slab_node(), flags);
Mel Gormancc9a6c82012-03-21 16:34:11 -07001701 for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) {
1702 struct kmem_cache_node *n;
Christoph Lameter81819f02007-05-06 14:49:36 -07001703
Mel Gormancc9a6c82012-03-21 16:34:11 -07001704 n = get_node(s, zone_to_nid(zone));
Christoph Lameter81819f02007-05-06 14:49:36 -07001705
Vladimir Davydovdee2f8a2014-12-12 16:58:28 -08001706 if (n && cpuset_zone_allowed(zone, flags) &&
Mel Gormancc9a6c82012-03-21 16:34:11 -07001707 n->nr_partial > s->min_partial) {
Joonsoo Kim8ba00bb2012-09-17 14:09:09 -07001708 object = get_partial_node(s, n, c, flags);
Mel Gormancc9a6c82012-03-21 16:34:11 -07001709 if (object) {
1710 /*
Mel Gormand26914d2014-04-03 14:47:24 -07001711 * Don't check read_mems_allowed_retry()
1712 * here - if mems_allowed was updated in
1713 * parallel, that was a harmless race
1714 * between allocation and the cpuset
1715 * update
Mel Gormancc9a6c82012-03-21 16:34:11 -07001716 */
Mel Gormancc9a6c82012-03-21 16:34:11 -07001717 return object;
1718 }
Miao Xiec0ff7452010-05-24 14:32:08 -07001719 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001720 }
Mel Gormand26914d2014-04-03 14:47:24 -07001721 } while (read_mems_allowed_retry(cpuset_mems_cookie));
Christoph Lameter81819f02007-05-06 14:49:36 -07001722#endif
1723 return NULL;
1724}
1725
1726/*
1727 * Get a partial page, lock it and return it.
1728 */
Christoph Lameter497b66f2011-08-09 16:12:26 -05001729static void *get_partial(struct kmem_cache *s, gfp_t flags, int node,
Christoph Lameteracd19fd2011-08-09 16:12:25 -05001730 struct kmem_cache_cpu *c)
Christoph Lameter81819f02007-05-06 14:49:36 -07001731{
Christoph Lameter497b66f2011-08-09 16:12:26 -05001732 void *object;
Joonsoo Kima561ce02014-10-09 15:26:15 -07001733 int searchnode = node;
1734
1735 if (node == NUMA_NO_NODE)
1736 searchnode = numa_mem_id();
1737 else if (!node_present_pages(node))
1738 searchnode = node_to_mem_node(node);
Christoph Lameter81819f02007-05-06 14:49:36 -07001739
Joonsoo Kim8ba00bb2012-09-17 14:09:09 -07001740 object = get_partial_node(s, get_node(s, searchnode), c, flags);
Christoph Lameter497b66f2011-08-09 16:12:26 -05001741 if (object || node != NUMA_NO_NODE)
1742 return object;
Christoph Lameter81819f02007-05-06 14:49:36 -07001743
Christoph Lameteracd19fd2011-08-09 16:12:25 -05001744 return get_any_partial(s, flags, c);
Christoph Lameter81819f02007-05-06 14:49:36 -07001745}
1746
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001747#ifdef CONFIG_PREEMPT
1748/*
1749 * Calculate the next globally unique transaction for disambiguiation
1750 * during cmpxchg. The transactions start with the cpu number and are then
1751 * incremented by CONFIG_NR_CPUS.
1752 */
1753#define TID_STEP roundup_pow_of_two(CONFIG_NR_CPUS)
1754#else
1755/*
1756 * No preemption supported therefore also no need to check for
1757 * different cpus.
1758 */
1759#define TID_STEP 1
1760#endif
1761
1762static inline unsigned long next_tid(unsigned long tid)
1763{
1764 return tid + TID_STEP;
1765}
1766
1767static inline unsigned int tid_to_cpu(unsigned long tid)
1768{
1769 return tid % TID_STEP;
1770}
1771
1772static inline unsigned long tid_to_event(unsigned long tid)
1773{
1774 return tid / TID_STEP;
1775}
1776
1777static inline unsigned int init_tid(int cpu)
1778{
1779 return cpu;
1780}
1781
1782static inline void note_cmpxchg_failure(const char *n,
1783 const struct kmem_cache *s, unsigned long tid)
1784{
1785#ifdef SLUB_DEBUG_CMPXCHG
1786 unsigned long actual_tid = __this_cpu_read(s->cpu_slab->tid);
1787
Fabian Frederickf9f58282014-06-04 16:06:34 -07001788 pr_info("%s %s: cmpxchg redo ", n, s->name);
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001789
1790#ifdef CONFIG_PREEMPT
1791 if (tid_to_cpu(tid) != tid_to_cpu(actual_tid))
Fabian Frederickf9f58282014-06-04 16:06:34 -07001792 pr_warn("due to cpu change %d -> %d\n",
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001793 tid_to_cpu(tid), tid_to_cpu(actual_tid));
1794 else
1795#endif
1796 if (tid_to_event(tid) != tid_to_event(actual_tid))
Fabian Frederickf9f58282014-06-04 16:06:34 -07001797 pr_warn("due to cpu running other code. Event %ld->%ld\n",
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001798 tid_to_event(tid), tid_to_event(actual_tid));
1799 else
Fabian Frederickf9f58282014-06-04 16:06:34 -07001800 pr_warn("for unknown reason: actual=%lx was=%lx target=%lx\n",
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001801 actual_tid, tid, next_tid(tid));
1802#endif
Christoph Lameter4fdccdf2011-03-22 13:35:00 -05001803 stat(s, CMPXCHG_DOUBLE_CPU_FAIL);
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001804}
1805
Fengguang Wu788e1aa2012-09-28 16:34:05 +08001806static void init_kmem_cache_cpus(struct kmem_cache *s)
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001807{
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001808 int cpu;
1809
1810 for_each_possible_cpu(cpu)
1811 per_cpu_ptr(s->cpu_slab, cpu)->tid = init_tid(cpu);
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001812}
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001813
1814/*
1815 * Remove the cpu slab
1816 */
Chen Gangd0e0ac92013-07-15 09:05:29 +08001817static void deactivate_slab(struct kmem_cache *s, struct page *page,
1818 void *freelist)
Christoph Lameter81819f02007-05-06 14:49:36 -07001819{
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001820 enum slab_modes { M_NONE, M_PARTIAL, M_FULL, M_FREE };
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001821 struct kmem_cache_node *n = get_node(s, page_to_nid(page));
1822 int lock = 0;
1823 enum slab_modes l = M_NONE, m = M_NONE;
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001824 void *nextfree;
Shaohua Li136333d2011-08-24 08:57:52 +08001825 int tail = DEACTIVATE_TO_HEAD;
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001826 struct page new;
1827 struct page old;
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001828
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001829 if (page->freelist) {
Christoph Lameter84e554e62009-12-18 16:26:23 -06001830 stat(s, DEACTIVATE_REMOTE_FREES);
Shaohua Li136333d2011-08-24 08:57:52 +08001831 tail = DEACTIVATE_TO_TAIL;
Christoph Lameter894b8782007-05-10 03:15:16 -07001832 }
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001833
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001834 /*
1835 * Stage one: Free all available per cpu objects back
1836 * to the page freelist while it is still frozen. Leave the
1837 * last one.
1838 *
1839 * There is no need to take the list->lock because the page
1840 * is still frozen.
1841 */
1842 while (freelist && (nextfree = get_freepointer(s, freelist))) {
1843 void *prior;
1844 unsigned long counters;
1845
1846 do {
1847 prior = page->freelist;
1848 counters = page->counters;
1849 set_freepointer(s, freelist, prior);
1850 new.counters = counters;
1851 new.inuse--;
Dave Hansena0132ac2014-01-29 14:05:50 -08001852 VM_BUG_ON(!new.frozen);
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001853
Christoph Lameter1d071712011-07-14 12:49:12 -05001854 } while (!__cmpxchg_double_slab(s, page,
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001855 prior, counters,
1856 freelist, new.counters,
1857 "drain percpu freelist"));
1858
1859 freelist = nextfree;
1860 }
1861
1862 /*
1863 * Stage two: Ensure that the page is unfrozen while the
1864 * list presence reflects the actual number of objects
1865 * during unfreeze.
1866 *
1867 * We setup the list membership and then perform a cmpxchg
1868 * with the count. If there is a mismatch then the page
1869 * is not unfrozen but the page is on the wrong list.
1870 *
1871 * Then we restart the process which may have to remove
1872 * the page from the list that we just put it on again
1873 * because the number of objects in the slab may have
1874 * changed.
1875 */
1876redo:
1877
1878 old.freelist = page->freelist;
1879 old.counters = page->counters;
Dave Hansena0132ac2014-01-29 14:05:50 -08001880 VM_BUG_ON(!old.frozen);
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001881
1882 /* Determine target state of the slab */
1883 new.counters = old.counters;
1884 if (freelist) {
1885 new.inuse--;
1886 set_freepointer(s, freelist, old.freelist);
1887 new.freelist = freelist;
1888 } else
1889 new.freelist = old.freelist;
1890
1891 new.frozen = 0;
1892
Joonsoo Kim8a5b20a2014-07-02 15:22:35 -07001893 if (!new.inuse && n->nr_partial >= s->min_partial)
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001894 m = M_FREE;
1895 else if (new.freelist) {
1896 m = M_PARTIAL;
1897 if (!lock) {
1898 lock = 1;
1899 /*
1900 * Taking the spinlock removes the possiblity
1901 * that acquire_slab() will see a slab page that
1902 * is frozen
1903 */
1904 spin_lock(&n->list_lock);
1905 }
1906 } else {
1907 m = M_FULL;
1908 if (kmem_cache_debug(s) && !lock) {
1909 lock = 1;
1910 /*
1911 * This also ensures that the scanning of full
1912 * slabs from diagnostic functions will not see
1913 * any frozen slabs.
1914 */
1915 spin_lock(&n->list_lock);
1916 }
1917 }
1918
1919 if (l != m) {
1920
1921 if (l == M_PARTIAL)
1922
1923 remove_partial(n, page);
1924
1925 else if (l == M_FULL)
1926
Peter Zijlstrac65c1872014-01-10 13:23:49 +01001927 remove_full(s, n, page);
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001928
1929 if (m == M_PARTIAL) {
1930
1931 add_partial(n, page, tail);
Shaohua Li136333d2011-08-24 08:57:52 +08001932 stat(s, tail);
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001933
1934 } else if (m == M_FULL) {
1935
1936 stat(s, DEACTIVATE_FULL);
1937 add_full(s, n, page);
1938
1939 }
1940 }
1941
1942 l = m;
Christoph Lameter1d071712011-07-14 12:49:12 -05001943 if (!__cmpxchg_double_slab(s, page,
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001944 old.freelist, old.counters,
1945 new.freelist, new.counters,
1946 "unfreezing slab"))
1947 goto redo;
1948
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001949 if (lock)
1950 spin_unlock(&n->list_lock);
1951
1952 if (m == M_FREE) {
1953 stat(s, DEACTIVATE_EMPTY);
1954 discard_slab(s, page);
1955 stat(s, FREE_SLAB);
1956 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001957}
1958
Joonsoo Kimd24ac772012-05-18 22:01:17 +09001959/*
1960 * Unfreeze all the cpu partial slabs.
1961 *
Christoph Lameter59a09912012-11-28 16:23:00 +00001962 * This function must be called with interrupts disabled
1963 * for the cpu using c (or some other guarantee must be there
1964 * to guarantee no concurrent accesses).
Joonsoo Kimd24ac772012-05-18 22:01:17 +09001965 */
Christoph Lameter59a09912012-11-28 16:23:00 +00001966static void unfreeze_partials(struct kmem_cache *s,
1967 struct kmem_cache_cpu *c)
Christoph Lameter49e22582011-08-09 16:12:27 -05001968{
Joonsoo Kim345c9052013-06-19 14:05:52 +09001969#ifdef CONFIG_SLUB_CPU_PARTIAL
Joonsoo Kim43d77862012-06-09 02:23:16 +09001970 struct kmem_cache_node *n = NULL, *n2 = NULL;
Shaohua Li9ada1932011-11-14 13:34:13 +08001971 struct page *page, *discard_page = NULL;
Christoph Lameter49e22582011-08-09 16:12:27 -05001972
1973 while ((page = c->partial)) {
Christoph Lameter49e22582011-08-09 16:12:27 -05001974 struct page new;
1975 struct page old;
1976
1977 c->partial = page->next;
Joonsoo Kim43d77862012-06-09 02:23:16 +09001978
1979 n2 = get_node(s, page_to_nid(page));
1980 if (n != n2) {
1981 if (n)
1982 spin_unlock(&n->list_lock);
1983
1984 n = n2;
1985 spin_lock(&n->list_lock);
1986 }
Christoph Lameter49e22582011-08-09 16:12:27 -05001987
1988 do {
1989
1990 old.freelist = page->freelist;
1991 old.counters = page->counters;
Dave Hansena0132ac2014-01-29 14:05:50 -08001992 VM_BUG_ON(!old.frozen);
Christoph Lameter49e22582011-08-09 16:12:27 -05001993
1994 new.counters = old.counters;
1995 new.freelist = old.freelist;
1996
1997 new.frozen = 0;
1998
Joonsoo Kimd24ac772012-05-18 22:01:17 +09001999 } while (!__cmpxchg_double_slab(s, page,
Christoph Lameter49e22582011-08-09 16:12:27 -05002000 old.freelist, old.counters,
2001 new.freelist, new.counters,
2002 "unfreezing slab"));
2003
Joonsoo Kim8a5b20a2014-07-02 15:22:35 -07002004 if (unlikely(!new.inuse && n->nr_partial >= s->min_partial)) {
Shaohua Li9ada1932011-11-14 13:34:13 +08002005 page->next = discard_page;
2006 discard_page = page;
Joonsoo Kim43d77862012-06-09 02:23:16 +09002007 } else {
2008 add_partial(n, page, DEACTIVATE_TO_TAIL);
2009 stat(s, FREE_ADD_PARTIAL);
Christoph Lameter49e22582011-08-09 16:12:27 -05002010 }
2011 }
2012
2013 if (n)
2014 spin_unlock(&n->list_lock);
Shaohua Li9ada1932011-11-14 13:34:13 +08002015
2016 while (discard_page) {
2017 page = discard_page;
2018 discard_page = discard_page->next;
2019
2020 stat(s, DEACTIVATE_EMPTY);
2021 discard_slab(s, page);
2022 stat(s, FREE_SLAB);
2023 }
Joonsoo Kim345c9052013-06-19 14:05:52 +09002024#endif
Christoph Lameter49e22582011-08-09 16:12:27 -05002025}
2026
2027/*
2028 * Put a page that was just frozen (in __slab_free) into a partial page
2029 * slot if available. This is done without interrupts disabled and without
2030 * preemption disabled. The cmpxchg is racy and may put the partial page
2031 * onto a random cpus partial slot.
2032 *
2033 * If we did not find a slot then simply move all the partials to the
2034 * per node partial list.
2035 */
Joonsoo Kim633b0762013-01-21 17:01:25 +09002036static void put_cpu_partial(struct kmem_cache *s, struct page *page, int drain)
Christoph Lameter49e22582011-08-09 16:12:27 -05002037{
Joonsoo Kim345c9052013-06-19 14:05:52 +09002038#ifdef CONFIG_SLUB_CPU_PARTIAL
Christoph Lameter49e22582011-08-09 16:12:27 -05002039 struct page *oldpage;
2040 int pages;
2041 int pobjects;
2042
Vladimir Davydovd6e0b7f2015-02-12 14:59:47 -08002043 preempt_disable();
Christoph Lameter49e22582011-08-09 16:12:27 -05002044 do {
2045 pages = 0;
2046 pobjects = 0;
2047 oldpage = this_cpu_read(s->cpu_slab->partial);
2048
2049 if (oldpage) {
2050 pobjects = oldpage->pobjects;
2051 pages = oldpage->pages;
2052 if (drain && pobjects > s->cpu_partial) {
2053 unsigned long flags;
2054 /*
2055 * partial array is full. Move the existing
2056 * set to the per node partial list.
2057 */
2058 local_irq_save(flags);
Christoph Lameter59a09912012-11-28 16:23:00 +00002059 unfreeze_partials(s, this_cpu_ptr(s->cpu_slab));
Christoph Lameter49e22582011-08-09 16:12:27 -05002060 local_irq_restore(flags);
Joonsoo Kime24fc412012-06-23 03:22:38 +09002061 oldpage = NULL;
Christoph Lameter49e22582011-08-09 16:12:27 -05002062 pobjects = 0;
2063 pages = 0;
Alex Shi8028dce2012-02-03 23:34:56 +08002064 stat(s, CPU_PARTIAL_DRAIN);
Christoph Lameter49e22582011-08-09 16:12:27 -05002065 }
2066 }
2067
2068 pages++;
2069 pobjects += page->objects - page->inuse;
2070
2071 page->pages = pages;
2072 page->pobjects = pobjects;
2073 page->next = oldpage;
2074
Chen Gangd0e0ac92013-07-15 09:05:29 +08002075 } while (this_cpu_cmpxchg(s->cpu_slab->partial, oldpage, page)
2076 != oldpage);
Vladimir Davydovd6e0b7f2015-02-12 14:59:47 -08002077 if (unlikely(!s->cpu_partial)) {
2078 unsigned long flags;
2079
2080 local_irq_save(flags);
2081 unfreeze_partials(s, this_cpu_ptr(s->cpu_slab));
2082 local_irq_restore(flags);
2083 }
2084 preempt_enable();
Joonsoo Kim345c9052013-06-19 14:05:52 +09002085#endif
Christoph Lameter49e22582011-08-09 16:12:27 -05002086}
2087
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002088static inline void flush_slab(struct kmem_cache *s, struct kmem_cache_cpu *c)
Christoph Lameter81819f02007-05-06 14:49:36 -07002089{
Christoph Lameter84e554e62009-12-18 16:26:23 -06002090 stat(s, CPUSLAB_FLUSH);
Christoph Lameterc17dda42012-05-09 10:09:57 -05002091 deactivate_slab(s, c->page, c->freelist);
2092
2093 c->tid = next_tid(c->tid);
2094 c->page = NULL;
2095 c->freelist = NULL;
Christoph Lameter81819f02007-05-06 14:49:36 -07002096}
2097
2098/*
2099 * Flush cpu slab.
Christoph Lameter6446faa2008-02-15 23:45:26 -08002100 *
Christoph Lameter81819f02007-05-06 14:49:36 -07002101 * Called from IPI handler with interrupts disabled.
2102 */
Christoph Lameter0c710012007-07-17 04:03:24 -07002103static inline void __flush_cpu_slab(struct kmem_cache *s, int cpu)
Christoph Lameter81819f02007-05-06 14:49:36 -07002104{
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06002105 struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu);
Christoph Lameter81819f02007-05-06 14:49:36 -07002106
Christoph Lameter49e22582011-08-09 16:12:27 -05002107 if (likely(c)) {
2108 if (c->page)
2109 flush_slab(s, c);
2110
Christoph Lameter59a09912012-11-28 16:23:00 +00002111 unfreeze_partials(s, c);
Christoph Lameter49e22582011-08-09 16:12:27 -05002112 }
Christoph Lameter81819f02007-05-06 14:49:36 -07002113}
2114
2115static void flush_cpu_slab(void *d)
2116{
2117 struct kmem_cache *s = d;
Christoph Lameter81819f02007-05-06 14:49:36 -07002118
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002119 __flush_cpu_slab(s, smp_processor_id());
Christoph Lameter81819f02007-05-06 14:49:36 -07002120}
2121
Gilad Ben-Yossefa8364d52012-03-28 14:42:44 -07002122static bool has_cpu_slab(int cpu, void *info)
2123{
2124 struct kmem_cache *s = info;
2125 struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu);
2126
majianpeng02e1a9c2012-05-17 17:03:26 -07002127 return c->page || c->partial;
Gilad Ben-Yossefa8364d52012-03-28 14:42:44 -07002128}
2129
Christoph Lameter81819f02007-05-06 14:49:36 -07002130static void flush_all(struct kmem_cache *s)
2131{
Gilad Ben-Yossefa8364d52012-03-28 14:42:44 -07002132 on_each_cpu_cond(has_cpu_slab, flush_cpu_slab, s, 1, GFP_ATOMIC);
Christoph Lameter81819f02007-05-06 14:49:36 -07002133}
2134
2135/*
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002136 * Check if the objects in a per cpu structure fit numa
2137 * locality expectations.
2138 */
Christoph Lameter57d437d2012-05-09 10:09:59 -05002139static inline int node_match(struct page *page, int node)
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002140{
2141#ifdef CONFIG_NUMA
Christoph Lameter4d7868e2013-01-23 21:45:47 +00002142 if (!page || (node != NUMA_NO_NODE && page_to_nid(page) != node))
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002143 return 0;
2144#endif
2145 return 1;
2146}
2147
David Rientjes9a02d692014-06-04 16:06:36 -07002148#ifdef CONFIG_SLUB_DEBUG
Pekka Enberg781b2ba2009-06-10 18:50:32 +03002149static int count_free(struct page *page)
2150{
2151 return page->objects - page->inuse;
2152}
2153
David Rientjes9a02d692014-06-04 16:06:36 -07002154static inline unsigned long node_nr_objs(struct kmem_cache_node *n)
2155{
2156 return atomic_long_read(&n->total_objects);
2157}
2158#endif /* CONFIG_SLUB_DEBUG */
2159
2160#if defined(CONFIG_SLUB_DEBUG) || defined(CONFIG_SYSFS)
Pekka Enberg781b2ba2009-06-10 18:50:32 +03002161static unsigned long count_partial(struct kmem_cache_node *n,
2162 int (*get_count)(struct page *))
2163{
2164 unsigned long flags;
2165 unsigned long x = 0;
2166 struct page *page;
2167
2168 spin_lock_irqsave(&n->list_lock, flags);
2169 list_for_each_entry(page, &n->partial, lru)
2170 x += get_count(page);
2171 spin_unlock_irqrestore(&n->list_lock, flags);
2172 return x;
2173}
David Rientjes9a02d692014-06-04 16:06:36 -07002174#endif /* CONFIG_SLUB_DEBUG || CONFIG_SYSFS */
Alexander Beregalov26c02cf2009-06-11 14:08:48 +04002175
Pekka Enberg781b2ba2009-06-10 18:50:32 +03002176static noinline void
2177slab_out_of_memory(struct kmem_cache *s, gfp_t gfpflags, int nid)
2178{
David Rientjes9a02d692014-06-04 16:06:36 -07002179#ifdef CONFIG_SLUB_DEBUG
2180 static DEFINE_RATELIMIT_STATE(slub_oom_rs, DEFAULT_RATELIMIT_INTERVAL,
2181 DEFAULT_RATELIMIT_BURST);
Pekka Enberg781b2ba2009-06-10 18:50:32 +03002182 int node;
Christoph Lameterfa45dc22014-08-06 16:04:09 -07002183 struct kmem_cache_node *n;
Pekka Enberg781b2ba2009-06-10 18:50:32 +03002184
David Rientjes9a02d692014-06-04 16:06:36 -07002185 if ((gfpflags & __GFP_NOWARN) || !__ratelimit(&slub_oom_rs))
2186 return;
2187
Fabian Frederickf9f58282014-06-04 16:06:34 -07002188 pr_warn("SLUB: Unable to allocate memory on node %d (gfp=0x%x)\n",
Pekka Enberg781b2ba2009-06-10 18:50:32 +03002189 nid, gfpflags);
Fabian Frederickf9f58282014-06-04 16:06:34 -07002190 pr_warn(" cache: %s, object size: %d, buffer size: %d, default order: %d, min order: %d\n",
2191 s->name, s->object_size, s->size, oo_order(s->oo),
2192 oo_order(s->min));
Pekka Enberg781b2ba2009-06-10 18:50:32 +03002193
Christoph Lameter3b0efdf2012-06-13 10:24:57 -05002194 if (oo_order(s->min) > get_order(s->object_size))
Fabian Frederickf9f58282014-06-04 16:06:34 -07002195 pr_warn(" %s debugging increased min order, use slub_debug=O to disable.\n",
2196 s->name);
David Rientjesfa5ec8a2009-07-07 00:14:14 -07002197
Christoph Lameterfa45dc22014-08-06 16:04:09 -07002198 for_each_kmem_cache_node(s, node, n) {
Pekka Enberg781b2ba2009-06-10 18:50:32 +03002199 unsigned long nr_slabs;
2200 unsigned long nr_objs;
2201 unsigned long nr_free;
2202
Alexander Beregalov26c02cf2009-06-11 14:08:48 +04002203 nr_free = count_partial(n, count_free);
2204 nr_slabs = node_nr_slabs(n);
2205 nr_objs = node_nr_objs(n);
Pekka Enberg781b2ba2009-06-10 18:50:32 +03002206
Fabian Frederickf9f58282014-06-04 16:06:34 -07002207 pr_warn(" node %d: slabs: %ld, objs: %ld, free: %ld\n",
Pekka Enberg781b2ba2009-06-10 18:50:32 +03002208 node, nr_slabs, nr_objs, nr_free);
2209 }
David Rientjes9a02d692014-06-04 16:06:36 -07002210#endif
Pekka Enberg781b2ba2009-06-10 18:50:32 +03002211}
2212
Christoph Lameter497b66f2011-08-09 16:12:26 -05002213static inline void *new_slab_objects(struct kmem_cache *s, gfp_t flags,
2214 int node, struct kmem_cache_cpu **pc)
2215{
Christoph Lameter6faa6832012-05-09 10:09:51 -05002216 void *freelist;
Christoph Lameter188fd062012-05-09 10:09:55 -05002217 struct kmem_cache_cpu *c = *pc;
2218 struct page *page;
Christoph Lameter497b66f2011-08-09 16:12:26 -05002219
Christoph Lameter188fd062012-05-09 10:09:55 -05002220 freelist = get_partial(s, flags, node, c);
2221
2222 if (freelist)
2223 return freelist;
2224
2225 page = new_slab(s, flags, node);
Christoph Lameter497b66f2011-08-09 16:12:26 -05002226 if (page) {
Christoph Lameter7c8e0182014-06-04 16:07:56 -07002227 c = raw_cpu_ptr(s->cpu_slab);
Christoph Lameter497b66f2011-08-09 16:12:26 -05002228 if (c->page)
2229 flush_slab(s, c);
2230
2231 /*
2232 * No other reference to the page yet so we can
2233 * muck around with it freely without cmpxchg
2234 */
Christoph Lameter6faa6832012-05-09 10:09:51 -05002235 freelist = page->freelist;
Christoph Lameter497b66f2011-08-09 16:12:26 -05002236 page->freelist = NULL;
2237
2238 stat(s, ALLOC_SLAB);
Christoph Lameter497b66f2011-08-09 16:12:26 -05002239 c->page = page;
2240 *pc = c;
2241 } else
Christoph Lameter6faa6832012-05-09 10:09:51 -05002242 freelist = NULL;
Christoph Lameter497b66f2011-08-09 16:12:26 -05002243
Christoph Lameter6faa6832012-05-09 10:09:51 -05002244 return freelist;
Christoph Lameter497b66f2011-08-09 16:12:26 -05002245}
2246
Mel Gorman072bb0a2012-07-31 16:43:58 -07002247static inline bool pfmemalloc_match(struct page *page, gfp_t gfpflags)
2248{
2249 if (unlikely(PageSlabPfmemalloc(page)))
2250 return gfp_pfmemalloc_allowed(gfpflags);
2251
2252 return true;
2253}
2254
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002255/*
Chen Gangd0e0ac92013-07-15 09:05:29 +08002256 * Check the page->freelist of a page and either transfer the freelist to the
2257 * per cpu freelist or deactivate the page.
Christoph Lameter213eeb92011-11-11 14:07:14 -06002258 *
2259 * The page is still frozen if the return value is not NULL.
2260 *
2261 * If this function returns NULL then the page has been unfrozen.
Joonsoo Kimd24ac772012-05-18 22:01:17 +09002262 *
2263 * This function must be called with interrupt disabled.
Christoph Lameter213eeb92011-11-11 14:07:14 -06002264 */
2265static inline void *get_freelist(struct kmem_cache *s, struct page *page)
2266{
2267 struct page new;
2268 unsigned long counters;
2269 void *freelist;
2270
2271 do {
2272 freelist = page->freelist;
2273 counters = page->counters;
Christoph Lameter6faa6832012-05-09 10:09:51 -05002274
Christoph Lameter213eeb92011-11-11 14:07:14 -06002275 new.counters = counters;
Dave Hansena0132ac2014-01-29 14:05:50 -08002276 VM_BUG_ON(!new.frozen);
Christoph Lameter213eeb92011-11-11 14:07:14 -06002277
2278 new.inuse = page->objects;
2279 new.frozen = freelist != NULL;
2280
Joonsoo Kimd24ac772012-05-18 22:01:17 +09002281 } while (!__cmpxchg_double_slab(s, page,
Christoph Lameter213eeb92011-11-11 14:07:14 -06002282 freelist, counters,
2283 NULL, new.counters,
2284 "get_freelist"));
2285
2286 return freelist;
2287}
2288
2289/*
Christoph Lameter894b8782007-05-10 03:15:16 -07002290 * Slow path. The lockless freelist is empty or we need to perform
2291 * debugging duties.
Christoph Lameter81819f02007-05-06 14:49:36 -07002292 *
Christoph Lameter894b8782007-05-10 03:15:16 -07002293 * Processing is still very fast if new objects have been freed to the
2294 * regular freelist. In that case we simply take over the regular freelist
2295 * as the lockless freelist and zap the regular freelist.
Christoph Lameter81819f02007-05-06 14:49:36 -07002296 *
Christoph Lameter894b8782007-05-10 03:15:16 -07002297 * If that is not working then we fall back to the partial lists. We take the
2298 * first element of the freelist as the object to allocate now and move the
2299 * rest of the freelist to the lockless freelist.
2300 *
2301 * And if we were unable to get a new slab from the partial slab lists then
Christoph Lameter6446faa2008-02-15 23:45:26 -08002302 * we need to allocate a new slab. This is the slowest path since it involves
2303 * a call to the page allocator and the setup of a new slab.
Christoph Lameter81819f02007-05-06 14:49:36 -07002304 */
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03002305static void *__slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node,
2306 unsigned long addr, struct kmem_cache_cpu *c)
Christoph Lameter81819f02007-05-06 14:49:36 -07002307{
Christoph Lameter6faa6832012-05-09 10:09:51 -05002308 void *freelist;
Christoph Lameterf6e7def2012-05-09 10:09:58 -05002309 struct page *page;
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002310 unsigned long flags;
2311
2312 local_irq_save(flags);
2313#ifdef CONFIG_PREEMPT
2314 /*
2315 * We may have been preempted and rescheduled on a different
2316 * cpu before disabling interrupts. Need to reload cpu area
2317 * pointer.
2318 */
2319 c = this_cpu_ptr(s->cpu_slab);
2320#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07002321
Christoph Lameterf6e7def2012-05-09 10:09:58 -05002322 page = c->page;
2323 if (!page)
Christoph Lameter81819f02007-05-06 14:49:36 -07002324 goto new_slab;
Christoph Lameter49e22582011-08-09 16:12:27 -05002325redo:
Christoph Lameter6faa6832012-05-09 10:09:51 -05002326
Christoph Lameter57d437d2012-05-09 10:09:59 -05002327 if (unlikely(!node_match(page, node))) {
Joonsoo Kima561ce02014-10-09 15:26:15 -07002328 int searchnode = node;
2329
2330 if (node != NUMA_NO_NODE && !node_present_pages(node))
2331 searchnode = node_to_mem_node(node);
2332
2333 if (unlikely(!node_match(page, searchnode))) {
2334 stat(s, ALLOC_NODE_MISMATCH);
2335 deactivate_slab(s, page, c->freelist);
2336 c->page = NULL;
2337 c->freelist = NULL;
2338 goto new_slab;
2339 }
Christoph Lameterfc59c052011-06-01 12:25:56 -05002340 }
Christoph Lameter6446faa2008-02-15 23:45:26 -08002341
Mel Gorman072bb0a2012-07-31 16:43:58 -07002342 /*
2343 * By rights, we should be searching for a slab page that was
2344 * PFMEMALLOC but right now, we are losing the pfmemalloc
2345 * information when the page leaves the per-cpu allocator
2346 */
2347 if (unlikely(!pfmemalloc_match(page, gfpflags))) {
2348 deactivate_slab(s, page, c->freelist);
2349 c->page = NULL;
2350 c->freelist = NULL;
2351 goto new_slab;
2352 }
2353
Eric Dumazet73736e02011-12-13 04:57:06 +01002354 /* must check again c->freelist in case of cpu migration or IRQ */
Christoph Lameter6faa6832012-05-09 10:09:51 -05002355 freelist = c->freelist;
2356 if (freelist)
Eric Dumazet73736e02011-12-13 04:57:06 +01002357 goto load_freelist;
2358
Christoph Lameterf6e7def2012-05-09 10:09:58 -05002359 freelist = get_freelist(s, page);
Christoph Lameter6446faa2008-02-15 23:45:26 -08002360
Christoph Lameter6faa6832012-05-09 10:09:51 -05002361 if (!freelist) {
Christoph Lameter03e404a2011-06-01 12:25:58 -05002362 c->page = NULL;
2363 stat(s, DEACTIVATE_BYPASS);
Christoph Lameterfc59c052011-06-01 12:25:56 -05002364 goto new_slab;
Christoph Lameter03e404a2011-06-01 12:25:58 -05002365 }
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002366
Christoph Lameter81819f02007-05-06 14:49:36 -07002367 stat(s, ALLOC_REFILL);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08002368
Christoph Lameter894b8782007-05-10 03:15:16 -07002369load_freelist:
Christoph Lameter507effe2012-05-09 10:09:52 -05002370 /*
2371 * freelist is pointing to the list of objects to be used.
2372 * page is pointing to the page from which the objects are obtained.
2373 * That page must be frozen for per cpu allocations to work.
2374 */
Dave Hansena0132ac2014-01-29 14:05:50 -08002375 VM_BUG_ON(!c->page->frozen);
Christoph Lameter6faa6832012-05-09 10:09:51 -05002376 c->freelist = get_freepointer(s, freelist);
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002377 c->tid = next_tid(c->tid);
2378 local_irq_restore(flags);
Christoph Lameter6faa6832012-05-09 10:09:51 -05002379 return freelist;
Christoph Lameter81819f02007-05-06 14:49:36 -07002380
Christoph Lameter81819f02007-05-06 14:49:36 -07002381new_slab:
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002382
Christoph Lameter49e22582011-08-09 16:12:27 -05002383 if (c->partial) {
Christoph Lameterf6e7def2012-05-09 10:09:58 -05002384 page = c->page = c->partial;
2385 c->partial = page->next;
Christoph Lameter49e22582011-08-09 16:12:27 -05002386 stat(s, CPU_PARTIAL_ALLOC);
2387 c->freelist = NULL;
2388 goto redo;
Christoph Lameter81819f02007-05-06 14:49:36 -07002389 }
2390
Christoph Lameter188fd062012-05-09 10:09:55 -05002391 freelist = new_slab_objects(s, gfpflags, node, &c);
Christoph Lameterb811c202007-10-16 23:25:51 -07002392
Christoph Lameterf46974362012-05-09 10:09:54 -05002393 if (unlikely(!freelist)) {
David Rientjes9a02d692014-06-04 16:06:36 -07002394 slab_out_of_memory(s, gfpflags, node);
Christoph Lameterf46974362012-05-09 10:09:54 -05002395 local_irq_restore(flags);
2396 return NULL;
Christoph Lameter81819f02007-05-06 14:49:36 -07002397 }
Christoph Lameter894b8782007-05-10 03:15:16 -07002398
Christoph Lameterf6e7def2012-05-09 10:09:58 -05002399 page = c->page;
Christoph Lameter5091b742012-07-31 16:44:00 -07002400 if (likely(!kmem_cache_debug(s) && pfmemalloc_match(page, gfpflags)))
Christoph Lameter81819f02007-05-06 14:49:36 -07002401 goto load_freelist;
Christoph Lameter894b8782007-05-10 03:15:16 -07002402
Christoph Lameter497b66f2011-08-09 16:12:26 -05002403 /* Only entered in the debug case */
Chen Gangd0e0ac92013-07-15 09:05:29 +08002404 if (kmem_cache_debug(s) &&
2405 !alloc_debug_processing(s, page, freelist, addr))
Christoph Lameter497b66f2011-08-09 16:12:26 -05002406 goto new_slab; /* Slab failed checks. Next slab needed */
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002407
Christoph Lameterf6e7def2012-05-09 10:09:58 -05002408 deactivate_slab(s, page, get_freepointer(s, freelist));
Christoph Lameterc17dda42012-05-09 10:09:57 -05002409 c->page = NULL;
2410 c->freelist = NULL;
Christoph Lametera71ae472011-05-25 09:47:43 -05002411 local_irq_restore(flags);
Christoph Lameter6faa6832012-05-09 10:09:51 -05002412 return freelist;
Christoph Lameter894b8782007-05-10 03:15:16 -07002413}
2414
2415/*
2416 * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc)
2417 * have the fastpath folded into their functions. So no function call
2418 * overhead for requests that can be satisfied on the fastpath.
2419 *
2420 * The fastpath works by first checking if the lockless freelist can be used.
2421 * If not then __slab_alloc is called for slow processing.
2422 *
2423 * Otherwise we can simply pick the next object from the lockless free list.
2424 */
Ezequiel Garcia2b847c32012-09-08 17:47:58 -03002425static __always_inline void *slab_alloc_node(struct kmem_cache *s,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03002426 gfp_t gfpflags, int node, unsigned long addr)
Christoph Lameter894b8782007-05-10 03:15:16 -07002427{
Christoph Lameter894b8782007-05-10 03:15:16 -07002428 void **object;
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002429 struct kmem_cache_cpu *c;
Christoph Lameter57d437d2012-05-09 10:09:59 -05002430 struct page *page;
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002431 unsigned long tid;
Christoph Lameter1f842602008-01-07 23:20:30 -08002432
Vladimir Davydov8135be52014-12-12 16:56:38 -08002433 s = slab_pre_alloc_hook(s, gfpflags);
2434 if (!s)
Akinobu Mita773ff602008-12-23 19:37:01 +09002435 return NULL;
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002436redo:
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002437 /*
2438 * Must read kmem_cache cpu data via this cpu ptr. Preemption is
2439 * enabled. We may switch back and forth between cpus while
2440 * reading from one cpu area. That does not matter as long
2441 * as we end up on the original cpu again when doing the cmpxchg.
Christoph Lameter7cccd802013-01-23 21:45:48 +00002442 *
Joonsoo Kim9aabf812015-02-10 14:09:32 -08002443 * We should guarantee that tid and kmem_cache are retrieved on
2444 * the same cpu. It could be different if CONFIG_PREEMPT so we need
2445 * to check if it is matched or not.
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002446 */
Joonsoo Kim9aabf812015-02-10 14:09:32 -08002447 do {
2448 tid = this_cpu_read(s->cpu_slab->tid);
2449 c = raw_cpu_ptr(s->cpu_slab);
Mark Rutland859b7a02015-03-25 15:55:23 -07002450 } while (IS_ENABLED(CONFIG_PREEMPT) &&
2451 unlikely(tid != READ_ONCE(c->tid)));
Joonsoo Kim9aabf812015-02-10 14:09:32 -08002452
2453 /*
2454 * Irqless object alloc/free algorithm used here depends on sequence
2455 * of fetching cpu_slab's data. tid should be fetched before anything
2456 * on c to guarantee that object and page associated with previous tid
2457 * won't be used with current tid. If we fetch tid first, object and
2458 * page could be one associated with next tid and our alloc/free
2459 * request will be failed. In this case, we will retry. So, no problem.
2460 */
2461 barrier();
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002462
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002463 /*
2464 * The transaction ids are globally unique per cpu and per operation on
2465 * a per cpu queue. Thus they can be guarantee that the cmpxchg_double
2466 * occurs on the right processor and that there was no operation on the
2467 * linked list in between.
2468 */
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002469
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06002470 object = c->freelist;
Christoph Lameter57d437d2012-05-09 10:09:59 -05002471 page = c->page;
Dave Hansen8eae1492014-06-04 16:06:37 -07002472 if (unlikely(!object || !node_match(page, node))) {
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002473 object = __slab_alloc(s, gfpflags, node, addr, c);
Dave Hansen8eae1492014-06-04 16:06:37 -07002474 stat(s, ALLOC_SLOWPATH);
2475 } else {
Eric Dumazet0ad95002011-12-16 16:25:34 +01002476 void *next_object = get_freepointer_safe(s, object);
2477
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002478 /*
Lucas De Marchi25985ed2011-03-30 22:57:33 -03002479 * The cmpxchg will only match if there was no additional
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002480 * operation and if we are on the right processor.
2481 *
Chen Gangd0e0ac92013-07-15 09:05:29 +08002482 * The cmpxchg does the following atomically (without lock
2483 * semantics!)
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002484 * 1. Relocate first pointer to the current per cpu area.
2485 * 2. Verify that tid and freelist have not been changed
2486 * 3. If they were not changed replace tid and freelist
2487 *
Chen Gangd0e0ac92013-07-15 09:05:29 +08002488 * Since this is without lock semantics the protection is only
2489 * against code executing on this cpu *not* from access by
2490 * other cpus.
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002491 */
Christoph Lameter933393f2011-12-22 11:58:51 -06002492 if (unlikely(!this_cpu_cmpxchg_double(
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002493 s->cpu_slab->freelist, s->cpu_slab->tid,
2494 object, tid,
Eric Dumazet0ad95002011-12-16 16:25:34 +01002495 next_object, next_tid(tid)))) {
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002496
2497 note_cmpxchg_failure("slab_alloc", s, tid);
2498 goto redo;
2499 }
Eric Dumazet0ad95002011-12-16 16:25:34 +01002500 prefetch_freepointer(s, next_object);
Christoph Lameter84e554e62009-12-18 16:26:23 -06002501 stat(s, ALLOC_FASTPATH);
Christoph Lameter894b8782007-05-10 03:15:16 -07002502 }
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002503
Pekka Enberg74e21342009-11-25 20:14:48 +02002504 if (unlikely(gfpflags & __GFP_ZERO) && object)
Christoph Lameter3b0efdf2012-06-13 10:24:57 -05002505 memset(object, 0, s->object_size);
Christoph Lameterd07dbea2007-07-17 04:03:23 -07002506
Christoph Lameterc016b0b2010-08-20 12:37:16 -05002507 slab_post_alloc_hook(s, gfpflags, object);
Vegard Nossum5a896d92008-04-04 00:54:48 +02002508
Christoph Lameter894b8782007-05-10 03:15:16 -07002509 return object;
Christoph Lameter81819f02007-05-06 14:49:36 -07002510}
2511
Ezequiel Garcia2b847c32012-09-08 17:47:58 -03002512static __always_inline void *slab_alloc(struct kmem_cache *s,
2513 gfp_t gfpflags, unsigned long addr)
2514{
2515 return slab_alloc_node(s, gfpflags, NUMA_NO_NODE, addr);
2516}
2517
Christoph Lameter81819f02007-05-06 14:49:36 -07002518void *kmem_cache_alloc(struct kmem_cache *s, gfp_t gfpflags)
2519{
Ezequiel Garcia2b847c32012-09-08 17:47:58 -03002520 void *ret = slab_alloc(s, gfpflags, _RET_IP_);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002521
Chen Gangd0e0ac92013-07-15 09:05:29 +08002522 trace_kmem_cache_alloc(_RET_IP_, ret, s->object_size,
2523 s->size, gfpflags);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002524
2525 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07002526}
2527EXPORT_SYMBOL(kmem_cache_alloc);
2528
Li Zefan0f24f122009-12-11 15:45:30 +08002529#ifdef CONFIG_TRACING
Richard Kennedy4a923792010-10-21 10:29:19 +01002530void *kmem_cache_alloc_trace(struct kmem_cache *s, gfp_t gfpflags, size_t size)
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002531{
Ezequiel Garcia2b847c32012-09-08 17:47:58 -03002532 void *ret = slab_alloc(s, gfpflags, _RET_IP_);
Richard Kennedy4a923792010-10-21 10:29:19 +01002533 trace_kmalloc(_RET_IP_, ret, size, s->size, gfpflags);
Andrey Ryabinin0316bec2015-02-13 14:39:42 -08002534 kasan_kmalloc(s, ret, size);
Richard Kennedy4a923792010-10-21 10:29:19 +01002535 return ret;
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002536}
Richard Kennedy4a923792010-10-21 10:29:19 +01002537EXPORT_SYMBOL(kmem_cache_alloc_trace);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002538#endif
2539
Christoph Lameter81819f02007-05-06 14:49:36 -07002540#ifdef CONFIG_NUMA
2541void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t gfpflags, int node)
2542{
Ezequiel Garcia2b847c32012-09-08 17:47:58 -03002543 void *ret = slab_alloc_node(s, gfpflags, node, _RET_IP_);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002544
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02002545 trace_kmem_cache_alloc_node(_RET_IP_, ret,
Christoph Lameter3b0efdf2012-06-13 10:24:57 -05002546 s->object_size, s->size, gfpflags, node);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002547
2548 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07002549}
2550EXPORT_SYMBOL(kmem_cache_alloc_node);
Christoph Lameter81819f02007-05-06 14:49:36 -07002551
Li Zefan0f24f122009-12-11 15:45:30 +08002552#ifdef CONFIG_TRACING
Richard Kennedy4a923792010-10-21 10:29:19 +01002553void *kmem_cache_alloc_node_trace(struct kmem_cache *s,
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002554 gfp_t gfpflags,
Richard Kennedy4a923792010-10-21 10:29:19 +01002555 int node, size_t size)
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002556{
Ezequiel Garcia2b847c32012-09-08 17:47:58 -03002557 void *ret = slab_alloc_node(s, gfpflags, node, _RET_IP_);
Richard Kennedy4a923792010-10-21 10:29:19 +01002558
2559 trace_kmalloc_node(_RET_IP_, ret,
2560 size, s->size, gfpflags, node);
Andrey Ryabinin0316bec2015-02-13 14:39:42 -08002561
2562 kasan_kmalloc(s, ret, size);
Richard Kennedy4a923792010-10-21 10:29:19 +01002563 return ret;
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002564}
Richard Kennedy4a923792010-10-21 10:29:19 +01002565EXPORT_SYMBOL(kmem_cache_alloc_node_trace);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002566#endif
Namhyung Kim5d1f57e2010-09-29 21:02:15 +09002567#endif
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002568
Christoph Lameter81819f02007-05-06 14:49:36 -07002569/*
Kim Phillips94e4d712015-02-10 14:09:37 -08002570 * Slow path handling. This may still be called frequently since objects
Christoph Lameter894b8782007-05-10 03:15:16 -07002571 * have a longer lifetime than the cpu slabs in most processing loads.
Christoph Lameter81819f02007-05-06 14:49:36 -07002572 *
Christoph Lameter894b8782007-05-10 03:15:16 -07002573 * So we still attempt to reduce cache line usage. Just take the slab
2574 * lock and free the item. If there is no additional partial page
2575 * handling required then we can return immediately.
Christoph Lameter81819f02007-05-06 14:49:36 -07002576 */
Christoph Lameter894b8782007-05-10 03:15:16 -07002577static void __slab_free(struct kmem_cache *s, struct page *page,
Christoph Lameterff120592009-12-18 16:26:22 -06002578 void *x, unsigned long addr)
Christoph Lameter81819f02007-05-06 14:49:36 -07002579{
2580 void *prior;
2581 void **object = (void *)x;
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002582 int was_frozen;
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002583 struct page new;
2584 unsigned long counters;
2585 struct kmem_cache_node *n = NULL;
Christoph Lameter61728d12011-06-01 12:25:51 -05002586 unsigned long uninitialized_var(flags);
Christoph Lameter81819f02007-05-06 14:49:36 -07002587
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002588 stat(s, FREE_SLOWPATH);
Christoph Lameter81819f02007-05-06 14:49:36 -07002589
Christoph Lameter19c7ff92012-05-30 12:54:46 -05002590 if (kmem_cache_debug(s) &&
2591 !(n = free_debug_processing(s, page, x, addr, &flags)))
Christoph Lameter80f08c12011-06-01 12:25:55 -05002592 return;
Christoph Lameter6446faa2008-02-15 23:45:26 -08002593
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002594 do {
Joonsoo Kim837d6782012-08-16 00:02:40 +09002595 if (unlikely(n)) {
2596 spin_unlock_irqrestore(&n->list_lock, flags);
2597 n = NULL;
2598 }
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002599 prior = page->freelist;
2600 counters = page->counters;
2601 set_freepointer(s, object, prior);
2602 new.counters = counters;
2603 was_frozen = new.frozen;
2604 new.inuse--;
Joonsoo Kim837d6782012-08-16 00:02:40 +09002605 if ((!new.inuse || !prior) && !was_frozen) {
Christoph Lameter49e22582011-08-09 16:12:27 -05002606
Peter Zijlstrac65c1872014-01-10 13:23:49 +01002607 if (kmem_cache_has_cpu_partial(s) && !prior) {
Christoph Lameter49e22582011-08-09 16:12:27 -05002608
2609 /*
Chen Gangd0e0ac92013-07-15 09:05:29 +08002610 * Slab was on no list before and will be
2611 * partially empty
2612 * We can defer the list move and instead
2613 * freeze it.
Christoph Lameter49e22582011-08-09 16:12:27 -05002614 */
2615 new.frozen = 1;
2616
Peter Zijlstrac65c1872014-01-10 13:23:49 +01002617 } else { /* Needs to be taken off a list */
Christoph Lameter49e22582011-08-09 16:12:27 -05002618
LQYMGTb455def2014-12-10 15:42:13 -08002619 n = get_node(s, page_to_nid(page));
Christoph Lameter49e22582011-08-09 16:12:27 -05002620 /*
2621 * Speculatively acquire the list_lock.
2622 * If the cmpxchg does not succeed then we may
2623 * drop the list_lock without any processing.
2624 *
2625 * Otherwise the list_lock will synchronize with
2626 * other processors updating the list of slabs.
2627 */
2628 spin_lock_irqsave(&n->list_lock, flags);
2629
2630 }
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002631 }
Christoph Lameter81819f02007-05-06 14:49:36 -07002632
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002633 } while (!cmpxchg_double_slab(s, page,
2634 prior, counters,
2635 object, new.counters,
2636 "__slab_free"));
Christoph Lameter81819f02007-05-06 14:49:36 -07002637
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002638 if (likely(!n)) {
Christoph Lameter49e22582011-08-09 16:12:27 -05002639
2640 /*
2641 * If we just froze the page then put it onto the
2642 * per cpu partial list.
2643 */
Alex Shi8028dce2012-02-03 23:34:56 +08002644 if (new.frozen && !was_frozen) {
Christoph Lameter49e22582011-08-09 16:12:27 -05002645 put_cpu_partial(s, page, 1);
Alex Shi8028dce2012-02-03 23:34:56 +08002646 stat(s, CPU_PARTIAL_FREE);
2647 }
Christoph Lameter49e22582011-08-09 16:12:27 -05002648 /*
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002649 * The list lock was not taken therefore no list
2650 * activity can be necessary.
2651 */
LQYMGTb455def2014-12-10 15:42:13 -08002652 if (was_frozen)
2653 stat(s, FREE_FROZEN);
2654 return;
2655 }
Christoph Lameter81819f02007-05-06 14:49:36 -07002656
Joonsoo Kim8a5b20a2014-07-02 15:22:35 -07002657 if (unlikely(!new.inuse && n->nr_partial >= s->min_partial))
Joonsoo Kim837d6782012-08-16 00:02:40 +09002658 goto slab_empty;
Christoph Lameter81819f02007-05-06 14:49:36 -07002659
Joonsoo Kim837d6782012-08-16 00:02:40 +09002660 /*
2661 * Objects left in the slab. If it was not on the partial list before
2662 * then add it.
2663 */
Joonsoo Kim345c9052013-06-19 14:05:52 +09002664 if (!kmem_cache_has_cpu_partial(s) && unlikely(!prior)) {
2665 if (kmem_cache_debug(s))
Peter Zijlstrac65c1872014-01-10 13:23:49 +01002666 remove_full(s, n, page);
Joonsoo Kim837d6782012-08-16 00:02:40 +09002667 add_partial(n, page, DEACTIVATE_TO_TAIL);
2668 stat(s, FREE_ADD_PARTIAL);
Christoph Lameter81819f02007-05-06 14:49:36 -07002669 }
Christoph Lameter80f08c12011-06-01 12:25:55 -05002670 spin_unlock_irqrestore(&n->list_lock, flags);
Christoph Lameter81819f02007-05-06 14:49:36 -07002671 return;
2672
2673slab_empty:
Christoph Lametera973e9d2008-03-01 13:40:44 -08002674 if (prior) {
Christoph Lameter81819f02007-05-06 14:49:36 -07002675 /*
Christoph Lameter6fbabb22011-08-08 11:16:56 -05002676 * Slab on the partial list.
Christoph Lameter81819f02007-05-06 14:49:36 -07002677 */
Christoph Lameter5cc6eee2011-06-01 12:25:50 -05002678 remove_partial(n, page);
Christoph Lameter84e554e62009-12-18 16:26:23 -06002679 stat(s, FREE_REMOVE_PARTIAL);
Peter Zijlstrac65c1872014-01-10 13:23:49 +01002680 } else {
Christoph Lameter6fbabb22011-08-08 11:16:56 -05002681 /* Slab must be on the full list */
Peter Zijlstrac65c1872014-01-10 13:23:49 +01002682 remove_full(s, n, page);
2683 }
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002684
Christoph Lameter80f08c12011-06-01 12:25:55 -05002685 spin_unlock_irqrestore(&n->list_lock, flags);
Christoph Lameter84e554e62009-12-18 16:26:23 -06002686 stat(s, FREE_SLAB);
Christoph Lameter81819f02007-05-06 14:49:36 -07002687 discard_slab(s, page);
Christoph Lameter81819f02007-05-06 14:49:36 -07002688}
2689
Christoph Lameter894b8782007-05-10 03:15:16 -07002690/*
2691 * Fastpath with forced inlining to produce a kfree and kmem_cache_free that
2692 * can perform fastpath freeing without additional function calls.
2693 *
2694 * The fastpath is only possible if we are freeing to the current cpu slab
2695 * of this processor. This typically the case if we have just allocated
2696 * the item before.
2697 *
2698 * If fastpath is not possible then fall back to __slab_free where we deal
2699 * with all sorts of special processing.
2700 */
Pekka Enberg06428782008-01-07 23:20:27 -08002701static __always_inline void slab_free(struct kmem_cache *s,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03002702 struct page *page, void *x, unsigned long addr)
Christoph Lameter894b8782007-05-10 03:15:16 -07002703{
2704 void **object = (void *)x;
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002705 struct kmem_cache_cpu *c;
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002706 unsigned long tid;
Christoph Lameter1f842602008-01-07 23:20:30 -08002707
Christoph Lameterc016b0b2010-08-20 12:37:16 -05002708 slab_free_hook(s, x);
2709
Christoph Lametera24c5a02011-03-15 12:45:21 -05002710redo:
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002711 /*
2712 * Determine the currently cpus per cpu slab.
2713 * The cpu may change afterward. However that does not matter since
2714 * data is retrieved via this pointer. If we are on the same cpu
2715 * during the cmpxchg then the free will succedd.
2716 */
Joonsoo Kim9aabf812015-02-10 14:09:32 -08002717 do {
2718 tid = this_cpu_read(s->cpu_slab->tid);
2719 c = raw_cpu_ptr(s->cpu_slab);
Mark Rutland859b7a02015-03-25 15:55:23 -07002720 } while (IS_ENABLED(CONFIG_PREEMPT) &&
2721 unlikely(tid != READ_ONCE(c->tid)));
Christoph Lameterc016b0b2010-08-20 12:37:16 -05002722
Joonsoo Kim9aabf812015-02-10 14:09:32 -08002723 /* Same with comment on barrier() in slab_alloc_node() */
2724 barrier();
Christoph Lameterc016b0b2010-08-20 12:37:16 -05002725
Christoph Lameter442b06b2011-05-17 16:29:31 -05002726 if (likely(page == c->page)) {
Christoph Lameterff120592009-12-18 16:26:22 -06002727 set_freepointer(s, object, c->freelist);
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002728
Christoph Lameter933393f2011-12-22 11:58:51 -06002729 if (unlikely(!this_cpu_cmpxchg_double(
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002730 s->cpu_slab->freelist, s->cpu_slab->tid,
2731 c->freelist, tid,
2732 object, next_tid(tid)))) {
2733
2734 note_cmpxchg_failure("slab_free", s, tid);
2735 goto redo;
2736 }
Christoph Lameter84e554e62009-12-18 16:26:23 -06002737 stat(s, FREE_FASTPATH);
Christoph Lameter894b8782007-05-10 03:15:16 -07002738 } else
Christoph Lameterff120592009-12-18 16:26:22 -06002739 __slab_free(s, page, x, addr);
Christoph Lameter894b8782007-05-10 03:15:16 -07002740
Christoph Lameter894b8782007-05-10 03:15:16 -07002741}
2742
Christoph Lameter81819f02007-05-06 14:49:36 -07002743void kmem_cache_free(struct kmem_cache *s, void *x)
2744{
Glauber Costab9ce5ef2012-12-18 14:22:46 -08002745 s = cache_from_obj(s, x);
2746 if (!s)
Christoph Lameter79576102012-09-04 23:06:14 +00002747 return;
Glauber Costab9ce5ef2012-12-18 14:22:46 -08002748 slab_free(s, virt_to_head_page(x), x, _RET_IP_);
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02002749 trace_kmem_cache_free(_RET_IP_, x);
Christoph Lameter81819f02007-05-06 14:49:36 -07002750}
2751EXPORT_SYMBOL(kmem_cache_free);
2752
Christoph Lameter81819f02007-05-06 14:49:36 -07002753/*
Christoph Lameter672bba32007-05-09 02:32:39 -07002754 * Object placement in a slab is made very easy because we always start at
2755 * offset 0. If we tune the size of the object to the alignment then we can
2756 * get the required alignment by putting one properly sized object after
2757 * another.
Christoph Lameter81819f02007-05-06 14:49:36 -07002758 *
2759 * Notice that the allocation order determines the sizes of the per cpu
2760 * caches. Each processor has always one slab available for allocations.
2761 * Increasing the allocation order reduces the number of times that slabs
Christoph Lameter672bba32007-05-09 02:32:39 -07002762 * must be moved on and off the partial lists and is therefore a factor in
Christoph Lameter81819f02007-05-06 14:49:36 -07002763 * locking overhead.
Christoph Lameter81819f02007-05-06 14:49:36 -07002764 */
2765
2766/*
2767 * Mininum / Maximum order of slab pages. This influences locking overhead
2768 * and slab fragmentation. A higher order reduces the number of partial slabs
2769 * and increases the number of allocations possible without having to
2770 * take the list_lock.
2771 */
2772static int slub_min_order;
Christoph Lameter114e9e82008-04-14 19:11:41 +03002773static int slub_max_order = PAGE_ALLOC_COSTLY_ORDER;
Christoph Lameter9b2cd502008-04-14 19:11:41 +03002774static int slub_min_objects;
Christoph Lameter81819f02007-05-06 14:49:36 -07002775
2776/*
Christoph Lameter81819f02007-05-06 14:49:36 -07002777 * Calculate the order of allocation given an slab object size.
2778 *
Christoph Lameter672bba32007-05-09 02:32:39 -07002779 * The order of allocation has significant impact on performance and other
2780 * system components. Generally order 0 allocations should be preferred since
2781 * order 0 does not cause fragmentation in the page allocator. Larger objects
2782 * be problematic to put into order 0 slabs because there may be too much
Christoph Lameterc124f5b2008-04-14 19:13:29 +03002783 * unused space left. We go to a higher order if more than 1/16th of the slab
Christoph Lameter672bba32007-05-09 02:32:39 -07002784 * would be wasted.
Christoph Lameter81819f02007-05-06 14:49:36 -07002785 *
Christoph Lameter672bba32007-05-09 02:32:39 -07002786 * In order to reach satisfactory performance we must ensure that a minimum
2787 * number of objects is in one slab. Otherwise we may generate too much
2788 * activity on the partial lists which requires taking the list_lock. This is
2789 * less a concern for large slabs though which are rarely used.
Christoph Lameter81819f02007-05-06 14:49:36 -07002790 *
Christoph Lameter672bba32007-05-09 02:32:39 -07002791 * slub_max_order specifies the order where we begin to stop considering the
2792 * number of objects in a slab as critical. If we reach slub_max_order then
2793 * we try to keep the page order as low as possible. So we accept more waste
2794 * of space in favor of a small page order.
2795 *
2796 * Higher order allocations also allow the placement of more objects in a
2797 * slab and thereby reduce object handling overhead. If the user has
2798 * requested a higher mininum order then we start with that one instead of
2799 * the smallest order which will fit the object.
Christoph Lameter81819f02007-05-06 14:49:36 -07002800 */
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002801static inline int slab_order(int size, int min_objects,
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002802 int max_order, int fract_leftover, int reserved)
Christoph Lameter81819f02007-05-06 14:49:36 -07002803{
2804 int order;
2805 int rem;
Christoph Lameter6300ea72007-07-17 04:03:20 -07002806 int min_order = slub_min_order;
Christoph Lameter81819f02007-05-06 14:49:36 -07002807
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002808 if (order_objects(min_order, size, reserved) > MAX_OBJS_PER_PAGE)
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +04002809 return get_order(size * MAX_OBJS_PER_PAGE) - 1;
Christoph Lameter39b26462008-04-14 19:11:30 +03002810
Christoph Lameter6300ea72007-07-17 04:03:20 -07002811 for (order = max(min_order,
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002812 fls(min_objects * size - 1) - PAGE_SHIFT);
2813 order <= max_order; order++) {
2814
Christoph Lameter81819f02007-05-06 14:49:36 -07002815 unsigned long slab_size = PAGE_SIZE << order;
2816
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002817 if (slab_size < min_objects * size + reserved)
Christoph Lameter81819f02007-05-06 14:49:36 -07002818 continue;
2819
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002820 rem = (slab_size - reserved) % size;
Christoph Lameter81819f02007-05-06 14:49:36 -07002821
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002822 if (rem <= slab_size / fract_leftover)
Christoph Lameter81819f02007-05-06 14:49:36 -07002823 break;
2824
2825 }
Christoph Lameter672bba32007-05-09 02:32:39 -07002826
Christoph Lameter81819f02007-05-06 14:49:36 -07002827 return order;
2828}
2829
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002830static inline int calculate_order(int size, int reserved)
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002831{
2832 int order;
2833 int min_objects;
2834 int fraction;
Zhang Yanmine8120ff2009-02-12 18:00:17 +02002835 int max_objects;
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002836
2837 /*
2838 * Attempt to find best configuration for a slab. This
2839 * works by first attempting to generate a layout with
2840 * the best configuration and backing off gradually.
2841 *
2842 * First we reduce the acceptable waste in a slab. Then
2843 * we reduce the minimum objects required in a slab.
2844 */
2845 min_objects = slub_min_objects;
Christoph Lameter9b2cd502008-04-14 19:11:41 +03002846 if (!min_objects)
2847 min_objects = 4 * (fls(nr_cpu_ids) + 1);
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002848 max_objects = order_objects(slub_max_order, size, reserved);
Zhang Yanmine8120ff2009-02-12 18:00:17 +02002849 min_objects = min(min_objects, max_objects);
2850
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002851 while (min_objects > 1) {
Christoph Lameterc124f5b2008-04-14 19:13:29 +03002852 fraction = 16;
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002853 while (fraction >= 4) {
2854 order = slab_order(size, min_objects,
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002855 slub_max_order, fraction, reserved);
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002856 if (order <= slub_max_order)
2857 return order;
2858 fraction /= 2;
2859 }
Amerigo Wang5086c389c2009-08-19 21:44:13 +03002860 min_objects--;
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002861 }
2862
2863 /*
2864 * We were unable to place multiple objects in a slab. Now
2865 * lets see if we can place a single object there.
2866 */
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002867 order = slab_order(size, 1, slub_max_order, 1, reserved);
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002868 if (order <= slub_max_order)
2869 return order;
2870
2871 /*
2872 * Doh this slab cannot be placed using slub_max_order.
2873 */
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002874 order = slab_order(size, 1, MAX_ORDER, 1, reserved);
David Rientjes818cf592009-04-23 09:58:22 +03002875 if (order < MAX_ORDER)
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002876 return order;
2877 return -ENOSYS;
2878}
2879
Pekka Enberg5595cff2008-08-05 09:28:47 +03002880static void
Joonsoo Kim40534972012-05-11 00:50:47 +09002881init_kmem_cache_node(struct kmem_cache_node *n)
Christoph Lameter81819f02007-05-06 14:49:36 -07002882{
2883 n->nr_partial = 0;
Christoph Lameter81819f02007-05-06 14:49:36 -07002884 spin_lock_init(&n->list_lock);
2885 INIT_LIST_HEAD(&n->partial);
Christoph Lameter8ab13722007-07-17 04:03:32 -07002886#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter0f389ec2008-04-14 18:53:02 +03002887 atomic_long_set(&n->nr_slabs, 0);
Salman Qazi02b71b72008-09-11 12:25:41 -07002888 atomic_long_set(&n->total_objects, 0);
Christoph Lameter643b1132007-05-06 14:49:42 -07002889 INIT_LIST_HEAD(&n->full);
Christoph Lameter8ab13722007-07-17 04:03:32 -07002890#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07002891}
2892
Christoph Lameter55136592010-08-20 12:37:13 -05002893static inline int alloc_kmem_cache_cpus(struct kmem_cache *s)
Christoph Lameter4c93c3552007-10-16 01:26:08 -07002894{
Christoph Lameter6c182dc2010-08-20 12:37:14 -05002895 BUILD_BUG_ON(PERCPU_DYNAMIC_EARLY_SIZE <
Christoph Lameter95a05b42013-01-10 19:14:19 +00002896 KMALLOC_SHIFT_HIGH * sizeof(struct kmem_cache_cpu));
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06002897
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002898 /*
Chris Metcalfd4d84fe2011-06-02 10:19:41 -04002899 * Must align to double word boundary for the double cmpxchg
2900 * instructions to work; see __pcpu_double_call_return_bool().
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002901 */
Chris Metcalfd4d84fe2011-06-02 10:19:41 -04002902 s->cpu_slab = __alloc_percpu(sizeof(struct kmem_cache_cpu),
2903 2 * sizeof(void *));
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06002904
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002905 if (!s->cpu_slab)
2906 return 0;
2907
2908 init_kmem_cache_cpus(s);
2909
2910 return 1;
Christoph Lameter4c93c3552007-10-16 01:26:08 -07002911}
Christoph Lameter4c93c3552007-10-16 01:26:08 -07002912
Christoph Lameter51df1142010-08-20 12:37:15 -05002913static struct kmem_cache *kmem_cache_node;
2914
Christoph Lameter81819f02007-05-06 14:49:36 -07002915/*
2916 * No kmalloc_node yet so do it by hand. We know that this is the first
2917 * slab on the node for this slabcache. There are no concurrent accesses
2918 * possible.
2919 *
Zhi Yong Wu721ae222013-11-08 20:47:37 +08002920 * Note that this function only works on the kmem_cache_node
2921 * when allocating for the kmem_cache_node. This is used for bootstrapping
Christoph Lameter4c93c3552007-10-16 01:26:08 -07002922 * memory on a fresh node that has no slab structures yet.
Christoph Lameter81819f02007-05-06 14:49:36 -07002923 */
Christoph Lameter55136592010-08-20 12:37:13 -05002924static void early_kmem_cache_node_alloc(int node)
Christoph Lameter81819f02007-05-06 14:49:36 -07002925{
2926 struct page *page;
2927 struct kmem_cache_node *n;
2928
Christoph Lameter51df1142010-08-20 12:37:15 -05002929 BUG_ON(kmem_cache_node->size < sizeof(struct kmem_cache_node));
Christoph Lameter81819f02007-05-06 14:49:36 -07002930
Christoph Lameter51df1142010-08-20 12:37:15 -05002931 page = new_slab(kmem_cache_node, GFP_NOWAIT, node);
Christoph Lameter81819f02007-05-06 14:49:36 -07002932
2933 BUG_ON(!page);
Christoph Lametera2f92ee2007-08-22 14:01:57 -07002934 if (page_to_nid(page) != node) {
Fabian Frederickf9f58282014-06-04 16:06:34 -07002935 pr_err("SLUB: Unable to allocate memory from node %d\n", node);
2936 pr_err("SLUB: Allocating a useless per node structure in order to be able to continue\n");
Christoph Lametera2f92ee2007-08-22 14:01:57 -07002937 }
2938
Christoph Lameter81819f02007-05-06 14:49:36 -07002939 n = page->freelist;
2940 BUG_ON(!n);
Christoph Lameter51df1142010-08-20 12:37:15 -05002941 page->freelist = get_freepointer(kmem_cache_node, n);
Christoph Lametere6e82ea2011-08-09 16:12:24 -05002942 page->inuse = 1;
Christoph Lameter8cb0a502011-06-01 12:25:46 -05002943 page->frozen = 0;
Christoph Lameter51df1142010-08-20 12:37:15 -05002944 kmem_cache_node->node[node] = n;
Christoph Lameter8ab13722007-07-17 04:03:32 -07002945#ifdef CONFIG_SLUB_DEBUG
Christoph Lameterf7cb1932010-09-29 07:15:01 -05002946 init_object(kmem_cache_node, n, SLUB_RED_ACTIVE);
Christoph Lameter51df1142010-08-20 12:37:15 -05002947 init_tracking(kmem_cache_node, n);
Christoph Lameter8ab13722007-07-17 04:03:32 -07002948#endif
Andrey Ryabinin0316bec2015-02-13 14:39:42 -08002949 kasan_kmalloc(kmem_cache_node, n, sizeof(struct kmem_cache_node));
Joonsoo Kim40534972012-05-11 00:50:47 +09002950 init_kmem_cache_node(n);
Christoph Lameter51df1142010-08-20 12:37:15 -05002951 inc_slabs_node(kmem_cache_node, node, page->objects);
Christoph Lameter6446faa2008-02-15 23:45:26 -08002952
Dave Hansen67b6c902014-01-24 07:20:23 -08002953 /*
Steven Rostedt1e4dd942014-02-10 14:25:46 -08002954 * No locks need to be taken here as it has just been
2955 * initialized and there is no concurrent access.
Dave Hansen67b6c902014-01-24 07:20:23 -08002956 */
Steven Rostedt1e4dd942014-02-10 14:25:46 -08002957 __add_partial(n, page, DEACTIVATE_TO_HEAD);
Christoph Lameter81819f02007-05-06 14:49:36 -07002958}
2959
2960static void free_kmem_cache_nodes(struct kmem_cache *s)
2961{
2962 int node;
Christoph Lameterfa45dc22014-08-06 16:04:09 -07002963 struct kmem_cache_node *n;
Christoph Lameter81819f02007-05-06 14:49:36 -07002964
Christoph Lameterfa45dc22014-08-06 16:04:09 -07002965 for_each_kmem_cache_node(s, node, n) {
2966 kmem_cache_free(kmem_cache_node, n);
Christoph Lameter81819f02007-05-06 14:49:36 -07002967 s->node[node] = NULL;
2968 }
2969}
2970
Christoph Lameter55136592010-08-20 12:37:13 -05002971static int init_kmem_cache_nodes(struct kmem_cache *s)
Christoph Lameter81819f02007-05-06 14:49:36 -07002972{
2973 int node;
Christoph Lameter81819f02007-05-06 14:49:36 -07002974
Christoph Lameterf64dc582007-10-16 01:25:33 -07002975 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter81819f02007-05-06 14:49:36 -07002976 struct kmem_cache_node *n;
2977
Alexander Duyck73367bd2010-05-21 14:41:35 -07002978 if (slab_state == DOWN) {
Christoph Lameter55136592010-08-20 12:37:13 -05002979 early_kmem_cache_node_alloc(node);
Alexander Duyck73367bd2010-05-21 14:41:35 -07002980 continue;
Christoph Lameter81819f02007-05-06 14:49:36 -07002981 }
Christoph Lameter51df1142010-08-20 12:37:15 -05002982 n = kmem_cache_alloc_node(kmem_cache_node,
Christoph Lameter55136592010-08-20 12:37:13 -05002983 GFP_KERNEL, node);
Alexander Duyck73367bd2010-05-21 14:41:35 -07002984
2985 if (!n) {
2986 free_kmem_cache_nodes(s);
2987 return 0;
2988 }
2989
Christoph Lameter81819f02007-05-06 14:49:36 -07002990 s->node[node] = n;
Joonsoo Kim40534972012-05-11 00:50:47 +09002991 init_kmem_cache_node(n);
Christoph Lameter81819f02007-05-06 14:49:36 -07002992 }
2993 return 1;
2994}
Christoph Lameter81819f02007-05-06 14:49:36 -07002995
David Rientjesc0bdb232009-02-25 09:16:35 +02002996static void set_min_partial(struct kmem_cache *s, unsigned long min)
David Rientjes3b89d7d2009-02-22 17:40:07 -08002997{
2998 if (min < MIN_PARTIAL)
2999 min = MIN_PARTIAL;
3000 else if (min > MAX_PARTIAL)
3001 min = MAX_PARTIAL;
3002 s->min_partial = min;
3003}
3004
Christoph Lameter81819f02007-05-06 14:49:36 -07003005/*
3006 * calculate_sizes() determines the order and the distribution of data within
3007 * a slab object.
3008 */
Christoph Lameter06b285d2008-04-14 19:11:41 +03003009static int calculate_sizes(struct kmem_cache *s, int forced_order)
Christoph Lameter81819f02007-05-06 14:49:36 -07003010{
3011 unsigned long flags = s->flags;
Christoph Lameter3b0efdf2012-06-13 10:24:57 -05003012 unsigned long size = s->object_size;
Christoph Lameter834f3d12008-04-14 19:11:31 +03003013 int order;
Christoph Lameter81819f02007-05-06 14:49:36 -07003014
3015 /*
Christoph Lameterd8b42bf2008-02-15 23:45:25 -08003016 * Round up object size to the next word boundary. We can only
3017 * place the free pointer at word boundaries and this determines
3018 * the possible location of the free pointer.
3019 */
3020 size = ALIGN(size, sizeof(void *));
3021
3022#ifdef CONFIG_SLUB_DEBUG
3023 /*
Christoph Lameter81819f02007-05-06 14:49:36 -07003024 * Determine if we can poison the object itself. If the user of
3025 * the slab may touch the object after free or before allocation
3026 * then we should never poison the object itself.
3027 */
3028 if ((flags & SLAB_POISON) && !(flags & SLAB_DESTROY_BY_RCU) &&
Christoph Lameterc59def92007-05-16 22:10:50 -07003029 !s->ctor)
Christoph Lameter81819f02007-05-06 14:49:36 -07003030 s->flags |= __OBJECT_POISON;
3031 else
3032 s->flags &= ~__OBJECT_POISON;
3033
Christoph Lameter81819f02007-05-06 14:49:36 -07003034
3035 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07003036 * If we are Redzoning then check if there is some space between the
Christoph Lameter81819f02007-05-06 14:49:36 -07003037 * end of the object and the free pointer. If not then add an
Christoph Lameter672bba32007-05-09 02:32:39 -07003038 * additional word to have some bytes to store Redzone information.
Christoph Lameter81819f02007-05-06 14:49:36 -07003039 */
Christoph Lameter3b0efdf2012-06-13 10:24:57 -05003040 if ((flags & SLAB_RED_ZONE) && size == s->object_size)
Christoph Lameter81819f02007-05-06 14:49:36 -07003041 size += sizeof(void *);
Christoph Lameter41ecc552007-05-09 02:32:44 -07003042#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07003043
3044 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07003045 * With that we have determined the number of bytes in actual use
3046 * by the object. This is the potential offset to the free pointer.
Christoph Lameter81819f02007-05-06 14:49:36 -07003047 */
3048 s->inuse = size;
3049
3050 if (((flags & (SLAB_DESTROY_BY_RCU | SLAB_POISON)) ||
Christoph Lameterc59def92007-05-16 22:10:50 -07003051 s->ctor)) {
Christoph Lameter81819f02007-05-06 14:49:36 -07003052 /*
3053 * Relocate free pointer after the object if it is not
3054 * permitted to overwrite the first word of the object on
3055 * kmem_cache_free.
3056 *
3057 * This is the case if we do RCU, have a constructor or
3058 * destructor or are poisoning the objects.
3059 */
3060 s->offset = size;
3061 size += sizeof(void *);
3062 }
3063
Christoph Lameterc12b3c62007-05-23 13:57:31 -07003064#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter81819f02007-05-06 14:49:36 -07003065 if (flags & SLAB_STORE_USER)
3066 /*
3067 * Need to store information about allocs and frees after
3068 * the object.
3069 */
3070 size += 2 * sizeof(struct track);
3071
Christoph Lameterbe7b3fb2007-05-09 02:32:36 -07003072 if (flags & SLAB_RED_ZONE)
Christoph Lameter81819f02007-05-06 14:49:36 -07003073 /*
3074 * Add some empty padding so that we can catch
3075 * overwrites from earlier objects rather than let
3076 * tracking information or the free pointer be
Frederik Schwarzer0211a9c2008-12-29 22:14:56 +01003077 * corrupted if a user writes before the start
Christoph Lameter81819f02007-05-06 14:49:36 -07003078 * of the object.
3079 */
3080 size += sizeof(void *);
Christoph Lameter41ecc552007-05-09 02:32:44 -07003081#endif
Christoph Lameter672bba32007-05-09 02:32:39 -07003082
Christoph Lameter81819f02007-05-06 14:49:36 -07003083 /*
Christoph Lameter81819f02007-05-06 14:49:36 -07003084 * SLUB stores one object immediately after another beginning from
3085 * offset 0. In order to align the objects we have to simply size
3086 * each object to conform to the alignment.
3087 */
Christoph Lameter45906852012-11-28 16:23:16 +00003088 size = ALIGN(size, s->align);
Christoph Lameter81819f02007-05-06 14:49:36 -07003089 s->size = size;
Christoph Lameter06b285d2008-04-14 19:11:41 +03003090 if (forced_order >= 0)
3091 order = forced_order;
3092 else
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08003093 order = calculate_order(size, s->reserved);
Christoph Lameter81819f02007-05-06 14:49:36 -07003094
Christoph Lameter834f3d12008-04-14 19:11:31 +03003095 if (order < 0)
Christoph Lameter81819f02007-05-06 14:49:36 -07003096 return 0;
3097
Christoph Lameterb7a49f02008-02-14 14:21:32 -08003098 s->allocflags = 0;
Christoph Lameter834f3d12008-04-14 19:11:31 +03003099 if (order)
Christoph Lameterb7a49f02008-02-14 14:21:32 -08003100 s->allocflags |= __GFP_COMP;
3101
3102 if (s->flags & SLAB_CACHE_DMA)
Christoph Lameter2c59dd62013-01-10 19:14:19 +00003103 s->allocflags |= GFP_DMA;
Christoph Lameterb7a49f02008-02-14 14:21:32 -08003104
3105 if (s->flags & SLAB_RECLAIM_ACCOUNT)
3106 s->allocflags |= __GFP_RECLAIMABLE;
3107
Christoph Lameter81819f02007-05-06 14:49:36 -07003108 /*
3109 * Determine the number of objects per slab
3110 */
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08003111 s->oo = oo_make(order, size, s->reserved);
3112 s->min = oo_make(get_order(size), size, s->reserved);
Christoph Lameter205ab992008-04-14 19:11:40 +03003113 if (oo_objects(s->oo) > oo_objects(s->max))
3114 s->max = s->oo;
Christoph Lameter81819f02007-05-06 14:49:36 -07003115
Christoph Lameter834f3d12008-04-14 19:11:31 +03003116 return !!oo_objects(s->oo);
Christoph Lameter81819f02007-05-06 14:49:36 -07003117}
3118
Christoph Lameter8a13a4c2012-09-04 23:18:33 +00003119static int kmem_cache_open(struct kmem_cache *s, unsigned long flags)
Christoph Lameter81819f02007-05-06 14:49:36 -07003120{
Christoph Lameter8a13a4c2012-09-04 23:18:33 +00003121 s->flags = kmem_cache_flags(s->size, flags, s->name, s->ctor);
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08003122 s->reserved = 0;
Christoph Lameter81819f02007-05-06 14:49:36 -07003123
Lai Jiangshanda9a6382011-03-10 15:22:00 +08003124 if (need_reserve_slab_rcu && (s->flags & SLAB_DESTROY_BY_RCU))
3125 s->reserved = sizeof(struct rcu_head);
Christoph Lameter81819f02007-05-06 14:49:36 -07003126
Christoph Lameter06b285d2008-04-14 19:11:41 +03003127 if (!calculate_sizes(s, -1))
Christoph Lameter81819f02007-05-06 14:49:36 -07003128 goto error;
David Rientjes3de47212009-07-27 18:30:35 -07003129 if (disable_higher_order_debug) {
3130 /*
3131 * Disable debugging flags that store metadata if the min slab
3132 * order increased.
3133 */
Christoph Lameter3b0efdf2012-06-13 10:24:57 -05003134 if (get_order(s->size) > get_order(s->object_size)) {
David Rientjes3de47212009-07-27 18:30:35 -07003135 s->flags &= ~DEBUG_METADATA_FLAGS;
3136 s->offset = 0;
3137 if (!calculate_sizes(s, -1))
3138 goto error;
3139 }
3140 }
Christoph Lameter81819f02007-05-06 14:49:36 -07003141
Heiko Carstens25654092012-01-12 17:17:33 -08003142#if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \
3143 defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
Christoph Lameterb789ef52011-06-01 12:25:49 -05003144 if (system_has_cmpxchg_double() && (s->flags & SLAB_DEBUG_FLAGS) == 0)
3145 /* Enable fast mode */
3146 s->flags |= __CMPXCHG_DOUBLE;
3147#endif
3148
David Rientjes3b89d7d2009-02-22 17:40:07 -08003149 /*
3150 * The larger the object size is, the more pages we want on the partial
3151 * list to avoid pounding the page allocator excessively.
3152 */
Christoph Lameter49e22582011-08-09 16:12:27 -05003153 set_min_partial(s, ilog2(s->size) / 2);
3154
3155 /*
3156 * cpu_partial determined the maximum number of objects kept in the
3157 * per cpu partial lists of a processor.
3158 *
3159 * Per cpu partial lists mainly contain slabs that just have one
3160 * object freed. If they are used for allocation then they can be
3161 * filled up again with minimal effort. The slab will never hit the
3162 * per node partial lists and therefore no locking will be required.
3163 *
3164 * This setting also determines
3165 *
3166 * A) The number of objects from per cpu partial slabs dumped to the
3167 * per node list when we reach the limit.
Alex Shi9f264902011-09-01 11:32:18 +08003168 * B) The number of objects in cpu partial slabs to extract from the
Chen Gangd0e0ac92013-07-15 09:05:29 +08003169 * per node list when we run out of per cpu objects. We only fetch
3170 * 50% to keep some capacity around for frees.
Christoph Lameter49e22582011-08-09 16:12:27 -05003171 */
Joonsoo Kim345c9052013-06-19 14:05:52 +09003172 if (!kmem_cache_has_cpu_partial(s))
Christoph Lameter8f1e33d2011-11-23 09:24:27 -06003173 s->cpu_partial = 0;
3174 else if (s->size >= PAGE_SIZE)
Christoph Lameter49e22582011-08-09 16:12:27 -05003175 s->cpu_partial = 2;
3176 else if (s->size >= 1024)
3177 s->cpu_partial = 6;
3178 else if (s->size >= 256)
3179 s->cpu_partial = 13;
3180 else
3181 s->cpu_partial = 30;
3182
Christoph Lameter81819f02007-05-06 14:49:36 -07003183#ifdef CONFIG_NUMA
Christoph Lametere2cb96b2008-08-19 08:51:22 -05003184 s->remote_node_defrag_ratio = 1000;
Christoph Lameter81819f02007-05-06 14:49:36 -07003185#endif
Christoph Lameter55136592010-08-20 12:37:13 -05003186 if (!init_kmem_cache_nodes(s))
Christoph Lameterdfb4f092007-10-16 01:26:05 -07003187 goto error;
Christoph Lameter81819f02007-05-06 14:49:36 -07003188
Christoph Lameter55136592010-08-20 12:37:13 -05003189 if (alloc_kmem_cache_cpus(s))
Christoph Lameter278b1bb2012-09-05 00:20:34 +00003190 return 0;
Christoph Lameterff120592009-12-18 16:26:22 -06003191
Christoph Lameter4c93c3552007-10-16 01:26:08 -07003192 free_kmem_cache_nodes(s);
Christoph Lameter81819f02007-05-06 14:49:36 -07003193error:
3194 if (flags & SLAB_PANIC)
3195 panic("Cannot create slab %s size=%lu realsize=%u "
3196 "order=%u offset=%u flags=%lx\n",
Chen Gangd0e0ac92013-07-15 09:05:29 +08003197 s->name, (unsigned long)s->size, s->size,
3198 oo_order(s->oo), s->offset, flags);
Christoph Lameter278b1bb2012-09-05 00:20:34 +00003199 return -EINVAL;
Christoph Lameter81819f02007-05-06 14:49:36 -07003200}
Christoph Lameter81819f02007-05-06 14:49:36 -07003201
Christoph Lameter33b12c32008-04-25 12:22:43 -07003202static void list_slab_objects(struct kmem_cache *s, struct page *page,
3203 const char *text)
Christoph Lameter81819f02007-05-06 14:49:36 -07003204{
Christoph Lameter33b12c32008-04-25 12:22:43 -07003205#ifdef CONFIG_SLUB_DEBUG
3206 void *addr = page_address(page);
3207 void *p;
Namhyung Kima5dd5c12010-09-29 21:02:13 +09003208 unsigned long *map = kzalloc(BITS_TO_LONGS(page->objects) *
3209 sizeof(long), GFP_ATOMIC);
Eric Dumazetbbd7d572010-03-24 22:25:47 +01003210 if (!map)
3211 return;
Christoph Lameter945cf2b2012-09-04 23:18:33 +00003212 slab_err(s, page, text, s->name);
Christoph Lameter33b12c32008-04-25 12:22:43 -07003213 slab_lock(page);
Christoph Lameter33b12c32008-04-25 12:22:43 -07003214
Christoph Lameter5f80b132011-04-15 14:48:13 -05003215 get_map(s, page, map);
Christoph Lameter33b12c32008-04-25 12:22:43 -07003216 for_each_object(p, s, addr, page->objects) {
3217
3218 if (!test_bit(slab_index(p, s, addr), map)) {
Fabian Frederickf9f58282014-06-04 16:06:34 -07003219 pr_err("INFO: Object 0x%p @offset=%tu\n", p, p - addr);
Christoph Lameter33b12c32008-04-25 12:22:43 -07003220 print_tracking(s, p);
3221 }
3222 }
3223 slab_unlock(page);
Eric Dumazetbbd7d572010-03-24 22:25:47 +01003224 kfree(map);
Christoph Lameter33b12c32008-04-25 12:22:43 -07003225#endif
3226}
3227
Christoph Lameter81819f02007-05-06 14:49:36 -07003228/*
Christoph Lameter599870b2008-04-23 12:36:52 -07003229 * Attempt to free all partial slabs on a node.
Christoph Lameter69cb8e62011-08-09 16:12:22 -05003230 * This is called from kmem_cache_close(). We must be the last thread
3231 * using the cache and therefore we do not need to lock anymore.
Christoph Lameter81819f02007-05-06 14:49:36 -07003232 */
Christoph Lameter599870b2008-04-23 12:36:52 -07003233static void free_partial(struct kmem_cache *s, struct kmem_cache_node *n)
Christoph Lameter81819f02007-05-06 14:49:36 -07003234{
Christoph Lameter81819f02007-05-06 14:49:36 -07003235 struct page *page, *h;
3236
Christoph Lameter33b12c32008-04-25 12:22:43 -07003237 list_for_each_entry_safe(page, h, &n->partial, lru) {
Christoph Lameter81819f02007-05-06 14:49:36 -07003238 if (!page->inuse) {
Steven Rostedt1e4dd942014-02-10 14:25:46 -08003239 __remove_partial(n, page);
Christoph Lameter81819f02007-05-06 14:49:36 -07003240 discard_slab(s, page);
Christoph Lameter33b12c32008-04-25 12:22:43 -07003241 } else {
3242 list_slab_objects(s, page,
Christoph Lameter945cf2b2012-09-04 23:18:33 +00003243 "Objects remaining in %s on kmem_cache_close()");
Christoph Lameter599870b2008-04-23 12:36:52 -07003244 }
Christoph Lameter33b12c32008-04-25 12:22:43 -07003245 }
Christoph Lameter81819f02007-05-06 14:49:36 -07003246}
3247
3248/*
Christoph Lameter672bba32007-05-09 02:32:39 -07003249 * Release all resources used by a slab cache.
Christoph Lameter81819f02007-05-06 14:49:36 -07003250 */
Christoph Lameter0c710012007-07-17 04:03:24 -07003251static inline int kmem_cache_close(struct kmem_cache *s)
Christoph Lameter81819f02007-05-06 14:49:36 -07003252{
3253 int node;
Christoph Lameterfa45dc22014-08-06 16:04:09 -07003254 struct kmem_cache_node *n;
Christoph Lameter81819f02007-05-06 14:49:36 -07003255
3256 flush_all(s);
Christoph Lameter81819f02007-05-06 14:49:36 -07003257 /* Attempt to free all objects */
Christoph Lameterfa45dc22014-08-06 16:04:09 -07003258 for_each_kmem_cache_node(s, node, n) {
Christoph Lameter599870b2008-04-23 12:36:52 -07003259 free_partial(s, n);
3260 if (n->nr_partial || slabs_node(s, node))
Christoph Lameter81819f02007-05-06 14:49:36 -07003261 return 1;
3262 }
Christoph Lameter945cf2b2012-09-04 23:18:33 +00003263 free_percpu(s->cpu_slab);
Christoph Lameter81819f02007-05-06 14:49:36 -07003264 free_kmem_cache_nodes(s);
3265 return 0;
3266}
3267
Christoph Lameter945cf2b2012-09-04 23:18:33 +00003268int __kmem_cache_shutdown(struct kmem_cache *s)
Christoph Lameter81819f02007-05-06 14:49:36 -07003269{
Christoph Lameter41a21282014-05-06 12:50:08 -07003270 return kmem_cache_close(s);
Christoph Lameter81819f02007-05-06 14:49:36 -07003271}
Christoph Lameter81819f02007-05-06 14:49:36 -07003272
3273/********************************************************************
3274 * Kmalloc subsystem
3275 *******************************************************************/
3276
Christoph Lameter81819f02007-05-06 14:49:36 -07003277static int __init setup_slub_min_order(char *str)
3278{
Pekka Enberg06428782008-01-07 23:20:27 -08003279 get_option(&str, &slub_min_order);
Christoph Lameter81819f02007-05-06 14:49:36 -07003280
3281 return 1;
3282}
3283
3284__setup("slub_min_order=", setup_slub_min_order);
3285
3286static int __init setup_slub_max_order(char *str)
3287{
Pekka Enberg06428782008-01-07 23:20:27 -08003288 get_option(&str, &slub_max_order);
David Rientjes818cf592009-04-23 09:58:22 +03003289 slub_max_order = min(slub_max_order, MAX_ORDER - 1);
Christoph Lameter81819f02007-05-06 14:49:36 -07003290
3291 return 1;
3292}
3293
3294__setup("slub_max_order=", setup_slub_max_order);
3295
3296static int __init setup_slub_min_objects(char *str)
3297{
Pekka Enberg06428782008-01-07 23:20:27 -08003298 get_option(&str, &slub_min_objects);
Christoph Lameter81819f02007-05-06 14:49:36 -07003299
3300 return 1;
3301}
3302
3303__setup("slub_min_objects=", setup_slub_min_objects);
3304
Christoph Lameter81819f02007-05-06 14:49:36 -07003305void *__kmalloc(size_t size, gfp_t flags)
3306{
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003307 struct kmem_cache *s;
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03003308 void *ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07003309
Christoph Lameter95a05b42013-01-10 19:14:19 +00003310 if (unlikely(size > KMALLOC_MAX_CACHE_SIZE))
Pekka Enbergeada35e2008-02-11 22:47:46 +02003311 return kmalloc_large(size, flags);
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003312
Christoph Lameter2c59dd62013-01-10 19:14:19 +00003313 s = kmalloc_slab(size, flags);
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003314
3315 if (unlikely(ZERO_OR_NULL_PTR(s)))
Christoph Lameter6cb8f912007-07-17 04:03:22 -07003316 return s;
3317
Ezequiel Garcia2b847c32012-09-08 17:47:58 -03003318 ret = slab_alloc(s, flags, _RET_IP_);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03003319
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02003320 trace_kmalloc(_RET_IP_, ret, size, s->size, flags);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03003321
Andrey Ryabinin0316bec2015-02-13 14:39:42 -08003322 kasan_kmalloc(s, ret, size);
3323
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03003324 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07003325}
3326EXPORT_SYMBOL(__kmalloc);
3327
Namhyung Kim5d1f57e2010-09-29 21:02:15 +09003328#ifdef CONFIG_NUMA
Christoph Lameterf619cfe2008-03-01 13:56:40 -08003329static void *kmalloc_large_node(size_t size, gfp_t flags, int node)
3330{
Vegard Nossumb1eeab62008-11-25 16:55:53 +01003331 struct page *page;
Catalin Marinase4f7c0b42009-07-07 10:32:59 +01003332 void *ptr = NULL;
Christoph Lameterf619cfe2008-03-01 13:56:40 -08003333
Vladimir Davydov52383432014-06-04 16:06:39 -07003334 flags |= __GFP_COMP | __GFP_NOTRACK;
3335 page = alloc_kmem_pages_node(node, flags, get_order(size));
Christoph Lameterf619cfe2008-03-01 13:56:40 -08003336 if (page)
Catalin Marinase4f7c0b42009-07-07 10:32:59 +01003337 ptr = page_address(page);
3338
Roman Bobnievd56791b2013-10-08 15:58:57 -07003339 kmalloc_large_node_hook(ptr, size, flags);
Catalin Marinase4f7c0b42009-07-07 10:32:59 +01003340 return ptr;
Christoph Lameterf619cfe2008-03-01 13:56:40 -08003341}
3342
Christoph Lameter81819f02007-05-06 14:49:36 -07003343void *__kmalloc_node(size_t size, gfp_t flags, int node)
3344{
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003345 struct kmem_cache *s;
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03003346 void *ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07003347
Christoph Lameter95a05b42013-01-10 19:14:19 +00003348 if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) {
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03003349 ret = kmalloc_large_node(size, flags, node);
3350
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02003351 trace_kmalloc_node(_RET_IP_, ret,
3352 size, PAGE_SIZE << get_order(size),
3353 flags, node);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03003354
3355 return ret;
3356 }
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003357
Christoph Lameter2c59dd62013-01-10 19:14:19 +00003358 s = kmalloc_slab(size, flags);
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003359
3360 if (unlikely(ZERO_OR_NULL_PTR(s)))
Christoph Lameter6cb8f912007-07-17 04:03:22 -07003361 return s;
3362
Ezequiel Garcia2b847c32012-09-08 17:47:58 -03003363 ret = slab_alloc_node(s, flags, node, _RET_IP_);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03003364
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02003365 trace_kmalloc_node(_RET_IP_, ret, size, s->size, flags, node);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03003366
Andrey Ryabinin0316bec2015-02-13 14:39:42 -08003367 kasan_kmalloc(s, ret, size);
3368
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03003369 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07003370}
3371EXPORT_SYMBOL(__kmalloc_node);
3372#endif
3373
Andrey Ryabinin0316bec2015-02-13 14:39:42 -08003374static size_t __ksize(const void *object)
Christoph Lameter81819f02007-05-06 14:49:36 -07003375{
Christoph Lameter272c1d22007-06-08 13:46:49 -07003376 struct page *page;
Christoph Lameter81819f02007-05-06 14:49:36 -07003377
Christoph Lameteref8b4522007-10-16 01:24:46 -07003378 if (unlikely(object == ZERO_SIZE_PTR))
Christoph Lameter272c1d22007-06-08 13:46:49 -07003379 return 0;
3380
Vegard Nossum294a80a2007-12-04 23:45:30 -08003381 page = virt_to_head_page(object);
Vegard Nossum294a80a2007-12-04 23:45:30 -08003382
Pekka Enberg76994412008-05-22 19:22:25 +03003383 if (unlikely(!PageSlab(page))) {
3384 WARN_ON(!PageCompound(page));
Vegard Nossum294a80a2007-12-04 23:45:30 -08003385 return PAGE_SIZE << compound_order(page);
Pekka Enberg76994412008-05-22 19:22:25 +03003386 }
Christoph Lameter81819f02007-05-06 14:49:36 -07003387
Glauber Costa1b4f59e32012-10-22 18:05:36 +04003388 return slab_ksize(page->slab_cache);
Christoph Lameter81819f02007-05-06 14:49:36 -07003389}
Andrey Ryabinin0316bec2015-02-13 14:39:42 -08003390
3391size_t ksize(const void *object)
3392{
3393 size_t size = __ksize(object);
3394 /* We assume that ksize callers could use whole allocated area,
3395 so we need unpoison this area. */
3396 kasan_krealloc(object, size);
3397 return size;
3398}
Kirill A. Shutemovb1aabec2009-02-10 15:21:44 +02003399EXPORT_SYMBOL(ksize);
Christoph Lameter81819f02007-05-06 14:49:36 -07003400
3401void kfree(const void *x)
3402{
Christoph Lameter81819f02007-05-06 14:49:36 -07003403 struct page *page;
Christoph Lameter5bb983b2008-02-07 17:47:41 -08003404 void *object = (void *)x;
Christoph Lameter81819f02007-05-06 14:49:36 -07003405
Pekka Enberg2121db72009-03-25 11:05:57 +02003406 trace_kfree(_RET_IP_, x);
3407
Satyam Sharma2408c552007-10-16 01:24:44 -07003408 if (unlikely(ZERO_OR_NULL_PTR(x)))
Christoph Lameter81819f02007-05-06 14:49:36 -07003409 return;
3410
Christoph Lameterb49af682007-05-06 14:49:41 -07003411 page = virt_to_head_page(x);
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003412 if (unlikely(!PageSlab(page))) {
Christoph Lameter09375022008-05-28 10:32:22 -07003413 BUG_ON(!PageCompound(page));
Roman Bobnievd56791b2013-10-08 15:58:57 -07003414 kfree_hook(x);
Vladimir Davydov52383432014-06-04 16:06:39 -07003415 __free_kmem_pages(page, compound_order(page));
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003416 return;
3417 }
Glauber Costa1b4f59e32012-10-22 18:05:36 +04003418 slab_free(page->slab_cache, page, object, _RET_IP_);
Christoph Lameter81819f02007-05-06 14:49:36 -07003419}
3420EXPORT_SYMBOL(kfree);
3421
Vladimir Davydov832f37f2015-02-12 14:59:41 -08003422#define SHRINK_PROMOTE_MAX 32
3423
Christoph Lameter2086d262007-05-06 14:49:46 -07003424/*
Vladimir Davydov832f37f2015-02-12 14:59:41 -08003425 * kmem_cache_shrink discards empty slabs and promotes the slabs filled
3426 * up most to the head of the partial lists. New allocations will then
3427 * fill those up and thus they can be removed from the partial lists.
Christoph Lameter672bba32007-05-09 02:32:39 -07003428 *
3429 * The slabs with the least items are placed last. This results in them
3430 * being allocated from last increasing the chance that the last objects
3431 * are freed in them.
Christoph Lameter2086d262007-05-06 14:49:46 -07003432 */
Vladimir Davydovd6e0b7f2015-02-12 14:59:47 -08003433int __kmem_cache_shrink(struct kmem_cache *s, bool deactivate)
Christoph Lameter2086d262007-05-06 14:49:46 -07003434{
3435 int node;
3436 int i;
3437 struct kmem_cache_node *n;
3438 struct page *page;
3439 struct page *t;
Vladimir Davydov832f37f2015-02-12 14:59:41 -08003440 struct list_head discard;
3441 struct list_head promote[SHRINK_PROMOTE_MAX];
Christoph Lameter2086d262007-05-06 14:49:46 -07003442 unsigned long flags;
Vladimir Davydovce3712d2015-02-12 14:59:44 -08003443 int ret = 0;
Christoph Lameter2086d262007-05-06 14:49:46 -07003444
Vladimir Davydovd6e0b7f2015-02-12 14:59:47 -08003445 if (deactivate) {
3446 /*
3447 * Disable empty slabs caching. Used to avoid pinning offline
3448 * memory cgroups by kmem pages that can be freed.
3449 */
3450 s->cpu_partial = 0;
3451 s->min_partial = 0;
3452
3453 /*
3454 * s->cpu_partial is checked locklessly (see put_cpu_partial),
3455 * so we have to make sure the change is visible.
3456 */
3457 kick_all_cpus_sync();
3458 }
3459
Christoph Lameter2086d262007-05-06 14:49:46 -07003460 flush_all(s);
Christoph Lameterfa45dc22014-08-06 16:04:09 -07003461 for_each_kmem_cache_node(s, node, n) {
Vladimir Davydov832f37f2015-02-12 14:59:41 -08003462 INIT_LIST_HEAD(&discard);
3463 for (i = 0; i < SHRINK_PROMOTE_MAX; i++)
3464 INIT_LIST_HEAD(promote + i);
Christoph Lameter2086d262007-05-06 14:49:46 -07003465
3466 spin_lock_irqsave(&n->list_lock, flags);
3467
3468 /*
Vladimir Davydov832f37f2015-02-12 14:59:41 -08003469 * Build lists of slabs to discard or promote.
Christoph Lameter2086d262007-05-06 14:49:46 -07003470 *
Christoph Lameter672bba32007-05-09 02:32:39 -07003471 * Note that concurrent frees may occur while we hold the
3472 * list_lock. page->inuse here is the upper limit.
Christoph Lameter2086d262007-05-06 14:49:46 -07003473 */
3474 list_for_each_entry_safe(page, t, &n->partial, lru) {
Vladimir Davydov832f37f2015-02-12 14:59:41 -08003475 int free = page->objects - page->inuse;
3476
3477 /* Do not reread page->inuse */
3478 barrier();
3479
3480 /* We do not keep full slabs on the list */
3481 BUG_ON(free <= 0);
3482
3483 if (free == page->objects) {
3484 list_move(&page->lru, &discard);
Christoph Lameter69cb8e62011-08-09 16:12:22 -05003485 n->nr_partial--;
Vladimir Davydov832f37f2015-02-12 14:59:41 -08003486 } else if (free <= SHRINK_PROMOTE_MAX)
3487 list_move(&page->lru, promote + free - 1);
Christoph Lameter2086d262007-05-06 14:49:46 -07003488 }
3489
Christoph Lameter2086d262007-05-06 14:49:46 -07003490 /*
Vladimir Davydov832f37f2015-02-12 14:59:41 -08003491 * Promote the slabs filled up most to the head of the
3492 * partial list.
Christoph Lameter2086d262007-05-06 14:49:46 -07003493 */
Vladimir Davydov832f37f2015-02-12 14:59:41 -08003494 for (i = SHRINK_PROMOTE_MAX - 1; i >= 0; i--)
3495 list_splice(promote + i, &n->partial);
Christoph Lameter2086d262007-05-06 14:49:46 -07003496
Christoph Lameter2086d262007-05-06 14:49:46 -07003497 spin_unlock_irqrestore(&n->list_lock, flags);
Christoph Lameter69cb8e62011-08-09 16:12:22 -05003498
3499 /* Release empty slabs */
Vladimir Davydov832f37f2015-02-12 14:59:41 -08003500 list_for_each_entry_safe(page, t, &discard, lru)
Christoph Lameter69cb8e62011-08-09 16:12:22 -05003501 discard_slab(s, page);
Vladimir Davydovce3712d2015-02-12 14:59:44 -08003502
3503 if (slabs_node(s, node))
3504 ret = 1;
Christoph Lameter2086d262007-05-06 14:49:46 -07003505 }
3506
Vladimir Davydovce3712d2015-02-12 14:59:44 -08003507 return ret;
Christoph Lameter2086d262007-05-06 14:49:46 -07003508}
Christoph Lameter2086d262007-05-06 14:49:46 -07003509
Yasunori Gotob9049e22007-10-21 16:41:37 -07003510static int slab_mem_going_offline_callback(void *arg)
3511{
3512 struct kmem_cache *s;
3513
Christoph Lameter18004c52012-07-06 15:25:12 -05003514 mutex_lock(&slab_mutex);
Yasunori Gotob9049e22007-10-21 16:41:37 -07003515 list_for_each_entry(s, &slab_caches, list)
Vladimir Davydovd6e0b7f2015-02-12 14:59:47 -08003516 __kmem_cache_shrink(s, false);
Christoph Lameter18004c52012-07-06 15:25:12 -05003517 mutex_unlock(&slab_mutex);
Yasunori Gotob9049e22007-10-21 16:41:37 -07003518
3519 return 0;
3520}
3521
3522static void slab_mem_offline_callback(void *arg)
3523{
3524 struct kmem_cache_node *n;
3525 struct kmem_cache *s;
3526 struct memory_notify *marg = arg;
3527 int offline_node;
3528
Lai Jiangshanb9d5ab22012-12-11 16:01:05 -08003529 offline_node = marg->status_change_nid_normal;
Yasunori Gotob9049e22007-10-21 16:41:37 -07003530
3531 /*
3532 * If the node still has available memory. we need kmem_cache_node
3533 * for it yet.
3534 */
3535 if (offline_node < 0)
3536 return;
3537
Christoph Lameter18004c52012-07-06 15:25:12 -05003538 mutex_lock(&slab_mutex);
Yasunori Gotob9049e22007-10-21 16:41:37 -07003539 list_for_each_entry(s, &slab_caches, list) {
3540 n = get_node(s, offline_node);
3541 if (n) {
3542 /*
3543 * if n->nr_slabs > 0, slabs still exist on the node
3544 * that is going down. We were unable to free them,
Adam Buchbinderc9404c92009-12-18 15:40:42 -05003545 * and offline_pages() function shouldn't call this
Yasunori Gotob9049e22007-10-21 16:41:37 -07003546 * callback. So, we must fail.
3547 */
Christoph Lameter0f389ec2008-04-14 18:53:02 +03003548 BUG_ON(slabs_node(s, offline_node));
Yasunori Gotob9049e22007-10-21 16:41:37 -07003549
3550 s->node[offline_node] = NULL;
Christoph Lameter8de66a02010-08-25 14:51:14 -05003551 kmem_cache_free(kmem_cache_node, n);
Yasunori Gotob9049e22007-10-21 16:41:37 -07003552 }
3553 }
Christoph Lameter18004c52012-07-06 15:25:12 -05003554 mutex_unlock(&slab_mutex);
Yasunori Gotob9049e22007-10-21 16:41:37 -07003555}
3556
3557static int slab_mem_going_online_callback(void *arg)
3558{
3559 struct kmem_cache_node *n;
3560 struct kmem_cache *s;
3561 struct memory_notify *marg = arg;
Lai Jiangshanb9d5ab22012-12-11 16:01:05 -08003562 int nid = marg->status_change_nid_normal;
Yasunori Gotob9049e22007-10-21 16:41:37 -07003563 int ret = 0;
3564
3565 /*
3566 * If the node's memory is already available, then kmem_cache_node is
3567 * already created. Nothing to do.
3568 */
3569 if (nid < 0)
3570 return 0;
3571
3572 /*
Christoph Lameter0121c6192008-04-29 16:11:12 -07003573 * We are bringing a node online. No memory is available yet. We must
Yasunori Gotob9049e22007-10-21 16:41:37 -07003574 * allocate a kmem_cache_node structure in order to bring the node
3575 * online.
3576 */
Christoph Lameter18004c52012-07-06 15:25:12 -05003577 mutex_lock(&slab_mutex);
Yasunori Gotob9049e22007-10-21 16:41:37 -07003578 list_for_each_entry(s, &slab_caches, list) {
3579 /*
3580 * XXX: kmem_cache_alloc_node will fallback to other nodes
3581 * since memory is not yet available from the node that
3582 * is brought up.
3583 */
Christoph Lameter8de66a02010-08-25 14:51:14 -05003584 n = kmem_cache_alloc(kmem_cache_node, GFP_KERNEL);
Yasunori Gotob9049e22007-10-21 16:41:37 -07003585 if (!n) {
3586 ret = -ENOMEM;
3587 goto out;
3588 }
Joonsoo Kim40534972012-05-11 00:50:47 +09003589 init_kmem_cache_node(n);
Yasunori Gotob9049e22007-10-21 16:41:37 -07003590 s->node[nid] = n;
3591 }
3592out:
Christoph Lameter18004c52012-07-06 15:25:12 -05003593 mutex_unlock(&slab_mutex);
Yasunori Gotob9049e22007-10-21 16:41:37 -07003594 return ret;
3595}
3596
3597static int slab_memory_callback(struct notifier_block *self,
3598 unsigned long action, void *arg)
3599{
3600 int ret = 0;
3601
3602 switch (action) {
3603 case MEM_GOING_ONLINE:
3604 ret = slab_mem_going_online_callback(arg);
3605 break;
3606 case MEM_GOING_OFFLINE:
3607 ret = slab_mem_going_offline_callback(arg);
3608 break;
3609 case MEM_OFFLINE:
3610 case MEM_CANCEL_ONLINE:
3611 slab_mem_offline_callback(arg);
3612 break;
3613 case MEM_ONLINE:
3614 case MEM_CANCEL_OFFLINE:
3615 break;
3616 }
KAMEZAWA Hiroyukidc19f9d2008-12-01 13:13:48 -08003617 if (ret)
3618 ret = notifier_from_errno(ret);
3619 else
3620 ret = NOTIFY_OK;
Yasunori Gotob9049e22007-10-21 16:41:37 -07003621 return ret;
3622}
3623
Andrew Morton3ac38fa2013-04-29 15:08:06 -07003624static struct notifier_block slab_memory_callback_nb = {
3625 .notifier_call = slab_memory_callback,
3626 .priority = SLAB_CALLBACK_PRI,
3627};
Yasunori Gotob9049e22007-10-21 16:41:37 -07003628
Christoph Lameter81819f02007-05-06 14:49:36 -07003629/********************************************************************
3630 * Basic setup of slabs
3631 *******************************************************************/
3632
Christoph Lameter51df1142010-08-20 12:37:15 -05003633/*
3634 * Used for early kmem_cache structures that were allocated using
Christoph Lameterdffb4d62012-11-28 16:23:07 +00003635 * the page allocator. Allocate them properly then fix up the pointers
3636 * that may be pointing to the wrong kmem_cache structure.
Christoph Lameter51df1142010-08-20 12:37:15 -05003637 */
3638
Christoph Lameterdffb4d62012-11-28 16:23:07 +00003639static struct kmem_cache * __init bootstrap(struct kmem_cache *static_cache)
Christoph Lameter51df1142010-08-20 12:37:15 -05003640{
3641 int node;
Christoph Lameterdffb4d62012-11-28 16:23:07 +00003642 struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT);
Christoph Lameterfa45dc22014-08-06 16:04:09 -07003643 struct kmem_cache_node *n;
Christoph Lameter51df1142010-08-20 12:37:15 -05003644
Christoph Lameterdffb4d62012-11-28 16:23:07 +00003645 memcpy(s, static_cache, kmem_cache->object_size);
Christoph Lameter51df1142010-08-20 12:37:15 -05003646
Glauber Costa7d557b32013-02-22 20:20:00 +04003647 /*
3648 * This runs very early, and only the boot processor is supposed to be
3649 * up. Even if it weren't true, IRQs are not up so we couldn't fire
3650 * IPIs around.
3651 */
3652 __flush_cpu_slab(s, smp_processor_id());
Christoph Lameterfa45dc22014-08-06 16:04:09 -07003653 for_each_kmem_cache_node(s, node, n) {
Christoph Lameter51df1142010-08-20 12:37:15 -05003654 struct page *p;
3655
Christoph Lameterfa45dc22014-08-06 16:04:09 -07003656 list_for_each_entry(p, &n->partial, lru)
3657 p->slab_cache = s;
Christoph Lameter51df1142010-08-20 12:37:15 -05003658
Li Zefan607bf322011-04-12 15:22:26 +08003659#ifdef CONFIG_SLUB_DEBUG
Christoph Lameterfa45dc22014-08-06 16:04:09 -07003660 list_for_each_entry(p, &n->full, lru)
3661 p->slab_cache = s;
Christoph Lameter51df1142010-08-20 12:37:15 -05003662#endif
Christoph Lameter51df1142010-08-20 12:37:15 -05003663 }
Vladimir Davydovf7ce3192015-02-12 14:59:20 -08003664 slab_init_memcg_params(s);
Christoph Lameterdffb4d62012-11-28 16:23:07 +00003665 list_add(&s->list, &slab_caches);
3666 return s;
Christoph Lameter51df1142010-08-20 12:37:15 -05003667}
3668
Christoph Lameter81819f02007-05-06 14:49:36 -07003669void __init kmem_cache_init(void)
3670{
Christoph Lameterdffb4d62012-11-28 16:23:07 +00003671 static __initdata struct kmem_cache boot_kmem_cache,
3672 boot_kmem_cache_node;
Christoph Lameter51df1142010-08-20 12:37:15 -05003673
Stanislaw Gruszkafc8d8622012-01-10 15:07:32 -08003674 if (debug_guardpage_minorder())
3675 slub_max_order = 0;
3676
Christoph Lameterdffb4d62012-11-28 16:23:07 +00003677 kmem_cache_node = &boot_kmem_cache_node;
3678 kmem_cache = &boot_kmem_cache;
Christoph Lameter51df1142010-08-20 12:37:15 -05003679
Christoph Lameterdffb4d62012-11-28 16:23:07 +00003680 create_boot_cache(kmem_cache_node, "kmem_cache_node",
3681 sizeof(struct kmem_cache_node), SLAB_HWCACHE_ALIGN);
Yasunori Gotob9049e22007-10-21 16:41:37 -07003682
Andrew Morton3ac38fa2013-04-29 15:08:06 -07003683 register_hotmemory_notifier(&slab_memory_callback_nb);
Christoph Lameter81819f02007-05-06 14:49:36 -07003684
3685 /* Able to allocate the per node structures */
3686 slab_state = PARTIAL;
3687
Christoph Lameterdffb4d62012-11-28 16:23:07 +00003688 create_boot_cache(kmem_cache, "kmem_cache",
3689 offsetof(struct kmem_cache, node) +
3690 nr_node_ids * sizeof(struct kmem_cache_node *),
3691 SLAB_HWCACHE_ALIGN);
Christoph Lameter8a13a4c2012-09-04 23:18:33 +00003692
Christoph Lameterdffb4d62012-11-28 16:23:07 +00003693 kmem_cache = bootstrap(&boot_kmem_cache);
Christoph Lameter81819f02007-05-06 14:49:36 -07003694
Christoph Lameter51df1142010-08-20 12:37:15 -05003695 /*
3696 * Allocate kmem_cache_node properly from the kmem_cache slab.
3697 * kmem_cache_node is separately allocated so no need to
3698 * update any list pointers.
3699 */
Christoph Lameterdffb4d62012-11-28 16:23:07 +00003700 kmem_cache_node = bootstrap(&boot_kmem_cache_node);
Christoph Lameter51df1142010-08-20 12:37:15 -05003701
3702 /* Now we can use the kmem_cache to allocate kmalloc slabs */
Christoph Lameterf97d5f62013-01-10 19:12:17 +00003703 create_kmalloc_caches(0);
Christoph Lameter81819f02007-05-06 14:49:36 -07003704
3705#ifdef CONFIG_SMP
3706 register_cpu_notifier(&slab_notifier);
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06003707#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07003708
Fabian Frederickf9f58282014-06-04 16:06:34 -07003709 pr_info("SLUB: HWalign=%d, Order=%d-%d, MinObjects=%d, CPUs=%d, Nodes=%d\n",
Christoph Lameterf97d5f62013-01-10 19:12:17 +00003710 cache_line_size(),
Christoph Lameter81819f02007-05-06 14:49:36 -07003711 slub_min_order, slub_max_order, slub_min_objects,
3712 nr_cpu_ids, nr_node_ids);
3713}
3714
Pekka Enberg7e85ee02009-06-12 14:03:06 +03003715void __init kmem_cache_init_late(void)
3716{
Pekka Enberg7e85ee02009-06-12 14:03:06 +03003717}
3718
Glauber Costa2633d7a2012-12-18 14:22:34 -08003719struct kmem_cache *
Vladimir Davydova44cb942014-04-07 15:39:23 -07003720__kmem_cache_alias(const char *name, size_t size, size_t align,
3721 unsigned long flags, void (*ctor)(void *))
Christoph Lameter81819f02007-05-06 14:49:36 -07003722{
Vladimir Davydov426589f2015-02-12 14:59:23 -08003723 struct kmem_cache *s, *c;
Christoph Lameter81819f02007-05-06 14:49:36 -07003724
Vladimir Davydova44cb942014-04-07 15:39:23 -07003725 s = find_mergeable(size, align, flags, name, ctor);
Christoph Lameter81819f02007-05-06 14:49:36 -07003726 if (s) {
3727 s->refcount++;
Vladimir Davydov84d0ddd2014-04-07 15:39:29 -07003728
Christoph Lameter81819f02007-05-06 14:49:36 -07003729 /*
3730 * Adjust the object sizes so that we clear
3731 * the complete object on kzalloc.
3732 */
Christoph Lameter3b0efdf2012-06-13 10:24:57 -05003733 s->object_size = max(s->object_size, (int)size);
Christoph Lameter81819f02007-05-06 14:49:36 -07003734 s->inuse = max_t(int, s->inuse, ALIGN(size, sizeof(void *)));
Christoph Lameter6446faa2008-02-15 23:45:26 -08003735
Vladimir Davydov426589f2015-02-12 14:59:23 -08003736 for_each_memcg_cache(c, s) {
Vladimir Davydov84d0ddd2014-04-07 15:39:29 -07003737 c->object_size = s->object_size;
3738 c->inuse = max_t(int, c->inuse,
3739 ALIGN(size, sizeof(void *)));
3740 }
3741
David Rientjes7b8f3b62008-12-17 22:09:46 -08003742 if (sysfs_slab_alias(s, name)) {
David Rientjes7b8f3b62008-12-17 22:09:46 -08003743 s->refcount--;
Christoph Lametercbb79692012-09-05 00:18:32 +00003744 s = NULL;
David Rientjes7b8f3b62008-12-17 22:09:46 -08003745 }
Christoph Lametera0e1d1b2007-07-17 04:03:31 -07003746 }
Christoph Lameter6446faa2008-02-15 23:45:26 -08003747
Christoph Lametercbb79692012-09-05 00:18:32 +00003748 return s;
3749}
Pekka Enberg84c1cf62010-09-14 23:21:12 +03003750
Christoph Lameter8a13a4c2012-09-04 23:18:33 +00003751int __kmem_cache_create(struct kmem_cache *s, unsigned long flags)
Christoph Lametercbb79692012-09-05 00:18:32 +00003752{
Pekka Enbergaac3a162012-09-05 12:07:44 +03003753 int err;
Christoph Lameter20cea962012-07-06 15:25:13 -05003754
Pekka Enbergaac3a162012-09-05 12:07:44 +03003755 err = kmem_cache_open(s, flags);
3756 if (err)
3757 return err;
Christoph Lameter20cea962012-07-06 15:25:13 -05003758
Christoph Lameter45530c42012-11-28 16:23:07 +00003759 /* Mutex is not taken during early boot */
3760 if (slab_state <= UP)
3761 return 0;
3762
Glauber Costa107dab52012-12-18 14:23:05 -08003763 memcg_propagate_slab_attrs(s);
Pekka Enbergaac3a162012-09-05 12:07:44 +03003764 err = sysfs_slab_add(s);
Pekka Enbergaac3a162012-09-05 12:07:44 +03003765 if (err)
3766 kmem_cache_close(s);
3767
3768 return err;
Christoph Lameter81819f02007-05-06 14:49:36 -07003769}
Christoph Lameter81819f02007-05-06 14:49:36 -07003770
Christoph Lameter81819f02007-05-06 14:49:36 -07003771#ifdef CONFIG_SMP
Christoph Lameter27390bc2007-06-01 00:47:09 -07003772/*
Christoph Lameter672bba32007-05-09 02:32:39 -07003773 * Use the cpu notifier to insure that the cpu slabs are flushed when
3774 * necessary.
Christoph Lameter81819f02007-05-06 14:49:36 -07003775 */
Paul Gortmaker0db06282013-06-19 14:53:51 -04003776static int slab_cpuup_callback(struct notifier_block *nfb,
Christoph Lameter81819f02007-05-06 14:49:36 -07003777 unsigned long action, void *hcpu)
3778{
3779 long cpu = (long)hcpu;
Christoph Lameter5b95a4ac2007-07-17 04:03:19 -07003780 struct kmem_cache *s;
3781 unsigned long flags;
Christoph Lameter81819f02007-05-06 14:49:36 -07003782
3783 switch (action) {
3784 case CPU_UP_CANCELED:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07003785 case CPU_UP_CANCELED_FROZEN:
Christoph Lameter81819f02007-05-06 14:49:36 -07003786 case CPU_DEAD:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07003787 case CPU_DEAD_FROZEN:
Christoph Lameter18004c52012-07-06 15:25:12 -05003788 mutex_lock(&slab_mutex);
Christoph Lameter5b95a4ac2007-07-17 04:03:19 -07003789 list_for_each_entry(s, &slab_caches, list) {
3790 local_irq_save(flags);
3791 __flush_cpu_slab(s, cpu);
3792 local_irq_restore(flags);
3793 }
Christoph Lameter18004c52012-07-06 15:25:12 -05003794 mutex_unlock(&slab_mutex);
Christoph Lameter81819f02007-05-06 14:49:36 -07003795 break;
3796 default:
3797 break;
3798 }
3799 return NOTIFY_OK;
3800}
3801
Paul Gortmaker0db06282013-06-19 14:53:51 -04003802static struct notifier_block slab_notifier = {
Ingo Molnar3adbefe2008-02-05 17:57:39 -08003803 .notifier_call = slab_cpuup_callback
Pekka Enberg06428782008-01-07 23:20:27 -08003804};
Christoph Lameter81819f02007-05-06 14:49:36 -07003805
3806#endif
3807
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03003808void *__kmalloc_track_caller(size_t size, gfp_t gfpflags, unsigned long caller)
Christoph Lameter81819f02007-05-06 14:49:36 -07003809{
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003810 struct kmem_cache *s;
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03003811 void *ret;
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003812
Christoph Lameter95a05b42013-01-10 19:14:19 +00003813 if (unlikely(size > KMALLOC_MAX_CACHE_SIZE))
Pekka Enbergeada35e2008-02-11 22:47:46 +02003814 return kmalloc_large(size, gfpflags);
3815
Christoph Lameter2c59dd62013-01-10 19:14:19 +00003816 s = kmalloc_slab(size, gfpflags);
Christoph Lameter81819f02007-05-06 14:49:36 -07003817
Satyam Sharma2408c552007-10-16 01:24:44 -07003818 if (unlikely(ZERO_OR_NULL_PTR(s)))
Christoph Lameter6cb8f912007-07-17 04:03:22 -07003819 return s;
Christoph Lameter81819f02007-05-06 14:49:36 -07003820
Ezequiel Garcia2b847c32012-09-08 17:47:58 -03003821 ret = slab_alloc(s, gfpflags, caller);
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03003822
Lucas De Marchi25985ed2011-03-30 22:57:33 -03003823 /* Honor the call site pointer we received. */
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02003824 trace_kmalloc(caller, ret, size, s->size, gfpflags);
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03003825
3826 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07003827}
3828
Namhyung Kim5d1f57e2010-09-29 21:02:15 +09003829#ifdef CONFIG_NUMA
Christoph Lameter81819f02007-05-06 14:49:36 -07003830void *__kmalloc_node_track_caller(size_t size, gfp_t gfpflags,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03003831 int node, unsigned long caller)
Christoph Lameter81819f02007-05-06 14:49:36 -07003832{
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003833 struct kmem_cache *s;
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03003834 void *ret;
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003835
Christoph Lameter95a05b42013-01-10 19:14:19 +00003836 if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) {
Xiaotian Fengd3e14aa2010-04-08 17:26:44 +08003837 ret = kmalloc_large_node(size, gfpflags, node);
3838
3839 trace_kmalloc_node(caller, ret,
3840 size, PAGE_SIZE << get_order(size),
3841 gfpflags, node);
3842
3843 return ret;
3844 }
Pekka Enbergeada35e2008-02-11 22:47:46 +02003845
Christoph Lameter2c59dd62013-01-10 19:14:19 +00003846 s = kmalloc_slab(size, gfpflags);
Christoph Lameter81819f02007-05-06 14:49:36 -07003847
Satyam Sharma2408c552007-10-16 01:24:44 -07003848 if (unlikely(ZERO_OR_NULL_PTR(s)))
Christoph Lameter6cb8f912007-07-17 04:03:22 -07003849 return s;
Christoph Lameter81819f02007-05-06 14:49:36 -07003850
Ezequiel Garcia2b847c32012-09-08 17:47:58 -03003851 ret = slab_alloc_node(s, gfpflags, node, caller);
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03003852
Lucas De Marchi25985ed2011-03-30 22:57:33 -03003853 /* Honor the call site pointer we received. */
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02003854 trace_kmalloc_node(caller, ret, size, s->size, gfpflags, node);
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03003855
3856 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07003857}
Namhyung Kim5d1f57e2010-09-29 21:02:15 +09003858#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07003859
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05003860#ifdef CONFIG_SYSFS
Christoph Lameter205ab992008-04-14 19:11:40 +03003861static int count_inuse(struct page *page)
3862{
3863 return page->inuse;
3864}
3865
3866static int count_total(struct page *page)
3867{
3868 return page->objects;
3869}
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05003870#endif
Christoph Lameter205ab992008-04-14 19:11:40 +03003871
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05003872#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter434e2452007-07-17 04:03:30 -07003873static int validate_slab(struct kmem_cache *s, struct page *page,
3874 unsigned long *map)
Christoph Lameter53e15af2007-05-06 14:49:43 -07003875{
3876 void *p;
Christoph Lametera973e9d2008-03-01 13:40:44 -08003877 void *addr = page_address(page);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003878
3879 if (!check_slab(s, page) ||
3880 !on_freelist(s, page, NULL))
3881 return 0;
3882
3883 /* Now we know that a valid freelist exists */
Christoph Lameter39b26462008-04-14 19:11:30 +03003884 bitmap_zero(map, page->objects);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003885
Christoph Lameter5f80b132011-04-15 14:48:13 -05003886 get_map(s, page, map);
3887 for_each_object(p, s, addr, page->objects) {
3888 if (test_bit(slab_index(p, s, addr), map))
3889 if (!check_object(s, page, p, SLUB_RED_INACTIVE))
3890 return 0;
Christoph Lameter53e15af2007-05-06 14:49:43 -07003891 }
3892
Christoph Lameter224a88b2008-04-14 19:11:31 +03003893 for_each_object(p, s, addr, page->objects)
Christoph Lameter7656c722007-05-09 02:32:40 -07003894 if (!test_bit(slab_index(p, s, addr), map))
Tero Roponen37d57442010-12-01 20:04:20 +02003895 if (!check_object(s, page, p, SLUB_RED_ACTIVE))
Christoph Lameter53e15af2007-05-06 14:49:43 -07003896 return 0;
3897 return 1;
3898}
3899
Christoph Lameter434e2452007-07-17 04:03:30 -07003900static void validate_slab_slab(struct kmem_cache *s, struct page *page,
3901 unsigned long *map)
Christoph Lameter53e15af2007-05-06 14:49:43 -07003902{
Christoph Lameter881db7f2011-06-01 12:25:53 -05003903 slab_lock(page);
3904 validate_slab(s, page, map);
3905 slab_unlock(page);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003906}
3907
Christoph Lameter434e2452007-07-17 04:03:30 -07003908static int validate_slab_node(struct kmem_cache *s,
3909 struct kmem_cache_node *n, unsigned long *map)
Christoph Lameter53e15af2007-05-06 14:49:43 -07003910{
3911 unsigned long count = 0;
3912 struct page *page;
3913 unsigned long flags;
3914
3915 spin_lock_irqsave(&n->list_lock, flags);
3916
3917 list_for_each_entry(page, &n->partial, lru) {
Christoph Lameter434e2452007-07-17 04:03:30 -07003918 validate_slab_slab(s, page, map);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003919 count++;
3920 }
3921 if (count != n->nr_partial)
Fabian Frederickf9f58282014-06-04 16:06:34 -07003922 pr_err("SLUB %s: %ld partial slabs counted but counter=%ld\n",
3923 s->name, count, n->nr_partial);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003924
3925 if (!(s->flags & SLAB_STORE_USER))
3926 goto out;
3927
3928 list_for_each_entry(page, &n->full, lru) {
Christoph Lameter434e2452007-07-17 04:03:30 -07003929 validate_slab_slab(s, page, map);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003930 count++;
3931 }
3932 if (count != atomic_long_read(&n->nr_slabs))
Fabian Frederickf9f58282014-06-04 16:06:34 -07003933 pr_err("SLUB: %s %ld slabs counted but counter=%ld\n",
3934 s->name, count, atomic_long_read(&n->nr_slabs));
Christoph Lameter53e15af2007-05-06 14:49:43 -07003935
3936out:
3937 spin_unlock_irqrestore(&n->list_lock, flags);
3938 return count;
3939}
3940
Christoph Lameter434e2452007-07-17 04:03:30 -07003941static long validate_slab_cache(struct kmem_cache *s)
Christoph Lameter53e15af2007-05-06 14:49:43 -07003942{
3943 int node;
3944 unsigned long count = 0;
Christoph Lameter205ab992008-04-14 19:11:40 +03003945 unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) *
Christoph Lameter434e2452007-07-17 04:03:30 -07003946 sizeof(unsigned long), GFP_KERNEL);
Christoph Lameterfa45dc22014-08-06 16:04:09 -07003947 struct kmem_cache_node *n;
Christoph Lameter434e2452007-07-17 04:03:30 -07003948
3949 if (!map)
3950 return -ENOMEM;
Christoph Lameter53e15af2007-05-06 14:49:43 -07003951
3952 flush_all(s);
Christoph Lameterfa45dc22014-08-06 16:04:09 -07003953 for_each_kmem_cache_node(s, node, n)
Christoph Lameter434e2452007-07-17 04:03:30 -07003954 count += validate_slab_node(s, n, map);
Christoph Lameter434e2452007-07-17 04:03:30 -07003955 kfree(map);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003956 return count;
3957}
Christoph Lameter88a420e2007-05-06 14:49:45 -07003958/*
Christoph Lameter672bba32007-05-09 02:32:39 -07003959 * Generate lists of code addresses where slabcache objects are allocated
Christoph Lameter88a420e2007-05-06 14:49:45 -07003960 * and freed.
3961 */
3962
3963struct location {
3964 unsigned long count;
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03003965 unsigned long addr;
Christoph Lameter45edfa52007-05-09 02:32:45 -07003966 long long sum_time;
3967 long min_time;
3968 long max_time;
3969 long min_pid;
3970 long max_pid;
Rusty Russell174596a2009-01-01 10:12:29 +10303971 DECLARE_BITMAP(cpus, NR_CPUS);
Christoph Lameter45edfa52007-05-09 02:32:45 -07003972 nodemask_t nodes;
Christoph Lameter88a420e2007-05-06 14:49:45 -07003973};
3974
3975struct loc_track {
3976 unsigned long max;
3977 unsigned long count;
3978 struct location *loc;
3979};
3980
3981static void free_loc_track(struct loc_track *t)
3982{
3983 if (t->max)
3984 free_pages((unsigned long)t->loc,
3985 get_order(sizeof(struct location) * t->max));
3986}
3987
Christoph Lameter68dff6a2007-07-17 04:03:20 -07003988static int alloc_loc_track(struct loc_track *t, unsigned long max, gfp_t flags)
Christoph Lameter88a420e2007-05-06 14:49:45 -07003989{
3990 struct location *l;
3991 int order;
3992
Christoph Lameter88a420e2007-05-06 14:49:45 -07003993 order = get_order(sizeof(struct location) * max);
3994
Christoph Lameter68dff6a2007-07-17 04:03:20 -07003995 l = (void *)__get_free_pages(flags, order);
Christoph Lameter88a420e2007-05-06 14:49:45 -07003996 if (!l)
3997 return 0;
3998
3999 if (t->count) {
4000 memcpy(l, t->loc, sizeof(struct location) * t->count);
4001 free_loc_track(t);
4002 }
4003 t->max = max;
4004 t->loc = l;
4005 return 1;
4006}
4007
4008static int add_location(struct loc_track *t, struct kmem_cache *s,
Christoph Lameter45edfa52007-05-09 02:32:45 -07004009 const struct track *track)
Christoph Lameter88a420e2007-05-06 14:49:45 -07004010{
4011 long start, end, pos;
4012 struct location *l;
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03004013 unsigned long caddr;
Christoph Lameter45edfa52007-05-09 02:32:45 -07004014 unsigned long age = jiffies - track->when;
Christoph Lameter88a420e2007-05-06 14:49:45 -07004015
4016 start = -1;
4017 end = t->count;
4018
4019 for ( ; ; ) {
4020 pos = start + (end - start + 1) / 2;
4021
4022 /*
4023 * There is nothing at "end". If we end up there
4024 * we need to add something to before end.
4025 */
4026 if (pos == end)
4027 break;
4028
4029 caddr = t->loc[pos].addr;
Christoph Lameter45edfa52007-05-09 02:32:45 -07004030 if (track->addr == caddr) {
4031
4032 l = &t->loc[pos];
4033 l->count++;
4034 if (track->when) {
4035 l->sum_time += age;
4036 if (age < l->min_time)
4037 l->min_time = age;
4038 if (age > l->max_time)
4039 l->max_time = age;
4040
4041 if (track->pid < l->min_pid)
4042 l->min_pid = track->pid;
4043 if (track->pid > l->max_pid)
4044 l->max_pid = track->pid;
4045
Rusty Russell174596a2009-01-01 10:12:29 +10304046 cpumask_set_cpu(track->cpu,
4047 to_cpumask(l->cpus));
Christoph Lameter45edfa52007-05-09 02:32:45 -07004048 }
4049 node_set(page_to_nid(virt_to_page(track)), l->nodes);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004050 return 1;
4051 }
4052
Christoph Lameter45edfa52007-05-09 02:32:45 -07004053 if (track->addr < caddr)
Christoph Lameter88a420e2007-05-06 14:49:45 -07004054 end = pos;
4055 else
4056 start = pos;
4057 }
4058
4059 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07004060 * Not found. Insert new tracking element.
Christoph Lameter88a420e2007-05-06 14:49:45 -07004061 */
Christoph Lameter68dff6a2007-07-17 04:03:20 -07004062 if (t->count >= t->max && !alloc_loc_track(t, 2 * t->max, GFP_ATOMIC))
Christoph Lameter88a420e2007-05-06 14:49:45 -07004063 return 0;
4064
4065 l = t->loc + pos;
4066 if (pos < t->count)
4067 memmove(l + 1, l,
4068 (t->count - pos) * sizeof(struct location));
4069 t->count++;
4070 l->count = 1;
Christoph Lameter45edfa52007-05-09 02:32:45 -07004071 l->addr = track->addr;
4072 l->sum_time = age;
4073 l->min_time = age;
4074 l->max_time = age;
4075 l->min_pid = track->pid;
4076 l->max_pid = track->pid;
Rusty Russell174596a2009-01-01 10:12:29 +10304077 cpumask_clear(to_cpumask(l->cpus));
4078 cpumask_set_cpu(track->cpu, to_cpumask(l->cpus));
Christoph Lameter45edfa52007-05-09 02:32:45 -07004079 nodes_clear(l->nodes);
4080 node_set(page_to_nid(virt_to_page(track)), l->nodes);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004081 return 1;
4082}
4083
4084static void process_slab(struct loc_track *t, struct kmem_cache *s,
Eric Dumazetbbd7d572010-03-24 22:25:47 +01004085 struct page *page, enum track_item alloc,
Namhyung Kima5dd5c12010-09-29 21:02:13 +09004086 unsigned long *map)
Christoph Lameter88a420e2007-05-06 14:49:45 -07004087{
Christoph Lametera973e9d2008-03-01 13:40:44 -08004088 void *addr = page_address(page);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004089 void *p;
4090
Christoph Lameter39b26462008-04-14 19:11:30 +03004091 bitmap_zero(map, page->objects);
Christoph Lameter5f80b132011-04-15 14:48:13 -05004092 get_map(s, page, map);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004093
Christoph Lameter224a88b2008-04-14 19:11:31 +03004094 for_each_object(p, s, addr, page->objects)
Christoph Lameter45edfa52007-05-09 02:32:45 -07004095 if (!test_bit(slab_index(p, s, addr), map))
4096 add_location(t, s, get_track(s, p, alloc));
Christoph Lameter88a420e2007-05-06 14:49:45 -07004097}
4098
4099static int list_locations(struct kmem_cache *s, char *buf,
4100 enum track_item alloc)
4101{
Harvey Harrisone374d482008-01-31 15:20:50 -08004102 int len = 0;
Christoph Lameter88a420e2007-05-06 14:49:45 -07004103 unsigned long i;
Christoph Lameter68dff6a2007-07-17 04:03:20 -07004104 struct loc_track t = { 0, 0, NULL };
Christoph Lameter88a420e2007-05-06 14:49:45 -07004105 int node;
Eric Dumazetbbd7d572010-03-24 22:25:47 +01004106 unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) *
4107 sizeof(unsigned long), GFP_KERNEL);
Christoph Lameterfa45dc22014-08-06 16:04:09 -07004108 struct kmem_cache_node *n;
Christoph Lameter88a420e2007-05-06 14:49:45 -07004109
Eric Dumazetbbd7d572010-03-24 22:25:47 +01004110 if (!map || !alloc_loc_track(&t, PAGE_SIZE / sizeof(struct location),
4111 GFP_TEMPORARY)) {
4112 kfree(map);
Christoph Lameter68dff6a2007-07-17 04:03:20 -07004113 return sprintf(buf, "Out of memory\n");
Eric Dumazetbbd7d572010-03-24 22:25:47 +01004114 }
Christoph Lameter88a420e2007-05-06 14:49:45 -07004115 /* Push back cpu slabs */
4116 flush_all(s);
4117
Christoph Lameterfa45dc22014-08-06 16:04:09 -07004118 for_each_kmem_cache_node(s, node, n) {
Christoph Lameter88a420e2007-05-06 14:49:45 -07004119 unsigned long flags;
4120 struct page *page;
4121
Christoph Lameter9e869432007-08-22 14:01:56 -07004122 if (!atomic_long_read(&n->nr_slabs))
Christoph Lameter88a420e2007-05-06 14:49:45 -07004123 continue;
4124
4125 spin_lock_irqsave(&n->list_lock, flags);
4126 list_for_each_entry(page, &n->partial, lru)
Eric Dumazetbbd7d572010-03-24 22:25:47 +01004127 process_slab(&t, s, page, alloc, map);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004128 list_for_each_entry(page, &n->full, lru)
Eric Dumazetbbd7d572010-03-24 22:25:47 +01004129 process_slab(&t, s, page, alloc, map);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004130 spin_unlock_irqrestore(&n->list_lock, flags);
4131 }
4132
4133 for (i = 0; i < t.count; i++) {
Christoph Lameter45edfa52007-05-09 02:32:45 -07004134 struct location *l = &t.loc[i];
Christoph Lameter88a420e2007-05-06 14:49:45 -07004135
Hugh Dickins9c246242008-12-09 13:14:27 -08004136 if (len > PAGE_SIZE - KSYM_SYMBOL_LEN - 100)
Christoph Lameter88a420e2007-05-06 14:49:45 -07004137 break;
Harvey Harrisone374d482008-01-31 15:20:50 -08004138 len += sprintf(buf + len, "%7ld ", l->count);
Christoph Lameter45edfa52007-05-09 02:32:45 -07004139
4140 if (l->addr)
Joe Perches62c70bc2011-01-13 15:45:52 -08004141 len += sprintf(buf + len, "%pS", (void *)l->addr);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004142 else
Harvey Harrisone374d482008-01-31 15:20:50 -08004143 len += sprintf(buf + len, "<not-available>");
Christoph Lameter45edfa52007-05-09 02:32:45 -07004144
4145 if (l->sum_time != l->min_time) {
Harvey Harrisone374d482008-01-31 15:20:50 -08004146 len += sprintf(buf + len, " age=%ld/%ld/%ld",
Roman Zippelf8bd2252008-05-01 04:34:31 -07004147 l->min_time,
4148 (long)div_u64(l->sum_time, l->count),
4149 l->max_time);
Christoph Lameter45edfa52007-05-09 02:32:45 -07004150 } else
Harvey Harrisone374d482008-01-31 15:20:50 -08004151 len += sprintf(buf + len, " age=%ld",
Christoph Lameter45edfa52007-05-09 02:32:45 -07004152 l->min_time);
4153
4154 if (l->min_pid != l->max_pid)
Harvey Harrisone374d482008-01-31 15:20:50 -08004155 len += sprintf(buf + len, " pid=%ld-%ld",
Christoph Lameter45edfa52007-05-09 02:32:45 -07004156 l->min_pid, l->max_pid);
4157 else
Harvey Harrisone374d482008-01-31 15:20:50 -08004158 len += sprintf(buf + len, " pid=%ld",
Christoph Lameter45edfa52007-05-09 02:32:45 -07004159 l->min_pid);
4160
Rusty Russell174596a2009-01-01 10:12:29 +10304161 if (num_online_cpus() > 1 &&
4162 !cpumask_empty(to_cpumask(l->cpus)) &&
Tejun Heo5024c1d2015-02-13 14:37:59 -08004163 len < PAGE_SIZE - 60)
4164 len += scnprintf(buf + len, PAGE_SIZE - len - 50,
4165 " cpus=%*pbl",
4166 cpumask_pr_args(to_cpumask(l->cpus)));
Christoph Lameter45edfa52007-05-09 02:32:45 -07004167
Christoph Lameter62bc62a2009-06-16 15:32:15 -07004168 if (nr_online_nodes > 1 && !nodes_empty(l->nodes) &&
Tejun Heo5024c1d2015-02-13 14:37:59 -08004169 len < PAGE_SIZE - 60)
4170 len += scnprintf(buf + len, PAGE_SIZE - len - 50,
4171 " nodes=%*pbl",
4172 nodemask_pr_args(&l->nodes));
Christoph Lameter45edfa52007-05-09 02:32:45 -07004173
Harvey Harrisone374d482008-01-31 15:20:50 -08004174 len += sprintf(buf + len, "\n");
Christoph Lameter88a420e2007-05-06 14:49:45 -07004175 }
4176
4177 free_loc_track(&t);
Eric Dumazetbbd7d572010-03-24 22:25:47 +01004178 kfree(map);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004179 if (!t.count)
Harvey Harrisone374d482008-01-31 15:20:50 -08004180 len += sprintf(buf, "No data\n");
4181 return len;
Christoph Lameter88a420e2007-05-06 14:49:45 -07004182}
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004183#endif
Christoph Lameter88a420e2007-05-06 14:49:45 -07004184
Christoph Lametera5a84752010-10-05 13:57:27 -05004185#ifdef SLUB_RESILIENCY_TEST
David Rientjesc07b8182014-08-06 16:04:16 -07004186static void __init resiliency_test(void)
Christoph Lametera5a84752010-10-05 13:57:27 -05004187{
4188 u8 *p;
4189
Christoph Lameter95a05b42013-01-10 19:14:19 +00004190 BUILD_BUG_ON(KMALLOC_MIN_SIZE > 16 || KMALLOC_SHIFT_HIGH < 10);
Christoph Lametera5a84752010-10-05 13:57:27 -05004191
Fabian Frederickf9f58282014-06-04 16:06:34 -07004192 pr_err("SLUB resiliency testing\n");
4193 pr_err("-----------------------\n");
4194 pr_err("A. Corruption after allocation\n");
Christoph Lametera5a84752010-10-05 13:57:27 -05004195
4196 p = kzalloc(16, GFP_KERNEL);
4197 p[16] = 0x12;
Fabian Frederickf9f58282014-06-04 16:06:34 -07004198 pr_err("\n1. kmalloc-16: Clobber Redzone/next pointer 0x12->0x%p\n\n",
4199 p + 16);
Christoph Lametera5a84752010-10-05 13:57:27 -05004200
4201 validate_slab_cache(kmalloc_caches[4]);
4202
4203 /* Hmmm... The next two are dangerous */
4204 p = kzalloc(32, GFP_KERNEL);
4205 p[32 + sizeof(void *)] = 0x34;
Fabian Frederickf9f58282014-06-04 16:06:34 -07004206 pr_err("\n2. kmalloc-32: Clobber next pointer/next slab 0x34 -> -0x%p\n",
4207 p);
4208 pr_err("If allocated object is overwritten then not detectable\n\n");
Christoph Lametera5a84752010-10-05 13:57:27 -05004209
4210 validate_slab_cache(kmalloc_caches[5]);
4211 p = kzalloc(64, GFP_KERNEL);
4212 p += 64 + (get_cycles() & 0xff) * sizeof(void *);
4213 *p = 0x56;
Fabian Frederickf9f58282014-06-04 16:06:34 -07004214 pr_err("\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n",
4215 p);
4216 pr_err("If allocated object is overwritten then not detectable\n\n");
Christoph Lametera5a84752010-10-05 13:57:27 -05004217 validate_slab_cache(kmalloc_caches[6]);
4218
Fabian Frederickf9f58282014-06-04 16:06:34 -07004219 pr_err("\nB. Corruption after free\n");
Christoph Lametera5a84752010-10-05 13:57:27 -05004220 p = kzalloc(128, GFP_KERNEL);
4221 kfree(p);
4222 *p = 0x78;
Fabian Frederickf9f58282014-06-04 16:06:34 -07004223 pr_err("1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p);
Christoph Lametera5a84752010-10-05 13:57:27 -05004224 validate_slab_cache(kmalloc_caches[7]);
4225
4226 p = kzalloc(256, GFP_KERNEL);
4227 kfree(p);
4228 p[50] = 0x9a;
Fabian Frederickf9f58282014-06-04 16:06:34 -07004229 pr_err("\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n", p);
Christoph Lametera5a84752010-10-05 13:57:27 -05004230 validate_slab_cache(kmalloc_caches[8]);
4231
4232 p = kzalloc(512, GFP_KERNEL);
4233 kfree(p);
4234 p[512] = 0xab;
Fabian Frederickf9f58282014-06-04 16:06:34 -07004235 pr_err("\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p);
Christoph Lametera5a84752010-10-05 13:57:27 -05004236 validate_slab_cache(kmalloc_caches[9]);
4237}
4238#else
4239#ifdef CONFIG_SYSFS
4240static void resiliency_test(void) {};
4241#endif
4242#endif
4243
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004244#ifdef CONFIG_SYSFS
Christoph Lameter81819f02007-05-06 14:49:36 -07004245enum slab_stat_type {
Christoph Lameter205ab992008-04-14 19:11:40 +03004246 SL_ALL, /* All slabs */
4247 SL_PARTIAL, /* Only partially allocated slabs */
4248 SL_CPU, /* Only slabs used for cpu caches */
4249 SL_OBJECTS, /* Determine allocated objects not slabs */
4250 SL_TOTAL /* Determine object capacity not slabs */
Christoph Lameter81819f02007-05-06 14:49:36 -07004251};
4252
Christoph Lameter205ab992008-04-14 19:11:40 +03004253#define SO_ALL (1 << SL_ALL)
Christoph Lameter81819f02007-05-06 14:49:36 -07004254#define SO_PARTIAL (1 << SL_PARTIAL)
4255#define SO_CPU (1 << SL_CPU)
4256#define SO_OBJECTS (1 << SL_OBJECTS)
Christoph Lameter205ab992008-04-14 19:11:40 +03004257#define SO_TOTAL (1 << SL_TOTAL)
Christoph Lameter81819f02007-05-06 14:49:36 -07004258
Cyrill Gorcunov62e5c4b2008-03-02 23:28:24 +03004259static ssize_t show_slab_objects(struct kmem_cache *s,
4260 char *buf, unsigned long flags)
Christoph Lameter81819f02007-05-06 14:49:36 -07004261{
4262 unsigned long total = 0;
Christoph Lameter81819f02007-05-06 14:49:36 -07004263 int node;
4264 int x;
4265 unsigned long *nodes;
Christoph Lameter81819f02007-05-06 14:49:36 -07004266
Chen Gange35e1a92013-07-12 08:23:48 +08004267 nodes = kzalloc(sizeof(unsigned long) * nr_node_ids, GFP_KERNEL);
Cyrill Gorcunov62e5c4b2008-03-02 23:28:24 +03004268 if (!nodes)
4269 return -ENOMEM;
Christoph Lameter81819f02007-05-06 14:49:36 -07004270
Christoph Lameter205ab992008-04-14 19:11:40 +03004271 if (flags & SO_CPU) {
4272 int cpu;
Christoph Lameter81819f02007-05-06 14:49:36 -07004273
Christoph Lameter205ab992008-04-14 19:11:40 +03004274 for_each_possible_cpu(cpu) {
Chen Gangd0e0ac92013-07-15 09:05:29 +08004275 struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab,
4276 cpu);
Christoph Lameterec3ab082012-05-09 10:09:56 -05004277 int node;
Christoph Lameter49e22582011-08-09 16:12:27 -05004278 struct page *page;
Christoph Lameterdfb4f092007-10-16 01:26:05 -07004279
Eric Dumazetbc6697d2011-11-22 16:02:02 +01004280 page = ACCESS_ONCE(c->page);
Christoph Lameterec3ab082012-05-09 10:09:56 -05004281 if (!page)
4282 continue;
Christoph Lameter205ab992008-04-14 19:11:40 +03004283
Christoph Lameterec3ab082012-05-09 10:09:56 -05004284 node = page_to_nid(page);
4285 if (flags & SO_TOTAL)
4286 x = page->objects;
4287 else if (flags & SO_OBJECTS)
4288 x = page->inuse;
4289 else
4290 x = 1;
Christoph Lameter49e22582011-08-09 16:12:27 -05004291
Christoph Lameterec3ab082012-05-09 10:09:56 -05004292 total += x;
4293 nodes[node] += x;
4294
4295 page = ACCESS_ONCE(c->partial);
Christoph Lameter49e22582011-08-09 16:12:27 -05004296 if (page) {
Li Zefan8afb1472013-09-10 11:43:37 +08004297 node = page_to_nid(page);
4298 if (flags & SO_TOTAL)
4299 WARN_ON_ONCE(1);
4300 else if (flags & SO_OBJECTS)
4301 WARN_ON_ONCE(1);
4302 else
4303 x = page->pages;
Eric Dumazetbc6697d2011-11-22 16:02:02 +01004304 total += x;
4305 nodes[node] += x;
Christoph Lameter49e22582011-08-09 16:12:27 -05004306 }
Christoph Lameter81819f02007-05-06 14:49:36 -07004307 }
4308 }
4309
Vladimir Davydovbfc8c902014-06-04 16:07:18 -07004310 get_online_mems();
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004311#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter205ab992008-04-14 19:11:40 +03004312 if (flags & SO_ALL) {
Christoph Lameterfa45dc22014-08-06 16:04:09 -07004313 struct kmem_cache_node *n;
4314
4315 for_each_kmem_cache_node(s, node, n) {
Christoph Lameter81819f02007-05-06 14:49:36 -07004316
Chen Gangd0e0ac92013-07-15 09:05:29 +08004317 if (flags & SO_TOTAL)
4318 x = atomic_long_read(&n->total_objects);
4319 else if (flags & SO_OBJECTS)
4320 x = atomic_long_read(&n->total_objects) -
4321 count_partial(n, count_free);
Christoph Lameter205ab992008-04-14 19:11:40 +03004322 else
4323 x = atomic_long_read(&n->nr_slabs);
4324 total += x;
4325 nodes[node] += x;
4326 }
4327
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004328 } else
4329#endif
4330 if (flags & SO_PARTIAL) {
Christoph Lameterfa45dc22014-08-06 16:04:09 -07004331 struct kmem_cache_node *n;
Christoph Lameter205ab992008-04-14 19:11:40 +03004332
Christoph Lameterfa45dc22014-08-06 16:04:09 -07004333 for_each_kmem_cache_node(s, node, n) {
Christoph Lameter205ab992008-04-14 19:11:40 +03004334 if (flags & SO_TOTAL)
4335 x = count_partial(n, count_total);
4336 else if (flags & SO_OBJECTS)
4337 x = count_partial(n, count_inuse);
Christoph Lameter81819f02007-05-06 14:49:36 -07004338 else
4339 x = n->nr_partial;
4340 total += x;
4341 nodes[node] += x;
4342 }
Christoph Lameter81819f02007-05-06 14:49:36 -07004343 }
Christoph Lameter81819f02007-05-06 14:49:36 -07004344 x = sprintf(buf, "%lu", total);
4345#ifdef CONFIG_NUMA
Christoph Lameterfa45dc22014-08-06 16:04:09 -07004346 for (node = 0; node < nr_node_ids; node++)
Christoph Lameter81819f02007-05-06 14:49:36 -07004347 if (nodes[node])
4348 x += sprintf(buf + x, " N%d=%lu",
4349 node, nodes[node]);
4350#endif
Vladimir Davydovbfc8c902014-06-04 16:07:18 -07004351 put_online_mems();
Christoph Lameter81819f02007-05-06 14:49:36 -07004352 kfree(nodes);
4353 return x + sprintf(buf + x, "\n");
4354}
4355
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004356#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter81819f02007-05-06 14:49:36 -07004357static int any_slab_objects(struct kmem_cache *s)
4358{
4359 int node;
Christoph Lameterfa45dc22014-08-06 16:04:09 -07004360 struct kmem_cache_node *n;
Christoph Lameterdfb4f092007-10-16 01:26:05 -07004361
Christoph Lameterfa45dc22014-08-06 16:04:09 -07004362 for_each_kmem_cache_node(s, node, n)
Benjamin Herrenschmidt4ea33e22008-05-06 20:42:39 -07004363 if (atomic_long_read(&n->total_objects))
Christoph Lameter81819f02007-05-06 14:49:36 -07004364 return 1;
Christoph Lameterfa45dc22014-08-06 16:04:09 -07004365
Christoph Lameter81819f02007-05-06 14:49:36 -07004366 return 0;
4367}
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004368#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07004369
4370#define to_slab_attr(n) container_of(n, struct slab_attribute, attr)
Phil Carmody497888c2011-07-14 15:07:13 +03004371#define to_slab(n) container_of(n, struct kmem_cache, kobj)
Christoph Lameter81819f02007-05-06 14:49:36 -07004372
4373struct slab_attribute {
4374 struct attribute attr;
4375 ssize_t (*show)(struct kmem_cache *s, char *buf);
4376 ssize_t (*store)(struct kmem_cache *s, const char *x, size_t count);
4377};
4378
4379#define SLAB_ATTR_RO(_name) \
Vasiliy Kulikovab067e92011-09-27 21:54:53 +04004380 static struct slab_attribute _name##_attr = \
4381 __ATTR(_name, 0400, _name##_show, NULL)
Christoph Lameter81819f02007-05-06 14:49:36 -07004382
4383#define SLAB_ATTR(_name) \
4384 static struct slab_attribute _name##_attr = \
Vasiliy Kulikovab067e92011-09-27 21:54:53 +04004385 __ATTR(_name, 0600, _name##_show, _name##_store)
Christoph Lameter81819f02007-05-06 14:49:36 -07004386
Christoph Lameter81819f02007-05-06 14:49:36 -07004387static ssize_t slab_size_show(struct kmem_cache *s, char *buf)
4388{
4389 return sprintf(buf, "%d\n", s->size);
4390}
4391SLAB_ATTR_RO(slab_size);
4392
4393static ssize_t align_show(struct kmem_cache *s, char *buf)
4394{
4395 return sprintf(buf, "%d\n", s->align);
4396}
4397SLAB_ATTR_RO(align);
4398
4399static ssize_t object_size_show(struct kmem_cache *s, char *buf)
4400{
Christoph Lameter3b0efdf2012-06-13 10:24:57 -05004401 return sprintf(buf, "%d\n", s->object_size);
Christoph Lameter81819f02007-05-06 14:49:36 -07004402}
4403SLAB_ATTR_RO(object_size);
4404
4405static ssize_t objs_per_slab_show(struct kmem_cache *s, char *buf)
4406{
Christoph Lameter834f3d12008-04-14 19:11:31 +03004407 return sprintf(buf, "%d\n", oo_objects(s->oo));
Christoph Lameter81819f02007-05-06 14:49:36 -07004408}
4409SLAB_ATTR_RO(objs_per_slab);
4410
Christoph Lameter06b285d2008-04-14 19:11:41 +03004411static ssize_t order_store(struct kmem_cache *s,
4412 const char *buf, size_t length)
4413{
Christoph Lameter0121c6192008-04-29 16:11:12 -07004414 unsigned long order;
4415 int err;
4416
Jingoo Han3dbb95f2013-09-11 14:20:25 -07004417 err = kstrtoul(buf, 10, &order);
Christoph Lameter0121c6192008-04-29 16:11:12 -07004418 if (err)
4419 return err;
Christoph Lameter06b285d2008-04-14 19:11:41 +03004420
4421 if (order > slub_max_order || order < slub_min_order)
4422 return -EINVAL;
4423
4424 calculate_sizes(s, order);
4425 return length;
4426}
4427
Christoph Lameter81819f02007-05-06 14:49:36 -07004428static ssize_t order_show(struct kmem_cache *s, char *buf)
4429{
Christoph Lameter834f3d12008-04-14 19:11:31 +03004430 return sprintf(buf, "%d\n", oo_order(s->oo));
Christoph Lameter81819f02007-05-06 14:49:36 -07004431}
Christoph Lameter06b285d2008-04-14 19:11:41 +03004432SLAB_ATTR(order);
Christoph Lameter81819f02007-05-06 14:49:36 -07004433
David Rientjes73d342b2009-02-22 17:40:09 -08004434static ssize_t min_partial_show(struct kmem_cache *s, char *buf)
4435{
4436 return sprintf(buf, "%lu\n", s->min_partial);
4437}
4438
4439static ssize_t min_partial_store(struct kmem_cache *s, const char *buf,
4440 size_t length)
4441{
4442 unsigned long min;
4443 int err;
4444
Jingoo Han3dbb95f2013-09-11 14:20:25 -07004445 err = kstrtoul(buf, 10, &min);
David Rientjes73d342b2009-02-22 17:40:09 -08004446 if (err)
4447 return err;
4448
David Rientjesc0bdb232009-02-25 09:16:35 +02004449 set_min_partial(s, min);
David Rientjes73d342b2009-02-22 17:40:09 -08004450 return length;
4451}
4452SLAB_ATTR(min_partial);
4453
Christoph Lameter49e22582011-08-09 16:12:27 -05004454static ssize_t cpu_partial_show(struct kmem_cache *s, char *buf)
4455{
4456 return sprintf(buf, "%u\n", s->cpu_partial);
4457}
4458
4459static ssize_t cpu_partial_store(struct kmem_cache *s, const char *buf,
4460 size_t length)
4461{
4462 unsigned long objects;
4463 int err;
4464
Jingoo Han3dbb95f2013-09-11 14:20:25 -07004465 err = kstrtoul(buf, 10, &objects);
Christoph Lameter49e22582011-08-09 16:12:27 -05004466 if (err)
4467 return err;
Joonsoo Kim345c9052013-06-19 14:05:52 +09004468 if (objects && !kmem_cache_has_cpu_partial(s))
David Rientjes74ee4ef2012-01-09 13:19:45 -08004469 return -EINVAL;
Christoph Lameter49e22582011-08-09 16:12:27 -05004470
4471 s->cpu_partial = objects;
4472 flush_all(s);
4473 return length;
4474}
4475SLAB_ATTR(cpu_partial);
4476
Christoph Lameter81819f02007-05-06 14:49:36 -07004477static ssize_t ctor_show(struct kmem_cache *s, char *buf)
4478{
Joe Perches62c70bc2011-01-13 15:45:52 -08004479 if (!s->ctor)
4480 return 0;
4481 return sprintf(buf, "%pS\n", s->ctor);
Christoph Lameter81819f02007-05-06 14:49:36 -07004482}
4483SLAB_ATTR_RO(ctor);
4484
Christoph Lameter81819f02007-05-06 14:49:36 -07004485static ssize_t aliases_show(struct kmem_cache *s, char *buf)
4486{
Gu Zheng4307c142014-08-06 16:04:51 -07004487 return sprintf(buf, "%d\n", s->refcount < 0 ? 0 : s->refcount - 1);
Christoph Lameter81819f02007-05-06 14:49:36 -07004488}
4489SLAB_ATTR_RO(aliases);
4490
Christoph Lameter81819f02007-05-06 14:49:36 -07004491static ssize_t partial_show(struct kmem_cache *s, char *buf)
4492{
Christoph Lameterd9acf4b2008-02-15 15:22:21 -08004493 return show_slab_objects(s, buf, SO_PARTIAL);
Christoph Lameter81819f02007-05-06 14:49:36 -07004494}
4495SLAB_ATTR_RO(partial);
4496
4497static ssize_t cpu_slabs_show(struct kmem_cache *s, char *buf)
4498{
Christoph Lameterd9acf4b2008-02-15 15:22:21 -08004499 return show_slab_objects(s, buf, SO_CPU);
Christoph Lameter81819f02007-05-06 14:49:36 -07004500}
4501SLAB_ATTR_RO(cpu_slabs);
4502
4503static ssize_t objects_show(struct kmem_cache *s, char *buf)
4504{
Christoph Lameter205ab992008-04-14 19:11:40 +03004505 return show_slab_objects(s, buf, SO_ALL|SO_OBJECTS);
Christoph Lameter81819f02007-05-06 14:49:36 -07004506}
4507SLAB_ATTR_RO(objects);
4508
Christoph Lameter205ab992008-04-14 19:11:40 +03004509static ssize_t objects_partial_show(struct kmem_cache *s, char *buf)
4510{
4511 return show_slab_objects(s, buf, SO_PARTIAL|SO_OBJECTS);
4512}
4513SLAB_ATTR_RO(objects_partial);
4514
Christoph Lameter49e22582011-08-09 16:12:27 -05004515static ssize_t slabs_cpu_partial_show(struct kmem_cache *s, char *buf)
4516{
4517 int objects = 0;
4518 int pages = 0;
4519 int cpu;
4520 int len;
4521
4522 for_each_online_cpu(cpu) {
4523 struct page *page = per_cpu_ptr(s->cpu_slab, cpu)->partial;
4524
4525 if (page) {
4526 pages += page->pages;
4527 objects += page->pobjects;
4528 }
4529 }
4530
4531 len = sprintf(buf, "%d(%d)", objects, pages);
4532
4533#ifdef CONFIG_SMP
4534 for_each_online_cpu(cpu) {
4535 struct page *page = per_cpu_ptr(s->cpu_slab, cpu) ->partial;
4536
4537 if (page && len < PAGE_SIZE - 20)
4538 len += sprintf(buf + len, " C%d=%d(%d)", cpu,
4539 page->pobjects, page->pages);
4540 }
4541#endif
4542 return len + sprintf(buf + len, "\n");
4543}
4544SLAB_ATTR_RO(slabs_cpu_partial);
4545
Christoph Lameter81819f02007-05-06 14:49:36 -07004546static ssize_t reclaim_account_show(struct kmem_cache *s, char *buf)
4547{
4548 return sprintf(buf, "%d\n", !!(s->flags & SLAB_RECLAIM_ACCOUNT));
4549}
4550
4551static ssize_t reclaim_account_store(struct kmem_cache *s,
4552 const char *buf, size_t length)
4553{
4554 s->flags &= ~SLAB_RECLAIM_ACCOUNT;
4555 if (buf[0] == '1')
4556 s->flags |= SLAB_RECLAIM_ACCOUNT;
4557 return length;
4558}
4559SLAB_ATTR(reclaim_account);
4560
4561static ssize_t hwcache_align_show(struct kmem_cache *s, char *buf)
4562{
Christoph Lameter5af60832007-05-06 14:49:56 -07004563 return sprintf(buf, "%d\n", !!(s->flags & SLAB_HWCACHE_ALIGN));
Christoph Lameter81819f02007-05-06 14:49:36 -07004564}
4565SLAB_ATTR_RO(hwcache_align);
4566
4567#ifdef CONFIG_ZONE_DMA
4568static ssize_t cache_dma_show(struct kmem_cache *s, char *buf)
4569{
4570 return sprintf(buf, "%d\n", !!(s->flags & SLAB_CACHE_DMA));
4571}
4572SLAB_ATTR_RO(cache_dma);
4573#endif
4574
4575static ssize_t destroy_by_rcu_show(struct kmem_cache *s, char *buf)
4576{
4577 return sprintf(buf, "%d\n", !!(s->flags & SLAB_DESTROY_BY_RCU));
4578}
4579SLAB_ATTR_RO(destroy_by_rcu);
4580
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08004581static ssize_t reserved_show(struct kmem_cache *s, char *buf)
4582{
4583 return sprintf(buf, "%d\n", s->reserved);
4584}
4585SLAB_ATTR_RO(reserved);
4586
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004587#ifdef CONFIG_SLUB_DEBUG
Christoph Lametera5a84752010-10-05 13:57:27 -05004588static ssize_t slabs_show(struct kmem_cache *s, char *buf)
4589{
4590 return show_slab_objects(s, buf, SO_ALL);
4591}
4592SLAB_ATTR_RO(slabs);
4593
4594static ssize_t total_objects_show(struct kmem_cache *s, char *buf)
4595{
4596 return show_slab_objects(s, buf, SO_ALL|SO_TOTAL);
4597}
4598SLAB_ATTR_RO(total_objects);
4599
4600static ssize_t sanity_checks_show(struct kmem_cache *s, char *buf)
4601{
4602 return sprintf(buf, "%d\n", !!(s->flags & SLAB_DEBUG_FREE));
4603}
4604
4605static ssize_t sanity_checks_store(struct kmem_cache *s,
4606 const char *buf, size_t length)
4607{
4608 s->flags &= ~SLAB_DEBUG_FREE;
Christoph Lameterb789ef52011-06-01 12:25:49 -05004609 if (buf[0] == '1') {
4610 s->flags &= ~__CMPXCHG_DOUBLE;
Christoph Lametera5a84752010-10-05 13:57:27 -05004611 s->flags |= SLAB_DEBUG_FREE;
Christoph Lameterb789ef52011-06-01 12:25:49 -05004612 }
Christoph Lametera5a84752010-10-05 13:57:27 -05004613 return length;
4614}
4615SLAB_ATTR(sanity_checks);
4616
4617static ssize_t trace_show(struct kmem_cache *s, char *buf)
4618{
4619 return sprintf(buf, "%d\n", !!(s->flags & SLAB_TRACE));
4620}
4621
4622static ssize_t trace_store(struct kmem_cache *s, const char *buf,
4623 size_t length)
4624{
Christoph Lameterc9e16132014-10-09 15:26:11 -07004625 /*
4626 * Tracing a merged cache is going to give confusing results
4627 * as well as cause other issues like converting a mergeable
4628 * cache into an umergeable one.
4629 */
4630 if (s->refcount > 1)
4631 return -EINVAL;
4632
Christoph Lametera5a84752010-10-05 13:57:27 -05004633 s->flags &= ~SLAB_TRACE;
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_TRACE;
Christoph Lameterb789ef52011-06-01 12:25:49 -05004637 }
Christoph Lametera5a84752010-10-05 13:57:27 -05004638 return length;
4639}
4640SLAB_ATTR(trace);
4641
Christoph Lameter81819f02007-05-06 14:49:36 -07004642static ssize_t red_zone_show(struct kmem_cache *s, char *buf)
4643{
4644 return sprintf(buf, "%d\n", !!(s->flags & SLAB_RED_ZONE));
4645}
4646
4647static ssize_t red_zone_store(struct kmem_cache *s,
4648 const char *buf, size_t length)
4649{
4650 if (any_slab_objects(s))
4651 return -EBUSY;
4652
4653 s->flags &= ~SLAB_RED_ZONE;
Christoph Lameterb789ef52011-06-01 12:25:49 -05004654 if (buf[0] == '1') {
4655 s->flags &= ~__CMPXCHG_DOUBLE;
Christoph Lameter81819f02007-05-06 14:49:36 -07004656 s->flags |= SLAB_RED_ZONE;
Christoph Lameterb789ef52011-06-01 12:25:49 -05004657 }
Christoph Lameter06b285d2008-04-14 19:11:41 +03004658 calculate_sizes(s, -1);
Christoph Lameter81819f02007-05-06 14:49:36 -07004659 return length;
4660}
4661SLAB_ATTR(red_zone);
4662
4663static ssize_t poison_show(struct kmem_cache *s, char *buf)
4664{
4665 return sprintf(buf, "%d\n", !!(s->flags & SLAB_POISON));
4666}
4667
4668static ssize_t poison_store(struct kmem_cache *s,
4669 const char *buf, size_t length)
4670{
4671 if (any_slab_objects(s))
4672 return -EBUSY;
4673
4674 s->flags &= ~SLAB_POISON;
Christoph Lameterb789ef52011-06-01 12:25:49 -05004675 if (buf[0] == '1') {
4676 s->flags &= ~__CMPXCHG_DOUBLE;
Christoph Lameter81819f02007-05-06 14:49:36 -07004677 s->flags |= SLAB_POISON;
Christoph Lameterb789ef52011-06-01 12:25:49 -05004678 }
Christoph Lameter06b285d2008-04-14 19:11:41 +03004679 calculate_sizes(s, -1);
Christoph Lameter81819f02007-05-06 14:49:36 -07004680 return length;
4681}
4682SLAB_ATTR(poison);
4683
4684static ssize_t store_user_show(struct kmem_cache *s, char *buf)
4685{
4686 return sprintf(buf, "%d\n", !!(s->flags & SLAB_STORE_USER));
4687}
4688
4689static ssize_t store_user_store(struct kmem_cache *s,
4690 const char *buf, size_t length)
4691{
4692 if (any_slab_objects(s))
4693 return -EBUSY;
4694
4695 s->flags &= ~SLAB_STORE_USER;
Christoph Lameterb789ef52011-06-01 12:25:49 -05004696 if (buf[0] == '1') {
4697 s->flags &= ~__CMPXCHG_DOUBLE;
Christoph Lameter81819f02007-05-06 14:49:36 -07004698 s->flags |= SLAB_STORE_USER;
Christoph Lameterb789ef52011-06-01 12:25:49 -05004699 }
Christoph Lameter06b285d2008-04-14 19:11:41 +03004700 calculate_sizes(s, -1);
Christoph Lameter81819f02007-05-06 14:49:36 -07004701 return length;
4702}
4703SLAB_ATTR(store_user);
4704
Christoph Lameter53e15af2007-05-06 14:49:43 -07004705static ssize_t validate_show(struct kmem_cache *s, char *buf)
4706{
4707 return 0;
4708}
4709
4710static ssize_t validate_store(struct kmem_cache *s,
4711 const char *buf, size_t length)
4712{
Christoph Lameter434e2452007-07-17 04:03:30 -07004713 int ret = -EINVAL;
4714
4715 if (buf[0] == '1') {
4716 ret = validate_slab_cache(s);
4717 if (ret >= 0)
4718 ret = length;
4719 }
4720 return ret;
Christoph Lameter53e15af2007-05-06 14:49:43 -07004721}
4722SLAB_ATTR(validate);
Christoph Lametera5a84752010-10-05 13:57:27 -05004723
4724static ssize_t alloc_calls_show(struct kmem_cache *s, char *buf)
4725{
4726 if (!(s->flags & SLAB_STORE_USER))
4727 return -ENOSYS;
4728 return list_locations(s, buf, TRACK_ALLOC);
4729}
4730SLAB_ATTR_RO(alloc_calls);
4731
4732static ssize_t free_calls_show(struct kmem_cache *s, char *buf)
4733{
4734 if (!(s->flags & SLAB_STORE_USER))
4735 return -ENOSYS;
4736 return list_locations(s, buf, TRACK_FREE);
4737}
4738SLAB_ATTR_RO(free_calls);
4739#endif /* CONFIG_SLUB_DEBUG */
4740
4741#ifdef CONFIG_FAILSLAB
4742static ssize_t failslab_show(struct kmem_cache *s, char *buf)
4743{
4744 return sprintf(buf, "%d\n", !!(s->flags & SLAB_FAILSLAB));
4745}
4746
4747static ssize_t failslab_store(struct kmem_cache *s, const char *buf,
4748 size_t length)
4749{
Christoph Lameterc9e16132014-10-09 15:26:11 -07004750 if (s->refcount > 1)
4751 return -EINVAL;
4752
Christoph Lametera5a84752010-10-05 13:57:27 -05004753 s->flags &= ~SLAB_FAILSLAB;
4754 if (buf[0] == '1')
4755 s->flags |= SLAB_FAILSLAB;
4756 return length;
4757}
4758SLAB_ATTR(failslab);
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004759#endif
Christoph Lameter53e15af2007-05-06 14:49:43 -07004760
Christoph Lameter2086d262007-05-06 14:49:46 -07004761static ssize_t shrink_show(struct kmem_cache *s, char *buf)
4762{
4763 return 0;
4764}
4765
4766static ssize_t shrink_store(struct kmem_cache *s,
4767 const char *buf, size_t length)
4768{
Vladimir Davydov832f37f2015-02-12 14:59:41 -08004769 if (buf[0] == '1')
4770 kmem_cache_shrink(s);
4771 else
Christoph Lameter2086d262007-05-06 14:49:46 -07004772 return -EINVAL;
4773 return length;
4774}
4775SLAB_ATTR(shrink);
4776
Christoph Lameter81819f02007-05-06 14:49:36 -07004777#ifdef CONFIG_NUMA
Christoph Lameter98246012008-01-07 23:20:26 -08004778static ssize_t remote_node_defrag_ratio_show(struct kmem_cache *s, char *buf)
Christoph Lameter81819f02007-05-06 14:49:36 -07004779{
Christoph Lameter98246012008-01-07 23:20:26 -08004780 return sprintf(buf, "%d\n", s->remote_node_defrag_ratio / 10);
Christoph Lameter81819f02007-05-06 14:49:36 -07004781}
4782
Christoph Lameter98246012008-01-07 23:20:26 -08004783static ssize_t remote_node_defrag_ratio_store(struct kmem_cache *s,
Christoph Lameter81819f02007-05-06 14:49:36 -07004784 const char *buf, size_t length)
4785{
Christoph Lameter0121c6192008-04-29 16:11:12 -07004786 unsigned long ratio;
4787 int err;
Christoph Lameter81819f02007-05-06 14:49:36 -07004788
Jingoo Han3dbb95f2013-09-11 14:20:25 -07004789 err = kstrtoul(buf, 10, &ratio);
Christoph Lameter0121c6192008-04-29 16:11:12 -07004790 if (err)
4791 return err;
4792
Christoph Lametere2cb96b2008-08-19 08:51:22 -05004793 if (ratio <= 100)
Christoph Lameter0121c6192008-04-29 16:11:12 -07004794 s->remote_node_defrag_ratio = ratio * 10;
4795
Christoph Lameter81819f02007-05-06 14:49:36 -07004796 return length;
4797}
Christoph Lameter98246012008-01-07 23:20:26 -08004798SLAB_ATTR(remote_node_defrag_ratio);
Christoph Lameter81819f02007-05-06 14:49:36 -07004799#endif
4800
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004801#ifdef CONFIG_SLUB_STATS
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004802static int show_stat(struct kmem_cache *s, char *buf, enum stat_item si)
4803{
4804 unsigned long sum = 0;
4805 int cpu;
4806 int len;
4807 int *data = kmalloc(nr_cpu_ids * sizeof(int), GFP_KERNEL);
4808
4809 if (!data)
4810 return -ENOMEM;
4811
4812 for_each_online_cpu(cpu) {
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06004813 unsigned x = per_cpu_ptr(s->cpu_slab, cpu)->stat[si];
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004814
4815 data[cpu] = x;
4816 sum += x;
4817 }
4818
4819 len = sprintf(buf, "%lu", sum);
4820
Christoph Lameter50ef37b2008-04-14 18:52:05 +03004821#ifdef CONFIG_SMP
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004822 for_each_online_cpu(cpu) {
4823 if (data[cpu] && len < PAGE_SIZE - 20)
Christoph Lameter50ef37b2008-04-14 18:52:05 +03004824 len += sprintf(buf + len, " C%d=%u", cpu, data[cpu]);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004825 }
Christoph Lameter50ef37b2008-04-14 18:52:05 +03004826#endif
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004827 kfree(data);
4828 return len + sprintf(buf + len, "\n");
4829}
4830
David Rientjes78eb00c2009-10-15 02:20:22 -07004831static void clear_stat(struct kmem_cache *s, enum stat_item si)
4832{
4833 int cpu;
4834
4835 for_each_online_cpu(cpu)
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06004836 per_cpu_ptr(s->cpu_slab, cpu)->stat[si] = 0;
David Rientjes78eb00c2009-10-15 02:20:22 -07004837}
4838
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004839#define STAT_ATTR(si, text) \
4840static ssize_t text##_show(struct kmem_cache *s, char *buf) \
4841{ \
4842 return show_stat(s, buf, si); \
4843} \
David Rientjes78eb00c2009-10-15 02:20:22 -07004844static ssize_t text##_store(struct kmem_cache *s, \
4845 const char *buf, size_t length) \
4846{ \
4847 if (buf[0] != '0') \
4848 return -EINVAL; \
4849 clear_stat(s, si); \
4850 return length; \
4851} \
4852SLAB_ATTR(text); \
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004853
4854STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath);
4855STAT_ATTR(ALLOC_SLOWPATH, alloc_slowpath);
4856STAT_ATTR(FREE_FASTPATH, free_fastpath);
4857STAT_ATTR(FREE_SLOWPATH, free_slowpath);
4858STAT_ATTR(FREE_FROZEN, free_frozen);
4859STAT_ATTR(FREE_ADD_PARTIAL, free_add_partial);
4860STAT_ATTR(FREE_REMOVE_PARTIAL, free_remove_partial);
4861STAT_ATTR(ALLOC_FROM_PARTIAL, alloc_from_partial);
4862STAT_ATTR(ALLOC_SLAB, alloc_slab);
4863STAT_ATTR(ALLOC_REFILL, alloc_refill);
Christoph Lametere36a2652011-06-01 12:25:57 -05004864STAT_ATTR(ALLOC_NODE_MISMATCH, alloc_node_mismatch);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004865STAT_ATTR(FREE_SLAB, free_slab);
4866STAT_ATTR(CPUSLAB_FLUSH, cpuslab_flush);
4867STAT_ATTR(DEACTIVATE_FULL, deactivate_full);
4868STAT_ATTR(DEACTIVATE_EMPTY, deactivate_empty);
4869STAT_ATTR(DEACTIVATE_TO_HEAD, deactivate_to_head);
4870STAT_ATTR(DEACTIVATE_TO_TAIL, deactivate_to_tail);
4871STAT_ATTR(DEACTIVATE_REMOTE_FREES, deactivate_remote_frees);
Christoph Lameter03e404a2011-06-01 12:25:58 -05004872STAT_ATTR(DEACTIVATE_BYPASS, deactivate_bypass);
Christoph Lameter65c33762008-04-14 19:11:40 +03004873STAT_ATTR(ORDER_FALLBACK, order_fallback);
Christoph Lameterb789ef52011-06-01 12:25:49 -05004874STAT_ATTR(CMPXCHG_DOUBLE_CPU_FAIL, cmpxchg_double_cpu_fail);
4875STAT_ATTR(CMPXCHG_DOUBLE_FAIL, cmpxchg_double_fail);
Christoph Lameter49e22582011-08-09 16:12:27 -05004876STAT_ATTR(CPU_PARTIAL_ALLOC, cpu_partial_alloc);
4877STAT_ATTR(CPU_PARTIAL_FREE, cpu_partial_free);
Alex Shi8028dce2012-02-03 23:34:56 +08004878STAT_ATTR(CPU_PARTIAL_NODE, cpu_partial_node);
4879STAT_ATTR(CPU_PARTIAL_DRAIN, cpu_partial_drain);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004880#endif
4881
Pekka Enberg06428782008-01-07 23:20:27 -08004882static struct attribute *slab_attrs[] = {
Christoph Lameter81819f02007-05-06 14:49:36 -07004883 &slab_size_attr.attr,
4884 &object_size_attr.attr,
4885 &objs_per_slab_attr.attr,
4886 &order_attr.attr,
David Rientjes73d342b2009-02-22 17:40:09 -08004887 &min_partial_attr.attr,
Christoph Lameter49e22582011-08-09 16:12:27 -05004888 &cpu_partial_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07004889 &objects_attr.attr,
Christoph Lameter205ab992008-04-14 19:11:40 +03004890 &objects_partial_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07004891 &partial_attr.attr,
4892 &cpu_slabs_attr.attr,
4893 &ctor_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07004894 &aliases_attr.attr,
4895 &align_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07004896 &hwcache_align_attr.attr,
4897 &reclaim_account_attr.attr,
4898 &destroy_by_rcu_attr.attr,
Christoph Lametera5a84752010-10-05 13:57:27 -05004899 &shrink_attr.attr,
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08004900 &reserved_attr.attr,
Christoph Lameter49e22582011-08-09 16:12:27 -05004901 &slabs_cpu_partial_attr.attr,
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004902#ifdef CONFIG_SLUB_DEBUG
Christoph Lametera5a84752010-10-05 13:57:27 -05004903 &total_objects_attr.attr,
4904 &slabs_attr.attr,
4905 &sanity_checks_attr.attr,
4906 &trace_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07004907 &red_zone_attr.attr,
4908 &poison_attr.attr,
4909 &store_user_attr.attr,
Christoph Lameter53e15af2007-05-06 14:49:43 -07004910 &validate_attr.attr,
Christoph Lameter88a420e2007-05-06 14:49:45 -07004911 &alloc_calls_attr.attr,
4912 &free_calls_attr.attr,
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004913#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07004914#ifdef CONFIG_ZONE_DMA
4915 &cache_dma_attr.attr,
4916#endif
4917#ifdef CONFIG_NUMA
Christoph Lameter98246012008-01-07 23:20:26 -08004918 &remote_node_defrag_ratio_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07004919#endif
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004920#ifdef CONFIG_SLUB_STATS
4921 &alloc_fastpath_attr.attr,
4922 &alloc_slowpath_attr.attr,
4923 &free_fastpath_attr.attr,
4924 &free_slowpath_attr.attr,
4925 &free_frozen_attr.attr,
4926 &free_add_partial_attr.attr,
4927 &free_remove_partial_attr.attr,
4928 &alloc_from_partial_attr.attr,
4929 &alloc_slab_attr.attr,
4930 &alloc_refill_attr.attr,
Christoph Lametere36a2652011-06-01 12:25:57 -05004931 &alloc_node_mismatch_attr.attr,
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004932 &free_slab_attr.attr,
4933 &cpuslab_flush_attr.attr,
4934 &deactivate_full_attr.attr,
4935 &deactivate_empty_attr.attr,
4936 &deactivate_to_head_attr.attr,
4937 &deactivate_to_tail_attr.attr,
4938 &deactivate_remote_frees_attr.attr,
Christoph Lameter03e404a2011-06-01 12:25:58 -05004939 &deactivate_bypass_attr.attr,
Christoph Lameter65c33762008-04-14 19:11:40 +03004940 &order_fallback_attr.attr,
Christoph Lameterb789ef52011-06-01 12:25:49 -05004941 &cmpxchg_double_fail_attr.attr,
4942 &cmpxchg_double_cpu_fail_attr.attr,
Christoph Lameter49e22582011-08-09 16:12:27 -05004943 &cpu_partial_alloc_attr.attr,
4944 &cpu_partial_free_attr.attr,
Alex Shi8028dce2012-02-03 23:34:56 +08004945 &cpu_partial_node_attr.attr,
4946 &cpu_partial_drain_attr.attr,
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004947#endif
Dmitry Monakhov4c13dd32010-02-26 09:36:12 +03004948#ifdef CONFIG_FAILSLAB
4949 &failslab_attr.attr,
4950#endif
4951
Christoph Lameter81819f02007-05-06 14:49:36 -07004952 NULL
4953};
4954
4955static struct attribute_group slab_attr_group = {
4956 .attrs = slab_attrs,
4957};
4958
4959static ssize_t slab_attr_show(struct kobject *kobj,
4960 struct attribute *attr,
4961 char *buf)
4962{
4963 struct slab_attribute *attribute;
4964 struct kmem_cache *s;
4965 int err;
4966
4967 attribute = to_slab_attr(attr);
4968 s = to_slab(kobj);
4969
4970 if (!attribute->show)
4971 return -EIO;
4972
4973 err = attribute->show(s, buf);
4974
4975 return err;
4976}
4977
4978static ssize_t slab_attr_store(struct kobject *kobj,
4979 struct attribute *attr,
4980 const char *buf, size_t len)
4981{
4982 struct slab_attribute *attribute;
4983 struct kmem_cache *s;
4984 int err;
4985
4986 attribute = to_slab_attr(attr);
4987 s = to_slab(kobj);
4988
4989 if (!attribute->store)
4990 return -EIO;
4991
4992 err = attribute->store(s, buf, len);
Glauber Costa107dab52012-12-18 14:23:05 -08004993#ifdef CONFIG_MEMCG_KMEM
4994 if (slab_state >= FULL && err >= 0 && is_root_cache(s)) {
Vladimir Davydov426589f2015-02-12 14:59:23 -08004995 struct kmem_cache *c;
Christoph Lameter81819f02007-05-06 14:49:36 -07004996
Glauber Costa107dab52012-12-18 14:23:05 -08004997 mutex_lock(&slab_mutex);
4998 if (s->max_attr_size < len)
4999 s->max_attr_size = len;
5000
Glauber Costaebe945c2012-12-18 14:23:10 -08005001 /*
5002 * This is a best effort propagation, so this function's return
5003 * value will be determined by the parent cache only. This is
5004 * basically because not all attributes will have a well
5005 * defined semantics for rollbacks - most of the actions will
5006 * have permanent effects.
5007 *
5008 * Returning the error value of any of the children that fail
5009 * is not 100 % defined, in the sense that users seeing the
5010 * error code won't be able to know anything about the state of
5011 * the cache.
5012 *
5013 * Only returning the error code for the parent cache at least
5014 * has well defined semantics. The cache being written to
5015 * directly either failed or succeeded, in which case we loop
5016 * through the descendants with best-effort propagation.
5017 */
Vladimir Davydov426589f2015-02-12 14:59:23 -08005018 for_each_memcg_cache(c, s)
5019 attribute->store(c, buf, len);
Glauber Costa107dab52012-12-18 14:23:05 -08005020 mutex_unlock(&slab_mutex);
5021 }
5022#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07005023 return err;
5024}
5025
Glauber Costa107dab52012-12-18 14:23:05 -08005026static void memcg_propagate_slab_attrs(struct kmem_cache *s)
5027{
5028#ifdef CONFIG_MEMCG_KMEM
5029 int i;
5030 char *buffer = NULL;
Vladimir Davydov93030d82014-05-06 12:49:59 -07005031 struct kmem_cache *root_cache;
Glauber Costa107dab52012-12-18 14:23:05 -08005032
Vladimir Davydov93030d82014-05-06 12:49:59 -07005033 if (is_root_cache(s))
Glauber Costa107dab52012-12-18 14:23:05 -08005034 return;
5035
Vladimir Davydovf7ce3192015-02-12 14:59:20 -08005036 root_cache = s->memcg_params.root_cache;
Vladimir Davydov93030d82014-05-06 12:49:59 -07005037
Glauber Costa107dab52012-12-18 14:23:05 -08005038 /*
5039 * This mean this cache had no attribute written. Therefore, no point
5040 * in copying default values around
5041 */
Vladimir Davydov93030d82014-05-06 12:49:59 -07005042 if (!root_cache->max_attr_size)
Glauber Costa107dab52012-12-18 14:23:05 -08005043 return;
5044
5045 for (i = 0; i < ARRAY_SIZE(slab_attrs); i++) {
5046 char mbuf[64];
5047 char *buf;
5048 struct slab_attribute *attr = to_slab_attr(slab_attrs[i]);
5049
5050 if (!attr || !attr->store || !attr->show)
5051 continue;
5052
5053 /*
5054 * It is really bad that we have to allocate here, so we will
5055 * do it only as a fallback. If we actually allocate, though,
5056 * we can just use the allocated buffer until the end.
5057 *
5058 * Most of the slub attributes will tend to be very small in
5059 * size, but sysfs allows buffers up to a page, so they can
5060 * theoretically happen.
5061 */
5062 if (buffer)
5063 buf = buffer;
Vladimir Davydov93030d82014-05-06 12:49:59 -07005064 else if (root_cache->max_attr_size < ARRAY_SIZE(mbuf))
Glauber Costa107dab52012-12-18 14:23:05 -08005065 buf = mbuf;
5066 else {
5067 buffer = (char *) get_zeroed_page(GFP_KERNEL);
5068 if (WARN_ON(!buffer))
5069 continue;
5070 buf = buffer;
5071 }
5072
Vladimir Davydov93030d82014-05-06 12:49:59 -07005073 attr->show(root_cache, buf);
Glauber Costa107dab52012-12-18 14:23:05 -08005074 attr->store(s, buf, strlen(buf));
5075 }
5076
5077 if (buffer)
5078 free_page((unsigned long)buffer);
5079#endif
5080}
5081
Christoph Lameter41a21282014-05-06 12:50:08 -07005082static void kmem_cache_release(struct kobject *k)
5083{
5084 slab_kmem_cache_release(to_slab(k));
5085}
5086
Emese Revfy52cf25d2010-01-19 02:58:23 +01005087static const struct sysfs_ops slab_sysfs_ops = {
Christoph Lameter81819f02007-05-06 14:49:36 -07005088 .show = slab_attr_show,
5089 .store = slab_attr_store,
5090};
5091
5092static struct kobj_type slab_ktype = {
5093 .sysfs_ops = &slab_sysfs_ops,
Christoph Lameter41a21282014-05-06 12:50:08 -07005094 .release = kmem_cache_release,
Christoph Lameter81819f02007-05-06 14:49:36 -07005095};
5096
5097static int uevent_filter(struct kset *kset, struct kobject *kobj)
5098{
5099 struct kobj_type *ktype = get_ktype(kobj);
5100
5101 if (ktype == &slab_ktype)
5102 return 1;
5103 return 0;
5104}
5105
Emese Revfy9cd43612009-12-31 14:52:51 +01005106static const struct kset_uevent_ops slab_uevent_ops = {
Christoph Lameter81819f02007-05-06 14:49:36 -07005107 .filter = uevent_filter,
5108};
5109
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06005110static struct kset *slab_kset;
Christoph Lameter81819f02007-05-06 14:49:36 -07005111
Vladimir Davydov9a417072014-04-07 15:39:31 -07005112static inline struct kset *cache_kset(struct kmem_cache *s)
5113{
5114#ifdef CONFIG_MEMCG_KMEM
5115 if (!is_root_cache(s))
Vladimir Davydovf7ce3192015-02-12 14:59:20 -08005116 return s->memcg_params.root_cache->memcg_kset;
Vladimir Davydov9a417072014-04-07 15:39:31 -07005117#endif
5118 return slab_kset;
5119}
5120
Christoph Lameter81819f02007-05-06 14:49:36 -07005121#define ID_STR_LENGTH 64
5122
5123/* Create a unique string id for a slab cache:
Christoph Lameter6446faa2008-02-15 23:45:26 -08005124 *
5125 * Format :[flags-]size
Christoph Lameter81819f02007-05-06 14:49:36 -07005126 */
5127static char *create_unique_id(struct kmem_cache *s)
5128{
5129 char *name = kmalloc(ID_STR_LENGTH, GFP_KERNEL);
5130 char *p = name;
5131
5132 BUG_ON(!name);
5133
5134 *p++ = ':';
5135 /*
5136 * First flags affecting slabcache operations. We will only
5137 * get here for aliasable slabs so we do not need to support
5138 * too many flags. The flags here must cover all flags that
5139 * are matched during merging to guarantee that the id is
5140 * unique.
5141 */
5142 if (s->flags & SLAB_CACHE_DMA)
5143 *p++ = 'd';
5144 if (s->flags & SLAB_RECLAIM_ACCOUNT)
5145 *p++ = 'a';
5146 if (s->flags & SLAB_DEBUG_FREE)
5147 *p++ = 'F';
Vegard Nossum5a896d92008-04-04 00:54:48 +02005148 if (!(s->flags & SLAB_NOTRACK))
5149 *p++ = 't';
Christoph Lameter81819f02007-05-06 14:49:36 -07005150 if (p != name + 1)
5151 *p++ = '-';
5152 p += sprintf(p, "%07d", s->size);
Glauber Costa2633d7a2012-12-18 14:22:34 -08005153
Christoph Lameter81819f02007-05-06 14:49:36 -07005154 BUG_ON(p > name + ID_STR_LENGTH - 1);
5155 return name;
5156}
5157
5158static int sysfs_slab_add(struct kmem_cache *s)
5159{
5160 int err;
5161 const char *name;
Christoph Lameter45530c42012-11-28 16:23:07 +00005162 int unmergeable = slab_unmergeable(s);
Christoph Lameter81819f02007-05-06 14:49:36 -07005163
Christoph Lameter81819f02007-05-06 14:49:36 -07005164 if (unmergeable) {
5165 /*
5166 * Slabcache can never be merged so we can use the name proper.
5167 * This is typically the case for debug situations. In that
5168 * case we can catch duplicate names easily.
5169 */
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06005170 sysfs_remove_link(&slab_kset->kobj, s->name);
Christoph Lameter81819f02007-05-06 14:49:36 -07005171 name = s->name;
5172 } else {
5173 /*
5174 * Create a unique name for the slab as a target
5175 * for the symlinks.
5176 */
5177 name = create_unique_id(s);
5178 }
5179
Vladimir Davydov9a417072014-04-07 15:39:31 -07005180 s->kobj.kset = cache_kset(s);
Tetsuo Handa26e4f202014-01-04 16:32:31 +09005181 err = kobject_init_and_add(&s->kobj, &slab_ktype, NULL, "%s", name);
Dave Jones54b6a732014-04-07 15:39:32 -07005182 if (err)
5183 goto out_put_kobj;
Christoph Lameter81819f02007-05-06 14:49:36 -07005184
5185 err = sysfs_create_group(&s->kobj, &slab_attr_group);
Dave Jones54b6a732014-04-07 15:39:32 -07005186 if (err)
5187 goto out_del_kobj;
Vladimir Davydov9a417072014-04-07 15:39:31 -07005188
5189#ifdef CONFIG_MEMCG_KMEM
5190 if (is_root_cache(s)) {
5191 s->memcg_kset = kset_create_and_add("cgroup", NULL, &s->kobj);
5192 if (!s->memcg_kset) {
Dave Jones54b6a732014-04-07 15:39:32 -07005193 err = -ENOMEM;
5194 goto out_del_kobj;
Vladimir Davydov9a417072014-04-07 15:39:31 -07005195 }
5196 }
5197#endif
5198
Christoph Lameter81819f02007-05-06 14:49:36 -07005199 kobject_uevent(&s->kobj, KOBJ_ADD);
5200 if (!unmergeable) {
5201 /* Setup first alias */
5202 sysfs_slab_alias(s, s->name);
Christoph Lameter81819f02007-05-06 14:49:36 -07005203 }
Dave Jones54b6a732014-04-07 15:39:32 -07005204out:
5205 if (!unmergeable)
5206 kfree(name);
5207 return err;
5208out_del_kobj:
5209 kobject_del(&s->kobj);
5210out_put_kobj:
5211 kobject_put(&s->kobj);
5212 goto out;
Christoph Lameter81819f02007-05-06 14:49:36 -07005213}
5214
Christoph Lameter41a21282014-05-06 12:50:08 -07005215void sysfs_slab_remove(struct kmem_cache *s)
Christoph Lameter81819f02007-05-06 14:49:36 -07005216{
Christoph Lameter97d06602012-07-06 15:25:11 -05005217 if (slab_state < FULL)
Christoph Lameter2bce6482010-07-19 11:39:11 -05005218 /*
5219 * Sysfs has not been setup yet so no need to remove the
5220 * cache from sysfs.
5221 */
5222 return;
5223
Vladimir Davydov9a417072014-04-07 15:39:31 -07005224#ifdef CONFIG_MEMCG_KMEM
5225 kset_unregister(s->memcg_kset);
5226#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07005227 kobject_uevent(&s->kobj, KOBJ_REMOVE);
5228 kobject_del(&s->kobj);
Christoph Lameter151c6022008-01-07 22:29:05 -08005229 kobject_put(&s->kobj);
Christoph Lameter81819f02007-05-06 14:49:36 -07005230}
5231
5232/*
5233 * Need to buffer aliases during bootup until sysfs becomes
Nick Andrew9f6c708e2008-12-05 14:08:08 +11005234 * available lest we lose that information.
Christoph Lameter81819f02007-05-06 14:49:36 -07005235 */
5236struct saved_alias {
5237 struct kmem_cache *s;
5238 const char *name;
5239 struct saved_alias *next;
5240};
5241
Adrian Bunk5af328a2007-07-17 04:03:27 -07005242static struct saved_alias *alias_list;
Christoph Lameter81819f02007-05-06 14:49:36 -07005243
5244static int sysfs_slab_alias(struct kmem_cache *s, const char *name)
5245{
5246 struct saved_alias *al;
5247
Christoph Lameter97d06602012-07-06 15:25:11 -05005248 if (slab_state == FULL) {
Christoph Lameter81819f02007-05-06 14:49:36 -07005249 /*
5250 * If we have a leftover link then remove it.
5251 */
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06005252 sysfs_remove_link(&slab_kset->kobj, name);
5253 return sysfs_create_link(&slab_kset->kobj, &s->kobj, name);
Christoph Lameter81819f02007-05-06 14:49:36 -07005254 }
5255
5256 al = kmalloc(sizeof(struct saved_alias), GFP_KERNEL);
5257 if (!al)
5258 return -ENOMEM;
5259
5260 al->s = s;
5261 al->name = name;
5262 al->next = alias_list;
5263 alias_list = al;
5264 return 0;
5265}
5266
5267static int __init slab_sysfs_init(void)
5268{
Christoph Lameter5b95a4ac2007-07-17 04:03:19 -07005269 struct kmem_cache *s;
Christoph Lameter81819f02007-05-06 14:49:36 -07005270 int err;
5271
Christoph Lameter18004c52012-07-06 15:25:12 -05005272 mutex_lock(&slab_mutex);
Christoph Lameter2bce6482010-07-19 11:39:11 -05005273
Greg Kroah-Hartman0ff21e42007-11-06 10:36:58 -08005274 slab_kset = kset_create_and_add("slab", &slab_uevent_ops, kernel_kobj);
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06005275 if (!slab_kset) {
Christoph Lameter18004c52012-07-06 15:25:12 -05005276 mutex_unlock(&slab_mutex);
Fabian Frederickf9f58282014-06-04 16:06:34 -07005277 pr_err("Cannot register slab subsystem.\n");
Christoph Lameter81819f02007-05-06 14:49:36 -07005278 return -ENOSYS;
5279 }
5280
Christoph Lameter97d06602012-07-06 15:25:11 -05005281 slab_state = FULL;
Christoph Lameter26a7bd02007-05-09 02:32:39 -07005282
Christoph Lameter5b95a4ac2007-07-17 04:03:19 -07005283 list_for_each_entry(s, &slab_caches, list) {
Christoph Lameter26a7bd02007-05-09 02:32:39 -07005284 err = sysfs_slab_add(s);
Christoph Lameter5d540fb2007-08-30 23:56:26 -07005285 if (err)
Fabian Frederickf9f58282014-06-04 16:06:34 -07005286 pr_err("SLUB: Unable to add boot slab %s to sysfs\n",
5287 s->name);
Christoph Lameter26a7bd02007-05-09 02:32:39 -07005288 }
Christoph Lameter81819f02007-05-06 14:49:36 -07005289
5290 while (alias_list) {
5291 struct saved_alias *al = alias_list;
5292
5293 alias_list = alias_list->next;
5294 err = sysfs_slab_alias(al->s, al->name);
Christoph Lameter5d540fb2007-08-30 23:56:26 -07005295 if (err)
Fabian Frederickf9f58282014-06-04 16:06:34 -07005296 pr_err("SLUB: Unable to add boot slab alias %s to sysfs\n",
5297 al->name);
Christoph Lameter81819f02007-05-06 14:49:36 -07005298 kfree(al);
5299 }
5300
Christoph Lameter18004c52012-07-06 15:25:12 -05005301 mutex_unlock(&slab_mutex);
Christoph Lameter81819f02007-05-06 14:49:36 -07005302 resiliency_test();
5303 return 0;
5304}
5305
5306__initcall(slab_sysfs_init);
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05005307#endif /* CONFIG_SYSFS */
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01005308
5309/*
5310 * The /proc/slabinfo ABI
5311 */
Linus Torvalds158a9622008-01-02 13:04:48 -08005312#ifdef CONFIG_SLABINFO
Glauber Costa0d7561c2012-10-19 18:20:27 +04005313void get_slabinfo(struct kmem_cache *s, struct slabinfo *sinfo)
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01005314{
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01005315 unsigned long nr_slabs = 0;
Christoph Lameter205ab992008-04-14 19:11:40 +03005316 unsigned long nr_objs = 0;
5317 unsigned long nr_free = 0;
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01005318 int node;
Christoph Lameterfa45dc22014-08-06 16:04:09 -07005319 struct kmem_cache_node *n;
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01005320
Christoph Lameterfa45dc22014-08-06 16:04:09 -07005321 for_each_kmem_cache_node(s, node, n) {
Wanpeng Lic17fd132013-07-04 08:33:26 +08005322 nr_slabs += node_nr_slabs(n);
5323 nr_objs += node_nr_objs(n);
Christoph Lameter205ab992008-04-14 19:11:40 +03005324 nr_free += count_partial(n, count_free);
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01005325 }
5326
Glauber Costa0d7561c2012-10-19 18:20:27 +04005327 sinfo->active_objs = nr_objs - nr_free;
5328 sinfo->num_objs = nr_objs;
5329 sinfo->active_slabs = nr_slabs;
5330 sinfo->num_slabs = nr_slabs;
5331 sinfo->objects_per_slab = oo_objects(s->oo);
5332 sinfo->cache_order = oo_order(s->oo);
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01005333}
5334
Glauber Costa0d7561c2012-10-19 18:20:27 +04005335void slabinfo_show_stats(struct seq_file *m, struct kmem_cache *s)
Alexey Dobriyan7b3c3a52008-10-06 02:42:17 +04005336{
Alexey Dobriyan7b3c3a52008-10-06 02:42:17 +04005337}
5338
Glauber Costab7454ad2012-10-19 18:20:25 +04005339ssize_t slabinfo_write(struct file *file, const char __user *buffer,
5340 size_t count, loff_t *ppos)
Alexey Dobriyan7b3c3a52008-10-06 02:42:17 +04005341{
Glauber Costab7454ad2012-10-19 18:20:25 +04005342 return -EIO;
Alexey Dobriyan7b3c3a52008-10-06 02:42:17 +04005343}
Linus Torvalds158a9622008-01-02 13:04:48 -08005344#endif /* CONFIG_SLABINFO */