blob: 084184e706c63184124bcf874cfe6702e0343950 [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
Thomas Gleixner588f8ba2015-09-04 15:45:48 -07001309static void setup_object(struct kmem_cache *s, struct page *page,
1310 void *object)
1311{
1312 setup_object_debug(s, page, object);
1313 if (unlikely(s->ctor)) {
1314 kasan_unpoison_object_data(s, object);
1315 s->ctor(object);
1316 kasan_poison_object_data(s, object);
1317 }
1318}
1319
Christoph Lameter81819f02007-05-06 14:49:36 -07001320/*
1321 * Slab allocation and freeing
1322 */
Vladimir Davydov5dfb4172014-06-04 16:06:38 -07001323static inline struct page *alloc_slab_page(struct kmem_cache *s,
1324 gfp_t flags, int node, struct kmem_cache_order_objects oo)
Christoph Lameter65c33762008-04-14 19:11:40 +03001325{
Vladimir Davydov5dfb4172014-06-04 16:06:38 -07001326 struct page *page;
Christoph Lameter65c33762008-04-14 19:11:40 +03001327 int order = oo_order(oo);
1328
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001329 flags |= __GFP_NOTRACK;
1330
Vladimir Davydov5dfb4172014-06-04 16:06:38 -07001331 if (memcg_charge_slab(s, flags, order))
1332 return NULL;
1333
Christoph Lameter2154a332010-07-09 14:07:10 -05001334 if (node == NUMA_NO_NODE)
Vladimir Davydov5dfb4172014-06-04 16:06:38 -07001335 page = alloc_pages(flags, order);
Christoph Lameter65c33762008-04-14 19:11:40 +03001336 else
Vladimir Davydov5dfb4172014-06-04 16:06:38 -07001337 page = alloc_pages_exact_node(node, flags, order);
1338
1339 if (!page)
1340 memcg_uncharge_slab(s, order);
1341
1342 return page;
Christoph Lameter65c33762008-04-14 19:11:40 +03001343}
1344
Christoph Lameter81819f02007-05-06 14:49:36 -07001345static struct page *allocate_slab(struct kmem_cache *s, gfp_t flags, int node)
1346{
Pekka Enberg06428782008-01-07 23:20:27 -08001347 struct page *page;
Christoph Lameter834f3d12008-04-14 19:11:31 +03001348 struct kmem_cache_order_objects oo = s->oo;
Pekka Enbergba522702009-06-24 21:59:51 +03001349 gfp_t alloc_gfp;
Thomas Gleixner588f8ba2015-09-04 15:45:48 -07001350 void *start, *p;
1351 int idx, order;
Christoph Lameter81819f02007-05-06 14:49:36 -07001352
Christoph Lameter7e0528d2011-06-01 12:25:44 -05001353 flags &= gfp_allowed_mask;
1354
1355 if (flags & __GFP_WAIT)
1356 local_irq_enable();
1357
Christoph Lameterb7a49f02008-02-14 14:21:32 -08001358 flags |= s->allocflags;
Mel Gormane12ba742007-10-16 01:25:52 -07001359
Pekka Enbergba522702009-06-24 21:59:51 +03001360 /*
1361 * Let the initial higher-order allocation fail under memory pressure
1362 * so we fall-back to the minimum order allocation.
1363 */
1364 alloc_gfp = (flags | __GFP_NOWARN | __GFP_NORETRY) & ~__GFP_NOFAIL;
Joonsoo Kim45eb00c2015-09-04 15:45:54 -07001365 if ((alloc_gfp & __GFP_WAIT) && oo_order(oo) > oo_order(s->min))
1366 alloc_gfp = (alloc_gfp | __GFP_NOMEMALLOC) & ~__GFP_WAIT;
Pekka Enbergba522702009-06-24 21:59:51 +03001367
Vladimir Davydov5dfb4172014-06-04 16:06:38 -07001368 page = alloc_slab_page(s, alloc_gfp, node, oo);
Christoph Lameter65c33762008-04-14 19:11:40 +03001369 if (unlikely(!page)) {
1370 oo = s->min;
Joonsoo Kim80c3a992014-03-12 17:26:20 +09001371 alloc_gfp = flags;
Christoph Lameter65c33762008-04-14 19:11:40 +03001372 /*
1373 * Allocation may have failed due to fragmentation.
1374 * Try a lower order alloc if possible
1375 */
Vladimir Davydov5dfb4172014-06-04 16:06:38 -07001376 page = alloc_slab_page(s, alloc_gfp, node, oo);
Thomas Gleixner588f8ba2015-09-04 15:45:48 -07001377 if (unlikely(!page))
1378 goto out;
1379 stat(s, ORDER_FALLBACK);
Christoph Lameter65c33762008-04-14 19:11:40 +03001380 }
Vegard Nossum5a896d92008-04-04 00:54:48 +02001381
Thomas Gleixner588f8ba2015-09-04 15:45:48 -07001382 if (kmemcheck_enabled &&
1383 !(s->flags & (SLAB_NOTRACK | DEBUG_DEFAULT_FLAGS))) {
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001384 int pages = 1 << oo_order(oo);
1385
Joonsoo Kim80c3a992014-03-12 17:26:20 +09001386 kmemcheck_alloc_shadow(page, oo_order(oo), alloc_gfp, node);
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001387
1388 /*
1389 * Objects from caches that have a constructor don't get
1390 * cleared when they're allocated, so we need to do it here.
1391 */
1392 if (s->ctor)
1393 kmemcheck_mark_uninitialized_pages(page, pages);
1394 else
1395 kmemcheck_mark_unallocated_pages(page, pages);
Vegard Nossum5a896d92008-04-04 00:54:48 +02001396 }
1397
Christoph Lameter834f3d12008-04-14 19:11:31 +03001398 page->objects = oo_objects(oo);
Christoph Lameter81819f02007-05-06 14:49:36 -07001399
Glauber Costa1f458cb2012-12-18 14:22:50 -08001400 order = compound_order(page);
Glauber Costa1b4f59e32012-10-22 18:05:36 +04001401 page->slab_cache = s;
Joonsoo Kimc03f94c2012-05-18 00:47:47 +09001402 __SetPageSlab(page);
Michal Hocko2f064f32015-08-21 14:11:51 -07001403 if (page_is_pfmemalloc(page))
Mel Gorman072bb0a2012-07-31 16:43:58 -07001404 SetPageSlabPfmemalloc(page);
Christoph Lameter81819f02007-05-06 14:49:36 -07001405
1406 start = page_address(page);
Christoph Lameter81819f02007-05-06 14:49:36 -07001407
1408 if (unlikely(s->flags & SLAB_POISON))
Glauber Costa1f458cb2012-12-18 14:22:50 -08001409 memset(start, POISON_INUSE, PAGE_SIZE << order);
Christoph Lameter81819f02007-05-06 14:49:36 -07001410
Andrey Ryabinin0316bec2015-02-13 14:39:42 -08001411 kasan_poison_slab(page);
1412
Wei Yang54266642014-08-06 16:04:42 -07001413 for_each_object_idx(p, idx, s, start, page->objects) {
1414 setup_object(s, page, p);
1415 if (likely(idx < page->objects))
1416 set_freepointer(s, p, p + s->size);
1417 else
1418 set_freepointer(s, p, NULL);
Christoph Lameter81819f02007-05-06 14:49:36 -07001419 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001420
1421 page->freelist = start;
Christoph Lametere6e82ea2011-08-09 16:12:24 -05001422 page->inuse = page->objects;
Christoph Lameter8cb0a502011-06-01 12:25:46 -05001423 page->frozen = 1;
Thomas Gleixner588f8ba2015-09-04 15:45:48 -07001424
Christoph Lameter81819f02007-05-06 14:49:36 -07001425out:
Thomas Gleixner588f8ba2015-09-04 15:45:48 -07001426 if (flags & __GFP_WAIT)
1427 local_irq_disable();
1428 if (!page)
1429 return NULL;
1430
1431 mod_zone_page_state(page_zone(page),
1432 (s->flags & SLAB_RECLAIM_ACCOUNT) ?
1433 NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE,
1434 1 << oo_order(oo));
1435
1436 inc_slabs_node(s, page_to_nid(page), page->objects);
1437
Christoph Lameter81819f02007-05-06 14:49:36 -07001438 return page;
1439}
1440
Thomas Gleixner588f8ba2015-09-04 15:45:48 -07001441static struct page *new_slab(struct kmem_cache *s, gfp_t flags, int node)
1442{
1443 if (unlikely(flags & GFP_SLAB_BUG_MASK)) {
1444 pr_emerg("gfp: %u\n", flags & GFP_SLAB_BUG_MASK);
1445 BUG();
1446 }
1447
1448 return allocate_slab(s,
1449 flags & (GFP_RECLAIM_MASK | GFP_CONSTRAINT_MASK), node);
1450}
1451
Christoph Lameter81819f02007-05-06 14:49:36 -07001452static void __free_slab(struct kmem_cache *s, struct page *page)
1453{
Christoph Lameter834f3d12008-04-14 19:11:31 +03001454 int order = compound_order(page);
1455 int pages = 1 << order;
Christoph Lameter81819f02007-05-06 14:49:36 -07001456
Christoph Lameteraf537b02010-07-09 14:07:14 -05001457 if (kmem_cache_debug(s)) {
Christoph Lameter81819f02007-05-06 14:49:36 -07001458 void *p;
1459
1460 slab_pad_check(s, page);
Christoph Lameter224a88b2008-04-14 19:11:31 +03001461 for_each_object(p, s, page_address(page),
1462 page->objects)
Christoph Lameterf7cb1932010-09-29 07:15:01 -05001463 check_object(s, page, p, SLUB_RED_INACTIVE);
Christoph Lameter81819f02007-05-06 14:49:36 -07001464 }
1465
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001466 kmemcheck_free_shadow(page, compound_order(page));
Vegard Nossum5a896d92008-04-04 00:54:48 +02001467
Christoph Lameter81819f02007-05-06 14:49:36 -07001468 mod_zone_page_state(page_zone(page),
1469 (s->flags & SLAB_RECLAIM_ACCOUNT) ?
1470 NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE,
Pekka Enberg06428782008-01-07 23:20:27 -08001471 -pages);
Christoph Lameter81819f02007-05-06 14:49:36 -07001472
Mel Gorman072bb0a2012-07-31 16:43:58 -07001473 __ClearPageSlabPfmemalloc(page);
Christoph Lameter49bd5222008-04-14 18:52:18 +03001474 __ClearPageSlab(page);
Glauber Costa1f458cb2012-12-18 14:22:50 -08001475
Mel Gorman22b751c2013-02-22 16:34:59 -08001476 page_mapcount_reset(page);
Nick Piggin1eb5ac62009-05-05 19:13:44 +10001477 if (current->reclaim_state)
1478 current->reclaim_state->reclaimed_slab += pages;
Vladimir Davydov5dfb4172014-06-04 16:06:38 -07001479 __free_pages(page, order);
1480 memcg_uncharge_slab(s, order);
Christoph Lameter81819f02007-05-06 14:49:36 -07001481}
1482
Lai Jiangshanda9a6382011-03-10 15:22:00 +08001483#define need_reserve_slab_rcu \
1484 (sizeof(((struct page *)NULL)->lru) < sizeof(struct rcu_head))
1485
Christoph Lameter81819f02007-05-06 14:49:36 -07001486static void rcu_free_slab(struct rcu_head *h)
1487{
1488 struct page *page;
1489
Lai Jiangshanda9a6382011-03-10 15:22:00 +08001490 if (need_reserve_slab_rcu)
1491 page = virt_to_head_page(h);
1492 else
1493 page = container_of((struct list_head *)h, struct page, lru);
1494
Glauber Costa1b4f59e32012-10-22 18:05:36 +04001495 __free_slab(page->slab_cache, page);
Christoph Lameter81819f02007-05-06 14:49:36 -07001496}
1497
1498static void free_slab(struct kmem_cache *s, struct page *page)
1499{
1500 if (unlikely(s->flags & SLAB_DESTROY_BY_RCU)) {
Lai Jiangshanda9a6382011-03-10 15:22:00 +08001501 struct rcu_head *head;
1502
1503 if (need_reserve_slab_rcu) {
1504 int order = compound_order(page);
1505 int offset = (PAGE_SIZE << order) - s->reserved;
1506
1507 VM_BUG_ON(s->reserved != sizeof(*head));
1508 head = page_address(page) + offset;
1509 } else {
1510 /*
1511 * RCU free overloads the RCU head over the LRU
1512 */
1513 head = (void *)&page->lru;
1514 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001515
1516 call_rcu(head, rcu_free_slab);
1517 } else
1518 __free_slab(s, page);
1519}
1520
1521static void discard_slab(struct kmem_cache *s, struct page *page)
1522{
Christoph Lameter205ab992008-04-14 19:11:40 +03001523 dec_slabs_node(s, page_to_nid(page), page->objects);
Christoph Lameter81819f02007-05-06 14:49:36 -07001524 free_slab(s, page);
1525}
1526
1527/*
Christoph Lameter5cc6eee2011-06-01 12:25:50 -05001528 * Management of partially allocated slabs.
Christoph Lameter81819f02007-05-06 14:49:36 -07001529 */
Steven Rostedt1e4dd942014-02-10 14:25:46 -08001530static inline void
1531__add_partial(struct kmem_cache_node *n, struct page *page, int tail)
Christoph Lameter81819f02007-05-06 14:49:36 -07001532{
Christoph Lametere95eed52007-05-06 14:49:44 -07001533 n->nr_partial++;
Shaohua Li136333d2011-08-24 08:57:52 +08001534 if (tail == DEACTIVATE_TO_TAIL)
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001535 list_add_tail(&page->lru, &n->partial);
1536 else
1537 list_add(&page->lru, &n->partial);
Christoph Lameter81819f02007-05-06 14:49:36 -07001538}
1539
Steven Rostedt1e4dd942014-02-10 14:25:46 -08001540static inline void add_partial(struct kmem_cache_node *n,
1541 struct page *page, int tail)
1542{
1543 lockdep_assert_held(&n->list_lock);
1544 __add_partial(n, page, tail);
1545}
1546
1547static inline void
1548__remove_partial(struct kmem_cache_node *n, struct page *page)
1549{
1550 list_del(&page->lru);
1551 n->nr_partial--;
1552}
1553
Christoph Lameter5cc6eee2011-06-01 12:25:50 -05001554static inline void remove_partial(struct kmem_cache_node *n,
Christoph Lameter62e346a2010-09-28 08:10:28 -05001555 struct page *page)
1556{
Peter Zijlstrac65c1872014-01-10 13:23:49 +01001557 lockdep_assert_held(&n->list_lock);
Steven Rostedt1e4dd942014-02-10 14:25:46 -08001558 __remove_partial(n, page);
Christoph Lameter62e346a2010-09-28 08:10:28 -05001559}
1560
Christoph Lameter81819f02007-05-06 14:49:36 -07001561/*
Christoph Lameter7ced3712012-05-09 10:09:53 -05001562 * Remove slab from the partial list, freeze it and
1563 * return the pointer to the freelist.
Christoph Lameter81819f02007-05-06 14:49:36 -07001564 *
Christoph Lameter497b66f2011-08-09 16:12:26 -05001565 * Returns a list of objects or NULL if it fails.
Christoph Lameter81819f02007-05-06 14:49:36 -07001566 */
Christoph Lameter497b66f2011-08-09 16:12:26 -05001567static inline void *acquire_slab(struct kmem_cache *s,
Christoph Lameteracd19fd2011-08-09 16:12:25 -05001568 struct kmem_cache_node *n, struct page *page,
Joonsoo Kim633b0762013-01-21 17:01:25 +09001569 int mode, int *objects)
Christoph Lameter81819f02007-05-06 14:49:36 -07001570{
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001571 void *freelist;
1572 unsigned long counters;
1573 struct page new;
1574
Peter Zijlstrac65c1872014-01-10 13:23:49 +01001575 lockdep_assert_held(&n->list_lock);
1576
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001577 /*
1578 * Zap the freelist and set the frozen bit.
1579 * The old freelist is the list of objects for the
1580 * per cpu allocation list.
1581 */
Christoph Lameter7ced3712012-05-09 10:09:53 -05001582 freelist = page->freelist;
1583 counters = page->counters;
1584 new.counters = counters;
Joonsoo Kim633b0762013-01-21 17:01:25 +09001585 *objects = new.objects - new.inuse;
Pekka Enberg23910c52012-06-04 10:14:58 +03001586 if (mode) {
Christoph Lameter7ced3712012-05-09 10:09:53 -05001587 new.inuse = page->objects;
Pekka Enberg23910c52012-06-04 10:14:58 +03001588 new.freelist = NULL;
1589 } else {
1590 new.freelist = freelist;
1591 }
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001592
Dave Hansena0132ac2014-01-29 14:05:50 -08001593 VM_BUG_ON(new.frozen);
Christoph Lameter7ced3712012-05-09 10:09:53 -05001594 new.frozen = 1;
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001595
Christoph Lameter7ced3712012-05-09 10:09:53 -05001596 if (!__cmpxchg_double_slab(s, page,
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001597 freelist, counters,
Joonsoo Kim02d76332012-05-17 00:13:02 +09001598 new.freelist, new.counters,
Christoph Lameter7ced3712012-05-09 10:09:53 -05001599 "acquire_slab"))
Christoph Lameter7ced3712012-05-09 10:09:53 -05001600 return NULL;
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001601
1602 remove_partial(n, page);
Christoph Lameter7ced3712012-05-09 10:09:53 -05001603 WARN_ON(!freelist);
Christoph Lameter49e22582011-08-09 16:12:27 -05001604 return freelist;
Christoph Lameter81819f02007-05-06 14:49:36 -07001605}
1606
Joonsoo Kim633b0762013-01-21 17:01:25 +09001607static void put_cpu_partial(struct kmem_cache *s, struct page *page, int drain);
Joonsoo Kim8ba00bb2012-09-17 14:09:09 -07001608static inline bool pfmemalloc_match(struct page *page, gfp_t gfpflags);
Christoph Lameter49e22582011-08-09 16:12:27 -05001609
Christoph Lameter81819f02007-05-06 14:49:36 -07001610/*
Christoph Lameter672bba32007-05-09 02:32:39 -07001611 * Try to allocate a partial slab from a specific node.
Christoph Lameter81819f02007-05-06 14:49:36 -07001612 */
Joonsoo Kim8ba00bb2012-09-17 14:09:09 -07001613static void *get_partial_node(struct kmem_cache *s, struct kmem_cache_node *n,
1614 struct kmem_cache_cpu *c, gfp_t flags)
Christoph Lameter81819f02007-05-06 14:49:36 -07001615{
Christoph Lameter49e22582011-08-09 16:12:27 -05001616 struct page *page, *page2;
1617 void *object = NULL;
Joonsoo Kim633b0762013-01-21 17:01:25 +09001618 int available = 0;
1619 int objects;
Christoph Lameter81819f02007-05-06 14:49:36 -07001620
1621 /*
1622 * Racy check. If we mistakenly see no partial slabs then we
1623 * just allocate an empty slab. If we mistakenly try to get a
Christoph Lameter672bba32007-05-09 02:32:39 -07001624 * partial slab and there is none available then get_partials()
1625 * will return NULL.
Christoph Lameter81819f02007-05-06 14:49:36 -07001626 */
1627 if (!n || !n->nr_partial)
1628 return NULL;
1629
1630 spin_lock(&n->list_lock);
Christoph Lameter49e22582011-08-09 16:12:27 -05001631 list_for_each_entry_safe(page, page2, &n->partial, lru) {
Joonsoo Kim8ba00bb2012-09-17 14:09:09 -07001632 void *t;
Christoph Lameter49e22582011-08-09 16:12:27 -05001633
Joonsoo Kim8ba00bb2012-09-17 14:09:09 -07001634 if (!pfmemalloc_match(page, flags))
1635 continue;
1636
Joonsoo Kim633b0762013-01-21 17:01:25 +09001637 t = acquire_slab(s, n, page, object == NULL, &objects);
Christoph Lameter49e22582011-08-09 16:12:27 -05001638 if (!t)
1639 break;
1640
Joonsoo Kim633b0762013-01-21 17:01:25 +09001641 available += objects;
Alex,Shi12d79632011-09-07 10:26:36 +08001642 if (!object) {
Christoph Lameter49e22582011-08-09 16:12:27 -05001643 c->page = page;
Christoph Lameter49e22582011-08-09 16:12:27 -05001644 stat(s, ALLOC_FROM_PARTIAL);
Christoph Lameter49e22582011-08-09 16:12:27 -05001645 object = t;
Christoph Lameter49e22582011-08-09 16:12:27 -05001646 } else {
Joonsoo Kim633b0762013-01-21 17:01:25 +09001647 put_cpu_partial(s, page, 0);
Alex Shi8028dce2012-02-03 23:34:56 +08001648 stat(s, CPU_PARTIAL_NODE);
Christoph Lameter49e22582011-08-09 16:12:27 -05001649 }
Joonsoo Kim345c9052013-06-19 14:05:52 +09001650 if (!kmem_cache_has_cpu_partial(s)
1651 || available > s->cpu_partial / 2)
Christoph Lameter49e22582011-08-09 16:12:27 -05001652 break;
1653
Christoph Lameter497b66f2011-08-09 16:12:26 -05001654 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001655 spin_unlock(&n->list_lock);
Christoph Lameter497b66f2011-08-09 16:12:26 -05001656 return object;
Christoph Lameter81819f02007-05-06 14:49:36 -07001657}
1658
1659/*
Christoph Lameter672bba32007-05-09 02:32:39 -07001660 * Get a page from somewhere. Search in increasing NUMA distances.
Christoph Lameter81819f02007-05-06 14:49:36 -07001661 */
Joonsoo Kimde3ec032012-01-27 00:12:23 -08001662static void *get_any_partial(struct kmem_cache *s, gfp_t flags,
Christoph Lameteracd19fd2011-08-09 16:12:25 -05001663 struct kmem_cache_cpu *c)
Christoph Lameter81819f02007-05-06 14:49:36 -07001664{
1665#ifdef CONFIG_NUMA
1666 struct zonelist *zonelist;
Mel Gormandd1a2392008-04-28 02:12:17 -07001667 struct zoneref *z;
Mel Gorman54a6eb52008-04-28 02:12:16 -07001668 struct zone *zone;
1669 enum zone_type high_zoneidx = gfp_zone(flags);
Christoph Lameter497b66f2011-08-09 16:12:26 -05001670 void *object;
Mel Gormancc9a6c82012-03-21 16:34:11 -07001671 unsigned int cpuset_mems_cookie;
Christoph Lameter81819f02007-05-06 14:49:36 -07001672
1673 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07001674 * The defrag ratio allows a configuration of the tradeoffs between
1675 * inter node defragmentation and node local allocations. A lower
1676 * defrag_ratio increases the tendency to do local allocations
1677 * instead of attempting to obtain partial slabs from other nodes.
Christoph Lameter81819f02007-05-06 14:49:36 -07001678 *
Christoph Lameter672bba32007-05-09 02:32:39 -07001679 * If the defrag_ratio is set to 0 then kmalloc() always
1680 * returns node local objects. If the ratio is higher then kmalloc()
1681 * may return off node objects because partial slabs are obtained
1682 * from other nodes and filled up.
Christoph Lameter81819f02007-05-06 14:49:36 -07001683 *
Christoph Lameter6446faa2008-02-15 23:45:26 -08001684 * If /sys/kernel/slab/xx/defrag_ratio is set to 100 (which makes
Christoph Lameter672bba32007-05-09 02:32:39 -07001685 * defrag_ratio = 1000) then every (well almost) allocation will
1686 * first attempt to defrag slab caches on other nodes. This means
1687 * scanning over all nodes to look for partial slabs which may be
1688 * expensive if we do it every time we are trying to find a slab
1689 * with available objects.
Christoph Lameter81819f02007-05-06 14:49:36 -07001690 */
Christoph Lameter98246012008-01-07 23:20:26 -08001691 if (!s->remote_node_defrag_ratio ||
1692 get_cycles() % 1024 > s->remote_node_defrag_ratio)
Christoph Lameter81819f02007-05-06 14:49:36 -07001693 return NULL;
1694
Mel Gormancc9a6c82012-03-21 16:34:11 -07001695 do {
Mel Gormand26914d2014-04-03 14:47:24 -07001696 cpuset_mems_cookie = read_mems_allowed_begin();
David Rientjes2a389612014-04-07 15:37:29 -07001697 zonelist = node_zonelist(mempolicy_slab_node(), flags);
Mel Gormancc9a6c82012-03-21 16:34:11 -07001698 for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) {
1699 struct kmem_cache_node *n;
Christoph Lameter81819f02007-05-06 14:49:36 -07001700
Mel Gormancc9a6c82012-03-21 16:34:11 -07001701 n = get_node(s, zone_to_nid(zone));
Christoph Lameter81819f02007-05-06 14:49:36 -07001702
Vladimir Davydovdee2f8a2014-12-12 16:58:28 -08001703 if (n && cpuset_zone_allowed(zone, flags) &&
Mel Gormancc9a6c82012-03-21 16:34:11 -07001704 n->nr_partial > s->min_partial) {
Joonsoo Kim8ba00bb2012-09-17 14:09:09 -07001705 object = get_partial_node(s, n, c, flags);
Mel Gormancc9a6c82012-03-21 16:34:11 -07001706 if (object) {
1707 /*
Mel Gormand26914d2014-04-03 14:47:24 -07001708 * Don't check read_mems_allowed_retry()
1709 * here - if mems_allowed was updated in
1710 * parallel, that was a harmless race
1711 * between allocation and the cpuset
1712 * update
Mel Gormancc9a6c82012-03-21 16:34:11 -07001713 */
Mel Gormancc9a6c82012-03-21 16:34:11 -07001714 return object;
1715 }
Miao Xiec0ff7452010-05-24 14:32:08 -07001716 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001717 }
Mel Gormand26914d2014-04-03 14:47:24 -07001718 } while (read_mems_allowed_retry(cpuset_mems_cookie));
Christoph Lameter81819f02007-05-06 14:49:36 -07001719#endif
1720 return NULL;
1721}
1722
1723/*
1724 * Get a partial page, lock it and return it.
1725 */
Christoph Lameter497b66f2011-08-09 16:12:26 -05001726static void *get_partial(struct kmem_cache *s, gfp_t flags, int node,
Christoph Lameteracd19fd2011-08-09 16:12:25 -05001727 struct kmem_cache_cpu *c)
Christoph Lameter81819f02007-05-06 14:49:36 -07001728{
Christoph Lameter497b66f2011-08-09 16:12:26 -05001729 void *object;
Joonsoo Kima561ce02014-10-09 15:26:15 -07001730 int searchnode = node;
1731
1732 if (node == NUMA_NO_NODE)
1733 searchnode = numa_mem_id();
1734 else if (!node_present_pages(node))
1735 searchnode = node_to_mem_node(node);
Christoph Lameter81819f02007-05-06 14:49:36 -07001736
Joonsoo Kim8ba00bb2012-09-17 14:09:09 -07001737 object = get_partial_node(s, get_node(s, searchnode), c, flags);
Christoph Lameter497b66f2011-08-09 16:12:26 -05001738 if (object || node != NUMA_NO_NODE)
1739 return object;
Christoph Lameter81819f02007-05-06 14:49:36 -07001740
Christoph Lameteracd19fd2011-08-09 16:12:25 -05001741 return get_any_partial(s, flags, c);
Christoph Lameter81819f02007-05-06 14:49:36 -07001742}
1743
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001744#ifdef CONFIG_PREEMPT
1745/*
1746 * Calculate the next globally unique transaction for disambiguiation
1747 * during cmpxchg. The transactions start with the cpu number and are then
1748 * incremented by CONFIG_NR_CPUS.
1749 */
1750#define TID_STEP roundup_pow_of_two(CONFIG_NR_CPUS)
1751#else
1752/*
1753 * No preemption supported therefore also no need to check for
1754 * different cpus.
1755 */
1756#define TID_STEP 1
1757#endif
1758
1759static inline unsigned long next_tid(unsigned long tid)
1760{
1761 return tid + TID_STEP;
1762}
1763
1764static inline unsigned int tid_to_cpu(unsigned long tid)
1765{
1766 return tid % TID_STEP;
1767}
1768
1769static inline unsigned long tid_to_event(unsigned long tid)
1770{
1771 return tid / TID_STEP;
1772}
1773
1774static inline unsigned int init_tid(int cpu)
1775{
1776 return cpu;
1777}
1778
1779static inline void note_cmpxchg_failure(const char *n,
1780 const struct kmem_cache *s, unsigned long tid)
1781{
1782#ifdef SLUB_DEBUG_CMPXCHG
1783 unsigned long actual_tid = __this_cpu_read(s->cpu_slab->tid);
1784
Fabian Frederickf9f58282014-06-04 16:06:34 -07001785 pr_info("%s %s: cmpxchg redo ", n, s->name);
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001786
1787#ifdef CONFIG_PREEMPT
1788 if (tid_to_cpu(tid) != tid_to_cpu(actual_tid))
Fabian Frederickf9f58282014-06-04 16:06:34 -07001789 pr_warn("due to cpu change %d -> %d\n",
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001790 tid_to_cpu(tid), tid_to_cpu(actual_tid));
1791 else
1792#endif
1793 if (tid_to_event(tid) != tid_to_event(actual_tid))
Fabian Frederickf9f58282014-06-04 16:06:34 -07001794 pr_warn("due to cpu running other code. Event %ld->%ld\n",
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001795 tid_to_event(tid), tid_to_event(actual_tid));
1796 else
Fabian Frederickf9f58282014-06-04 16:06:34 -07001797 pr_warn("for unknown reason: actual=%lx was=%lx target=%lx\n",
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001798 actual_tid, tid, next_tid(tid));
1799#endif
Christoph Lameter4fdccdf2011-03-22 13:35:00 -05001800 stat(s, CMPXCHG_DOUBLE_CPU_FAIL);
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001801}
1802
Fengguang Wu788e1aa2012-09-28 16:34:05 +08001803static void init_kmem_cache_cpus(struct kmem_cache *s)
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001804{
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001805 int cpu;
1806
1807 for_each_possible_cpu(cpu)
1808 per_cpu_ptr(s->cpu_slab, cpu)->tid = init_tid(cpu);
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001809}
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001810
1811/*
1812 * Remove the cpu slab
1813 */
Chen Gangd0e0ac92013-07-15 09:05:29 +08001814static void deactivate_slab(struct kmem_cache *s, struct page *page,
1815 void *freelist)
Christoph Lameter81819f02007-05-06 14:49:36 -07001816{
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001817 enum slab_modes { M_NONE, M_PARTIAL, M_FULL, M_FREE };
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001818 struct kmem_cache_node *n = get_node(s, page_to_nid(page));
1819 int lock = 0;
1820 enum slab_modes l = M_NONE, m = M_NONE;
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001821 void *nextfree;
Shaohua Li136333d2011-08-24 08:57:52 +08001822 int tail = DEACTIVATE_TO_HEAD;
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001823 struct page new;
1824 struct page old;
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08001825
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001826 if (page->freelist) {
Christoph Lameter84e554e62009-12-18 16:26:23 -06001827 stat(s, DEACTIVATE_REMOTE_FREES);
Shaohua Li136333d2011-08-24 08:57:52 +08001828 tail = DEACTIVATE_TO_TAIL;
Christoph Lameter894b8782007-05-10 03:15:16 -07001829 }
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001830
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001831 /*
1832 * Stage one: Free all available per cpu objects back
1833 * to the page freelist while it is still frozen. Leave the
1834 * last one.
1835 *
1836 * There is no need to take the list->lock because the page
1837 * is still frozen.
1838 */
1839 while (freelist && (nextfree = get_freepointer(s, freelist))) {
1840 void *prior;
1841 unsigned long counters;
1842
1843 do {
1844 prior = page->freelist;
1845 counters = page->counters;
1846 set_freepointer(s, freelist, prior);
1847 new.counters = counters;
1848 new.inuse--;
Dave Hansena0132ac2014-01-29 14:05:50 -08001849 VM_BUG_ON(!new.frozen);
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001850
Christoph Lameter1d071712011-07-14 12:49:12 -05001851 } while (!__cmpxchg_double_slab(s, page,
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001852 prior, counters,
1853 freelist, new.counters,
1854 "drain percpu freelist"));
1855
1856 freelist = nextfree;
1857 }
1858
1859 /*
1860 * Stage two: Ensure that the page is unfrozen while the
1861 * list presence reflects the actual number of objects
1862 * during unfreeze.
1863 *
1864 * We setup the list membership and then perform a cmpxchg
1865 * with the count. If there is a mismatch then the page
1866 * is not unfrozen but the page is on the wrong list.
1867 *
1868 * Then we restart the process which may have to remove
1869 * the page from the list that we just put it on again
1870 * because the number of objects in the slab may have
1871 * changed.
1872 */
1873redo:
1874
1875 old.freelist = page->freelist;
1876 old.counters = page->counters;
Dave Hansena0132ac2014-01-29 14:05:50 -08001877 VM_BUG_ON(!old.frozen);
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001878
1879 /* Determine target state of the slab */
1880 new.counters = old.counters;
1881 if (freelist) {
1882 new.inuse--;
1883 set_freepointer(s, freelist, old.freelist);
1884 new.freelist = freelist;
1885 } else
1886 new.freelist = old.freelist;
1887
1888 new.frozen = 0;
1889
Joonsoo Kim8a5b20a2014-07-02 15:22:35 -07001890 if (!new.inuse && n->nr_partial >= s->min_partial)
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001891 m = M_FREE;
1892 else if (new.freelist) {
1893 m = M_PARTIAL;
1894 if (!lock) {
1895 lock = 1;
1896 /*
1897 * Taking the spinlock removes the possiblity
1898 * that acquire_slab() will see a slab page that
1899 * is frozen
1900 */
1901 spin_lock(&n->list_lock);
1902 }
1903 } else {
1904 m = M_FULL;
1905 if (kmem_cache_debug(s) && !lock) {
1906 lock = 1;
1907 /*
1908 * This also ensures that the scanning of full
1909 * slabs from diagnostic functions will not see
1910 * any frozen slabs.
1911 */
1912 spin_lock(&n->list_lock);
1913 }
1914 }
1915
1916 if (l != m) {
1917
1918 if (l == M_PARTIAL)
1919
1920 remove_partial(n, page);
1921
1922 else if (l == M_FULL)
1923
Peter Zijlstrac65c1872014-01-10 13:23:49 +01001924 remove_full(s, n, page);
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001925
1926 if (m == M_PARTIAL) {
1927
1928 add_partial(n, page, tail);
Shaohua Li136333d2011-08-24 08:57:52 +08001929 stat(s, tail);
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001930
1931 } else if (m == M_FULL) {
1932
1933 stat(s, DEACTIVATE_FULL);
1934 add_full(s, n, page);
1935
1936 }
1937 }
1938
1939 l = m;
Christoph Lameter1d071712011-07-14 12:49:12 -05001940 if (!__cmpxchg_double_slab(s, page,
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001941 old.freelist, old.counters,
1942 new.freelist, new.counters,
1943 "unfreezing slab"))
1944 goto redo;
1945
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001946 if (lock)
1947 spin_unlock(&n->list_lock);
1948
1949 if (m == M_FREE) {
1950 stat(s, DEACTIVATE_EMPTY);
1951 discard_slab(s, page);
1952 stat(s, FREE_SLAB);
1953 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001954}
1955
Joonsoo Kimd24ac772012-05-18 22:01:17 +09001956/*
1957 * Unfreeze all the cpu partial slabs.
1958 *
Christoph Lameter59a09912012-11-28 16:23:00 +00001959 * This function must be called with interrupts disabled
1960 * for the cpu using c (or some other guarantee must be there
1961 * to guarantee no concurrent accesses).
Joonsoo Kimd24ac772012-05-18 22:01:17 +09001962 */
Christoph Lameter59a09912012-11-28 16:23:00 +00001963static void unfreeze_partials(struct kmem_cache *s,
1964 struct kmem_cache_cpu *c)
Christoph Lameter49e22582011-08-09 16:12:27 -05001965{
Joonsoo Kim345c9052013-06-19 14:05:52 +09001966#ifdef CONFIG_SLUB_CPU_PARTIAL
Joonsoo Kim43d77862012-06-09 02:23:16 +09001967 struct kmem_cache_node *n = NULL, *n2 = NULL;
Shaohua Li9ada1932011-11-14 13:34:13 +08001968 struct page *page, *discard_page = NULL;
Christoph Lameter49e22582011-08-09 16:12:27 -05001969
1970 while ((page = c->partial)) {
Christoph Lameter49e22582011-08-09 16:12:27 -05001971 struct page new;
1972 struct page old;
1973
1974 c->partial = page->next;
Joonsoo Kim43d77862012-06-09 02:23:16 +09001975
1976 n2 = get_node(s, page_to_nid(page));
1977 if (n != n2) {
1978 if (n)
1979 spin_unlock(&n->list_lock);
1980
1981 n = n2;
1982 spin_lock(&n->list_lock);
1983 }
Christoph Lameter49e22582011-08-09 16:12:27 -05001984
1985 do {
1986
1987 old.freelist = page->freelist;
1988 old.counters = page->counters;
Dave Hansena0132ac2014-01-29 14:05:50 -08001989 VM_BUG_ON(!old.frozen);
Christoph Lameter49e22582011-08-09 16:12:27 -05001990
1991 new.counters = old.counters;
1992 new.freelist = old.freelist;
1993
1994 new.frozen = 0;
1995
Joonsoo Kimd24ac772012-05-18 22:01:17 +09001996 } while (!__cmpxchg_double_slab(s, page,
Christoph Lameter49e22582011-08-09 16:12:27 -05001997 old.freelist, old.counters,
1998 new.freelist, new.counters,
1999 "unfreezing slab"));
2000
Joonsoo Kim8a5b20a2014-07-02 15:22:35 -07002001 if (unlikely(!new.inuse && n->nr_partial >= s->min_partial)) {
Shaohua Li9ada1932011-11-14 13:34:13 +08002002 page->next = discard_page;
2003 discard_page = page;
Joonsoo Kim43d77862012-06-09 02:23:16 +09002004 } else {
2005 add_partial(n, page, DEACTIVATE_TO_TAIL);
2006 stat(s, FREE_ADD_PARTIAL);
Christoph Lameter49e22582011-08-09 16:12:27 -05002007 }
2008 }
2009
2010 if (n)
2011 spin_unlock(&n->list_lock);
Shaohua Li9ada1932011-11-14 13:34:13 +08002012
2013 while (discard_page) {
2014 page = discard_page;
2015 discard_page = discard_page->next;
2016
2017 stat(s, DEACTIVATE_EMPTY);
2018 discard_slab(s, page);
2019 stat(s, FREE_SLAB);
2020 }
Joonsoo Kim345c9052013-06-19 14:05:52 +09002021#endif
Christoph Lameter49e22582011-08-09 16:12:27 -05002022}
2023
2024/*
2025 * Put a page that was just frozen (in __slab_free) into a partial page
2026 * slot if available. This is done without interrupts disabled and without
2027 * preemption disabled. The cmpxchg is racy and may put the partial page
2028 * onto a random cpus partial slot.
2029 *
2030 * If we did not find a slot then simply move all the partials to the
2031 * per node partial list.
2032 */
Joonsoo Kim633b0762013-01-21 17:01:25 +09002033static void put_cpu_partial(struct kmem_cache *s, struct page *page, int drain)
Christoph Lameter49e22582011-08-09 16:12:27 -05002034{
Joonsoo Kim345c9052013-06-19 14:05:52 +09002035#ifdef CONFIG_SLUB_CPU_PARTIAL
Christoph Lameter49e22582011-08-09 16:12:27 -05002036 struct page *oldpage;
2037 int pages;
2038 int pobjects;
2039
Vladimir Davydovd6e0b7f2015-02-12 14:59:47 -08002040 preempt_disable();
Christoph Lameter49e22582011-08-09 16:12:27 -05002041 do {
2042 pages = 0;
2043 pobjects = 0;
2044 oldpage = this_cpu_read(s->cpu_slab->partial);
2045
2046 if (oldpage) {
2047 pobjects = oldpage->pobjects;
2048 pages = oldpage->pages;
2049 if (drain && pobjects > s->cpu_partial) {
2050 unsigned long flags;
2051 /*
2052 * partial array is full. Move the existing
2053 * set to the per node partial list.
2054 */
2055 local_irq_save(flags);
Christoph Lameter59a09912012-11-28 16:23:00 +00002056 unfreeze_partials(s, this_cpu_ptr(s->cpu_slab));
Christoph Lameter49e22582011-08-09 16:12:27 -05002057 local_irq_restore(flags);
Joonsoo Kime24fc412012-06-23 03:22:38 +09002058 oldpage = NULL;
Christoph Lameter49e22582011-08-09 16:12:27 -05002059 pobjects = 0;
2060 pages = 0;
Alex Shi8028dce2012-02-03 23:34:56 +08002061 stat(s, CPU_PARTIAL_DRAIN);
Christoph Lameter49e22582011-08-09 16:12:27 -05002062 }
2063 }
2064
2065 pages++;
2066 pobjects += page->objects - page->inuse;
2067
2068 page->pages = pages;
2069 page->pobjects = pobjects;
2070 page->next = oldpage;
2071
Chen Gangd0e0ac92013-07-15 09:05:29 +08002072 } while (this_cpu_cmpxchg(s->cpu_slab->partial, oldpage, page)
2073 != oldpage);
Vladimir Davydovd6e0b7f2015-02-12 14:59:47 -08002074 if (unlikely(!s->cpu_partial)) {
2075 unsigned long flags;
2076
2077 local_irq_save(flags);
2078 unfreeze_partials(s, this_cpu_ptr(s->cpu_slab));
2079 local_irq_restore(flags);
2080 }
2081 preempt_enable();
Joonsoo Kim345c9052013-06-19 14:05:52 +09002082#endif
Christoph Lameter49e22582011-08-09 16:12:27 -05002083}
2084
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002085static inline void flush_slab(struct kmem_cache *s, struct kmem_cache_cpu *c)
Christoph Lameter81819f02007-05-06 14:49:36 -07002086{
Christoph Lameter84e554e62009-12-18 16:26:23 -06002087 stat(s, CPUSLAB_FLUSH);
Christoph Lameterc17dda42012-05-09 10:09:57 -05002088 deactivate_slab(s, c->page, c->freelist);
2089
2090 c->tid = next_tid(c->tid);
2091 c->page = NULL;
2092 c->freelist = NULL;
Christoph Lameter81819f02007-05-06 14:49:36 -07002093}
2094
2095/*
2096 * Flush cpu slab.
Christoph Lameter6446faa2008-02-15 23:45:26 -08002097 *
Christoph Lameter81819f02007-05-06 14:49:36 -07002098 * Called from IPI handler with interrupts disabled.
2099 */
Christoph Lameter0c710012007-07-17 04:03:24 -07002100static inline void __flush_cpu_slab(struct kmem_cache *s, int cpu)
Christoph Lameter81819f02007-05-06 14:49:36 -07002101{
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06002102 struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu);
Christoph Lameter81819f02007-05-06 14:49:36 -07002103
Christoph Lameter49e22582011-08-09 16:12:27 -05002104 if (likely(c)) {
2105 if (c->page)
2106 flush_slab(s, c);
2107
Christoph Lameter59a09912012-11-28 16:23:00 +00002108 unfreeze_partials(s, c);
Christoph Lameter49e22582011-08-09 16:12:27 -05002109 }
Christoph Lameter81819f02007-05-06 14:49:36 -07002110}
2111
2112static void flush_cpu_slab(void *d)
2113{
2114 struct kmem_cache *s = d;
Christoph Lameter81819f02007-05-06 14:49:36 -07002115
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002116 __flush_cpu_slab(s, smp_processor_id());
Christoph Lameter81819f02007-05-06 14:49:36 -07002117}
2118
Gilad Ben-Yossefa8364d52012-03-28 14:42:44 -07002119static bool has_cpu_slab(int cpu, void *info)
2120{
2121 struct kmem_cache *s = info;
2122 struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu);
2123
majianpeng02e1a9c2012-05-17 17:03:26 -07002124 return c->page || c->partial;
Gilad Ben-Yossefa8364d52012-03-28 14:42:44 -07002125}
2126
Christoph Lameter81819f02007-05-06 14:49:36 -07002127static void flush_all(struct kmem_cache *s)
2128{
Gilad Ben-Yossefa8364d52012-03-28 14:42:44 -07002129 on_each_cpu_cond(has_cpu_slab, flush_cpu_slab, s, 1, GFP_ATOMIC);
Christoph Lameter81819f02007-05-06 14:49:36 -07002130}
2131
2132/*
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002133 * Check if the objects in a per cpu structure fit numa
2134 * locality expectations.
2135 */
Christoph Lameter57d437d2012-05-09 10:09:59 -05002136static inline int node_match(struct page *page, int node)
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002137{
2138#ifdef CONFIG_NUMA
Christoph Lameter4d7868e2013-01-23 21:45:47 +00002139 if (!page || (node != NUMA_NO_NODE && page_to_nid(page) != node))
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002140 return 0;
2141#endif
2142 return 1;
2143}
2144
David Rientjes9a02d692014-06-04 16:06:36 -07002145#ifdef CONFIG_SLUB_DEBUG
Pekka Enberg781b2ba2009-06-10 18:50:32 +03002146static int count_free(struct page *page)
2147{
2148 return page->objects - page->inuse;
2149}
2150
David Rientjes9a02d692014-06-04 16:06:36 -07002151static inline unsigned long node_nr_objs(struct kmem_cache_node *n)
2152{
2153 return atomic_long_read(&n->total_objects);
2154}
2155#endif /* CONFIG_SLUB_DEBUG */
2156
2157#if defined(CONFIG_SLUB_DEBUG) || defined(CONFIG_SYSFS)
Pekka Enberg781b2ba2009-06-10 18:50:32 +03002158static unsigned long count_partial(struct kmem_cache_node *n,
2159 int (*get_count)(struct page *))
2160{
2161 unsigned long flags;
2162 unsigned long x = 0;
2163 struct page *page;
2164
2165 spin_lock_irqsave(&n->list_lock, flags);
2166 list_for_each_entry(page, &n->partial, lru)
2167 x += get_count(page);
2168 spin_unlock_irqrestore(&n->list_lock, flags);
2169 return x;
2170}
David Rientjes9a02d692014-06-04 16:06:36 -07002171#endif /* CONFIG_SLUB_DEBUG || CONFIG_SYSFS */
Alexander Beregalov26c02cf2009-06-11 14:08:48 +04002172
Pekka Enberg781b2ba2009-06-10 18:50:32 +03002173static noinline void
2174slab_out_of_memory(struct kmem_cache *s, gfp_t gfpflags, int nid)
2175{
David Rientjes9a02d692014-06-04 16:06:36 -07002176#ifdef CONFIG_SLUB_DEBUG
2177 static DEFINE_RATELIMIT_STATE(slub_oom_rs, DEFAULT_RATELIMIT_INTERVAL,
2178 DEFAULT_RATELIMIT_BURST);
Pekka Enberg781b2ba2009-06-10 18:50:32 +03002179 int node;
Christoph Lameterfa45dc22014-08-06 16:04:09 -07002180 struct kmem_cache_node *n;
Pekka Enberg781b2ba2009-06-10 18:50:32 +03002181
David Rientjes9a02d692014-06-04 16:06:36 -07002182 if ((gfpflags & __GFP_NOWARN) || !__ratelimit(&slub_oom_rs))
2183 return;
2184
Fabian Frederickf9f58282014-06-04 16:06:34 -07002185 pr_warn("SLUB: Unable to allocate memory on node %d (gfp=0x%x)\n",
Pekka Enberg781b2ba2009-06-10 18:50:32 +03002186 nid, gfpflags);
Fabian Frederickf9f58282014-06-04 16:06:34 -07002187 pr_warn(" cache: %s, object size: %d, buffer size: %d, default order: %d, min order: %d\n",
2188 s->name, s->object_size, s->size, oo_order(s->oo),
2189 oo_order(s->min));
Pekka Enberg781b2ba2009-06-10 18:50:32 +03002190
Christoph Lameter3b0efdf2012-06-13 10:24:57 -05002191 if (oo_order(s->min) > get_order(s->object_size))
Fabian Frederickf9f58282014-06-04 16:06:34 -07002192 pr_warn(" %s debugging increased min order, use slub_debug=O to disable.\n",
2193 s->name);
David Rientjesfa5ec8a2009-07-07 00:14:14 -07002194
Christoph Lameterfa45dc22014-08-06 16:04:09 -07002195 for_each_kmem_cache_node(s, node, n) {
Pekka Enberg781b2ba2009-06-10 18:50:32 +03002196 unsigned long nr_slabs;
2197 unsigned long nr_objs;
2198 unsigned long nr_free;
2199
Alexander Beregalov26c02cf2009-06-11 14:08:48 +04002200 nr_free = count_partial(n, count_free);
2201 nr_slabs = node_nr_slabs(n);
2202 nr_objs = node_nr_objs(n);
Pekka Enberg781b2ba2009-06-10 18:50:32 +03002203
Fabian Frederickf9f58282014-06-04 16:06:34 -07002204 pr_warn(" node %d: slabs: %ld, objs: %ld, free: %ld\n",
Pekka Enberg781b2ba2009-06-10 18:50:32 +03002205 node, nr_slabs, nr_objs, nr_free);
2206 }
David Rientjes9a02d692014-06-04 16:06:36 -07002207#endif
Pekka Enberg781b2ba2009-06-10 18:50:32 +03002208}
2209
Christoph Lameter497b66f2011-08-09 16:12:26 -05002210static inline void *new_slab_objects(struct kmem_cache *s, gfp_t flags,
2211 int node, struct kmem_cache_cpu **pc)
2212{
Christoph Lameter6faa6832012-05-09 10:09:51 -05002213 void *freelist;
Christoph Lameter188fd062012-05-09 10:09:55 -05002214 struct kmem_cache_cpu *c = *pc;
2215 struct page *page;
Christoph Lameter497b66f2011-08-09 16:12:26 -05002216
Christoph Lameter188fd062012-05-09 10:09:55 -05002217 freelist = get_partial(s, flags, node, c);
2218
2219 if (freelist)
2220 return freelist;
2221
2222 page = new_slab(s, flags, node);
Christoph Lameter497b66f2011-08-09 16:12:26 -05002223 if (page) {
Christoph Lameter7c8e0182014-06-04 16:07:56 -07002224 c = raw_cpu_ptr(s->cpu_slab);
Christoph Lameter497b66f2011-08-09 16:12:26 -05002225 if (c->page)
2226 flush_slab(s, c);
2227
2228 /*
2229 * No other reference to the page yet so we can
2230 * muck around with it freely without cmpxchg
2231 */
Christoph Lameter6faa6832012-05-09 10:09:51 -05002232 freelist = page->freelist;
Christoph Lameter497b66f2011-08-09 16:12:26 -05002233 page->freelist = NULL;
2234
2235 stat(s, ALLOC_SLAB);
Christoph Lameter497b66f2011-08-09 16:12:26 -05002236 c->page = page;
2237 *pc = c;
2238 } else
Christoph Lameter6faa6832012-05-09 10:09:51 -05002239 freelist = NULL;
Christoph Lameter497b66f2011-08-09 16:12:26 -05002240
Christoph Lameter6faa6832012-05-09 10:09:51 -05002241 return freelist;
Christoph Lameter497b66f2011-08-09 16:12:26 -05002242}
2243
Mel Gorman072bb0a2012-07-31 16:43:58 -07002244static inline bool pfmemalloc_match(struct page *page, gfp_t gfpflags)
2245{
2246 if (unlikely(PageSlabPfmemalloc(page)))
2247 return gfp_pfmemalloc_allowed(gfpflags);
2248
2249 return true;
2250}
2251
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002252/*
Chen Gangd0e0ac92013-07-15 09:05:29 +08002253 * Check the page->freelist of a page and either transfer the freelist to the
2254 * per cpu freelist or deactivate the page.
Christoph Lameter213eeb92011-11-11 14:07:14 -06002255 *
2256 * The page is still frozen if the return value is not NULL.
2257 *
2258 * If this function returns NULL then the page has been unfrozen.
Joonsoo Kimd24ac772012-05-18 22:01:17 +09002259 *
2260 * This function must be called with interrupt disabled.
Christoph Lameter213eeb92011-11-11 14:07:14 -06002261 */
2262static inline void *get_freelist(struct kmem_cache *s, struct page *page)
2263{
2264 struct page new;
2265 unsigned long counters;
2266 void *freelist;
2267
2268 do {
2269 freelist = page->freelist;
2270 counters = page->counters;
Christoph Lameter6faa6832012-05-09 10:09:51 -05002271
Christoph Lameter213eeb92011-11-11 14:07:14 -06002272 new.counters = counters;
Dave Hansena0132ac2014-01-29 14:05:50 -08002273 VM_BUG_ON(!new.frozen);
Christoph Lameter213eeb92011-11-11 14:07:14 -06002274
2275 new.inuse = page->objects;
2276 new.frozen = freelist != NULL;
2277
Joonsoo Kimd24ac772012-05-18 22:01:17 +09002278 } while (!__cmpxchg_double_slab(s, page,
Christoph Lameter213eeb92011-11-11 14:07:14 -06002279 freelist, counters,
2280 NULL, new.counters,
2281 "get_freelist"));
2282
2283 return freelist;
2284}
2285
2286/*
Christoph Lameter894b8782007-05-10 03:15:16 -07002287 * Slow path. The lockless freelist is empty or we need to perform
2288 * debugging duties.
Christoph Lameter81819f02007-05-06 14:49:36 -07002289 *
Christoph Lameter894b8782007-05-10 03:15:16 -07002290 * Processing is still very fast if new objects have been freed to the
2291 * regular freelist. In that case we simply take over the regular freelist
2292 * as the lockless freelist and zap the regular freelist.
Christoph Lameter81819f02007-05-06 14:49:36 -07002293 *
Christoph Lameter894b8782007-05-10 03:15:16 -07002294 * If that is not working then we fall back to the partial lists. We take the
2295 * first element of the freelist as the object to allocate now and move the
2296 * rest of the freelist to the lockless freelist.
2297 *
2298 * And if we were unable to get a new slab from the partial slab lists then
Christoph Lameter6446faa2008-02-15 23:45:26 -08002299 * we need to allocate a new slab. This is the slowest path since it involves
2300 * a call to the page allocator and the setup of a new slab.
Christoph Lameter81819f02007-05-06 14:49:36 -07002301 */
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03002302static void *__slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node,
2303 unsigned long addr, struct kmem_cache_cpu *c)
Christoph Lameter81819f02007-05-06 14:49:36 -07002304{
Christoph Lameter6faa6832012-05-09 10:09:51 -05002305 void *freelist;
Christoph Lameterf6e7def2012-05-09 10:09:58 -05002306 struct page *page;
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002307 unsigned long flags;
2308
2309 local_irq_save(flags);
2310#ifdef CONFIG_PREEMPT
2311 /*
2312 * We may have been preempted and rescheduled on a different
2313 * cpu before disabling interrupts. Need to reload cpu area
2314 * pointer.
2315 */
2316 c = this_cpu_ptr(s->cpu_slab);
2317#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07002318
Christoph Lameterf6e7def2012-05-09 10:09:58 -05002319 page = c->page;
2320 if (!page)
Christoph Lameter81819f02007-05-06 14:49:36 -07002321 goto new_slab;
Christoph Lameter49e22582011-08-09 16:12:27 -05002322redo:
Christoph Lameter6faa6832012-05-09 10:09:51 -05002323
Christoph Lameter57d437d2012-05-09 10:09:59 -05002324 if (unlikely(!node_match(page, node))) {
Joonsoo Kima561ce02014-10-09 15:26:15 -07002325 int searchnode = node;
2326
2327 if (node != NUMA_NO_NODE && !node_present_pages(node))
2328 searchnode = node_to_mem_node(node);
2329
2330 if (unlikely(!node_match(page, searchnode))) {
2331 stat(s, ALLOC_NODE_MISMATCH);
2332 deactivate_slab(s, page, c->freelist);
2333 c->page = NULL;
2334 c->freelist = NULL;
2335 goto new_slab;
2336 }
Christoph Lameterfc59c052011-06-01 12:25:56 -05002337 }
Christoph Lameter6446faa2008-02-15 23:45:26 -08002338
Mel Gorman072bb0a2012-07-31 16:43:58 -07002339 /*
2340 * By rights, we should be searching for a slab page that was
2341 * PFMEMALLOC but right now, we are losing the pfmemalloc
2342 * information when the page leaves the per-cpu allocator
2343 */
2344 if (unlikely(!pfmemalloc_match(page, gfpflags))) {
2345 deactivate_slab(s, page, c->freelist);
2346 c->page = NULL;
2347 c->freelist = NULL;
2348 goto new_slab;
2349 }
2350
Eric Dumazet73736e02011-12-13 04:57:06 +01002351 /* must check again c->freelist in case of cpu migration or IRQ */
Christoph Lameter6faa6832012-05-09 10:09:51 -05002352 freelist = c->freelist;
2353 if (freelist)
Eric Dumazet73736e02011-12-13 04:57:06 +01002354 goto load_freelist;
2355
Christoph Lameterf6e7def2012-05-09 10:09:58 -05002356 freelist = get_freelist(s, page);
Christoph Lameter6446faa2008-02-15 23:45:26 -08002357
Christoph Lameter6faa6832012-05-09 10:09:51 -05002358 if (!freelist) {
Christoph Lameter03e404a2011-06-01 12:25:58 -05002359 c->page = NULL;
2360 stat(s, DEACTIVATE_BYPASS);
Christoph Lameterfc59c052011-06-01 12:25:56 -05002361 goto new_slab;
Christoph Lameter03e404a2011-06-01 12:25:58 -05002362 }
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002363
Christoph Lameter81819f02007-05-06 14:49:36 -07002364 stat(s, ALLOC_REFILL);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08002365
Christoph Lameter894b8782007-05-10 03:15:16 -07002366load_freelist:
Christoph Lameter507effe2012-05-09 10:09:52 -05002367 /*
2368 * freelist is pointing to the list of objects to be used.
2369 * page is pointing to the page from which the objects are obtained.
2370 * That page must be frozen for per cpu allocations to work.
2371 */
Dave Hansena0132ac2014-01-29 14:05:50 -08002372 VM_BUG_ON(!c->page->frozen);
Christoph Lameter6faa6832012-05-09 10:09:51 -05002373 c->freelist = get_freepointer(s, freelist);
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002374 c->tid = next_tid(c->tid);
2375 local_irq_restore(flags);
Christoph Lameter6faa6832012-05-09 10:09:51 -05002376 return freelist;
Christoph Lameter81819f02007-05-06 14:49:36 -07002377
Christoph Lameter81819f02007-05-06 14:49:36 -07002378new_slab:
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002379
Christoph Lameter49e22582011-08-09 16:12:27 -05002380 if (c->partial) {
Christoph Lameterf6e7def2012-05-09 10:09:58 -05002381 page = c->page = c->partial;
2382 c->partial = page->next;
Christoph Lameter49e22582011-08-09 16:12:27 -05002383 stat(s, CPU_PARTIAL_ALLOC);
2384 c->freelist = NULL;
2385 goto redo;
Christoph Lameter81819f02007-05-06 14:49:36 -07002386 }
2387
Christoph Lameter188fd062012-05-09 10:09:55 -05002388 freelist = new_slab_objects(s, gfpflags, node, &c);
Christoph Lameterb811c202007-10-16 23:25:51 -07002389
Christoph Lameterf46974362012-05-09 10:09:54 -05002390 if (unlikely(!freelist)) {
David Rientjes9a02d692014-06-04 16:06:36 -07002391 slab_out_of_memory(s, gfpflags, node);
Christoph Lameterf46974362012-05-09 10:09:54 -05002392 local_irq_restore(flags);
2393 return NULL;
Christoph Lameter81819f02007-05-06 14:49:36 -07002394 }
Christoph Lameter894b8782007-05-10 03:15:16 -07002395
Christoph Lameterf6e7def2012-05-09 10:09:58 -05002396 page = c->page;
Christoph Lameter5091b742012-07-31 16:44:00 -07002397 if (likely(!kmem_cache_debug(s) && pfmemalloc_match(page, gfpflags)))
Christoph Lameter81819f02007-05-06 14:49:36 -07002398 goto load_freelist;
Christoph Lameter894b8782007-05-10 03:15:16 -07002399
Christoph Lameter497b66f2011-08-09 16:12:26 -05002400 /* Only entered in the debug case */
Chen Gangd0e0ac92013-07-15 09:05:29 +08002401 if (kmem_cache_debug(s) &&
2402 !alloc_debug_processing(s, page, freelist, addr))
Christoph Lameter497b66f2011-08-09 16:12:26 -05002403 goto new_slab; /* Slab failed checks. Next slab needed */
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002404
Christoph Lameterf6e7def2012-05-09 10:09:58 -05002405 deactivate_slab(s, page, get_freepointer(s, freelist));
Christoph Lameterc17dda42012-05-09 10:09:57 -05002406 c->page = NULL;
2407 c->freelist = NULL;
Christoph Lametera71ae472011-05-25 09:47:43 -05002408 local_irq_restore(flags);
Christoph Lameter6faa6832012-05-09 10:09:51 -05002409 return freelist;
Christoph Lameter894b8782007-05-10 03:15:16 -07002410}
2411
2412/*
2413 * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc)
2414 * have the fastpath folded into their functions. So no function call
2415 * overhead for requests that can be satisfied on the fastpath.
2416 *
2417 * The fastpath works by first checking if the lockless freelist can be used.
2418 * If not then __slab_alloc is called for slow processing.
2419 *
2420 * Otherwise we can simply pick the next object from the lockless free list.
2421 */
Ezequiel Garcia2b847c32012-09-08 17:47:58 -03002422static __always_inline void *slab_alloc_node(struct kmem_cache *s,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03002423 gfp_t gfpflags, int node, unsigned long addr)
Christoph Lameter894b8782007-05-10 03:15:16 -07002424{
Christoph Lameter894b8782007-05-10 03:15:16 -07002425 void **object;
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002426 struct kmem_cache_cpu *c;
Christoph Lameter57d437d2012-05-09 10:09:59 -05002427 struct page *page;
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002428 unsigned long tid;
Christoph Lameter1f842602008-01-07 23:20:30 -08002429
Vladimir Davydov8135be52014-12-12 16:56:38 -08002430 s = slab_pre_alloc_hook(s, gfpflags);
2431 if (!s)
Akinobu Mita773ff602008-12-23 19:37:01 +09002432 return NULL;
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002433redo:
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002434 /*
2435 * Must read kmem_cache cpu data via this cpu ptr. Preemption is
2436 * enabled. We may switch back and forth between cpus while
2437 * reading from one cpu area. That does not matter as long
2438 * as we end up on the original cpu again when doing the cmpxchg.
Christoph Lameter7cccd802013-01-23 21:45:48 +00002439 *
Joonsoo Kim9aabf812015-02-10 14:09:32 -08002440 * We should guarantee that tid and kmem_cache are retrieved on
2441 * the same cpu. It could be different if CONFIG_PREEMPT so we need
2442 * to check if it is matched or not.
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002443 */
Joonsoo Kim9aabf812015-02-10 14:09:32 -08002444 do {
2445 tid = this_cpu_read(s->cpu_slab->tid);
2446 c = raw_cpu_ptr(s->cpu_slab);
Mark Rutland859b7a02015-03-25 15:55:23 -07002447 } while (IS_ENABLED(CONFIG_PREEMPT) &&
2448 unlikely(tid != READ_ONCE(c->tid)));
Joonsoo Kim9aabf812015-02-10 14:09:32 -08002449
2450 /*
2451 * Irqless object alloc/free algorithm used here depends on sequence
2452 * of fetching cpu_slab's data. tid should be fetched before anything
2453 * on c to guarantee that object and page associated with previous tid
2454 * won't be used with current tid. If we fetch tid first, object and
2455 * page could be one associated with next tid and our alloc/free
2456 * request will be failed. In this case, we will retry. So, no problem.
2457 */
2458 barrier();
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002459
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002460 /*
2461 * The transaction ids are globally unique per cpu and per operation on
2462 * a per cpu queue. Thus they can be guarantee that the cmpxchg_double
2463 * occurs on the right processor and that there was no operation on the
2464 * linked list in between.
2465 */
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002466
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06002467 object = c->freelist;
Christoph Lameter57d437d2012-05-09 10:09:59 -05002468 page = c->page;
Dave Hansen8eae1492014-06-04 16:06:37 -07002469 if (unlikely(!object || !node_match(page, node))) {
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002470 object = __slab_alloc(s, gfpflags, node, addr, c);
Dave Hansen8eae1492014-06-04 16:06:37 -07002471 stat(s, ALLOC_SLOWPATH);
2472 } else {
Eric Dumazet0ad95002011-12-16 16:25:34 +01002473 void *next_object = get_freepointer_safe(s, object);
2474
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002475 /*
Lucas De Marchi25985ed2011-03-30 22:57:33 -03002476 * The cmpxchg will only match if there was no additional
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002477 * operation and if we are on the right processor.
2478 *
Chen Gangd0e0ac92013-07-15 09:05:29 +08002479 * The cmpxchg does the following atomically (without lock
2480 * semantics!)
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002481 * 1. Relocate first pointer to the current per cpu area.
2482 * 2. Verify that tid and freelist have not been changed
2483 * 3. If they were not changed replace tid and freelist
2484 *
Chen Gangd0e0ac92013-07-15 09:05:29 +08002485 * Since this is without lock semantics the protection is only
2486 * against code executing on this cpu *not* from access by
2487 * other cpus.
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002488 */
Christoph Lameter933393f2011-12-22 11:58:51 -06002489 if (unlikely(!this_cpu_cmpxchg_double(
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002490 s->cpu_slab->freelist, s->cpu_slab->tid,
2491 object, tid,
Eric Dumazet0ad95002011-12-16 16:25:34 +01002492 next_object, next_tid(tid)))) {
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002493
2494 note_cmpxchg_failure("slab_alloc", s, tid);
2495 goto redo;
2496 }
Eric Dumazet0ad95002011-12-16 16:25:34 +01002497 prefetch_freepointer(s, next_object);
Christoph Lameter84e554e62009-12-18 16:26:23 -06002498 stat(s, ALLOC_FASTPATH);
Christoph Lameter894b8782007-05-10 03:15:16 -07002499 }
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002500
Pekka Enberg74e21342009-11-25 20:14:48 +02002501 if (unlikely(gfpflags & __GFP_ZERO) && object)
Christoph Lameter3b0efdf2012-06-13 10:24:57 -05002502 memset(object, 0, s->object_size);
Christoph Lameterd07dbea2007-07-17 04:03:23 -07002503
Christoph Lameterc016b0b2010-08-20 12:37:16 -05002504 slab_post_alloc_hook(s, gfpflags, object);
Vegard Nossum5a896d92008-04-04 00:54:48 +02002505
Christoph Lameter894b8782007-05-10 03:15:16 -07002506 return object;
Christoph Lameter81819f02007-05-06 14:49:36 -07002507}
2508
Ezequiel Garcia2b847c32012-09-08 17:47:58 -03002509static __always_inline void *slab_alloc(struct kmem_cache *s,
2510 gfp_t gfpflags, unsigned long addr)
2511{
2512 return slab_alloc_node(s, gfpflags, NUMA_NO_NODE, addr);
2513}
2514
Christoph Lameter81819f02007-05-06 14:49:36 -07002515void *kmem_cache_alloc(struct kmem_cache *s, gfp_t gfpflags)
2516{
Ezequiel Garcia2b847c32012-09-08 17:47:58 -03002517 void *ret = slab_alloc(s, gfpflags, _RET_IP_);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002518
Chen Gangd0e0ac92013-07-15 09:05:29 +08002519 trace_kmem_cache_alloc(_RET_IP_, ret, s->object_size,
2520 s->size, gfpflags);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002521
2522 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07002523}
2524EXPORT_SYMBOL(kmem_cache_alloc);
2525
Li Zefan0f24f122009-12-11 15:45:30 +08002526#ifdef CONFIG_TRACING
Richard Kennedy4a923792010-10-21 10:29:19 +01002527void *kmem_cache_alloc_trace(struct kmem_cache *s, gfp_t gfpflags, size_t size)
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002528{
Ezequiel Garcia2b847c32012-09-08 17:47:58 -03002529 void *ret = slab_alloc(s, gfpflags, _RET_IP_);
Richard Kennedy4a923792010-10-21 10:29:19 +01002530 trace_kmalloc(_RET_IP_, ret, size, s->size, gfpflags);
Andrey Ryabinin0316bec2015-02-13 14:39:42 -08002531 kasan_kmalloc(s, ret, size);
Richard Kennedy4a923792010-10-21 10:29:19 +01002532 return ret;
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002533}
Richard Kennedy4a923792010-10-21 10:29:19 +01002534EXPORT_SYMBOL(kmem_cache_alloc_trace);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002535#endif
2536
Christoph Lameter81819f02007-05-06 14:49:36 -07002537#ifdef CONFIG_NUMA
2538void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t gfpflags, int node)
2539{
Ezequiel Garcia2b847c32012-09-08 17:47:58 -03002540 void *ret = slab_alloc_node(s, gfpflags, node, _RET_IP_);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002541
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02002542 trace_kmem_cache_alloc_node(_RET_IP_, ret,
Christoph Lameter3b0efdf2012-06-13 10:24:57 -05002543 s->object_size, s->size, gfpflags, node);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002544
2545 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07002546}
2547EXPORT_SYMBOL(kmem_cache_alloc_node);
Christoph Lameter81819f02007-05-06 14:49:36 -07002548
Li Zefan0f24f122009-12-11 15:45:30 +08002549#ifdef CONFIG_TRACING
Richard Kennedy4a923792010-10-21 10:29:19 +01002550void *kmem_cache_alloc_node_trace(struct kmem_cache *s,
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002551 gfp_t gfpflags,
Richard Kennedy4a923792010-10-21 10:29:19 +01002552 int node, size_t size)
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002553{
Ezequiel Garcia2b847c32012-09-08 17:47:58 -03002554 void *ret = slab_alloc_node(s, gfpflags, node, _RET_IP_);
Richard Kennedy4a923792010-10-21 10:29:19 +01002555
2556 trace_kmalloc_node(_RET_IP_, ret,
2557 size, s->size, gfpflags, node);
Andrey Ryabinin0316bec2015-02-13 14:39:42 -08002558
2559 kasan_kmalloc(s, ret, size);
Richard Kennedy4a923792010-10-21 10:29:19 +01002560 return ret;
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002561}
Richard Kennedy4a923792010-10-21 10:29:19 +01002562EXPORT_SYMBOL(kmem_cache_alloc_node_trace);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002563#endif
Namhyung Kim5d1f57e2010-09-29 21:02:15 +09002564#endif
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002565
Christoph Lameter81819f02007-05-06 14:49:36 -07002566/*
Kim Phillips94e4d712015-02-10 14:09:37 -08002567 * Slow path handling. This may still be called frequently since objects
Christoph Lameter894b8782007-05-10 03:15:16 -07002568 * have a longer lifetime than the cpu slabs in most processing loads.
Christoph Lameter81819f02007-05-06 14:49:36 -07002569 *
Christoph Lameter894b8782007-05-10 03:15:16 -07002570 * So we still attempt to reduce cache line usage. Just take the slab
2571 * lock and free the item. If there is no additional partial page
2572 * handling required then we can return immediately.
Christoph Lameter81819f02007-05-06 14:49:36 -07002573 */
Christoph Lameter894b8782007-05-10 03:15:16 -07002574static void __slab_free(struct kmem_cache *s, struct page *page,
Christoph Lameterff120592009-12-18 16:26:22 -06002575 void *x, unsigned long addr)
Christoph Lameter81819f02007-05-06 14:49:36 -07002576{
2577 void *prior;
2578 void **object = (void *)x;
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002579 int was_frozen;
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002580 struct page new;
2581 unsigned long counters;
2582 struct kmem_cache_node *n = NULL;
Christoph Lameter61728d12011-06-01 12:25:51 -05002583 unsigned long uninitialized_var(flags);
Christoph Lameter81819f02007-05-06 14:49:36 -07002584
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002585 stat(s, FREE_SLOWPATH);
Christoph Lameter81819f02007-05-06 14:49:36 -07002586
Christoph Lameter19c7ff92012-05-30 12:54:46 -05002587 if (kmem_cache_debug(s) &&
2588 !(n = free_debug_processing(s, page, x, addr, &flags)))
Christoph Lameter80f08c12011-06-01 12:25:55 -05002589 return;
Christoph Lameter6446faa2008-02-15 23:45:26 -08002590
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002591 do {
Joonsoo Kim837d6782012-08-16 00:02:40 +09002592 if (unlikely(n)) {
2593 spin_unlock_irqrestore(&n->list_lock, flags);
2594 n = NULL;
2595 }
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002596 prior = page->freelist;
2597 counters = page->counters;
2598 set_freepointer(s, object, prior);
2599 new.counters = counters;
2600 was_frozen = new.frozen;
2601 new.inuse--;
Joonsoo Kim837d6782012-08-16 00:02:40 +09002602 if ((!new.inuse || !prior) && !was_frozen) {
Christoph Lameter49e22582011-08-09 16:12:27 -05002603
Peter Zijlstrac65c1872014-01-10 13:23:49 +01002604 if (kmem_cache_has_cpu_partial(s) && !prior) {
Christoph Lameter49e22582011-08-09 16:12:27 -05002605
2606 /*
Chen Gangd0e0ac92013-07-15 09:05:29 +08002607 * Slab was on no list before and will be
2608 * partially empty
2609 * We can defer the list move and instead
2610 * freeze it.
Christoph Lameter49e22582011-08-09 16:12:27 -05002611 */
2612 new.frozen = 1;
2613
Peter Zijlstrac65c1872014-01-10 13:23:49 +01002614 } else { /* Needs to be taken off a list */
Christoph Lameter49e22582011-08-09 16:12:27 -05002615
LQYMGTb455def2014-12-10 15:42:13 -08002616 n = get_node(s, page_to_nid(page));
Christoph Lameter49e22582011-08-09 16:12:27 -05002617 /*
2618 * Speculatively acquire the list_lock.
2619 * If the cmpxchg does not succeed then we may
2620 * drop the list_lock without any processing.
2621 *
2622 * Otherwise the list_lock will synchronize with
2623 * other processors updating the list of slabs.
2624 */
2625 spin_lock_irqsave(&n->list_lock, flags);
2626
2627 }
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002628 }
Christoph Lameter81819f02007-05-06 14:49:36 -07002629
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002630 } while (!cmpxchg_double_slab(s, page,
2631 prior, counters,
2632 object, new.counters,
2633 "__slab_free"));
Christoph Lameter81819f02007-05-06 14:49:36 -07002634
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002635 if (likely(!n)) {
Christoph Lameter49e22582011-08-09 16:12:27 -05002636
2637 /*
2638 * If we just froze the page then put it onto the
2639 * per cpu partial list.
2640 */
Alex Shi8028dce2012-02-03 23:34:56 +08002641 if (new.frozen && !was_frozen) {
Christoph Lameter49e22582011-08-09 16:12:27 -05002642 put_cpu_partial(s, page, 1);
Alex Shi8028dce2012-02-03 23:34:56 +08002643 stat(s, CPU_PARTIAL_FREE);
2644 }
Christoph Lameter49e22582011-08-09 16:12:27 -05002645 /*
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002646 * The list lock was not taken therefore no list
2647 * activity can be necessary.
2648 */
LQYMGTb455def2014-12-10 15:42:13 -08002649 if (was_frozen)
2650 stat(s, FREE_FROZEN);
2651 return;
2652 }
Christoph Lameter81819f02007-05-06 14:49:36 -07002653
Joonsoo Kim8a5b20a2014-07-02 15:22:35 -07002654 if (unlikely(!new.inuse && n->nr_partial >= s->min_partial))
Joonsoo Kim837d6782012-08-16 00:02:40 +09002655 goto slab_empty;
Christoph Lameter81819f02007-05-06 14:49:36 -07002656
Joonsoo Kim837d6782012-08-16 00:02:40 +09002657 /*
2658 * Objects left in the slab. If it was not on the partial list before
2659 * then add it.
2660 */
Joonsoo Kim345c9052013-06-19 14:05:52 +09002661 if (!kmem_cache_has_cpu_partial(s) && unlikely(!prior)) {
2662 if (kmem_cache_debug(s))
Peter Zijlstrac65c1872014-01-10 13:23:49 +01002663 remove_full(s, n, page);
Joonsoo Kim837d6782012-08-16 00:02:40 +09002664 add_partial(n, page, DEACTIVATE_TO_TAIL);
2665 stat(s, FREE_ADD_PARTIAL);
Christoph Lameter81819f02007-05-06 14:49:36 -07002666 }
Christoph Lameter80f08c12011-06-01 12:25:55 -05002667 spin_unlock_irqrestore(&n->list_lock, flags);
Christoph Lameter81819f02007-05-06 14:49:36 -07002668 return;
2669
2670slab_empty:
Christoph Lametera973e9d2008-03-01 13:40:44 -08002671 if (prior) {
Christoph Lameter81819f02007-05-06 14:49:36 -07002672 /*
Christoph Lameter6fbabb22011-08-08 11:16:56 -05002673 * Slab on the partial list.
Christoph Lameter81819f02007-05-06 14:49:36 -07002674 */
Christoph Lameter5cc6eee2011-06-01 12:25:50 -05002675 remove_partial(n, page);
Christoph Lameter84e554e62009-12-18 16:26:23 -06002676 stat(s, FREE_REMOVE_PARTIAL);
Peter Zijlstrac65c1872014-01-10 13:23:49 +01002677 } else {
Christoph Lameter6fbabb22011-08-08 11:16:56 -05002678 /* Slab must be on the full list */
Peter Zijlstrac65c1872014-01-10 13:23:49 +01002679 remove_full(s, n, page);
2680 }
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002681
Christoph Lameter80f08c12011-06-01 12:25:55 -05002682 spin_unlock_irqrestore(&n->list_lock, flags);
Christoph Lameter84e554e62009-12-18 16:26:23 -06002683 stat(s, FREE_SLAB);
Christoph Lameter81819f02007-05-06 14:49:36 -07002684 discard_slab(s, page);
Christoph Lameter81819f02007-05-06 14:49:36 -07002685}
2686
Christoph Lameter894b8782007-05-10 03:15:16 -07002687/*
2688 * Fastpath with forced inlining to produce a kfree and kmem_cache_free that
2689 * can perform fastpath freeing without additional function calls.
2690 *
2691 * The fastpath is only possible if we are freeing to the current cpu slab
2692 * of this processor. This typically the case if we have just allocated
2693 * the item before.
2694 *
2695 * If fastpath is not possible then fall back to __slab_free where we deal
2696 * with all sorts of special processing.
2697 */
Pekka Enberg06428782008-01-07 23:20:27 -08002698static __always_inline void slab_free(struct kmem_cache *s,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03002699 struct page *page, void *x, unsigned long addr)
Christoph Lameter894b8782007-05-10 03:15:16 -07002700{
2701 void **object = (void *)x;
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002702 struct kmem_cache_cpu *c;
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002703 unsigned long tid;
Christoph Lameter1f842602008-01-07 23:20:30 -08002704
Christoph Lameterc016b0b2010-08-20 12:37:16 -05002705 slab_free_hook(s, x);
2706
Christoph Lametera24c5a02011-03-15 12:45:21 -05002707redo:
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002708 /*
2709 * Determine the currently cpus per cpu slab.
2710 * The cpu may change afterward. However that does not matter since
2711 * data is retrieved via this pointer. If we are on the same cpu
Jesper Dangaard Brouer2ae44002015-09-04 15:45:31 -07002712 * during the cmpxchg then the free will succeed.
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002713 */
Joonsoo Kim9aabf812015-02-10 14:09:32 -08002714 do {
2715 tid = this_cpu_read(s->cpu_slab->tid);
2716 c = raw_cpu_ptr(s->cpu_slab);
Mark Rutland859b7a02015-03-25 15:55:23 -07002717 } while (IS_ENABLED(CONFIG_PREEMPT) &&
2718 unlikely(tid != READ_ONCE(c->tid)));
Christoph Lameterc016b0b2010-08-20 12:37:16 -05002719
Joonsoo Kim9aabf812015-02-10 14:09:32 -08002720 /* Same with comment on barrier() in slab_alloc_node() */
2721 barrier();
Christoph Lameterc016b0b2010-08-20 12:37:16 -05002722
Christoph Lameter442b06b2011-05-17 16:29:31 -05002723 if (likely(page == c->page)) {
Christoph Lameterff120592009-12-18 16:26:22 -06002724 set_freepointer(s, object, c->freelist);
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002725
Christoph Lameter933393f2011-12-22 11:58:51 -06002726 if (unlikely(!this_cpu_cmpxchg_double(
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002727 s->cpu_slab->freelist, s->cpu_slab->tid,
2728 c->freelist, tid,
2729 object, next_tid(tid)))) {
2730
2731 note_cmpxchg_failure("slab_free", s, tid);
2732 goto redo;
2733 }
Christoph Lameter84e554e62009-12-18 16:26:23 -06002734 stat(s, FREE_FASTPATH);
Christoph Lameter894b8782007-05-10 03:15:16 -07002735 } else
Christoph Lameterff120592009-12-18 16:26:22 -06002736 __slab_free(s, page, x, addr);
Christoph Lameter894b8782007-05-10 03:15:16 -07002737
Christoph Lameter894b8782007-05-10 03:15:16 -07002738}
2739
Christoph Lameter81819f02007-05-06 14:49:36 -07002740void kmem_cache_free(struct kmem_cache *s, void *x)
2741{
Glauber Costab9ce5ef2012-12-18 14:22:46 -08002742 s = cache_from_obj(s, x);
2743 if (!s)
Christoph Lameter79576102012-09-04 23:06:14 +00002744 return;
Glauber Costab9ce5ef2012-12-18 14:22:46 -08002745 slab_free(s, virt_to_head_page(x), x, _RET_IP_);
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02002746 trace_kmem_cache_free(_RET_IP_, x);
Christoph Lameter81819f02007-05-06 14:49:36 -07002747}
2748EXPORT_SYMBOL(kmem_cache_free);
2749
Jesper Dangaard Brouer994eb762015-09-04 15:45:37 -07002750/* Note that interrupts must be enabled when calling this function. */
Christoph Lameter484748f2015-09-04 15:45:34 -07002751void kmem_cache_free_bulk(struct kmem_cache *s, size_t size, void **p)
2752{
Jesper Dangaard Brouerfbd02632015-09-04 15:45:43 -07002753 struct kmem_cache_cpu *c;
2754 struct page *page;
2755 int i;
2756
Jesper Dangaard Brouerfbd02632015-09-04 15:45:43 -07002757 local_irq_disable();
2758 c = this_cpu_ptr(s->cpu_slab);
2759
2760 for (i = 0; i < size; i++) {
2761 void *object = p[i];
2762
2763 BUG_ON(!object);
Jesper Dangaard Brouer3eed0342015-09-04 15:45:45 -07002764 /* kmem cache debug support */
2765 s = cache_from_obj(s, object);
2766 if (unlikely(!s))
2767 goto exit;
2768 slab_free_hook(s, object);
2769
Jesper Dangaard Brouerfbd02632015-09-04 15:45:43 -07002770 page = virt_to_head_page(object);
Jesper Dangaard Brouerfbd02632015-09-04 15:45:43 -07002771
2772 if (c->page == page) {
2773 /* Fastpath: local CPU free */
2774 set_freepointer(s, object, c->freelist);
2775 c->freelist = object;
2776 } else {
2777 c->tid = next_tid(c->tid);
2778 local_irq_enable();
2779 /* Slowpath: overhead locked cmpxchg_double_slab */
2780 __slab_free(s, page, object, _RET_IP_);
2781 local_irq_disable();
2782 c = this_cpu_ptr(s->cpu_slab);
2783 }
2784 }
Jesper Dangaard Brouer3eed0342015-09-04 15:45:45 -07002785exit:
Jesper Dangaard Brouerfbd02632015-09-04 15:45:43 -07002786 c->tid = next_tid(c->tid);
2787 local_irq_enable();
Christoph Lameter484748f2015-09-04 15:45:34 -07002788}
2789EXPORT_SYMBOL(kmem_cache_free_bulk);
2790
Jesper Dangaard Brouer994eb762015-09-04 15:45:37 -07002791/* Note that interrupts must be enabled when calling this function. */
Christoph Lameter484748f2015-09-04 15:45:34 -07002792bool kmem_cache_alloc_bulk(struct kmem_cache *s, gfp_t flags, size_t size,
Jesper Dangaard Brouer994eb762015-09-04 15:45:37 -07002793 void **p)
Christoph Lameter484748f2015-09-04 15:45:34 -07002794{
Jesper Dangaard Brouer994eb762015-09-04 15:45:37 -07002795 struct kmem_cache_cpu *c;
2796 int i;
2797
Jesper Dangaard Brouer994eb762015-09-04 15:45:37 -07002798 /*
2799 * Drain objects in the per cpu slab, while disabling local
2800 * IRQs, which protects against PREEMPT and interrupts
2801 * handlers invoking normal fastpath.
2802 */
2803 local_irq_disable();
2804 c = this_cpu_ptr(s->cpu_slab);
2805
2806 for (i = 0; i < size; i++) {
2807 void *object = c->freelist;
2808
Jesper Dangaard Brouerebe909e2015-09-04 15:45:40 -07002809 if (unlikely(!object)) {
2810 local_irq_enable();
2811 /*
2812 * Invoking slow path likely have side-effect
2813 * of re-populating per CPU c->freelist
2814 */
2815 p[i] = __slab_alloc(s, flags, NUMA_NO_NODE,
2816 _RET_IP_, c);
2817 if (unlikely(!p[i])) {
2818 __kmem_cache_free_bulk(s, i, p);
2819 return false;
2820 }
2821 local_irq_disable();
2822 c = this_cpu_ptr(s->cpu_slab);
2823 continue; /* goto for-loop */
2824 }
Jesper Dangaard Brouer994eb762015-09-04 15:45:37 -07002825
Jesper Dangaard Brouer3eed0342015-09-04 15:45:45 -07002826 /* kmem_cache debug support */
2827 s = slab_pre_alloc_hook(s, flags);
2828 if (unlikely(!s)) {
2829 __kmem_cache_free_bulk(s, i, p);
2830 c->tid = next_tid(c->tid);
2831 local_irq_enable();
2832 return false;
2833 }
2834
Jesper Dangaard Brouer994eb762015-09-04 15:45:37 -07002835 c->freelist = get_freepointer(s, object);
2836 p[i] = object;
Jesper Dangaard Brouer3eed0342015-09-04 15:45:45 -07002837
2838 /* kmem_cache debug support */
2839 slab_post_alloc_hook(s, flags, object);
Jesper Dangaard Brouer994eb762015-09-04 15:45:37 -07002840 }
2841 c->tid = next_tid(c->tid);
2842 local_irq_enable();
2843
2844 /* Clear memory outside IRQ disabled fastpath loop */
2845 if (unlikely(flags & __GFP_ZERO)) {
2846 int j;
2847
2848 for (j = 0; j < i; j++)
2849 memset(p[j], 0, s->object_size);
2850 }
2851
Jesper Dangaard Brouer994eb762015-09-04 15:45:37 -07002852 return true;
Christoph Lameter484748f2015-09-04 15:45:34 -07002853}
2854EXPORT_SYMBOL(kmem_cache_alloc_bulk);
2855
2856
Christoph Lameter81819f02007-05-06 14:49:36 -07002857/*
Christoph Lameter672bba32007-05-09 02:32:39 -07002858 * Object placement in a slab is made very easy because we always start at
2859 * offset 0. If we tune the size of the object to the alignment then we can
2860 * get the required alignment by putting one properly sized object after
2861 * another.
Christoph Lameter81819f02007-05-06 14:49:36 -07002862 *
2863 * Notice that the allocation order determines the sizes of the per cpu
2864 * caches. Each processor has always one slab available for allocations.
2865 * Increasing the allocation order reduces the number of times that slabs
Christoph Lameter672bba32007-05-09 02:32:39 -07002866 * must be moved on and off the partial lists and is therefore a factor in
Christoph Lameter81819f02007-05-06 14:49:36 -07002867 * locking overhead.
Christoph Lameter81819f02007-05-06 14:49:36 -07002868 */
2869
2870/*
2871 * Mininum / Maximum order of slab pages. This influences locking overhead
2872 * and slab fragmentation. A higher order reduces the number of partial slabs
2873 * and increases the number of allocations possible without having to
2874 * take the list_lock.
2875 */
2876static int slub_min_order;
Christoph Lameter114e9e82008-04-14 19:11:41 +03002877static int slub_max_order = PAGE_ALLOC_COSTLY_ORDER;
Christoph Lameter9b2cd502008-04-14 19:11:41 +03002878static int slub_min_objects;
Christoph Lameter81819f02007-05-06 14:49:36 -07002879
2880/*
Christoph Lameter81819f02007-05-06 14:49:36 -07002881 * Calculate the order of allocation given an slab object size.
2882 *
Christoph Lameter672bba32007-05-09 02:32:39 -07002883 * The order of allocation has significant impact on performance and other
2884 * system components. Generally order 0 allocations should be preferred since
2885 * order 0 does not cause fragmentation in the page allocator. Larger objects
2886 * be problematic to put into order 0 slabs because there may be too much
Christoph Lameterc124f5b2008-04-14 19:13:29 +03002887 * unused space left. We go to a higher order if more than 1/16th of the slab
Christoph Lameter672bba32007-05-09 02:32:39 -07002888 * would be wasted.
Christoph Lameter81819f02007-05-06 14:49:36 -07002889 *
Christoph Lameter672bba32007-05-09 02:32:39 -07002890 * In order to reach satisfactory performance we must ensure that a minimum
2891 * number of objects is in one slab. Otherwise we may generate too much
2892 * activity on the partial lists which requires taking the list_lock. This is
2893 * less a concern for large slabs though which are rarely used.
Christoph Lameter81819f02007-05-06 14:49:36 -07002894 *
Christoph Lameter672bba32007-05-09 02:32:39 -07002895 * slub_max_order specifies the order where we begin to stop considering the
2896 * number of objects in a slab as critical. If we reach slub_max_order then
2897 * we try to keep the page order as low as possible. So we accept more waste
2898 * of space in favor of a small page order.
2899 *
2900 * Higher order allocations also allow the placement of more objects in a
2901 * slab and thereby reduce object handling overhead. If the user has
2902 * requested a higher mininum order then we start with that one instead of
2903 * the smallest order which will fit the object.
Christoph Lameter81819f02007-05-06 14:49:36 -07002904 */
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002905static inline int slab_order(int size, int min_objects,
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002906 int max_order, int fract_leftover, int reserved)
Christoph Lameter81819f02007-05-06 14:49:36 -07002907{
2908 int order;
2909 int rem;
Christoph Lameter6300ea72007-07-17 04:03:20 -07002910 int min_order = slub_min_order;
Christoph Lameter81819f02007-05-06 14:49:36 -07002911
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002912 if (order_objects(min_order, size, reserved) > MAX_OBJS_PER_PAGE)
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +04002913 return get_order(size * MAX_OBJS_PER_PAGE) - 1;
Christoph Lameter39b26462008-04-14 19:11:30 +03002914
Christoph Lameter6300ea72007-07-17 04:03:20 -07002915 for (order = max(min_order,
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002916 fls(min_objects * size - 1) - PAGE_SHIFT);
2917 order <= max_order; order++) {
2918
Christoph Lameter81819f02007-05-06 14:49:36 -07002919 unsigned long slab_size = PAGE_SIZE << order;
2920
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002921 if (slab_size < min_objects * size + reserved)
Christoph Lameter81819f02007-05-06 14:49:36 -07002922 continue;
2923
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002924 rem = (slab_size - reserved) % size;
Christoph Lameter81819f02007-05-06 14:49:36 -07002925
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002926 if (rem <= slab_size / fract_leftover)
Christoph Lameter81819f02007-05-06 14:49:36 -07002927 break;
2928
2929 }
Christoph Lameter672bba32007-05-09 02:32:39 -07002930
Christoph Lameter81819f02007-05-06 14:49:36 -07002931 return order;
2932}
2933
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002934static inline int calculate_order(int size, int reserved)
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002935{
2936 int order;
2937 int min_objects;
2938 int fraction;
Zhang Yanmine8120ff2009-02-12 18:00:17 +02002939 int max_objects;
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002940
2941 /*
2942 * Attempt to find best configuration for a slab. This
2943 * works by first attempting to generate a layout with
2944 * the best configuration and backing off gradually.
2945 *
2946 * First we reduce the acceptable waste in a slab. Then
2947 * we reduce the minimum objects required in a slab.
2948 */
2949 min_objects = slub_min_objects;
Christoph Lameter9b2cd502008-04-14 19:11:41 +03002950 if (!min_objects)
2951 min_objects = 4 * (fls(nr_cpu_ids) + 1);
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002952 max_objects = order_objects(slub_max_order, size, reserved);
Zhang Yanmine8120ff2009-02-12 18:00:17 +02002953 min_objects = min(min_objects, max_objects);
2954
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002955 while (min_objects > 1) {
Christoph Lameterc124f5b2008-04-14 19:13:29 +03002956 fraction = 16;
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002957 while (fraction >= 4) {
2958 order = slab_order(size, min_objects,
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002959 slub_max_order, fraction, reserved);
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002960 if (order <= slub_max_order)
2961 return order;
2962 fraction /= 2;
2963 }
Amerigo Wang5086c389c2009-08-19 21:44:13 +03002964 min_objects--;
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002965 }
2966
2967 /*
2968 * We were unable to place multiple objects in a slab. Now
2969 * lets see if we can place a single object there.
2970 */
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002971 order = slab_order(size, 1, slub_max_order, 1, reserved);
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002972 if (order <= slub_max_order)
2973 return order;
2974
2975 /*
2976 * Doh this slab cannot be placed using slub_max_order.
2977 */
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08002978 order = slab_order(size, 1, MAX_ORDER, 1, reserved);
David Rientjes818cf592009-04-23 09:58:22 +03002979 if (order < MAX_ORDER)
Christoph Lameter5e6d4442007-05-09 02:32:46 -07002980 return order;
2981 return -ENOSYS;
2982}
2983
Pekka Enberg5595cff2008-08-05 09:28:47 +03002984static void
Joonsoo Kim40534972012-05-11 00:50:47 +09002985init_kmem_cache_node(struct kmem_cache_node *n)
Christoph Lameter81819f02007-05-06 14:49:36 -07002986{
2987 n->nr_partial = 0;
Christoph Lameter81819f02007-05-06 14:49:36 -07002988 spin_lock_init(&n->list_lock);
2989 INIT_LIST_HEAD(&n->partial);
Christoph Lameter8ab13722007-07-17 04:03:32 -07002990#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter0f389ec2008-04-14 18:53:02 +03002991 atomic_long_set(&n->nr_slabs, 0);
Salman Qazi02b71b72008-09-11 12:25:41 -07002992 atomic_long_set(&n->total_objects, 0);
Christoph Lameter643b1132007-05-06 14:49:42 -07002993 INIT_LIST_HEAD(&n->full);
Christoph Lameter8ab13722007-07-17 04:03:32 -07002994#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07002995}
2996
Christoph Lameter55136592010-08-20 12:37:13 -05002997static inline int alloc_kmem_cache_cpus(struct kmem_cache *s)
Christoph Lameter4c93c3552007-10-16 01:26:08 -07002998{
Christoph Lameter6c182dc2010-08-20 12:37:14 -05002999 BUILD_BUG_ON(PERCPU_DYNAMIC_EARLY_SIZE <
Christoph Lameter95a05b42013-01-10 19:14:19 +00003000 KMALLOC_SHIFT_HIGH * sizeof(struct kmem_cache_cpu));
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06003001
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06003002 /*
Chris Metcalfd4d84fe2011-06-02 10:19:41 -04003003 * Must align to double word boundary for the double cmpxchg
3004 * instructions to work; see __pcpu_double_call_return_bool().
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06003005 */
Chris Metcalfd4d84fe2011-06-02 10:19:41 -04003006 s->cpu_slab = __alloc_percpu(sizeof(struct kmem_cache_cpu),
3007 2 * sizeof(void *));
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06003008
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06003009 if (!s->cpu_slab)
3010 return 0;
3011
3012 init_kmem_cache_cpus(s);
3013
3014 return 1;
Christoph Lameter4c93c3552007-10-16 01:26:08 -07003015}
Christoph Lameter4c93c3552007-10-16 01:26:08 -07003016
Christoph Lameter51df1142010-08-20 12:37:15 -05003017static struct kmem_cache *kmem_cache_node;
3018
Christoph Lameter81819f02007-05-06 14:49:36 -07003019/*
3020 * No kmalloc_node yet so do it by hand. We know that this is the first
3021 * slab on the node for this slabcache. There are no concurrent accesses
3022 * possible.
3023 *
Zhi Yong Wu721ae222013-11-08 20:47:37 +08003024 * Note that this function only works on the kmem_cache_node
3025 * when allocating for the kmem_cache_node. This is used for bootstrapping
Christoph Lameter4c93c3552007-10-16 01:26:08 -07003026 * memory on a fresh node that has no slab structures yet.
Christoph Lameter81819f02007-05-06 14:49:36 -07003027 */
Christoph Lameter55136592010-08-20 12:37:13 -05003028static void early_kmem_cache_node_alloc(int node)
Christoph Lameter81819f02007-05-06 14:49:36 -07003029{
3030 struct page *page;
3031 struct kmem_cache_node *n;
3032
Christoph Lameter51df1142010-08-20 12:37:15 -05003033 BUG_ON(kmem_cache_node->size < sizeof(struct kmem_cache_node));
Christoph Lameter81819f02007-05-06 14:49:36 -07003034
Christoph Lameter51df1142010-08-20 12:37:15 -05003035 page = new_slab(kmem_cache_node, GFP_NOWAIT, node);
Christoph Lameter81819f02007-05-06 14:49:36 -07003036
3037 BUG_ON(!page);
Christoph Lametera2f92ee2007-08-22 14:01:57 -07003038 if (page_to_nid(page) != node) {
Fabian Frederickf9f58282014-06-04 16:06:34 -07003039 pr_err("SLUB: Unable to allocate memory from node %d\n", node);
3040 pr_err("SLUB: Allocating a useless per node structure in order to be able to continue\n");
Christoph Lametera2f92ee2007-08-22 14:01:57 -07003041 }
3042
Christoph Lameter81819f02007-05-06 14:49:36 -07003043 n = page->freelist;
3044 BUG_ON(!n);
Christoph Lameter51df1142010-08-20 12:37:15 -05003045 page->freelist = get_freepointer(kmem_cache_node, n);
Christoph Lametere6e82ea2011-08-09 16:12:24 -05003046 page->inuse = 1;
Christoph Lameter8cb0a502011-06-01 12:25:46 -05003047 page->frozen = 0;
Christoph Lameter51df1142010-08-20 12:37:15 -05003048 kmem_cache_node->node[node] = n;
Christoph Lameter8ab13722007-07-17 04:03:32 -07003049#ifdef CONFIG_SLUB_DEBUG
Christoph Lameterf7cb1932010-09-29 07:15:01 -05003050 init_object(kmem_cache_node, n, SLUB_RED_ACTIVE);
Christoph Lameter51df1142010-08-20 12:37:15 -05003051 init_tracking(kmem_cache_node, n);
Christoph Lameter8ab13722007-07-17 04:03:32 -07003052#endif
Andrey Ryabinin0316bec2015-02-13 14:39:42 -08003053 kasan_kmalloc(kmem_cache_node, n, sizeof(struct kmem_cache_node));
Joonsoo Kim40534972012-05-11 00:50:47 +09003054 init_kmem_cache_node(n);
Christoph Lameter51df1142010-08-20 12:37:15 -05003055 inc_slabs_node(kmem_cache_node, node, page->objects);
Christoph Lameter6446faa2008-02-15 23:45:26 -08003056
Dave Hansen67b6c902014-01-24 07:20:23 -08003057 /*
Steven Rostedt1e4dd942014-02-10 14:25:46 -08003058 * No locks need to be taken here as it has just been
3059 * initialized and there is no concurrent access.
Dave Hansen67b6c902014-01-24 07:20:23 -08003060 */
Steven Rostedt1e4dd942014-02-10 14:25:46 -08003061 __add_partial(n, page, DEACTIVATE_TO_HEAD);
Christoph Lameter81819f02007-05-06 14:49:36 -07003062}
3063
3064static void free_kmem_cache_nodes(struct kmem_cache *s)
3065{
3066 int node;
Christoph Lameterfa45dc22014-08-06 16:04:09 -07003067 struct kmem_cache_node *n;
Christoph Lameter81819f02007-05-06 14:49:36 -07003068
Christoph Lameterfa45dc22014-08-06 16:04:09 -07003069 for_each_kmem_cache_node(s, node, n) {
3070 kmem_cache_free(kmem_cache_node, n);
Christoph Lameter81819f02007-05-06 14:49:36 -07003071 s->node[node] = NULL;
3072 }
3073}
3074
Christoph Lameter55136592010-08-20 12:37:13 -05003075static int init_kmem_cache_nodes(struct kmem_cache *s)
Christoph Lameter81819f02007-05-06 14:49:36 -07003076{
3077 int node;
Christoph Lameter81819f02007-05-06 14:49:36 -07003078
Christoph Lameterf64dc582007-10-16 01:25:33 -07003079 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter81819f02007-05-06 14:49:36 -07003080 struct kmem_cache_node *n;
3081
Alexander Duyck73367bd2010-05-21 14:41:35 -07003082 if (slab_state == DOWN) {
Christoph Lameter55136592010-08-20 12:37:13 -05003083 early_kmem_cache_node_alloc(node);
Alexander Duyck73367bd2010-05-21 14:41:35 -07003084 continue;
Christoph Lameter81819f02007-05-06 14:49:36 -07003085 }
Christoph Lameter51df1142010-08-20 12:37:15 -05003086 n = kmem_cache_alloc_node(kmem_cache_node,
Christoph Lameter55136592010-08-20 12:37:13 -05003087 GFP_KERNEL, node);
Alexander Duyck73367bd2010-05-21 14:41:35 -07003088
3089 if (!n) {
3090 free_kmem_cache_nodes(s);
3091 return 0;
3092 }
3093
Christoph Lameter81819f02007-05-06 14:49:36 -07003094 s->node[node] = n;
Joonsoo Kim40534972012-05-11 00:50:47 +09003095 init_kmem_cache_node(n);
Christoph Lameter81819f02007-05-06 14:49:36 -07003096 }
3097 return 1;
3098}
Christoph Lameter81819f02007-05-06 14:49:36 -07003099
David Rientjesc0bdb232009-02-25 09:16:35 +02003100static void set_min_partial(struct kmem_cache *s, unsigned long min)
David Rientjes3b89d7d2009-02-22 17:40:07 -08003101{
3102 if (min < MIN_PARTIAL)
3103 min = MIN_PARTIAL;
3104 else if (min > MAX_PARTIAL)
3105 min = MAX_PARTIAL;
3106 s->min_partial = min;
3107}
3108
Christoph Lameter81819f02007-05-06 14:49:36 -07003109/*
3110 * calculate_sizes() determines the order and the distribution of data within
3111 * a slab object.
3112 */
Christoph Lameter06b285d2008-04-14 19:11:41 +03003113static int calculate_sizes(struct kmem_cache *s, int forced_order)
Christoph Lameter81819f02007-05-06 14:49:36 -07003114{
3115 unsigned long flags = s->flags;
Christoph Lameter3b0efdf2012-06-13 10:24:57 -05003116 unsigned long size = s->object_size;
Christoph Lameter834f3d12008-04-14 19:11:31 +03003117 int order;
Christoph Lameter81819f02007-05-06 14:49:36 -07003118
3119 /*
Christoph Lameterd8b42bf2008-02-15 23:45:25 -08003120 * Round up object size to the next word boundary. We can only
3121 * place the free pointer at word boundaries and this determines
3122 * the possible location of the free pointer.
3123 */
3124 size = ALIGN(size, sizeof(void *));
3125
3126#ifdef CONFIG_SLUB_DEBUG
3127 /*
Christoph Lameter81819f02007-05-06 14:49:36 -07003128 * Determine if we can poison the object itself. If the user of
3129 * the slab may touch the object after free or before allocation
3130 * then we should never poison the object itself.
3131 */
3132 if ((flags & SLAB_POISON) && !(flags & SLAB_DESTROY_BY_RCU) &&
Christoph Lameterc59def92007-05-16 22:10:50 -07003133 !s->ctor)
Christoph Lameter81819f02007-05-06 14:49:36 -07003134 s->flags |= __OBJECT_POISON;
3135 else
3136 s->flags &= ~__OBJECT_POISON;
3137
Christoph Lameter81819f02007-05-06 14:49:36 -07003138
3139 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07003140 * If we are Redzoning then check if there is some space between the
Christoph Lameter81819f02007-05-06 14:49:36 -07003141 * end of the object and the free pointer. If not then add an
Christoph Lameter672bba32007-05-09 02:32:39 -07003142 * additional word to have some bytes to store Redzone information.
Christoph Lameter81819f02007-05-06 14:49:36 -07003143 */
Christoph Lameter3b0efdf2012-06-13 10:24:57 -05003144 if ((flags & SLAB_RED_ZONE) && size == s->object_size)
Christoph Lameter81819f02007-05-06 14:49:36 -07003145 size += sizeof(void *);
Christoph Lameter41ecc552007-05-09 02:32:44 -07003146#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07003147
3148 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07003149 * With that we have determined the number of bytes in actual use
3150 * by the object. This is the potential offset to the free pointer.
Christoph Lameter81819f02007-05-06 14:49:36 -07003151 */
3152 s->inuse = size;
3153
3154 if (((flags & (SLAB_DESTROY_BY_RCU | SLAB_POISON)) ||
Christoph Lameterc59def92007-05-16 22:10:50 -07003155 s->ctor)) {
Christoph Lameter81819f02007-05-06 14:49:36 -07003156 /*
3157 * Relocate free pointer after the object if it is not
3158 * permitted to overwrite the first word of the object on
3159 * kmem_cache_free.
3160 *
3161 * This is the case if we do RCU, have a constructor or
3162 * destructor or are poisoning the objects.
3163 */
3164 s->offset = size;
3165 size += sizeof(void *);
3166 }
3167
Christoph Lameterc12b3c62007-05-23 13:57:31 -07003168#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter81819f02007-05-06 14:49:36 -07003169 if (flags & SLAB_STORE_USER)
3170 /*
3171 * Need to store information about allocs and frees after
3172 * the object.
3173 */
3174 size += 2 * sizeof(struct track);
3175
Christoph Lameterbe7b3fb2007-05-09 02:32:36 -07003176 if (flags & SLAB_RED_ZONE)
Christoph Lameter81819f02007-05-06 14:49:36 -07003177 /*
3178 * Add some empty padding so that we can catch
3179 * overwrites from earlier objects rather than let
3180 * tracking information or the free pointer be
Frederik Schwarzer0211a9c2008-12-29 22:14:56 +01003181 * corrupted if a user writes before the start
Christoph Lameter81819f02007-05-06 14:49:36 -07003182 * of the object.
3183 */
3184 size += sizeof(void *);
Christoph Lameter41ecc552007-05-09 02:32:44 -07003185#endif
Christoph Lameter672bba32007-05-09 02:32:39 -07003186
Christoph Lameter81819f02007-05-06 14:49:36 -07003187 /*
Christoph Lameter81819f02007-05-06 14:49:36 -07003188 * SLUB stores one object immediately after another beginning from
3189 * offset 0. In order to align the objects we have to simply size
3190 * each object to conform to the alignment.
3191 */
Christoph Lameter45906852012-11-28 16:23:16 +00003192 size = ALIGN(size, s->align);
Christoph Lameter81819f02007-05-06 14:49:36 -07003193 s->size = size;
Christoph Lameter06b285d2008-04-14 19:11:41 +03003194 if (forced_order >= 0)
3195 order = forced_order;
3196 else
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08003197 order = calculate_order(size, s->reserved);
Christoph Lameter81819f02007-05-06 14:49:36 -07003198
Christoph Lameter834f3d12008-04-14 19:11:31 +03003199 if (order < 0)
Christoph Lameter81819f02007-05-06 14:49:36 -07003200 return 0;
3201
Christoph Lameterb7a49f02008-02-14 14:21:32 -08003202 s->allocflags = 0;
Christoph Lameter834f3d12008-04-14 19:11:31 +03003203 if (order)
Christoph Lameterb7a49f02008-02-14 14:21:32 -08003204 s->allocflags |= __GFP_COMP;
3205
3206 if (s->flags & SLAB_CACHE_DMA)
Christoph Lameter2c59dd62013-01-10 19:14:19 +00003207 s->allocflags |= GFP_DMA;
Christoph Lameterb7a49f02008-02-14 14:21:32 -08003208
3209 if (s->flags & SLAB_RECLAIM_ACCOUNT)
3210 s->allocflags |= __GFP_RECLAIMABLE;
3211
Christoph Lameter81819f02007-05-06 14:49:36 -07003212 /*
3213 * Determine the number of objects per slab
3214 */
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08003215 s->oo = oo_make(order, size, s->reserved);
3216 s->min = oo_make(get_order(size), size, s->reserved);
Christoph Lameter205ab992008-04-14 19:11:40 +03003217 if (oo_objects(s->oo) > oo_objects(s->max))
3218 s->max = s->oo;
Christoph Lameter81819f02007-05-06 14:49:36 -07003219
Christoph Lameter834f3d12008-04-14 19:11:31 +03003220 return !!oo_objects(s->oo);
Christoph Lameter81819f02007-05-06 14:49:36 -07003221}
3222
Christoph Lameter8a13a4c2012-09-04 23:18:33 +00003223static int kmem_cache_open(struct kmem_cache *s, unsigned long flags)
Christoph Lameter81819f02007-05-06 14:49:36 -07003224{
Christoph Lameter8a13a4c2012-09-04 23:18:33 +00003225 s->flags = kmem_cache_flags(s->size, flags, s->name, s->ctor);
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08003226 s->reserved = 0;
Christoph Lameter81819f02007-05-06 14:49:36 -07003227
Lai Jiangshanda9a6382011-03-10 15:22:00 +08003228 if (need_reserve_slab_rcu && (s->flags & SLAB_DESTROY_BY_RCU))
3229 s->reserved = sizeof(struct rcu_head);
Christoph Lameter81819f02007-05-06 14:49:36 -07003230
Christoph Lameter06b285d2008-04-14 19:11:41 +03003231 if (!calculate_sizes(s, -1))
Christoph Lameter81819f02007-05-06 14:49:36 -07003232 goto error;
David Rientjes3de47212009-07-27 18:30:35 -07003233 if (disable_higher_order_debug) {
3234 /*
3235 * Disable debugging flags that store metadata if the min slab
3236 * order increased.
3237 */
Christoph Lameter3b0efdf2012-06-13 10:24:57 -05003238 if (get_order(s->size) > get_order(s->object_size)) {
David Rientjes3de47212009-07-27 18:30:35 -07003239 s->flags &= ~DEBUG_METADATA_FLAGS;
3240 s->offset = 0;
3241 if (!calculate_sizes(s, -1))
3242 goto error;
3243 }
3244 }
Christoph Lameter81819f02007-05-06 14:49:36 -07003245
Heiko Carstens25654092012-01-12 17:17:33 -08003246#if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \
3247 defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
Christoph Lameterb789ef52011-06-01 12:25:49 -05003248 if (system_has_cmpxchg_double() && (s->flags & SLAB_DEBUG_FLAGS) == 0)
3249 /* Enable fast mode */
3250 s->flags |= __CMPXCHG_DOUBLE;
3251#endif
3252
David Rientjes3b89d7d2009-02-22 17:40:07 -08003253 /*
3254 * The larger the object size is, the more pages we want on the partial
3255 * list to avoid pounding the page allocator excessively.
3256 */
Christoph Lameter49e22582011-08-09 16:12:27 -05003257 set_min_partial(s, ilog2(s->size) / 2);
3258
3259 /*
3260 * cpu_partial determined the maximum number of objects kept in the
3261 * per cpu partial lists of a processor.
3262 *
3263 * Per cpu partial lists mainly contain slabs that just have one
3264 * object freed. If they are used for allocation then they can be
3265 * filled up again with minimal effort. The slab will never hit the
3266 * per node partial lists and therefore no locking will be required.
3267 *
3268 * This setting also determines
3269 *
3270 * A) The number of objects from per cpu partial slabs dumped to the
3271 * per node list when we reach the limit.
Alex Shi9f264902011-09-01 11:32:18 +08003272 * B) The number of objects in cpu partial slabs to extract from the
Chen Gangd0e0ac92013-07-15 09:05:29 +08003273 * per node list when we run out of per cpu objects. We only fetch
3274 * 50% to keep some capacity around for frees.
Christoph Lameter49e22582011-08-09 16:12:27 -05003275 */
Joonsoo Kim345c9052013-06-19 14:05:52 +09003276 if (!kmem_cache_has_cpu_partial(s))
Christoph Lameter8f1e33d2011-11-23 09:24:27 -06003277 s->cpu_partial = 0;
3278 else if (s->size >= PAGE_SIZE)
Christoph Lameter49e22582011-08-09 16:12:27 -05003279 s->cpu_partial = 2;
3280 else if (s->size >= 1024)
3281 s->cpu_partial = 6;
3282 else if (s->size >= 256)
3283 s->cpu_partial = 13;
3284 else
3285 s->cpu_partial = 30;
3286
Christoph Lameter81819f02007-05-06 14:49:36 -07003287#ifdef CONFIG_NUMA
Christoph Lametere2cb96b2008-08-19 08:51:22 -05003288 s->remote_node_defrag_ratio = 1000;
Christoph Lameter81819f02007-05-06 14:49:36 -07003289#endif
Christoph Lameter55136592010-08-20 12:37:13 -05003290 if (!init_kmem_cache_nodes(s))
Christoph Lameterdfb4f092007-10-16 01:26:05 -07003291 goto error;
Christoph Lameter81819f02007-05-06 14:49:36 -07003292
Christoph Lameter55136592010-08-20 12:37:13 -05003293 if (alloc_kmem_cache_cpus(s))
Christoph Lameter278b1bb2012-09-05 00:20:34 +00003294 return 0;
Christoph Lameterff120592009-12-18 16:26:22 -06003295
Christoph Lameter4c93c3552007-10-16 01:26:08 -07003296 free_kmem_cache_nodes(s);
Christoph Lameter81819f02007-05-06 14:49:36 -07003297error:
3298 if (flags & SLAB_PANIC)
3299 panic("Cannot create slab %s size=%lu realsize=%u "
3300 "order=%u offset=%u flags=%lx\n",
Chen Gangd0e0ac92013-07-15 09:05:29 +08003301 s->name, (unsigned long)s->size, s->size,
3302 oo_order(s->oo), s->offset, flags);
Christoph Lameter278b1bb2012-09-05 00:20:34 +00003303 return -EINVAL;
Christoph Lameter81819f02007-05-06 14:49:36 -07003304}
Christoph Lameter81819f02007-05-06 14:49:36 -07003305
Christoph Lameter33b12c32008-04-25 12:22:43 -07003306static void list_slab_objects(struct kmem_cache *s, struct page *page,
3307 const char *text)
Christoph Lameter81819f02007-05-06 14:49:36 -07003308{
Christoph Lameter33b12c32008-04-25 12:22:43 -07003309#ifdef CONFIG_SLUB_DEBUG
3310 void *addr = page_address(page);
3311 void *p;
Namhyung Kima5dd5c12010-09-29 21:02:13 +09003312 unsigned long *map = kzalloc(BITS_TO_LONGS(page->objects) *
3313 sizeof(long), GFP_ATOMIC);
Eric Dumazetbbd7d572010-03-24 22:25:47 +01003314 if (!map)
3315 return;
Christoph Lameter945cf2b2012-09-04 23:18:33 +00003316 slab_err(s, page, text, s->name);
Christoph Lameter33b12c32008-04-25 12:22:43 -07003317 slab_lock(page);
Christoph Lameter33b12c32008-04-25 12:22:43 -07003318
Christoph Lameter5f80b132011-04-15 14:48:13 -05003319 get_map(s, page, map);
Christoph Lameter33b12c32008-04-25 12:22:43 -07003320 for_each_object(p, s, addr, page->objects) {
3321
3322 if (!test_bit(slab_index(p, s, addr), map)) {
Fabian Frederickf9f58282014-06-04 16:06:34 -07003323 pr_err("INFO: Object 0x%p @offset=%tu\n", p, p - addr);
Christoph Lameter33b12c32008-04-25 12:22:43 -07003324 print_tracking(s, p);
3325 }
3326 }
3327 slab_unlock(page);
Eric Dumazetbbd7d572010-03-24 22:25:47 +01003328 kfree(map);
Christoph Lameter33b12c32008-04-25 12:22:43 -07003329#endif
3330}
3331
Christoph Lameter81819f02007-05-06 14:49:36 -07003332/*
Christoph Lameter599870b2008-04-23 12:36:52 -07003333 * Attempt to free all partial slabs on a node.
Christoph Lameter69cb8e62011-08-09 16:12:22 -05003334 * This is called from kmem_cache_close(). We must be the last thread
3335 * using the cache and therefore we do not need to lock anymore.
Christoph Lameter81819f02007-05-06 14:49:36 -07003336 */
Christoph Lameter599870b2008-04-23 12:36:52 -07003337static void free_partial(struct kmem_cache *s, struct kmem_cache_node *n)
Christoph Lameter81819f02007-05-06 14:49:36 -07003338{
Christoph Lameter81819f02007-05-06 14:49:36 -07003339 struct page *page, *h;
3340
Christoph Lameter33b12c32008-04-25 12:22:43 -07003341 list_for_each_entry_safe(page, h, &n->partial, lru) {
Christoph Lameter81819f02007-05-06 14:49:36 -07003342 if (!page->inuse) {
Steven Rostedt1e4dd942014-02-10 14:25:46 -08003343 __remove_partial(n, page);
Christoph Lameter81819f02007-05-06 14:49:36 -07003344 discard_slab(s, page);
Christoph Lameter33b12c32008-04-25 12:22:43 -07003345 } else {
3346 list_slab_objects(s, page,
Christoph Lameter945cf2b2012-09-04 23:18:33 +00003347 "Objects remaining in %s on kmem_cache_close()");
Christoph Lameter599870b2008-04-23 12:36:52 -07003348 }
Christoph Lameter33b12c32008-04-25 12:22:43 -07003349 }
Christoph Lameter81819f02007-05-06 14:49:36 -07003350}
3351
3352/*
Christoph Lameter672bba32007-05-09 02:32:39 -07003353 * Release all resources used by a slab cache.
Christoph Lameter81819f02007-05-06 14:49:36 -07003354 */
Christoph Lameter0c710012007-07-17 04:03:24 -07003355static inline int kmem_cache_close(struct kmem_cache *s)
Christoph Lameter81819f02007-05-06 14:49:36 -07003356{
3357 int node;
Christoph Lameterfa45dc22014-08-06 16:04:09 -07003358 struct kmem_cache_node *n;
Christoph Lameter81819f02007-05-06 14:49:36 -07003359
3360 flush_all(s);
Christoph Lameter81819f02007-05-06 14:49:36 -07003361 /* Attempt to free all objects */
Christoph Lameterfa45dc22014-08-06 16:04:09 -07003362 for_each_kmem_cache_node(s, node, n) {
Christoph Lameter599870b2008-04-23 12:36:52 -07003363 free_partial(s, n);
3364 if (n->nr_partial || slabs_node(s, node))
Christoph Lameter81819f02007-05-06 14:49:36 -07003365 return 1;
3366 }
Christoph Lameter945cf2b2012-09-04 23:18:33 +00003367 free_percpu(s->cpu_slab);
Christoph Lameter81819f02007-05-06 14:49:36 -07003368 free_kmem_cache_nodes(s);
3369 return 0;
3370}
3371
Christoph Lameter945cf2b2012-09-04 23:18:33 +00003372int __kmem_cache_shutdown(struct kmem_cache *s)
Christoph Lameter81819f02007-05-06 14:49:36 -07003373{
Christoph Lameter41a21282014-05-06 12:50:08 -07003374 return kmem_cache_close(s);
Christoph Lameter81819f02007-05-06 14:49:36 -07003375}
Christoph Lameter81819f02007-05-06 14:49:36 -07003376
3377/********************************************************************
3378 * Kmalloc subsystem
3379 *******************************************************************/
3380
Christoph Lameter81819f02007-05-06 14:49:36 -07003381static int __init setup_slub_min_order(char *str)
3382{
Pekka Enberg06428782008-01-07 23:20:27 -08003383 get_option(&str, &slub_min_order);
Christoph Lameter81819f02007-05-06 14:49:36 -07003384
3385 return 1;
3386}
3387
3388__setup("slub_min_order=", setup_slub_min_order);
3389
3390static int __init setup_slub_max_order(char *str)
3391{
Pekka Enberg06428782008-01-07 23:20:27 -08003392 get_option(&str, &slub_max_order);
David Rientjes818cf592009-04-23 09:58:22 +03003393 slub_max_order = min(slub_max_order, MAX_ORDER - 1);
Christoph Lameter81819f02007-05-06 14:49:36 -07003394
3395 return 1;
3396}
3397
3398__setup("slub_max_order=", setup_slub_max_order);
3399
3400static int __init setup_slub_min_objects(char *str)
3401{
Pekka Enberg06428782008-01-07 23:20:27 -08003402 get_option(&str, &slub_min_objects);
Christoph Lameter81819f02007-05-06 14:49:36 -07003403
3404 return 1;
3405}
3406
3407__setup("slub_min_objects=", setup_slub_min_objects);
3408
Christoph Lameter81819f02007-05-06 14:49:36 -07003409void *__kmalloc(size_t size, gfp_t flags)
3410{
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003411 struct kmem_cache *s;
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03003412 void *ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07003413
Christoph Lameter95a05b42013-01-10 19:14:19 +00003414 if (unlikely(size > KMALLOC_MAX_CACHE_SIZE))
Pekka Enbergeada35e2008-02-11 22:47:46 +02003415 return kmalloc_large(size, flags);
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003416
Christoph Lameter2c59dd62013-01-10 19:14:19 +00003417 s = kmalloc_slab(size, flags);
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003418
3419 if (unlikely(ZERO_OR_NULL_PTR(s)))
Christoph Lameter6cb8f912007-07-17 04:03:22 -07003420 return s;
3421
Ezequiel Garcia2b847c32012-09-08 17:47:58 -03003422 ret = slab_alloc(s, flags, _RET_IP_);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03003423
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02003424 trace_kmalloc(_RET_IP_, ret, size, s->size, flags);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03003425
Andrey Ryabinin0316bec2015-02-13 14:39:42 -08003426 kasan_kmalloc(s, ret, size);
3427
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03003428 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07003429}
3430EXPORT_SYMBOL(__kmalloc);
3431
Namhyung Kim5d1f57e2010-09-29 21:02:15 +09003432#ifdef CONFIG_NUMA
Christoph Lameterf619cfe2008-03-01 13:56:40 -08003433static void *kmalloc_large_node(size_t size, gfp_t flags, int node)
3434{
Vegard Nossumb1eeab62008-11-25 16:55:53 +01003435 struct page *page;
Catalin Marinase4f7c0b42009-07-07 10:32:59 +01003436 void *ptr = NULL;
Christoph Lameterf619cfe2008-03-01 13:56:40 -08003437
Vladimir Davydov52383432014-06-04 16:06:39 -07003438 flags |= __GFP_COMP | __GFP_NOTRACK;
3439 page = alloc_kmem_pages_node(node, flags, get_order(size));
Christoph Lameterf619cfe2008-03-01 13:56:40 -08003440 if (page)
Catalin Marinase4f7c0b42009-07-07 10:32:59 +01003441 ptr = page_address(page);
3442
Roman Bobnievd56791b2013-10-08 15:58:57 -07003443 kmalloc_large_node_hook(ptr, size, flags);
Catalin Marinase4f7c0b42009-07-07 10:32:59 +01003444 return ptr;
Christoph Lameterf619cfe2008-03-01 13:56:40 -08003445}
3446
Christoph Lameter81819f02007-05-06 14:49:36 -07003447void *__kmalloc_node(size_t size, gfp_t flags, int node)
3448{
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003449 struct kmem_cache *s;
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03003450 void *ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07003451
Christoph Lameter95a05b42013-01-10 19:14:19 +00003452 if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) {
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03003453 ret = kmalloc_large_node(size, flags, node);
3454
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02003455 trace_kmalloc_node(_RET_IP_, ret,
3456 size, PAGE_SIZE << get_order(size),
3457 flags, node);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03003458
3459 return ret;
3460 }
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003461
Christoph Lameter2c59dd62013-01-10 19:14:19 +00003462 s = kmalloc_slab(size, flags);
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003463
3464 if (unlikely(ZERO_OR_NULL_PTR(s)))
Christoph Lameter6cb8f912007-07-17 04:03:22 -07003465 return s;
3466
Ezequiel Garcia2b847c32012-09-08 17:47:58 -03003467 ret = slab_alloc_node(s, flags, node, _RET_IP_);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03003468
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02003469 trace_kmalloc_node(_RET_IP_, ret, size, s->size, flags, node);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03003470
Andrey Ryabinin0316bec2015-02-13 14:39:42 -08003471 kasan_kmalloc(s, ret, size);
3472
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03003473 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07003474}
3475EXPORT_SYMBOL(__kmalloc_node);
3476#endif
3477
Andrey Ryabinin0316bec2015-02-13 14:39:42 -08003478static size_t __ksize(const void *object)
Christoph Lameter81819f02007-05-06 14:49:36 -07003479{
Christoph Lameter272c1d22007-06-08 13:46:49 -07003480 struct page *page;
Christoph Lameter81819f02007-05-06 14:49:36 -07003481
Christoph Lameteref8b4522007-10-16 01:24:46 -07003482 if (unlikely(object == ZERO_SIZE_PTR))
Christoph Lameter272c1d22007-06-08 13:46:49 -07003483 return 0;
3484
Vegard Nossum294a80a2007-12-04 23:45:30 -08003485 page = virt_to_head_page(object);
Vegard Nossum294a80a2007-12-04 23:45:30 -08003486
Pekka Enberg76994412008-05-22 19:22:25 +03003487 if (unlikely(!PageSlab(page))) {
3488 WARN_ON(!PageCompound(page));
Vegard Nossum294a80a2007-12-04 23:45:30 -08003489 return PAGE_SIZE << compound_order(page);
Pekka Enberg76994412008-05-22 19:22:25 +03003490 }
Christoph Lameter81819f02007-05-06 14:49:36 -07003491
Glauber Costa1b4f59e32012-10-22 18:05:36 +04003492 return slab_ksize(page->slab_cache);
Christoph Lameter81819f02007-05-06 14:49:36 -07003493}
Andrey Ryabinin0316bec2015-02-13 14:39:42 -08003494
3495size_t ksize(const void *object)
3496{
3497 size_t size = __ksize(object);
3498 /* We assume that ksize callers could use whole allocated area,
3499 so we need unpoison this area. */
3500 kasan_krealloc(object, size);
3501 return size;
3502}
Kirill A. Shutemovb1aabec2009-02-10 15:21:44 +02003503EXPORT_SYMBOL(ksize);
Christoph Lameter81819f02007-05-06 14:49:36 -07003504
3505void kfree(const void *x)
3506{
Christoph Lameter81819f02007-05-06 14:49:36 -07003507 struct page *page;
Christoph Lameter5bb983b2008-02-07 17:47:41 -08003508 void *object = (void *)x;
Christoph Lameter81819f02007-05-06 14:49:36 -07003509
Pekka Enberg2121db72009-03-25 11:05:57 +02003510 trace_kfree(_RET_IP_, x);
3511
Satyam Sharma2408c552007-10-16 01:24:44 -07003512 if (unlikely(ZERO_OR_NULL_PTR(x)))
Christoph Lameter81819f02007-05-06 14:49:36 -07003513 return;
3514
Christoph Lameterb49af682007-05-06 14:49:41 -07003515 page = virt_to_head_page(x);
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003516 if (unlikely(!PageSlab(page))) {
Christoph Lameter09375022008-05-28 10:32:22 -07003517 BUG_ON(!PageCompound(page));
Roman Bobnievd56791b2013-10-08 15:58:57 -07003518 kfree_hook(x);
Vladimir Davydov52383432014-06-04 16:06:39 -07003519 __free_kmem_pages(page, compound_order(page));
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003520 return;
3521 }
Glauber Costa1b4f59e32012-10-22 18:05:36 +04003522 slab_free(page->slab_cache, page, object, _RET_IP_);
Christoph Lameter81819f02007-05-06 14:49:36 -07003523}
3524EXPORT_SYMBOL(kfree);
3525
Vladimir Davydov832f37f2015-02-12 14:59:41 -08003526#define SHRINK_PROMOTE_MAX 32
3527
Christoph Lameter2086d262007-05-06 14:49:46 -07003528/*
Vladimir Davydov832f37f2015-02-12 14:59:41 -08003529 * kmem_cache_shrink discards empty slabs and promotes the slabs filled
3530 * up most to the head of the partial lists. New allocations will then
3531 * fill those up and thus they can be removed from the partial lists.
Christoph Lameter672bba32007-05-09 02:32:39 -07003532 *
3533 * The slabs with the least items are placed last. This results in them
3534 * being allocated from last increasing the chance that the last objects
3535 * are freed in them.
Christoph Lameter2086d262007-05-06 14:49:46 -07003536 */
Vladimir Davydovd6e0b7f2015-02-12 14:59:47 -08003537int __kmem_cache_shrink(struct kmem_cache *s, bool deactivate)
Christoph Lameter2086d262007-05-06 14:49:46 -07003538{
3539 int node;
3540 int i;
3541 struct kmem_cache_node *n;
3542 struct page *page;
3543 struct page *t;
Vladimir Davydov832f37f2015-02-12 14:59:41 -08003544 struct list_head discard;
3545 struct list_head promote[SHRINK_PROMOTE_MAX];
Christoph Lameter2086d262007-05-06 14:49:46 -07003546 unsigned long flags;
Vladimir Davydovce3712d2015-02-12 14:59:44 -08003547 int ret = 0;
Christoph Lameter2086d262007-05-06 14:49:46 -07003548
Vladimir Davydovd6e0b7f2015-02-12 14:59:47 -08003549 if (deactivate) {
3550 /*
3551 * Disable empty slabs caching. Used to avoid pinning offline
3552 * memory cgroups by kmem pages that can be freed.
3553 */
3554 s->cpu_partial = 0;
3555 s->min_partial = 0;
3556
3557 /*
3558 * s->cpu_partial is checked locklessly (see put_cpu_partial),
3559 * so we have to make sure the change is visible.
3560 */
3561 kick_all_cpus_sync();
3562 }
3563
Christoph Lameter2086d262007-05-06 14:49:46 -07003564 flush_all(s);
Christoph Lameterfa45dc22014-08-06 16:04:09 -07003565 for_each_kmem_cache_node(s, node, n) {
Vladimir Davydov832f37f2015-02-12 14:59:41 -08003566 INIT_LIST_HEAD(&discard);
3567 for (i = 0; i < SHRINK_PROMOTE_MAX; i++)
3568 INIT_LIST_HEAD(promote + i);
Christoph Lameter2086d262007-05-06 14:49:46 -07003569
3570 spin_lock_irqsave(&n->list_lock, flags);
3571
3572 /*
Vladimir Davydov832f37f2015-02-12 14:59:41 -08003573 * Build lists of slabs to discard or promote.
Christoph Lameter2086d262007-05-06 14:49:46 -07003574 *
Christoph Lameter672bba32007-05-09 02:32:39 -07003575 * Note that concurrent frees may occur while we hold the
3576 * list_lock. page->inuse here is the upper limit.
Christoph Lameter2086d262007-05-06 14:49:46 -07003577 */
3578 list_for_each_entry_safe(page, t, &n->partial, lru) {
Vladimir Davydov832f37f2015-02-12 14:59:41 -08003579 int free = page->objects - page->inuse;
3580
3581 /* Do not reread page->inuse */
3582 barrier();
3583
3584 /* We do not keep full slabs on the list */
3585 BUG_ON(free <= 0);
3586
3587 if (free == page->objects) {
3588 list_move(&page->lru, &discard);
Christoph Lameter69cb8e62011-08-09 16:12:22 -05003589 n->nr_partial--;
Vladimir Davydov832f37f2015-02-12 14:59:41 -08003590 } else if (free <= SHRINK_PROMOTE_MAX)
3591 list_move(&page->lru, promote + free - 1);
Christoph Lameter2086d262007-05-06 14:49:46 -07003592 }
3593
Christoph Lameter2086d262007-05-06 14:49:46 -07003594 /*
Vladimir Davydov832f37f2015-02-12 14:59:41 -08003595 * Promote the slabs filled up most to the head of the
3596 * partial list.
Christoph Lameter2086d262007-05-06 14:49:46 -07003597 */
Vladimir Davydov832f37f2015-02-12 14:59:41 -08003598 for (i = SHRINK_PROMOTE_MAX - 1; i >= 0; i--)
3599 list_splice(promote + i, &n->partial);
Christoph Lameter2086d262007-05-06 14:49:46 -07003600
Christoph Lameter2086d262007-05-06 14:49:46 -07003601 spin_unlock_irqrestore(&n->list_lock, flags);
Christoph Lameter69cb8e62011-08-09 16:12:22 -05003602
3603 /* Release empty slabs */
Vladimir Davydov832f37f2015-02-12 14:59:41 -08003604 list_for_each_entry_safe(page, t, &discard, lru)
Christoph Lameter69cb8e62011-08-09 16:12:22 -05003605 discard_slab(s, page);
Vladimir Davydovce3712d2015-02-12 14:59:44 -08003606
3607 if (slabs_node(s, node))
3608 ret = 1;
Christoph Lameter2086d262007-05-06 14:49:46 -07003609 }
3610
Vladimir Davydovce3712d2015-02-12 14:59:44 -08003611 return ret;
Christoph Lameter2086d262007-05-06 14:49:46 -07003612}
Christoph Lameter2086d262007-05-06 14:49:46 -07003613
Yasunori Gotob9049e22007-10-21 16:41:37 -07003614static int slab_mem_going_offline_callback(void *arg)
3615{
3616 struct kmem_cache *s;
3617
Christoph Lameter18004c52012-07-06 15:25:12 -05003618 mutex_lock(&slab_mutex);
Yasunori Gotob9049e22007-10-21 16:41:37 -07003619 list_for_each_entry(s, &slab_caches, list)
Vladimir Davydovd6e0b7f2015-02-12 14:59:47 -08003620 __kmem_cache_shrink(s, false);
Christoph Lameter18004c52012-07-06 15:25:12 -05003621 mutex_unlock(&slab_mutex);
Yasunori Gotob9049e22007-10-21 16:41:37 -07003622
3623 return 0;
3624}
3625
3626static void slab_mem_offline_callback(void *arg)
3627{
3628 struct kmem_cache_node *n;
3629 struct kmem_cache *s;
3630 struct memory_notify *marg = arg;
3631 int offline_node;
3632
Lai Jiangshanb9d5ab22012-12-11 16:01:05 -08003633 offline_node = marg->status_change_nid_normal;
Yasunori Gotob9049e22007-10-21 16:41:37 -07003634
3635 /*
3636 * If the node still has available memory. we need kmem_cache_node
3637 * for it yet.
3638 */
3639 if (offline_node < 0)
3640 return;
3641
Christoph Lameter18004c52012-07-06 15:25:12 -05003642 mutex_lock(&slab_mutex);
Yasunori Gotob9049e22007-10-21 16:41:37 -07003643 list_for_each_entry(s, &slab_caches, list) {
3644 n = get_node(s, offline_node);
3645 if (n) {
3646 /*
3647 * if n->nr_slabs > 0, slabs still exist on the node
3648 * that is going down. We were unable to free them,
Adam Buchbinderc9404c92009-12-18 15:40:42 -05003649 * and offline_pages() function shouldn't call this
Yasunori Gotob9049e22007-10-21 16:41:37 -07003650 * callback. So, we must fail.
3651 */
Christoph Lameter0f389ec2008-04-14 18:53:02 +03003652 BUG_ON(slabs_node(s, offline_node));
Yasunori Gotob9049e22007-10-21 16:41:37 -07003653
3654 s->node[offline_node] = NULL;
Christoph Lameter8de66a02010-08-25 14:51:14 -05003655 kmem_cache_free(kmem_cache_node, n);
Yasunori Gotob9049e22007-10-21 16:41:37 -07003656 }
3657 }
Christoph Lameter18004c52012-07-06 15:25:12 -05003658 mutex_unlock(&slab_mutex);
Yasunori Gotob9049e22007-10-21 16:41:37 -07003659}
3660
3661static int slab_mem_going_online_callback(void *arg)
3662{
3663 struct kmem_cache_node *n;
3664 struct kmem_cache *s;
3665 struct memory_notify *marg = arg;
Lai Jiangshanb9d5ab22012-12-11 16:01:05 -08003666 int nid = marg->status_change_nid_normal;
Yasunori Gotob9049e22007-10-21 16:41:37 -07003667 int ret = 0;
3668
3669 /*
3670 * If the node's memory is already available, then kmem_cache_node is
3671 * already created. Nothing to do.
3672 */
3673 if (nid < 0)
3674 return 0;
3675
3676 /*
Christoph Lameter0121c6192008-04-29 16:11:12 -07003677 * We are bringing a node online. No memory is available yet. We must
Yasunori Gotob9049e22007-10-21 16:41:37 -07003678 * allocate a kmem_cache_node structure in order to bring the node
3679 * online.
3680 */
Christoph Lameter18004c52012-07-06 15:25:12 -05003681 mutex_lock(&slab_mutex);
Yasunori Gotob9049e22007-10-21 16:41:37 -07003682 list_for_each_entry(s, &slab_caches, list) {
3683 /*
3684 * XXX: kmem_cache_alloc_node will fallback to other nodes
3685 * since memory is not yet available from the node that
3686 * is brought up.
3687 */
Christoph Lameter8de66a02010-08-25 14:51:14 -05003688 n = kmem_cache_alloc(kmem_cache_node, GFP_KERNEL);
Yasunori Gotob9049e22007-10-21 16:41:37 -07003689 if (!n) {
3690 ret = -ENOMEM;
3691 goto out;
3692 }
Joonsoo Kim40534972012-05-11 00:50:47 +09003693 init_kmem_cache_node(n);
Yasunori Gotob9049e22007-10-21 16:41:37 -07003694 s->node[nid] = n;
3695 }
3696out:
Christoph Lameter18004c52012-07-06 15:25:12 -05003697 mutex_unlock(&slab_mutex);
Yasunori Gotob9049e22007-10-21 16:41:37 -07003698 return ret;
3699}
3700
3701static int slab_memory_callback(struct notifier_block *self,
3702 unsigned long action, void *arg)
3703{
3704 int ret = 0;
3705
3706 switch (action) {
3707 case MEM_GOING_ONLINE:
3708 ret = slab_mem_going_online_callback(arg);
3709 break;
3710 case MEM_GOING_OFFLINE:
3711 ret = slab_mem_going_offline_callback(arg);
3712 break;
3713 case MEM_OFFLINE:
3714 case MEM_CANCEL_ONLINE:
3715 slab_mem_offline_callback(arg);
3716 break;
3717 case MEM_ONLINE:
3718 case MEM_CANCEL_OFFLINE:
3719 break;
3720 }
KAMEZAWA Hiroyukidc19f9d2008-12-01 13:13:48 -08003721 if (ret)
3722 ret = notifier_from_errno(ret);
3723 else
3724 ret = NOTIFY_OK;
Yasunori Gotob9049e22007-10-21 16:41:37 -07003725 return ret;
3726}
3727
Andrew Morton3ac38fa2013-04-29 15:08:06 -07003728static struct notifier_block slab_memory_callback_nb = {
3729 .notifier_call = slab_memory_callback,
3730 .priority = SLAB_CALLBACK_PRI,
3731};
Yasunori Gotob9049e22007-10-21 16:41:37 -07003732
Christoph Lameter81819f02007-05-06 14:49:36 -07003733/********************************************************************
3734 * Basic setup of slabs
3735 *******************************************************************/
3736
Christoph Lameter51df1142010-08-20 12:37:15 -05003737/*
3738 * Used for early kmem_cache structures that were allocated using
Christoph Lameterdffb4d62012-11-28 16:23:07 +00003739 * the page allocator. Allocate them properly then fix up the pointers
3740 * that may be pointing to the wrong kmem_cache structure.
Christoph Lameter51df1142010-08-20 12:37:15 -05003741 */
3742
Christoph Lameterdffb4d62012-11-28 16:23:07 +00003743static struct kmem_cache * __init bootstrap(struct kmem_cache *static_cache)
Christoph Lameter51df1142010-08-20 12:37:15 -05003744{
3745 int node;
Christoph Lameterdffb4d62012-11-28 16:23:07 +00003746 struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT);
Christoph Lameterfa45dc22014-08-06 16:04:09 -07003747 struct kmem_cache_node *n;
Christoph Lameter51df1142010-08-20 12:37:15 -05003748
Christoph Lameterdffb4d62012-11-28 16:23:07 +00003749 memcpy(s, static_cache, kmem_cache->object_size);
Christoph Lameter51df1142010-08-20 12:37:15 -05003750
Glauber Costa7d557b32013-02-22 20:20:00 +04003751 /*
3752 * This runs very early, and only the boot processor is supposed to be
3753 * up. Even if it weren't true, IRQs are not up so we couldn't fire
3754 * IPIs around.
3755 */
3756 __flush_cpu_slab(s, smp_processor_id());
Christoph Lameterfa45dc22014-08-06 16:04:09 -07003757 for_each_kmem_cache_node(s, node, n) {
Christoph Lameter51df1142010-08-20 12:37:15 -05003758 struct page *p;
3759
Christoph Lameterfa45dc22014-08-06 16:04:09 -07003760 list_for_each_entry(p, &n->partial, lru)
3761 p->slab_cache = s;
Christoph Lameter51df1142010-08-20 12:37:15 -05003762
Li Zefan607bf322011-04-12 15:22:26 +08003763#ifdef CONFIG_SLUB_DEBUG
Christoph Lameterfa45dc22014-08-06 16:04:09 -07003764 list_for_each_entry(p, &n->full, lru)
3765 p->slab_cache = s;
Christoph Lameter51df1142010-08-20 12:37:15 -05003766#endif
Christoph Lameter51df1142010-08-20 12:37:15 -05003767 }
Vladimir Davydovf7ce3192015-02-12 14:59:20 -08003768 slab_init_memcg_params(s);
Christoph Lameterdffb4d62012-11-28 16:23:07 +00003769 list_add(&s->list, &slab_caches);
3770 return s;
Christoph Lameter51df1142010-08-20 12:37:15 -05003771}
3772
Christoph Lameter81819f02007-05-06 14:49:36 -07003773void __init kmem_cache_init(void)
3774{
Christoph Lameterdffb4d62012-11-28 16:23:07 +00003775 static __initdata struct kmem_cache boot_kmem_cache,
3776 boot_kmem_cache_node;
Christoph Lameter51df1142010-08-20 12:37:15 -05003777
Stanislaw Gruszkafc8d8622012-01-10 15:07:32 -08003778 if (debug_guardpage_minorder())
3779 slub_max_order = 0;
3780
Christoph Lameterdffb4d62012-11-28 16:23:07 +00003781 kmem_cache_node = &boot_kmem_cache_node;
3782 kmem_cache = &boot_kmem_cache;
Christoph Lameter51df1142010-08-20 12:37:15 -05003783
Christoph Lameterdffb4d62012-11-28 16:23:07 +00003784 create_boot_cache(kmem_cache_node, "kmem_cache_node",
3785 sizeof(struct kmem_cache_node), SLAB_HWCACHE_ALIGN);
Yasunori Gotob9049e22007-10-21 16:41:37 -07003786
Andrew Morton3ac38fa2013-04-29 15:08:06 -07003787 register_hotmemory_notifier(&slab_memory_callback_nb);
Christoph Lameter81819f02007-05-06 14:49:36 -07003788
3789 /* Able to allocate the per node structures */
3790 slab_state = PARTIAL;
3791
Christoph Lameterdffb4d62012-11-28 16:23:07 +00003792 create_boot_cache(kmem_cache, "kmem_cache",
3793 offsetof(struct kmem_cache, node) +
3794 nr_node_ids * sizeof(struct kmem_cache_node *),
3795 SLAB_HWCACHE_ALIGN);
Christoph Lameter8a13a4c2012-09-04 23:18:33 +00003796
Christoph Lameterdffb4d62012-11-28 16:23:07 +00003797 kmem_cache = bootstrap(&boot_kmem_cache);
Christoph Lameter81819f02007-05-06 14:49:36 -07003798
Christoph Lameter51df1142010-08-20 12:37:15 -05003799 /*
3800 * Allocate kmem_cache_node properly from the kmem_cache slab.
3801 * kmem_cache_node is separately allocated so no need to
3802 * update any list pointers.
3803 */
Christoph Lameterdffb4d62012-11-28 16:23:07 +00003804 kmem_cache_node = bootstrap(&boot_kmem_cache_node);
Christoph Lameter51df1142010-08-20 12:37:15 -05003805
3806 /* Now we can use the kmem_cache to allocate kmalloc slabs */
Daniel Sanders34cc6992015-06-24 16:55:57 -07003807 setup_kmalloc_cache_index_table();
Christoph Lameterf97d5f62013-01-10 19:12:17 +00003808 create_kmalloc_caches(0);
Christoph Lameter81819f02007-05-06 14:49:36 -07003809
3810#ifdef CONFIG_SMP
3811 register_cpu_notifier(&slab_notifier);
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06003812#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07003813
Fabian Frederickf9f58282014-06-04 16:06:34 -07003814 pr_info("SLUB: HWalign=%d, Order=%d-%d, MinObjects=%d, CPUs=%d, Nodes=%d\n",
Christoph Lameterf97d5f62013-01-10 19:12:17 +00003815 cache_line_size(),
Christoph Lameter81819f02007-05-06 14:49:36 -07003816 slub_min_order, slub_max_order, slub_min_objects,
3817 nr_cpu_ids, nr_node_ids);
3818}
3819
Pekka Enberg7e85ee02009-06-12 14:03:06 +03003820void __init kmem_cache_init_late(void)
3821{
Pekka Enberg7e85ee02009-06-12 14:03:06 +03003822}
3823
Glauber Costa2633d7a2012-12-18 14:22:34 -08003824struct kmem_cache *
Vladimir Davydova44cb942014-04-07 15:39:23 -07003825__kmem_cache_alias(const char *name, size_t size, size_t align,
3826 unsigned long flags, void (*ctor)(void *))
Christoph Lameter81819f02007-05-06 14:49:36 -07003827{
Vladimir Davydov426589f2015-02-12 14:59:23 -08003828 struct kmem_cache *s, *c;
Christoph Lameter81819f02007-05-06 14:49:36 -07003829
Vladimir Davydova44cb942014-04-07 15:39:23 -07003830 s = find_mergeable(size, align, flags, name, ctor);
Christoph Lameter81819f02007-05-06 14:49:36 -07003831 if (s) {
3832 s->refcount++;
Vladimir Davydov84d0ddd2014-04-07 15:39:29 -07003833
Christoph Lameter81819f02007-05-06 14:49:36 -07003834 /*
3835 * Adjust the object sizes so that we clear
3836 * the complete object on kzalloc.
3837 */
Christoph Lameter3b0efdf2012-06-13 10:24:57 -05003838 s->object_size = max(s->object_size, (int)size);
Christoph Lameter81819f02007-05-06 14:49:36 -07003839 s->inuse = max_t(int, s->inuse, ALIGN(size, sizeof(void *)));
Christoph Lameter6446faa2008-02-15 23:45:26 -08003840
Vladimir Davydov426589f2015-02-12 14:59:23 -08003841 for_each_memcg_cache(c, s) {
Vladimir Davydov84d0ddd2014-04-07 15:39:29 -07003842 c->object_size = s->object_size;
3843 c->inuse = max_t(int, c->inuse,
3844 ALIGN(size, sizeof(void *)));
3845 }
3846
David Rientjes7b8f3b62008-12-17 22:09:46 -08003847 if (sysfs_slab_alias(s, name)) {
David Rientjes7b8f3b62008-12-17 22:09:46 -08003848 s->refcount--;
Christoph Lametercbb79692012-09-05 00:18:32 +00003849 s = NULL;
David Rientjes7b8f3b62008-12-17 22:09:46 -08003850 }
Christoph Lametera0e1d1b2007-07-17 04:03:31 -07003851 }
Christoph Lameter6446faa2008-02-15 23:45:26 -08003852
Christoph Lametercbb79692012-09-05 00:18:32 +00003853 return s;
3854}
Pekka Enberg84c1cf62010-09-14 23:21:12 +03003855
Christoph Lameter8a13a4c2012-09-04 23:18:33 +00003856int __kmem_cache_create(struct kmem_cache *s, unsigned long flags)
Christoph Lametercbb79692012-09-05 00:18:32 +00003857{
Pekka Enbergaac3a162012-09-05 12:07:44 +03003858 int err;
Christoph Lameter20cea962012-07-06 15:25:13 -05003859
Pekka Enbergaac3a162012-09-05 12:07:44 +03003860 err = kmem_cache_open(s, flags);
3861 if (err)
3862 return err;
Christoph Lameter20cea962012-07-06 15:25:13 -05003863
Christoph Lameter45530c42012-11-28 16:23:07 +00003864 /* Mutex is not taken during early boot */
3865 if (slab_state <= UP)
3866 return 0;
3867
Glauber Costa107dab52012-12-18 14:23:05 -08003868 memcg_propagate_slab_attrs(s);
Pekka Enbergaac3a162012-09-05 12:07:44 +03003869 err = sysfs_slab_add(s);
Pekka Enbergaac3a162012-09-05 12:07:44 +03003870 if (err)
3871 kmem_cache_close(s);
3872
3873 return err;
Christoph Lameter81819f02007-05-06 14:49:36 -07003874}
Christoph Lameter81819f02007-05-06 14:49:36 -07003875
Christoph Lameter81819f02007-05-06 14:49:36 -07003876#ifdef CONFIG_SMP
Christoph Lameter27390bc2007-06-01 00:47:09 -07003877/*
Christoph Lameter672bba32007-05-09 02:32:39 -07003878 * Use the cpu notifier to insure that the cpu slabs are flushed when
3879 * necessary.
Christoph Lameter81819f02007-05-06 14:49:36 -07003880 */
Paul Gortmaker0db06282013-06-19 14:53:51 -04003881static int slab_cpuup_callback(struct notifier_block *nfb,
Christoph Lameter81819f02007-05-06 14:49:36 -07003882 unsigned long action, void *hcpu)
3883{
3884 long cpu = (long)hcpu;
Christoph Lameter5b95a4ac2007-07-17 04:03:19 -07003885 struct kmem_cache *s;
3886 unsigned long flags;
Christoph Lameter81819f02007-05-06 14:49:36 -07003887
3888 switch (action) {
3889 case CPU_UP_CANCELED:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07003890 case CPU_UP_CANCELED_FROZEN:
Christoph Lameter81819f02007-05-06 14:49:36 -07003891 case CPU_DEAD:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07003892 case CPU_DEAD_FROZEN:
Christoph Lameter18004c52012-07-06 15:25:12 -05003893 mutex_lock(&slab_mutex);
Christoph Lameter5b95a4ac2007-07-17 04:03:19 -07003894 list_for_each_entry(s, &slab_caches, list) {
3895 local_irq_save(flags);
3896 __flush_cpu_slab(s, cpu);
3897 local_irq_restore(flags);
3898 }
Christoph Lameter18004c52012-07-06 15:25:12 -05003899 mutex_unlock(&slab_mutex);
Christoph Lameter81819f02007-05-06 14:49:36 -07003900 break;
3901 default:
3902 break;
3903 }
3904 return NOTIFY_OK;
3905}
3906
Paul Gortmaker0db06282013-06-19 14:53:51 -04003907static struct notifier_block slab_notifier = {
Ingo Molnar3adbefe2008-02-05 17:57:39 -08003908 .notifier_call = slab_cpuup_callback
Pekka Enberg06428782008-01-07 23:20:27 -08003909};
Christoph Lameter81819f02007-05-06 14:49:36 -07003910
3911#endif
3912
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03003913void *__kmalloc_track_caller(size_t size, gfp_t gfpflags, unsigned long caller)
Christoph Lameter81819f02007-05-06 14:49:36 -07003914{
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003915 struct kmem_cache *s;
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03003916 void *ret;
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003917
Christoph Lameter95a05b42013-01-10 19:14:19 +00003918 if (unlikely(size > KMALLOC_MAX_CACHE_SIZE))
Pekka Enbergeada35e2008-02-11 22:47:46 +02003919 return kmalloc_large(size, gfpflags);
3920
Christoph Lameter2c59dd62013-01-10 19:14:19 +00003921 s = kmalloc_slab(size, gfpflags);
Christoph Lameter81819f02007-05-06 14:49:36 -07003922
Satyam Sharma2408c552007-10-16 01:24:44 -07003923 if (unlikely(ZERO_OR_NULL_PTR(s)))
Christoph Lameter6cb8f912007-07-17 04:03:22 -07003924 return s;
Christoph Lameter81819f02007-05-06 14:49:36 -07003925
Ezequiel Garcia2b847c32012-09-08 17:47:58 -03003926 ret = slab_alloc(s, gfpflags, caller);
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03003927
Lucas De Marchi25985ed2011-03-30 22:57:33 -03003928 /* Honor the call site pointer we received. */
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02003929 trace_kmalloc(caller, ret, size, s->size, gfpflags);
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03003930
3931 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07003932}
3933
Namhyung Kim5d1f57e2010-09-29 21:02:15 +09003934#ifdef CONFIG_NUMA
Christoph Lameter81819f02007-05-06 14:49:36 -07003935void *__kmalloc_node_track_caller(size_t size, gfp_t gfpflags,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03003936 int node, unsigned long caller)
Christoph Lameter81819f02007-05-06 14:49:36 -07003937{
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003938 struct kmem_cache *s;
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03003939 void *ret;
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003940
Christoph Lameter95a05b42013-01-10 19:14:19 +00003941 if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) {
Xiaotian Fengd3e14aa2010-04-08 17:26:44 +08003942 ret = kmalloc_large_node(size, gfpflags, node);
3943
3944 trace_kmalloc_node(caller, ret,
3945 size, PAGE_SIZE << get_order(size),
3946 gfpflags, node);
3947
3948 return ret;
3949 }
Pekka Enbergeada35e2008-02-11 22:47:46 +02003950
Christoph Lameter2c59dd62013-01-10 19:14:19 +00003951 s = kmalloc_slab(size, gfpflags);
Christoph Lameter81819f02007-05-06 14:49:36 -07003952
Satyam Sharma2408c552007-10-16 01:24:44 -07003953 if (unlikely(ZERO_OR_NULL_PTR(s)))
Christoph Lameter6cb8f912007-07-17 04:03:22 -07003954 return s;
Christoph Lameter81819f02007-05-06 14:49:36 -07003955
Ezequiel Garcia2b847c32012-09-08 17:47:58 -03003956 ret = slab_alloc_node(s, gfpflags, node, caller);
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03003957
Lucas De Marchi25985ed2011-03-30 22:57:33 -03003958 /* Honor the call site pointer we received. */
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02003959 trace_kmalloc_node(caller, ret, size, s->size, gfpflags, node);
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03003960
3961 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07003962}
Namhyung Kim5d1f57e2010-09-29 21:02:15 +09003963#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07003964
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05003965#ifdef CONFIG_SYSFS
Christoph Lameter205ab992008-04-14 19:11:40 +03003966static int count_inuse(struct page *page)
3967{
3968 return page->inuse;
3969}
3970
3971static int count_total(struct page *page)
3972{
3973 return page->objects;
3974}
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05003975#endif
Christoph Lameter205ab992008-04-14 19:11:40 +03003976
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05003977#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter434e2452007-07-17 04:03:30 -07003978static int validate_slab(struct kmem_cache *s, struct page *page,
3979 unsigned long *map)
Christoph Lameter53e15af2007-05-06 14:49:43 -07003980{
3981 void *p;
Christoph Lametera973e9d2008-03-01 13:40:44 -08003982 void *addr = page_address(page);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003983
3984 if (!check_slab(s, page) ||
3985 !on_freelist(s, page, NULL))
3986 return 0;
3987
3988 /* Now we know that a valid freelist exists */
Christoph Lameter39b26462008-04-14 19:11:30 +03003989 bitmap_zero(map, page->objects);
Christoph Lameter53e15af2007-05-06 14:49:43 -07003990
Christoph Lameter5f80b132011-04-15 14:48:13 -05003991 get_map(s, page, map);
3992 for_each_object(p, s, addr, page->objects) {
3993 if (test_bit(slab_index(p, s, addr), map))
3994 if (!check_object(s, page, p, SLUB_RED_INACTIVE))
3995 return 0;
Christoph Lameter53e15af2007-05-06 14:49:43 -07003996 }
3997
Christoph Lameter224a88b2008-04-14 19:11:31 +03003998 for_each_object(p, s, addr, page->objects)
Christoph Lameter7656c722007-05-09 02:32:40 -07003999 if (!test_bit(slab_index(p, s, addr), map))
Tero Roponen37d57442010-12-01 20:04:20 +02004000 if (!check_object(s, page, p, SLUB_RED_ACTIVE))
Christoph Lameter53e15af2007-05-06 14:49:43 -07004001 return 0;
4002 return 1;
4003}
4004
Christoph Lameter434e2452007-07-17 04:03:30 -07004005static void validate_slab_slab(struct kmem_cache *s, struct page *page,
4006 unsigned long *map)
Christoph Lameter53e15af2007-05-06 14:49:43 -07004007{
Christoph Lameter881db7f2011-06-01 12:25:53 -05004008 slab_lock(page);
4009 validate_slab(s, page, map);
4010 slab_unlock(page);
Christoph Lameter53e15af2007-05-06 14:49:43 -07004011}
4012
Christoph Lameter434e2452007-07-17 04:03:30 -07004013static int validate_slab_node(struct kmem_cache *s,
4014 struct kmem_cache_node *n, unsigned long *map)
Christoph Lameter53e15af2007-05-06 14:49:43 -07004015{
4016 unsigned long count = 0;
4017 struct page *page;
4018 unsigned long flags;
4019
4020 spin_lock_irqsave(&n->list_lock, flags);
4021
4022 list_for_each_entry(page, &n->partial, lru) {
Christoph Lameter434e2452007-07-17 04:03:30 -07004023 validate_slab_slab(s, page, map);
Christoph Lameter53e15af2007-05-06 14:49:43 -07004024 count++;
4025 }
4026 if (count != n->nr_partial)
Fabian Frederickf9f58282014-06-04 16:06:34 -07004027 pr_err("SLUB %s: %ld partial slabs counted but counter=%ld\n",
4028 s->name, count, n->nr_partial);
Christoph Lameter53e15af2007-05-06 14:49:43 -07004029
4030 if (!(s->flags & SLAB_STORE_USER))
4031 goto out;
4032
4033 list_for_each_entry(page, &n->full, lru) {
Christoph Lameter434e2452007-07-17 04:03:30 -07004034 validate_slab_slab(s, page, map);
Christoph Lameter53e15af2007-05-06 14:49:43 -07004035 count++;
4036 }
4037 if (count != atomic_long_read(&n->nr_slabs))
Fabian Frederickf9f58282014-06-04 16:06:34 -07004038 pr_err("SLUB: %s %ld slabs counted but counter=%ld\n",
4039 s->name, count, atomic_long_read(&n->nr_slabs));
Christoph Lameter53e15af2007-05-06 14:49:43 -07004040
4041out:
4042 spin_unlock_irqrestore(&n->list_lock, flags);
4043 return count;
4044}
4045
Christoph Lameter434e2452007-07-17 04:03:30 -07004046static long validate_slab_cache(struct kmem_cache *s)
Christoph Lameter53e15af2007-05-06 14:49:43 -07004047{
4048 int node;
4049 unsigned long count = 0;
Christoph Lameter205ab992008-04-14 19:11:40 +03004050 unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) *
Christoph Lameter434e2452007-07-17 04:03:30 -07004051 sizeof(unsigned long), GFP_KERNEL);
Christoph Lameterfa45dc22014-08-06 16:04:09 -07004052 struct kmem_cache_node *n;
Christoph Lameter434e2452007-07-17 04:03:30 -07004053
4054 if (!map)
4055 return -ENOMEM;
Christoph Lameter53e15af2007-05-06 14:49:43 -07004056
4057 flush_all(s);
Christoph Lameterfa45dc22014-08-06 16:04:09 -07004058 for_each_kmem_cache_node(s, node, n)
Christoph Lameter434e2452007-07-17 04:03:30 -07004059 count += validate_slab_node(s, n, map);
Christoph Lameter434e2452007-07-17 04:03:30 -07004060 kfree(map);
Christoph Lameter53e15af2007-05-06 14:49:43 -07004061 return count;
4062}
Christoph Lameter88a420e2007-05-06 14:49:45 -07004063/*
Christoph Lameter672bba32007-05-09 02:32:39 -07004064 * Generate lists of code addresses where slabcache objects are allocated
Christoph Lameter88a420e2007-05-06 14:49:45 -07004065 * and freed.
4066 */
4067
4068struct location {
4069 unsigned long count;
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03004070 unsigned long addr;
Christoph Lameter45edfa52007-05-09 02:32:45 -07004071 long long sum_time;
4072 long min_time;
4073 long max_time;
4074 long min_pid;
4075 long max_pid;
Rusty Russell174596a2009-01-01 10:12:29 +10304076 DECLARE_BITMAP(cpus, NR_CPUS);
Christoph Lameter45edfa52007-05-09 02:32:45 -07004077 nodemask_t nodes;
Christoph Lameter88a420e2007-05-06 14:49:45 -07004078};
4079
4080struct loc_track {
4081 unsigned long max;
4082 unsigned long count;
4083 struct location *loc;
4084};
4085
4086static void free_loc_track(struct loc_track *t)
4087{
4088 if (t->max)
4089 free_pages((unsigned long)t->loc,
4090 get_order(sizeof(struct location) * t->max));
4091}
4092
Christoph Lameter68dff6a2007-07-17 04:03:20 -07004093static int alloc_loc_track(struct loc_track *t, unsigned long max, gfp_t flags)
Christoph Lameter88a420e2007-05-06 14:49:45 -07004094{
4095 struct location *l;
4096 int order;
4097
Christoph Lameter88a420e2007-05-06 14:49:45 -07004098 order = get_order(sizeof(struct location) * max);
4099
Christoph Lameter68dff6a2007-07-17 04:03:20 -07004100 l = (void *)__get_free_pages(flags, order);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004101 if (!l)
4102 return 0;
4103
4104 if (t->count) {
4105 memcpy(l, t->loc, sizeof(struct location) * t->count);
4106 free_loc_track(t);
4107 }
4108 t->max = max;
4109 t->loc = l;
4110 return 1;
4111}
4112
4113static int add_location(struct loc_track *t, struct kmem_cache *s,
Christoph Lameter45edfa52007-05-09 02:32:45 -07004114 const struct track *track)
Christoph Lameter88a420e2007-05-06 14:49:45 -07004115{
4116 long start, end, pos;
4117 struct location *l;
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03004118 unsigned long caddr;
Christoph Lameter45edfa52007-05-09 02:32:45 -07004119 unsigned long age = jiffies - track->when;
Christoph Lameter88a420e2007-05-06 14:49:45 -07004120
4121 start = -1;
4122 end = t->count;
4123
4124 for ( ; ; ) {
4125 pos = start + (end - start + 1) / 2;
4126
4127 /*
4128 * There is nothing at "end". If we end up there
4129 * we need to add something to before end.
4130 */
4131 if (pos == end)
4132 break;
4133
4134 caddr = t->loc[pos].addr;
Christoph Lameter45edfa52007-05-09 02:32:45 -07004135 if (track->addr == caddr) {
4136
4137 l = &t->loc[pos];
4138 l->count++;
4139 if (track->when) {
4140 l->sum_time += age;
4141 if (age < l->min_time)
4142 l->min_time = age;
4143 if (age > l->max_time)
4144 l->max_time = age;
4145
4146 if (track->pid < l->min_pid)
4147 l->min_pid = track->pid;
4148 if (track->pid > l->max_pid)
4149 l->max_pid = track->pid;
4150
Rusty Russell174596a2009-01-01 10:12:29 +10304151 cpumask_set_cpu(track->cpu,
4152 to_cpumask(l->cpus));
Christoph Lameter45edfa52007-05-09 02:32:45 -07004153 }
4154 node_set(page_to_nid(virt_to_page(track)), l->nodes);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004155 return 1;
4156 }
4157
Christoph Lameter45edfa52007-05-09 02:32:45 -07004158 if (track->addr < caddr)
Christoph Lameter88a420e2007-05-06 14:49:45 -07004159 end = pos;
4160 else
4161 start = pos;
4162 }
4163
4164 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07004165 * Not found. Insert new tracking element.
Christoph Lameter88a420e2007-05-06 14:49:45 -07004166 */
Christoph Lameter68dff6a2007-07-17 04:03:20 -07004167 if (t->count >= t->max && !alloc_loc_track(t, 2 * t->max, GFP_ATOMIC))
Christoph Lameter88a420e2007-05-06 14:49:45 -07004168 return 0;
4169
4170 l = t->loc + pos;
4171 if (pos < t->count)
4172 memmove(l + 1, l,
4173 (t->count - pos) * sizeof(struct location));
4174 t->count++;
4175 l->count = 1;
Christoph Lameter45edfa52007-05-09 02:32:45 -07004176 l->addr = track->addr;
4177 l->sum_time = age;
4178 l->min_time = age;
4179 l->max_time = age;
4180 l->min_pid = track->pid;
4181 l->max_pid = track->pid;
Rusty Russell174596a2009-01-01 10:12:29 +10304182 cpumask_clear(to_cpumask(l->cpus));
4183 cpumask_set_cpu(track->cpu, to_cpumask(l->cpus));
Christoph Lameter45edfa52007-05-09 02:32:45 -07004184 nodes_clear(l->nodes);
4185 node_set(page_to_nid(virt_to_page(track)), l->nodes);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004186 return 1;
4187}
4188
4189static void process_slab(struct loc_track *t, struct kmem_cache *s,
Eric Dumazetbbd7d572010-03-24 22:25:47 +01004190 struct page *page, enum track_item alloc,
Namhyung Kima5dd5c12010-09-29 21:02:13 +09004191 unsigned long *map)
Christoph Lameter88a420e2007-05-06 14:49:45 -07004192{
Christoph Lametera973e9d2008-03-01 13:40:44 -08004193 void *addr = page_address(page);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004194 void *p;
4195
Christoph Lameter39b26462008-04-14 19:11:30 +03004196 bitmap_zero(map, page->objects);
Christoph Lameter5f80b132011-04-15 14:48:13 -05004197 get_map(s, page, map);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004198
Christoph Lameter224a88b2008-04-14 19:11:31 +03004199 for_each_object(p, s, addr, page->objects)
Christoph Lameter45edfa52007-05-09 02:32:45 -07004200 if (!test_bit(slab_index(p, s, addr), map))
4201 add_location(t, s, get_track(s, p, alloc));
Christoph Lameter88a420e2007-05-06 14:49:45 -07004202}
4203
4204static int list_locations(struct kmem_cache *s, char *buf,
4205 enum track_item alloc)
4206{
Harvey Harrisone374d482008-01-31 15:20:50 -08004207 int len = 0;
Christoph Lameter88a420e2007-05-06 14:49:45 -07004208 unsigned long i;
Christoph Lameter68dff6a2007-07-17 04:03:20 -07004209 struct loc_track t = { 0, 0, NULL };
Christoph Lameter88a420e2007-05-06 14:49:45 -07004210 int node;
Eric Dumazetbbd7d572010-03-24 22:25:47 +01004211 unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) *
4212 sizeof(unsigned long), GFP_KERNEL);
Christoph Lameterfa45dc22014-08-06 16:04:09 -07004213 struct kmem_cache_node *n;
Christoph Lameter88a420e2007-05-06 14:49:45 -07004214
Eric Dumazetbbd7d572010-03-24 22:25:47 +01004215 if (!map || !alloc_loc_track(&t, PAGE_SIZE / sizeof(struct location),
4216 GFP_TEMPORARY)) {
4217 kfree(map);
Christoph Lameter68dff6a2007-07-17 04:03:20 -07004218 return sprintf(buf, "Out of memory\n");
Eric Dumazetbbd7d572010-03-24 22:25:47 +01004219 }
Christoph Lameter88a420e2007-05-06 14:49:45 -07004220 /* Push back cpu slabs */
4221 flush_all(s);
4222
Christoph Lameterfa45dc22014-08-06 16:04:09 -07004223 for_each_kmem_cache_node(s, node, n) {
Christoph Lameter88a420e2007-05-06 14:49:45 -07004224 unsigned long flags;
4225 struct page *page;
4226
Christoph Lameter9e869432007-08-22 14:01:56 -07004227 if (!atomic_long_read(&n->nr_slabs))
Christoph Lameter88a420e2007-05-06 14:49:45 -07004228 continue;
4229
4230 spin_lock_irqsave(&n->list_lock, flags);
4231 list_for_each_entry(page, &n->partial, lru)
Eric Dumazetbbd7d572010-03-24 22:25:47 +01004232 process_slab(&t, s, page, alloc, map);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004233 list_for_each_entry(page, &n->full, lru)
Eric Dumazetbbd7d572010-03-24 22:25:47 +01004234 process_slab(&t, s, page, alloc, map);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004235 spin_unlock_irqrestore(&n->list_lock, flags);
4236 }
4237
4238 for (i = 0; i < t.count; i++) {
Christoph Lameter45edfa52007-05-09 02:32:45 -07004239 struct location *l = &t.loc[i];
Christoph Lameter88a420e2007-05-06 14:49:45 -07004240
Hugh Dickins9c246242008-12-09 13:14:27 -08004241 if (len > PAGE_SIZE - KSYM_SYMBOL_LEN - 100)
Christoph Lameter88a420e2007-05-06 14:49:45 -07004242 break;
Harvey Harrisone374d482008-01-31 15:20:50 -08004243 len += sprintf(buf + len, "%7ld ", l->count);
Christoph Lameter45edfa52007-05-09 02:32:45 -07004244
4245 if (l->addr)
Joe Perches62c70bc2011-01-13 15:45:52 -08004246 len += sprintf(buf + len, "%pS", (void *)l->addr);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004247 else
Harvey Harrisone374d482008-01-31 15:20:50 -08004248 len += sprintf(buf + len, "<not-available>");
Christoph Lameter45edfa52007-05-09 02:32:45 -07004249
4250 if (l->sum_time != l->min_time) {
Harvey Harrisone374d482008-01-31 15:20:50 -08004251 len += sprintf(buf + len, " age=%ld/%ld/%ld",
Roman Zippelf8bd2252008-05-01 04:34:31 -07004252 l->min_time,
4253 (long)div_u64(l->sum_time, l->count),
4254 l->max_time);
Christoph Lameter45edfa52007-05-09 02:32:45 -07004255 } else
Harvey Harrisone374d482008-01-31 15:20:50 -08004256 len += sprintf(buf + len, " age=%ld",
Christoph Lameter45edfa52007-05-09 02:32:45 -07004257 l->min_time);
4258
4259 if (l->min_pid != l->max_pid)
Harvey Harrisone374d482008-01-31 15:20:50 -08004260 len += sprintf(buf + len, " pid=%ld-%ld",
Christoph Lameter45edfa52007-05-09 02:32:45 -07004261 l->min_pid, l->max_pid);
4262 else
Harvey Harrisone374d482008-01-31 15:20:50 -08004263 len += sprintf(buf + len, " pid=%ld",
Christoph Lameter45edfa52007-05-09 02:32:45 -07004264 l->min_pid);
4265
Rusty Russell174596a2009-01-01 10:12:29 +10304266 if (num_online_cpus() > 1 &&
4267 !cpumask_empty(to_cpumask(l->cpus)) &&
Tejun Heo5024c1d2015-02-13 14:37:59 -08004268 len < PAGE_SIZE - 60)
4269 len += scnprintf(buf + len, PAGE_SIZE - len - 50,
4270 " cpus=%*pbl",
4271 cpumask_pr_args(to_cpumask(l->cpus)));
Christoph Lameter45edfa52007-05-09 02:32:45 -07004272
Christoph Lameter62bc62a2009-06-16 15:32:15 -07004273 if (nr_online_nodes > 1 && !nodes_empty(l->nodes) &&
Tejun Heo5024c1d2015-02-13 14:37:59 -08004274 len < PAGE_SIZE - 60)
4275 len += scnprintf(buf + len, PAGE_SIZE - len - 50,
4276 " nodes=%*pbl",
4277 nodemask_pr_args(&l->nodes));
Christoph Lameter45edfa52007-05-09 02:32:45 -07004278
Harvey Harrisone374d482008-01-31 15:20:50 -08004279 len += sprintf(buf + len, "\n");
Christoph Lameter88a420e2007-05-06 14:49:45 -07004280 }
4281
4282 free_loc_track(&t);
Eric Dumazetbbd7d572010-03-24 22:25:47 +01004283 kfree(map);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004284 if (!t.count)
Harvey Harrisone374d482008-01-31 15:20:50 -08004285 len += sprintf(buf, "No data\n");
4286 return len;
Christoph Lameter88a420e2007-05-06 14:49:45 -07004287}
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004288#endif
Christoph Lameter88a420e2007-05-06 14:49:45 -07004289
Christoph Lametera5a84752010-10-05 13:57:27 -05004290#ifdef SLUB_RESILIENCY_TEST
David Rientjesc07b8182014-08-06 16:04:16 -07004291static void __init resiliency_test(void)
Christoph Lametera5a84752010-10-05 13:57:27 -05004292{
4293 u8 *p;
4294
Christoph Lameter95a05b42013-01-10 19:14:19 +00004295 BUILD_BUG_ON(KMALLOC_MIN_SIZE > 16 || KMALLOC_SHIFT_HIGH < 10);
Christoph Lametera5a84752010-10-05 13:57:27 -05004296
Fabian Frederickf9f58282014-06-04 16:06:34 -07004297 pr_err("SLUB resiliency testing\n");
4298 pr_err("-----------------------\n");
4299 pr_err("A. Corruption after allocation\n");
Christoph Lametera5a84752010-10-05 13:57:27 -05004300
4301 p = kzalloc(16, GFP_KERNEL);
4302 p[16] = 0x12;
Fabian Frederickf9f58282014-06-04 16:06:34 -07004303 pr_err("\n1. kmalloc-16: Clobber Redzone/next pointer 0x12->0x%p\n\n",
4304 p + 16);
Christoph Lametera5a84752010-10-05 13:57:27 -05004305
4306 validate_slab_cache(kmalloc_caches[4]);
4307
4308 /* Hmmm... The next two are dangerous */
4309 p = kzalloc(32, GFP_KERNEL);
4310 p[32 + sizeof(void *)] = 0x34;
Fabian Frederickf9f58282014-06-04 16:06:34 -07004311 pr_err("\n2. kmalloc-32: Clobber next pointer/next slab 0x34 -> -0x%p\n",
4312 p);
4313 pr_err("If allocated object is overwritten then not detectable\n\n");
Christoph Lametera5a84752010-10-05 13:57:27 -05004314
4315 validate_slab_cache(kmalloc_caches[5]);
4316 p = kzalloc(64, GFP_KERNEL);
4317 p += 64 + (get_cycles() & 0xff) * sizeof(void *);
4318 *p = 0x56;
Fabian Frederickf9f58282014-06-04 16:06:34 -07004319 pr_err("\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n",
4320 p);
4321 pr_err("If allocated object is overwritten then not detectable\n\n");
Christoph Lametera5a84752010-10-05 13:57:27 -05004322 validate_slab_cache(kmalloc_caches[6]);
4323
Fabian Frederickf9f58282014-06-04 16:06:34 -07004324 pr_err("\nB. Corruption after free\n");
Christoph Lametera5a84752010-10-05 13:57:27 -05004325 p = kzalloc(128, GFP_KERNEL);
4326 kfree(p);
4327 *p = 0x78;
Fabian Frederickf9f58282014-06-04 16:06:34 -07004328 pr_err("1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p);
Christoph Lametera5a84752010-10-05 13:57:27 -05004329 validate_slab_cache(kmalloc_caches[7]);
4330
4331 p = kzalloc(256, GFP_KERNEL);
4332 kfree(p);
4333 p[50] = 0x9a;
Fabian Frederickf9f58282014-06-04 16:06:34 -07004334 pr_err("\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n", p);
Christoph Lametera5a84752010-10-05 13:57:27 -05004335 validate_slab_cache(kmalloc_caches[8]);
4336
4337 p = kzalloc(512, GFP_KERNEL);
4338 kfree(p);
4339 p[512] = 0xab;
Fabian Frederickf9f58282014-06-04 16:06:34 -07004340 pr_err("\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p);
Christoph Lametera5a84752010-10-05 13:57:27 -05004341 validate_slab_cache(kmalloc_caches[9]);
4342}
4343#else
4344#ifdef CONFIG_SYSFS
4345static void resiliency_test(void) {};
4346#endif
4347#endif
4348
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004349#ifdef CONFIG_SYSFS
Christoph Lameter81819f02007-05-06 14:49:36 -07004350enum slab_stat_type {
Christoph Lameter205ab992008-04-14 19:11:40 +03004351 SL_ALL, /* All slabs */
4352 SL_PARTIAL, /* Only partially allocated slabs */
4353 SL_CPU, /* Only slabs used for cpu caches */
4354 SL_OBJECTS, /* Determine allocated objects not slabs */
4355 SL_TOTAL /* Determine object capacity not slabs */
Christoph Lameter81819f02007-05-06 14:49:36 -07004356};
4357
Christoph Lameter205ab992008-04-14 19:11:40 +03004358#define SO_ALL (1 << SL_ALL)
Christoph Lameter81819f02007-05-06 14:49:36 -07004359#define SO_PARTIAL (1 << SL_PARTIAL)
4360#define SO_CPU (1 << SL_CPU)
4361#define SO_OBJECTS (1 << SL_OBJECTS)
Christoph Lameter205ab992008-04-14 19:11:40 +03004362#define SO_TOTAL (1 << SL_TOTAL)
Christoph Lameter81819f02007-05-06 14:49:36 -07004363
Cyrill Gorcunov62e5c4b2008-03-02 23:28:24 +03004364static ssize_t show_slab_objects(struct kmem_cache *s,
4365 char *buf, unsigned long flags)
Christoph Lameter81819f02007-05-06 14:49:36 -07004366{
4367 unsigned long total = 0;
Christoph Lameter81819f02007-05-06 14:49:36 -07004368 int node;
4369 int x;
4370 unsigned long *nodes;
Christoph Lameter81819f02007-05-06 14:49:36 -07004371
Chen Gange35e1a92013-07-12 08:23:48 +08004372 nodes = kzalloc(sizeof(unsigned long) * nr_node_ids, GFP_KERNEL);
Cyrill Gorcunov62e5c4b2008-03-02 23:28:24 +03004373 if (!nodes)
4374 return -ENOMEM;
Christoph Lameter81819f02007-05-06 14:49:36 -07004375
Christoph Lameter205ab992008-04-14 19:11:40 +03004376 if (flags & SO_CPU) {
4377 int cpu;
Christoph Lameter81819f02007-05-06 14:49:36 -07004378
Christoph Lameter205ab992008-04-14 19:11:40 +03004379 for_each_possible_cpu(cpu) {
Chen Gangd0e0ac92013-07-15 09:05:29 +08004380 struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab,
4381 cpu);
Christoph Lameterec3ab082012-05-09 10:09:56 -05004382 int node;
Christoph Lameter49e22582011-08-09 16:12:27 -05004383 struct page *page;
Christoph Lameterdfb4f092007-10-16 01:26:05 -07004384
Jason Low4db0c3c2015-04-15 16:14:08 -07004385 page = READ_ONCE(c->page);
Christoph Lameterec3ab082012-05-09 10:09:56 -05004386 if (!page)
4387 continue;
Christoph Lameter205ab992008-04-14 19:11:40 +03004388
Christoph Lameterec3ab082012-05-09 10:09:56 -05004389 node = page_to_nid(page);
4390 if (flags & SO_TOTAL)
4391 x = page->objects;
4392 else if (flags & SO_OBJECTS)
4393 x = page->inuse;
4394 else
4395 x = 1;
Christoph Lameter49e22582011-08-09 16:12:27 -05004396
Christoph Lameterec3ab082012-05-09 10:09:56 -05004397 total += x;
4398 nodes[node] += x;
4399
Jason Low4db0c3c2015-04-15 16:14:08 -07004400 page = READ_ONCE(c->partial);
Christoph Lameter49e22582011-08-09 16:12:27 -05004401 if (page) {
Li Zefan8afb1472013-09-10 11:43:37 +08004402 node = page_to_nid(page);
4403 if (flags & SO_TOTAL)
4404 WARN_ON_ONCE(1);
4405 else if (flags & SO_OBJECTS)
4406 WARN_ON_ONCE(1);
4407 else
4408 x = page->pages;
Eric Dumazetbc6697d2011-11-22 16:02:02 +01004409 total += x;
4410 nodes[node] += x;
Christoph Lameter49e22582011-08-09 16:12:27 -05004411 }
Christoph Lameter81819f02007-05-06 14:49:36 -07004412 }
4413 }
4414
Vladimir Davydovbfc8c902014-06-04 16:07:18 -07004415 get_online_mems();
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004416#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter205ab992008-04-14 19:11:40 +03004417 if (flags & SO_ALL) {
Christoph Lameterfa45dc22014-08-06 16:04:09 -07004418 struct kmem_cache_node *n;
4419
4420 for_each_kmem_cache_node(s, node, n) {
Christoph Lameter81819f02007-05-06 14:49:36 -07004421
Chen Gangd0e0ac92013-07-15 09:05:29 +08004422 if (flags & SO_TOTAL)
4423 x = atomic_long_read(&n->total_objects);
4424 else if (flags & SO_OBJECTS)
4425 x = atomic_long_read(&n->total_objects) -
4426 count_partial(n, count_free);
Christoph Lameter205ab992008-04-14 19:11:40 +03004427 else
4428 x = atomic_long_read(&n->nr_slabs);
4429 total += x;
4430 nodes[node] += x;
4431 }
4432
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004433 } else
4434#endif
4435 if (flags & SO_PARTIAL) {
Christoph Lameterfa45dc22014-08-06 16:04:09 -07004436 struct kmem_cache_node *n;
Christoph Lameter205ab992008-04-14 19:11:40 +03004437
Christoph Lameterfa45dc22014-08-06 16:04:09 -07004438 for_each_kmem_cache_node(s, node, n) {
Christoph Lameter205ab992008-04-14 19:11:40 +03004439 if (flags & SO_TOTAL)
4440 x = count_partial(n, count_total);
4441 else if (flags & SO_OBJECTS)
4442 x = count_partial(n, count_inuse);
Christoph Lameter81819f02007-05-06 14:49:36 -07004443 else
4444 x = n->nr_partial;
4445 total += x;
4446 nodes[node] += x;
4447 }
Christoph Lameter81819f02007-05-06 14:49:36 -07004448 }
Christoph Lameter81819f02007-05-06 14:49:36 -07004449 x = sprintf(buf, "%lu", total);
4450#ifdef CONFIG_NUMA
Christoph Lameterfa45dc22014-08-06 16:04:09 -07004451 for (node = 0; node < nr_node_ids; node++)
Christoph Lameter81819f02007-05-06 14:49:36 -07004452 if (nodes[node])
4453 x += sprintf(buf + x, " N%d=%lu",
4454 node, nodes[node]);
4455#endif
Vladimir Davydovbfc8c902014-06-04 16:07:18 -07004456 put_online_mems();
Christoph Lameter81819f02007-05-06 14:49:36 -07004457 kfree(nodes);
4458 return x + sprintf(buf + x, "\n");
4459}
4460
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004461#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter81819f02007-05-06 14:49:36 -07004462static int any_slab_objects(struct kmem_cache *s)
4463{
4464 int node;
Christoph Lameterfa45dc22014-08-06 16:04:09 -07004465 struct kmem_cache_node *n;
Christoph Lameterdfb4f092007-10-16 01:26:05 -07004466
Christoph Lameterfa45dc22014-08-06 16:04:09 -07004467 for_each_kmem_cache_node(s, node, n)
Benjamin Herrenschmidt4ea33e22008-05-06 20:42:39 -07004468 if (atomic_long_read(&n->total_objects))
Christoph Lameter81819f02007-05-06 14:49:36 -07004469 return 1;
Christoph Lameterfa45dc22014-08-06 16:04:09 -07004470
Christoph Lameter81819f02007-05-06 14:49:36 -07004471 return 0;
4472}
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004473#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07004474
4475#define to_slab_attr(n) container_of(n, struct slab_attribute, attr)
Phil Carmody497888c2011-07-14 15:07:13 +03004476#define to_slab(n) container_of(n, struct kmem_cache, kobj)
Christoph Lameter81819f02007-05-06 14:49:36 -07004477
4478struct slab_attribute {
4479 struct attribute attr;
4480 ssize_t (*show)(struct kmem_cache *s, char *buf);
4481 ssize_t (*store)(struct kmem_cache *s, const char *x, size_t count);
4482};
4483
4484#define SLAB_ATTR_RO(_name) \
Vasiliy Kulikovab067e92011-09-27 21:54:53 +04004485 static struct slab_attribute _name##_attr = \
4486 __ATTR(_name, 0400, _name##_show, NULL)
Christoph Lameter81819f02007-05-06 14:49:36 -07004487
4488#define SLAB_ATTR(_name) \
4489 static struct slab_attribute _name##_attr = \
Vasiliy Kulikovab067e92011-09-27 21:54:53 +04004490 __ATTR(_name, 0600, _name##_show, _name##_store)
Christoph Lameter81819f02007-05-06 14:49:36 -07004491
Christoph Lameter81819f02007-05-06 14:49:36 -07004492static ssize_t slab_size_show(struct kmem_cache *s, char *buf)
4493{
4494 return sprintf(buf, "%d\n", s->size);
4495}
4496SLAB_ATTR_RO(slab_size);
4497
4498static ssize_t align_show(struct kmem_cache *s, char *buf)
4499{
4500 return sprintf(buf, "%d\n", s->align);
4501}
4502SLAB_ATTR_RO(align);
4503
4504static ssize_t object_size_show(struct kmem_cache *s, char *buf)
4505{
Christoph Lameter3b0efdf2012-06-13 10:24:57 -05004506 return sprintf(buf, "%d\n", s->object_size);
Christoph Lameter81819f02007-05-06 14:49:36 -07004507}
4508SLAB_ATTR_RO(object_size);
4509
4510static ssize_t objs_per_slab_show(struct kmem_cache *s, char *buf)
4511{
Christoph Lameter834f3d12008-04-14 19:11:31 +03004512 return sprintf(buf, "%d\n", oo_objects(s->oo));
Christoph Lameter81819f02007-05-06 14:49:36 -07004513}
4514SLAB_ATTR_RO(objs_per_slab);
4515
Christoph Lameter06b285d2008-04-14 19:11:41 +03004516static ssize_t order_store(struct kmem_cache *s,
4517 const char *buf, size_t length)
4518{
Christoph Lameter0121c6192008-04-29 16:11:12 -07004519 unsigned long order;
4520 int err;
4521
Jingoo Han3dbb95f2013-09-11 14:20:25 -07004522 err = kstrtoul(buf, 10, &order);
Christoph Lameter0121c6192008-04-29 16:11:12 -07004523 if (err)
4524 return err;
Christoph Lameter06b285d2008-04-14 19:11:41 +03004525
4526 if (order > slub_max_order || order < slub_min_order)
4527 return -EINVAL;
4528
4529 calculate_sizes(s, order);
4530 return length;
4531}
4532
Christoph Lameter81819f02007-05-06 14:49:36 -07004533static ssize_t order_show(struct kmem_cache *s, char *buf)
4534{
Christoph Lameter834f3d12008-04-14 19:11:31 +03004535 return sprintf(buf, "%d\n", oo_order(s->oo));
Christoph Lameter81819f02007-05-06 14:49:36 -07004536}
Christoph Lameter06b285d2008-04-14 19:11:41 +03004537SLAB_ATTR(order);
Christoph Lameter81819f02007-05-06 14:49:36 -07004538
David Rientjes73d342b2009-02-22 17:40:09 -08004539static ssize_t min_partial_show(struct kmem_cache *s, char *buf)
4540{
4541 return sprintf(buf, "%lu\n", s->min_partial);
4542}
4543
4544static ssize_t min_partial_store(struct kmem_cache *s, const char *buf,
4545 size_t length)
4546{
4547 unsigned long min;
4548 int err;
4549
Jingoo Han3dbb95f2013-09-11 14:20:25 -07004550 err = kstrtoul(buf, 10, &min);
David Rientjes73d342b2009-02-22 17:40:09 -08004551 if (err)
4552 return err;
4553
David Rientjesc0bdb232009-02-25 09:16:35 +02004554 set_min_partial(s, min);
David Rientjes73d342b2009-02-22 17:40:09 -08004555 return length;
4556}
4557SLAB_ATTR(min_partial);
4558
Christoph Lameter49e22582011-08-09 16:12:27 -05004559static ssize_t cpu_partial_show(struct kmem_cache *s, char *buf)
4560{
4561 return sprintf(buf, "%u\n", s->cpu_partial);
4562}
4563
4564static ssize_t cpu_partial_store(struct kmem_cache *s, const char *buf,
4565 size_t length)
4566{
4567 unsigned long objects;
4568 int err;
4569
Jingoo Han3dbb95f2013-09-11 14:20:25 -07004570 err = kstrtoul(buf, 10, &objects);
Christoph Lameter49e22582011-08-09 16:12:27 -05004571 if (err)
4572 return err;
Joonsoo Kim345c9052013-06-19 14:05:52 +09004573 if (objects && !kmem_cache_has_cpu_partial(s))
David Rientjes74ee4ef2012-01-09 13:19:45 -08004574 return -EINVAL;
Christoph Lameter49e22582011-08-09 16:12:27 -05004575
4576 s->cpu_partial = objects;
4577 flush_all(s);
4578 return length;
4579}
4580SLAB_ATTR(cpu_partial);
4581
Christoph Lameter81819f02007-05-06 14:49:36 -07004582static ssize_t ctor_show(struct kmem_cache *s, char *buf)
4583{
Joe Perches62c70bc2011-01-13 15:45:52 -08004584 if (!s->ctor)
4585 return 0;
4586 return sprintf(buf, "%pS\n", s->ctor);
Christoph Lameter81819f02007-05-06 14:49:36 -07004587}
4588SLAB_ATTR_RO(ctor);
4589
Christoph Lameter81819f02007-05-06 14:49:36 -07004590static ssize_t aliases_show(struct kmem_cache *s, char *buf)
4591{
Gu Zheng4307c142014-08-06 16:04:51 -07004592 return sprintf(buf, "%d\n", s->refcount < 0 ? 0 : s->refcount - 1);
Christoph Lameter81819f02007-05-06 14:49:36 -07004593}
4594SLAB_ATTR_RO(aliases);
4595
Christoph Lameter81819f02007-05-06 14:49:36 -07004596static ssize_t partial_show(struct kmem_cache *s, char *buf)
4597{
Christoph Lameterd9acf4b2008-02-15 15:22:21 -08004598 return show_slab_objects(s, buf, SO_PARTIAL);
Christoph Lameter81819f02007-05-06 14:49:36 -07004599}
4600SLAB_ATTR_RO(partial);
4601
4602static ssize_t cpu_slabs_show(struct kmem_cache *s, char *buf)
4603{
Christoph Lameterd9acf4b2008-02-15 15:22:21 -08004604 return show_slab_objects(s, buf, SO_CPU);
Christoph Lameter81819f02007-05-06 14:49:36 -07004605}
4606SLAB_ATTR_RO(cpu_slabs);
4607
4608static ssize_t objects_show(struct kmem_cache *s, char *buf)
4609{
Christoph Lameter205ab992008-04-14 19:11:40 +03004610 return show_slab_objects(s, buf, SO_ALL|SO_OBJECTS);
Christoph Lameter81819f02007-05-06 14:49:36 -07004611}
4612SLAB_ATTR_RO(objects);
4613
Christoph Lameter205ab992008-04-14 19:11:40 +03004614static ssize_t objects_partial_show(struct kmem_cache *s, char *buf)
4615{
4616 return show_slab_objects(s, buf, SO_PARTIAL|SO_OBJECTS);
4617}
4618SLAB_ATTR_RO(objects_partial);
4619
Christoph Lameter49e22582011-08-09 16:12:27 -05004620static ssize_t slabs_cpu_partial_show(struct kmem_cache *s, char *buf)
4621{
4622 int objects = 0;
4623 int pages = 0;
4624 int cpu;
4625 int len;
4626
4627 for_each_online_cpu(cpu) {
4628 struct page *page = per_cpu_ptr(s->cpu_slab, cpu)->partial;
4629
4630 if (page) {
4631 pages += page->pages;
4632 objects += page->pobjects;
4633 }
4634 }
4635
4636 len = sprintf(buf, "%d(%d)", objects, pages);
4637
4638#ifdef CONFIG_SMP
4639 for_each_online_cpu(cpu) {
4640 struct page *page = per_cpu_ptr(s->cpu_slab, cpu) ->partial;
4641
4642 if (page && len < PAGE_SIZE - 20)
4643 len += sprintf(buf + len, " C%d=%d(%d)", cpu,
4644 page->pobjects, page->pages);
4645 }
4646#endif
4647 return len + sprintf(buf + len, "\n");
4648}
4649SLAB_ATTR_RO(slabs_cpu_partial);
4650
Christoph Lameter81819f02007-05-06 14:49:36 -07004651static ssize_t reclaim_account_show(struct kmem_cache *s, char *buf)
4652{
4653 return sprintf(buf, "%d\n", !!(s->flags & SLAB_RECLAIM_ACCOUNT));
4654}
4655
4656static ssize_t reclaim_account_store(struct kmem_cache *s,
4657 const char *buf, size_t length)
4658{
4659 s->flags &= ~SLAB_RECLAIM_ACCOUNT;
4660 if (buf[0] == '1')
4661 s->flags |= SLAB_RECLAIM_ACCOUNT;
4662 return length;
4663}
4664SLAB_ATTR(reclaim_account);
4665
4666static ssize_t hwcache_align_show(struct kmem_cache *s, char *buf)
4667{
Christoph Lameter5af60832007-05-06 14:49:56 -07004668 return sprintf(buf, "%d\n", !!(s->flags & SLAB_HWCACHE_ALIGN));
Christoph Lameter81819f02007-05-06 14:49:36 -07004669}
4670SLAB_ATTR_RO(hwcache_align);
4671
4672#ifdef CONFIG_ZONE_DMA
4673static ssize_t cache_dma_show(struct kmem_cache *s, char *buf)
4674{
4675 return sprintf(buf, "%d\n", !!(s->flags & SLAB_CACHE_DMA));
4676}
4677SLAB_ATTR_RO(cache_dma);
4678#endif
4679
4680static ssize_t destroy_by_rcu_show(struct kmem_cache *s, char *buf)
4681{
4682 return sprintf(buf, "%d\n", !!(s->flags & SLAB_DESTROY_BY_RCU));
4683}
4684SLAB_ATTR_RO(destroy_by_rcu);
4685
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08004686static ssize_t reserved_show(struct kmem_cache *s, char *buf)
4687{
4688 return sprintf(buf, "%d\n", s->reserved);
4689}
4690SLAB_ATTR_RO(reserved);
4691
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004692#ifdef CONFIG_SLUB_DEBUG
Christoph Lametera5a84752010-10-05 13:57:27 -05004693static ssize_t slabs_show(struct kmem_cache *s, char *buf)
4694{
4695 return show_slab_objects(s, buf, SO_ALL);
4696}
4697SLAB_ATTR_RO(slabs);
4698
4699static ssize_t total_objects_show(struct kmem_cache *s, char *buf)
4700{
4701 return show_slab_objects(s, buf, SO_ALL|SO_TOTAL);
4702}
4703SLAB_ATTR_RO(total_objects);
4704
4705static ssize_t sanity_checks_show(struct kmem_cache *s, char *buf)
4706{
4707 return sprintf(buf, "%d\n", !!(s->flags & SLAB_DEBUG_FREE));
4708}
4709
4710static ssize_t sanity_checks_store(struct kmem_cache *s,
4711 const char *buf, size_t length)
4712{
4713 s->flags &= ~SLAB_DEBUG_FREE;
Christoph Lameterb789ef52011-06-01 12:25:49 -05004714 if (buf[0] == '1') {
4715 s->flags &= ~__CMPXCHG_DOUBLE;
Christoph Lametera5a84752010-10-05 13:57:27 -05004716 s->flags |= SLAB_DEBUG_FREE;
Christoph Lameterb789ef52011-06-01 12:25:49 -05004717 }
Christoph Lametera5a84752010-10-05 13:57:27 -05004718 return length;
4719}
4720SLAB_ATTR(sanity_checks);
4721
4722static ssize_t trace_show(struct kmem_cache *s, char *buf)
4723{
4724 return sprintf(buf, "%d\n", !!(s->flags & SLAB_TRACE));
4725}
4726
4727static ssize_t trace_store(struct kmem_cache *s, const char *buf,
4728 size_t length)
4729{
Christoph Lameterc9e16132014-10-09 15:26:11 -07004730 /*
4731 * Tracing a merged cache is going to give confusing results
4732 * as well as cause other issues like converting a mergeable
4733 * cache into an umergeable one.
4734 */
4735 if (s->refcount > 1)
4736 return -EINVAL;
4737
Christoph Lametera5a84752010-10-05 13:57:27 -05004738 s->flags &= ~SLAB_TRACE;
Christoph Lameterb789ef52011-06-01 12:25:49 -05004739 if (buf[0] == '1') {
4740 s->flags &= ~__CMPXCHG_DOUBLE;
Christoph Lametera5a84752010-10-05 13:57:27 -05004741 s->flags |= SLAB_TRACE;
Christoph Lameterb789ef52011-06-01 12:25:49 -05004742 }
Christoph Lametera5a84752010-10-05 13:57:27 -05004743 return length;
4744}
4745SLAB_ATTR(trace);
4746
Christoph Lameter81819f02007-05-06 14:49:36 -07004747static ssize_t red_zone_show(struct kmem_cache *s, char *buf)
4748{
4749 return sprintf(buf, "%d\n", !!(s->flags & SLAB_RED_ZONE));
4750}
4751
4752static ssize_t red_zone_store(struct kmem_cache *s,
4753 const char *buf, size_t length)
4754{
4755 if (any_slab_objects(s))
4756 return -EBUSY;
4757
4758 s->flags &= ~SLAB_RED_ZONE;
Christoph Lameterb789ef52011-06-01 12:25:49 -05004759 if (buf[0] == '1') {
4760 s->flags &= ~__CMPXCHG_DOUBLE;
Christoph Lameter81819f02007-05-06 14:49:36 -07004761 s->flags |= SLAB_RED_ZONE;
Christoph Lameterb789ef52011-06-01 12:25:49 -05004762 }
Christoph Lameter06b285d2008-04-14 19:11:41 +03004763 calculate_sizes(s, -1);
Christoph Lameter81819f02007-05-06 14:49:36 -07004764 return length;
4765}
4766SLAB_ATTR(red_zone);
4767
4768static ssize_t poison_show(struct kmem_cache *s, char *buf)
4769{
4770 return sprintf(buf, "%d\n", !!(s->flags & SLAB_POISON));
4771}
4772
4773static ssize_t poison_store(struct kmem_cache *s,
4774 const char *buf, size_t length)
4775{
4776 if (any_slab_objects(s))
4777 return -EBUSY;
4778
4779 s->flags &= ~SLAB_POISON;
Christoph Lameterb789ef52011-06-01 12:25:49 -05004780 if (buf[0] == '1') {
4781 s->flags &= ~__CMPXCHG_DOUBLE;
Christoph Lameter81819f02007-05-06 14:49:36 -07004782 s->flags |= SLAB_POISON;
Christoph Lameterb789ef52011-06-01 12:25:49 -05004783 }
Christoph Lameter06b285d2008-04-14 19:11:41 +03004784 calculate_sizes(s, -1);
Christoph Lameter81819f02007-05-06 14:49:36 -07004785 return length;
4786}
4787SLAB_ATTR(poison);
4788
4789static ssize_t store_user_show(struct kmem_cache *s, char *buf)
4790{
4791 return sprintf(buf, "%d\n", !!(s->flags & SLAB_STORE_USER));
4792}
4793
4794static ssize_t store_user_store(struct kmem_cache *s,
4795 const char *buf, size_t length)
4796{
4797 if (any_slab_objects(s))
4798 return -EBUSY;
4799
4800 s->flags &= ~SLAB_STORE_USER;
Christoph Lameterb789ef52011-06-01 12:25:49 -05004801 if (buf[0] == '1') {
4802 s->flags &= ~__CMPXCHG_DOUBLE;
Christoph Lameter81819f02007-05-06 14:49:36 -07004803 s->flags |= SLAB_STORE_USER;
Christoph Lameterb789ef52011-06-01 12:25:49 -05004804 }
Christoph Lameter06b285d2008-04-14 19:11:41 +03004805 calculate_sizes(s, -1);
Christoph Lameter81819f02007-05-06 14:49:36 -07004806 return length;
4807}
4808SLAB_ATTR(store_user);
4809
Christoph Lameter53e15af2007-05-06 14:49:43 -07004810static ssize_t validate_show(struct kmem_cache *s, char *buf)
4811{
4812 return 0;
4813}
4814
4815static ssize_t validate_store(struct kmem_cache *s,
4816 const char *buf, size_t length)
4817{
Christoph Lameter434e2452007-07-17 04:03:30 -07004818 int ret = -EINVAL;
4819
4820 if (buf[0] == '1') {
4821 ret = validate_slab_cache(s);
4822 if (ret >= 0)
4823 ret = length;
4824 }
4825 return ret;
Christoph Lameter53e15af2007-05-06 14:49:43 -07004826}
4827SLAB_ATTR(validate);
Christoph Lametera5a84752010-10-05 13:57:27 -05004828
4829static ssize_t alloc_calls_show(struct kmem_cache *s, char *buf)
4830{
4831 if (!(s->flags & SLAB_STORE_USER))
4832 return -ENOSYS;
4833 return list_locations(s, buf, TRACK_ALLOC);
4834}
4835SLAB_ATTR_RO(alloc_calls);
4836
4837static ssize_t free_calls_show(struct kmem_cache *s, char *buf)
4838{
4839 if (!(s->flags & SLAB_STORE_USER))
4840 return -ENOSYS;
4841 return list_locations(s, buf, TRACK_FREE);
4842}
4843SLAB_ATTR_RO(free_calls);
4844#endif /* CONFIG_SLUB_DEBUG */
4845
4846#ifdef CONFIG_FAILSLAB
4847static ssize_t failslab_show(struct kmem_cache *s, char *buf)
4848{
4849 return sprintf(buf, "%d\n", !!(s->flags & SLAB_FAILSLAB));
4850}
4851
4852static ssize_t failslab_store(struct kmem_cache *s, const char *buf,
4853 size_t length)
4854{
Christoph Lameterc9e16132014-10-09 15:26:11 -07004855 if (s->refcount > 1)
4856 return -EINVAL;
4857
Christoph Lametera5a84752010-10-05 13:57:27 -05004858 s->flags &= ~SLAB_FAILSLAB;
4859 if (buf[0] == '1')
4860 s->flags |= SLAB_FAILSLAB;
4861 return length;
4862}
4863SLAB_ATTR(failslab);
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004864#endif
Christoph Lameter53e15af2007-05-06 14:49:43 -07004865
Christoph Lameter2086d262007-05-06 14:49:46 -07004866static ssize_t shrink_show(struct kmem_cache *s, char *buf)
4867{
4868 return 0;
4869}
4870
4871static ssize_t shrink_store(struct kmem_cache *s,
4872 const char *buf, size_t length)
4873{
Vladimir Davydov832f37f2015-02-12 14:59:41 -08004874 if (buf[0] == '1')
4875 kmem_cache_shrink(s);
4876 else
Christoph Lameter2086d262007-05-06 14:49:46 -07004877 return -EINVAL;
4878 return length;
4879}
4880SLAB_ATTR(shrink);
4881
Christoph Lameter81819f02007-05-06 14:49:36 -07004882#ifdef CONFIG_NUMA
Christoph Lameter98246012008-01-07 23:20:26 -08004883static ssize_t remote_node_defrag_ratio_show(struct kmem_cache *s, char *buf)
Christoph Lameter81819f02007-05-06 14:49:36 -07004884{
Christoph Lameter98246012008-01-07 23:20:26 -08004885 return sprintf(buf, "%d\n", s->remote_node_defrag_ratio / 10);
Christoph Lameter81819f02007-05-06 14:49:36 -07004886}
4887
Christoph Lameter98246012008-01-07 23:20:26 -08004888static ssize_t remote_node_defrag_ratio_store(struct kmem_cache *s,
Christoph Lameter81819f02007-05-06 14:49:36 -07004889 const char *buf, size_t length)
4890{
Christoph Lameter0121c6192008-04-29 16:11:12 -07004891 unsigned long ratio;
4892 int err;
Christoph Lameter81819f02007-05-06 14:49:36 -07004893
Jingoo Han3dbb95f2013-09-11 14:20:25 -07004894 err = kstrtoul(buf, 10, &ratio);
Christoph Lameter0121c6192008-04-29 16:11:12 -07004895 if (err)
4896 return err;
4897
Christoph Lametere2cb96b2008-08-19 08:51:22 -05004898 if (ratio <= 100)
Christoph Lameter0121c6192008-04-29 16:11:12 -07004899 s->remote_node_defrag_ratio = ratio * 10;
4900
Christoph Lameter81819f02007-05-06 14:49:36 -07004901 return length;
4902}
Christoph Lameter98246012008-01-07 23:20:26 -08004903SLAB_ATTR(remote_node_defrag_ratio);
Christoph Lameter81819f02007-05-06 14:49:36 -07004904#endif
4905
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004906#ifdef CONFIG_SLUB_STATS
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004907static int show_stat(struct kmem_cache *s, char *buf, enum stat_item si)
4908{
4909 unsigned long sum = 0;
4910 int cpu;
4911 int len;
4912 int *data = kmalloc(nr_cpu_ids * sizeof(int), GFP_KERNEL);
4913
4914 if (!data)
4915 return -ENOMEM;
4916
4917 for_each_online_cpu(cpu) {
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06004918 unsigned x = per_cpu_ptr(s->cpu_slab, cpu)->stat[si];
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004919
4920 data[cpu] = x;
4921 sum += x;
4922 }
4923
4924 len = sprintf(buf, "%lu", sum);
4925
Christoph Lameter50ef37b2008-04-14 18:52:05 +03004926#ifdef CONFIG_SMP
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004927 for_each_online_cpu(cpu) {
4928 if (data[cpu] && len < PAGE_SIZE - 20)
Christoph Lameter50ef37b2008-04-14 18:52:05 +03004929 len += sprintf(buf + len, " C%d=%u", cpu, data[cpu]);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004930 }
Christoph Lameter50ef37b2008-04-14 18:52:05 +03004931#endif
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004932 kfree(data);
4933 return len + sprintf(buf + len, "\n");
4934}
4935
David Rientjes78eb00c2009-10-15 02:20:22 -07004936static void clear_stat(struct kmem_cache *s, enum stat_item si)
4937{
4938 int cpu;
4939
4940 for_each_online_cpu(cpu)
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06004941 per_cpu_ptr(s->cpu_slab, cpu)->stat[si] = 0;
David Rientjes78eb00c2009-10-15 02:20:22 -07004942}
4943
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004944#define STAT_ATTR(si, text) \
4945static ssize_t text##_show(struct kmem_cache *s, char *buf) \
4946{ \
4947 return show_stat(s, buf, si); \
4948} \
David Rientjes78eb00c2009-10-15 02:20:22 -07004949static ssize_t text##_store(struct kmem_cache *s, \
4950 const char *buf, size_t length) \
4951{ \
4952 if (buf[0] != '0') \
4953 return -EINVAL; \
4954 clear_stat(s, si); \
4955 return length; \
4956} \
4957SLAB_ATTR(text); \
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004958
4959STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath);
4960STAT_ATTR(ALLOC_SLOWPATH, alloc_slowpath);
4961STAT_ATTR(FREE_FASTPATH, free_fastpath);
4962STAT_ATTR(FREE_SLOWPATH, free_slowpath);
4963STAT_ATTR(FREE_FROZEN, free_frozen);
4964STAT_ATTR(FREE_ADD_PARTIAL, free_add_partial);
4965STAT_ATTR(FREE_REMOVE_PARTIAL, free_remove_partial);
4966STAT_ATTR(ALLOC_FROM_PARTIAL, alloc_from_partial);
4967STAT_ATTR(ALLOC_SLAB, alloc_slab);
4968STAT_ATTR(ALLOC_REFILL, alloc_refill);
Christoph Lametere36a2652011-06-01 12:25:57 -05004969STAT_ATTR(ALLOC_NODE_MISMATCH, alloc_node_mismatch);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004970STAT_ATTR(FREE_SLAB, free_slab);
4971STAT_ATTR(CPUSLAB_FLUSH, cpuslab_flush);
4972STAT_ATTR(DEACTIVATE_FULL, deactivate_full);
4973STAT_ATTR(DEACTIVATE_EMPTY, deactivate_empty);
4974STAT_ATTR(DEACTIVATE_TO_HEAD, deactivate_to_head);
4975STAT_ATTR(DEACTIVATE_TO_TAIL, deactivate_to_tail);
4976STAT_ATTR(DEACTIVATE_REMOTE_FREES, deactivate_remote_frees);
Christoph Lameter03e404a2011-06-01 12:25:58 -05004977STAT_ATTR(DEACTIVATE_BYPASS, deactivate_bypass);
Christoph Lameter65c33762008-04-14 19:11:40 +03004978STAT_ATTR(ORDER_FALLBACK, order_fallback);
Christoph Lameterb789ef52011-06-01 12:25:49 -05004979STAT_ATTR(CMPXCHG_DOUBLE_CPU_FAIL, cmpxchg_double_cpu_fail);
4980STAT_ATTR(CMPXCHG_DOUBLE_FAIL, cmpxchg_double_fail);
Christoph Lameter49e22582011-08-09 16:12:27 -05004981STAT_ATTR(CPU_PARTIAL_ALLOC, cpu_partial_alloc);
4982STAT_ATTR(CPU_PARTIAL_FREE, cpu_partial_free);
Alex Shi8028dce2012-02-03 23:34:56 +08004983STAT_ATTR(CPU_PARTIAL_NODE, cpu_partial_node);
4984STAT_ATTR(CPU_PARTIAL_DRAIN, cpu_partial_drain);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08004985#endif
4986
Pekka Enberg06428782008-01-07 23:20:27 -08004987static struct attribute *slab_attrs[] = {
Christoph Lameter81819f02007-05-06 14:49:36 -07004988 &slab_size_attr.attr,
4989 &object_size_attr.attr,
4990 &objs_per_slab_attr.attr,
4991 &order_attr.attr,
David Rientjes73d342b2009-02-22 17:40:09 -08004992 &min_partial_attr.attr,
Christoph Lameter49e22582011-08-09 16:12:27 -05004993 &cpu_partial_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07004994 &objects_attr.attr,
Christoph Lameter205ab992008-04-14 19:11:40 +03004995 &objects_partial_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07004996 &partial_attr.attr,
4997 &cpu_slabs_attr.attr,
4998 &ctor_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07004999 &aliases_attr.attr,
5000 &align_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07005001 &hwcache_align_attr.attr,
5002 &reclaim_account_attr.attr,
5003 &destroy_by_rcu_attr.attr,
Christoph Lametera5a84752010-10-05 13:57:27 -05005004 &shrink_attr.attr,
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08005005 &reserved_attr.attr,
Christoph Lameter49e22582011-08-09 16:12:27 -05005006 &slabs_cpu_partial_attr.attr,
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05005007#ifdef CONFIG_SLUB_DEBUG
Christoph Lametera5a84752010-10-05 13:57:27 -05005008 &total_objects_attr.attr,
5009 &slabs_attr.attr,
5010 &sanity_checks_attr.attr,
5011 &trace_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07005012 &red_zone_attr.attr,
5013 &poison_attr.attr,
5014 &store_user_attr.attr,
Christoph Lameter53e15af2007-05-06 14:49:43 -07005015 &validate_attr.attr,
Christoph Lameter88a420e2007-05-06 14:49:45 -07005016 &alloc_calls_attr.attr,
5017 &free_calls_attr.attr,
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05005018#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07005019#ifdef CONFIG_ZONE_DMA
5020 &cache_dma_attr.attr,
5021#endif
5022#ifdef CONFIG_NUMA
Christoph Lameter98246012008-01-07 23:20:26 -08005023 &remote_node_defrag_ratio_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07005024#endif
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08005025#ifdef CONFIG_SLUB_STATS
5026 &alloc_fastpath_attr.attr,
5027 &alloc_slowpath_attr.attr,
5028 &free_fastpath_attr.attr,
5029 &free_slowpath_attr.attr,
5030 &free_frozen_attr.attr,
5031 &free_add_partial_attr.attr,
5032 &free_remove_partial_attr.attr,
5033 &alloc_from_partial_attr.attr,
5034 &alloc_slab_attr.attr,
5035 &alloc_refill_attr.attr,
Christoph Lametere36a2652011-06-01 12:25:57 -05005036 &alloc_node_mismatch_attr.attr,
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08005037 &free_slab_attr.attr,
5038 &cpuslab_flush_attr.attr,
5039 &deactivate_full_attr.attr,
5040 &deactivate_empty_attr.attr,
5041 &deactivate_to_head_attr.attr,
5042 &deactivate_to_tail_attr.attr,
5043 &deactivate_remote_frees_attr.attr,
Christoph Lameter03e404a2011-06-01 12:25:58 -05005044 &deactivate_bypass_attr.attr,
Christoph Lameter65c33762008-04-14 19:11:40 +03005045 &order_fallback_attr.attr,
Christoph Lameterb789ef52011-06-01 12:25:49 -05005046 &cmpxchg_double_fail_attr.attr,
5047 &cmpxchg_double_cpu_fail_attr.attr,
Christoph Lameter49e22582011-08-09 16:12:27 -05005048 &cpu_partial_alloc_attr.attr,
5049 &cpu_partial_free_attr.attr,
Alex Shi8028dce2012-02-03 23:34:56 +08005050 &cpu_partial_node_attr.attr,
5051 &cpu_partial_drain_attr.attr,
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08005052#endif
Dmitry Monakhov4c13dd32010-02-26 09:36:12 +03005053#ifdef CONFIG_FAILSLAB
5054 &failslab_attr.attr,
5055#endif
5056
Christoph Lameter81819f02007-05-06 14:49:36 -07005057 NULL
5058};
5059
5060static struct attribute_group slab_attr_group = {
5061 .attrs = slab_attrs,
5062};
5063
5064static ssize_t slab_attr_show(struct kobject *kobj,
5065 struct attribute *attr,
5066 char *buf)
5067{
5068 struct slab_attribute *attribute;
5069 struct kmem_cache *s;
5070 int err;
5071
5072 attribute = to_slab_attr(attr);
5073 s = to_slab(kobj);
5074
5075 if (!attribute->show)
5076 return -EIO;
5077
5078 err = attribute->show(s, buf);
5079
5080 return err;
5081}
5082
5083static ssize_t slab_attr_store(struct kobject *kobj,
5084 struct attribute *attr,
5085 const char *buf, size_t len)
5086{
5087 struct slab_attribute *attribute;
5088 struct kmem_cache *s;
5089 int err;
5090
5091 attribute = to_slab_attr(attr);
5092 s = to_slab(kobj);
5093
5094 if (!attribute->store)
5095 return -EIO;
5096
5097 err = attribute->store(s, buf, len);
Glauber Costa107dab52012-12-18 14:23:05 -08005098#ifdef CONFIG_MEMCG_KMEM
5099 if (slab_state >= FULL && err >= 0 && is_root_cache(s)) {
Vladimir Davydov426589f2015-02-12 14:59:23 -08005100 struct kmem_cache *c;
Christoph Lameter81819f02007-05-06 14:49:36 -07005101
Glauber Costa107dab52012-12-18 14:23:05 -08005102 mutex_lock(&slab_mutex);
5103 if (s->max_attr_size < len)
5104 s->max_attr_size = len;
5105
Glauber Costaebe945c2012-12-18 14:23:10 -08005106 /*
5107 * This is a best effort propagation, so this function's return
5108 * value will be determined by the parent cache only. This is
5109 * basically because not all attributes will have a well
5110 * defined semantics for rollbacks - most of the actions will
5111 * have permanent effects.
5112 *
5113 * Returning the error value of any of the children that fail
5114 * is not 100 % defined, in the sense that users seeing the
5115 * error code won't be able to know anything about the state of
5116 * the cache.
5117 *
5118 * Only returning the error code for the parent cache at least
5119 * has well defined semantics. The cache being written to
5120 * directly either failed or succeeded, in which case we loop
5121 * through the descendants with best-effort propagation.
5122 */
Vladimir Davydov426589f2015-02-12 14:59:23 -08005123 for_each_memcg_cache(c, s)
5124 attribute->store(c, buf, len);
Glauber Costa107dab52012-12-18 14:23:05 -08005125 mutex_unlock(&slab_mutex);
5126 }
5127#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07005128 return err;
5129}
5130
Glauber Costa107dab52012-12-18 14:23:05 -08005131static void memcg_propagate_slab_attrs(struct kmem_cache *s)
5132{
5133#ifdef CONFIG_MEMCG_KMEM
5134 int i;
5135 char *buffer = NULL;
Vladimir Davydov93030d82014-05-06 12:49:59 -07005136 struct kmem_cache *root_cache;
Glauber Costa107dab52012-12-18 14:23:05 -08005137
Vladimir Davydov93030d82014-05-06 12:49:59 -07005138 if (is_root_cache(s))
Glauber Costa107dab52012-12-18 14:23:05 -08005139 return;
5140
Vladimir Davydovf7ce3192015-02-12 14:59:20 -08005141 root_cache = s->memcg_params.root_cache;
Vladimir Davydov93030d82014-05-06 12:49:59 -07005142
Glauber Costa107dab52012-12-18 14:23:05 -08005143 /*
5144 * This mean this cache had no attribute written. Therefore, no point
5145 * in copying default values around
5146 */
Vladimir Davydov93030d82014-05-06 12:49:59 -07005147 if (!root_cache->max_attr_size)
Glauber Costa107dab52012-12-18 14:23:05 -08005148 return;
5149
5150 for (i = 0; i < ARRAY_SIZE(slab_attrs); i++) {
5151 char mbuf[64];
5152 char *buf;
5153 struct slab_attribute *attr = to_slab_attr(slab_attrs[i]);
5154
5155 if (!attr || !attr->store || !attr->show)
5156 continue;
5157
5158 /*
5159 * It is really bad that we have to allocate here, so we will
5160 * do it only as a fallback. If we actually allocate, though,
5161 * we can just use the allocated buffer until the end.
5162 *
5163 * Most of the slub attributes will tend to be very small in
5164 * size, but sysfs allows buffers up to a page, so they can
5165 * theoretically happen.
5166 */
5167 if (buffer)
5168 buf = buffer;
Vladimir Davydov93030d82014-05-06 12:49:59 -07005169 else if (root_cache->max_attr_size < ARRAY_SIZE(mbuf))
Glauber Costa107dab52012-12-18 14:23:05 -08005170 buf = mbuf;
5171 else {
5172 buffer = (char *) get_zeroed_page(GFP_KERNEL);
5173 if (WARN_ON(!buffer))
5174 continue;
5175 buf = buffer;
5176 }
5177
Vladimir Davydov93030d82014-05-06 12:49:59 -07005178 attr->show(root_cache, buf);
Glauber Costa107dab52012-12-18 14:23:05 -08005179 attr->store(s, buf, strlen(buf));
5180 }
5181
5182 if (buffer)
5183 free_page((unsigned long)buffer);
5184#endif
5185}
5186
Christoph Lameter41a21282014-05-06 12:50:08 -07005187static void kmem_cache_release(struct kobject *k)
5188{
5189 slab_kmem_cache_release(to_slab(k));
5190}
5191
Emese Revfy52cf25d2010-01-19 02:58:23 +01005192static const struct sysfs_ops slab_sysfs_ops = {
Christoph Lameter81819f02007-05-06 14:49:36 -07005193 .show = slab_attr_show,
5194 .store = slab_attr_store,
5195};
5196
5197static struct kobj_type slab_ktype = {
5198 .sysfs_ops = &slab_sysfs_ops,
Christoph Lameter41a21282014-05-06 12:50:08 -07005199 .release = kmem_cache_release,
Christoph Lameter81819f02007-05-06 14:49:36 -07005200};
5201
5202static int uevent_filter(struct kset *kset, struct kobject *kobj)
5203{
5204 struct kobj_type *ktype = get_ktype(kobj);
5205
5206 if (ktype == &slab_ktype)
5207 return 1;
5208 return 0;
5209}
5210
Emese Revfy9cd43612009-12-31 14:52:51 +01005211static const struct kset_uevent_ops slab_uevent_ops = {
Christoph Lameter81819f02007-05-06 14:49:36 -07005212 .filter = uevent_filter,
5213};
5214
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06005215static struct kset *slab_kset;
Christoph Lameter81819f02007-05-06 14:49:36 -07005216
Vladimir Davydov9a417072014-04-07 15:39:31 -07005217static inline struct kset *cache_kset(struct kmem_cache *s)
5218{
5219#ifdef CONFIG_MEMCG_KMEM
5220 if (!is_root_cache(s))
Vladimir Davydovf7ce3192015-02-12 14:59:20 -08005221 return s->memcg_params.root_cache->memcg_kset;
Vladimir Davydov9a417072014-04-07 15:39:31 -07005222#endif
5223 return slab_kset;
5224}
5225
Christoph Lameter81819f02007-05-06 14:49:36 -07005226#define ID_STR_LENGTH 64
5227
5228/* Create a unique string id for a slab cache:
Christoph Lameter6446faa2008-02-15 23:45:26 -08005229 *
5230 * Format :[flags-]size
Christoph Lameter81819f02007-05-06 14:49:36 -07005231 */
5232static char *create_unique_id(struct kmem_cache *s)
5233{
5234 char *name = kmalloc(ID_STR_LENGTH, GFP_KERNEL);
5235 char *p = name;
5236
5237 BUG_ON(!name);
5238
5239 *p++ = ':';
5240 /*
5241 * First flags affecting slabcache operations. We will only
5242 * get here for aliasable slabs so we do not need to support
5243 * too many flags. The flags here must cover all flags that
5244 * are matched during merging to guarantee that the id is
5245 * unique.
5246 */
5247 if (s->flags & SLAB_CACHE_DMA)
5248 *p++ = 'd';
5249 if (s->flags & SLAB_RECLAIM_ACCOUNT)
5250 *p++ = 'a';
5251 if (s->flags & SLAB_DEBUG_FREE)
5252 *p++ = 'F';
Vegard Nossum5a896d92008-04-04 00:54:48 +02005253 if (!(s->flags & SLAB_NOTRACK))
5254 *p++ = 't';
Christoph Lameter81819f02007-05-06 14:49:36 -07005255 if (p != name + 1)
5256 *p++ = '-';
5257 p += sprintf(p, "%07d", s->size);
Glauber Costa2633d7a2012-12-18 14:22:34 -08005258
Christoph Lameter81819f02007-05-06 14:49:36 -07005259 BUG_ON(p > name + ID_STR_LENGTH - 1);
5260 return name;
5261}
5262
5263static int sysfs_slab_add(struct kmem_cache *s)
5264{
5265 int err;
5266 const char *name;
Christoph Lameter45530c42012-11-28 16:23:07 +00005267 int unmergeable = slab_unmergeable(s);
Christoph Lameter81819f02007-05-06 14:49:36 -07005268
Christoph Lameter81819f02007-05-06 14:49:36 -07005269 if (unmergeable) {
5270 /*
5271 * Slabcache can never be merged so we can use the name proper.
5272 * This is typically the case for debug situations. In that
5273 * case we can catch duplicate names easily.
5274 */
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06005275 sysfs_remove_link(&slab_kset->kobj, s->name);
Christoph Lameter81819f02007-05-06 14:49:36 -07005276 name = s->name;
5277 } else {
5278 /*
5279 * Create a unique name for the slab as a target
5280 * for the symlinks.
5281 */
5282 name = create_unique_id(s);
5283 }
5284
Vladimir Davydov9a417072014-04-07 15:39:31 -07005285 s->kobj.kset = cache_kset(s);
Tetsuo Handa26e4f202014-01-04 16:32:31 +09005286 err = kobject_init_and_add(&s->kobj, &slab_ktype, NULL, "%s", name);
Dave Jones54b6a732014-04-07 15:39:32 -07005287 if (err)
Konstantin Khlebnikov80da0262015-09-04 15:45:51 -07005288 goto out;
Christoph Lameter81819f02007-05-06 14:49:36 -07005289
5290 err = sysfs_create_group(&s->kobj, &slab_attr_group);
Dave Jones54b6a732014-04-07 15:39:32 -07005291 if (err)
5292 goto out_del_kobj;
Vladimir Davydov9a417072014-04-07 15:39:31 -07005293
5294#ifdef CONFIG_MEMCG_KMEM
5295 if (is_root_cache(s)) {
5296 s->memcg_kset = kset_create_and_add("cgroup", NULL, &s->kobj);
5297 if (!s->memcg_kset) {
Dave Jones54b6a732014-04-07 15:39:32 -07005298 err = -ENOMEM;
5299 goto out_del_kobj;
Vladimir Davydov9a417072014-04-07 15:39:31 -07005300 }
5301 }
5302#endif
5303
Christoph Lameter81819f02007-05-06 14:49:36 -07005304 kobject_uevent(&s->kobj, KOBJ_ADD);
5305 if (!unmergeable) {
5306 /* Setup first alias */
5307 sysfs_slab_alias(s, s->name);
Christoph Lameter81819f02007-05-06 14:49:36 -07005308 }
Dave Jones54b6a732014-04-07 15:39:32 -07005309out:
5310 if (!unmergeable)
5311 kfree(name);
5312 return err;
5313out_del_kobj:
5314 kobject_del(&s->kobj);
Dave Jones54b6a732014-04-07 15:39:32 -07005315 goto out;
Christoph Lameter81819f02007-05-06 14:49:36 -07005316}
5317
Christoph Lameter41a21282014-05-06 12:50:08 -07005318void sysfs_slab_remove(struct kmem_cache *s)
Christoph Lameter81819f02007-05-06 14:49:36 -07005319{
Christoph Lameter97d06602012-07-06 15:25:11 -05005320 if (slab_state < FULL)
Christoph Lameter2bce6482010-07-19 11:39:11 -05005321 /*
5322 * Sysfs has not been setup yet so no need to remove the
5323 * cache from sysfs.
5324 */
5325 return;
5326
Vladimir Davydov9a417072014-04-07 15:39:31 -07005327#ifdef CONFIG_MEMCG_KMEM
5328 kset_unregister(s->memcg_kset);
5329#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07005330 kobject_uevent(&s->kobj, KOBJ_REMOVE);
5331 kobject_del(&s->kobj);
Christoph Lameter151c6022008-01-07 22:29:05 -08005332 kobject_put(&s->kobj);
Christoph Lameter81819f02007-05-06 14:49:36 -07005333}
5334
5335/*
5336 * Need to buffer aliases during bootup until sysfs becomes
Nick Andrew9f6c708e2008-12-05 14:08:08 +11005337 * available lest we lose that information.
Christoph Lameter81819f02007-05-06 14:49:36 -07005338 */
5339struct saved_alias {
5340 struct kmem_cache *s;
5341 const char *name;
5342 struct saved_alias *next;
5343};
5344
Adrian Bunk5af328a2007-07-17 04:03:27 -07005345static struct saved_alias *alias_list;
Christoph Lameter81819f02007-05-06 14:49:36 -07005346
5347static int sysfs_slab_alias(struct kmem_cache *s, const char *name)
5348{
5349 struct saved_alias *al;
5350
Christoph Lameter97d06602012-07-06 15:25:11 -05005351 if (slab_state == FULL) {
Christoph Lameter81819f02007-05-06 14:49:36 -07005352 /*
5353 * If we have a leftover link then remove it.
5354 */
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06005355 sysfs_remove_link(&slab_kset->kobj, name);
5356 return sysfs_create_link(&slab_kset->kobj, &s->kobj, name);
Christoph Lameter81819f02007-05-06 14:49:36 -07005357 }
5358
5359 al = kmalloc(sizeof(struct saved_alias), GFP_KERNEL);
5360 if (!al)
5361 return -ENOMEM;
5362
5363 al->s = s;
5364 al->name = name;
5365 al->next = alias_list;
5366 alias_list = al;
5367 return 0;
5368}
5369
5370static int __init slab_sysfs_init(void)
5371{
Christoph Lameter5b95a4ac2007-07-17 04:03:19 -07005372 struct kmem_cache *s;
Christoph Lameter81819f02007-05-06 14:49:36 -07005373 int err;
5374
Christoph Lameter18004c52012-07-06 15:25:12 -05005375 mutex_lock(&slab_mutex);
Christoph Lameter2bce6482010-07-19 11:39:11 -05005376
Greg Kroah-Hartman0ff21e42007-11-06 10:36:58 -08005377 slab_kset = kset_create_and_add("slab", &slab_uevent_ops, kernel_kobj);
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06005378 if (!slab_kset) {
Christoph Lameter18004c52012-07-06 15:25:12 -05005379 mutex_unlock(&slab_mutex);
Fabian Frederickf9f58282014-06-04 16:06:34 -07005380 pr_err("Cannot register slab subsystem.\n");
Christoph Lameter81819f02007-05-06 14:49:36 -07005381 return -ENOSYS;
5382 }
5383
Christoph Lameter97d06602012-07-06 15:25:11 -05005384 slab_state = FULL;
Christoph Lameter26a7bd02007-05-09 02:32:39 -07005385
Christoph Lameter5b95a4ac2007-07-17 04:03:19 -07005386 list_for_each_entry(s, &slab_caches, list) {
Christoph Lameter26a7bd02007-05-09 02:32:39 -07005387 err = sysfs_slab_add(s);
Christoph Lameter5d540fb2007-08-30 23:56:26 -07005388 if (err)
Fabian Frederickf9f58282014-06-04 16:06:34 -07005389 pr_err("SLUB: Unable to add boot slab %s to sysfs\n",
5390 s->name);
Christoph Lameter26a7bd02007-05-09 02:32:39 -07005391 }
Christoph Lameter81819f02007-05-06 14:49:36 -07005392
5393 while (alias_list) {
5394 struct saved_alias *al = alias_list;
5395
5396 alias_list = alias_list->next;
5397 err = sysfs_slab_alias(al->s, al->name);
Christoph Lameter5d540fb2007-08-30 23:56:26 -07005398 if (err)
Fabian Frederickf9f58282014-06-04 16:06:34 -07005399 pr_err("SLUB: Unable to add boot slab alias %s to sysfs\n",
5400 al->name);
Christoph Lameter81819f02007-05-06 14:49:36 -07005401 kfree(al);
5402 }
5403
Christoph Lameter18004c52012-07-06 15:25:12 -05005404 mutex_unlock(&slab_mutex);
Christoph Lameter81819f02007-05-06 14:49:36 -07005405 resiliency_test();
5406 return 0;
5407}
5408
5409__initcall(slab_sysfs_init);
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05005410#endif /* CONFIG_SYSFS */
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01005411
5412/*
5413 * The /proc/slabinfo ABI
5414 */
Linus Torvalds158a9622008-01-02 13:04:48 -08005415#ifdef CONFIG_SLABINFO
Glauber Costa0d7561c2012-10-19 18:20:27 +04005416void get_slabinfo(struct kmem_cache *s, struct slabinfo *sinfo)
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01005417{
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01005418 unsigned long nr_slabs = 0;
Christoph Lameter205ab992008-04-14 19:11:40 +03005419 unsigned long nr_objs = 0;
5420 unsigned long nr_free = 0;
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01005421 int node;
Christoph Lameterfa45dc22014-08-06 16:04:09 -07005422 struct kmem_cache_node *n;
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01005423
Christoph Lameterfa45dc22014-08-06 16:04:09 -07005424 for_each_kmem_cache_node(s, node, n) {
Wanpeng Lic17fd132013-07-04 08:33:26 +08005425 nr_slabs += node_nr_slabs(n);
5426 nr_objs += node_nr_objs(n);
Christoph Lameter205ab992008-04-14 19:11:40 +03005427 nr_free += count_partial(n, count_free);
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01005428 }
5429
Glauber Costa0d7561c2012-10-19 18:20:27 +04005430 sinfo->active_objs = nr_objs - nr_free;
5431 sinfo->num_objs = nr_objs;
5432 sinfo->active_slabs = nr_slabs;
5433 sinfo->num_slabs = nr_slabs;
5434 sinfo->objects_per_slab = oo_objects(s->oo);
5435 sinfo->cache_order = oo_order(s->oo);
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01005436}
5437
Glauber Costa0d7561c2012-10-19 18:20:27 +04005438void slabinfo_show_stats(struct seq_file *m, struct kmem_cache *s)
Alexey Dobriyan7b3c3a52008-10-06 02:42:17 +04005439{
Alexey Dobriyan7b3c3a52008-10-06 02:42:17 +04005440}
5441
Glauber Costab7454ad2012-10-19 18:20:25 +04005442ssize_t slabinfo_write(struct file *file, const char __user *buffer,
5443 size_t count, loff_t *ppos)
Alexey Dobriyan7b3c3a52008-10-06 02:42:17 +04005444{
Glauber Costab7454ad2012-10-19 18:20:25 +04005445 return -EIO;
Alexey Dobriyan7b3c3a52008-10-06 02:42:17 +04005446}
Linus Torvalds158a9622008-01-02 13:04:48 -08005447#endif /* CONFIG_SLABINFO */