blob: e0ce5dec84ba5fb641d4ac464f3ce440b8a5bbd9 [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
Geert Uytterhoeven117d54d2016-08-04 15:31:55 -0700127void *fixup_red_left(struct kmem_cache *s, void *p)
Joonsoo Kimd86bd1b2016-03-15 14:55:12 -0700128{
129 if (kmem_cache_debug(s) && s->flags & SLAB_RED_ZONE)
130 p += s->red_left_pad;
131
132 return p;
133}
134
Joonsoo Kim345c9052013-06-19 14:05:52 +0900135static inline bool kmem_cache_has_cpu_partial(struct kmem_cache *s)
136{
137#ifdef CONFIG_SLUB_CPU_PARTIAL
138 return !kmem_cache_debug(s);
139#else
140 return false;
141#endif
142}
143
Christoph Lameter81819f02007-05-06 14:49:36 -0700144/*
145 * Issues still to be resolved:
146 *
Christoph Lameter81819f02007-05-06 14:49:36 -0700147 * - Support PAGE_ALLOC_DEBUG. Should be easy to do.
148 *
Christoph Lameter81819f02007-05-06 14:49:36 -0700149 * - Variable sizing of the per node arrays
150 */
151
152/* Enable to test recovery from slab corruption on boot */
153#undef SLUB_RESILIENCY_TEST
154
Christoph Lameterb789ef52011-06-01 12:25:49 -0500155/* Enable to log cmpxchg failures */
156#undef SLUB_DEBUG_CMPXCHG
157
Christoph Lameter81819f02007-05-06 14:49:36 -0700158/*
Christoph Lameter2086d262007-05-06 14:49:46 -0700159 * Mininum number of partial slabs. These will be left on the partial
160 * lists even if they are empty. kmem_cache_shrink may reclaim them.
161 */
Christoph Lameter76be8952007-12-21 14:37:37 -0800162#define MIN_PARTIAL 5
Christoph Lametere95eed52007-05-06 14:49:44 -0700163
Christoph Lameter2086d262007-05-06 14:49:46 -0700164/*
165 * Maximum number of desirable partial slabs.
166 * The existence of more partial slabs makes kmem_cache_shrink
Zhi Yong Wu721ae222013-11-08 20:47:37 +0800167 * sort the partial list by the number of objects in use.
Christoph Lameter2086d262007-05-06 14:49:46 -0700168 */
169#define MAX_PARTIAL 10
170
Laura Abbottbecfda62016-03-15 14:55:06 -0700171#define DEBUG_DEFAULT_FLAGS (SLAB_CONSISTENCY_CHECKS | SLAB_RED_ZONE | \
Christoph Lameter81819f02007-05-06 14:49:36 -0700172 SLAB_POISON | SLAB_STORE_USER)
Christoph Lameter672bba32007-05-09 02:32:39 -0700173
Christoph Lameter81819f02007-05-06 14:49:36 -0700174/*
Laura Abbott149daaf2016-03-15 14:55:09 -0700175 * These debug flags cannot use CMPXCHG because there might be consistency
176 * issues when checking or reading debug information
177 */
178#define SLAB_NO_CMPXCHG (SLAB_CONSISTENCY_CHECKS | SLAB_STORE_USER | \
179 SLAB_TRACE)
180
181
182/*
David Rientjes3de47212009-07-27 18:30:35 -0700183 * Debugging flags that require metadata to be stored in the slab. These get
184 * disabled when slub_debug=O is used and a cache's min order increases with
185 * metadata.
David Rientjesfa5ec8a2009-07-07 00:14:14 -0700186 */
David Rientjes3de47212009-07-27 18:30:35 -0700187#define DEBUG_METADATA_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER)
David Rientjesfa5ec8a2009-07-07 00:14:14 -0700188
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +0400189#define OO_SHIFT 16
190#define OO_MASK ((1 << OO_SHIFT) - 1)
Christoph Lameter50d5c412011-06-01 12:25:45 -0500191#define MAX_OBJS_PER_PAGE 32767 /* since page.objects is u15 */
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +0400192
Christoph Lameter81819f02007-05-06 14:49:36 -0700193/* Internal SLUB flags */
Christoph Lameterf90ec392010-07-09 14:07:11 -0500194#define __OBJECT_POISON 0x80000000UL /* Poison object */
Christoph Lameterb789ef52011-06-01 12:25:49 -0500195#define __CMPXCHG_DOUBLE 0x40000000UL /* Use cmpxchg_double */
Christoph Lameter81819f02007-05-06 14:49:36 -0700196
Christoph Lameter02cbc872007-05-09 02:32:43 -0700197/*
198 * Tracking user of a slab.
199 */
Ben Greeard6543e32011-07-07 11:36:36 -0700200#define TRACK_ADDRS_COUNT 16
Christoph Lameter02cbc872007-05-09 02:32:43 -0700201struct track {
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300202 unsigned long addr; /* Called from address */
Ben Greeard6543e32011-07-07 11:36:36 -0700203#ifdef CONFIG_STACKTRACE
204 unsigned long addrs[TRACK_ADDRS_COUNT]; /* Called from address */
205#endif
Christoph Lameter02cbc872007-05-09 02:32:43 -0700206 int cpu; /* Was running on cpu */
207 int pid; /* Pid context */
208 unsigned long when; /* When did the operation occur */
209};
210
211enum track_item { TRACK_ALLOC, TRACK_FREE };
212
Christoph Lameterab4d5ed2010-10-05 13:57:26 -0500213#ifdef CONFIG_SYSFS
Christoph Lameter81819f02007-05-06 14:49:36 -0700214static int sysfs_slab_add(struct kmem_cache *);
215static int sysfs_slab_alias(struct kmem_cache *, const char *);
Glauber Costa107dab52012-12-18 14:23:05 -0800216static void memcg_propagate_slab_attrs(struct kmem_cache *s);
Christoph Lameter81819f02007-05-06 14:49:36 -0700217#else
Christoph Lameter0c710012007-07-17 04:03:24 -0700218static inline int sysfs_slab_add(struct kmem_cache *s) { return 0; }
219static inline int sysfs_slab_alias(struct kmem_cache *s, const char *p)
220 { return 0; }
Glauber Costa107dab52012-12-18 14:23:05 -0800221static inline void memcg_propagate_slab_attrs(struct kmem_cache *s) { }
Christoph Lameter81819f02007-05-06 14:49:36 -0700222#endif
223
Christoph Lameter4fdccdf2011-03-22 13:35:00 -0500224static inline void stat(const struct kmem_cache *s, enum stat_item si)
Christoph Lameter8ff12cf2008-02-07 17:47:41 -0800225{
226#ifdef CONFIG_SLUB_STATS
Christoph Lameter88da03a2014-04-07 15:39:42 -0700227 /*
228 * The rmw is racy on a preemptible kernel but this is acceptable, so
229 * avoid this_cpu_add()'s irq-disable overhead.
230 */
231 raw_cpu_inc(s->cpu_slab->stat[si]);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -0800232#endif
233}
234
Christoph Lameter81819f02007-05-06 14:49:36 -0700235/********************************************************************
236 * Core slab cache functions
237 *******************************************************************/
238
Christoph Lameter7656c722007-05-09 02:32:40 -0700239static inline void *get_freepointer(struct kmem_cache *s, void *object)
240{
241 return *(void **)(object + s->offset);
242}
243
Eric Dumazet0ad95002011-12-16 16:25:34 +0100244static void prefetch_freepointer(const struct kmem_cache *s, void *object)
245{
246 prefetch(object + s->offset);
247}
248
Christoph Lameter1393d9a2011-05-16 15:26:08 -0500249static inline void *get_freepointer_safe(struct kmem_cache *s, void *object)
250{
251 void *p;
252
Joonsoo Kim922d5662016-03-17 14:17:53 -0700253 if (!debug_pagealloc_enabled())
254 return get_freepointer(s, object);
255
Christoph Lameter1393d9a2011-05-16 15:26:08 -0500256 probe_kernel_read(&p, (void **)(object + s->offset), sizeof(p));
Christoph Lameter1393d9a2011-05-16 15:26:08 -0500257 return p;
258}
259
Christoph Lameter7656c722007-05-09 02:32:40 -0700260static inline void set_freepointer(struct kmem_cache *s, void *object, void *fp)
261{
262 *(void **)(object + s->offset) = fp;
263}
264
265/* Loop over all objects in a slab */
Christoph Lameter224a88b2008-04-14 19:11:31 +0300266#define for_each_object(__p, __s, __addr, __objects) \
Joonsoo Kimd86bd1b2016-03-15 14:55:12 -0700267 for (__p = fixup_red_left(__s, __addr); \
268 __p < (__addr) + (__objects) * (__s)->size; \
269 __p += (__s)->size)
Christoph Lameter7656c722007-05-09 02:32:40 -0700270
Wei Yang54266642014-08-06 16:04:42 -0700271#define for_each_object_idx(__p, __idx, __s, __addr, __objects) \
Joonsoo Kimd86bd1b2016-03-15 14:55:12 -0700272 for (__p = fixup_red_left(__s, __addr), __idx = 1; \
273 __idx <= __objects; \
274 __p += (__s)->size, __idx++)
Wei Yang54266642014-08-06 16:04:42 -0700275
Christoph Lameter7656c722007-05-09 02:32:40 -0700276/* Determine object index from a given position */
277static inline int slab_index(void *p, struct kmem_cache *s, void *addr)
278{
279 return (p - addr) / s->size;
280}
281
Lai Jiangshanab9a0f12011-03-10 15:21:48 +0800282static inline int order_objects(int order, unsigned long size, int reserved)
283{
284 return ((PAGE_SIZE << order) - reserved) / size;
285}
286
Christoph Lameter834f3d12008-04-14 19:11:31 +0300287static inline struct kmem_cache_order_objects oo_make(int order,
Lai Jiangshanab9a0f12011-03-10 15:21:48 +0800288 unsigned long size, int reserved)
Christoph Lameter834f3d12008-04-14 19:11:31 +0300289{
290 struct kmem_cache_order_objects x = {
Lai Jiangshanab9a0f12011-03-10 15:21:48 +0800291 (order << OO_SHIFT) + order_objects(order, size, reserved)
Christoph Lameter834f3d12008-04-14 19:11:31 +0300292 };
293
294 return x;
295}
296
297static inline int oo_order(struct kmem_cache_order_objects x)
298{
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +0400299 return x.x >> OO_SHIFT;
Christoph Lameter834f3d12008-04-14 19:11:31 +0300300}
301
302static inline int oo_objects(struct kmem_cache_order_objects x)
303{
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +0400304 return x.x & OO_MASK;
Christoph Lameter834f3d12008-04-14 19:11:31 +0300305}
306
Christoph Lameter881db7f2011-06-01 12:25:53 -0500307/*
308 * Per slab locking using the pagelock
309 */
310static __always_inline void slab_lock(struct page *page)
311{
Kirill A. Shutemov48c935a2016-01-15 16:51:24 -0800312 VM_BUG_ON_PAGE(PageTail(page), page);
Christoph Lameter881db7f2011-06-01 12:25:53 -0500313 bit_spin_lock(PG_locked, &page->flags);
314}
315
316static __always_inline void slab_unlock(struct page *page)
317{
Kirill A. Shutemov48c935a2016-01-15 16:51:24 -0800318 VM_BUG_ON_PAGE(PageTail(page), page);
Christoph Lameter881db7f2011-06-01 12:25:53 -0500319 __bit_spin_unlock(PG_locked, &page->flags);
320}
321
Dave Hansena0320862014-01-30 15:46:09 -0800322static inline void set_page_slub_counters(struct page *page, unsigned long counters_new)
323{
324 struct page tmp;
325 tmp.counters = counters_new;
326 /*
327 * page->counters can cover frozen/inuse/objects as well
Joonsoo Kim0139aa72016-05-19 17:10:49 -0700328 * as page->_refcount. If we assign to ->counters directly
329 * we run the risk of losing updates to page->_refcount, so
Dave Hansena0320862014-01-30 15:46:09 -0800330 * be careful and only assign to the fields we need.
331 */
332 page->frozen = tmp.frozen;
333 page->inuse = tmp.inuse;
334 page->objects = tmp.objects;
335}
336
Christoph Lameter1d071712011-07-14 12:49:12 -0500337/* Interrupts must be disabled (for the fallback code to work right) */
338static inline bool __cmpxchg_double_slab(struct kmem_cache *s, struct page *page,
Christoph Lameterb789ef52011-06-01 12:25:49 -0500339 void *freelist_old, unsigned long counters_old,
340 void *freelist_new, unsigned long counters_new,
341 const char *n)
342{
Christoph Lameter1d071712011-07-14 12:49:12 -0500343 VM_BUG_ON(!irqs_disabled());
Heiko Carstens25654092012-01-12 17:17:33 -0800344#if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \
345 defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
Christoph Lameterb789ef52011-06-01 12:25:49 -0500346 if (s->flags & __CMPXCHG_DOUBLE) {
Jan Beulichcdcd6292012-01-02 17:02:18 +0000347 if (cmpxchg_double(&page->freelist, &page->counters,
Dan Carpenter0aa9a132014-08-06 16:04:48 -0700348 freelist_old, counters_old,
349 freelist_new, counters_new))
Joe Perches6f6528a2015-04-14 15:44:31 -0700350 return true;
Christoph Lameterb789ef52011-06-01 12:25:49 -0500351 } else
352#endif
353 {
Christoph Lameter881db7f2011-06-01 12:25:53 -0500354 slab_lock(page);
Chen Gangd0e0ac92013-07-15 09:05:29 +0800355 if (page->freelist == freelist_old &&
356 page->counters == counters_old) {
Christoph Lameterb789ef52011-06-01 12:25:49 -0500357 page->freelist = freelist_new;
Dave Hansena0320862014-01-30 15:46:09 -0800358 set_page_slub_counters(page, counters_new);
Christoph Lameter881db7f2011-06-01 12:25:53 -0500359 slab_unlock(page);
Joe Perches6f6528a2015-04-14 15:44:31 -0700360 return true;
Christoph Lameterb789ef52011-06-01 12:25:49 -0500361 }
Christoph Lameter881db7f2011-06-01 12:25:53 -0500362 slab_unlock(page);
Christoph Lameterb789ef52011-06-01 12:25:49 -0500363 }
364
365 cpu_relax();
366 stat(s, CMPXCHG_DOUBLE_FAIL);
367
368#ifdef SLUB_DEBUG_CMPXCHG
Fabian Frederickf9f58282014-06-04 16:06:34 -0700369 pr_info("%s %s: cmpxchg double redo ", n, s->name);
Christoph Lameterb789ef52011-06-01 12:25:49 -0500370#endif
371
Joe Perches6f6528a2015-04-14 15:44:31 -0700372 return false;
Christoph Lameterb789ef52011-06-01 12:25:49 -0500373}
374
Christoph Lameter1d071712011-07-14 12:49:12 -0500375static inline bool cmpxchg_double_slab(struct kmem_cache *s, struct page *page,
376 void *freelist_old, unsigned long counters_old,
377 void *freelist_new, unsigned long counters_new,
378 const char *n)
379{
Heiko Carstens25654092012-01-12 17:17:33 -0800380#if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \
381 defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
Christoph Lameter1d071712011-07-14 12:49:12 -0500382 if (s->flags & __CMPXCHG_DOUBLE) {
Jan Beulichcdcd6292012-01-02 17:02:18 +0000383 if (cmpxchg_double(&page->freelist, &page->counters,
Dan Carpenter0aa9a132014-08-06 16:04:48 -0700384 freelist_old, counters_old,
385 freelist_new, counters_new))
Joe Perches6f6528a2015-04-14 15:44:31 -0700386 return true;
Christoph Lameter1d071712011-07-14 12:49:12 -0500387 } else
388#endif
389 {
390 unsigned long flags;
391
392 local_irq_save(flags);
393 slab_lock(page);
Chen Gangd0e0ac92013-07-15 09:05:29 +0800394 if (page->freelist == freelist_old &&
395 page->counters == counters_old) {
Christoph Lameter1d071712011-07-14 12:49:12 -0500396 page->freelist = freelist_new;
Dave Hansena0320862014-01-30 15:46:09 -0800397 set_page_slub_counters(page, counters_new);
Christoph Lameter1d071712011-07-14 12:49:12 -0500398 slab_unlock(page);
399 local_irq_restore(flags);
Joe Perches6f6528a2015-04-14 15:44:31 -0700400 return true;
Christoph Lameter1d071712011-07-14 12:49:12 -0500401 }
402 slab_unlock(page);
403 local_irq_restore(flags);
404 }
405
406 cpu_relax();
407 stat(s, CMPXCHG_DOUBLE_FAIL);
408
409#ifdef SLUB_DEBUG_CMPXCHG
Fabian Frederickf9f58282014-06-04 16:06:34 -0700410 pr_info("%s %s: cmpxchg double redo ", n, s->name);
Christoph Lameter1d071712011-07-14 12:49:12 -0500411#endif
412
Joe Perches6f6528a2015-04-14 15:44:31 -0700413 return false;
Christoph Lameter1d071712011-07-14 12:49:12 -0500414}
415
Christoph Lameter41ecc552007-05-09 02:32:44 -0700416#ifdef CONFIG_SLUB_DEBUG
417/*
Christoph Lameter5f80b132011-04-15 14:48:13 -0500418 * Determine a map of object in use on a page.
419 *
Christoph Lameter881db7f2011-06-01 12:25:53 -0500420 * Node listlock must be held to guarantee that the page does
Christoph Lameter5f80b132011-04-15 14:48:13 -0500421 * not vanish from under us.
422 */
423static void get_map(struct kmem_cache *s, struct page *page, unsigned long *map)
424{
425 void *p;
426 void *addr = page_address(page);
427
428 for (p = page->freelist; p; p = get_freepointer(s, p))
429 set_bit(slab_index(p, s, addr), map);
430}
431
Joonsoo Kimd86bd1b2016-03-15 14:55:12 -0700432static inline int size_from_object(struct kmem_cache *s)
433{
434 if (s->flags & SLAB_RED_ZONE)
435 return s->size - s->red_left_pad;
436
437 return s->size;
438}
439
440static inline void *restore_red_left(struct kmem_cache *s, void *p)
441{
442 if (s->flags & SLAB_RED_ZONE)
443 p -= s->red_left_pad;
444
445 return p;
446}
447
Christoph Lameter41ecc552007-05-09 02:32:44 -0700448/*
449 * Debug settings:
450 */
Andrey Ryabinin89d3c872015-11-05 18:51:23 -0800451#if defined(CONFIG_SLUB_DEBUG_ON)
Christoph Lameterf0630ff2007-07-15 23:38:14 -0700452static int slub_debug = DEBUG_DEFAULT_FLAGS;
453#else
Christoph Lameter41ecc552007-05-09 02:32:44 -0700454static int slub_debug;
Christoph Lameterf0630ff2007-07-15 23:38:14 -0700455#endif
Christoph Lameter41ecc552007-05-09 02:32:44 -0700456
457static char *slub_debug_slabs;
David Rientjesfa5ec8a2009-07-07 00:14:14 -0700458static int disable_higher_order_debug;
Christoph Lameter41ecc552007-05-09 02:32:44 -0700459
Christoph Lameter7656c722007-05-09 02:32:40 -0700460/*
Andrey Ryabinina79316c2015-02-13 14:39:38 -0800461 * slub is about to manipulate internal object metadata. This memory lies
462 * outside the range of the allocated object, so accessing it would normally
463 * be reported by kasan as a bounds error. metadata_access_enable() is used
464 * to tell kasan that these accesses are OK.
465 */
466static inline void metadata_access_enable(void)
467{
468 kasan_disable_current();
469}
470
471static inline void metadata_access_disable(void)
472{
473 kasan_enable_current();
474}
475
476/*
Christoph Lameter81819f02007-05-06 14:49:36 -0700477 * Object debugging
478 */
Joonsoo Kimd86bd1b2016-03-15 14:55:12 -0700479
480/* Verify that a pointer has an address that is valid within a slab page */
481static inline int check_valid_pointer(struct kmem_cache *s,
482 struct page *page, void *object)
483{
484 void *base;
485
486 if (!object)
487 return 1;
488
489 base = page_address(page);
490 object = restore_red_left(s, object);
491 if (object < base || object >= base + page->objects * s->size ||
492 (object - base) % s->size) {
493 return 0;
494 }
495
496 return 1;
497}
498
Daniel Thompson74b41632017-01-24 15:18:02 -0800499static void print_section(char *level, char *text, u8 *addr,
500 unsigned int length)
Christoph Lameter81819f02007-05-06 14:49:36 -0700501{
Andrey Ryabinina79316c2015-02-13 14:39:38 -0800502 metadata_access_enable();
Daniel Thompson74b41632017-01-24 15:18:02 -0800503 print_hex_dump(level, text, DUMP_PREFIX_ADDRESS, 16, 1, addr,
Sebastian Andrzej Siewiorffc79d22011-07-29 14:10:20 +0200504 length, 1);
Andrey Ryabinina79316c2015-02-13 14:39:38 -0800505 metadata_access_disable();
Christoph Lameter81819f02007-05-06 14:49:36 -0700506}
507
Christoph Lameter81819f02007-05-06 14:49:36 -0700508static struct track *get_track(struct kmem_cache *s, void *object,
509 enum track_item alloc)
510{
511 struct track *p;
512
513 if (s->offset)
514 p = object + s->offset + sizeof(void *);
515 else
516 p = object + s->inuse;
517
518 return p + alloc;
519}
520
521static void set_track(struct kmem_cache *s, void *object,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300522 enum track_item alloc, unsigned long addr)
Christoph Lameter81819f02007-05-06 14:49:36 -0700523{
Akinobu Mita1a00df42009-03-07 00:36:21 +0900524 struct track *p = get_track(s, object, alloc);
Christoph Lameter81819f02007-05-06 14:49:36 -0700525
Christoph Lameter81819f02007-05-06 14:49:36 -0700526 if (addr) {
Ben Greeard6543e32011-07-07 11:36:36 -0700527#ifdef CONFIG_STACKTRACE
528 struct stack_trace trace;
529 int i;
530
531 trace.nr_entries = 0;
532 trace.max_entries = TRACK_ADDRS_COUNT;
533 trace.entries = p->addrs;
534 trace.skip = 3;
Andrey Ryabinina79316c2015-02-13 14:39:38 -0800535 metadata_access_enable();
Ben Greeard6543e32011-07-07 11:36:36 -0700536 save_stack_trace(&trace);
Andrey Ryabinina79316c2015-02-13 14:39:38 -0800537 metadata_access_disable();
Ben Greeard6543e32011-07-07 11:36:36 -0700538
539 /* See rant in lockdep.c */
540 if (trace.nr_entries != 0 &&
541 trace.entries[trace.nr_entries - 1] == ULONG_MAX)
542 trace.nr_entries--;
543
544 for (i = trace.nr_entries; i < TRACK_ADDRS_COUNT; i++)
545 p->addrs[i] = 0;
546#endif
Christoph Lameter81819f02007-05-06 14:49:36 -0700547 p->addr = addr;
548 p->cpu = smp_processor_id();
Alexey Dobriyan88e4ccf2008-06-23 02:58:37 +0400549 p->pid = current->pid;
Christoph Lameter81819f02007-05-06 14:49:36 -0700550 p->when = jiffies;
551 } else
552 memset(p, 0, sizeof(struct track));
553}
554
Christoph Lameter81819f02007-05-06 14:49:36 -0700555static void init_tracking(struct kmem_cache *s, void *object)
556{
Christoph Lameter24922682007-07-17 04:03:18 -0700557 if (!(s->flags & SLAB_STORE_USER))
558 return;
559
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +0300560 set_track(s, object, TRACK_FREE, 0UL);
561 set_track(s, object, TRACK_ALLOC, 0UL);
Christoph Lameter81819f02007-05-06 14:49:36 -0700562}
563
564static void print_track(const char *s, struct track *t)
565{
566 if (!t->addr)
567 return;
568
Fabian Frederickf9f58282014-06-04 16:06:34 -0700569 pr_err("INFO: %s in %pS age=%lu cpu=%u pid=%d\n",
570 s, (void *)t->addr, jiffies - t->when, t->cpu, t->pid);
Ben Greeard6543e32011-07-07 11:36:36 -0700571#ifdef CONFIG_STACKTRACE
572 {
573 int i;
574 for (i = 0; i < TRACK_ADDRS_COUNT; i++)
575 if (t->addrs[i])
Fabian Frederickf9f58282014-06-04 16:06:34 -0700576 pr_err("\t%pS\n", (void *)t->addrs[i]);
Ben Greeard6543e32011-07-07 11:36:36 -0700577 else
578 break;
579 }
580#endif
Christoph Lameter81819f02007-05-06 14:49:36 -0700581}
582
Christoph Lameter24922682007-07-17 04:03:18 -0700583static void print_tracking(struct kmem_cache *s, void *object)
584{
585 if (!(s->flags & SLAB_STORE_USER))
586 return;
587
588 print_track("Allocated", get_track(s, object, TRACK_ALLOC));
589 print_track("Freed", get_track(s, object, TRACK_FREE));
590}
591
592static void print_page_info(struct page *page)
593{
Fabian Frederickf9f58282014-06-04 16:06:34 -0700594 pr_err("INFO: Slab 0x%p objects=%u used=%u fp=0x%p flags=0x%04lx\n",
Chen Gangd0e0ac92013-07-15 09:05:29 +0800595 page, page->objects, page->inuse, page->freelist, page->flags);
Christoph Lameter24922682007-07-17 04:03:18 -0700596
597}
598
599static void slab_bug(struct kmem_cache *s, char *fmt, ...)
600{
Fabian Frederickecc42fb2014-06-04 16:06:35 -0700601 struct va_format vaf;
Christoph Lameter24922682007-07-17 04:03:18 -0700602 va_list args;
Christoph Lameter24922682007-07-17 04:03:18 -0700603
604 va_start(args, fmt);
Fabian Frederickecc42fb2014-06-04 16:06:35 -0700605 vaf.fmt = fmt;
606 vaf.va = &args;
Fabian Frederickf9f58282014-06-04 16:06:34 -0700607 pr_err("=============================================================================\n");
Fabian Frederickecc42fb2014-06-04 16:06:35 -0700608 pr_err("BUG %s (%s): %pV\n", s->name, print_tainted(), &vaf);
Fabian Frederickf9f58282014-06-04 16:06:34 -0700609 pr_err("-----------------------------------------------------------------------------\n\n");
Dave Jones645df232012-09-18 15:54:12 -0400610
Rusty Russell373d4d02013-01-21 17:17:39 +1030611 add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
Fabian Frederickecc42fb2014-06-04 16:06:35 -0700612 va_end(args);
Christoph Lameter24922682007-07-17 04:03:18 -0700613}
614
615static void slab_fix(struct kmem_cache *s, char *fmt, ...)
616{
Fabian Frederickecc42fb2014-06-04 16:06:35 -0700617 struct va_format vaf;
Christoph Lameter24922682007-07-17 04:03:18 -0700618 va_list args;
Christoph Lameter24922682007-07-17 04:03:18 -0700619
620 va_start(args, fmt);
Fabian Frederickecc42fb2014-06-04 16:06:35 -0700621 vaf.fmt = fmt;
622 vaf.va = &args;
623 pr_err("FIX %s: %pV\n", s->name, &vaf);
Christoph Lameter24922682007-07-17 04:03:18 -0700624 va_end(args);
Christoph Lameter24922682007-07-17 04:03:18 -0700625}
626
627static void print_trailer(struct kmem_cache *s, struct page *page, u8 *p)
Christoph Lameter81819f02007-05-06 14:49:36 -0700628{
629 unsigned int off; /* Offset of last byte */
Christoph Lametera973e9d2008-03-01 13:40:44 -0800630 u8 *addr = page_address(page);
Christoph Lameter24922682007-07-17 04:03:18 -0700631
632 print_tracking(s, p);
633
634 print_page_info(page);
635
Fabian Frederickf9f58282014-06-04 16:06:34 -0700636 pr_err("INFO: Object 0x%p @offset=%tu fp=0x%p\n\n",
637 p, p - addr, get_freepointer(s, p));
Christoph Lameter24922682007-07-17 04:03:18 -0700638
Joonsoo Kimd86bd1b2016-03-15 14:55:12 -0700639 if (s->flags & SLAB_RED_ZONE)
Daniel Thompson74b41632017-01-24 15:18:02 -0800640 print_section(KERN_ERR, "Redzone ", p - s->red_left_pad,
641 s->red_left_pad);
Joonsoo Kimd86bd1b2016-03-15 14:55:12 -0700642 else if (p > addr + 16)
Daniel Thompson74b41632017-01-24 15:18:02 -0800643 print_section(KERN_ERR, "Bytes b4 ", p - 16, 16);
Christoph Lameter24922682007-07-17 04:03:18 -0700644
Daniel Thompson74b41632017-01-24 15:18:02 -0800645 print_section(KERN_ERR, "Object ", p,
646 min_t(unsigned long, s->object_size, PAGE_SIZE));
Christoph Lameter81819f02007-05-06 14:49:36 -0700647 if (s->flags & SLAB_RED_ZONE)
Daniel Thompson74b41632017-01-24 15:18:02 -0800648 print_section(KERN_ERR, "Redzone ", p + s->object_size,
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500649 s->inuse - s->object_size);
Christoph Lameter81819f02007-05-06 14:49:36 -0700650
Christoph Lameter81819f02007-05-06 14:49:36 -0700651 if (s->offset)
652 off = s->offset + sizeof(void *);
653 else
654 off = s->inuse;
655
Christoph Lameter24922682007-07-17 04:03:18 -0700656 if (s->flags & SLAB_STORE_USER)
Christoph Lameter81819f02007-05-06 14:49:36 -0700657 off += 2 * sizeof(struct track);
Christoph Lameter81819f02007-05-06 14:49:36 -0700658
Alexander Potapenko80a92012016-07-28 15:49:07 -0700659 off += kasan_metadata_size(s);
660
Joonsoo Kimd86bd1b2016-03-15 14:55:12 -0700661 if (off != size_from_object(s))
Christoph Lameter81819f02007-05-06 14:49:36 -0700662 /* Beginning of the filler is the free pointer */
Daniel Thompson74b41632017-01-24 15:18:02 -0800663 print_section(KERN_ERR, "Padding ", p + off,
664 size_from_object(s) - off);
Christoph Lameter24922682007-07-17 04:03:18 -0700665
666 dump_stack();
Christoph Lameter81819f02007-05-06 14:49:36 -0700667}
668
Andrey Ryabinin75c66de2015-02-13 14:39:35 -0800669void object_err(struct kmem_cache *s, struct page *page,
Christoph Lameter81819f02007-05-06 14:49:36 -0700670 u8 *object, char *reason)
671{
Christoph Lameter3dc50632008-04-23 12:28:01 -0700672 slab_bug(s, "%s", reason);
Christoph Lameter24922682007-07-17 04:03:18 -0700673 print_trailer(s, page, object);
Christoph Lameter81819f02007-05-06 14:49:36 -0700674}
675
Mathieu Malaterrec99dbd92018-06-07 17:05:17 -0700676static __printf(3, 4) void slab_err(struct kmem_cache *s, struct page *page,
Chen Gangd0e0ac92013-07-15 09:05:29 +0800677 const char *fmt, ...)
Christoph Lameter81819f02007-05-06 14:49:36 -0700678{
679 va_list args;
680 char buf[100];
681
Christoph Lameter24922682007-07-17 04:03:18 -0700682 va_start(args, fmt);
683 vsnprintf(buf, sizeof(buf), fmt, args);
Christoph Lameter81819f02007-05-06 14:49:36 -0700684 va_end(args);
Christoph Lameter3dc50632008-04-23 12:28:01 -0700685 slab_bug(s, "%s", buf);
Christoph Lameter24922682007-07-17 04:03:18 -0700686 print_page_info(page);
Christoph Lameter81819f02007-05-06 14:49:36 -0700687 dump_stack();
688}
689
Christoph Lameterf7cb1932010-09-29 07:15:01 -0500690static void init_object(struct kmem_cache *s, void *object, u8 val)
Christoph Lameter81819f02007-05-06 14:49:36 -0700691{
692 u8 *p = object;
693
Joonsoo Kimd86bd1b2016-03-15 14:55:12 -0700694 if (s->flags & SLAB_RED_ZONE)
695 memset(p - s->red_left_pad, val, s->red_left_pad);
696
Christoph Lameter81819f02007-05-06 14:49:36 -0700697 if (s->flags & __OBJECT_POISON) {
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500698 memset(p, POISON_FREE, s->object_size - 1);
699 p[s->object_size - 1] = POISON_END;
Christoph Lameter81819f02007-05-06 14:49:36 -0700700 }
701
702 if (s->flags & SLAB_RED_ZONE)
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500703 memset(p + s->object_size, val, s->inuse - s->object_size);
Christoph Lameter81819f02007-05-06 14:49:36 -0700704}
705
Christoph Lameter24922682007-07-17 04:03:18 -0700706static void restore_bytes(struct kmem_cache *s, char *message, u8 data,
707 void *from, void *to)
708{
709 slab_fix(s, "Restoring 0x%p-0x%p=0x%x\n", from, to - 1, data);
710 memset(from, data, to - from);
711}
712
713static int check_bytes_and_report(struct kmem_cache *s, struct page *page,
714 u8 *object, char *what,
Pekka Enberg06428782008-01-07 23:20:27 -0800715 u8 *start, unsigned int value, unsigned int bytes)
Christoph Lameter24922682007-07-17 04:03:18 -0700716{
717 u8 *fault;
718 u8 *end;
719
Andrey Ryabinina79316c2015-02-13 14:39:38 -0800720 metadata_access_enable();
Akinobu Mita798248202011-10-31 17:08:07 -0700721 fault = memchr_inv(start, value, bytes);
Andrey Ryabinina79316c2015-02-13 14:39:38 -0800722 metadata_access_disable();
Christoph Lameter24922682007-07-17 04:03:18 -0700723 if (!fault)
724 return 1;
725
726 end = start + bytes;
727 while (end > fault && end[-1] == value)
728 end--;
729
730 slab_bug(s, "%s overwritten", what);
Fabian Frederickf9f58282014-06-04 16:06:34 -0700731 pr_err("INFO: 0x%p-0x%p. First byte 0x%x instead of 0x%x\n",
Christoph Lameter24922682007-07-17 04:03:18 -0700732 fault, end - 1, fault[0], value);
733 print_trailer(s, page, object);
734
735 restore_bytes(s, what, value, fault, end);
736 return 0;
Christoph Lameter81819f02007-05-06 14:49:36 -0700737}
738
Christoph Lameter81819f02007-05-06 14:49:36 -0700739/*
740 * Object layout:
741 *
742 * object address
743 * Bytes of the object to be managed.
744 * If the freepointer may overlay the object then the free
745 * pointer is the first word of the object.
Christoph Lameter672bba32007-05-09 02:32:39 -0700746 *
Christoph Lameter81819f02007-05-06 14:49:36 -0700747 * Poisoning uses 0x6b (POISON_FREE) and the last byte is
748 * 0xa5 (POISON_END)
749 *
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500750 * object + s->object_size
Christoph Lameter81819f02007-05-06 14:49:36 -0700751 * Padding to reach word boundary. This is also used for Redzoning.
Christoph Lameter672bba32007-05-09 02:32:39 -0700752 * Padding is extended by another word if Redzoning is enabled and
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500753 * object_size == inuse.
Christoph Lameter672bba32007-05-09 02:32:39 -0700754 *
Christoph Lameter81819f02007-05-06 14:49:36 -0700755 * We fill with 0xbb (RED_INACTIVE) for inactive objects and with
756 * 0xcc (RED_ACTIVE) for objects in use.
757 *
758 * object + s->inuse
Christoph Lameter672bba32007-05-09 02:32:39 -0700759 * Meta data starts here.
760 *
Christoph Lameter81819f02007-05-06 14:49:36 -0700761 * A. Free pointer (if we cannot overwrite object on free)
762 * B. Tracking data for SLAB_STORE_USER
Christoph Lameter672bba32007-05-09 02:32:39 -0700763 * C. Padding to reach required alignment boundary or at mininum
Christoph Lameter6446faa2008-02-15 23:45:26 -0800764 * one word if debugging is on to be able to detect writes
Christoph Lameter672bba32007-05-09 02:32:39 -0700765 * before the word boundary.
766 *
767 * Padding is done using 0x5a (POISON_INUSE)
Christoph Lameter81819f02007-05-06 14:49:36 -0700768 *
769 * object + s->size
Christoph Lameter672bba32007-05-09 02:32:39 -0700770 * Nothing is used beyond s->size.
Christoph Lameter81819f02007-05-06 14:49:36 -0700771 *
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500772 * If slabcaches are merged then the object_size and inuse boundaries are mostly
Christoph Lameter672bba32007-05-09 02:32:39 -0700773 * ignored. And therefore no slab options that rely on these boundaries
Christoph Lameter81819f02007-05-06 14:49:36 -0700774 * may be used with merged slabcaches.
775 */
776
Christoph Lameter81819f02007-05-06 14:49:36 -0700777static int check_pad_bytes(struct kmem_cache *s, struct page *page, u8 *p)
778{
779 unsigned long off = s->inuse; /* The end of info */
780
781 if (s->offset)
782 /* Freepointer is placed after the object. */
783 off += sizeof(void *);
784
785 if (s->flags & SLAB_STORE_USER)
786 /* We also have user information there */
787 off += 2 * sizeof(struct track);
788
Alexander Potapenko80a92012016-07-28 15:49:07 -0700789 off += kasan_metadata_size(s);
790
Joonsoo Kimd86bd1b2016-03-15 14:55:12 -0700791 if (size_from_object(s) == off)
Christoph Lameter81819f02007-05-06 14:49:36 -0700792 return 1;
793
Christoph Lameter24922682007-07-17 04:03:18 -0700794 return check_bytes_and_report(s, page, p, "Object padding",
Joonsoo Kimd86bd1b2016-03-15 14:55:12 -0700795 p + off, POISON_INUSE, size_from_object(s) - off);
Christoph Lameter81819f02007-05-06 14:49:36 -0700796}
797
Christoph Lameter39b26462008-04-14 19:11:30 +0300798/* Check the pad bytes at the end of a slab page */
Christoph Lameter81819f02007-05-06 14:49:36 -0700799static int slab_pad_check(struct kmem_cache *s, struct page *page)
800{
Christoph Lameter24922682007-07-17 04:03:18 -0700801 u8 *start;
802 u8 *fault;
803 u8 *end;
804 int length;
805 int remainder;
Christoph Lameter81819f02007-05-06 14:49:36 -0700806
807 if (!(s->flags & SLAB_POISON))
808 return 1;
809
Christoph Lametera973e9d2008-03-01 13:40:44 -0800810 start = page_address(page);
Lai Jiangshanab9a0f12011-03-10 15:21:48 +0800811 length = (PAGE_SIZE << compound_order(page)) - s->reserved;
Christoph Lameter39b26462008-04-14 19:11:30 +0300812 end = start + length;
813 remainder = length % s->size;
Christoph Lameter81819f02007-05-06 14:49:36 -0700814 if (!remainder)
815 return 1;
816
Andrey Ryabinina79316c2015-02-13 14:39:38 -0800817 metadata_access_enable();
Akinobu Mita798248202011-10-31 17:08:07 -0700818 fault = memchr_inv(end - remainder, POISON_INUSE, remainder);
Andrey Ryabinina79316c2015-02-13 14:39:38 -0800819 metadata_access_disable();
Christoph Lameter24922682007-07-17 04:03:18 -0700820 if (!fault)
821 return 1;
822 while (end > fault && end[-1] == POISON_INUSE)
823 end--;
824
825 slab_err(s, page, "Padding overwritten. 0x%p-0x%p", fault, end - 1);
Daniel Thompson74b41632017-01-24 15:18:02 -0800826 print_section(KERN_ERR, "Padding ", end - remainder, remainder);
Christoph Lameter24922682007-07-17 04:03:18 -0700827
Eric Dumazet8a3d2712009-09-03 16:08:06 +0200828 restore_bytes(s, "slab padding", POISON_INUSE, end - remainder, end);
Christoph Lameter24922682007-07-17 04:03:18 -0700829 return 0;
Christoph Lameter81819f02007-05-06 14:49:36 -0700830}
831
832static int check_object(struct kmem_cache *s, struct page *page,
Christoph Lameterf7cb1932010-09-29 07:15:01 -0500833 void *object, u8 val)
Christoph Lameter81819f02007-05-06 14:49:36 -0700834{
835 u8 *p = object;
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500836 u8 *endobject = object + s->object_size;
Christoph Lameter81819f02007-05-06 14:49:36 -0700837
838 if (s->flags & SLAB_RED_ZONE) {
Christoph Lameter24922682007-07-17 04:03:18 -0700839 if (!check_bytes_and_report(s, page, object, "Redzone",
Joonsoo Kimd86bd1b2016-03-15 14:55:12 -0700840 object - s->red_left_pad, val, s->red_left_pad))
841 return 0;
842
843 if (!check_bytes_and_report(s, page, object, "Redzone",
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500844 endobject, val, s->inuse - s->object_size))
Christoph Lameter81819f02007-05-06 14:49:36 -0700845 return 0;
Christoph Lameter81819f02007-05-06 14:49:36 -0700846 } else {
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500847 if ((s->flags & SLAB_POISON) && s->object_size < s->inuse) {
Ingo Molnar3adbefe2008-02-05 17:57:39 -0800848 check_bytes_and_report(s, page, p, "Alignment padding",
Chen Gangd0e0ac92013-07-15 09:05:29 +0800849 endobject, POISON_INUSE,
850 s->inuse - s->object_size);
Ingo Molnar3adbefe2008-02-05 17:57:39 -0800851 }
Christoph Lameter81819f02007-05-06 14:49:36 -0700852 }
853
854 if (s->flags & SLAB_POISON) {
Christoph Lameterf7cb1932010-09-29 07:15:01 -0500855 if (val != SLUB_RED_ACTIVE && (s->flags & __OBJECT_POISON) &&
Christoph Lameter24922682007-07-17 04:03:18 -0700856 (!check_bytes_and_report(s, page, p, "Poison", p,
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500857 POISON_FREE, s->object_size - 1) ||
Christoph Lameter24922682007-07-17 04:03:18 -0700858 !check_bytes_and_report(s, page, p, "Poison",
Christoph Lameter3b0efdf2012-06-13 10:24:57 -0500859 p + s->object_size - 1, POISON_END, 1)))
Christoph Lameter81819f02007-05-06 14:49:36 -0700860 return 0;
Christoph Lameter81819f02007-05-06 14:49:36 -0700861 /*
862 * check_pad_bytes cleans up on its own.
863 */
864 check_pad_bytes(s, page, p);
865 }
866
Christoph Lameterf7cb1932010-09-29 07:15:01 -0500867 if (!s->offset && val == SLUB_RED_ACTIVE)
Christoph Lameter81819f02007-05-06 14:49:36 -0700868 /*
869 * Object and freepointer overlap. Cannot check
870 * freepointer while object is allocated.
871 */
872 return 1;
873
874 /* Check free pointer validity */
875 if (!check_valid_pointer(s, page, get_freepointer(s, p))) {
876 object_err(s, page, p, "Freepointer corrupt");
877 /*
Nick Andrew9f6c708e2008-12-05 14:08:08 +1100878 * No choice but to zap it and thus lose the remainder
Christoph Lameter81819f02007-05-06 14:49:36 -0700879 * of the free objects in this slab. May cause
Christoph Lameter672bba32007-05-09 02:32:39 -0700880 * another error because the object count is now wrong.
Christoph Lameter81819f02007-05-06 14:49:36 -0700881 */
Christoph Lametera973e9d2008-03-01 13:40:44 -0800882 set_freepointer(s, p, NULL);
Christoph Lameter81819f02007-05-06 14:49:36 -0700883 return 0;
884 }
885 return 1;
886}
887
888static int check_slab(struct kmem_cache *s, struct page *page)
889{
Christoph Lameter39b26462008-04-14 19:11:30 +0300890 int maxobj;
891
Christoph Lameter81819f02007-05-06 14:49:36 -0700892 VM_BUG_ON(!irqs_disabled());
893
894 if (!PageSlab(page)) {
Christoph Lameter24922682007-07-17 04:03:18 -0700895 slab_err(s, page, "Not a valid slab page");
Christoph Lameter81819f02007-05-06 14:49:36 -0700896 return 0;
897 }
Christoph Lameter39b26462008-04-14 19:11:30 +0300898
Lai Jiangshanab9a0f12011-03-10 15:21:48 +0800899 maxobj = order_objects(compound_order(page), s->size, s->reserved);
Christoph Lameter39b26462008-04-14 19:11:30 +0300900 if (page->objects > maxobj) {
901 slab_err(s, page, "objects %u > max %u",
Andrey Ryabininf6edde92014-12-10 15:42:22 -0800902 page->objects, maxobj);
Christoph Lameter39b26462008-04-14 19:11:30 +0300903 return 0;
904 }
905 if (page->inuse > page->objects) {
Christoph Lameter24922682007-07-17 04:03:18 -0700906 slab_err(s, page, "inuse %u > max %u",
Andrey Ryabininf6edde92014-12-10 15:42:22 -0800907 page->inuse, page->objects);
Christoph Lameter81819f02007-05-06 14:49:36 -0700908 return 0;
909 }
910 /* Slab_pad_check fixes things up after itself */
911 slab_pad_check(s, page);
912 return 1;
913}
914
915/*
Christoph Lameter672bba32007-05-09 02:32:39 -0700916 * Determine if a certain object on a page is on the freelist. Must hold the
917 * slab lock to guarantee that the chains are in a consistent state.
Christoph Lameter81819f02007-05-06 14:49:36 -0700918 */
919static int on_freelist(struct kmem_cache *s, struct page *page, void *search)
920{
921 int nr = 0;
Christoph Lameter881db7f2011-06-01 12:25:53 -0500922 void *fp;
Christoph Lameter81819f02007-05-06 14:49:36 -0700923 void *object = NULL;
Andrey Ryabininf6edde92014-12-10 15:42:22 -0800924 int max_objects;
Christoph Lameter81819f02007-05-06 14:49:36 -0700925
Christoph Lameter881db7f2011-06-01 12:25:53 -0500926 fp = page->freelist;
Christoph Lameter39b26462008-04-14 19:11:30 +0300927 while (fp && nr <= page->objects) {
Christoph Lameter81819f02007-05-06 14:49:36 -0700928 if (fp == search)
929 return 1;
930 if (!check_valid_pointer(s, page, fp)) {
931 if (object) {
932 object_err(s, page, object,
933 "Freechain corrupt");
Christoph Lametera973e9d2008-03-01 13:40:44 -0800934 set_freepointer(s, object, NULL);
Christoph Lameter81819f02007-05-06 14:49:36 -0700935 } else {
Christoph Lameter24922682007-07-17 04:03:18 -0700936 slab_err(s, page, "Freepointer corrupt");
Christoph Lametera973e9d2008-03-01 13:40:44 -0800937 page->freelist = NULL;
Christoph Lameter39b26462008-04-14 19:11:30 +0300938 page->inuse = page->objects;
Christoph Lameter24922682007-07-17 04:03:18 -0700939 slab_fix(s, "Freelist cleared");
Christoph Lameter81819f02007-05-06 14:49:36 -0700940 return 0;
941 }
942 break;
943 }
944 object = fp;
945 fp = get_freepointer(s, object);
946 nr++;
947 }
948
Lai Jiangshanab9a0f12011-03-10 15:21:48 +0800949 max_objects = order_objects(compound_order(page), s->size, s->reserved);
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +0400950 if (max_objects > MAX_OBJS_PER_PAGE)
951 max_objects = MAX_OBJS_PER_PAGE;
Christoph Lameter224a88b2008-04-14 19:11:31 +0300952
953 if (page->objects != max_objects) {
Joe Perches756a025f02016-03-17 14:19:47 -0700954 slab_err(s, page, "Wrong number of objects. Found %d but should be %d",
955 page->objects, max_objects);
Christoph Lameter224a88b2008-04-14 19:11:31 +0300956 page->objects = max_objects;
957 slab_fix(s, "Number of objects adjusted.");
958 }
Christoph Lameter39b26462008-04-14 19:11:30 +0300959 if (page->inuse != page->objects - nr) {
Joe Perches756a025f02016-03-17 14:19:47 -0700960 slab_err(s, page, "Wrong object count. Counter is %d but counted were %d",
961 page->inuse, page->objects - nr);
Christoph Lameter39b26462008-04-14 19:11:30 +0300962 page->inuse = page->objects - nr;
Christoph Lameter24922682007-07-17 04:03:18 -0700963 slab_fix(s, "Object count adjusted.");
Christoph Lameter81819f02007-05-06 14:49:36 -0700964 }
965 return search == NULL;
966}
967
Christoph Lameter0121c6192008-04-29 16:11:12 -0700968static void trace(struct kmem_cache *s, struct page *page, void *object,
969 int alloc)
Christoph Lameter3ec09742007-05-16 22:11:00 -0700970{
971 if (s->flags & SLAB_TRACE) {
Fabian Frederickf9f58282014-06-04 16:06:34 -0700972 pr_info("TRACE %s %s 0x%p inuse=%d fp=0x%p\n",
Christoph Lameter3ec09742007-05-16 22:11:00 -0700973 s->name,
974 alloc ? "alloc" : "free",
975 object, page->inuse,
976 page->freelist);
977
978 if (!alloc)
Daniel Thompson74b41632017-01-24 15:18:02 -0800979 print_section(KERN_INFO, "Object ", (void *)object,
Chen Gangd0e0ac92013-07-15 09:05:29 +0800980 s->object_size);
Christoph Lameter3ec09742007-05-16 22:11:00 -0700981
982 dump_stack();
983 }
984}
985
Christoph Lameter643b1132007-05-06 14:49:42 -0700986/*
Christoph Lameter672bba32007-05-09 02:32:39 -0700987 * Tracking of fully allocated slabs for debugging purposes.
Christoph Lameter643b1132007-05-06 14:49:42 -0700988 */
Christoph Lameter5cc6eee2011-06-01 12:25:50 -0500989static void add_full(struct kmem_cache *s,
990 struct kmem_cache_node *n, struct page *page)
Christoph Lameter643b1132007-05-06 14:49:42 -0700991{
Christoph Lameter643b1132007-05-06 14:49:42 -0700992 if (!(s->flags & SLAB_STORE_USER))
993 return;
994
David Rientjes255d0882014-02-10 14:25:39 -0800995 lockdep_assert_held(&n->list_lock);
Christoph Lameter5cc6eee2011-06-01 12:25:50 -0500996 list_add(&page->lru, &n->full);
997}
Christoph Lameter643b1132007-05-06 14:49:42 -0700998
Peter Zijlstrac65c1872014-01-10 13:23:49 +0100999static void remove_full(struct kmem_cache *s, struct kmem_cache_node *n, struct page *page)
Christoph Lameter5cc6eee2011-06-01 12:25:50 -05001000{
1001 if (!(s->flags & SLAB_STORE_USER))
1002 return;
1003
David Rientjes255d0882014-02-10 14:25:39 -08001004 lockdep_assert_held(&n->list_lock);
Christoph Lameter643b1132007-05-06 14:49:42 -07001005 list_del(&page->lru);
Christoph Lameter643b1132007-05-06 14:49:42 -07001006}
1007
Christoph Lameter0f389ec2008-04-14 18:53:02 +03001008/* Tracking of the number of slabs for debugging purposes */
1009static inline unsigned long slabs_node(struct kmem_cache *s, int node)
1010{
1011 struct kmem_cache_node *n = get_node(s, node);
1012
1013 return atomic_long_read(&n->nr_slabs);
1014}
1015
Alexander Beregalov26c02cf2009-06-11 14:08:48 +04001016static inline unsigned long node_nr_slabs(struct kmem_cache_node *n)
1017{
1018 return atomic_long_read(&n->nr_slabs);
1019}
1020
Christoph Lameter205ab992008-04-14 19:11:40 +03001021static inline void inc_slabs_node(struct kmem_cache *s, int node, int objects)
Christoph Lameter0f389ec2008-04-14 18:53:02 +03001022{
1023 struct kmem_cache_node *n = get_node(s, node);
1024
1025 /*
1026 * May be called early in order to allocate a slab for the
1027 * kmem_cache_node structure. Solve the chicken-egg
1028 * dilemma by deferring the increment of the count during
1029 * bootstrap (see early_kmem_cache_node_alloc).
1030 */
Joonsoo Kim338b2642013-01-21 17:01:27 +09001031 if (likely(n)) {
Christoph Lameter0f389ec2008-04-14 18:53:02 +03001032 atomic_long_inc(&n->nr_slabs);
Christoph Lameter205ab992008-04-14 19:11:40 +03001033 atomic_long_add(objects, &n->total_objects);
1034 }
Christoph Lameter0f389ec2008-04-14 18:53:02 +03001035}
Christoph Lameter205ab992008-04-14 19:11:40 +03001036static inline void dec_slabs_node(struct kmem_cache *s, int node, int objects)
Christoph Lameter0f389ec2008-04-14 18:53:02 +03001037{
1038 struct kmem_cache_node *n = get_node(s, node);
1039
1040 atomic_long_dec(&n->nr_slabs);
Christoph Lameter205ab992008-04-14 19:11:40 +03001041 atomic_long_sub(objects, &n->total_objects);
Christoph Lameter0f389ec2008-04-14 18:53:02 +03001042}
1043
1044/* Object debug checks for alloc/free paths */
Christoph Lameter3ec09742007-05-16 22:11:00 -07001045static void setup_object_debug(struct kmem_cache *s, struct page *page,
1046 void *object)
1047{
1048 if (!(s->flags & (SLAB_STORE_USER|SLAB_RED_ZONE|__OBJECT_POISON)))
1049 return;
1050
Christoph Lameterf7cb1932010-09-29 07:15:01 -05001051 init_object(s, object, SLUB_RED_INACTIVE);
Christoph Lameter3ec09742007-05-16 22:11:00 -07001052 init_tracking(s, object);
1053}
1054
Laura Abbottbecfda62016-03-15 14:55:06 -07001055static inline int alloc_consistency_checks(struct kmem_cache *s,
Chen Gangd0e0ac92013-07-15 09:05:29 +08001056 struct page *page,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03001057 void *object, unsigned long addr)
Christoph Lameter81819f02007-05-06 14:49:36 -07001058{
1059 if (!check_slab(s, page))
Laura Abbottbecfda62016-03-15 14:55:06 -07001060 return 0;
Christoph Lameter81819f02007-05-06 14:49:36 -07001061
Christoph Lameter81819f02007-05-06 14:49:36 -07001062 if (!check_valid_pointer(s, page, object)) {
1063 object_err(s, page, object, "Freelist Pointer check fails");
Laura Abbottbecfda62016-03-15 14:55:06 -07001064 return 0;
Christoph Lameter81819f02007-05-06 14:49:36 -07001065 }
1066
Christoph Lameterf7cb1932010-09-29 07:15:01 -05001067 if (!check_object(s, page, object, SLUB_RED_INACTIVE))
Laura Abbottbecfda62016-03-15 14:55:06 -07001068 return 0;
1069
1070 return 1;
1071}
1072
1073static noinline int alloc_debug_processing(struct kmem_cache *s,
1074 struct page *page,
1075 void *object, unsigned long addr)
1076{
1077 if (s->flags & SLAB_CONSISTENCY_CHECKS) {
1078 if (!alloc_consistency_checks(s, page, object, addr))
1079 goto bad;
1080 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001081
Christoph Lameter3ec09742007-05-16 22:11:00 -07001082 /* Success perform special debug activities for allocs */
1083 if (s->flags & SLAB_STORE_USER)
1084 set_track(s, object, TRACK_ALLOC, addr);
1085 trace(s, page, object, 1);
Christoph Lameterf7cb1932010-09-29 07:15:01 -05001086 init_object(s, object, SLUB_RED_ACTIVE);
Christoph Lameter81819f02007-05-06 14:49:36 -07001087 return 1;
Christoph Lameter3ec09742007-05-16 22:11:00 -07001088
Christoph Lameter81819f02007-05-06 14:49:36 -07001089bad:
1090 if (PageSlab(page)) {
1091 /*
1092 * If this is a slab page then lets do the best we can
1093 * to avoid issues in the future. Marking all objects
Christoph Lameter672bba32007-05-09 02:32:39 -07001094 * as used avoids touching the remaining objects.
Christoph Lameter81819f02007-05-06 14:49:36 -07001095 */
Christoph Lameter24922682007-07-17 04:03:18 -07001096 slab_fix(s, "Marking all objects used");
Christoph Lameter39b26462008-04-14 19:11:30 +03001097 page->inuse = page->objects;
Christoph Lametera973e9d2008-03-01 13:40:44 -08001098 page->freelist = NULL;
Christoph Lameter81819f02007-05-06 14:49:36 -07001099 }
1100 return 0;
1101}
1102
Laura Abbottbecfda62016-03-15 14:55:06 -07001103static inline int free_consistency_checks(struct kmem_cache *s,
1104 struct page *page, void *object, unsigned long addr)
1105{
1106 if (!check_valid_pointer(s, page, object)) {
1107 slab_err(s, page, "Invalid object pointer 0x%p", object);
1108 return 0;
1109 }
1110
1111 if (on_freelist(s, page, object)) {
1112 object_err(s, page, object, "Object already free");
1113 return 0;
1114 }
1115
1116 if (!check_object(s, page, object, SLUB_RED_ACTIVE))
1117 return 0;
1118
1119 if (unlikely(s != page->slab_cache)) {
1120 if (!PageSlab(page)) {
Joe Perches756a025f02016-03-17 14:19:47 -07001121 slab_err(s, page, "Attempt to free object(0x%p) outside of slab",
1122 object);
Laura Abbottbecfda62016-03-15 14:55:06 -07001123 } else if (!page->slab_cache) {
1124 pr_err("SLUB <none>: no slab for object 0x%p.\n",
1125 object);
1126 dump_stack();
1127 } else
1128 object_err(s, page, object,
1129 "page slab pointer corrupt.");
1130 return 0;
1131 }
1132 return 1;
1133}
1134
Jesper Dangaard Brouer81084652015-11-20 15:57:46 -08001135/* Supports checking bulk free of a constructed freelist */
Laura Abbott282acb42016-03-15 14:54:59 -07001136static noinline int free_debug_processing(
Jesper Dangaard Brouer81084652015-11-20 15:57:46 -08001137 struct kmem_cache *s, struct page *page,
1138 void *head, void *tail, int bulk_cnt,
Laura Abbott282acb42016-03-15 14:54:59 -07001139 unsigned long addr)
Christoph Lameter81819f02007-05-06 14:49:36 -07001140{
Christoph Lameter19c7ff92012-05-30 12:54:46 -05001141 struct kmem_cache_node *n = get_node(s, page_to_nid(page));
Jesper Dangaard Brouer81084652015-11-20 15:57:46 -08001142 void *object = head;
1143 int cnt = 0;
Laura Abbott282acb42016-03-15 14:54:59 -07001144 unsigned long uninitialized_var(flags);
Laura Abbott804aa132016-03-15 14:55:02 -07001145 int ret = 0;
Christoph Lameter5c2e4bb2011-06-01 12:25:54 -05001146
Laura Abbott282acb42016-03-15 14:54:59 -07001147 spin_lock_irqsave(&n->list_lock, flags);
Christoph Lameter881db7f2011-06-01 12:25:53 -05001148 slab_lock(page);
1149
Laura Abbottbecfda62016-03-15 14:55:06 -07001150 if (s->flags & SLAB_CONSISTENCY_CHECKS) {
1151 if (!check_slab(s, page))
1152 goto out;
1153 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001154
Jesper Dangaard Brouer81084652015-11-20 15:57:46 -08001155next_object:
1156 cnt++;
1157
Laura Abbottbecfda62016-03-15 14:55:06 -07001158 if (s->flags & SLAB_CONSISTENCY_CHECKS) {
1159 if (!free_consistency_checks(s, page, object, addr))
1160 goto out;
Christoph Lameter81819f02007-05-06 14:49:36 -07001161 }
Christoph Lameter3ec09742007-05-16 22:11:00 -07001162
Christoph Lameter3ec09742007-05-16 22:11:00 -07001163 if (s->flags & SLAB_STORE_USER)
1164 set_track(s, object, TRACK_FREE, addr);
1165 trace(s, page, object, 0);
Jesper Dangaard Brouer81084652015-11-20 15:57:46 -08001166 /* Freepointer not overwritten by init_object(), SLAB_POISON moved it */
Christoph Lameterf7cb1932010-09-29 07:15:01 -05001167 init_object(s, object, SLUB_RED_INACTIVE);
Jesper Dangaard Brouer81084652015-11-20 15:57:46 -08001168
1169 /* Reached end of constructed freelist yet? */
1170 if (object != tail) {
1171 object = get_freepointer(s, object);
1172 goto next_object;
1173 }
Laura Abbott804aa132016-03-15 14:55:02 -07001174 ret = 1;
1175
Christoph Lameter5c2e4bb2011-06-01 12:25:54 -05001176out:
Jesper Dangaard Brouer81084652015-11-20 15:57:46 -08001177 if (cnt != bulk_cnt)
1178 slab_err(s, page, "Bulk freelist count(%d) invalid(%d)\n",
1179 bulk_cnt, cnt);
1180
Christoph Lameter881db7f2011-06-01 12:25:53 -05001181 slab_unlock(page);
Laura Abbott282acb42016-03-15 14:54:59 -07001182 spin_unlock_irqrestore(&n->list_lock, flags);
Laura Abbott804aa132016-03-15 14:55:02 -07001183 if (!ret)
1184 slab_fix(s, "Object at 0x%p not freed", object);
1185 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07001186}
1187
Christoph Lameter41ecc552007-05-09 02:32:44 -07001188static int __init setup_slub_debug(char *str)
1189{
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001190 slub_debug = DEBUG_DEFAULT_FLAGS;
1191 if (*str++ != '=' || !*str)
1192 /*
1193 * No options specified. Switch on full debugging.
1194 */
1195 goto out;
Christoph Lameter41ecc552007-05-09 02:32:44 -07001196
1197 if (*str == ',')
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001198 /*
1199 * No options but restriction on slabs. This means full
1200 * debugging for slabs matching a pattern.
1201 */
1202 goto check_slabs;
1203
1204 slub_debug = 0;
1205 if (*str == '-')
1206 /*
1207 * Switch off all debugging measures.
1208 */
1209 goto out;
1210
1211 /*
1212 * Determine which debug features should be switched on
1213 */
Pekka Enberg06428782008-01-07 23:20:27 -08001214 for (; *str && *str != ','; str++) {
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001215 switch (tolower(*str)) {
1216 case 'f':
Laura Abbottbecfda62016-03-15 14:55:06 -07001217 slub_debug |= SLAB_CONSISTENCY_CHECKS;
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001218 break;
1219 case 'z':
1220 slub_debug |= SLAB_RED_ZONE;
1221 break;
1222 case 'p':
1223 slub_debug |= SLAB_POISON;
1224 break;
1225 case 'u':
1226 slub_debug |= SLAB_STORE_USER;
1227 break;
1228 case 't':
1229 slub_debug |= SLAB_TRACE;
1230 break;
Dmitry Monakhov4c13dd32010-02-26 09:36:12 +03001231 case 'a':
1232 slub_debug |= SLAB_FAILSLAB;
1233 break;
Chris J Arges08303a72015-04-14 15:44:25 -07001234 case 'o':
1235 /*
1236 * Avoid enabling debugging on caches if its minimum
1237 * order would increase as a result.
1238 */
1239 disable_higher_order_debug = 1;
1240 break;
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001241 default:
Fabian Frederickf9f58282014-06-04 16:06:34 -07001242 pr_err("slub_debug option '%c' unknown. skipped\n",
1243 *str);
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001244 }
1245 }
1246
1247check_slabs:
1248 if (*str == ',')
Christoph Lameter41ecc552007-05-09 02:32:44 -07001249 slub_debug_slabs = str + 1;
Christoph Lameterf0630ff2007-07-15 23:38:14 -07001250out:
Christoph Lameter41ecc552007-05-09 02:32:44 -07001251 return 1;
1252}
1253
1254__setup("slub_debug", setup_slub_debug);
1255
Joonsoo Kim423c9292014-10-09 15:26:22 -07001256unsigned long kmem_cache_flags(unsigned long object_size,
Christoph Lameterba0268a2007-09-11 15:24:11 -07001257 unsigned long flags, const char *name,
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07001258 void (*ctor)(void *))
Christoph Lameter41ecc552007-05-09 02:32:44 -07001259{
1260 /*
Christoph Lametere1533622008-02-15 23:45:24 -08001261 * Enable debugging if selected on the kernel commandline.
Christoph Lameter41ecc552007-05-09 02:32:44 -07001262 */
Christoph Lameterc6f58d92013-11-07 16:29:15 +00001263 if (slub_debug && (!slub_debug_slabs || (name &&
1264 !strncmp(slub_debug_slabs, name, strlen(slub_debug_slabs)))))
David Rientjes3de47212009-07-27 18:30:35 -07001265 flags |= slub_debug;
Christoph Lameterba0268a2007-09-11 15:24:11 -07001266
1267 return flags;
Christoph Lameter41ecc552007-05-09 02:32:44 -07001268}
Jesper Dangaard Brouerb4a64712015-11-20 15:57:41 -08001269#else /* !CONFIG_SLUB_DEBUG */
Christoph Lameter3ec09742007-05-16 22:11:00 -07001270static inline void setup_object_debug(struct kmem_cache *s,
1271 struct page *page, void *object) {}
Christoph Lameter41ecc552007-05-09 02:32:44 -07001272
Christoph Lameter3ec09742007-05-16 22:11:00 -07001273static inline int alloc_debug_processing(struct kmem_cache *s,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03001274 struct page *page, void *object, unsigned long addr) { return 0; }
Christoph Lameter41ecc552007-05-09 02:32:44 -07001275
Laura Abbott282acb42016-03-15 14:54:59 -07001276static inline int free_debug_processing(
Jesper Dangaard Brouer81084652015-11-20 15:57:46 -08001277 struct kmem_cache *s, struct page *page,
1278 void *head, void *tail, int bulk_cnt,
Laura Abbott282acb42016-03-15 14:54:59 -07001279 unsigned long addr) { return 0; }
Christoph Lameter41ecc552007-05-09 02:32:44 -07001280
Christoph Lameter41ecc552007-05-09 02:32:44 -07001281static inline int slab_pad_check(struct kmem_cache *s, struct page *page)
1282 { return 1; }
1283static inline int check_object(struct kmem_cache *s, struct page *page,
Christoph Lameterf7cb1932010-09-29 07:15:01 -05001284 void *object, u8 val) { return 1; }
Christoph Lameter5cc6eee2011-06-01 12:25:50 -05001285static inline void add_full(struct kmem_cache *s, struct kmem_cache_node *n,
1286 struct page *page) {}
Peter Zijlstrac65c1872014-01-10 13:23:49 +01001287static inline void remove_full(struct kmem_cache *s, struct kmem_cache_node *n,
1288 struct page *page) {}
Joonsoo Kim423c9292014-10-09 15:26:22 -07001289unsigned long kmem_cache_flags(unsigned long object_size,
Christoph Lameterba0268a2007-09-11 15:24:11 -07001290 unsigned long flags, const char *name,
Alexey Dobriyan51cc5062008-07-25 19:45:34 -07001291 void (*ctor)(void *))
Christoph Lameterba0268a2007-09-11 15:24:11 -07001292{
1293 return flags;
1294}
Christoph Lameter41ecc552007-05-09 02:32:44 -07001295#define slub_debug 0
Christoph Lameter0f389ec2008-04-14 18:53:02 +03001296
Ingo Molnarfdaa45e2009-09-15 11:00:26 +02001297#define disable_higher_order_debug 0
1298
Christoph Lameter0f389ec2008-04-14 18:53:02 +03001299static inline unsigned long slabs_node(struct kmem_cache *s, int node)
1300 { return 0; }
Alexander Beregalov26c02cf2009-06-11 14:08:48 +04001301static inline unsigned long node_nr_slabs(struct kmem_cache_node *n)
1302 { return 0; }
Christoph Lameter205ab992008-04-14 19:11:40 +03001303static inline void inc_slabs_node(struct kmem_cache *s, int node,
1304 int objects) {}
1305static inline void dec_slabs_node(struct kmem_cache *s, int node,
1306 int objects) {}
Christoph Lameter7d550c52010-08-25 14:07:16 -05001307
Andrey Ryabinin02e72cc2014-08-06 16:04:18 -07001308#endif /* CONFIG_SLUB_DEBUG */
1309
1310/*
1311 * Hooks for other subsystems that check memory allocations. In a typical
1312 * production configuration these hooks all should produce no code at all.
1313 */
Roman Bobnievd56791b2013-10-08 15:58:57 -07001314static inline void kmalloc_large_node_hook(void *ptr, size_t size, gfp_t flags)
1315{
1316 kmemleak_alloc(ptr, size, 1, flags);
Alexander Potapenko505f5dc2016-03-25 14:22:02 -07001317 kasan_kmalloc_large(ptr, size, flags);
Roman Bobnievd56791b2013-10-08 15:58:57 -07001318}
1319
1320static inline void kfree_hook(const void *x)
1321{
1322 kmemleak_free(x);
Andrey Ryabinin0316bec2015-02-13 14:39:42 -08001323 kasan_kfree_large(x);
Roman Bobnievd56791b2013-10-08 15:58:57 -07001324}
1325
Alexander Potapenko80a92012016-07-28 15:49:07 -07001326static inline void *slab_free_hook(struct kmem_cache *s, void *x)
Roman Bobnievd56791b2013-10-08 15:58:57 -07001327{
Alexander Potapenko80a92012016-07-28 15:49:07 -07001328 void *freeptr;
1329
Roman Bobnievd56791b2013-10-08 15:58:57 -07001330 kmemleak_free_recursive(x, s->flags);
Christoph Lameter7d550c52010-08-25 14:07:16 -05001331
Andrey Ryabinin02e72cc2014-08-06 16:04:18 -07001332 /*
1333 * Trouble is that we may no longer disable interrupts in the fast path
1334 * So in order to make the debug calls that expect irqs to be
1335 * disabled we need to disable interrupts temporarily.
1336 */
1337#if defined(CONFIG_KMEMCHECK) || defined(CONFIG_LOCKDEP)
1338 {
1339 unsigned long flags;
1340
1341 local_irq_save(flags);
1342 kmemcheck_slab_free(s, x, s->object_size);
1343 debug_check_no_locks_freed(x, s->object_size);
1344 local_irq_restore(flags);
1345 }
1346#endif
1347 if (!(s->flags & SLAB_DEBUG_OBJECTS))
1348 debug_check_no_obj_freed(x, s->object_size);
Andrey Ryabinin0316bec2015-02-13 14:39:42 -08001349
Alexander Potapenko80a92012016-07-28 15:49:07 -07001350 freeptr = get_freepointer(s, x);
1351 /*
1352 * kasan_slab_free() may put x into memory quarantine, delaying its
1353 * reuse. In this case the object's freelist pointer is changed.
1354 */
Andrey Ryabinin0316bec2015-02-13 14:39:42 -08001355 kasan_slab_free(s, x);
Alexander Potapenko80a92012016-07-28 15:49:07 -07001356 return freeptr;
Andrey Ryabinin02e72cc2014-08-06 16:04:18 -07001357}
Christoph Lameter205ab992008-04-14 19:11:40 +03001358
Jesper Dangaard Brouer81084652015-11-20 15:57:46 -08001359static inline void slab_free_freelist_hook(struct kmem_cache *s,
1360 void *head, void *tail)
1361{
1362/*
1363 * Compiler cannot detect this function can be removed if slab_free_hook()
1364 * evaluates to nothing. Thus, catch all relevant config debug options here.
1365 */
1366#if defined(CONFIG_KMEMCHECK) || \
1367 defined(CONFIG_LOCKDEP) || \
1368 defined(CONFIG_DEBUG_KMEMLEAK) || \
1369 defined(CONFIG_DEBUG_OBJECTS_FREE) || \
1370 defined(CONFIG_KASAN)
1371
1372 void *object = head;
1373 void *tail_obj = tail ? : head;
Alexander Potapenko80a92012016-07-28 15:49:07 -07001374 void *freeptr;
Jesper Dangaard Brouer81084652015-11-20 15:57:46 -08001375
1376 do {
Alexander Potapenko80a92012016-07-28 15:49:07 -07001377 freeptr = slab_free_hook(s, object);
1378 } while ((object != tail_obj) && (object = freeptr));
Jesper Dangaard Brouer81084652015-11-20 15:57:46 -08001379#endif
1380}
1381
Thomas Gleixner588f8ba2015-09-04 15:45:48 -07001382static void setup_object(struct kmem_cache *s, struct page *page,
1383 void *object)
1384{
1385 setup_object_debug(s, page, object);
Andrey Ryabininb3cbd9b2016-08-02 14:02:52 -07001386 kasan_init_slab_obj(s, object);
Thomas Gleixner588f8ba2015-09-04 15:45:48 -07001387 if (unlikely(s->ctor)) {
1388 kasan_unpoison_object_data(s, object);
1389 s->ctor(object);
1390 kasan_poison_object_data(s, object);
1391 }
1392}
1393
Christoph Lameter81819f02007-05-06 14:49:36 -07001394/*
1395 * Slab allocation and freeing
1396 */
Vladimir Davydov5dfb4172014-06-04 16:06:38 -07001397static inline struct page *alloc_slab_page(struct kmem_cache *s,
1398 gfp_t flags, int node, struct kmem_cache_order_objects oo)
Christoph Lameter65c33762008-04-14 19:11:40 +03001399{
Vladimir Davydov5dfb4172014-06-04 16:06:38 -07001400 struct page *page;
Christoph Lameter65c33762008-04-14 19:11:40 +03001401 int order = oo_order(oo);
1402
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001403 flags |= __GFP_NOTRACK;
1404
Christoph Lameter2154a332010-07-09 14:07:10 -05001405 if (node == NUMA_NO_NODE)
Vladimir Davydov5dfb4172014-06-04 16:06:38 -07001406 page = alloc_pages(flags, order);
Christoph Lameter65c33762008-04-14 19:11:40 +03001407 else
Vlastimil Babka96db8002015-09-08 15:03:50 -07001408 page = __alloc_pages_node(node, flags, order);
Vladimir Davydov5dfb4172014-06-04 16:06:38 -07001409
Vladimir Davydovf3ccb2c42015-11-05 18:49:01 -08001410 if (page && memcg_charge_slab(page, flags, order, s)) {
1411 __free_pages(page, order);
1412 page = NULL;
1413 }
Vladimir Davydov5dfb4172014-06-04 16:06:38 -07001414
1415 return page;
Christoph Lameter65c33762008-04-14 19:11:40 +03001416}
1417
Thomas Garnier210e7a42016-07-26 15:21:59 -07001418#ifdef CONFIG_SLAB_FREELIST_RANDOM
1419/* Pre-initialize the random sequence cache */
1420static int init_cache_random_seq(struct kmem_cache *s)
1421{
1422 int err;
1423 unsigned long i, count = oo_objects(s->oo);
1424
Sean Rees49f68cc2017-02-08 14:30:59 -08001425 /* Bailout if already initialised */
1426 if (s->random_seq)
1427 return 0;
1428
Thomas Garnier210e7a42016-07-26 15:21:59 -07001429 err = cache_random_seq_create(s, count, GFP_KERNEL);
1430 if (err) {
1431 pr_err("SLUB: Unable to initialize free list for %s\n",
1432 s->name);
1433 return err;
1434 }
1435
1436 /* Transform to an offset on the set of pages */
1437 if (s->random_seq) {
1438 for (i = 0; i < count; i++)
1439 s->random_seq[i] *= s->size;
1440 }
1441 return 0;
1442}
1443
1444/* Initialize each random sequence freelist per cache */
1445static void __init init_freelist_randomization(void)
1446{
1447 struct kmem_cache *s;
1448
1449 mutex_lock(&slab_mutex);
1450
1451 list_for_each_entry(s, &slab_caches, list)
1452 init_cache_random_seq(s);
1453
1454 mutex_unlock(&slab_mutex);
1455}
1456
1457/* Get the next entry on the pre-computed freelist randomized */
1458static void *next_freelist_entry(struct kmem_cache *s, struct page *page,
1459 unsigned long *pos, void *start,
1460 unsigned long page_limit,
1461 unsigned long freelist_count)
1462{
1463 unsigned int idx;
1464
1465 /*
1466 * If the target page allocation failed, the number of objects on the
1467 * page might be smaller than the usual size defined by the cache.
1468 */
1469 do {
1470 idx = s->random_seq[*pos];
1471 *pos += 1;
1472 if (*pos >= freelist_count)
1473 *pos = 0;
1474 } while (unlikely(idx >= page_limit));
1475
1476 return (char *)start + idx;
1477}
1478
1479/* Shuffle the single linked freelist based on a random pre-computed sequence */
1480static bool shuffle_freelist(struct kmem_cache *s, struct page *page)
1481{
1482 void *start;
1483 void *cur;
1484 void *next;
1485 unsigned long idx, pos, page_limit, freelist_count;
1486
1487 if (page->objects < 2 || !s->random_seq)
1488 return false;
1489
1490 freelist_count = oo_objects(s->oo);
1491 pos = get_random_int() % freelist_count;
1492
1493 page_limit = page->objects * s->size;
1494 start = fixup_red_left(s, page_address(page));
1495
1496 /* First entry is used as the base of the freelist */
1497 cur = next_freelist_entry(s, page, &pos, start, page_limit,
1498 freelist_count);
1499 page->freelist = cur;
1500
1501 for (idx = 1; idx < page->objects; idx++) {
1502 setup_object(s, page, cur);
1503 next = next_freelist_entry(s, page, &pos, start, page_limit,
1504 freelist_count);
1505 set_freepointer(s, cur, next);
1506 cur = next;
1507 }
1508 setup_object(s, page, cur);
1509 set_freepointer(s, cur, NULL);
1510
1511 return true;
1512}
1513#else
1514static inline int init_cache_random_seq(struct kmem_cache *s)
1515{
1516 return 0;
1517}
1518static inline void init_freelist_randomization(void) { }
1519static inline bool shuffle_freelist(struct kmem_cache *s, struct page *page)
1520{
1521 return false;
1522}
1523#endif /* CONFIG_SLAB_FREELIST_RANDOM */
1524
Christoph Lameter81819f02007-05-06 14:49:36 -07001525static struct page *allocate_slab(struct kmem_cache *s, gfp_t flags, int node)
1526{
Pekka Enberg06428782008-01-07 23:20:27 -08001527 struct page *page;
Christoph Lameter834f3d12008-04-14 19:11:31 +03001528 struct kmem_cache_order_objects oo = s->oo;
Pekka Enbergba522702009-06-24 21:59:51 +03001529 gfp_t alloc_gfp;
Thomas Gleixner588f8ba2015-09-04 15:45:48 -07001530 void *start, *p;
1531 int idx, order;
Thomas Garnier210e7a42016-07-26 15:21:59 -07001532 bool shuffle;
Christoph Lameter81819f02007-05-06 14:49:36 -07001533
Christoph Lameter7e0528d2011-06-01 12:25:44 -05001534 flags &= gfp_allowed_mask;
1535
Mel Gormand0164ad2015-11-06 16:28:21 -08001536 if (gfpflags_allow_blocking(flags))
Christoph Lameter7e0528d2011-06-01 12:25:44 -05001537 local_irq_enable();
1538
Christoph Lameterb7a49f02008-02-14 14:21:32 -08001539 flags |= s->allocflags;
Mel Gormane12ba742007-10-16 01:25:52 -07001540
Pekka Enbergba522702009-06-24 21:59:51 +03001541 /*
1542 * Let the initial higher-order allocation fail under memory pressure
1543 * so we fall-back to the minimum order allocation.
1544 */
1545 alloc_gfp = (flags | __GFP_NOWARN | __GFP_NORETRY) & ~__GFP_NOFAIL;
Mel Gormand0164ad2015-11-06 16:28:21 -08001546 if ((alloc_gfp & __GFP_DIRECT_RECLAIM) && oo_order(oo) > oo_order(s->min))
Mel Gorman444eb2a42016-03-17 14:19:23 -07001547 alloc_gfp = (alloc_gfp | __GFP_NOMEMALLOC) & ~(__GFP_RECLAIM|__GFP_NOFAIL);
Pekka Enbergba522702009-06-24 21:59:51 +03001548
Vladimir Davydov5dfb4172014-06-04 16:06:38 -07001549 page = alloc_slab_page(s, alloc_gfp, node, oo);
Christoph Lameter65c33762008-04-14 19:11:40 +03001550 if (unlikely(!page)) {
1551 oo = s->min;
Joonsoo Kim80c3a992014-03-12 17:26:20 +09001552 alloc_gfp = flags;
Christoph Lameter65c33762008-04-14 19:11:40 +03001553 /*
1554 * Allocation may have failed due to fragmentation.
1555 * Try a lower order alloc if possible
1556 */
Vladimir Davydov5dfb4172014-06-04 16:06:38 -07001557 page = alloc_slab_page(s, alloc_gfp, node, oo);
Thomas Gleixner588f8ba2015-09-04 15:45:48 -07001558 if (unlikely(!page))
1559 goto out;
1560 stat(s, ORDER_FALLBACK);
Christoph Lameter65c33762008-04-14 19:11:40 +03001561 }
Vegard Nossum5a896d92008-04-04 00:54:48 +02001562
Thomas Gleixner588f8ba2015-09-04 15:45:48 -07001563 if (kmemcheck_enabled &&
1564 !(s->flags & (SLAB_NOTRACK | DEBUG_DEFAULT_FLAGS))) {
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001565 int pages = 1 << oo_order(oo);
1566
Joonsoo Kim80c3a992014-03-12 17:26:20 +09001567 kmemcheck_alloc_shadow(page, oo_order(oo), alloc_gfp, node);
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001568
1569 /*
1570 * Objects from caches that have a constructor don't get
1571 * cleared when they're allocated, so we need to do it here.
1572 */
1573 if (s->ctor)
1574 kmemcheck_mark_uninitialized_pages(page, pages);
1575 else
1576 kmemcheck_mark_unallocated_pages(page, pages);
Vegard Nossum5a896d92008-04-04 00:54:48 +02001577 }
1578
Christoph Lameter834f3d12008-04-14 19:11:31 +03001579 page->objects = oo_objects(oo);
Christoph Lameter81819f02007-05-06 14:49:36 -07001580
Glauber Costa1f458cb2012-12-18 14:22:50 -08001581 order = compound_order(page);
Glauber Costa1b4f59e32012-10-22 18:05:36 +04001582 page->slab_cache = s;
Joonsoo Kimc03f94c2012-05-18 00:47:47 +09001583 __SetPageSlab(page);
Michal Hocko2f064f32015-08-21 14:11:51 -07001584 if (page_is_pfmemalloc(page))
Mel Gorman072bb0a2012-07-31 16:43:58 -07001585 SetPageSlabPfmemalloc(page);
Christoph Lameter81819f02007-05-06 14:49:36 -07001586
1587 start = page_address(page);
Christoph Lameter81819f02007-05-06 14:49:36 -07001588
1589 if (unlikely(s->flags & SLAB_POISON))
Glauber Costa1f458cb2012-12-18 14:22:50 -08001590 memset(start, POISON_INUSE, PAGE_SIZE << order);
Christoph Lameter81819f02007-05-06 14:49:36 -07001591
Andrey Ryabinin0316bec2015-02-13 14:39:42 -08001592 kasan_poison_slab(page);
1593
Thomas Garnier210e7a42016-07-26 15:21:59 -07001594 shuffle = shuffle_freelist(s, page);
1595
1596 if (!shuffle) {
1597 for_each_object_idx(p, idx, s, start, page->objects) {
1598 setup_object(s, page, p);
1599 if (likely(idx < page->objects))
1600 set_freepointer(s, p, p + s->size);
1601 else
1602 set_freepointer(s, p, NULL);
1603 }
1604 page->freelist = fixup_red_left(s, start);
Christoph Lameter81819f02007-05-06 14:49:36 -07001605 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001606
Christoph Lametere6e82ea2011-08-09 16:12:24 -05001607 page->inuse = page->objects;
Christoph Lameter8cb0a502011-06-01 12:25:46 -05001608 page->frozen = 1;
Thomas Gleixner588f8ba2015-09-04 15:45:48 -07001609
Christoph Lameter81819f02007-05-06 14:49:36 -07001610out:
Mel Gormand0164ad2015-11-06 16:28:21 -08001611 if (gfpflags_allow_blocking(flags))
Thomas Gleixner588f8ba2015-09-04 15:45:48 -07001612 local_irq_disable();
1613 if (!page)
1614 return NULL;
1615
1616 mod_zone_page_state(page_zone(page),
1617 (s->flags & SLAB_RECLAIM_ACCOUNT) ?
1618 NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE,
1619 1 << oo_order(oo));
1620
1621 inc_slabs_node(s, page_to_nid(page), page->objects);
1622
Christoph Lameter81819f02007-05-06 14:49:36 -07001623 return page;
1624}
1625
Thomas Gleixner588f8ba2015-09-04 15:45:48 -07001626static struct page *new_slab(struct kmem_cache *s, gfp_t flags, int node)
1627{
1628 if (unlikely(flags & GFP_SLAB_BUG_MASK)) {
Michal Hockobacdcb32016-07-26 15:22:02 -07001629 gfp_t invalid_mask = flags & GFP_SLAB_BUG_MASK;
Michal Hocko72baeef0c2016-07-26 15:22:05 -07001630 flags &= ~GFP_SLAB_BUG_MASK;
1631 pr_warn("Unexpected gfp: %#x (%pGg). Fixing up to gfp: %#x (%pGg). Fix your code!\n",
1632 invalid_mask, &invalid_mask, flags, &flags);
Thomas Gleixner588f8ba2015-09-04 15:45:48 -07001633 }
1634
1635 return allocate_slab(s,
1636 flags & (GFP_RECLAIM_MASK | GFP_CONSTRAINT_MASK), node);
1637}
1638
Christoph Lameter81819f02007-05-06 14:49:36 -07001639static void __free_slab(struct kmem_cache *s, struct page *page)
1640{
Christoph Lameter834f3d12008-04-14 19:11:31 +03001641 int order = compound_order(page);
1642 int pages = 1 << order;
Christoph Lameter81819f02007-05-06 14:49:36 -07001643
Laura Abbottbecfda62016-03-15 14:55:06 -07001644 if (s->flags & SLAB_CONSISTENCY_CHECKS) {
Christoph Lameter81819f02007-05-06 14:49:36 -07001645 void *p;
1646
1647 slab_pad_check(s, page);
Christoph Lameter224a88b2008-04-14 19:11:31 +03001648 for_each_object(p, s, page_address(page),
1649 page->objects)
Christoph Lameterf7cb1932010-09-29 07:15:01 -05001650 check_object(s, page, p, SLUB_RED_INACTIVE);
Christoph Lameter81819f02007-05-06 14:49:36 -07001651 }
1652
Vegard Nossumb1eeab62008-11-25 16:55:53 +01001653 kmemcheck_free_shadow(page, compound_order(page));
Vegard Nossum5a896d92008-04-04 00:54:48 +02001654
Christoph Lameter81819f02007-05-06 14:49:36 -07001655 mod_zone_page_state(page_zone(page),
1656 (s->flags & SLAB_RECLAIM_ACCOUNT) ?
1657 NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE,
Pekka Enberg06428782008-01-07 23:20:27 -08001658 -pages);
Christoph Lameter81819f02007-05-06 14:49:36 -07001659
Mel Gorman072bb0a2012-07-31 16:43:58 -07001660 __ClearPageSlabPfmemalloc(page);
Christoph Lameter49bd5222008-04-14 18:52:18 +03001661 __ClearPageSlab(page);
Glauber Costa1f458cb2012-12-18 14:22:50 -08001662
Mel Gorman22b751c2013-02-22 16:34:59 -08001663 page_mapcount_reset(page);
Nick Piggin1eb5ac62009-05-05 19:13:44 +10001664 if (current->reclaim_state)
1665 current->reclaim_state->reclaimed_slab += pages;
Vladimir Davydov27ee57c2016-03-17 14:17:35 -07001666 memcg_uncharge_slab(page, order, s);
1667 __free_pages(page, order);
Christoph Lameter81819f02007-05-06 14:49:36 -07001668}
1669
Lai Jiangshanda9a6382011-03-10 15:22:00 +08001670#define need_reserve_slab_rcu \
1671 (sizeof(((struct page *)NULL)->lru) < sizeof(struct rcu_head))
1672
Christoph Lameter81819f02007-05-06 14:49:36 -07001673static void rcu_free_slab(struct rcu_head *h)
1674{
1675 struct page *page;
1676
Lai Jiangshanda9a6382011-03-10 15:22:00 +08001677 if (need_reserve_slab_rcu)
1678 page = virt_to_head_page(h);
1679 else
1680 page = container_of((struct list_head *)h, struct page, lru);
1681
Glauber Costa1b4f59e32012-10-22 18:05:36 +04001682 __free_slab(page->slab_cache, page);
Christoph Lameter81819f02007-05-06 14:49:36 -07001683}
1684
1685static void free_slab(struct kmem_cache *s, struct page *page)
1686{
1687 if (unlikely(s->flags & SLAB_DESTROY_BY_RCU)) {
Lai Jiangshanda9a6382011-03-10 15:22:00 +08001688 struct rcu_head *head;
1689
1690 if (need_reserve_slab_rcu) {
1691 int order = compound_order(page);
1692 int offset = (PAGE_SIZE << order) - s->reserved;
1693
1694 VM_BUG_ON(s->reserved != sizeof(*head));
1695 head = page_address(page) + offset;
1696 } else {
Kirill A. Shutemovbc4f6102015-11-06 16:29:44 -08001697 head = &page->rcu_head;
Lai Jiangshanda9a6382011-03-10 15:22:00 +08001698 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001699
1700 call_rcu(head, rcu_free_slab);
1701 } else
1702 __free_slab(s, page);
1703}
1704
1705static void discard_slab(struct kmem_cache *s, struct page *page)
1706{
Christoph Lameter205ab992008-04-14 19:11:40 +03001707 dec_slabs_node(s, page_to_nid(page), page->objects);
Christoph Lameter81819f02007-05-06 14:49:36 -07001708 free_slab(s, page);
1709}
1710
1711/*
Christoph Lameter5cc6eee2011-06-01 12:25:50 -05001712 * Management of partially allocated slabs.
Christoph Lameter81819f02007-05-06 14:49:36 -07001713 */
Steven Rostedt1e4dd942014-02-10 14:25:46 -08001714static inline void
1715__add_partial(struct kmem_cache_node *n, struct page *page, int tail)
Christoph Lameter81819f02007-05-06 14:49:36 -07001716{
Christoph Lametere95eed52007-05-06 14:49:44 -07001717 n->nr_partial++;
Shaohua Li136333d2011-08-24 08:57:52 +08001718 if (tail == DEACTIVATE_TO_TAIL)
Christoph Lameter7c2e1322008-01-07 23:20:27 -08001719 list_add_tail(&page->lru, &n->partial);
1720 else
1721 list_add(&page->lru, &n->partial);
Christoph Lameter81819f02007-05-06 14:49:36 -07001722}
1723
Steven Rostedt1e4dd942014-02-10 14:25:46 -08001724static inline void add_partial(struct kmem_cache_node *n,
1725 struct page *page, int tail)
1726{
1727 lockdep_assert_held(&n->list_lock);
1728 __add_partial(n, page, tail);
1729}
1730
Christoph Lameter5cc6eee2011-06-01 12:25:50 -05001731static inline void remove_partial(struct kmem_cache_node *n,
Christoph Lameter62e346a2010-09-28 08:10:28 -05001732 struct page *page)
1733{
Peter Zijlstrac65c1872014-01-10 13:23:49 +01001734 lockdep_assert_held(&n->list_lock);
Dmitry Safonov52b4b952016-02-17 13:11:37 -08001735 list_del(&page->lru);
1736 n->nr_partial--;
Christoph Lameter62e346a2010-09-28 08:10:28 -05001737}
1738
Christoph Lameter81819f02007-05-06 14:49:36 -07001739/*
Christoph Lameter7ced3712012-05-09 10:09:53 -05001740 * Remove slab from the partial list, freeze it and
1741 * return the pointer to the freelist.
Christoph Lameter81819f02007-05-06 14:49:36 -07001742 *
Christoph Lameter497b66f2011-08-09 16:12:26 -05001743 * Returns a list of objects or NULL if it fails.
Christoph Lameter81819f02007-05-06 14:49:36 -07001744 */
Christoph Lameter497b66f2011-08-09 16:12:26 -05001745static inline void *acquire_slab(struct kmem_cache *s,
Christoph Lameteracd19fd2011-08-09 16:12:25 -05001746 struct kmem_cache_node *n, struct page *page,
Joonsoo Kim633b0762013-01-21 17:01:25 +09001747 int mode, int *objects)
Christoph Lameter81819f02007-05-06 14:49:36 -07001748{
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001749 void *freelist;
1750 unsigned long counters;
1751 struct page new;
1752
Peter Zijlstrac65c1872014-01-10 13:23:49 +01001753 lockdep_assert_held(&n->list_lock);
1754
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001755 /*
1756 * Zap the freelist and set the frozen bit.
1757 * The old freelist is the list of objects for the
1758 * per cpu allocation list.
1759 */
Christoph Lameter7ced3712012-05-09 10:09:53 -05001760 freelist = page->freelist;
1761 counters = page->counters;
1762 new.counters = counters;
Joonsoo Kim633b0762013-01-21 17:01:25 +09001763 *objects = new.objects - new.inuse;
Pekka Enberg23910c52012-06-04 10:14:58 +03001764 if (mode) {
Christoph Lameter7ced3712012-05-09 10:09:53 -05001765 new.inuse = page->objects;
Pekka Enberg23910c52012-06-04 10:14:58 +03001766 new.freelist = NULL;
1767 } else {
1768 new.freelist = freelist;
1769 }
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001770
Dave Hansena0132ac2014-01-29 14:05:50 -08001771 VM_BUG_ON(new.frozen);
Christoph Lameter7ced3712012-05-09 10:09:53 -05001772 new.frozen = 1;
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001773
Christoph Lameter7ced3712012-05-09 10:09:53 -05001774 if (!__cmpxchg_double_slab(s, page,
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001775 freelist, counters,
Joonsoo Kim02d76332012-05-17 00:13:02 +09001776 new.freelist, new.counters,
Christoph Lameter7ced3712012-05-09 10:09:53 -05001777 "acquire_slab"))
Christoph Lameter7ced3712012-05-09 10:09:53 -05001778 return NULL;
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001779
1780 remove_partial(n, page);
Christoph Lameter7ced3712012-05-09 10:09:53 -05001781 WARN_ON(!freelist);
Christoph Lameter49e22582011-08-09 16:12:27 -05001782 return freelist;
Christoph Lameter81819f02007-05-06 14:49:36 -07001783}
1784
Joonsoo Kim633b0762013-01-21 17:01:25 +09001785static void put_cpu_partial(struct kmem_cache *s, struct page *page, int drain);
Joonsoo Kim8ba00bb2012-09-17 14:09:09 -07001786static inline bool pfmemalloc_match(struct page *page, gfp_t gfpflags);
Christoph Lameter49e22582011-08-09 16:12:27 -05001787
Christoph Lameter81819f02007-05-06 14:49:36 -07001788/*
Christoph Lameter672bba32007-05-09 02:32:39 -07001789 * Try to allocate a partial slab from a specific node.
Christoph Lameter81819f02007-05-06 14:49:36 -07001790 */
Joonsoo Kim8ba00bb2012-09-17 14:09:09 -07001791static void *get_partial_node(struct kmem_cache *s, struct kmem_cache_node *n,
1792 struct kmem_cache_cpu *c, gfp_t flags)
Christoph Lameter81819f02007-05-06 14:49:36 -07001793{
Christoph Lameter49e22582011-08-09 16:12:27 -05001794 struct page *page, *page2;
1795 void *object = NULL;
Joonsoo Kim633b0762013-01-21 17:01:25 +09001796 int available = 0;
1797 int objects;
Christoph Lameter81819f02007-05-06 14:49:36 -07001798
1799 /*
1800 * Racy check. If we mistakenly see no partial slabs then we
1801 * just allocate an empty slab. If we mistakenly try to get a
Christoph Lameter672bba32007-05-09 02:32:39 -07001802 * partial slab and there is none available then get_partials()
1803 * will return NULL.
Christoph Lameter81819f02007-05-06 14:49:36 -07001804 */
1805 if (!n || !n->nr_partial)
1806 return NULL;
1807
1808 spin_lock(&n->list_lock);
Christoph Lameter49e22582011-08-09 16:12:27 -05001809 list_for_each_entry_safe(page, page2, &n->partial, lru) {
Joonsoo Kim8ba00bb2012-09-17 14:09:09 -07001810 void *t;
Christoph Lameter49e22582011-08-09 16:12:27 -05001811
Joonsoo Kim8ba00bb2012-09-17 14:09:09 -07001812 if (!pfmemalloc_match(page, flags))
1813 continue;
1814
Joonsoo Kim633b0762013-01-21 17:01:25 +09001815 t = acquire_slab(s, n, page, object == NULL, &objects);
Christoph Lameter49e22582011-08-09 16:12:27 -05001816 if (!t)
1817 break;
1818
Joonsoo Kim633b0762013-01-21 17:01:25 +09001819 available += objects;
Alex,Shi12d79632011-09-07 10:26:36 +08001820 if (!object) {
Christoph Lameter49e22582011-08-09 16:12:27 -05001821 c->page = page;
Christoph Lameter49e22582011-08-09 16:12:27 -05001822 stat(s, ALLOC_FROM_PARTIAL);
Christoph Lameter49e22582011-08-09 16:12:27 -05001823 object = t;
Christoph Lameter49e22582011-08-09 16:12:27 -05001824 } else {
Joonsoo Kim633b0762013-01-21 17:01:25 +09001825 put_cpu_partial(s, page, 0);
Alex Shi8028dce2012-02-03 23:34:56 +08001826 stat(s, CPU_PARTIAL_NODE);
Christoph Lameter49e22582011-08-09 16:12:27 -05001827 }
Joonsoo Kim345c9052013-06-19 14:05:52 +09001828 if (!kmem_cache_has_cpu_partial(s)
1829 || available > s->cpu_partial / 2)
Christoph Lameter49e22582011-08-09 16:12:27 -05001830 break;
1831
Christoph Lameter497b66f2011-08-09 16:12:26 -05001832 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001833 spin_unlock(&n->list_lock);
Christoph Lameter497b66f2011-08-09 16:12:26 -05001834 return object;
Christoph Lameter81819f02007-05-06 14:49:36 -07001835}
1836
1837/*
Christoph Lameter672bba32007-05-09 02:32:39 -07001838 * Get a page from somewhere. Search in increasing NUMA distances.
Christoph Lameter81819f02007-05-06 14:49:36 -07001839 */
Joonsoo Kimde3ec032012-01-27 00:12:23 -08001840static void *get_any_partial(struct kmem_cache *s, gfp_t flags,
Christoph Lameteracd19fd2011-08-09 16:12:25 -05001841 struct kmem_cache_cpu *c)
Christoph Lameter81819f02007-05-06 14:49:36 -07001842{
1843#ifdef CONFIG_NUMA
1844 struct zonelist *zonelist;
Mel Gormandd1a2392008-04-28 02:12:17 -07001845 struct zoneref *z;
Mel Gorman54a6eb52008-04-28 02:12:16 -07001846 struct zone *zone;
1847 enum zone_type high_zoneidx = gfp_zone(flags);
Christoph Lameter497b66f2011-08-09 16:12:26 -05001848 void *object;
Mel Gormancc9a6c82012-03-21 16:34:11 -07001849 unsigned int cpuset_mems_cookie;
Christoph Lameter81819f02007-05-06 14:49:36 -07001850
1851 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07001852 * The defrag ratio allows a configuration of the tradeoffs between
1853 * inter node defragmentation and node local allocations. A lower
1854 * defrag_ratio increases the tendency to do local allocations
1855 * instead of attempting to obtain partial slabs from other nodes.
Christoph Lameter81819f02007-05-06 14:49:36 -07001856 *
Christoph Lameter672bba32007-05-09 02:32:39 -07001857 * If the defrag_ratio is set to 0 then kmalloc() always
1858 * returns node local objects. If the ratio is higher then kmalloc()
1859 * may return off node objects because partial slabs are obtained
1860 * from other nodes and filled up.
Christoph Lameter81819f02007-05-06 14:49:36 -07001861 *
Li Peng43efd3e2016-05-19 17:10:43 -07001862 * If /sys/kernel/slab/xx/remote_node_defrag_ratio is set to 100
1863 * (which makes defrag_ratio = 1000) then every (well almost)
1864 * allocation will first attempt to defrag slab caches on other nodes.
1865 * This means scanning over all nodes to look for partial slabs which
1866 * may be expensive if we do it every time we are trying to find a slab
Christoph Lameter672bba32007-05-09 02:32:39 -07001867 * with available objects.
Christoph Lameter81819f02007-05-06 14:49:36 -07001868 */
Christoph Lameter98246012008-01-07 23:20:26 -08001869 if (!s->remote_node_defrag_ratio ||
1870 get_cycles() % 1024 > s->remote_node_defrag_ratio)
Christoph Lameter81819f02007-05-06 14:49:36 -07001871 return NULL;
1872
Mel Gormancc9a6c82012-03-21 16:34:11 -07001873 do {
Mel Gormand26914d2014-04-03 14:47:24 -07001874 cpuset_mems_cookie = read_mems_allowed_begin();
David Rientjes2a389612014-04-07 15:37:29 -07001875 zonelist = node_zonelist(mempolicy_slab_node(), flags);
Mel Gormancc9a6c82012-03-21 16:34:11 -07001876 for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) {
1877 struct kmem_cache_node *n;
Christoph Lameter81819f02007-05-06 14:49:36 -07001878
Mel Gormancc9a6c82012-03-21 16:34:11 -07001879 n = get_node(s, zone_to_nid(zone));
Christoph Lameter81819f02007-05-06 14:49:36 -07001880
Vladimir Davydovdee2f8a2014-12-12 16:58:28 -08001881 if (n && cpuset_zone_allowed(zone, flags) &&
Mel Gormancc9a6c82012-03-21 16:34:11 -07001882 n->nr_partial > s->min_partial) {
Joonsoo Kim8ba00bb2012-09-17 14:09:09 -07001883 object = get_partial_node(s, n, c, flags);
Mel Gormancc9a6c82012-03-21 16:34:11 -07001884 if (object) {
1885 /*
Mel Gormand26914d2014-04-03 14:47:24 -07001886 * Don't check read_mems_allowed_retry()
1887 * here - if mems_allowed was updated in
1888 * parallel, that was a harmless race
1889 * between allocation and the cpuset
1890 * update
Mel Gormancc9a6c82012-03-21 16:34:11 -07001891 */
Mel Gormancc9a6c82012-03-21 16:34:11 -07001892 return object;
1893 }
Miao Xiec0ff7452010-05-24 14:32:08 -07001894 }
Christoph Lameter81819f02007-05-06 14:49:36 -07001895 }
Mel Gormand26914d2014-04-03 14:47:24 -07001896 } while (read_mems_allowed_retry(cpuset_mems_cookie));
Christoph Lameter81819f02007-05-06 14:49:36 -07001897#endif
1898 return NULL;
1899}
1900
1901/*
1902 * Get a partial page, lock it and return it.
1903 */
Christoph Lameter497b66f2011-08-09 16:12:26 -05001904static void *get_partial(struct kmem_cache *s, gfp_t flags, int node,
Christoph Lameteracd19fd2011-08-09 16:12:25 -05001905 struct kmem_cache_cpu *c)
Christoph Lameter81819f02007-05-06 14:49:36 -07001906{
Christoph Lameter497b66f2011-08-09 16:12:26 -05001907 void *object;
Joonsoo Kima561ce02014-10-09 15:26:15 -07001908 int searchnode = node;
1909
1910 if (node == NUMA_NO_NODE)
1911 searchnode = numa_mem_id();
1912 else if (!node_present_pages(node))
1913 searchnode = node_to_mem_node(node);
Christoph Lameter81819f02007-05-06 14:49:36 -07001914
Joonsoo Kim8ba00bb2012-09-17 14:09:09 -07001915 object = get_partial_node(s, get_node(s, searchnode), c, flags);
Christoph Lameter497b66f2011-08-09 16:12:26 -05001916 if (object || node != NUMA_NO_NODE)
1917 return object;
Christoph Lameter81819f02007-05-06 14:49:36 -07001918
Christoph Lameteracd19fd2011-08-09 16:12:25 -05001919 return get_any_partial(s, flags, c);
Christoph Lameter81819f02007-05-06 14:49:36 -07001920}
1921
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001922#ifdef CONFIG_PREEMPT
1923/*
1924 * Calculate the next globally unique transaction for disambiguiation
1925 * during cmpxchg. The transactions start with the cpu number and are then
1926 * incremented by CONFIG_NR_CPUS.
1927 */
1928#define TID_STEP roundup_pow_of_two(CONFIG_NR_CPUS)
1929#else
1930/*
1931 * No preemption supported therefore also no need to check for
1932 * different cpus.
1933 */
1934#define TID_STEP 1
1935#endif
1936
1937static inline unsigned long next_tid(unsigned long tid)
1938{
1939 return tid + TID_STEP;
1940}
1941
1942static inline unsigned int tid_to_cpu(unsigned long tid)
1943{
1944 return tid % TID_STEP;
1945}
1946
1947static inline unsigned long tid_to_event(unsigned long tid)
1948{
1949 return tid / TID_STEP;
1950}
1951
1952static inline unsigned int init_tid(int cpu)
1953{
1954 return cpu;
1955}
1956
1957static inline void note_cmpxchg_failure(const char *n,
1958 const struct kmem_cache *s, unsigned long tid)
1959{
1960#ifdef SLUB_DEBUG_CMPXCHG
1961 unsigned long actual_tid = __this_cpu_read(s->cpu_slab->tid);
1962
Fabian Frederickf9f58282014-06-04 16:06:34 -07001963 pr_info("%s %s: cmpxchg redo ", n, s->name);
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001964
1965#ifdef CONFIG_PREEMPT
1966 if (tid_to_cpu(tid) != tid_to_cpu(actual_tid))
Fabian Frederickf9f58282014-06-04 16:06:34 -07001967 pr_warn("due to cpu change %d -> %d\n",
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001968 tid_to_cpu(tid), tid_to_cpu(actual_tid));
1969 else
1970#endif
1971 if (tid_to_event(tid) != tid_to_event(actual_tid))
Fabian Frederickf9f58282014-06-04 16:06:34 -07001972 pr_warn("due to cpu running other code. Event %ld->%ld\n",
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001973 tid_to_event(tid), tid_to_event(actual_tid));
1974 else
Fabian Frederickf9f58282014-06-04 16:06:34 -07001975 pr_warn("for unknown reason: actual=%lx was=%lx target=%lx\n",
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001976 actual_tid, tid, next_tid(tid));
1977#endif
Christoph Lameter4fdccdf2011-03-22 13:35:00 -05001978 stat(s, CMPXCHG_DOUBLE_CPU_FAIL);
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001979}
1980
Fengguang Wu788e1aa2012-09-28 16:34:05 +08001981static void init_kmem_cache_cpus(struct kmem_cache *s)
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001982{
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001983 int cpu;
1984
1985 for_each_possible_cpu(cpu)
1986 per_cpu_ptr(s->cpu_slab, cpu)->tid = init_tid(cpu);
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06001987}
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001988
1989/*
1990 * Remove the cpu slab
1991 */
Chen Gangd0e0ac92013-07-15 09:05:29 +08001992static void deactivate_slab(struct kmem_cache *s, struct page *page,
1993 void *freelist)
Christoph Lameter81819f02007-05-06 14:49:36 -07001994{
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001995 enum slab_modes { M_NONE, M_PARTIAL, M_FULL, M_FREE };
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001996 struct kmem_cache_node *n = get_node(s, page_to_nid(page));
1997 int lock = 0;
1998 enum slab_modes l = M_NONE, m = M_NONE;
Christoph Lameter2cfb7452011-06-01 12:25:52 -05001999 void *nextfree;
Shaohua Li136333d2011-08-24 08:57:52 +08002000 int tail = DEACTIVATE_TO_HEAD;
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002001 struct page new;
2002 struct page old;
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08002003
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002004 if (page->freelist) {
Christoph Lameter84e554e62009-12-18 16:26:23 -06002005 stat(s, DEACTIVATE_REMOTE_FREES);
Shaohua Li136333d2011-08-24 08:57:52 +08002006 tail = DEACTIVATE_TO_TAIL;
Christoph Lameter894b8782007-05-10 03:15:16 -07002007 }
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002008
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002009 /*
2010 * Stage one: Free all available per cpu objects back
2011 * to the page freelist while it is still frozen. Leave the
2012 * last one.
2013 *
2014 * There is no need to take the list->lock because the page
2015 * is still frozen.
2016 */
2017 while (freelist && (nextfree = get_freepointer(s, freelist))) {
2018 void *prior;
2019 unsigned long counters;
2020
2021 do {
2022 prior = page->freelist;
2023 counters = page->counters;
2024 set_freepointer(s, freelist, prior);
2025 new.counters = counters;
2026 new.inuse--;
Dave Hansena0132ac2014-01-29 14:05:50 -08002027 VM_BUG_ON(!new.frozen);
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002028
Christoph Lameter1d071712011-07-14 12:49:12 -05002029 } while (!__cmpxchg_double_slab(s, page,
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002030 prior, counters,
2031 freelist, new.counters,
2032 "drain percpu freelist"));
2033
2034 freelist = nextfree;
2035 }
2036
2037 /*
2038 * Stage two: Ensure that the page is unfrozen while the
2039 * list presence reflects the actual number of objects
2040 * during unfreeze.
2041 *
2042 * We setup the list membership and then perform a cmpxchg
2043 * with the count. If there is a mismatch then the page
2044 * is not unfrozen but the page is on the wrong list.
2045 *
2046 * Then we restart the process which may have to remove
2047 * the page from the list that we just put it on again
2048 * because the number of objects in the slab may have
2049 * changed.
2050 */
2051redo:
2052
2053 old.freelist = page->freelist;
2054 old.counters = page->counters;
Dave Hansena0132ac2014-01-29 14:05:50 -08002055 VM_BUG_ON(!old.frozen);
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002056
2057 /* Determine target state of the slab */
2058 new.counters = old.counters;
2059 if (freelist) {
2060 new.inuse--;
2061 set_freepointer(s, freelist, old.freelist);
2062 new.freelist = freelist;
2063 } else
2064 new.freelist = old.freelist;
2065
2066 new.frozen = 0;
2067
Joonsoo Kim8a5b20a2014-07-02 15:22:35 -07002068 if (!new.inuse && n->nr_partial >= s->min_partial)
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002069 m = M_FREE;
2070 else if (new.freelist) {
2071 m = M_PARTIAL;
2072 if (!lock) {
2073 lock = 1;
2074 /*
2075 * Taking the spinlock removes the possiblity
2076 * that acquire_slab() will see a slab page that
2077 * is frozen
2078 */
2079 spin_lock(&n->list_lock);
2080 }
2081 } else {
2082 m = M_FULL;
2083 if (kmem_cache_debug(s) && !lock) {
2084 lock = 1;
2085 /*
2086 * This also ensures that the scanning of full
2087 * slabs from diagnostic functions will not see
2088 * any frozen slabs.
2089 */
2090 spin_lock(&n->list_lock);
2091 }
2092 }
2093
2094 if (l != m) {
2095
2096 if (l == M_PARTIAL)
2097
2098 remove_partial(n, page);
2099
2100 else if (l == M_FULL)
2101
Peter Zijlstrac65c1872014-01-10 13:23:49 +01002102 remove_full(s, n, page);
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002103
2104 if (m == M_PARTIAL) {
2105
2106 add_partial(n, page, tail);
Shaohua Li136333d2011-08-24 08:57:52 +08002107 stat(s, tail);
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002108
2109 } else if (m == M_FULL) {
2110
2111 stat(s, DEACTIVATE_FULL);
2112 add_full(s, n, page);
2113
2114 }
2115 }
2116
2117 l = m;
Christoph Lameter1d071712011-07-14 12:49:12 -05002118 if (!__cmpxchg_double_slab(s, page,
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002119 old.freelist, old.counters,
2120 new.freelist, new.counters,
2121 "unfreezing slab"))
2122 goto redo;
2123
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002124 if (lock)
2125 spin_unlock(&n->list_lock);
2126
2127 if (m == M_FREE) {
2128 stat(s, DEACTIVATE_EMPTY);
2129 discard_slab(s, page);
2130 stat(s, FREE_SLAB);
2131 }
Christoph Lameter81819f02007-05-06 14:49:36 -07002132}
2133
Joonsoo Kimd24ac772012-05-18 22:01:17 +09002134/*
2135 * Unfreeze all the cpu partial slabs.
2136 *
Christoph Lameter59a09912012-11-28 16:23:00 +00002137 * This function must be called with interrupts disabled
2138 * for the cpu using c (or some other guarantee must be there
2139 * to guarantee no concurrent accesses).
Joonsoo Kimd24ac772012-05-18 22:01:17 +09002140 */
Christoph Lameter59a09912012-11-28 16:23:00 +00002141static void unfreeze_partials(struct kmem_cache *s,
2142 struct kmem_cache_cpu *c)
Christoph Lameter49e22582011-08-09 16:12:27 -05002143{
Joonsoo Kim345c9052013-06-19 14:05:52 +09002144#ifdef CONFIG_SLUB_CPU_PARTIAL
Joonsoo Kim43d77862012-06-09 02:23:16 +09002145 struct kmem_cache_node *n = NULL, *n2 = NULL;
Shaohua Li9ada1932011-11-14 13:34:13 +08002146 struct page *page, *discard_page = NULL;
Christoph Lameter49e22582011-08-09 16:12:27 -05002147
2148 while ((page = c->partial)) {
Christoph Lameter49e22582011-08-09 16:12:27 -05002149 struct page new;
2150 struct page old;
2151
2152 c->partial = page->next;
Joonsoo Kim43d77862012-06-09 02:23:16 +09002153
2154 n2 = get_node(s, page_to_nid(page));
2155 if (n != n2) {
2156 if (n)
2157 spin_unlock(&n->list_lock);
2158
2159 n = n2;
2160 spin_lock(&n->list_lock);
2161 }
Christoph Lameter49e22582011-08-09 16:12:27 -05002162
2163 do {
2164
2165 old.freelist = page->freelist;
2166 old.counters = page->counters;
Dave Hansena0132ac2014-01-29 14:05:50 -08002167 VM_BUG_ON(!old.frozen);
Christoph Lameter49e22582011-08-09 16:12:27 -05002168
2169 new.counters = old.counters;
2170 new.freelist = old.freelist;
2171
2172 new.frozen = 0;
2173
Joonsoo Kimd24ac772012-05-18 22:01:17 +09002174 } while (!__cmpxchg_double_slab(s, page,
Christoph Lameter49e22582011-08-09 16:12:27 -05002175 old.freelist, old.counters,
2176 new.freelist, new.counters,
2177 "unfreezing slab"));
2178
Joonsoo Kim8a5b20a2014-07-02 15:22:35 -07002179 if (unlikely(!new.inuse && n->nr_partial >= s->min_partial)) {
Shaohua Li9ada1932011-11-14 13:34:13 +08002180 page->next = discard_page;
2181 discard_page = page;
Joonsoo Kim43d77862012-06-09 02:23:16 +09002182 } else {
2183 add_partial(n, page, DEACTIVATE_TO_TAIL);
2184 stat(s, FREE_ADD_PARTIAL);
Christoph Lameter49e22582011-08-09 16:12:27 -05002185 }
2186 }
2187
2188 if (n)
2189 spin_unlock(&n->list_lock);
Shaohua Li9ada1932011-11-14 13:34:13 +08002190
2191 while (discard_page) {
2192 page = discard_page;
2193 discard_page = discard_page->next;
2194
2195 stat(s, DEACTIVATE_EMPTY);
2196 discard_slab(s, page);
2197 stat(s, FREE_SLAB);
2198 }
Joonsoo Kim345c9052013-06-19 14:05:52 +09002199#endif
Christoph Lameter49e22582011-08-09 16:12:27 -05002200}
2201
2202/*
2203 * Put a page that was just frozen (in __slab_free) into a partial page
2204 * slot if available. This is done without interrupts disabled and without
2205 * preemption disabled. The cmpxchg is racy and may put the partial page
2206 * onto a random cpus partial slot.
2207 *
2208 * If we did not find a slot then simply move all the partials to the
2209 * per node partial list.
2210 */
Joonsoo Kim633b0762013-01-21 17:01:25 +09002211static void put_cpu_partial(struct kmem_cache *s, struct page *page, int drain)
Christoph Lameter49e22582011-08-09 16:12:27 -05002212{
Joonsoo Kim345c9052013-06-19 14:05:52 +09002213#ifdef CONFIG_SLUB_CPU_PARTIAL
Christoph Lameter49e22582011-08-09 16:12:27 -05002214 struct page *oldpage;
2215 int pages;
2216 int pobjects;
2217
Vladimir Davydovd6e0b7f2015-02-12 14:59:47 -08002218 preempt_disable();
Christoph Lameter49e22582011-08-09 16:12:27 -05002219 do {
2220 pages = 0;
2221 pobjects = 0;
2222 oldpage = this_cpu_read(s->cpu_slab->partial);
2223
2224 if (oldpage) {
2225 pobjects = oldpage->pobjects;
2226 pages = oldpage->pages;
2227 if (drain && pobjects > s->cpu_partial) {
2228 unsigned long flags;
2229 /*
2230 * partial array is full. Move the existing
2231 * set to the per node partial list.
2232 */
2233 local_irq_save(flags);
Christoph Lameter59a09912012-11-28 16:23:00 +00002234 unfreeze_partials(s, this_cpu_ptr(s->cpu_slab));
Christoph Lameter49e22582011-08-09 16:12:27 -05002235 local_irq_restore(flags);
Joonsoo Kime24fc412012-06-23 03:22:38 +09002236 oldpage = NULL;
Christoph Lameter49e22582011-08-09 16:12:27 -05002237 pobjects = 0;
2238 pages = 0;
Alex Shi8028dce2012-02-03 23:34:56 +08002239 stat(s, CPU_PARTIAL_DRAIN);
Christoph Lameter49e22582011-08-09 16:12:27 -05002240 }
2241 }
2242
2243 pages++;
2244 pobjects += page->objects - page->inuse;
2245
2246 page->pages = pages;
2247 page->pobjects = pobjects;
2248 page->next = oldpage;
2249
Chen Gangd0e0ac92013-07-15 09:05:29 +08002250 } while (this_cpu_cmpxchg(s->cpu_slab->partial, oldpage, page)
2251 != oldpage);
Vladimir Davydovd6e0b7f2015-02-12 14:59:47 -08002252 if (unlikely(!s->cpu_partial)) {
2253 unsigned long flags;
2254
2255 local_irq_save(flags);
2256 unfreeze_partials(s, this_cpu_ptr(s->cpu_slab));
2257 local_irq_restore(flags);
2258 }
2259 preempt_enable();
Joonsoo Kim345c9052013-06-19 14:05:52 +09002260#endif
Christoph Lameter49e22582011-08-09 16:12:27 -05002261}
2262
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002263static inline void flush_slab(struct kmem_cache *s, struct kmem_cache_cpu *c)
Christoph Lameter81819f02007-05-06 14:49:36 -07002264{
Christoph Lameter84e554e62009-12-18 16:26:23 -06002265 stat(s, CPUSLAB_FLUSH);
Christoph Lameterc17dda42012-05-09 10:09:57 -05002266 deactivate_slab(s, c->page, c->freelist);
2267
2268 c->tid = next_tid(c->tid);
2269 c->page = NULL;
2270 c->freelist = NULL;
Christoph Lameter81819f02007-05-06 14:49:36 -07002271}
2272
2273/*
2274 * Flush cpu slab.
Christoph Lameter6446faa2008-02-15 23:45:26 -08002275 *
Christoph Lameter81819f02007-05-06 14:49:36 -07002276 * Called from IPI handler with interrupts disabled.
2277 */
Christoph Lameter0c710012007-07-17 04:03:24 -07002278static inline void __flush_cpu_slab(struct kmem_cache *s, int cpu)
Christoph Lameter81819f02007-05-06 14:49:36 -07002279{
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06002280 struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu);
Christoph Lameter81819f02007-05-06 14:49:36 -07002281
Christoph Lameter49e22582011-08-09 16:12:27 -05002282 if (likely(c)) {
2283 if (c->page)
2284 flush_slab(s, c);
2285
Christoph Lameter59a09912012-11-28 16:23:00 +00002286 unfreeze_partials(s, c);
Christoph Lameter49e22582011-08-09 16:12:27 -05002287 }
Christoph Lameter81819f02007-05-06 14:49:36 -07002288}
2289
2290static void flush_cpu_slab(void *d)
2291{
2292 struct kmem_cache *s = d;
Christoph Lameter81819f02007-05-06 14:49:36 -07002293
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002294 __flush_cpu_slab(s, smp_processor_id());
Christoph Lameter81819f02007-05-06 14:49:36 -07002295}
2296
Gilad Ben-Yossefa8364d52012-03-28 14:42:44 -07002297static bool has_cpu_slab(int cpu, void *info)
2298{
2299 struct kmem_cache *s = info;
2300 struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab, cpu);
2301
majianpeng02e1a9c2012-05-17 17:03:26 -07002302 return c->page || c->partial;
Gilad Ben-Yossefa8364d52012-03-28 14:42:44 -07002303}
2304
Christoph Lameter81819f02007-05-06 14:49:36 -07002305static void flush_all(struct kmem_cache *s)
2306{
Gilad Ben-Yossefa8364d52012-03-28 14:42:44 -07002307 on_each_cpu_cond(has_cpu_slab, flush_cpu_slab, s, 1, GFP_ATOMIC);
Christoph Lameter81819f02007-05-06 14:49:36 -07002308}
2309
2310/*
Sebastian Andrzej Siewiora96a87b2016-08-18 14:57:19 +02002311 * Use the cpu notifier to insure that the cpu slabs are flushed when
2312 * necessary.
2313 */
2314static int slub_cpu_dead(unsigned int cpu)
2315{
2316 struct kmem_cache *s;
2317 unsigned long flags;
2318
2319 mutex_lock(&slab_mutex);
2320 list_for_each_entry(s, &slab_caches, list) {
2321 local_irq_save(flags);
2322 __flush_cpu_slab(s, cpu);
2323 local_irq_restore(flags);
2324 }
2325 mutex_unlock(&slab_mutex);
2326 return 0;
2327}
2328
2329/*
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002330 * Check if the objects in a per cpu structure fit numa
2331 * locality expectations.
2332 */
Christoph Lameter57d437d2012-05-09 10:09:59 -05002333static inline int node_match(struct page *page, int node)
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002334{
2335#ifdef CONFIG_NUMA
Christoph Lameter4d7868e2013-01-23 21:45:47 +00002336 if (!page || (node != NUMA_NO_NODE && page_to_nid(page) != node))
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002337 return 0;
2338#endif
2339 return 1;
2340}
2341
David Rientjes9a02d692014-06-04 16:06:36 -07002342#ifdef CONFIG_SLUB_DEBUG
Pekka Enberg781b2ba2009-06-10 18:50:32 +03002343static int count_free(struct page *page)
2344{
2345 return page->objects - page->inuse;
2346}
2347
David Rientjes9a02d692014-06-04 16:06:36 -07002348static inline unsigned long node_nr_objs(struct kmem_cache_node *n)
2349{
2350 return atomic_long_read(&n->total_objects);
2351}
2352#endif /* CONFIG_SLUB_DEBUG */
2353
2354#if defined(CONFIG_SLUB_DEBUG) || defined(CONFIG_SYSFS)
Pekka Enberg781b2ba2009-06-10 18:50:32 +03002355static unsigned long count_partial(struct kmem_cache_node *n,
2356 int (*get_count)(struct page *))
2357{
2358 unsigned long flags;
2359 unsigned long x = 0;
2360 struct page *page;
2361
2362 spin_lock_irqsave(&n->list_lock, flags);
2363 list_for_each_entry(page, &n->partial, lru)
2364 x += get_count(page);
2365 spin_unlock_irqrestore(&n->list_lock, flags);
2366 return x;
2367}
David Rientjes9a02d692014-06-04 16:06:36 -07002368#endif /* CONFIG_SLUB_DEBUG || CONFIG_SYSFS */
Alexander Beregalov26c02cf2009-06-11 14:08:48 +04002369
Pekka Enberg781b2ba2009-06-10 18:50:32 +03002370static noinline void
2371slab_out_of_memory(struct kmem_cache *s, gfp_t gfpflags, int nid)
2372{
David Rientjes9a02d692014-06-04 16:06:36 -07002373#ifdef CONFIG_SLUB_DEBUG
2374 static DEFINE_RATELIMIT_STATE(slub_oom_rs, DEFAULT_RATELIMIT_INTERVAL,
2375 DEFAULT_RATELIMIT_BURST);
Pekka Enberg781b2ba2009-06-10 18:50:32 +03002376 int node;
Christoph Lameterfa45dc22014-08-06 16:04:09 -07002377 struct kmem_cache_node *n;
Pekka Enberg781b2ba2009-06-10 18:50:32 +03002378
David Rientjes9a02d692014-06-04 16:06:36 -07002379 if ((gfpflags & __GFP_NOWARN) || !__ratelimit(&slub_oom_rs))
2380 return;
2381
Vlastimil Babka5b3810e2016-03-15 14:56:33 -07002382 pr_warn("SLUB: Unable to allocate memory on node %d, gfp=%#x(%pGg)\n",
2383 nid, gfpflags, &gfpflags);
Fabian Frederickf9f58282014-06-04 16:06:34 -07002384 pr_warn(" cache: %s, object size: %d, buffer size: %d, default order: %d, min order: %d\n",
2385 s->name, s->object_size, s->size, oo_order(s->oo),
2386 oo_order(s->min));
Pekka Enberg781b2ba2009-06-10 18:50:32 +03002387
Christoph Lameter3b0efdf2012-06-13 10:24:57 -05002388 if (oo_order(s->min) > get_order(s->object_size))
Fabian Frederickf9f58282014-06-04 16:06:34 -07002389 pr_warn(" %s debugging increased min order, use slub_debug=O to disable.\n",
2390 s->name);
David Rientjesfa5ec8a2009-07-07 00:14:14 -07002391
Christoph Lameterfa45dc22014-08-06 16:04:09 -07002392 for_each_kmem_cache_node(s, node, n) {
Pekka Enberg781b2ba2009-06-10 18:50:32 +03002393 unsigned long nr_slabs;
2394 unsigned long nr_objs;
2395 unsigned long nr_free;
2396
Alexander Beregalov26c02cf2009-06-11 14:08:48 +04002397 nr_free = count_partial(n, count_free);
2398 nr_slabs = node_nr_slabs(n);
2399 nr_objs = node_nr_objs(n);
Pekka Enberg781b2ba2009-06-10 18:50:32 +03002400
Fabian Frederickf9f58282014-06-04 16:06:34 -07002401 pr_warn(" node %d: slabs: %ld, objs: %ld, free: %ld\n",
Pekka Enberg781b2ba2009-06-10 18:50:32 +03002402 node, nr_slabs, nr_objs, nr_free);
2403 }
David Rientjes9a02d692014-06-04 16:06:36 -07002404#endif
Pekka Enberg781b2ba2009-06-10 18:50:32 +03002405}
2406
Christoph Lameter497b66f2011-08-09 16:12:26 -05002407static inline void *new_slab_objects(struct kmem_cache *s, gfp_t flags,
2408 int node, struct kmem_cache_cpu **pc)
2409{
Christoph Lameter6faa6832012-05-09 10:09:51 -05002410 void *freelist;
Christoph Lameter188fd062012-05-09 10:09:55 -05002411 struct kmem_cache_cpu *c = *pc;
2412 struct page *page;
Christoph Lameter497b66f2011-08-09 16:12:26 -05002413
Christoph Lameter188fd062012-05-09 10:09:55 -05002414 freelist = get_partial(s, flags, node, c);
2415
2416 if (freelist)
2417 return freelist;
2418
2419 page = new_slab(s, flags, node);
Christoph Lameter497b66f2011-08-09 16:12:26 -05002420 if (page) {
Christoph Lameter7c8e0182014-06-04 16:07:56 -07002421 c = raw_cpu_ptr(s->cpu_slab);
Christoph Lameter497b66f2011-08-09 16:12:26 -05002422 if (c->page)
2423 flush_slab(s, c);
2424
2425 /*
2426 * No other reference to the page yet so we can
2427 * muck around with it freely without cmpxchg
2428 */
Christoph Lameter6faa6832012-05-09 10:09:51 -05002429 freelist = page->freelist;
Christoph Lameter497b66f2011-08-09 16:12:26 -05002430 page->freelist = NULL;
2431
2432 stat(s, ALLOC_SLAB);
Christoph Lameter497b66f2011-08-09 16:12:26 -05002433 c->page = page;
2434 *pc = c;
2435 } else
Christoph Lameter6faa6832012-05-09 10:09:51 -05002436 freelist = NULL;
Christoph Lameter497b66f2011-08-09 16:12:26 -05002437
Christoph Lameter6faa6832012-05-09 10:09:51 -05002438 return freelist;
Christoph Lameter497b66f2011-08-09 16:12:26 -05002439}
2440
Mel Gorman072bb0a2012-07-31 16:43:58 -07002441static inline bool pfmemalloc_match(struct page *page, gfp_t gfpflags)
2442{
2443 if (unlikely(PageSlabPfmemalloc(page)))
2444 return gfp_pfmemalloc_allowed(gfpflags);
2445
2446 return true;
2447}
2448
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002449/*
Chen Gangd0e0ac92013-07-15 09:05:29 +08002450 * Check the page->freelist of a page and either transfer the freelist to the
2451 * per cpu freelist or deactivate the page.
Christoph Lameter213eeb92011-11-11 14:07:14 -06002452 *
2453 * The page is still frozen if the return value is not NULL.
2454 *
2455 * If this function returns NULL then the page has been unfrozen.
Joonsoo Kimd24ac772012-05-18 22:01:17 +09002456 *
2457 * This function must be called with interrupt disabled.
Christoph Lameter213eeb92011-11-11 14:07:14 -06002458 */
2459static inline void *get_freelist(struct kmem_cache *s, struct page *page)
2460{
2461 struct page new;
2462 unsigned long counters;
2463 void *freelist;
2464
2465 do {
2466 freelist = page->freelist;
2467 counters = page->counters;
Christoph Lameter6faa6832012-05-09 10:09:51 -05002468
Christoph Lameter213eeb92011-11-11 14:07:14 -06002469 new.counters = counters;
Dave Hansena0132ac2014-01-29 14:05:50 -08002470 VM_BUG_ON(!new.frozen);
Christoph Lameter213eeb92011-11-11 14:07:14 -06002471
2472 new.inuse = page->objects;
2473 new.frozen = freelist != NULL;
2474
Joonsoo Kimd24ac772012-05-18 22:01:17 +09002475 } while (!__cmpxchg_double_slab(s, page,
Christoph Lameter213eeb92011-11-11 14:07:14 -06002476 freelist, counters,
2477 NULL, new.counters,
2478 "get_freelist"));
2479
2480 return freelist;
2481}
2482
2483/*
Christoph Lameter894b8782007-05-10 03:15:16 -07002484 * Slow path. The lockless freelist is empty or we need to perform
2485 * debugging duties.
Christoph Lameter81819f02007-05-06 14:49:36 -07002486 *
Christoph Lameter894b8782007-05-10 03:15:16 -07002487 * Processing is still very fast if new objects have been freed to the
2488 * regular freelist. In that case we simply take over the regular freelist
2489 * as the lockless freelist and zap the regular freelist.
Christoph Lameter81819f02007-05-06 14:49:36 -07002490 *
Christoph Lameter894b8782007-05-10 03:15:16 -07002491 * If that is not working then we fall back to the partial lists. We take the
2492 * first element of the freelist as the object to allocate now and move the
2493 * rest of the freelist to the lockless freelist.
2494 *
2495 * And if we were unable to get a new slab from the partial slab lists then
Christoph Lameter6446faa2008-02-15 23:45:26 -08002496 * we need to allocate a new slab. This is the slowest path since it involves
2497 * a call to the page allocator and the setup of a new slab.
Christoph Lametera380a3c2015-11-20 15:57:35 -08002498 *
2499 * Version of __slab_alloc to use when we know that interrupts are
2500 * already disabled (which is the case for bulk allocation).
Christoph Lameter81819f02007-05-06 14:49:36 -07002501 */
Christoph Lametera380a3c2015-11-20 15:57:35 -08002502static void *___slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03002503 unsigned long addr, struct kmem_cache_cpu *c)
Christoph Lameter81819f02007-05-06 14:49:36 -07002504{
Christoph Lameter6faa6832012-05-09 10:09:51 -05002505 void *freelist;
Christoph Lameterf6e7def2012-05-09 10:09:58 -05002506 struct page *page;
Christoph Lameter81819f02007-05-06 14:49:36 -07002507
Christoph Lameterf6e7def2012-05-09 10:09:58 -05002508 page = c->page;
2509 if (!page)
Christoph Lameter81819f02007-05-06 14:49:36 -07002510 goto new_slab;
Christoph Lameter49e22582011-08-09 16:12:27 -05002511redo:
Christoph Lameter6faa6832012-05-09 10:09:51 -05002512
Christoph Lameter57d437d2012-05-09 10:09:59 -05002513 if (unlikely(!node_match(page, node))) {
Joonsoo Kima561ce02014-10-09 15:26:15 -07002514 int searchnode = node;
2515
2516 if (node != NUMA_NO_NODE && !node_present_pages(node))
2517 searchnode = node_to_mem_node(node);
2518
2519 if (unlikely(!node_match(page, searchnode))) {
2520 stat(s, ALLOC_NODE_MISMATCH);
2521 deactivate_slab(s, page, c->freelist);
2522 c->page = NULL;
2523 c->freelist = NULL;
2524 goto new_slab;
2525 }
Christoph Lameterfc59c052011-06-01 12:25:56 -05002526 }
Christoph Lameter6446faa2008-02-15 23:45:26 -08002527
Mel Gorman072bb0a2012-07-31 16:43:58 -07002528 /*
2529 * By rights, we should be searching for a slab page that was
2530 * PFMEMALLOC but right now, we are losing the pfmemalloc
2531 * information when the page leaves the per-cpu allocator
2532 */
2533 if (unlikely(!pfmemalloc_match(page, gfpflags))) {
2534 deactivate_slab(s, page, c->freelist);
2535 c->page = NULL;
2536 c->freelist = NULL;
2537 goto new_slab;
2538 }
2539
Eric Dumazet73736e02011-12-13 04:57:06 +01002540 /* must check again c->freelist in case of cpu migration or IRQ */
Christoph Lameter6faa6832012-05-09 10:09:51 -05002541 freelist = c->freelist;
2542 if (freelist)
Eric Dumazet73736e02011-12-13 04:57:06 +01002543 goto load_freelist;
2544
Christoph Lameterf6e7def2012-05-09 10:09:58 -05002545 freelist = get_freelist(s, page);
Christoph Lameter6446faa2008-02-15 23:45:26 -08002546
Christoph Lameter6faa6832012-05-09 10:09:51 -05002547 if (!freelist) {
Christoph Lameter03e404a2011-06-01 12:25:58 -05002548 c->page = NULL;
2549 stat(s, DEACTIVATE_BYPASS);
Christoph Lameterfc59c052011-06-01 12:25:56 -05002550 goto new_slab;
Christoph Lameter03e404a2011-06-01 12:25:58 -05002551 }
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002552
Christoph Lameter81819f02007-05-06 14:49:36 -07002553 stat(s, ALLOC_REFILL);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08002554
Christoph Lameter894b8782007-05-10 03:15:16 -07002555load_freelist:
Christoph Lameter507effe2012-05-09 10:09:52 -05002556 /*
2557 * freelist is pointing to the list of objects to be used.
2558 * page is pointing to the page from which the objects are obtained.
2559 * That page must be frozen for per cpu allocations to work.
2560 */
Dave Hansena0132ac2014-01-29 14:05:50 -08002561 VM_BUG_ON(!c->page->frozen);
Christoph Lameter6faa6832012-05-09 10:09:51 -05002562 c->freelist = get_freepointer(s, freelist);
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002563 c->tid = next_tid(c->tid);
Christoph Lameter6faa6832012-05-09 10:09:51 -05002564 return freelist;
Christoph Lameter81819f02007-05-06 14:49:36 -07002565
Christoph Lameter81819f02007-05-06 14:49:36 -07002566new_slab:
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002567
Christoph Lameter49e22582011-08-09 16:12:27 -05002568 if (c->partial) {
Christoph Lameterf6e7def2012-05-09 10:09:58 -05002569 page = c->page = c->partial;
2570 c->partial = page->next;
Christoph Lameter49e22582011-08-09 16:12:27 -05002571 stat(s, CPU_PARTIAL_ALLOC);
2572 c->freelist = NULL;
2573 goto redo;
Christoph Lameter81819f02007-05-06 14:49:36 -07002574 }
2575
Christoph Lameter188fd062012-05-09 10:09:55 -05002576 freelist = new_slab_objects(s, gfpflags, node, &c);
Christoph Lameterb811c202007-10-16 23:25:51 -07002577
Christoph Lameterf46974362012-05-09 10:09:54 -05002578 if (unlikely(!freelist)) {
David Rientjes9a02d692014-06-04 16:06:36 -07002579 slab_out_of_memory(s, gfpflags, node);
Christoph Lameterf46974362012-05-09 10:09:54 -05002580 return NULL;
Christoph Lameter81819f02007-05-06 14:49:36 -07002581 }
Christoph Lameter894b8782007-05-10 03:15:16 -07002582
Christoph Lameterf6e7def2012-05-09 10:09:58 -05002583 page = c->page;
Christoph Lameter5091b742012-07-31 16:44:00 -07002584 if (likely(!kmem_cache_debug(s) && pfmemalloc_match(page, gfpflags)))
Christoph Lameter81819f02007-05-06 14:49:36 -07002585 goto load_freelist;
Christoph Lameter894b8782007-05-10 03:15:16 -07002586
Christoph Lameter497b66f2011-08-09 16:12:26 -05002587 /* Only entered in the debug case */
Chen Gangd0e0ac92013-07-15 09:05:29 +08002588 if (kmem_cache_debug(s) &&
2589 !alloc_debug_processing(s, page, freelist, addr))
Christoph Lameter497b66f2011-08-09 16:12:26 -05002590 goto new_slab; /* Slab failed checks. Next slab needed */
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002591
Christoph Lameterf6e7def2012-05-09 10:09:58 -05002592 deactivate_slab(s, page, get_freepointer(s, freelist));
Christoph Lameterc17dda42012-05-09 10:09:57 -05002593 c->page = NULL;
2594 c->freelist = NULL;
Christoph Lameter6faa6832012-05-09 10:09:51 -05002595 return freelist;
Christoph Lameter894b8782007-05-10 03:15:16 -07002596}
2597
2598/*
Christoph Lametera380a3c2015-11-20 15:57:35 -08002599 * Another one that disabled interrupt and compensates for possible
2600 * cpu changes by refetching the per cpu area pointer.
2601 */
2602static void *__slab_alloc(struct kmem_cache *s, gfp_t gfpflags, int node,
2603 unsigned long addr, struct kmem_cache_cpu *c)
2604{
2605 void *p;
2606 unsigned long flags;
2607
2608 local_irq_save(flags);
2609#ifdef CONFIG_PREEMPT
2610 /*
2611 * We may have been preempted and rescheduled on a different
2612 * cpu before disabling interrupts. Need to reload cpu area
2613 * pointer.
2614 */
2615 c = this_cpu_ptr(s->cpu_slab);
2616#endif
2617
2618 p = ___slab_alloc(s, gfpflags, node, addr, c);
2619 local_irq_restore(flags);
2620 return p;
2621}
2622
2623/*
Christoph Lameter894b8782007-05-10 03:15:16 -07002624 * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc)
2625 * have the fastpath folded into their functions. So no function call
2626 * overhead for requests that can be satisfied on the fastpath.
2627 *
2628 * The fastpath works by first checking if the lockless freelist can be used.
2629 * If not then __slab_alloc is called for slow processing.
2630 *
2631 * Otherwise we can simply pick the next object from the lockless free list.
2632 */
Ezequiel Garcia2b847c32012-09-08 17:47:58 -03002633static __always_inline void *slab_alloc_node(struct kmem_cache *s,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03002634 gfp_t gfpflags, int node, unsigned long addr)
Christoph Lameter894b8782007-05-10 03:15:16 -07002635{
Jesper Dangaard Brouer03ec0ed2015-11-20 15:57:52 -08002636 void *object;
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002637 struct kmem_cache_cpu *c;
Christoph Lameter57d437d2012-05-09 10:09:59 -05002638 struct page *page;
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002639 unsigned long tid;
Christoph Lameter1f842602008-01-07 23:20:30 -08002640
Vladimir Davydov8135be52014-12-12 16:56:38 -08002641 s = slab_pre_alloc_hook(s, gfpflags);
2642 if (!s)
Akinobu Mita773ff602008-12-23 19:37:01 +09002643 return NULL;
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002644redo:
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002645 /*
2646 * Must read kmem_cache cpu data via this cpu ptr. Preemption is
2647 * enabled. We may switch back and forth between cpus while
2648 * reading from one cpu area. That does not matter as long
2649 * as we end up on the original cpu again when doing the cmpxchg.
Christoph Lameter7cccd802013-01-23 21:45:48 +00002650 *
Joonsoo Kim9aabf812015-02-10 14:09:32 -08002651 * We should guarantee that tid and kmem_cache are retrieved on
2652 * the same cpu. It could be different if CONFIG_PREEMPT so we need
2653 * to check if it is matched or not.
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002654 */
Joonsoo Kim9aabf812015-02-10 14:09:32 -08002655 do {
2656 tid = this_cpu_read(s->cpu_slab->tid);
2657 c = raw_cpu_ptr(s->cpu_slab);
Mark Rutland859b7a02015-03-25 15:55:23 -07002658 } while (IS_ENABLED(CONFIG_PREEMPT) &&
2659 unlikely(tid != READ_ONCE(c->tid)));
Joonsoo Kim9aabf812015-02-10 14:09:32 -08002660
2661 /*
2662 * Irqless object alloc/free algorithm used here depends on sequence
2663 * of fetching cpu_slab's data. tid should be fetched before anything
2664 * on c to guarantee that object and page associated with previous tid
2665 * won't be used with current tid. If we fetch tid first, object and
2666 * page could be one associated with next tid and our alloc/free
2667 * request will be failed. In this case, we will retry. So, no problem.
2668 */
2669 barrier();
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002670
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002671 /*
2672 * The transaction ids are globally unique per cpu and per operation on
2673 * a per cpu queue. Thus they can be guarantee that the cmpxchg_double
2674 * occurs on the right processor and that there was no operation on the
2675 * linked list in between.
2676 */
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002677
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06002678 object = c->freelist;
Christoph Lameter57d437d2012-05-09 10:09:59 -05002679 page = c->page;
Dave Hansen8eae1492014-06-04 16:06:37 -07002680 if (unlikely(!object || !node_match(page, node))) {
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002681 object = __slab_alloc(s, gfpflags, node, addr, c);
Dave Hansen8eae1492014-06-04 16:06:37 -07002682 stat(s, ALLOC_SLOWPATH);
2683 } else {
Eric Dumazet0ad95002011-12-16 16:25:34 +01002684 void *next_object = get_freepointer_safe(s, object);
2685
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002686 /*
Lucas De Marchi25985ed2011-03-30 22:57:33 -03002687 * The cmpxchg will only match if there was no additional
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002688 * operation and if we are on the right processor.
2689 *
Chen Gangd0e0ac92013-07-15 09:05:29 +08002690 * The cmpxchg does the following atomically (without lock
2691 * semantics!)
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002692 * 1. Relocate first pointer to the current per cpu area.
2693 * 2. Verify that tid and freelist have not been changed
2694 * 3. If they were not changed replace tid and freelist
2695 *
Chen Gangd0e0ac92013-07-15 09:05:29 +08002696 * Since this is without lock semantics the protection is only
2697 * against code executing on this cpu *not* from access by
2698 * other cpus.
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002699 */
Christoph Lameter933393f2011-12-22 11:58:51 -06002700 if (unlikely(!this_cpu_cmpxchg_double(
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002701 s->cpu_slab->freelist, s->cpu_slab->tid,
2702 object, tid,
Eric Dumazet0ad95002011-12-16 16:25:34 +01002703 next_object, next_tid(tid)))) {
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002704
2705 note_cmpxchg_failure("slab_alloc", s, tid);
2706 goto redo;
2707 }
Eric Dumazet0ad95002011-12-16 16:25:34 +01002708 prefetch_freepointer(s, next_object);
Christoph Lameter84e554e62009-12-18 16:26:23 -06002709 stat(s, ALLOC_FASTPATH);
Christoph Lameter894b8782007-05-10 03:15:16 -07002710 }
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002711
Pekka Enberg74e21342009-11-25 20:14:48 +02002712 if (unlikely(gfpflags & __GFP_ZERO) && object)
Christoph Lameter3b0efdf2012-06-13 10:24:57 -05002713 memset(object, 0, s->object_size);
Christoph Lameterd07dbea2007-07-17 04:03:23 -07002714
Jesper Dangaard Brouer03ec0ed2015-11-20 15:57:52 -08002715 slab_post_alloc_hook(s, gfpflags, 1, &object);
Vegard Nossum5a896d92008-04-04 00:54:48 +02002716
Christoph Lameter894b8782007-05-10 03:15:16 -07002717 return object;
Christoph Lameter81819f02007-05-06 14:49:36 -07002718}
2719
Ezequiel Garcia2b847c32012-09-08 17:47:58 -03002720static __always_inline void *slab_alloc(struct kmem_cache *s,
2721 gfp_t gfpflags, unsigned long addr)
2722{
2723 return slab_alloc_node(s, gfpflags, NUMA_NO_NODE, addr);
2724}
2725
Christoph Lameter81819f02007-05-06 14:49:36 -07002726void *kmem_cache_alloc(struct kmem_cache *s, gfp_t gfpflags)
2727{
Ezequiel Garcia2b847c32012-09-08 17:47:58 -03002728 void *ret = slab_alloc(s, gfpflags, _RET_IP_);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002729
Chen Gangd0e0ac92013-07-15 09:05:29 +08002730 trace_kmem_cache_alloc(_RET_IP_, ret, s->object_size,
2731 s->size, gfpflags);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002732
2733 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07002734}
2735EXPORT_SYMBOL(kmem_cache_alloc);
2736
Li Zefan0f24f122009-12-11 15:45:30 +08002737#ifdef CONFIG_TRACING
Richard Kennedy4a923792010-10-21 10:29:19 +01002738void *kmem_cache_alloc_trace(struct kmem_cache *s, gfp_t gfpflags, size_t size)
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002739{
Ezequiel Garcia2b847c32012-09-08 17:47:58 -03002740 void *ret = slab_alloc(s, gfpflags, _RET_IP_);
Richard Kennedy4a923792010-10-21 10:29:19 +01002741 trace_kmalloc(_RET_IP_, ret, size, s->size, gfpflags);
Alexander Potapenko505f5dc2016-03-25 14:22:02 -07002742 kasan_kmalloc(s, ret, size, gfpflags);
Richard Kennedy4a923792010-10-21 10:29:19 +01002743 return ret;
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002744}
Richard Kennedy4a923792010-10-21 10:29:19 +01002745EXPORT_SYMBOL(kmem_cache_alloc_trace);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002746#endif
2747
Christoph Lameter81819f02007-05-06 14:49:36 -07002748#ifdef CONFIG_NUMA
2749void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t gfpflags, int node)
2750{
Ezequiel Garcia2b847c32012-09-08 17:47:58 -03002751 void *ret = slab_alloc_node(s, gfpflags, node, _RET_IP_);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002752
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02002753 trace_kmem_cache_alloc_node(_RET_IP_, ret,
Christoph Lameter3b0efdf2012-06-13 10:24:57 -05002754 s->object_size, s->size, gfpflags, node);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002755
2756 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07002757}
2758EXPORT_SYMBOL(kmem_cache_alloc_node);
Christoph Lameter81819f02007-05-06 14:49:36 -07002759
Li Zefan0f24f122009-12-11 15:45:30 +08002760#ifdef CONFIG_TRACING
Richard Kennedy4a923792010-10-21 10:29:19 +01002761void *kmem_cache_alloc_node_trace(struct kmem_cache *s,
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002762 gfp_t gfpflags,
Richard Kennedy4a923792010-10-21 10:29:19 +01002763 int node, size_t size)
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002764{
Ezequiel Garcia2b847c32012-09-08 17:47:58 -03002765 void *ret = slab_alloc_node(s, gfpflags, node, _RET_IP_);
Richard Kennedy4a923792010-10-21 10:29:19 +01002766
2767 trace_kmalloc_node(_RET_IP_, ret,
2768 size, s->size, gfpflags, node);
Andrey Ryabinin0316bec2015-02-13 14:39:42 -08002769
Alexander Potapenko505f5dc2016-03-25 14:22:02 -07002770 kasan_kmalloc(s, ret, size, gfpflags);
Richard Kennedy4a923792010-10-21 10:29:19 +01002771 return ret;
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002772}
Richard Kennedy4a923792010-10-21 10:29:19 +01002773EXPORT_SYMBOL(kmem_cache_alloc_node_trace);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002774#endif
Namhyung Kim5d1f57e2010-09-29 21:02:15 +09002775#endif
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03002776
Christoph Lameter81819f02007-05-06 14:49:36 -07002777/*
Kim Phillips94e4d712015-02-10 14:09:37 -08002778 * Slow path handling. This may still be called frequently since objects
Christoph Lameter894b8782007-05-10 03:15:16 -07002779 * have a longer lifetime than the cpu slabs in most processing loads.
Christoph Lameter81819f02007-05-06 14:49:36 -07002780 *
Christoph Lameter894b8782007-05-10 03:15:16 -07002781 * So we still attempt to reduce cache line usage. Just take the slab
2782 * lock and free the item. If there is no additional partial page
2783 * handling required then we can return immediately.
Christoph Lameter81819f02007-05-06 14:49:36 -07002784 */
Christoph Lameter894b8782007-05-10 03:15:16 -07002785static void __slab_free(struct kmem_cache *s, struct page *page,
Jesper Dangaard Brouer81084652015-11-20 15:57:46 -08002786 void *head, void *tail, int cnt,
2787 unsigned long addr)
2788
Christoph Lameter81819f02007-05-06 14:49:36 -07002789{
2790 void *prior;
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002791 int was_frozen;
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002792 struct page new;
2793 unsigned long counters;
2794 struct kmem_cache_node *n = NULL;
Christoph Lameter61728d12011-06-01 12:25:51 -05002795 unsigned long uninitialized_var(flags);
Christoph Lameter81819f02007-05-06 14:49:36 -07002796
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002797 stat(s, FREE_SLOWPATH);
Christoph Lameter81819f02007-05-06 14:49:36 -07002798
Christoph Lameter19c7ff92012-05-30 12:54:46 -05002799 if (kmem_cache_debug(s) &&
Laura Abbott282acb42016-03-15 14:54:59 -07002800 !free_debug_processing(s, page, head, tail, cnt, addr))
Christoph Lameter80f08c12011-06-01 12:25:55 -05002801 return;
Christoph Lameter6446faa2008-02-15 23:45:26 -08002802
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002803 do {
Joonsoo Kim837d6782012-08-16 00:02:40 +09002804 if (unlikely(n)) {
2805 spin_unlock_irqrestore(&n->list_lock, flags);
2806 n = NULL;
2807 }
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002808 prior = page->freelist;
2809 counters = page->counters;
Jesper Dangaard Brouer81084652015-11-20 15:57:46 -08002810 set_freepointer(s, tail, prior);
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002811 new.counters = counters;
2812 was_frozen = new.frozen;
Jesper Dangaard Brouer81084652015-11-20 15:57:46 -08002813 new.inuse -= cnt;
Joonsoo Kim837d6782012-08-16 00:02:40 +09002814 if ((!new.inuse || !prior) && !was_frozen) {
Christoph Lameter49e22582011-08-09 16:12:27 -05002815
Peter Zijlstrac65c1872014-01-10 13:23:49 +01002816 if (kmem_cache_has_cpu_partial(s) && !prior) {
Christoph Lameter49e22582011-08-09 16:12:27 -05002817
2818 /*
Chen Gangd0e0ac92013-07-15 09:05:29 +08002819 * Slab was on no list before and will be
2820 * partially empty
2821 * We can defer the list move and instead
2822 * freeze it.
Christoph Lameter49e22582011-08-09 16:12:27 -05002823 */
2824 new.frozen = 1;
2825
Peter Zijlstrac65c1872014-01-10 13:23:49 +01002826 } else { /* Needs to be taken off a list */
Christoph Lameter49e22582011-08-09 16:12:27 -05002827
LQYMGTb455def2014-12-10 15:42:13 -08002828 n = get_node(s, page_to_nid(page));
Christoph Lameter49e22582011-08-09 16:12:27 -05002829 /*
2830 * Speculatively acquire the list_lock.
2831 * If the cmpxchg does not succeed then we may
2832 * drop the list_lock without any processing.
2833 *
2834 * Otherwise the list_lock will synchronize with
2835 * other processors updating the list of slabs.
2836 */
2837 spin_lock_irqsave(&n->list_lock, flags);
2838
2839 }
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002840 }
Christoph Lameter81819f02007-05-06 14:49:36 -07002841
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002842 } while (!cmpxchg_double_slab(s, page,
2843 prior, counters,
Jesper Dangaard Brouer81084652015-11-20 15:57:46 -08002844 head, new.counters,
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002845 "__slab_free"));
Christoph Lameter81819f02007-05-06 14:49:36 -07002846
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002847 if (likely(!n)) {
Christoph Lameter49e22582011-08-09 16:12:27 -05002848
2849 /*
2850 * If we just froze the page then put it onto the
2851 * per cpu partial list.
2852 */
Alex Shi8028dce2012-02-03 23:34:56 +08002853 if (new.frozen && !was_frozen) {
Christoph Lameter49e22582011-08-09 16:12:27 -05002854 put_cpu_partial(s, page, 1);
Alex Shi8028dce2012-02-03 23:34:56 +08002855 stat(s, CPU_PARTIAL_FREE);
2856 }
Christoph Lameter49e22582011-08-09 16:12:27 -05002857 /*
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002858 * The list lock was not taken therefore no list
2859 * activity can be necessary.
2860 */
LQYMGTb455def2014-12-10 15:42:13 -08002861 if (was_frozen)
2862 stat(s, FREE_FROZEN);
2863 return;
2864 }
Christoph Lameter81819f02007-05-06 14:49:36 -07002865
Joonsoo Kim8a5b20a2014-07-02 15:22:35 -07002866 if (unlikely(!new.inuse && n->nr_partial >= s->min_partial))
Joonsoo Kim837d6782012-08-16 00:02:40 +09002867 goto slab_empty;
Christoph Lameter81819f02007-05-06 14:49:36 -07002868
Joonsoo Kim837d6782012-08-16 00:02:40 +09002869 /*
2870 * Objects left in the slab. If it was not on the partial list before
2871 * then add it.
2872 */
Joonsoo Kim345c9052013-06-19 14:05:52 +09002873 if (!kmem_cache_has_cpu_partial(s) && unlikely(!prior)) {
2874 if (kmem_cache_debug(s))
Peter Zijlstrac65c1872014-01-10 13:23:49 +01002875 remove_full(s, n, page);
Joonsoo Kim837d6782012-08-16 00:02:40 +09002876 add_partial(n, page, DEACTIVATE_TO_TAIL);
2877 stat(s, FREE_ADD_PARTIAL);
Christoph Lameter81819f02007-05-06 14:49:36 -07002878 }
Christoph Lameter80f08c12011-06-01 12:25:55 -05002879 spin_unlock_irqrestore(&n->list_lock, flags);
Christoph Lameter81819f02007-05-06 14:49:36 -07002880 return;
2881
2882slab_empty:
Christoph Lametera973e9d2008-03-01 13:40:44 -08002883 if (prior) {
Christoph Lameter81819f02007-05-06 14:49:36 -07002884 /*
Christoph Lameter6fbabb22011-08-08 11:16:56 -05002885 * Slab on the partial list.
Christoph Lameter81819f02007-05-06 14:49:36 -07002886 */
Christoph Lameter5cc6eee2011-06-01 12:25:50 -05002887 remove_partial(n, page);
Christoph Lameter84e554e62009-12-18 16:26:23 -06002888 stat(s, FREE_REMOVE_PARTIAL);
Peter Zijlstrac65c1872014-01-10 13:23:49 +01002889 } else {
Christoph Lameter6fbabb22011-08-08 11:16:56 -05002890 /* Slab must be on the full list */
Peter Zijlstrac65c1872014-01-10 13:23:49 +01002891 remove_full(s, n, page);
2892 }
Christoph Lameter2cfb7452011-06-01 12:25:52 -05002893
Christoph Lameter80f08c12011-06-01 12:25:55 -05002894 spin_unlock_irqrestore(&n->list_lock, flags);
Christoph Lameter84e554e62009-12-18 16:26:23 -06002895 stat(s, FREE_SLAB);
Christoph Lameter81819f02007-05-06 14:49:36 -07002896 discard_slab(s, page);
Christoph Lameter81819f02007-05-06 14:49:36 -07002897}
2898
Christoph Lameter894b8782007-05-10 03:15:16 -07002899/*
2900 * Fastpath with forced inlining to produce a kfree and kmem_cache_free that
2901 * can perform fastpath freeing without additional function calls.
2902 *
2903 * The fastpath is only possible if we are freeing to the current cpu slab
2904 * of this processor. This typically the case if we have just allocated
2905 * the item before.
2906 *
2907 * If fastpath is not possible then fall back to __slab_free where we deal
2908 * with all sorts of special processing.
Jesper Dangaard Brouer81084652015-11-20 15:57:46 -08002909 *
2910 * Bulk free of a freelist with several objects (all pointing to the
2911 * same page) possible by specifying head and tail ptr, plus objects
2912 * count (cnt). Bulk free indicated by tail pointer being set.
Christoph Lameter894b8782007-05-10 03:15:16 -07002913 */
Alexander Potapenko80a92012016-07-28 15:49:07 -07002914static __always_inline void do_slab_free(struct kmem_cache *s,
2915 struct page *page, void *head, void *tail,
2916 int cnt, unsigned long addr)
Christoph Lameter894b8782007-05-10 03:15:16 -07002917{
Jesper Dangaard Brouer81084652015-11-20 15:57:46 -08002918 void *tail_obj = tail ? : head;
Christoph Lameterdfb4f092007-10-16 01:26:05 -07002919 struct kmem_cache_cpu *c;
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002920 unsigned long tid;
Christoph Lametera24c5a02011-03-15 12:45:21 -05002921redo:
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002922 /*
2923 * Determine the currently cpus per cpu slab.
2924 * The cpu may change afterward. However that does not matter since
2925 * data is retrieved via this pointer. If we are on the same cpu
Jesper Dangaard Brouer2ae44002015-09-04 15:45:31 -07002926 * during the cmpxchg then the free will succeed.
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002927 */
Joonsoo Kim9aabf812015-02-10 14:09:32 -08002928 do {
2929 tid = this_cpu_read(s->cpu_slab->tid);
2930 c = raw_cpu_ptr(s->cpu_slab);
Mark Rutland859b7a02015-03-25 15:55:23 -07002931 } while (IS_ENABLED(CONFIG_PREEMPT) &&
2932 unlikely(tid != READ_ONCE(c->tid)));
Christoph Lameterc016b0b2010-08-20 12:37:16 -05002933
Joonsoo Kim9aabf812015-02-10 14:09:32 -08002934 /* Same with comment on barrier() in slab_alloc_node() */
2935 barrier();
Christoph Lameterc016b0b2010-08-20 12:37:16 -05002936
Christoph Lameter442b06b2011-05-17 16:29:31 -05002937 if (likely(page == c->page)) {
Jesper Dangaard Brouer81084652015-11-20 15:57:46 -08002938 set_freepointer(s, tail_obj, c->freelist);
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002939
Christoph Lameter933393f2011-12-22 11:58:51 -06002940 if (unlikely(!this_cpu_cmpxchg_double(
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002941 s->cpu_slab->freelist, s->cpu_slab->tid,
2942 c->freelist, tid,
Jesper Dangaard Brouer81084652015-11-20 15:57:46 -08002943 head, next_tid(tid)))) {
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06002944
2945 note_cmpxchg_failure("slab_free", s, tid);
2946 goto redo;
2947 }
Christoph Lameter84e554e62009-12-18 16:26:23 -06002948 stat(s, FREE_FASTPATH);
Christoph Lameter894b8782007-05-10 03:15:16 -07002949 } else
Jesper Dangaard Brouer81084652015-11-20 15:57:46 -08002950 __slab_free(s, page, head, tail_obj, cnt, addr);
Christoph Lameter894b8782007-05-10 03:15:16 -07002951
Christoph Lameter894b8782007-05-10 03:15:16 -07002952}
2953
Alexander Potapenko80a92012016-07-28 15:49:07 -07002954static __always_inline void slab_free(struct kmem_cache *s, struct page *page,
2955 void *head, void *tail, int cnt,
2956 unsigned long addr)
2957{
2958 slab_free_freelist_hook(s, head, tail);
2959 /*
2960 * slab_free_freelist_hook() could have put the items into quarantine.
2961 * If so, no need to free them.
2962 */
2963 if (s->flags & SLAB_KASAN && !(s->flags & SLAB_DESTROY_BY_RCU))
2964 return;
2965 do_slab_free(s, page, head, tail, cnt, addr);
2966}
2967
2968#ifdef CONFIG_KASAN
2969void ___cache_free(struct kmem_cache *cache, void *x, unsigned long addr)
2970{
2971 do_slab_free(cache, virt_to_head_page(x), x, NULL, 1, addr);
2972}
2973#endif
2974
Christoph Lameter81819f02007-05-06 14:49:36 -07002975void kmem_cache_free(struct kmem_cache *s, void *x)
2976{
Glauber Costab9ce5ef2012-12-18 14:22:46 -08002977 s = cache_from_obj(s, x);
2978 if (!s)
Christoph Lameter79576102012-09-04 23:06:14 +00002979 return;
Jesper Dangaard Brouer81084652015-11-20 15:57:46 -08002980 slab_free(s, virt_to_head_page(x), x, NULL, 1, _RET_IP_);
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02002981 trace_kmem_cache_free(_RET_IP_, x);
Christoph Lameter81819f02007-05-06 14:49:36 -07002982}
2983EXPORT_SYMBOL(kmem_cache_free);
2984
Jesper Dangaard Brouerd0ecd892015-11-20 15:57:49 -08002985struct detached_freelist {
2986 struct page *page;
2987 void *tail;
2988 void *freelist;
2989 int cnt;
Jesper Dangaard Brouer376bf122016-03-15 14:53:32 -07002990 struct kmem_cache *s;
Jesper Dangaard Brouerd0ecd892015-11-20 15:57:49 -08002991};
2992
2993/*
2994 * This function progressively scans the array with free objects (with
2995 * a limited look ahead) and extract objects belonging to the same
2996 * page. It builds a detached freelist directly within the given
2997 * page/objects. This can happen without any need for
2998 * synchronization, because the objects are owned by running process.
2999 * The freelist is build up as a single linked list in the objects.
3000 * The idea is, that this detached freelist can then be bulk
3001 * transferred to the real freelist(s), but only requiring a single
3002 * synchronization primitive. Look ahead in the array is limited due
3003 * to performance reasons.
3004 */
Jesper Dangaard Brouer376bf122016-03-15 14:53:32 -07003005static inline
3006int build_detached_freelist(struct kmem_cache *s, size_t size,
3007 void **p, struct detached_freelist *df)
Jesper Dangaard Brouerd0ecd892015-11-20 15:57:49 -08003008{
3009 size_t first_skipped_index = 0;
3010 int lookahead = 3;
3011 void *object;
Jesper Dangaard Brouerca257192016-03-15 14:54:00 -07003012 struct page *page;
Jesper Dangaard Brouerd0ecd892015-11-20 15:57:49 -08003013
3014 /* Always re-init detached_freelist */
3015 df->page = NULL;
3016
3017 do {
3018 object = p[--size];
Jesper Dangaard Brouerca257192016-03-15 14:54:00 -07003019 /* Do we need !ZERO_OR_NULL_PTR(object) here? (for kfree) */
Jesper Dangaard Brouerd0ecd892015-11-20 15:57:49 -08003020 } while (!object && size);
3021
3022 if (!object)
3023 return 0;
3024
Jesper Dangaard Brouerca257192016-03-15 14:54:00 -07003025 page = virt_to_head_page(object);
3026 if (!s) {
3027 /* Handle kalloc'ed objects */
3028 if (unlikely(!PageSlab(page))) {
3029 BUG_ON(!PageCompound(page));
3030 kfree_hook(object);
Vladimir Davydov49491482016-07-26 15:24:24 -07003031 __free_pages(page, compound_order(page));
Jesper Dangaard Brouerca257192016-03-15 14:54:00 -07003032 p[size] = NULL; /* mark object processed */
3033 return size;
3034 }
3035 /* Derive kmem_cache from object */
3036 df->s = page->slab_cache;
3037 } else {
3038 df->s = cache_from_obj(s, object); /* Support for memcg */
3039 }
Jesper Dangaard Brouer376bf122016-03-15 14:53:32 -07003040
Jesper Dangaard Brouerd0ecd892015-11-20 15:57:49 -08003041 /* Start new detached freelist */
Jesper Dangaard Brouerca257192016-03-15 14:54:00 -07003042 df->page = page;
Jesper Dangaard Brouer376bf122016-03-15 14:53:32 -07003043 set_freepointer(df->s, object, NULL);
Jesper Dangaard Brouerd0ecd892015-11-20 15:57:49 -08003044 df->tail = object;
3045 df->freelist = object;
3046 p[size] = NULL; /* mark object processed */
3047 df->cnt = 1;
3048
3049 while (size) {
3050 object = p[--size];
3051 if (!object)
3052 continue; /* Skip processed objects */
3053
3054 /* df->page is always set at this point */
3055 if (df->page == virt_to_head_page(object)) {
3056 /* Opportunity build freelist */
Jesper Dangaard Brouer376bf122016-03-15 14:53:32 -07003057 set_freepointer(df->s, object, df->freelist);
Jesper Dangaard Brouerd0ecd892015-11-20 15:57:49 -08003058 df->freelist = object;
3059 df->cnt++;
3060 p[size] = NULL; /* mark object processed */
3061
3062 continue;
3063 }
3064
3065 /* Limit look ahead search */
3066 if (!--lookahead)
3067 break;
3068
3069 if (!first_skipped_index)
3070 first_skipped_index = size + 1;
3071 }
3072
3073 return first_skipped_index;
3074}
3075
Jesper Dangaard Brouer994eb762015-09-04 15:45:37 -07003076/* Note that interrupts must be enabled when calling this function. */
Jesper Dangaard Brouer376bf122016-03-15 14:53:32 -07003077void kmem_cache_free_bulk(struct kmem_cache *s, size_t size, void **p)
Christoph Lameter484748f2015-09-04 15:45:34 -07003078{
Jesper Dangaard Brouerd0ecd892015-11-20 15:57:49 -08003079 if (WARN_ON(!size))
3080 return;
Jesper Dangaard Brouerfbd02632015-09-04 15:45:43 -07003081
Jesper Dangaard Brouerd0ecd892015-11-20 15:57:49 -08003082 do {
3083 struct detached_freelist df;
Jesper Dangaard Brouerfbd02632015-09-04 15:45:43 -07003084
Jesper Dangaard Brouerd0ecd892015-11-20 15:57:49 -08003085 size = build_detached_freelist(s, size, p, &df);
3086 if (unlikely(!df.page))
3087 continue;
Jesper Dangaard Brouerfbd02632015-09-04 15:45:43 -07003088
Jesper Dangaard Brouer376bf122016-03-15 14:53:32 -07003089 slab_free(df.s, df.page, df.freelist, df.tail, df.cnt,_RET_IP_);
Jesper Dangaard Brouerd0ecd892015-11-20 15:57:49 -08003090 } while (likely(size));
Christoph Lameter484748f2015-09-04 15:45:34 -07003091}
3092EXPORT_SYMBOL(kmem_cache_free_bulk);
3093
Jesper Dangaard Brouer994eb762015-09-04 15:45:37 -07003094/* Note that interrupts must be enabled when calling this function. */
Jesper Dangaard Brouer865762a2015-11-20 15:57:58 -08003095int kmem_cache_alloc_bulk(struct kmem_cache *s, gfp_t flags, size_t size,
3096 void **p)
Christoph Lameter484748f2015-09-04 15:45:34 -07003097{
Jesper Dangaard Brouer994eb762015-09-04 15:45:37 -07003098 struct kmem_cache_cpu *c;
3099 int i;
3100
Jesper Dangaard Brouer03ec0ed2015-11-20 15:57:52 -08003101 /* memcg and kmem_cache debug support */
3102 s = slab_pre_alloc_hook(s, flags);
3103 if (unlikely(!s))
3104 return false;
Jesper Dangaard Brouer994eb762015-09-04 15:45:37 -07003105 /*
3106 * Drain objects in the per cpu slab, while disabling local
3107 * IRQs, which protects against PREEMPT and interrupts
3108 * handlers invoking normal fastpath.
3109 */
3110 local_irq_disable();
3111 c = this_cpu_ptr(s->cpu_slab);
3112
3113 for (i = 0; i < size; i++) {
3114 void *object = c->freelist;
3115
Jesper Dangaard Brouerebe909e2015-09-04 15:45:40 -07003116 if (unlikely(!object)) {
Jesper Dangaard Brouerebe909e2015-09-04 15:45:40 -07003117 /*
3118 * Invoking slow path likely have side-effect
3119 * of re-populating per CPU c->freelist
3120 */
Christoph Lameter87098372015-11-20 15:57:38 -08003121 p[i] = ___slab_alloc(s, flags, NUMA_NO_NODE,
Jesper Dangaard Brouerebe909e2015-09-04 15:45:40 -07003122 _RET_IP_, c);
Christoph Lameter87098372015-11-20 15:57:38 -08003123 if (unlikely(!p[i]))
3124 goto error;
3125
Jesper Dangaard Brouerebe909e2015-09-04 15:45:40 -07003126 c = this_cpu_ptr(s->cpu_slab);
3127 continue; /* goto for-loop */
3128 }
Jesper Dangaard Brouer994eb762015-09-04 15:45:37 -07003129 c->freelist = get_freepointer(s, object);
3130 p[i] = object;
3131 }
3132 c->tid = next_tid(c->tid);
3133 local_irq_enable();
3134
3135 /* Clear memory outside IRQ disabled fastpath loop */
3136 if (unlikely(flags & __GFP_ZERO)) {
3137 int j;
3138
3139 for (j = 0; j < i; j++)
3140 memset(p[j], 0, s->object_size);
3141 }
3142
Jesper Dangaard Brouer03ec0ed2015-11-20 15:57:52 -08003143 /* memcg and kmem_cache debug support */
3144 slab_post_alloc_hook(s, flags, size, p);
Jesper Dangaard Brouer865762a2015-11-20 15:57:58 -08003145 return i;
Christoph Lameter87098372015-11-20 15:57:38 -08003146error:
Christoph Lameter87098372015-11-20 15:57:38 -08003147 local_irq_enable();
Jesper Dangaard Brouer03ec0ed2015-11-20 15:57:52 -08003148 slab_post_alloc_hook(s, flags, i, p);
3149 __kmem_cache_free_bulk(s, i, p);
Jesper Dangaard Brouer865762a2015-11-20 15:57:58 -08003150 return 0;
Christoph Lameter484748f2015-09-04 15:45:34 -07003151}
3152EXPORT_SYMBOL(kmem_cache_alloc_bulk);
3153
3154
Christoph Lameter81819f02007-05-06 14:49:36 -07003155/*
Christoph Lameter672bba32007-05-09 02:32:39 -07003156 * Object placement in a slab is made very easy because we always start at
3157 * offset 0. If we tune the size of the object to the alignment then we can
3158 * get the required alignment by putting one properly sized object after
3159 * another.
Christoph Lameter81819f02007-05-06 14:49:36 -07003160 *
3161 * Notice that the allocation order determines the sizes of the per cpu
3162 * caches. Each processor has always one slab available for allocations.
3163 * Increasing the allocation order reduces the number of times that slabs
Christoph Lameter672bba32007-05-09 02:32:39 -07003164 * must be moved on and off the partial lists and is therefore a factor in
Christoph Lameter81819f02007-05-06 14:49:36 -07003165 * locking overhead.
Christoph Lameter81819f02007-05-06 14:49:36 -07003166 */
3167
3168/*
3169 * Mininum / Maximum order of slab pages. This influences locking overhead
3170 * and slab fragmentation. A higher order reduces the number of partial slabs
3171 * and increases the number of allocations possible without having to
3172 * take the list_lock.
3173 */
3174static int slub_min_order;
Christoph Lameter114e9e82008-04-14 19:11:41 +03003175static int slub_max_order = PAGE_ALLOC_COSTLY_ORDER;
Christoph Lameter9b2cd502008-04-14 19:11:41 +03003176static int slub_min_objects;
Christoph Lameter81819f02007-05-06 14:49:36 -07003177
3178/*
Christoph Lameter81819f02007-05-06 14:49:36 -07003179 * Calculate the order of allocation given an slab object size.
3180 *
Christoph Lameter672bba32007-05-09 02:32:39 -07003181 * The order of allocation has significant impact on performance and other
3182 * system components. Generally order 0 allocations should be preferred since
3183 * order 0 does not cause fragmentation in the page allocator. Larger objects
3184 * be problematic to put into order 0 slabs because there may be too much
Christoph Lameterc124f5b2008-04-14 19:13:29 +03003185 * unused space left. We go to a higher order if more than 1/16th of the slab
Christoph Lameter672bba32007-05-09 02:32:39 -07003186 * would be wasted.
Christoph Lameter81819f02007-05-06 14:49:36 -07003187 *
Christoph Lameter672bba32007-05-09 02:32:39 -07003188 * In order to reach satisfactory performance we must ensure that a minimum
3189 * number of objects is in one slab. Otherwise we may generate too much
3190 * activity on the partial lists which requires taking the list_lock. This is
3191 * less a concern for large slabs though which are rarely used.
Christoph Lameter81819f02007-05-06 14:49:36 -07003192 *
Christoph Lameter672bba32007-05-09 02:32:39 -07003193 * slub_max_order specifies the order where we begin to stop considering the
3194 * number of objects in a slab as critical. If we reach slub_max_order then
3195 * we try to keep the page order as low as possible. So we accept more waste
3196 * of space in favor of a small page order.
3197 *
3198 * Higher order allocations also allow the placement of more objects in a
3199 * slab and thereby reduce object handling overhead. If the user has
3200 * requested a higher mininum order then we start with that one instead of
3201 * the smallest order which will fit the object.
Christoph Lameter81819f02007-05-06 14:49:36 -07003202 */
Christoph Lameter5e6d4442007-05-09 02:32:46 -07003203static inline int slab_order(int size, int min_objects,
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08003204 int max_order, int fract_leftover, int reserved)
Christoph Lameter81819f02007-05-06 14:49:36 -07003205{
3206 int order;
3207 int rem;
Christoph Lameter6300ea72007-07-17 04:03:20 -07003208 int min_order = slub_min_order;
Christoph Lameter81819f02007-05-06 14:49:36 -07003209
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08003210 if (order_objects(min_order, size, reserved) > MAX_OBJS_PER_PAGE)
Cyrill Gorcunov210b5c02008-10-22 23:00:38 +04003211 return get_order(size * MAX_OBJS_PER_PAGE) - 1;
Christoph Lameter39b26462008-04-14 19:11:30 +03003212
Wei Yang9f835702015-11-05 18:45:51 -08003213 for (order = max(min_order, get_order(min_objects * size + reserved));
Christoph Lameter5e6d4442007-05-09 02:32:46 -07003214 order <= max_order; order++) {
3215
Christoph Lameter81819f02007-05-06 14:49:36 -07003216 unsigned long slab_size = PAGE_SIZE << order;
3217
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08003218 rem = (slab_size - reserved) % size;
Christoph Lameter81819f02007-05-06 14:49:36 -07003219
Christoph Lameter5e6d4442007-05-09 02:32:46 -07003220 if (rem <= slab_size / fract_leftover)
Christoph Lameter81819f02007-05-06 14:49:36 -07003221 break;
Christoph Lameter81819f02007-05-06 14:49:36 -07003222 }
Christoph Lameter672bba32007-05-09 02:32:39 -07003223
Christoph Lameter81819f02007-05-06 14:49:36 -07003224 return order;
3225}
3226
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08003227static inline int calculate_order(int size, int reserved)
Christoph Lameter5e6d4442007-05-09 02:32:46 -07003228{
3229 int order;
3230 int min_objects;
3231 int fraction;
Zhang Yanmine8120ff2009-02-12 18:00:17 +02003232 int max_objects;
Christoph Lameter5e6d4442007-05-09 02:32:46 -07003233
3234 /*
3235 * Attempt to find best configuration for a slab. This
3236 * works by first attempting to generate a layout with
3237 * the best configuration and backing off gradually.
3238 *
Wei Yang422ff4d2015-11-05 18:45:46 -08003239 * First we increase the acceptable waste in a slab. Then
Christoph Lameter5e6d4442007-05-09 02:32:46 -07003240 * we reduce the minimum objects required in a slab.
3241 */
3242 min_objects = slub_min_objects;
Christoph Lameter9b2cd502008-04-14 19:11:41 +03003243 if (!min_objects)
3244 min_objects = 4 * (fls(nr_cpu_ids) + 1);
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08003245 max_objects = order_objects(slub_max_order, size, reserved);
Zhang Yanmine8120ff2009-02-12 18:00:17 +02003246 min_objects = min(min_objects, max_objects);
3247
Christoph Lameter5e6d4442007-05-09 02:32:46 -07003248 while (min_objects > 1) {
Christoph Lameterc124f5b2008-04-14 19:13:29 +03003249 fraction = 16;
Christoph Lameter5e6d4442007-05-09 02:32:46 -07003250 while (fraction >= 4) {
3251 order = slab_order(size, min_objects,
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08003252 slub_max_order, fraction, reserved);
Christoph Lameter5e6d4442007-05-09 02:32:46 -07003253 if (order <= slub_max_order)
3254 return order;
3255 fraction /= 2;
3256 }
Amerigo Wang5086c389c2009-08-19 21:44:13 +03003257 min_objects--;
Christoph Lameter5e6d4442007-05-09 02:32:46 -07003258 }
3259
3260 /*
3261 * We were unable to place multiple objects in a slab. Now
3262 * lets see if we can place a single object there.
3263 */
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08003264 order = slab_order(size, 1, slub_max_order, 1, reserved);
Christoph Lameter5e6d4442007-05-09 02:32:46 -07003265 if (order <= slub_max_order)
3266 return order;
3267
3268 /*
3269 * Doh this slab cannot be placed using slub_max_order.
3270 */
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08003271 order = slab_order(size, 1, MAX_ORDER, 1, reserved);
David Rientjes818cf592009-04-23 09:58:22 +03003272 if (order < MAX_ORDER)
Christoph Lameter5e6d4442007-05-09 02:32:46 -07003273 return order;
3274 return -ENOSYS;
3275}
3276
Pekka Enberg5595cff2008-08-05 09:28:47 +03003277static void
Joonsoo Kim40534972012-05-11 00:50:47 +09003278init_kmem_cache_node(struct kmem_cache_node *n)
Christoph Lameter81819f02007-05-06 14:49:36 -07003279{
3280 n->nr_partial = 0;
Christoph Lameter81819f02007-05-06 14:49:36 -07003281 spin_lock_init(&n->list_lock);
3282 INIT_LIST_HEAD(&n->partial);
Christoph Lameter8ab13722007-07-17 04:03:32 -07003283#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter0f389ec2008-04-14 18:53:02 +03003284 atomic_long_set(&n->nr_slabs, 0);
Salman Qazi02b71b72008-09-11 12:25:41 -07003285 atomic_long_set(&n->total_objects, 0);
Christoph Lameter643b1132007-05-06 14:49:42 -07003286 INIT_LIST_HEAD(&n->full);
Christoph Lameter8ab13722007-07-17 04:03:32 -07003287#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07003288}
3289
Christoph Lameter55136592010-08-20 12:37:13 -05003290static inline int alloc_kmem_cache_cpus(struct kmem_cache *s)
Christoph Lameter4c93c3552007-10-16 01:26:08 -07003291{
Christoph Lameter6c182dc2010-08-20 12:37:14 -05003292 BUILD_BUG_ON(PERCPU_DYNAMIC_EARLY_SIZE <
Christoph Lameter95a05b42013-01-10 19:14:19 +00003293 KMALLOC_SHIFT_HIGH * sizeof(struct kmem_cache_cpu));
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06003294
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06003295 /*
Chris Metcalfd4d84fe2011-06-02 10:19:41 -04003296 * Must align to double word boundary for the double cmpxchg
3297 * instructions to work; see __pcpu_double_call_return_bool().
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06003298 */
Chris Metcalfd4d84fe2011-06-02 10:19:41 -04003299 s->cpu_slab = __alloc_percpu(sizeof(struct kmem_cache_cpu),
3300 2 * sizeof(void *));
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06003301
Christoph Lameter8a5ec0b2011-02-25 11:38:54 -06003302 if (!s->cpu_slab)
3303 return 0;
3304
3305 init_kmem_cache_cpus(s);
3306
3307 return 1;
Christoph Lameter4c93c3552007-10-16 01:26:08 -07003308}
Christoph Lameter4c93c3552007-10-16 01:26:08 -07003309
Christoph Lameter51df1142010-08-20 12:37:15 -05003310static struct kmem_cache *kmem_cache_node;
3311
Christoph Lameter81819f02007-05-06 14:49:36 -07003312/*
3313 * No kmalloc_node yet so do it by hand. We know that this is the first
3314 * slab on the node for this slabcache. There are no concurrent accesses
3315 * possible.
3316 *
Zhi Yong Wu721ae222013-11-08 20:47:37 +08003317 * Note that this function only works on the kmem_cache_node
3318 * when allocating for the kmem_cache_node. This is used for bootstrapping
Christoph Lameter4c93c3552007-10-16 01:26:08 -07003319 * memory on a fresh node that has no slab structures yet.
Christoph Lameter81819f02007-05-06 14:49:36 -07003320 */
Christoph Lameter55136592010-08-20 12:37:13 -05003321static void early_kmem_cache_node_alloc(int node)
Christoph Lameter81819f02007-05-06 14:49:36 -07003322{
3323 struct page *page;
3324 struct kmem_cache_node *n;
3325
Christoph Lameter51df1142010-08-20 12:37:15 -05003326 BUG_ON(kmem_cache_node->size < sizeof(struct kmem_cache_node));
Christoph Lameter81819f02007-05-06 14:49:36 -07003327
Christoph Lameter51df1142010-08-20 12:37:15 -05003328 page = new_slab(kmem_cache_node, GFP_NOWAIT, node);
Christoph Lameter81819f02007-05-06 14:49:36 -07003329
3330 BUG_ON(!page);
Christoph Lametera2f92ee2007-08-22 14:01:57 -07003331 if (page_to_nid(page) != node) {
Fabian Frederickf9f58282014-06-04 16:06:34 -07003332 pr_err("SLUB: Unable to allocate memory from node %d\n", node);
3333 pr_err("SLUB: Allocating a useless per node structure in order to be able to continue\n");
Christoph Lametera2f92ee2007-08-22 14:01:57 -07003334 }
3335
Christoph Lameter81819f02007-05-06 14:49:36 -07003336 n = page->freelist;
3337 BUG_ON(!n);
Christoph Lameter51df1142010-08-20 12:37:15 -05003338 page->freelist = get_freepointer(kmem_cache_node, n);
Christoph Lametere6e82ea2011-08-09 16:12:24 -05003339 page->inuse = 1;
Christoph Lameter8cb0a502011-06-01 12:25:46 -05003340 page->frozen = 0;
Christoph Lameter51df1142010-08-20 12:37:15 -05003341 kmem_cache_node->node[node] = n;
Christoph Lameter8ab13722007-07-17 04:03:32 -07003342#ifdef CONFIG_SLUB_DEBUG
Christoph Lameterf7cb1932010-09-29 07:15:01 -05003343 init_object(kmem_cache_node, n, SLUB_RED_ACTIVE);
Christoph Lameter51df1142010-08-20 12:37:15 -05003344 init_tracking(kmem_cache_node, n);
Christoph Lameter8ab13722007-07-17 04:03:32 -07003345#endif
Alexander Potapenko505f5dc2016-03-25 14:22:02 -07003346 kasan_kmalloc(kmem_cache_node, n, sizeof(struct kmem_cache_node),
3347 GFP_KERNEL);
Joonsoo Kim40534972012-05-11 00:50:47 +09003348 init_kmem_cache_node(n);
Christoph Lameter51df1142010-08-20 12:37:15 -05003349 inc_slabs_node(kmem_cache_node, node, page->objects);
Christoph Lameter6446faa2008-02-15 23:45:26 -08003350
Dave Hansen67b6c902014-01-24 07:20:23 -08003351 /*
Steven Rostedt1e4dd942014-02-10 14:25:46 -08003352 * No locks need to be taken here as it has just been
3353 * initialized and there is no concurrent access.
Dave Hansen67b6c902014-01-24 07:20:23 -08003354 */
Steven Rostedt1e4dd942014-02-10 14:25:46 -08003355 __add_partial(n, page, DEACTIVATE_TO_HEAD);
Christoph Lameter81819f02007-05-06 14:49:36 -07003356}
3357
3358static void free_kmem_cache_nodes(struct kmem_cache *s)
3359{
3360 int node;
Christoph Lameterfa45dc22014-08-06 16:04:09 -07003361 struct kmem_cache_node *n;
Christoph Lameter81819f02007-05-06 14:49:36 -07003362
Christoph Lameterfa45dc22014-08-06 16:04:09 -07003363 for_each_kmem_cache_node(s, node, n) {
3364 kmem_cache_free(kmem_cache_node, n);
Christoph Lameter81819f02007-05-06 14:49:36 -07003365 s->node[node] = NULL;
3366 }
3367}
3368
Dmitry Safonov52b4b952016-02-17 13:11:37 -08003369void __kmem_cache_release(struct kmem_cache *s)
3370{
Thomas Garnier210e7a42016-07-26 15:21:59 -07003371 cache_random_seq_destroy(s);
Dmitry Safonov52b4b952016-02-17 13:11:37 -08003372 free_percpu(s->cpu_slab);
3373 free_kmem_cache_nodes(s);
3374}
3375
Christoph Lameter55136592010-08-20 12:37:13 -05003376static int init_kmem_cache_nodes(struct kmem_cache *s)
Christoph Lameter81819f02007-05-06 14:49:36 -07003377{
3378 int node;
Christoph Lameter81819f02007-05-06 14:49:36 -07003379
Christoph Lameterf64dc582007-10-16 01:25:33 -07003380 for_each_node_state(node, N_NORMAL_MEMORY) {
Christoph Lameter81819f02007-05-06 14:49:36 -07003381 struct kmem_cache_node *n;
3382
Alexander Duyck73367bd2010-05-21 14:41:35 -07003383 if (slab_state == DOWN) {
Christoph Lameter55136592010-08-20 12:37:13 -05003384 early_kmem_cache_node_alloc(node);
Alexander Duyck73367bd2010-05-21 14:41:35 -07003385 continue;
Christoph Lameter81819f02007-05-06 14:49:36 -07003386 }
Christoph Lameter51df1142010-08-20 12:37:15 -05003387 n = kmem_cache_alloc_node(kmem_cache_node,
Christoph Lameter55136592010-08-20 12:37:13 -05003388 GFP_KERNEL, node);
Alexander Duyck73367bd2010-05-21 14:41:35 -07003389
3390 if (!n) {
3391 free_kmem_cache_nodes(s);
3392 return 0;
3393 }
3394
Christoph Lameter81819f02007-05-06 14:49:36 -07003395 s->node[node] = n;
Joonsoo Kim40534972012-05-11 00:50:47 +09003396 init_kmem_cache_node(n);
Christoph Lameter81819f02007-05-06 14:49:36 -07003397 }
3398 return 1;
3399}
Christoph Lameter81819f02007-05-06 14:49:36 -07003400
David Rientjesc0bdb232009-02-25 09:16:35 +02003401static void set_min_partial(struct kmem_cache *s, unsigned long min)
David Rientjes3b89d7d2009-02-22 17:40:07 -08003402{
3403 if (min < MIN_PARTIAL)
3404 min = MIN_PARTIAL;
3405 else if (min > MAX_PARTIAL)
3406 min = MAX_PARTIAL;
3407 s->min_partial = min;
3408}
3409
Christoph Lameter81819f02007-05-06 14:49:36 -07003410/*
3411 * calculate_sizes() determines the order and the distribution of data within
3412 * a slab object.
3413 */
Christoph Lameter06b285d2008-04-14 19:11:41 +03003414static int calculate_sizes(struct kmem_cache *s, int forced_order)
Christoph Lameter81819f02007-05-06 14:49:36 -07003415{
3416 unsigned long flags = s->flags;
Alexander Potapenko80a92012016-07-28 15:49:07 -07003417 size_t size = s->object_size;
Christoph Lameter834f3d12008-04-14 19:11:31 +03003418 int order;
Christoph Lameter81819f02007-05-06 14:49:36 -07003419
3420 /*
Christoph Lameterd8b42bf2008-02-15 23:45:25 -08003421 * Round up object size to the next word boundary. We can only
3422 * place the free pointer at word boundaries and this determines
3423 * the possible location of the free pointer.
3424 */
3425 size = ALIGN(size, sizeof(void *));
3426
3427#ifdef CONFIG_SLUB_DEBUG
3428 /*
Christoph Lameter81819f02007-05-06 14:49:36 -07003429 * Determine if we can poison the object itself. If the user of
3430 * the slab may touch the object after free or before allocation
3431 * then we should never poison the object itself.
3432 */
3433 if ((flags & SLAB_POISON) && !(flags & SLAB_DESTROY_BY_RCU) &&
Christoph Lameterc59def92007-05-16 22:10:50 -07003434 !s->ctor)
Christoph Lameter81819f02007-05-06 14:49:36 -07003435 s->flags |= __OBJECT_POISON;
3436 else
3437 s->flags &= ~__OBJECT_POISON;
3438
Christoph Lameter81819f02007-05-06 14:49:36 -07003439
3440 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07003441 * If we are Redzoning then check if there is some space between the
Christoph Lameter81819f02007-05-06 14:49:36 -07003442 * end of the object and the free pointer. If not then add an
Christoph Lameter672bba32007-05-09 02:32:39 -07003443 * additional word to have some bytes to store Redzone information.
Christoph Lameter81819f02007-05-06 14:49:36 -07003444 */
Christoph Lameter3b0efdf2012-06-13 10:24:57 -05003445 if ((flags & SLAB_RED_ZONE) && size == s->object_size)
Christoph Lameter81819f02007-05-06 14:49:36 -07003446 size += sizeof(void *);
Christoph Lameter41ecc552007-05-09 02:32:44 -07003447#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07003448
3449 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07003450 * With that we have determined the number of bytes in actual use
3451 * by the object. This is the potential offset to the free pointer.
Christoph Lameter81819f02007-05-06 14:49:36 -07003452 */
3453 s->inuse = size;
3454
3455 if (((flags & (SLAB_DESTROY_BY_RCU | SLAB_POISON)) ||
Christoph Lameterc59def92007-05-16 22:10:50 -07003456 s->ctor)) {
Christoph Lameter81819f02007-05-06 14:49:36 -07003457 /*
3458 * Relocate free pointer after the object if it is not
3459 * permitted to overwrite the first word of the object on
3460 * kmem_cache_free.
3461 *
3462 * This is the case if we do RCU, have a constructor or
3463 * destructor or are poisoning the objects.
3464 */
3465 s->offset = size;
3466 size += sizeof(void *);
3467 }
3468
Christoph Lameterc12b3c62007-05-23 13:57:31 -07003469#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter81819f02007-05-06 14:49:36 -07003470 if (flags & SLAB_STORE_USER)
3471 /*
3472 * Need to store information about allocs and frees after
3473 * the object.
3474 */
3475 size += 2 * sizeof(struct track);
Alexander Potapenko80a92012016-07-28 15:49:07 -07003476#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07003477
Alexander Potapenko80a92012016-07-28 15:49:07 -07003478 kasan_cache_create(s, &size, &s->flags);
3479#ifdef CONFIG_SLUB_DEBUG
Joonsoo Kimd86bd1b2016-03-15 14:55:12 -07003480 if (flags & SLAB_RED_ZONE) {
Christoph Lameter81819f02007-05-06 14:49:36 -07003481 /*
3482 * Add some empty padding so that we can catch
3483 * overwrites from earlier objects rather than let
3484 * tracking information or the free pointer be
Frederik Schwarzer0211a9c2008-12-29 22:14:56 +01003485 * corrupted if a user writes before the start
Christoph Lameter81819f02007-05-06 14:49:36 -07003486 * of the object.
3487 */
3488 size += sizeof(void *);
Joonsoo Kimd86bd1b2016-03-15 14:55:12 -07003489
3490 s->red_left_pad = sizeof(void *);
3491 s->red_left_pad = ALIGN(s->red_left_pad, s->align);
3492 size += s->red_left_pad;
3493 }
Christoph Lameter41ecc552007-05-09 02:32:44 -07003494#endif
Christoph Lameter672bba32007-05-09 02:32:39 -07003495
Christoph Lameter81819f02007-05-06 14:49:36 -07003496 /*
Christoph Lameter81819f02007-05-06 14:49:36 -07003497 * SLUB stores one object immediately after another beginning from
3498 * offset 0. In order to align the objects we have to simply size
3499 * each object to conform to the alignment.
3500 */
Christoph Lameter45906852012-11-28 16:23:16 +00003501 size = ALIGN(size, s->align);
Christoph Lameter81819f02007-05-06 14:49:36 -07003502 s->size = size;
Christoph Lameter06b285d2008-04-14 19:11:41 +03003503 if (forced_order >= 0)
3504 order = forced_order;
3505 else
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08003506 order = calculate_order(size, s->reserved);
Christoph Lameter81819f02007-05-06 14:49:36 -07003507
Christoph Lameter834f3d12008-04-14 19:11:31 +03003508 if (order < 0)
Christoph Lameter81819f02007-05-06 14:49:36 -07003509 return 0;
3510
Christoph Lameterb7a49f02008-02-14 14:21:32 -08003511 s->allocflags = 0;
Christoph Lameter834f3d12008-04-14 19:11:31 +03003512 if (order)
Christoph Lameterb7a49f02008-02-14 14:21:32 -08003513 s->allocflags |= __GFP_COMP;
3514
3515 if (s->flags & SLAB_CACHE_DMA)
Christoph Lameter2c59dd62013-01-10 19:14:19 +00003516 s->allocflags |= GFP_DMA;
Christoph Lameterb7a49f02008-02-14 14:21:32 -08003517
3518 if (s->flags & SLAB_RECLAIM_ACCOUNT)
3519 s->allocflags |= __GFP_RECLAIMABLE;
3520
Christoph Lameter81819f02007-05-06 14:49:36 -07003521 /*
3522 * Determine the number of objects per slab
3523 */
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08003524 s->oo = oo_make(order, size, s->reserved);
3525 s->min = oo_make(get_order(size), size, s->reserved);
Christoph Lameter205ab992008-04-14 19:11:40 +03003526 if (oo_objects(s->oo) > oo_objects(s->max))
3527 s->max = s->oo;
Christoph Lameter81819f02007-05-06 14:49:36 -07003528
Christoph Lameter834f3d12008-04-14 19:11:31 +03003529 return !!oo_objects(s->oo);
Christoph Lameter81819f02007-05-06 14:49:36 -07003530}
3531
Christoph Lameter8a13a4c2012-09-04 23:18:33 +00003532static int kmem_cache_open(struct kmem_cache *s, unsigned long flags)
Christoph Lameter81819f02007-05-06 14:49:36 -07003533{
Christoph Lameter8a13a4c2012-09-04 23:18:33 +00003534 s->flags = kmem_cache_flags(s->size, flags, s->name, s->ctor);
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08003535 s->reserved = 0;
Christoph Lameter81819f02007-05-06 14:49:36 -07003536
Lai Jiangshanda9a6382011-03-10 15:22:00 +08003537 if (need_reserve_slab_rcu && (s->flags & SLAB_DESTROY_BY_RCU))
3538 s->reserved = sizeof(struct rcu_head);
Christoph Lameter81819f02007-05-06 14:49:36 -07003539
Christoph Lameter06b285d2008-04-14 19:11:41 +03003540 if (!calculate_sizes(s, -1))
Christoph Lameter81819f02007-05-06 14:49:36 -07003541 goto error;
David Rientjes3de47212009-07-27 18:30:35 -07003542 if (disable_higher_order_debug) {
3543 /*
3544 * Disable debugging flags that store metadata if the min slab
3545 * order increased.
3546 */
Christoph Lameter3b0efdf2012-06-13 10:24:57 -05003547 if (get_order(s->size) > get_order(s->object_size)) {
David Rientjes3de47212009-07-27 18:30:35 -07003548 s->flags &= ~DEBUG_METADATA_FLAGS;
3549 s->offset = 0;
3550 if (!calculate_sizes(s, -1))
3551 goto error;
3552 }
3553 }
Christoph Lameter81819f02007-05-06 14:49:36 -07003554
Heiko Carstens25654092012-01-12 17:17:33 -08003555#if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \
3556 defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
Laura Abbott149daaf2016-03-15 14:55:09 -07003557 if (system_has_cmpxchg_double() && (s->flags & SLAB_NO_CMPXCHG) == 0)
Christoph Lameterb789ef52011-06-01 12:25:49 -05003558 /* Enable fast mode */
3559 s->flags |= __CMPXCHG_DOUBLE;
3560#endif
3561
David Rientjes3b89d7d2009-02-22 17:40:07 -08003562 /*
3563 * The larger the object size is, the more pages we want on the partial
3564 * list to avoid pounding the page allocator excessively.
3565 */
Christoph Lameter49e22582011-08-09 16:12:27 -05003566 set_min_partial(s, ilog2(s->size) / 2);
3567
3568 /*
3569 * cpu_partial determined the maximum number of objects kept in the
3570 * per cpu partial lists of a processor.
3571 *
3572 * Per cpu partial lists mainly contain slabs that just have one
3573 * object freed. If they are used for allocation then they can be
3574 * filled up again with minimal effort. The slab will never hit the
3575 * per node partial lists and therefore no locking will be required.
3576 *
3577 * This setting also determines
3578 *
3579 * A) The number of objects from per cpu partial slabs dumped to the
3580 * per node list when we reach the limit.
Alex Shi9f264902011-09-01 11:32:18 +08003581 * B) The number of objects in cpu partial slabs to extract from the
Chen Gangd0e0ac92013-07-15 09:05:29 +08003582 * per node list when we run out of per cpu objects. We only fetch
3583 * 50% to keep some capacity around for frees.
Christoph Lameter49e22582011-08-09 16:12:27 -05003584 */
Joonsoo Kim345c9052013-06-19 14:05:52 +09003585 if (!kmem_cache_has_cpu_partial(s))
Christoph Lameter8f1e33d2011-11-23 09:24:27 -06003586 s->cpu_partial = 0;
3587 else if (s->size >= PAGE_SIZE)
Christoph Lameter49e22582011-08-09 16:12:27 -05003588 s->cpu_partial = 2;
3589 else if (s->size >= 1024)
3590 s->cpu_partial = 6;
3591 else if (s->size >= 256)
3592 s->cpu_partial = 13;
3593 else
3594 s->cpu_partial = 30;
3595
Christoph Lameter81819f02007-05-06 14:49:36 -07003596#ifdef CONFIG_NUMA
Christoph Lametere2cb96b2008-08-19 08:51:22 -05003597 s->remote_node_defrag_ratio = 1000;
Christoph Lameter81819f02007-05-06 14:49:36 -07003598#endif
Thomas Garnier210e7a42016-07-26 15:21:59 -07003599
3600 /* Initialize the pre-computed randomized freelist if slab is up */
3601 if (slab_state >= UP) {
3602 if (init_cache_random_seq(s))
3603 goto error;
3604 }
3605
Christoph Lameter55136592010-08-20 12:37:13 -05003606 if (!init_kmem_cache_nodes(s))
Christoph Lameterdfb4f092007-10-16 01:26:05 -07003607 goto error;
Christoph Lameter81819f02007-05-06 14:49:36 -07003608
Christoph Lameter55136592010-08-20 12:37:13 -05003609 if (alloc_kmem_cache_cpus(s))
Christoph Lameter278b1bb2012-09-05 00:20:34 +00003610 return 0;
Christoph Lameterff120592009-12-18 16:26:22 -06003611
Christoph Lameter4c93c3552007-10-16 01:26:08 -07003612 free_kmem_cache_nodes(s);
Christoph Lameter81819f02007-05-06 14:49:36 -07003613error:
3614 if (flags & SLAB_PANIC)
Joe Perches756a025f02016-03-17 14:19:47 -07003615 panic("Cannot create slab %s size=%lu realsize=%u order=%u offset=%u flags=%lx\n",
3616 s->name, (unsigned long)s->size, s->size,
3617 oo_order(s->oo), s->offset, flags);
Christoph Lameter278b1bb2012-09-05 00:20:34 +00003618 return -EINVAL;
Christoph Lameter81819f02007-05-06 14:49:36 -07003619}
Christoph Lameter81819f02007-05-06 14:49:36 -07003620
Christoph Lameter33b12c32008-04-25 12:22:43 -07003621static void list_slab_objects(struct kmem_cache *s, struct page *page,
3622 const char *text)
Christoph Lameter81819f02007-05-06 14:49:36 -07003623{
Christoph Lameter33b12c32008-04-25 12:22:43 -07003624#ifdef CONFIG_SLUB_DEBUG
3625 void *addr = page_address(page);
3626 void *p;
Namhyung Kima5dd5c12010-09-29 21:02:13 +09003627 unsigned long *map = kzalloc(BITS_TO_LONGS(page->objects) *
3628 sizeof(long), GFP_ATOMIC);
Eric Dumazetbbd7d572010-03-24 22:25:47 +01003629 if (!map)
3630 return;
Christoph Lameter945cf2b2012-09-04 23:18:33 +00003631 slab_err(s, page, text, s->name);
Christoph Lameter33b12c32008-04-25 12:22:43 -07003632 slab_lock(page);
Christoph Lameter33b12c32008-04-25 12:22:43 -07003633
Christoph Lameter5f80b132011-04-15 14:48:13 -05003634 get_map(s, page, map);
Christoph Lameter33b12c32008-04-25 12:22:43 -07003635 for_each_object(p, s, addr, page->objects) {
3636
3637 if (!test_bit(slab_index(p, s, addr), map)) {
Fabian Frederickf9f58282014-06-04 16:06:34 -07003638 pr_err("INFO: Object 0x%p @offset=%tu\n", p, p - addr);
Christoph Lameter33b12c32008-04-25 12:22:43 -07003639 print_tracking(s, p);
3640 }
3641 }
3642 slab_unlock(page);
Eric Dumazetbbd7d572010-03-24 22:25:47 +01003643 kfree(map);
Christoph Lameter33b12c32008-04-25 12:22:43 -07003644#endif
3645}
3646
Christoph Lameter81819f02007-05-06 14:49:36 -07003647/*
Christoph Lameter599870b2008-04-23 12:36:52 -07003648 * Attempt to free all partial slabs on a node.
Dmitry Safonov52b4b952016-02-17 13:11:37 -08003649 * This is called from __kmem_cache_shutdown(). We must take list_lock
3650 * because sysfs file might still access partial list after the shutdowning.
Christoph Lameter81819f02007-05-06 14:49:36 -07003651 */
Christoph Lameter599870b2008-04-23 12:36:52 -07003652static void free_partial(struct kmem_cache *s, struct kmem_cache_node *n)
Christoph Lameter81819f02007-05-06 14:49:36 -07003653{
Chris Wilson60398922016-08-10 16:27:58 -07003654 LIST_HEAD(discard);
Christoph Lameter81819f02007-05-06 14:49:36 -07003655 struct page *page, *h;
3656
Dmitry Safonov52b4b952016-02-17 13:11:37 -08003657 BUG_ON(irqs_disabled());
3658 spin_lock_irq(&n->list_lock);
Christoph Lameter33b12c32008-04-25 12:22:43 -07003659 list_for_each_entry_safe(page, h, &n->partial, lru) {
Christoph Lameter81819f02007-05-06 14:49:36 -07003660 if (!page->inuse) {
Dmitry Safonov52b4b952016-02-17 13:11:37 -08003661 remove_partial(n, page);
Chris Wilson60398922016-08-10 16:27:58 -07003662 list_add(&page->lru, &discard);
Christoph Lameter33b12c32008-04-25 12:22:43 -07003663 } else {
3664 list_slab_objects(s, page,
Dmitry Safonov52b4b952016-02-17 13:11:37 -08003665 "Objects remaining in %s on __kmem_cache_shutdown()");
Christoph Lameter599870b2008-04-23 12:36:52 -07003666 }
Christoph Lameter33b12c32008-04-25 12:22:43 -07003667 }
Dmitry Safonov52b4b952016-02-17 13:11:37 -08003668 spin_unlock_irq(&n->list_lock);
Chris Wilson60398922016-08-10 16:27:58 -07003669
3670 list_for_each_entry_safe(page, h, &discard, lru)
3671 discard_slab(s, page);
Christoph Lameter81819f02007-05-06 14:49:36 -07003672}
3673
3674/*
Christoph Lameter672bba32007-05-09 02:32:39 -07003675 * Release all resources used by a slab cache.
Christoph Lameter81819f02007-05-06 14:49:36 -07003676 */
Dmitry Safonov52b4b952016-02-17 13:11:37 -08003677int __kmem_cache_shutdown(struct kmem_cache *s)
Christoph Lameter81819f02007-05-06 14:49:36 -07003678{
3679 int node;
Christoph Lameterfa45dc22014-08-06 16:04:09 -07003680 struct kmem_cache_node *n;
Christoph Lameter81819f02007-05-06 14:49:36 -07003681
3682 flush_all(s);
Christoph Lameter81819f02007-05-06 14:49:36 -07003683 /* Attempt to free all objects */
Christoph Lameterfa45dc22014-08-06 16:04:09 -07003684 for_each_kmem_cache_node(s, node, n) {
Christoph Lameter599870b2008-04-23 12:36:52 -07003685 free_partial(s, n);
3686 if (n->nr_partial || slabs_node(s, node))
Christoph Lameter81819f02007-05-06 14:49:36 -07003687 return 1;
3688 }
Christoph Lameter81819f02007-05-06 14:49:36 -07003689 return 0;
3690}
3691
Christoph Lameter81819f02007-05-06 14:49:36 -07003692/********************************************************************
3693 * Kmalloc subsystem
3694 *******************************************************************/
3695
Christoph Lameter81819f02007-05-06 14:49:36 -07003696static int __init setup_slub_min_order(char *str)
3697{
Pekka Enberg06428782008-01-07 23:20:27 -08003698 get_option(&str, &slub_min_order);
Christoph Lameter81819f02007-05-06 14:49:36 -07003699
3700 return 1;
3701}
3702
3703__setup("slub_min_order=", setup_slub_min_order);
3704
3705static int __init setup_slub_max_order(char *str)
3706{
Pekka Enberg06428782008-01-07 23:20:27 -08003707 get_option(&str, &slub_max_order);
David Rientjes818cf592009-04-23 09:58:22 +03003708 slub_max_order = min(slub_max_order, MAX_ORDER - 1);
Christoph Lameter81819f02007-05-06 14:49:36 -07003709
3710 return 1;
3711}
3712
3713__setup("slub_max_order=", setup_slub_max_order);
3714
3715static int __init setup_slub_min_objects(char *str)
3716{
Pekka Enberg06428782008-01-07 23:20:27 -08003717 get_option(&str, &slub_min_objects);
Christoph Lameter81819f02007-05-06 14:49:36 -07003718
3719 return 1;
3720}
3721
3722__setup("slub_min_objects=", setup_slub_min_objects);
3723
Christoph Lameter81819f02007-05-06 14:49:36 -07003724void *__kmalloc(size_t size, gfp_t flags)
3725{
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003726 struct kmem_cache *s;
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03003727 void *ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07003728
Christoph Lameter95a05b42013-01-10 19:14:19 +00003729 if (unlikely(size > KMALLOC_MAX_CACHE_SIZE))
Pekka Enbergeada35e2008-02-11 22:47:46 +02003730 return kmalloc_large(size, flags);
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003731
Christoph Lameter2c59dd62013-01-10 19:14:19 +00003732 s = kmalloc_slab(size, flags);
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003733
3734 if (unlikely(ZERO_OR_NULL_PTR(s)))
Christoph Lameter6cb8f912007-07-17 04:03:22 -07003735 return s;
3736
Ezequiel Garcia2b847c32012-09-08 17:47:58 -03003737 ret = slab_alloc(s, flags, _RET_IP_);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03003738
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02003739 trace_kmalloc(_RET_IP_, ret, size, s->size, flags);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03003740
Alexander Potapenko505f5dc2016-03-25 14:22:02 -07003741 kasan_kmalloc(s, ret, size, flags);
Andrey Ryabinin0316bec2015-02-13 14:39:42 -08003742
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03003743 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07003744}
3745EXPORT_SYMBOL(__kmalloc);
3746
Namhyung Kim5d1f57e2010-09-29 21:02:15 +09003747#ifdef CONFIG_NUMA
Christoph Lameterf619cfe2008-03-01 13:56:40 -08003748static void *kmalloc_large_node(size_t size, gfp_t flags, int node)
3749{
Vegard Nossumb1eeab62008-11-25 16:55:53 +01003750 struct page *page;
Catalin Marinase4f7c0b42009-07-07 10:32:59 +01003751 void *ptr = NULL;
Christoph Lameterf619cfe2008-03-01 13:56:40 -08003752
Vladimir Davydov52383432014-06-04 16:06:39 -07003753 flags |= __GFP_COMP | __GFP_NOTRACK;
Vladimir Davydov49491482016-07-26 15:24:24 -07003754 page = alloc_pages_node(node, flags, get_order(size));
Christoph Lameterf619cfe2008-03-01 13:56:40 -08003755 if (page)
Catalin Marinase4f7c0b42009-07-07 10:32:59 +01003756 ptr = page_address(page);
3757
Roman Bobnievd56791b2013-10-08 15:58:57 -07003758 kmalloc_large_node_hook(ptr, size, flags);
Catalin Marinase4f7c0b42009-07-07 10:32:59 +01003759 return ptr;
Christoph Lameterf619cfe2008-03-01 13:56:40 -08003760}
3761
Christoph Lameter81819f02007-05-06 14:49:36 -07003762void *__kmalloc_node(size_t size, gfp_t flags, int node)
3763{
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003764 struct kmem_cache *s;
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03003765 void *ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07003766
Christoph Lameter95a05b42013-01-10 19:14:19 +00003767 if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) {
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03003768 ret = kmalloc_large_node(size, flags, node);
3769
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02003770 trace_kmalloc_node(_RET_IP_, ret,
3771 size, PAGE_SIZE << get_order(size),
3772 flags, node);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03003773
3774 return ret;
3775 }
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003776
Christoph Lameter2c59dd62013-01-10 19:14:19 +00003777 s = kmalloc_slab(size, flags);
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003778
3779 if (unlikely(ZERO_OR_NULL_PTR(s)))
Christoph Lameter6cb8f912007-07-17 04:03:22 -07003780 return s;
3781
Ezequiel Garcia2b847c32012-09-08 17:47:58 -03003782 ret = slab_alloc_node(s, flags, node, _RET_IP_);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03003783
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02003784 trace_kmalloc_node(_RET_IP_, ret, size, s->size, flags, node);
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03003785
Alexander Potapenko505f5dc2016-03-25 14:22:02 -07003786 kasan_kmalloc(s, ret, size, flags);
Andrey Ryabinin0316bec2015-02-13 14:39:42 -08003787
Eduard - Gabriel Munteanu5b882be2008-08-19 20:43:26 +03003788 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07003789}
3790EXPORT_SYMBOL(__kmalloc_node);
3791#endif
3792
Kees Cooked18adc2016-06-23 15:24:05 -07003793#ifdef CONFIG_HARDENED_USERCOPY
3794/*
3795 * Rejects objects that are incorrectly sized.
3796 *
3797 * Returns NULL if check passes, otherwise const char * to name of cache
3798 * to indicate an error.
3799 */
3800const char *__check_heap_object(const void *ptr, unsigned long n,
3801 struct page *page)
3802{
3803 struct kmem_cache *s;
3804 unsigned long offset;
3805 size_t object_size;
3806
3807 /* Find object and usable object size. */
3808 s = page->slab_cache;
3809 object_size = slab_ksize(s);
3810
3811 /* Reject impossible pointers. */
3812 if (ptr < page_address(page))
3813 return s->name;
3814
3815 /* Find offset within object. */
3816 offset = (ptr - page_address(page)) % s->size;
3817
3818 /* Adjust for redzone and reject if within the redzone. */
3819 if (kmem_cache_debug(s) && s->flags & SLAB_RED_ZONE) {
3820 if (offset < s->red_left_pad)
3821 return s->name;
3822 offset -= s->red_left_pad;
3823 }
3824
3825 /* Allow address range falling entirely within object size. */
3826 if (offset <= object_size && n <= object_size - offset)
3827 return NULL;
3828
3829 return s->name;
3830}
3831#endif /* CONFIG_HARDENED_USERCOPY */
3832
Andrey Ryabinin0316bec2015-02-13 14:39:42 -08003833static size_t __ksize(const void *object)
Christoph Lameter81819f02007-05-06 14:49:36 -07003834{
Christoph Lameter272c1d22007-06-08 13:46:49 -07003835 struct page *page;
Christoph Lameter81819f02007-05-06 14:49:36 -07003836
Christoph Lameteref8b4522007-10-16 01:24:46 -07003837 if (unlikely(object == ZERO_SIZE_PTR))
Christoph Lameter272c1d22007-06-08 13:46:49 -07003838 return 0;
3839
Vegard Nossum294a80a2007-12-04 23:45:30 -08003840 page = virt_to_head_page(object);
Vegard Nossum294a80a2007-12-04 23:45:30 -08003841
Pekka Enberg76994412008-05-22 19:22:25 +03003842 if (unlikely(!PageSlab(page))) {
3843 WARN_ON(!PageCompound(page));
Vegard Nossum294a80a2007-12-04 23:45:30 -08003844 return PAGE_SIZE << compound_order(page);
Pekka Enberg76994412008-05-22 19:22:25 +03003845 }
Christoph Lameter81819f02007-05-06 14:49:36 -07003846
Glauber Costa1b4f59e32012-10-22 18:05:36 +04003847 return slab_ksize(page->slab_cache);
Christoph Lameter81819f02007-05-06 14:49:36 -07003848}
Andrey Ryabinin0316bec2015-02-13 14:39:42 -08003849
3850size_t ksize(const void *object)
3851{
3852 size_t size = __ksize(object);
3853 /* We assume that ksize callers could use whole allocated area,
Alexander Potapenko4ebb31a42016-05-20 16:59:14 -07003854 * so we need to unpoison this area.
3855 */
3856 kasan_unpoison_shadow(object, size);
Andrey Ryabinin0316bec2015-02-13 14:39:42 -08003857 return size;
3858}
Kirill A. Shutemovb1aabec2009-02-10 15:21:44 +02003859EXPORT_SYMBOL(ksize);
Christoph Lameter81819f02007-05-06 14:49:36 -07003860
3861void kfree(const void *x)
3862{
Christoph Lameter81819f02007-05-06 14:49:36 -07003863 struct page *page;
Christoph Lameter5bb983b2008-02-07 17:47:41 -08003864 void *object = (void *)x;
Christoph Lameter81819f02007-05-06 14:49:36 -07003865
Pekka Enberg2121db72009-03-25 11:05:57 +02003866 trace_kfree(_RET_IP_, x);
3867
Satyam Sharma2408c552007-10-16 01:24:44 -07003868 if (unlikely(ZERO_OR_NULL_PTR(x)))
Christoph Lameter81819f02007-05-06 14:49:36 -07003869 return;
3870
Christoph Lameterb49af682007-05-06 14:49:41 -07003871 page = virt_to_head_page(x);
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003872 if (unlikely(!PageSlab(page))) {
Christoph Lameter09375022008-05-28 10:32:22 -07003873 BUG_ON(!PageCompound(page));
Roman Bobnievd56791b2013-10-08 15:58:57 -07003874 kfree_hook(x);
Vladimir Davydov49491482016-07-26 15:24:24 -07003875 __free_pages(page, compound_order(page));
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07003876 return;
3877 }
Jesper Dangaard Brouer81084652015-11-20 15:57:46 -08003878 slab_free(page->slab_cache, page, object, NULL, 1, _RET_IP_);
Christoph Lameter81819f02007-05-06 14:49:36 -07003879}
3880EXPORT_SYMBOL(kfree);
3881
Vladimir Davydov832f37f2015-02-12 14:59:41 -08003882#define SHRINK_PROMOTE_MAX 32
3883
Christoph Lameter2086d262007-05-06 14:49:46 -07003884/*
Vladimir Davydov832f37f2015-02-12 14:59:41 -08003885 * kmem_cache_shrink discards empty slabs and promotes the slabs filled
3886 * up most to the head of the partial lists. New allocations will then
3887 * fill those up and thus they can be removed from the partial lists.
Christoph Lameter672bba32007-05-09 02:32:39 -07003888 *
3889 * The slabs with the least items are placed last. This results in them
3890 * being allocated from last increasing the chance that the last objects
3891 * are freed in them.
Christoph Lameter2086d262007-05-06 14:49:46 -07003892 */
Vladimir Davydovbc01eb92017-03-17 00:48:31 +00003893int __kmem_cache_shrink(struct kmem_cache *s)
Christoph Lameter2086d262007-05-06 14:49:46 -07003894{
3895 int node;
3896 int i;
3897 struct kmem_cache_node *n;
3898 struct page *page;
3899 struct page *t;
Vladimir Davydov832f37f2015-02-12 14:59:41 -08003900 struct list_head discard;
3901 struct list_head promote[SHRINK_PROMOTE_MAX];
Christoph Lameter2086d262007-05-06 14:49:46 -07003902 unsigned long flags;
Vladimir Davydovce3712d2015-02-12 14:59:44 -08003903 int ret = 0;
Christoph Lameter2086d262007-05-06 14:49:46 -07003904
Christoph Lameter2086d262007-05-06 14:49:46 -07003905 flush_all(s);
Christoph Lameterfa45dc22014-08-06 16:04:09 -07003906 for_each_kmem_cache_node(s, node, n) {
Vladimir Davydov832f37f2015-02-12 14:59:41 -08003907 INIT_LIST_HEAD(&discard);
3908 for (i = 0; i < SHRINK_PROMOTE_MAX; i++)
3909 INIT_LIST_HEAD(promote + i);
Christoph Lameter2086d262007-05-06 14:49:46 -07003910
3911 spin_lock_irqsave(&n->list_lock, flags);
3912
3913 /*
Vladimir Davydov832f37f2015-02-12 14:59:41 -08003914 * Build lists of slabs to discard or promote.
Christoph Lameter2086d262007-05-06 14:49:46 -07003915 *
Christoph Lameter672bba32007-05-09 02:32:39 -07003916 * Note that concurrent frees may occur while we hold the
3917 * list_lock. page->inuse here is the upper limit.
Christoph Lameter2086d262007-05-06 14:49:46 -07003918 */
3919 list_for_each_entry_safe(page, t, &n->partial, lru) {
Vladimir Davydov832f37f2015-02-12 14:59:41 -08003920 int free = page->objects - page->inuse;
3921
3922 /* Do not reread page->inuse */
3923 barrier();
3924
3925 /* We do not keep full slabs on the list */
3926 BUG_ON(free <= 0);
3927
3928 if (free == page->objects) {
3929 list_move(&page->lru, &discard);
Christoph Lameter69cb8e62011-08-09 16:12:22 -05003930 n->nr_partial--;
Vladimir Davydov832f37f2015-02-12 14:59:41 -08003931 } else if (free <= SHRINK_PROMOTE_MAX)
3932 list_move(&page->lru, promote + free - 1);
Christoph Lameter2086d262007-05-06 14:49:46 -07003933 }
3934
Christoph Lameter2086d262007-05-06 14:49:46 -07003935 /*
Vladimir Davydov832f37f2015-02-12 14:59:41 -08003936 * Promote the slabs filled up most to the head of the
3937 * partial list.
Christoph Lameter2086d262007-05-06 14:49:46 -07003938 */
Vladimir Davydov832f37f2015-02-12 14:59:41 -08003939 for (i = SHRINK_PROMOTE_MAX - 1; i >= 0; i--)
3940 list_splice(promote + i, &n->partial);
Christoph Lameter2086d262007-05-06 14:49:46 -07003941
Christoph Lameter2086d262007-05-06 14:49:46 -07003942 spin_unlock_irqrestore(&n->list_lock, flags);
Christoph Lameter69cb8e62011-08-09 16:12:22 -05003943
3944 /* Release empty slabs */
Vladimir Davydov832f37f2015-02-12 14:59:41 -08003945 list_for_each_entry_safe(page, t, &discard, lru)
Christoph Lameter69cb8e62011-08-09 16:12:22 -05003946 discard_slab(s, page);
Vladimir Davydovce3712d2015-02-12 14:59:44 -08003947
3948 if (slabs_node(s, node))
3949 ret = 1;
Christoph Lameter2086d262007-05-06 14:49:46 -07003950 }
3951
Vladimir Davydovce3712d2015-02-12 14:59:44 -08003952 return ret;
Christoph Lameter2086d262007-05-06 14:49:46 -07003953}
Christoph Lameter2086d262007-05-06 14:49:46 -07003954
Yasunori Gotob9049e22007-10-21 16:41:37 -07003955static int slab_mem_going_offline_callback(void *arg)
3956{
3957 struct kmem_cache *s;
3958
Christoph Lameter18004c52012-07-06 15:25:12 -05003959 mutex_lock(&slab_mutex);
Yasunori Gotob9049e22007-10-21 16:41:37 -07003960 list_for_each_entry(s, &slab_caches, list)
Vladimir Davydovbc01eb92017-03-17 00:48:31 +00003961 __kmem_cache_shrink(s);
Christoph Lameter18004c52012-07-06 15:25:12 -05003962 mutex_unlock(&slab_mutex);
Yasunori Gotob9049e22007-10-21 16:41:37 -07003963
3964 return 0;
3965}
3966
3967static void slab_mem_offline_callback(void *arg)
3968{
3969 struct kmem_cache_node *n;
3970 struct kmem_cache *s;
3971 struct memory_notify *marg = arg;
3972 int offline_node;
3973
Lai Jiangshanb9d5ab22012-12-11 16:01:05 -08003974 offline_node = marg->status_change_nid_normal;
Yasunori Gotob9049e22007-10-21 16:41:37 -07003975
3976 /*
3977 * If the node still has available memory. we need kmem_cache_node
3978 * for it yet.
3979 */
3980 if (offline_node < 0)
3981 return;
3982
Christoph Lameter18004c52012-07-06 15:25:12 -05003983 mutex_lock(&slab_mutex);
Yasunori Gotob9049e22007-10-21 16:41:37 -07003984 list_for_each_entry(s, &slab_caches, list) {
3985 n = get_node(s, offline_node);
3986 if (n) {
3987 /*
3988 * if n->nr_slabs > 0, slabs still exist on the node
3989 * that is going down. We were unable to free them,
Adam Buchbinderc9404c92009-12-18 15:40:42 -05003990 * and offline_pages() function shouldn't call this
Yasunori Gotob9049e22007-10-21 16:41:37 -07003991 * callback. So, we must fail.
3992 */
Christoph Lameter0f389ec2008-04-14 18:53:02 +03003993 BUG_ON(slabs_node(s, offline_node));
Yasunori Gotob9049e22007-10-21 16:41:37 -07003994
3995 s->node[offline_node] = NULL;
Christoph Lameter8de66a02010-08-25 14:51:14 -05003996 kmem_cache_free(kmem_cache_node, n);
Yasunori Gotob9049e22007-10-21 16:41:37 -07003997 }
3998 }
Christoph Lameter18004c52012-07-06 15:25:12 -05003999 mutex_unlock(&slab_mutex);
Yasunori Gotob9049e22007-10-21 16:41:37 -07004000}
4001
4002static int slab_mem_going_online_callback(void *arg)
4003{
4004 struct kmem_cache_node *n;
4005 struct kmem_cache *s;
4006 struct memory_notify *marg = arg;
Lai Jiangshanb9d5ab22012-12-11 16:01:05 -08004007 int nid = marg->status_change_nid_normal;
Yasunori Gotob9049e22007-10-21 16:41:37 -07004008 int ret = 0;
4009
4010 /*
4011 * If the node's memory is already available, then kmem_cache_node is
4012 * already created. Nothing to do.
4013 */
4014 if (nid < 0)
4015 return 0;
4016
4017 /*
Christoph Lameter0121c6192008-04-29 16:11:12 -07004018 * We are bringing a node online. No memory is available yet. We must
Yasunori Gotob9049e22007-10-21 16:41:37 -07004019 * allocate a kmem_cache_node structure in order to bring the node
4020 * online.
4021 */
Christoph Lameter18004c52012-07-06 15:25:12 -05004022 mutex_lock(&slab_mutex);
Yasunori Gotob9049e22007-10-21 16:41:37 -07004023 list_for_each_entry(s, &slab_caches, list) {
4024 /*
4025 * XXX: kmem_cache_alloc_node will fallback to other nodes
4026 * since memory is not yet available from the node that
4027 * is brought up.
4028 */
Christoph Lameter8de66a02010-08-25 14:51:14 -05004029 n = kmem_cache_alloc(kmem_cache_node, GFP_KERNEL);
Yasunori Gotob9049e22007-10-21 16:41:37 -07004030 if (!n) {
4031 ret = -ENOMEM;
4032 goto out;
4033 }
Joonsoo Kim40534972012-05-11 00:50:47 +09004034 init_kmem_cache_node(n);
Yasunori Gotob9049e22007-10-21 16:41:37 -07004035 s->node[nid] = n;
4036 }
4037out:
Christoph Lameter18004c52012-07-06 15:25:12 -05004038 mutex_unlock(&slab_mutex);
Yasunori Gotob9049e22007-10-21 16:41:37 -07004039 return ret;
4040}
4041
4042static int slab_memory_callback(struct notifier_block *self,
4043 unsigned long action, void *arg)
4044{
4045 int ret = 0;
4046
4047 switch (action) {
4048 case MEM_GOING_ONLINE:
4049 ret = slab_mem_going_online_callback(arg);
4050 break;
4051 case MEM_GOING_OFFLINE:
4052 ret = slab_mem_going_offline_callback(arg);
4053 break;
4054 case MEM_OFFLINE:
4055 case MEM_CANCEL_ONLINE:
4056 slab_mem_offline_callback(arg);
4057 break;
4058 case MEM_ONLINE:
4059 case MEM_CANCEL_OFFLINE:
4060 break;
4061 }
KAMEZAWA Hiroyukidc19f9d2008-12-01 13:13:48 -08004062 if (ret)
4063 ret = notifier_from_errno(ret);
4064 else
4065 ret = NOTIFY_OK;
Yasunori Gotob9049e22007-10-21 16:41:37 -07004066 return ret;
4067}
4068
Andrew Morton3ac38fa2013-04-29 15:08:06 -07004069static struct notifier_block slab_memory_callback_nb = {
4070 .notifier_call = slab_memory_callback,
4071 .priority = SLAB_CALLBACK_PRI,
4072};
Yasunori Gotob9049e22007-10-21 16:41:37 -07004073
Christoph Lameter81819f02007-05-06 14:49:36 -07004074/********************************************************************
4075 * Basic setup of slabs
4076 *******************************************************************/
4077
Christoph Lameter51df1142010-08-20 12:37:15 -05004078/*
4079 * Used for early kmem_cache structures that were allocated using
Christoph Lameterdffb4d62012-11-28 16:23:07 +00004080 * the page allocator. Allocate them properly then fix up the pointers
4081 * that may be pointing to the wrong kmem_cache structure.
Christoph Lameter51df1142010-08-20 12:37:15 -05004082 */
4083
Christoph Lameterdffb4d62012-11-28 16:23:07 +00004084static struct kmem_cache * __init bootstrap(struct kmem_cache *static_cache)
Christoph Lameter51df1142010-08-20 12:37:15 -05004085{
4086 int node;
Christoph Lameterdffb4d62012-11-28 16:23:07 +00004087 struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT);
Christoph Lameterfa45dc22014-08-06 16:04:09 -07004088 struct kmem_cache_node *n;
Christoph Lameter51df1142010-08-20 12:37:15 -05004089
Christoph Lameterdffb4d62012-11-28 16:23:07 +00004090 memcpy(s, static_cache, kmem_cache->object_size);
Christoph Lameter51df1142010-08-20 12:37:15 -05004091
Glauber Costa7d557b32013-02-22 20:20:00 +04004092 /*
4093 * This runs very early, and only the boot processor is supposed to be
4094 * up. Even if it weren't true, IRQs are not up so we couldn't fire
4095 * IPIs around.
4096 */
4097 __flush_cpu_slab(s, smp_processor_id());
Christoph Lameterfa45dc22014-08-06 16:04:09 -07004098 for_each_kmem_cache_node(s, node, n) {
Christoph Lameter51df1142010-08-20 12:37:15 -05004099 struct page *p;
4100
Christoph Lameterfa45dc22014-08-06 16:04:09 -07004101 list_for_each_entry(p, &n->partial, lru)
4102 p->slab_cache = s;
Christoph Lameter51df1142010-08-20 12:37:15 -05004103
Li Zefan607bf322011-04-12 15:22:26 +08004104#ifdef CONFIG_SLUB_DEBUG
Christoph Lameterfa45dc22014-08-06 16:04:09 -07004105 list_for_each_entry(p, &n->full, lru)
4106 p->slab_cache = s;
Christoph Lameter51df1142010-08-20 12:37:15 -05004107#endif
Christoph Lameter51df1142010-08-20 12:37:15 -05004108 }
Vladimir Davydovf7ce3192015-02-12 14:59:20 -08004109 slab_init_memcg_params(s);
Christoph Lameterdffb4d62012-11-28 16:23:07 +00004110 list_add(&s->list, &slab_caches);
4111 return s;
Christoph Lameter51df1142010-08-20 12:37:15 -05004112}
4113
Christoph Lameter81819f02007-05-06 14:49:36 -07004114void __init kmem_cache_init(void)
4115{
Christoph Lameterdffb4d62012-11-28 16:23:07 +00004116 static __initdata struct kmem_cache boot_kmem_cache,
4117 boot_kmem_cache_node;
Christoph Lameter51df1142010-08-20 12:37:15 -05004118
Stanislaw Gruszkafc8d8622012-01-10 15:07:32 -08004119 if (debug_guardpage_minorder())
4120 slub_max_order = 0;
4121
Christoph Lameterdffb4d62012-11-28 16:23:07 +00004122 kmem_cache_node = &boot_kmem_cache_node;
4123 kmem_cache = &boot_kmem_cache;
Christoph Lameter51df1142010-08-20 12:37:15 -05004124
Christoph Lameterdffb4d62012-11-28 16:23:07 +00004125 create_boot_cache(kmem_cache_node, "kmem_cache_node",
4126 sizeof(struct kmem_cache_node), SLAB_HWCACHE_ALIGN);
Yasunori Gotob9049e22007-10-21 16:41:37 -07004127
Andrew Morton3ac38fa2013-04-29 15:08:06 -07004128 register_hotmemory_notifier(&slab_memory_callback_nb);
Christoph Lameter81819f02007-05-06 14:49:36 -07004129
4130 /* Able to allocate the per node structures */
4131 slab_state = PARTIAL;
4132
Christoph Lameterdffb4d62012-11-28 16:23:07 +00004133 create_boot_cache(kmem_cache, "kmem_cache",
4134 offsetof(struct kmem_cache, node) +
4135 nr_node_ids * sizeof(struct kmem_cache_node *),
4136 SLAB_HWCACHE_ALIGN);
Christoph Lameter8a13a4c2012-09-04 23:18:33 +00004137
Christoph Lameterdffb4d62012-11-28 16:23:07 +00004138 kmem_cache = bootstrap(&boot_kmem_cache);
Christoph Lameter81819f02007-05-06 14:49:36 -07004139
Christoph Lameter51df1142010-08-20 12:37:15 -05004140 /*
4141 * Allocate kmem_cache_node properly from the kmem_cache slab.
4142 * kmem_cache_node is separately allocated so no need to
4143 * update any list pointers.
4144 */
Christoph Lameterdffb4d62012-11-28 16:23:07 +00004145 kmem_cache_node = bootstrap(&boot_kmem_cache_node);
Christoph Lameter51df1142010-08-20 12:37:15 -05004146
4147 /* Now we can use the kmem_cache to allocate kmalloc slabs */
Daniel Sanders34cc6992015-06-24 16:55:57 -07004148 setup_kmalloc_cache_index_table();
Christoph Lameterf97d5f62013-01-10 19:12:17 +00004149 create_kmalloc_caches(0);
Christoph Lameter81819f02007-05-06 14:49:36 -07004150
Thomas Garnier210e7a42016-07-26 15:21:59 -07004151 /* Setup random freelists for each cache */
4152 init_freelist_randomization();
4153
Sebastian Andrzej Siewiora96a87b2016-08-18 14:57:19 +02004154 cpuhp_setup_state_nocalls(CPUHP_SLUB_DEAD, "slub:dead", NULL,
4155 slub_cpu_dead);
Christoph Lameter81819f02007-05-06 14:49:36 -07004156
Fabian Frederickf9f58282014-06-04 16:06:34 -07004157 pr_info("SLUB: HWalign=%d, Order=%d-%d, MinObjects=%d, CPUs=%d, Nodes=%d\n",
Christoph Lameterf97d5f62013-01-10 19:12:17 +00004158 cache_line_size(),
Christoph Lameter81819f02007-05-06 14:49:36 -07004159 slub_min_order, slub_max_order, slub_min_objects,
4160 nr_cpu_ids, nr_node_ids);
4161}
4162
Pekka Enberg7e85ee02009-06-12 14:03:06 +03004163void __init kmem_cache_init_late(void)
4164{
Pekka Enberg7e85ee02009-06-12 14:03:06 +03004165}
4166
Glauber Costa2633d7a2012-12-18 14:22:34 -08004167struct kmem_cache *
Vladimir Davydova44cb942014-04-07 15:39:23 -07004168__kmem_cache_alias(const char *name, size_t size, size_t align,
4169 unsigned long flags, void (*ctor)(void *))
Christoph Lameter81819f02007-05-06 14:49:36 -07004170{
Vladimir Davydov426589f2015-02-12 14:59:23 -08004171 struct kmem_cache *s, *c;
Christoph Lameter81819f02007-05-06 14:49:36 -07004172
Vladimir Davydova44cb942014-04-07 15:39:23 -07004173 s = find_mergeable(size, align, flags, name, ctor);
Christoph Lameter81819f02007-05-06 14:49:36 -07004174 if (s) {
4175 s->refcount++;
Vladimir Davydov84d0ddd2014-04-07 15:39:29 -07004176
Christoph Lameter81819f02007-05-06 14:49:36 -07004177 /*
4178 * Adjust the object sizes so that we clear
4179 * the complete object on kzalloc.
4180 */
Christoph Lameter3b0efdf2012-06-13 10:24:57 -05004181 s->object_size = max(s->object_size, (int)size);
Christoph Lameter81819f02007-05-06 14:49:36 -07004182 s->inuse = max_t(int, s->inuse, ALIGN(size, sizeof(void *)));
Christoph Lameter6446faa2008-02-15 23:45:26 -08004183
Vladimir Davydov426589f2015-02-12 14:59:23 -08004184 for_each_memcg_cache(c, s) {
Vladimir Davydov84d0ddd2014-04-07 15:39:29 -07004185 c->object_size = s->object_size;
4186 c->inuse = max_t(int, c->inuse,
4187 ALIGN(size, sizeof(void *)));
4188 }
4189
David Rientjes7b8f3b62008-12-17 22:09:46 -08004190 if (sysfs_slab_alias(s, name)) {
David Rientjes7b8f3b62008-12-17 22:09:46 -08004191 s->refcount--;
Christoph Lametercbb79692012-09-05 00:18:32 +00004192 s = NULL;
David Rientjes7b8f3b62008-12-17 22:09:46 -08004193 }
Christoph Lametera0e1d1b2007-07-17 04:03:31 -07004194 }
Christoph Lameter6446faa2008-02-15 23:45:26 -08004195
Christoph Lametercbb79692012-09-05 00:18:32 +00004196 return s;
4197}
Pekka Enberg84c1cf62010-09-14 23:21:12 +03004198
Christoph Lameter8a13a4c2012-09-04 23:18:33 +00004199int __kmem_cache_create(struct kmem_cache *s, unsigned long flags)
Christoph Lametercbb79692012-09-05 00:18:32 +00004200{
Pekka Enbergaac3a162012-09-05 12:07:44 +03004201 int err;
Christoph Lameter20cea962012-07-06 15:25:13 -05004202
Pekka Enbergaac3a162012-09-05 12:07:44 +03004203 err = kmem_cache_open(s, flags);
4204 if (err)
4205 return err;
Christoph Lameter20cea962012-07-06 15:25:13 -05004206
Christoph Lameter45530c42012-11-28 16:23:07 +00004207 /* Mutex is not taken during early boot */
4208 if (slab_state <= UP)
4209 return 0;
4210
Glauber Costa107dab52012-12-18 14:23:05 -08004211 memcg_propagate_slab_attrs(s);
Pekka Enbergaac3a162012-09-05 12:07:44 +03004212 err = sysfs_slab_add(s);
Pekka Enbergaac3a162012-09-05 12:07:44 +03004213 if (err)
Dmitry Safonov52b4b952016-02-17 13:11:37 -08004214 __kmem_cache_release(s);
Pekka Enbergaac3a162012-09-05 12:07:44 +03004215
4216 return err;
Christoph Lameter81819f02007-05-06 14:49:36 -07004217}
Christoph Lameter81819f02007-05-06 14:49:36 -07004218
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03004219void *__kmalloc_track_caller(size_t size, gfp_t gfpflags, unsigned long caller)
Christoph Lameter81819f02007-05-06 14:49:36 -07004220{
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07004221 struct kmem_cache *s;
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03004222 void *ret;
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07004223
Christoph Lameter95a05b42013-01-10 19:14:19 +00004224 if (unlikely(size > KMALLOC_MAX_CACHE_SIZE))
Pekka Enbergeada35e2008-02-11 22:47:46 +02004225 return kmalloc_large(size, gfpflags);
4226
Christoph Lameter2c59dd62013-01-10 19:14:19 +00004227 s = kmalloc_slab(size, gfpflags);
Christoph Lameter81819f02007-05-06 14:49:36 -07004228
Satyam Sharma2408c552007-10-16 01:24:44 -07004229 if (unlikely(ZERO_OR_NULL_PTR(s)))
Christoph Lameter6cb8f912007-07-17 04:03:22 -07004230 return s;
Christoph Lameter81819f02007-05-06 14:49:36 -07004231
Ezequiel Garcia2b847c32012-09-08 17:47:58 -03004232 ret = slab_alloc(s, gfpflags, caller);
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03004233
Lucas De Marchi25985ed2011-03-30 22:57:33 -03004234 /* Honor the call site pointer we received. */
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02004235 trace_kmalloc(caller, ret, size, s->size, gfpflags);
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03004236
4237 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07004238}
4239
Namhyung Kim5d1f57e2010-09-29 21:02:15 +09004240#ifdef CONFIG_NUMA
Christoph Lameter81819f02007-05-06 14:49:36 -07004241void *__kmalloc_node_track_caller(size_t size, gfp_t gfpflags,
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03004242 int node, unsigned long caller)
Christoph Lameter81819f02007-05-06 14:49:36 -07004243{
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07004244 struct kmem_cache *s;
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03004245 void *ret;
Christoph Lameteraadb4bc2007-10-16 01:24:38 -07004246
Christoph Lameter95a05b42013-01-10 19:14:19 +00004247 if (unlikely(size > KMALLOC_MAX_CACHE_SIZE)) {
Xiaotian Fengd3e14aa2010-04-08 17:26:44 +08004248 ret = kmalloc_large_node(size, gfpflags, node);
4249
4250 trace_kmalloc_node(caller, ret,
4251 size, PAGE_SIZE << get_order(size),
4252 gfpflags, node);
4253
4254 return ret;
4255 }
Pekka Enbergeada35e2008-02-11 22:47:46 +02004256
Christoph Lameter2c59dd62013-01-10 19:14:19 +00004257 s = kmalloc_slab(size, gfpflags);
Christoph Lameter81819f02007-05-06 14:49:36 -07004258
Satyam Sharma2408c552007-10-16 01:24:44 -07004259 if (unlikely(ZERO_OR_NULL_PTR(s)))
Christoph Lameter6cb8f912007-07-17 04:03:22 -07004260 return s;
Christoph Lameter81819f02007-05-06 14:49:36 -07004261
Ezequiel Garcia2b847c32012-09-08 17:47:58 -03004262 ret = slab_alloc_node(s, gfpflags, node, caller);
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03004263
Lucas De Marchi25985ed2011-03-30 22:57:33 -03004264 /* Honor the call site pointer we received. */
Eduard - Gabriel Munteanuca2b84cb2009-03-23 15:12:24 +02004265 trace_kmalloc_node(caller, ret, size, s->size, gfpflags, node);
Eduard - Gabriel Munteanu94b528d2008-08-24 20:49:35 +03004266
4267 return ret;
Christoph Lameter81819f02007-05-06 14:49:36 -07004268}
Namhyung Kim5d1f57e2010-09-29 21:02:15 +09004269#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07004270
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004271#ifdef CONFIG_SYSFS
Christoph Lameter205ab992008-04-14 19:11:40 +03004272static int count_inuse(struct page *page)
4273{
4274 return page->inuse;
4275}
4276
4277static int count_total(struct page *page)
4278{
4279 return page->objects;
4280}
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004281#endif
Christoph Lameter205ab992008-04-14 19:11:40 +03004282
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004283#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter434e2452007-07-17 04:03:30 -07004284static int validate_slab(struct kmem_cache *s, struct page *page,
4285 unsigned long *map)
Christoph Lameter53e15af2007-05-06 14:49:43 -07004286{
4287 void *p;
Christoph Lametera973e9d2008-03-01 13:40:44 -08004288 void *addr = page_address(page);
Christoph Lameter53e15af2007-05-06 14:49:43 -07004289
4290 if (!check_slab(s, page) ||
4291 !on_freelist(s, page, NULL))
4292 return 0;
4293
4294 /* Now we know that a valid freelist exists */
Christoph Lameter39b26462008-04-14 19:11:30 +03004295 bitmap_zero(map, page->objects);
Christoph Lameter53e15af2007-05-06 14:49:43 -07004296
Christoph Lameter5f80b132011-04-15 14:48:13 -05004297 get_map(s, page, map);
4298 for_each_object(p, s, addr, page->objects) {
4299 if (test_bit(slab_index(p, s, addr), map))
4300 if (!check_object(s, page, p, SLUB_RED_INACTIVE))
4301 return 0;
Christoph Lameter53e15af2007-05-06 14:49:43 -07004302 }
4303
Christoph Lameter224a88b2008-04-14 19:11:31 +03004304 for_each_object(p, s, addr, page->objects)
Christoph Lameter7656c722007-05-09 02:32:40 -07004305 if (!test_bit(slab_index(p, s, addr), map))
Tero Roponen37d57442010-12-01 20:04:20 +02004306 if (!check_object(s, page, p, SLUB_RED_ACTIVE))
Christoph Lameter53e15af2007-05-06 14:49:43 -07004307 return 0;
4308 return 1;
4309}
4310
Christoph Lameter434e2452007-07-17 04:03:30 -07004311static void validate_slab_slab(struct kmem_cache *s, struct page *page,
4312 unsigned long *map)
Christoph Lameter53e15af2007-05-06 14:49:43 -07004313{
Christoph Lameter881db7f2011-06-01 12:25:53 -05004314 slab_lock(page);
4315 validate_slab(s, page, map);
4316 slab_unlock(page);
Christoph Lameter53e15af2007-05-06 14:49:43 -07004317}
4318
Christoph Lameter434e2452007-07-17 04:03:30 -07004319static int validate_slab_node(struct kmem_cache *s,
4320 struct kmem_cache_node *n, unsigned long *map)
Christoph Lameter53e15af2007-05-06 14:49:43 -07004321{
4322 unsigned long count = 0;
4323 struct page *page;
4324 unsigned long flags;
4325
4326 spin_lock_irqsave(&n->list_lock, flags);
4327
4328 list_for_each_entry(page, &n->partial, lru) {
Christoph Lameter434e2452007-07-17 04:03:30 -07004329 validate_slab_slab(s, page, map);
Christoph Lameter53e15af2007-05-06 14:49:43 -07004330 count++;
4331 }
4332 if (count != n->nr_partial)
Fabian Frederickf9f58282014-06-04 16:06:34 -07004333 pr_err("SLUB %s: %ld partial slabs counted but counter=%ld\n",
4334 s->name, count, n->nr_partial);
Christoph Lameter53e15af2007-05-06 14:49:43 -07004335
4336 if (!(s->flags & SLAB_STORE_USER))
4337 goto out;
4338
4339 list_for_each_entry(page, &n->full, lru) {
Christoph Lameter434e2452007-07-17 04:03:30 -07004340 validate_slab_slab(s, page, map);
Christoph Lameter53e15af2007-05-06 14:49:43 -07004341 count++;
4342 }
4343 if (count != atomic_long_read(&n->nr_slabs))
Fabian Frederickf9f58282014-06-04 16:06:34 -07004344 pr_err("SLUB: %s %ld slabs counted but counter=%ld\n",
4345 s->name, count, atomic_long_read(&n->nr_slabs));
Christoph Lameter53e15af2007-05-06 14:49:43 -07004346
4347out:
4348 spin_unlock_irqrestore(&n->list_lock, flags);
4349 return count;
4350}
4351
Christoph Lameter434e2452007-07-17 04:03:30 -07004352static long validate_slab_cache(struct kmem_cache *s)
Christoph Lameter53e15af2007-05-06 14:49:43 -07004353{
4354 int node;
4355 unsigned long count = 0;
Christoph Lameter205ab992008-04-14 19:11:40 +03004356 unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) *
Christoph Lameter434e2452007-07-17 04:03:30 -07004357 sizeof(unsigned long), GFP_KERNEL);
Christoph Lameterfa45dc22014-08-06 16:04:09 -07004358 struct kmem_cache_node *n;
Christoph Lameter434e2452007-07-17 04:03:30 -07004359
4360 if (!map)
4361 return -ENOMEM;
Christoph Lameter53e15af2007-05-06 14:49:43 -07004362
4363 flush_all(s);
Christoph Lameterfa45dc22014-08-06 16:04:09 -07004364 for_each_kmem_cache_node(s, node, n)
Christoph Lameter434e2452007-07-17 04:03:30 -07004365 count += validate_slab_node(s, n, map);
Christoph Lameter434e2452007-07-17 04:03:30 -07004366 kfree(map);
Christoph Lameter53e15af2007-05-06 14:49:43 -07004367 return count;
4368}
Christoph Lameter88a420e2007-05-06 14:49:45 -07004369/*
Christoph Lameter672bba32007-05-09 02:32:39 -07004370 * Generate lists of code addresses where slabcache objects are allocated
Christoph Lameter88a420e2007-05-06 14:49:45 -07004371 * and freed.
4372 */
4373
4374struct location {
4375 unsigned long count;
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03004376 unsigned long addr;
Christoph Lameter45edfa52007-05-09 02:32:45 -07004377 long long sum_time;
4378 long min_time;
4379 long max_time;
4380 long min_pid;
4381 long max_pid;
Rusty Russell174596a2009-01-01 10:12:29 +10304382 DECLARE_BITMAP(cpus, NR_CPUS);
Christoph Lameter45edfa52007-05-09 02:32:45 -07004383 nodemask_t nodes;
Christoph Lameter88a420e2007-05-06 14:49:45 -07004384};
4385
4386struct loc_track {
4387 unsigned long max;
4388 unsigned long count;
4389 struct location *loc;
4390};
4391
4392static void free_loc_track(struct loc_track *t)
4393{
4394 if (t->max)
4395 free_pages((unsigned long)t->loc,
4396 get_order(sizeof(struct location) * t->max));
4397}
4398
Christoph Lameter68dff6a2007-07-17 04:03:20 -07004399static int alloc_loc_track(struct loc_track *t, unsigned long max, gfp_t flags)
Christoph Lameter88a420e2007-05-06 14:49:45 -07004400{
4401 struct location *l;
4402 int order;
4403
Christoph Lameter88a420e2007-05-06 14:49:45 -07004404 order = get_order(sizeof(struct location) * max);
4405
Christoph Lameter68dff6a2007-07-17 04:03:20 -07004406 l = (void *)__get_free_pages(flags, order);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004407 if (!l)
4408 return 0;
4409
4410 if (t->count) {
4411 memcpy(l, t->loc, sizeof(struct location) * t->count);
4412 free_loc_track(t);
4413 }
4414 t->max = max;
4415 t->loc = l;
4416 return 1;
4417}
4418
4419static int add_location(struct loc_track *t, struct kmem_cache *s,
Christoph Lameter45edfa52007-05-09 02:32:45 -07004420 const struct track *track)
Christoph Lameter88a420e2007-05-06 14:49:45 -07004421{
4422 long start, end, pos;
4423 struct location *l;
Eduard - Gabriel Munteanuce71e272008-08-19 20:43:25 +03004424 unsigned long caddr;
Christoph Lameter45edfa52007-05-09 02:32:45 -07004425 unsigned long age = jiffies - track->when;
Christoph Lameter88a420e2007-05-06 14:49:45 -07004426
4427 start = -1;
4428 end = t->count;
4429
4430 for ( ; ; ) {
4431 pos = start + (end - start + 1) / 2;
4432
4433 /*
4434 * There is nothing at "end". If we end up there
4435 * we need to add something to before end.
4436 */
4437 if (pos == end)
4438 break;
4439
4440 caddr = t->loc[pos].addr;
Christoph Lameter45edfa52007-05-09 02:32:45 -07004441 if (track->addr == caddr) {
4442
4443 l = &t->loc[pos];
4444 l->count++;
4445 if (track->when) {
4446 l->sum_time += age;
4447 if (age < l->min_time)
4448 l->min_time = age;
4449 if (age > l->max_time)
4450 l->max_time = age;
4451
4452 if (track->pid < l->min_pid)
4453 l->min_pid = track->pid;
4454 if (track->pid > l->max_pid)
4455 l->max_pid = track->pid;
4456
Rusty Russell174596a2009-01-01 10:12:29 +10304457 cpumask_set_cpu(track->cpu,
4458 to_cpumask(l->cpus));
Christoph Lameter45edfa52007-05-09 02:32:45 -07004459 }
4460 node_set(page_to_nid(virt_to_page(track)), l->nodes);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004461 return 1;
4462 }
4463
Christoph Lameter45edfa52007-05-09 02:32:45 -07004464 if (track->addr < caddr)
Christoph Lameter88a420e2007-05-06 14:49:45 -07004465 end = pos;
4466 else
4467 start = pos;
4468 }
4469
4470 /*
Christoph Lameter672bba32007-05-09 02:32:39 -07004471 * Not found. Insert new tracking element.
Christoph Lameter88a420e2007-05-06 14:49:45 -07004472 */
Christoph Lameter68dff6a2007-07-17 04:03:20 -07004473 if (t->count >= t->max && !alloc_loc_track(t, 2 * t->max, GFP_ATOMIC))
Christoph Lameter88a420e2007-05-06 14:49:45 -07004474 return 0;
4475
4476 l = t->loc + pos;
4477 if (pos < t->count)
4478 memmove(l + 1, l,
4479 (t->count - pos) * sizeof(struct location));
4480 t->count++;
4481 l->count = 1;
Christoph Lameter45edfa52007-05-09 02:32:45 -07004482 l->addr = track->addr;
4483 l->sum_time = age;
4484 l->min_time = age;
4485 l->max_time = age;
4486 l->min_pid = track->pid;
4487 l->max_pid = track->pid;
Rusty Russell174596a2009-01-01 10:12:29 +10304488 cpumask_clear(to_cpumask(l->cpus));
4489 cpumask_set_cpu(track->cpu, to_cpumask(l->cpus));
Christoph Lameter45edfa52007-05-09 02:32:45 -07004490 nodes_clear(l->nodes);
4491 node_set(page_to_nid(virt_to_page(track)), l->nodes);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004492 return 1;
4493}
4494
4495static void process_slab(struct loc_track *t, struct kmem_cache *s,
Eric Dumazetbbd7d572010-03-24 22:25:47 +01004496 struct page *page, enum track_item alloc,
Namhyung Kima5dd5c12010-09-29 21:02:13 +09004497 unsigned long *map)
Christoph Lameter88a420e2007-05-06 14:49:45 -07004498{
Christoph Lametera973e9d2008-03-01 13:40:44 -08004499 void *addr = page_address(page);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004500 void *p;
4501
Christoph Lameter39b26462008-04-14 19:11:30 +03004502 bitmap_zero(map, page->objects);
Christoph Lameter5f80b132011-04-15 14:48:13 -05004503 get_map(s, page, map);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004504
Christoph Lameter224a88b2008-04-14 19:11:31 +03004505 for_each_object(p, s, addr, page->objects)
Christoph Lameter45edfa52007-05-09 02:32:45 -07004506 if (!test_bit(slab_index(p, s, addr), map))
4507 add_location(t, s, get_track(s, p, alloc));
Christoph Lameter88a420e2007-05-06 14:49:45 -07004508}
4509
4510static int list_locations(struct kmem_cache *s, char *buf,
4511 enum track_item alloc)
4512{
Harvey Harrisone374d482008-01-31 15:20:50 -08004513 int len = 0;
Christoph Lameter88a420e2007-05-06 14:49:45 -07004514 unsigned long i;
Christoph Lameter68dff6a2007-07-17 04:03:20 -07004515 struct loc_track t = { 0, 0, NULL };
Christoph Lameter88a420e2007-05-06 14:49:45 -07004516 int node;
Eric Dumazetbbd7d572010-03-24 22:25:47 +01004517 unsigned long *map = kmalloc(BITS_TO_LONGS(oo_objects(s->max)) *
4518 sizeof(unsigned long), GFP_KERNEL);
Christoph Lameterfa45dc22014-08-06 16:04:09 -07004519 struct kmem_cache_node *n;
Christoph Lameter88a420e2007-05-06 14:49:45 -07004520
Eric Dumazetbbd7d572010-03-24 22:25:47 +01004521 if (!map || !alloc_loc_track(&t, PAGE_SIZE / sizeof(struct location),
4522 GFP_TEMPORARY)) {
4523 kfree(map);
Christoph Lameter68dff6a2007-07-17 04:03:20 -07004524 return sprintf(buf, "Out of memory\n");
Eric Dumazetbbd7d572010-03-24 22:25:47 +01004525 }
Christoph Lameter88a420e2007-05-06 14:49:45 -07004526 /* Push back cpu slabs */
4527 flush_all(s);
4528
Christoph Lameterfa45dc22014-08-06 16:04:09 -07004529 for_each_kmem_cache_node(s, node, n) {
Christoph Lameter88a420e2007-05-06 14:49:45 -07004530 unsigned long flags;
4531 struct page *page;
4532
Christoph Lameter9e869432007-08-22 14:01:56 -07004533 if (!atomic_long_read(&n->nr_slabs))
Christoph Lameter88a420e2007-05-06 14:49:45 -07004534 continue;
4535
4536 spin_lock_irqsave(&n->list_lock, flags);
4537 list_for_each_entry(page, &n->partial, lru)
Eric Dumazetbbd7d572010-03-24 22:25:47 +01004538 process_slab(&t, s, page, alloc, map);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004539 list_for_each_entry(page, &n->full, lru)
Eric Dumazetbbd7d572010-03-24 22:25:47 +01004540 process_slab(&t, s, page, alloc, map);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004541 spin_unlock_irqrestore(&n->list_lock, flags);
4542 }
4543
4544 for (i = 0; i < t.count; i++) {
Christoph Lameter45edfa52007-05-09 02:32:45 -07004545 struct location *l = &t.loc[i];
Christoph Lameter88a420e2007-05-06 14:49:45 -07004546
Hugh Dickins9c246242008-12-09 13:14:27 -08004547 if (len > PAGE_SIZE - KSYM_SYMBOL_LEN - 100)
Christoph Lameter88a420e2007-05-06 14:49:45 -07004548 break;
Harvey Harrisone374d482008-01-31 15:20:50 -08004549 len += sprintf(buf + len, "%7ld ", l->count);
Christoph Lameter45edfa52007-05-09 02:32:45 -07004550
4551 if (l->addr)
Joe Perches62c70bc2011-01-13 15:45:52 -08004552 len += sprintf(buf + len, "%pS", (void *)l->addr);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004553 else
Harvey Harrisone374d482008-01-31 15:20:50 -08004554 len += sprintf(buf + len, "<not-available>");
Christoph Lameter45edfa52007-05-09 02:32:45 -07004555
4556 if (l->sum_time != l->min_time) {
Harvey Harrisone374d482008-01-31 15:20:50 -08004557 len += sprintf(buf + len, " age=%ld/%ld/%ld",
Roman Zippelf8bd2252008-05-01 04:34:31 -07004558 l->min_time,
4559 (long)div_u64(l->sum_time, l->count),
4560 l->max_time);
Christoph Lameter45edfa52007-05-09 02:32:45 -07004561 } else
Harvey Harrisone374d482008-01-31 15:20:50 -08004562 len += sprintf(buf + len, " age=%ld",
Christoph Lameter45edfa52007-05-09 02:32:45 -07004563 l->min_time);
4564
4565 if (l->min_pid != l->max_pid)
Harvey Harrisone374d482008-01-31 15:20:50 -08004566 len += sprintf(buf + len, " pid=%ld-%ld",
Christoph Lameter45edfa52007-05-09 02:32:45 -07004567 l->min_pid, l->max_pid);
4568 else
Harvey Harrisone374d482008-01-31 15:20:50 -08004569 len += sprintf(buf + len, " pid=%ld",
Christoph Lameter45edfa52007-05-09 02:32:45 -07004570 l->min_pid);
4571
Rusty Russell174596a2009-01-01 10:12:29 +10304572 if (num_online_cpus() > 1 &&
4573 !cpumask_empty(to_cpumask(l->cpus)) &&
Tejun Heo5024c1d2015-02-13 14:37:59 -08004574 len < PAGE_SIZE - 60)
4575 len += scnprintf(buf + len, PAGE_SIZE - len - 50,
4576 " cpus=%*pbl",
4577 cpumask_pr_args(to_cpumask(l->cpus)));
Christoph Lameter45edfa52007-05-09 02:32:45 -07004578
Christoph Lameter62bc62a2009-06-16 15:32:15 -07004579 if (nr_online_nodes > 1 && !nodes_empty(l->nodes) &&
Tejun Heo5024c1d2015-02-13 14:37:59 -08004580 len < PAGE_SIZE - 60)
4581 len += scnprintf(buf + len, PAGE_SIZE - len - 50,
4582 " nodes=%*pbl",
4583 nodemask_pr_args(&l->nodes));
Christoph Lameter45edfa52007-05-09 02:32:45 -07004584
Harvey Harrisone374d482008-01-31 15:20:50 -08004585 len += sprintf(buf + len, "\n");
Christoph Lameter88a420e2007-05-06 14:49:45 -07004586 }
4587
4588 free_loc_track(&t);
Eric Dumazetbbd7d572010-03-24 22:25:47 +01004589 kfree(map);
Christoph Lameter88a420e2007-05-06 14:49:45 -07004590 if (!t.count)
Harvey Harrisone374d482008-01-31 15:20:50 -08004591 len += sprintf(buf, "No data\n");
4592 return len;
Christoph Lameter88a420e2007-05-06 14:49:45 -07004593}
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004594#endif
Christoph Lameter88a420e2007-05-06 14:49:45 -07004595
Christoph Lametera5a84752010-10-05 13:57:27 -05004596#ifdef SLUB_RESILIENCY_TEST
David Rientjesc07b8182014-08-06 16:04:16 -07004597static void __init resiliency_test(void)
Christoph Lametera5a84752010-10-05 13:57:27 -05004598{
4599 u8 *p;
4600
Christoph Lameter95a05b42013-01-10 19:14:19 +00004601 BUILD_BUG_ON(KMALLOC_MIN_SIZE > 16 || KMALLOC_SHIFT_HIGH < 10);
Christoph Lametera5a84752010-10-05 13:57:27 -05004602
Fabian Frederickf9f58282014-06-04 16:06:34 -07004603 pr_err("SLUB resiliency testing\n");
4604 pr_err("-----------------------\n");
4605 pr_err("A. Corruption after allocation\n");
Christoph Lametera5a84752010-10-05 13:57:27 -05004606
4607 p = kzalloc(16, GFP_KERNEL);
4608 p[16] = 0x12;
Fabian Frederickf9f58282014-06-04 16:06:34 -07004609 pr_err("\n1. kmalloc-16: Clobber Redzone/next pointer 0x12->0x%p\n\n",
4610 p + 16);
Christoph Lametera5a84752010-10-05 13:57:27 -05004611
4612 validate_slab_cache(kmalloc_caches[4]);
4613
4614 /* Hmmm... The next two are dangerous */
4615 p = kzalloc(32, GFP_KERNEL);
4616 p[32 + sizeof(void *)] = 0x34;
Fabian Frederickf9f58282014-06-04 16:06:34 -07004617 pr_err("\n2. kmalloc-32: Clobber next pointer/next slab 0x34 -> -0x%p\n",
4618 p);
4619 pr_err("If allocated object is overwritten then not detectable\n\n");
Christoph Lametera5a84752010-10-05 13:57:27 -05004620
4621 validate_slab_cache(kmalloc_caches[5]);
4622 p = kzalloc(64, GFP_KERNEL);
4623 p += 64 + (get_cycles() & 0xff) * sizeof(void *);
4624 *p = 0x56;
Fabian Frederickf9f58282014-06-04 16:06:34 -07004625 pr_err("\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n",
4626 p);
4627 pr_err("If allocated object is overwritten then not detectable\n\n");
Christoph Lametera5a84752010-10-05 13:57:27 -05004628 validate_slab_cache(kmalloc_caches[6]);
4629
Fabian Frederickf9f58282014-06-04 16:06:34 -07004630 pr_err("\nB. Corruption after free\n");
Christoph Lametera5a84752010-10-05 13:57:27 -05004631 p = kzalloc(128, GFP_KERNEL);
4632 kfree(p);
4633 *p = 0x78;
Fabian Frederickf9f58282014-06-04 16:06:34 -07004634 pr_err("1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p);
Christoph Lametera5a84752010-10-05 13:57:27 -05004635 validate_slab_cache(kmalloc_caches[7]);
4636
4637 p = kzalloc(256, GFP_KERNEL);
4638 kfree(p);
4639 p[50] = 0x9a;
Fabian Frederickf9f58282014-06-04 16:06:34 -07004640 pr_err("\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n", p);
Christoph Lametera5a84752010-10-05 13:57:27 -05004641 validate_slab_cache(kmalloc_caches[8]);
4642
4643 p = kzalloc(512, GFP_KERNEL);
4644 kfree(p);
4645 p[512] = 0xab;
Fabian Frederickf9f58282014-06-04 16:06:34 -07004646 pr_err("\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p);
Christoph Lametera5a84752010-10-05 13:57:27 -05004647 validate_slab_cache(kmalloc_caches[9]);
4648}
4649#else
4650#ifdef CONFIG_SYSFS
4651static void resiliency_test(void) {};
4652#endif
4653#endif
4654
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004655#ifdef CONFIG_SYSFS
Christoph Lameter81819f02007-05-06 14:49:36 -07004656enum slab_stat_type {
Christoph Lameter205ab992008-04-14 19:11:40 +03004657 SL_ALL, /* All slabs */
4658 SL_PARTIAL, /* Only partially allocated slabs */
4659 SL_CPU, /* Only slabs used for cpu caches */
4660 SL_OBJECTS, /* Determine allocated objects not slabs */
4661 SL_TOTAL /* Determine object capacity not slabs */
Christoph Lameter81819f02007-05-06 14:49:36 -07004662};
4663
Christoph Lameter205ab992008-04-14 19:11:40 +03004664#define SO_ALL (1 << SL_ALL)
Christoph Lameter81819f02007-05-06 14:49:36 -07004665#define SO_PARTIAL (1 << SL_PARTIAL)
4666#define SO_CPU (1 << SL_CPU)
4667#define SO_OBJECTS (1 << SL_OBJECTS)
Christoph Lameter205ab992008-04-14 19:11:40 +03004668#define SO_TOTAL (1 << SL_TOTAL)
Christoph Lameter81819f02007-05-06 14:49:36 -07004669
Cyrill Gorcunov62e5c4b2008-03-02 23:28:24 +03004670static ssize_t show_slab_objects(struct kmem_cache *s,
4671 char *buf, unsigned long flags)
Christoph Lameter81819f02007-05-06 14:49:36 -07004672{
4673 unsigned long total = 0;
Christoph Lameter81819f02007-05-06 14:49:36 -07004674 int node;
4675 int x;
4676 unsigned long *nodes;
Christoph Lameter81819f02007-05-06 14:49:36 -07004677
Chen Gange35e1a92013-07-12 08:23:48 +08004678 nodes = kzalloc(sizeof(unsigned long) * nr_node_ids, GFP_KERNEL);
Cyrill Gorcunov62e5c4b2008-03-02 23:28:24 +03004679 if (!nodes)
4680 return -ENOMEM;
Christoph Lameter81819f02007-05-06 14:49:36 -07004681
Christoph Lameter205ab992008-04-14 19:11:40 +03004682 if (flags & SO_CPU) {
4683 int cpu;
Christoph Lameter81819f02007-05-06 14:49:36 -07004684
Christoph Lameter205ab992008-04-14 19:11:40 +03004685 for_each_possible_cpu(cpu) {
Chen Gangd0e0ac92013-07-15 09:05:29 +08004686 struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab,
4687 cpu);
Christoph Lameterec3ab082012-05-09 10:09:56 -05004688 int node;
Christoph Lameter49e22582011-08-09 16:12:27 -05004689 struct page *page;
Christoph Lameterdfb4f092007-10-16 01:26:05 -07004690
Jason Low4db0c3c2015-04-15 16:14:08 -07004691 page = READ_ONCE(c->page);
Christoph Lameterec3ab082012-05-09 10:09:56 -05004692 if (!page)
4693 continue;
Christoph Lameter205ab992008-04-14 19:11:40 +03004694
Christoph Lameterec3ab082012-05-09 10:09:56 -05004695 node = page_to_nid(page);
4696 if (flags & SO_TOTAL)
4697 x = page->objects;
4698 else if (flags & SO_OBJECTS)
4699 x = page->inuse;
4700 else
4701 x = 1;
Christoph Lameter49e22582011-08-09 16:12:27 -05004702
Christoph Lameterec3ab082012-05-09 10:09:56 -05004703 total += x;
4704 nodes[node] += x;
4705
Jason Low4db0c3c2015-04-15 16:14:08 -07004706 page = READ_ONCE(c->partial);
Christoph Lameter49e22582011-08-09 16:12:27 -05004707 if (page) {
Li Zefan8afb1472013-09-10 11:43:37 +08004708 node = page_to_nid(page);
4709 if (flags & SO_TOTAL)
4710 WARN_ON_ONCE(1);
4711 else if (flags & SO_OBJECTS)
4712 WARN_ON_ONCE(1);
4713 else
4714 x = page->pages;
Eric Dumazetbc6697d2011-11-22 16:02:02 +01004715 total += x;
4716 nodes[node] += x;
Christoph Lameter49e22582011-08-09 16:12:27 -05004717 }
Christoph Lameter81819f02007-05-06 14:49:36 -07004718 }
4719 }
4720
Vladimir Davydovbfc8c902014-06-04 16:07:18 -07004721 get_online_mems();
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004722#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter205ab992008-04-14 19:11:40 +03004723 if (flags & SO_ALL) {
Christoph Lameterfa45dc22014-08-06 16:04:09 -07004724 struct kmem_cache_node *n;
4725
4726 for_each_kmem_cache_node(s, node, n) {
Christoph Lameter81819f02007-05-06 14:49:36 -07004727
Chen Gangd0e0ac92013-07-15 09:05:29 +08004728 if (flags & SO_TOTAL)
4729 x = atomic_long_read(&n->total_objects);
4730 else if (flags & SO_OBJECTS)
4731 x = atomic_long_read(&n->total_objects) -
4732 count_partial(n, count_free);
Christoph Lameter205ab992008-04-14 19:11:40 +03004733 else
4734 x = atomic_long_read(&n->nr_slabs);
4735 total += x;
4736 nodes[node] += x;
4737 }
4738
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004739 } else
4740#endif
4741 if (flags & SO_PARTIAL) {
Christoph Lameterfa45dc22014-08-06 16:04:09 -07004742 struct kmem_cache_node *n;
Christoph Lameter205ab992008-04-14 19:11:40 +03004743
Christoph Lameterfa45dc22014-08-06 16:04:09 -07004744 for_each_kmem_cache_node(s, node, n) {
Christoph Lameter205ab992008-04-14 19:11:40 +03004745 if (flags & SO_TOTAL)
4746 x = count_partial(n, count_total);
4747 else if (flags & SO_OBJECTS)
4748 x = count_partial(n, count_inuse);
Christoph Lameter81819f02007-05-06 14:49:36 -07004749 else
4750 x = n->nr_partial;
4751 total += x;
4752 nodes[node] += x;
4753 }
Christoph Lameter81819f02007-05-06 14:49:36 -07004754 }
Christoph Lameter81819f02007-05-06 14:49:36 -07004755 x = sprintf(buf, "%lu", total);
4756#ifdef CONFIG_NUMA
Christoph Lameterfa45dc22014-08-06 16:04:09 -07004757 for (node = 0; node < nr_node_ids; node++)
Christoph Lameter81819f02007-05-06 14:49:36 -07004758 if (nodes[node])
4759 x += sprintf(buf + x, " N%d=%lu",
4760 node, nodes[node]);
4761#endif
Vladimir Davydovbfc8c902014-06-04 16:07:18 -07004762 put_online_mems();
Christoph Lameter81819f02007-05-06 14:49:36 -07004763 kfree(nodes);
4764 return x + sprintf(buf + x, "\n");
4765}
4766
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004767#ifdef CONFIG_SLUB_DEBUG
Christoph Lameter81819f02007-05-06 14:49:36 -07004768static int any_slab_objects(struct kmem_cache *s)
4769{
4770 int node;
Christoph Lameterfa45dc22014-08-06 16:04:09 -07004771 struct kmem_cache_node *n;
Christoph Lameterdfb4f092007-10-16 01:26:05 -07004772
Christoph Lameterfa45dc22014-08-06 16:04:09 -07004773 for_each_kmem_cache_node(s, node, n)
Benjamin Herrenschmidt4ea33e22008-05-06 20:42:39 -07004774 if (atomic_long_read(&n->total_objects))
Christoph Lameter81819f02007-05-06 14:49:36 -07004775 return 1;
Christoph Lameterfa45dc22014-08-06 16:04:09 -07004776
Christoph Lameter81819f02007-05-06 14:49:36 -07004777 return 0;
4778}
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004779#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07004780
4781#define to_slab_attr(n) container_of(n, struct slab_attribute, attr)
Phil Carmody497888c2011-07-14 15:07:13 +03004782#define to_slab(n) container_of(n, struct kmem_cache, kobj)
Christoph Lameter81819f02007-05-06 14:49:36 -07004783
4784struct slab_attribute {
4785 struct attribute attr;
4786 ssize_t (*show)(struct kmem_cache *s, char *buf);
4787 ssize_t (*store)(struct kmem_cache *s, const char *x, size_t count);
4788};
4789
4790#define SLAB_ATTR_RO(_name) \
Vasiliy Kulikovab067e92011-09-27 21:54:53 +04004791 static struct slab_attribute _name##_attr = \
4792 __ATTR(_name, 0400, _name##_show, NULL)
Christoph Lameter81819f02007-05-06 14:49:36 -07004793
4794#define SLAB_ATTR(_name) \
4795 static struct slab_attribute _name##_attr = \
Vasiliy Kulikovab067e92011-09-27 21:54:53 +04004796 __ATTR(_name, 0600, _name##_show, _name##_store)
Christoph Lameter81819f02007-05-06 14:49:36 -07004797
Christoph Lameter81819f02007-05-06 14:49:36 -07004798static ssize_t slab_size_show(struct kmem_cache *s, char *buf)
4799{
4800 return sprintf(buf, "%d\n", s->size);
4801}
4802SLAB_ATTR_RO(slab_size);
4803
4804static ssize_t align_show(struct kmem_cache *s, char *buf)
4805{
4806 return sprintf(buf, "%d\n", s->align);
4807}
4808SLAB_ATTR_RO(align);
4809
4810static ssize_t object_size_show(struct kmem_cache *s, char *buf)
4811{
Christoph Lameter3b0efdf2012-06-13 10:24:57 -05004812 return sprintf(buf, "%d\n", s->object_size);
Christoph Lameter81819f02007-05-06 14:49:36 -07004813}
4814SLAB_ATTR_RO(object_size);
4815
4816static ssize_t objs_per_slab_show(struct kmem_cache *s, char *buf)
4817{
Christoph Lameter834f3d12008-04-14 19:11:31 +03004818 return sprintf(buf, "%d\n", oo_objects(s->oo));
Christoph Lameter81819f02007-05-06 14:49:36 -07004819}
4820SLAB_ATTR_RO(objs_per_slab);
4821
Christoph Lameter06b285d2008-04-14 19:11:41 +03004822static ssize_t order_store(struct kmem_cache *s,
4823 const char *buf, size_t length)
4824{
Christoph Lameter0121c6192008-04-29 16:11:12 -07004825 unsigned long order;
4826 int err;
4827
Jingoo Han3dbb95f2013-09-11 14:20:25 -07004828 err = kstrtoul(buf, 10, &order);
Christoph Lameter0121c6192008-04-29 16:11:12 -07004829 if (err)
4830 return err;
Christoph Lameter06b285d2008-04-14 19:11:41 +03004831
4832 if (order > slub_max_order || order < slub_min_order)
4833 return -EINVAL;
4834
4835 calculate_sizes(s, order);
4836 return length;
4837}
4838
Christoph Lameter81819f02007-05-06 14:49:36 -07004839static ssize_t order_show(struct kmem_cache *s, char *buf)
4840{
Christoph Lameter834f3d12008-04-14 19:11:31 +03004841 return sprintf(buf, "%d\n", oo_order(s->oo));
Christoph Lameter81819f02007-05-06 14:49:36 -07004842}
Christoph Lameter06b285d2008-04-14 19:11:41 +03004843SLAB_ATTR(order);
Christoph Lameter81819f02007-05-06 14:49:36 -07004844
David Rientjes73d342b2009-02-22 17:40:09 -08004845static ssize_t min_partial_show(struct kmem_cache *s, char *buf)
4846{
4847 return sprintf(buf, "%lu\n", s->min_partial);
4848}
4849
4850static ssize_t min_partial_store(struct kmem_cache *s, const char *buf,
4851 size_t length)
4852{
4853 unsigned long min;
4854 int err;
4855
Jingoo Han3dbb95f2013-09-11 14:20:25 -07004856 err = kstrtoul(buf, 10, &min);
David Rientjes73d342b2009-02-22 17:40:09 -08004857 if (err)
4858 return err;
4859
David Rientjesc0bdb232009-02-25 09:16:35 +02004860 set_min_partial(s, min);
David Rientjes73d342b2009-02-22 17:40:09 -08004861 return length;
4862}
4863SLAB_ATTR(min_partial);
4864
Christoph Lameter49e22582011-08-09 16:12:27 -05004865static ssize_t cpu_partial_show(struct kmem_cache *s, char *buf)
4866{
4867 return sprintf(buf, "%u\n", s->cpu_partial);
4868}
4869
4870static ssize_t cpu_partial_store(struct kmem_cache *s, const char *buf,
4871 size_t length)
4872{
4873 unsigned long objects;
4874 int err;
4875
Jingoo Han3dbb95f2013-09-11 14:20:25 -07004876 err = kstrtoul(buf, 10, &objects);
Christoph Lameter49e22582011-08-09 16:12:27 -05004877 if (err)
4878 return err;
Joonsoo Kim345c9052013-06-19 14:05:52 +09004879 if (objects && !kmem_cache_has_cpu_partial(s))
David Rientjes74ee4ef2012-01-09 13:19:45 -08004880 return -EINVAL;
Christoph Lameter49e22582011-08-09 16:12:27 -05004881
4882 s->cpu_partial = objects;
4883 flush_all(s);
4884 return length;
4885}
4886SLAB_ATTR(cpu_partial);
4887
Christoph Lameter81819f02007-05-06 14:49:36 -07004888static ssize_t ctor_show(struct kmem_cache *s, char *buf)
4889{
Joe Perches62c70bc2011-01-13 15:45:52 -08004890 if (!s->ctor)
4891 return 0;
4892 return sprintf(buf, "%pS\n", s->ctor);
Christoph Lameter81819f02007-05-06 14:49:36 -07004893}
4894SLAB_ATTR_RO(ctor);
4895
Christoph Lameter81819f02007-05-06 14:49:36 -07004896static ssize_t aliases_show(struct kmem_cache *s, char *buf)
4897{
Gu Zheng4307c142014-08-06 16:04:51 -07004898 return sprintf(buf, "%d\n", s->refcount < 0 ? 0 : s->refcount - 1);
Christoph Lameter81819f02007-05-06 14:49:36 -07004899}
4900SLAB_ATTR_RO(aliases);
4901
Christoph Lameter81819f02007-05-06 14:49:36 -07004902static ssize_t partial_show(struct kmem_cache *s, char *buf)
4903{
Christoph Lameterd9acf4b2008-02-15 15:22:21 -08004904 return show_slab_objects(s, buf, SO_PARTIAL);
Christoph Lameter81819f02007-05-06 14:49:36 -07004905}
4906SLAB_ATTR_RO(partial);
4907
4908static ssize_t cpu_slabs_show(struct kmem_cache *s, char *buf)
4909{
Christoph Lameterd9acf4b2008-02-15 15:22:21 -08004910 return show_slab_objects(s, buf, SO_CPU);
Christoph Lameter81819f02007-05-06 14:49:36 -07004911}
4912SLAB_ATTR_RO(cpu_slabs);
4913
4914static ssize_t objects_show(struct kmem_cache *s, char *buf)
4915{
Christoph Lameter205ab992008-04-14 19:11:40 +03004916 return show_slab_objects(s, buf, SO_ALL|SO_OBJECTS);
Christoph Lameter81819f02007-05-06 14:49:36 -07004917}
4918SLAB_ATTR_RO(objects);
4919
Christoph Lameter205ab992008-04-14 19:11:40 +03004920static ssize_t objects_partial_show(struct kmem_cache *s, char *buf)
4921{
4922 return show_slab_objects(s, buf, SO_PARTIAL|SO_OBJECTS);
4923}
4924SLAB_ATTR_RO(objects_partial);
4925
Christoph Lameter49e22582011-08-09 16:12:27 -05004926static ssize_t slabs_cpu_partial_show(struct kmem_cache *s, char *buf)
4927{
4928 int objects = 0;
4929 int pages = 0;
4930 int cpu;
4931 int len;
4932
4933 for_each_online_cpu(cpu) {
4934 struct page *page = per_cpu_ptr(s->cpu_slab, cpu)->partial;
4935
4936 if (page) {
4937 pages += page->pages;
4938 objects += page->pobjects;
4939 }
4940 }
4941
4942 len = sprintf(buf, "%d(%d)", objects, pages);
4943
4944#ifdef CONFIG_SMP
4945 for_each_online_cpu(cpu) {
4946 struct page *page = per_cpu_ptr(s->cpu_slab, cpu) ->partial;
4947
4948 if (page && len < PAGE_SIZE - 20)
4949 len += sprintf(buf + len, " C%d=%d(%d)", cpu,
4950 page->pobjects, page->pages);
4951 }
4952#endif
4953 return len + sprintf(buf + len, "\n");
4954}
4955SLAB_ATTR_RO(slabs_cpu_partial);
4956
Christoph Lameter81819f02007-05-06 14:49:36 -07004957static ssize_t reclaim_account_show(struct kmem_cache *s, char *buf)
4958{
4959 return sprintf(buf, "%d\n", !!(s->flags & SLAB_RECLAIM_ACCOUNT));
4960}
4961
4962static ssize_t reclaim_account_store(struct kmem_cache *s,
4963 const char *buf, size_t length)
4964{
4965 s->flags &= ~SLAB_RECLAIM_ACCOUNT;
4966 if (buf[0] == '1')
4967 s->flags |= SLAB_RECLAIM_ACCOUNT;
4968 return length;
4969}
4970SLAB_ATTR(reclaim_account);
4971
4972static ssize_t hwcache_align_show(struct kmem_cache *s, char *buf)
4973{
Christoph Lameter5af60832007-05-06 14:49:56 -07004974 return sprintf(buf, "%d\n", !!(s->flags & SLAB_HWCACHE_ALIGN));
Christoph Lameter81819f02007-05-06 14:49:36 -07004975}
4976SLAB_ATTR_RO(hwcache_align);
4977
4978#ifdef CONFIG_ZONE_DMA
4979static ssize_t cache_dma_show(struct kmem_cache *s, char *buf)
4980{
4981 return sprintf(buf, "%d\n", !!(s->flags & SLAB_CACHE_DMA));
4982}
4983SLAB_ATTR_RO(cache_dma);
4984#endif
4985
4986static ssize_t destroy_by_rcu_show(struct kmem_cache *s, char *buf)
4987{
4988 return sprintf(buf, "%d\n", !!(s->flags & SLAB_DESTROY_BY_RCU));
4989}
4990SLAB_ATTR_RO(destroy_by_rcu);
4991
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08004992static ssize_t reserved_show(struct kmem_cache *s, char *buf)
4993{
4994 return sprintf(buf, "%d\n", s->reserved);
4995}
4996SLAB_ATTR_RO(reserved);
4997
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05004998#ifdef CONFIG_SLUB_DEBUG
Christoph Lametera5a84752010-10-05 13:57:27 -05004999static ssize_t slabs_show(struct kmem_cache *s, char *buf)
5000{
5001 return show_slab_objects(s, buf, SO_ALL);
5002}
5003SLAB_ATTR_RO(slabs);
5004
5005static ssize_t total_objects_show(struct kmem_cache *s, char *buf)
5006{
5007 return show_slab_objects(s, buf, SO_ALL|SO_TOTAL);
5008}
5009SLAB_ATTR_RO(total_objects);
5010
5011static ssize_t sanity_checks_show(struct kmem_cache *s, char *buf)
5012{
Laura Abbottbecfda62016-03-15 14:55:06 -07005013 return sprintf(buf, "%d\n", !!(s->flags & SLAB_CONSISTENCY_CHECKS));
Christoph Lametera5a84752010-10-05 13:57:27 -05005014}
5015
5016static ssize_t sanity_checks_store(struct kmem_cache *s,
5017 const char *buf, size_t length)
5018{
Laura Abbottbecfda62016-03-15 14:55:06 -07005019 s->flags &= ~SLAB_CONSISTENCY_CHECKS;
Christoph Lameterb789ef52011-06-01 12:25:49 -05005020 if (buf[0] == '1') {
5021 s->flags &= ~__CMPXCHG_DOUBLE;
Laura Abbottbecfda62016-03-15 14:55:06 -07005022 s->flags |= SLAB_CONSISTENCY_CHECKS;
Christoph Lameterb789ef52011-06-01 12:25:49 -05005023 }
Christoph Lametera5a84752010-10-05 13:57:27 -05005024 return length;
5025}
5026SLAB_ATTR(sanity_checks);
5027
5028static ssize_t trace_show(struct kmem_cache *s, char *buf)
5029{
5030 return sprintf(buf, "%d\n", !!(s->flags & SLAB_TRACE));
5031}
5032
5033static ssize_t trace_store(struct kmem_cache *s, const char *buf,
5034 size_t length)
5035{
Christoph Lameterc9e16132014-10-09 15:26:11 -07005036 /*
5037 * Tracing a merged cache is going to give confusing results
5038 * as well as cause other issues like converting a mergeable
5039 * cache into an umergeable one.
5040 */
5041 if (s->refcount > 1)
5042 return -EINVAL;
5043
Christoph Lametera5a84752010-10-05 13:57:27 -05005044 s->flags &= ~SLAB_TRACE;
Christoph Lameterb789ef52011-06-01 12:25:49 -05005045 if (buf[0] == '1') {
5046 s->flags &= ~__CMPXCHG_DOUBLE;
Christoph Lametera5a84752010-10-05 13:57:27 -05005047 s->flags |= SLAB_TRACE;
Christoph Lameterb789ef52011-06-01 12:25:49 -05005048 }
Christoph Lametera5a84752010-10-05 13:57:27 -05005049 return length;
5050}
5051SLAB_ATTR(trace);
5052
Christoph Lameter81819f02007-05-06 14:49:36 -07005053static ssize_t red_zone_show(struct kmem_cache *s, char *buf)
5054{
5055 return sprintf(buf, "%d\n", !!(s->flags & SLAB_RED_ZONE));
5056}
5057
5058static ssize_t red_zone_store(struct kmem_cache *s,
5059 const char *buf, size_t length)
5060{
5061 if (any_slab_objects(s))
5062 return -EBUSY;
5063
5064 s->flags &= ~SLAB_RED_ZONE;
Christoph Lameterb789ef52011-06-01 12:25:49 -05005065 if (buf[0] == '1') {
Christoph Lameter81819f02007-05-06 14:49:36 -07005066 s->flags |= SLAB_RED_ZONE;
Christoph Lameterb789ef52011-06-01 12:25:49 -05005067 }
Christoph Lameter06b285d2008-04-14 19:11:41 +03005068 calculate_sizes(s, -1);
Christoph Lameter81819f02007-05-06 14:49:36 -07005069 return length;
5070}
5071SLAB_ATTR(red_zone);
5072
5073static ssize_t poison_show(struct kmem_cache *s, char *buf)
5074{
5075 return sprintf(buf, "%d\n", !!(s->flags & SLAB_POISON));
5076}
5077
5078static ssize_t poison_store(struct kmem_cache *s,
5079 const char *buf, size_t length)
5080{
5081 if (any_slab_objects(s))
5082 return -EBUSY;
5083
5084 s->flags &= ~SLAB_POISON;
Christoph Lameterb789ef52011-06-01 12:25:49 -05005085 if (buf[0] == '1') {
Christoph Lameter81819f02007-05-06 14:49:36 -07005086 s->flags |= SLAB_POISON;
Christoph Lameterb789ef52011-06-01 12:25:49 -05005087 }
Christoph Lameter06b285d2008-04-14 19:11:41 +03005088 calculate_sizes(s, -1);
Christoph Lameter81819f02007-05-06 14:49:36 -07005089 return length;
5090}
5091SLAB_ATTR(poison);
5092
5093static ssize_t store_user_show(struct kmem_cache *s, char *buf)
5094{
5095 return sprintf(buf, "%d\n", !!(s->flags & SLAB_STORE_USER));
5096}
5097
5098static ssize_t store_user_store(struct kmem_cache *s,
5099 const char *buf, size_t length)
5100{
5101 if (any_slab_objects(s))
5102 return -EBUSY;
5103
5104 s->flags &= ~SLAB_STORE_USER;
Christoph Lameterb789ef52011-06-01 12:25:49 -05005105 if (buf[0] == '1') {
5106 s->flags &= ~__CMPXCHG_DOUBLE;
Christoph Lameter81819f02007-05-06 14:49:36 -07005107 s->flags |= SLAB_STORE_USER;
Christoph Lameterb789ef52011-06-01 12:25:49 -05005108 }
Christoph Lameter06b285d2008-04-14 19:11:41 +03005109 calculate_sizes(s, -1);
Christoph Lameter81819f02007-05-06 14:49:36 -07005110 return length;
5111}
5112SLAB_ATTR(store_user);
5113
Christoph Lameter53e15af2007-05-06 14:49:43 -07005114static ssize_t validate_show(struct kmem_cache *s, char *buf)
5115{
5116 return 0;
5117}
5118
5119static ssize_t validate_store(struct kmem_cache *s,
5120 const char *buf, size_t length)
5121{
Christoph Lameter434e2452007-07-17 04:03:30 -07005122 int ret = -EINVAL;
5123
5124 if (buf[0] == '1') {
5125 ret = validate_slab_cache(s);
5126 if (ret >= 0)
5127 ret = length;
5128 }
5129 return ret;
Christoph Lameter53e15af2007-05-06 14:49:43 -07005130}
5131SLAB_ATTR(validate);
Christoph Lametera5a84752010-10-05 13:57:27 -05005132
5133static ssize_t alloc_calls_show(struct kmem_cache *s, char *buf)
5134{
5135 if (!(s->flags & SLAB_STORE_USER))
5136 return -ENOSYS;
5137 return list_locations(s, buf, TRACK_ALLOC);
5138}
5139SLAB_ATTR_RO(alloc_calls);
5140
5141static ssize_t free_calls_show(struct kmem_cache *s, char *buf)
5142{
5143 if (!(s->flags & SLAB_STORE_USER))
5144 return -ENOSYS;
5145 return list_locations(s, buf, TRACK_FREE);
5146}
5147SLAB_ATTR_RO(free_calls);
5148#endif /* CONFIG_SLUB_DEBUG */
5149
5150#ifdef CONFIG_FAILSLAB
5151static ssize_t failslab_show(struct kmem_cache *s, char *buf)
5152{
5153 return sprintf(buf, "%d\n", !!(s->flags & SLAB_FAILSLAB));
5154}
5155
5156static ssize_t failslab_store(struct kmem_cache *s, const char *buf,
5157 size_t length)
5158{
Christoph Lameterc9e16132014-10-09 15:26:11 -07005159 if (s->refcount > 1)
5160 return -EINVAL;
5161
Christoph Lametera5a84752010-10-05 13:57:27 -05005162 s->flags &= ~SLAB_FAILSLAB;
5163 if (buf[0] == '1')
5164 s->flags |= SLAB_FAILSLAB;
5165 return length;
5166}
5167SLAB_ATTR(failslab);
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05005168#endif
Christoph Lameter53e15af2007-05-06 14:49:43 -07005169
Christoph Lameter2086d262007-05-06 14:49:46 -07005170static ssize_t shrink_show(struct kmem_cache *s, char *buf)
5171{
5172 return 0;
5173}
5174
5175static ssize_t shrink_store(struct kmem_cache *s,
5176 const char *buf, size_t length)
5177{
Vladimir Davydov832f37f2015-02-12 14:59:41 -08005178 if (buf[0] == '1')
5179 kmem_cache_shrink(s);
5180 else
Christoph Lameter2086d262007-05-06 14:49:46 -07005181 return -EINVAL;
5182 return length;
5183}
5184SLAB_ATTR(shrink);
5185
Christoph Lameter81819f02007-05-06 14:49:36 -07005186#ifdef CONFIG_NUMA
Christoph Lameter98246012008-01-07 23:20:26 -08005187static ssize_t remote_node_defrag_ratio_show(struct kmem_cache *s, char *buf)
Christoph Lameter81819f02007-05-06 14:49:36 -07005188{
Christoph Lameter98246012008-01-07 23:20:26 -08005189 return sprintf(buf, "%d\n", s->remote_node_defrag_ratio / 10);
Christoph Lameter81819f02007-05-06 14:49:36 -07005190}
5191
Christoph Lameter98246012008-01-07 23:20:26 -08005192static ssize_t remote_node_defrag_ratio_store(struct kmem_cache *s,
Christoph Lameter81819f02007-05-06 14:49:36 -07005193 const char *buf, size_t length)
5194{
Christoph Lameter0121c6192008-04-29 16:11:12 -07005195 unsigned long ratio;
5196 int err;
Christoph Lameter81819f02007-05-06 14:49:36 -07005197
Jingoo Han3dbb95f2013-09-11 14:20:25 -07005198 err = kstrtoul(buf, 10, &ratio);
Christoph Lameter0121c6192008-04-29 16:11:12 -07005199 if (err)
5200 return err;
5201
Christoph Lametere2cb96b2008-08-19 08:51:22 -05005202 if (ratio <= 100)
Christoph Lameter0121c6192008-04-29 16:11:12 -07005203 s->remote_node_defrag_ratio = ratio * 10;
5204
Christoph Lameter81819f02007-05-06 14:49:36 -07005205 return length;
5206}
Christoph Lameter98246012008-01-07 23:20:26 -08005207SLAB_ATTR(remote_node_defrag_ratio);
Christoph Lameter81819f02007-05-06 14:49:36 -07005208#endif
5209
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08005210#ifdef CONFIG_SLUB_STATS
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08005211static int show_stat(struct kmem_cache *s, char *buf, enum stat_item si)
5212{
5213 unsigned long sum = 0;
5214 int cpu;
5215 int len;
5216 int *data = kmalloc(nr_cpu_ids * sizeof(int), GFP_KERNEL);
5217
5218 if (!data)
5219 return -ENOMEM;
5220
5221 for_each_online_cpu(cpu) {
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06005222 unsigned x = per_cpu_ptr(s->cpu_slab, cpu)->stat[si];
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08005223
5224 data[cpu] = x;
5225 sum += x;
5226 }
5227
5228 len = sprintf(buf, "%lu", sum);
5229
Christoph Lameter50ef37b2008-04-14 18:52:05 +03005230#ifdef CONFIG_SMP
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08005231 for_each_online_cpu(cpu) {
5232 if (data[cpu] && len < PAGE_SIZE - 20)
Christoph Lameter50ef37b2008-04-14 18:52:05 +03005233 len += sprintf(buf + len, " C%d=%u", cpu, data[cpu]);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08005234 }
Christoph Lameter50ef37b2008-04-14 18:52:05 +03005235#endif
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08005236 kfree(data);
5237 return len + sprintf(buf + len, "\n");
5238}
5239
David Rientjes78eb00c2009-10-15 02:20:22 -07005240static void clear_stat(struct kmem_cache *s, enum stat_item si)
5241{
5242 int cpu;
5243
5244 for_each_online_cpu(cpu)
Christoph Lameter9dfc6e62009-12-18 16:26:20 -06005245 per_cpu_ptr(s->cpu_slab, cpu)->stat[si] = 0;
David Rientjes78eb00c2009-10-15 02:20:22 -07005246}
5247
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08005248#define STAT_ATTR(si, text) \
5249static ssize_t text##_show(struct kmem_cache *s, char *buf) \
5250{ \
5251 return show_stat(s, buf, si); \
5252} \
David Rientjes78eb00c2009-10-15 02:20:22 -07005253static ssize_t text##_store(struct kmem_cache *s, \
5254 const char *buf, size_t length) \
5255{ \
5256 if (buf[0] != '0') \
5257 return -EINVAL; \
5258 clear_stat(s, si); \
5259 return length; \
5260} \
5261SLAB_ATTR(text); \
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08005262
5263STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath);
5264STAT_ATTR(ALLOC_SLOWPATH, alloc_slowpath);
5265STAT_ATTR(FREE_FASTPATH, free_fastpath);
5266STAT_ATTR(FREE_SLOWPATH, free_slowpath);
5267STAT_ATTR(FREE_FROZEN, free_frozen);
5268STAT_ATTR(FREE_ADD_PARTIAL, free_add_partial);
5269STAT_ATTR(FREE_REMOVE_PARTIAL, free_remove_partial);
5270STAT_ATTR(ALLOC_FROM_PARTIAL, alloc_from_partial);
5271STAT_ATTR(ALLOC_SLAB, alloc_slab);
5272STAT_ATTR(ALLOC_REFILL, alloc_refill);
Christoph Lametere36a2652011-06-01 12:25:57 -05005273STAT_ATTR(ALLOC_NODE_MISMATCH, alloc_node_mismatch);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08005274STAT_ATTR(FREE_SLAB, free_slab);
5275STAT_ATTR(CPUSLAB_FLUSH, cpuslab_flush);
5276STAT_ATTR(DEACTIVATE_FULL, deactivate_full);
5277STAT_ATTR(DEACTIVATE_EMPTY, deactivate_empty);
5278STAT_ATTR(DEACTIVATE_TO_HEAD, deactivate_to_head);
5279STAT_ATTR(DEACTIVATE_TO_TAIL, deactivate_to_tail);
5280STAT_ATTR(DEACTIVATE_REMOTE_FREES, deactivate_remote_frees);
Christoph Lameter03e404a2011-06-01 12:25:58 -05005281STAT_ATTR(DEACTIVATE_BYPASS, deactivate_bypass);
Christoph Lameter65c33762008-04-14 19:11:40 +03005282STAT_ATTR(ORDER_FALLBACK, order_fallback);
Christoph Lameterb789ef52011-06-01 12:25:49 -05005283STAT_ATTR(CMPXCHG_DOUBLE_CPU_FAIL, cmpxchg_double_cpu_fail);
5284STAT_ATTR(CMPXCHG_DOUBLE_FAIL, cmpxchg_double_fail);
Christoph Lameter49e22582011-08-09 16:12:27 -05005285STAT_ATTR(CPU_PARTIAL_ALLOC, cpu_partial_alloc);
5286STAT_ATTR(CPU_PARTIAL_FREE, cpu_partial_free);
Alex Shi8028dce2012-02-03 23:34:56 +08005287STAT_ATTR(CPU_PARTIAL_NODE, cpu_partial_node);
5288STAT_ATTR(CPU_PARTIAL_DRAIN, cpu_partial_drain);
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08005289#endif
5290
Pekka Enberg06428782008-01-07 23:20:27 -08005291static struct attribute *slab_attrs[] = {
Christoph Lameter81819f02007-05-06 14:49:36 -07005292 &slab_size_attr.attr,
5293 &object_size_attr.attr,
5294 &objs_per_slab_attr.attr,
5295 &order_attr.attr,
David Rientjes73d342b2009-02-22 17:40:09 -08005296 &min_partial_attr.attr,
Christoph Lameter49e22582011-08-09 16:12:27 -05005297 &cpu_partial_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07005298 &objects_attr.attr,
Christoph Lameter205ab992008-04-14 19:11:40 +03005299 &objects_partial_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07005300 &partial_attr.attr,
5301 &cpu_slabs_attr.attr,
5302 &ctor_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07005303 &aliases_attr.attr,
5304 &align_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07005305 &hwcache_align_attr.attr,
5306 &reclaim_account_attr.attr,
5307 &destroy_by_rcu_attr.attr,
Christoph Lametera5a84752010-10-05 13:57:27 -05005308 &shrink_attr.attr,
Lai Jiangshanab9a0f12011-03-10 15:21:48 +08005309 &reserved_attr.attr,
Christoph Lameter49e22582011-08-09 16:12:27 -05005310 &slabs_cpu_partial_attr.attr,
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05005311#ifdef CONFIG_SLUB_DEBUG
Christoph Lametera5a84752010-10-05 13:57:27 -05005312 &total_objects_attr.attr,
5313 &slabs_attr.attr,
5314 &sanity_checks_attr.attr,
5315 &trace_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07005316 &red_zone_attr.attr,
5317 &poison_attr.attr,
5318 &store_user_attr.attr,
Christoph Lameter53e15af2007-05-06 14:49:43 -07005319 &validate_attr.attr,
Christoph Lameter88a420e2007-05-06 14:49:45 -07005320 &alloc_calls_attr.attr,
5321 &free_calls_attr.attr,
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05005322#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07005323#ifdef CONFIG_ZONE_DMA
5324 &cache_dma_attr.attr,
5325#endif
5326#ifdef CONFIG_NUMA
Christoph Lameter98246012008-01-07 23:20:26 -08005327 &remote_node_defrag_ratio_attr.attr,
Christoph Lameter81819f02007-05-06 14:49:36 -07005328#endif
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08005329#ifdef CONFIG_SLUB_STATS
5330 &alloc_fastpath_attr.attr,
5331 &alloc_slowpath_attr.attr,
5332 &free_fastpath_attr.attr,
5333 &free_slowpath_attr.attr,
5334 &free_frozen_attr.attr,
5335 &free_add_partial_attr.attr,
5336 &free_remove_partial_attr.attr,
5337 &alloc_from_partial_attr.attr,
5338 &alloc_slab_attr.attr,
5339 &alloc_refill_attr.attr,
Christoph Lametere36a2652011-06-01 12:25:57 -05005340 &alloc_node_mismatch_attr.attr,
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08005341 &free_slab_attr.attr,
5342 &cpuslab_flush_attr.attr,
5343 &deactivate_full_attr.attr,
5344 &deactivate_empty_attr.attr,
5345 &deactivate_to_head_attr.attr,
5346 &deactivate_to_tail_attr.attr,
5347 &deactivate_remote_frees_attr.attr,
Christoph Lameter03e404a2011-06-01 12:25:58 -05005348 &deactivate_bypass_attr.attr,
Christoph Lameter65c33762008-04-14 19:11:40 +03005349 &order_fallback_attr.attr,
Christoph Lameterb789ef52011-06-01 12:25:49 -05005350 &cmpxchg_double_fail_attr.attr,
5351 &cmpxchg_double_cpu_fail_attr.attr,
Christoph Lameter49e22582011-08-09 16:12:27 -05005352 &cpu_partial_alloc_attr.attr,
5353 &cpu_partial_free_attr.attr,
Alex Shi8028dce2012-02-03 23:34:56 +08005354 &cpu_partial_node_attr.attr,
5355 &cpu_partial_drain_attr.attr,
Christoph Lameter8ff12cf2008-02-07 17:47:41 -08005356#endif
Dmitry Monakhov4c13dd32010-02-26 09:36:12 +03005357#ifdef CONFIG_FAILSLAB
5358 &failslab_attr.attr,
5359#endif
5360
Christoph Lameter81819f02007-05-06 14:49:36 -07005361 NULL
5362};
5363
5364static struct attribute_group slab_attr_group = {
5365 .attrs = slab_attrs,
5366};
5367
5368static ssize_t slab_attr_show(struct kobject *kobj,
5369 struct attribute *attr,
5370 char *buf)
5371{
5372 struct slab_attribute *attribute;
5373 struct kmem_cache *s;
5374 int err;
5375
5376 attribute = to_slab_attr(attr);
5377 s = to_slab(kobj);
5378
5379 if (!attribute->show)
5380 return -EIO;
5381
5382 err = attribute->show(s, buf);
5383
5384 return err;
5385}
5386
5387static ssize_t slab_attr_store(struct kobject *kobj,
5388 struct attribute *attr,
5389 const char *buf, size_t len)
5390{
5391 struct slab_attribute *attribute;
5392 struct kmem_cache *s;
5393 int err;
5394
5395 attribute = to_slab_attr(attr);
5396 s = to_slab(kobj);
5397
5398 if (!attribute->store)
5399 return -EIO;
5400
5401 err = attribute->store(s, buf, len);
Johannes Weiner127424c2016-01-20 15:02:32 -08005402#ifdef CONFIG_MEMCG
Glauber Costa107dab52012-12-18 14:23:05 -08005403 if (slab_state >= FULL && err >= 0 && is_root_cache(s)) {
Vladimir Davydov426589f2015-02-12 14:59:23 -08005404 struct kmem_cache *c;
Christoph Lameter81819f02007-05-06 14:49:36 -07005405
Glauber Costa107dab52012-12-18 14:23:05 -08005406 mutex_lock(&slab_mutex);
5407 if (s->max_attr_size < len)
5408 s->max_attr_size = len;
5409
Glauber Costaebe945c2012-12-18 14:23:10 -08005410 /*
5411 * This is a best effort propagation, so this function's return
5412 * value will be determined by the parent cache only. This is
5413 * basically because not all attributes will have a well
5414 * defined semantics for rollbacks - most of the actions will
5415 * have permanent effects.
5416 *
5417 * Returning the error value of any of the children that fail
5418 * is not 100 % defined, in the sense that users seeing the
5419 * error code won't be able to know anything about the state of
5420 * the cache.
5421 *
5422 * Only returning the error code for the parent cache at least
5423 * has well defined semantics. The cache being written to
5424 * directly either failed or succeeded, in which case we loop
5425 * through the descendants with best-effort propagation.
5426 */
Vladimir Davydov426589f2015-02-12 14:59:23 -08005427 for_each_memcg_cache(c, s)
5428 attribute->store(c, buf, len);
Glauber Costa107dab52012-12-18 14:23:05 -08005429 mutex_unlock(&slab_mutex);
5430 }
5431#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07005432 return err;
5433}
5434
Glauber Costa107dab52012-12-18 14:23:05 -08005435static void memcg_propagate_slab_attrs(struct kmem_cache *s)
5436{
Johannes Weiner127424c2016-01-20 15:02:32 -08005437#ifdef CONFIG_MEMCG
Glauber Costa107dab52012-12-18 14:23:05 -08005438 int i;
5439 char *buffer = NULL;
Vladimir Davydov93030d82014-05-06 12:49:59 -07005440 struct kmem_cache *root_cache;
Glauber Costa107dab52012-12-18 14:23:05 -08005441
Vladimir Davydov93030d82014-05-06 12:49:59 -07005442 if (is_root_cache(s))
Glauber Costa107dab52012-12-18 14:23:05 -08005443 return;
5444
Vladimir Davydovf7ce3192015-02-12 14:59:20 -08005445 root_cache = s->memcg_params.root_cache;
Vladimir Davydov93030d82014-05-06 12:49:59 -07005446
Glauber Costa107dab52012-12-18 14:23:05 -08005447 /*
5448 * This mean this cache had no attribute written. Therefore, no point
5449 * in copying default values around
5450 */
Vladimir Davydov93030d82014-05-06 12:49:59 -07005451 if (!root_cache->max_attr_size)
Glauber Costa107dab52012-12-18 14:23:05 -08005452 return;
5453
5454 for (i = 0; i < ARRAY_SIZE(slab_attrs); i++) {
5455 char mbuf[64];
5456 char *buf;
5457 struct slab_attribute *attr = to_slab_attr(slab_attrs[i]);
Thomas Gleixnerc1bb2a82017-06-02 14:46:25 -07005458 ssize_t len;
Glauber Costa107dab52012-12-18 14:23:05 -08005459
5460 if (!attr || !attr->store || !attr->show)
5461 continue;
5462
5463 /*
5464 * It is really bad that we have to allocate here, so we will
5465 * do it only as a fallback. If we actually allocate, though,
5466 * we can just use the allocated buffer until the end.
5467 *
5468 * Most of the slub attributes will tend to be very small in
5469 * size, but sysfs allows buffers up to a page, so they can
5470 * theoretically happen.
5471 */
5472 if (buffer)
5473 buf = buffer;
Vladimir Davydov93030d82014-05-06 12:49:59 -07005474 else if (root_cache->max_attr_size < ARRAY_SIZE(mbuf))
Glauber Costa107dab52012-12-18 14:23:05 -08005475 buf = mbuf;
5476 else {
5477 buffer = (char *) get_zeroed_page(GFP_KERNEL);
5478 if (WARN_ON(!buffer))
5479 continue;
5480 buf = buffer;
5481 }
5482
Thomas Gleixnerc1bb2a82017-06-02 14:46:25 -07005483 len = attr->show(root_cache, buf);
5484 if (len > 0)
5485 attr->store(s, buf, len);
Glauber Costa107dab52012-12-18 14:23:05 -08005486 }
5487
5488 if (buffer)
5489 free_page((unsigned long)buffer);
5490#endif
5491}
5492
Christoph Lameter41a21282014-05-06 12:50:08 -07005493static void kmem_cache_release(struct kobject *k)
5494{
5495 slab_kmem_cache_release(to_slab(k));
5496}
5497
Emese Revfy52cf25d2010-01-19 02:58:23 +01005498static const struct sysfs_ops slab_sysfs_ops = {
Christoph Lameter81819f02007-05-06 14:49:36 -07005499 .show = slab_attr_show,
5500 .store = slab_attr_store,
5501};
5502
5503static struct kobj_type slab_ktype = {
5504 .sysfs_ops = &slab_sysfs_ops,
Christoph Lameter41a21282014-05-06 12:50:08 -07005505 .release = kmem_cache_release,
Christoph Lameter81819f02007-05-06 14:49:36 -07005506};
5507
5508static int uevent_filter(struct kset *kset, struct kobject *kobj)
5509{
5510 struct kobj_type *ktype = get_ktype(kobj);
5511
5512 if (ktype == &slab_ktype)
5513 return 1;
5514 return 0;
5515}
5516
Emese Revfy9cd43612009-12-31 14:52:51 +01005517static const struct kset_uevent_ops slab_uevent_ops = {
Christoph Lameter81819f02007-05-06 14:49:36 -07005518 .filter = uevent_filter,
5519};
5520
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06005521static struct kset *slab_kset;
Christoph Lameter81819f02007-05-06 14:49:36 -07005522
Vladimir Davydov9a417072014-04-07 15:39:31 -07005523static inline struct kset *cache_kset(struct kmem_cache *s)
5524{
Johannes Weiner127424c2016-01-20 15:02:32 -08005525#ifdef CONFIG_MEMCG
Vladimir Davydov9a417072014-04-07 15:39:31 -07005526 if (!is_root_cache(s))
Vladimir Davydovf7ce3192015-02-12 14:59:20 -08005527 return s->memcg_params.root_cache->memcg_kset;
Vladimir Davydov9a417072014-04-07 15:39:31 -07005528#endif
5529 return slab_kset;
5530}
5531
Christoph Lameter81819f02007-05-06 14:49:36 -07005532#define ID_STR_LENGTH 64
5533
5534/* Create a unique string id for a slab cache:
Christoph Lameter6446faa2008-02-15 23:45:26 -08005535 *
5536 * Format :[flags-]size
Christoph Lameter81819f02007-05-06 14:49:36 -07005537 */
5538static char *create_unique_id(struct kmem_cache *s)
5539{
5540 char *name = kmalloc(ID_STR_LENGTH, GFP_KERNEL);
5541 char *p = name;
5542
5543 BUG_ON(!name);
5544
5545 *p++ = ':';
5546 /*
5547 * First flags affecting slabcache operations. We will only
5548 * get here for aliasable slabs so we do not need to support
5549 * too many flags. The flags here must cover all flags that
5550 * are matched during merging to guarantee that the id is
5551 * unique.
5552 */
5553 if (s->flags & SLAB_CACHE_DMA)
5554 *p++ = 'd';
5555 if (s->flags & SLAB_RECLAIM_ACCOUNT)
5556 *p++ = 'a';
Laura Abbottbecfda62016-03-15 14:55:06 -07005557 if (s->flags & SLAB_CONSISTENCY_CHECKS)
Christoph Lameter81819f02007-05-06 14:49:36 -07005558 *p++ = 'F';
Vegard Nossum5a896d92008-04-04 00:54:48 +02005559 if (!(s->flags & SLAB_NOTRACK))
5560 *p++ = 't';
Vladimir Davydov230e9fc2016-01-14 15:18:15 -08005561 if (s->flags & SLAB_ACCOUNT)
5562 *p++ = 'A';
Christoph Lameter81819f02007-05-06 14:49:36 -07005563 if (p != name + 1)
5564 *p++ = '-';
5565 p += sprintf(p, "%07d", s->size);
Glauber Costa2633d7a2012-12-18 14:22:34 -08005566
Christoph Lameter81819f02007-05-06 14:49:36 -07005567 BUG_ON(p > name + ID_STR_LENGTH - 1);
5568 return name;
5569}
5570
5571static int sysfs_slab_add(struct kmem_cache *s)
5572{
5573 int err;
5574 const char *name;
Christoph Lameter45530c42012-11-28 16:23:07 +00005575 int unmergeable = slab_unmergeable(s);
Christoph Lameter81819f02007-05-06 14:49:36 -07005576
Christoph Lameter81819f02007-05-06 14:49:36 -07005577 if (unmergeable) {
5578 /*
5579 * Slabcache can never be merged so we can use the name proper.
5580 * This is typically the case for debug situations. In that
5581 * case we can catch duplicate names easily.
5582 */
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06005583 sysfs_remove_link(&slab_kset->kobj, s->name);
Christoph Lameter81819f02007-05-06 14:49:36 -07005584 name = s->name;
5585 } else {
5586 /*
5587 * Create a unique name for the slab as a target
5588 * for the symlinks.
5589 */
5590 name = create_unique_id(s);
5591 }
5592
Vladimir Davydov9a417072014-04-07 15:39:31 -07005593 s->kobj.kset = cache_kset(s);
Tetsuo Handa26e4f202014-01-04 16:32:31 +09005594 err = kobject_init_and_add(&s->kobj, &slab_ktype, NULL, "%s", name);
Dave Jones54b6a732014-04-07 15:39:32 -07005595 if (err)
Konstantin Khlebnikov80da0262015-09-04 15:45:51 -07005596 goto out;
Christoph Lameter81819f02007-05-06 14:49:36 -07005597
5598 err = sysfs_create_group(&s->kobj, &slab_attr_group);
Dave Jones54b6a732014-04-07 15:39:32 -07005599 if (err)
5600 goto out_del_kobj;
Vladimir Davydov9a417072014-04-07 15:39:31 -07005601
Johannes Weiner127424c2016-01-20 15:02:32 -08005602#ifdef CONFIG_MEMCG
Vladimir Davydov9a417072014-04-07 15:39:31 -07005603 if (is_root_cache(s)) {
5604 s->memcg_kset = kset_create_and_add("cgroup", NULL, &s->kobj);
5605 if (!s->memcg_kset) {
Dave Jones54b6a732014-04-07 15:39:32 -07005606 err = -ENOMEM;
5607 goto out_del_kobj;
Vladimir Davydov9a417072014-04-07 15:39:31 -07005608 }
5609 }
5610#endif
5611
Christoph Lameter81819f02007-05-06 14:49:36 -07005612 kobject_uevent(&s->kobj, KOBJ_ADD);
5613 if (!unmergeable) {
5614 /* Setup first alias */
5615 sysfs_slab_alias(s, s->name);
Christoph Lameter81819f02007-05-06 14:49:36 -07005616 }
Dave Jones54b6a732014-04-07 15:39:32 -07005617out:
5618 if (!unmergeable)
5619 kfree(name);
5620 return err;
5621out_del_kobj:
5622 kobject_del(&s->kobj);
Dave Jones54b6a732014-04-07 15:39:32 -07005623 goto out;
Christoph Lameter81819f02007-05-06 14:49:36 -07005624}
5625
Christoph Lameter41a21282014-05-06 12:50:08 -07005626void sysfs_slab_remove(struct kmem_cache *s)
Christoph Lameter81819f02007-05-06 14:49:36 -07005627{
Christoph Lameter97d06602012-07-06 15:25:11 -05005628 if (slab_state < FULL)
Christoph Lameter2bce6482010-07-19 11:39:11 -05005629 /*
5630 * Sysfs has not been setup yet so no need to remove the
5631 * cache from sysfs.
5632 */
5633 return;
5634
Johannes Weiner127424c2016-01-20 15:02:32 -08005635#ifdef CONFIG_MEMCG
Vladimir Davydov9a417072014-04-07 15:39:31 -07005636 kset_unregister(s->memcg_kset);
5637#endif
Christoph Lameter81819f02007-05-06 14:49:36 -07005638 kobject_uevent(&s->kobj, KOBJ_REMOVE);
5639 kobject_del(&s->kobj);
Christoph Lameter151c6022008-01-07 22:29:05 -08005640 kobject_put(&s->kobj);
Christoph Lameter81819f02007-05-06 14:49:36 -07005641}
5642
5643/*
5644 * Need to buffer aliases during bootup until sysfs becomes
Nick Andrew9f6c708e2008-12-05 14:08:08 +11005645 * available lest we lose that information.
Christoph Lameter81819f02007-05-06 14:49:36 -07005646 */
5647struct saved_alias {
5648 struct kmem_cache *s;
5649 const char *name;
5650 struct saved_alias *next;
5651};
5652
Adrian Bunk5af328a2007-07-17 04:03:27 -07005653static struct saved_alias *alias_list;
Christoph Lameter81819f02007-05-06 14:49:36 -07005654
5655static int sysfs_slab_alias(struct kmem_cache *s, const char *name)
5656{
5657 struct saved_alias *al;
5658
Christoph Lameter97d06602012-07-06 15:25:11 -05005659 if (slab_state == FULL) {
Christoph Lameter81819f02007-05-06 14:49:36 -07005660 /*
5661 * If we have a leftover link then remove it.
5662 */
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06005663 sysfs_remove_link(&slab_kset->kobj, name);
5664 return sysfs_create_link(&slab_kset->kobj, &s->kobj, name);
Christoph Lameter81819f02007-05-06 14:49:36 -07005665 }
5666
5667 al = kmalloc(sizeof(struct saved_alias), GFP_KERNEL);
5668 if (!al)
5669 return -ENOMEM;
5670
5671 al->s = s;
5672 al->name = name;
5673 al->next = alias_list;
5674 alias_list = al;
5675 return 0;
5676}
5677
5678static int __init slab_sysfs_init(void)
5679{
Christoph Lameter5b95a4ac2007-07-17 04:03:19 -07005680 struct kmem_cache *s;
Christoph Lameter81819f02007-05-06 14:49:36 -07005681 int err;
5682
Christoph Lameter18004c52012-07-06 15:25:12 -05005683 mutex_lock(&slab_mutex);
Christoph Lameter2bce6482010-07-19 11:39:11 -05005684
Greg Kroah-Hartman0ff21e42007-11-06 10:36:58 -08005685 slab_kset = kset_create_and_add("slab", &slab_uevent_ops, kernel_kobj);
Greg Kroah-Hartman27c3a312007-11-01 09:29:06 -06005686 if (!slab_kset) {
Christoph Lameter18004c52012-07-06 15:25:12 -05005687 mutex_unlock(&slab_mutex);
Fabian Frederickf9f58282014-06-04 16:06:34 -07005688 pr_err("Cannot register slab subsystem.\n");
Christoph Lameter81819f02007-05-06 14:49:36 -07005689 return -ENOSYS;
5690 }
5691
Christoph Lameter97d06602012-07-06 15:25:11 -05005692 slab_state = FULL;
Christoph Lameter26a7bd02007-05-09 02:32:39 -07005693
Christoph Lameter5b95a4ac2007-07-17 04:03:19 -07005694 list_for_each_entry(s, &slab_caches, list) {
Christoph Lameter26a7bd02007-05-09 02:32:39 -07005695 err = sysfs_slab_add(s);
Christoph Lameter5d540fb2007-08-30 23:56:26 -07005696 if (err)
Fabian Frederickf9f58282014-06-04 16:06:34 -07005697 pr_err("SLUB: Unable to add boot slab %s to sysfs\n",
5698 s->name);
Christoph Lameter26a7bd02007-05-09 02:32:39 -07005699 }
Christoph Lameter81819f02007-05-06 14:49:36 -07005700
5701 while (alias_list) {
5702 struct saved_alias *al = alias_list;
5703
5704 alias_list = alias_list->next;
5705 err = sysfs_slab_alias(al->s, al->name);
Christoph Lameter5d540fb2007-08-30 23:56:26 -07005706 if (err)
Fabian Frederickf9f58282014-06-04 16:06:34 -07005707 pr_err("SLUB: Unable to add boot slab alias %s to sysfs\n",
5708 al->name);
Christoph Lameter81819f02007-05-06 14:49:36 -07005709 kfree(al);
5710 }
5711
Christoph Lameter18004c52012-07-06 15:25:12 -05005712 mutex_unlock(&slab_mutex);
Christoph Lameter81819f02007-05-06 14:49:36 -07005713 resiliency_test();
5714 return 0;
5715}
5716
5717__initcall(slab_sysfs_init);
Christoph Lameterab4d5ed2010-10-05 13:57:26 -05005718#endif /* CONFIG_SYSFS */
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01005719
5720/*
5721 * The /proc/slabinfo ABI
5722 */
Linus Torvalds158a9622008-01-02 13:04:48 -08005723#ifdef CONFIG_SLABINFO
Glauber Costa0d7561c2012-10-19 18:20:27 +04005724void get_slabinfo(struct kmem_cache *s, struct slabinfo *sinfo)
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01005725{
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01005726 unsigned long nr_slabs = 0;
Christoph Lameter205ab992008-04-14 19:11:40 +03005727 unsigned long nr_objs = 0;
5728 unsigned long nr_free = 0;
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01005729 int node;
Christoph Lameterfa45dc22014-08-06 16:04:09 -07005730 struct kmem_cache_node *n;
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01005731
Christoph Lameterfa45dc22014-08-06 16:04:09 -07005732 for_each_kmem_cache_node(s, node, n) {
Wanpeng Lic17fd132013-07-04 08:33:26 +08005733 nr_slabs += node_nr_slabs(n);
5734 nr_objs += node_nr_objs(n);
Christoph Lameter205ab992008-04-14 19:11:40 +03005735 nr_free += count_partial(n, count_free);
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01005736 }
5737
Glauber Costa0d7561c2012-10-19 18:20:27 +04005738 sinfo->active_objs = nr_objs - nr_free;
5739 sinfo->num_objs = nr_objs;
5740 sinfo->active_slabs = nr_slabs;
5741 sinfo->num_slabs = nr_slabs;
5742 sinfo->objects_per_slab = oo_objects(s->oo);
5743 sinfo->cache_order = oo_order(s->oo);
Pekka J Enberg57ed3ed2008-01-01 17:23:28 +01005744}
5745
Glauber Costa0d7561c2012-10-19 18:20:27 +04005746void slabinfo_show_stats(struct seq_file *m, struct kmem_cache *s)
Alexey Dobriyan7b3c3a52008-10-06 02:42:17 +04005747{
Alexey Dobriyan7b3c3a52008-10-06 02:42:17 +04005748}
5749
Glauber Costab7454ad2012-10-19 18:20:25 +04005750ssize_t slabinfo_write(struct file *file, const char __user *buffer,
5751 size_t count, loff_t *ppos)
Alexey Dobriyan7b3c3a52008-10-06 02:42:17 +04005752{
Glauber Costab7454ad2012-10-19 18:20:25 +04005753 return -EIO;
Alexey Dobriyan7b3c3a52008-10-06 02:42:17 +04005754}
Linus Torvalds158a9622008-01-02 13:04:48 -08005755#endif /* CONFIG_SLABINFO */