2 * SLUB: A slab allocator that limits cache line use instead of queuing
3 * objects in per cpu and per node lists.
5 * The allocator synchronizes using per slab locks or atomic operatios
6 * and only uses a centralized lock to manage a pool of partial slabs.
8 * (C) 2007 SGI, Christoph Lameter
9 * (C) 2011 Linux Foundation, Christoph Lameter
13 #include <linux/swap.h> /* struct reclaim_state */
14 #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>
20 #include <linux/proc_fs.h>
21 #include <linux/notifier.h>
22 #include <linux/seq_file.h>
23 #include <linux/kmemcheck.h>
24 #include <linux/cpu.h>
25 #include <linux/cpuset.h>
26 #include <linux/mempolicy.h>
27 #include <linux/ctype.h>
28 #include <linux/debugobjects.h>
29 #include <linux/kallsyms.h>
30 #include <linux/memory.h>
31 #include <linux/math64.h>
32 #include <linux/fault-inject.h>
33 #include <linux/stacktrace.h>
34 #include <linux/prefetch.h>
35 #include <linux/memcontrol.h>
37 #include <trace/events/kmem.h>
43 * 1. slab_mutex (Global Mutex)
45 * 3. slab_lock(page) (Only on some arches and for debugging)
49 * The role of the slab_mutex is to protect the list of all the slabs
50 * and to synchronize major metadata changes to slab cache structures.
52 * The slab_lock is only used for debugging and on arches that do not
53 * have the ability to do a cmpxchg_double. It only protects the second
54 * double word in the page struct. Meaning
55 * A. page->freelist -> List of object free in a page
56 * B. page->counters -> Counters of objects
57 * C. page->frozen -> frozen state
59 * If a slab is frozen then it is exempt from list management. It is not
60 * on any list. The processor that froze the slab is the one who can
61 * perform list operations on the page. Other processors may put objects
62 * onto the freelist but the processor that froze the slab is the only
63 * one that can retrieve the objects from the page's freelist.
65 * The list_lock protects the partial and full list on each node and
66 * the partial slab counter. If taken then no new slabs may be added or
67 * removed from the lists nor make the number of partial slabs be modified.
68 * (Note that the total number of slabs is an atomic value that may be
69 * modified without taking the list lock).
71 * The list_lock is a centralized lock and thus we avoid taking it as
72 * much as possible. As long as SLUB does not have to handle partial
73 * slabs, operations can continue without any centralized lock. F.e.
74 * allocating a long series of objects that fill up slabs does not require
76 * Interrupts are disabled during allocation and deallocation in order to
77 * make the slab allocator safe to use in the context of an irq. In addition
78 * interrupts are disabled to ensure that the processor does not change
79 * while handling per_cpu slabs, due to kernel preemption.
81 * SLUB assigns one slab for allocation to each processor.
82 * Allocations only occur from these slabs called cpu slabs.
84 * Slabs with free elements are kept on a partial list and during regular
85 * operations no list for full slabs is used. If an object in a full slab is
86 * freed then the slab will show up again on the partial lists.
87 * We track full slabs for debugging purposes though because otherwise we
88 * cannot scan all objects.
90 * Slabs are freed when they become empty. Teardown and setup is
91 * minimal so we rely on the page allocators per cpu caches for
92 * fast frees and allocs.
94 * Overloading of page flags that are otherwise used for LRU management.
96 * PageActive The slab is frozen and exempt from list processing.
97 * This means that the slab is dedicated to a purpose
98 * such as satisfying allocations for a specific
99 * processor. Objects may be freed in the slab while
100 * it is frozen but slab_free will then skip the usual
101 * list operations. It is up to the processor holding
102 * the slab to integrate the slab into the slab lists
103 * when the slab is no longer needed.
105 * One use of this flag is to mark slabs that are
106 * used for allocations. Then such a slab becomes a cpu
107 * slab. The cpu slab may be equipped with an additional
108 * freelist that allows lockless access to
109 * free objects in addition to the regular freelist
110 * that requires the slab lock.
112 * PageError Slab requires special handling due to debug
113 * options set. This moves slab handling out of
114 * the fast path and disables lockless freelists.
117 static inline int kmem_cache_debug(struct kmem_cache
*s
)
119 #ifdef CONFIG_SLUB_DEBUG
120 return unlikely(s
->flags
& SLAB_DEBUG_FLAGS
);
126 static inline bool kmem_cache_has_cpu_partial(struct kmem_cache
*s
)
128 #ifdef CONFIG_SLUB_CPU_PARTIAL
129 return !kmem_cache_debug(s
);
136 * Issues still to be resolved:
138 * - Support PAGE_ALLOC_DEBUG. Should be easy to do.
140 * - Variable sizing of the per node arrays
143 /* Enable to test recovery from slab corruption on boot */
144 #undef SLUB_RESILIENCY_TEST
146 /* Enable to log cmpxchg failures */
147 #undef SLUB_DEBUG_CMPXCHG
150 * Mininum number of partial slabs. These will be left on the partial
151 * lists even if they are empty. kmem_cache_shrink may reclaim them.
153 #define MIN_PARTIAL 5
156 * Maximum number of desirable partial slabs.
157 * The existence of more partial slabs makes kmem_cache_shrink
158 * sort the partial list by the number of objects in use.
160 #define MAX_PARTIAL 10
162 #define DEBUG_DEFAULT_FLAGS (SLAB_DEBUG_FREE | SLAB_RED_ZONE | \
163 SLAB_POISON | SLAB_STORE_USER)
166 * Debugging flags that require metadata to be stored in the slab. These get
167 * disabled when slub_debug=O is used and a cache's min order increases with
170 #define DEBUG_METADATA_FLAGS (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER)
173 #define OO_MASK ((1 << OO_SHIFT) - 1)
174 #define MAX_OBJS_PER_PAGE 32767 /* since page.objects is u15 */
176 /* Internal SLUB flags */
177 #define __OBJECT_POISON 0x80000000UL /* Poison object */
178 #define __CMPXCHG_DOUBLE 0x40000000UL /* Use cmpxchg_double */
181 static struct notifier_block slab_notifier
;
185 * Tracking user of a slab.
187 #define TRACK_ADDRS_COUNT 16
189 unsigned long addr
; /* Called from address */
190 #ifdef CONFIG_STACKTRACE
191 unsigned long addrs
[TRACK_ADDRS_COUNT
]; /* Called from address */
193 int cpu
; /* Was running on cpu */
194 int pid
; /* Pid context */
195 unsigned long when
; /* When did the operation occur */
198 enum track_item
{ TRACK_ALLOC
, TRACK_FREE
};
201 static int sysfs_slab_add(struct kmem_cache
*);
202 static int sysfs_slab_alias(struct kmem_cache
*, const char *);
203 static void memcg_propagate_slab_attrs(struct kmem_cache
*s
);
205 static inline int sysfs_slab_add(struct kmem_cache
*s
) { return 0; }
206 static inline int sysfs_slab_alias(struct kmem_cache
*s
, const char *p
)
208 static inline void memcg_propagate_slab_attrs(struct kmem_cache
*s
) { }
211 static inline void stat(const struct kmem_cache
*s
, enum stat_item si
)
213 #ifdef CONFIG_SLUB_STATS
215 * The rmw is racy on a preemptible kernel but this is acceptable, so
216 * avoid this_cpu_add()'s irq-disable overhead.
218 raw_cpu_inc(s
->cpu_slab
->stat
[si
]);
222 /********************************************************************
223 * Core slab cache functions
224 *******************************************************************/
226 /* Verify that a pointer has an address that is valid within a slab page */
227 static inline int check_valid_pointer(struct kmem_cache
*s
,
228 struct page
*page
, const void *object
)
235 base
= page_address(page
);
236 if (object
< base
|| object
>= base
+ page
->objects
* s
->size
||
237 (object
- base
) % s
->size
) {
244 static inline void *get_freepointer(struct kmem_cache
*s
, void *object
)
246 return *(void **)(object
+ s
->offset
);
249 static void prefetch_freepointer(const struct kmem_cache
*s
, void *object
)
251 prefetch(object
+ s
->offset
);
254 static inline void *get_freepointer_safe(struct kmem_cache
*s
, void *object
)
258 #ifdef CONFIG_DEBUG_PAGEALLOC
259 probe_kernel_read(&p
, (void **)(object
+ s
->offset
), sizeof(p
));
261 p
= get_freepointer(s
, object
);
266 static inline void set_freepointer(struct kmem_cache
*s
, void *object
, void *fp
)
268 *(void **)(object
+ s
->offset
) = fp
;
271 /* Loop over all objects in a slab */
272 #define for_each_object(__p, __s, __addr, __objects) \
273 for (__p = (__addr); __p < (__addr) + (__objects) * (__s)->size;\
276 #define for_each_object_idx(__p, __idx, __s, __addr, __objects) \
277 for (__p = (__addr), __idx = 1; __idx <= __objects;\
278 __p += (__s)->size, __idx++)
280 /* Determine object index from a given position */
281 static inline int slab_index(void *p
, struct kmem_cache
*s
, void *addr
)
283 return (p
- addr
) / s
->size
;
286 static inline size_t slab_ksize(const struct kmem_cache
*s
)
288 #ifdef CONFIG_SLUB_DEBUG
290 * Debugging requires use of the padding between object
291 * and whatever may come after it.
293 if (s
->flags
& (SLAB_RED_ZONE
| SLAB_POISON
))
294 return s
->object_size
;
298 * If we have the need to store the freelist pointer
299 * back there or track user information then we can
300 * only use the space before that information.
302 if (s
->flags
& (SLAB_DESTROY_BY_RCU
| SLAB_STORE_USER
))
305 * Else we can use all the padding etc for the allocation
310 static inline int order_objects(int order
, unsigned long size
, int reserved
)
312 return ((PAGE_SIZE
<< order
) - reserved
) / size
;
315 static inline struct kmem_cache_order_objects
oo_make(int order
,
316 unsigned long size
, int reserved
)
318 struct kmem_cache_order_objects x
= {
319 (order
<< OO_SHIFT
) + order_objects(order
, size
, reserved
)
325 static inline int oo_order(struct kmem_cache_order_objects x
)
327 return x
.x
>> OO_SHIFT
;
330 static inline int oo_objects(struct kmem_cache_order_objects x
)
332 return x
.x
& OO_MASK
;
336 * Per slab locking using the pagelock
338 static __always_inline
void slab_lock(struct page
*page
)
340 bit_spin_lock(PG_locked
, &page
->flags
);
343 static __always_inline
void slab_unlock(struct page
*page
)
345 __bit_spin_unlock(PG_locked
, &page
->flags
);
348 static inline void set_page_slub_counters(struct page
*page
, unsigned long counters_new
)
351 tmp
.counters
= counters_new
;
353 * page->counters can cover frozen/inuse/objects as well
354 * as page->_count. If we assign to ->counters directly
355 * we run the risk of losing updates to page->_count, so
356 * be careful and only assign to the fields we need.
358 page
->frozen
= tmp
.frozen
;
359 page
->inuse
= tmp
.inuse
;
360 page
->objects
= tmp
.objects
;
363 /* Interrupts must be disabled (for the fallback code to work right) */
364 static inline bool __cmpxchg_double_slab(struct kmem_cache
*s
, struct page
*page
,
365 void *freelist_old
, unsigned long counters_old
,
366 void *freelist_new
, unsigned long counters_new
,
369 VM_BUG_ON(!irqs_disabled());
370 #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \
371 defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
372 if (s
->flags
& __CMPXCHG_DOUBLE
) {
373 if (cmpxchg_double(&page
->freelist
, &page
->counters
,
374 freelist_old
, counters_old
,
375 freelist_new
, counters_new
))
381 if (page
->freelist
== freelist_old
&&
382 page
->counters
== counters_old
) {
383 page
->freelist
= freelist_new
;
384 set_page_slub_counters(page
, counters_new
);
392 stat(s
, CMPXCHG_DOUBLE_FAIL
);
394 #ifdef SLUB_DEBUG_CMPXCHG
395 pr_info("%s %s: cmpxchg double redo ", n
, s
->name
);
401 static inline bool cmpxchg_double_slab(struct kmem_cache
*s
, struct page
*page
,
402 void *freelist_old
, unsigned long counters_old
,
403 void *freelist_new
, unsigned long counters_new
,
406 #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \
407 defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
408 if (s
->flags
& __CMPXCHG_DOUBLE
) {
409 if (cmpxchg_double(&page
->freelist
, &page
->counters
,
410 freelist_old
, counters_old
,
411 freelist_new
, counters_new
))
418 local_irq_save(flags
);
420 if (page
->freelist
== freelist_old
&&
421 page
->counters
== counters_old
) {
422 page
->freelist
= freelist_new
;
423 set_page_slub_counters(page
, counters_new
);
425 local_irq_restore(flags
);
429 local_irq_restore(flags
);
433 stat(s
, CMPXCHG_DOUBLE_FAIL
);
435 #ifdef SLUB_DEBUG_CMPXCHG
436 pr_info("%s %s: cmpxchg double redo ", n
, s
->name
);
442 #ifdef CONFIG_SLUB_DEBUG
444 * Determine a map of object in use on a page.
446 * Node listlock must be held to guarantee that the page does
447 * not vanish from under us.
449 static void get_map(struct kmem_cache
*s
, struct page
*page
, unsigned long *map
)
452 void *addr
= page_address(page
);
454 for (p
= page
->freelist
; p
; p
= get_freepointer(s
, p
))
455 set_bit(slab_index(p
, s
, addr
), map
);
461 #ifdef CONFIG_SLUB_DEBUG_ON
462 static int slub_debug
= DEBUG_DEFAULT_FLAGS
;
464 static int slub_debug
;
467 static char *slub_debug_slabs
;
468 static int disable_higher_order_debug
;
473 static void print_section(char *text
, u8
*addr
, unsigned int length
)
475 print_hex_dump(KERN_ERR
, text
, DUMP_PREFIX_ADDRESS
, 16, 1, addr
,
479 static struct track
*get_track(struct kmem_cache
*s
, void *object
,
480 enum track_item alloc
)
485 p
= object
+ s
->offset
+ sizeof(void *);
487 p
= object
+ s
->inuse
;
492 static void set_track(struct kmem_cache
*s
, void *object
,
493 enum track_item alloc
, unsigned long addr
)
495 struct track
*p
= get_track(s
, object
, alloc
);
498 #ifdef CONFIG_STACKTRACE
499 struct stack_trace trace
;
502 trace
.nr_entries
= 0;
503 trace
.max_entries
= TRACK_ADDRS_COUNT
;
504 trace
.entries
= p
->addrs
;
506 save_stack_trace(&trace
);
508 /* See rant in lockdep.c */
509 if (trace
.nr_entries
!= 0 &&
510 trace
.entries
[trace
.nr_entries
- 1] == ULONG_MAX
)
513 for (i
= trace
.nr_entries
; i
< TRACK_ADDRS_COUNT
; i
++)
517 p
->cpu
= smp_processor_id();
518 p
->pid
= current
->pid
;
521 memset(p
, 0, sizeof(struct track
));
524 static void init_tracking(struct kmem_cache
*s
, void *object
)
526 if (!(s
->flags
& SLAB_STORE_USER
))
529 set_track(s
, object
, TRACK_FREE
, 0UL);
530 set_track(s
, object
, TRACK_ALLOC
, 0UL);
533 static void print_track(const char *s
, struct track
*t
)
538 pr_err("INFO: %s in %pS age=%lu cpu=%u pid=%d\n",
539 s
, (void *)t
->addr
, jiffies
- t
->when
, t
->cpu
, t
->pid
);
540 #ifdef CONFIG_STACKTRACE
543 for (i
= 0; i
< TRACK_ADDRS_COUNT
; i
++)
545 pr_err("\t%pS\n", (void *)t
->addrs
[i
]);
552 static void print_tracking(struct kmem_cache
*s
, void *object
)
554 if (!(s
->flags
& SLAB_STORE_USER
))
557 print_track("Allocated", get_track(s
, object
, TRACK_ALLOC
));
558 print_track("Freed", get_track(s
, object
, TRACK_FREE
));
561 static void print_page_info(struct page
*page
)
563 pr_err("INFO: Slab 0x%p objects=%u used=%u fp=0x%p flags=0x%04lx\n",
564 page
, page
->objects
, page
->inuse
, page
->freelist
, page
->flags
);
568 static void slab_bug(struct kmem_cache
*s
, char *fmt
, ...)
570 struct va_format vaf
;
576 pr_err("=============================================================================\n");
577 pr_err("BUG %s (%s): %pV\n", s
->name
, print_tainted(), &vaf
);
578 pr_err("-----------------------------------------------------------------------------\n\n");
580 add_taint(TAINT_BAD_PAGE
, LOCKDEP_NOW_UNRELIABLE
);
584 static void slab_fix(struct kmem_cache
*s
, char *fmt
, ...)
586 struct va_format vaf
;
592 pr_err("FIX %s: %pV\n", s
->name
, &vaf
);
596 static void print_trailer(struct kmem_cache
*s
, struct page
*page
, u8
*p
)
598 unsigned int off
; /* Offset of last byte */
599 u8
*addr
= page_address(page
);
601 print_tracking(s
, p
);
603 print_page_info(page
);
605 pr_err("INFO: Object 0x%p @offset=%tu fp=0x%p\n\n",
606 p
, p
- addr
, get_freepointer(s
, p
));
609 print_section("Bytes b4 ", p
- 16, 16);
611 print_section("Object ", p
, min_t(unsigned long, s
->object_size
,
613 if (s
->flags
& SLAB_RED_ZONE
)
614 print_section("Redzone ", p
+ s
->object_size
,
615 s
->inuse
- s
->object_size
);
618 off
= s
->offset
+ sizeof(void *);
622 if (s
->flags
& SLAB_STORE_USER
)
623 off
+= 2 * sizeof(struct track
);
626 /* Beginning of the filler is the free pointer */
627 print_section("Padding ", p
+ off
, s
->size
- off
);
632 static void object_err(struct kmem_cache
*s
, struct page
*page
,
633 u8
*object
, char *reason
)
635 slab_bug(s
, "%s", reason
);
636 print_trailer(s
, page
, object
);
639 static void slab_err(struct kmem_cache
*s
, struct page
*page
,
640 const char *fmt
, ...)
646 vsnprintf(buf
, sizeof(buf
), fmt
, args
);
648 slab_bug(s
, "%s", buf
);
649 print_page_info(page
);
653 static void init_object(struct kmem_cache
*s
, void *object
, u8 val
)
657 if (s
->flags
& __OBJECT_POISON
) {
658 memset(p
, POISON_FREE
, s
->object_size
- 1);
659 p
[s
->object_size
- 1] = POISON_END
;
662 if (s
->flags
& SLAB_RED_ZONE
)
663 memset(p
+ s
->object_size
, val
, s
->inuse
- s
->object_size
);
666 static void restore_bytes(struct kmem_cache
*s
, char *message
, u8 data
,
667 void *from
, void *to
)
669 slab_fix(s
, "Restoring 0x%p-0x%p=0x%x\n", from
, to
- 1, data
);
670 memset(from
, data
, to
- from
);
673 static int check_bytes_and_report(struct kmem_cache
*s
, struct page
*page
,
674 u8
*object
, char *what
,
675 u8
*start
, unsigned int value
, unsigned int bytes
)
680 fault
= memchr_inv(start
, value
, bytes
);
685 while (end
> fault
&& end
[-1] == value
)
688 slab_bug(s
, "%s overwritten", what
);
689 pr_err("INFO: 0x%p-0x%p. First byte 0x%x instead of 0x%x\n",
690 fault
, end
- 1, fault
[0], value
);
691 print_trailer(s
, page
, object
);
693 restore_bytes(s
, what
, value
, fault
, end
);
701 * Bytes of the object to be managed.
702 * If the freepointer may overlay the object then the free
703 * pointer is the first word of the object.
705 * Poisoning uses 0x6b (POISON_FREE) and the last byte is
708 * object + s->object_size
709 * Padding to reach word boundary. This is also used for Redzoning.
710 * Padding is extended by another word if Redzoning is enabled and
711 * object_size == inuse.
713 * We fill with 0xbb (RED_INACTIVE) for inactive objects and with
714 * 0xcc (RED_ACTIVE) for objects in use.
717 * Meta data starts here.
719 * A. Free pointer (if we cannot overwrite object on free)
720 * B. Tracking data for SLAB_STORE_USER
721 * C. Padding to reach required alignment boundary or at mininum
722 * one word if debugging is on to be able to detect writes
723 * before the word boundary.
725 * Padding is done using 0x5a (POISON_INUSE)
728 * Nothing is used beyond s->size.
730 * If slabcaches are merged then the object_size and inuse boundaries are mostly
731 * ignored. And therefore no slab options that rely on these boundaries
732 * may be used with merged slabcaches.
735 static int check_pad_bytes(struct kmem_cache
*s
, struct page
*page
, u8
*p
)
737 unsigned long off
= s
->inuse
; /* The end of info */
740 /* Freepointer is placed after the object. */
741 off
+= sizeof(void *);
743 if (s
->flags
& SLAB_STORE_USER
)
744 /* We also have user information there */
745 off
+= 2 * sizeof(struct track
);
750 return check_bytes_and_report(s
, page
, p
, "Object padding",
751 p
+ off
, POISON_INUSE
, s
->size
- off
);
754 /* Check the pad bytes at the end of a slab page */
755 static int slab_pad_check(struct kmem_cache
*s
, struct page
*page
)
763 if (!(s
->flags
& SLAB_POISON
))
766 start
= page_address(page
);
767 length
= (PAGE_SIZE
<< compound_order(page
)) - s
->reserved
;
768 end
= start
+ length
;
769 remainder
= length
% s
->size
;
773 fault
= memchr_inv(end
- remainder
, POISON_INUSE
, remainder
);
776 while (end
> fault
&& end
[-1] == POISON_INUSE
)
779 slab_err(s
, page
, "Padding overwritten. 0x%p-0x%p", fault
, end
- 1);
780 print_section("Padding ", end
- remainder
, remainder
);
782 restore_bytes(s
, "slab padding", POISON_INUSE
, end
- remainder
, end
);
786 static int check_object(struct kmem_cache
*s
, struct page
*page
,
787 void *object
, u8 val
)
790 u8
*endobject
= object
+ s
->object_size
;
792 if (s
->flags
& SLAB_RED_ZONE
) {
793 if (!check_bytes_and_report(s
, page
, object
, "Redzone",
794 endobject
, val
, s
->inuse
- s
->object_size
))
797 if ((s
->flags
& SLAB_POISON
) && s
->object_size
< s
->inuse
) {
798 check_bytes_and_report(s
, page
, p
, "Alignment padding",
799 endobject
, POISON_INUSE
,
800 s
->inuse
- s
->object_size
);
804 if (s
->flags
& SLAB_POISON
) {
805 if (val
!= SLUB_RED_ACTIVE
&& (s
->flags
& __OBJECT_POISON
) &&
806 (!check_bytes_and_report(s
, page
, p
, "Poison", p
,
807 POISON_FREE
, s
->object_size
- 1) ||
808 !check_bytes_and_report(s
, page
, p
, "Poison",
809 p
+ s
->object_size
- 1, POISON_END
, 1)))
812 * check_pad_bytes cleans up on its own.
814 check_pad_bytes(s
, page
, p
);
817 if (!s
->offset
&& val
== SLUB_RED_ACTIVE
)
819 * Object and freepointer overlap. Cannot check
820 * freepointer while object is allocated.
824 /* Check free pointer validity */
825 if (!check_valid_pointer(s
, page
, get_freepointer(s
, p
))) {
826 object_err(s
, page
, p
, "Freepointer corrupt");
828 * No choice but to zap it and thus lose the remainder
829 * of the free objects in this slab. May cause
830 * another error because the object count is now wrong.
832 set_freepointer(s
, p
, NULL
);
838 static int check_slab(struct kmem_cache
*s
, struct page
*page
)
842 VM_BUG_ON(!irqs_disabled());
844 if (!PageSlab(page
)) {
845 slab_err(s
, page
, "Not a valid slab page");
849 maxobj
= order_objects(compound_order(page
), s
->size
, s
->reserved
);
850 if (page
->objects
> maxobj
) {
851 slab_err(s
, page
, "objects %u > max %u",
852 page
->objects
, maxobj
);
855 if (page
->inuse
> page
->objects
) {
856 slab_err(s
, page
, "inuse %u > max %u",
857 page
->inuse
, page
->objects
);
860 /* Slab_pad_check fixes things up after itself */
861 slab_pad_check(s
, page
);
866 * Determine if a certain object on a page is on the freelist. Must hold the
867 * slab lock to guarantee that the chains are in a consistent state.
869 static int on_freelist(struct kmem_cache
*s
, struct page
*page
, void *search
)
877 while (fp
&& nr
<= page
->objects
) {
880 if (!check_valid_pointer(s
, page
, fp
)) {
882 object_err(s
, page
, object
,
883 "Freechain corrupt");
884 set_freepointer(s
, object
, NULL
);
886 slab_err(s
, page
, "Freepointer corrupt");
887 page
->freelist
= NULL
;
888 page
->inuse
= page
->objects
;
889 slab_fix(s
, "Freelist cleared");
895 fp
= get_freepointer(s
, object
);
899 max_objects
= order_objects(compound_order(page
), s
->size
, s
->reserved
);
900 if (max_objects
> MAX_OBJS_PER_PAGE
)
901 max_objects
= MAX_OBJS_PER_PAGE
;
903 if (page
->objects
!= max_objects
) {
904 slab_err(s
, page
, "Wrong number of objects. Found %d but "
905 "should be %d", page
->objects
, max_objects
);
906 page
->objects
= max_objects
;
907 slab_fix(s
, "Number of objects adjusted.");
909 if (page
->inuse
!= page
->objects
- nr
) {
910 slab_err(s
, page
, "Wrong object count. Counter is %d but "
911 "counted were %d", page
->inuse
, page
->objects
- nr
);
912 page
->inuse
= page
->objects
- nr
;
913 slab_fix(s
, "Object count adjusted.");
915 return search
== NULL
;
918 static void trace(struct kmem_cache
*s
, struct page
*page
, void *object
,
921 if (s
->flags
& SLAB_TRACE
) {
922 pr_info("TRACE %s %s 0x%p inuse=%d fp=0x%p\n",
924 alloc
? "alloc" : "free",
929 print_section("Object ", (void *)object
,
937 * Tracking of fully allocated slabs for debugging purposes.
939 static void add_full(struct kmem_cache
*s
,
940 struct kmem_cache_node
*n
, struct page
*page
)
942 if (!(s
->flags
& SLAB_STORE_USER
))
945 lockdep_assert_held(&n
->list_lock
);
946 list_add(&page
->lru
, &n
->full
);
949 static void remove_full(struct kmem_cache
*s
, struct kmem_cache_node
*n
, struct page
*page
)
951 if (!(s
->flags
& SLAB_STORE_USER
))
954 lockdep_assert_held(&n
->list_lock
);
955 list_del(&page
->lru
);
958 /* Tracking of the number of slabs for debugging purposes */
959 static inline unsigned long slabs_node(struct kmem_cache
*s
, int node
)
961 struct kmem_cache_node
*n
= get_node(s
, node
);
963 return atomic_long_read(&n
->nr_slabs
);
966 static inline unsigned long node_nr_slabs(struct kmem_cache_node
*n
)
968 return atomic_long_read(&n
->nr_slabs
);
971 static inline void inc_slabs_node(struct kmem_cache
*s
, int node
, int objects
)
973 struct kmem_cache_node
*n
= get_node(s
, node
);
976 * May be called early in order to allocate a slab for the
977 * kmem_cache_node structure. Solve the chicken-egg
978 * dilemma by deferring the increment of the count during
979 * bootstrap (see early_kmem_cache_node_alloc).
982 atomic_long_inc(&n
->nr_slabs
);
983 atomic_long_add(objects
, &n
->total_objects
);
986 static inline void dec_slabs_node(struct kmem_cache
*s
, int node
, int objects
)
988 struct kmem_cache_node
*n
= get_node(s
, node
);
990 atomic_long_dec(&n
->nr_slabs
);
991 atomic_long_sub(objects
, &n
->total_objects
);
994 /* Object debug checks for alloc/free paths */
995 static void setup_object_debug(struct kmem_cache
*s
, struct page
*page
,
998 if (!(s
->flags
& (SLAB_STORE_USER
|SLAB_RED_ZONE
|__OBJECT_POISON
)))
1001 init_object(s
, object
, SLUB_RED_INACTIVE
);
1002 init_tracking(s
, object
);
1005 static noinline
int alloc_debug_processing(struct kmem_cache
*s
,
1007 void *object
, unsigned long addr
)
1009 if (!check_slab(s
, page
))
1012 if (!check_valid_pointer(s
, page
, object
)) {
1013 object_err(s
, page
, object
, "Freelist Pointer check fails");
1017 if (!check_object(s
, page
, object
, SLUB_RED_INACTIVE
))
1020 /* Success perform special debug activities for allocs */
1021 if (s
->flags
& SLAB_STORE_USER
)
1022 set_track(s
, object
, TRACK_ALLOC
, addr
);
1023 trace(s
, page
, object
, 1);
1024 init_object(s
, object
, SLUB_RED_ACTIVE
);
1028 if (PageSlab(page
)) {
1030 * If this is a slab page then lets do the best we can
1031 * to avoid issues in the future. Marking all objects
1032 * as used avoids touching the remaining objects.
1034 slab_fix(s
, "Marking all objects used");
1035 page
->inuse
= page
->objects
;
1036 page
->freelist
= NULL
;
1041 static noinline
struct kmem_cache_node
*free_debug_processing(
1042 struct kmem_cache
*s
, struct page
*page
, void *object
,
1043 unsigned long addr
, unsigned long *flags
)
1045 struct kmem_cache_node
*n
= get_node(s
, page_to_nid(page
));
1047 spin_lock_irqsave(&n
->list_lock
, *flags
);
1050 if (!check_slab(s
, page
))
1053 if (!check_valid_pointer(s
, page
, object
)) {
1054 slab_err(s
, page
, "Invalid object pointer 0x%p", object
);
1058 if (on_freelist(s
, page
, object
)) {
1059 object_err(s
, page
, object
, "Object already free");
1063 if (!check_object(s
, page
, object
, SLUB_RED_ACTIVE
))
1066 if (unlikely(s
!= page
->slab_cache
)) {
1067 if (!PageSlab(page
)) {
1068 slab_err(s
, page
, "Attempt to free object(0x%p) "
1069 "outside of slab", object
);
1070 } else if (!page
->slab_cache
) {
1071 pr_err("SLUB <none>: no slab for object 0x%p.\n",
1075 object_err(s
, page
, object
,
1076 "page slab pointer corrupt.");
1080 if (s
->flags
& SLAB_STORE_USER
)
1081 set_track(s
, object
, TRACK_FREE
, addr
);
1082 trace(s
, page
, object
, 0);
1083 init_object(s
, object
, SLUB_RED_INACTIVE
);
1087 * Keep node_lock to preserve integrity
1088 * until the object is actually freed
1094 spin_unlock_irqrestore(&n
->list_lock
, *flags
);
1095 slab_fix(s
, "Object at 0x%p not freed", object
);
1099 static int __init
setup_slub_debug(char *str
)
1101 slub_debug
= DEBUG_DEFAULT_FLAGS
;
1102 if (*str
++ != '=' || !*str
)
1104 * No options specified. Switch on full debugging.
1110 * No options but restriction on slabs. This means full
1111 * debugging for slabs matching a pattern.
1115 if (tolower(*str
) == 'o') {
1117 * Avoid enabling debugging on caches if its minimum order
1118 * would increase as a result.
1120 disable_higher_order_debug
= 1;
1127 * Switch off all debugging measures.
1132 * Determine which debug features should be switched on
1134 for (; *str
&& *str
!= ','; str
++) {
1135 switch (tolower(*str
)) {
1137 slub_debug
|= SLAB_DEBUG_FREE
;
1140 slub_debug
|= SLAB_RED_ZONE
;
1143 slub_debug
|= SLAB_POISON
;
1146 slub_debug
|= SLAB_STORE_USER
;
1149 slub_debug
|= SLAB_TRACE
;
1152 slub_debug
|= SLAB_FAILSLAB
;
1155 pr_err("slub_debug option '%c' unknown. skipped\n",
1162 slub_debug_slabs
= str
+ 1;
1167 __setup("slub_debug", setup_slub_debug
);
1169 unsigned long kmem_cache_flags(unsigned long object_size
,
1170 unsigned long flags
, const char *name
,
1171 void (*ctor
)(void *))
1174 * Enable debugging if selected on the kernel commandline.
1176 if (slub_debug
&& (!slub_debug_slabs
|| (name
&&
1177 !strncmp(slub_debug_slabs
, name
, strlen(slub_debug_slabs
)))))
1178 flags
|= slub_debug
;
1183 static inline void setup_object_debug(struct kmem_cache
*s
,
1184 struct page
*page
, void *object
) {}
1186 static inline int alloc_debug_processing(struct kmem_cache
*s
,
1187 struct page
*page
, void *object
, unsigned long addr
) { return 0; }
1189 static inline struct kmem_cache_node
*free_debug_processing(
1190 struct kmem_cache
*s
, struct page
*page
, void *object
,
1191 unsigned long addr
, unsigned long *flags
) { return NULL
; }
1193 static inline int slab_pad_check(struct kmem_cache
*s
, struct page
*page
)
1195 static inline int check_object(struct kmem_cache
*s
, struct page
*page
,
1196 void *object
, u8 val
) { return 1; }
1197 static inline void add_full(struct kmem_cache
*s
, struct kmem_cache_node
*n
,
1198 struct page
*page
) {}
1199 static inline void remove_full(struct kmem_cache
*s
, struct kmem_cache_node
*n
,
1200 struct page
*page
) {}
1201 unsigned long kmem_cache_flags(unsigned long object_size
,
1202 unsigned long flags
, const char *name
,
1203 void (*ctor
)(void *))
1207 #define slub_debug 0
1209 #define disable_higher_order_debug 0
1211 static inline unsigned long slabs_node(struct kmem_cache
*s
, int node
)
1213 static inline unsigned long node_nr_slabs(struct kmem_cache_node
*n
)
1215 static inline void inc_slabs_node(struct kmem_cache
*s
, int node
,
1217 static inline void dec_slabs_node(struct kmem_cache
*s
, int node
,
1220 #endif /* CONFIG_SLUB_DEBUG */
1223 * Hooks for other subsystems that check memory allocations. In a typical
1224 * production configuration these hooks all should produce no code at all.
1226 static inline void kmalloc_large_node_hook(void *ptr
, size_t size
, gfp_t flags
)
1228 kmemleak_alloc(ptr
, size
, 1, flags
);
1231 static inline void kfree_hook(const void *x
)
1236 static inline struct kmem_cache
*slab_pre_alloc_hook(struct kmem_cache
*s
,
1239 flags
&= gfp_allowed_mask
;
1240 lockdep_trace_alloc(flags
);
1241 might_sleep_if(flags
& __GFP_WAIT
);
1243 if (should_failslab(s
->object_size
, flags
, s
->flags
))
1246 return memcg_kmem_get_cache(s
, flags
);
1249 static inline void slab_post_alloc_hook(struct kmem_cache
*s
,
1250 gfp_t flags
, void *object
)
1252 flags
&= gfp_allowed_mask
;
1253 kmemcheck_slab_alloc(s
, flags
, object
, slab_ksize(s
));
1254 kmemleak_alloc_recursive(object
, s
->object_size
, 1, s
->flags
, flags
);
1255 memcg_kmem_put_cache(s
);
1258 static inline void slab_free_hook(struct kmem_cache
*s
, void *x
)
1260 kmemleak_free_recursive(x
, s
->flags
);
1263 * Trouble is that we may no longer disable interrupts in the fast path
1264 * So in order to make the debug calls that expect irqs to be
1265 * disabled we need to disable interrupts temporarily.
1267 #if defined(CONFIG_KMEMCHECK) || defined(CONFIG_LOCKDEP)
1269 unsigned long flags
;
1271 local_irq_save(flags
);
1272 kmemcheck_slab_free(s
, x
, s
->object_size
);
1273 debug_check_no_locks_freed(x
, s
->object_size
);
1274 local_irq_restore(flags
);
1277 if (!(s
->flags
& SLAB_DEBUG_OBJECTS
))
1278 debug_check_no_obj_freed(x
, s
->object_size
);
1282 * Slab allocation and freeing
1284 static inline struct page
*alloc_slab_page(struct kmem_cache
*s
,
1285 gfp_t flags
, int node
, struct kmem_cache_order_objects oo
)
1288 int order
= oo_order(oo
);
1290 flags
|= __GFP_NOTRACK
;
1292 if (memcg_charge_slab(s
, flags
, order
))
1295 if (node
== NUMA_NO_NODE
)
1296 page
= alloc_pages(flags
, order
);
1298 page
= alloc_pages_exact_node(node
, flags
, order
);
1301 memcg_uncharge_slab(s
, order
);
1306 static struct page
*allocate_slab(struct kmem_cache
*s
, gfp_t flags
, int node
)
1309 struct kmem_cache_order_objects oo
= s
->oo
;
1312 flags
&= gfp_allowed_mask
;
1314 if (flags
& __GFP_WAIT
)
1317 flags
|= s
->allocflags
;
1320 * Let the initial higher-order allocation fail under memory pressure
1321 * so we fall-back to the minimum order allocation.
1323 alloc_gfp
= (flags
| __GFP_NOWARN
| __GFP_NORETRY
) & ~__GFP_NOFAIL
;
1325 page
= alloc_slab_page(s
, alloc_gfp
, node
, oo
);
1326 if (unlikely(!page
)) {
1330 * Allocation may have failed due to fragmentation.
1331 * Try a lower order alloc if possible
1333 page
= alloc_slab_page(s
, alloc_gfp
, node
, oo
);
1336 stat(s
, ORDER_FALLBACK
);
1339 if (kmemcheck_enabled
&& page
1340 && !(s
->flags
& (SLAB_NOTRACK
| DEBUG_DEFAULT_FLAGS
))) {
1341 int pages
= 1 << oo_order(oo
);
1343 kmemcheck_alloc_shadow(page
, oo_order(oo
), alloc_gfp
, node
);
1346 * Objects from caches that have a constructor don't get
1347 * cleared when they're allocated, so we need to do it here.
1350 kmemcheck_mark_uninitialized_pages(page
, pages
);
1352 kmemcheck_mark_unallocated_pages(page
, pages
);
1355 if (flags
& __GFP_WAIT
)
1356 local_irq_disable();
1360 page
->objects
= oo_objects(oo
);
1361 mod_zone_page_state(page_zone(page
),
1362 (s
->flags
& SLAB_RECLAIM_ACCOUNT
) ?
1363 NR_SLAB_RECLAIMABLE
: NR_SLAB_UNRECLAIMABLE
,
1369 static void setup_object(struct kmem_cache
*s
, struct page
*page
,
1372 setup_object_debug(s
, page
, object
);
1373 if (unlikely(s
->ctor
))
1377 static struct page
*new_slab(struct kmem_cache
*s
, gfp_t flags
, int node
)
1385 if (unlikely(flags
& GFP_SLAB_BUG_MASK
)) {
1386 pr_emerg("gfp: %u\n", flags
& GFP_SLAB_BUG_MASK
);
1390 page
= allocate_slab(s
,
1391 flags
& (GFP_RECLAIM_MASK
| GFP_CONSTRAINT_MASK
), node
);
1395 order
= compound_order(page
);
1396 inc_slabs_node(s
, page_to_nid(page
), page
->objects
);
1397 page
->slab_cache
= s
;
1398 __SetPageSlab(page
);
1399 if (page
->pfmemalloc
)
1400 SetPageSlabPfmemalloc(page
);
1402 start
= page_address(page
);
1404 if (unlikely(s
->flags
& SLAB_POISON
))
1405 memset(start
, POISON_INUSE
, PAGE_SIZE
<< order
);
1407 for_each_object_idx(p
, idx
, s
, start
, page
->objects
) {
1408 setup_object(s
, page
, p
);
1409 if (likely(idx
< page
->objects
))
1410 set_freepointer(s
, p
, p
+ s
->size
);
1412 set_freepointer(s
, p
, NULL
);
1415 page
->freelist
= start
;
1416 page
->inuse
= page
->objects
;
1422 static void __free_slab(struct kmem_cache
*s
, struct page
*page
)
1424 int order
= compound_order(page
);
1425 int pages
= 1 << order
;
1427 if (kmem_cache_debug(s
)) {
1430 slab_pad_check(s
, page
);
1431 for_each_object(p
, s
, page_address(page
),
1433 check_object(s
, page
, p
, SLUB_RED_INACTIVE
);
1436 kmemcheck_free_shadow(page
, compound_order(page
));
1438 mod_zone_page_state(page_zone(page
),
1439 (s
->flags
& SLAB_RECLAIM_ACCOUNT
) ?
1440 NR_SLAB_RECLAIMABLE
: NR_SLAB_UNRECLAIMABLE
,
1443 __ClearPageSlabPfmemalloc(page
);
1444 __ClearPageSlab(page
);
1446 page_mapcount_reset(page
);
1447 if (current
->reclaim_state
)
1448 current
->reclaim_state
->reclaimed_slab
+= pages
;
1449 __free_pages(page
, order
);
1450 memcg_uncharge_slab(s
, order
);
1453 #define need_reserve_slab_rcu \
1454 (sizeof(((struct page *)NULL)->lru) < sizeof(struct rcu_head))
1456 static void rcu_free_slab(struct rcu_head
*h
)
1460 if (need_reserve_slab_rcu
)
1461 page
= virt_to_head_page(h
);
1463 page
= container_of((struct list_head
*)h
, struct page
, lru
);
1465 __free_slab(page
->slab_cache
, page
);
1468 static void free_slab(struct kmem_cache
*s
, struct page
*page
)
1470 if (unlikely(s
->flags
& SLAB_DESTROY_BY_RCU
)) {
1471 struct rcu_head
*head
;
1473 if (need_reserve_slab_rcu
) {
1474 int order
= compound_order(page
);
1475 int offset
= (PAGE_SIZE
<< order
) - s
->reserved
;
1477 VM_BUG_ON(s
->reserved
!= sizeof(*head
));
1478 head
= page_address(page
) + offset
;
1481 * RCU free overloads the RCU head over the LRU
1483 head
= (void *)&page
->lru
;
1486 call_rcu(head
, rcu_free_slab
);
1488 __free_slab(s
, page
);
1491 static void discard_slab(struct kmem_cache
*s
, struct page
*page
)
1493 dec_slabs_node(s
, page_to_nid(page
), page
->objects
);
1498 * Management of partially allocated slabs.
1501 __add_partial(struct kmem_cache_node
*n
, struct page
*page
, int tail
)
1504 if (tail
== DEACTIVATE_TO_TAIL
)
1505 list_add_tail(&page
->lru
, &n
->partial
);
1507 list_add(&page
->lru
, &n
->partial
);
1510 static inline void add_partial(struct kmem_cache_node
*n
,
1511 struct page
*page
, int tail
)
1513 lockdep_assert_held(&n
->list_lock
);
1514 __add_partial(n
, page
, tail
);
1518 __remove_partial(struct kmem_cache_node
*n
, struct page
*page
)
1520 list_del(&page
->lru
);
1524 static inline void remove_partial(struct kmem_cache_node
*n
,
1527 lockdep_assert_held(&n
->list_lock
);
1528 __remove_partial(n
, page
);
1532 * Remove slab from the partial list, freeze it and
1533 * return the pointer to the freelist.
1535 * Returns a list of objects or NULL if it fails.
1537 static inline void *acquire_slab(struct kmem_cache
*s
,
1538 struct kmem_cache_node
*n
, struct page
*page
,
1539 int mode
, int *objects
)
1542 unsigned long counters
;
1545 lockdep_assert_held(&n
->list_lock
);
1548 * Zap the freelist and set the frozen bit.
1549 * The old freelist is the list of objects for the
1550 * per cpu allocation list.
1552 freelist
= page
->freelist
;
1553 counters
= page
->counters
;
1554 new.counters
= counters
;
1555 *objects
= new.objects
- new.inuse
;
1557 new.inuse
= page
->objects
;
1558 new.freelist
= NULL
;
1560 new.freelist
= freelist
;
1563 VM_BUG_ON(new.frozen
);
1566 if (!__cmpxchg_double_slab(s
, page
,
1568 new.freelist
, new.counters
,
1572 remove_partial(n
, page
);
1577 static void put_cpu_partial(struct kmem_cache
*s
, struct page
*page
, int drain
);
1578 static inline bool pfmemalloc_match(struct page
*page
, gfp_t gfpflags
);
1581 * Try to allocate a partial slab from a specific node.
1583 static void *get_partial_node(struct kmem_cache
*s
, struct kmem_cache_node
*n
,
1584 struct kmem_cache_cpu
*c
, gfp_t flags
)
1586 struct page
*page
, *page2
;
1587 void *object
= NULL
;
1592 * Racy check. If we mistakenly see no partial slabs then we
1593 * just allocate an empty slab. If we mistakenly try to get a
1594 * partial slab and there is none available then get_partials()
1597 if (!n
|| !n
->nr_partial
)
1600 spin_lock(&n
->list_lock
);
1601 list_for_each_entry_safe(page
, page2
, &n
->partial
, lru
) {
1604 if (!pfmemalloc_match(page
, flags
))
1607 t
= acquire_slab(s
, n
, page
, object
== NULL
, &objects
);
1611 available
+= objects
;
1614 stat(s
, ALLOC_FROM_PARTIAL
);
1617 put_cpu_partial(s
, page
, 0);
1618 stat(s
, CPU_PARTIAL_NODE
);
1620 if (!kmem_cache_has_cpu_partial(s
)
1621 || available
> s
->cpu_partial
/ 2)
1625 spin_unlock(&n
->list_lock
);
1630 * Get a page from somewhere. Search in increasing NUMA distances.
1632 static void *get_any_partial(struct kmem_cache
*s
, gfp_t flags
,
1633 struct kmem_cache_cpu
*c
)
1636 struct zonelist
*zonelist
;
1639 enum zone_type high_zoneidx
= gfp_zone(flags
);
1641 unsigned int cpuset_mems_cookie
;
1644 * The defrag ratio allows a configuration of the tradeoffs between
1645 * inter node defragmentation and node local allocations. A lower
1646 * defrag_ratio increases the tendency to do local allocations
1647 * instead of attempting to obtain partial slabs from other nodes.
1649 * If the defrag_ratio is set to 0 then kmalloc() always
1650 * returns node local objects. If the ratio is higher then kmalloc()
1651 * may return off node objects because partial slabs are obtained
1652 * from other nodes and filled up.
1654 * If /sys/kernel/slab/xx/defrag_ratio is set to 100 (which makes
1655 * defrag_ratio = 1000) then every (well almost) allocation will
1656 * first attempt to defrag slab caches on other nodes. This means
1657 * scanning over all nodes to look for partial slabs which may be
1658 * expensive if we do it every time we are trying to find a slab
1659 * with available objects.
1661 if (!s
->remote_node_defrag_ratio
||
1662 get_cycles() % 1024 > s
->remote_node_defrag_ratio
)
1666 cpuset_mems_cookie
= read_mems_allowed_begin();
1667 zonelist
= node_zonelist(mempolicy_slab_node(), flags
);
1668 for_each_zone_zonelist(zone
, z
, zonelist
, high_zoneidx
) {
1669 struct kmem_cache_node
*n
;
1671 n
= get_node(s
, zone_to_nid(zone
));
1673 if (n
&& cpuset_zone_allowed(zone
, flags
) &&
1674 n
->nr_partial
> s
->min_partial
) {
1675 object
= get_partial_node(s
, n
, c
, flags
);
1678 * Don't check read_mems_allowed_retry()
1679 * here - if mems_allowed was updated in
1680 * parallel, that was a harmless race
1681 * between allocation and the cpuset
1688 } while (read_mems_allowed_retry(cpuset_mems_cookie
));
1694 * Get a partial page, lock it and return it.
1696 static void *get_partial(struct kmem_cache
*s
, gfp_t flags
, int node
,
1697 struct kmem_cache_cpu
*c
)
1700 int searchnode
= node
;
1702 if (node
== NUMA_NO_NODE
)
1703 searchnode
= numa_mem_id();
1704 else if (!node_present_pages(node
))
1705 searchnode
= node_to_mem_node(node
);
1707 object
= get_partial_node(s
, get_node(s
, searchnode
), c
, flags
);
1708 if (object
|| node
!= NUMA_NO_NODE
)
1711 return get_any_partial(s
, flags
, c
);
1714 #ifdef CONFIG_PREEMPT
1716 * Calculate the next globally unique transaction for disambiguiation
1717 * during cmpxchg. The transactions start with the cpu number and are then
1718 * incremented by CONFIG_NR_CPUS.
1720 #define TID_STEP roundup_pow_of_two(CONFIG_NR_CPUS)
1723 * No preemption supported therefore also no need to check for
1729 static inline unsigned long next_tid(unsigned long tid
)
1731 return tid
+ TID_STEP
;
1734 static inline unsigned int tid_to_cpu(unsigned long tid
)
1736 return tid
% TID_STEP
;
1739 static inline unsigned long tid_to_event(unsigned long tid
)
1741 return tid
/ TID_STEP
;
1744 static inline unsigned int init_tid(int cpu
)
1749 static inline void note_cmpxchg_failure(const char *n
,
1750 const struct kmem_cache
*s
, unsigned long tid
)
1752 #ifdef SLUB_DEBUG_CMPXCHG
1753 unsigned long actual_tid
= __this_cpu_read(s
->cpu_slab
->tid
);
1755 pr_info("%s %s: cmpxchg redo ", n
, s
->name
);
1757 #ifdef CONFIG_PREEMPT
1758 if (tid_to_cpu(tid
) != tid_to_cpu(actual_tid
))
1759 pr_warn("due to cpu change %d -> %d\n",
1760 tid_to_cpu(tid
), tid_to_cpu(actual_tid
));
1763 if (tid_to_event(tid
) != tid_to_event(actual_tid
))
1764 pr_warn("due to cpu running other code. Event %ld->%ld\n",
1765 tid_to_event(tid
), tid_to_event(actual_tid
));
1767 pr_warn("for unknown reason: actual=%lx was=%lx target=%lx\n",
1768 actual_tid
, tid
, next_tid(tid
));
1770 stat(s
, CMPXCHG_DOUBLE_CPU_FAIL
);
1773 static void init_kmem_cache_cpus(struct kmem_cache
*s
)
1777 for_each_possible_cpu(cpu
)
1778 per_cpu_ptr(s
->cpu_slab
, cpu
)->tid
= init_tid(cpu
);
1782 * Remove the cpu slab
1784 static void deactivate_slab(struct kmem_cache
*s
, struct page
*page
,
1787 enum slab_modes
{ M_NONE
, M_PARTIAL
, M_FULL
, M_FREE
};
1788 struct kmem_cache_node
*n
= get_node(s
, page_to_nid(page
));
1790 enum slab_modes l
= M_NONE
, m
= M_NONE
;
1792 int tail
= DEACTIVATE_TO_HEAD
;
1796 if (page
->freelist
) {
1797 stat(s
, DEACTIVATE_REMOTE_FREES
);
1798 tail
= DEACTIVATE_TO_TAIL
;
1802 * Stage one: Free all available per cpu objects back
1803 * to the page freelist while it is still frozen. Leave the
1806 * There is no need to take the list->lock because the page
1809 while (freelist
&& (nextfree
= get_freepointer(s
, freelist
))) {
1811 unsigned long counters
;
1814 prior
= page
->freelist
;
1815 counters
= page
->counters
;
1816 set_freepointer(s
, freelist
, prior
);
1817 new.counters
= counters
;
1819 VM_BUG_ON(!new.frozen
);
1821 } while (!__cmpxchg_double_slab(s
, page
,
1823 freelist
, new.counters
,
1824 "drain percpu freelist"));
1826 freelist
= nextfree
;
1830 * Stage two: Ensure that the page is unfrozen while the
1831 * list presence reflects the actual number of objects
1834 * We setup the list membership and then perform a cmpxchg
1835 * with the count. If there is a mismatch then the page
1836 * is not unfrozen but the page is on the wrong list.
1838 * Then we restart the process which may have to remove
1839 * the page from the list that we just put it on again
1840 * because the number of objects in the slab may have
1845 old
.freelist
= page
->freelist
;
1846 old
.counters
= page
->counters
;
1847 VM_BUG_ON(!old
.frozen
);
1849 /* Determine target state of the slab */
1850 new.counters
= old
.counters
;
1853 set_freepointer(s
, freelist
, old
.freelist
);
1854 new.freelist
= freelist
;
1856 new.freelist
= old
.freelist
;
1860 if (!new.inuse
&& n
->nr_partial
>= s
->min_partial
)
1862 else if (new.freelist
) {
1867 * Taking the spinlock removes the possiblity
1868 * that acquire_slab() will see a slab page that
1871 spin_lock(&n
->list_lock
);
1875 if (kmem_cache_debug(s
) && !lock
) {
1878 * This also ensures that the scanning of full
1879 * slabs from diagnostic functions will not see
1882 spin_lock(&n
->list_lock
);
1890 remove_partial(n
, page
);
1892 else if (l
== M_FULL
)
1894 remove_full(s
, n
, page
);
1896 if (m
== M_PARTIAL
) {
1898 add_partial(n
, page
, tail
);
1901 } else if (m
== M_FULL
) {
1903 stat(s
, DEACTIVATE_FULL
);
1904 add_full(s
, n
, page
);
1910 if (!__cmpxchg_double_slab(s
, page
,
1911 old
.freelist
, old
.counters
,
1912 new.freelist
, new.counters
,
1917 spin_unlock(&n
->list_lock
);
1920 stat(s
, DEACTIVATE_EMPTY
);
1921 discard_slab(s
, page
);
1927 * Unfreeze all the cpu partial slabs.
1929 * This function must be called with interrupts disabled
1930 * for the cpu using c (or some other guarantee must be there
1931 * to guarantee no concurrent accesses).
1933 static void unfreeze_partials(struct kmem_cache
*s
,
1934 struct kmem_cache_cpu
*c
)
1936 #ifdef CONFIG_SLUB_CPU_PARTIAL
1937 struct kmem_cache_node
*n
= NULL
, *n2
= NULL
;
1938 struct page
*page
, *discard_page
= NULL
;
1940 while ((page
= c
->partial
)) {
1944 c
->partial
= page
->next
;
1946 n2
= get_node(s
, page_to_nid(page
));
1949 spin_unlock(&n
->list_lock
);
1952 spin_lock(&n
->list_lock
);
1957 old
.freelist
= page
->freelist
;
1958 old
.counters
= page
->counters
;
1959 VM_BUG_ON(!old
.frozen
);
1961 new.counters
= old
.counters
;
1962 new.freelist
= old
.freelist
;
1966 } while (!__cmpxchg_double_slab(s
, page
,
1967 old
.freelist
, old
.counters
,
1968 new.freelist
, new.counters
,
1969 "unfreezing slab"));
1971 if (unlikely(!new.inuse
&& n
->nr_partial
>= s
->min_partial
)) {
1972 page
->next
= discard_page
;
1973 discard_page
= page
;
1975 add_partial(n
, page
, DEACTIVATE_TO_TAIL
);
1976 stat(s
, FREE_ADD_PARTIAL
);
1981 spin_unlock(&n
->list_lock
);
1983 while (discard_page
) {
1984 page
= discard_page
;
1985 discard_page
= discard_page
->next
;
1987 stat(s
, DEACTIVATE_EMPTY
);
1988 discard_slab(s
, page
);
1995 * Put a page that was just frozen (in __slab_free) into a partial page
1996 * slot if available. This is done without interrupts disabled and without
1997 * preemption disabled. The cmpxchg is racy and may put the partial page
1998 * onto a random cpus partial slot.
2000 * If we did not find a slot then simply move all the partials to the
2001 * per node partial list.
2003 static void put_cpu_partial(struct kmem_cache
*s
, struct page
*page
, int drain
)
2005 #ifdef CONFIG_SLUB_CPU_PARTIAL
2006 struct page
*oldpage
;
2014 oldpage
= this_cpu_read(s
->cpu_slab
->partial
);
2017 pobjects
= oldpage
->pobjects
;
2018 pages
= oldpage
->pages
;
2019 if (drain
&& pobjects
> s
->cpu_partial
) {
2020 unsigned long flags
;
2022 * partial array is full. Move the existing
2023 * set to the per node partial list.
2025 local_irq_save(flags
);
2026 unfreeze_partials(s
, this_cpu_ptr(s
->cpu_slab
));
2027 local_irq_restore(flags
);
2031 stat(s
, CPU_PARTIAL_DRAIN
);
2036 pobjects
+= page
->objects
- page
->inuse
;
2038 page
->pages
= pages
;
2039 page
->pobjects
= pobjects
;
2040 page
->next
= oldpage
;
2042 } while (this_cpu_cmpxchg(s
->cpu_slab
->partial
, oldpage
, page
)
2044 if (unlikely(!s
->cpu_partial
)) {
2045 unsigned long flags
;
2047 local_irq_save(flags
);
2048 unfreeze_partials(s
, this_cpu_ptr(s
->cpu_slab
));
2049 local_irq_restore(flags
);
2055 static inline void flush_slab(struct kmem_cache
*s
, struct kmem_cache_cpu
*c
)
2057 stat(s
, CPUSLAB_FLUSH
);
2058 deactivate_slab(s
, c
->page
, c
->freelist
);
2060 c
->tid
= next_tid(c
->tid
);
2068 * Called from IPI handler with interrupts disabled.
2070 static inline void __flush_cpu_slab(struct kmem_cache
*s
, int cpu
)
2072 struct kmem_cache_cpu
*c
= per_cpu_ptr(s
->cpu_slab
, cpu
);
2078 unfreeze_partials(s
, c
);
2082 static void flush_cpu_slab(void *d
)
2084 struct kmem_cache
*s
= d
;
2086 __flush_cpu_slab(s
, smp_processor_id());
2089 static bool has_cpu_slab(int cpu
, void *info
)
2091 struct kmem_cache
*s
= info
;
2092 struct kmem_cache_cpu
*c
= per_cpu_ptr(s
->cpu_slab
, cpu
);
2094 return c
->page
|| c
->partial
;
2097 static void flush_all(struct kmem_cache
*s
)
2099 on_each_cpu_cond(has_cpu_slab
, flush_cpu_slab
, s
, 1, GFP_ATOMIC
);
2103 * Check if the objects in a per cpu structure fit numa
2104 * locality expectations.
2106 static inline int node_match(struct page
*page
, int node
)
2109 if (!page
|| (node
!= NUMA_NO_NODE
&& page_to_nid(page
) != node
))
2115 #ifdef CONFIG_SLUB_DEBUG
2116 static int count_free(struct page
*page
)
2118 return page
->objects
- page
->inuse
;
2121 static inline unsigned long node_nr_objs(struct kmem_cache_node
*n
)
2123 return atomic_long_read(&n
->total_objects
);
2125 #endif /* CONFIG_SLUB_DEBUG */
2127 #if defined(CONFIG_SLUB_DEBUG) || defined(CONFIG_SYSFS)
2128 static unsigned long count_partial(struct kmem_cache_node
*n
,
2129 int (*get_count
)(struct page
*))
2131 unsigned long flags
;
2132 unsigned long x
= 0;
2135 spin_lock_irqsave(&n
->list_lock
, flags
);
2136 list_for_each_entry(page
, &n
->partial
, lru
)
2137 x
+= get_count(page
);
2138 spin_unlock_irqrestore(&n
->list_lock
, flags
);
2141 #endif /* CONFIG_SLUB_DEBUG || CONFIG_SYSFS */
2143 static noinline
void
2144 slab_out_of_memory(struct kmem_cache
*s
, gfp_t gfpflags
, int nid
)
2146 #ifdef CONFIG_SLUB_DEBUG
2147 static DEFINE_RATELIMIT_STATE(slub_oom_rs
, DEFAULT_RATELIMIT_INTERVAL
,
2148 DEFAULT_RATELIMIT_BURST
);
2150 struct kmem_cache_node
*n
;
2152 if ((gfpflags
& __GFP_NOWARN
) || !__ratelimit(&slub_oom_rs
))
2155 pr_warn("SLUB: Unable to allocate memory on node %d (gfp=0x%x)\n",
2157 pr_warn(" cache: %s, object size: %d, buffer size: %d, default order: %d, min order: %d\n",
2158 s
->name
, s
->object_size
, s
->size
, oo_order(s
->oo
),
2161 if (oo_order(s
->min
) > get_order(s
->object_size
))
2162 pr_warn(" %s debugging increased min order, use slub_debug=O to disable.\n",
2165 for_each_kmem_cache_node(s
, node
, n
) {
2166 unsigned long nr_slabs
;
2167 unsigned long nr_objs
;
2168 unsigned long nr_free
;
2170 nr_free
= count_partial(n
, count_free
);
2171 nr_slabs
= node_nr_slabs(n
);
2172 nr_objs
= node_nr_objs(n
);
2174 pr_warn(" node %d: slabs: %ld, objs: %ld, free: %ld\n",
2175 node
, nr_slabs
, nr_objs
, nr_free
);
2180 static inline void *new_slab_objects(struct kmem_cache
*s
, gfp_t flags
,
2181 int node
, struct kmem_cache_cpu
**pc
)
2184 struct kmem_cache_cpu
*c
= *pc
;
2187 freelist
= get_partial(s
, flags
, node
, c
);
2192 page
= new_slab(s
, flags
, node
);
2194 c
= raw_cpu_ptr(s
->cpu_slab
);
2199 * No other reference to the page yet so we can
2200 * muck around with it freely without cmpxchg
2202 freelist
= page
->freelist
;
2203 page
->freelist
= NULL
;
2205 stat(s
, ALLOC_SLAB
);
2214 static inline bool pfmemalloc_match(struct page
*page
, gfp_t gfpflags
)
2216 if (unlikely(PageSlabPfmemalloc(page
)))
2217 return gfp_pfmemalloc_allowed(gfpflags
);
2223 * Check the page->freelist of a page and either transfer the freelist to the
2224 * per cpu freelist or deactivate the page.
2226 * The page is still frozen if the return value is not NULL.
2228 * If this function returns NULL then the page has been unfrozen.
2230 * This function must be called with interrupt disabled.
2232 static inline void *get_freelist(struct kmem_cache
*s
, struct page
*page
)
2235 unsigned long counters
;
2239 freelist
= page
->freelist
;
2240 counters
= page
->counters
;
2242 new.counters
= counters
;
2243 VM_BUG_ON(!new.frozen
);
2245 new.inuse
= page
->objects
;
2246 new.frozen
= freelist
!= NULL
;
2248 } while (!__cmpxchg_double_slab(s
, page
,
2257 * Slow path. The lockless freelist is empty or we need to perform
2260 * Processing is still very fast if new objects have been freed to the
2261 * regular freelist. In that case we simply take over the regular freelist
2262 * as the lockless freelist and zap the regular freelist.
2264 * If that is not working then we fall back to the partial lists. We take the
2265 * first element of the freelist as the object to allocate now and move the
2266 * rest of the freelist to the lockless freelist.
2268 * And if we were unable to get a new slab from the partial slab lists then
2269 * we need to allocate a new slab. This is the slowest path since it involves
2270 * a call to the page allocator and the setup of a new slab.
2272 static void *__slab_alloc(struct kmem_cache
*s
, gfp_t gfpflags
, int node
,
2273 unsigned long addr
, struct kmem_cache_cpu
*c
)
2277 unsigned long flags
;
2279 local_irq_save(flags
);
2280 #ifdef CONFIG_PREEMPT
2282 * We may have been preempted and rescheduled on a different
2283 * cpu before disabling interrupts. Need to reload cpu area
2286 c
= this_cpu_ptr(s
->cpu_slab
);
2294 if (unlikely(!node_match(page
, node
))) {
2295 int searchnode
= node
;
2297 if (node
!= NUMA_NO_NODE
&& !node_present_pages(node
))
2298 searchnode
= node_to_mem_node(node
);
2300 if (unlikely(!node_match(page
, searchnode
))) {
2301 stat(s
, ALLOC_NODE_MISMATCH
);
2302 deactivate_slab(s
, page
, c
->freelist
);
2310 * By rights, we should be searching for a slab page that was
2311 * PFMEMALLOC but right now, we are losing the pfmemalloc
2312 * information when the page leaves the per-cpu allocator
2314 if (unlikely(!pfmemalloc_match(page
, gfpflags
))) {
2315 deactivate_slab(s
, page
, c
->freelist
);
2321 /* must check again c->freelist in case of cpu migration or IRQ */
2322 freelist
= c
->freelist
;
2326 freelist
= get_freelist(s
, page
);
2330 stat(s
, DEACTIVATE_BYPASS
);
2334 stat(s
, ALLOC_REFILL
);
2338 * freelist is pointing to the list of objects to be used.
2339 * page is pointing to the page from which the objects are obtained.
2340 * That page must be frozen for per cpu allocations to work.
2342 VM_BUG_ON(!c
->page
->frozen
);
2343 c
->freelist
= get_freepointer(s
, freelist
);
2344 c
->tid
= next_tid(c
->tid
);
2345 local_irq_restore(flags
);
2351 page
= c
->page
= c
->partial
;
2352 c
->partial
= page
->next
;
2353 stat(s
, CPU_PARTIAL_ALLOC
);
2358 freelist
= new_slab_objects(s
, gfpflags
, node
, &c
);
2360 if (unlikely(!freelist
)) {
2361 slab_out_of_memory(s
, gfpflags
, node
);
2362 local_irq_restore(flags
);
2367 if (likely(!kmem_cache_debug(s
) && pfmemalloc_match(page
, gfpflags
)))
2370 /* Only entered in the debug case */
2371 if (kmem_cache_debug(s
) &&
2372 !alloc_debug_processing(s
, page
, freelist
, addr
))
2373 goto new_slab
; /* Slab failed checks. Next slab needed */
2375 deactivate_slab(s
, page
, get_freepointer(s
, freelist
));
2378 local_irq_restore(flags
);
2383 * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc)
2384 * have the fastpath folded into their functions. So no function call
2385 * overhead for requests that can be satisfied on the fastpath.
2387 * The fastpath works by first checking if the lockless freelist can be used.
2388 * If not then __slab_alloc is called for slow processing.
2390 * Otherwise we can simply pick the next object from the lockless free list.
2392 static __always_inline
void *slab_alloc_node(struct kmem_cache
*s
,
2393 gfp_t gfpflags
, int node
, unsigned long addr
)
2396 struct kmem_cache_cpu
*c
;
2400 s
= slab_pre_alloc_hook(s
, gfpflags
);
2405 * Must read kmem_cache cpu data via this cpu ptr. Preemption is
2406 * enabled. We may switch back and forth between cpus while
2407 * reading from one cpu area. That does not matter as long
2408 * as we end up on the original cpu again when doing the cmpxchg.
2410 * We should guarantee that tid and kmem_cache are retrieved on
2411 * the same cpu. It could be different if CONFIG_PREEMPT so we need
2412 * to check if it is matched or not.
2415 tid
= this_cpu_read(s
->cpu_slab
->tid
);
2416 c
= raw_cpu_ptr(s
->cpu_slab
);
2417 } while (IS_ENABLED(CONFIG_PREEMPT
) && unlikely(tid
!= c
->tid
));
2420 * Irqless object alloc/free algorithm used here depends on sequence
2421 * of fetching cpu_slab's data. tid should be fetched before anything
2422 * on c to guarantee that object and page associated with previous tid
2423 * won't be used with current tid. If we fetch tid first, object and
2424 * page could be one associated with next tid and our alloc/free
2425 * request will be failed. In this case, we will retry. So, no problem.
2430 * The transaction ids are globally unique per cpu and per operation on
2431 * a per cpu queue. Thus they can be guarantee that the cmpxchg_double
2432 * occurs on the right processor and that there was no operation on the
2433 * linked list in between.
2436 object
= c
->freelist
;
2438 if (unlikely(!object
|| !node_match(page
, node
))) {
2439 object
= __slab_alloc(s
, gfpflags
, node
, addr
, c
);
2440 stat(s
, ALLOC_SLOWPATH
);
2442 void *next_object
= get_freepointer_safe(s
, object
);
2445 * The cmpxchg will only match if there was no additional
2446 * operation and if we are on the right processor.
2448 * The cmpxchg does the following atomically (without lock
2450 * 1. Relocate first pointer to the current per cpu area.
2451 * 2. Verify that tid and freelist have not been changed
2452 * 3. If they were not changed replace tid and freelist
2454 * Since this is without lock semantics the protection is only
2455 * against code executing on this cpu *not* from access by
2458 if (unlikely(!this_cpu_cmpxchg_double(
2459 s
->cpu_slab
->freelist
, s
->cpu_slab
->tid
,
2461 next_object
, next_tid(tid
)))) {
2463 note_cmpxchg_failure("slab_alloc", s
, tid
);
2466 prefetch_freepointer(s
, next_object
);
2467 stat(s
, ALLOC_FASTPATH
);
2470 if (unlikely(gfpflags
& __GFP_ZERO
) && object
)
2471 memset(object
, 0, s
->object_size
);
2473 slab_post_alloc_hook(s
, gfpflags
, object
);
2478 static __always_inline
void *slab_alloc(struct kmem_cache
*s
,
2479 gfp_t gfpflags
, unsigned long addr
)
2481 return slab_alloc_node(s
, gfpflags
, NUMA_NO_NODE
, addr
);
2484 void *kmem_cache_alloc(struct kmem_cache
*s
, gfp_t gfpflags
)
2486 void *ret
= slab_alloc(s
, gfpflags
, _RET_IP_
);
2488 trace_kmem_cache_alloc(_RET_IP_
, ret
, s
->object_size
,
2493 EXPORT_SYMBOL(kmem_cache_alloc
);
2495 #ifdef CONFIG_TRACING
2496 void *kmem_cache_alloc_trace(struct kmem_cache
*s
, gfp_t gfpflags
, size_t size
)
2498 void *ret
= slab_alloc(s
, gfpflags
, _RET_IP_
);
2499 trace_kmalloc(_RET_IP_
, ret
, size
, s
->size
, gfpflags
);
2502 EXPORT_SYMBOL(kmem_cache_alloc_trace
);
2506 void *kmem_cache_alloc_node(struct kmem_cache
*s
, gfp_t gfpflags
, int node
)
2508 void *ret
= slab_alloc_node(s
, gfpflags
, node
, _RET_IP_
);
2510 trace_kmem_cache_alloc_node(_RET_IP_
, ret
,
2511 s
->object_size
, s
->size
, gfpflags
, node
);
2515 EXPORT_SYMBOL(kmem_cache_alloc_node
);
2517 #ifdef CONFIG_TRACING
2518 void *kmem_cache_alloc_node_trace(struct kmem_cache
*s
,
2520 int node
, size_t size
)
2522 void *ret
= slab_alloc_node(s
, gfpflags
, node
, _RET_IP_
);
2524 trace_kmalloc_node(_RET_IP_
, ret
,
2525 size
, s
->size
, gfpflags
, node
);
2528 EXPORT_SYMBOL(kmem_cache_alloc_node_trace
);
2533 * Slow path handling. This may still be called frequently since objects
2534 * have a longer lifetime than the cpu slabs in most processing loads.
2536 * So we still attempt to reduce cache line usage. Just take the slab
2537 * lock and free the item. If there is no additional partial page
2538 * handling required then we can return immediately.
2540 static void __slab_free(struct kmem_cache
*s
, struct page
*page
,
2541 void *x
, unsigned long addr
)
2544 void **object
= (void *)x
;
2547 unsigned long counters
;
2548 struct kmem_cache_node
*n
= NULL
;
2549 unsigned long uninitialized_var(flags
);
2551 stat(s
, FREE_SLOWPATH
);
2553 if (kmem_cache_debug(s
) &&
2554 !(n
= free_debug_processing(s
, page
, x
, addr
, &flags
)))
2559 spin_unlock_irqrestore(&n
->list_lock
, flags
);
2562 prior
= page
->freelist
;
2563 counters
= page
->counters
;
2564 set_freepointer(s
, object
, prior
);
2565 new.counters
= counters
;
2566 was_frozen
= new.frozen
;
2568 if ((!new.inuse
|| !prior
) && !was_frozen
) {
2570 if (kmem_cache_has_cpu_partial(s
) && !prior
) {
2573 * Slab was on no list before and will be
2575 * We can defer the list move and instead
2580 } else { /* Needs to be taken off a list */
2582 n
= get_node(s
, page_to_nid(page
));
2584 * Speculatively acquire the list_lock.
2585 * If the cmpxchg does not succeed then we may
2586 * drop the list_lock without any processing.
2588 * Otherwise the list_lock will synchronize with
2589 * other processors updating the list of slabs.
2591 spin_lock_irqsave(&n
->list_lock
, flags
);
2596 } while (!cmpxchg_double_slab(s
, page
,
2598 object
, new.counters
,
2604 * If we just froze the page then put it onto the
2605 * per cpu partial list.
2607 if (new.frozen
&& !was_frozen
) {
2608 put_cpu_partial(s
, page
, 1);
2609 stat(s
, CPU_PARTIAL_FREE
);
2612 * The list lock was not taken therefore no list
2613 * activity can be necessary.
2616 stat(s
, FREE_FROZEN
);
2620 if (unlikely(!new.inuse
&& n
->nr_partial
>= s
->min_partial
))
2624 * Objects left in the slab. If it was not on the partial list before
2627 if (!kmem_cache_has_cpu_partial(s
) && unlikely(!prior
)) {
2628 if (kmem_cache_debug(s
))
2629 remove_full(s
, n
, page
);
2630 add_partial(n
, page
, DEACTIVATE_TO_TAIL
);
2631 stat(s
, FREE_ADD_PARTIAL
);
2633 spin_unlock_irqrestore(&n
->list_lock
, flags
);
2639 * Slab on the partial list.
2641 remove_partial(n
, page
);
2642 stat(s
, FREE_REMOVE_PARTIAL
);
2644 /* Slab must be on the full list */
2645 remove_full(s
, n
, page
);
2648 spin_unlock_irqrestore(&n
->list_lock
, flags
);
2650 discard_slab(s
, page
);
2654 * Fastpath with forced inlining to produce a kfree and kmem_cache_free that
2655 * can perform fastpath freeing without additional function calls.
2657 * The fastpath is only possible if we are freeing to the current cpu slab
2658 * of this processor. This typically the case if we have just allocated
2661 * If fastpath is not possible then fall back to __slab_free where we deal
2662 * with all sorts of special processing.
2664 static __always_inline
void slab_free(struct kmem_cache
*s
,
2665 struct page
*page
, void *x
, unsigned long addr
)
2667 void **object
= (void *)x
;
2668 struct kmem_cache_cpu
*c
;
2671 slab_free_hook(s
, x
);
2675 * Determine the currently cpus per cpu slab.
2676 * The cpu may change afterward. However that does not matter since
2677 * data is retrieved via this pointer. If we are on the same cpu
2678 * during the cmpxchg then the free will succedd.
2681 tid
= this_cpu_read(s
->cpu_slab
->tid
);
2682 c
= raw_cpu_ptr(s
->cpu_slab
);
2683 } while (IS_ENABLED(CONFIG_PREEMPT
) && unlikely(tid
!= c
->tid
));
2685 /* Same with comment on barrier() in slab_alloc_node() */
2688 if (likely(page
== c
->page
)) {
2689 set_freepointer(s
, object
, c
->freelist
);
2691 if (unlikely(!this_cpu_cmpxchg_double(
2692 s
->cpu_slab
->freelist
, s
->cpu_slab
->tid
,
2694 object
, next_tid(tid
)))) {
2696 note_cmpxchg_failure("slab_free", s
, tid
);
2699 stat(s
, FREE_FASTPATH
);
2701 __slab_free(s
, page
, x
, addr
);
2705 void kmem_cache_free(struct kmem_cache
*s
, void *x
)
2707 s
= cache_from_obj(s
, x
);
2710 slab_free(s
, virt_to_head_page(x
), x
, _RET_IP_
);
2711 trace_kmem_cache_free(_RET_IP_
, x
);
2713 EXPORT_SYMBOL(kmem_cache_free
);
2716 * Object placement in a slab is made very easy because we always start at
2717 * offset 0. If we tune the size of the object to the alignment then we can
2718 * get the required alignment by putting one properly sized object after
2721 * Notice that the allocation order determines the sizes of the per cpu
2722 * caches. Each processor has always one slab available for allocations.
2723 * Increasing the allocation order reduces the number of times that slabs
2724 * must be moved on and off the partial lists and is therefore a factor in
2729 * Mininum / Maximum order of slab pages. This influences locking overhead
2730 * and slab fragmentation. A higher order reduces the number of partial slabs
2731 * and increases the number of allocations possible without having to
2732 * take the list_lock.
2734 static int slub_min_order
;
2735 static int slub_max_order
= PAGE_ALLOC_COSTLY_ORDER
;
2736 static int slub_min_objects
;
2739 * Calculate the order of allocation given an slab object size.
2741 * The order of allocation has significant impact on performance and other
2742 * system components. Generally order 0 allocations should be preferred since
2743 * order 0 does not cause fragmentation in the page allocator. Larger objects
2744 * be problematic to put into order 0 slabs because there may be too much
2745 * unused space left. We go to a higher order if more than 1/16th of the slab
2748 * In order to reach satisfactory performance we must ensure that a minimum
2749 * number of objects is in one slab. Otherwise we may generate too much
2750 * activity on the partial lists which requires taking the list_lock. This is
2751 * less a concern for large slabs though which are rarely used.
2753 * slub_max_order specifies the order where we begin to stop considering the
2754 * number of objects in a slab as critical. If we reach slub_max_order then
2755 * we try to keep the page order as low as possible. So we accept more waste
2756 * of space in favor of a small page order.
2758 * Higher order allocations also allow the placement of more objects in a
2759 * slab and thereby reduce object handling overhead. If the user has
2760 * requested a higher mininum order then we start with that one instead of
2761 * the smallest order which will fit the object.
2763 static inline int slab_order(int size
, int min_objects
,
2764 int max_order
, int fract_leftover
, int reserved
)
2768 int min_order
= slub_min_order
;
2770 if (order_objects(min_order
, size
, reserved
) > MAX_OBJS_PER_PAGE
)
2771 return get_order(size
* MAX_OBJS_PER_PAGE
) - 1;
2773 for (order
= max(min_order
,
2774 fls(min_objects
* size
- 1) - PAGE_SHIFT
);
2775 order
<= max_order
; order
++) {
2777 unsigned long slab_size
= PAGE_SIZE
<< order
;
2779 if (slab_size
< min_objects
* size
+ reserved
)
2782 rem
= (slab_size
- reserved
) % size
;
2784 if (rem
<= slab_size
/ fract_leftover
)
2792 static inline int calculate_order(int size
, int reserved
)
2800 * Attempt to find best configuration for a slab. This
2801 * works by first attempting to generate a layout with
2802 * the best configuration and backing off gradually.
2804 * First we reduce the acceptable waste in a slab. Then
2805 * we reduce the minimum objects required in a slab.
2807 min_objects
= slub_min_objects
;
2809 min_objects
= 4 * (fls(nr_cpu_ids
) + 1);
2810 max_objects
= order_objects(slub_max_order
, size
, reserved
);
2811 min_objects
= min(min_objects
, max_objects
);
2813 while (min_objects
> 1) {
2815 while (fraction
>= 4) {
2816 order
= slab_order(size
, min_objects
,
2817 slub_max_order
, fraction
, reserved
);
2818 if (order
<= slub_max_order
)
2826 * We were unable to place multiple objects in a slab. Now
2827 * lets see if we can place a single object there.
2829 order
= slab_order(size
, 1, slub_max_order
, 1, reserved
);
2830 if (order
<= slub_max_order
)
2834 * Doh this slab cannot be placed using slub_max_order.
2836 order
= slab_order(size
, 1, MAX_ORDER
, 1, reserved
);
2837 if (order
< MAX_ORDER
)
2843 init_kmem_cache_node(struct kmem_cache_node
*n
)
2846 spin_lock_init(&n
->list_lock
);
2847 INIT_LIST_HEAD(&n
->partial
);
2848 #ifdef CONFIG_SLUB_DEBUG
2849 atomic_long_set(&n
->nr_slabs
, 0);
2850 atomic_long_set(&n
->total_objects
, 0);
2851 INIT_LIST_HEAD(&n
->full
);
2855 static inline int alloc_kmem_cache_cpus(struct kmem_cache
*s
)
2857 BUILD_BUG_ON(PERCPU_DYNAMIC_EARLY_SIZE
<
2858 KMALLOC_SHIFT_HIGH
* sizeof(struct kmem_cache_cpu
));
2861 * Must align to double word boundary for the double cmpxchg
2862 * instructions to work; see __pcpu_double_call_return_bool().
2864 s
->cpu_slab
= __alloc_percpu(sizeof(struct kmem_cache_cpu
),
2865 2 * sizeof(void *));
2870 init_kmem_cache_cpus(s
);
2875 static struct kmem_cache
*kmem_cache_node
;
2878 * No kmalloc_node yet so do it by hand. We know that this is the first
2879 * slab on the node for this slabcache. There are no concurrent accesses
2882 * Note that this function only works on the kmem_cache_node
2883 * when allocating for the kmem_cache_node. This is used for bootstrapping
2884 * memory on a fresh node that has no slab structures yet.
2886 static void early_kmem_cache_node_alloc(int node
)
2889 struct kmem_cache_node
*n
;
2891 BUG_ON(kmem_cache_node
->size
< sizeof(struct kmem_cache_node
));
2893 page
= new_slab(kmem_cache_node
, GFP_NOWAIT
, node
);
2896 if (page_to_nid(page
) != node
) {
2897 pr_err("SLUB: Unable to allocate memory from node %d\n", node
);
2898 pr_err("SLUB: Allocating a useless per node structure in order to be able to continue\n");
2903 page
->freelist
= get_freepointer(kmem_cache_node
, n
);
2906 kmem_cache_node
->node
[node
] = n
;
2907 #ifdef CONFIG_SLUB_DEBUG
2908 init_object(kmem_cache_node
, n
, SLUB_RED_ACTIVE
);
2909 init_tracking(kmem_cache_node
, n
);
2911 init_kmem_cache_node(n
);
2912 inc_slabs_node(kmem_cache_node
, node
, page
->objects
);
2915 * No locks need to be taken here as it has just been
2916 * initialized and there is no concurrent access.
2918 __add_partial(n
, page
, DEACTIVATE_TO_HEAD
);
2921 static void free_kmem_cache_nodes(struct kmem_cache
*s
)
2924 struct kmem_cache_node
*n
;
2926 for_each_kmem_cache_node(s
, node
, n
) {
2927 kmem_cache_free(kmem_cache_node
, n
);
2928 s
->node
[node
] = NULL
;
2932 static int init_kmem_cache_nodes(struct kmem_cache
*s
)
2936 for_each_node_state(node
, N_NORMAL_MEMORY
) {
2937 struct kmem_cache_node
*n
;
2939 if (slab_state
== DOWN
) {
2940 early_kmem_cache_node_alloc(node
);
2943 n
= kmem_cache_alloc_node(kmem_cache_node
,
2947 free_kmem_cache_nodes(s
);
2952 init_kmem_cache_node(n
);
2957 static void set_min_partial(struct kmem_cache
*s
, unsigned long min
)
2959 if (min
< MIN_PARTIAL
)
2961 else if (min
> MAX_PARTIAL
)
2963 s
->min_partial
= min
;
2967 * calculate_sizes() determines the order and the distribution of data within
2970 static int calculate_sizes(struct kmem_cache
*s
, int forced_order
)
2972 unsigned long flags
= s
->flags
;
2973 unsigned long size
= s
->object_size
;
2977 * Round up object size to the next word boundary. We can only
2978 * place the free pointer at word boundaries and this determines
2979 * the possible location of the free pointer.
2981 size
= ALIGN(size
, sizeof(void *));
2983 #ifdef CONFIG_SLUB_DEBUG
2985 * Determine if we can poison the object itself. If the user of
2986 * the slab may touch the object after free or before allocation
2987 * then we should never poison the object itself.
2989 if ((flags
& SLAB_POISON
) && !(flags
& SLAB_DESTROY_BY_RCU
) &&
2991 s
->flags
|= __OBJECT_POISON
;
2993 s
->flags
&= ~__OBJECT_POISON
;
2997 * If we are Redzoning then check if there is some space between the
2998 * end of the object and the free pointer. If not then add an
2999 * additional word to have some bytes to store Redzone information.
3001 if ((flags
& SLAB_RED_ZONE
) && size
== s
->object_size
)
3002 size
+= sizeof(void *);
3006 * With that we have determined the number of bytes in actual use
3007 * by the object. This is the potential offset to the free pointer.
3011 if (((flags
& (SLAB_DESTROY_BY_RCU
| SLAB_POISON
)) ||
3014 * Relocate free pointer after the object if it is not
3015 * permitted to overwrite the first word of the object on
3018 * This is the case if we do RCU, have a constructor or
3019 * destructor or are poisoning the objects.
3022 size
+= sizeof(void *);
3025 #ifdef CONFIG_SLUB_DEBUG
3026 if (flags
& SLAB_STORE_USER
)
3028 * Need to store information about allocs and frees after
3031 size
+= 2 * sizeof(struct track
);
3033 if (flags
& SLAB_RED_ZONE
)
3035 * Add some empty padding so that we can catch
3036 * overwrites from earlier objects rather than let
3037 * tracking information or the free pointer be
3038 * corrupted if a user writes before the start
3041 size
+= sizeof(void *);
3045 * SLUB stores one object immediately after another beginning from
3046 * offset 0. In order to align the objects we have to simply size
3047 * each object to conform to the alignment.
3049 size
= ALIGN(size
, s
->align
);
3051 if (forced_order
>= 0)
3052 order
= forced_order
;
3054 order
= calculate_order(size
, s
->reserved
);
3061 s
->allocflags
|= __GFP_COMP
;
3063 if (s
->flags
& SLAB_CACHE_DMA
)
3064 s
->allocflags
|= GFP_DMA
;
3066 if (s
->flags
& SLAB_RECLAIM_ACCOUNT
)
3067 s
->allocflags
|= __GFP_RECLAIMABLE
;
3070 * Determine the number of objects per slab
3072 s
->oo
= oo_make(order
, size
, s
->reserved
);
3073 s
->min
= oo_make(get_order(size
), size
, s
->reserved
);
3074 if (oo_objects(s
->oo
) > oo_objects(s
->max
))
3077 return !!oo_objects(s
->oo
);
3080 static int kmem_cache_open(struct kmem_cache
*s
, unsigned long flags
)
3082 s
->flags
= kmem_cache_flags(s
->size
, flags
, s
->name
, s
->ctor
);
3085 if (need_reserve_slab_rcu
&& (s
->flags
& SLAB_DESTROY_BY_RCU
))
3086 s
->reserved
= sizeof(struct rcu_head
);
3088 if (!calculate_sizes(s
, -1))
3090 if (disable_higher_order_debug
) {
3092 * Disable debugging flags that store metadata if the min slab
3095 if (get_order(s
->size
) > get_order(s
->object_size
)) {
3096 s
->flags
&= ~DEBUG_METADATA_FLAGS
;
3098 if (!calculate_sizes(s
, -1))
3103 #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \
3104 defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
3105 if (system_has_cmpxchg_double() && (s
->flags
& SLAB_DEBUG_FLAGS
) == 0)
3106 /* Enable fast mode */
3107 s
->flags
|= __CMPXCHG_DOUBLE
;
3111 * The larger the object size is, the more pages we want on the partial
3112 * list to avoid pounding the page allocator excessively.
3114 set_min_partial(s
, ilog2(s
->size
) / 2);
3117 * cpu_partial determined the maximum number of objects kept in the
3118 * per cpu partial lists of a processor.
3120 * Per cpu partial lists mainly contain slabs that just have one
3121 * object freed. If they are used for allocation then they can be
3122 * filled up again with minimal effort. The slab will never hit the
3123 * per node partial lists and therefore no locking will be required.
3125 * This setting also determines
3127 * A) The number of objects from per cpu partial slabs dumped to the
3128 * per node list when we reach the limit.
3129 * B) The number of objects in cpu partial slabs to extract from the
3130 * per node list when we run out of per cpu objects. We only fetch
3131 * 50% to keep some capacity around for frees.
3133 if (!kmem_cache_has_cpu_partial(s
))
3135 else if (s
->size
>= PAGE_SIZE
)
3137 else if (s
->size
>= 1024)
3139 else if (s
->size
>= 256)
3140 s
->cpu_partial
= 13;
3142 s
->cpu_partial
= 30;
3145 s
->remote_node_defrag_ratio
= 1000;
3147 if (!init_kmem_cache_nodes(s
))
3150 if (alloc_kmem_cache_cpus(s
))
3153 free_kmem_cache_nodes(s
);
3155 if (flags
& SLAB_PANIC
)
3156 panic("Cannot create slab %s size=%lu realsize=%u "
3157 "order=%u offset=%u flags=%lx\n",
3158 s
->name
, (unsigned long)s
->size
, s
->size
,
3159 oo_order(s
->oo
), s
->offset
, flags
);
3163 static void list_slab_objects(struct kmem_cache
*s
, struct page
*page
,
3166 #ifdef CONFIG_SLUB_DEBUG
3167 void *addr
= page_address(page
);
3169 unsigned long *map
= kzalloc(BITS_TO_LONGS(page
->objects
) *
3170 sizeof(long), GFP_ATOMIC
);
3173 slab_err(s
, page
, text
, s
->name
);
3176 get_map(s
, page
, map
);
3177 for_each_object(p
, s
, addr
, page
->objects
) {
3179 if (!test_bit(slab_index(p
, s
, addr
), map
)) {
3180 pr_err("INFO: Object 0x%p @offset=%tu\n", p
, p
- addr
);
3181 print_tracking(s
, p
);
3190 * Attempt to free all partial slabs on a node.
3191 * This is called from kmem_cache_close(). We must be the last thread
3192 * using the cache and therefore we do not need to lock anymore.
3194 static void free_partial(struct kmem_cache
*s
, struct kmem_cache_node
*n
)
3196 struct page
*page
, *h
;
3198 list_for_each_entry_safe(page
, h
, &n
->partial
, lru
) {
3200 __remove_partial(n
, page
);
3201 discard_slab(s
, page
);
3203 list_slab_objects(s
, page
,
3204 "Objects remaining in %s on kmem_cache_close()");
3210 * Release all resources used by a slab cache.
3212 static inline int kmem_cache_close(struct kmem_cache
*s
)
3215 struct kmem_cache_node
*n
;
3218 /* Attempt to free all objects */
3219 for_each_kmem_cache_node(s
, node
, n
) {
3221 if (n
->nr_partial
|| slabs_node(s
, node
))
3224 free_percpu(s
->cpu_slab
);
3225 free_kmem_cache_nodes(s
);
3229 int __kmem_cache_shutdown(struct kmem_cache
*s
)
3231 return kmem_cache_close(s
);
3234 /********************************************************************
3236 *******************************************************************/
3238 static int __init
setup_slub_min_order(char *str
)
3240 get_option(&str
, &slub_min_order
);
3245 __setup("slub_min_order=", setup_slub_min_order
);
3247 static int __init
setup_slub_max_order(char *str
)
3249 get_option(&str
, &slub_max_order
);
3250 slub_max_order
= min(slub_max_order
, MAX_ORDER
- 1);
3255 __setup("slub_max_order=", setup_slub_max_order
);
3257 static int __init
setup_slub_min_objects(char *str
)
3259 get_option(&str
, &slub_min_objects
);
3264 __setup("slub_min_objects=", setup_slub_min_objects
);
3266 void *__kmalloc(size_t size
, gfp_t flags
)
3268 struct kmem_cache
*s
;
3271 if (unlikely(size
> KMALLOC_MAX_CACHE_SIZE
))
3272 return kmalloc_large(size
, flags
);
3274 s
= kmalloc_slab(size
, flags
);
3276 if (unlikely(ZERO_OR_NULL_PTR(s
)))
3279 ret
= slab_alloc(s
, flags
, _RET_IP_
);
3281 trace_kmalloc(_RET_IP_
, ret
, size
, s
->size
, flags
);
3285 EXPORT_SYMBOL(__kmalloc
);
3288 static void *kmalloc_large_node(size_t size
, gfp_t flags
, int node
)
3293 flags
|= __GFP_COMP
| __GFP_NOTRACK
;
3294 page
= alloc_kmem_pages_node(node
, flags
, get_order(size
));
3296 ptr
= page_address(page
);
3298 kmalloc_large_node_hook(ptr
, size
, flags
);
3302 void *__kmalloc_node(size_t size
, gfp_t flags
, int node
)
3304 struct kmem_cache
*s
;
3307 if (unlikely(size
> KMALLOC_MAX_CACHE_SIZE
)) {
3308 ret
= kmalloc_large_node(size
, flags
, node
);
3310 trace_kmalloc_node(_RET_IP_
, ret
,
3311 size
, PAGE_SIZE
<< get_order(size
),
3317 s
= kmalloc_slab(size
, flags
);
3319 if (unlikely(ZERO_OR_NULL_PTR(s
)))
3322 ret
= slab_alloc_node(s
, flags
, node
, _RET_IP_
);
3324 trace_kmalloc_node(_RET_IP_
, ret
, size
, s
->size
, flags
, node
);
3328 EXPORT_SYMBOL(__kmalloc_node
);
3331 size_t ksize(const void *object
)
3335 if (unlikely(object
== ZERO_SIZE_PTR
))
3338 page
= virt_to_head_page(object
);
3340 if (unlikely(!PageSlab(page
))) {
3341 WARN_ON(!PageCompound(page
));
3342 return PAGE_SIZE
<< compound_order(page
);
3345 return slab_ksize(page
->slab_cache
);
3347 EXPORT_SYMBOL(ksize
);
3349 void kfree(const void *x
)
3352 void *object
= (void *)x
;
3354 trace_kfree(_RET_IP_
, x
);
3356 if (unlikely(ZERO_OR_NULL_PTR(x
)))
3359 page
= virt_to_head_page(x
);
3360 if (unlikely(!PageSlab(page
))) {
3361 BUG_ON(!PageCompound(page
));
3363 __free_kmem_pages(page
, compound_order(page
));
3366 slab_free(page
->slab_cache
, page
, object
, _RET_IP_
);
3368 EXPORT_SYMBOL(kfree
);
3370 #define SHRINK_PROMOTE_MAX 32
3373 * kmem_cache_shrink discards empty slabs and promotes the slabs filled
3374 * up most to the head of the partial lists. New allocations will then
3375 * fill those up and thus they can be removed from the partial lists.
3377 * The slabs with the least items are placed last. This results in them
3378 * being allocated from last increasing the chance that the last objects
3379 * are freed in them.
3381 int __kmem_cache_shrink(struct kmem_cache
*s
, bool deactivate
)
3385 struct kmem_cache_node
*n
;
3388 struct list_head discard
;
3389 struct list_head promote
[SHRINK_PROMOTE_MAX
];
3390 unsigned long flags
;
3395 * Disable empty slabs caching. Used to avoid pinning offline
3396 * memory cgroups by kmem pages that can be freed.
3402 * s->cpu_partial is checked locklessly (see put_cpu_partial),
3403 * so we have to make sure the change is visible.
3405 kick_all_cpus_sync();
3409 for_each_kmem_cache_node(s
, node
, n
) {
3410 INIT_LIST_HEAD(&discard
);
3411 for (i
= 0; i
< SHRINK_PROMOTE_MAX
; i
++)
3412 INIT_LIST_HEAD(promote
+ i
);
3414 spin_lock_irqsave(&n
->list_lock
, flags
);
3417 * Build lists of slabs to discard or promote.
3419 * Note that concurrent frees may occur while we hold the
3420 * list_lock. page->inuse here is the upper limit.
3422 list_for_each_entry_safe(page
, t
, &n
->partial
, lru
) {
3423 int free
= page
->objects
- page
->inuse
;
3425 /* Do not reread page->inuse */
3428 /* We do not keep full slabs on the list */
3431 if (free
== page
->objects
) {
3432 list_move(&page
->lru
, &discard
);
3434 } else if (free
<= SHRINK_PROMOTE_MAX
)
3435 list_move(&page
->lru
, promote
+ free
- 1);
3439 * Promote the slabs filled up most to the head of the
3442 for (i
= SHRINK_PROMOTE_MAX
- 1; i
>= 0; i
--)
3443 list_splice(promote
+ i
, &n
->partial
);
3445 spin_unlock_irqrestore(&n
->list_lock
, flags
);
3447 /* Release empty slabs */
3448 list_for_each_entry_safe(page
, t
, &discard
, lru
)
3449 discard_slab(s
, page
);
3451 if (slabs_node(s
, node
))
3458 static int slab_mem_going_offline_callback(void *arg
)
3460 struct kmem_cache
*s
;
3462 mutex_lock(&slab_mutex
);
3463 list_for_each_entry(s
, &slab_caches
, list
)
3464 __kmem_cache_shrink(s
, false);
3465 mutex_unlock(&slab_mutex
);
3470 static void slab_mem_offline_callback(void *arg
)
3472 struct kmem_cache_node
*n
;
3473 struct kmem_cache
*s
;
3474 struct memory_notify
*marg
= arg
;
3477 offline_node
= marg
->status_change_nid_normal
;
3480 * If the node still has available memory. we need kmem_cache_node
3483 if (offline_node
< 0)
3486 mutex_lock(&slab_mutex
);
3487 list_for_each_entry(s
, &slab_caches
, list
) {
3488 n
= get_node(s
, offline_node
);
3491 * if n->nr_slabs > 0, slabs still exist on the node
3492 * that is going down. We were unable to free them,
3493 * and offline_pages() function shouldn't call this
3494 * callback. So, we must fail.
3496 BUG_ON(slabs_node(s
, offline_node
));
3498 s
->node
[offline_node
] = NULL
;
3499 kmem_cache_free(kmem_cache_node
, n
);
3502 mutex_unlock(&slab_mutex
);
3505 static int slab_mem_going_online_callback(void *arg
)
3507 struct kmem_cache_node
*n
;
3508 struct kmem_cache
*s
;
3509 struct memory_notify
*marg
= arg
;
3510 int nid
= marg
->status_change_nid_normal
;
3514 * If the node's memory is already available, then kmem_cache_node is
3515 * already created. Nothing to do.
3521 * We are bringing a node online. No memory is available yet. We must
3522 * allocate a kmem_cache_node structure in order to bring the node
3525 mutex_lock(&slab_mutex
);
3526 list_for_each_entry(s
, &slab_caches
, list
) {
3528 * XXX: kmem_cache_alloc_node will fallback to other nodes
3529 * since memory is not yet available from the node that
3532 n
= kmem_cache_alloc(kmem_cache_node
, GFP_KERNEL
);
3537 init_kmem_cache_node(n
);
3541 mutex_unlock(&slab_mutex
);
3545 static int slab_memory_callback(struct notifier_block
*self
,
3546 unsigned long action
, void *arg
)
3551 case MEM_GOING_ONLINE
:
3552 ret
= slab_mem_going_online_callback(arg
);
3554 case MEM_GOING_OFFLINE
:
3555 ret
= slab_mem_going_offline_callback(arg
);
3558 case MEM_CANCEL_ONLINE
:
3559 slab_mem_offline_callback(arg
);
3562 case MEM_CANCEL_OFFLINE
:
3566 ret
= notifier_from_errno(ret
);
3572 static struct notifier_block slab_memory_callback_nb
= {
3573 .notifier_call
= slab_memory_callback
,
3574 .priority
= SLAB_CALLBACK_PRI
,
3577 /********************************************************************
3578 * Basic setup of slabs
3579 *******************************************************************/
3582 * Used for early kmem_cache structures that were allocated using
3583 * the page allocator. Allocate them properly then fix up the pointers
3584 * that may be pointing to the wrong kmem_cache structure.
3587 static struct kmem_cache
* __init
bootstrap(struct kmem_cache
*static_cache
)
3590 struct kmem_cache
*s
= kmem_cache_zalloc(kmem_cache
, GFP_NOWAIT
);
3591 struct kmem_cache_node
*n
;
3593 memcpy(s
, static_cache
, kmem_cache
->object_size
);
3596 * This runs very early, and only the boot processor is supposed to be
3597 * up. Even if it weren't true, IRQs are not up so we couldn't fire
3600 __flush_cpu_slab(s
, smp_processor_id());
3601 for_each_kmem_cache_node(s
, node
, n
) {
3604 list_for_each_entry(p
, &n
->partial
, lru
)
3607 #ifdef CONFIG_SLUB_DEBUG
3608 list_for_each_entry(p
, &n
->full
, lru
)
3612 slab_init_memcg_params(s
);
3613 list_add(&s
->list
, &slab_caches
);
3617 void __init
kmem_cache_init(void)
3619 static __initdata
struct kmem_cache boot_kmem_cache
,
3620 boot_kmem_cache_node
;
3622 if (debug_guardpage_minorder())
3625 kmem_cache_node
= &boot_kmem_cache_node
;
3626 kmem_cache
= &boot_kmem_cache
;
3628 create_boot_cache(kmem_cache_node
, "kmem_cache_node",
3629 sizeof(struct kmem_cache_node
), SLAB_HWCACHE_ALIGN
);
3631 register_hotmemory_notifier(&slab_memory_callback_nb
);
3633 /* Able to allocate the per node structures */
3634 slab_state
= PARTIAL
;
3636 create_boot_cache(kmem_cache
, "kmem_cache",
3637 offsetof(struct kmem_cache
, node
) +
3638 nr_node_ids
* sizeof(struct kmem_cache_node
*),
3639 SLAB_HWCACHE_ALIGN
);
3641 kmem_cache
= bootstrap(&boot_kmem_cache
);
3644 * Allocate kmem_cache_node properly from the kmem_cache slab.
3645 * kmem_cache_node is separately allocated so no need to
3646 * update any list pointers.
3648 kmem_cache_node
= bootstrap(&boot_kmem_cache_node
);
3650 /* Now we can use the kmem_cache to allocate kmalloc slabs */
3651 create_kmalloc_caches(0);
3654 register_cpu_notifier(&slab_notifier
);
3657 pr_info("SLUB: HWalign=%d, Order=%d-%d, MinObjects=%d, CPUs=%d, Nodes=%d\n",
3659 slub_min_order
, slub_max_order
, slub_min_objects
,
3660 nr_cpu_ids
, nr_node_ids
);
3663 void __init
kmem_cache_init_late(void)
3668 __kmem_cache_alias(const char *name
, size_t size
, size_t align
,
3669 unsigned long flags
, void (*ctor
)(void *))
3671 struct kmem_cache
*s
, *c
;
3673 s
= find_mergeable(size
, align
, flags
, name
, ctor
);
3678 * Adjust the object sizes so that we clear
3679 * the complete object on kzalloc.
3681 s
->object_size
= max(s
->object_size
, (int)size
);
3682 s
->inuse
= max_t(int, s
->inuse
, ALIGN(size
, sizeof(void *)));
3684 for_each_memcg_cache(c
, s
) {
3685 c
->object_size
= s
->object_size
;
3686 c
->inuse
= max_t(int, c
->inuse
,
3687 ALIGN(size
, sizeof(void *)));
3690 if (sysfs_slab_alias(s
, name
)) {
3699 int __kmem_cache_create(struct kmem_cache
*s
, unsigned long flags
)
3703 err
= kmem_cache_open(s
, flags
);
3707 /* Mutex is not taken during early boot */
3708 if (slab_state
<= UP
)
3711 memcg_propagate_slab_attrs(s
);
3712 err
= sysfs_slab_add(s
);
3714 kmem_cache_close(s
);
3721 * Use the cpu notifier to insure that the cpu slabs are flushed when
3724 static int slab_cpuup_callback(struct notifier_block
*nfb
,
3725 unsigned long action
, void *hcpu
)
3727 long cpu
= (long)hcpu
;
3728 struct kmem_cache
*s
;
3729 unsigned long flags
;
3732 case CPU_UP_CANCELED
:
3733 case CPU_UP_CANCELED_FROZEN
:
3735 case CPU_DEAD_FROZEN
:
3736 mutex_lock(&slab_mutex
);
3737 list_for_each_entry(s
, &slab_caches
, list
) {
3738 local_irq_save(flags
);
3739 __flush_cpu_slab(s
, cpu
);
3740 local_irq_restore(flags
);
3742 mutex_unlock(&slab_mutex
);
3750 static struct notifier_block slab_notifier
= {
3751 .notifier_call
= slab_cpuup_callback
3756 void *__kmalloc_track_caller(size_t size
, gfp_t gfpflags
, unsigned long caller
)
3758 struct kmem_cache
*s
;
3761 if (unlikely(size
> KMALLOC_MAX_CACHE_SIZE
))
3762 return kmalloc_large(size
, gfpflags
);
3764 s
= kmalloc_slab(size
, gfpflags
);
3766 if (unlikely(ZERO_OR_NULL_PTR(s
)))
3769 ret
= slab_alloc(s
, gfpflags
, caller
);
3771 /* Honor the call site pointer we received. */
3772 trace_kmalloc(caller
, ret
, size
, s
->size
, gfpflags
);
3778 void *__kmalloc_node_track_caller(size_t size
, gfp_t gfpflags
,
3779 int node
, unsigned long caller
)
3781 struct kmem_cache
*s
;
3784 if (unlikely(size
> KMALLOC_MAX_CACHE_SIZE
)) {
3785 ret
= kmalloc_large_node(size
, gfpflags
, node
);
3787 trace_kmalloc_node(caller
, ret
,
3788 size
, PAGE_SIZE
<< get_order(size
),
3794 s
= kmalloc_slab(size
, gfpflags
);
3796 if (unlikely(ZERO_OR_NULL_PTR(s
)))
3799 ret
= slab_alloc_node(s
, gfpflags
, node
, caller
);
3801 /* Honor the call site pointer we received. */
3802 trace_kmalloc_node(caller
, ret
, size
, s
->size
, gfpflags
, node
);
3809 static int count_inuse(struct page
*page
)
3814 static int count_total(struct page
*page
)
3816 return page
->objects
;
3820 #ifdef CONFIG_SLUB_DEBUG
3821 static int validate_slab(struct kmem_cache
*s
, struct page
*page
,
3825 void *addr
= page_address(page
);
3827 if (!check_slab(s
, page
) ||
3828 !on_freelist(s
, page
, NULL
))
3831 /* Now we know that a valid freelist exists */
3832 bitmap_zero(map
, page
->objects
);
3834 get_map(s
, page
, map
);
3835 for_each_object(p
, s
, addr
, page
->objects
) {
3836 if (test_bit(slab_index(p
, s
, addr
), map
))
3837 if (!check_object(s
, page
, p
, SLUB_RED_INACTIVE
))
3841 for_each_object(p
, s
, addr
, page
->objects
)
3842 if (!test_bit(slab_index(p
, s
, addr
), map
))
3843 if (!check_object(s
, page
, p
, SLUB_RED_ACTIVE
))
3848 static void validate_slab_slab(struct kmem_cache
*s
, struct page
*page
,
3852 validate_slab(s
, page
, map
);
3856 static int validate_slab_node(struct kmem_cache
*s
,
3857 struct kmem_cache_node
*n
, unsigned long *map
)
3859 unsigned long count
= 0;
3861 unsigned long flags
;
3863 spin_lock_irqsave(&n
->list_lock
, flags
);
3865 list_for_each_entry(page
, &n
->partial
, lru
) {
3866 validate_slab_slab(s
, page
, map
);
3869 if (count
!= n
->nr_partial
)
3870 pr_err("SLUB %s: %ld partial slabs counted but counter=%ld\n",
3871 s
->name
, count
, n
->nr_partial
);
3873 if (!(s
->flags
& SLAB_STORE_USER
))
3876 list_for_each_entry(page
, &n
->full
, lru
) {
3877 validate_slab_slab(s
, page
, map
);
3880 if (count
!= atomic_long_read(&n
->nr_slabs
))
3881 pr_err("SLUB: %s %ld slabs counted but counter=%ld\n",
3882 s
->name
, count
, atomic_long_read(&n
->nr_slabs
));
3885 spin_unlock_irqrestore(&n
->list_lock
, flags
);
3889 static long validate_slab_cache(struct kmem_cache
*s
)
3892 unsigned long count
= 0;
3893 unsigned long *map
= kmalloc(BITS_TO_LONGS(oo_objects(s
->max
)) *
3894 sizeof(unsigned long), GFP_KERNEL
);
3895 struct kmem_cache_node
*n
;
3901 for_each_kmem_cache_node(s
, node
, n
)
3902 count
+= validate_slab_node(s
, n
, map
);
3907 * Generate lists of code addresses where slabcache objects are allocated
3912 unsigned long count
;
3919 DECLARE_BITMAP(cpus
, NR_CPUS
);
3925 unsigned long count
;
3926 struct location
*loc
;
3929 static void free_loc_track(struct loc_track
*t
)
3932 free_pages((unsigned long)t
->loc
,
3933 get_order(sizeof(struct location
) * t
->max
));
3936 static int alloc_loc_track(struct loc_track
*t
, unsigned long max
, gfp_t flags
)
3941 order
= get_order(sizeof(struct location
) * max
);
3943 l
= (void *)__get_free_pages(flags
, order
);
3948 memcpy(l
, t
->loc
, sizeof(struct location
) * t
->count
);
3956 static int add_location(struct loc_track
*t
, struct kmem_cache
*s
,
3957 const struct track
*track
)
3959 long start
, end
, pos
;
3961 unsigned long caddr
;
3962 unsigned long age
= jiffies
- track
->when
;
3968 pos
= start
+ (end
- start
+ 1) / 2;
3971 * There is nothing at "end". If we end up there
3972 * we need to add something to before end.
3977 caddr
= t
->loc
[pos
].addr
;
3978 if (track
->addr
== caddr
) {
3984 if (age
< l
->min_time
)
3986 if (age
> l
->max_time
)
3989 if (track
->pid
< l
->min_pid
)
3990 l
->min_pid
= track
->pid
;
3991 if (track
->pid
> l
->max_pid
)
3992 l
->max_pid
= track
->pid
;
3994 cpumask_set_cpu(track
->cpu
,
3995 to_cpumask(l
->cpus
));
3997 node_set(page_to_nid(virt_to_page(track
)), l
->nodes
);
4001 if (track
->addr
< caddr
)
4008 * Not found. Insert new tracking element.
4010 if (t
->count
>= t
->max
&& !alloc_loc_track(t
, 2 * t
->max
, GFP_ATOMIC
))
4016 (t
->count
- pos
) * sizeof(struct location
));
4019 l
->addr
= track
->addr
;
4023 l
->min_pid
= track
->pid
;
4024 l
->max_pid
= track
->pid
;
4025 cpumask_clear(to_cpumask(l
->cpus
));
4026 cpumask_set_cpu(track
->cpu
, to_cpumask(l
->cpus
));
4027 nodes_clear(l
->nodes
);
4028 node_set(page_to_nid(virt_to_page(track
)), l
->nodes
);
4032 static void process_slab(struct loc_track
*t
, struct kmem_cache
*s
,
4033 struct page
*page
, enum track_item alloc
,
4036 void *addr
= page_address(page
);
4039 bitmap_zero(map
, page
->objects
);
4040 get_map(s
, page
, map
);
4042 for_each_object(p
, s
, addr
, page
->objects
)
4043 if (!test_bit(slab_index(p
, s
, addr
), map
))
4044 add_location(t
, s
, get_track(s
, p
, alloc
));
4047 static int list_locations(struct kmem_cache
*s
, char *buf
,
4048 enum track_item alloc
)
4052 struct loc_track t
= { 0, 0, NULL
};
4054 unsigned long *map
= kmalloc(BITS_TO_LONGS(oo_objects(s
->max
)) *
4055 sizeof(unsigned long), GFP_KERNEL
);
4056 struct kmem_cache_node
*n
;
4058 if (!map
|| !alloc_loc_track(&t
, PAGE_SIZE
/ sizeof(struct location
),
4061 return sprintf(buf
, "Out of memory\n");
4063 /* Push back cpu slabs */
4066 for_each_kmem_cache_node(s
, node
, n
) {
4067 unsigned long flags
;
4070 if (!atomic_long_read(&n
->nr_slabs
))
4073 spin_lock_irqsave(&n
->list_lock
, flags
);
4074 list_for_each_entry(page
, &n
->partial
, lru
)
4075 process_slab(&t
, s
, page
, alloc
, map
);
4076 list_for_each_entry(page
, &n
->full
, lru
)
4077 process_slab(&t
, s
, page
, alloc
, map
);
4078 spin_unlock_irqrestore(&n
->list_lock
, flags
);
4081 for (i
= 0; i
< t
.count
; i
++) {
4082 struct location
*l
= &t
.loc
[i
];
4084 if (len
> PAGE_SIZE
- KSYM_SYMBOL_LEN
- 100)
4086 len
+= sprintf(buf
+ len
, "%7ld ", l
->count
);
4089 len
+= sprintf(buf
+ len
, "%pS", (void *)l
->addr
);
4091 len
+= sprintf(buf
+ len
, "<not-available>");
4093 if (l
->sum_time
!= l
->min_time
) {
4094 len
+= sprintf(buf
+ len
, " age=%ld/%ld/%ld",
4096 (long)div_u64(l
->sum_time
, l
->count
),
4099 len
+= sprintf(buf
+ len
, " age=%ld",
4102 if (l
->min_pid
!= l
->max_pid
)
4103 len
+= sprintf(buf
+ len
, " pid=%ld-%ld",
4104 l
->min_pid
, l
->max_pid
);
4106 len
+= sprintf(buf
+ len
, " pid=%ld",
4109 if (num_online_cpus() > 1 &&
4110 !cpumask_empty(to_cpumask(l
->cpus
)) &&
4111 len
< PAGE_SIZE
- 60) {
4112 len
+= sprintf(buf
+ len
, " cpus=");
4113 len
+= cpulist_scnprintf(buf
+ len
,
4114 PAGE_SIZE
- len
- 50,
4115 to_cpumask(l
->cpus
));
4118 if (nr_online_nodes
> 1 && !nodes_empty(l
->nodes
) &&
4119 len
< PAGE_SIZE
- 60) {
4120 len
+= sprintf(buf
+ len
, " nodes=");
4121 len
+= nodelist_scnprintf(buf
+ len
,
4122 PAGE_SIZE
- len
- 50,
4126 len
+= sprintf(buf
+ len
, "\n");
4132 len
+= sprintf(buf
, "No data\n");
4137 #ifdef SLUB_RESILIENCY_TEST
4138 static void __init
resiliency_test(void)
4142 BUILD_BUG_ON(KMALLOC_MIN_SIZE
> 16 || KMALLOC_SHIFT_HIGH
< 10);
4144 pr_err("SLUB resiliency testing\n");
4145 pr_err("-----------------------\n");
4146 pr_err("A. Corruption after allocation\n");
4148 p
= kzalloc(16, GFP_KERNEL
);
4150 pr_err("\n1. kmalloc-16: Clobber Redzone/next pointer 0x12->0x%p\n\n",
4153 validate_slab_cache(kmalloc_caches
[4]);
4155 /* Hmmm... The next two are dangerous */
4156 p
= kzalloc(32, GFP_KERNEL
);
4157 p
[32 + sizeof(void *)] = 0x34;
4158 pr_err("\n2. kmalloc-32: Clobber next pointer/next slab 0x34 -> -0x%p\n",
4160 pr_err("If allocated object is overwritten then not detectable\n\n");
4162 validate_slab_cache(kmalloc_caches
[5]);
4163 p
= kzalloc(64, GFP_KERNEL
);
4164 p
+= 64 + (get_cycles() & 0xff) * sizeof(void *);
4166 pr_err("\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n",
4168 pr_err("If allocated object is overwritten then not detectable\n\n");
4169 validate_slab_cache(kmalloc_caches
[6]);
4171 pr_err("\nB. Corruption after free\n");
4172 p
= kzalloc(128, GFP_KERNEL
);
4175 pr_err("1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p
);
4176 validate_slab_cache(kmalloc_caches
[7]);
4178 p
= kzalloc(256, GFP_KERNEL
);
4181 pr_err("\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n", p
);
4182 validate_slab_cache(kmalloc_caches
[8]);
4184 p
= kzalloc(512, GFP_KERNEL
);
4187 pr_err("\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p
);
4188 validate_slab_cache(kmalloc_caches
[9]);
4192 static void resiliency_test(void) {};
4197 enum slab_stat_type
{
4198 SL_ALL
, /* All slabs */
4199 SL_PARTIAL
, /* Only partially allocated slabs */
4200 SL_CPU
, /* Only slabs used for cpu caches */
4201 SL_OBJECTS
, /* Determine allocated objects not slabs */
4202 SL_TOTAL
/* Determine object capacity not slabs */
4205 #define SO_ALL (1 << SL_ALL)
4206 #define SO_PARTIAL (1 << SL_PARTIAL)
4207 #define SO_CPU (1 << SL_CPU)
4208 #define SO_OBJECTS (1 << SL_OBJECTS)
4209 #define SO_TOTAL (1 << SL_TOTAL)
4211 static ssize_t
show_slab_objects(struct kmem_cache
*s
,
4212 char *buf
, unsigned long flags
)
4214 unsigned long total
= 0;
4217 unsigned long *nodes
;
4219 nodes
= kzalloc(sizeof(unsigned long) * nr_node_ids
, GFP_KERNEL
);
4223 if (flags
& SO_CPU
) {
4226 for_each_possible_cpu(cpu
) {
4227 struct kmem_cache_cpu
*c
= per_cpu_ptr(s
->cpu_slab
,
4232 page
= ACCESS_ONCE(c
->page
);
4236 node
= page_to_nid(page
);
4237 if (flags
& SO_TOTAL
)
4239 else if (flags
& SO_OBJECTS
)
4247 page
= ACCESS_ONCE(c
->partial
);
4249 node
= page_to_nid(page
);
4250 if (flags
& SO_TOTAL
)
4252 else if (flags
& SO_OBJECTS
)
4263 #ifdef CONFIG_SLUB_DEBUG
4264 if (flags
& SO_ALL
) {
4265 struct kmem_cache_node
*n
;
4267 for_each_kmem_cache_node(s
, node
, n
) {
4269 if (flags
& SO_TOTAL
)
4270 x
= atomic_long_read(&n
->total_objects
);
4271 else if (flags
& SO_OBJECTS
)
4272 x
= atomic_long_read(&n
->total_objects
) -
4273 count_partial(n
, count_free
);
4275 x
= atomic_long_read(&n
->nr_slabs
);
4282 if (flags
& SO_PARTIAL
) {
4283 struct kmem_cache_node
*n
;
4285 for_each_kmem_cache_node(s
, node
, n
) {
4286 if (flags
& SO_TOTAL
)
4287 x
= count_partial(n
, count_total
);
4288 else if (flags
& SO_OBJECTS
)
4289 x
= count_partial(n
, count_inuse
);
4296 x
= sprintf(buf
, "%lu", total
);
4298 for (node
= 0; node
< nr_node_ids
; node
++)
4300 x
+= sprintf(buf
+ x
, " N%d=%lu",
4305 return x
+ sprintf(buf
+ x
, "\n");
4308 #ifdef CONFIG_SLUB_DEBUG
4309 static int any_slab_objects(struct kmem_cache
*s
)
4312 struct kmem_cache_node
*n
;
4314 for_each_kmem_cache_node(s
, node
, n
)
4315 if (atomic_long_read(&n
->total_objects
))
4322 #define to_slab_attr(n) container_of(n, struct slab_attribute, attr)
4323 #define to_slab(n) container_of(n, struct kmem_cache, kobj)
4325 struct slab_attribute
{
4326 struct attribute attr
;
4327 ssize_t (*show
)(struct kmem_cache
*s
, char *buf
);
4328 ssize_t (*store
)(struct kmem_cache
*s
, const char *x
, size_t count
);
4331 #define SLAB_ATTR_RO(_name) \
4332 static struct slab_attribute _name##_attr = \
4333 __ATTR(_name, 0400, _name##_show, NULL)
4335 #define SLAB_ATTR(_name) \
4336 static struct slab_attribute _name##_attr = \
4337 __ATTR(_name, 0600, _name##_show, _name##_store)
4339 static ssize_t
slab_size_show(struct kmem_cache
*s
, char *buf
)
4341 return sprintf(buf
, "%d\n", s
->size
);
4343 SLAB_ATTR_RO(slab_size
);
4345 static ssize_t
align_show(struct kmem_cache
*s
, char *buf
)
4347 return sprintf(buf
, "%d\n", s
->align
);
4349 SLAB_ATTR_RO(align
);
4351 static ssize_t
object_size_show(struct kmem_cache
*s
, char *buf
)
4353 return sprintf(buf
, "%d\n", s
->object_size
);
4355 SLAB_ATTR_RO(object_size
);
4357 static ssize_t
objs_per_slab_show(struct kmem_cache
*s
, char *buf
)
4359 return sprintf(buf
, "%d\n", oo_objects(s
->oo
));
4361 SLAB_ATTR_RO(objs_per_slab
);
4363 static ssize_t
order_store(struct kmem_cache
*s
,
4364 const char *buf
, size_t length
)
4366 unsigned long order
;
4369 err
= kstrtoul(buf
, 10, &order
);
4373 if (order
> slub_max_order
|| order
< slub_min_order
)
4376 calculate_sizes(s
, order
);
4380 static ssize_t
order_show(struct kmem_cache
*s
, char *buf
)
4382 return sprintf(buf
, "%d\n", oo_order(s
->oo
));
4386 static ssize_t
min_partial_show(struct kmem_cache
*s
, char *buf
)
4388 return sprintf(buf
, "%lu\n", s
->min_partial
);
4391 static ssize_t
min_partial_store(struct kmem_cache
*s
, const char *buf
,
4397 err
= kstrtoul(buf
, 10, &min
);
4401 set_min_partial(s
, min
);
4404 SLAB_ATTR(min_partial
);
4406 static ssize_t
cpu_partial_show(struct kmem_cache
*s
, char *buf
)
4408 return sprintf(buf
, "%u\n", s
->cpu_partial
);
4411 static ssize_t
cpu_partial_store(struct kmem_cache
*s
, const char *buf
,
4414 unsigned long objects
;
4417 err
= kstrtoul(buf
, 10, &objects
);
4420 if (objects
&& !kmem_cache_has_cpu_partial(s
))
4423 s
->cpu_partial
= objects
;
4427 SLAB_ATTR(cpu_partial
);
4429 static ssize_t
ctor_show(struct kmem_cache
*s
, char *buf
)
4433 return sprintf(buf
, "%pS\n", s
->ctor
);
4437 static ssize_t
aliases_show(struct kmem_cache
*s
, char *buf
)
4439 return sprintf(buf
, "%d\n", s
->refcount
< 0 ? 0 : s
->refcount
- 1);
4441 SLAB_ATTR_RO(aliases
);
4443 static ssize_t
partial_show(struct kmem_cache
*s
, char *buf
)
4445 return show_slab_objects(s
, buf
, SO_PARTIAL
);
4447 SLAB_ATTR_RO(partial
);
4449 static ssize_t
cpu_slabs_show(struct kmem_cache
*s
, char *buf
)
4451 return show_slab_objects(s
, buf
, SO_CPU
);
4453 SLAB_ATTR_RO(cpu_slabs
);
4455 static ssize_t
objects_show(struct kmem_cache
*s
, char *buf
)
4457 return show_slab_objects(s
, buf
, SO_ALL
|SO_OBJECTS
);
4459 SLAB_ATTR_RO(objects
);
4461 static ssize_t
objects_partial_show(struct kmem_cache
*s
, char *buf
)
4463 return show_slab_objects(s
, buf
, SO_PARTIAL
|SO_OBJECTS
);
4465 SLAB_ATTR_RO(objects_partial
);
4467 static ssize_t
slabs_cpu_partial_show(struct kmem_cache
*s
, char *buf
)
4474 for_each_online_cpu(cpu
) {
4475 struct page
*page
= per_cpu_ptr(s
->cpu_slab
, cpu
)->partial
;
4478 pages
+= page
->pages
;
4479 objects
+= page
->pobjects
;
4483 len
= sprintf(buf
, "%d(%d)", objects
, pages
);
4486 for_each_online_cpu(cpu
) {
4487 struct page
*page
= per_cpu_ptr(s
->cpu_slab
, cpu
) ->partial
;
4489 if (page
&& len
< PAGE_SIZE
- 20)
4490 len
+= sprintf(buf
+ len
, " C%d=%d(%d)", cpu
,
4491 page
->pobjects
, page
->pages
);
4494 return len
+ sprintf(buf
+ len
, "\n");
4496 SLAB_ATTR_RO(slabs_cpu_partial
);
4498 static ssize_t
reclaim_account_show(struct kmem_cache
*s
, char *buf
)
4500 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_RECLAIM_ACCOUNT
));
4503 static ssize_t
reclaim_account_store(struct kmem_cache
*s
,
4504 const char *buf
, size_t length
)
4506 s
->flags
&= ~SLAB_RECLAIM_ACCOUNT
;
4508 s
->flags
|= SLAB_RECLAIM_ACCOUNT
;
4511 SLAB_ATTR(reclaim_account
);
4513 static ssize_t
hwcache_align_show(struct kmem_cache
*s
, char *buf
)
4515 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_HWCACHE_ALIGN
));
4517 SLAB_ATTR_RO(hwcache_align
);
4519 #ifdef CONFIG_ZONE_DMA
4520 static ssize_t
cache_dma_show(struct kmem_cache
*s
, char *buf
)
4522 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_CACHE_DMA
));
4524 SLAB_ATTR_RO(cache_dma
);
4527 static ssize_t
destroy_by_rcu_show(struct kmem_cache
*s
, char *buf
)
4529 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_DESTROY_BY_RCU
));
4531 SLAB_ATTR_RO(destroy_by_rcu
);
4533 static ssize_t
reserved_show(struct kmem_cache
*s
, char *buf
)
4535 return sprintf(buf
, "%d\n", s
->reserved
);
4537 SLAB_ATTR_RO(reserved
);
4539 #ifdef CONFIG_SLUB_DEBUG
4540 static ssize_t
slabs_show(struct kmem_cache
*s
, char *buf
)
4542 return show_slab_objects(s
, buf
, SO_ALL
);
4544 SLAB_ATTR_RO(slabs
);
4546 static ssize_t
total_objects_show(struct kmem_cache
*s
, char *buf
)
4548 return show_slab_objects(s
, buf
, SO_ALL
|SO_TOTAL
);
4550 SLAB_ATTR_RO(total_objects
);
4552 static ssize_t
sanity_checks_show(struct kmem_cache
*s
, char *buf
)
4554 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_DEBUG_FREE
));
4557 static ssize_t
sanity_checks_store(struct kmem_cache
*s
,
4558 const char *buf
, size_t length
)
4560 s
->flags
&= ~SLAB_DEBUG_FREE
;
4561 if (buf
[0] == '1') {
4562 s
->flags
&= ~__CMPXCHG_DOUBLE
;
4563 s
->flags
|= SLAB_DEBUG_FREE
;
4567 SLAB_ATTR(sanity_checks
);
4569 static ssize_t
trace_show(struct kmem_cache
*s
, char *buf
)
4571 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_TRACE
));
4574 static ssize_t
trace_store(struct kmem_cache
*s
, const char *buf
,
4578 * Tracing a merged cache is going to give confusing results
4579 * as well as cause other issues like converting a mergeable
4580 * cache into an umergeable one.
4582 if (s
->refcount
> 1)
4585 s
->flags
&= ~SLAB_TRACE
;
4586 if (buf
[0] == '1') {
4587 s
->flags
&= ~__CMPXCHG_DOUBLE
;
4588 s
->flags
|= SLAB_TRACE
;
4594 static ssize_t
red_zone_show(struct kmem_cache
*s
, char *buf
)
4596 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_RED_ZONE
));
4599 static ssize_t
red_zone_store(struct kmem_cache
*s
,
4600 const char *buf
, size_t length
)
4602 if (any_slab_objects(s
))
4605 s
->flags
&= ~SLAB_RED_ZONE
;
4606 if (buf
[0] == '1') {
4607 s
->flags
&= ~__CMPXCHG_DOUBLE
;
4608 s
->flags
|= SLAB_RED_ZONE
;
4610 calculate_sizes(s
, -1);
4613 SLAB_ATTR(red_zone
);
4615 static ssize_t
poison_show(struct kmem_cache
*s
, char *buf
)
4617 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_POISON
));
4620 static ssize_t
poison_store(struct kmem_cache
*s
,
4621 const char *buf
, size_t length
)
4623 if (any_slab_objects(s
))
4626 s
->flags
&= ~SLAB_POISON
;
4627 if (buf
[0] == '1') {
4628 s
->flags
&= ~__CMPXCHG_DOUBLE
;
4629 s
->flags
|= SLAB_POISON
;
4631 calculate_sizes(s
, -1);
4636 static ssize_t
store_user_show(struct kmem_cache
*s
, char *buf
)
4638 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_STORE_USER
));
4641 static ssize_t
store_user_store(struct kmem_cache
*s
,
4642 const char *buf
, size_t length
)
4644 if (any_slab_objects(s
))
4647 s
->flags
&= ~SLAB_STORE_USER
;
4648 if (buf
[0] == '1') {
4649 s
->flags
&= ~__CMPXCHG_DOUBLE
;
4650 s
->flags
|= SLAB_STORE_USER
;
4652 calculate_sizes(s
, -1);
4655 SLAB_ATTR(store_user
);
4657 static ssize_t
validate_show(struct kmem_cache
*s
, char *buf
)
4662 static ssize_t
validate_store(struct kmem_cache
*s
,
4663 const char *buf
, size_t length
)
4667 if (buf
[0] == '1') {
4668 ret
= validate_slab_cache(s
);
4674 SLAB_ATTR(validate
);
4676 static ssize_t
alloc_calls_show(struct kmem_cache
*s
, char *buf
)
4678 if (!(s
->flags
& SLAB_STORE_USER
))
4680 return list_locations(s
, buf
, TRACK_ALLOC
);
4682 SLAB_ATTR_RO(alloc_calls
);
4684 static ssize_t
free_calls_show(struct kmem_cache
*s
, char *buf
)
4686 if (!(s
->flags
& SLAB_STORE_USER
))
4688 return list_locations(s
, buf
, TRACK_FREE
);
4690 SLAB_ATTR_RO(free_calls
);
4691 #endif /* CONFIG_SLUB_DEBUG */
4693 #ifdef CONFIG_FAILSLAB
4694 static ssize_t
failslab_show(struct kmem_cache
*s
, char *buf
)
4696 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_FAILSLAB
));
4699 static ssize_t
failslab_store(struct kmem_cache
*s
, const char *buf
,
4702 if (s
->refcount
> 1)
4705 s
->flags
&= ~SLAB_FAILSLAB
;
4707 s
->flags
|= SLAB_FAILSLAB
;
4710 SLAB_ATTR(failslab
);
4713 static ssize_t
shrink_show(struct kmem_cache
*s
, char *buf
)
4718 static ssize_t
shrink_store(struct kmem_cache
*s
,
4719 const char *buf
, size_t length
)
4722 kmem_cache_shrink(s
);
4730 static ssize_t
remote_node_defrag_ratio_show(struct kmem_cache
*s
, char *buf
)
4732 return sprintf(buf
, "%d\n", s
->remote_node_defrag_ratio
/ 10);
4735 static ssize_t
remote_node_defrag_ratio_store(struct kmem_cache
*s
,
4736 const char *buf
, size_t length
)
4738 unsigned long ratio
;
4741 err
= kstrtoul(buf
, 10, &ratio
);
4746 s
->remote_node_defrag_ratio
= ratio
* 10;
4750 SLAB_ATTR(remote_node_defrag_ratio
);
4753 #ifdef CONFIG_SLUB_STATS
4754 static int show_stat(struct kmem_cache
*s
, char *buf
, enum stat_item si
)
4756 unsigned long sum
= 0;
4759 int *data
= kmalloc(nr_cpu_ids
* sizeof(int), GFP_KERNEL
);
4764 for_each_online_cpu(cpu
) {
4765 unsigned x
= per_cpu_ptr(s
->cpu_slab
, cpu
)->stat
[si
];
4771 len
= sprintf(buf
, "%lu", sum
);
4774 for_each_online_cpu(cpu
) {
4775 if (data
[cpu
] && len
< PAGE_SIZE
- 20)
4776 len
+= sprintf(buf
+ len
, " C%d=%u", cpu
, data
[cpu
]);
4780 return len
+ sprintf(buf
+ len
, "\n");
4783 static void clear_stat(struct kmem_cache
*s
, enum stat_item si
)
4787 for_each_online_cpu(cpu
)
4788 per_cpu_ptr(s
->cpu_slab
, cpu
)->stat
[si
] = 0;
4791 #define STAT_ATTR(si, text) \
4792 static ssize_t text##_show(struct kmem_cache *s, char *buf) \
4794 return show_stat(s, buf, si); \
4796 static ssize_t text##_store(struct kmem_cache *s, \
4797 const char *buf, size_t length) \
4799 if (buf[0] != '0') \
4801 clear_stat(s, si); \
4806 STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath);
4807 STAT_ATTR(ALLOC_SLOWPATH
, alloc_slowpath
);
4808 STAT_ATTR(FREE_FASTPATH
, free_fastpath
);
4809 STAT_ATTR(FREE_SLOWPATH
, free_slowpath
);
4810 STAT_ATTR(FREE_FROZEN
, free_frozen
);
4811 STAT_ATTR(FREE_ADD_PARTIAL
, free_add_partial
);
4812 STAT_ATTR(FREE_REMOVE_PARTIAL
, free_remove_partial
);
4813 STAT_ATTR(ALLOC_FROM_PARTIAL
, alloc_from_partial
);
4814 STAT_ATTR(ALLOC_SLAB
, alloc_slab
);
4815 STAT_ATTR(ALLOC_REFILL
, alloc_refill
);
4816 STAT_ATTR(ALLOC_NODE_MISMATCH
, alloc_node_mismatch
);
4817 STAT_ATTR(FREE_SLAB
, free_slab
);
4818 STAT_ATTR(CPUSLAB_FLUSH
, cpuslab_flush
);
4819 STAT_ATTR(DEACTIVATE_FULL
, deactivate_full
);
4820 STAT_ATTR(DEACTIVATE_EMPTY
, deactivate_empty
);
4821 STAT_ATTR(DEACTIVATE_TO_HEAD
, deactivate_to_head
);
4822 STAT_ATTR(DEACTIVATE_TO_TAIL
, deactivate_to_tail
);
4823 STAT_ATTR(DEACTIVATE_REMOTE_FREES
, deactivate_remote_frees
);
4824 STAT_ATTR(DEACTIVATE_BYPASS
, deactivate_bypass
);
4825 STAT_ATTR(ORDER_FALLBACK
, order_fallback
);
4826 STAT_ATTR(CMPXCHG_DOUBLE_CPU_FAIL
, cmpxchg_double_cpu_fail
);
4827 STAT_ATTR(CMPXCHG_DOUBLE_FAIL
, cmpxchg_double_fail
);
4828 STAT_ATTR(CPU_PARTIAL_ALLOC
, cpu_partial_alloc
);
4829 STAT_ATTR(CPU_PARTIAL_FREE
, cpu_partial_free
);
4830 STAT_ATTR(CPU_PARTIAL_NODE
, cpu_partial_node
);
4831 STAT_ATTR(CPU_PARTIAL_DRAIN
, cpu_partial_drain
);
4834 static struct attribute
*slab_attrs
[] = {
4835 &slab_size_attr
.attr
,
4836 &object_size_attr
.attr
,
4837 &objs_per_slab_attr
.attr
,
4839 &min_partial_attr
.attr
,
4840 &cpu_partial_attr
.attr
,
4842 &objects_partial_attr
.attr
,
4844 &cpu_slabs_attr
.attr
,
4848 &hwcache_align_attr
.attr
,
4849 &reclaim_account_attr
.attr
,
4850 &destroy_by_rcu_attr
.attr
,
4852 &reserved_attr
.attr
,
4853 &slabs_cpu_partial_attr
.attr
,
4854 #ifdef CONFIG_SLUB_DEBUG
4855 &total_objects_attr
.attr
,
4857 &sanity_checks_attr
.attr
,
4859 &red_zone_attr
.attr
,
4861 &store_user_attr
.attr
,
4862 &validate_attr
.attr
,
4863 &alloc_calls_attr
.attr
,
4864 &free_calls_attr
.attr
,
4866 #ifdef CONFIG_ZONE_DMA
4867 &cache_dma_attr
.attr
,
4870 &remote_node_defrag_ratio_attr
.attr
,
4872 #ifdef CONFIG_SLUB_STATS
4873 &alloc_fastpath_attr
.attr
,
4874 &alloc_slowpath_attr
.attr
,
4875 &free_fastpath_attr
.attr
,
4876 &free_slowpath_attr
.attr
,
4877 &free_frozen_attr
.attr
,
4878 &free_add_partial_attr
.attr
,
4879 &free_remove_partial_attr
.attr
,
4880 &alloc_from_partial_attr
.attr
,
4881 &alloc_slab_attr
.attr
,
4882 &alloc_refill_attr
.attr
,
4883 &alloc_node_mismatch_attr
.attr
,
4884 &free_slab_attr
.attr
,
4885 &cpuslab_flush_attr
.attr
,
4886 &deactivate_full_attr
.attr
,
4887 &deactivate_empty_attr
.attr
,
4888 &deactivate_to_head_attr
.attr
,
4889 &deactivate_to_tail_attr
.attr
,
4890 &deactivate_remote_frees_attr
.attr
,
4891 &deactivate_bypass_attr
.attr
,
4892 &order_fallback_attr
.attr
,
4893 &cmpxchg_double_fail_attr
.attr
,
4894 &cmpxchg_double_cpu_fail_attr
.attr
,
4895 &cpu_partial_alloc_attr
.attr
,
4896 &cpu_partial_free_attr
.attr
,
4897 &cpu_partial_node_attr
.attr
,
4898 &cpu_partial_drain_attr
.attr
,
4900 #ifdef CONFIG_FAILSLAB
4901 &failslab_attr
.attr
,
4907 static struct attribute_group slab_attr_group
= {
4908 .attrs
= slab_attrs
,
4911 static ssize_t
slab_attr_show(struct kobject
*kobj
,
4912 struct attribute
*attr
,
4915 struct slab_attribute
*attribute
;
4916 struct kmem_cache
*s
;
4919 attribute
= to_slab_attr(attr
);
4922 if (!attribute
->show
)
4925 err
= attribute
->show(s
, buf
);
4930 static ssize_t
slab_attr_store(struct kobject
*kobj
,
4931 struct attribute
*attr
,
4932 const char *buf
, size_t len
)
4934 struct slab_attribute
*attribute
;
4935 struct kmem_cache
*s
;
4938 attribute
= to_slab_attr(attr
);
4941 if (!attribute
->store
)
4944 err
= attribute
->store(s
, buf
, len
);
4945 #ifdef CONFIG_MEMCG_KMEM
4946 if (slab_state
>= FULL
&& err
>= 0 && is_root_cache(s
)) {
4947 struct kmem_cache
*c
;
4949 mutex_lock(&slab_mutex
);
4950 if (s
->max_attr_size
< len
)
4951 s
->max_attr_size
= len
;
4954 * This is a best effort propagation, so this function's return
4955 * value will be determined by the parent cache only. This is
4956 * basically because not all attributes will have a well
4957 * defined semantics for rollbacks - most of the actions will
4958 * have permanent effects.
4960 * Returning the error value of any of the children that fail
4961 * is not 100 % defined, in the sense that users seeing the
4962 * error code won't be able to know anything about the state of
4965 * Only returning the error code for the parent cache at least
4966 * has well defined semantics. The cache being written to
4967 * directly either failed or succeeded, in which case we loop
4968 * through the descendants with best-effort propagation.
4970 for_each_memcg_cache(c
, s
)
4971 attribute
->store(c
, buf
, len
);
4972 mutex_unlock(&slab_mutex
);
4978 static void memcg_propagate_slab_attrs(struct kmem_cache
*s
)
4980 #ifdef CONFIG_MEMCG_KMEM
4982 char *buffer
= NULL
;
4983 struct kmem_cache
*root_cache
;
4985 if (is_root_cache(s
))
4988 root_cache
= s
->memcg_params
.root_cache
;
4991 * This mean this cache had no attribute written. Therefore, no point
4992 * in copying default values around
4994 if (!root_cache
->max_attr_size
)
4997 for (i
= 0; i
< ARRAY_SIZE(slab_attrs
); i
++) {
5000 struct slab_attribute
*attr
= to_slab_attr(slab_attrs
[i
]);
5002 if (!attr
|| !attr
->store
|| !attr
->show
)
5006 * It is really bad that we have to allocate here, so we will
5007 * do it only as a fallback. If we actually allocate, though,
5008 * we can just use the allocated buffer until the end.
5010 * Most of the slub attributes will tend to be very small in
5011 * size, but sysfs allows buffers up to a page, so they can
5012 * theoretically happen.
5016 else if (root_cache
->max_attr_size
< ARRAY_SIZE(mbuf
))
5019 buffer
= (char *) get_zeroed_page(GFP_KERNEL
);
5020 if (WARN_ON(!buffer
))
5025 attr
->show(root_cache
, buf
);
5026 attr
->store(s
, buf
, strlen(buf
));
5030 free_page((unsigned long)buffer
);
5034 static void kmem_cache_release(struct kobject
*k
)
5036 slab_kmem_cache_release(to_slab(k
));
5039 static const struct sysfs_ops slab_sysfs_ops
= {
5040 .show
= slab_attr_show
,
5041 .store
= slab_attr_store
,
5044 static struct kobj_type slab_ktype
= {
5045 .sysfs_ops
= &slab_sysfs_ops
,
5046 .release
= kmem_cache_release
,
5049 static int uevent_filter(struct kset
*kset
, struct kobject
*kobj
)
5051 struct kobj_type
*ktype
= get_ktype(kobj
);
5053 if (ktype
== &slab_ktype
)
5058 static const struct kset_uevent_ops slab_uevent_ops
= {
5059 .filter
= uevent_filter
,
5062 static struct kset
*slab_kset
;
5064 static inline struct kset
*cache_kset(struct kmem_cache
*s
)
5066 #ifdef CONFIG_MEMCG_KMEM
5067 if (!is_root_cache(s
))
5068 return s
->memcg_params
.root_cache
->memcg_kset
;
5073 #define ID_STR_LENGTH 64
5075 /* Create a unique string id for a slab cache:
5077 * Format :[flags-]size
5079 static char *create_unique_id(struct kmem_cache
*s
)
5081 char *name
= kmalloc(ID_STR_LENGTH
, GFP_KERNEL
);
5088 * First flags affecting slabcache operations. We will only
5089 * get here for aliasable slabs so we do not need to support
5090 * too many flags. The flags here must cover all flags that
5091 * are matched during merging to guarantee that the id is
5094 if (s
->flags
& SLAB_CACHE_DMA
)
5096 if (s
->flags
& SLAB_RECLAIM_ACCOUNT
)
5098 if (s
->flags
& SLAB_DEBUG_FREE
)
5100 if (!(s
->flags
& SLAB_NOTRACK
))
5104 p
+= sprintf(p
, "%07d", s
->size
);
5106 BUG_ON(p
> name
+ ID_STR_LENGTH
- 1);
5110 static int sysfs_slab_add(struct kmem_cache
*s
)
5114 int unmergeable
= slab_unmergeable(s
);
5118 * Slabcache can never be merged so we can use the name proper.
5119 * This is typically the case for debug situations. In that
5120 * case we can catch duplicate names easily.
5122 sysfs_remove_link(&slab_kset
->kobj
, s
->name
);
5126 * Create a unique name for the slab as a target
5129 name
= create_unique_id(s
);
5132 s
->kobj
.kset
= cache_kset(s
);
5133 err
= kobject_init_and_add(&s
->kobj
, &slab_ktype
, NULL
, "%s", name
);
5137 err
= sysfs_create_group(&s
->kobj
, &slab_attr_group
);
5141 #ifdef CONFIG_MEMCG_KMEM
5142 if (is_root_cache(s
)) {
5143 s
->memcg_kset
= kset_create_and_add("cgroup", NULL
, &s
->kobj
);
5144 if (!s
->memcg_kset
) {
5151 kobject_uevent(&s
->kobj
, KOBJ_ADD
);
5153 /* Setup first alias */
5154 sysfs_slab_alias(s
, s
->name
);
5161 kobject_del(&s
->kobj
);
5163 kobject_put(&s
->kobj
);
5167 void sysfs_slab_remove(struct kmem_cache
*s
)
5169 if (slab_state
< FULL
)
5171 * Sysfs has not been setup yet so no need to remove the
5176 #ifdef CONFIG_MEMCG_KMEM
5177 kset_unregister(s
->memcg_kset
);
5179 kobject_uevent(&s
->kobj
, KOBJ_REMOVE
);
5180 kobject_del(&s
->kobj
);
5181 kobject_put(&s
->kobj
);
5185 * Need to buffer aliases during bootup until sysfs becomes
5186 * available lest we lose that information.
5188 struct saved_alias
{
5189 struct kmem_cache
*s
;
5191 struct saved_alias
*next
;
5194 static struct saved_alias
*alias_list
;
5196 static int sysfs_slab_alias(struct kmem_cache
*s
, const char *name
)
5198 struct saved_alias
*al
;
5200 if (slab_state
== FULL
) {
5202 * If we have a leftover link then remove it.
5204 sysfs_remove_link(&slab_kset
->kobj
, name
);
5205 return sysfs_create_link(&slab_kset
->kobj
, &s
->kobj
, name
);
5208 al
= kmalloc(sizeof(struct saved_alias
), GFP_KERNEL
);
5214 al
->next
= alias_list
;
5219 static int __init
slab_sysfs_init(void)
5221 struct kmem_cache
*s
;
5224 mutex_lock(&slab_mutex
);
5226 slab_kset
= kset_create_and_add("slab", &slab_uevent_ops
, kernel_kobj
);
5228 mutex_unlock(&slab_mutex
);
5229 pr_err("Cannot register slab subsystem.\n");
5235 list_for_each_entry(s
, &slab_caches
, list
) {
5236 err
= sysfs_slab_add(s
);
5238 pr_err("SLUB: Unable to add boot slab %s to sysfs\n",
5242 while (alias_list
) {
5243 struct saved_alias
*al
= alias_list
;
5245 alias_list
= alias_list
->next
;
5246 err
= sysfs_slab_alias(al
->s
, al
->name
);
5248 pr_err("SLUB: Unable to add boot slab alias %s to sysfs\n",
5253 mutex_unlock(&slab_mutex
);
5258 __initcall(slab_sysfs_init
);
5259 #endif /* CONFIG_SYSFS */
5262 * The /proc/slabinfo ABI
5264 #ifdef CONFIG_SLABINFO
5265 void get_slabinfo(struct kmem_cache
*s
, struct slabinfo
*sinfo
)
5267 unsigned long nr_slabs
= 0;
5268 unsigned long nr_objs
= 0;
5269 unsigned long nr_free
= 0;
5271 struct kmem_cache_node
*n
;
5273 for_each_kmem_cache_node(s
, node
, n
) {
5274 nr_slabs
+= node_nr_slabs(n
);
5275 nr_objs
+= node_nr_objs(n
);
5276 nr_free
+= count_partial(n
, count_free
);
5279 sinfo
->active_objs
= nr_objs
- nr_free
;
5280 sinfo
->num_objs
= nr_objs
;
5281 sinfo
->active_slabs
= nr_slabs
;
5282 sinfo
->num_slabs
= nr_slabs
;
5283 sinfo
->objects_per_slab
= oo_objects(s
->oo
);
5284 sinfo
->cache_order
= oo_order(s
->oo
);
5287 void slabinfo_show_stats(struct seq_file
*m
, struct kmem_cache
*s
)
5291 ssize_t
slabinfo_write(struct file
*file
, const char __user
*buffer
,
5292 size_t count
, loff_t
*ppos
)
5296 #endif /* CONFIG_SLABINFO */