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
;
2013 oldpage
= this_cpu_read(s
->cpu_slab
->partial
);
2016 pobjects
= oldpage
->pobjects
;
2017 pages
= oldpage
->pages
;
2018 if (drain
&& pobjects
> s
->cpu_partial
) {
2019 unsigned long flags
;
2021 * partial array is full. Move the existing
2022 * set to the per node partial list.
2024 local_irq_save(flags
);
2025 unfreeze_partials(s
, this_cpu_ptr(s
->cpu_slab
));
2026 local_irq_restore(flags
);
2030 stat(s
, CPU_PARTIAL_DRAIN
);
2035 pobjects
+= page
->objects
- page
->inuse
;
2037 page
->pages
= pages
;
2038 page
->pobjects
= pobjects
;
2039 page
->next
= oldpage
;
2041 } while (this_cpu_cmpxchg(s
->cpu_slab
->partial
, oldpage
, page
)
2046 static inline void flush_slab(struct kmem_cache
*s
, struct kmem_cache_cpu
*c
)
2048 stat(s
, CPUSLAB_FLUSH
);
2049 deactivate_slab(s
, c
->page
, c
->freelist
);
2051 c
->tid
= next_tid(c
->tid
);
2059 * Called from IPI handler with interrupts disabled.
2061 static inline void __flush_cpu_slab(struct kmem_cache
*s
, int cpu
)
2063 struct kmem_cache_cpu
*c
= per_cpu_ptr(s
->cpu_slab
, cpu
);
2069 unfreeze_partials(s
, c
);
2073 static void flush_cpu_slab(void *d
)
2075 struct kmem_cache
*s
= d
;
2077 __flush_cpu_slab(s
, smp_processor_id());
2080 static bool has_cpu_slab(int cpu
, void *info
)
2082 struct kmem_cache
*s
= info
;
2083 struct kmem_cache_cpu
*c
= per_cpu_ptr(s
->cpu_slab
, cpu
);
2085 return c
->page
|| c
->partial
;
2088 static void flush_all(struct kmem_cache
*s
)
2090 on_each_cpu_cond(has_cpu_slab
, flush_cpu_slab
, s
, 1, GFP_ATOMIC
);
2094 * Check if the objects in a per cpu structure fit numa
2095 * locality expectations.
2097 static inline int node_match(struct page
*page
, int node
)
2100 if (!page
|| (node
!= NUMA_NO_NODE
&& page_to_nid(page
) != node
))
2106 #ifdef CONFIG_SLUB_DEBUG
2107 static int count_free(struct page
*page
)
2109 return page
->objects
- page
->inuse
;
2112 static inline unsigned long node_nr_objs(struct kmem_cache_node
*n
)
2114 return atomic_long_read(&n
->total_objects
);
2116 #endif /* CONFIG_SLUB_DEBUG */
2118 #if defined(CONFIG_SLUB_DEBUG) || defined(CONFIG_SYSFS)
2119 static unsigned long count_partial(struct kmem_cache_node
*n
,
2120 int (*get_count
)(struct page
*))
2122 unsigned long flags
;
2123 unsigned long x
= 0;
2126 spin_lock_irqsave(&n
->list_lock
, flags
);
2127 list_for_each_entry(page
, &n
->partial
, lru
)
2128 x
+= get_count(page
);
2129 spin_unlock_irqrestore(&n
->list_lock
, flags
);
2132 #endif /* CONFIG_SLUB_DEBUG || CONFIG_SYSFS */
2134 static noinline
void
2135 slab_out_of_memory(struct kmem_cache
*s
, gfp_t gfpflags
, int nid
)
2137 #ifdef CONFIG_SLUB_DEBUG
2138 static DEFINE_RATELIMIT_STATE(slub_oom_rs
, DEFAULT_RATELIMIT_INTERVAL
,
2139 DEFAULT_RATELIMIT_BURST
);
2141 struct kmem_cache_node
*n
;
2143 if ((gfpflags
& __GFP_NOWARN
) || !__ratelimit(&slub_oom_rs
))
2146 pr_warn("SLUB: Unable to allocate memory on node %d (gfp=0x%x)\n",
2148 pr_warn(" cache: %s, object size: %d, buffer size: %d, default order: %d, min order: %d\n",
2149 s
->name
, s
->object_size
, s
->size
, oo_order(s
->oo
),
2152 if (oo_order(s
->min
) > get_order(s
->object_size
))
2153 pr_warn(" %s debugging increased min order, use slub_debug=O to disable.\n",
2156 for_each_kmem_cache_node(s
, node
, n
) {
2157 unsigned long nr_slabs
;
2158 unsigned long nr_objs
;
2159 unsigned long nr_free
;
2161 nr_free
= count_partial(n
, count_free
);
2162 nr_slabs
= node_nr_slabs(n
);
2163 nr_objs
= node_nr_objs(n
);
2165 pr_warn(" node %d: slabs: %ld, objs: %ld, free: %ld\n",
2166 node
, nr_slabs
, nr_objs
, nr_free
);
2171 static inline void *new_slab_objects(struct kmem_cache
*s
, gfp_t flags
,
2172 int node
, struct kmem_cache_cpu
**pc
)
2175 struct kmem_cache_cpu
*c
= *pc
;
2178 freelist
= get_partial(s
, flags
, node
, c
);
2183 page
= new_slab(s
, flags
, node
);
2185 c
= raw_cpu_ptr(s
->cpu_slab
);
2190 * No other reference to the page yet so we can
2191 * muck around with it freely without cmpxchg
2193 freelist
= page
->freelist
;
2194 page
->freelist
= NULL
;
2196 stat(s
, ALLOC_SLAB
);
2205 static inline bool pfmemalloc_match(struct page
*page
, gfp_t gfpflags
)
2207 if (unlikely(PageSlabPfmemalloc(page
)))
2208 return gfp_pfmemalloc_allowed(gfpflags
);
2214 * Check the page->freelist of a page and either transfer the freelist to the
2215 * per cpu freelist or deactivate the page.
2217 * The page is still frozen if the return value is not NULL.
2219 * If this function returns NULL then the page has been unfrozen.
2221 * This function must be called with interrupt disabled.
2223 static inline void *get_freelist(struct kmem_cache
*s
, struct page
*page
)
2226 unsigned long counters
;
2230 freelist
= page
->freelist
;
2231 counters
= page
->counters
;
2233 new.counters
= counters
;
2234 VM_BUG_ON(!new.frozen
);
2236 new.inuse
= page
->objects
;
2237 new.frozen
= freelist
!= NULL
;
2239 } while (!__cmpxchg_double_slab(s
, page
,
2248 * Slow path. The lockless freelist is empty or we need to perform
2251 * Processing is still very fast if new objects have been freed to the
2252 * regular freelist. In that case we simply take over the regular freelist
2253 * as the lockless freelist and zap the regular freelist.
2255 * If that is not working then we fall back to the partial lists. We take the
2256 * first element of the freelist as the object to allocate now and move the
2257 * rest of the freelist to the lockless freelist.
2259 * And if we were unable to get a new slab from the partial slab lists then
2260 * we need to allocate a new slab. This is the slowest path since it involves
2261 * a call to the page allocator and the setup of a new slab.
2263 static void *__slab_alloc(struct kmem_cache
*s
, gfp_t gfpflags
, int node
,
2264 unsigned long addr
, struct kmem_cache_cpu
*c
)
2268 unsigned long flags
;
2270 local_irq_save(flags
);
2271 #ifdef CONFIG_PREEMPT
2273 * We may have been preempted and rescheduled on a different
2274 * cpu before disabling interrupts. Need to reload cpu area
2277 c
= this_cpu_ptr(s
->cpu_slab
);
2285 if (unlikely(!node_match(page
, node
))) {
2286 int searchnode
= node
;
2288 if (node
!= NUMA_NO_NODE
&& !node_present_pages(node
))
2289 searchnode
= node_to_mem_node(node
);
2291 if (unlikely(!node_match(page
, searchnode
))) {
2292 stat(s
, ALLOC_NODE_MISMATCH
);
2293 deactivate_slab(s
, page
, c
->freelist
);
2301 * By rights, we should be searching for a slab page that was
2302 * PFMEMALLOC but right now, we are losing the pfmemalloc
2303 * information when the page leaves the per-cpu allocator
2305 if (unlikely(!pfmemalloc_match(page
, gfpflags
))) {
2306 deactivate_slab(s
, page
, c
->freelist
);
2312 /* must check again c->freelist in case of cpu migration or IRQ */
2313 freelist
= c
->freelist
;
2317 freelist
= get_freelist(s
, page
);
2321 stat(s
, DEACTIVATE_BYPASS
);
2325 stat(s
, ALLOC_REFILL
);
2329 * freelist is pointing to the list of objects to be used.
2330 * page is pointing to the page from which the objects are obtained.
2331 * That page must be frozen for per cpu allocations to work.
2333 VM_BUG_ON(!c
->page
->frozen
);
2334 c
->freelist
= get_freepointer(s
, freelist
);
2335 c
->tid
= next_tid(c
->tid
);
2336 local_irq_restore(flags
);
2342 page
= c
->page
= c
->partial
;
2343 c
->partial
= page
->next
;
2344 stat(s
, CPU_PARTIAL_ALLOC
);
2349 freelist
= new_slab_objects(s
, gfpflags
, node
, &c
);
2351 if (unlikely(!freelist
)) {
2352 slab_out_of_memory(s
, gfpflags
, node
);
2353 local_irq_restore(flags
);
2358 if (likely(!kmem_cache_debug(s
) && pfmemalloc_match(page
, gfpflags
)))
2361 /* Only entered in the debug case */
2362 if (kmem_cache_debug(s
) &&
2363 !alloc_debug_processing(s
, page
, freelist
, addr
))
2364 goto new_slab
; /* Slab failed checks. Next slab needed */
2366 deactivate_slab(s
, page
, get_freepointer(s
, freelist
));
2369 local_irq_restore(flags
);
2374 * Inlined fastpath so that allocation functions (kmalloc, kmem_cache_alloc)
2375 * have the fastpath folded into their functions. So no function call
2376 * overhead for requests that can be satisfied on the fastpath.
2378 * The fastpath works by first checking if the lockless freelist can be used.
2379 * If not then __slab_alloc is called for slow processing.
2381 * Otherwise we can simply pick the next object from the lockless free list.
2383 static __always_inline
void *slab_alloc_node(struct kmem_cache
*s
,
2384 gfp_t gfpflags
, int node
, unsigned long addr
)
2387 struct kmem_cache_cpu
*c
;
2391 s
= slab_pre_alloc_hook(s
, gfpflags
);
2396 * Must read kmem_cache cpu data via this cpu ptr. Preemption is
2397 * enabled. We may switch back and forth between cpus while
2398 * reading from one cpu area. That does not matter as long
2399 * as we end up on the original cpu again when doing the cmpxchg.
2401 * We should guarantee that tid and kmem_cache are retrieved on
2402 * the same cpu. It could be different if CONFIG_PREEMPT so we need
2403 * to check if it is matched or not.
2406 tid
= this_cpu_read(s
->cpu_slab
->tid
);
2407 c
= raw_cpu_ptr(s
->cpu_slab
);
2408 } while (IS_ENABLED(CONFIG_PREEMPT
) && unlikely(tid
!= c
->tid
));
2411 * Irqless object alloc/free algorithm used here depends on sequence
2412 * of fetching cpu_slab's data. tid should be fetched before anything
2413 * on c to guarantee that object and page associated with previous tid
2414 * won't be used with current tid. If we fetch tid first, object and
2415 * page could be one associated with next tid and our alloc/free
2416 * request will be failed. In this case, we will retry. So, no problem.
2421 * The transaction ids are globally unique per cpu and per operation on
2422 * a per cpu queue. Thus they can be guarantee that the cmpxchg_double
2423 * occurs on the right processor and that there was no operation on the
2424 * linked list in between.
2427 object
= c
->freelist
;
2429 if (unlikely(!object
|| !node_match(page
, node
))) {
2430 object
= __slab_alloc(s
, gfpflags
, node
, addr
, c
);
2431 stat(s
, ALLOC_SLOWPATH
);
2433 void *next_object
= get_freepointer_safe(s
, object
);
2436 * The cmpxchg will only match if there was no additional
2437 * operation and if we are on the right processor.
2439 * The cmpxchg does the following atomically (without lock
2441 * 1. Relocate first pointer to the current per cpu area.
2442 * 2. Verify that tid and freelist have not been changed
2443 * 3. If they were not changed replace tid and freelist
2445 * Since this is without lock semantics the protection is only
2446 * against code executing on this cpu *not* from access by
2449 if (unlikely(!this_cpu_cmpxchg_double(
2450 s
->cpu_slab
->freelist
, s
->cpu_slab
->tid
,
2452 next_object
, next_tid(tid
)))) {
2454 note_cmpxchg_failure("slab_alloc", s
, tid
);
2457 prefetch_freepointer(s
, next_object
);
2458 stat(s
, ALLOC_FASTPATH
);
2461 if (unlikely(gfpflags
& __GFP_ZERO
) && object
)
2462 memset(object
, 0, s
->object_size
);
2464 slab_post_alloc_hook(s
, gfpflags
, object
);
2469 static __always_inline
void *slab_alloc(struct kmem_cache
*s
,
2470 gfp_t gfpflags
, unsigned long addr
)
2472 return slab_alloc_node(s
, gfpflags
, NUMA_NO_NODE
, addr
);
2475 void *kmem_cache_alloc(struct kmem_cache
*s
, gfp_t gfpflags
)
2477 void *ret
= slab_alloc(s
, gfpflags
, _RET_IP_
);
2479 trace_kmem_cache_alloc(_RET_IP_
, ret
, s
->object_size
,
2484 EXPORT_SYMBOL(kmem_cache_alloc
);
2486 #ifdef CONFIG_TRACING
2487 void *kmem_cache_alloc_trace(struct kmem_cache
*s
, gfp_t gfpflags
, size_t size
)
2489 void *ret
= slab_alloc(s
, gfpflags
, _RET_IP_
);
2490 trace_kmalloc(_RET_IP_
, ret
, size
, s
->size
, gfpflags
);
2493 EXPORT_SYMBOL(kmem_cache_alloc_trace
);
2497 void *kmem_cache_alloc_node(struct kmem_cache
*s
, gfp_t gfpflags
, int node
)
2499 void *ret
= slab_alloc_node(s
, gfpflags
, node
, _RET_IP_
);
2501 trace_kmem_cache_alloc_node(_RET_IP_
, ret
,
2502 s
->object_size
, s
->size
, gfpflags
, node
);
2506 EXPORT_SYMBOL(kmem_cache_alloc_node
);
2508 #ifdef CONFIG_TRACING
2509 void *kmem_cache_alloc_node_trace(struct kmem_cache
*s
,
2511 int node
, size_t size
)
2513 void *ret
= slab_alloc_node(s
, gfpflags
, node
, _RET_IP_
);
2515 trace_kmalloc_node(_RET_IP_
, ret
,
2516 size
, s
->size
, gfpflags
, node
);
2519 EXPORT_SYMBOL(kmem_cache_alloc_node_trace
);
2524 * Slow path handling. This may still be called frequently since objects
2525 * have a longer lifetime than the cpu slabs in most processing loads.
2527 * So we still attempt to reduce cache line usage. Just take the slab
2528 * lock and free the item. If there is no additional partial page
2529 * handling required then we can return immediately.
2531 static void __slab_free(struct kmem_cache
*s
, struct page
*page
,
2532 void *x
, unsigned long addr
)
2535 void **object
= (void *)x
;
2538 unsigned long counters
;
2539 struct kmem_cache_node
*n
= NULL
;
2540 unsigned long uninitialized_var(flags
);
2542 stat(s
, FREE_SLOWPATH
);
2544 if (kmem_cache_debug(s
) &&
2545 !(n
= free_debug_processing(s
, page
, x
, addr
, &flags
)))
2550 spin_unlock_irqrestore(&n
->list_lock
, flags
);
2553 prior
= page
->freelist
;
2554 counters
= page
->counters
;
2555 set_freepointer(s
, object
, prior
);
2556 new.counters
= counters
;
2557 was_frozen
= new.frozen
;
2559 if ((!new.inuse
|| !prior
) && !was_frozen
) {
2561 if (kmem_cache_has_cpu_partial(s
) && !prior
) {
2564 * Slab was on no list before and will be
2566 * We can defer the list move and instead
2571 } else { /* Needs to be taken off a list */
2573 n
= get_node(s
, page_to_nid(page
));
2575 * Speculatively acquire the list_lock.
2576 * If the cmpxchg does not succeed then we may
2577 * drop the list_lock without any processing.
2579 * Otherwise the list_lock will synchronize with
2580 * other processors updating the list of slabs.
2582 spin_lock_irqsave(&n
->list_lock
, flags
);
2587 } while (!cmpxchg_double_slab(s
, page
,
2589 object
, new.counters
,
2595 * If we just froze the page then put it onto the
2596 * per cpu partial list.
2598 if (new.frozen
&& !was_frozen
) {
2599 put_cpu_partial(s
, page
, 1);
2600 stat(s
, CPU_PARTIAL_FREE
);
2603 * The list lock was not taken therefore no list
2604 * activity can be necessary.
2607 stat(s
, FREE_FROZEN
);
2611 if (unlikely(!new.inuse
&& n
->nr_partial
>= s
->min_partial
))
2615 * Objects left in the slab. If it was not on the partial list before
2618 if (!kmem_cache_has_cpu_partial(s
) && unlikely(!prior
)) {
2619 if (kmem_cache_debug(s
))
2620 remove_full(s
, n
, page
);
2621 add_partial(n
, page
, DEACTIVATE_TO_TAIL
);
2622 stat(s
, FREE_ADD_PARTIAL
);
2624 spin_unlock_irqrestore(&n
->list_lock
, flags
);
2630 * Slab on the partial list.
2632 remove_partial(n
, page
);
2633 stat(s
, FREE_REMOVE_PARTIAL
);
2635 /* Slab must be on the full list */
2636 remove_full(s
, n
, page
);
2639 spin_unlock_irqrestore(&n
->list_lock
, flags
);
2641 discard_slab(s
, page
);
2645 * Fastpath with forced inlining to produce a kfree and kmem_cache_free that
2646 * can perform fastpath freeing without additional function calls.
2648 * The fastpath is only possible if we are freeing to the current cpu slab
2649 * of this processor. This typically the case if we have just allocated
2652 * If fastpath is not possible then fall back to __slab_free where we deal
2653 * with all sorts of special processing.
2655 static __always_inline
void slab_free(struct kmem_cache
*s
,
2656 struct page
*page
, void *x
, unsigned long addr
)
2658 void **object
= (void *)x
;
2659 struct kmem_cache_cpu
*c
;
2662 slab_free_hook(s
, x
);
2666 * Determine the currently cpus per cpu slab.
2667 * The cpu may change afterward. However that does not matter since
2668 * data is retrieved via this pointer. If we are on the same cpu
2669 * during the cmpxchg then the free will succedd.
2672 tid
= this_cpu_read(s
->cpu_slab
->tid
);
2673 c
= raw_cpu_ptr(s
->cpu_slab
);
2674 } while (IS_ENABLED(CONFIG_PREEMPT
) && unlikely(tid
!= c
->tid
));
2676 /* Same with comment on barrier() in slab_alloc_node() */
2679 if (likely(page
== c
->page
)) {
2680 set_freepointer(s
, object
, c
->freelist
);
2682 if (unlikely(!this_cpu_cmpxchg_double(
2683 s
->cpu_slab
->freelist
, s
->cpu_slab
->tid
,
2685 object
, next_tid(tid
)))) {
2687 note_cmpxchg_failure("slab_free", s
, tid
);
2690 stat(s
, FREE_FASTPATH
);
2692 __slab_free(s
, page
, x
, addr
);
2696 void kmem_cache_free(struct kmem_cache
*s
, void *x
)
2698 s
= cache_from_obj(s
, x
);
2701 slab_free(s
, virt_to_head_page(x
), x
, _RET_IP_
);
2702 trace_kmem_cache_free(_RET_IP_
, x
);
2704 EXPORT_SYMBOL(kmem_cache_free
);
2707 * Object placement in a slab is made very easy because we always start at
2708 * offset 0. If we tune the size of the object to the alignment then we can
2709 * get the required alignment by putting one properly sized object after
2712 * Notice that the allocation order determines the sizes of the per cpu
2713 * caches. Each processor has always one slab available for allocations.
2714 * Increasing the allocation order reduces the number of times that slabs
2715 * must be moved on and off the partial lists and is therefore a factor in
2720 * Mininum / Maximum order of slab pages. This influences locking overhead
2721 * and slab fragmentation. A higher order reduces the number of partial slabs
2722 * and increases the number of allocations possible without having to
2723 * take the list_lock.
2725 static int slub_min_order
;
2726 static int slub_max_order
= PAGE_ALLOC_COSTLY_ORDER
;
2727 static int slub_min_objects
;
2730 * Calculate the order of allocation given an slab object size.
2732 * The order of allocation has significant impact on performance and other
2733 * system components. Generally order 0 allocations should be preferred since
2734 * order 0 does not cause fragmentation in the page allocator. Larger objects
2735 * be problematic to put into order 0 slabs because there may be too much
2736 * unused space left. We go to a higher order if more than 1/16th of the slab
2739 * In order to reach satisfactory performance we must ensure that a minimum
2740 * number of objects is in one slab. Otherwise we may generate too much
2741 * activity on the partial lists which requires taking the list_lock. This is
2742 * less a concern for large slabs though which are rarely used.
2744 * slub_max_order specifies the order where we begin to stop considering the
2745 * number of objects in a slab as critical. If we reach slub_max_order then
2746 * we try to keep the page order as low as possible. So we accept more waste
2747 * of space in favor of a small page order.
2749 * Higher order allocations also allow the placement of more objects in a
2750 * slab and thereby reduce object handling overhead. If the user has
2751 * requested a higher mininum order then we start with that one instead of
2752 * the smallest order which will fit the object.
2754 static inline int slab_order(int size
, int min_objects
,
2755 int max_order
, int fract_leftover
, int reserved
)
2759 int min_order
= slub_min_order
;
2761 if (order_objects(min_order
, size
, reserved
) > MAX_OBJS_PER_PAGE
)
2762 return get_order(size
* MAX_OBJS_PER_PAGE
) - 1;
2764 for (order
= max(min_order
,
2765 fls(min_objects
* size
- 1) - PAGE_SHIFT
);
2766 order
<= max_order
; order
++) {
2768 unsigned long slab_size
= PAGE_SIZE
<< order
;
2770 if (slab_size
< min_objects
* size
+ reserved
)
2773 rem
= (slab_size
- reserved
) % size
;
2775 if (rem
<= slab_size
/ fract_leftover
)
2783 static inline int calculate_order(int size
, int reserved
)
2791 * Attempt to find best configuration for a slab. This
2792 * works by first attempting to generate a layout with
2793 * the best configuration and backing off gradually.
2795 * First we reduce the acceptable waste in a slab. Then
2796 * we reduce the minimum objects required in a slab.
2798 min_objects
= slub_min_objects
;
2800 min_objects
= 4 * (fls(nr_cpu_ids
) + 1);
2801 max_objects
= order_objects(slub_max_order
, size
, reserved
);
2802 min_objects
= min(min_objects
, max_objects
);
2804 while (min_objects
> 1) {
2806 while (fraction
>= 4) {
2807 order
= slab_order(size
, min_objects
,
2808 slub_max_order
, fraction
, reserved
);
2809 if (order
<= slub_max_order
)
2817 * We were unable to place multiple objects in a slab. Now
2818 * lets see if we can place a single object there.
2820 order
= slab_order(size
, 1, slub_max_order
, 1, reserved
);
2821 if (order
<= slub_max_order
)
2825 * Doh this slab cannot be placed using slub_max_order.
2827 order
= slab_order(size
, 1, MAX_ORDER
, 1, reserved
);
2828 if (order
< MAX_ORDER
)
2834 init_kmem_cache_node(struct kmem_cache_node
*n
)
2837 spin_lock_init(&n
->list_lock
);
2838 INIT_LIST_HEAD(&n
->partial
);
2839 #ifdef CONFIG_SLUB_DEBUG
2840 atomic_long_set(&n
->nr_slabs
, 0);
2841 atomic_long_set(&n
->total_objects
, 0);
2842 INIT_LIST_HEAD(&n
->full
);
2846 static inline int alloc_kmem_cache_cpus(struct kmem_cache
*s
)
2848 BUILD_BUG_ON(PERCPU_DYNAMIC_EARLY_SIZE
<
2849 KMALLOC_SHIFT_HIGH
* sizeof(struct kmem_cache_cpu
));
2852 * Must align to double word boundary for the double cmpxchg
2853 * instructions to work; see __pcpu_double_call_return_bool().
2855 s
->cpu_slab
= __alloc_percpu(sizeof(struct kmem_cache_cpu
),
2856 2 * sizeof(void *));
2861 init_kmem_cache_cpus(s
);
2866 static struct kmem_cache
*kmem_cache_node
;
2869 * No kmalloc_node yet so do it by hand. We know that this is the first
2870 * slab on the node for this slabcache. There are no concurrent accesses
2873 * Note that this function only works on the kmem_cache_node
2874 * when allocating for the kmem_cache_node. This is used for bootstrapping
2875 * memory on a fresh node that has no slab structures yet.
2877 static void early_kmem_cache_node_alloc(int node
)
2880 struct kmem_cache_node
*n
;
2882 BUG_ON(kmem_cache_node
->size
< sizeof(struct kmem_cache_node
));
2884 page
= new_slab(kmem_cache_node
, GFP_NOWAIT
, node
);
2887 if (page_to_nid(page
) != node
) {
2888 pr_err("SLUB: Unable to allocate memory from node %d\n", node
);
2889 pr_err("SLUB: Allocating a useless per node structure in order to be able to continue\n");
2894 page
->freelist
= get_freepointer(kmem_cache_node
, n
);
2897 kmem_cache_node
->node
[node
] = n
;
2898 #ifdef CONFIG_SLUB_DEBUG
2899 init_object(kmem_cache_node
, n
, SLUB_RED_ACTIVE
);
2900 init_tracking(kmem_cache_node
, n
);
2902 init_kmem_cache_node(n
);
2903 inc_slabs_node(kmem_cache_node
, node
, page
->objects
);
2906 * No locks need to be taken here as it has just been
2907 * initialized and there is no concurrent access.
2909 __add_partial(n
, page
, DEACTIVATE_TO_HEAD
);
2912 static void free_kmem_cache_nodes(struct kmem_cache
*s
)
2915 struct kmem_cache_node
*n
;
2917 for_each_kmem_cache_node(s
, node
, n
) {
2918 kmem_cache_free(kmem_cache_node
, n
);
2919 s
->node
[node
] = NULL
;
2923 static int init_kmem_cache_nodes(struct kmem_cache
*s
)
2927 for_each_node_state(node
, N_NORMAL_MEMORY
) {
2928 struct kmem_cache_node
*n
;
2930 if (slab_state
== DOWN
) {
2931 early_kmem_cache_node_alloc(node
);
2934 n
= kmem_cache_alloc_node(kmem_cache_node
,
2938 free_kmem_cache_nodes(s
);
2943 init_kmem_cache_node(n
);
2948 static void set_min_partial(struct kmem_cache
*s
, unsigned long min
)
2950 if (min
< MIN_PARTIAL
)
2952 else if (min
> MAX_PARTIAL
)
2954 s
->min_partial
= min
;
2958 * calculate_sizes() determines the order and the distribution of data within
2961 static int calculate_sizes(struct kmem_cache
*s
, int forced_order
)
2963 unsigned long flags
= s
->flags
;
2964 unsigned long size
= s
->object_size
;
2968 * Round up object size to the next word boundary. We can only
2969 * place the free pointer at word boundaries and this determines
2970 * the possible location of the free pointer.
2972 size
= ALIGN(size
, sizeof(void *));
2974 #ifdef CONFIG_SLUB_DEBUG
2976 * Determine if we can poison the object itself. If the user of
2977 * the slab may touch the object after free or before allocation
2978 * then we should never poison the object itself.
2980 if ((flags
& SLAB_POISON
) && !(flags
& SLAB_DESTROY_BY_RCU
) &&
2982 s
->flags
|= __OBJECT_POISON
;
2984 s
->flags
&= ~__OBJECT_POISON
;
2988 * If we are Redzoning then check if there is some space between the
2989 * end of the object and the free pointer. If not then add an
2990 * additional word to have some bytes to store Redzone information.
2992 if ((flags
& SLAB_RED_ZONE
) && size
== s
->object_size
)
2993 size
+= sizeof(void *);
2997 * With that we have determined the number of bytes in actual use
2998 * by the object. This is the potential offset to the free pointer.
3002 if (((flags
& (SLAB_DESTROY_BY_RCU
| SLAB_POISON
)) ||
3005 * Relocate free pointer after the object if it is not
3006 * permitted to overwrite the first word of the object on
3009 * This is the case if we do RCU, have a constructor or
3010 * destructor or are poisoning the objects.
3013 size
+= sizeof(void *);
3016 #ifdef CONFIG_SLUB_DEBUG
3017 if (flags
& SLAB_STORE_USER
)
3019 * Need to store information about allocs and frees after
3022 size
+= 2 * sizeof(struct track
);
3024 if (flags
& SLAB_RED_ZONE
)
3026 * Add some empty padding so that we can catch
3027 * overwrites from earlier objects rather than let
3028 * tracking information or the free pointer be
3029 * corrupted if a user writes before the start
3032 size
+= sizeof(void *);
3036 * SLUB stores one object immediately after another beginning from
3037 * offset 0. In order to align the objects we have to simply size
3038 * each object to conform to the alignment.
3040 size
= ALIGN(size
, s
->align
);
3042 if (forced_order
>= 0)
3043 order
= forced_order
;
3045 order
= calculate_order(size
, s
->reserved
);
3052 s
->allocflags
|= __GFP_COMP
;
3054 if (s
->flags
& SLAB_CACHE_DMA
)
3055 s
->allocflags
|= GFP_DMA
;
3057 if (s
->flags
& SLAB_RECLAIM_ACCOUNT
)
3058 s
->allocflags
|= __GFP_RECLAIMABLE
;
3061 * Determine the number of objects per slab
3063 s
->oo
= oo_make(order
, size
, s
->reserved
);
3064 s
->min
= oo_make(get_order(size
), size
, s
->reserved
);
3065 if (oo_objects(s
->oo
) > oo_objects(s
->max
))
3068 return !!oo_objects(s
->oo
);
3071 static int kmem_cache_open(struct kmem_cache
*s
, unsigned long flags
)
3073 s
->flags
= kmem_cache_flags(s
->size
, flags
, s
->name
, s
->ctor
);
3076 if (need_reserve_slab_rcu
&& (s
->flags
& SLAB_DESTROY_BY_RCU
))
3077 s
->reserved
= sizeof(struct rcu_head
);
3079 if (!calculate_sizes(s
, -1))
3081 if (disable_higher_order_debug
) {
3083 * Disable debugging flags that store metadata if the min slab
3086 if (get_order(s
->size
) > get_order(s
->object_size
)) {
3087 s
->flags
&= ~DEBUG_METADATA_FLAGS
;
3089 if (!calculate_sizes(s
, -1))
3094 #if defined(CONFIG_HAVE_CMPXCHG_DOUBLE) && \
3095 defined(CONFIG_HAVE_ALIGNED_STRUCT_PAGE)
3096 if (system_has_cmpxchg_double() && (s
->flags
& SLAB_DEBUG_FLAGS
) == 0)
3097 /* Enable fast mode */
3098 s
->flags
|= __CMPXCHG_DOUBLE
;
3102 * The larger the object size is, the more pages we want on the partial
3103 * list to avoid pounding the page allocator excessively.
3105 set_min_partial(s
, ilog2(s
->size
) / 2);
3108 * cpu_partial determined the maximum number of objects kept in the
3109 * per cpu partial lists of a processor.
3111 * Per cpu partial lists mainly contain slabs that just have one
3112 * object freed. If they are used for allocation then they can be
3113 * filled up again with minimal effort. The slab will never hit the
3114 * per node partial lists and therefore no locking will be required.
3116 * This setting also determines
3118 * A) The number of objects from per cpu partial slabs dumped to the
3119 * per node list when we reach the limit.
3120 * B) The number of objects in cpu partial slabs to extract from the
3121 * per node list when we run out of per cpu objects. We only fetch
3122 * 50% to keep some capacity around for frees.
3124 if (!kmem_cache_has_cpu_partial(s
))
3126 else if (s
->size
>= PAGE_SIZE
)
3128 else if (s
->size
>= 1024)
3130 else if (s
->size
>= 256)
3131 s
->cpu_partial
= 13;
3133 s
->cpu_partial
= 30;
3136 s
->remote_node_defrag_ratio
= 1000;
3138 if (!init_kmem_cache_nodes(s
))
3141 if (alloc_kmem_cache_cpus(s
))
3144 free_kmem_cache_nodes(s
);
3146 if (flags
& SLAB_PANIC
)
3147 panic("Cannot create slab %s size=%lu realsize=%u "
3148 "order=%u offset=%u flags=%lx\n",
3149 s
->name
, (unsigned long)s
->size
, s
->size
,
3150 oo_order(s
->oo
), s
->offset
, flags
);
3154 static void list_slab_objects(struct kmem_cache
*s
, struct page
*page
,
3157 #ifdef CONFIG_SLUB_DEBUG
3158 void *addr
= page_address(page
);
3160 unsigned long *map
= kzalloc(BITS_TO_LONGS(page
->objects
) *
3161 sizeof(long), GFP_ATOMIC
);
3164 slab_err(s
, page
, text
, s
->name
);
3167 get_map(s
, page
, map
);
3168 for_each_object(p
, s
, addr
, page
->objects
) {
3170 if (!test_bit(slab_index(p
, s
, addr
), map
)) {
3171 pr_err("INFO: Object 0x%p @offset=%tu\n", p
, p
- addr
);
3172 print_tracking(s
, p
);
3181 * Attempt to free all partial slabs on a node.
3182 * This is called from kmem_cache_close(). We must be the last thread
3183 * using the cache and therefore we do not need to lock anymore.
3185 static void free_partial(struct kmem_cache
*s
, struct kmem_cache_node
*n
)
3187 struct page
*page
, *h
;
3189 list_for_each_entry_safe(page
, h
, &n
->partial
, lru
) {
3191 __remove_partial(n
, page
);
3192 discard_slab(s
, page
);
3194 list_slab_objects(s
, page
,
3195 "Objects remaining in %s on kmem_cache_close()");
3201 * Release all resources used by a slab cache.
3203 static inline int kmem_cache_close(struct kmem_cache
*s
)
3206 struct kmem_cache_node
*n
;
3209 /* Attempt to free all objects */
3210 for_each_kmem_cache_node(s
, node
, n
) {
3212 if (n
->nr_partial
|| slabs_node(s
, node
))
3215 free_percpu(s
->cpu_slab
);
3216 free_kmem_cache_nodes(s
);
3220 int __kmem_cache_shutdown(struct kmem_cache
*s
)
3222 return kmem_cache_close(s
);
3225 /********************************************************************
3227 *******************************************************************/
3229 static int __init
setup_slub_min_order(char *str
)
3231 get_option(&str
, &slub_min_order
);
3236 __setup("slub_min_order=", setup_slub_min_order
);
3238 static int __init
setup_slub_max_order(char *str
)
3240 get_option(&str
, &slub_max_order
);
3241 slub_max_order
= min(slub_max_order
, MAX_ORDER
- 1);
3246 __setup("slub_max_order=", setup_slub_max_order
);
3248 static int __init
setup_slub_min_objects(char *str
)
3250 get_option(&str
, &slub_min_objects
);
3255 __setup("slub_min_objects=", setup_slub_min_objects
);
3257 void *__kmalloc(size_t size
, gfp_t flags
)
3259 struct kmem_cache
*s
;
3262 if (unlikely(size
> KMALLOC_MAX_CACHE_SIZE
))
3263 return kmalloc_large(size
, flags
);
3265 s
= kmalloc_slab(size
, flags
);
3267 if (unlikely(ZERO_OR_NULL_PTR(s
)))
3270 ret
= slab_alloc(s
, flags
, _RET_IP_
);
3272 trace_kmalloc(_RET_IP_
, ret
, size
, s
->size
, flags
);
3276 EXPORT_SYMBOL(__kmalloc
);
3279 static void *kmalloc_large_node(size_t size
, gfp_t flags
, int node
)
3284 flags
|= __GFP_COMP
| __GFP_NOTRACK
;
3285 page
= alloc_kmem_pages_node(node
, flags
, get_order(size
));
3287 ptr
= page_address(page
);
3289 kmalloc_large_node_hook(ptr
, size
, flags
);
3293 void *__kmalloc_node(size_t size
, gfp_t flags
, int node
)
3295 struct kmem_cache
*s
;
3298 if (unlikely(size
> KMALLOC_MAX_CACHE_SIZE
)) {
3299 ret
= kmalloc_large_node(size
, flags
, node
);
3301 trace_kmalloc_node(_RET_IP_
, ret
,
3302 size
, PAGE_SIZE
<< get_order(size
),
3308 s
= kmalloc_slab(size
, flags
);
3310 if (unlikely(ZERO_OR_NULL_PTR(s
)))
3313 ret
= slab_alloc_node(s
, flags
, node
, _RET_IP_
);
3315 trace_kmalloc_node(_RET_IP_
, ret
, size
, s
->size
, flags
, node
);
3319 EXPORT_SYMBOL(__kmalloc_node
);
3322 size_t ksize(const void *object
)
3326 if (unlikely(object
== ZERO_SIZE_PTR
))
3329 page
= virt_to_head_page(object
);
3331 if (unlikely(!PageSlab(page
))) {
3332 WARN_ON(!PageCompound(page
));
3333 return PAGE_SIZE
<< compound_order(page
);
3336 return slab_ksize(page
->slab_cache
);
3338 EXPORT_SYMBOL(ksize
);
3340 void kfree(const void *x
)
3343 void *object
= (void *)x
;
3345 trace_kfree(_RET_IP_
, x
);
3347 if (unlikely(ZERO_OR_NULL_PTR(x
)))
3350 page
= virt_to_head_page(x
);
3351 if (unlikely(!PageSlab(page
))) {
3352 BUG_ON(!PageCompound(page
));
3354 __free_kmem_pages(page
, compound_order(page
));
3357 slab_free(page
->slab_cache
, page
, object
, _RET_IP_
);
3359 EXPORT_SYMBOL(kfree
);
3362 * kmem_cache_shrink removes empty slabs from the partial lists and sorts
3363 * the remaining slabs by the number of items in use. The slabs with the
3364 * most items in use come first. New allocations will then fill those up
3365 * and thus they can be removed from the partial lists.
3367 * The slabs with the least items are placed last. This results in them
3368 * being allocated from last increasing the chance that the last objects
3369 * are freed in them.
3371 int __kmem_cache_shrink(struct kmem_cache
*s
)
3375 struct kmem_cache_node
*n
;
3378 int objects
= oo_objects(s
->max
);
3379 struct list_head
*slabs_by_inuse
=
3380 kmalloc(sizeof(struct list_head
) * objects
, GFP_KERNEL
);
3381 unsigned long flags
;
3383 if (!slabs_by_inuse
)
3387 for_each_kmem_cache_node(s
, node
, n
) {
3391 for (i
= 0; i
< objects
; i
++)
3392 INIT_LIST_HEAD(slabs_by_inuse
+ i
);
3394 spin_lock_irqsave(&n
->list_lock
, flags
);
3397 * Build lists indexed by the items in use in each slab.
3399 * Note that concurrent frees may occur while we hold the
3400 * list_lock. page->inuse here is the upper limit.
3402 list_for_each_entry_safe(page
, t
, &n
->partial
, lru
) {
3403 list_move(&page
->lru
, slabs_by_inuse
+ page
->inuse
);
3409 * Rebuild the partial list with the slabs filled up most
3410 * first and the least used slabs at the end.
3412 for (i
= objects
- 1; i
> 0; i
--)
3413 list_splice(slabs_by_inuse
+ i
, n
->partial
.prev
);
3415 spin_unlock_irqrestore(&n
->list_lock
, flags
);
3417 /* Release empty slabs */
3418 list_for_each_entry_safe(page
, t
, slabs_by_inuse
, lru
)
3419 discard_slab(s
, page
);
3422 kfree(slabs_by_inuse
);
3426 static int slab_mem_going_offline_callback(void *arg
)
3428 struct kmem_cache
*s
;
3430 mutex_lock(&slab_mutex
);
3431 list_for_each_entry(s
, &slab_caches
, list
)
3432 __kmem_cache_shrink(s
);
3433 mutex_unlock(&slab_mutex
);
3438 static void slab_mem_offline_callback(void *arg
)
3440 struct kmem_cache_node
*n
;
3441 struct kmem_cache
*s
;
3442 struct memory_notify
*marg
= arg
;
3445 offline_node
= marg
->status_change_nid_normal
;
3448 * If the node still has available memory. we need kmem_cache_node
3451 if (offline_node
< 0)
3454 mutex_lock(&slab_mutex
);
3455 list_for_each_entry(s
, &slab_caches
, list
) {
3456 n
= get_node(s
, offline_node
);
3459 * if n->nr_slabs > 0, slabs still exist on the node
3460 * that is going down. We were unable to free them,
3461 * and offline_pages() function shouldn't call this
3462 * callback. So, we must fail.
3464 BUG_ON(slabs_node(s
, offline_node
));
3466 s
->node
[offline_node
] = NULL
;
3467 kmem_cache_free(kmem_cache_node
, n
);
3470 mutex_unlock(&slab_mutex
);
3473 static int slab_mem_going_online_callback(void *arg
)
3475 struct kmem_cache_node
*n
;
3476 struct kmem_cache
*s
;
3477 struct memory_notify
*marg
= arg
;
3478 int nid
= marg
->status_change_nid_normal
;
3482 * If the node's memory is already available, then kmem_cache_node is
3483 * already created. Nothing to do.
3489 * We are bringing a node online. No memory is available yet. We must
3490 * allocate a kmem_cache_node structure in order to bring the node
3493 mutex_lock(&slab_mutex
);
3494 list_for_each_entry(s
, &slab_caches
, list
) {
3496 * XXX: kmem_cache_alloc_node will fallback to other nodes
3497 * since memory is not yet available from the node that
3500 n
= kmem_cache_alloc(kmem_cache_node
, GFP_KERNEL
);
3505 init_kmem_cache_node(n
);
3509 mutex_unlock(&slab_mutex
);
3513 static int slab_memory_callback(struct notifier_block
*self
,
3514 unsigned long action
, void *arg
)
3519 case MEM_GOING_ONLINE
:
3520 ret
= slab_mem_going_online_callback(arg
);
3522 case MEM_GOING_OFFLINE
:
3523 ret
= slab_mem_going_offline_callback(arg
);
3526 case MEM_CANCEL_ONLINE
:
3527 slab_mem_offline_callback(arg
);
3530 case MEM_CANCEL_OFFLINE
:
3534 ret
= notifier_from_errno(ret
);
3540 static struct notifier_block slab_memory_callback_nb
= {
3541 .notifier_call
= slab_memory_callback
,
3542 .priority
= SLAB_CALLBACK_PRI
,
3545 /********************************************************************
3546 * Basic setup of slabs
3547 *******************************************************************/
3550 * Used for early kmem_cache structures that were allocated using
3551 * the page allocator. Allocate them properly then fix up the pointers
3552 * that may be pointing to the wrong kmem_cache structure.
3555 static struct kmem_cache
* __init
bootstrap(struct kmem_cache
*static_cache
)
3558 struct kmem_cache
*s
= kmem_cache_zalloc(kmem_cache
, GFP_NOWAIT
);
3559 struct kmem_cache_node
*n
;
3561 memcpy(s
, static_cache
, kmem_cache
->object_size
);
3564 * This runs very early, and only the boot processor is supposed to be
3565 * up. Even if it weren't true, IRQs are not up so we couldn't fire
3568 __flush_cpu_slab(s
, smp_processor_id());
3569 for_each_kmem_cache_node(s
, node
, n
) {
3572 list_for_each_entry(p
, &n
->partial
, lru
)
3575 #ifdef CONFIG_SLUB_DEBUG
3576 list_for_each_entry(p
, &n
->full
, lru
)
3580 list_add(&s
->list
, &slab_caches
);
3584 void __init
kmem_cache_init(void)
3586 static __initdata
struct kmem_cache boot_kmem_cache
,
3587 boot_kmem_cache_node
;
3589 if (debug_guardpage_minorder())
3592 kmem_cache_node
= &boot_kmem_cache_node
;
3593 kmem_cache
= &boot_kmem_cache
;
3595 create_boot_cache(kmem_cache_node
, "kmem_cache_node",
3596 sizeof(struct kmem_cache_node
), SLAB_HWCACHE_ALIGN
);
3598 register_hotmemory_notifier(&slab_memory_callback_nb
);
3600 /* Able to allocate the per node structures */
3601 slab_state
= PARTIAL
;
3603 create_boot_cache(kmem_cache
, "kmem_cache",
3604 offsetof(struct kmem_cache
, node
) +
3605 nr_node_ids
* sizeof(struct kmem_cache_node
*),
3606 SLAB_HWCACHE_ALIGN
);
3608 kmem_cache
= bootstrap(&boot_kmem_cache
);
3611 * Allocate kmem_cache_node properly from the kmem_cache slab.
3612 * kmem_cache_node is separately allocated so no need to
3613 * update any list pointers.
3615 kmem_cache_node
= bootstrap(&boot_kmem_cache_node
);
3617 /* Now we can use the kmem_cache to allocate kmalloc slabs */
3618 create_kmalloc_caches(0);
3621 register_cpu_notifier(&slab_notifier
);
3624 pr_info("SLUB: HWalign=%d, Order=%d-%d, MinObjects=%d, CPUs=%d, Nodes=%d\n",
3626 slub_min_order
, slub_max_order
, slub_min_objects
,
3627 nr_cpu_ids
, nr_node_ids
);
3630 void __init
kmem_cache_init_late(void)
3635 __kmem_cache_alias(const char *name
, size_t size
, size_t align
,
3636 unsigned long flags
, void (*ctor
)(void *))
3638 struct kmem_cache
*s
;
3640 s
= find_mergeable(size
, align
, flags
, name
, ctor
);
3643 struct kmem_cache
*c
;
3648 * Adjust the object sizes so that we clear
3649 * the complete object on kzalloc.
3651 s
->object_size
= max(s
->object_size
, (int)size
);
3652 s
->inuse
= max_t(int, s
->inuse
, ALIGN(size
, sizeof(void *)));
3654 for_each_memcg_cache_index(i
) {
3655 c
= cache_from_memcg_idx(s
, i
);
3658 c
->object_size
= s
->object_size
;
3659 c
->inuse
= max_t(int, c
->inuse
,
3660 ALIGN(size
, sizeof(void *)));
3663 if (sysfs_slab_alias(s
, name
)) {
3672 int __kmem_cache_create(struct kmem_cache
*s
, unsigned long flags
)
3676 err
= kmem_cache_open(s
, flags
);
3680 /* Mutex is not taken during early boot */
3681 if (slab_state
<= UP
)
3684 memcg_propagate_slab_attrs(s
);
3685 err
= sysfs_slab_add(s
);
3687 kmem_cache_close(s
);
3694 * Use the cpu notifier to insure that the cpu slabs are flushed when
3697 static int slab_cpuup_callback(struct notifier_block
*nfb
,
3698 unsigned long action
, void *hcpu
)
3700 long cpu
= (long)hcpu
;
3701 struct kmem_cache
*s
;
3702 unsigned long flags
;
3705 case CPU_UP_CANCELED
:
3706 case CPU_UP_CANCELED_FROZEN
:
3708 case CPU_DEAD_FROZEN
:
3709 mutex_lock(&slab_mutex
);
3710 list_for_each_entry(s
, &slab_caches
, list
) {
3711 local_irq_save(flags
);
3712 __flush_cpu_slab(s
, cpu
);
3713 local_irq_restore(flags
);
3715 mutex_unlock(&slab_mutex
);
3723 static struct notifier_block slab_notifier
= {
3724 .notifier_call
= slab_cpuup_callback
3729 void *__kmalloc_track_caller(size_t size
, gfp_t gfpflags
, unsigned long caller
)
3731 struct kmem_cache
*s
;
3734 if (unlikely(size
> KMALLOC_MAX_CACHE_SIZE
))
3735 return kmalloc_large(size
, gfpflags
);
3737 s
= kmalloc_slab(size
, gfpflags
);
3739 if (unlikely(ZERO_OR_NULL_PTR(s
)))
3742 ret
= slab_alloc(s
, gfpflags
, caller
);
3744 /* Honor the call site pointer we received. */
3745 trace_kmalloc(caller
, ret
, size
, s
->size
, gfpflags
);
3751 void *__kmalloc_node_track_caller(size_t size
, gfp_t gfpflags
,
3752 int node
, unsigned long caller
)
3754 struct kmem_cache
*s
;
3757 if (unlikely(size
> KMALLOC_MAX_CACHE_SIZE
)) {
3758 ret
= kmalloc_large_node(size
, gfpflags
, node
);
3760 trace_kmalloc_node(caller
, ret
,
3761 size
, PAGE_SIZE
<< get_order(size
),
3767 s
= kmalloc_slab(size
, gfpflags
);
3769 if (unlikely(ZERO_OR_NULL_PTR(s
)))
3772 ret
= slab_alloc_node(s
, gfpflags
, node
, caller
);
3774 /* Honor the call site pointer we received. */
3775 trace_kmalloc_node(caller
, ret
, size
, s
->size
, gfpflags
, node
);
3782 static int count_inuse(struct page
*page
)
3787 static int count_total(struct page
*page
)
3789 return page
->objects
;
3793 #ifdef CONFIG_SLUB_DEBUG
3794 static int validate_slab(struct kmem_cache
*s
, struct page
*page
,
3798 void *addr
= page_address(page
);
3800 if (!check_slab(s
, page
) ||
3801 !on_freelist(s
, page
, NULL
))
3804 /* Now we know that a valid freelist exists */
3805 bitmap_zero(map
, page
->objects
);
3807 get_map(s
, page
, map
);
3808 for_each_object(p
, s
, addr
, page
->objects
) {
3809 if (test_bit(slab_index(p
, s
, addr
), map
))
3810 if (!check_object(s
, page
, p
, SLUB_RED_INACTIVE
))
3814 for_each_object(p
, s
, addr
, page
->objects
)
3815 if (!test_bit(slab_index(p
, s
, addr
), map
))
3816 if (!check_object(s
, page
, p
, SLUB_RED_ACTIVE
))
3821 static void validate_slab_slab(struct kmem_cache
*s
, struct page
*page
,
3825 validate_slab(s
, page
, map
);
3829 static int validate_slab_node(struct kmem_cache
*s
,
3830 struct kmem_cache_node
*n
, unsigned long *map
)
3832 unsigned long count
= 0;
3834 unsigned long flags
;
3836 spin_lock_irqsave(&n
->list_lock
, flags
);
3838 list_for_each_entry(page
, &n
->partial
, lru
) {
3839 validate_slab_slab(s
, page
, map
);
3842 if (count
!= n
->nr_partial
)
3843 pr_err("SLUB %s: %ld partial slabs counted but counter=%ld\n",
3844 s
->name
, count
, n
->nr_partial
);
3846 if (!(s
->flags
& SLAB_STORE_USER
))
3849 list_for_each_entry(page
, &n
->full
, lru
) {
3850 validate_slab_slab(s
, page
, map
);
3853 if (count
!= atomic_long_read(&n
->nr_slabs
))
3854 pr_err("SLUB: %s %ld slabs counted but counter=%ld\n",
3855 s
->name
, count
, atomic_long_read(&n
->nr_slabs
));
3858 spin_unlock_irqrestore(&n
->list_lock
, flags
);
3862 static long validate_slab_cache(struct kmem_cache
*s
)
3865 unsigned long count
= 0;
3866 unsigned long *map
= kmalloc(BITS_TO_LONGS(oo_objects(s
->max
)) *
3867 sizeof(unsigned long), GFP_KERNEL
);
3868 struct kmem_cache_node
*n
;
3874 for_each_kmem_cache_node(s
, node
, n
)
3875 count
+= validate_slab_node(s
, n
, map
);
3880 * Generate lists of code addresses where slabcache objects are allocated
3885 unsigned long count
;
3892 DECLARE_BITMAP(cpus
, NR_CPUS
);
3898 unsigned long count
;
3899 struct location
*loc
;
3902 static void free_loc_track(struct loc_track
*t
)
3905 free_pages((unsigned long)t
->loc
,
3906 get_order(sizeof(struct location
) * t
->max
));
3909 static int alloc_loc_track(struct loc_track
*t
, unsigned long max
, gfp_t flags
)
3914 order
= get_order(sizeof(struct location
) * max
);
3916 l
= (void *)__get_free_pages(flags
, order
);
3921 memcpy(l
, t
->loc
, sizeof(struct location
) * t
->count
);
3929 static int add_location(struct loc_track
*t
, struct kmem_cache
*s
,
3930 const struct track
*track
)
3932 long start
, end
, pos
;
3934 unsigned long caddr
;
3935 unsigned long age
= jiffies
- track
->when
;
3941 pos
= start
+ (end
- start
+ 1) / 2;
3944 * There is nothing at "end". If we end up there
3945 * we need to add something to before end.
3950 caddr
= t
->loc
[pos
].addr
;
3951 if (track
->addr
== caddr
) {
3957 if (age
< l
->min_time
)
3959 if (age
> l
->max_time
)
3962 if (track
->pid
< l
->min_pid
)
3963 l
->min_pid
= track
->pid
;
3964 if (track
->pid
> l
->max_pid
)
3965 l
->max_pid
= track
->pid
;
3967 cpumask_set_cpu(track
->cpu
,
3968 to_cpumask(l
->cpus
));
3970 node_set(page_to_nid(virt_to_page(track
)), l
->nodes
);
3974 if (track
->addr
< caddr
)
3981 * Not found. Insert new tracking element.
3983 if (t
->count
>= t
->max
&& !alloc_loc_track(t
, 2 * t
->max
, GFP_ATOMIC
))
3989 (t
->count
- pos
) * sizeof(struct location
));
3992 l
->addr
= track
->addr
;
3996 l
->min_pid
= track
->pid
;
3997 l
->max_pid
= track
->pid
;
3998 cpumask_clear(to_cpumask(l
->cpus
));
3999 cpumask_set_cpu(track
->cpu
, to_cpumask(l
->cpus
));
4000 nodes_clear(l
->nodes
);
4001 node_set(page_to_nid(virt_to_page(track
)), l
->nodes
);
4005 static void process_slab(struct loc_track
*t
, struct kmem_cache
*s
,
4006 struct page
*page
, enum track_item alloc
,
4009 void *addr
= page_address(page
);
4012 bitmap_zero(map
, page
->objects
);
4013 get_map(s
, page
, map
);
4015 for_each_object(p
, s
, addr
, page
->objects
)
4016 if (!test_bit(slab_index(p
, s
, addr
), map
))
4017 add_location(t
, s
, get_track(s
, p
, alloc
));
4020 static int list_locations(struct kmem_cache
*s
, char *buf
,
4021 enum track_item alloc
)
4025 struct loc_track t
= { 0, 0, NULL
};
4027 unsigned long *map
= kmalloc(BITS_TO_LONGS(oo_objects(s
->max
)) *
4028 sizeof(unsigned long), GFP_KERNEL
);
4029 struct kmem_cache_node
*n
;
4031 if (!map
|| !alloc_loc_track(&t
, PAGE_SIZE
/ sizeof(struct location
),
4034 return sprintf(buf
, "Out of memory\n");
4036 /* Push back cpu slabs */
4039 for_each_kmem_cache_node(s
, node
, n
) {
4040 unsigned long flags
;
4043 if (!atomic_long_read(&n
->nr_slabs
))
4046 spin_lock_irqsave(&n
->list_lock
, flags
);
4047 list_for_each_entry(page
, &n
->partial
, lru
)
4048 process_slab(&t
, s
, page
, alloc
, map
);
4049 list_for_each_entry(page
, &n
->full
, lru
)
4050 process_slab(&t
, s
, page
, alloc
, map
);
4051 spin_unlock_irqrestore(&n
->list_lock
, flags
);
4054 for (i
= 0; i
< t
.count
; i
++) {
4055 struct location
*l
= &t
.loc
[i
];
4057 if (len
> PAGE_SIZE
- KSYM_SYMBOL_LEN
- 100)
4059 len
+= sprintf(buf
+ len
, "%7ld ", l
->count
);
4062 len
+= sprintf(buf
+ len
, "%pS", (void *)l
->addr
);
4064 len
+= sprintf(buf
+ len
, "<not-available>");
4066 if (l
->sum_time
!= l
->min_time
) {
4067 len
+= sprintf(buf
+ len
, " age=%ld/%ld/%ld",
4069 (long)div_u64(l
->sum_time
, l
->count
),
4072 len
+= sprintf(buf
+ len
, " age=%ld",
4075 if (l
->min_pid
!= l
->max_pid
)
4076 len
+= sprintf(buf
+ len
, " pid=%ld-%ld",
4077 l
->min_pid
, l
->max_pid
);
4079 len
+= sprintf(buf
+ len
, " pid=%ld",
4082 if (num_online_cpus() > 1 &&
4083 !cpumask_empty(to_cpumask(l
->cpus
)) &&
4084 len
< PAGE_SIZE
- 60) {
4085 len
+= sprintf(buf
+ len
, " cpus=");
4086 len
+= cpulist_scnprintf(buf
+ len
,
4087 PAGE_SIZE
- len
- 50,
4088 to_cpumask(l
->cpus
));
4091 if (nr_online_nodes
> 1 && !nodes_empty(l
->nodes
) &&
4092 len
< PAGE_SIZE
- 60) {
4093 len
+= sprintf(buf
+ len
, " nodes=");
4094 len
+= nodelist_scnprintf(buf
+ len
,
4095 PAGE_SIZE
- len
- 50,
4099 len
+= sprintf(buf
+ len
, "\n");
4105 len
+= sprintf(buf
, "No data\n");
4110 #ifdef SLUB_RESILIENCY_TEST
4111 static void __init
resiliency_test(void)
4115 BUILD_BUG_ON(KMALLOC_MIN_SIZE
> 16 || KMALLOC_SHIFT_HIGH
< 10);
4117 pr_err("SLUB resiliency testing\n");
4118 pr_err("-----------------------\n");
4119 pr_err("A. Corruption after allocation\n");
4121 p
= kzalloc(16, GFP_KERNEL
);
4123 pr_err("\n1. kmalloc-16: Clobber Redzone/next pointer 0x12->0x%p\n\n",
4126 validate_slab_cache(kmalloc_caches
[4]);
4128 /* Hmmm... The next two are dangerous */
4129 p
= kzalloc(32, GFP_KERNEL
);
4130 p
[32 + sizeof(void *)] = 0x34;
4131 pr_err("\n2. kmalloc-32: Clobber next pointer/next slab 0x34 -> -0x%p\n",
4133 pr_err("If allocated object is overwritten then not detectable\n\n");
4135 validate_slab_cache(kmalloc_caches
[5]);
4136 p
= kzalloc(64, GFP_KERNEL
);
4137 p
+= 64 + (get_cycles() & 0xff) * sizeof(void *);
4139 pr_err("\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n",
4141 pr_err("If allocated object is overwritten then not detectable\n\n");
4142 validate_slab_cache(kmalloc_caches
[6]);
4144 pr_err("\nB. Corruption after free\n");
4145 p
= kzalloc(128, GFP_KERNEL
);
4148 pr_err("1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p
);
4149 validate_slab_cache(kmalloc_caches
[7]);
4151 p
= kzalloc(256, GFP_KERNEL
);
4154 pr_err("\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n", p
);
4155 validate_slab_cache(kmalloc_caches
[8]);
4157 p
= kzalloc(512, GFP_KERNEL
);
4160 pr_err("\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p
);
4161 validate_slab_cache(kmalloc_caches
[9]);
4165 static void resiliency_test(void) {};
4170 enum slab_stat_type
{
4171 SL_ALL
, /* All slabs */
4172 SL_PARTIAL
, /* Only partially allocated slabs */
4173 SL_CPU
, /* Only slabs used for cpu caches */
4174 SL_OBJECTS
, /* Determine allocated objects not slabs */
4175 SL_TOTAL
/* Determine object capacity not slabs */
4178 #define SO_ALL (1 << SL_ALL)
4179 #define SO_PARTIAL (1 << SL_PARTIAL)
4180 #define SO_CPU (1 << SL_CPU)
4181 #define SO_OBJECTS (1 << SL_OBJECTS)
4182 #define SO_TOTAL (1 << SL_TOTAL)
4184 static ssize_t
show_slab_objects(struct kmem_cache
*s
,
4185 char *buf
, unsigned long flags
)
4187 unsigned long total
= 0;
4190 unsigned long *nodes
;
4192 nodes
= kzalloc(sizeof(unsigned long) * nr_node_ids
, GFP_KERNEL
);
4196 if (flags
& SO_CPU
) {
4199 for_each_possible_cpu(cpu
) {
4200 struct kmem_cache_cpu
*c
= per_cpu_ptr(s
->cpu_slab
,
4205 page
= ACCESS_ONCE(c
->page
);
4209 node
= page_to_nid(page
);
4210 if (flags
& SO_TOTAL
)
4212 else if (flags
& SO_OBJECTS
)
4220 page
= ACCESS_ONCE(c
->partial
);
4222 node
= page_to_nid(page
);
4223 if (flags
& SO_TOTAL
)
4225 else if (flags
& SO_OBJECTS
)
4236 #ifdef CONFIG_SLUB_DEBUG
4237 if (flags
& SO_ALL
) {
4238 struct kmem_cache_node
*n
;
4240 for_each_kmem_cache_node(s
, node
, n
) {
4242 if (flags
& SO_TOTAL
)
4243 x
= atomic_long_read(&n
->total_objects
);
4244 else if (flags
& SO_OBJECTS
)
4245 x
= atomic_long_read(&n
->total_objects
) -
4246 count_partial(n
, count_free
);
4248 x
= atomic_long_read(&n
->nr_slabs
);
4255 if (flags
& SO_PARTIAL
) {
4256 struct kmem_cache_node
*n
;
4258 for_each_kmem_cache_node(s
, node
, n
) {
4259 if (flags
& SO_TOTAL
)
4260 x
= count_partial(n
, count_total
);
4261 else if (flags
& SO_OBJECTS
)
4262 x
= count_partial(n
, count_inuse
);
4269 x
= sprintf(buf
, "%lu", total
);
4271 for (node
= 0; node
< nr_node_ids
; node
++)
4273 x
+= sprintf(buf
+ x
, " N%d=%lu",
4278 return x
+ sprintf(buf
+ x
, "\n");
4281 #ifdef CONFIG_SLUB_DEBUG
4282 static int any_slab_objects(struct kmem_cache
*s
)
4285 struct kmem_cache_node
*n
;
4287 for_each_kmem_cache_node(s
, node
, n
)
4288 if (atomic_long_read(&n
->total_objects
))
4295 #define to_slab_attr(n) container_of(n, struct slab_attribute, attr)
4296 #define to_slab(n) container_of(n, struct kmem_cache, kobj)
4298 struct slab_attribute
{
4299 struct attribute attr
;
4300 ssize_t (*show
)(struct kmem_cache
*s
, char *buf
);
4301 ssize_t (*store
)(struct kmem_cache
*s
, const char *x
, size_t count
);
4304 #define SLAB_ATTR_RO(_name) \
4305 static struct slab_attribute _name##_attr = \
4306 __ATTR(_name, 0400, _name##_show, NULL)
4308 #define SLAB_ATTR(_name) \
4309 static struct slab_attribute _name##_attr = \
4310 __ATTR(_name, 0600, _name##_show, _name##_store)
4312 static ssize_t
slab_size_show(struct kmem_cache
*s
, char *buf
)
4314 return sprintf(buf
, "%d\n", s
->size
);
4316 SLAB_ATTR_RO(slab_size
);
4318 static ssize_t
align_show(struct kmem_cache
*s
, char *buf
)
4320 return sprintf(buf
, "%d\n", s
->align
);
4322 SLAB_ATTR_RO(align
);
4324 static ssize_t
object_size_show(struct kmem_cache
*s
, char *buf
)
4326 return sprintf(buf
, "%d\n", s
->object_size
);
4328 SLAB_ATTR_RO(object_size
);
4330 static ssize_t
objs_per_slab_show(struct kmem_cache
*s
, char *buf
)
4332 return sprintf(buf
, "%d\n", oo_objects(s
->oo
));
4334 SLAB_ATTR_RO(objs_per_slab
);
4336 static ssize_t
order_store(struct kmem_cache
*s
,
4337 const char *buf
, size_t length
)
4339 unsigned long order
;
4342 err
= kstrtoul(buf
, 10, &order
);
4346 if (order
> slub_max_order
|| order
< slub_min_order
)
4349 calculate_sizes(s
, order
);
4353 static ssize_t
order_show(struct kmem_cache
*s
, char *buf
)
4355 return sprintf(buf
, "%d\n", oo_order(s
->oo
));
4359 static ssize_t
min_partial_show(struct kmem_cache
*s
, char *buf
)
4361 return sprintf(buf
, "%lu\n", s
->min_partial
);
4364 static ssize_t
min_partial_store(struct kmem_cache
*s
, const char *buf
,
4370 err
= kstrtoul(buf
, 10, &min
);
4374 set_min_partial(s
, min
);
4377 SLAB_ATTR(min_partial
);
4379 static ssize_t
cpu_partial_show(struct kmem_cache
*s
, char *buf
)
4381 return sprintf(buf
, "%u\n", s
->cpu_partial
);
4384 static ssize_t
cpu_partial_store(struct kmem_cache
*s
, const char *buf
,
4387 unsigned long objects
;
4390 err
= kstrtoul(buf
, 10, &objects
);
4393 if (objects
&& !kmem_cache_has_cpu_partial(s
))
4396 s
->cpu_partial
= objects
;
4400 SLAB_ATTR(cpu_partial
);
4402 static ssize_t
ctor_show(struct kmem_cache
*s
, char *buf
)
4406 return sprintf(buf
, "%pS\n", s
->ctor
);
4410 static ssize_t
aliases_show(struct kmem_cache
*s
, char *buf
)
4412 return sprintf(buf
, "%d\n", s
->refcount
< 0 ? 0 : s
->refcount
- 1);
4414 SLAB_ATTR_RO(aliases
);
4416 static ssize_t
partial_show(struct kmem_cache
*s
, char *buf
)
4418 return show_slab_objects(s
, buf
, SO_PARTIAL
);
4420 SLAB_ATTR_RO(partial
);
4422 static ssize_t
cpu_slabs_show(struct kmem_cache
*s
, char *buf
)
4424 return show_slab_objects(s
, buf
, SO_CPU
);
4426 SLAB_ATTR_RO(cpu_slabs
);
4428 static ssize_t
objects_show(struct kmem_cache
*s
, char *buf
)
4430 return show_slab_objects(s
, buf
, SO_ALL
|SO_OBJECTS
);
4432 SLAB_ATTR_RO(objects
);
4434 static ssize_t
objects_partial_show(struct kmem_cache
*s
, char *buf
)
4436 return show_slab_objects(s
, buf
, SO_PARTIAL
|SO_OBJECTS
);
4438 SLAB_ATTR_RO(objects_partial
);
4440 static ssize_t
slabs_cpu_partial_show(struct kmem_cache
*s
, char *buf
)
4447 for_each_online_cpu(cpu
) {
4448 struct page
*page
= per_cpu_ptr(s
->cpu_slab
, cpu
)->partial
;
4451 pages
+= page
->pages
;
4452 objects
+= page
->pobjects
;
4456 len
= sprintf(buf
, "%d(%d)", objects
, pages
);
4459 for_each_online_cpu(cpu
) {
4460 struct page
*page
= per_cpu_ptr(s
->cpu_slab
, cpu
) ->partial
;
4462 if (page
&& len
< PAGE_SIZE
- 20)
4463 len
+= sprintf(buf
+ len
, " C%d=%d(%d)", cpu
,
4464 page
->pobjects
, page
->pages
);
4467 return len
+ sprintf(buf
+ len
, "\n");
4469 SLAB_ATTR_RO(slabs_cpu_partial
);
4471 static ssize_t
reclaim_account_show(struct kmem_cache
*s
, char *buf
)
4473 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_RECLAIM_ACCOUNT
));
4476 static ssize_t
reclaim_account_store(struct kmem_cache
*s
,
4477 const char *buf
, size_t length
)
4479 s
->flags
&= ~SLAB_RECLAIM_ACCOUNT
;
4481 s
->flags
|= SLAB_RECLAIM_ACCOUNT
;
4484 SLAB_ATTR(reclaim_account
);
4486 static ssize_t
hwcache_align_show(struct kmem_cache
*s
, char *buf
)
4488 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_HWCACHE_ALIGN
));
4490 SLAB_ATTR_RO(hwcache_align
);
4492 #ifdef CONFIG_ZONE_DMA
4493 static ssize_t
cache_dma_show(struct kmem_cache
*s
, char *buf
)
4495 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_CACHE_DMA
));
4497 SLAB_ATTR_RO(cache_dma
);
4500 static ssize_t
destroy_by_rcu_show(struct kmem_cache
*s
, char *buf
)
4502 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_DESTROY_BY_RCU
));
4504 SLAB_ATTR_RO(destroy_by_rcu
);
4506 static ssize_t
reserved_show(struct kmem_cache
*s
, char *buf
)
4508 return sprintf(buf
, "%d\n", s
->reserved
);
4510 SLAB_ATTR_RO(reserved
);
4512 #ifdef CONFIG_SLUB_DEBUG
4513 static ssize_t
slabs_show(struct kmem_cache
*s
, char *buf
)
4515 return show_slab_objects(s
, buf
, SO_ALL
);
4517 SLAB_ATTR_RO(slabs
);
4519 static ssize_t
total_objects_show(struct kmem_cache
*s
, char *buf
)
4521 return show_slab_objects(s
, buf
, SO_ALL
|SO_TOTAL
);
4523 SLAB_ATTR_RO(total_objects
);
4525 static ssize_t
sanity_checks_show(struct kmem_cache
*s
, char *buf
)
4527 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_DEBUG_FREE
));
4530 static ssize_t
sanity_checks_store(struct kmem_cache
*s
,
4531 const char *buf
, size_t length
)
4533 s
->flags
&= ~SLAB_DEBUG_FREE
;
4534 if (buf
[0] == '1') {
4535 s
->flags
&= ~__CMPXCHG_DOUBLE
;
4536 s
->flags
|= SLAB_DEBUG_FREE
;
4540 SLAB_ATTR(sanity_checks
);
4542 static ssize_t
trace_show(struct kmem_cache
*s
, char *buf
)
4544 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_TRACE
));
4547 static ssize_t
trace_store(struct kmem_cache
*s
, const char *buf
,
4551 * Tracing a merged cache is going to give confusing results
4552 * as well as cause other issues like converting a mergeable
4553 * cache into an umergeable one.
4555 if (s
->refcount
> 1)
4558 s
->flags
&= ~SLAB_TRACE
;
4559 if (buf
[0] == '1') {
4560 s
->flags
&= ~__CMPXCHG_DOUBLE
;
4561 s
->flags
|= SLAB_TRACE
;
4567 static ssize_t
red_zone_show(struct kmem_cache
*s
, char *buf
)
4569 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_RED_ZONE
));
4572 static ssize_t
red_zone_store(struct kmem_cache
*s
,
4573 const char *buf
, size_t length
)
4575 if (any_slab_objects(s
))
4578 s
->flags
&= ~SLAB_RED_ZONE
;
4579 if (buf
[0] == '1') {
4580 s
->flags
&= ~__CMPXCHG_DOUBLE
;
4581 s
->flags
|= SLAB_RED_ZONE
;
4583 calculate_sizes(s
, -1);
4586 SLAB_ATTR(red_zone
);
4588 static ssize_t
poison_show(struct kmem_cache
*s
, char *buf
)
4590 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_POISON
));
4593 static ssize_t
poison_store(struct kmem_cache
*s
,
4594 const char *buf
, size_t length
)
4596 if (any_slab_objects(s
))
4599 s
->flags
&= ~SLAB_POISON
;
4600 if (buf
[0] == '1') {
4601 s
->flags
&= ~__CMPXCHG_DOUBLE
;
4602 s
->flags
|= SLAB_POISON
;
4604 calculate_sizes(s
, -1);
4609 static ssize_t
store_user_show(struct kmem_cache
*s
, char *buf
)
4611 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_STORE_USER
));
4614 static ssize_t
store_user_store(struct kmem_cache
*s
,
4615 const char *buf
, size_t length
)
4617 if (any_slab_objects(s
))
4620 s
->flags
&= ~SLAB_STORE_USER
;
4621 if (buf
[0] == '1') {
4622 s
->flags
&= ~__CMPXCHG_DOUBLE
;
4623 s
->flags
|= SLAB_STORE_USER
;
4625 calculate_sizes(s
, -1);
4628 SLAB_ATTR(store_user
);
4630 static ssize_t
validate_show(struct kmem_cache
*s
, char *buf
)
4635 static ssize_t
validate_store(struct kmem_cache
*s
,
4636 const char *buf
, size_t length
)
4640 if (buf
[0] == '1') {
4641 ret
= validate_slab_cache(s
);
4647 SLAB_ATTR(validate
);
4649 static ssize_t
alloc_calls_show(struct kmem_cache
*s
, char *buf
)
4651 if (!(s
->flags
& SLAB_STORE_USER
))
4653 return list_locations(s
, buf
, TRACK_ALLOC
);
4655 SLAB_ATTR_RO(alloc_calls
);
4657 static ssize_t
free_calls_show(struct kmem_cache
*s
, char *buf
)
4659 if (!(s
->flags
& SLAB_STORE_USER
))
4661 return list_locations(s
, buf
, TRACK_FREE
);
4663 SLAB_ATTR_RO(free_calls
);
4664 #endif /* CONFIG_SLUB_DEBUG */
4666 #ifdef CONFIG_FAILSLAB
4667 static ssize_t
failslab_show(struct kmem_cache
*s
, char *buf
)
4669 return sprintf(buf
, "%d\n", !!(s
->flags
& SLAB_FAILSLAB
));
4672 static ssize_t
failslab_store(struct kmem_cache
*s
, const char *buf
,
4675 if (s
->refcount
> 1)
4678 s
->flags
&= ~SLAB_FAILSLAB
;
4680 s
->flags
|= SLAB_FAILSLAB
;
4683 SLAB_ATTR(failslab
);
4686 static ssize_t
shrink_show(struct kmem_cache
*s
, char *buf
)
4691 static ssize_t
shrink_store(struct kmem_cache
*s
,
4692 const char *buf
, size_t length
)
4694 if (buf
[0] == '1') {
4695 int rc
= kmem_cache_shrink(s
);
4706 static ssize_t
remote_node_defrag_ratio_show(struct kmem_cache
*s
, char *buf
)
4708 return sprintf(buf
, "%d\n", s
->remote_node_defrag_ratio
/ 10);
4711 static ssize_t
remote_node_defrag_ratio_store(struct kmem_cache
*s
,
4712 const char *buf
, size_t length
)
4714 unsigned long ratio
;
4717 err
= kstrtoul(buf
, 10, &ratio
);
4722 s
->remote_node_defrag_ratio
= ratio
* 10;
4726 SLAB_ATTR(remote_node_defrag_ratio
);
4729 #ifdef CONFIG_SLUB_STATS
4730 static int show_stat(struct kmem_cache
*s
, char *buf
, enum stat_item si
)
4732 unsigned long sum
= 0;
4735 int *data
= kmalloc(nr_cpu_ids
* sizeof(int), GFP_KERNEL
);
4740 for_each_online_cpu(cpu
) {
4741 unsigned x
= per_cpu_ptr(s
->cpu_slab
, cpu
)->stat
[si
];
4747 len
= sprintf(buf
, "%lu", sum
);
4750 for_each_online_cpu(cpu
) {
4751 if (data
[cpu
] && len
< PAGE_SIZE
- 20)
4752 len
+= sprintf(buf
+ len
, " C%d=%u", cpu
, data
[cpu
]);
4756 return len
+ sprintf(buf
+ len
, "\n");
4759 static void clear_stat(struct kmem_cache
*s
, enum stat_item si
)
4763 for_each_online_cpu(cpu
)
4764 per_cpu_ptr(s
->cpu_slab
, cpu
)->stat
[si
] = 0;
4767 #define STAT_ATTR(si, text) \
4768 static ssize_t text##_show(struct kmem_cache *s, char *buf) \
4770 return show_stat(s, buf, si); \
4772 static ssize_t text##_store(struct kmem_cache *s, \
4773 const char *buf, size_t length) \
4775 if (buf[0] != '0') \
4777 clear_stat(s, si); \
4782 STAT_ATTR(ALLOC_FASTPATH, alloc_fastpath);
4783 STAT_ATTR(ALLOC_SLOWPATH
, alloc_slowpath
);
4784 STAT_ATTR(FREE_FASTPATH
, free_fastpath
);
4785 STAT_ATTR(FREE_SLOWPATH
, free_slowpath
);
4786 STAT_ATTR(FREE_FROZEN
, free_frozen
);
4787 STAT_ATTR(FREE_ADD_PARTIAL
, free_add_partial
);
4788 STAT_ATTR(FREE_REMOVE_PARTIAL
, free_remove_partial
);
4789 STAT_ATTR(ALLOC_FROM_PARTIAL
, alloc_from_partial
);
4790 STAT_ATTR(ALLOC_SLAB
, alloc_slab
);
4791 STAT_ATTR(ALLOC_REFILL
, alloc_refill
);
4792 STAT_ATTR(ALLOC_NODE_MISMATCH
, alloc_node_mismatch
);
4793 STAT_ATTR(FREE_SLAB
, free_slab
);
4794 STAT_ATTR(CPUSLAB_FLUSH
, cpuslab_flush
);
4795 STAT_ATTR(DEACTIVATE_FULL
, deactivate_full
);
4796 STAT_ATTR(DEACTIVATE_EMPTY
, deactivate_empty
);
4797 STAT_ATTR(DEACTIVATE_TO_HEAD
, deactivate_to_head
);
4798 STAT_ATTR(DEACTIVATE_TO_TAIL
, deactivate_to_tail
);
4799 STAT_ATTR(DEACTIVATE_REMOTE_FREES
, deactivate_remote_frees
);
4800 STAT_ATTR(DEACTIVATE_BYPASS
, deactivate_bypass
);
4801 STAT_ATTR(ORDER_FALLBACK
, order_fallback
);
4802 STAT_ATTR(CMPXCHG_DOUBLE_CPU_FAIL
, cmpxchg_double_cpu_fail
);
4803 STAT_ATTR(CMPXCHG_DOUBLE_FAIL
, cmpxchg_double_fail
);
4804 STAT_ATTR(CPU_PARTIAL_ALLOC
, cpu_partial_alloc
);
4805 STAT_ATTR(CPU_PARTIAL_FREE
, cpu_partial_free
);
4806 STAT_ATTR(CPU_PARTIAL_NODE
, cpu_partial_node
);
4807 STAT_ATTR(CPU_PARTIAL_DRAIN
, cpu_partial_drain
);
4810 static struct attribute
*slab_attrs
[] = {
4811 &slab_size_attr
.attr
,
4812 &object_size_attr
.attr
,
4813 &objs_per_slab_attr
.attr
,
4815 &min_partial_attr
.attr
,
4816 &cpu_partial_attr
.attr
,
4818 &objects_partial_attr
.attr
,
4820 &cpu_slabs_attr
.attr
,
4824 &hwcache_align_attr
.attr
,
4825 &reclaim_account_attr
.attr
,
4826 &destroy_by_rcu_attr
.attr
,
4828 &reserved_attr
.attr
,
4829 &slabs_cpu_partial_attr
.attr
,
4830 #ifdef CONFIG_SLUB_DEBUG
4831 &total_objects_attr
.attr
,
4833 &sanity_checks_attr
.attr
,
4835 &red_zone_attr
.attr
,
4837 &store_user_attr
.attr
,
4838 &validate_attr
.attr
,
4839 &alloc_calls_attr
.attr
,
4840 &free_calls_attr
.attr
,
4842 #ifdef CONFIG_ZONE_DMA
4843 &cache_dma_attr
.attr
,
4846 &remote_node_defrag_ratio_attr
.attr
,
4848 #ifdef CONFIG_SLUB_STATS
4849 &alloc_fastpath_attr
.attr
,
4850 &alloc_slowpath_attr
.attr
,
4851 &free_fastpath_attr
.attr
,
4852 &free_slowpath_attr
.attr
,
4853 &free_frozen_attr
.attr
,
4854 &free_add_partial_attr
.attr
,
4855 &free_remove_partial_attr
.attr
,
4856 &alloc_from_partial_attr
.attr
,
4857 &alloc_slab_attr
.attr
,
4858 &alloc_refill_attr
.attr
,
4859 &alloc_node_mismatch_attr
.attr
,
4860 &free_slab_attr
.attr
,
4861 &cpuslab_flush_attr
.attr
,
4862 &deactivate_full_attr
.attr
,
4863 &deactivate_empty_attr
.attr
,
4864 &deactivate_to_head_attr
.attr
,
4865 &deactivate_to_tail_attr
.attr
,
4866 &deactivate_remote_frees_attr
.attr
,
4867 &deactivate_bypass_attr
.attr
,
4868 &order_fallback_attr
.attr
,
4869 &cmpxchg_double_fail_attr
.attr
,
4870 &cmpxchg_double_cpu_fail_attr
.attr
,
4871 &cpu_partial_alloc_attr
.attr
,
4872 &cpu_partial_free_attr
.attr
,
4873 &cpu_partial_node_attr
.attr
,
4874 &cpu_partial_drain_attr
.attr
,
4876 #ifdef CONFIG_FAILSLAB
4877 &failslab_attr
.attr
,
4883 static struct attribute_group slab_attr_group
= {
4884 .attrs
= slab_attrs
,
4887 static ssize_t
slab_attr_show(struct kobject
*kobj
,
4888 struct attribute
*attr
,
4891 struct slab_attribute
*attribute
;
4892 struct kmem_cache
*s
;
4895 attribute
= to_slab_attr(attr
);
4898 if (!attribute
->show
)
4901 err
= attribute
->show(s
, buf
);
4906 static ssize_t
slab_attr_store(struct kobject
*kobj
,
4907 struct attribute
*attr
,
4908 const char *buf
, size_t len
)
4910 struct slab_attribute
*attribute
;
4911 struct kmem_cache
*s
;
4914 attribute
= to_slab_attr(attr
);
4917 if (!attribute
->store
)
4920 err
= attribute
->store(s
, buf
, len
);
4921 #ifdef CONFIG_MEMCG_KMEM
4922 if (slab_state
>= FULL
&& err
>= 0 && is_root_cache(s
)) {
4925 mutex_lock(&slab_mutex
);
4926 if (s
->max_attr_size
< len
)
4927 s
->max_attr_size
= len
;
4930 * This is a best effort propagation, so this function's return
4931 * value will be determined by the parent cache only. This is
4932 * basically because not all attributes will have a well
4933 * defined semantics for rollbacks - most of the actions will
4934 * have permanent effects.
4936 * Returning the error value of any of the children that fail
4937 * is not 100 % defined, in the sense that users seeing the
4938 * error code won't be able to know anything about the state of
4941 * Only returning the error code for the parent cache at least
4942 * has well defined semantics. The cache being written to
4943 * directly either failed or succeeded, in which case we loop
4944 * through the descendants with best-effort propagation.
4946 for_each_memcg_cache_index(i
) {
4947 struct kmem_cache
*c
= cache_from_memcg_idx(s
, i
);
4949 attribute
->store(c
, buf
, len
);
4951 mutex_unlock(&slab_mutex
);
4957 static void memcg_propagate_slab_attrs(struct kmem_cache
*s
)
4959 #ifdef CONFIG_MEMCG_KMEM
4961 char *buffer
= NULL
;
4962 struct kmem_cache
*root_cache
;
4964 if (is_root_cache(s
))
4967 root_cache
= s
->memcg_params
->root_cache
;
4970 * This mean this cache had no attribute written. Therefore, no point
4971 * in copying default values around
4973 if (!root_cache
->max_attr_size
)
4976 for (i
= 0; i
< ARRAY_SIZE(slab_attrs
); i
++) {
4979 struct slab_attribute
*attr
= to_slab_attr(slab_attrs
[i
]);
4981 if (!attr
|| !attr
->store
|| !attr
->show
)
4985 * It is really bad that we have to allocate here, so we will
4986 * do it only as a fallback. If we actually allocate, though,
4987 * we can just use the allocated buffer until the end.
4989 * Most of the slub attributes will tend to be very small in
4990 * size, but sysfs allows buffers up to a page, so they can
4991 * theoretically happen.
4995 else if (root_cache
->max_attr_size
< ARRAY_SIZE(mbuf
))
4998 buffer
= (char *) get_zeroed_page(GFP_KERNEL
);
4999 if (WARN_ON(!buffer
))
5004 attr
->show(root_cache
, buf
);
5005 attr
->store(s
, buf
, strlen(buf
));
5009 free_page((unsigned long)buffer
);
5013 static void kmem_cache_release(struct kobject
*k
)
5015 slab_kmem_cache_release(to_slab(k
));
5018 static const struct sysfs_ops slab_sysfs_ops
= {
5019 .show
= slab_attr_show
,
5020 .store
= slab_attr_store
,
5023 static struct kobj_type slab_ktype
= {
5024 .sysfs_ops
= &slab_sysfs_ops
,
5025 .release
= kmem_cache_release
,
5028 static int uevent_filter(struct kset
*kset
, struct kobject
*kobj
)
5030 struct kobj_type
*ktype
= get_ktype(kobj
);
5032 if (ktype
== &slab_ktype
)
5037 static const struct kset_uevent_ops slab_uevent_ops
= {
5038 .filter
= uevent_filter
,
5041 static struct kset
*slab_kset
;
5043 static inline struct kset
*cache_kset(struct kmem_cache
*s
)
5045 #ifdef CONFIG_MEMCG_KMEM
5046 if (!is_root_cache(s
))
5047 return s
->memcg_params
->root_cache
->memcg_kset
;
5052 #define ID_STR_LENGTH 64
5054 /* Create a unique string id for a slab cache:
5056 * Format :[flags-]size
5058 static char *create_unique_id(struct kmem_cache
*s
)
5060 char *name
= kmalloc(ID_STR_LENGTH
, GFP_KERNEL
);
5067 * First flags affecting slabcache operations. We will only
5068 * get here for aliasable slabs so we do not need to support
5069 * too many flags. The flags here must cover all flags that
5070 * are matched during merging to guarantee that the id is
5073 if (s
->flags
& SLAB_CACHE_DMA
)
5075 if (s
->flags
& SLAB_RECLAIM_ACCOUNT
)
5077 if (s
->flags
& SLAB_DEBUG_FREE
)
5079 if (!(s
->flags
& SLAB_NOTRACK
))
5083 p
+= sprintf(p
, "%07d", s
->size
);
5085 BUG_ON(p
> name
+ ID_STR_LENGTH
- 1);
5089 static int sysfs_slab_add(struct kmem_cache
*s
)
5093 int unmergeable
= slab_unmergeable(s
);
5097 * Slabcache can never be merged so we can use the name proper.
5098 * This is typically the case for debug situations. In that
5099 * case we can catch duplicate names easily.
5101 sysfs_remove_link(&slab_kset
->kobj
, s
->name
);
5105 * Create a unique name for the slab as a target
5108 name
= create_unique_id(s
);
5111 s
->kobj
.kset
= cache_kset(s
);
5112 err
= kobject_init_and_add(&s
->kobj
, &slab_ktype
, NULL
, "%s", name
);
5116 err
= sysfs_create_group(&s
->kobj
, &slab_attr_group
);
5120 #ifdef CONFIG_MEMCG_KMEM
5121 if (is_root_cache(s
)) {
5122 s
->memcg_kset
= kset_create_and_add("cgroup", NULL
, &s
->kobj
);
5123 if (!s
->memcg_kset
) {
5130 kobject_uevent(&s
->kobj
, KOBJ_ADD
);
5132 /* Setup first alias */
5133 sysfs_slab_alias(s
, s
->name
);
5140 kobject_del(&s
->kobj
);
5142 kobject_put(&s
->kobj
);
5146 void sysfs_slab_remove(struct kmem_cache
*s
)
5148 if (slab_state
< FULL
)
5150 * Sysfs has not been setup yet so no need to remove the
5155 #ifdef CONFIG_MEMCG_KMEM
5156 kset_unregister(s
->memcg_kset
);
5158 kobject_uevent(&s
->kobj
, KOBJ_REMOVE
);
5159 kobject_del(&s
->kobj
);
5160 kobject_put(&s
->kobj
);
5164 * Need to buffer aliases during bootup until sysfs becomes
5165 * available lest we lose that information.
5167 struct saved_alias
{
5168 struct kmem_cache
*s
;
5170 struct saved_alias
*next
;
5173 static struct saved_alias
*alias_list
;
5175 static int sysfs_slab_alias(struct kmem_cache
*s
, const char *name
)
5177 struct saved_alias
*al
;
5179 if (slab_state
== FULL
) {
5181 * If we have a leftover link then remove it.
5183 sysfs_remove_link(&slab_kset
->kobj
, name
);
5184 return sysfs_create_link(&slab_kset
->kobj
, &s
->kobj
, name
);
5187 al
= kmalloc(sizeof(struct saved_alias
), GFP_KERNEL
);
5193 al
->next
= alias_list
;
5198 static int __init
slab_sysfs_init(void)
5200 struct kmem_cache
*s
;
5203 mutex_lock(&slab_mutex
);
5205 slab_kset
= kset_create_and_add("slab", &slab_uevent_ops
, kernel_kobj
);
5207 mutex_unlock(&slab_mutex
);
5208 pr_err("Cannot register slab subsystem.\n");
5214 list_for_each_entry(s
, &slab_caches
, list
) {
5215 err
= sysfs_slab_add(s
);
5217 pr_err("SLUB: Unable to add boot slab %s to sysfs\n",
5221 while (alias_list
) {
5222 struct saved_alias
*al
= alias_list
;
5224 alias_list
= alias_list
->next
;
5225 err
= sysfs_slab_alias(al
->s
, al
->name
);
5227 pr_err("SLUB: Unable to add boot slab alias %s to sysfs\n",
5232 mutex_unlock(&slab_mutex
);
5237 __initcall(slab_sysfs_init
);
5238 #endif /* CONFIG_SYSFS */
5241 * The /proc/slabinfo ABI
5243 #ifdef CONFIG_SLABINFO
5244 void get_slabinfo(struct kmem_cache
*s
, struct slabinfo
*sinfo
)
5246 unsigned long nr_slabs
= 0;
5247 unsigned long nr_objs
= 0;
5248 unsigned long nr_free
= 0;
5250 struct kmem_cache_node
*n
;
5252 for_each_kmem_cache_node(s
, node
, n
) {
5253 nr_slabs
+= node_nr_slabs(n
);
5254 nr_objs
+= node_nr_objs(n
);
5255 nr_free
+= count_partial(n
, count_free
);
5258 sinfo
->active_objs
= nr_objs
- nr_free
;
5259 sinfo
->num_objs
= nr_objs
;
5260 sinfo
->active_slabs
= nr_slabs
;
5261 sinfo
->num_slabs
= nr_slabs
;
5262 sinfo
->objects_per_slab
= oo_objects(s
->oo
);
5263 sinfo
->cache_order
= oo_order(s
->oo
);
5266 void slabinfo_show_stats(struct seq_file
*m
, struct kmem_cache
*s
)
5270 ssize_t
slabinfo_write(struct file
*file
, const char __user
*buffer
,
5271 size_t count
, loff_t
*ppos
)
5275 #endif /* CONFIG_SLABINFO */