2 * Slab allocator functions that are independent of the allocator strategy
4 * (C) 2012 Christoph Lameter <cl@linux.com>
6 #include <linux/slab.h>
9 #include <linux/poison.h>
10 #include <linux/interrupt.h>
11 #include <linux/memory.h>
12 #include <linux/compiler.h>
13 #include <linux/module.h>
14 #include <linux/cpu.h>
15 #include <linux/uaccess.h>
16 #include <linux/seq_file.h>
17 #include <linux/proc_fs.h>
18 #include <asm/cacheflush.h>
19 #include <asm/tlbflush.h>
21 #include <linux/memcontrol.h>
25 enum slab_state slab_state
;
26 LIST_HEAD(slab_caches
);
27 DEFINE_MUTEX(slab_mutex
);
28 struct kmem_cache
*kmem_cache
;
30 #ifdef CONFIG_DEBUG_VM
31 static int kmem_cache_sanity_check(struct mem_cgroup
*memcg
, const char *name
,
34 struct kmem_cache
*s
= NULL
;
36 if (!name
|| in_interrupt() || size
< sizeof(void *) ||
37 size
> KMALLOC_MAX_SIZE
) {
38 pr_err("kmem_cache_create(%s) integrity check failed\n", name
);
42 list_for_each_entry(s
, &slab_caches
, list
) {
47 * This happens when the module gets unloaded and doesn't
48 * destroy its slab cache and no-one else reuses the vmalloc
49 * area of the module. Print a warning.
51 res
= probe_kernel_address(s
->name
, tmp
);
53 pr_err("Slab cache with size %d has lost its name\n",
59 * For simplicity, we won't check this in the list of memcg
60 * caches. We have control over memcg naming, and if there
61 * aren't duplicates in the global list, there won't be any
62 * duplicates in the memcg lists as well.
64 if (!memcg
&& !strcmp(s
->name
, name
)) {
65 pr_err("%s (%s): Cache name already exists.\n",
73 WARN_ON(strchr(name
, ' ')); /* It confuses parsers */
77 static inline int kmem_cache_sanity_check(struct mem_cgroup
*memcg
,
78 const char *name
, size_t size
)
84 #ifdef CONFIG_MEMCG_KMEM
85 int memcg_update_all_caches(int num_memcgs
)
89 mutex_lock(&slab_mutex
);
91 list_for_each_entry(s
, &slab_caches
, list
) {
92 if (!is_root_cache(s
))
95 ret
= memcg_update_cache_size(s
, num_memcgs
);
97 * See comment in memcontrol.c, memcg_update_cache_size:
98 * Instead of freeing the memory, we'll just leave the caches
99 * up to this point in an updated state.
105 memcg_update_array_size(num_memcgs
);
107 mutex_unlock(&slab_mutex
);
113 * Figure out what the alignment of the objects will be given a set of
114 * flags, a user specified alignment and the size of the objects.
116 unsigned long calculate_alignment(unsigned long flags
,
117 unsigned long align
, unsigned long size
)
120 * If the user wants hardware cache aligned objects then follow that
121 * suggestion if the object is sufficiently large.
123 * The hardware cache alignment cannot override the specified
124 * alignment though. If that is greater then use it.
126 if (flags
& SLAB_HWCACHE_ALIGN
) {
127 unsigned long ralign
= cache_line_size();
128 while (size
<= ralign
/ 2)
130 align
= max(align
, ralign
);
133 if (align
< ARCH_SLAB_MINALIGN
)
134 align
= ARCH_SLAB_MINALIGN
;
136 return ALIGN(align
, sizeof(void *));
141 * kmem_cache_create - Create a cache.
142 * @name: A string which is used in /proc/slabinfo to identify this cache.
143 * @size: The size of objects to be created in this cache.
144 * @align: The required alignment for the objects.
146 * @ctor: A constructor for the objects.
148 * Returns a ptr to the cache on success, NULL on failure.
149 * Cannot be called within a interrupt, but can be interrupted.
150 * The @ctor is run when new pages are allocated by the cache.
154 * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5)
155 * to catch references to uninitialised memory.
157 * %SLAB_RED_ZONE - Insert `Red' zones around the allocated memory to check
158 * for buffer overruns.
160 * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware
161 * cacheline. This can be beneficial if you're counting cycles as closely
166 kmem_cache_create_memcg(struct mem_cgroup
*memcg
, const char *name
, size_t size
,
167 size_t align
, unsigned long flags
, void (*ctor
)(void *))
169 struct kmem_cache
*s
= NULL
;
173 mutex_lock(&slab_mutex
);
175 if (!kmem_cache_sanity_check(memcg
, name
, size
) == 0)
179 * Some allocators will constraint the set of valid flags to a subset
180 * of all flags. We expect them to define CACHE_CREATE_MASK in this
181 * case, and we'll just provide them with a sanitized version of the
184 flags
&= CACHE_CREATE_MASK
;
186 s
= __kmem_cache_alias(memcg
, name
, size
, align
, flags
, ctor
);
190 s
= kmem_cache_zalloc(kmem_cache
, GFP_KERNEL
);
192 s
->object_size
= s
->size
= size
;
193 s
->align
= calculate_alignment(flags
, align
, size
);
196 if (memcg_register_cache(memcg
, s
)) {
197 kmem_cache_free(kmem_cache
, s
);
202 s
->name
= kstrdup(name
, GFP_KERNEL
);
204 kmem_cache_free(kmem_cache
, s
);
209 err
= __kmem_cache_create(s
, flags
);
212 list_add(&s
->list
, &slab_caches
);
213 memcg_cache_list_add(memcg
, s
);
216 kmem_cache_free(kmem_cache
, s
);
222 mutex_unlock(&slab_mutex
);
227 if (flags
& SLAB_PANIC
)
228 panic("kmem_cache_create: Failed to create slab '%s'. Error %d\n",
231 printk(KERN_WARNING
"kmem_cache_create(%s) failed with error %d",
243 kmem_cache_create(const char *name
, size_t size
, size_t align
,
244 unsigned long flags
, void (*ctor
)(void *))
246 return kmem_cache_create_memcg(NULL
, name
, size
, align
, flags
, ctor
);
248 EXPORT_SYMBOL(kmem_cache_create
);
250 void kmem_cache_destroy(struct kmem_cache
*s
)
253 mutex_lock(&slab_mutex
);
258 if (!__kmem_cache_shutdown(s
)) {
259 mutex_unlock(&slab_mutex
);
260 if (s
->flags
& SLAB_DESTROY_BY_RCU
)
263 memcg_release_cache(s
);
265 kmem_cache_free(kmem_cache
, s
);
267 list_add(&s
->list
, &slab_caches
);
268 mutex_unlock(&slab_mutex
);
269 printk(KERN_ERR
"kmem_cache_destroy %s: Slab cache still has objects\n",
274 mutex_unlock(&slab_mutex
);
278 EXPORT_SYMBOL(kmem_cache_destroy
);
280 int slab_is_available(void)
282 return slab_state
>= UP
;
286 /* Create a cache during boot when no slab services are available yet */
287 void __init
create_boot_cache(struct kmem_cache
*s
, const char *name
, size_t size
,
293 s
->size
= s
->object_size
= size
;
294 s
->align
= calculate_alignment(flags
, ARCH_KMALLOC_MINALIGN
, size
);
295 err
= __kmem_cache_create(s
, flags
);
298 panic("Creation of kmalloc slab %s size=%zd failed. Reason %d\n",
301 s
->refcount
= -1; /* Exempt from merging for now */
304 struct kmem_cache
*__init
create_kmalloc_cache(const char *name
, size_t size
,
307 struct kmem_cache
*s
= kmem_cache_zalloc(kmem_cache
, GFP_NOWAIT
);
310 panic("Out of memory when creating slab %s\n", name
);
312 create_boot_cache(s
, name
, size
, flags
);
313 list_add(&s
->list
, &slab_caches
);
318 #endif /* !CONFIG_SLOB */
321 #ifdef CONFIG_SLABINFO
322 static void print_slabinfo_header(struct seq_file
*m
)
325 * Output format version, so at least we can change it
326 * without _too_ many complaints.
328 #ifdef CONFIG_DEBUG_SLAB
329 seq_puts(m
, "slabinfo - version: 2.1 (statistics)\n");
331 seq_puts(m
, "slabinfo - version: 2.1\n");
333 seq_puts(m
, "# name <active_objs> <num_objs> <objsize> "
334 "<objperslab> <pagesperslab>");
335 seq_puts(m
, " : tunables <limit> <batchcount> <sharedfactor>");
336 seq_puts(m
, " : slabdata <active_slabs> <num_slabs> <sharedavail>");
337 #ifdef CONFIG_DEBUG_SLAB
338 seq_puts(m
, " : globalstat <listallocs> <maxobjs> <grown> <reaped> "
339 "<error> <maxfreeable> <nodeallocs> <remotefrees> <alienoverflow>");
340 seq_puts(m
, " : cpustat <allochit> <allocmiss> <freehit> <freemiss>");
345 static void *s_start(struct seq_file
*m
, loff_t
*pos
)
349 mutex_lock(&slab_mutex
);
351 print_slabinfo_header(m
);
353 return seq_list_start(&slab_caches
, *pos
);
356 static void *s_next(struct seq_file
*m
, void *p
, loff_t
*pos
)
358 return seq_list_next(p
, &slab_caches
, pos
);
361 static void s_stop(struct seq_file
*m
, void *p
)
363 mutex_unlock(&slab_mutex
);
366 static int s_show(struct seq_file
*m
, void *p
)
368 struct kmem_cache
*s
= list_entry(p
, struct kmem_cache
, list
);
369 struct slabinfo sinfo
;
371 memset(&sinfo
, 0, sizeof(sinfo
));
372 get_slabinfo(s
, &sinfo
);
374 seq_printf(m
, "%-17s %6lu %6lu %6u %4u %4d",
375 s
->name
, sinfo
.active_objs
, sinfo
.num_objs
, s
->size
,
376 sinfo
.objects_per_slab
, (1 << sinfo
.cache_order
));
378 seq_printf(m
, " : tunables %4u %4u %4u",
379 sinfo
.limit
, sinfo
.batchcount
, sinfo
.shared
);
380 seq_printf(m
, " : slabdata %6lu %6lu %6lu",
381 sinfo
.active_slabs
, sinfo
.num_slabs
, sinfo
.shared_avail
);
382 slabinfo_show_stats(m
, s
);
388 * slabinfo_op - iterator that generates /proc/slabinfo
398 * + further values on SMP and with statistics enabled
400 static const struct seq_operations slabinfo_op
= {
407 static int slabinfo_open(struct inode
*inode
, struct file
*file
)
409 return seq_open(file
, &slabinfo_op
);
412 static const struct file_operations proc_slabinfo_operations
= {
413 .open
= slabinfo_open
,
415 .write
= slabinfo_write
,
417 .release
= seq_release
,
420 static int __init
slab_proc_init(void)
422 proc_create("slabinfo", S_IRUSR
, NULL
, &proc_slabinfo_operations
);
425 module_init(slab_proc_init
);
426 #endif /* CONFIG_SLABINFO */