Commit | Line | Data |
---|---|---|
039363f3 CL |
1 | /* |
2 | * Slab allocator functions that are independent of the allocator strategy | |
3 | * | |
4 | * (C) 2012 Christoph Lameter <cl@linux.com> | |
5 | */ | |
6 | #include <linux/slab.h> | |
7 | ||
8 | #include <linux/mm.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> | |
20cea968 CL |
14 | #include <linux/cpu.h> |
15 | #include <linux/uaccess.h> | |
b7454ad3 GC |
16 | #include <linux/seq_file.h> |
17 | #include <linux/proc_fs.h> | |
039363f3 CL |
18 | #include <asm/cacheflush.h> |
19 | #include <asm/tlbflush.h> | |
20 | #include <asm/page.h> | |
2633d7a0 | 21 | #include <linux/memcontrol.h> |
039363f3 | 22 | |
97d06609 CL |
23 | #include "slab.h" |
24 | ||
25 | enum slab_state slab_state; | |
18004c5d CL |
26 | LIST_HEAD(slab_caches); |
27 | DEFINE_MUTEX(slab_mutex); | |
9b030cb8 | 28 | struct kmem_cache *kmem_cache; |
97d06609 | 29 | |
77be4b13 | 30 | #ifdef CONFIG_DEBUG_VM |
2633d7a0 GC |
31 | static int kmem_cache_sanity_check(struct mem_cgroup *memcg, const char *name, |
32 | size_t size) | |
039363f3 CL |
33 | { |
34 | struct kmem_cache *s = NULL; | |
35 | ||
039363f3 CL |
36 | if (!name || in_interrupt() || size < sizeof(void *) || |
37 | size > KMALLOC_MAX_SIZE) { | |
77be4b13 SK |
38 | pr_err("kmem_cache_create(%s) integrity check failed\n", name); |
39 | return -EINVAL; | |
039363f3 | 40 | } |
b920536a | 41 | |
20cea968 CL |
42 | list_for_each_entry(s, &slab_caches, list) { |
43 | char tmp; | |
44 | int res; | |
45 | ||
46 | /* | |
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. | |
50 | */ | |
51 | res = probe_kernel_address(s->name, tmp); | |
52 | if (res) { | |
77be4b13 | 53 | pr_err("Slab cache with size %d has lost its name\n", |
20cea968 CL |
54 | s->object_size); |
55 | continue; | |
56 | } | |
57 | ||
2633d7a0 GC |
58 | /* |
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. | |
63 | */ | |
64 | if (!memcg && !strcmp(s->name, name)) { | |
77be4b13 SK |
65 | pr_err("%s (%s): Cache name already exists.\n", |
66 | __func__, name); | |
20cea968 CL |
67 | dump_stack(); |
68 | s = NULL; | |
77be4b13 | 69 | return -EINVAL; |
20cea968 CL |
70 | } |
71 | } | |
72 | ||
73 | WARN_ON(strchr(name, ' ')); /* It confuses parsers */ | |
77be4b13 SK |
74 | return 0; |
75 | } | |
76 | #else | |
2633d7a0 GC |
77 | static inline int kmem_cache_sanity_check(struct mem_cgroup *memcg, |
78 | const char *name, size_t size) | |
77be4b13 SK |
79 | { |
80 | return 0; | |
81 | } | |
20cea968 CL |
82 | #endif |
83 | ||
55007d84 GC |
84 | #ifdef CONFIG_MEMCG_KMEM |
85 | int memcg_update_all_caches(int num_memcgs) | |
86 | { | |
87 | struct kmem_cache *s; | |
88 | int ret = 0; | |
89 | mutex_lock(&slab_mutex); | |
90 | ||
91 | list_for_each_entry(s, &slab_caches, list) { | |
92 | if (!is_root_cache(s)) | |
93 | continue; | |
94 | ||
95 | ret = memcg_update_cache_size(s, num_memcgs); | |
96 | /* | |
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. | |
100 | */ | |
101 | if (ret) | |
102 | goto out; | |
103 | } | |
104 | ||
105 | memcg_update_array_size(num_memcgs); | |
106 | out: | |
107 | mutex_unlock(&slab_mutex); | |
108 | return ret; | |
109 | } | |
110 | #endif | |
111 | ||
45906855 CL |
112 | /* |
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. | |
115 | */ | |
116 | unsigned long calculate_alignment(unsigned long flags, | |
117 | unsigned long align, unsigned long size) | |
118 | { | |
119 | /* | |
120 | * If the user wants hardware cache aligned objects then follow that | |
121 | * suggestion if the object is sufficiently large. | |
122 | * | |
123 | * The hardware cache alignment cannot override the specified | |
124 | * alignment though. If that is greater then use it. | |
125 | */ | |
126 | if (flags & SLAB_HWCACHE_ALIGN) { | |
127 | unsigned long ralign = cache_line_size(); | |
128 | while (size <= ralign / 2) | |
129 | ralign /= 2; | |
130 | align = max(align, ralign); | |
131 | } | |
132 | ||
133 | if (align < ARCH_SLAB_MINALIGN) | |
134 | align = ARCH_SLAB_MINALIGN; | |
135 | ||
136 | return ALIGN(align, sizeof(void *)); | |
137 | } | |
138 | ||
139 | ||
77be4b13 SK |
140 | /* |
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. | |
145 | * @flags: SLAB flags | |
146 | * @ctor: A constructor for the objects. | |
147 | * | |
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. | |
151 | * | |
152 | * The flags are | |
153 | * | |
154 | * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5) | |
155 | * to catch references to uninitialised memory. | |
156 | * | |
157 | * %SLAB_RED_ZONE - Insert `Red' zones around the allocated memory to check | |
158 | * for buffer overruns. | |
159 | * | |
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 | |
162 | * as davem. | |
163 | */ | |
164 | ||
2633d7a0 GC |
165 | struct kmem_cache * |
166 | kmem_cache_create_memcg(struct mem_cgroup *memcg, const char *name, size_t size, | |
943a451a GC |
167 | size_t align, unsigned long flags, void (*ctor)(void *), |
168 | struct kmem_cache *parent_cache) | |
77be4b13 SK |
169 | { |
170 | struct kmem_cache *s = NULL; | |
686d550d | 171 | int err = 0; |
039363f3 | 172 | |
77be4b13 SK |
173 | get_online_cpus(); |
174 | mutex_lock(&slab_mutex); | |
686d550d | 175 | |
2633d7a0 | 176 | if (!kmem_cache_sanity_check(memcg, name, size) == 0) |
686d550d CL |
177 | goto out_locked; |
178 | ||
d8843922 GC |
179 | /* |
180 | * Some allocators will constraint the set of valid flags to a subset | |
181 | * of all flags. We expect them to define CACHE_CREATE_MASK in this | |
182 | * case, and we'll just provide them with a sanitized version of the | |
183 | * passed flags. | |
184 | */ | |
185 | flags &= CACHE_CREATE_MASK; | |
686d550d | 186 | |
2633d7a0 | 187 | s = __kmem_cache_alias(memcg, name, size, align, flags, ctor); |
cbb79694 CL |
188 | if (s) |
189 | goto out_locked; | |
190 | ||
278b1bb1 | 191 | s = kmem_cache_zalloc(kmem_cache, GFP_KERNEL); |
db265eca | 192 | if (s) { |
8a13a4cc | 193 | s->object_size = s->size = size; |
45906855 | 194 | s->align = calculate_alignment(flags, align, size); |
8a13a4cc | 195 | s->ctor = ctor; |
2633d7a0 | 196 | |
943a451a | 197 | if (memcg_register_cache(memcg, s, parent_cache)) { |
2633d7a0 GC |
198 | kmem_cache_free(kmem_cache, s); |
199 | err = -ENOMEM; | |
200 | goto out_locked; | |
201 | } | |
202 | ||
8a13a4cc CL |
203 | s->name = kstrdup(name, GFP_KERNEL); |
204 | if (!s->name) { | |
205 | kmem_cache_free(kmem_cache, s); | |
206 | err = -ENOMEM; | |
207 | goto out_locked; | |
208 | } | |
209 | ||
210 | err = __kmem_cache_create(s, flags); | |
cce89f4f | 211 | if (!err) { |
cce89f4f | 212 | s->refcount = 1; |
db265eca | 213 | list_add(&s->list, &slab_caches); |
2633d7a0 | 214 | memcg_cache_list_add(memcg, s); |
cce89f4f | 215 | } else { |
8a13a4cc | 216 | kfree(s->name); |
278b1bb1 CL |
217 | kmem_cache_free(kmem_cache, s); |
218 | } | |
8a13a4cc | 219 | } else |
278b1bb1 | 220 | err = -ENOMEM; |
7c9adf5a | 221 | |
686d550d | 222 | out_locked: |
20cea968 CL |
223 | mutex_unlock(&slab_mutex); |
224 | put_online_cpus(); | |
225 | ||
686d550d CL |
226 | if (err) { |
227 | ||
228 | if (flags & SLAB_PANIC) | |
229 | panic("kmem_cache_create: Failed to create slab '%s'. Error %d\n", | |
230 | name, err); | |
231 | else { | |
232 | printk(KERN_WARNING "kmem_cache_create(%s) failed with error %d", | |
233 | name, err); | |
234 | dump_stack(); | |
235 | } | |
236 | ||
237 | return NULL; | |
238 | } | |
039363f3 CL |
239 | |
240 | return s; | |
241 | } | |
2633d7a0 GC |
242 | |
243 | struct kmem_cache * | |
244 | kmem_cache_create(const char *name, size_t size, size_t align, | |
245 | unsigned long flags, void (*ctor)(void *)) | |
246 | { | |
943a451a | 247 | return kmem_cache_create_memcg(NULL, name, size, align, flags, ctor, NULL); |
2633d7a0 | 248 | } |
039363f3 | 249 | EXPORT_SYMBOL(kmem_cache_create); |
97d06609 | 250 | |
945cf2b6 CL |
251 | void kmem_cache_destroy(struct kmem_cache *s) |
252 | { | |
7cf27982 GC |
253 | /* Destroy all the children caches if we aren't a memcg cache */ |
254 | kmem_cache_destroy_memcg_children(s); | |
255 | ||
945cf2b6 CL |
256 | get_online_cpus(); |
257 | mutex_lock(&slab_mutex); | |
258 | s->refcount--; | |
259 | if (!s->refcount) { | |
260 | list_del(&s->list); | |
261 | ||
262 | if (!__kmem_cache_shutdown(s)) { | |
210ed9de | 263 | mutex_unlock(&slab_mutex); |
945cf2b6 CL |
264 | if (s->flags & SLAB_DESTROY_BY_RCU) |
265 | rcu_barrier(); | |
266 | ||
2633d7a0 | 267 | memcg_release_cache(s); |
db265eca | 268 | kfree(s->name); |
8f4c765c | 269 | kmem_cache_free(kmem_cache, s); |
945cf2b6 CL |
270 | } else { |
271 | list_add(&s->list, &slab_caches); | |
210ed9de | 272 | mutex_unlock(&slab_mutex); |
945cf2b6 CL |
273 | printk(KERN_ERR "kmem_cache_destroy %s: Slab cache still has objects\n", |
274 | s->name); | |
275 | dump_stack(); | |
276 | } | |
210ed9de JK |
277 | } else { |
278 | mutex_unlock(&slab_mutex); | |
945cf2b6 | 279 | } |
945cf2b6 CL |
280 | put_online_cpus(); |
281 | } | |
282 | EXPORT_SYMBOL(kmem_cache_destroy); | |
283 | ||
97d06609 CL |
284 | int slab_is_available(void) |
285 | { | |
286 | return slab_state >= UP; | |
287 | } | |
b7454ad3 | 288 | |
45530c44 CL |
289 | #ifndef CONFIG_SLOB |
290 | /* Create a cache during boot when no slab services are available yet */ | |
291 | void __init create_boot_cache(struct kmem_cache *s, const char *name, size_t size, | |
292 | unsigned long flags) | |
293 | { | |
294 | int err; | |
295 | ||
296 | s->name = name; | |
297 | s->size = s->object_size = size; | |
45906855 | 298 | s->align = calculate_alignment(flags, ARCH_KMALLOC_MINALIGN, size); |
45530c44 CL |
299 | err = __kmem_cache_create(s, flags); |
300 | ||
301 | if (err) | |
302 | panic("Creation of kmalloc slab %s size=%zd failed. Reason %d\n", | |
303 | name, size, err); | |
304 | ||
305 | s->refcount = -1; /* Exempt from merging for now */ | |
306 | } | |
307 | ||
308 | struct kmem_cache *__init create_kmalloc_cache(const char *name, size_t size, | |
309 | unsigned long flags) | |
310 | { | |
311 | struct kmem_cache *s = kmem_cache_zalloc(kmem_cache, GFP_NOWAIT); | |
312 | ||
313 | if (!s) | |
314 | panic("Out of memory when creating slab %s\n", name); | |
315 | ||
316 | create_boot_cache(s, name, size, flags); | |
317 | list_add(&s->list, &slab_caches); | |
318 | s->refcount = 1; | |
319 | return s; | |
320 | } | |
321 | ||
322 | #endif /* !CONFIG_SLOB */ | |
323 | ||
324 | ||
b7454ad3 | 325 | #ifdef CONFIG_SLABINFO |
749c5415 | 326 | void print_slabinfo_header(struct seq_file *m) |
bcee6e2a GC |
327 | { |
328 | /* | |
329 | * Output format version, so at least we can change it | |
330 | * without _too_ many complaints. | |
331 | */ | |
332 | #ifdef CONFIG_DEBUG_SLAB | |
333 | seq_puts(m, "slabinfo - version: 2.1 (statistics)\n"); | |
334 | #else | |
335 | seq_puts(m, "slabinfo - version: 2.1\n"); | |
336 | #endif | |
337 | seq_puts(m, "# name <active_objs> <num_objs> <objsize> " | |
338 | "<objperslab> <pagesperslab>"); | |
339 | seq_puts(m, " : tunables <limit> <batchcount> <sharedfactor>"); | |
340 | seq_puts(m, " : slabdata <active_slabs> <num_slabs> <sharedavail>"); | |
341 | #ifdef CONFIG_DEBUG_SLAB | |
342 | seq_puts(m, " : globalstat <listallocs> <maxobjs> <grown> <reaped> " | |
343 | "<error> <maxfreeable> <nodeallocs> <remotefrees> <alienoverflow>"); | |
344 | seq_puts(m, " : cpustat <allochit> <allocmiss> <freehit> <freemiss>"); | |
345 | #endif | |
346 | seq_putc(m, '\n'); | |
347 | } | |
348 | ||
b7454ad3 GC |
349 | static void *s_start(struct seq_file *m, loff_t *pos) |
350 | { | |
351 | loff_t n = *pos; | |
352 | ||
353 | mutex_lock(&slab_mutex); | |
354 | if (!n) | |
355 | print_slabinfo_header(m); | |
356 | ||
357 | return seq_list_start(&slab_caches, *pos); | |
358 | } | |
359 | ||
360 | static void *s_next(struct seq_file *m, void *p, loff_t *pos) | |
361 | { | |
362 | return seq_list_next(p, &slab_caches, pos); | |
363 | } | |
364 | ||
365 | static void s_stop(struct seq_file *m, void *p) | |
366 | { | |
367 | mutex_unlock(&slab_mutex); | |
368 | } | |
369 | ||
749c5415 GC |
370 | static void |
371 | memcg_accumulate_slabinfo(struct kmem_cache *s, struct slabinfo *info) | |
372 | { | |
373 | struct kmem_cache *c; | |
374 | struct slabinfo sinfo; | |
375 | int i; | |
376 | ||
377 | if (!is_root_cache(s)) | |
378 | return; | |
379 | ||
380 | for_each_memcg_cache_index(i) { | |
381 | c = cache_from_memcg(s, i); | |
382 | if (!c) | |
383 | continue; | |
384 | ||
385 | memset(&sinfo, 0, sizeof(sinfo)); | |
386 | get_slabinfo(c, &sinfo); | |
387 | ||
388 | info->active_slabs += sinfo.active_slabs; | |
389 | info->num_slabs += sinfo.num_slabs; | |
390 | info->shared_avail += sinfo.shared_avail; | |
391 | info->active_objs += sinfo.active_objs; | |
392 | info->num_objs += sinfo.num_objs; | |
393 | } | |
394 | } | |
395 | ||
396 | int cache_show(struct kmem_cache *s, struct seq_file *m) | |
b7454ad3 | 397 | { |
0d7561c6 GC |
398 | struct slabinfo sinfo; |
399 | ||
400 | memset(&sinfo, 0, sizeof(sinfo)); | |
401 | get_slabinfo(s, &sinfo); | |
402 | ||
749c5415 GC |
403 | memcg_accumulate_slabinfo(s, &sinfo); |
404 | ||
0d7561c6 | 405 | seq_printf(m, "%-17s %6lu %6lu %6u %4u %4d", |
749c5415 | 406 | cache_name(s), sinfo.active_objs, sinfo.num_objs, s->size, |
0d7561c6 GC |
407 | sinfo.objects_per_slab, (1 << sinfo.cache_order)); |
408 | ||
409 | seq_printf(m, " : tunables %4u %4u %4u", | |
410 | sinfo.limit, sinfo.batchcount, sinfo.shared); | |
411 | seq_printf(m, " : slabdata %6lu %6lu %6lu", | |
412 | sinfo.active_slabs, sinfo.num_slabs, sinfo.shared_avail); | |
413 | slabinfo_show_stats(m, s); | |
414 | seq_putc(m, '\n'); | |
415 | return 0; | |
b7454ad3 GC |
416 | } |
417 | ||
749c5415 GC |
418 | static int s_show(struct seq_file *m, void *p) |
419 | { | |
420 | struct kmem_cache *s = list_entry(p, struct kmem_cache, list); | |
421 | ||
422 | if (!is_root_cache(s)) | |
423 | return 0; | |
424 | return cache_show(s, m); | |
425 | } | |
426 | ||
b7454ad3 GC |
427 | /* |
428 | * slabinfo_op - iterator that generates /proc/slabinfo | |
429 | * | |
430 | * Output layout: | |
431 | * cache-name | |
432 | * num-active-objs | |
433 | * total-objs | |
434 | * object size | |
435 | * num-active-slabs | |
436 | * total-slabs | |
437 | * num-pages-per-slab | |
438 | * + further values on SMP and with statistics enabled | |
439 | */ | |
440 | static const struct seq_operations slabinfo_op = { | |
441 | .start = s_start, | |
442 | .next = s_next, | |
443 | .stop = s_stop, | |
444 | .show = s_show, | |
445 | }; | |
446 | ||
447 | static int slabinfo_open(struct inode *inode, struct file *file) | |
448 | { | |
449 | return seq_open(file, &slabinfo_op); | |
450 | } | |
451 | ||
452 | static const struct file_operations proc_slabinfo_operations = { | |
453 | .open = slabinfo_open, | |
454 | .read = seq_read, | |
455 | .write = slabinfo_write, | |
456 | .llseek = seq_lseek, | |
457 | .release = seq_release, | |
458 | }; | |
459 | ||
460 | static int __init slab_proc_init(void) | |
461 | { | |
462 | proc_create("slabinfo", S_IRUSR, NULL, &proc_slabinfo_operations); | |
463 | return 0; | |
464 | } | |
465 | module_init(slab_proc_init); | |
466 | #endif /* CONFIG_SLABINFO */ |