Linux內存管理之slab機制（創建cache）

該函數的執行流程：

1，從全局cache_cache中獲得cache結構，因為全局cache_cache初始化對象的大小就是kmem_cache結構的大小，所以返回的指針正好可以轉換為cache結構；調用 kmem_cache_zalloc(&cache_cache, gfp);

2，獲得slab中碎片大小，由函數calculate_slab_order()實現；

3，計算並初始化cache的各種屬性，如果是外置式，需要用kmem_find_general_cachep(slab_size, 0u)指定cachep->slabp_cache，用於存放slab對象和kmem_bufctl_t[]數組；

4，設置每個CPU上得本地cache，setup_cpu_cache();

5，cache創建完畢，將其加入到全局slab cache鏈表中；

一、主實現

1. /**
2. * kmem_cache_create - Create a cache.
3. * @name: A string which is used in /proc/slabinfo to identify this cache.
4. * @size: The size of objects to be created in this cache.
5. * @align: The required alignment for the objects.
6. * @flags: SLAB flags
7. * @ctor: A constructor for the objects.
8. *
9. * Returns a ptr to the cache on success, NULL on failure.
10. * Cannot be called within a int, but can be interrupted.
11. * The @ctor is run when new pages are allocated by the cache.
12. *
13. * @name must be valid until the cache is destroyed. This implies that
14. * the module calling this has to destroy the cache before getting unloaded.
15. * Note that kmem_cache_name() is not guaranteed to return the same pointer,
16. * therefore applications must manage it themselves.
17. *
18. * The flags are
19. *
20. * %SLAB_POISON - Poison the slab with a known test pattern (a5a5a5a5)
21. * to catch references to uninitialised memory.
22. *
23. * %SLAB_RED_ZONE - Insert `Red' zones around the allocated memory to check
24. * for buffer overruns.
25. *
26. * %SLAB_HWCACHE_ALIGN - Align the objects in this cache to a hardware
27. * cacheline. This can be beneficial if you're counting cycles as closely
28. * as davem.
29. */
30. /*創建slab系統頂層的cache節點。創建完成後，cache
31. 裡並沒有任何slab以及對象，只有當分配對象
32. ，並且cache中沒有空閒對象時，才會創建新的slab。*/
33. struct kmem_cache *
34. kmem_cache_create (const char *name, size_t size, size_t align,
35. unsigned long flags, void (*ctor)(void *))
36. {
37. size_t left_over, slab_size, ralign;
38. struct kmem_cache *cachep = NULL, *pc;
39. gfp_t gfp;
41. /*
42. * Sanity checks... these are all serious usage bugs.
43. *//* 安全性檢查 */
44. if (!name || in_interrupt() || (size < BYTES_PER_WORD) ||
45. size > KMALLOC_MAX_SIZE) {
46. printk(KERN_ERR "%s: Early error in slab %s\n", __func__,
47. name);
48. BUG();
49. }
51. /*
52. * We use cache_chain_mutex to ensure a consistent view of
53. * cpu_online_mask as well. Please see cpuup_callback
54. */
55. /* slab分配器是否已經初始化好，如果是內核啟動階段
56. ，則只有一個cpu執行slab分配器的初始化動作，無需加鎖，否則需要加鎖 */
57. if (slab_is_available()) {
58. get_online_cpus();
59. mutex_lock(&cache_chain_mutex);
60. }
61. /* 遍歷cache鏈，做些校驗工作 */
62. list_for_each_entry(pc, &cache_chain, next) {
63. char tmp;
64. int res;
66. /*
67. * This happens when the module gets unloaded and doesn't
68. * destroy its slab cache and no-one else reuses the vmalloc
69. * area of the module. Print a warning.
70. */
71. /* 檢查cache鏈表中的cache是否都有名字 */
72. res = probe_kernel_address(pc->name, tmp);
73. if (res) {/*沒有名字，報錯*/
74. printk(KERN_ERR
75. "SLAB: cache with size %d has lost its name\n",
76. pc->buffer_size);
77. continue;
78. }
79. /* 檢查cache鏈表中是否已經存在相同名字的cache */
80. if (!strcmp(pc->name, name)) {
81. printk(KERN_ERR
82. "kmem_cache_create: duplicate cache %s\n", name);
83. dump_stack();
84. goto oops;
85. }
86. }
88. #if DEBUG
89. WARN_ON(strchr(name, ' ')); /* It confuses parsers */
90. #if FORCED_DEBUG
91. /*
92. * Enable redzoning and last user accounting, except for caches with
93. * large objects, if the increased size would increase the object size
94. * above the next power of two: caches with object sizes just above a
95. * power of two have a significant amount of internal fragmentation.
96. */
97. if (size < 4096 || fls(size - 1) == fls(size-1 + REDZONE_ALIGN +
98. 2 * sizeof(unsigned long long)))
99. flags |= SLAB_RED_ZONE | SLAB_STORE_USER;
100. if (!(flags & SLAB_DESTROY_BY_RCU))
101. flags |= SLAB_POISON;
102. #endif
103. if (flags & SLAB_DESTROY_BY_RCU)
104. BUG_ON(flags & SLAB_POISON);
105. #endif
106. /*
107. * Always checks flags, a caller might be expecting debug support which
108. * isn't available.
109. */
110. BUG_ON(flags & ~CREATE_MASK);
112. /*
113. * Check that size is in terms of words. This is needed to avoid
114. * unaligned accesses for some archs when redzoning is used, and makes
115. * sure any on-slab bufctl's are also correctly aligned.
116. */
117. if (size & (BYTES_PER_WORD - 1)) {
118. size += (BYTES_PER_WORD - 1);
119. size &= ~(BYTES_PER_WORD - 1);
120. }
122. /* calculate the final buffer alignment: */
124. /* 1) arch recommendation: can be overridden for debug */
125. if (flags & SLAB_HWCACHE_ALIGN) {
126. /*
127. * Default alignment: as specified by the arch code. Except if
128. * an object is really small, then squeeze multiple objects into
129. * one cacheline.
130. */
131. ralign = cache_line_size();
132. while (size <= ralign / 2)
133. ralign /= 2;
134. } else {
135. ralign = BYTES_PER_WORD;
136. }
138. /*
139. * Redzoning and user store require word alignment or possibly larger.
140. * Note this will be overridden by architecture or caller mandated
141. * alignment if either is greater than BYTES_PER_WORD.
142. */
143. if (flags & SLAB_STORE_USER)
144. ralign = BYTES_PER_WORD;
146. if (flags & SLAB_RED_ZONE) {
147. ralign = REDZONE_ALIGN;
148. /* If redzoning, ensure that the second redzone is suitably
149. * aligned, by adjusting the object size accordingly. */
150. size += REDZONE_ALIGN - 1;
151. size &= ~(REDZONE_ALIGN - 1);
152. }
154. /* 2) arch mandated alignment */
155. if (ralign < ARCH_SLAB_MINALIGN) {
156. ralign = ARCH_SLAB_MINALIGN;
157. }
158. /* 3) caller mandated alignment */
159. if (ralign < align) {
160. ralign = align;
161. }
162. /* disable debug if necessary */
163. if (ralign > __alignof__(unsigned long long))
164. flags &= ~(SLAB_RED_ZONE | SLAB_STORE_USER);
165. /*
166. * 4) Store it.
167. */
168. align = ralign;
169. /* slab分配器是否已經可用 */
170. if (slab_is_available())
171. gfp = GFP_KERNEL;
172. else
173. /* slab初始化好之前，不允許阻塞，且只能在低端內存區分配 */
174. gfp = GFP_NOWAIT;
176. /* Get cache's description obj. */
177. /* 獲得struct kmem_cache對象 ,為什麼能從cache中獲得的對象是
178. kmem_cache結構呢，因為這裡的全局變量cache_cache的對象大小
179. 就是kmem_cache結構大小*/
180. cachep = kmem_cache_zalloc(&cache_cache, gfp);
181. if (!cachep)
182. goto oops;
184. #if DEBUG
185. cachep->obj_size = size;
187. /*
188. * Both debugging options require word-alignment which is calculated
189. * into align above.
190. */
191. if (flags & SLAB_RED_ZONE) {
192. /* add space for red zone words */
193. cachep->obj_offset += sizeof(unsigned long long);
194. size += 2 * sizeof(unsigned long long);
195. }
196. if (flags & SLAB_STORE_USER) {
197. /* user store requires one word storage behind the end of
198. * the real object. But if the second red zone needs to be
199. * aligned to 64 bits, we must allow that much space.
200. */
201. if (flags & SLAB_RED_ZONE)
202. size += REDZONE_ALIGN;
203. else
204. size += BYTES_PER_WORD;
205. }
206. #if FORCED_DEBUG && defined(CONFIG_DEBUG_PAGEALLOC)
207. if (size >= malloc_sizes[INDEX_L3 + 1].cs_size
208. && cachep->obj_size > cache_line_size() && size < PAGE_SIZE) {
209. cachep->obj_offset += PAGE_SIZE - size;
210. size = PAGE_SIZE;
211. }
212. #endif
213. #endif
215. /*
216. * Determine if the slab management is 'on' or 'off' slab.
217. * (bootstrapping cannot cope with offslab caches so don't do
218. * it too early on.)
219. */
220. /* 確定slab管理對象的存儲方式：內置還是外置
221. 。通常，當對象大於等於512時，使用外置方式
222. 。初始化階段采用內置式。
223. slab_early_init 參見kmem_cache_init函數 */
224. if ((size >= (PAGE_SIZE >> 3)) && !slab_early_init)
225. /*
226. * Size is large, assume best to place the slab management obj
227. * off-slab (should allow better packing of objs).
228. */
229. flags |= CFLGS_OFF_SLAB;
231. size = ALIGN(size, align);
232. /* 獲得slab中碎片的大小 */
233. left_over = calculate_slab_order(cachep, size, align, flags);
234. /* cachep->num為該cache中每個slab的對象數，為0，表示為該對象創建cache失敗 */
235. if (!cachep->num) {
236. printk(KERN_ERR
237. "kmem_cache_create: couldn't create cache %s.\n", name);
238. kmem_cache_free(&cache_cache, cachep);
239. cachep = NULL;
240. goto oops;
241. }
242. /* 計算slab管理對象的大小，包括struct slab對象和kmem_bufctl_t數組 */
243. slab_size = ALIGN(cachep->num * sizeof(kmem_bufctl_t)
244. + sizeof(struct slab), align);
246. /*
247. * If the slab has been placed off-slab, and we have enough space then
248. * move it on-slab. This is at the expense of any extra colouring.
249. */
251. /* 如果這是一個外置式slab，並且碎片大小大於slab管理對象的大小
252. ，則可將slab管理對象移到slab中，改造成一個內置式slab */
253. if (flags & CFLGS_OFF_SLAB && left_over >= slab_size) {
254. /* 除去off-slab標志位 */
255. flags &= ~CFLGS_OFF_SLAB;
256. /* 更新碎片大小 */
257. left_over -= slab_size;
258. }
260. if (flags & CFLGS_OFF_SLAB) {
261. /* really off slab. No need for manual alignment */
262. /* align是針對slab對象的，如果slab管理對象是外置存儲
263. ，自然不會像內置那樣影響到後面slab對象的存儲位置
264. ，也就不需要對齊了 */
265. slab_size =
266. cachep->num * sizeof(kmem_bufctl_t) + sizeof(struct slab);
268. #ifdef CONFIG_PAGE_POISONING
269. /* If we're going to use the generic kernel_map_pages()
270. * poisoning, then it's going to smash the contents of
271. * the redzone and userword anyhow, so switch them off.
272. */
273. if (size % PAGE_SIZE == 0 && flags & SLAB_POISON)
274. flags &= ~(SLAB_RED_ZONE | SLAB_STORE_USER);
275. #endif
276. }
277. /* cache的著色塊的單位大小 */
278. cachep->colour_off = cache_line_size();
279. /* Offset must be a multiple of the alignment. */
280. /* 著色塊大小必須是對象要求對齊方式的倍數 */
281. if (cachep->colour_off < align)
282. cachep->colour_off = align;
283. /* 計算碎片區需要多少個著色快 */
284. cachep->colour = left_over / cachep->colour_off;
285. /* slab管理對象的大小 */
286. cachep->slab_size = slab_size;
287. cachep->flags = flags;
288. cachep->gfpflags = 0;
289. if (CONFIG_ZONE_DMA_FLAG && (flags & SLAB_CACHE_DMA))
290. cachep->gfpflags |= GFP_DMA;
291. /* slab對象的大小 */
292. cachep->buffer_size = size;
293. /* 計算對象在slab中索引時用，參見obj_to_index函數 */
294. cachep->reciprocal_buffer_size = reciprocal_value(size);
296. if (flags & CFLGS_OFF_SLAB) {
297. /* 分配一個slab管理區域對象，保存在slabp_cache中，
298. 這個函數傳入的大小為slab_size,也就是分配slab_size大小的cache
299. ,在slab創建的時候如果是外置式，那麼需要從分配的這裡面
300. 分配出slab對象，剩下的空間放kmem_bufctl_t[]數組，
301. 如果是內置式的slab，此指針為空 */
302. cachep->slabp_cache = kmem_find_general_cachep(slab_size, 0u);
303. /*
304. * This is a possibility for one of the malloc_sizes caches.
305. * But since we go off slab only for object size greater than
306. * PAGE_SIZE/8, and malloc_sizes gets created in ascending order,
307. * this should not happen at all.
308. * But leave a BUG_ON for some lucky dude.
309. */
310. BUG_ON(ZERO_OR_NULL_PTR(cachep->slabp_cache));
311. }
312. cachep->ctor = ctor;
313. cachep->name = name;
314. /* 設置每個cpu上的local cache */
315. if (setup_cpu_cache(cachep, gfp)) {
316. __kmem_cache_destroy(cachep);
317. cachep = NULL;
318. goto oops;
319. }
321. /* cache setup completed, link it into the list */
322. /* cache創建完畢，將其加入到全局slab cache鏈表中 */
323. list_add(&cachep->next, &cache_chain);
324. oops:
325. if (!cachep && (flags & SLAB_PANIC))
326. panic("kmem_cache_create(): failed to create slab `%s'\n",
327. name);
328. if (slab_is_available()) {
329. mutex_unlock(&cache_chain_mutex);
330. put_online_cpus();
331. }
332. return cachep;
333. }

其中，cache_cache

1. /* internal cache of cache description objs */
2. static struct kmem_cache cache_cache = {
3. .batchcount = 1,
4. .limit = BOOT_CPUCACHE_ENTRIES,
5. .shared = 1,
6. .buffer_size = sizeof(struct kmem_cache),/*大小為cache結構，難怪名稱為cache_cache*/
7. .name = "kmem_cache",
8. };