]>
Commit | Line | Data |
---|---|---|
10cef602 MM |
1 | /* |
2 | * SLOB Allocator: Simple List Of Blocks | |
3 | * | |
4 | * Matt Mackall <mpm@selenic.com> 12/30/03 | |
5 | * | |
6193a2ff PM |
6 | * NUMA support by Paul Mundt, 2007. |
7 | * | |
10cef602 MM |
8 | * How SLOB works: |
9 | * | |
10 | * The core of SLOB is a traditional K&R style heap allocator, with | |
11 | * support for returning aligned objects. The granularity of this | |
55394849 NP |
12 | * allocator is as little as 2 bytes, however typically most architectures |
13 | * will require 4 bytes on 32-bit and 8 bytes on 64-bit. | |
95b35127 | 14 | * |
20cecbae MM |
15 | * The slob heap is a set of linked list of pages from alloc_pages(), |
16 | * and within each page, there is a singly-linked list of free blocks | |
17 | * (slob_t). The heap is grown on demand. To reduce fragmentation, | |
18 | * heap pages are segregated into three lists, with objects less than | |
19 | * 256 bytes, objects less than 1024 bytes, and all other objects. | |
20 | * | |
21 | * Allocation from heap involves first searching for a page with | |
22 | * sufficient free blocks (using a next-fit-like approach) followed by | |
23 | * a first-fit scan of the page. Deallocation inserts objects back | |
24 | * into the free list in address order, so this is effectively an | |
25 | * address-ordered first fit. | |
10cef602 MM |
26 | * |
27 | * Above this is an implementation of kmalloc/kfree. Blocks returned | |
55394849 | 28 | * from kmalloc are prepended with a 4-byte header with the kmalloc size. |
10cef602 | 29 | * If kmalloc is asked for objects of PAGE_SIZE or larger, it calls |
6193a2ff | 30 | * alloc_pages() directly, allocating compound pages so the page order |
d87a133f NP |
31 | * does not have to be separately tracked, and also stores the exact |
32 | * allocation size in page->private so that it can be used to accurately | |
33 | * provide ksize(). These objects are detected in kfree() because slob_page() | |
34 | * is false for them. | |
10cef602 MM |
35 | * |
36 | * SLAB is emulated on top of SLOB by simply calling constructors and | |
95b35127 NP |
37 | * destructors for every SLAB allocation. Objects are returned with the |
38 | * 4-byte alignment unless the SLAB_HWCACHE_ALIGN flag is set, in which | |
39 | * case the low-level allocator will fragment blocks to create the proper | |
40 | * alignment. Again, objects of page-size or greater are allocated by | |
6193a2ff | 41 | * calling alloc_pages(). As SLAB objects know their size, no separate |
95b35127 | 42 | * size bookkeeping is necessary and there is essentially no allocation |
d87a133f NP |
43 | * space overhead, and compound pages aren't needed for multi-page |
44 | * allocations. | |
6193a2ff PM |
45 | * |
46 | * NUMA support in SLOB is fairly simplistic, pushing most of the real | |
47 | * logic down to the page allocator, and simply doing the node accounting | |
48 | * on the upper levels. In the event that a node id is explicitly | |
49 | * provided, alloc_pages_node() with the specified node id is used | |
50 | * instead. The common case (or when the node id isn't explicitly provided) | |
51 | * will default to the current node, as per numa_node_id(). | |
52 | * | |
53 | * Node aware pages are still inserted in to the global freelist, and | |
54 | * these are scanned for by matching against the node id encoded in the | |
55 | * page flags. As a result, block allocations that can be satisfied from | |
56 | * the freelist will only be done so on pages residing on the same node, | |
57 | * in order to prevent random node placement. | |
10cef602 MM |
58 | */ |
59 | ||
95b35127 | 60 | #include <linux/kernel.h> |
10cef602 MM |
61 | #include <linux/slab.h> |
62 | #include <linux/mm.h> | |
63 | #include <linux/cache.h> | |
64 | #include <linux/init.h> | |
65 | #include <linux/module.h> | |
afc0cedb | 66 | #include <linux/rcupdate.h> |
95b35127 | 67 | #include <linux/list.h> |
3eae2cb2 | 68 | #include <linux/kmemtrace.h> |
95b35127 NP |
69 | #include <asm/atomic.h> |
70 | ||
95b35127 NP |
71 | /* |
72 | * slob_block has a field 'units', which indicates size of block if +ve, | |
73 | * or offset of next block if -ve (in SLOB_UNITs). | |
74 | * | |
75 | * Free blocks of size 1 unit simply contain the offset of the next block. | |
76 | * Those with larger size contain their size in the first SLOB_UNIT of | |
77 | * memory, and the offset of the next free block in the second SLOB_UNIT. | |
78 | */ | |
55394849 | 79 | #if PAGE_SIZE <= (32767 * 2) |
95b35127 NP |
80 | typedef s16 slobidx_t; |
81 | #else | |
82 | typedef s32 slobidx_t; | |
83 | #endif | |
84 | ||
10cef602 | 85 | struct slob_block { |
95b35127 | 86 | slobidx_t units; |
55394849 | 87 | }; |
10cef602 MM |
88 | typedef struct slob_block slob_t; |
89 | ||
95b35127 NP |
90 | /* |
91 | * We use struct page fields to manage some slob allocation aspects, | |
92 | * however to avoid the horrible mess in include/linux/mm_types.h, we'll | |
93 | * just define our own struct page type variant here. | |
94 | */ | |
95 | struct slob_page { | |
96 | union { | |
97 | struct { | |
98 | unsigned long flags; /* mandatory */ | |
99 | atomic_t _count; /* mandatory */ | |
100 | slobidx_t units; /* free units left in page */ | |
101 | unsigned long pad[2]; | |
102 | slob_t *free; /* first free slob_t in page */ | |
103 | struct list_head list; /* linked list of free pages */ | |
104 | }; | |
105 | struct page page; | |
106 | }; | |
107 | }; | |
108 | static inline void struct_slob_page_wrong_size(void) | |
109 | { BUILD_BUG_ON(sizeof(struct slob_page) != sizeof(struct page)); } | |
110 | ||
111 | /* | |
112 | * free_slob_page: call before a slob_page is returned to the page allocator. | |
113 | */ | |
114 | static inline void free_slob_page(struct slob_page *sp) | |
115 | { | |
116 | reset_page_mapcount(&sp->page); | |
117 | sp->page.mapping = NULL; | |
118 | } | |
119 | ||
120 | /* | |
20cecbae | 121 | * All partially free slob pages go on these lists. |
95b35127 | 122 | */ |
20cecbae MM |
123 | #define SLOB_BREAK1 256 |
124 | #define SLOB_BREAK2 1024 | |
125 | static LIST_HEAD(free_slob_small); | |
126 | static LIST_HEAD(free_slob_medium); | |
127 | static LIST_HEAD(free_slob_large); | |
95b35127 NP |
128 | |
129 | /* | |
130 | * slob_page: True for all slob pages (false for bigblock pages) | |
131 | */ | |
132 | static inline int slob_page(struct slob_page *sp) | |
133 | { | |
9023cb7e | 134 | return PageSlobPage((struct page *)sp); |
95b35127 NP |
135 | } |
136 | ||
137 | static inline void set_slob_page(struct slob_page *sp) | |
138 | { | |
9023cb7e | 139 | __SetPageSlobPage((struct page *)sp); |
95b35127 NP |
140 | } |
141 | ||
142 | static inline void clear_slob_page(struct slob_page *sp) | |
143 | { | |
9023cb7e | 144 | __ClearPageSlobPage((struct page *)sp); |
95b35127 NP |
145 | } |
146 | ||
147 | /* | |
148 | * slob_page_free: true for pages on free_slob_pages list. | |
149 | */ | |
150 | static inline int slob_page_free(struct slob_page *sp) | |
151 | { | |
9023cb7e | 152 | return PageSlobFree((struct page *)sp); |
95b35127 NP |
153 | } |
154 | ||
20cecbae | 155 | static void set_slob_page_free(struct slob_page *sp, struct list_head *list) |
95b35127 | 156 | { |
20cecbae | 157 | list_add(&sp->list, list); |
9023cb7e | 158 | __SetPageSlobFree((struct page *)sp); |
95b35127 NP |
159 | } |
160 | ||
161 | static inline void clear_slob_page_free(struct slob_page *sp) | |
162 | { | |
163 | list_del(&sp->list); | |
9023cb7e | 164 | __ClearPageSlobFree((struct page *)sp); |
95b35127 NP |
165 | } |
166 | ||
10cef602 MM |
167 | #define SLOB_UNIT sizeof(slob_t) |
168 | #define SLOB_UNITS(size) (((size) + SLOB_UNIT - 1)/SLOB_UNIT) | |
169 | #define SLOB_ALIGN L1_CACHE_BYTES | |
170 | ||
afc0cedb NP |
171 | /* |
172 | * struct slob_rcu is inserted at the tail of allocated slob blocks, which | |
173 | * were created with a SLAB_DESTROY_BY_RCU slab. slob_rcu is used to free | |
174 | * the block using call_rcu. | |
175 | */ | |
176 | struct slob_rcu { | |
177 | struct rcu_head head; | |
178 | int size; | |
179 | }; | |
180 | ||
95b35127 NP |
181 | /* |
182 | * slob_lock protects all slob allocator structures. | |
183 | */ | |
10cef602 | 184 | static DEFINE_SPINLOCK(slob_lock); |
10cef602 | 185 | |
95b35127 NP |
186 | /* |
187 | * Encode the given size and next info into a free slob block s. | |
188 | */ | |
189 | static void set_slob(slob_t *s, slobidx_t size, slob_t *next) | |
190 | { | |
191 | slob_t *base = (slob_t *)((unsigned long)s & PAGE_MASK); | |
192 | slobidx_t offset = next - base; | |
bcb4ddb4 | 193 | |
95b35127 NP |
194 | if (size > 1) { |
195 | s[0].units = size; | |
196 | s[1].units = offset; | |
197 | } else | |
198 | s[0].units = -offset; | |
199 | } | |
10cef602 | 200 | |
95b35127 NP |
201 | /* |
202 | * Return the size of a slob block. | |
203 | */ | |
204 | static slobidx_t slob_units(slob_t *s) | |
205 | { | |
206 | if (s->units > 0) | |
207 | return s->units; | |
208 | return 1; | |
209 | } | |
210 | ||
211 | /* | |
212 | * Return the next free slob block pointer after this one. | |
213 | */ | |
214 | static slob_t *slob_next(slob_t *s) | |
215 | { | |
216 | slob_t *base = (slob_t *)((unsigned long)s & PAGE_MASK); | |
217 | slobidx_t next; | |
218 | ||
219 | if (s[0].units < 0) | |
220 | next = -s[0].units; | |
221 | else | |
222 | next = s[1].units; | |
223 | return base+next; | |
224 | } | |
225 | ||
226 | /* | |
227 | * Returns true if s is the last free block in its page. | |
228 | */ | |
229 | static int slob_last(slob_t *s) | |
230 | { | |
231 | return !((unsigned long)slob_next(s) & ~PAGE_MASK); | |
232 | } | |
233 | ||
6193a2ff PM |
234 | static void *slob_new_page(gfp_t gfp, int order, int node) |
235 | { | |
236 | void *page; | |
237 | ||
238 | #ifdef CONFIG_NUMA | |
239 | if (node != -1) | |
240 | page = alloc_pages_node(node, gfp, order); | |
241 | else | |
242 | #endif | |
243 | page = alloc_pages(gfp, order); | |
244 | ||
245 | if (!page) | |
246 | return NULL; | |
247 | ||
248 | return page_address(page); | |
249 | } | |
250 | ||
95b35127 NP |
251 | /* |
252 | * Allocate a slob block within a given slob_page sp. | |
253 | */ | |
254 | static void *slob_page_alloc(struct slob_page *sp, size_t size, int align) | |
10cef602 MM |
255 | { |
256 | slob_t *prev, *cur, *aligned = 0; | |
257 | int delta = 0, units = SLOB_UNITS(size); | |
10cef602 | 258 | |
95b35127 NP |
259 | for (prev = NULL, cur = sp->free; ; prev = cur, cur = slob_next(cur)) { |
260 | slobidx_t avail = slob_units(cur); | |
261 | ||
10cef602 MM |
262 | if (align) { |
263 | aligned = (slob_t *)ALIGN((unsigned long)cur, align); | |
264 | delta = aligned - cur; | |
265 | } | |
95b35127 NP |
266 | if (avail >= units + delta) { /* room enough? */ |
267 | slob_t *next; | |
268 | ||
10cef602 | 269 | if (delta) { /* need to fragment head to align? */ |
95b35127 NP |
270 | next = slob_next(cur); |
271 | set_slob(aligned, avail - delta, next); | |
272 | set_slob(cur, delta, aligned); | |
10cef602 MM |
273 | prev = cur; |
274 | cur = aligned; | |
95b35127 | 275 | avail = slob_units(cur); |
10cef602 MM |
276 | } |
277 | ||
95b35127 NP |
278 | next = slob_next(cur); |
279 | if (avail == units) { /* exact fit? unlink. */ | |
280 | if (prev) | |
281 | set_slob(prev, slob_units(prev), next); | |
282 | else | |
283 | sp->free = next; | |
284 | } else { /* fragment */ | |
285 | if (prev) | |
286 | set_slob(prev, slob_units(prev), cur + units); | |
287 | else | |
288 | sp->free = cur + units; | |
289 | set_slob(cur + units, avail - units, next); | |
10cef602 MM |
290 | } |
291 | ||
95b35127 NP |
292 | sp->units -= units; |
293 | if (!sp->units) | |
294 | clear_slob_page_free(sp); | |
10cef602 MM |
295 | return cur; |
296 | } | |
95b35127 NP |
297 | if (slob_last(cur)) |
298 | return NULL; | |
299 | } | |
300 | } | |
10cef602 | 301 | |
95b35127 NP |
302 | /* |
303 | * slob_alloc: entry point into the slob allocator. | |
304 | */ | |
6193a2ff | 305 | static void *slob_alloc(size_t size, gfp_t gfp, int align, int node) |
95b35127 NP |
306 | { |
307 | struct slob_page *sp; | |
d6269543 | 308 | struct list_head *prev; |
20cecbae | 309 | struct list_head *slob_list; |
95b35127 NP |
310 | slob_t *b = NULL; |
311 | unsigned long flags; | |
10cef602 | 312 | |
20cecbae MM |
313 | if (size < SLOB_BREAK1) |
314 | slob_list = &free_slob_small; | |
315 | else if (size < SLOB_BREAK2) | |
316 | slob_list = &free_slob_medium; | |
317 | else | |
318 | slob_list = &free_slob_large; | |
319 | ||
95b35127 NP |
320 | spin_lock_irqsave(&slob_lock, flags); |
321 | /* Iterate through each partially free page, try to find room */ | |
20cecbae | 322 | list_for_each_entry(sp, slob_list, list) { |
6193a2ff PM |
323 | #ifdef CONFIG_NUMA |
324 | /* | |
325 | * If there's a node specification, search for a partial | |
326 | * page with a matching node id in the freelist. | |
327 | */ | |
328 | if (node != -1 && page_to_nid(&sp->page) != node) | |
329 | continue; | |
330 | #endif | |
d6269543 MM |
331 | /* Enough room on this page? */ |
332 | if (sp->units < SLOB_UNITS(size)) | |
333 | continue; | |
6193a2ff | 334 | |
d6269543 MM |
335 | /* Attempt to alloc */ |
336 | prev = sp->list.prev; | |
337 | b = slob_page_alloc(sp, size, align); | |
338 | if (!b) | |
339 | continue; | |
340 | ||
341 | /* Improve fragment distribution and reduce our average | |
342 | * search time by starting our next search here. (see | |
343 | * Knuth vol 1, sec 2.5, pg 449) */ | |
20cecbae MM |
344 | if (prev != slob_list->prev && |
345 | slob_list->next != prev->next) | |
346 | list_move_tail(slob_list, prev->next); | |
d6269543 | 347 | break; |
10cef602 | 348 | } |
95b35127 NP |
349 | spin_unlock_irqrestore(&slob_lock, flags); |
350 | ||
351 | /* Not enough space: must allocate a new page */ | |
352 | if (!b) { | |
7fd27255 | 353 | b = slob_new_page(gfp & ~__GFP_ZERO, 0, node); |
95b35127 NP |
354 | if (!b) |
355 | return 0; | |
356 | sp = (struct slob_page *)virt_to_page(b); | |
357 | set_slob_page(sp); | |
358 | ||
359 | spin_lock_irqsave(&slob_lock, flags); | |
360 | sp->units = SLOB_UNITS(PAGE_SIZE); | |
361 | sp->free = b; | |
362 | INIT_LIST_HEAD(&sp->list); | |
363 | set_slob(b, SLOB_UNITS(PAGE_SIZE), b + SLOB_UNITS(PAGE_SIZE)); | |
20cecbae | 364 | set_slob_page_free(sp, slob_list); |
95b35127 NP |
365 | b = slob_page_alloc(sp, size, align); |
366 | BUG_ON(!b); | |
367 | spin_unlock_irqrestore(&slob_lock, flags); | |
368 | } | |
d07dbea4 CL |
369 | if (unlikely((gfp & __GFP_ZERO) && b)) |
370 | memset(b, 0, size); | |
95b35127 | 371 | return b; |
10cef602 MM |
372 | } |
373 | ||
95b35127 NP |
374 | /* |
375 | * slob_free: entry point into the slob allocator. | |
376 | */ | |
10cef602 MM |
377 | static void slob_free(void *block, int size) |
378 | { | |
95b35127 NP |
379 | struct slob_page *sp; |
380 | slob_t *prev, *next, *b = (slob_t *)block; | |
381 | slobidx_t units; | |
10cef602 MM |
382 | unsigned long flags; |
383 | ||
2408c550 | 384 | if (unlikely(ZERO_OR_NULL_PTR(block))) |
10cef602 | 385 | return; |
95b35127 | 386 | BUG_ON(!size); |
10cef602 | 387 | |
95b35127 NP |
388 | sp = (struct slob_page *)virt_to_page(block); |
389 | units = SLOB_UNITS(size); | |
10cef602 | 390 | |
10cef602 | 391 | spin_lock_irqsave(&slob_lock, flags); |
10cef602 | 392 | |
95b35127 NP |
393 | if (sp->units + units == SLOB_UNITS(PAGE_SIZE)) { |
394 | /* Go directly to page allocator. Do not pass slob allocator */ | |
395 | if (slob_page_free(sp)) | |
396 | clear_slob_page_free(sp); | |
397 | clear_slob_page(sp); | |
398 | free_slob_page(sp); | |
399 | free_page((unsigned long)b); | |
400 | goto out; | |
401 | } | |
10cef602 | 402 | |
95b35127 NP |
403 | if (!slob_page_free(sp)) { |
404 | /* This slob page is about to become partially free. Easy! */ | |
405 | sp->units = units; | |
406 | sp->free = b; | |
407 | set_slob(b, units, | |
408 | (void *)((unsigned long)(b + | |
409 | SLOB_UNITS(PAGE_SIZE)) & PAGE_MASK)); | |
20cecbae | 410 | set_slob_page_free(sp, &free_slob_small); |
95b35127 NP |
411 | goto out; |
412 | } | |
413 | ||
414 | /* | |
415 | * Otherwise the page is already partially free, so find reinsertion | |
416 | * point. | |
417 | */ | |
418 | sp->units += units; | |
10cef602 | 419 | |
95b35127 | 420 | if (b < sp->free) { |
679299b3 MM |
421 | if (b + units == sp->free) { |
422 | units += slob_units(sp->free); | |
423 | sp->free = slob_next(sp->free); | |
424 | } | |
95b35127 NP |
425 | set_slob(b, units, sp->free); |
426 | sp->free = b; | |
427 | } else { | |
428 | prev = sp->free; | |
429 | next = slob_next(prev); | |
430 | while (b > next) { | |
431 | prev = next; | |
432 | next = slob_next(prev); | |
433 | } | |
10cef602 | 434 | |
95b35127 NP |
435 | if (!slob_last(prev) && b + units == next) { |
436 | units += slob_units(next); | |
437 | set_slob(b, units, slob_next(next)); | |
438 | } else | |
439 | set_slob(b, units, next); | |
440 | ||
441 | if (prev + slob_units(prev) == b) { | |
442 | units = slob_units(b) + slob_units(prev); | |
443 | set_slob(prev, units, slob_next(b)); | |
444 | } else | |
445 | set_slob(prev, slob_units(prev), b); | |
446 | } | |
447 | out: | |
10cef602 MM |
448 | spin_unlock_irqrestore(&slob_lock, flags); |
449 | } | |
450 | ||
95b35127 NP |
451 | /* |
452 | * End of slob allocator proper. Begin kmem_cache_alloc and kmalloc frontend. | |
453 | */ | |
454 | ||
55394849 NP |
455 | #ifndef ARCH_KMALLOC_MINALIGN |
456 | #define ARCH_KMALLOC_MINALIGN __alignof__(unsigned long) | |
457 | #endif | |
458 | ||
459 | #ifndef ARCH_SLAB_MINALIGN | |
460 | #define ARCH_SLAB_MINALIGN __alignof__(unsigned long) | |
461 | #endif | |
462 | ||
6193a2ff | 463 | void *__kmalloc_node(size_t size, gfp_t gfp, int node) |
10cef602 | 464 | { |
6cb8f913 | 465 | unsigned int *m; |
55394849 | 466 | int align = max(ARCH_KMALLOC_MINALIGN, ARCH_SLAB_MINALIGN); |
3eae2cb2 | 467 | void *ret; |
55394849 NP |
468 | |
469 | if (size < PAGE_SIZE - align) { | |
6cb8f913 CL |
470 | if (!size) |
471 | return ZERO_SIZE_PTR; | |
472 | ||
6193a2ff | 473 | m = slob_alloc(size + align, gfp, align, node); |
3eae2cb2 | 474 | |
239f49c0 MK |
475 | if (!m) |
476 | return NULL; | |
477 | *m = size; | |
3eae2cb2 EGM |
478 | ret = (void *)m + align; |
479 | ||
480 | kmemtrace_mark_alloc_node(KMEMTRACE_TYPE_KMALLOC, | |
481 | _RET_IP_, ret, | |
482 | size, size + align, gfp, node); | |
d87a133f | 483 | } else { |
3eae2cb2 | 484 | unsigned int order = get_order(size); |
d87a133f | 485 | |
3eae2cb2 | 486 | ret = slob_new_page(gfp | __GFP_COMP, order, node); |
d87a133f NP |
487 | if (ret) { |
488 | struct page *page; | |
489 | page = virt_to_page(ret); | |
490 | page->private = size; | |
491 | } | |
3eae2cb2 EGM |
492 | |
493 | kmemtrace_mark_alloc_node(KMEMTRACE_TYPE_KMALLOC, | |
494 | _RET_IP_, ret, | |
495 | size, PAGE_SIZE << order, gfp, node); | |
10cef602 | 496 | } |
3eae2cb2 EGM |
497 | |
498 | return ret; | |
10cef602 | 499 | } |
6193a2ff | 500 | EXPORT_SYMBOL(__kmalloc_node); |
10cef602 MM |
501 | |
502 | void kfree(const void *block) | |
503 | { | |
95b35127 | 504 | struct slob_page *sp; |
10cef602 | 505 | |
2408c550 | 506 | if (unlikely(ZERO_OR_NULL_PTR(block))) |
10cef602 MM |
507 | return; |
508 | ||
95b35127 | 509 | sp = (struct slob_page *)virt_to_page(block); |
d87a133f | 510 | if (slob_page(sp)) { |
55394849 NP |
511 | int align = max(ARCH_KMALLOC_MINALIGN, ARCH_SLAB_MINALIGN); |
512 | unsigned int *m = (unsigned int *)(block - align); | |
513 | slob_free(m, *m + align); | |
d87a133f NP |
514 | } else |
515 | put_page(&sp->page); | |
3eae2cb2 EGM |
516 | |
517 | kmemtrace_mark_free(KMEMTRACE_TYPE_KMALLOC, _RET_IP_, block); | |
10cef602 | 518 | } |
10cef602 MM |
519 | EXPORT_SYMBOL(kfree); |
520 | ||
d87a133f | 521 | /* can't use ksize for kmem_cache_alloc memory, only kmalloc */ |
fd76bab2 | 522 | size_t ksize(const void *block) |
10cef602 | 523 | { |
95b35127 | 524 | struct slob_page *sp; |
10cef602 | 525 | |
ef8b4520 CL |
526 | BUG_ON(!block); |
527 | if (unlikely(block == ZERO_SIZE_PTR)) | |
10cef602 MM |
528 | return 0; |
529 | ||
95b35127 | 530 | sp = (struct slob_page *)virt_to_page(block); |
70096a56 MM |
531 | if (slob_page(sp)) { |
532 | int align = max(ARCH_KMALLOC_MINALIGN, ARCH_SLAB_MINALIGN); | |
533 | unsigned int *m = (unsigned int *)(block - align); | |
534 | return SLOB_UNITS(*m) * SLOB_UNIT; | |
535 | } else | |
d87a133f | 536 | return sp->page.private; |
10cef602 MM |
537 | } |
538 | ||
539 | struct kmem_cache { | |
540 | unsigned int size, align; | |
afc0cedb | 541 | unsigned long flags; |
10cef602 | 542 | const char *name; |
51cc5068 | 543 | void (*ctor)(void *); |
10cef602 MM |
544 | }; |
545 | ||
546 | struct kmem_cache *kmem_cache_create(const char *name, size_t size, | |
51cc5068 | 547 | size_t align, unsigned long flags, void (*ctor)(void *)) |
10cef602 MM |
548 | { |
549 | struct kmem_cache *c; | |
550 | ||
0701a9e6 | 551 | c = slob_alloc(sizeof(struct kmem_cache), |
5e18e2b8 | 552 | GFP_KERNEL, ARCH_KMALLOC_MINALIGN, -1); |
10cef602 MM |
553 | |
554 | if (c) { | |
555 | c->name = name; | |
556 | c->size = size; | |
afc0cedb | 557 | if (flags & SLAB_DESTROY_BY_RCU) { |
afc0cedb NP |
558 | /* leave room for rcu footer at the end of object */ |
559 | c->size += sizeof(struct slob_rcu); | |
560 | } | |
561 | c->flags = flags; | |
10cef602 | 562 | c->ctor = ctor; |
10cef602 | 563 | /* ignore alignment unless it's forced */ |
5af60839 | 564 | c->align = (flags & SLAB_HWCACHE_ALIGN) ? SLOB_ALIGN : 0; |
55394849 NP |
565 | if (c->align < ARCH_SLAB_MINALIGN) |
566 | c->align = ARCH_SLAB_MINALIGN; | |
10cef602 MM |
567 | if (c->align < align) |
568 | c->align = align; | |
bc0055ae AM |
569 | } else if (flags & SLAB_PANIC) |
570 | panic("Cannot create slab cache %s\n", name); | |
10cef602 MM |
571 | |
572 | return c; | |
573 | } | |
574 | EXPORT_SYMBOL(kmem_cache_create); | |
575 | ||
133d205a | 576 | void kmem_cache_destroy(struct kmem_cache *c) |
10cef602 MM |
577 | { |
578 | slob_free(c, sizeof(struct kmem_cache)); | |
10cef602 MM |
579 | } |
580 | EXPORT_SYMBOL(kmem_cache_destroy); | |
581 | ||
6193a2ff | 582 | void *kmem_cache_alloc_node(struct kmem_cache *c, gfp_t flags, int node) |
10cef602 MM |
583 | { |
584 | void *b; | |
585 | ||
3eae2cb2 | 586 | if (c->size < PAGE_SIZE) { |
6193a2ff | 587 | b = slob_alloc(c->size, flags, c->align, node); |
3eae2cb2 EGM |
588 | kmemtrace_mark_alloc_node(KMEMTRACE_TYPE_CACHE, |
589 | _RET_IP_, b, c->size, | |
590 | SLOB_UNITS(c->size) * SLOB_UNIT, | |
591 | flags, node); | |
592 | } else { | |
6193a2ff | 593 | b = slob_new_page(flags, get_order(c->size), node); |
3eae2cb2 EGM |
594 | kmemtrace_mark_alloc_node(KMEMTRACE_TYPE_CACHE, |
595 | _RET_IP_, b, c->size, | |
596 | PAGE_SIZE << get_order(c->size), | |
597 | flags, node); | |
598 | } | |
10cef602 MM |
599 | |
600 | if (c->ctor) | |
51cc5068 | 601 | c->ctor(b); |
10cef602 MM |
602 | |
603 | return b; | |
604 | } | |
6193a2ff | 605 | EXPORT_SYMBOL(kmem_cache_alloc_node); |
10cef602 | 606 | |
afc0cedb | 607 | static void __kmem_cache_free(void *b, int size) |
10cef602 | 608 | { |
afc0cedb NP |
609 | if (size < PAGE_SIZE) |
610 | slob_free(b, size); | |
10cef602 | 611 | else |
afc0cedb NP |
612 | free_pages((unsigned long)b, get_order(size)); |
613 | } | |
614 | ||
615 | static void kmem_rcu_free(struct rcu_head *head) | |
616 | { | |
617 | struct slob_rcu *slob_rcu = (struct slob_rcu *)head; | |
618 | void *b = (void *)slob_rcu - (slob_rcu->size - sizeof(struct slob_rcu)); | |
619 | ||
620 | __kmem_cache_free(b, slob_rcu->size); | |
621 | } | |
622 | ||
623 | void kmem_cache_free(struct kmem_cache *c, void *b) | |
624 | { | |
625 | if (unlikely(c->flags & SLAB_DESTROY_BY_RCU)) { | |
626 | struct slob_rcu *slob_rcu; | |
627 | slob_rcu = b + (c->size - sizeof(struct slob_rcu)); | |
628 | INIT_RCU_HEAD(&slob_rcu->head); | |
629 | slob_rcu->size = c->size; | |
630 | call_rcu(&slob_rcu->head, kmem_rcu_free); | |
631 | } else { | |
afc0cedb NP |
632 | __kmem_cache_free(b, c->size); |
633 | } | |
3eae2cb2 EGM |
634 | |
635 | kmemtrace_mark_free(KMEMTRACE_TYPE_CACHE, _RET_IP_, b); | |
10cef602 MM |
636 | } |
637 | EXPORT_SYMBOL(kmem_cache_free); | |
638 | ||
639 | unsigned int kmem_cache_size(struct kmem_cache *c) | |
640 | { | |
641 | return c->size; | |
642 | } | |
643 | EXPORT_SYMBOL(kmem_cache_size); | |
644 | ||
645 | const char *kmem_cache_name(struct kmem_cache *c) | |
646 | { | |
647 | return c->name; | |
648 | } | |
649 | EXPORT_SYMBOL(kmem_cache_name); | |
650 | ||
2e892f43 CL |
651 | int kmem_cache_shrink(struct kmem_cache *d) |
652 | { | |
653 | return 0; | |
654 | } | |
655 | EXPORT_SYMBOL(kmem_cache_shrink); | |
656 | ||
55935a34 | 657 | int kmem_ptr_validate(struct kmem_cache *a, const void *b) |
2e892f43 CL |
658 | { |
659 | return 0; | |
660 | } | |
661 | ||
84a01c2f PM |
662 | static unsigned int slob_ready __read_mostly; |
663 | ||
664 | int slab_is_available(void) | |
665 | { | |
666 | return slob_ready; | |
667 | } | |
668 | ||
bcb4ddb4 DG |
669 | void __init kmem_cache_init(void) |
670 | { | |
84a01c2f | 671 | slob_ready = 1; |
10cef602 | 672 | } |