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> | |
1f0532eb | 63 | #include <linux/swap.h> /* struct reclaim_state */ |
10cef602 MM |
64 | #include <linux/cache.h> |
65 | #include <linux/init.h> | |
66 | #include <linux/module.h> | |
afc0cedb | 67 | #include <linux/rcupdate.h> |
95b35127 | 68 | #include <linux/list.h> |
36994e58 | 69 | #include <trace/kmemtrace.h> |
95b35127 NP |
70 | #include <asm/atomic.h> |
71 | ||
95b35127 NP |
72 | /* |
73 | * slob_block has a field 'units', which indicates size of block if +ve, | |
74 | * or offset of next block if -ve (in SLOB_UNITs). | |
75 | * | |
76 | * Free blocks of size 1 unit simply contain the offset of the next block. | |
77 | * Those with larger size contain their size in the first SLOB_UNIT of | |
78 | * memory, and the offset of the next free block in the second SLOB_UNIT. | |
79 | */ | |
55394849 | 80 | #if PAGE_SIZE <= (32767 * 2) |
95b35127 NP |
81 | typedef s16 slobidx_t; |
82 | #else | |
83 | typedef s32 slobidx_t; | |
84 | #endif | |
85 | ||
10cef602 | 86 | struct slob_block { |
95b35127 | 87 | slobidx_t units; |
55394849 | 88 | }; |
10cef602 MM |
89 | typedef struct slob_block slob_t; |
90 | ||
95b35127 NP |
91 | /* |
92 | * We use struct page fields to manage some slob allocation aspects, | |
93 | * however to avoid the horrible mess in include/linux/mm_types.h, we'll | |
94 | * just define our own struct page type variant here. | |
95 | */ | |
96 | struct slob_page { | |
97 | union { | |
98 | struct { | |
99 | unsigned long flags; /* mandatory */ | |
100 | atomic_t _count; /* mandatory */ | |
101 | slobidx_t units; /* free units left in page */ | |
102 | unsigned long pad[2]; | |
103 | slob_t *free; /* first free slob_t in page */ | |
104 | struct list_head list; /* linked list of free pages */ | |
105 | }; | |
106 | struct page page; | |
107 | }; | |
108 | }; | |
109 | static inline void struct_slob_page_wrong_size(void) | |
110 | { BUILD_BUG_ON(sizeof(struct slob_page) != sizeof(struct page)); } | |
111 | ||
112 | /* | |
113 | * free_slob_page: call before a slob_page is returned to the page allocator. | |
114 | */ | |
115 | static inline void free_slob_page(struct slob_page *sp) | |
116 | { | |
117 | reset_page_mapcount(&sp->page); | |
118 | sp->page.mapping = NULL; | |
119 | } | |
120 | ||
121 | /* | |
20cecbae | 122 | * All partially free slob pages go on these lists. |
95b35127 | 123 | */ |
20cecbae MM |
124 | #define SLOB_BREAK1 256 |
125 | #define SLOB_BREAK2 1024 | |
126 | static LIST_HEAD(free_slob_small); | |
127 | static LIST_HEAD(free_slob_medium); | |
128 | static LIST_HEAD(free_slob_large); | |
95b35127 NP |
129 | |
130 | /* | |
6e9ed0cc | 131 | * is_slob_page: True for all slob pages (false for bigblock pages) |
95b35127 | 132 | */ |
6e9ed0cc | 133 | static inline int is_slob_page(struct slob_page *sp) |
95b35127 | 134 | { |
9023cb7e | 135 | return PageSlobPage((struct page *)sp); |
95b35127 NP |
136 | } |
137 | ||
138 | static inline void set_slob_page(struct slob_page *sp) | |
139 | { | |
9023cb7e | 140 | __SetPageSlobPage((struct page *)sp); |
95b35127 NP |
141 | } |
142 | ||
143 | static inline void clear_slob_page(struct slob_page *sp) | |
144 | { | |
9023cb7e | 145 | __ClearPageSlobPage((struct page *)sp); |
95b35127 NP |
146 | } |
147 | ||
6e9ed0cc AW |
148 | static inline struct slob_page *slob_page(const void *addr) |
149 | { | |
150 | return (struct slob_page *)virt_to_page(addr); | |
151 | } | |
152 | ||
95b35127 NP |
153 | /* |
154 | * slob_page_free: true for pages on free_slob_pages list. | |
155 | */ | |
156 | static inline int slob_page_free(struct slob_page *sp) | |
157 | { | |
9023cb7e | 158 | return PageSlobFree((struct page *)sp); |
95b35127 NP |
159 | } |
160 | ||
20cecbae | 161 | static void set_slob_page_free(struct slob_page *sp, struct list_head *list) |
95b35127 | 162 | { |
20cecbae | 163 | list_add(&sp->list, list); |
9023cb7e | 164 | __SetPageSlobFree((struct page *)sp); |
95b35127 NP |
165 | } |
166 | ||
167 | static inline void clear_slob_page_free(struct slob_page *sp) | |
168 | { | |
169 | list_del(&sp->list); | |
9023cb7e | 170 | __ClearPageSlobFree((struct page *)sp); |
95b35127 NP |
171 | } |
172 | ||
10cef602 MM |
173 | #define SLOB_UNIT sizeof(slob_t) |
174 | #define SLOB_UNITS(size) (((size) + SLOB_UNIT - 1)/SLOB_UNIT) | |
175 | #define SLOB_ALIGN L1_CACHE_BYTES | |
176 | ||
afc0cedb NP |
177 | /* |
178 | * struct slob_rcu is inserted at the tail of allocated slob blocks, which | |
179 | * were created with a SLAB_DESTROY_BY_RCU slab. slob_rcu is used to free | |
180 | * the block using call_rcu. | |
181 | */ | |
182 | struct slob_rcu { | |
183 | struct rcu_head head; | |
184 | int size; | |
185 | }; | |
186 | ||
95b35127 NP |
187 | /* |
188 | * slob_lock protects all slob allocator structures. | |
189 | */ | |
10cef602 | 190 | static DEFINE_SPINLOCK(slob_lock); |
10cef602 | 191 | |
95b35127 NP |
192 | /* |
193 | * Encode the given size and next info into a free slob block s. | |
194 | */ | |
195 | static void set_slob(slob_t *s, slobidx_t size, slob_t *next) | |
196 | { | |
197 | slob_t *base = (slob_t *)((unsigned long)s & PAGE_MASK); | |
198 | slobidx_t offset = next - base; | |
bcb4ddb4 | 199 | |
95b35127 NP |
200 | if (size > 1) { |
201 | s[0].units = size; | |
202 | s[1].units = offset; | |
203 | } else | |
204 | s[0].units = -offset; | |
205 | } | |
10cef602 | 206 | |
95b35127 NP |
207 | /* |
208 | * Return the size of a slob block. | |
209 | */ | |
210 | static slobidx_t slob_units(slob_t *s) | |
211 | { | |
212 | if (s->units > 0) | |
213 | return s->units; | |
214 | return 1; | |
215 | } | |
216 | ||
217 | /* | |
218 | * Return the next free slob block pointer after this one. | |
219 | */ | |
220 | static slob_t *slob_next(slob_t *s) | |
221 | { | |
222 | slob_t *base = (slob_t *)((unsigned long)s & PAGE_MASK); | |
223 | slobidx_t next; | |
224 | ||
225 | if (s[0].units < 0) | |
226 | next = -s[0].units; | |
227 | else | |
228 | next = s[1].units; | |
229 | return base+next; | |
230 | } | |
231 | ||
232 | /* | |
233 | * Returns true if s is the last free block in its page. | |
234 | */ | |
235 | static int slob_last(slob_t *s) | |
236 | { | |
237 | return !((unsigned long)slob_next(s) & ~PAGE_MASK); | |
238 | } | |
239 | ||
6e9ed0cc | 240 | static void *slob_new_pages(gfp_t gfp, int order, int node) |
6193a2ff PM |
241 | { |
242 | void *page; | |
243 | ||
244 | #ifdef CONFIG_NUMA | |
245 | if (node != -1) | |
246 | page = alloc_pages_node(node, gfp, order); | |
247 | else | |
248 | #endif | |
249 | page = alloc_pages(gfp, order); | |
250 | ||
251 | if (!page) | |
252 | return NULL; | |
253 | ||
254 | return page_address(page); | |
255 | } | |
256 | ||
6e9ed0cc AW |
257 | static void slob_free_pages(void *b, int order) |
258 | { | |
1f0532eb NP |
259 | if (current->reclaim_state) |
260 | current->reclaim_state->reclaimed_slab += 1 << order; | |
6e9ed0cc AW |
261 | free_pages((unsigned long)b, order); |
262 | } | |
263 | ||
95b35127 NP |
264 | /* |
265 | * Allocate a slob block within a given slob_page sp. | |
266 | */ | |
267 | static void *slob_page_alloc(struct slob_page *sp, size_t size, int align) | |
10cef602 | 268 | { |
6e9ed0cc | 269 | slob_t *prev, *cur, *aligned = NULL; |
10cef602 | 270 | int delta = 0, units = SLOB_UNITS(size); |
10cef602 | 271 | |
95b35127 NP |
272 | for (prev = NULL, cur = sp->free; ; prev = cur, cur = slob_next(cur)) { |
273 | slobidx_t avail = slob_units(cur); | |
274 | ||
10cef602 MM |
275 | if (align) { |
276 | aligned = (slob_t *)ALIGN((unsigned long)cur, align); | |
277 | delta = aligned - cur; | |
278 | } | |
95b35127 NP |
279 | if (avail >= units + delta) { /* room enough? */ |
280 | slob_t *next; | |
281 | ||
10cef602 | 282 | if (delta) { /* need to fragment head to align? */ |
95b35127 NP |
283 | next = slob_next(cur); |
284 | set_slob(aligned, avail - delta, next); | |
285 | set_slob(cur, delta, aligned); | |
10cef602 MM |
286 | prev = cur; |
287 | cur = aligned; | |
95b35127 | 288 | avail = slob_units(cur); |
10cef602 MM |
289 | } |
290 | ||
95b35127 NP |
291 | next = slob_next(cur); |
292 | if (avail == units) { /* exact fit? unlink. */ | |
293 | if (prev) | |
294 | set_slob(prev, slob_units(prev), next); | |
295 | else | |
296 | sp->free = next; | |
297 | } else { /* fragment */ | |
298 | if (prev) | |
299 | set_slob(prev, slob_units(prev), cur + units); | |
300 | else | |
301 | sp->free = cur + units; | |
302 | set_slob(cur + units, avail - units, next); | |
10cef602 MM |
303 | } |
304 | ||
95b35127 NP |
305 | sp->units -= units; |
306 | if (!sp->units) | |
307 | clear_slob_page_free(sp); | |
10cef602 MM |
308 | return cur; |
309 | } | |
95b35127 NP |
310 | if (slob_last(cur)) |
311 | return NULL; | |
312 | } | |
313 | } | |
10cef602 | 314 | |
95b35127 NP |
315 | /* |
316 | * slob_alloc: entry point into the slob allocator. | |
317 | */ | |
6193a2ff | 318 | static void *slob_alloc(size_t size, gfp_t gfp, int align, int node) |
95b35127 NP |
319 | { |
320 | struct slob_page *sp; | |
d6269543 | 321 | struct list_head *prev; |
20cecbae | 322 | struct list_head *slob_list; |
95b35127 NP |
323 | slob_t *b = NULL; |
324 | unsigned long flags; | |
10cef602 | 325 | |
20cecbae MM |
326 | if (size < SLOB_BREAK1) |
327 | slob_list = &free_slob_small; | |
328 | else if (size < SLOB_BREAK2) | |
329 | slob_list = &free_slob_medium; | |
330 | else | |
331 | slob_list = &free_slob_large; | |
332 | ||
95b35127 NP |
333 | spin_lock_irqsave(&slob_lock, flags); |
334 | /* Iterate through each partially free page, try to find room */ | |
20cecbae | 335 | list_for_each_entry(sp, slob_list, list) { |
6193a2ff PM |
336 | #ifdef CONFIG_NUMA |
337 | /* | |
338 | * If there's a node specification, search for a partial | |
339 | * page with a matching node id in the freelist. | |
340 | */ | |
341 | if (node != -1 && page_to_nid(&sp->page) != node) | |
342 | continue; | |
343 | #endif | |
d6269543 MM |
344 | /* Enough room on this page? */ |
345 | if (sp->units < SLOB_UNITS(size)) | |
346 | continue; | |
6193a2ff | 347 | |
d6269543 MM |
348 | /* Attempt to alloc */ |
349 | prev = sp->list.prev; | |
350 | b = slob_page_alloc(sp, size, align); | |
351 | if (!b) | |
352 | continue; | |
353 | ||
354 | /* Improve fragment distribution and reduce our average | |
355 | * search time by starting our next search here. (see | |
356 | * Knuth vol 1, sec 2.5, pg 449) */ | |
20cecbae MM |
357 | if (prev != slob_list->prev && |
358 | slob_list->next != prev->next) | |
359 | list_move_tail(slob_list, prev->next); | |
d6269543 | 360 | break; |
10cef602 | 361 | } |
95b35127 NP |
362 | spin_unlock_irqrestore(&slob_lock, flags); |
363 | ||
364 | /* Not enough space: must allocate a new page */ | |
365 | if (!b) { | |
6e9ed0cc | 366 | b = slob_new_pages(gfp & ~__GFP_ZERO, 0, node); |
95b35127 | 367 | if (!b) |
6e9ed0cc AW |
368 | return NULL; |
369 | sp = slob_page(b); | |
95b35127 NP |
370 | set_slob_page(sp); |
371 | ||
372 | spin_lock_irqsave(&slob_lock, flags); | |
373 | sp->units = SLOB_UNITS(PAGE_SIZE); | |
374 | sp->free = b; | |
375 | INIT_LIST_HEAD(&sp->list); | |
376 | set_slob(b, SLOB_UNITS(PAGE_SIZE), b + SLOB_UNITS(PAGE_SIZE)); | |
20cecbae | 377 | set_slob_page_free(sp, slob_list); |
95b35127 NP |
378 | b = slob_page_alloc(sp, size, align); |
379 | BUG_ON(!b); | |
380 | spin_unlock_irqrestore(&slob_lock, flags); | |
381 | } | |
d07dbea4 CL |
382 | if (unlikely((gfp & __GFP_ZERO) && b)) |
383 | memset(b, 0, size); | |
95b35127 | 384 | return b; |
10cef602 MM |
385 | } |
386 | ||
95b35127 NP |
387 | /* |
388 | * slob_free: entry point into the slob allocator. | |
389 | */ | |
10cef602 MM |
390 | static void slob_free(void *block, int size) |
391 | { | |
95b35127 NP |
392 | struct slob_page *sp; |
393 | slob_t *prev, *next, *b = (slob_t *)block; | |
394 | slobidx_t units; | |
10cef602 MM |
395 | unsigned long flags; |
396 | ||
2408c550 | 397 | if (unlikely(ZERO_OR_NULL_PTR(block))) |
10cef602 | 398 | return; |
95b35127 | 399 | BUG_ON(!size); |
10cef602 | 400 | |
6e9ed0cc | 401 | sp = slob_page(block); |
95b35127 | 402 | units = SLOB_UNITS(size); |
10cef602 | 403 | |
10cef602 | 404 | spin_lock_irqsave(&slob_lock, flags); |
10cef602 | 405 | |
95b35127 NP |
406 | if (sp->units + units == SLOB_UNITS(PAGE_SIZE)) { |
407 | /* Go directly to page allocator. Do not pass slob allocator */ | |
408 | if (slob_page_free(sp)) | |
409 | clear_slob_page_free(sp); | |
6fb8f424 | 410 | spin_unlock_irqrestore(&slob_lock, flags); |
95b35127 NP |
411 | clear_slob_page(sp); |
412 | free_slob_page(sp); | |
1f0532eb | 413 | slob_free_pages(b, 0); |
6fb8f424 | 414 | return; |
95b35127 | 415 | } |
10cef602 | 416 | |
95b35127 NP |
417 | if (!slob_page_free(sp)) { |
418 | /* This slob page is about to become partially free. Easy! */ | |
419 | sp->units = units; | |
420 | sp->free = b; | |
421 | set_slob(b, units, | |
422 | (void *)((unsigned long)(b + | |
423 | SLOB_UNITS(PAGE_SIZE)) & PAGE_MASK)); | |
20cecbae | 424 | set_slob_page_free(sp, &free_slob_small); |
95b35127 NP |
425 | goto out; |
426 | } | |
427 | ||
428 | /* | |
429 | * Otherwise the page is already partially free, so find reinsertion | |
430 | * point. | |
431 | */ | |
432 | sp->units += units; | |
10cef602 | 433 | |
95b35127 | 434 | if (b < sp->free) { |
679299b3 MM |
435 | if (b + units == sp->free) { |
436 | units += slob_units(sp->free); | |
437 | sp->free = slob_next(sp->free); | |
438 | } | |
95b35127 NP |
439 | set_slob(b, units, sp->free); |
440 | sp->free = b; | |
441 | } else { | |
442 | prev = sp->free; | |
443 | next = slob_next(prev); | |
444 | while (b > next) { | |
445 | prev = next; | |
446 | next = slob_next(prev); | |
447 | } | |
10cef602 | 448 | |
95b35127 NP |
449 | if (!slob_last(prev) && b + units == next) { |
450 | units += slob_units(next); | |
451 | set_slob(b, units, slob_next(next)); | |
452 | } else | |
453 | set_slob(b, units, next); | |
454 | ||
455 | if (prev + slob_units(prev) == b) { | |
456 | units = slob_units(b) + slob_units(prev); | |
457 | set_slob(prev, units, slob_next(b)); | |
458 | } else | |
459 | set_slob(prev, slob_units(prev), b); | |
460 | } | |
461 | out: | |
10cef602 MM |
462 | spin_unlock_irqrestore(&slob_lock, flags); |
463 | } | |
464 | ||
95b35127 NP |
465 | /* |
466 | * End of slob allocator proper. Begin kmem_cache_alloc and kmalloc frontend. | |
467 | */ | |
468 | ||
55394849 NP |
469 | #ifndef ARCH_KMALLOC_MINALIGN |
470 | #define ARCH_KMALLOC_MINALIGN __alignof__(unsigned long) | |
471 | #endif | |
472 | ||
473 | #ifndef ARCH_SLAB_MINALIGN | |
474 | #define ARCH_SLAB_MINALIGN __alignof__(unsigned long) | |
475 | #endif | |
476 | ||
6193a2ff | 477 | void *__kmalloc_node(size_t size, gfp_t gfp, int node) |
10cef602 | 478 | { |
6cb8f913 | 479 | unsigned int *m; |
55394849 | 480 | int align = max(ARCH_KMALLOC_MINALIGN, ARCH_SLAB_MINALIGN); |
3eae2cb2 | 481 | void *ret; |
55394849 | 482 | |
19cefdff | 483 | lockdep_trace_alloc(gfp); |
cf40bd16 | 484 | |
55394849 | 485 | if (size < PAGE_SIZE - align) { |
6cb8f913 CL |
486 | if (!size) |
487 | return ZERO_SIZE_PTR; | |
488 | ||
6193a2ff | 489 | m = slob_alloc(size + align, gfp, align, node); |
3eae2cb2 | 490 | |
239f49c0 MK |
491 | if (!m) |
492 | return NULL; | |
493 | *m = size; | |
3eae2cb2 EGM |
494 | ret = (void *)m + align; |
495 | ||
ca2b84cb EGM |
496 | trace_kmalloc_node(_RET_IP_, ret, |
497 | size, size + align, gfp, node); | |
d87a133f | 498 | } else { |
3eae2cb2 | 499 | unsigned int order = get_order(size); |
d87a133f | 500 | |
6e9ed0cc | 501 | ret = slob_new_pages(gfp | __GFP_COMP, get_order(size), node); |
d87a133f NP |
502 | if (ret) { |
503 | struct page *page; | |
504 | page = virt_to_page(ret); | |
505 | page->private = size; | |
506 | } | |
3eae2cb2 | 507 | |
ca2b84cb EGM |
508 | trace_kmalloc_node(_RET_IP_, ret, |
509 | size, PAGE_SIZE << order, gfp, node); | |
10cef602 | 510 | } |
3eae2cb2 EGM |
511 | |
512 | return ret; | |
10cef602 | 513 | } |
6193a2ff | 514 | EXPORT_SYMBOL(__kmalloc_node); |
10cef602 MM |
515 | |
516 | void kfree(const void *block) | |
517 | { | |
95b35127 | 518 | struct slob_page *sp; |
10cef602 | 519 | |
2121db74 PE |
520 | trace_kfree(_RET_IP_, block); |
521 | ||
2408c550 | 522 | if (unlikely(ZERO_OR_NULL_PTR(block))) |
10cef602 MM |
523 | return; |
524 | ||
6e9ed0cc AW |
525 | sp = slob_page(block); |
526 | if (is_slob_page(sp)) { | |
55394849 NP |
527 | int align = max(ARCH_KMALLOC_MINALIGN, ARCH_SLAB_MINALIGN); |
528 | unsigned int *m = (unsigned int *)(block - align); | |
529 | slob_free(m, *m + align); | |
d87a133f NP |
530 | } else |
531 | put_page(&sp->page); | |
10cef602 | 532 | } |
10cef602 MM |
533 | EXPORT_SYMBOL(kfree); |
534 | ||
d87a133f | 535 | /* can't use ksize for kmem_cache_alloc memory, only kmalloc */ |
fd76bab2 | 536 | size_t ksize(const void *block) |
10cef602 | 537 | { |
95b35127 | 538 | struct slob_page *sp; |
10cef602 | 539 | |
ef8b4520 CL |
540 | BUG_ON(!block); |
541 | if (unlikely(block == ZERO_SIZE_PTR)) | |
10cef602 MM |
542 | return 0; |
543 | ||
6e9ed0cc AW |
544 | sp = slob_page(block); |
545 | if (is_slob_page(sp)) { | |
70096a56 MM |
546 | int align = max(ARCH_KMALLOC_MINALIGN, ARCH_SLAB_MINALIGN); |
547 | unsigned int *m = (unsigned int *)(block - align); | |
548 | return SLOB_UNITS(*m) * SLOB_UNIT; | |
549 | } else | |
d87a133f | 550 | return sp->page.private; |
10cef602 | 551 | } |
b1aabecd | 552 | EXPORT_SYMBOL(ksize); |
10cef602 MM |
553 | |
554 | struct kmem_cache { | |
555 | unsigned int size, align; | |
afc0cedb | 556 | unsigned long flags; |
10cef602 | 557 | const char *name; |
51cc5068 | 558 | void (*ctor)(void *); |
10cef602 MM |
559 | }; |
560 | ||
561 | struct kmem_cache *kmem_cache_create(const char *name, size_t size, | |
51cc5068 | 562 | size_t align, unsigned long flags, void (*ctor)(void *)) |
10cef602 MM |
563 | { |
564 | struct kmem_cache *c; | |
565 | ||
0701a9e6 | 566 | c = slob_alloc(sizeof(struct kmem_cache), |
5e18e2b8 | 567 | GFP_KERNEL, ARCH_KMALLOC_MINALIGN, -1); |
10cef602 MM |
568 | |
569 | if (c) { | |
570 | c->name = name; | |
571 | c->size = size; | |
afc0cedb | 572 | if (flags & SLAB_DESTROY_BY_RCU) { |
afc0cedb NP |
573 | /* leave room for rcu footer at the end of object */ |
574 | c->size += sizeof(struct slob_rcu); | |
575 | } | |
576 | c->flags = flags; | |
10cef602 | 577 | c->ctor = ctor; |
10cef602 | 578 | /* ignore alignment unless it's forced */ |
5af60839 | 579 | c->align = (flags & SLAB_HWCACHE_ALIGN) ? SLOB_ALIGN : 0; |
55394849 NP |
580 | if (c->align < ARCH_SLAB_MINALIGN) |
581 | c->align = ARCH_SLAB_MINALIGN; | |
10cef602 MM |
582 | if (c->align < align) |
583 | c->align = align; | |
bc0055ae AM |
584 | } else if (flags & SLAB_PANIC) |
585 | panic("Cannot create slab cache %s\n", name); | |
10cef602 MM |
586 | |
587 | return c; | |
588 | } | |
589 | EXPORT_SYMBOL(kmem_cache_create); | |
590 | ||
133d205a | 591 | void kmem_cache_destroy(struct kmem_cache *c) |
10cef602 MM |
592 | { |
593 | slob_free(c, sizeof(struct kmem_cache)); | |
10cef602 MM |
594 | } |
595 | EXPORT_SYMBOL(kmem_cache_destroy); | |
596 | ||
6193a2ff | 597 | void *kmem_cache_alloc_node(struct kmem_cache *c, gfp_t flags, int node) |
10cef602 MM |
598 | { |
599 | void *b; | |
600 | ||
3eae2cb2 | 601 | if (c->size < PAGE_SIZE) { |
6193a2ff | 602 | b = slob_alloc(c->size, flags, c->align, node); |
ca2b84cb EGM |
603 | trace_kmem_cache_alloc_node(_RET_IP_, b, c->size, |
604 | SLOB_UNITS(c->size) * SLOB_UNIT, | |
605 | flags, node); | |
3eae2cb2 | 606 | } else { |
6e9ed0cc | 607 | b = slob_new_pages(flags, get_order(c->size), node); |
ca2b84cb EGM |
608 | trace_kmem_cache_alloc_node(_RET_IP_, b, c->size, |
609 | PAGE_SIZE << get_order(c->size), | |
610 | flags, node); | |
3eae2cb2 | 611 | } |
10cef602 MM |
612 | |
613 | if (c->ctor) | |
51cc5068 | 614 | c->ctor(b); |
10cef602 MM |
615 | |
616 | return b; | |
617 | } | |
6193a2ff | 618 | EXPORT_SYMBOL(kmem_cache_alloc_node); |
10cef602 | 619 | |
afc0cedb | 620 | static void __kmem_cache_free(void *b, int size) |
10cef602 | 621 | { |
afc0cedb NP |
622 | if (size < PAGE_SIZE) |
623 | slob_free(b, size); | |
10cef602 | 624 | else |
6e9ed0cc | 625 | slob_free_pages(b, get_order(size)); |
afc0cedb NP |
626 | } |
627 | ||
628 | static void kmem_rcu_free(struct rcu_head *head) | |
629 | { | |
630 | struct slob_rcu *slob_rcu = (struct slob_rcu *)head; | |
631 | void *b = (void *)slob_rcu - (slob_rcu->size - sizeof(struct slob_rcu)); | |
632 | ||
633 | __kmem_cache_free(b, slob_rcu->size); | |
634 | } | |
635 | ||
636 | void kmem_cache_free(struct kmem_cache *c, void *b) | |
637 | { | |
638 | if (unlikely(c->flags & SLAB_DESTROY_BY_RCU)) { | |
639 | struct slob_rcu *slob_rcu; | |
640 | slob_rcu = b + (c->size - sizeof(struct slob_rcu)); | |
641 | INIT_RCU_HEAD(&slob_rcu->head); | |
642 | slob_rcu->size = c->size; | |
643 | call_rcu(&slob_rcu->head, kmem_rcu_free); | |
644 | } else { | |
afc0cedb NP |
645 | __kmem_cache_free(b, c->size); |
646 | } | |
3eae2cb2 | 647 | |
ca2b84cb | 648 | trace_kmem_cache_free(_RET_IP_, b); |
10cef602 MM |
649 | } |
650 | EXPORT_SYMBOL(kmem_cache_free); | |
651 | ||
652 | unsigned int kmem_cache_size(struct kmem_cache *c) | |
653 | { | |
654 | return c->size; | |
655 | } | |
656 | EXPORT_SYMBOL(kmem_cache_size); | |
657 | ||
658 | const char *kmem_cache_name(struct kmem_cache *c) | |
659 | { | |
660 | return c->name; | |
661 | } | |
662 | EXPORT_SYMBOL(kmem_cache_name); | |
663 | ||
2e892f43 CL |
664 | int kmem_cache_shrink(struct kmem_cache *d) |
665 | { | |
666 | return 0; | |
667 | } | |
668 | EXPORT_SYMBOL(kmem_cache_shrink); | |
669 | ||
55935a34 | 670 | int kmem_ptr_validate(struct kmem_cache *a, const void *b) |
2e892f43 CL |
671 | { |
672 | return 0; | |
673 | } | |
674 | ||
84a01c2f PM |
675 | static unsigned int slob_ready __read_mostly; |
676 | ||
677 | int slab_is_available(void) | |
678 | { | |
679 | return slob_ready; | |
680 | } | |
681 | ||
bcb4ddb4 DG |
682 | void __init kmem_cache_init(void) |
683 | { | |
84a01c2f | 684 | slob_ready = 1; |
10cef602 | 685 | } |