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 |
999d8795 EG |
31 | * does not have to be separately tracked. |
32 | * These objects are detected in kfree() because PageSlab() | |
d87a133f | 33 | * is false for them. |
10cef602 MM |
34 | * |
35 | * SLAB is emulated on top of SLOB by simply calling constructors and | |
95b35127 NP |
36 | * destructors for every SLAB allocation. Objects are returned with the |
37 | * 4-byte alignment unless the SLAB_HWCACHE_ALIGN flag is set, in which | |
38 | * case the low-level allocator will fragment blocks to create the proper | |
39 | * alignment. Again, objects of page-size or greater are allocated by | |
6193a2ff | 40 | * calling alloc_pages(). As SLAB objects know their size, no separate |
95b35127 | 41 | * size bookkeeping is necessary and there is essentially no allocation |
d87a133f NP |
42 | * space overhead, and compound pages aren't needed for multi-page |
43 | * allocations. | |
6193a2ff PM |
44 | * |
45 | * NUMA support in SLOB is fairly simplistic, pushing most of the real | |
46 | * logic down to the page allocator, and simply doing the node accounting | |
47 | * on the upper levels. In the event that a node id is explicitly | |
6484eb3e | 48 | * provided, alloc_pages_exact_node() with the specified node id is used |
6193a2ff PM |
49 | * instead. The common case (or when the node id isn't explicitly provided) |
50 | * will default to the current node, as per numa_node_id(). | |
51 | * | |
52 | * Node aware pages are still inserted in to the global freelist, and | |
53 | * these are scanned for by matching against the node id encoded in the | |
54 | * page flags. As a result, block allocations that can be satisfied from | |
55 | * the freelist will only be done so on pages residing on the same node, | |
56 | * in order to prevent random node placement. | |
10cef602 MM |
57 | */ |
58 | ||
95b35127 | 59 | #include <linux/kernel.h> |
10cef602 | 60 | #include <linux/slab.h> |
97d06609 CL |
61 | #include "slab.h" |
62 | ||
10cef602 | 63 | #include <linux/mm.h> |
1f0532eb | 64 | #include <linux/swap.h> /* struct reclaim_state */ |
10cef602 MM |
65 | #include <linux/cache.h> |
66 | #include <linux/init.h> | |
b95f1b31 | 67 | #include <linux/export.h> |
afc0cedb | 68 | #include <linux/rcupdate.h> |
95b35127 | 69 | #include <linux/list.h> |
4374e616 | 70 | #include <linux/kmemleak.h> |
039ca4e7 LZ |
71 | |
72 | #include <trace/events/kmem.h> | |
73 | ||
60063497 | 74 | #include <linux/atomic.h> |
95b35127 | 75 | |
95b35127 NP |
76 | /* |
77 | * slob_block has a field 'units', which indicates size of block if +ve, | |
78 | * or offset of next block if -ve (in SLOB_UNITs). | |
79 | * | |
80 | * Free blocks of size 1 unit simply contain the offset of the next block. | |
81 | * Those with larger size contain their size in the first SLOB_UNIT of | |
82 | * memory, and the offset of the next free block in the second SLOB_UNIT. | |
83 | */ | |
55394849 | 84 | #if PAGE_SIZE <= (32767 * 2) |
95b35127 NP |
85 | typedef s16 slobidx_t; |
86 | #else | |
87 | typedef s32 slobidx_t; | |
88 | #endif | |
89 | ||
10cef602 | 90 | struct slob_block { |
95b35127 | 91 | slobidx_t units; |
55394849 | 92 | }; |
10cef602 MM |
93 | typedef struct slob_block slob_t; |
94 | ||
95b35127 | 95 | /* |
20cecbae | 96 | * All partially free slob pages go on these lists. |
95b35127 | 97 | */ |
20cecbae MM |
98 | #define SLOB_BREAK1 256 |
99 | #define SLOB_BREAK2 1024 | |
100 | static LIST_HEAD(free_slob_small); | |
101 | static LIST_HEAD(free_slob_medium); | |
102 | static LIST_HEAD(free_slob_large); | |
95b35127 | 103 | |
95b35127 NP |
104 | /* |
105 | * slob_page_free: true for pages on free_slob_pages list. | |
106 | */ | |
b8c24c4a | 107 | static inline int slob_page_free(struct page *sp) |
95b35127 | 108 | { |
b8c24c4a | 109 | return PageSlobFree(sp); |
95b35127 NP |
110 | } |
111 | ||
b8c24c4a | 112 | static void set_slob_page_free(struct page *sp, struct list_head *list) |
95b35127 | 113 | { |
20cecbae | 114 | list_add(&sp->list, list); |
b8c24c4a | 115 | __SetPageSlobFree(sp); |
95b35127 NP |
116 | } |
117 | ||
b8c24c4a | 118 | static inline void clear_slob_page_free(struct page *sp) |
95b35127 NP |
119 | { |
120 | list_del(&sp->list); | |
b8c24c4a | 121 | __ClearPageSlobFree(sp); |
95b35127 NP |
122 | } |
123 | ||
10cef602 MM |
124 | #define SLOB_UNIT sizeof(slob_t) |
125 | #define SLOB_UNITS(size) (((size) + SLOB_UNIT - 1)/SLOB_UNIT) | |
126 | #define SLOB_ALIGN L1_CACHE_BYTES | |
127 | ||
afc0cedb NP |
128 | /* |
129 | * struct slob_rcu is inserted at the tail of allocated slob blocks, which | |
130 | * were created with a SLAB_DESTROY_BY_RCU slab. slob_rcu is used to free | |
131 | * the block using call_rcu. | |
132 | */ | |
133 | struct slob_rcu { | |
134 | struct rcu_head head; | |
135 | int size; | |
136 | }; | |
137 | ||
95b35127 NP |
138 | /* |
139 | * slob_lock protects all slob allocator structures. | |
140 | */ | |
10cef602 | 141 | static DEFINE_SPINLOCK(slob_lock); |
10cef602 | 142 | |
95b35127 NP |
143 | /* |
144 | * Encode the given size and next info into a free slob block s. | |
145 | */ | |
146 | static void set_slob(slob_t *s, slobidx_t size, slob_t *next) | |
147 | { | |
148 | slob_t *base = (slob_t *)((unsigned long)s & PAGE_MASK); | |
149 | slobidx_t offset = next - base; | |
bcb4ddb4 | 150 | |
95b35127 NP |
151 | if (size > 1) { |
152 | s[0].units = size; | |
153 | s[1].units = offset; | |
154 | } else | |
155 | s[0].units = -offset; | |
156 | } | |
10cef602 | 157 | |
95b35127 NP |
158 | /* |
159 | * Return the size of a slob block. | |
160 | */ | |
161 | static slobidx_t slob_units(slob_t *s) | |
162 | { | |
163 | if (s->units > 0) | |
164 | return s->units; | |
165 | return 1; | |
166 | } | |
167 | ||
168 | /* | |
169 | * Return the next free slob block pointer after this one. | |
170 | */ | |
171 | static slob_t *slob_next(slob_t *s) | |
172 | { | |
173 | slob_t *base = (slob_t *)((unsigned long)s & PAGE_MASK); | |
174 | slobidx_t next; | |
175 | ||
176 | if (s[0].units < 0) | |
177 | next = -s[0].units; | |
178 | else | |
179 | next = s[1].units; | |
180 | return base+next; | |
181 | } | |
182 | ||
183 | /* | |
184 | * Returns true if s is the last free block in its page. | |
185 | */ | |
186 | static int slob_last(slob_t *s) | |
187 | { | |
188 | return !((unsigned long)slob_next(s) & ~PAGE_MASK); | |
189 | } | |
190 | ||
6e9ed0cc | 191 | static void *slob_new_pages(gfp_t gfp, int order, int node) |
6193a2ff PM |
192 | { |
193 | void *page; | |
194 | ||
195 | #ifdef CONFIG_NUMA | |
90f2cbbc | 196 | if (node != NUMA_NO_NODE) |
6484eb3e | 197 | page = alloc_pages_exact_node(node, gfp, order); |
6193a2ff PM |
198 | else |
199 | #endif | |
200 | page = alloc_pages(gfp, order); | |
201 | ||
202 | if (!page) | |
203 | return NULL; | |
204 | ||
205 | return page_address(page); | |
206 | } | |
207 | ||
6e9ed0cc AW |
208 | static void slob_free_pages(void *b, int order) |
209 | { | |
1f0532eb NP |
210 | if (current->reclaim_state) |
211 | current->reclaim_state->reclaimed_slab += 1 << order; | |
6e9ed0cc AW |
212 | free_pages((unsigned long)b, order); |
213 | } | |
214 | ||
95b35127 NP |
215 | /* |
216 | * Allocate a slob block within a given slob_page sp. | |
217 | */ | |
b8c24c4a | 218 | static void *slob_page_alloc(struct page *sp, size_t size, int align) |
10cef602 | 219 | { |
6e9ed0cc | 220 | slob_t *prev, *cur, *aligned = NULL; |
10cef602 | 221 | int delta = 0, units = SLOB_UNITS(size); |
10cef602 | 222 | |
b8c24c4a | 223 | for (prev = NULL, cur = sp->freelist; ; prev = cur, cur = slob_next(cur)) { |
95b35127 NP |
224 | slobidx_t avail = slob_units(cur); |
225 | ||
10cef602 MM |
226 | if (align) { |
227 | aligned = (slob_t *)ALIGN((unsigned long)cur, align); | |
228 | delta = aligned - cur; | |
229 | } | |
95b35127 NP |
230 | if (avail >= units + delta) { /* room enough? */ |
231 | slob_t *next; | |
232 | ||
10cef602 | 233 | if (delta) { /* need to fragment head to align? */ |
95b35127 NP |
234 | next = slob_next(cur); |
235 | set_slob(aligned, avail - delta, next); | |
236 | set_slob(cur, delta, aligned); | |
10cef602 MM |
237 | prev = cur; |
238 | cur = aligned; | |
95b35127 | 239 | avail = slob_units(cur); |
10cef602 MM |
240 | } |
241 | ||
95b35127 NP |
242 | next = slob_next(cur); |
243 | if (avail == units) { /* exact fit? unlink. */ | |
244 | if (prev) | |
245 | set_slob(prev, slob_units(prev), next); | |
246 | else | |
b8c24c4a | 247 | sp->freelist = next; |
95b35127 NP |
248 | } else { /* fragment */ |
249 | if (prev) | |
250 | set_slob(prev, slob_units(prev), cur + units); | |
251 | else | |
b8c24c4a | 252 | sp->freelist = cur + units; |
95b35127 | 253 | set_slob(cur + units, avail - units, next); |
10cef602 MM |
254 | } |
255 | ||
95b35127 NP |
256 | sp->units -= units; |
257 | if (!sp->units) | |
258 | clear_slob_page_free(sp); | |
10cef602 MM |
259 | return cur; |
260 | } | |
95b35127 NP |
261 | if (slob_last(cur)) |
262 | return NULL; | |
263 | } | |
264 | } | |
10cef602 | 265 | |
95b35127 NP |
266 | /* |
267 | * slob_alloc: entry point into the slob allocator. | |
268 | */ | |
6193a2ff | 269 | static void *slob_alloc(size_t size, gfp_t gfp, int align, int node) |
95b35127 | 270 | { |
b8c24c4a | 271 | struct page *sp; |
d6269543 | 272 | struct list_head *prev; |
20cecbae | 273 | struct list_head *slob_list; |
95b35127 NP |
274 | slob_t *b = NULL; |
275 | unsigned long flags; | |
10cef602 | 276 | |
20cecbae MM |
277 | if (size < SLOB_BREAK1) |
278 | slob_list = &free_slob_small; | |
279 | else if (size < SLOB_BREAK2) | |
280 | slob_list = &free_slob_medium; | |
281 | else | |
282 | slob_list = &free_slob_large; | |
283 | ||
95b35127 NP |
284 | spin_lock_irqsave(&slob_lock, flags); |
285 | /* Iterate through each partially free page, try to find room */ | |
20cecbae | 286 | list_for_each_entry(sp, slob_list, list) { |
6193a2ff PM |
287 | #ifdef CONFIG_NUMA |
288 | /* | |
289 | * If there's a node specification, search for a partial | |
290 | * page with a matching node id in the freelist. | |
291 | */ | |
90f2cbbc | 292 | if (node != NUMA_NO_NODE && page_to_nid(sp) != node) |
6193a2ff PM |
293 | continue; |
294 | #endif | |
d6269543 MM |
295 | /* Enough room on this page? */ |
296 | if (sp->units < SLOB_UNITS(size)) | |
297 | continue; | |
6193a2ff | 298 | |
d6269543 MM |
299 | /* Attempt to alloc */ |
300 | prev = sp->list.prev; | |
301 | b = slob_page_alloc(sp, size, align); | |
302 | if (!b) | |
303 | continue; | |
304 | ||
305 | /* Improve fragment distribution and reduce our average | |
306 | * search time by starting our next search here. (see | |
307 | * Knuth vol 1, sec 2.5, pg 449) */ | |
20cecbae MM |
308 | if (prev != slob_list->prev && |
309 | slob_list->next != prev->next) | |
310 | list_move_tail(slob_list, prev->next); | |
d6269543 | 311 | break; |
10cef602 | 312 | } |
95b35127 NP |
313 | spin_unlock_irqrestore(&slob_lock, flags); |
314 | ||
315 | /* Not enough space: must allocate a new page */ | |
316 | if (!b) { | |
6e9ed0cc | 317 | b = slob_new_pages(gfp & ~__GFP_ZERO, 0, node); |
95b35127 | 318 | if (!b) |
6e9ed0cc | 319 | return NULL; |
b5568280 CL |
320 | sp = virt_to_page(b); |
321 | __SetPageSlab(sp); | |
95b35127 NP |
322 | |
323 | spin_lock_irqsave(&slob_lock, flags); | |
324 | sp->units = SLOB_UNITS(PAGE_SIZE); | |
b8c24c4a | 325 | sp->freelist = b; |
95b35127 NP |
326 | INIT_LIST_HEAD(&sp->list); |
327 | set_slob(b, SLOB_UNITS(PAGE_SIZE), b + SLOB_UNITS(PAGE_SIZE)); | |
20cecbae | 328 | set_slob_page_free(sp, slob_list); |
95b35127 NP |
329 | b = slob_page_alloc(sp, size, align); |
330 | BUG_ON(!b); | |
331 | spin_unlock_irqrestore(&slob_lock, flags); | |
332 | } | |
d07dbea4 CL |
333 | if (unlikely((gfp & __GFP_ZERO) && b)) |
334 | memset(b, 0, size); | |
95b35127 | 335 | return b; |
10cef602 MM |
336 | } |
337 | ||
95b35127 NP |
338 | /* |
339 | * slob_free: entry point into the slob allocator. | |
340 | */ | |
10cef602 MM |
341 | static void slob_free(void *block, int size) |
342 | { | |
b8c24c4a | 343 | struct page *sp; |
95b35127 NP |
344 | slob_t *prev, *next, *b = (slob_t *)block; |
345 | slobidx_t units; | |
10cef602 | 346 | unsigned long flags; |
d602daba | 347 | struct list_head *slob_list; |
10cef602 | 348 | |
2408c550 | 349 | if (unlikely(ZERO_OR_NULL_PTR(block))) |
10cef602 | 350 | return; |
95b35127 | 351 | BUG_ON(!size); |
10cef602 | 352 | |
b5568280 | 353 | sp = virt_to_page(block); |
95b35127 | 354 | units = SLOB_UNITS(size); |
10cef602 | 355 | |
10cef602 | 356 | spin_lock_irqsave(&slob_lock, flags); |
10cef602 | 357 | |
95b35127 NP |
358 | if (sp->units + units == SLOB_UNITS(PAGE_SIZE)) { |
359 | /* Go directly to page allocator. Do not pass slob allocator */ | |
360 | if (slob_page_free(sp)) | |
361 | clear_slob_page_free(sp); | |
6fb8f424 | 362 | spin_unlock_irqrestore(&slob_lock, flags); |
b5568280 CL |
363 | __ClearPageSlab(sp); |
364 | reset_page_mapcount(sp); | |
1f0532eb | 365 | slob_free_pages(b, 0); |
6fb8f424 | 366 | return; |
95b35127 | 367 | } |
10cef602 | 368 | |
95b35127 NP |
369 | if (!slob_page_free(sp)) { |
370 | /* This slob page is about to become partially free. Easy! */ | |
371 | sp->units = units; | |
b8c24c4a | 372 | sp->freelist = b; |
95b35127 NP |
373 | set_slob(b, units, |
374 | (void *)((unsigned long)(b + | |
375 | SLOB_UNITS(PAGE_SIZE)) & PAGE_MASK)); | |
d602daba BL |
376 | if (size < SLOB_BREAK1) |
377 | slob_list = &free_slob_small; | |
378 | else if (size < SLOB_BREAK2) | |
379 | slob_list = &free_slob_medium; | |
380 | else | |
381 | slob_list = &free_slob_large; | |
382 | set_slob_page_free(sp, slob_list); | |
95b35127 NP |
383 | goto out; |
384 | } | |
385 | ||
386 | /* | |
387 | * Otherwise the page is already partially free, so find reinsertion | |
388 | * point. | |
389 | */ | |
390 | sp->units += units; | |
10cef602 | 391 | |
b8c24c4a CL |
392 | if (b < (slob_t *)sp->freelist) { |
393 | if (b + units == sp->freelist) { | |
394 | units += slob_units(sp->freelist); | |
395 | sp->freelist = slob_next(sp->freelist); | |
679299b3 | 396 | } |
b8c24c4a CL |
397 | set_slob(b, units, sp->freelist); |
398 | sp->freelist = b; | |
95b35127 | 399 | } else { |
b8c24c4a | 400 | prev = sp->freelist; |
95b35127 NP |
401 | next = slob_next(prev); |
402 | while (b > next) { | |
403 | prev = next; | |
404 | next = slob_next(prev); | |
405 | } | |
10cef602 | 406 | |
95b35127 NP |
407 | if (!slob_last(prev) && b + units == next) { |
408 | units += slob_units(next); | |
409 | set_slob(b, units, slob_next(next)); | |
410 | } else | |
411 | set_slob(b, units, next); | |
412 | ||
413 | if (prev + slob_units(prev) == b) { | |
414 | units = slob_units(b) + slob_units(prev); | |
415 | set_slob(prev, units, slob_next(b)); | |
416 | } else | |
417 | set_slob(prev, slob_units(prev), b); | |
418 | } | |
419 | out: | |
10cef602 MM |
420 | spin_unlock_irqrestore(&slob_lock, flags); |
421 | } | |
422 | ||
95b35127 NP |
423 | /* |
424 | * End of slob allocator proper. Begin kmem_cache_alloc and kmalloc frontend. | |
425 | */ | |
426 | ||
f3f74101 EG |
427 | static __always_inline void * |
428 | __do_kmalloc_node(size_t size, gfp_t gfp, int node, unsigned long caller) | |
10cef602 | 429 | { |
6cb8f913 | 430 | unsigned int *m; |
55394849 | 431 | int align = max(ARCH_KMALLOC_MINALIGN, ARCH_SLAB_MINALIGN); |
3eae2cb2 | 432 | void *ret; |
55394849 | 433 | |
bd50cfa8 SR |
434 | gfp &= gfp_allowed_mask; |
435 | ||
19cefdff | 436 | lockdep_trace_alloc(gfp); |
cf40bd16 | 437 | |
55394849 | 438 | if (size < PAGE_SIZE - align) { |
6cb8f913 CL |
439 | if (!size) |
440 | return ZERO_SIZE_PTR; | |
441 | ||
6193a2ff | 442 | m = slob_alloc(size + align, gfp, align, node); |
3eae2cb2 | 443 | |
239f49c0 MK |
444 | if (!m) |
445 | return NULL; | |
446 | *m = size; | |
3eae2cb2 EGM |
447 | ret = (void *)m + align; |
448 | ||
f3f74101 | 449 | trace_kmalloc_node(caller, ret, |
ca2b84cb | 450 | size, size + align, gfp, node); |
d87a133f | 451 | } else { |
3eae2cb2 | 452 | unsigned int order = get_order(size); |
d87a133f | 453 | |
8df275af DR |
454 | if (likely(order)) |
455 | gfp |= __GFP_COMP; | |
456 | ret = slob_new_pages(gfp, order, node); | |
3eae2cb2 | 457 | |
f3f74101 | 458 | trace_kmalloc_node(caller, ret, |
ca2b84cb | 459 | size, PAGE_SIZE << order, gfp, node); |
10cef602 | 460 | } |
3eae2cb2 | 461 | |
4374e616 | 462 | kmemleak_alloc(ret, size, 1, gfp); |
3eae2cb2 | 463 | return ret; |
10cef602 | 464 | } |
f3f74101 EG |
465 | |
466 | void *__kmalloc_node(size_t size, gfp_t gfp, int node) | |
467 | { | |
468 | return __do_kmalloc_node(size, gfp, node, _RET_IP_); | |
469 | } | |
6193a2ff | 470 | EXPORT_SYMBOL(__kmalloc_node); |
10cef602 | 471 | |
f3f74101 EG |
472 | #ifdef CONFIG_TRACING |
473 | void *__kmalloc_track_caller(size_t size, gfp_t gfp, unsigned long caller) | |
474 | { | |
475 | return __do_kmalloc_node(size, gfp, NUMA_NO_NODE, caller); | |
476 | } | |
477 | ||
478 | #ifdef CONFIG_NUMA | |
82bd5508 | 479 | void *__kmalloc_node_track_caller(size_t size, gfp_t gfp, |
f3f74101 EG |
480 | int node, unsigned long caller) |
481 | { | |
482 | return __do_kmalloc_node(size, gfp, node, caller); | |
483 | } | |
484 | #endif | |
485 | #endif | |
486 | ||
10cef602 MM |
487 | void kfree(const void *block) |
488 | { | |
b8c24c4a | 489 | struct page *sp; |
10cef602 | 490 | |
2121db74 PE |
491 | trace_kfree(_RET_IP_, block); |
492 | ||
2408c550 | 493 | if (unlikely(ZERO_OR_NULL_PTR(block))) |
10cef602 | 494 | return; |
4374e616 | 495 | kmemleak_free(block); |
10cef602 | 496 | |
b5568280 CL |
497 | sp = virt_to_page(block); |
498 | if (PageSlab(sp)) { | |
55394849 NP |
499 | int align = max(ARCH_KMALLOC_MINALIGN, ARCH_SLAB_MINALIGN); |
500 | unsigned int *m = (unsigned int *)(block - align); | |
501 | slob_free(m, *m + align); | |
d87a133f | 502 | } else |
8cf9864b | 503 | __free_pages(sp, compound_order(sp)); |
10cef602 | 504 | } |
10cef602 MM |
505 | EXPORT_SYMBOL(kfree); |
506 | ||
d87a133f | 507 | /* can't use ksize for kmem_cache_alloc memory, only kmalloc */ |
fd76bab2 | 508 | size_t ksize(const void *block) |
10cef602 | 509 | { |
b8c24c4a | 510 | struct page *sp; |
999d8795 EG |
511 | int align; |
512 | unsigned int *m; | |
10cef602 | 513 | |
ef8b4520 CL |
514 | BUG_ON(!block); |
515 | if (unlikely(block == ZERO_SIZE_PTR)) | |
10cef602 MM |
516 | return 0; |
517 | ||
b5568280 | 518 | sp = virt_to_page(block); |
999d8795 EG |
519 | if (unlikely(!PageSlab(sp))) |
520 | return PAGE_SIZE << compound_order(sp); | |
521 | ||
522 | align = max(ARCH_KMALLOC_MINALIGN, ARCH_SLAB_MINALIGN); | |
523 | m = (unsigned int *)(block - align); | |
524 | return SLOB_UNITS(*m) * SLOB_UNIT; | |
10cef602 | 525 | } |
b1aabecd | 526 | EXPORT_SYMBOL(ksize); |
10cef602 | 527 | |
8a13a4cc | 528 | int __kmem_cache_create(struct kmem_cache *c, unsigned long flags) |
10cef602 | 529 | { |
8a13a4cc | 530 | size_t align = c->size; |
10cef602 | 531 | |
278b1bb1 CL |
532 | if (flags & SLAB_DESTROY_BY_RCU) { |
533 | /* leave room for rcu footer at the end of object */ | |
534 | c->size += sizeof(struct slob_rcu); | |
039363f3 | 535 | } |
278b1bb1 | 536 | c->flags = flags; |
278b1bb1 CL |
537 | /* ignore alignment unless it's forced */ |
538 | c->align = (flags & SLAB_HWCACHE_ALIGN) ? SLOB_ALIGN : 0; | |
539 | if (c->align < ARCH_SLAB_MINALIGN) | |
540 | c->align = ARCH_SLAB_MINALIGN; | |
541 | if (c->align < align) | |
542 | c->align = align; | |
10cef602 | 543 | |
278b1bb1 | 544 | return 0; |
10cef602 | 545 | } |
10cef602 | 546 | |
6193a2ff | 547 | void *kmem_cache_alloc_node(struct kmem_cache *c, gfp_t flags, int node) |
10cef602 MM |
548 | { |
549 | void *b; | |
550 | ||
bd50cfa8 SR |
551 | flags &= gfp_allowed_mask; |
552 | ||
553 | lockdep_trace_alloc(flags); | |
554 | ||
3eae2cb2 | 555 | if (c->size < PAGE_SIZE) { |
6193a2ff | 556 | b = slob_alloc(c->size, flags, c->align, node); |
fe74fe2b | 557 | trace_kmem_cache_alloc_node(_RET_IP_, b, c->object_size, |
ca2b84cb EGM |
558 | SLOB_UNITS(c->size) * SLOB_UNIT, |
559 | flags, node); | |
3eae2cb2 | 560 | } else { |
6e9ed0cc | 561 | b = slob_new_pages(flags, get_order(c->size), node); |
fe74fe2b | 562 | trace_kmem_cache_alloc_node(_RET_IP_, b, c->object_size, |
ca2b84cb EGM |
563 | PAGE_SIZE << get_order(c->size), |
564 | flags, node); | |
3eae2cb2 | 565 | } |
10cef602 MM |
566 | |
567 | if (c->ctor) | |
51cc5068 | 568 | c->ctor(b); |
10cef602 | 569 | |
4374e616 | 570 | kmemleak_alloc_recursive(b, c->size, 1, c->flags, flags); |
10cef602 MM |
571 | return b; |
572 | } | |
6193a2ff | 573 | EXPORT_SYMBOL(kmem_cache_alloc_node); |
10cef602 | 574 | |
afc0cedb | 575 | static void __kmem_cache_free(void *b, int size) |
10cef602 | 576 | { |
afc0cedb NP |
577 | if (size < PAGE_SIZE) |
578 | slob_free(b, size); | |
10cef602 | 579 | else |
6e9ed0cc | 580 | slob_free_pages(b, get_order(size)); |
afc0cedb NP |
581 | } |
582 | ||
583 | static void kmem_rcu_free(struct rcu_head *head) | |
584 | { | |
585 | struct slob_rcu *slob_rcu = (struct slob_rcu *)head; | |
586 | void *b = (void *)slob_rcu - (slob_rcu->size - sizeof(struct slob_rcu)); | |
587 | ||
588 | __kmem_cache_free(b, slob_rcu->size); | |
589 | } | |
590 | ||
591 | void kmem_cache_free(struct kmem_cache *c, void *b) | |
592 | { | |
4374e616 | 593 | kmemleak_free_recursive(b, c->flags); |
afc0cedb NP |
594 | if (unlikely(c->flags & SLAB_DESTROY_BY_RCU)) { |
595 | struct slob_rcu *slob_rcu; | |
596 | slob_rcu = b + (c->size - sizeof(struct slob_rcu)); | |
afc0cedb NP |
597 | slob_rcu->size = c->size; |
598 | call_rcu(&slob_rcu->head, kmem_rcu_free); | |
599 | } else { | |
afc0cedb NP |
600 | __kmem_cache_free(b, c->size); |
601 | } | |
3eae2cb2 | 602 | |
ca2b84cb | 603 | trace_kmem_cache_free(_RET_IP_, b); |
10cef602 MM |
604 | } |
605 | EXPORT_SYMBOL(kmem_cache_free); | |
606 | ||
945cf2b6 CL |
607 | int __kmem_cache_shutdown(struct kmem_cache *c) |
608 | { | |
609 | /* No way to check for remaining objects */ | |
610 | return 0; | |
611 | } | |
612 | ||
2e892f43 CL |
613 | int kmem_cache_shrink(struct kmem_cache *d) |
614 | { | |
615 | return 0; | |
616 | } | |
617 | EXPORT_SYMBOL(kmem_cache_shrink); | |
618 | ||
9b030cb8 CL |
619 | struct kmem_cache kmem_cache_boot = { |
620 | .name = "kmem_cache", | |
621 | .size = sizeof(struct kmem_cache), | |
622 | .flags = SLAB_PANIC, | |
623 | .align = ARCH_KMALLOC_MINALIGN, | |
624 | }; | |
625 | ||
bcb4ddb4 DG |
626 | void __init kmem_cache_init(void) |
627 | { | |
9b030cb8 | 628 | kmem_cache = &kmem_cache_boot; |
97d06609 | 629 | slab_state = UP; |
10cef602 | 630 | } |
bbff2e43 WF |
631 | |
632 | void __init kmem_cache_init_late(void) | |
633 | { | |
97d06609 | 634 | slab_state = FULL; |
bbff2e43 | 635 | } |