]>
Commit | Line | Data |
---|---|---|
9cc06bf8 DM |
1 | /* |
2 | * zcache.c | |
3 | * | |
4 | * Copyright (c) 2010,2011, Dan Magenheimer, Oracle Corp. | |
5 | * Copyright (c) 2010,2011, Nitin Gupta | |
6 | * | |
7 | * Zcache provides an in-kernel "host implementation" for transcendent memory | |
8 | * and, thus indirectly, for cleancache and frontswap. Zcache includes two | |
9 | * page-accessible memory [1] interfaces, both utilizing lzo1x compression: | |
10 | * 1) "compression buddies" ("zbud") is used for ephemeral pages | |
11 | * 2) xvmalloc is used for persistent pages. | |
12 | * Xvmalloc (based on the TLSF allocator) has very low fragmentation | |
13 | * so maximizes space efficiency, while zbud allows pairs (and potentially, | |
14 | * in the future, more than a pair of) compressed pages to be closely linked | |
15 | * so that reclaiming can be done via the kernel's physical-page-oriented | |
16 | * "shrinker" interface. | |
17 | * | |
18 | * [1] For a definition of page-accessible memory (aka PAM), see: | |
19 | * http://marc.info/?l=linux-mm&m=127811271605009 | |
20 | */ | |
21 | ||
12623f07 | 22 | #include <linux/module.h> |
9cc06bf8 DM |
23 | #include <linux/cpu.h> |
24 | #include <linux/highmem.h> | |
25 | #include <linux/list.h> | |
26 | #include <linux/lzo.h> | |
27 | #include <linux/slab.h> | |
28 | #include <linux/spinlock.h> | |
29 | #include <linux/types.h> | |
30 | #include <linux/atomic.h> | |
3ca15c44 | 31 | #include <linux/math64.h> |
9cc06bf8 DM |
32 | #include "tmem.h" |
33 | ||
34 | #include "../zram/xvmalloc.h" /* if built in drivers/staging */ | |
35 | ||
36 | #if (!defined(CONFIG_CLEANCACHE) && !defined(CONFIG_FRONTSWAP)) | |
37 | #error "zcache is useless without CONFIG_CLEANCACHE or CONFIG_FRONTSWAP" | |
38 | #endif | |
39 | #ifdef CONFIG_CLEANCACHE | |
40 | #include <linux/cleancache.h> | |
41 | #endif | |
42 | #ifdef CONFIG_FRONTSWAP | |
43 | #include <linux/frontswap.h> | |
44 | #endif | |
45 | ||
46 | #if 0 | |
47 | /* this is more aggressive but may cause other problems? */ | |
48 | #define ZCACHE_GFP_MASK (GFP_ATOMIC | __GFP_NORETRY | __GFP_NOWARN) | |
49 | #else | |
50 | #define ZCACHE_GFP_MASK \ | |
51 | (__GFP_FS | __GFP_NORETRY | __GFP_NOWARN | __GFP_NOMEMALLOC) | |
52 | #endif | |
53 | ||
966b9016 DM |
54 | #define MAX_POOLS_PER_CLIENT 16 |
55 | ||
56 | #define MAX_CLIENTS 16 | |
57 | #define LOCAL_CLIENT ((uint16_t)-1) | |
fd6b68bb TLSC |
58 | |
59 | MODULE_LICENSE("GPL"); | |
60 | ||
966b9016 DM |
61 | struct zcache_client { |
62 | struct tmem_pool *tmem_pools[MAX_POOLS_PER_CLIENT]; | |
63 | struct xv_pool *xvpool; | |
64 | bool allocated; | |
65 | atomic_t refcount; | |
66 | }; | |
67 | ||
68 | static struct zcache_client zcache_host; | |
69 | static struct zcache_client zcache_clients[MAX_CLIENTS]; | |
70 | ||
71 | static inline uint16_t get_client_id_from_client(struct zcache_client *cli) | |
72 | { | |
73 | BUG_ON(cli == NULL); | |
74 | if (cli == &zcache_host) | |
75 | return LOCAL_CLIENT; | |
76 | return cli - &zcache_clients[0]; | |
77 | } | |
78 | ||
79 | static inline bool is_local_client(struct zcache_client *cli) | |
80 | { | |
81 | return cli == &zcache_host; | |
82 | } | |
83 | ||
9cc06bf8 DM |
84 | /********** |
85 | * Compression buddies ("zbud") provides for packing two (or, possibly | |
86 | * in the future, more) compressed ephemeral pages into a single "raw" | |
87 | * (physical) page and tracking them with data structures so that | |
88 | * the raw pages can be easily reclaimed. | |
89 | * | |
90 | * A zbud page ("zbpg") is an aligned page containing a list_head, | |
91 | * a lock, and two "zbud headers". The remainder of the physical | |
92 | * page is divided up into aligned 64-byte "chunks" which contain | |
93 | * the compressed data for zero, one, or two zbuds. Each zbpg | |
94 | * resides on: (1) an "unused list" if it has no zbuds; (2) a | |
95 | * "buddied" list if it is fully populated with two zbuds; or | |
96 | * (3) one of PAGE_SIZE/64 "unbuddied" lists indexed by how many chunks | |
97 | * the one unbuddied zbud uses. The data inside a zbpg cannot be | |
98 | * read or written unless the zbpg's lock is held. | |
99 | */ | |
100 | ||
101 | #define ZBH_SENTINEL 0x43214321 | |
102 | #define ZBPG_SENTINEL 0xdeadbeef | |
103 | ||
104 | #define ZBUD_MAX_BUDS 2 | |
105 | ||
106 | struct zbud_hdr { | |
966b9016 DM |
107 | uint16_t client_id; |
108 | uint16_t pool_id; | |
9cc06bf8 DM |
109 | struct tmem_oid oid; |
110 | uint32_t index; | |
111 | uint16_t size; /* compressed size in bytes, zero means unused */ | |
112 | DECL_SENTINEL | |
113 | }; | |
114 | ||
115 | struct zbud_page { | |
116 | struct list_head bud_list; | |
117 | spinlock_t lock; | |
118 | struct zbud_hdr buddy[ZBUD_MAX_BUDS]; | |
119 | DECL_SENTINEL | |
120 | /* followed by NUM_CHUNK aligned CHUNK_SIZE-byte chunks */ | |
121 | }; | |
122 | ||
123 | #define CHUNK_SHIFT 6 | |
124 | #define CHUNK_SIZE (1 << CHUNK_SHIFT) | |
125 | #define CHUNK_MASK (~(CHUNK_SIZE-1)) | |
126 | #define NCHUNKS (((PAGE_SIZE - sizeof(struct zbud_page)) & \ | |
127 | CHUNK_MASK) >> CHUNK_SHIFT) | |
128 | #define MAX_CHUNK (NCHUNKS-1) | |
129 | ||
130 | static struct { | |
131 | struct list_head list; | |
132 | unsigned count; | |
133 | } zbud_unbuddied[NCHUNKS]; | |
134 | /* list N contains pages with N chunks USED and NCHUNKS-N unused */ | |
135 | /* element 0 is never used but optimizing that isn't worth it */ | |
136 | static unsigned long zbud_cumul_chunk_counts[NCHUNKS]; | |
137 | ||
138 | struct list_head zbud_buddied_list; | |
139 | static unsigned long zcache_zbud_buddied_count; | |
140 | ||
141 | /* protects the buddied list and all unbuddied lists */ | |
142 | static DEFINE_SPINLOCK(zbud_budlists_spinlock); | |
143 | ||
144 | static LIST_HEAD(zbpg_unused_list); | |
145 | static unsigned long zcache_zbpg_unused_list_count; | |
146 | ||
147 | /* protects the unused page list */ | |
148 | static DEFINE_SPINLOCK(zbpg_unused_list_spinlock); | |
149 | ||
150 | static atomic_t zcache_zbud_curr_raw_pages; | |
151 | static atomic_t zcache_zbud_curr_zpages; | |
152 | static unsigned long zcache_zbud_curr_zbytes; | |
153 | static unsigned long zcache_zbud_cumul_zpages; | |
154 | static unsigned long zcache_zbud_cumul_zbytes; | |
155 | static unsigned long zcache_compress_poor; | |
966b9016 | 156 | static unsigned long zcache_mean_compress_poor; |
9cc06bf8 DM |
157 | |
158 | /* forward references */ | |
159 | static void *zcache_get_free_page(void); | |
160 | static void zcache_free_page(void *p); | |
161 | ||
162 | /* | |
163 | * zbud helper functions | |
164 | */ | |
165 | ||
166 | static inline unsigned zbud_max_buddy_size(void) | |
167 | { | |
168 | return MAX_CHUNK << CHUNK_SHIFT; | |
169 | } | |
170 | ||
171 | static inline unsigned zbud_size_to_chunks(unsigned size) | |
172 | { | |
173 | BUG_ON(size == 0 || size > zbud_max_buddy_size()); | |
174 | return (size + CHUNK_SIZE - 1) >> CHUNK_SHIFT; | |
175 | } | |
176 | ||
177 | static inline int zbud_budnum(struct zbud_hdr *zh) | |
178 | { | |
179 | unsigned offset = (unsigned long)zh & (PAGE_SIZE - 1); | |
180 | struct zbud_page *zbpg = NULL; | |
181 | unsigned budnum = -1U; | |
182 | int i; | |
183 | ||
184 | for (i = 0; i < ZBUD_MAX_BUDS; i++) | |
185 | if (offset == offsetof(typeof(*zbpg), buddy[i])) { | |
186 | budnum = i; | |
187 | break; | |
188 | } | |
189 | BUG_ON(budnum == -1U); | |
190 | return budnum; | |
191 | } | |
192 | ||
193 | static char *zbud_data(struct zbud_hdr *zh, unsigned size) | |
194 | { | |
195 | struct zbud_page *zbpg; | |
196 | char *p; | |
197 | unsigned budnum; | |
198 | ||
199 | ASSERT_SENTINEL(zh, ZBH); | |
200 | budnum = zbud_budnum(zh); | |
201 | BUG_ON(size == 0 || size > zbud_max_buddy_size()); | |
202 | zbpg = container_of(zh, struct zbud_page, buddy[budnum]); | |
203 | ASSERT_SPINLOCK(&zbpg->lock); | |
204 | p = (char *)zbpg; | |
205 | if (budnum == 0) | |
206 | p += ((sizeof(struct zbud_page) + CHUNK_SIZE - 1) & | |
207 | CHUNK_MASK); | |
208 | else if (budnum == 1) | |
209 | p += PAGE_SIZE - ((size + CHUNK_SIZE - 1) & CHUNK_MASK); | |
210 | return p; | |
211 | } | |
212 | ||
213 | /* | |
214 | * zbud raw page management | |
215 | */ | |
216 | ||
217 | static struct zbud_page *zbud_alloc_raw_page(void) | |
218 | { | |
219 | struct zbud_page *zbpg = NULL; | |
220 | struct zbud_hdr *zh0, *zh1; | |
221 | bool recycled = 0; | |
222 | ||
223 | /* if any pages on the zbpg list, use one */ | |
224 | spin_lock(&zbpg_unused_list_spinlock); | |
225 | if (!list_empty(&zbpg_unused_list)) { | |
226 | zbpg = list_first_entry(&zbpg_unused_list, | |
227 | struct zbud_page, bud_list); | |
228 | list_del_init(&zbpg->bud_list); | |
229 | zcache_zbpg_unused_list_count--; | |
230 | recycled = 1; | |
231 | } | |
232 | spin_unlock(&zbpg_unused_list_spinlock); | |
233 | if (zbpg == NULL) | |
234 | /* none on zbpg list, try to get a kernel page */ | |
235 | zbpg = zcache_get_free_page(); | |
236 | if (likely(zbpg != NULL)) { | |
237 | INIT_LIST_HEAD(&zbpg->bud_list); | |
238 | zh0 = &zbpg->buddy[0]; zh1 = &zbpg->buddy[1]; | |
239 | spin_lock_init(&zbpg->lock); | |
240 | if (recycled) { | |
241 | ASSERT_INVERTED_SENTINEL(zbpg, ZBPG); | |
242 | SET_SENTINEL(zbpg, ZBPG); | |
243 | BUG_ON(zh0->size != 0 || tmem_oid_valid(&zh0->oid)); | |
244 | BUG_ON(zh1->size != 0 || tmem_oid_valid(&zh1->oid)); | |
245 | } else { | |
246 | atomic_inc(&zcache_zbud_curr_raw_pages); | |
247 | INIT_LIST_HEAD(&zbpg->bud_list); | |
248 | SET_SENTINEL(zbpg, ZBPG); | |
249 | zh0->size = 0; zh1->size = 0; | |
250 | tmem_oid_set_invalid(&zh0->oid); | |
251 | tmem_oid_set_invalid(&zh1->oid); | |
252 | } | |
253 | } | |
254 | return zbpg; | |
255 | } | |
256 | ||
257 | static void zbud_free_raw_page(struct zbud_page *zbpg) | |
258 | { | |
259 | struct zbud_hdr *zh0 = &zbpg->buddy[0], *zh1 = &zbpg->buddy[1]; | |
260 | ||
261 | ASSERT_SENTINEL(zbpg, ZBPG); | |
262 | BUG_ON(!list_empty(&zbpg->bud_list)); | |
263 | ASSERT_SPINLOCK(&zbpg->lock); | |
264 | BUG_ON(zh0->size != 0 || tmem_oid_valid(&zh0->oid)); | |
265 | BUG_ON(zh1->size != 0 || tmem_oid_valid(&zh1->oid)); | |
266 | INVERT_SENTINEL(zbpg, ZBPG); | |
267 | spin_unlock(&zbpg->lock); | |
268 | spin_lock(&zbpg_unused_list_spinlock); | |
269 | list_add(&zbpg->bud_list, &zbpg_unused_list); | |
270 | zcache_zbpg_unused_list_count++; | |
271 | spin_unlock(&zbpg_unused_list_spinlock); | |
272 | } | |
273 | ||
274 | /* | |
275 | * core zbud handling routines | |
276 | */ | |
277 | ||
278 | static unsigned zbud_free(struct zbud_hdr *zh) | |
279 | { | |
280 | unsigned size; | |
281 | ||
282 | ASSERT_SENTINEL(zh, ZBH); | |
283 | BUG_ON(!tmem_oid_valid(&zh->oid)); | |
284 | size = zh->size; | |
285 | BUG_ON(zh->size == 0 || zh->size > zbud_max_buddy_size()); | |
286 | zh->size = 0; | |
287 | tmem_oid_set_invalid(&zh->oid); | |
288 | INVERT_SENTINEL(zh, ZBH); | |
289 | zcache_zbud_curr_zbytes -= size; | |
290 | atomic_dec(&zcache_zbud_curr_zpages); | |
291 | return size; | |
292 | } | |
293 | ||
294 | static void zbud_free_and_delist(struct zbud_hdr *zh) | |
295 | { | |
296 | unsigned chunks; | |
297 | struct zbud_hdr *zh_other; | |
298 | unsigned budnum = zbud_budnum(zh), size; | |
299 | struct zbud_page *zbpg = | |
300 | container_of(zh, struct zbud_page, buddy[budnum]); | |
301 | ||
302 | spin_lock(&zbpg->lock); | |
303 | if (list_empty(&zbpg->bud_list)) { | |
304 | /* ignore zombie page... see zbud_evict_pages() */ | |
305 | spin_unlock(&zbpg->lock); | |
306 | return; | |
307 | } | |
308 | size = zbud_free(zh); | |
309 | ASSERT_SPINLOCK(&zbpg->lock); | |
310 | zh_other = &zbpg->buddy[(budnum == 0) ? 1 : 0]; | |
311 | if (zh_other->size == 0) { /* was unbuddied: unlist and free */ | |
312 | chunks = zbud_size_to_chunks(size) ; | |
313 | spin_lock(&zbud_budlists_spinlock); | |
314 | BUG_ON(list_empty(&zbud_unbuddied[chunks].list)); | |
315 | list_del_init(&zbpg->bud_list); | |
316 | zbud_unbuddied[chunks].count--; | |
317 | spin_unlock(&zbud_budlists_spinlock); | |
318 | zbud_free_raw_page(zbpg); | |
319 | } else { /* was buddied: move remaining buddy to unbuddied list */ | |
320 | chunks = zbud_size_to_chunks(zh_other->size) ; | |
321 | spin_lock(&zbud_budlists_spinlock); | |
322 | list_del_init(&zbpg->bud_list); | |
323 | zcache_zbud_buddied_count--; | |
324 | list_add_tail(&zbpg->bud_list, &zbud_unbuddied[chunks].list); | |
325 | zbud_unbuddied[chunks].count++; | |
326 | spin_unlock(&zbud_budlists_spinlock); | |
327 | spin_unlock(&zbpg->lock); | |
328 | } | |
329 | } | |
330 | ||
966b9016 DM |
331 | static struct zbud_hdr *zbud_create(uint16_t client_id, uint16_t pool_id, |
332 | struct tmem_oid *oid, | |
9cc06bf8 DM |
333 | uint32_t index, struct page *page, |
334 | void *cdata, unsigned size) | |
335 | { | |
336 | struct zbud_hdr *zh0, *zh1, *zh = NULL; | |
337 | struct zbud_page *zbpg = NULL, *ztmp; | |
338 | unsigned nchunks; | |
339 | char *to; | |
340 | int i, found_good_buddy = 0; | |
341 | ||
342 | nchunks = zbud_size_to_chunks(size) ; | |
343 | for (i = MAX_CHUNK - nchunks + 1; i > 0; i--) { | |
344 | spin_lock(&zbud_budlists_spinlock); | |
345 | if (!list_empty(&zbud_unbuddied[i].list)) { | |
346 | list_for_each_entry_safe(zbpg, ztmp, | |
347 | &zbud_unbuddied[i].list, bud_list) { | |
348 | if (spin_trylock(&zbpg->lock)) { | |
349 | found_good_buddy = i; | |
350 | goto found_unbuddied; | |
351 | } | |
352 | } | |
353 | } | |
354 | spin_unlock(&zbud_budlists_spinlock); | |
355 | } | |
356 | /* didn't find a good buddy, try allocating a new page */ | |
357 | zbpg = zbud_alloc_raw_page(); | |
358 | if (unlikely(zbpg == NULL)) | |
359 | goto out; | |
360 | /* ok, have a page, now compress the data before taking locks */ | |
361 | spin_lock(&zbpg->lock); | |
362 | spin_lock(&zbud_budlists_spinlock); | |
363 | list_add_tail(&zbpg->bud_list, &zbud_unbuddied[nchunks].list); | |
364 | zbud_unbuddied[nchunks].count++; | |
365 | zh = &zbpg->buddy[0]; | |
366 | goto init_zh; | |
367 | ||
368 | found_unbuddied: | |
369 | ASSERT_SPINLOCK(&zbpg->lock); | |
370 | zh0 = &zbpg->buddy[0]; zh1 = &zbpg->buddy[1]; | |
371 | BUG_ON(!((zh0->size == 0) ^ (zh1->size == 0))); | |
372 | if (zh0->size != 0) { /* buddy0 in use, buddy1 is vacant */ | |
373 | ASSERT_SENTINEL(zh0, ZBH); | |
374 | zh = zh1; | |
375 | } else if (zh1->size != 0) { /* buddy1 in use, buddy0 is vacant */ | |
376 | ASSERT_SENTINEL(zh1, ZBH); | |
377 | zh = zh0; | |
378 | } else | |
379 | BUG(); | |
380 | list_del_init(&zbpg->bud_list); | |
381 | zbud_unbuddied[found_good_buddy].count--; | |
382 | list_add_tail(&zbpg->bud_list, &zbud_buddied_list); | |
383 | zcache_zbud_buddied_count++; | |
384 | ||
385 | init_zh: | |
386 | SET_SENTINEL(zh, ZBH); | |
387 | zh->size = size; | |
388 | zh->index = index; | |
389 | zh->oid = *oid; | |
390 | zh->pool_id = pool_id; | |
966b9016 | 391 | zh->client_id = client_id; |
9cc06bf8 DM |
392 | /* can wait to copy the data until the list locks are dropped */ |
393 | spin_unlock(&zbud_budlists_spinlock); | |
394 | ||
395 | to = zbud_data(zh, size); | |
396 | memcpy(to, cdata, size); | |
397 | spin_unlock(&zbpg->lock); | |
398 | zbud_cumul_chunk_counts[nchunks]++; | |
399 | atomic_inc(&zcache_zbud_curr_zpages); | |
400 | zcache_zbud_cumul_zpages++; | |
401 | zcache_zbud_curr_zbytes += size; | |
402 | zcache_zbud_cumul_zbytes += size; | |
403 | out: | |
404 | return zh; | |
405 | } | |
406 | ||
407 | static int zbud_decompress(struct page *page, struct zbud_hdr *zh) | |
408 | { | |
409 | struct zbud_page *zbpg; | |
410 | unsigned budnum = zbud_budnum(zh); | |
411 | size_t out_len = PAGE_SIZE; | |
412 | char *to_va, *from_va; | |
413 | unsigned size; | |
414 | int ret = 0; | |
415 | ||
416 | zbpg = container_of(zh, struct zbud_page, buddy[budnum]); | |
417 | spin_lock(&zbpg->lock); | |
418 | if (list_empty(&zbpg->bud_list)) { | |
419 | /* ignore zombie page... see zbud_evict_pages() */ | |
420 | ret = -EINVAL; | |
421 | goto out; | |
422 | } | |
423 | ASSERT_SENTINEL(zh, ZBH); | |
424 | BUG_ON(zh->size == 0 || zh->size > zbud_max_buddy_size()); | |
425 | to_va = kmap_atomic(page, KM_USER0); | |
426 | size = zh->size; | |
427 | from_va = zbud_data(zh, size); | |
428 | ret = lzo1x_decompress_safe(from_va, size, to_va, &out_len); | |
429 | BUG_ON(ret != LZO_E_OK); | |
430 | BUG_ON(out_len != PAGE_SIZE); | |
431 | kunmap_atomic(to_va, KM_USER0); | |
432 | out: | |
433 | spin_unlock(&zbpg->lock); | |
434 | return ret; | |
435 | } | |
436 | ||
437 | /* | |
438 | * The following routines handle shrinking of ephemeral pages by evicting | |
439 | * pages "least valuable" first. | |
440 | */ | |
441 | ||
442 | static unsigned long zcache_evicted_raw_pages; | |
443 | static unsigned long zcache_evicted_buddied_pages; | |
444 | static unsigned long zcache_evicted_unbuddied_pages; | |
445 | ||
966b9016 DM |
446 | static struct tmem_pool *zcache_get_pool_by_id(uint16_t cli_id, |
447 | uint16_t poolid); | |
9cc06bf8 DM |
448 | static void zcache_put_pool(struct tmem_pool *pool); |
449 | ||
450 | /* | |
451 | * Flush and free all zbuds in a zbpg, then free the pageframe | |
452 | */ | |
453 | static void zbud_evict_zbpg(struct zbud_page *zbpg) | |
454 | { | |
455 | struct zbud_hdr *zh; | |
456 | int i, j; | |
966b9016 DM |
457 | uint32_t pool_id[ZBUD_MAX_BUDS], client_id[ZBUD_MAX_BUDS]; |
458 | uint32_t index[ZBUD_MAX_BUDS]; | |
9cc06bf8 DM |
459 | struct tmem_oid oid[ZBUD_MAX_BUDS]; |
460 | struct tmem_pool *pool; | |
461 | ||
462 | ASSERT_SPINLOCK(&zbpg->lock); | |
463 | BUG_ON(!list_empty(&zbpg->bud_list)); | |
464 | for (i = 0, j = 0; i < ZBUD_MAX_BUDS; i++) { | |
465 | zh = &zbpg->buddy[i]; | |
466 | if (zh->size) { | |
966b9016 | 467 | client_id[j] = zh->client_id; |
9cc06bf8 DM |
468 | pool_id[j] = zh->pool_id; |
469 | oid[j] = zh->oid; | |
470 | index[j] = zh->index; | |
471 | j++; | |
472 | zbud_free(zh); | |
473 | } | |
474 | } | |
475 | spin_unlock(&zbpg->lock); | |
476 | for (i = 0; i < j; i++) { | |
966b9016 | 477 | pool = zcache_get_pool_by_id(client_id[i], pool_id[i]); |
9cc06bf8 DM |
478 | if (pool != NULL) { |
479 | tmem_flush_page(pool, &oid[i], index[i]); | |
480 | zcache_put_pool(pool); | |
481 | } | |
482 | } | |
483 | ASSERT_SENTINEL(zbpg, ZBPG); | |
484 | spin_lock(&zbpg->lock); | |
485 | zbud_free_raw_page(zbpg); | |
486 | } | |
487 | ||
488 | /* | |
489 | * Free nr pages. This code is funky because we want to hold the locks | |
490 | * protecting various lists for as short a time as possible, and in some | |
491 | * circumstances the list may change asynchronously when the list lock is | |
492 | * not held. In some cases we also trylock not only to avoid waiting on a | |
493 | * page in use by another cpu, but also to avoid potential deadlock due to | |
494 | * lock inversion. | |
495 | */ | |
496 | static void zbud_evict_pages(int nr) | |
497 | { | |
498 | struct zbud_page *zbpg; | |
499 | int i; | |
500 | ||
501 | /* first try freeing any pages on unused list */ | |
502 | retry_unused_list: | |
503 | spin_lock_bh(&zbpg_unused_list_spinlock); | |
504 | if (!list_empty(&zbpg_unused_list)) { | |
505 | /* can't walk list here, since it may change when unlocked */ | |
506 | zbpg = list_first_entry(&zbpg_unused_list, | |
507 | struct zbud_page, bud_list); | |
508 | list_del_init(&zbpg->bud_list); | |
509 | zcache_zbpg_unused_list_count--; | |
510 | atomic_dec(&zcache_zbud_curr_raw_pages); | |
511 | spin_unlock_bh(&zbpg_unused_list_spinlock); | |
512 | zcache_free_page(zbpg); | |
513 | zcache_evicted_raw_pages++; | |
514 | if (--nr <= 0) | |
515 | goto out; | |
516 | goto retry_unused_list; | |
517 | } | |
518 | spin_unlock_bh(&zbpg_unused_list_spinlock); | |
519 | ||
520 | /* now try freeing unbuddied pages, starting with least space avail */ | |
521 | for (i = 0; i < MAX_CHUNK; i++) { | |
522 | retry_unbud_list_i: | |
523 | spin_lock_bh(&zbud_budlists_spinlock); | |
524 | if (list_empty(&zbud_unbuddied[i].list)) { | |
525 | spin_unlock_bh(&zbud_budlists_spinlock); | |
526 | continue; | |
527 | } | |
528 | list_for_each_entry(zbpg, &zbud_unbuddied[i].list, bud_list) { | |
529 | if (unlikely(!spin_trylock(&zbpg->lock))) | |
530 | continue; | |
531 | list_del_init(&zbpg->bud_list); | |
532 | zbud_unbuddied[i].count--; | |
533 | spin_unlock(&zbud_budlists_spinlock); | |
534 | zcache_evicted_unbuddied_pages++; | |
535 | /* want budlists unlocked when doing zbpg eviction */ | |
536 | zbud_evict_zbpg(zbpg); | |
537 | local_bh_enable(); | |
538 | if (--nr <= 0) | |
539 | goto out; | |
540 | goto retry_unbud_list_i; | |
541 | } | |
542 | spin_unlock_bh(&zbud_budlists_spinlock); | |
543 | } | |
544 | ||
545 | /* as a last resort, free buddied pages */ | |
546 | retry_bud_list: | |
547 | spin_lock_bh(&zbud_budlists_spinlock); | |
548 | if (list_empty(&zbud_buddied_list)) { | |
549 | spin_unlock_bh(&zbud_budlists_spinlock); | |
550 | goto out; | |
551 | } | |
552 | list_for_each_entry(zbpg, &zbud_buddied_list, bud_list) { | |
553 | if (unlikely(!spin_trylock(&zbpg->lock))) | |
554 | continue; | |
555 | list_del_init(&zbpg->bud_list); | |
556 | zcache_zbud_buddied_count--; | |
557 | spin_unlock(&zbud_budlists_spinlock); | |
558 | zcache_evicted_buddied_pages++; | |
559 | /* want budlists unlocked when doing zbpg eviction */ | |
560 | zbud_evict_zbpg(zbpg); | |
561 | local_bh_enable(); | |
562 | if (--nr <= 0) | |
563 | goto out; | |
564 | goto retry_bud_list; | |
565 | } | |
566 | spin_unlock_bh(&zbud_budlists_spinlock); | |
567 | out: | |
568 | return; | |
569 | } | |
570 | ||
571 | static void zbud_init(void) | |
572 | { | |
573 | int i; | |
574 | ||
575 | INIT_LIST_HEAD(&zbud_buddied_list); | |
576 | zcache_zbud_buddied_count = 0; | |
577 | for (i = 0; i < NCHUNKS; i++) { | |
578 | INIT_LIST_HEAD(&zbud_unbuddied[i].list); | |
579 | zbud_unbuddied[i].count = 0; | |
580 | } | |
581 | } | |
582 | ||
583 | #ifdef CONFIG_SYSFS | |
584 | /* | |
585 | * These sysfs routines show a nice distribution of how many zbpg's are | |
586 | * currently (and have ever been placed) in each unbuddied list. It's fun | |
587 | * to watch but can probably go away before final merge. | |
588 | */ | |
589 | static int zbud_show_unbuddied_list_counts(char *buf) | |
590 | { | |
591 | int i; | |
592 | char *p = buf; | |
593 | ||
966b9016 | 594 | for (i = 0; i < NCHUNKS; i++) |
9cc06bf8 | 595 | p += sprintf(p, "%u ", zbud_unbuddied[i].count); |
9cc06bf8 DM |
596 | return p - buf; |
597 | } | |
598 | ||
599 | static int zbud_show_cumul_chunk_counts(char *buf) | |
600 | { | |
601 | unsigned long i, chunks = 0, total_chunks = 0, sum_total_chunks = 0; | |
602 | unsigned long total_chunks_lte_21 = 0, total_chunks_lte_32 = 0; | |
603 | unsigned long total_chunks_lte_42 = 0; | |
604 | char *p = buf; | |
605 | ||
606 | for (i = 0; i < NCHUNKS; i++) { | |
607 | p += sprintf(p, "%lu ", zbud_cumul_chunk_counts[i]); | |
608 | chunks += zbud_cumul_chunk_counts[i]; | |
609 | total_chunks += zbud_cumul_chunk_counts[i]; | |
610 | sum_total_chunks += i * zbud_cumul_chunk_counts[i]; | |
611 | if (i == 21) | |
612 | total_chunks_lte_21 = total_chunks; | |
613 | if (i == 32) | |
614 | total_chunks_lte_32 = total_chunks; | |
615 | if (i == 42) | |
616 | total_chunks_lte_42 = total_chunks; | |
617 | } | |
618 | p += sprintf(p, "<=21:%lu <=32:%lu <=42:%lu, mean:%lu\n", | |
619 | total_chunks_lte_21, total_chunks_lte_32, total_chunks_lte_42, | |
620 | chunks == 0 ? 0 : sum_total_chunks / chunks); | |
621 | return p - buf; | |
622 | } | |
623 | #endif | |
624 | ||
625 | /********** | |
626 | * This "zv" PAM implementation combines the TLSF-based xvMalloc | |
627 | * with lzo1x compression to maximize the amount of data that can | |
628 | * be packed into a physical page. | |
629 | * | |
630 | * Zv represents a PAM page with the index and object (plus a "size" value | |
631 | * necessary for decompression) immediately preceding the compressed data. | |
632 | */ | |
633 | ||
634 | #define ZVH_SENTINEL 0x43214321 | |
635 | ||
636 | struct zv_hdr { | |
637 | uint32_t pool_id; | |
638 | struct tmem_oid oid; | |
639 | uint32_t index; | |
640 | DECL_SENTINEL | |
641 | }; | |
642 | ||
966b9016 DM |
643 | /* rudimentary policy limits */ |
644 | /* total number of persistent pages may not exceed this percentage */ | |
645 | static unsigned int zv_page_count_policy_percent = 75; | |
646 | /* | |
647 | * byte count defining poor compression; pages with greater zsize will be | |
648 | * rejected | |
649 | */ | |
650 | static unsigned int zv_max_zsize = (PAGE_SIZE / 8) * 7; | |
651 | /* | |
652 | * byte count defining poor *mean* compression; pages with greater zsize | |
653 | * will be rejected until sufficient better-compressed pages are accepted | |
0428fec3 | 654 | * driving the mean below this threshold |
966b9016 DM |
655 | */ |
656 | static unsigned int zv_max_mean_zsize = (PAGE_SIZE / 8) * 5; | |
657 | ||
658 | static unsigned long zv_curr_dist_counts[NCHUNKS]; | |
659 | static unsigned long zv_cumul_dist_counts[NCHUNKS]; | |
9cc06bf8 DM |
660 | |
661 | static struct zv_hdr *zv_create(struct xv_pool *xvpool, uint32_t pool_id, | |
662 | struct tmem_oid *oid, uint32_t index, | |
663 | void *cdata, unsigned clen) | |
664 | { | |
665 | struct page *page; | |
666 | struct zv_hdr *zv = NULL; | |
667 | uint32_t offset; | |
966b9016 DM |
668 | int alloc_size = clen + sizeof(struct zv_hdr); |
669 | int chunks = (alloc_size + (CHUNK_SIZE - 1)) >> CHUNK_SHIFT; | |
9cc06bf8 DM |
670 | int ret; |
671 | ||
672 | BUG_ON(!irqs_disabled()); | |
966b9016 DM |
673 | BUG_ON(chunks >= NCHUNKS); |
674 | ret = xv_malloc(xvpool, alloc_size, | |
9cc06bf8 DM |
675 | &page, &offset, ZCACHE_GFP_MASK); |
676 | if (unlikely(ret)) | |
677 | goto out; | |
966b9016 DM |
678 | zv_curr_dist_counts[chunks]++; |
679 | zv_cumul_dist_counts[chunks]++; | |
9cc06bf8 DM |
680 | zv = kmap_atomic(page, KM_USER0) + offset; |
681 | zv->index = index; | |
682 | zv->oid = *oid; | |
683 | zv->pool_id = pool_id; | |
684 | SET_SENTINEL(zv, ZVH); | |
685 | memcpy((char *)zv + sizeof(struct zv_hdr), cdata, clen); | |
686 | kunmap_atomic(zv, KM_USER0); | |
687 | out: | |
688 | return zv; | |
689 | } | |
690 | ||
691 | static void zv_free(struct xv_pool *xvpool, struct zv_hdr *zv) | |
692 | { | |
693 | unsigned long flags; | |
694 | struct page *page; | |
695 | uint32_t offset; | |
966b9016 DM |
696 | uint16_t size = xv_get_object_size(zv); |
697 | int chunks = (size + (CHUNK_SIZE - 1)) >> CHUNK_SHIFT; | |
9cc06bf8 DM |
698 | |
699 | ASSERT_SENTINEL(zv, ZVH); | |
966b9016 DM |
700 | BUG_ON(chunks >= NCHUNKS); |
701 | zv_curr_dist_counts[chunks]--; | |
702 | size -= sizeof(*zv); | |
703 | BUG_ON(size == 0); | |
9cc06bf8 DM |
704 | INVERT_SENTINEL(zv, ZVH); |
705 | page = virt_to_page(zv); | |
706 | offset = (unsigned long)zv & ~PAGE_MASK; | |
707 | local_irq_save(flags); | |
708 | xv_free(xvpool, page, offset); | |
709 | local_irq_restore(flags); | |
710 | } | |
711 | ||
712 | static void zv_decompress(struct page *page, struct zv_hdr *zv) | |
713 | { | |
714 | size_t clen = PAGE_SIZE; | |
715 | char *to_va; | |
716 | unsigned size; | |
717 | int ret; | |
718 | ||
719 | ASSERT_SENTINEL(zv, ZVH); | |
720 | size = xv_get_object_size(zv) - sizeof(*zv); | |
966b9016 | 721 | BUG_ON(size == 0); |
9cc06bf8 DM |
722 | to_va = kmap_atomic(page, KM_USER0); |
723 | ret = lzo1x_decompress_safe((char *)zv + sizeof(*zv), | |
724 | size, to_va, &clen); | |
725 | kunmap_atomic(to_va, KM_USER0); | |
726 | BUG_ON(ret != LZO_E_OK); | |
727 | BUG_ON(clen != PAGE_SIZE); | |
728 | } | |
729 | ||
966b9016 DM |
730 | #ifdef CONFIG_SYSFS |
731 | /* | |
732 | * show a distribution of compression stats for zv pages. | |
733 | */ | |
734 | ||
735 | static int zv_curr_dist_counts_show(char *buf) | |
736 | { | |
737 | unsigned long i, n, chunks = 0, sum_total_chunks = 0; | |
738 | char *p = buf; | |
739 | ||
740 | for (i = 0; i < NCHUNKS; i++) { | |
741 | n = zv_curr_dist_counts[i]; | |
742 | p += sprintf(p, "%lu ", n); | |
743 | chunks += n; | |
744 | sum_total_chunks += i * n; | |
745 | } | |
746 | p += sprintf(p, "mean:%lu\n", | |
747 | chunks == 0 ? 0 : sum_total_chunks / chunks); | |
748 | return p - buf; | |
749 | } | |
750 | ||
751 | static int zv_cumul_dist_counts_show(char *buf) | |
752 | { | |
753 | unsigned long i, n, chunks = 0, sum_total_chunks = 0; | |
754 | char *p = buf; | |
755 | ||
756 | for (i = 0; i < NCHUNKS; i++) { | |
757 | n = zv_cumul_dist_counts[i]; | |
758 | p += sprintf(p, "%lu ", n); | |
759 | chunks += n; | |
760 | sum_total_chunks += i * n; | |
761 | } | |
762 | p += sprintf(p, "mean:%lu\n", | |
763 | chunks == 0 ? 0 : sum_total_chunks / chunks); | |
764 | return p - buf; | |
765 | } | |
766 | ||
767 | /* | |
768 | * setting zv_max_zsize via sysfs causes all persistent (e.g. swap) | |
769 | * pages that don't compress to less than this value (including metadata | |
770 | * overhead) to be rejected. We don't allow the value to get too close | |
771 | * to PAGE_SIZE. | |
772 | */ | |
773 | static ssize_t zv_max_zsize_show(struct kobject *kobj, | |
774 | struct kobj_attribute *attr, | |
775 | char *buf) | |
776 | { | |
777 | return sprintf(buf, "%u\n", zv_max_zsize); | |
778 | } | |
779 | ||
780 | static ssize_t zv_max_zsize_store(struct kobject *kobj, | |
781 | struct kobj_attribute *attr, | |
782 | const char *buf, size_t count) | |
783 | { | |
784 | unsigned long val; | |
785 | int err; | |
786 | ||
787 | if (!capable(CAP_SYS_ADMIN)) | |
788 | return -EPERM; | |
789 | ||
ebadb730 | 790 | err = kstrtoul(buf, 10, &val); |
966b9016 DM |
791 | if (err || (val == 0) || (val > (PAGE_SIZE / 8) * 7)) |
792 | return -EINVAL; | |
793 | zv_max_zsize = val; | |
794 | return count; | |
795 | } | |
796 | ||
797 | /* | |
798 | * setting zv_max_mean_zsize via sysfs causes all persistent (e.g. swap) | |
799 | * pages that don't compress to less than this value (including metadata | |
800 | * overhead) to be rejected UNLESS the mean compression is also smaller | |
801 | * than this value. In other words, we are load-balancing-by-zsize the | |
802 | * accepted pages. Again, we don't allow the value to get too close | |
803 | * to PAGE_SIZE. | |
804 | */ | |
805 | static ssize_t zv_max_mean_zsize_show(struct kobject *kobj, | |
806 | struct kobj_attribute *attr, | |
807 | char *buf) | |
808 | { | |
809 | return sprintf(buf, "%u\n", zv_max_mean_zsize); | |
810 | } | |
811 | ||
812 | static ssize_t zv_max_mean_zsize_store(struct kobject *kobj, | |
813 | struct kobj_attribute *attr, | |
814 | const char *buf, size_t count) | |
815 | { | |
816 | unsigned long val; | |
817 | int err; | |
818 | ||
819 | if (!capable(CAP_SYS_ADMIN)) | |
820 | return -EPERM; | |
821 | ||
ebadb730 | 822 | err = kstrtoul(buf, 10, &val); |
966b9016 DM |
823 | if (err || (val == 0) || (val > (PAGE_SIZE / 8) * 7)) |
824 | return -EINVAL; | |
825 | zv_max_mean_zsize = val; | |
826 | return count; | |
827 | } | |
828 | ||
829 | /* | |
830 | * setting zv_page_count_policy_percent via sysfs sets an upper bound of | |
831 | * persistent (e.g. swap) pages that will be retained according to: | |
832 | * (zv_page_count_policy_percent * totalram_pages) / 100) | |
833 | * when that limit is reached, further puts will be rejected (until | |
834 | * some pages have been flushed). Note that, due to compression, | |
835 | * this number may exceed 100; it defaults to 75 and we set an | |
836 | * arbitary limit of 150. A poor choice will almost certainly result | |
837 | * in OOM's, so this value should only be changed prudently. | |
838 | */ | |
839 | static ssize_t zv_page_count_policy_percent_show(struct kobject *kobj, | |
840 | struct kobj_attribute *attr, | |
841 | char *buf) | |
842 | { | |
843 | return sprintf(buf, "%u\n", zv_page_count_policy_percent); | |
844 | } | |
845 | ||
846 | static ssize_t zv_page_count_policy_percent_store(struct kobject *kobj, | |
847 | struct kobj_attribute *attr, | |
848 | const char *buf, size_t count) | |
849 | { | |
850 | unsigned long val; | |
851 | int err; | |
852 | ||
853 | if (!capable(CAP_SYS_ADMIN)) | |
854 | return -EPERM; | |
855 | ||
ebadb730 | 856 | err = kstrtoul(buf, 10, &val); |
966b9016 DM |
857 | if (err || (val == 0) || (val > 150)) |
858 | return -EINVAL; | |
859 | zv_page_count_policy_percent = val; | |
860 | return count; | |
861 | } | |
862 | ||
863 | static struct kobj_attribute zcache_zv_max_zsize_attr = { | |
864 | .attr = { .name = "zv_max_zsize", .mode = 0644 }, | |
865 | .show = zv_max_zsize_show, | |
866 | .store = zv_max_zsize_store, | |
867 | }; | |
868 | ||
869 | static struct kobj_attribute zcache_zv_max_mean_zsize_attr = { | |
870 | .attr = { .name = "zv_max_mean_zsize", .mode = 0644 }, | |
871 | .show = zv_max_mean_zsize_show, | |
872 | .store = zv_max_mean_zsize_store, | |
873 | }; | |
874 | ||
875 | static struct kobj_attribute zcache_zv_page_count_policy_percent_attr = { | |
876 | .attr = { .name = "zv_page_count_policy_percent", | |
877 | .mode = 0644 }, | |
878 | .show = zv_page_count_policy_percent_show, | |
879 | .store = zv_page_count_policy_percent_store, | |
880 | }; | |
881 | #endif | |
882 | ||
9cc06bf8 DM |
883 | /* |
884 | * zcache core code starts here | |
885 | */ | |
886 | ||
887 | /* useful stats not collected by cleancache or frontswap */ | |
888 | static unsigned long zcache_flush_total; | |
889 | static unsigned long zcache_flush_found; | |
890 | static unsigned long zcache_flobj_total; | |
891 | static unsigned long zcache_flobj_found; | |
892 | static unsigned long zcache_failed_eph_puts; | |
893 | static unsigned long zcache_failed_pers_puts; | |
894 | ||
9cc06bf8 DM |
895 | /* |
896 | * Tmem operations assume the poolid implies the invoking client. | |
966b9016 DM |
897 | * Zcache only has one client (the kernel itself): LOCAL_CLIENT. |
898 | * RAMster has each client numbered by cluster node, and a KVM version | |
9cc06bf8 DM |
899 | * of zcache would have one client per guest and each client might |
900 | * have a poolid==N. | |
901 | */ | |
966b9016 | 902 | static struct tmem_pool *zcache_get_pool_by_id(uint16_t cli_id, uint16_t poolid) |
9cc06bf8 DM |
903 | { |
904 | struct tmem_pool *pool = NULL; | |
966b9016 | 905 | struct zcache_client *cli = NULL; |
9cc06bf8 | 906 | |
966b9016 DM |
907 | if (cli_id == LOCAL_CLIENT) |
908 | cli = &zcache_host; | |
909 | else { | |
910 | if (cli_id >= MAX_CLIENTS) | |
911 | goto out; | |
912 | cli = &zcache_clients[cli_id]; | |
913 | if (cli == NULL) | |
914 | goto out; | |
915 | atomic_inc(&cli->refcount); | |
916 | } | |
917 | if (poolid < MAX_POOLS_PER_CLIENT) { | |
918 | pool = cli->tmem_pools[poolid]; | |
9cc06bf8 DM |
919 | if (pool != NULL) |
920 | atomic_inc(&pool->refcount); | |
921 | } | |
966b9016 | 922 | out: |
9cc06bf8 DM |
923 | return pool; |
924 | } | |
925 | ||
926 | static void zcache_put_pool(struct tmem_pool *pool) | |
927 | { | |
966b9016 DM |
928 | struct zcache_client *cli = NULL; |
929 | ||
930 | if (pool == NULL) | |
931 | BUG(); | |
932 | cli = pool->client; | |
933 | atomic_dec(&pool->refcount); | |
934 | atomic_dec(&cli->refcount); | |
935 | } | |
936 | ||
937 | int zcache_new_client(uint16_t cli_id) | |
938 | { | |
939 | struct zcache_client *cli = NULL; | |
940 | int ret = -1; | |
941 | ||
942 | if (cli_id == LOCAL_CLIENT) | |
943 | cli = &zcache_host; | |
944 | else if ((unsigned int)cli_id < MAX_CLIENTS) | |
945 | cli = &zcache_clients[cli_id]; | |
946 | if (cli == NULL) | |
947 | goto out; | |
948 | if (cli->allocated) | |
949 | goto out; | |
950 | cli->allocated = 1; | |
951 | #ifdef CONFIG_FRONTSWAP | |
952 | cli->xvpool = xv_create_pool(); | |
953 | if (cli->xvpool == NULL) | |
954 | goto out; | |
955 | #endif | |
956 | ret = 0; | |
957 | out: | |
958 | return ret; | |
9cc06bf8 DM |
959 | } |
960 | ||
961 | /* counters for debugging */ | |
962 | static unsigned long zcache_failed_get_free_pages; | |
963 | static unsigned long zcache_failed_alloc; | |
964 | static unsigned long zcache_put_to_flush; | |
9cc06bf8 DM |
965 | |
966 | /* | |
967 | * for now, used named slabs so can easily track usage; later can | |
968 | * either just use kmalloc, or perhaps add a slab-like allocator | |
969 | * to more carefully manage total memory utilization | |
970 | */ | |
971 | static struct kmem_cache *zcache_objnode_cache; | |
972 | static struct kmem_cache *zcache_obj_cache; | |
973 | static atomic_t zcache_curr_obj_count = ATOMIC_INIT(0); | |
974 | static unsigned long zcache_curr_obj_count_max; | |
975 | static atomic_t zcache_curr_objnode_count = ATOMIC_INIT(0); | |
976 | static unsigned long zcache_curr_objnode_count_max; | |
977 | ||
978 | /* | |
979 | * to avoid memory allocation recursion (e.g. due to direct reclaim), we | |
980 | * preload all necessary data structures so the hostops callbacks never | |
981 | * actually do a malloc | |
982 | */ | |
983 | struct zcache_preload { | |
984 | void *page; | |
985 | struct tmem_obj *obj; | |
986 | int nr; | |
987 | struct tmem_objnode *objnodes[OBJNODE_TREE_MAX_PATH]; | |
988 | }; | |
989 | static DEFINE_PER_CPU(struct zcache_preload, zcache_preloads) = { 0, }; | |
990 | ||
991 | static int zcache_do_preload(struct tmem_pool *pool) | |
992 | { | |
993 | struct zcache_preload *kp; | |
994 | struct tmem_objnode *objnode; | |
995 | struct tmem_obj *obj; | |
996 | void *page; | |
997 | int ret = -ENOMEM; | |
998 | ||
999 | if (unlikely(zcache_objnode_cache == NULL)) | |
1000 | goto out; | |
1001 | if (unlikely(zcache_obj_cache == NULL)) | |
1002 | goto out; | |
9cc06bf8 DM |
1003 | preempt_disable(); |
1004 | kp = &__get_cpu_var(zcache_preloads); | |
1005 | while (kp->nr < ARRAY_SIZE(kp->objnodes)) { | |
1006 | preempt_enable_no_resched(); | |
1007 | objnode = kmem_cache_alloc(zcache_objnode_cache, | |
1008 | ZCACHE_GFP_MASK); | |
1009 | if (unlikely(objnode == NULL)) { | |
1010 | zcache_failed_alloc++; | |
00bf2560 | 1011 | goto out; |
9cc06bf8 DM |
1012 | } |
1013 | preempt_disable(); | |
1014 | kp = &__get_cpu_var(zcache_preloads); | |
1015 | if (kp->nr < ARRAY_SIZE(kp->objnodes)) | |
1016 | kp->objnodes[kp->nr++] = objnode; | |
1017 | else | |
1018 | kmem_cache_free(zcache_objnode_cache, objnode); | |
1019 | } | |
1020 | preempt_enable_no_resched(); | |
1021 | obj = kmem_cache_alloc(zcache_obj_cache, ZCACHE_GFP_MASK); | |
1022 | if (unlikely(obj == NULL)) { | |
1023 | zcache_failed_alloc++; | |
00bf2560 | 1024 | goto out; |
9cc06bf8 DM |
1025 | } |
1026 | page = (void *)__get_free_page(ZCACHE_GFP_MASK); | |
1027 | if (unlikely(page == NULL)) { | |
1028 | zcache_failed_get_free_pages++; | |
69648bed | 1029 | kmem_cache_free(zcache_obj_cache, obj); |
00bf2560 | 1030 | goto out; |
9cc06bf8 DM |
1031 | } |
1032 | preempt_disable(); | |
1033 | kp = &__get_cpu_var(zcache_preloads); | |
1034 | if (kp->obj == NULL) | |
1035 | kp->obj = obj; | |
1036 | else | |
1037 | kmem_cache_free(zcache_obj_cache, obj); | |
1038 | if (kp->page == NULL) | |
1039 | kp->page = page; | |
1040 | else | |
1041 | free_page((unsigned long)page); | |
1042 | ret = 0; | |
9cc06bf8 DM |
1043 | out: |
1044 | return ret; | |
1045 | } | |
1046 | ||
1047 | static void *zcache_get_free_page(void) | |
1048 | { | |
1049 | struct zcache_preload *kp; | |
1050 | void *page; | |
1051 | ||
1052 | kp = &__get_cpu_var(zcache_preloads); | |
1053 | page = kp->page; | |
1054 | BUG_ON(page == NULL); | |
1055 | kp->page = NULL; | |
1056 | return page; | |
1057 | } | |
1058 | ||
1059 | static void zcache_free_page(void *p) | |
1060 | { | |
1061 | free_page((unsigned long)p); | |
1062 | } | |
1063 | ||
1064 | /* | |
1065 | * zcache implementation for tmem host ops | |
1066 | */ | |
1067 | ||
1068 | static struct tmem_objnode *zcache_objnode_alloc(struct tmem_pool *pool) | |
1069 | { | |
1070 | struct tmem_objnode *objnode = NULL; | |
1071 | unsigned long count; | |
1072 | struct zcache_preload *kp; | |
1073 | ||
1074 | kp = &__get_cpu_var(zcache_preloads); | |
1075 | if (kp->nr <= 0) | |
1076 | goto out; | |
1077 | objnode = kp->objnodes[kp->nr - 1]; | |
1078 | BUG_ON(objnode == NULL); | |
1079 | kp->objnodes[kp->nr - 1] = NULL; | |
1080 | kp->nr--; | |
1081 | count = atomic_inc_return(&zcache_curr_objnode_count); | |
1082 | if (count > zcache_curr_objnode_count_max) | |
1083 | zcache_curr_objnode_count_max = count; | |
1084 | out: | |
1085 | return objnode; | |
1086 | } | |
1087 | ||
1088 | static void zcache_objnode_free(struct tmem_objnode *objnode, | |
1089 | struct tmem_pool *pool) | |
1090 | { | |
1091 | atomic_dec(&zcache_curr_objnode_count); | |
1092 | BUG_ON(atomic_read(&zcache_curr_objnode_count) < 0); | |
1093 | kmem_cache_free(zcache_objnode_cache, objnode); | |
1094 | } | |
1095 | ||
1096 | static struct tmem_obj *zcache_obj_alloc(struct tmem_pool *pool) | |
1097 | { | |
1098 | struct tmem_obj *obj = NULL; | |
1099 | unsigned long count; | |
1100 | struct zcache_preload *kp; | |
1101 | ||
1102 | kp = &__get_cpu_var(zcache_preloads); | |
1103 | obj = kp->obj; | |
1104 | BUG_ON(obj == NULL); | |
1105 | kp->obj = NULL; | |
1106 | count = atomic_inc_return(&zcache_curr_obj_count); | |
1107 | if (count > zcache_curr_obj_count_max) | |
1108 | zcache_curr_obj_count_max = count; | |
1109 | return obj; | |
1110 | } | |
1111 | ||
1112 | static void zcache_obj_free(struct tmem_obj *obj, struct tmem_pool *pool) | |
1113 | { | |
1114 | atomic_dec(&zcache_curr_obj_count); | |
1115 | BUG_ON(atomic_read(&zcache_curr_obj_count) < 0); | |
1116 | kmem_cache_free(zcache_obj_cache, obj); | |
1117 | } | |
1118 | ||
1119 | static struct tmem_hostops zcache_hostops = { | |
1120 | .obj_alloc = zcache_obj_alloc, | |
1121 | .obj_free = zcache_obj_free, | |
1122 | .objnode_alloc = zcache_objnode_alloc, | |
1123 | .objnode_free = zcache_objnode_free, | |
1124 | }; | |
1125 | ||
1126 | /* | |
1127 | * zcache implementations for PAM page descriptor ops | |
1128 | */ | |
1129 | ||
1130 | static atomic_t zcache_curr_eph_pampd_count = ATOMIC_INIT(0); | |
1131 | static unsigned long zcache_curr_eph_pampd_count_max; | |
1132 | static atomic_t zcache_curr_pers_pampd_count = ATOMIC_INIT(0); | |
1133 | static unsigned long zcache_curr_pers_pampd_count_max; | |
1134 | ||
1135 | /* forward reference */ | |
1136 | static int zcache_compress(struct page *from, void **out_va, size_t *out_len); | |
1137 | ||
966b9016 DM |
1138 | static void *zcache_pampd_create(char *data, size_t size, bool raw, int eph, |
1139 | struct tmem_pool *pool, struct tmem_oid *oid, | |
1140 | uint32_t index) | |
9cc06bf8 DM |
1141 | { |
1142 | void *pampd = NULL, *cdata; | |
1143 | size_t clen; | |
1144 | int ret; | |
9cc06bf8 | 1145 | unsigned long count; |
c5f5c4db | 1146 | struct page *page = (struct page *)(data); |
966b9016 DM |
1147 | struct zcache_client *cli = pool->client; |
1148 | uint16_t client_id = get_client_id_from_client(cli); | |
1149 | unsigned long zv_mean_zsize; | |
1150 | unsigned long curr_pers_pampd_count; | |
3ca15c44 | 1151 | u64 total_zsize; |
9cc06bf8 | 1152 | |
966b9016 | 1153 | if (eph) { |
9cc06bf8 DM |
1154 | ret = zcache_compress(page, &cdata, &clen); |
1155 | if (ret == 0) | |
9cc06bf8 DM |
1156 | goto out; |
1157 | if (clen == 0 || clen > zbud_max_buddy_size()) { | |
1158 | zcache_compress_poor++; | |
1159 | goto out; | |
1160 | } | |
966b9016 DM |
1161 | pampd = (void *)zbud_create(client_id, pool->pool_id, oid, |
1162 | index, page, cdata, clen); | |
9cc06bf8 DM |
1163 | if (pampd != NULL) { |
1164 | count = atomic_inc_return(&zcache_curr_eph_pampd_count); | |
1165 | if (count > zcache_curr_eph_pampd_count_max) | |
1166 | zcache_curr_eph_pampd_count_max = count; | |
1167 | } | |
1168 | } else { | |
966b9016 DM |
1169 | curr_pers_pampd_count = |
1170 | atomic_read(&zcache_curr_pers_pampd_count); | |
1171 | if (curr_pers_pampd_count > | |
1172 | (zv_page_count_policy_percent * totalram_pages) / 100) | |
9cc06bf8 DM |
1173 | goto out; |
1174 | ret = zcache_compress(page, &cdata, &clen); | |
1175 | if (ret == 0) | |
1176 | goto out; | |
966b9016 DM |
1177 | /* reject if compression is too poor */ |
1178 | if (clen > zv_max_zsize) { | |
9cc06bf8 DM |
1179 | zcache_compress_poor++; |
1180 | goto out; | |
1181 | } | |
966b9016 DM |
1182 | /* reject if mean compression is too poor */ |
1183 | if ((clen > zv_max_mean_zsize) && (curr_pers_pampd_count > 0)) { | |
3ca15c44 TLSC |
1184 | total_zsize = xv_get_total_size_bytes(cli->xvpool); |
1185 | zv_mean_zsize = div_u64(total_zsize, | |
1186 | curr_pers_pampd_count); | |
966b9016 DM |
1187 | if (zv_mean_zsize > zv_max_mean_zsize) { |
1188 | zcache_mean_compress_poor++; | |
1189 | goto out; | |
1190 | } | |
1191 | } | |
1192 | pampd = (void *)zv_create(cli->xvpool, pool->pool_id, | |
9cc06bf8 DM |
1193 | oid, index, cdata, clen); |
1194 | if (pampd == NULL) | |
1195 | goto out; | |
1196 | count = atomic_inc_return(&zcache_curr_pers_pampd_count); | |
1197 | if (count > zcache_curr_pers_pampd_count_max) | |
1198 | zcache_curr_pers_pampd_count_max = count; | |
1199 | } | |
1200 | out: | |
1201 | return pampd; | |
1202 | } | |
1203 | ||
1204 | /* | |
1205 | * fill the pageframe corresponding to the struct page with the data | |
1206 | * from the passed pampd | |
1207 | */ | |
966b9016 DM |
1208 | static int zcache_pampd_get_data(char *data, size_t *bufsize, bool raw, |
1209 | void *pampd, struct tmem_pool *pool, | |
1210 | struct tmem_oid *oid, uint32_t index) | |
9cc06bf8 DM |
1211 | { |
1212 | int ret = 0; | |
1213 | ||
966b9016 | 1214 | BUG_ON(is_ephemeral(pool)); |
c5f5c4db | 1215 | zv_decompress((struct page *)(data), pampd); |
966b9016 DM |
1216 | return ret; |
1217 | } | |
1218 | ||
1219 | /* | |
1220 | * fill the pageframe corresponding to the struct page with the data | |
1221 | * from the passed pampd | |
1222 | */ | |
1223 | static int zcache_pampd_get_data_and_free(char *data, size_t *bufsize, bool raw, | |
1224 | void *pampd, struct tmem_pool *pool, | |
1225 | struct tmem_oid *oid, uint32_t index) | |
1226 | { | |
1227 | int ret = 0; | |
1228 | ||
1229 | BUG_ON(!is_ephemeral(pool)); | |
80976804 | 1230 | zbud_decompress((struct page *)(data), pampd); |
966b9016 DM |
1231 | zbud_free_and_delist((struct zbud_hdr *)pampd); |
1232 | atomic_dec(&zcache_curr_eph_pampd_count); | |
9cc06bf8 DM |
1233 | return ret; |
1234 | } | |
1235 | ||
1236 | /* | |
1237 | * free the pampd and remove it from any zcache lists | |
1238 | * pampd must no longer be pointed to from any tmem data structures! | |
1239 | */ | |
966b9016 DM |
1240 | static void zcache_pampd_free(void *pampd, struct tmem_pool *pool, |
1241 | struct tmem_oid *oid, uint32_t index) | |
9cc06bf8 | 1242 | { |
966b9016 DM |
1243 | struct zcache_client *cli = pool->client; |
1244 | ||
9cc06bf8 DM |
1245 | if (is_ephemeral(pool)) { |
1246 | zbud_free_and_delist((struct zbud_hdr *)pampd); | |
1247 | atomic_dec(&zcache_curr_eph_pampd_count); | |
1248 | BUG_ON(atomic_read(&zcache_curr_eph_pampd_count) < 0); | |
1249 | } else { | |
966b9016 | 1250 | zv_free(cli->xvpool, (struct zv_hdr *)pampd); |
9cc06bf8 DM |
1251 | atomic_dec(&zcache_curr_pers_pampd_count); |
1252 | BUG_ON(atomic_read(&zcache_curr_pers_pampd_count) < 0); | |
1253 | } | |
1254 | } | |
1255 | ||
966b9016 DM |
1256 | static void zcache_pampd_free_obj(struct tmem_pool *pool, struct tmem_obj *obj) |
1257 | { | |
1258 | } | |
1259 | ||
1260 | static void zcache_pampd_new_obj(struct tmem_obj *obj) | |
1261 | { | |
1262 | } | |
1263 | ||
1264 | static int zcache_pampd_replace_in_obj(void *pampd, struct tmem_obj *obj) | |
1265 | { | |
1266 | return -1; | |
1267 | } | |
1268 | ||
1269 | static bool zcache_pampd_is_remote(void *pampd) | |
1270 | { | |
1271 | return 0; | |
1272 | } | |
1273 | ||
9cc06bf8 DM |
1274 | static struct tmem_pamops zcache_pamops = { |
1275 | .create = zcache_pampd_create, | |
1276 | .get_data = zcache_pampd_get_data, | |
966b9016 | 1277 | .get_data_and_free = zcache_pampd_get_data_and_free, |
9cc06bf8 | 1278 | .free = zcache_pampd_free, |
966b9016 DM |
1279 | .free_obj = zcache_pampd_free_obj, |
1280 | .new_obj = zcache_pampd_new_obj, | |
1281 | .replace_in_obj = zcache_pampd_replace_in_obj, | |
1282 | .is_remote = zcache_pampd_is_remote, | |
9cc06bf8 DM |
1283 | }; |
1284 | ||
1285 | /* | |
1286 | * zcache compression/decompression and related per-cpu stuff | |
1287 | */ | |
1288 | ||
1289 | #define LZO_WORKMEM_BYTES LZO1X_1_MEM_COMPRESS | |
1290 | #define LZO_DSTMEM_PAGE_ORDER 1 | |
1291 | static DEFINE_PER_CPU(unsigned char *, zcache_workmem); | |
1292 | static DEFINE_PER_CPU(unsigned char *, zcache_dstmem); | |
1293 | ||
1294 | static int zcache_compress(struct page *from, void **out_va, size_t *out_len) | |
1295 | { | |
1296 | int ret = 0; | |
1297 | unsigned char *dmem = __get_cpu_var(zcache_dstmem); | |
1298 | unsigned char *wmem = __get_cpu_var(zcache_workmem); | |
1299 | char *from_va; | |
1300 | ||
1301 | BUG_ON(!irqs_disabled()); | |
1302 | if (unlikely(dmem == NULL || wmem == NULL)) | |
1303 | goto out; /* no buffer, so can't compress */ | |
1304 | from_va = kmap_atomic(from, KM_USER0); | |
1305 | mb(); | |
1306 | ret = lzo1x_1_compress(from_va, PAGE_SIZE, dmem, out_len, wmem); | |
1307 | BUG_ON(ret != LZO_E_OK); | |
1308 | *out_va = dmem; | |
1309 | kunmap_atomic(from_va, KM_USER0); | |
1310 | ret = 1; | |
1311 | out: | |
1312 | return ret; | |
1313 | } | |
1314 | ||
1315 | ||
1316 | static int zcache_cpu_notifier(struct notifier_block *nb, | |
1317 | unsigned long action, void *pcpu) | |
1318 | { | |
1319 | int cpu = (long)pcpu; | |
1320 | struct zcache_preload *kp; | |
1321 | ||
1322 | switch (action) { | |
1323 | case CPU_UP_PREPARE: | |
1324 | per_cpu(zcache_dstmem, cpu) = (void *)__get_free_pages( | |
1325 | GFP_KERNEL | __GFP_REPEAT, | |
1326 | LZO_DSTMEM_PAGE_ORDER), | |
1327 | per_cpu(zcache_workmem, cpu) = | |
1328 | kzalloc(LZO1X_MEM_COMPRESS, | |
1329 | GFP_KERNEL | __GFP_REPEAT); | |
1330 | break; | |
1331 | case CPU_DEAD: | |
1332 | case CPU_UP_CANCELED: | |
1333 | free_pages((unsigned long)per_cpu(zcache_dstmem, cpu), | |
1334 | LZO_DSTMEM_PAGE_ORDER); | |
1335 | per_cpu(zcache_dstmem, cpu) = NULL; | |
1336 | kfree(per_cpu(zcache_workmem, cpu)); | |
1337 | per_cpu(zcache_workmem, cpu) = NULL; | |
1338 | kp = &per_cpu(zcache_preloads, cpu); | |
1339 | while (kp->nr) { | |
1340 | kmem_cache_free(zcache_objnode_cache, | |
1341 | kp->objnodes[kp->nr - 1]); | |
1342 | kp->objnodes[kp->nr - 1] = NULL; | |
1343 | kp->nr--; | |
1344 | } | |
8550be08 SJ |
1345 | if (kp->obj) { |
1346 | kmem_cache_free(zcache_obj_cache, kp->obj); | |
1347 | kp->obj = NULL; | |
1348 | } | |
1349 | if (kp->page) { | |
1350 | free_page((unsigned long)kp->page); | |
1351 | kp->page = NULL; | |
1352 | } | |
9cc06bf8 DM |
1353 | break; |
1354 | default: | |
1355 | break; | |
1356 | } | |
1357 | return NOTIFY_OK; | |
1358 | } | |
1359 | ||
1360 | static struct notifier_block zcache_cpu_notifier_block = { | |
1361 | .notifier_call = zcache_cpu_notifier | |
1362 | }; | |
1363 | ||
1364 | #ifdef CONFIG_SYSFS | |
1365 | #define ZCACHE_SYSFS_RO(_name) \ | |
1366 | static ssize_t zcache_##_name##_show(struct kobject *kobj, \ | |
1367 | struct kobj_attribute *attr, char *buf) \ | |
1368 | { \ | |
1369 | return sprintf(buf, "%lu\n", zcache_##_name); \ | |
1370 | } \ | |
1371 | static struct kobj_attribute zcache_##_name##_attr = { \ | |
1372 | .attr = { .name = __stringify(_name), .mode = 0444 }, \ | |
1373 | .show = zcache_##_name##_show, \ | |
1374 | } | |
1375 | ||
1376 | #define ZCACHE_SYSFS_RO_ATOMIC(_name) \ | |
1377 | static ssize_t zcache_##_name##_show(struct kobject *kobj, \ | |
1378 | struct kobj_attribute *attr, char *buf) \ | |
1379 | { \ | |
1380 | return sprintf(buf, "%d\n", atomic_read(&zcache_##_name)); \ | |
1381 | } \ | |
1382 | static struct kobj_attribute zcache_##_name##_attr = { \ | |
1383 | .attr = { .name = __stringify(_name), .mode = 0444 }, \ | |
1384 | .show = zcache_##_name##_show, \ | |
1385 | } | |
1386 | ||
1387 | #define ZCACHE_SYSFS_RO_CUSTOM(_name, _func) \ | |
1388 | static ssize_t zcache_##_name##_show(struct kobject *kobj, \ | |
1389 | struct kobj_attribute *attr, char *buf) \ | |
1390 | { \ | |
1391 | return _func(buf); \ | |
1392 | } \ | |
1393 | static struct kobj_attribute zcache_##_name##_attr = { \ | |
1394 | .attr = { .name = __stringify(_name), .mode = 0444 }, \ | |
1395 | .show = zcache_##_name##_show, \ | |
1396 | } | |
1397 | ||
1398 | ZCACHE_SYSFS_RO(curr_obj_count_max); | |
1399 | ZCACHE_SYSFS_RO(curr_objnode_count_max); | |
1400 | ZCACHE_SYSFS_RO(flush_total); | |
1401 | ZCACHE_SYSFS_RO(flush_found); | |
1402 | ZCACHE_SYSFS_RO(flobj_total); | |
1403 | ZCACHE_SYSFS_RO(flobj_found); | |
1404 | ZCACHE_SYSFS_RO(failed_eph_puts); | |
1405 | ZCACHE_SYSFS_RO(failed_pers_puts); | |
1406 | ZCACHE_SYSFS_RO(zbud_curr_zbytes); | |
1407 | ZCACHE_SYSFS_RO(zbud_cumul_zpages); | |
1408 | ZCACHE_SYSFS_RO(zbud_cumul_zbytes); | |
1409 | ZCACHE_SYSFS_RO(zbud_buddied_count); | |
1410 | ZCACHE_SYSFS_RO(zbpg_unused_list_count); | |
1411 | ZCACHE_SYSFS_RO(evicted_raw_pages); | |
1412 | ZCACHE_SYSFS_RO(evicted_unbuddied_pages); | |
1413 | ZCACHE_SYSFS_RO(evicted_buddied_pages); | |
1414 | ZCACHE_SYSFS_RO(failed_get_free_pages); | |
1415 | ZCACHE_SYSFS_RO(failed_alloc); | |
1416 | ZCACHE_SYSFS_RO(put_to_flush); | |
9cc06bf8 | 1417 | ZCACHE_SYSFS_RO(compress_poor); |
966b9016 | 1418 | ZCACHE_SYSFS_RO(mean_compress_poor); |
9cc06bf8 DM |
1419 | ZCACHE_SYSFS_RO_ATOMIC(zbud_curr_raw_pages); |
1420 | ZCACHE_SYSFS_RO_ATOMIC(zbud_curr_zpages); | |
1421 | ZCACHE_SYSFS_RO_ATOMIC(curr_obj_count); | |
1422 | ZCACHE_SYSFS_RO_ATOMIC(curr_objnode_count); | |
1423 | ZCACHE_SYSFS_RO_CUSTOM(zbud_unbuddied_list_counts, | |
1424 | zbud_show_unbuddied_list_counts); | |
1425 | ZCACHE_SYSFS_RO_CUSTOM(zbud_cumul_chunk_counts, | |
1426 | zbud_show_cumul_chunk_counts); | |
966b9016 DM |
1427 | ZCACHE_SYSFS_RO_CUSTOM(zv_curr_dist_counts, |
1428 | zv_curr_dist_counts_show); | |
1429 | ZCACHE_SYSFS_RO_CUSTOM(zv_cumul_dist_counts, | |
1430 | zv_cumul_dist_counts_show); | |
9cc06bf8 DM |
1431 | |
1432 | static struct attribute *zcache_attrs[] = { | |
1433 | &zcache_curr_obj_count_attr.attr, | |
1434 | &zcache_curr_obj_count_max_attr.attr, | |
1435 | &zcache_curr_objnode_count_attr.attr, | |
1436 | &zcache_curr_objnode_count_max_attr.attr, | |
1437 | &zcache_flush_total_attr.attr, | |
1438 | &zcache_flobj_total_attr.attr, | |
1439 | &zcache_flush_found_attr.attr, | |
1440 | &zcache_flobj_found_attr.attr, | |
1441 | &zcache_failed_eph_puts_attr.attr, | |
1442 | &zcache_failed_pers_puts_attr.attr, | |
1443 | &zcache_compress_poor_attr.attr, | |
966b9016 | 1444 | &zcache_mean_compress_poor_attr.attr, |
9cc06bf8 DM |
1445 | &zcache_zbud_curr_raw_pages_attr.attr, |
1446 | &zcache_zbud_curr_zpages_attr.attr, | |
1447 | &zcache_zbud_curr_zbytes_attr.attr, | |
1448 | &zcache_zbud_cumul_zpages_attr.attr, | |
1449 | &zcache_zbud_cumul_zbytes_attr.attr, | |
1450 | &zcache_zbud_buddied_count_attr.attr, | |
1451 | &zcache_zbpg_unused_list_count_attr.attr, | |
1452 | &zcache_evicted_raw_pages_attr.attr, | |
1453 | &zcache_evicted_unbuddied_pages_attr.attr, | |
1454 | &zcache_evicted_buddied_pages_attr.attr, | |
1455 | &zcache_failed_get_free_pages_attr.attr, | |
1456 | &zcache_failed_alloc_attr.attr, | |
1457 | &zcache_put_to_flush_attr.attr, | |
9cc06bf8 DM |
1458 | &zcache_zbud_unbuddied_list_counts_attr.attr, |
1459 | &zcache_zbud_cumul_chunk_counts_attr.attr, | |
966b9016 DM |
1460 | &zcache_zv_curr_dist_counts_attr.attr, |
1461 | &zcache_zv_cumul_dist_counts_attr.attr, | |
1462 | &zcache_zv_max_zsize_attr.attr, | |
1463 | &zcache_zv_max_mean_zsize_attr.attr, | |
1464 | &zcache_zv_page_count_policy_percent_attr.attr, | |
9cc06bf8 DM |
1465 | NULL, |
1466 | }; | |
1467 | ||
1468 | static struct attribute_group zcache_attr_group = { | |
1469 | .attrs = zcache_attrs, | |
1470 | .name = "zcache", | |
1471 | }; | |
1472 | ||
1473 | #endif /* CONFIG_SYSFS */ | |
1474 | /* | |
1475 | * When zcache is disabled ("frozen"), pools can be created and destroyed, | |
1476 | * but all puts (and thus all other operations that require memory allocation) | |
1477 | * must fail. If zcache is unfrozen, accepts puts, then frozen again, | |
1478 | * data consistency requires all puts while frozen to be converted into | |
1479 | * flushes. | |
1480 | */ | |
1481 | static bool zcache_freeze; | |
1482 | ||
1483 | /* | |
1484 | * zcache shrinker interface (only useful for ephemeral pages, so zbud only) | |
1485 | */ | |
1495f230 YH |
1486 | static int shrink_zcache_memory(struct shrinker *shrink, |
1487 | struct shrink_control *sc) | |
9cc06bf8 DM |
1488 | { |
1489 | int ret = -1; | |
1495f230 YH |
1490 | int nr = sc->nr_to_scan; |
1491 | gfp_t gfp_mask = sc->gfp_mask; | |
9cc06bf8 DM |
1492 | |
1493 | if (nr >= 0) { | |
1494 | if (!(gfp_mask & __GFP_FS)) | |
1495 | /* does this case really need to be skipped? */ | |
1496 | goto out; | |
00bf2560 | 1497 | zbud_evict_pages(nr); |
9cc06bf8 DM |
1498 | } |
1499 | ret = (int)atomic_read(&zcache_zbud_curr_raw_pages); | |
1500 | out: | |
1501 | return ret; | |
1502 | } | |
1503 | ||
1504 | static struct shrinker zcache_shrinker = { | |
1505 | .shrink = shrink_zcache_memory, | |
1506 | .seeks = DEFAULT_SEEKS, | |
1507 | }; | |
1508 | ||
1509 | /* | |
1510 | * zcache shims between cleancache/frontswap ops and tmem | |
1511 | */ | |
1512 | ||
966b9016 | 1513 | static int zcache_put_page(int cli_id, int pool_id, struct tmem_oid *oidp, |
9cc06bf8 DM |
1514 | uint32_t index, struct page *page) |
1515 | { | |
1516 | struct tmem_pool *pool; | |
1517 | int ret = -1; | |
1518 | ||
1519 | BUG_ON(!irqs_disabled()); | |
966b9016 | 1520 | pool = zcache_get_pool_by_id(cli_id, pool_id); |
9cc06bf8 DM |
1521 | if (unlikely(pool == NULL)) |
1522 | goto out; | |
1523 | if (!zcache_freeze && zcache_do_preload(pool) == 0) { | |
1524 | /* preload does preempt_disable on success */ | |
c5f5c4db | 1525 | ret = tmem_put(pool, oidp, index, (char *)(page), |
966b9016 | 1526 | PAGE_SIZE, 0, is_ephemeral(pool)); |
9cc06bf8 DM |
1527 | if (ret < 0) { |
1528 | if (is_ephemeral(pool)) | |
1529 | zcache_failed_eph_puts++; | |
1530 | else | |
1531 | zcache_failed_pers_puts++; | |
1532 | } | |
1533 | zcache_put_pool(pool); | |
1534 | preempt_enable_no_resched(); | |
1535 | } else { | |
1536 | zcache_put_to_flush++; | |
1537 | if (atomic_read(&pool->obj_count) > 0) | |
1538 | /* the put fails whether the flush succeeds or not */ | |
1539 | (void)tmem_flush_page(pool, oidp, index); | |
1540 | zcache_put_pool(pool); | |
1541 | } | |
1542 | out: | |
1543 | return ret; | |
1544 | } | |
1545 | ||
966b9016 | 1546 | static int zcache_get_page(int cli_id, int pool_id, struct tmem_oid *oidp, |
9cc06bf8 DM |
1547 | uint32_t index, struct page *page) |
1548 | { | |
1549 | struct tmem_pool *pool; | |
1550 | int ret = -1; | |
1551 | unsigned long flags; | |
966b9016 | 1552 | size_t size = PAGE_SIZE; |
9cc06bf8 DM |
1553 | |
1554 | local_irq_save(flags); | |
966b9016 | 1555 | pool = zcache_get_pool_by_id(cli_id, pool_id); |
9cc06bf8 DM |
1556 | if (likely(pool != NULL)) { |
1557 | if (atomic_read(&pool->obj_count) > 0) | |
c5f5c4db | 1558 | ret = tmem_get(pool, oidp, index, (char *)(page), |
966b9016 | 1559 | &size, 0, is_ephemeral(pool)); |
9cc06bf8 DM |
1560 | zcache_put_pool(pool); |
1561 | } | |
1562 | local_irq_restore(flags); | |
1563 | return ret; | |
1564 | } | |
1565 | ||
966b9016 DM |
1566 | static int zcache_flush_page(int cli_id, int pool_id, |
1567 | struct tmem_oid *oidp, uint32_t index) | |
9cc06bf8 DM |
1568 | { |
1569 | struct tmem_pool *pool; | |
1570 | int ret = -1; | |
1571 | unsigned long flags; | |
1572 | ||
1573 | local_irq_save(flags); | |
1574 | zcache_flush_total++; | |
966b9016 | 1575 | pool = zcache_get_pool_by_id(cli_id, pool_id); |
9cc06bf8 DM |
1576 | if (likely(pool != NULL)) { |
1577 | if (atomic_read(&pool->obj_count) > 0) | |
1578 | ret = tmem_flush_page(pool, oidp, index); | |
1579 | zcache_put_pool(pool); | |
1580 | } | |
1581 | if (ret >= 0) | |
1582 | zcache_flush_found++; | |
1583 | local_irq_restore(flags); | |
1584 | return ret; | |
1585 | } | |
1586 | ||
966b9016 DM |
1587 | static int zcache_flush_object(int cli_id, int pool_id, |
1588 | struct tmem_oid *oidp) | |
9cc06bf8 DM |
1589 | { |
1590 | struct tmem_pool *pool; | |
1591 | int ret = -1; | |
1592 | unsigned long flags; | |
1593 | ||
1594 | local_irq_save(flags); | |
1595 | zcache_flobj_total++; | |
966b9016 | 1596 | pool = zcache_get_pool_by_id(cli_id, pool_id); |
9cc06bf8 DM |
1597 | if (likely(pool != NULL)) { |
1598 | if (atomic_read(&pool->obj_count) > 0) | |
1599 | ret = tmem_flush_object(pool, oidp); | |
1600 | zcache_put_pool(pool); | |
1601 | } | |
1602 | if (ret >= 0) | |
1603 | zcache_flobj_found++; | |
1604 | local_irq_restore(flags); | |
1605 | return ret; | |
1606 | } | |
1607 | ||
966b9016 | 1608 | static int zcache_destroy_pool(int cli_id, int pool_id) |
9cc06bf8 DM |
1609 | { |
1610 | struct tmem_pool *pool = NULL; | |
966b9016 | 1611 | struct zcache_client *cli = NULL; |
9cc06bf8 DM |
1612 | int ret = -1; |
1613 | ||
1614 | if (pool_id < 0) | |
1615 | goto out; | |
966b9016 DM |
1616 | if (cli_id == LOCAL_CLIENT) |
1617 | cli = &zcache_host; | |
1618 | else if ((unsigned int)cli_id < MAX_CLIENTS) | |
1619 | cli = &zcache_clients[cli_id]; | |
1620 | if (cli == NULL) | |
1621 | goto out; | |
1622 | atomic_inc(&cli->refcount); | |
1623 | pool = cli->tmem_pools[pool_id]; | |
9cc06bf8 DM |
1624 | if (pool == NULL) |
1625 | goto out; | |
966b9016 | 1626 | cli->tmem_pools[pool_id] = NULL; |
9cc06bf8 DM |
1627 | /* wait for pool activity on other cpus to quiesce */ |
1628 | while (atomic_read(&pool->refcount) != 0) | |
1629 | ; | |
966b9016 | 1630 | atomic_dec(&cli->refcount); |
9cc06bf8 DM |
1631 | local_bh_disable(); |
1632 | ret = tmem_destroy_pool(pool); | |
1633 | local_bh_enable(); | |
1634 | kfree(pool); | |
966b9016 DM |
1635 | pr_info("zcache: destroyed pool id=%d, cli_id=%d\n", |
1636 | pool_id, cli_id); | |
9cc06bf8 DM |
1637 | out: |
1638 | return ret; | |
1639 | } | |
1640 | ||
966b9016 | 1641 | static int zcache_new_pool(uint16_t cli_id, uint32_t flags) |
9cc06bf8 DM |
1642 | { |
1643 | int poolid = -1; | |
1644 | struct tmem_pool *pool; | |
966b9016 | 1645 | struct zcache_client *cli = NULL; |
9cc06bf8 | 1646 | |
966b9016 DM |
1647 | if (cli_id == LOCAL_CLIENT) |
1648 | cli = &zcache_host; | |
1649 | else if ((unsigned int)cli_id < MAX_CLIENTS) | |
1650 | cli = &zcache_clients[cli_id]; | |
1651 | if (cli == NULL) | |
1652 | goto out; | |
1653 | atomic_inc(&cli->refcount); | |
dbe82eb1 | 1654 | pool = kmalloc(sizeof(struct tmem_pool), GFP_ATOMIC); |
9cc06bf8 DM |
1655 | if (pool == NULL) { |
1656 | pr_info("zcache: pool creation failed: out of memory\n"); | |
1657 | goto out; | |
1658 | } | |
1659 | ||
1660 | for (poolid = 0; poolid < MAX_POOLS_PER_CLIENT; poolid++) | |
966b9016 | 1661 | if (cli->tmem_pools[poolid] == NULL) |
9cc06bf8 DM |
1662 | break; |
1663 | if (poolid >= MAX_POOLS_PER_CLIENT) { | |
1664 | pr_info("zcache: pool creation failed: max exceeded\n"); | |
1665 | kfree(pool); | |
1666 | poolid = -1; | |
1667 | goto out; | |
1668 | } | |
1669 | atomic_set(&pool->refcount, 0); | |
966b9016 | 1670 | pool->client = cli; |
9cc06bf8 DM |
1671 | pool->pool_id = poolid; |
1672 | tmem_new_pool(pool, flags); | |
966b9016 DM |
1673 | cli->tmem_pools[poolid] = pool; |
1674 | pr_info("zcache: created %s tmem pool, id=%d, client=%d\n", | |
9cc06bf8 | 1675 | flags & TMEM_POOL_PERSIST ? "persistent" : "ephemeral", |
966b9016 | 1676 | poolid, cli_id); |
9cc06bf8 | 1677 | out: |
966b9016 DM |
1678 | if (cli != NULL) |
1679 | atomic_dec(&cli->refcount); | |
9cc06bf8 DM |
1680 | return poolid; |
1681 | } | |
1682 | ||
1683 | /********** | |
1684 | * Two kernel functionalities currently can be layered on top of tmem. | |
1685 | * These are "cleancache" which is used as a second-chance cache for clean | |
1686 | * page cache pages; and "frontswap" which is used for swap pages | |
1687 | * to avoid writes to disk. A generic "shim" is provided here for each | |
1688 | * to translate in-kernel semantics to zcache semantics. | |
1689 | */ | |
1690 | ||
1691 | #ifdef CONFIG_CLEANCACHE | |
1692 | static void zcache_cleancache_put_page(int pool_id, | |
1693 | struct cleancache_filekey key, | |
1694 | pgoff_t index, struct page *page) | |
1695 | { | |
1696 | u32 ind = (u32) index; | |
1697 | struct tmem_oid oid = *(struct tmem_oid *)&key; | |
1698 | ||
1699 | if (likely(ind == index)) | |
966b9016 | 1700 | (void)zcache_put_page(LOCAL_CLIENT, pool_id, &oid, index, page); |
9cc06bf8 DM |
1701 | } |
1702 | ||
1703 | static int zcache_cleancache_get_page(int pool_id, | |
1704 | struct cleancache_filekey key, | |
1705 | pgoff_t index, struct page *page) | |
1706 | { | |
1707 | u32 ind = (u32) index; | |
1708 | struct tmem_oid oid = *(struct tmem_oid *)&key; | |
1709 | int ret = -1; | |
1710 | ||
1711 | if (likely(ind == index)) | |
966b9016 | 1712 | ret = zcache_get_page(LOCAL_CLIENT, pool_id, &oid, index, page); |
9cc06bf8 DM |
1713 | return ret; |
1714 | } | |
1715 | ||
1716 | static void zcache_cleancache_flush_page(int pool_id, | |
1717 | struct cleancache_filekey key, | |
1718 | pgoff_t index) | |
1719 | { | |
1720 | u32 ind = (u32) index; | |
1721 | struct tmem_oid oid = *(struct tmem_oid *)&key; | |
1722 | ||
1723 | if (likely(ind == index)) | |
966b9016 | 1724 | (void)zcache_flush_page(LOCAL_CLIENT, pool_id, &oid, ind); |
9cc06bf8 DM |
1725 | } |
1726 | ||
1727 | static void zcache_cleancache_flush_inode(int pool_id, | |
1728 | struct cleancache_filekey key) | |
1729 | { | |
1730 | struct tmem_oid oid = *(struct tmem_oid *)&key; | |
1731 | ||
966b9016 | 1732 | (void)zcache_flush_object(LOCAL_CLIENT, pool_id, &oid); |
9cc06bf8 DM |
1733 | } |
1734 | ||
1735 | static void zcache_cleancache_flush_fs(int pool_id) | |
1736 | { | |
1737 | if (pool_id >= 0) | |
966b9016 | 1738 | (void)zcache_destroy_pool(LOCAL_CLIENT, pool_id); |
9cc06bf8 DM |
1739 | } |
1740 | ||
1741 | static int zcache_cleancache_init_fs(size_t pagesize) | |
1742 | { | |
1743 | BUG_ON(sizeof(struct cleancache_filekey) != | |
1744 | sizeof(struct tmem_oid)); | |
1745 | BUG_ON(pagesize != PAGE_SIZE); | |
966b9016 | 1746 | return zcache_new_pool(LOCAL_CLIENT, 0); |
9cc06bf8 DM |
1747 | } |
1748 | ||
1749 | static int zcache_cleancache_init_shared_fs(char *uuid, size_t pagesize) | |
1750 | { | |
1751 | /* shared pools are unsupported and map to private */ | |
1752 | BUG_ON(sizeof(struct cleancache_filekey) != | |
1753 | sizeof(struct tmem_oid)); | |
1754 | BUG_ON(pagesize != PAGE_SIZE); | |
966b9016 | 1755 | return zcache_new_pool(LOCAL_CLIENT, 0); |
9cc06bf8 DM |
1756 | } |
1757 | ||
1758 | static struct cleancache_ops zcache_cleancache_ops = { | |
1759 | .put_page = zcache_cleancache_put_page, | |
1760 | .get_page = zcache_cleancache_get_page, | |
1761 | .flush_page = zcache_cleancache_flush_page, | |
1762 | .flush_inode = zcache_cleancache_flush_inode, | |
1763 | .flush_fs = zcache_cleancache_flush_fs, | |
1764 | .init_shared_fs = zcache_cleancache_init_shared_fs, | |
1765 | .init_fs = zcache_cleancache_init_fs | |
1766 | }; | |
1767 | ||
1768 | struct cleancache_ops zcache_cleancache_register_ops(void) | |
1769 | { | |
1770 | struct cleancache_ops old_ops = | |
1771 | cleancache_register_ops(&zcache_cleancache_ops); | |
1772 | ||
1773 | return old_ops; | |
1774 | } | |
1775 | #endif | |
1776 | ||
1777 | #ifdef CONFIG_FRONTSWAP | |
1778 | /* a single tmem poolid is used for all frontswap "types" (swapfiles) */ | |
1779 | static int zcache_frontswap_poolid = -1; | |
1780 | ||
1781 | /* | |
1782 | * Swizzling increases objects per swaptype, increasing tmem concurrency | |
1783 | * for heavy swaploads. Later, larger nr_cpus -> larger SWIZ_BITS | |
3d65c85f SJ |
1784 | * Setting SWIZ_BITS to 27 basically reconstructs the swap entry from |
1785 | * frontswap_get_page() | |
9cc06bf8 | 1786 | */ |
3d65c85f | 1787 | #define SWIZ_BITS 27 |
9cc06bf8 DM |
1788 | #define SWIZ_MASK ((1 << SWIZ_BITS) - 1) |
1789 | #define _oswiz(_type, _ind) ((_type << SWIZ_BITS) | (_ind & SWIZ_MASK)) | |
1790 | #define iswiz(_ind) (_ind >> SWIZ_BITS) | |
1791 | ||
1792 | static inline struct tmem_oid oswiz(unsigned type, u32 ind) | |
1793 | { | |
1794 | struct tmem_oid oid = { .oid = { 0 } }; | |
1795 | oid.oid[0] = _oswiz(type, ind); | |
1796 | return oid; | |
1797 | } | |
1798 | ||
1799 | static int zcache_frontswap_put_page(unsigned type, pgoff_t offset, | |
1800 | struct page *page) | |
1801 | { | |
1802 | u64 ind64 = (u64)offset; | |
1803 | u32 ind = (u32)offset; | |
1804 | struct tmem_oid oid = oswiz(type, ind); | |
1805 | int ret = -1; | |
1806 | unsigned long flags; | |
1807 | ||
1808 | BUG_ON(!PageLocked(page)); | |
1809 | if (likely(ind64 == ind)) { | |
1810 | local_irq_save(flags); | |
966b9016 DM |
1811 | ret = zcache_put_page(LOCAL_CLIENT, zcache_frontswap_poolid, |
1812 | &oid, iswiz(ind), page); | |
9cc06bf8 DM |
1813 | local_irq_restore(flags); |
1814 | } | |
1815 | return ret; | |
1816 | } | |
1817 | ||
1818 | /* returns 0 if the page was successfully gotten from frontswap, -1 if | |
1819 | * was not present (should never happen!) */ | |
1820 | static int zcache_frontswap_get_page(unsigned type, pgoff_t offset, | |
1821 | struct page *page) | |
1822 | { | |
1823 | u64 ind64 = (u64)offset; | |
1824 | u32 ind = (u32)offset; | |
1825 | struct tmem_oid oid = oswiz(type, ind); | |
1826 | int ret = -1; | |
1827 | ||
1828 | BUG_ON(!PageLocked(page)); | |
1829 | if (likely(ind64 == ind)) | |
966b9016 DM |
1830 | ret = zcache_get_page(LOCAL_CLIENT, zcache_frontswap_poolid, |
1831 | &oid, iswiz(ind), page); | |
9cc06bf8 DM |
1832 | return ret; |
1833 | } | |
1834 | ||
1835 | /* flush a single page from frontswap */ | |
1836 | static void zcache_frontswap_flush_page(unsigned type, pgoff_t offset) | |
1837 | { | |
1838 | u64 ind64 = (u64)offset; | |
1839 | u32 ind = (u32)offset; | |
1840 | struct tmem_oid oid = oswiz(type, ind); | |
1841 | ||
1842 | if (likely(ind64 == ind)) | |
966b9016 DM |
1843 | (void)zcache_flush_page(LOCAL_CLIENT, zcache_frontswap_poolid, |
1844 | &oid, iswiz(ind)); | |
9cc06bf8 DM |
1845 | } |
1846 | ||
1847 | /* flush all pages from the passed swaptype */ | |
1848 | static void zcache_frontswap_flush_area(unsigned type) | |
1849 | { | |
1850 | struct tmem_oid oid; | |
1851 | int ind; | |
1852 | ||
1853 | for (ind = SWIZ_MASK; ind >= 0; ind--) { | |
1854 | oid = oswiz(type, ind); | |
966b9016 DM |
1855 | (void)zcache_flush_object(LOCAL_CLIENT, |
1856 | zcache_frontswap_poolid, &oid); | |
9cc06bf8 DM |
1857 | } |
1858 | } | |
1859 | ||
1860 | static void zcache_frontswap_init(unsigned ignored) | |
1861 | { | |
1862 | /* a single tmem poolid is used for all frontswap "types" (swapfiles) */ | |
1863 | if (zcache_frontswap_poolid < 0) | |
966b9016 DM |
1864 | zcache_frontswap_poolid = |
1865 | zcache_new_pool(LOCAL_CLIENT, TMEM_POOL_PERSIST); | |
9cc06bf8 DM |
1866 | } |
1867 | ||
1868 | static struct frontswap_ops zcache_frontswap_ops = { | |
1869 | .put_page = zcache_frontswap_put_page, | |
1870 | .get_page = zcache_frontswap_get_page, | |
1871 | .flush_page = zcache_frontswap_flush_page, | |
1872 | .flush_area = zcache_frontswap_flush_area, | |
1873 | .init = zcache_frontswap_init | |
1874 | }; | |
1875 | ||
1876 | struct frontswap_ops zcache_frontswap_register_ops(void) | |
1877 | { | |
1878 | struct frontswap_ops old_ops = | |
1879 | frontswap_register_ops(&zcache_frontswap_ops); | |
1880 | ||
1881 | return old_ops; | |
1882 | } | |
1883 | #endif | |
1884 | ||
1885 | /* | |
1886 | * zcache initialization | |
1887 | * NOTE FOR NOW zcache MUST BE PROVIDED AS A KERNEL BOOT PARAMETER OR | |
1888 | * NOTHING HAPPENS! | |
1889 | */ | |
1890 | ||
1891 | static int zcache_enabled; | |
1892 | ||
1893 | static int __init enable_zcache(char *s) | |
1894 | { | |
1895 | zcache_enabled = 1; | |
1896 | return 1; | |
1897 | } | |
1898 | __setup("zcache", enable_zcache); | |
1899 | ||
1900 | /* allow independent dynamic disabling of cleancache and frontswap */ | |
1901 | ||
1902 | static int use_cleancache = 1; | |
1903 | ||
1904 | static int __init no_cleancache(char *s) | |
1905 | { | |
1906 | use_cleancache = 0; | |
1907 | return 1; | |
1908 | } | |
1909 | ||
1910 | __setup("nocleancache", no_cleancache); | |
1911 | ||
1912 | static int use_frontswap = 1; | |
1913 | ||
1914 | static int __init no_frontswap(char *s) | |
1915 | { | |
1916 | use_frontswap = 0; | |
1917 | return 1; | |
1918 | } | |
1919 | ||
1920 | __setup("nofrontswap", no_frontswap); | |
1921 | ||
1922 | static int __init zcache_init(void) | |
1923 | { | |
9cc06bf8 DM |
1924 | int ret = 0; |
1925 | ||
d8c778fd | 1926 | #ifdef CONFIG_SYSFS |
9cc06bf8 DM |
1927 | ret = sysfs_create_group(mm_kobj, &zcache_attr_group); |
1928 | if (ret) { | |
1929 | pr_err("zcache: can't create sysfs\n"); | |
1930 | goto out; | |
1931 | } | |
1932 | #endif /* CONFIG_SYSFS */ | |
1933 | #if defined(CONFIG_CLEANCACHE) || defined(CONFIG_FRONTSWAP) | |
1934 | if (zcache_enabled) { | |
1935 | unsigned int cpu; | |
1936 | ||
1937 | tmem_register_hostops(&zcache_hostops); | |
1938 | tmem_register_pamops(&zcache_pamops); | |
1939 | ret = register_cpu_notifier(&zcache_cpu_notifier_block); | |
1940 | if (ret) { | |
1941 | pr_err("zcache: can't register cpu notifier\n"); | |
1942 | goto out; | |
1943 | } | |
1944 | for_each_online_cpu(cpu) { | |
1945 | void *pcpu = (void *)(long)cpu; | |
1946 | zcache_cpu_notifier(&zcache_cpu_notifier_block, | |
1947 | CPU_UP_PREPARE, pcpu); | |
1948 | } | |
1949 | } | |
1950 | zcache_objnode_cache = kmem_cache_create("zcache_objnode", | |
1951 | sizeof(struct tmem_objnode), 0, 0, NULL); | |
1952 | zcache_obj_cache = kmem_cache_create("zcache_obj", | |
1953 | sizeof(struct tmem_obj), 0, 0, NULL); | |
966b9016 DM |
1954 | ret = zcache_new_client(LOCAL_CLIENT); |
1955 | if (ret) { | |
1956 | pr_err("zcache: can't create client\n"); | |
1957 | goto out; | |
1958 | } | |
9cc06bf8 DM |
1959 | #endif |
1960 | #ifdef CONFIG_CLEANCACHE | |
1961 | if (zcache_enabled && use_cleancache) { | |
1962 | struct cleancache_ops old_ops; | |
1963 | ||
1964 | zbud_init(); | |
1965 | register_shrinker(&zcache_shrinker); | |
1966 | old_ops = zcache_cleancache_register_ops(); | |
1967 | pr_info("zcache: cleancache enabled using kernel " | |
1968 | "transcendent memory and compression buddies\n"); | |
1969 | if (old_ops.init_fs != NULL) | |
1970 | pr_warning("zcache: cleancache_ops overridden"); | |
1971 | } | |
1972 | #endif | |
1973 | #ifdef CONFIG_FRONTSWAP | |
1974 | if (zcache_enabled && use_frontswap) { | |
1975 | struct frontswap_ops old_ops; | |
1976 | ||
9cc06bf8 DM |
1977 | old_ops = zcache_frontswap_register_ops(); |
1978 | pr_info("zcache: frontswap enabled using kernel " | |
1979 | "transcendent memory and xvmalloc\n"); | |
1980 | if (old_ops.init != NULL) | |
0428fec3 | 1981 | pr_warning("zcache: frontswap_ops overridden"); |
9cc06bf8 DM |
1982 | } |
1983 | #endif | |
1984 | out: | |
1985 | return ret; | |
1986 | } | |
1987 | ||
1988 | module_init(zcache_init) |