Root/drivers/staging/zcache/zcache-main.c

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 the crypto compression
10 * API:
11 * 1) "compression buddies" ("zbud") is used for ephemeral pages
12 * 2) zsmalloc is used for persistent pages.
13 * Xvmalloc (based on the TLSF allocator) has very low fragmentation
14 * so maximizes space efficiency, while zbud allows pairs (and potentially,
15 * in the future, more than a pair of) compressed pages to be closely linked
16 * so that reclaiming can be done via the kernel's physical-page-oriented
17 * "shrinker" interface.
18 *
19 * [1] For a definition of page-accessible memory (aka PAM), see:
20 * http://marc.info/?l=linux-mm&m=127811271605009
21 */
22
23#include <linux/module.h>
24#include <linux/cpu.h>
25#include <linux/highmem.h>
26#include <linux/list.h>
27#include <linux/slab.h>
28#include <linux/spinlock.h>
29#include <linux/types.h>
30#include <linux/atomic.h>
31#include <linux/math64.h>
32#include <linux/crypto.h>
33#include <linux/string.h>
34#include <linux/idr.h>
35#include "tmem.h"
36
37#include "../zsmalloc/zsmalloc.h"
38
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
54#define MAX_CLIENTS 16
55#define LOCAL_CLIENT ((uint16_t)-1)
56
57MODULE_LICENSE("GPL");
58
59struct zcache_client {
60    struct idr tmem_pools;
61    struct zs_pool *zspool;
62    bool allocated;
63    atomic_t refcount;
64};
65
66static struct zcache_client zcache_host;
67static struct zcache_client zcache_clients[MAX_CLIENTS];
68
69static inline uint16_t get_client_id_from_client(struct zcache_client *cli)
70{
71    BUG_ON(cli == NULL);
72    if (cli == &zcache_host)
73        return LOCAL_CLIENT;
74    return cli - &zcache_clients[0];
75}
76
77static struct zcache_client *get_zcache_client(uint16_t cli_id)
78{
79    if (cli_id == LOCAL_CLIENT)
80        return &zcache_host;
81
82    if ((unsigned int)cli_id < MAX_CLIENTS)
83        return &zcache_clients[cli_id];
84
85    return NULL;
86}
87
88static inline bool is_local_client(struct zcache_client *cli)
89{
90    return cli == &zcache_host;
91}
92
93/* crypto API for zcache */
94#define ZCACHE_COMP_NAME_SZ CRYPTO_MAX_ALG_NAME
95static char zcache_comp_name[ZCACHE_COMP_NAME_SZ];
96static struct crypto_comp * __percpu *zcache_comp_pcpu_tfms;
97
98enum comp_op {
99    ZCACHE_COMPOP_COMPRESS,
100    ZCACHE_COMPOP_DECOMPRESS
101};
102
103static inline int zcache_comp_op(enum comp_op op,
104                const u8 *src, unsigned int slen,
105                u8 *dst, unsigned int *dlen)
106{
107    struct crypto_comp *tfm;
108    int ret;
109
110    BUG_ON(!zcache_comp_pcpu_tfms);
111    tfm = *per_cpu_ptr(zcache_comp_pcpu_tfms, get_cpu());
112    BUG_ON(!tfm);
113    switch (op) {
114    case ZCACHE_COMPOP_COMPRESS:
115        ret = crypto_comp_compress(tfm, src, slen, dst, dlen);
116        break;
117    case ZCACHE_COMPOP_DECOMPRESS:
118        ret = crypto_comp_decompress(tfm, src, slen, dst, dlen);
119        break;
120    default:
121        ret = -EINVAL;
122    }
123    put_cpu();
124    return ret;
125}
126
127/**********
128 * Compression buddies ("zbud") provides for packing two (or, possibly
129 * in the future, more) compressed ephemeral pages into a single "raw"
130 * (physical) page and tracking them with data structures so that
131 * the raw pages can be easily reclaimed.
132 *
133 * A zbud page ("zbpg") is an aligned page containing a list_head,
134 * a lock, and two "zbud headers". The remainder of the physical
135 * page is divided up into aligned 64-byte "chunks" which contain
136 * the compressed data for zero, one, or two zbuds. Each zbpg
137 * resides on: (1) an "unused list" if it has no zbuds; (2) a
138 * "buddied" list if it is fully populated with two zbuds; or
139 * (3) one of PAGE_SIZE/64 "unbuddied" lists indexed by how many chunks
140 * the one unbuddied zbud uses. The data inside a zbpg cannot be
141 * read or written unless the zbpg's lock is held.
142 */
143
144#define ZBH_SENTINEL 0x43214321
145#define ZBPG_SENTINEL 0xdeadbeef
146
147#define ZBUD_MAX_BUDS 2
148
149struct zbud_hdr {
150    uint16_t client_id;
151    uint16_t pool_id;
152    struct tmem_oid oid;
153    uint32_t index;
154    uint16_t size; /* compressed size in bytes, zero means unused */
155    DECL_SENTINEL
156};
157
158struct zbud_page {
159    struct list_head bud_list;
160    spinlock_t lock;
161    struct zbud_hdr buddy[ZBUD_MAX_BUDS];
162    DECL_SENTINEL
163    /* followed by NUM_CHUNK aligned CHUNK_SIZE-byte chunks */
164};
165
166#define CHUNK_SHIFT 6
167#define CHUNK_SIZE (1 << CHUNK_SHIFT)
168#define CHUNK_MASK (~(CHUNK_SIZE-1))
169#define NCHUNKS (((PAGE_SIZE - sizeof(struct zbud_page)) & \
170                CHUNK_MASK) >> CHUNK_SHIFT)
171#define MAX_CHUNK (NCHUNKS-1)
172
173static struct {
174    struct list_head list;
175    unsigned count;
176} zbud_unbuddied[NCHUNKS];
177/* list N contains pages with N chunks USED and NCHUNKS-N unused */
178/* element 0 is never used but optimizing that isn't worth it */
179static unsigned long zbud_cumul_chunk_counts[NCHUNKS];
180
181struct list_head zbud_buddied_list;
182static unsigned long zcache_zbud_buddied_count;
183
184/* protects the buddied list and all unbuddied lists */
185static DEFINE_SPINLOCK(zbud_budlists_spinlock);
186
187static LIST_HEAD(zbpg_unused_list);
188static unsigned long zcache_zbpg_unused_list_count;
189
190/* protects the unused page list */
191static DEFINE_SPINLOCK(zbpg_unused_list_spinlock);
192
193static atomic_t zcache_zbud_curr_raw_pages;
194static atomic_t zcache_zbud_curr_zpages;
195static unsigned long zcache_zbud_curr_zbytes;
196static unsigned long zcache_zbud_cumul_zpages;
197static unsigned long zcache_zbud_cumul_zbytes;
198static unsigned long zcache_compress_poor;
199static unsigned long zcache_mean_compress_poor;
200
201/* forward references */
202static void *zcache_get_free_page(void);
203static void zcache_free_page(void *p);
204
205/*
206 * zbud helper functions
207 */
208
209static inline unsigned zbud_max_buddy_size(void)
210{
211    return MAX_CHUNK << CHUNK_SHIFT;
212}
213
214static inline unsigned zbud_size_to_chunks(unsigned size)
215{
216    BUG_ON(size == 0 || size > zbud_max_buddy_size());
217    return (size + CHUNK_SIZE - 1) >> CHUNK_SHIFT;
218}
219
220static inline int zbud_budnum(struct zbud_hdr *zh)
221{
222    unsigned offset = (unsigned long)zh & (PAGE_SIZE - 1);
223    struct zbud_page *zbpg = NULL;
224    unsigned budnum = -1U;
225    int i;
226
227    for (i = 0; i < ZBUD_MAX_BUDS; i++)
228        if (offset == offsetof(typeof(*zbpg), buddy[i])) {
229            budnum = i;
230            break;
231        }
232    BUG_ON(budnum == -1U);
233    return budnum;
234}
235
236static char *zbud_data(struct zbud_hdr *zh, unsigned size)
237{
238    struct zbud_page *zbpg;
239    char *p;
240    unsigned budnum;
241
242    ASSERT_SENTINEL(zh, ZBH);
243    budnum = zbud_budnum(zh);
244    BUG_ON(size == 0 || size > zbud_max_buddy_size());
245    zbpg = container_of(zh, struct zbud_page, buddy[budnum]);
246    ASSERT_SPINLOCK(&zbpg->lock);
247    p = (char *)zbpg;
248    if (budnum == 0)
249        p += ((sizeof(struct zbud_page) + CHUNK_SIZE - 1) &
250                            CHUNK_MASK);
251    else if (budnum == 1)
252        p += PAGE_SIZE - ((size + CHUNK_SIZE - 1) & CHUNK_MASK);
253    return p;
254}
255
256/*
257 * zbud raw page management
258 */
259
260static struct zbud_page *zbud_alloc_raw_page(void)
261{
262    struct zbud_page *zbpg = NULL;
263    struct zbud_hdr *zh0, *zh1;
264    bool recycled = 0;
265
266    /* if any pages on the zbpg list, use one */
267    spin_lock(&zbpg_unused_list_spinlock);
268    if (!list_empty(&zbpg_unused_list)) {
269        zbpg = list_first_entry(&zbpg_unused_list,
270                struct zbud_page, bud_list);
271        list_del_init(&zbpg->bud_list);
272        zcache_zbpg_unused_list_count--;
273        recycled = 1;
274    }
275    spin_unlock(&zbpg_unused_list_spinlock);
276    if (zbpg == NULL)
277        /* none on zbpg list, try to get a kernel page */
278        zbpg = zcache_get_free_page();
279    if (likely(zbpg != NULL)) {
280        INIT_LIST_HEAD(&zbpg->bud_list);
281        zh0 = &zbpg->buddy[0]; zh1 = &zbpg->buddy[1];
282        spin_lock_init(&zbpg->lock);
283        if (recycled) {
284            ASSERT_INVERTED_SENTINEL(zbpg, ZBPG);
285            SET_SENTINEL(zbpg, ZBPG);
286            BUG_ON(zh0->size != 0 || tmem_oid_valid(&zh0->oid));
287            BUG_ON(zh1->size != 0 || tmem_oid_valid(&zh1->oid));
288        } else {
289            atomic_inc(&zcache_zbud_curr_raw_pages);
290            INIT_LIST_HEAD(&zbpg->bud_list);
291            SET_SENTINEL(zbpg, ZBPG);
292            zh0->size = 0; zh1->size = 0;
293            tmem_oid_set_invalid(&zh0->oid);
294            tmem_oid_set_invalid(&zh1->oid);
295        }
296    }
297    return zbpg;
298}
299
300static void zbud_free_raw_page(struct zbud_page *zbpg)
301{
302    struct zbud_hdr *zh0 = &zbpg->buddy[0], *zh1 = &zbpg->buddy[1];
303
304    ASSERT_SENTINEL(zbpg, ZBPG);
305    BUG_ON(!list_empty(&zbpg->bud_list));
306    ASSERT_SPINLOCK(&zbpg->lock);
307    BUG_ON(zh0->size != 0 || tmem_oid_valid(&zh0->oid));
308    BUG_ON(zh1->size != 0 || tmem_oid_valid(&zh1->oid));
309    INVERT_SENTINEL(zbpg, ZBPG);
310    spin_unlock(&zbpg->lock);
311    spin_lock(&zbpg_unused_list_spinlock);
312    list_add(&zbpg->bud_list, &zbpg_unused_list);
313    zcache_zbpg_unused_list_count++;
314    spin_unlock(&zbpg_unused_list_spinlock);
315}
316
317/*
318 * core zbud handling routines
319 */
320
321static unsigned zbud_free(struct zbud_hdr *zh)
322{
323    unsigned size;
324
325    ASSERT_SENTINEL(zh, ZBH);
326    BUG_ON(!tmem_oid_valid(&zh->oid));
327    size = zh->size;
328    BUG_ON(zh->size == 0 || zh->size > zbud_max_buddy_size());
329    zh->size = 0;
330    tmem_oid_set_invalid(&zh->oid);
331    INVERT_SENTINEL(zh, ZBH);
332    zcache_zbud_curr_zbytes -= size;
333    atomic_dec(&zcache_zbud_curr_zpages);
334    return size;
335}
336
337static void zbud_free_and_delist(struct zbud_hdr *zh)
338{
339    unsigned chunks;
340    struct zbud_hdr *zh_other;
341    unsigned budnum = zbud_budnum(zh), size;
342    struct zbud_page *zbpg =
343        container_of(zh, struct zbud_page, buddy[budnum]);
344
345    spin_lock(&zbud_budlists_spinlock);
346    spin_lock(&zbpg->lock);
347    if (list_empty(&zbpg->bud_list)) {
348        /* ignore zombie page... see zbud_evict_pages() */
349        spin_unlock(&zbpg->lock);
350        spin_unlock(&zbud_budlists_spinlock);
351        return;
352    }
353    size = zbud_free(zh);
354    ASSERT_SPINLOCK(&zbpg->lock);
355    zh_other = &zbpg->buddy[(budnum == 0) ? 1 : 0];
356    if (zh_other->size == 0) { /* was unbuddied: unlist and free */
357        chunks = zbud_size_to_chunks(size) ;
358        BUG_ON(list_empty(&zbud_unbuddied[chunks].list));
359        list_del_init(&zbpg->bud_list);
360        zbud_unbuddied[chunks].count--;
361        spin_unlock(&zbud_budlists_spinlock);
362        zbud_free_raw_page(zbpg);
363    } else { /* was buddied: move remaining buddy to unbuddied list */
364        chunks = zbud_size_to_chunks(zh_other->size) ;
365        list_del_init(&zbpg->bud_list);
366        zcache_zbud_buddied_count--;
367        list_add_tail(&zbpg->bud_list, &zbud_unbuddied[chunks].list);
368        zbud_unbuddied[chunks].count++;
369        spin_unlock(&zbud_budlists_spinlock);
370        spin_unlock(&zbpg->lock);
371    }
372}
373
374static struct zbud_hdr *zbud_create(uint16_t client_id, uint16_t pool_id,
375                    struct tmem_oid *oid,
376                    uint32_t index, struct page *page,
377                    void *cdata, unsigned size)
378{
379    struct zbud_hdr *zh0, *zh1, *zh = NULL;
380    struct zbud_page *zbpg = NULL, *ztmp;
381    unsigned nchunks;
382    char *to;
383    int i, found_good_buddy = 0;
384
385    nchunks = zbud_size_to_chunks(size) ;
386    for (i = MAX_CHUNK - nchunks + 1; i > 0; i--) {
387        spin_lock(&zbud_budlists_spinlock);
388        if (!list_empty(&zbud_unbuddied[i].list)) {
389            list_for_each_entry_safe(zbpg, ztmp,
390                    &zbud_unbuddied[i].list, bud_list) {
391                if (spin_trylock(&zbpg->lock)) {
392                    found_good_buddy = i;
393                    goto found_unbuddied;
394                }
395            }
396        }
397        spin_unlock(&zbud_budlists_spinlock);
398    }
399    /* didn't find a good buddy, try allocating a new page */
400    zbpg = zbud_alloc_raw_page();
401    if (unlikely(zbpg == NULL))
402        goto out;
403    /* ok, have a page, now compress the data before taking locks */
404    spin_lock(&zbud_budlists_spinlock);
405    spin_lock(&zbpg->lock);
406    list_add_tail(&zbpg->bud_list, &zbud_unbuddied[nchunks].list);
407    zbud_unbuddied[nchunks].count++;
408    zh = &zbpg->buddy[0];
409    goto init_zh;
410
411found_unbuddied:
412    ASSERT_SPINLOCK(&zbpg->lock);
413    zh0 = &zbpg->buddy[0]; zh1 = &zbpg->buddy[1];
414    BUG_ON(!((zh0->size == 0) ^ (zh1->size == 0)));
415    if (zh0->size != 0) { /* buddy0 in use, buddy1 is vacant */
416        ASSERT_SENTINEL(zh0, ZBH);
417        zh = zh1;
418    } else if (zh1->size != 0) { /* buddy1 in use, buddy0 is vacant */
419        ASSERT_SENTINEL(zh1, ZBH);
420        zh = zh0;
421    } else
422        BUG();
423    list_del_init(&zbpg->bud_list);
424    zbud_unbuddied[found_good_buddy].count--;
425    list_add_tail(&zbpg->bud_list, &zbud_buddied_list);
426    zcache_zbud_buddied_count++;
427
428init_zh:
429    SET_SENTINEL(zh, ZBH);
430    zh->size = size;
431    zh->index = index;
432    zh->oid = *oid;
433    zh->pool_id = pool_id;
434    zh->client_id = client_id;
435    to = zbud_data(zh, size);
436    memcpy(to, cdata, size);
437    spin_unlock(&zbpg->lock);
438    spin_unlock(&zbud_budlists_spinlock);
439
440    zbud_cumul_chunk_counts[nchunks]++;
441    atomic_inc(&zcache_zbud_curr_zpages);
442    zcache_zbud_cumul_zpages++;
443    zcache_zbud_curr_zbytes += size;
444    zcache_zbud_cumul_zbytes += size;
445out:
446    return zh;
447}
448
449static int zbud_decompress(struct page *page, struct zbud_hdr *zh)
450{
451    struct zbud_page *zbpg;
452    unsigned budnum = zbud_budnum(zh);
453    unsigned int out_len = PAGE_SIZE;
454    char *to_va, *from_va;
455    unsigned size;
456    int ret = 0;
457
458    zbpg = container_of(zh, struct zbud_page, buddy[budnum]);
459    spin_lock(&zbpg->lock);
460    if (list_empty(&zbpg->bud_list)) {
461        /* ignore zombie page... see zbud_evict_pages() */
462        ret = -EINVAL;
463        goto out;
464    }
465    ASSERT_SENTINEL(zh, ZBH);
466    BUG_ON(zh->size == 0 || zh->size > zbud_max_buddy_size());
467    to_va = kmap_atomic(page);
468    size = zh->size;
469    from_va = zbud_data(zh, size);
470    ret = zcache_comp_op(ZCACHE_COMPOP_DECOMPRESS, from_va, size,
471                to_va, &out_len);
472    BUG_ON(ret);
473    BUG_ON(out_len != PAGE_SIZE);
474    kunmap_atomic(to_va);
475out:
476    spin_unlock(&zbpg->lock);
477    return ret;
478}
479
480/*
481 * The following routines handle shrinking of ephemeral pages by evicting
482 * pages "least valuable" first.
483 */
484
485static unsigned long zcache_evicted_raw_pages;
486static unsigned long zcache_evicted_buddied_pages;
487static unsigned long zcache_evicted_unbuddied_pages;
488
489static struct tmem_pool *zcache_get_pool_by_id(uint16_t cli_id,
490                        uint16_t poolid);
491static void zcache_put_pool(struct tmem_pool *pool);
492
493/*
494 * Flush and free all zbuds in a zbpg, then free the pageframe
495 */
496static void zbud_evict_zbpg(struct zbud_page *zbpg)
497{
498    struct zbud_hdr *zh;
499    int i, j;
500    uint32_t pool_id[ZBUD_MAX_BUDS], client_id[ZBUD_MAX_BUDS];
501    uint32_t index[ZBUD_MAX_BUDS];
502    struct tmem_oid oid[ZBUD_MAX_BUDS];
503    struct tmem_pool *pool;
504
505    ASSERT_SPINLOCK(&zbpg->lock);
506    BUG_ON(!list_empty(&zbpg->bud_list));
507    for (i = 0, j = 0; i < ZBUD_MAX_BUDS; i++) {
508        zh = &zbpg->buddy[i];
509        if (zh->size) {
510            client_id[j] = zh->client_id;
511            pool_id[j] = zh->pool_id;
512            oid[j] = zh->oid;
513            index[j] = zh->index;
514            j++;
515            zbud_free(zh);
516        }
517    }
518    spin_unlock(&zbpg->lock);
519    for (i = 0; i < j; i++) {
520        pool = zcache_get_pool_by_id(client_id[i], pool_id[i]);
521        if (pool != NULL) {
522            tmem_flush_page(pool, &oid[i], index[i]);
523            zcache_put_pool(pool);
524        }
525    }
526    ASSERT_SENTINEL(zbpg, ZBPG);
527    spin_lock(&zbpg->lock);
528    zbud_free_raw_page(zbpg);
529}
530
531/*
532 * Free nr pages. This code is funky because we want to hold the locks
533 * protecting various lists for as short a time as possible, and in some
534 * circumstances the list may change asynchronously when the list lock is
535 * not held. In some cases we also trylock not only to avoid waiting on a
536 * page in use by another cpu, but also to avoid potential deadlock due to
537 * lock inversion.
538 */
539static void zbud_evict_pages(int nr)
540{
541    struct zbud_page *zbpg;
542    int i;
543
544    /* first try freeing any pages on unused list */
545retry_unused_list:
546    spin_lock_bh(&zbpg_unused_list_spinlock);
547    if (!list_empty(&zbpg_unused_list)) {
548        /* can't walk list here, since it may change when unlocked */
549        zbpg = list_first_entry(&zbpg_unused_list,
550                struct zbud_page, bud_list);
551        list_del_init(&zbpg->bud_list);
552        zcache_zbpg_unused_list_count--;
553        atomic_dec(&zcache_zbud_curr_raw_pages);
554        spin_unlock_bh(&zbpg_unused_list_spinlock);
555        zcache_free_page(zbpg);
556        zcache_evicted_raw_pages++;
557        if (--nr <= 0)
558            goto out;
559        goto retry_unused_list;
560    }
561    spin_unlock_bh(&zbpg_unused_list_spinlock);
562
563    /* now try freeing unbuddied pages, starting with least space avail */
564    for (i = 0; i < MAX_CHUNK; i++) {
565retry_unbud_list_i:
566        spin_lock_bh(&zbud_budlists_spinlock);
567        if (list_empty(&zbud_unbuddied[i].list)) {
568            spin_unlock_bh(&zbud_budlists_spinlock);
569            continue;
570        }
571        list_for_each_entry(zbpg, &zbud_unbuddied[i].list, bud_list) {
572            if (unlikely(!spin_trylock(&zbpg->lock)))
573                continue;
574            list_del_init(&zbpg->bud_list);
575            zbud_unbuddied[i].count--;
576            spin_unlock(&zbud_budlists_spinlock);
577            zcache_evicted_unbuddied_pages++;
578            /* want budlists unlocked when doing zbpg eviction */
579            zbud_evict_zbpg(zbpg);
580            local_bh_enable();
581            if (--nr <= 0)
582                goto out;
583            goto retry_unbud_list_i;
584        }
585        spin_unlock_bh(&zbud_budlists_spinlock);
586    }
587
588    /* as a last resort, free buddied pages */
589retry_bud_list:
590    spin_lock_bh(&zbud_budlists_spinlock);
591    if (list_empty(&zbud_buddied_list)) {
592        spin_unlock_bh(&zbud_budlists_spinlock);
593        goto out;
594    }
595    list_for_each_entry(zbpg, &zbud_buddied_list, bud_list) {
596        if (unlikely(!spin_trylock(&zbpg->lock)))
597            continue;
598        list_del_init(&zbpg->bud_list);
599        zcache_zbud_buddied_count--;
600        spin_unlock(&zbud_budlists_spinlock);
601        zcache_evicted_buddied_pages++;
602        /* want budlists unlocked when doing zbpg eviction */
603        zbud_evict_zbpg(zbpg);
604        local_bh_enable();
605        if (--nr <= 0)
606            goto out;
607        goto retry_bud_list;
608    }
609    spin_unlock_bh(&zbud_budlists_spinlock);
610out:
611    return;
612}
613
614static void __init zbud_init(void)
615{
616    int i;
617
618    INIT_LIST_HEAD(&zbud_buddied_list);
619
620    for (i = 0; i < NCHUNKS; i++)
621        INIT_LIST_HEAD(&zbud_unbuddied[i].list);
622}
623
624#ifdef CONFIG_SYSFS
625/*
626 * These sysfs routines show a nice distribution of how many zbpg's are
627 * currently (and have ever been placed) in each unbuddied list. It's fun
628 * to watch but can probably go away before final merge.
629 */
630static int zbud_show_unbuddied_list_counts(char *buf)
631{
632    int i;
633    char *p = buf;
634
635    for (i = 0; i < NCHUNKS; i++)
636        p += sprintf(p, "%u ", zbud_unbuddied[i].count);
637    return p - buf;
638}
639
640static int zbud_show_cumul_chunk_counts(char *buf)
641{
642    unsigned long i, chunks = 0, total_chunks = 0, sum_total_chunks = 0;
643    unsigned long total_chunks_lte_21 = 0, total_chunks_lte_32 = 0;
644    unsigned long total_chunks_lte_42 = 0;
645    char *p = buf;
646
647    for (i = 0; i < NCHUNKS; i++) {
648        p += sprintf(p, "%lu ", zbud_cumul_chunk_counts[i]);
649        chunks += zbud_cumul_chunk_counts[i];
650        total_chunks += zbud_cumul_chunk_counts[i];
651        sum_total_chunks += i * zbud_cumul_chunk_counts[i];
652        if (i == 21)
653            total_chunks_lte_21 = total_chunks;
654        if (i == 32)
655            total_chunks_lte_32 = total_chunks;
656        if (i == 42)
657            total_chunks_lte_42 = total_chunks;
658    }
659    p += sprintf(p, "<=21:%lu <=32:%lu <=42:%lu, mean:%lu\n",
660        total_chunks_lte_21, total_chunks_lte_32, total_chunks_lte_42,
661        chunks == 0 ? 0 : sum_total_chunks / chunks);
662    return p - buf;
663}
664#endif
665
666/**********
667 * This "zv" PAM implementation combines the slab-based zsmalloc
668 * with the crypto compression API to maximize the amount of data that can
669 * be packed into a physical page.
670 *
671 * Zv represents a PAM page with the index and object (plus a "size" value
672 * necessary for decompression) immediately preceding the compressed data.
673 */
674
675#define ZVH_SENTINEL 0x43214321
676
677struct zv_hdr {
678    uint32_t pool_id;
679    struct tmem_oid oid;
680    uint32_t index;
681    size_t size;
682    DECL_SENTINEL
683};
684
685/* rudimentary policy limits */
686/* total number of persistent pages may not exceed this percentage */
687static unsigned int zv_page_count_policy_percent = 75;
688/*
689 * byte count defining poor compression; pages with greater zsize will be
690 * rejected
691 */
692static unsigned int zv_max_zsize = (PAGE_SIZE / 8) * 7;
693/*
694 * byte count defining poor *mean* compression; pages with greater zsize
695 * will be rejected until sufficient better-compressed pages are accepted
696 * driving the mean below this threshold
697 */
698static unsigned int zv_max_mean_zsize = (PAGE_SIZE / 8) * 5;
699
700static atomic_t zv_curr_dist_counts[NCHUNKS];
701static atomic_t zv_cumul_dist_counts[NCHUNKS];
702
703static unsigned long zv_create(struct zs_pool *pool, uint32_t pool_id,
704                struct tmem_oid *oid, uint32_t index,
705                void *cdata, unsigned clen)
706{
707    struct zv_hdr *zv;
708    u32 size = clen + sizeof(struct zv_hdr);
709    int chunks = (size + (CHUNK_SIZE - 1)) >> CHUNK_SHIFT;
710    unsigned long handle = 0;
711
712    BUG_ON(!irqs_disabled());
713    BUG_ON(chunks >= NCHUNKS);
714    handle = zs_malloc(pool, size);
715    if (!handle)
716        goto out;
717    atomic_inc(&zv_curr_dist_counts[chunks]);
718    atomic_inc(&zv_cumul_dist_counts[chunks]);
719    zv = zs_map_object(pool, handle, ZS_MM_WO);
720    zv->index = index;
721    zv->oid = *oid;
722    zv->pool_id = pool_id;
723    zv->size = clen;
724    SET_SENTINEL(zv, ZVH);
725    memcpy((char *)zv + sizeof(struct zv_hdr), cdata, clen);
726    zs_unmap_object(pool, handle);
727out:
728    return handle;
729}
730
731static void zv_free(struct zs_pool *pool, unsigned long handle)
732{
733    unsigned long flags;
734    struct zv_hdr *zv;
735    uint16_t size;
736    int chunks;
737
738    zv = zs_map_object(pool, handle, ZS_MM_RW);
739    ASSERT_SENTINEL(zv, ZVH);
740    size = zv->size + sizeof(struct zv_hdr);
741    INVERT_SENTINEL(zv, ZVH);
742    zs_unmap_object(pool, handle);
743
744    chunks = (size + (CHUNK_SIZE - 1)) >> CHUNK_SHIFT;
745    BUG_ON(chunks >= NCHUNKS);
746    atomic_dec(&zv_curr_dist_counts[chunks]);
747
748    local_irq_save(flags);
749    zs_free(pool, handle);
750    local_irq_restore(flags);
751}
752
753static void zv_decompress(struct page *page, unsigned long handle)
754{
755    unsigned int clen = PAGE_SIZE;
756    char *to_va;
757    int ret;
758    struct zv_hdr *zv;
759
760    zv = zs_map_object(zcache_host.zspool, handle, ZS_MM_RO);
761    BUG_ON(zv->size == 0);
762    ASSERT_SENTINEL(zv, ZVH);
763    to_va = kmap_atomic(page);
764    ret = zcache_comp_op(ZCACHE_COMPOP_DECOMPRESS, (char *)zv + sizeof(*zv),
765                zv->size, to_va, &clen);
766    kunmap_atomic(to_va);
767    zs_unmap_object(zcache_host.zspool, handle);
768    BUG_ON(ret);
769    BUG_ON(clen != PAGE_SIZE);
770}
771
772#ifdef CONFIG_SYSFS
773/*
774 * show a distribution of compression stats for zv pages.
775 */
776
777static int zv_curr_dist_counts_show(char *buf)
778{
779    unsigned long i, n, chunks = 0, sum_total_chunks = 0;
780    char *p = buf;
781
782    for (i = 0; i < NCHUNKS; i++) {
783        n = atomic_read(&zv_curr_dist_counts[i]);
784        p += sprintf(p, "%lu ", n);
785        chunks += n;
786        sum_total_chunks += i * n;
787    }
788    p += sprintf(p, "mean:%lu\n",
789        chunks == 0 ? 0 : sum_total_chunks / chunks);
790    return p - buf;
791}
792
793static int zv_cumul_dist_counts_show(char *buf)
794{
795    unsigned long i, n, chunks = 0, sum_total_chunks = 0;
796    char *p = buf;
797
798    for (i = 0; i < NCHUNKS; i++) {
799        n = atomic_read(&zv_cumul_dist_counts[i]);
800        p += sprintf(p, "%lu ", n);
801        chunks += n;
802        sum_total_chunks += i * n;
803    }
804    p += sprintf(p, "mean:%lu\n",
805        chunks == 0 ? 0 : sum_total_chunks / chunks);
806    return p - buf;
807}
808
809/*
810 * setting zv_max_zsize via sysfs causes all persistent (e.g. swap)
811 * pages that don't compress to less than this value (including metadata
812 * overhead) to be rejected. We don't allow the value to get too close
813 * to PAGE_SIZE.
814 */
815static ssize_t zv_max_zsize_show(struct kobject *kobj,
816                    struct kobj_attribute *attr,
817                    char *buf)
818{
819    return sprintf(buf, "%u\n", zv_max_zsize);
820}
821
822static ssize_t zv_max_zsize_store(struct kobject *kobj,
823                    struct kobj_attribute *attr,
824                    const char *buf, size_t count)
825{
826    unsigned long val;
827    int err;
828
829    if (!capable(CAP_SYS_ADMIN))
830        return -EPERM;
831
832    err = kstrtoul(buf, 10, &val);
833    if (err || (val == 0) || (val > (PAGE_SIZE / 8) * 7))
834        return -EINVAL;
835    zv_max_zsize = val;
836    return count;
837}
838
839/*
840 * setting zv_max_mean_zsize via sysfs causes all persistent (e.g. swap)
841 * pages that don't compress to less than this value (including metadata
842 * overhead) to be rejected UNLESS the mean compression is also smaller
843 * than this value. In other words, we are load-balancing-by-zsize the
844 * accepted pages. Again, we don't allow the value to get too close
845 * to PAGE_SIZE.
846 */
847static ssize_t zv_max_mean_zsize_show(struct kobject *kobj,
848                    struct kobj_attribute *attr,
849                    char *buf)
850{
851    return sprintf(buf, "%u\n", zv_max_mean_zsize);
852}
853
854static ssize_t zv_max_mean_zsize_store(struct kobject *kobj,
855                    struct kobj_attribute *attr,
856                    const char *buf, size_t count)
857{
858    unsigned long val;
859    int err;
860
861    if (!capable(CAP_SYS_ADMIN))
862        return -EPERM;
863
864    err = kstrtoul(buf, 10, &val);
865    if (err || (val == 0) || (val > (PAGE_SIZE / 8) * 7))
866        return -EINVAL;
867    zv_max_mean_zsize = val;
868    return count;
869}
870
871/*
872 * setting zv_page_count_policy_percent via sysfs sets an upper bound of
873 * persistent (e.g. swap) pages that will be retained according to:
874 * (zv_page_count_policy_percent * totalram_pages) / 100)
875 * when that limit is reached, further puts will be rejected (until
876 * some pages have been flushed). Note that, due to compression,
877 * this number may exceed 100; it defaults to 75 and we set an
878 * arbitary limit of 150. A poor choice will almost certainly result
879 * in OOM's, so this value should only be changed prudently.
880 */
881static ssize_t zv_page_count_policy_percent_show(struct kobject *kobj,
882                         struct kobj_attribute *attr,
883                         char *buf)
884{
885    return sprintf(buf, "%u\n", zv_page_count_policy_percent);
886}
887
888static ssize_t zv_page_count_policy_percent_store(struct kobject *kobj,
889                          struct kobj_attribute *attr,
890                          const char *buf, size_t count)
891{
892    unsigned long val;
893    int err;
894
895    if (!capable(CAP_SYS_ADMIN))
896        return -EPERM;
897
898    err = kstrtoul(buf, 10, &val);
899    if (err || (val == 0) || (val > 150))
900        return -EINVAL;
901    zv_page_count_policy_percent = val;
902    return count;
903}
904
905static struct kobj_attribute zcache_zv_max_zsize_attr = {
906        .attr = { .name = "zv_max_zsize", .mode = 0644 },
907        .show = zv_max_zsize_show,
908        .store = zv_max_zsize_store,
909};
910
911static struct kobj_attribute zcache_zv_max_mean_zsize_attr = {
912        .attr = { .name = "zv_max_mean_zsize", .mode = 0644 },
913        .show = zv_max_mean_zsize_show,
914        .store = zv_max_mean_zsize_store,
915};
916
917static struct kobj_attribute zcache_zv_page_count_policy_percent_attr = {
918        .attr = { .name = "zv_page_count_policy_percent",
919              .mode = 0644 },
920        .show = zv_page_count_policy_percent_show,
921        .store = zv_page_count_policy_percent_store,
922};
923#endif
924
925/*
926 * zcache core code starts here
927 */
928
929/* useful stats not collected by cleancache or frontswap */
930static unsigned long zcache_flush_total;
931static unsigned long zcache_flush_found;
932static unsigned long zcache_flobj_total;
933static unsigned long zcache_flobj_found;
934static unsigned long zcache_failed_eph_puts;
935static unsigned long zcache_failed_pers_puts;
936
937/*
938 * Tmem operations assume the poolid implies the invoking client.
939 * Zcache only has one client (the kernel itself): LOCAL_CLIENT.
940 * RAMster has each client numbered by cluster node, and a KVM version
941 * of zcache would have one client per guest and each client might
942 * have a poolid==N.
943 */
944static struct tmem_pool *zcache_get_pool_by_id(uint16_t cli_id, uint16_t poolid)
945{
946    struct tmem_pool *pool = NULL;
947    struct zcache_client *cli = NULL;
948
949    cli = get_zcache_client(cli_id);
950    if (!cli)
951        goto out;
952
953    atomic_inc(&cli->refcount);
954    pool = idr_find(&cli->tmem_pools, poolid);
955    if (pool != NULL)
956        atomic_inc(&pool->refcount);
957out:
958    return pool;
959}
960
961static void zcache_put_pool(struct tmem_pool *pool)
962{
963    struct zcache_client *cli = NULL;
964
965    if (pool == NULL)
966        BUG();
967    cli = pool->client;
968    atomic_dec(&pool->refcount);
969    atomic_dec(&cli->refcount);
970}
971
972int zcache_new_client(uint16_t cli_id)
973{
974    struct zcache_client *cli;
975    int ret = -1;
976
977    cli = get_zcache_client(cli_id);
978
979    if (cli == NULL)
980        goto out;
981    if (cli->allocated)
982        goto out;
983    cli->allocated = 1;
984#ifdef CONFIG_FRONTSWAP
985    cli->zspool = zs_create_pool("zcache", ZCACHE_GFP_MASK);
986    if (cli->zspool == NULL)
987        goto out;
988    idr_init(&cli->tmem_pools);
989#endif
990    ret = 0;
991out:
992    return ret;
993}
994
995/* counters for debugging */
996static unsigned long zcache_failed_get_free_pages;
997static unsigned long zcache_failed_alloc;
998static unsigned long zcache_put_to_flush;
999
1000/*
1001 * for now, used named slabs so can easily track usage; later can
1002 * either just use kmalloc, or perhaps add a slab-like allocator
1003 * to more carefully manage total memory utilization
1004 */
1005static struct kmem_cache *zcache_objnode_cache;
1006static struct kmem_cache *zcache_obj_cache;
1007static atomic_t zcache_curr_obj_count = ATOMIC_INIT(0);
1008static unsigned long zcache_curr_obj_count_max;
1009static atomic_t zcache_curr_objnode_count = ATOMIC_INIT(0);
1010static unsigned long zcache_curr_objnode_count_max;
1011
1012/*
1013 * to avoid memory allocation recursion (e.g. due to direct reclaim), we
1014 * preload all necessary data structures so the hostops callbacks never
1015 * actually do a malloc
1016 */
1017struct zcache_preload {
1018    void *page;
1019    struct tmem_obj *obj;
1020    int nr;
1021    struct tmem_objnode *objnodes[OBJNODE_TREE_MAX_PATH];
1022};
1023static DEFINE_PER_CPU(struct zcache_preload, zcache_preloads) = { 0, };
1024
1025static int zcache_do_preload(struct tmem_pool *pool)
1026{
1027    struct zcache_preload *kp;
1028    struct tmem_objnode *objnode;
1029    struct tmem_obj *obj;
1030    void *page;
1031    int ret = -ENOMEM;
1032
1033    if (unlikely(zcache_objnode_cache == NULL))
1034        goto out;
1035    if (unlikely(zcache_obj_cache == NULL))
1036        goto out;
1037
1038    /* IRQ has already been disabled. */
1039    kp = &__get_cpu_var(zcache_preloads);
1040    while (kp->nr < ARRAY_SIZE(kp->objnodes)) {
1041        objnode = kmem_cache_alloc(zcache_objnode_cache,
1042                ZCACHE_GFP_MASK);
1043        if (unlikely(objnode == NULL)) {
1044            zcache_failed_alloc++;
1045            goto out;
1046        }
1047
1048        kp->objnodes[kp->nr++] = objnode;
1049    }
1050
1051    if (!kp->obj) {
1052        obj = kmem_cache_alloc(zcache_obj_cache, ZCACHE_GFP_MASK);
1053        if (unlikely(obj == NULL)) {
1054            zcache_failed_alloc++;
1055            goto out;
1056        }
1057        kp->obj = obj;
1058    }
1059
1060    if (!kp->page) {
1061        page = (void *)__get_free_page(ZCACHE_GFP_MASK);
1062        if (unlikely(page == NULL)) {
1063            zcache_failed_get_free_pages++;
1064            goto out;
1065        }
1066        kp->page = page;
1067    }
1068
1069    ret = 0;
1070out:
1071    return ret;
1072}
1073
1074static void *zcache_get_free_page(void)
1075{
1076    struct zcache_preload *kp;
1077    void *page;
1078
1079    kp = &__get_cpu_var(zcache_preloads);
1080    page = kp->page;
1081    BUG_ON(page == NULL);
1082    kp->page = NULL;
1083    return page;
1084}
1085
1086static void zcache_free_page(void *p)
1087{
1088    free_page((unsigned long)p);
1089}
1090
1091/*
1092 * zcache implementation for tmem host ops
1093 */
1094
1095static struct tmem_objnode *zcache_objnode_alloc(struct tmem_pool *pool)
1096{
1097    struct tmem_objnode *objnode = NULL;
1098    unsigned long count;
1099    struct zcache_preload *kp;
1100
1101    kp = &__get_cpu_var(zcache_preloads);
1102    if (kp->nr <= 0)
1103        goto out;
1104    objnode = kp->objnodes[kp->nr - 1];
1105    BUG_ON(objnode == NULL);
1106    kp->objnodes[kp->nr - 1] = NULL;
1107    kp->nr--;
1108    count = atomic_inc_return(&zcache_curr_objnode_count);
1109    if (count > zcache_curr_objnode_count_max)
1110        zcache_curr_objnode_count_max = count;
1111out:
1112    return objnode;
1113}
1114
1115static void zcache_objnode_free(struct tmem_objnode *objnode,
1116                    struct tmem_pool *pool)
1117{
1118    atomic_dec(&zcache_curr_objnode_count);
1119    BUG_ON(atomic_read(&zcache_curr_objnode_count) < 0);
1120    kmem_cache_free(zcache_objnode_cache, objnode);
1121}
1122
1123static struct tmem_obj *zcache_obj_alloc(struct tmem_pool *pool)
1124{
1125    struct tmem_obj *obj = NULL;
1126    unsigned long count;
1127    struct zcache_preload *kp;
1128
1129    kp = &__get_cpu_var(zcache_preloads);
1130    obj = kp->obj;
1131    BUG_ON(obj == NULL);
1132    kp->obj = NULL;
1133    count = atomic_inc_return(&zcache_curr_obj_count);
1134    if (count > zcache_curr_obj_count_max)
1135        zcache_curr_obj_count_max = count;
1136    return obj;
1137}
1138
1139static void zcache_obj_free(struct tmem_obj *obj, struct tmem_pool *pool)
1140{
1141    atomic_dec(&zcache_curr_obj_count);
1142    BUG_ON(atomic_read(&zcache_curr_obj_count) < 0);
1143    kmem_cache_free(zcache_obj_cache, obj);
1144}
1145
1146static struct tmem_hostops zcache_hostops = {
1147    .obj_alloc = zcache_obj_alloc,
1148    .obj_free = zcache_obj_free,
1149    .objnode_alloc = zcache_objnode_alloc,
1150    .objnode_free = zcache_objnode_free,
1151};
1152
1153/*
1154 * zcache implementations for PAM page descriptor ops
1155 */
1156
1157static atomic_t zcache_curr_eph_pampd_count = ATOMIC_INIT(0);
1158static unsigned long zcache_curr_eph_pampd_count_max;
1159static atomic_t zcache_curr_pers_pampd_count = ATOMIC_INIT(0);
1160static unsigned long zcache_curr_pers_pampd_count_max;
1161
1162/* forward reference */
1163static int zcache_compress(struct page *from, void **out_va, unsigned *out_len);
1164
1165static void *zcache_pampd_create(char *data, size_t size, bool raw, int eph,
1166                struct tmem_pool *pool, struct tmem_oid *oid,
1167                 uint32_t index)
1168{
1169    void *pampd = NULL, *cdata;
1170    unsigned clen;
1171    int ret;
1172    unsigned long count;
1173    struct page *page = (struct page *)(data);
1174    struct zcache_client *cli = pool->client;
1175    uint16_t client_id = get_client_id_from_client(cli);
1176    unsigned long zv_mean_zsize;
1177    unsigned long curr_pers_pampd_count;
1178    u64 total_zsize;
1179
1180    if (eph) {
1181        ret = zcache_compress(page, &cdata, &clen);
1182        if (ret == 0)
1183            goto out;
1184        if (clen == 0 || clen > zbud_max_buddy_size()) {
1185            zcache_compress_poor++;
1186            goto out;
1187        }
1188        pampd = (void *)zbud_create(client_id, pool->pool_id, oid,
1189                        index, page, cdata, clen);
1190        if (pampd != NULL) {
1191            count = atomic_inc_return(&zcache_curr_eph_pampd_count);
1192            if (count > zcache_curr_eph_pampd_count_max)
1193                zcache_curr_eph_pampd_count_max = count;
1194        }
1195    } else {
1196        curr_pers_pampd_count =
1197            atomic_read(&zcache_curr_pers_pampd_count);
1198        if (curr_pers_pampd_count >
1199            (zv_page_count_policy_percent * totalram_pages) / 100)
1200            goto out;
1201        ret = zcache_compress(page, &cdata, &clen);
1202        if (ret == 0)
1203            goto out;
1204        /* reject if compression is too poor */
1205        if (clen > zv_max_zsize) {
1206            zcache_compress_poor++;
1207            goto out;
1208        }
1209        /* reject if mean compression is too poor */
1210        if ((clen > zv_max_mean_zsize) && (curr_pers_pampd_count > 0)) {
1211            total_zsize = zs_get_total_size_bytes(cli->zspool);
1212            zv_mean_zsize = div_u64(total_zsize,
1213                        curr_pers_pampd_count);
1214            if (zv_mean_zsize > zv_max_mean_zsize) {
1215                zcache_mean_compress_poor++;
1216                goto out;
1217            }
1218        }
1219        pampd = (void *)zv_create(cli->zspool, pool->pool_id,
1220                        oid, index, cdata, clen);
1221        if (pampd == NULL)
1222            goto out;
1223        count = atomic_inc_return(&zcache_curr_pers_pampd_count);
1224        if (count > zcache_curr_pers_pampd_count_max)
1225            zcache_curr_pers_pampd_count_max = count;
1226    }
1227out:
1228    return pampd;
1229}
1230
1231/*
1232 * fill the pageframe corresponding to the struct page with the data
1233 * from the passed pampd
1234 */
1235static int zcache_pampd_get_data(char *data, size_t *bufsize, bool raw,
1236                    void *pampd, struct tmem_pool *pool,
1237                    struct tmem_oid *oid, uint32_t index)
1238{
1239    int ret = 0;
1240
1241    BUG_ON(is_ephemeral(pool));
1242    zv_decompress((struct page *)(data), (unsigned long)pampd);
1243    return ret;
1244}
1245
1246/*
1247 * fill the pageframe corresponding to the struct page with the data
1248 * from the passed pampd
1249 */
1250static int zcache_pampd_get_data_and_free(char *data, size_t *bufsize, bool raw,
1251                    void *pampd, struct tmem_pool *pool,
1252                    struct tmem_oid *oid, uint32_t index)
1253{
1254    BUG_ON(!is_ephemeral(pool));
1255    if (zbud_decompress((struct page *)(data), pampd) < 0)
1256        return -EINVAL;
1257    zbud_free_and_delist((struct zbud_hdr *)pampd);
1258    atomic_dec(&zcache_curr_eph_pampd_count);
1259    return 0;
1260}
1261
1262/*
1263 * free the pampd and remove it from any zcache lists
1264 * pampd must no longer be pointed to from any tmem data structures!
1265 */
1266static void zcache_pampd_free(void *pampd, struct tmem_pool *pool,
1267                struct tmem_oid *oid, uint32_t index)
1268{
1269    struct zcache_client *cli = pool->client;
1270
1271    if (is_ephemeral(pool)) {
1272        zbud_free_and_delist((struct zbud_hdr *)pampd);
1273        atomic_dec(&zcache_curr_eph_pampd_count);
1274        BUG_ON(atomic_read(&zcache_curr_eph_pampd_count) < 0);
1275    } else {
1276        zv_free(cli->zspool, (unsigned long)pampd);
1277        atomic_dec(&zcache_curr_pers_pampd_count);
1278        BUG_ON(atomic_read(&zcache_curr_pers_pampd_count) < 0);
1279    }
1280}
1281
1282static void zcache_pampd_free_obj(struct tmem_pool *pool, struct tmem_obj *obj)
1283{
1284}
1285
1286static void zcache_pampd_new_obj(struct tmem_obj *obj)
1287{
1288}
1289
1290static int zcache_pampd_replace_in_obj(void *pampd, struct tmem_obj *obj)
1291{
1292    return -1;
1293}
1294
1295static bool zcache_pampd_is_remote(void *pampd)
1296{
1297    return 0;
1298}
1299
1300static struct tmem_pamops zcache_pamops = {
1301    .create = zcache_pampd_create,
1302    .get_data = zcache_pampd_get_data,
1303    .get_data_and_free = zcache_pampd_get_data_and_free,
1304    .free = zcache_pampd_free,
1305    .free_obj = zcache_pampd_free_obj,
1306    .new_obj = zcache_pampd_new_obj,
1307    .replace_in_obj = zcache_pampd_replace_in_obj,
1308    .is_remote = zcache_pampd_is_remote,
1309};
1310
1311/*
1312 * zcache compression/decompression and related per-cpu stuff
1313 */
1314
1315static DEFINE_PER_CPU(unsigned char *, zcache_dstmem);
1316#define ZCACHE_DSTMEM_ORDER 1
1317
1318static int zcache_compress(struct page *from, void **out_va, unsigned *out_len)
1319{
1320    int ret = 0;
1321    unsigned char *dmem = __get_cpu_var(zcache_dstmem);
1322    char *from_va;
1323
1324    BUG_ON(!irqs_disabled());
1325    if (unlikely(dmem == NULL))
1326        goto out; /* no buffer or no compressor so can't compress */
1327    *out_len = PAGE_SIZE << ZCACHE_DSTMEM_ORDER;
1328    from_va = kmap_atomic(from);
1329    mb();
1330    ret = zcache_comp_op(ZCACHE_COMPOP_COMPRESS, from_va, PAGE_SIZE, dmem,
1331                out_len);
1332    BUG_ON(ret);
1333    *out_va = dmem;
1334    kunmap_atomic(from_va);
1335    ret = 1;
1336out:
1337    return ret;
1338}
1339
1340static int zcache_comp_cpu_up(int cpu)
1341{
1342    struct crypto_comp *tfm;
1343
1344    tfm = crypto_alloc_comp(zcache_comp_name, 0, 0);
1345    if (IS_ERR(tfm))
1346        return NOTIFY_BAD;
1347    *per_cpu_ptr(zcache_comp_pcpu_tfms, cpu) = tfm;
1348    return NOTIFY_OK;
1349}
1350
1351static void zcache_comp_cpu_down(int cpu)
1352{
1353    struct crypto_comp *tfm;
1354
1355    tfm = *per_cpu_ptr(zcache_comp_pcpu_tfms, cpu);
1356    crypto_free_comp(tfm);
1357    *per_cpu_ptr(zcache_comp_pcpu_tfms, cpu) = NULL;
1358}
1359
1360static int zcache_cpu_notifier(struct notifier_block *nb,
1361                unsigned long action, void *pcpu)
1362{
1363    int ret, cpu = (long)pcpu;
1364    struct zcache_preload *kp;
1365
1366    switch (action) {
1367    case CPU_UP_PREPARE:
1368        ret = zcache_comp_cpu_up(cpu);
1369        if (ret != NOTIFY_OK) {
1370            pr_err("zcache: can't allocate compressor transform\n");
1371            return ret;
1372        }
1373        per_cpu(zcache_dstmem, cpu) = (void *)__get_free_pages(
1374            GFP_KERNEL | __GFP_REPEAT, ZCACHE_DSTMEM_ORDER);
1375        break;
1376    case CPU_DEAD:
1377    case CPU_UP_CANCELED:
1378        zcache_comp_cpu_down(cpu);
1379        free_pages((unsigned long)per_cpu(zcache_dstmem, cpu),
1380            ZCACHE_DSTMEM_ORDER);
1381        per_cpu(zcache_dstmem, cpu) = NULL;
1382        kp = &per_cpu(zcache_preloads, cpu);
1383        while (kp->nr) {
1384            kmem_cache_free(zcache_objnode_cache,
1385                    kp->objnodes[kp->nr - 1]);
1386            kp->objnodes[kp->nr - 1] = NULL;
1387            kp->nr--;
1388        }
1389        if (kp->obj) {
1390            kmem_cache_free(zcache_obj_cache, kp->obj);
1391            kp->obj = NULL;
1392        }
1393        if (kp->page) {
1394            free_page((unsigned long)kp->page);
1395            kp->page = NULL;
1396        }
1397        break;
1398    default:
1399        break;
1400    }
1401    return NOTIFY_OK;
1402}
1403
1404static struct notifier_block zcache_cpu_notifier_block = {
1405    .notifier_call = zcache_cpu_notifier
1406};
1407
1408#ifdef CONFIG_SYSFS
1409#define ZCACHE_SYSFS_RO(_name) \
1410    static ssize_t zcache_##_name##_show(struct kobject *kobj, \
1411                struct kobj_attribute *attr, char *buf) \
1412    { \
1413        return sprintf(buf, "%lu\n", zcache_##_name); \
1414    } \
1415    static struct kobj_attribute zcache_##_name##_attr = { \
1416        .attr = { .name = __stringify(_name), .mode = 0444 }, \
1417        .show = zcache_##_name##_show, \
1418    }
1419
1420#define ZCACHE_SYSFS_RO_ATOMIC(_name) \
1421    static ssize_t zcache_##_name##_show(struct kobject *kobj, \
1422                struct kobj_attribute *attr, char *buf) \
1423    { \
1424        return sprintf(buf, "%d\n", atomic_read(&zcache_##_name)); \
1425    } \
1426    static struct kobj_attribute zcache_##_name##_attr = { \
1427        .attr = { .name = __stringify(_name), .mode = 0444 }, \
1428        .show = zcache_##_name##_show, \
1429    }
1430
1431#define ZCACHE_SYSFS_RO_CUSTOM(_name, _func) \
1432    static ssize_t zcache_##_name##_show(struct kobject *kobj, \
1433                struct kobj_attribute *attr, char *buf) \
1434    { \
1435        return _func(buf); \
1436    } \
1437    static struct kobj_attribute zcache_##_name##_attr = { \
1438        .attr = { .name = __stringify(_name), .mode = 0444 }, \
1439        .show = zcache_##_name##_show, \
1440    }
1441
1442ZCACHE_SYSFS_RO(curr_obj_count_max);
1443ZCACHE_SYSFS_RO(curr_objnode_count_max);
1444ZCACHE_SYSFS_RO(flush_total);
1445ZCACHE_SYSFS_RO(flush_found);
1446ZCACHE_SYSFS_RO(flobj_total);
1447ZCACHE_SYSFS_RO(flobj_found);
1448ZCACHE_SYSFS_RO(failed_eph_puts);
1449ZCACHE_SYSFS_RO(failed_pers_puts);
1450ZCACHE_SYSFS_RO(zbud_curr_zbytes);
1451ZCACHE_SYSFS_RO(zbud_cumul_zpages);
1452ZCACHE_SYSFS_RO(zbud_cumul_zbytes);
1453ZCACHE_SYSFS_RO(zbud_buddied_count);
1454ZCACHE_SYSFS_RO(zbpg_unused_list_count);
1455ZCACHE_SYSFS_RO(evicted_raw_pages);
1456ZCACHE_SYSFS_RO(evicted_unbuddied_pages);
1457ZCACHE_SYSFS_RO(evicted_buddied_pages);
1458ZCACHE_SYSFS_RO(failed_get_free_pages);
1459ZCACHE_SYSFS_RO(failed_alloc);
1460ZCACHE_SYSFS_RO(put_to_flush);
1461ZCACHE_SYSFS_RO(compress_poor);
1462ZCACHE_SYSFS_RO(mean_compress_poor);
1463ZCACHE_SYSFS_RO_ATOMIC(zbud_curr_raw_pages);
1464ZCACHE_SYSFS_RO_ATOMIC(zbud_curr_zpages);
1465ZCACHE_SYSFS_RO_ATOMIC(curr_obj_count);
1466ZCACHE_SYSFS_RO_ATOMIC(curr_objnode_count);
1467ZCACHE_SYSFS_RO_CUSTOM(zbud_unbuddied_list_counts,
1468            zbud_show_unbuddied_list_counts);
1469ZCACHE_SYSFS_RO_CUSTOM(zbud_cumul_chunk_counts,
1470            zbud_show_cumul_chunk_counts);
1471ZCACHE_SYSFS_RO_CUSTOM(zv_curr_dist_counts,
1472            zv_curr_dist_counts_show);
1473ZCACHE_SYSFS_RO_CUSTOM(zv_cumul_dist_counts,
1474            zv_cumul_dist_counts_show);
1475
1476static struct attribute *zcache_attrs[] = {
1477    &zcache_curr_obj_count_attr.attr,
1478    &zcache_curr_obj_count_max_attr.attr,
1479    &zcache_curr_objnode_count_attr.attr,
1480    &zcache_curr_objnode_count_max_attr.attr,
1481    &zcache_flush_total_attr.attr,
1482    &zcache_flobj_total_attr.attr,
1483    &zcache_flush_found_attr.attr,
1484    &zcache_flobj_found_attr.attr,
1485    &zcache_failed_eph_puts_attr.attr,
1486    &zcache_failed_pers_puts_attr.attr,
1487    &zcache_compress_poor_attr.attr,
1488    &zcache_mean_compress_poor_attr.attr,
1489    &zcache_zbud_curr_raw_pages_attr.attr,
1490    &zcache_zbud_curr_zpages_attr.attr,
1491    &zcache_zbud_curr_zbytes_attr.attr,
1492    &zcache_zbud_cumul_zpages_attr.attr,
1493    &zcache_zbud_cumul_zbytes_attr.attr,
1494    &zcache_zbud_buddied_count_attr.attr,
1495    &zcache_zbpg_unused_list_count_attr.attr,
1496    &zcache_evicted_raw_pages_attr.attr,
1497    &zcache_evicted_unbuddied_pages_attr.attr,
1498    &zcache_evicted_buddied_pages_attr.attr,
1499    &zcache_failed_get_free_pages_attr.attr,
1500    &zcache_failed_alloc_attr.attr,
1501    &zcache_put_to_flush_attr.attr,
1502    &zcache_zbud_unbuddied_list_counts_attr.attr,
1503    &zcache_zbud_cumul_chunk_counts_attr.attr,
1504    &zcache_zv_curr_dist_counts_attr.attr,
1505    &zcache_zv_cumul_dist_counts_attr.attr,
1506    &zcache_zv_max_zsize_attr.attr,
1507    &zcache_zv_max_mean_zsize_attr.attr,
1508    &zcache_zv_page_count_policy_percent_attr.attr,
1509    NULL,
1510};
1511
1512static struct attribute_group zcache_attr_group = {
1513    .attrs = zcache_attrs,
1514    .name = "zcache",
1515};
1516
1517#endif /* CONFIG_SYSFS */
1518/*
1519 * When zcache is disabled ("frozen"), pools can be created and destroyed,
1520 * but all puts (and thus all other operations that require memory allocation)
1521 * must fail. If zcache is unfrozen, accepts puts, then frozen again,
1522 * data consistency requires all puts while frozen to be converted into
1523 * flushes.
1524 */
1525static bool zcache_freeze;
1526
1527/*
1528 * zcache shrinker interface (only useful for ephemeral pages, so zbud only)
1529 */
1530static int shrink_zcache_memory(struct shrinker *shrink,
1531                struct shrink_control *sc)
1532{
1533    int ret = -1;
1534    int nr = sc->nr_to_scan;
1535    gfp_t gfp_mask = sc->gfp_mask;
1536
1537    if (nr >= 0) {
1538        if (!(gfp_mask & __GFP_FS))
1539            /* does this case really need to be skipped? */
1540            goto out;
1541        zbud_evict_pages(nr);
1542    }
1543    ret = (int)atomic_read(&zcache_zbud_curr_raw_pages);
1544out:
1545    return ret;
1546}
1547
1548static struct shrinker zcache_shrinker = {
1549    .shrink = shrink_zcache_memory,
1550    .seeks = DEFAULT_SEEKS,
1551};
1552
1553/*
1554 * zcache shims between cleancache/frontswap ops and tmem
1555 */
1556
1557static int zcache_put_page(int cli_id, int pool_id, struct tmem_oid *oidp,
1558                uint32_t index, struct page *page)
1559{
1560    struct tmem_pool *pool;
1561    int ret = -1;
1562
1563    BUG_ON(!irqs_disabled());
1564    pool = zcache_get_pool_by_id(cli_id, pool_id);
1565    if (unlikely(pool == NULL))
1566        goto out;
1567    if (!zcache_freeze && zcache_do_preload(pool) == 0) {
1568        /* preload does preempt_disable on success */
1569        ret = tmem_put(pool, oidp, index, (char *)(page),
1570                PAGE_SIZE, 0, is_ephemeral(pool));
1571        if (ret < 0) {
1572            if (is_ephemeral(pool))
1573                zcache_failed_eph_puts++;
1574            else
1575                zcache_failed_pers_puts++;
1576        }
1577    } else {
1578        zcache_put_to_flush++;
1579        if (atomic_read(&pool->obj_count) > 0)
1580            /* the put fails whether the flush succeeds or not */
1581            (void)tmem_flush_page(pool, oidp, index);
1582    }
1583
1584    zcache_put_pool(pool);
1585out:
1586    return ret;
1587}
1588
1589static int zcache_get_page(int cli_id, int pool_id, struct tmem_oid *oidp,
1590                uint32_t index, struct page *page)
1591{
1592    struct tmem_pool *pool;
1593    int ret = -1;
1594    unsigned long flags;
1595    size_t size = PAGE_SIZE;
1596
1597    local_irq_save(flags);
1598    pool = zcache_get_pool_by_id(cli_id, pool_id);
1599    if (likely(pool != NULL)) {
1600        if (atomic_read(&pool->obj_count) > 0)
1601            ret = tmem_get(pool, oidp, index, (char *)(page),
1602                    &size, 0, is_ephemeral(pool));
1603        zcache_put_pool(pool);
1604    }
1605    local_irq_restore(flags);
1606    return ret;
1607}
1608
1609static int zcache_flush_page(int cli_id, int pool_id,
1610                struct tmem_oid *oidp, uint32_t index)
1611{
1612    struct tmem_pool *pool;
1613    int ret = -1;
1614    unsigned long flags;
1615
1616    local_irq_save(flags);
1617    zcache_flush_total++;
1618    pool = zcache_get_pool_by_id(cli_id, pool_id);
1619    if (likely(pool != NULL)) {
1620        if (atomic_read(&pool->obj_count) > 0)
1621            ret = tmem_flush_page(pool, oidp, index);
1622        zcache_put_pool(pool);
1623    }
1624    if (ret >= 0)
1625        zcache_flush_found++;
1626    local_irq_restore(flags);
1627    return ret;
1628}
1629
1630static int zcache_flush_object(int cli_id, int pool_id,
1631                struct tmem_oid *oidp)
1632{
1633    struct tmem_pool *pool;
1634    int ret = -1;
1635    unsigned long flags;
1636
1637    local_irq_save(flags);
1638    zcache_flobj_total++;
1639    pool = zcache_get_pool_by_id(cli_id, pool_id);
1640    if (likely(pool != NULL)) {
1641        if (atomic_read(&pool->obj_count) > 0)
1642            ret = tmem_flush_object(pool, oidp);
1643        zcache_put_pool(pool);
1644    }
1645    if (ret >= 0)
1646        zcache_flobj_found++;
1647    local_irq_restore(flags);
1648    return ret;
1649}
1650
1651static int zcache_destroy_pool(int cli_id, int pool_id)
1652{
1653    struct tmem_pool *pool = NULL;
1654    struct zcache_client *cli;
1655    int ret = -1;
1656
1657    if (pool_id < 0)
1658        goto out;
1659
1660    cli = get_zcache_client(cli_id);
1661    if (cli == NULL)
1662        goto out;
1663
1664    atomic_inc(&cli->refcount);
1665    pool = idr_find(&cli->tmem_pools, pool_id);
1666    if (pool == NULL)
1667        goto out;
1668    idr_remove(&cli->tmem_pools, pool_id);
1669    /* wait for pool activity on other cpus to quiesce */
1670    while (atomic_read(&pool->refcount) != 0)
1671        ;
1672    atomic_dec(&cli->refcount);
1673    local_bh_disable();
1674    ret = tmem_destroy_pool(pool);
1675    local_bh_enable();
1676    kfree(pool);
1677    pr_info("zcache: destroyed pool id=%d, cli_id=%d\n",
1678            pool_id, cli_id);
1679out:
1680    return ret;
1681}
1682
1683static int zcache_new_pool(uint16_t cli_id, uint32_t flags)
1684{
1685    int poolid = -1;
1686    struct tmem_pool *pool;
1687    struct zcache_client *cli = NULL;
1688    int r;
1689
1690    cli = get_zcache_client(cli_id);
1691    if (cli == NULL)
1692        goto out;
1693
1694    atomic_inc(&cli->refcount);
1695    pool = kmalloc(sizeof(struct tmem_pool), GFP_ATOMIC);
1696    if (pool == NULL) {
1697        pr_info("zcache: pool creation failed: out of memory\n");
1698        goto out;
1699    }
1700
1701    do {
1702        r = idr_pre_get(&cli->tmem_pools, GFP_ATOMIC);
1703        if (r != 1) {
1704            kfree(pool);
1705            pr_info("zcache: pool creation failed: out of memory\n");
1706            goto out;
1707        }
1708        r = idr_get_new(&cli->tmem_pools, pool, &poolid);
1709    } while (r == -EAGAIN);
1710    if (r) {
1711        pr_info("zcache: pool creation failed: error %d\n", r);
1712        kfree(pool);
1713        goto out;
1714    }
1715
1716    atomic_set(&pool->refcount, 0);
1717    pool->client = cli;
1718    pool->pool_id = poolid;
1719    tmem_new_pool(pool, flags);
1720    pr_info("zcache: created %s tmem pool, id=%d, client=%d\n",
1721        flags & TMEM_POOL_PERSIST ? "persistent" : "ephemeral",
1722        poolid, cli_id);
1723out:
1724    if (cli != NULL)
1725        atomic_dec(&cli->refcount);
1726    return poolid;
1727}
1728
1729/**********
1730 * Two kernel functionalities currently can be layered on top of tmem.
1731 * These are "cleancache" which is used as a second-chance cache for clean
1732 * page cache pages; and "frontswap" which is used for swap pages
1733 * to avoid writes to disk. A generic "shim" is provided here for each
1734 * to translate in-kernel semantics to zcache semantics.
1735 */
1736
1737#ifdef CONFIG_CLEANCACHE
1738static void zcache_cleancache_put_page(int pool_id,
1739                    struct cleancache_filekey key,
1740                    pgoff_t index, struct page *page)
1741{
1742    u32 ind = (u32) index;
1743    struct tmem_oid oid = *(struct tmem_oid *)&key;
1744
1745    if (likely(ind == index))
1746        (void)zcache_put_page(LOCAL_CLIENT, pool_id, &oid, index, page);
1747}
1748
1749static int zcache_cleancache_get_page(int pool_id,
1750                    struct cleancache_filekey key,
1751                    pgoff_t index, struct page *page)
1752{
1753    u32 ind = (u32) index;
1754    struct tmem_oid oid = *(struct tmem_oid *)&key;
1755    int ret = -1;
1756
1757    if (likely(ind == index))
1758        ret = zcache_get_page(LOCAL_CLIENT, pool_id, &oid, index, page);
1759    return ret;
1760}
1761
1762static void zcache_cleancache_flush_page(int pool_id,
1763                    struct cleancache_filekey key,
1764                    pgoff_t index)
1765{
1766    u32 ind = (u32) index;
1767    struct tmem_oid oid = *(struct tmem_oid *)&key;
1768
1769    if (likely(ind == index))
1770        (void)zcache_flush_page(LOCAL_CLIENT, pool_id, &oid, ind);
1771}
1772
1773static void zcache_cleancache_flush_inode(int pool_id,
1774                    struct cleancache_filekey key)
1775{
1776    struct tmem_oid oid = *(struct tmem_oid *)&key;
1777
1778    (void)zcache_flush_object(LOCAL_CLIENT, pool_id, &oid);
1779}
1780
1781static void zcache_cleancache_flush_fs(int pool_id)
1782{
1783    if (pool_id >= 0)
1784        (void)zcache_destroy_pool(LOCAL_CLIENT, pool_id);
1785}
1786
1787static int zcache_cleancache_init_fs(size_t pagesize)
1788{
1789    BUG_ON(sizeof(struct cleancache_filekey) !=
1790                sizeof(struct tmem_oid));
1791    BUG_ON(pagesize != PAGE_SIZE);
1792    return zcache_new_pool(LOCAL_CLIENT, 0);
1793}
1794
1795static int zcache_cleancache_init_shared_fs(char *uuid, size_t pagesize)
1796{
1797    /* shared pools are unsupported and map to private */
1798    BUG_ON(sizeof(struct cleancache_filekey) !=
1799                sizeof(struct tmem_oid));
1800    BUG_ON(pagesize != PAGE_SIZE);
1801    return zcache_new_pool(LOCAL_CLIENT, 0);
1802}
1803
1804static struct cleancache_ops zcache_cleancache_ops = {
1805    .put_page = zcache_cleancache_put_page,
1806    .get_page = zcache_cleancache_get_page,
1807    .invalidate_page = zcache_cleancache_flush_page,
1808    .invalidate_inode = zcache_cleancache_flush_inode,
1809    .invalidate_fs = zcache_cleancache_flush_fs,
1810    .init_shared_fs = zcache_cleancache_init_shared_fs,
1811    .init_fs = zcache_cleancache_init_fs
1812};
1813
1814struct cleancache_ops zcache_cleancache_register_ops(void)
1815{
1816    struct cleancache_ops old_ops =
1817        cleancache_register_ops(&zcache_cleancache_ops);
1818
1819    return old_ops;
1820}
1821#endif
1822
1823#ifdef CONFIG_FRONTSWAP
1824/* a single tmem poolid is used for all frontswap "types" (swapfiles) */
1825static int zcache_frontswap_poolid = -1;
1826
1827/*
1828 * Swizzling increases objects per swaptype, increasing tmem concurrency
1829 * for heavy swaploads. Later, larger nr_cpus -> larger SWIZ_BITS
1830 * Setting SWIZ_BITS to 27 basically reconstructs the swap entry from
1831 * frontswap_load(), but has side-effects. Hence using 8.
1832 */
1833#define SWIZ_BITS 8
1834#define SWIZ_MASK ((1 << SWIZ_BITS) - 1)
1835#define _oswiz(_type, _ind) ((_type << SWIZ_BITS) | (_ind & SWIZ_MASK))
1836#define iswiz(_ind) (_ind >> SWIZ_BITS)
1837
1838static inline struct tmem_oid oswiz(unsigned type, u32 ind)
1839{
1840    struct tmem_oid oid = { .oid = { 0 } };
1841    oid.oid[0] = _oswiz(type, ind);
1842    return oid;
1843}
1844
1845static int zcache_frontswap_store(unsigned type, pgoff_t offset,
1846                   struct page *page)
1847{
1848    u64 ind64 = (u64)offset;
1849    u32 ind = (u32)offset;
1850    struct tmem_oid oid = oswiz(type, ind);
1851    int ret = -1;
1852    unsigned long flags;
1853
1854    BUG_ON(!PageLocked(page));
1855    if (likely(ind64 == ind)) {
1856        local_irq_save(flags);
1857        ret = zcache_put_page(LOCAL_CLIENT, zcache_frontswap_poolid,
1858                    &oid, iswiz(ind), page);
1859        local_irq_restore(flags);
1860    }
1861    return ret;
1862}
1863
1864/* returns 0 if the page was successfully gotten from frontswap, -1 if
1865 * was not present (should never happen!) */
1866static int zcache_frontswap_load(unsigned type, pgoff_t offset,
1867                   struct page *page)
1868{
1869    u64 ind64 = (u64)offset;
1870    u32 ind = (u32)offset;
1871    struct tmem_oid oid = oswiz(type, ind);
1872    int ret = -1;
1873
1874    BUG_ON(!PageLocked(page));
1875    if (likely(ind64 == ind))
1876        ret = zcache_get_page(LOCAL_CLIENT, zcache_frontswap_poolid,
1877                    &oid, iswiz(ind), page);
1878    return ret;
1879}
1880
1881/* flush a single page from frontswap */
1882static void zcache_frontswap_flush_page(unsigned type, pgoff_t offset)
1883{
1884    u64 ind64 = (u64)offset;
1885    u32 ind = (u32)offset;
1886    struct tmem_oid oid = oswiz(type, ind);
1887
1888    if (likely(ind64 == ind))
1889        (void)zcache_flush_page(LOCAL_CLIENT, zcache_frontswap_poolid,
1890                    &oid, iswiz(ind));
1891}
1892
1893/* flush all pages from the passed swaptype */
1894static void zcache_frontswap_flush_area(unsigned type)
1895{
1896    struct tmem_oid oid;
1897    int ind;
1898
1899    for (ind = SWIZ_MASK; ind >= 0; ind--) {
1900        oid = oswiz(type, ind);
1901        (void)zcache_flush_object(LOCAL_CLIENT,
1902                        zcache_frontswap_poolid, &oid);
1903    }
1904}
1905
1906static void zcache_frontswap_init(unsigned ignored)
1907{
1908    /* a single tmem poolid is used for all frontswap "types" (swapfiles) */
1909    if (zcache_frontswap_poolid < 0)
1910        zcache_frontswap_poolid =
1911            zcache_new_pool(LOCAL_CLIENT, TMEM_POOL_PERSIST);
1912}
1913
1914static struct frontswap_ops zcache_frontswap_ops = {
1915    .store = zcache_frontswap_store,
1916    .load = zcache_frontswap_load,
1917    .invalidate_page = zcache_frontswap_flush_page,
1918    .invalidate_area = zcache_frontswap_flush_area,
1919    .init = zcache_frontswap_init
1920};
1921
1922struct frontswap_ops zcache_frontswap_register_ops(void)
1923{
1924    struct frontswap_ops old_ops =
1925        frontswap_register_ops(&zcache_frontswap_ops);
1926
1927    return old_ops;
1928}
1929#endif
1930
1931/*
1932 * zcache initialization
1933 * NOTE FOR NOW zcache MUST BE PROVIDED AS A KERNEL BOOT PARAMETER OR
1934 * NOTHING HAPPENS!
1935 */
1936
1937static int zcache_enabled;
1938
1939static int __init enable_zcache(char *s)
1940{
1941    zcache_enabled = 1;
1942    return 1;
1943}
1944__setup("zcache", enable_zcache);
1945
1946/* allow independent dynamic disabling of cleancache and frontswap */
1947
1948static int use_cleancache = 1;
1949
1950static int __init no_cleancache(char *s)
1951{
1952    use_cleancache = 0;
1953    return 1;
1954}
1955
1956__setup("nocleancache", no_cleancache);
1957
1958static int use_frontswap = 1;
1959
1960static int __init no_frontswap(char *s)
1961{
1962    use_frontswap = 0;
1963    return 1;
1964}
1965
1966__setup("nofrontswap", no_frontswap);
1967
1968static int __init enable_zcache_compressor(char *s)
1969{
1970    strncpy(zcache_comp_name, s, ZCACHE_COMP_NAME_SZ);
1971    zcache_enabled = 1;
1972    return 1;
1973}
1974__setup("zcache=", enable_zcache_compressor);
1975
1976
1977static int __init zcache_comp_init(void)
1978{
1979    int ret = 0;
1980
1981    /* check crypto algorithm */
1982    if (*zcache_comp_name != '\0') {
1983        ret = crypto_has_comp(zcache_comp_name, 0, 0);
1984        if (!ret)
1985            pr_info("zcache: %s not supported\n",
1986                    zcache_comp_name);
1987    }
1988    if (!ret)
1989        strcpy(zcache_comp_name, "lzo");
1990    ret = crypto_has_comp(zcache_comp_name, 0, 0);
1991    if (!ret) {
1992        ret = 1;
1993        goto out;
1994    }
1995    pr_info("zcache: using %s compressor\n", zcache_comp_name);
1996
1997    /* alloc percpu transforms */
1998    ret = 0;
1999    zcache_comp_pcpu_tfms = alloc_percpu(struct crypto_comp *);
2000    if (!zcache_comp_pcpu_tfms)
2001        ret = 1;
2002out:
2003    return ret;
2004}
2005
2006static int __init zcache_init(void)
2007{
2008    int ret = 0;
2009
2010#ifdef CONFIG_SYSFS
2011    ret = sysfs_create_group(mm_kobj, &zcache_attr_group);
2012    if (ret) {
2013        pr_err("zcache: can't create sysfs\n");
2014        goto out;
2015    }
2016#endif /* CONFIG_SYSFS */
2017
2018    if (zcache_enabled) {
2019        unsigned int cpu;
2020
2021        tmem_register_hostops(&zcache_hostops);
2022        tmem_register_pamops(&zcache_pamops);
2023        ret = register_cpu_notifier(&zcache_cpu_notifier_block);
2024        if (ret) {
2025            pr_err("zcache: can't register cpu notifier\n");
2026            goto out;
2027        }
2028        ret = zcache_comp_init();
2029        if (ret) {
2030            pr_err("zcache: compressor initialization failed\n");
2031            goto out;
2032        }
2033        for_each_online_cpu(cpu) {
2034            void *pcpu = (void *)(long)cpu;
2035            zcache_cpu_notifier(&zcache_cpu_notifier_block,
2036                CPU_UP_PREPARE, pcpu);
2037        }
2038    }
2039    zcache_objnode_cache = kmem_cache_create("zcache_objnode",
2040                sizeof(struct tmem_objnode), 0, 0, NULL);
2041    zcache_obj_cache = kmem_cache_create("zcache_obj",
2042                sizeof(struct tmem_obj), 0, 0, NULL);
2043    ret = zcache_new_client(LOCAL_CLIENT);
2044    if (ret) {
2045        pr_err("zcache: can't create client\n");
2046        goto out;
2047    }
2048
2049#ifdef CONFIG_CLEANCACHE
2050    if (zcache_enabled && use_cleancache) {
2051        struct cleancache_ops old_ops;
2052
2053        zbud_init();
2054        register_shrinker(&zcache_shrinker);
2055        old_ops = zcache_cleancache_register_ops();
2056        pr_info("zcache: cleancache enabled using kernel "
2057            "transcendent memory and compression buddies\n");
2058        if (old_ops.init_fs != NULL)
2059            pr_warning("zcache: cleancache_ops overridden");
2060    }
2061#endif
2062#ifdef CONFIG_FRONTSWAP
2063    if (zcache_enabled && use_frontswap) {
2064        struct frontswap_ops old_ops;
2065
2066        old_ops = zcache_frontswap_register_ops();
2067        pr_info("zcache: frontswap enabled using kernel "
2068            "transcendent memory and zsmalloc\n");
2069        if (old_ops.init != NULL)
2070            pr_warning("zcache: frontswap_ops overridden");
2071    }
2072#endif
2073out:
2074    return ret;
2075}
2076
2077module_init(zcache_init)
2078

Archive Download this file



interactive