blob: ae8cd6d5af8b2998ec34fafe9a51d358ab6db552 [file] [log] [blame]
Javier Gonzáleza4bd2172017-04-15 20:55:50 +02001/*
2 * Copyright (C) 2015 IT University of Copenhagen (rrpc.c)
3 * Copyright (C) 2016 CNEX Labs
4 * Initial release: Javier Gonzalez <javier@cnexlabs.com>
5 * Matias Bjorling <matias@cnexlabs.com>
6 *
7 * This program is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU General Public License version
9 * 2 as published by the Free Software Foundation.
10 *
11 * This program is distributed in the hope that it will be useful, but
12 * WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 * General Public License for more details.
15 *
16 * Implementation of a physical block-device target for Open-channel SSDs.
17 *
18 * pblk-init.c - pblk's initialization.
19 */
20
21#include "pblk.h"
22
23static struct kmem_cache *pblk_blk_ws_cache, *pblk_rec_cache, *pblk_r_rq_cache,
24 *pblk_w_rq_cache, *pblk_line_meta_cache;
25static DECLARE_RWSEM(pblk_lock);
26
27static int pblk_rw_io(struct request_queue *q, struct pblk *pblk,
28 struct bio *bio)
29{
30 int ret;
31
32 /* Read requests must be <= 256kb due to NVMe's 64 bit completion bitmap
33 * constraint. Writes can be of arbitrary size.
34 */
35 if (bio_data_dir(bio) == READ) {
36 blk_queue_split(q, &bio, q->bio_split);
37 ret = pblk_submit_read(pblk, bio);
38 if (ret == NVM_IO_DONE && bio_flagged(bio, BIO_CLONED))
39 bio_put(bio);
40
41 return ret;
42 }
43
44 /* Prevent deadlock in the case of a modest LUN configuration and large
45 * user I/Os. Unless stalled, the rate limiter leaves at least 256KB
46 * available for user I/O.
47 */
48 if (unlikely(pblk_get_secs(bio) >= pblk_rl_sysfs_rate_show(&pblk->rl)))
49 blk_queue_split(q, &bio, q->bio_split);
50
51 return pblk_write_to_cache(pblk, bio, PBLK_IOTYPE_USER);
52}
53
54static blk_qc_t pblk_make_rq(struct request_queue *q, struct bio *bio)
55{
56 struct pblk *pblk = q->queuedata;
57
58 if (bio_op(bio) == REQ_OP_DISCARD) {
59 pblk_discard(pblk, bio);
60 if (!(bio->bi_opf & REQ_PREFLUSH)) {
61 bio_endio(bio);
62 return BLK_QC_T_NONE;
63 }
64 }
65
66 switch (pblk_rw_io(q, pblk, bio)) {
67 case NVM_IO_ERR:
68 bio_io_error(bio);
69 break;
70 case NVM_IO_DONE:
71 bio_endio(bio);
72 break;
73 }
74
75 return BLK_QC_T_NONE;
76}
77
78static void pblk_l2p_free(struct pblk *pblk)
79{
80 vfree(pblk->trans_map);
81}
82
83static int pblk_l2p_init(struct pblk *pblk)
84{
85 sector_t i;
86 struct ppa_addr ppa;
87 int entry_size = 8;
88
89 if (pblk->ppaf_bitsize < 32)
90 entry_size = 4;
91
92 pblk->trans_map = vmalloc(entry_size * pblk->rl.nr_secs);
93 if (!pblk->trans_map)
94 return -ENOMEM;
95
96 pblk_ppa_set_empty(&ppa);
97
98 for (i = 0; i < pblk->rl.nr_secs; i++)
99 pblk_trans_map_set(pblk, i, ppa);
100
101 return 0;
102}
103
104static void pblk_rwb_free(struct pblk *pblk)
105{
106 if (pblk_rb_tear_down_check(&pblk->rwb))
107 pr_err("pblk: write buffer error on tear down\n");
108
109 pblk_rb_data_free(&pblk->rwb);
110 vfree(pblk_rb_entries_ref(&pblk->rwb));
111}
112
113static int pblk_rwb_init(struct pblk *pblk)
114{
115 struct nvm_tgt_dev *dev = pblk->dev;
116 struct nvm_geo *geo = &dev->geo;
117 struct pblk_rb_entry *entries;
118 unsigned long nr_entries;
119 unsigned int power_size, power_seg_sz;
120
121 nr_entries = pblk_rb_calculate_size(pblk->pgs_in_buffer);
122
123 entries = vzalloc(nr_entries * sizeof(struct pblk_rb_entry));
124 if (!entries)
125 return -ENOMEM;
126
127 power_size = get_count_order(nr_entries);
128 power_seg_sz = get_count_order(geo->sec_size);
129
130 return pblk_rb_init(&pblk->rwb, entries, power_size, power_seg_sz);
131}
132
133/* Minimum pages needed within a lun */
134#define PAGE_POOL_SIZE 16
135#define ADDR_POOL_SIZE 64
136
137static int pblk_set_ppaf(struct pblk *pblk)
138{
139 struct nvm_tgt_dev *dev = pblk->dev;
140 struct nvm_geo *geo = &dev->geo;
141 struct nvm_addr_format ppaf = geo->ppaf;
142 int power_len;
143
144 /* Re-calculate channel and lun format to adapt to configuration */
145 power_len = get_count_order(geo->nr_chnls);
146 if (1 << power_len != geo->nr_chnls) {
147 pr_err("pblk: supports only power-of-two channel config.\n");
148 return -EINVAL;
149 }
150 ppaf.ch_len = power_len;
151
152 power_len = get_count_order(geo->luns_per_chnl);
153 if (1 << power_len != geo->luns_per_chnl) {
154 pr_err("pblk: supports only power-of-two LUN config.\n");
155 return -EINVAL;
156 }
157 ppaf.lun_len = power_len;
158
159 pblk->ppaf.sec_offset = 0;
160 pblk->ppaf.pln_offset = ppaf.sect_len;
161 pblk->ppaf.ch_offset = pblk->ppaf.pln_offset + ppaf.pln_len;
162 pblk->ppaf.lun_offset = pblk->ppaf.ch_offset + ppaf.ch_len;
163 pblk->ppaf.pg_offset = pblk->ppaf.lun_offset + ppaf.lun_len;
164 pblk->ppaf.blk_offset = pblk->ppaf.pg_offset + ppaf.pg_len;
165 pblk->ppaf.sec_mask = (1ULL << ppaf.sect_len) - 1;
166 pblk->ppaf.pln_mask = ((1ULL << ppaf.pln_len) - 1) <<
167 pblk->ppaf.pln_offset;
168 pblk->ppaf.ch_mask = ((1ULL << ppaf.ch_len) - 1) <<
169 pblk->ppaf.ch_offset;
170 pblk->ppaf.lun_mask = ((1ULL << ppaf.lun_len) - 1) <<
171 pblk->ppaf.lun_offset;
172 pblk->ppaf.pg_mask = ((1ULL << ppaf.pg_len) - 1) <<
173 pblk->ppaf.pg_offset;
174 pblk->ppaf.blk_mask = ((1ULL << ppaf.blk_len) - 1) <<
175 pblk->ppaf.blk_offset;
176
177 pblk->ppaf_bitsize = pblk->ppaf.blk_offset + ppaf.blk_len;
178
179 return 0;
180}
181
182static int pblk_init_global_caches(struct pblk *pblk)
183{
184 char cache_name[PBLK_CACHE_NAME_LEN];
185
186 down_write(&pblk_lock);
187 pblk_blk_ws_cache = kmem_cache_create("pblk_blk_ws",
188 sizeof(struct pblk_line_ws), 0, 0, NULL);
189 if (!pblk_blk_ws_cache) {
190 up_write(&pblk_lock);
191 return -ENOMEM;
192 }
193
194 pblk_rec_cache = kmem_cache_create("pblk_rec",
195 sizeof(struct pblk_rec_ctx), 0, 0, NULL);
196 if (!pblk_rec_cache) {
197 kmem_cache_destroy(pblk_blk_ws_cache);
198 up_write(&pblk_lock);
199 return -ENOMEM;
200 }
201
202 pblk_r_rq_cache = kmem_cache_create("pblk_r_rq", pblk_r_rq_size,
203 0, 0, NULL);
204 if (!pblk_r_rq_cache) {
205 kmem_cache_destroy(pblk_blk_ws_cache);
206 kmem_cache_destroy(pblk_rec_cache);
207 up_write(&pblk_lock);
208 return -ENOMEM;
209 }
210
211 pblk_w_rq_cache = kmem_cache_create("pblk_w_rq", pblk_w_rq_size,
212 0, 0, NULL);
213 if (!pblk_w_rq_cache) {
214 kmem_cache_destroy(pblk_blk_ws_cache);
215 kmem_cache_destroy(pblk_rec_cache);
216 kmem_cache_destroy(pblk_r_rq_cache);
217 up_write(&pblk_lock);
218 return -ENOMEM;
219 }
220
221 snprintf(cache_name, sizeof(cache_name), "pblk_line_m_%s",
222 pblk->disk->disk_name);
223 pblk_line_meta_cache = kmem_cache_create(cache_name,
224 pblk->lm.sec_bitmap_len, 0, 0, NULL);
225 if (!pblk_line_meta_cache) {
226 kmem_cache_destroy(pblk_blk_ws_cache);
227 kmem_cache_destroy(pblk_rec_cache);
228 kmem_cache_destroy(pblk_r_rq_cache);
229 kmem_cache_destroy(pblk_w_rq_cache);
230 up_write(&pblk_lock);
231 return -ENOMEM;
232 }
233 up_write(&pblk_lock);
234
235 return 0;
236}
237
238static int pblk_core_init(struct pblk *pblk)
239{
240 struct nvm_tgt_dev *dev = pblk->dev;
241 struct nvm_geo *geo = &dev->geo;
242 int max_write_ppas;
243 int mod;
244
245 pblk->min_write_pgs = geo->sec_per_pl * (geo->sec_size / PAGE_SIZE);
246 max_write_ppas = pblk->min_write_pgs * geo->nr_luns;
247 pblk->max_write_pgs = (max_write_ppas < nvm_max_phys_sects(dev)) ?
248 max_write_ppas : nvm_max_phys_sects(dev);
249 pblk->pgs_in_buffer = NVM_MEM_PAGE_WRITE * geo->sec_per_pg *
250 geo->nr_planes * geo->nr_luns;
251
252 if (pblk->max_write_pgs > PBLK_MAX_REQ_ADDRS) {
253 pr_err("pblk: cannot support device max_phys_sect\n");
254 return -EINVAL;
255 }
256
257 div_u64_rem(geo->sec_per_blk, pblk->min_write_pgs, &mod);
258 if (mod) {
259 pr_err("pblk: bad configuration of sectors/pages\n");
260 return -EINVAL;
261 }
262
263 if (pblk_init_global_caches(pblk))
264 return -ENOMEM;
265
266 pblk->page_pool = mempool_create_page_pool(PAGE_POOL_SIZE, 0);
267 if (!pblk->page_pool)
268 return -ENOMEM;
269
270 pblk->line_ws_pool = mempool_create_slab_pool(geo->nr_luns,
271 pblk_blk_ws_cache);
272 if (!pblk->line_ws_pool)
273 goto free_page_pool;
274
275 pblk->rec_pool = mempool_create_slab_pool(geo->nr_luns, pblk_rec_cache);
276 if (!pblk->rec_pool)
277 goto free_blk_ws_pool;
278
279 pblk->r_rq_pool = mempool_create_slab_pool(64, pblk_r_rq_cache);
280 if (!pblk->r_rq_pool)
281 goto free_rec_pool;
282
283 pblk->w_rq_pool = mempool_create_slab_pool(64, pblk_w_rq_cache);
284 if (!pblk->w_rq_pool)
285 goto free_r_rq_pool;
286
287 pblk->line_meta_pool =
288 mempool_create_slab_pool(16, pblk_line_meta_cache);
289 if (!pblk->line_meta_pool)
290 goto free_w_rq_pool;
291
292 pblk->kw_wq = alloc_workqueue("pblk-aux-wq",
293 WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
294 if (!pblk->kw_wq)
295 goto free_line_meta_pool;
296
297 if (pblk_set_ppaf(pblk))
298 goto free_kw_wq;
299
300 if (pblk_rwb_init(pblk))
301 goto free_kw_wq;
302
303 INIT_LIST_HEAD(&pblk->compl_list);
304 return 0;
305
306free_kw_wq:
307 destroy_workqueue(pblk->kw_wq);
308free_line_meta_pool:
309 mempool_destroy(pblk->line_meta_pool);
310free_w_rq_pool:
311 mempool_destroy(pblk->w_rq_pool);
312free_r_rq_pool:
313 mempool_destroy(pblk->r_rq_pool);
314free_rec_pool:
315 mempool_destroy(pblk->rec_pool);
316free_blk_ws_pool:
317 mempool_destroy(pblk->line_ws_pool);
318free_page_pool:
319 mempool_destroy(pblk->page_pool);
320 return -ENOMEM;
321}
322
323static void pblk_core_free(struct pblk *pblk)
324{
325 if (pblk->kw_wq)
326 destroy_workqueue(pblk->kw_wq);
327
328 mempool_destroy(pblk->page_pool);
329 mempool_destroy(pblk->line_ws_pool);
330 mempool_destroy(pblk->rec_pool);
331 mempool_destroy(pblk->r_rq_pool);
332 mempool_destroy(pblk->w_rq_pool);
333 mempool_destroy(pblk->line_meta_pool);
334
335 kmem_cache_destroy(pblk_blk_ws_cache);
336 kmem_cache_destroy(pblk_rec_cache);
337 kmem_cache_destroy(pblk_r_rq_cache);
338 kmem_cache_destroy(pblk_w_rq_cache);
339 kmem_cache_destroy(pblk_line_meta_cache);
340}
341
342static void pblk_luns_free(struct pblk *pblk)
343{
344 kfree(pblk->luns);
345}
346
347static void pblk_lines_free(struct pblk *pblk)
348{
349 struct pblk_line_mgmt *l_mg = &pblk->l_mg;
350 struct pblk_line *line;
351 int i;
352
353 spin_lock(&l_mg->free_lock);
354 for (i = 0; i < l_mg->nr_lines; i++) {
355 line = &pblk->lines[i];
356
357 pblk_line_free(pblk, line);
358 kfree(line->blk_bitmap);
359 kfree(line->erase_bitmap);
360 }
361 spin_unlock(&l_mg->free_lock);
362}
363
364static void pblk_line_meta_free(struct pblk *pblk)
365{
366 struct pblk_line_mgmt *l_mg = &pblk->l_mg;
367 int i;
368
369 kfree(l_mg->bb_template);
370 kfree(l_mg->bb_aux);
371
372 for (i = 0; i < PBLK_DATA_LINES; i++) {
373 pblk_mfree(l_mg->sline_meta[i].meta, l_mg->smeta_alloc_type);
374 pblk_mfree(l_mg->eline_meta[i].meta, l_mg->emeta_alloc_type);
375 }
376
377 kfree(pblk->lines);
378}
379
380static int pblk_bb_discovery(struct nvm_tgt_dev *dev, struct pblk_lun *rlun)
381{
382 struct nvm_geo *geo = &dev->geo;
383 struct ppa_addr ppa;
384 u8 *blks;
385 int nr_blks, ret;
386
387 nr_blks = geo->blks_per_lun * geo->plane_mode;
388 blks = kmalloc(nr_blks, GFP_KERNEL);
389 if (!blks)
390 return -ENOMEM;
391
392 ppa.ppa = 0;
393 ppa.g.ch = rlun->bppa.g.ch;
394 ppa.g.lun = rlun->bppa.g.lun;
395
396 ret = nvm_get_tgt_bb_tbl(dev, ppa, blks);
397 if (ret)
398 goto out;
399
400 nr_blks = nvm_bb_tbl_fold(dev->parent, blks, nr_blks);
401 if (nr_blks < 0) {
Javier Gonzáleza4bd2172017-04-15 20:55:50 +0200402 ret = nr_blks;
Wei Yongjun5136a4f2017-04-25 16:15:30 +0000403 goto out;
Javier Gonzáleza4bd2172017-04-15 20:55:50 +0200404 }
405
406 rlun->bb_list = blks;
407
Wei Yongjun5136a4f2017-04-25 16:15:30 +0000408 return 0;
Javier Gonzáleza4bd2172017-04-15 20:55:50 +0200409out:
Wei Yongjun5136a4f2017-04-25 16:15:30 +0000410 kfree(blks);
Javier Gonzáleza4bd2172017-04-15 20:55:50 +0200411 return ret;
412}
413
414static int pblk_bb_line(struct pblk *pblk, struct pblk_line *line)
415{
416 struct pblk_line_meta *lm = &pblk->lm;
417 struct pblk_lun *rlun;
418 int bb_cnt = 0;
419 int i;
420
421 line->blk_bitmap = kzalloc(lm->blk_bitmap_len, GFP_KERNEL);
422 if (!line->blk_bitmap)
423 return -ENOMEM;
424
425 line->erase_bitmap = kzalloc(lm->blk_bitmap_len, GFP_KERNEL);
426 if (!line->erase_bitmap) {
427 kfree(line->blk_bitmap);
428 return -ENOMEM;
429 }
430
431 for (i = 0; i < lm->blk_per_line; i++) {
432 rlun = &pblk->luns[i];
433 if (rlun->bb_list[line->id] == NVM_BLK_T_FREE)
434 continue;
435
436 set_bit(i, line->blk_bitmap);
437 bb_cnt++;
438 }
439
440 return bb_cnt;
441}
442
443static int pblk_luns_init(struct pblk *pblk, struct ppa_addr *luns)
444{
445 struct nvm_tgt_dev *dev = pblk->dev;
446 struct nvm_geo *geo = &dev->geo;
447 struct pblk_lun *rlun;
448 int i, ret;
449
450 /* TODO: Implement unbalanced LUN support */
451 if (geo->luns_per_chnl < 0) {
452 pr_err("pblk: unbalanced LUN config.\n");
453 return -EINVAL;
454 }
455
456 pblk->luns = kcalloc(geo->nr_luns, sizeof(struct pblk_lun), GFP_KERNEL);
457 if (!pblk->luns)
458 return -ENOMEM;
459
460 for (i = 0; i < geo->nr_luns; i++) {
461 /* Stripe across channels */
462 int ch = i % geo->nr_chnls;
463 int lun_raw = i / geo->nr_chnls;
464 int lunid = lun_raw + ch * geo->luns_per_chnl;
465
466 rlun = &pblk->luns[i];
467 rlun->bppa = luns[lunid];
468
469 sema_init(&rlun->wr_sem, 1);
470
471 ret = pblk_bb_discovery(dev, rlun);
472 if (ret) {
473 while (--i >= 0)
474 kfree(pblk->luns[i].bb_list);
475 return ret;
476 }
477 }
478
479 return 0;
480}
481
482static int pblk_lines_configure(struct pblk *pblk, int flags)
483{
484 struct pblk_line *line = NULL;
485 int ret = 0;
486
487 if (!(flags & NVM_TARGET_FACTORY)) {
488 line = pblk_recov_l2p(pblk);
489 if (IS_ERR(line)) {
490 pr_err("pblk: could not recover l2p table\n");
491 ret = -EFAULT;
492 }
493 }
494
495 if (!line) {
496 /* Configure next line for user data */
497 line = pblk_line_get_first_data(pblk);
498 if (!line) {
499 pr_err("pblk: line list corrupted\n");
500 ret = -EFAULT;
501 }
502 }
503
504 return ret;
505}
506
507/* See comment over struct line_emeta definition */
508static unsigned int calc_emeta_len(struct pblk *pblk, struct pblk_line_meta *lm)
509{
510 return (sizeof(struct line_emeta) +
511 ((lm->sec_per_line - lm->emeta_sec) * sizeof(u64)) +
512 (pblk->l_mg.nr_lines * sizeof(u32)) +
513 lm->blk_bitmap_len);
514}
515
516static void pblk_set_provision(struct pblk *pblk, long nr_free_blks)
517{
518 struct nvm_tgt_dev *dev = pblk->dev;
519 struct nvm_geo *geo = &dev->geo;
520 sector_t provisioned;
521
522 pblk->over_pct = 20;
523
524 provisioned = nr_free_blks;
525 provisioned *= (100 - pblk->over_pct);
526 sector_div(provisioned, 100);
527
528 /* Internally pblk manages all free blocks, but all calculations based
529 * on user capacity consider only provisioned blocks
530 */
531 pblk->rl.total_blocks = nr_free_blks;
532 pblk->rl.nr_secs = nr_free_blks * geo->sec_per_blk;
533 pblk->capacity = provisioned * geo->sec_per_blk;
534 atomic_set(&pblk->rl.free_blocks, nr_free_blks);
535}
536
537static int pblk_lines_init(struct pblk *pblk)
538{
539 struct nvm_tgt_dev *dev = pblk->dev;
540 struct nvm_geo *geo = &dev->geo;
541 struct pblk_line_mgmt *l_mg = &pblk->l_mg;
542 struct pblk_line_meta *lm = &pblk->lm;
543 struct pblk_line *line;
544 unsigned int smeta_len, emeta_len;
545 long nr_bad_blks, nr_meta_blks, nr_free_blks;
546 int bb_distance;
547 int i;
Dan Carpenter1c6286f2017-04-15 20:55:53 +0200548 int ret;
Javier Gonzáleza4bd2172017-04-15 20:55:50 +0200549
550 lm->sec_per_line = geo->sec_per_blk * geo->nr_luns;
551 lm->blk_per_line = geo->nr_luns;
552 lm->blk_bitmap_len = BITS_TO_LONGS(geo->nr_luns) * sizeof(long);
553 lm->sec_bitmap_len = BITS_TO_LONGS(lm->sec_per_line) * sizeof(long);
554 lm->lun_bitmap_len = BITS_TO_LONGS(geo->nr_luns) * sizeof(long);
555 lm->high_thrs = lm->sec_per_line / 2;
556 lm->mid_thrs = lm->sec_per_line / 4;
557
558 /* Calculate necessary pages for smeta. See comment over struct
559 * line_smeta definition
560 */
561 lm->smeta_len = sizeof(struct line_smeta) +
562 PBLK_LINE_NR_LUN_BITMAP * lm->lun_bitmap_len;
563
564 i = 1;
565add_smeta_page:
566 lm->smeta_sec = i * geo->sec_per_pl;
567 lm->smeta_len = lm->smeta_sec * geo->sec_size;
568
569 smeta_len = sizeof(struct line_smeta) +
570 PBLK_LINE_NR_LUN_BITMAP * lm->lun_bitmap_len;
571 if (smeta_len > lm->smeta_len) {
572 i++;
573 goto add_smeta_page;
574 }
575
576 /* Calculate necessary pages for emeta. See comment over struct
577 * line_emeta definition
578 */
579 i = 1;
580add_emeta_page:
581 lm->emeta_sec = i * geo->sec_per_pl;
582 lm->emeta_len = lm->emeta_sec * geo->sec_size;
583
584 emeta_len = calc_emeta_len(pblk, lm);
585 if (emeta_len > lm->emeta_len) {
586 i++;
587 goto add_emeta_page;
588 }
589 lm->emeta_bb = geo->nr_luns - i;
590
591 nr_meta_blks = (lm->smeta_sec + lm->emeta_sec +
592 (geo->sec_per_blk / 2)) / geo->sec_per_blk;
593 lm->min_blk_line = nr_meta_blks + 1;
594
595 l_mg->nr_lines = geo->blks_per_lun;
596 l_mg->log_line = l_mg->data_line = NULL;
597 l_mg->l_seq_nr = l_mg->d_seq_nr = 0;
598 l_mg->nr_free_lines = 0;
599 bitmap_zero(&l_mg->meta_bitmap, PBLK_DATA_LINES);
600
601 /* smeta is always small enough to fit on a kmalloc memory allocation,
602 * emeta depends on the number of LUNs allocated to the pblk instance
603 */
604 l_mg->smeta_alloc_type = PBLK_KMALLOC_META;
605 for (i = 0; i < PBLK_DATA_LINES; i++) {
606 l_mg->sline_meta[i].meta = kmalloc(lm->smeta_len, GFP_KERNEL);
607 if (!l_mg->sline_meta[i].meta)
608 while (--i >= 0) {
609 kfree(l_mg->sline_meta[i].meta);
610 ret = -ENOMEM;
611 goto fail;
612 }
613 }
614
615 if (lm->emeta_len > KMALLOC_MAX_CACHE_SIZE) {
616 l_mg->emeta_alloc_type = PBLK_VMALLOC_META;
617
618 for (i = 0; i < PBLK_DATA_LINES; i++) {
619 l_mg->eline_meta[i].meta = vmalloc(lm->emeta_len);
620 if (!l_mg->eline_meta[i].meta)
621 while (--i >= 0) {
622 vfree(l_mg->eline_meta[i].meta);
623 ret = -ENOMEM;
624 goto fail;
625 }
626 }
627 } else {
628 l_mg->emeta_alloc_type = PBLK_KMALLOC_META;
629
630 for (i = 0; i < PBLK_DATA_LINES; i++) {
631 l_mg->eline_meta[i].meta =
632 kmalloc(lm->emeta_len, GFP_KERNEL);
633 if (!l_mg->eline_meta[i].meta)
634 while (--i >= 0) {
635 kfree(l_mg->eline_meta[i].meta);
636 ret = -ENOMEM;
637 goto fail;
638 }
639 }
640 }
641
642 l_mg->bb_template = kzalloc(lm->sec_bitmap_len, GFP_KERNEL);
Dan Carpenter1c6286f2017-04-15 20:55:53 +0200643 if (!l_mg->bb_template) {
644 ret = -ENOMEM;
Javier Gonzáleza4bd2172017-04-15 20:55:50 +0200645 goto fail_free_meta;
Dan Carpenter1c6286f2017-04-15 20:55:53 +0200646 }
Javier Gonzáleza4bd2172017-04-15 20:55:50 +0200647
648 l_mg->bb_aux = kzalloc(lm->sec_bitmap_len, GFP_KERNEL);
Dan Carpenter1c6286f2017-04-15 20:55:53 +0200649 if (!l_mg->bb_aux) {
650 ret = -ENOMEM;
Javier Gonzáleza4bd2172017-04-15 20:55:50 +0200651 goto fail_free_bb_template;
Dan Carpenter1c6286f2017-04-15 20:55:53 +0200652 }
Javier Gonzáleza4bd2172017-04-15 20:55:50 +0200653
654 bb_distance = (geo->nr_luns) * geo->sec_per_pl;
655 for (i = 0; i < lm->sec_per_line; i += bb_distance)
656 bitmap_set(l_mg->bb_template, i, geo->sec_per_pl);
657
658 INIT_LIST_HEAD(&l_mg->free_list);
659 INIT_LIST_HEAD(&l_mg->corrupt_list);
660 INIT_LIST_HEAD(&l_mg->bad_list);
661 INIT_LIST_HEAD(&l_mg->gc_full_list);
662 INIT_LIST_HEAD(&l_mg->gc_high_list);
663 INIT_LIST_HEAD(&l_mg->gc_mid_list);
664 INIT_LIST_HEAD(&l_mg->gc_low_list);
665 INIT_LIST_HEAD(&l_mg->gc_empty_list);
666
667 l_mg->gc_lists[0] = &l_mg->gc_high_list;
668 l_mg->gc_lists[1] = &l_mg->gc_mid_list;
669 l_mg->gc_lists[2] = &l_mg->gc_low_list;
670
671 spin_lock_init(&l_mg->free_lock);
672 spin_lock_init(&l_mg->gc_lock);
673
674 pblk->lines = kcalloc(l_mg->nr_lines, sizeof(struct pblk_line),
675 GFP_KERNEL);
Dan Carpenter1c6286f2017-04-15 20:55:53 +0200676 if (!pblk->lines) {
677 ret = -ENOMEM;
Javier Gonzáleza4bd2172017-04-15 20:55:50 +0200678 goto fail_free_bb_aux;
Dan Carpenter1c6286f2017-04-15 20:55:53 +0200679 }
Javier Gonzáleza4bd2172017-04-15 20:55:50 +0200680
681 nr_free_blks = 0;
682 for (i = 0; i < l_mg->nr_lines; i++) {
Javier Gonzáleza44f53f2017-04-22 01:32:49 +0200683 int blk_in_line;
684
Javier Gonzáleza4bd2172017-04-15 20:55:50 +0200685 line = &pblk->lines[i];
686
687 line->pblk = pblk;
688 line->id = i;
689 line->type = PBLK_LINETYPE_FREE;
690 line->state = PBLK_LINESTATE_FREE;
691 line->gc_group = PBLK_LINEGC_NONE;
692 spin_lock_init(&line->lock);
693
694 nr_bad_blks = pblk_bb_line(pblk, line);
Dan Carpenter1c6286f2017-04-15 20:55:53 +0200695 if (nr_bad_blks < 0 || nr_bad_blks > lm->blk_per_line) {
696 ret = -EINVAL;
Javier Gonzáleza4bd2172017-04-15 20:55:50 +0200697 goto fail_free_lines;
Dan Carpenter1c6286f2017-04-15 20:55:53 +0200698 }
Javier Gonzáleza4bd2172017-04-15 20:55:50 +0200699
Javier Gonzáleza44f53f2017-04-22 01:32:49 +0200700 blk_in_line = lm->blk_per_line - nr_bad_blks;
701 if (blk_in_line < lm->min_blk_line) {
Javier Gonzáleza4bd2172017-04-15 20:55:50 +0200702 line->state = PBLK_LINESTATE_BAD;
703 list_add_tail(&line->list, &l_mg->bad_list);
704 continue;
705 }
706
Javier Gonzáleza44f53f2017-04-22 01:32:49 +0200707 nr_free_blks += blk_in_line;
708 atomic_set(&line->blk_in_line, blk_in_line);
Javier Gonzáleza4bd2172017-04-15 20:55:50 +0200709
710 l_mg->nr_free_lines++;
711 list_add_tail(&line->list, &l_mg->free_list);
712 }
713
714 pblk_set_provision(pblk, nr_free_blks);
715
716 sema_init(&pblk->erase_sem, 1);
717
718 /* Cleanup per-LUN bad block lists - managed within lines on run-time */
719 for (i = 0; i < geo->nr_luns; i++)
720 kfree(pblk->luns[i].bb_list);
721
722 return 0;
723fail_free_lines:
724 kfree(pblk->lines);
725fail_free_bb_aux:
726 kfree(l_mg->bb_aux);
727fail_free_bb_template:
728 kfree(l_mg->bb_template);
729fail_free_meta:
730 for (i = 0; i < PBLK_DATA_LINES; i++) {
731 pblk_mfree(l_mg->sline_meta[i].meta, l_mg->smeta_alloc_type);
732 pblk_mfree(l_mg->eline_meta[i].meta, l_mg->emeta_alloc_type);
733 }
734fail:
735 for (i = 0; i < geo->nr_luns; i++)
736 kfree(pblk->luns[i].bb_list);
737
738 return ret;
739}
740
741static int pblk_writer_init(struct pblk *pblk)
742{
743 setup_timer(&pblk->wtimer, pblk_write_timer_fn, (unsigned long)pblk);
744 mod_timer(&pblk->wtimer, jiffies + msecs_to_jiffies(100));
745
746 pblk->writer_ts = kthread_create(pblk_write_ts, pblk, "pblk-writer-t");
747 if (IS_ERR(pblk->writer_ts)) {
748 pr_err("pblk: could not allocate writer kthread\n");
Dan Carpenter1c6286f2017-04-15 20:55:53 +0200749 return PTR_ERR(pblk->writer_ts);
Javier Gonzáleza4bd2172017-04-15 20:55:50 +0200750 }
751
752 return 0;
753}
754
755static void pblk_writer_stop(struct pblk *pblk)
756{
757 if (pblk->writer_ts)
758 kthread_stop(pblk->writer_ts);
759 del_timer(&pblk->wtimer);
760}
761
762static void pblk_free(struct pblk *pblk)
763{
764 pblk_luns_free(pblk);
765 pblk_lines_free(pblk);
766 pblk_line_meta_free(pblk);
767 pblk_core_free(pblk);
768 pblk_l2p_free(pblk);
769
770 kfree(pblk);
771}
772
773static void pblk_tear_down(struct pblk *pblk)
774{
775 pblk_flush_writer(pblk);
776 pblk_writer_stop(pblk);
777 pblk_rb_sync_l2p(&pblk->rwb);
778 pblk_recov_pad(pblk);
779 pblk_rwb_free(pblk);
780 pblk_rl_free(&pblk->rl);
781
782 pr_debug("pblk: consistent tear down\n");
783}
784
785static void pblk_exit(void *private)
786{
787 struct pblk *pblk = private;
788
789 down_write(&pblk_lock);
790 pblk_gc_exit(pblk);
791 pblk_tear_down(pblk);
792 pblk_free(pblk);
793 up_write(&pblk_lock);
794}
795
796static sector_t pblk_capacity(void *private)
797{
798 struct pblk *pblk = private;
799
800 return pblk->capacity * NR_PHY_IN_LOG;
801}
802
803static void *pblk_init(struct nvm_tgt_dev *dev, struct gendisk *tdisk,
804 int flags)
805{
806 struct nvm_geo *geo = &dev->geo;
807 struct request_queue *bqueue = dev->q;
808 struct request_queue *tqueue = tdisk->queue;
809 struct pblk *pblk;
810 int ret;
811
812 if (dev->identity.dom & NVM_RSP_L2P) {
813 pr_err("pblk: device-side L2P table not supported. (%x)\n",
814 dev->identity.dom);
815 return ERR_PTR(-EINVAL);
816 }
817
818 pblk = kzalloc(sizeof(struct pblk), GFP_KERNEL);
819 if (!pblk)
820 return ERR_PTR(-ENOMEM);
821
822 pblk->dev = dev;
823 pblk->disk = tdisk;
824
825 spin_lock_init(&pblk->trans_lock);
826 spin_lock_init(&pblk->lock);
827
828 if (flags & NVM_TARGET_FACTORY)
829 pblk_setup_uuid(pblk);
830
831#ifdef CONFIG_NVM_DEBUG
832 atomic_long_set(&pblk->inflight_writes, 0);
833 atomic_long_set(&pblk->padded_writes, 0);
834 atomic_long_set(&pblk->padded_wb, 0);
835 atomic_long_set(&pblk->nr_flush, 0);
836 atomic_long_set(&pblk->req_writes, 0);
837 atomic_long_set(&pblk->sub_writes, 0);
838 atomic_long_set(&pblk->sync_writes, 0);
839 atomic_long_set(&pblk->compl_writes, 0);
840 atomic_long_set(&pblk->inflight_reads, 0);
841 atomic_long_set(&pblk->sync_reads, 0);
842 atomic_long_set(&pblk->recov_writes, 0);
843 atomic_long_set(&pblk->recov_writes, 0);
844 atomic_long_set(&pblk->recov_gc_writes, 0);
845#endif
846
847 atomic_long_set(&pblk->read_failed, 0);
848 atomic_long_set(&pblk->read_empty, 0);
849 atomic_long_set(&pblk->read_high_ecc, 0);
850 atomic_long_set(&pblk->read_failed_gc, 0);
851 atomic_long_set(&pblk->write_failed, 0);
852 atomic_long_set(&pblk->erase_failed, 0);
853
854 ret = pblk_luns_init(pblk, dev->luns);
855 if (ret) {
856 pr_err("pblk: could not initialize luns\n");
857 goto fail;
858 }
859
860 ret = pblk_lines_init(pblk);
861 if (ret) {
862 pr_err("pblk: could not initialize lines\n");
863 goto fail_free_luns;
864 }
865
866 ret = pblk_core_init(pblk);
867 if (ret) {
868 pr_err("pblk: could not initialize core\n");
869 goto fail_free_line_meta;
870 }
871
872 ret = pblk_l2p_init(pblk);
873 if (ret) {
874 pr_err("pblk: could not initialize maps\n");
875 goto fail_free_core;
876 }
877
878 ret = pblk_lines_configure(pblk, flags);
879 if (ret) {
880 pr_err("pblk: could not configure lines\n");
881 goto fail_free_l2p;
882 }
883
884 ret = pblk_writer_init(pblk);
885 if (ret) {
886 pr_err("pblk: could not initialize write thread\n");
887 goto fail_free_lines;
888 }
889
890 ret = pblk_gc_init(pblk);
891 if (ret) {
892 pr_err("pblk: could not initialize gc\n");
893 goto fail_stop_writer;
894 }
895
896 /* inherit the size from the underlying device */
897 blk_queue_logical_block_size(tqueue, queue_physical_block_size(bqueue));
898 blk_queue_max_hw_sectors(tqueue, queue_max_hw_sectors(bqueue));
899
900 blk_queue_write_cache(tqueue, true, false);
901
902 tqueue->limits.discard_granularity = geo->pgs_per_blk * geo->pfpg_size;
903 tqueue->limits.discard_alignment = 0;
904 blk_queue_max_discard_sectors(tqueue, UINT_MAX >> 9);
905 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, tqueue);
906
907 pr_info("pblk init: luns:%u, lines:%d, secs:%llu, buf entries:%u\n",
908 geo->nr_luns, pblk->l_mg.nr_lines,
909 (unsigned long long)pblk->rl.nr_secs,
910 pblk->rwb.nr_entries);
911
912 wake_up_process(pblk->writer_ts);
913 return pblk;
914
915fail_stop_writer:
916 pblk_writer_stop(pblk);
917fail_free_lines:
918 pblk_lines_free(pblk);
919fail_free_l2p:
920 pblk_l2p_free(pblk);
921fail_free_core:
922 pblk_core_free(pblk);
923fail_free_line_meta:
924 pblk_line_meta_free(pblk);
925fail_free_luns:
926 pblk_luns_free(pblk);
927fail:
928 kfree(pblk);
929 return ERR_PTR(ret);
930}
931
932/* physical block device target */
933static struct nvm_tgt_type tt_pblk = {
934 .name = "pblk",
935 .version = {1, 0, 0},
936
937 .make_rq = pblk_make_rq,
938 .capacity = pblk_capacity,
939
940 .init = pblk_init,
941 .exit = pblk_exit,
942
943 .sysfs_init = pblk_sysfs_init,
944 .sysfs_exit = pblk_sysfs_exit,
945};
946
947static int __init pblk_module_init(void)
948{
949 return nvm_register_tgt_type(&tt_pblk);
950}
951
952static void pblk_module_exit(void)
953{
954 nvm_unregister_tgt_type(&tt_pblk);
955}
956
957module_init(pblk_module_init);
958module_exit(pblk_module_exit);
959MODULE_AUTHOR("Javier Gonzalez <javier@cnexlabs.com>");
960MODULE_AUTHOR("Matias Bjorling <matias@cnexlabs.com>");
961MODULE_LICENSE("GPL v2");
962MODULE_DESCRIPTION("Physical Block-Device for Open-Channel SSDs");