Loading...
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Copyright (C) 2015 IT University of Copenhagen (rrpc.c)
4 * Copyright (C) 2016 CNEX Labs
5 * Initial release: Javier Gonzalez <javier@cnexlabs.com>
6 * Matias Bjorling <matias@cnexlabs.com>
7 *
8 * This program is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU General Public License version
10 * 2 as published by the Free Software Foundation.
11 *
12 * This program is distributed in the hope that it will be useful, but
13 * WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License for more details.
16 *
17 * Implementation of a physical block-device target for Open-channel SSDs.
18 *
19 * pblk-init.c - pblk's initialization.
20 */
21
22#include "pblk.h"
23#include "pblk-trace.h"
24
25static unsigned int write_buffer_size;
26
27module_param(write_buffer_size, uint, 0644);
28MODULE_PARM_DESC(write_buffer_size, "number of entries in a write buffer");
29
30struct pblk_global_caches {
31 struct kmem_cache *ws;
32 struct kmem_cache *rec;
33 struct kmem_cache *g_rq;
34 struct kmem_cache *w_rq;
35
36 struct kref kref;
37
38 struct mutex mutex; /* Ensures consistency between
39 * caches and kref
40 */
41};
42
43static struct pblk_global_caches pblk_caches = {
44 .mutex = __MUTEX_INITIALIZER(pblk_caches.mutex),
45 .kref = KREF_INIT(0),
46};
47
48struct bio_set pblk_bio_set;
49
50static blk_qc_t pblk_submit_bio(struct bio *bio)
51{
52 struct pblk *pblk = bio->bi_disk->queue->queuedata;
53
54 if (bio_op(bio) == REQ_OP_DISCARD) {
55 pblk_discard(pblk, bio);
56 if (!(bio->bi_opf & REQ_PREFLUSH)) {
57 bio_endio(bio);
58 return BLK_QC_T_NONE;
59 }
60 }
61
62 /* Read requests must be <= 256kb due to NVMe's 64 bit completion bitmap
63 * constraint. Writes can be of arbitrary size.
64 */
65 if (bio_data_dir(bio) == READ) {
66 blk_queue_split(&bio);
67 pblk_submit_read(pblk, bio);
68 } else {
69 /* Prevent deadlock in the case of a modest LUN configuration
70 * and large user I/Os. Unless stalled, the rate limiter
71 * leaves at least 256KB available for user I/O.
72 */
73 if (pblk_get_secs(bio) > pblk_rl_max_io(&pblk->rl))
74 blk_queue_split(&bio);
75
76 pblk_write_to_cache(pblk, bio, PBLK_IOTYPE_USER);
77 }
78
79 return BLK_QC_T_NONE;
80}
81
82static const struct block_device_operations pblk_bops = {
83 .owner = THIS_MODULE,
84 .submit_bio = pblk_submit_bio,
85};
86
87
88static size_t pblk_trans_map_size(struct pblk *pblk)
89{
90 int entry_size = 8;
91
92 if (pblk->addrf_len < 32)
93 entry_size = 4;
94
95 return entry_size * pblk->capacity;
96}
97
98#ifdef CONFIG_NVM_PBLK_DEBUG
99static u32 pblk_l2p_crc(struct pblk *pblk)
100{
101 size_t map_size;
102 u32 crc = ~(u32)0;
103
104 map_size = pblk_trans_map_size(pblk);
105 crc = crc32_le(crc, pblk->trans_map, map_size);
106 return crc;
107}
108#endif
109
110static void pblk_l2p_free(struct pblk *pblk)
111{
112 vfree(pblk->trans_map);
113}
114
115static int pblk_l2p_recover(struct pblk *pblk, bool factory_init)
116{
117 struct pblk_line *line = NULL;
118
119 if (factory_init) {
120 guid_gen(&pblk->instance_uuid);
121 } else {
122 line = pblk_recov_l2p(pblk);
123 if (IS_ERR(line)) {
124 pblk_err(pblk, "could not recover l2p table\n");
125 return -EFAULT;
126 }
127 }
128
129#ifdef CONFIG_NVM_PBLK_DEBUG
130 pblk_info(pblk, "init: L2P CRC: %x\n", pblk_l2p_crc(pblk));
131#endif
132
133 /* Free full lines directly as GC has not been started yet */
134 pblk_gc_free_full_lines(pblk);
135
136 if (!line) {
137 /* Configure next line for user data */
138 line = pblk_line_get_first_data(pblk);
139 if (!line)
140 return -EFAULT;
141 }
142
143 return 0;
144}
145
146static int pblk_l2p_init(struct pblk *pblk, bool factory_init)
147{
148 sector_t i;
149 struct ppa_addr ppa;
150 size_t map_size;
151 int ret = 0;
152
153 map_size = pblk_trans_map_size(pblk);
154 pblk->trans_map = __vmalloc(map_size, GFP_KERNEL | __GFP_NOWARN |
155 __GFP_RETRY_MAYFAIL | __GFP_HIGHMEM);
156 if (!pblk->trans_map) {
157 pblk_err(pblk, "failed to allocate L2P (need %zu of memory)\n",
158 map_size);
159 return -ENOMEM;
160 }
161
162 pblk_ppa_set_empty(&ppa);
163
164 for (i = 0; i < pblk->capacity; i++)
165 pblk_trans_map_set(pblk, i, ppa);
166
167 ret = pblk_l2p_recover(pblk, factory_init);
168 if (ret)
169 vfree(pblk->trans_map);
170
171 return ret;
172}
173
174static void pblk_rwb_free(struct pblk *pblk)
175{
176 if (pblk_rb_tear_down_check(&pblk->rwb))
177 pblk_err(pblk, "write buffer error on tear down\n");
178
179 pblk_rb_free(&pblk->rwb);
180}
181
182static int pblk_rwb_init(struct pblk *pblk)
183{
184 struct nvm_tgt_dev *dev = pblk->dev;
185 struct nvm_geo *geo = &dev->geo;
186 unsigned long buffer_size;
187 int pgs_in_buffer, threshold;
188
189 threshold = geo->mw_cunits * geo->all_luns;
190 pgs_in_buffer = (max(geo->mw_cunits, geo->ws_opt) + geo->ws_opt)
191 * geo->all_luns;
192
193 if (write_buffer_size && (write_buffer_size > pgs_in_buffer))
194 buffer_size = write_buffer_size;
195 else
196 buffer_size = pgs_in_buffer;
197
198 return pblk_rb_init(&pblk->rwb, buffer_size, threshold, geo->csecs);
199}
200
201static int pblk_set_addrf_12(struct pblk *pblk, struct nvm_geo *geo,
202 struct nvm_addrf_12 *dst)
203{
204 struct nvm_addrf_12 *src = (struct nvm_addrf_12 *)&geo->addrf;
205 int power_len;
206
207 /* Re-calculate channel and lun format to adapt to configuration */
208 power_len = get_count_order(geo->num_ch);
209 if (1 << power_len != geo->num_ch) {
210 pblk_err(pblk, "supports only power-of-two channel config.\n");
211 return -EINVAL;
212 }
213 dst->ch_len = power_len;
214
215 power_len = get_count_order(geo->num_lun);
216 if (1 << power_len != geo->num_lun) {
217 pblk_err(pblk, "supports only power-of-two LUN config.\n");
218 return -EINVAL;
219 }
220 dst->lun_len = power_len;
221
222 dst->blk_len = src->blk_len;
223 dst->pg_len = src->pg_len;
224 dst->pln_len = src->pln_len;
225 dst->sec_len = src->sec_len;
226
227 dst->sec_offset = 0;
228 dst->pln_offset = dst->sec_len;
229 dst->ch_offset = dst->pln_offset + dst->pln_len;
230 dst->lun_offset = dst->ch_offset + dst->ch_len;
231 dst->pg_offset = dst->lun_offset + dst->lun_len;
232 dst->blk_offset = dst->pg_offset + dst->pg_len;
233
234 dst->sec_mask = ((1ULL << dst->sec_len) - 1) << dst->sec_offset;
235 dst->pln_mask = ((1ULL << dst->pln_len) - 1) << dst->pln_offset;
236 dst->ch_mask = ((1ULL << dst->ch_len) - 1) << dst->ch_offset;
237 dst->lun_mask = ((1ULL << dst->lun_len) - 1) << dst->lun_offset;
238 dst->pg_mask = ((1ULL << dst->pg_len) - 1) << dst->pg_offset;
239 dst->blk_mask = ((1ULL << dst->blk_len) - 1) << dst->blk_offset;
240
241 return dst->blk_offset + src->blk_len;
242}
243
244static int pblk_set_addrf_20(struct nvm_geo *geo, struct nvm_addrf *adst,
245 struct pblk_addrf *udst)
246{
247 struct nvm_addrf *src = &geo->addrf;
248
249 adst->ch_len = get_count_order(geo->num_ch);
250 adst->lun_len = get_count_order(geo->num_lun);
251 adst->chk_len = src->chk_len;
252 adst->sec_len = src->sec_len;
253
254 adst->sec_offset = 0;
255 adst->ch_offset = adst->sec_len;
256 adst->lun_offset = adst->ch_offset + adst->ch_len;
257 adst->chk_offset = adst->lun_offset + adst->lun_len;
258
259 adst->sec_mask = ((1ULL << adst->sec_len) - 1) << adst->sec_offset;
260 adst->chk_mask = ((1ULL << adst->chk_len) - 1) << adst->chk_offset;
261 adst->lun_mask = ((1ULL << adst->lun_len) - 1) << adst->lun_offset;
262 adst->ch_mask = ((1ULL << adst->ch_len) - 1) << adst->ch_offset;
263
264 udst->sec_stripe = geo->ws_opt;
265 udst->ch_stripe = geo->num_ch;
266 udst->lun_stripe = geo->num_lun;
267
268 udst->sec_lun_stripe = udst->sec_stripe * udst->ch_stripe;
269 udst->sec_ws_stripe = udst->sec_lun_stripe * udst->lun_stripe;
270
271 return adst->chk_offset + adst->chk_len;
272}
273
274static int pblk_set_addrf(struct pblk *pblk)
275{
276 struct nvm_tgt_dev *dev = pblk->dev;
277 struct nvm_geo *geo = &dev->geo;
278 int mod;
279
280 switch (geo->version) {
281 case NVM_OCSSD_SPEC_12:
282 div_u64_rem(geo->clba, pblk->min_write_pgs, &mod);
283 if (mod) {
284 pblk_err(pblk, "bad configuration of sectors/pages\n");
285 return -EINVAL;
286 }
287
288 pblk->addrf_len = pblk_set_addrf_12(pblk, geo,
289 (void *)&pblk->addrf);
290 break;
291 case NVM_OCSSD_SPEC_20:
292 pblk->addrf_len = pblk_set_addrf_20(geo, (void *)&pblk->addrf,
293 &pblk->uaddrf);
294 break;
295 default:
296 pblk_err(pblk, "OCSSD revision not supported (%d)\n",
297 geo->version);
298 return -EINVAL;
299 }
300
301 return 0;
302}
303
304static int pblk_create_global_caches(void)
305{
306
307 pblk_caches.ws = kmem_cache_create("pblk_blk_ws",
308 sizeof(struct pblk_line_ws), 0, 0, NULL);
309 if (!pblk_caches.ws)
310 return -ENOMEM;
311
312 pblk_caches.rec = kmem_cache_create("pblk_rec",
313 sizeof(struct pblk_rec_ctx), 0, 0, NULL);
314 if (!pblk_caches.rec)
315 goto fail_destroy_ws;
316
317 pblk_caches.g_rq = kmem_cache_create("pblk_g_rq", pblk_g_rq_size,
318 0, 0, NULL);
319 if (!pblk_caches.g_rq)
320 goto fail_destroy_rec;
321
322 pblk_caches.w_rq = kmem_cache_create("pblk_w_rq", pblk_w_rq_size,
323 0, 0, NULL);
324 if (!pblk_caches.w_rq)
325 goto fail_destroy_g_rq;
326
327 return 0;
328
329fail_destroy_g_rq:
330 kmem_cache_destroy(pblk_caches.g_rq);
331fail_destroy_rec:
332 kmem_cache_destroy(pblk_caches.rec);
333fail_destroy_ws:
334 kmem_cache_destroy(pblk_caches.ws);
335
336 return -ENOMEM;
337}
338
339static int pblk_get_global_caches(void)
340{
341 int ret = 0;
342
343 mutex_lock(&pblk_caches.mutex);
344
345 if (kref_get_unless_zero(&pblk_caches.kref))
346 goto out;
347
348 ret = pblk_create_global_caches();
349 if (!ret)
350 kref_init(&pblk_caches.kref);
351
352out:
353 mutex_unlock(&pblk_caches.mutex);
354 return ret;
355}
356
357static void pblk_destroy_global_caches(struct kref *ref)
358{
359 struct pblk_global_caches *c;
360
361 c = container_of(ref, struct pblk_global_caches, kref);
362
363 kmem_cache_destroy(c->ws);
364 kmem_cache_destroy(c->rec);
365 kmem_cache_destroy(c->g_rq);
366 kmem_cache_destroy(c->w_rq);
367}
368
369static void pblk_put_global_caches(void)
370{
371 mutex_lock(&pblk_caches.mutex);
372 kref_put(&pblk_caches.kref, pblk_destroy_global_caches);
373 mutex_unlock(&pblk_caches.mutex);
374}
375
376static int pblk_core_init(struct pblk *pblk)
377{
378 struct nvm_tgt_dev *dev = pblk->dev;
379 struct nvm_geo *geo = &dev->geo;
380 int ret, max_write_ppas;
381
382 atomic64_set(&pblk->user_wa, 0);
383 atomic64_set(&pblk->pad_wa, 0);
384 atomic64_set(&pblk->gc_wa, 0);
385 pblk->user_rst_wa = 0;
386 pblk->pad_rst_wa = 0;
387 pblk->gc_rst_wa = 0;
388
389 atomic64_set(&pblk->nr_flush, 0);
390 pblk->nr_flush_rst = 0;
391
392 pblk->min_write_pgs = geo->ws_opt;
393 pblk->min_write_pgs_data = pblk->min_write_pgs;
394 max_write_ppas = pblk->min_write_pgs * geo->all_luns;
395 pblk->max_write_pgs = min_t(int, max_write_ppas, NVM_MAX_VLBA);
396 pblk->max_write_pgs = min_t(int, pblk->max_write_pgs,
397 queue_max_hw_sectors(dev->q) / (geo->csecs >> SECTOR_SHIFT));
398 pblk_set_sec_per_write(pblk, pblk->min_write_pgs);
399
400 pblk->oob_meta_size = geo->sos;
401 if (!pblk_is_oob_meta_supported(pblk)) {
402 /* For drives which does not have OOB metadata feature
403 * in order to support recovery feature we need to use
404 * so called packed metadata. Packed metada will store
405 * the same information as OOB metadata (l2p table mapping,
406 * but in the form of the single page at the end of
407 * every write request.
408 */
409 if (pblk->min_write_pgs
410 * sizeof(struct pblk_sec_meta) > PAGE_SIZE) {
411 /* We want to keep all the packed metadata on single
412 * page per write requests. So we need to ensure that
413 * it will fit.
414 *
415 * This is more like sanity check, since there is
416 * no device with such a big minimal write size
417 * (above 1 metabytes).
418 */
419 pblk_err(pblk, "Not supported min write size\n");
420 return -EINVAL;
421 }
422 /* For packed meta approach we do some simplification.
423 * On read path we always issue requests which size
424 * equal to max_write_pgs, with all pages filled with
425 * user payload except of last one page which will be
426 * filled with packed metadata.
427 */
428 pblk->max_write_pgs = pblk->min_write_pgs;
429 pblk->min_write_pgs_data = pblk->min_write_pgs - 1;
430 }
431
432 pblk->pad_dist = kcalloc(pblk->min_write_pgs - 1, sizeof(atomic64_t),
433 GFP_KERNEL);
434 if (!pblk->pad_dist)
435 return -ENOMEM;
436
437 if (pblk_get_global_caches())
438 goto fail_free_pad_dist;
439
440 /* Internal bios can be at most the sectors signaled by the device. */
441 ret = mempool_init_page_pool(&pblk->page_bio_pool, NVM_MAX_VLBA, 0);
442 if (ret)
443 goto free_global_caches;
444
445 ret = mempool_init_slab_pool(&pblk->gen_ws_pool, PBLK_GEN_WS_POOL_SIZE,
446 pblk_caches.ws);
447 if (ret)
448 goto free_page_bio_pool;
449
450 ret = mempool_init_slab_pool(&pblk->rec_pool, geo->all_luns,
451 pblk_caches.rec);
452 if (ret)
453 goto free_gen_ws_pool;
454
455 ret = mempool_init_slab_pool(&pblk->r_rq_pool, geo->all_luns,
456 pblk_caches.g_rq);
457 if (ret)
458 goto free_rec_pool;
459
460 ret = mempool_init_slab_pool(&pblk->e_rq_pool, geo->all_luns,
461 pblk_caches.g_rq);
462 if (ret)
463 goto free_r_rq_pool;
464
465 ret = mempool_init_slab_pool(&pblk->w_rq_pool, geo->all_luns,
466 pblk_caches.w_rq);
467 if (ret)
468 goto free_e_rq_pool;
469
470 pblk->close_wq = alloc_workqueue("pblk-close-wq",
471 WQ_MEM_RECLAIM | WQ_UNBOUND, PBLK_NR_CLOSE_JOBS);
472 if (!pblk->close_wq)
473 goto free_w_rq_pool;
474
475 pblk->bb_wq = alloc_workqueue("pblk-bb-wq",
476 WQ_MEM_RECLAIM | WQ_UNBOUND, 0);
477 if (!pblk->bb_wq)
478 goto free_close_wq;
479
480 pblk->r_end_wq = alloc_workqueue("pblk-read-end-wq",
481 WQ_MEM_RECLAIM | WQ_UNBOUND, 0);
482 if (!pblk->r_end_wq)
483 goto free_bb_wq;
484
485 if (pblk_set_addrf(pblk))
486 goto free_r_end_wq;
487
488 INIT_LIST_HEAD(&pblk->compl_list);
489 INIT_LIST_HEAD(&pblk->resubmit_list);
490
491 return 0;
492
493free_r_end_wq:
494 destroy_workqueue(pblk->r_end_wq);
495free_bb_wq:
496 destroy_workqueue(pblk->bb_wq);
497free_close_wq:
498 destroy_workqueue(pblk->close_wq);
499free_w_rq_pool:
500 mempool_exit(&pblk->w_rq_pool);
501free_e_rq_pool:
502 mempool_exit(&pblk->e_rq_pool);
503free_r_rq_pool:
504 mempool_exit(&pblk->r_rq_pool);
505free_rec_pool:
506 mempool_exit(&pblk->rec_pool);
507free_gen_ws_pool:
508 mempool_exit(&pblk->gen_ws_pool);
509free_page_bio_pool:
510 mempool_exit(&pblk->page_bio_pool);
511free_global_caches:
512 pblk_put_global_caches();
513fail_free_pad_dist:
514 kfree(pblk->pad_dist);
515 return -ENOMEM;
516}
517
518static void pblk_core_free(struct pblk *pblk)
519{
520 if (pblk->close_wq)
521 destroy_workqueue(pblk->close_wq);
522
523 if (pblk->r_end_wq)
524 destroy_workqueue(pblk->r_end_wq);
525
526 if (pblk->bb_wq)
527 destroy_workqueue(pblk->bb_wq);
528
529 mempool_exit(&pblk->page_bio_pool);
530 mempool_exit(&pblk->gen_ws_pool);
531 mempool_exit(&pblk->rec_pool);
532 mempool_exit(&pblk->r_rq_pool);
533 mempool_exit(&pblk->e_rq_pool);
534 mempool_exit(&pblk->w_rq_pool);
535
536 pblk_put_global_caches();
537 kfree(pblk->pad_dist);
538}
539
540static void pblk_line_mg_free(struct pblk *pblk)
541{
542 struct pblk_line_mgmt *l_mg = &pblk->l_mg;
543 int i;
544
545 kfree(l_mg->bb_template);
546 kfree(l_mg->bb_aux);
547 kfree(l_mg->vsc_list);
548
549 for (i = 0; i < PBLK_DATA_LINES; i++) {
550 kfree(l_mg->sline_meta[i]);
551 kvfree(l_mg->eline_meta[i]->buf);
552 kfree(l_mg->eline_meta[i]);
553 }
554
555 mempool_destroy(l_mg->bitmap_pool);
556 kmem_cache_destroy(l_mg->bitmap_cache);
557}
558
559static void pblk_line_meta_free(struct pblk_line_mgmt *l_mg,
560 struct pblk_line *line)
561{
562 struct pblk_w_err_gc *w_err_gc = line->w_err_gc;
563
564 kfree(line->blk_bitmap);
565 kfree(line->erase_bitmap);
566 kfree(line->chks);
567
568 kvfree(w_err_gc->lba_list);
569 kfree(w_err_gc);
570}
571
572static void pblk_lines_free(struct pblk *pblk)
573{
574 struct pblk_line_mgmt *l_mg = &pblk->l_mg;
575 struct pblk_line *line;
576 int i;
577
578 for (i = 0; i < l_mg->nr_lines; i++) {
579 line = &pblk->lines[i];
580
581 pblk_line_free(line);
582 pblk_line_meta_free(l_mg, line);
583 }
584
585 pblk_line_mg_free(pblk);
586
587 kfree(pblk->luns);
588 kfree(pblk->lines);
589}
590
591static int pblk_luns_init(struct pblk *pblk)
592{
593 struct nvm_tgt_dev *dev = pblk->dev;
594 struct nvm_geo *geo = &dev->geo;
595 struct pblk_lun *rlun;
596 int i;
597
598 /* TODO: Implement unbalanced LUN support */
599 if (geo->num_lun < 0) {
600 pblk_err(pblk, "unbalanced LUN config.\n");
601 return -EINVAL;
602 }
603
604 pblk->luns = kcalloc(geo->all_luns, sizeof(struct pblk_lun),
605 GFP_KERNEL);
606 if (!pblk->luns)
607 return -ENOMEM;
608
609 for (i = 0; i < geo->all_luns; i++) {
610 /* Stripe across channels */
611 int ch = i % geo->num_ch;
612 int lun_raw = i / geo->num_ch;
613 int lunid = lun_raw + ch * geo->num_lun;
614
615 rlun = &pblk->luns[i];
616 rlun->bppa = dev->luns[lunid];
617
618 sema_init(&rlun->wr_sem, 1);
619 }
620
621 return 0;
622}
623
624/* See comment over struct line_emeta definition */
625static unsigned int calc_emeta_len(struct pblk *pblk)
626{
627 struct pblk_line_meta *lm = &pblk->lm;
628 struct pblk_line_mgmt *l_mg = &pblk->l_mg;
629 struct nvm_tgt_dev *dev = pblk->dev;
630 struct nvm_geo *geo = &dev->geo;
631
632 /* Round to sector size so that lba_list starts on its own sector */
633 lm->emeta_sec[1] = DIV_ROUND_UP(
634 sizeof(struct line_emeta) + lm->blk_bitmap_len +
635 sizeof(struct wa_counters), geo->csecs);
636 lm->emeta_len[1] = lm->emeta_sec[1] * geo->csecs;
637
638 /* Round to sector size so that vsc_list starts on its own sector */
639 lm->dsec_per_line = lm->sec_per_line - lm->emeta_sec[0];
640 lm->emeta_sec[2] = DIV_ROUND_UP(lm->dsec_per_line * sizeof(u64),
641 geo->csecs);
642 lm->emeta_len[2] = lm->emeta_sec[2] * geo->csecs;
643
644 lm->emeta_sec[3] = DIV_ROUND_UP(l_mg->nr_lines * sizeof(u32),
645 geo->csecs);
646 lm->emeta_len[3] = lm->emeta_sec[3] * geo->csecs;
647
648 lm->vsc_list_len = l_mg->nr_lines * sizeof(u32);
649
650 return (lm->emeta_len[1] + lm->emeta_len[2] + lm->emeta_len[3]);
651}
652
653static int pblk_set_provision(struct pblk *pblk, int nr_free_chks)
654{
655 struct nvm_tgt_dev *dev = pblk->dev;
656 struct pblk_line_mgmt *l_mg = &pblk->l_mg;
657 struct pblk_line_meta *lm = &pblk->lm;
658 struct nvm_geo *geo = &dev->geo;
659 sector_t provisioned;
660 int sec_meta, blk_meta, clba;
661 int minimum;
662
663 if (geo->op == NVM_TARGET_DEFAULT_OP)
664 pblk->op = PBLK_DEFAULT_OP;
665 else
666 pblk->op = geo->op;
667
668 minimum = pblk_get_min_chks(pblk);
669 provisioned = nr_free_chks;
670 provisioned *= (100 - pblk->op);
671 sector_div(provisioned, 100);
672
673 if ((nr_free_chks - provisioned) < minimum) {
674 if (geo->op != NVM_TARGET_DEFAULT_OP) {
675 pblk_err(pblk, "OP too small to create a sane instance\n");
676 return -EINTR;
677 }
678
679 /* If the user did not specify an OP value, and PBLK_DEFAULT_OP
680 * is not enough, calculate and set sane value
681 */
682
683 provisioned = nr_free_chks - minimum;
684 pblk->op = (100 * minimum) / nr_free_chks;
685 pblk_info(pblk, "Default OP insufficient, adjusting OP to %d\n",
686 pblk->op);
687 }
688
689 pblk->op_blks = nr_free_chks - provisioned;
690
691 /* Internally pblk manages all free blocks, but all calculations based
692 * on user capacity consider only provisioned blocks
693 */
694 pblk->rl.total_blocks = nr_free_chks;
695
696 /* Consider sectors used for metadata */
697 sec_meta = (lm->smeta_sec + lm->emeta_sec[0]) * l_mg->nr_free_lines;
698 blk_meta = DIV_ROUND_UP(sec_meta, geo->clba);
699
700 clba = (geo->clba / pblk->min_write_pgs) * pblk->min_write_pgs_data;
701 pblk->capacity = (provisioned - blk_meta) * clba;
702
703 atomic_set(&pblk->rl.free_blocks, nr_free_chks);
704 atomic_set(&pblk->rl.free_user_blocks, nr_free_chks);
705
706 return 0;
707}
708
709static int pblk_setup_line_meta_chk(struct pblk *pblk, struct pblk_line *line,
710 struct nvm_chk_meta *meta)
711{
712 struct nvm_tgt_dev *dev = pblk->dev;
713 struct nvm_geo *geo = &dev->geo;
714 struct pblk_line_meta *lm = &pblk->lm;
715 int i, nr_bad_chks = 0;
716
717 for (i = 0; i < lm->blk_per_line; i++) {
718 struct pblk_lun *rlun = &pblk->luns[i];
719 struct nvm_chk_meta *chunk;
720 struct nvm_chk_meta *chunk_meta;
721 struct ppa_addr ppa;
722 int pos;
723
724 ppa = rlun->bppa;
725 pos = pblk_ppa_to_pos(geo, ppa);
726 chunk = &line->chks[pos];
727
728 ppa.m.chk = line->id;
729 chunk_meta = pblk_chunk_get_off(pblk, meta, ppa);
730
731 chunk->state = chunk_meta->state;
732 chunk->type = chunk_meta->type;
733 chunk->wi = chunk_meta->wi;
734 chunk->slba = chunk_meta->slba;
735 chunk->cnlb = chunk_meta->cnlb;
736 chunk->wp = chunk_meta->wp;
737
738 trace_pblk_chunk_state(pblk_disk_name(pblk), &ppa,
739 chunk->state);
740
741 if (chunk->type & NVM_CHK_TP_SZ_SPEC) {
742 WARN_ONCE(1, "pblk: custom-sized chunks unsupported\n");
743 continue;
744 }
745
746 if (!(chunk->state & NVM_CHK_ST_OFFLINE))
747 continue;
748
749 set_bit(pos, line->blk_bitmap);
750 nr_bad_chks++;
751 }
752
753 return nr_bad_chks;
754}
755
756static long pblk_setup_line_meta(struct pblk *pblk, struct pblk_line *line,
757 void *chunk_meta, int line_id)
758{
759 struct pblk_line_mgmt *l_mg = &pblk->l_mg;
760 struct pblk_line_meta *lm = &pblk->lm;
761 long nr_bad_chks, chk_in_line;
762
763 line->pblk = pblk;
764 line->id = line_id;
765 line->type = PBLK_LINETYPE_FREE;
766 line->state = PBLK_LINESTATE_NEW;
767 line->gc_group = PBLK_LINEGC_NONE;
768 line->vsc = &l_mg->vsc_list[line_id];
769 spin_lock_init(&line->lock);
770
771 nr_bad_chks = pblk_setup_line_meta_chk(pblk, line, chunk_meta);
772
773 chk_in_line = lm->blk_per_line - nr_bad_chks;
774 if (nr_bad_chks < 0 || nr_bad_chks > lm->blk_per_line ||
775 chk_in_line < lm->min_blk_line) {
776 line->state = PBLK_LINESTATE_BAD;
777 list_add_tail(&line->list, &l_mg->bad_list);
778 return 0;
779 }
780
781 atomic_set(&line->blk_in_line, chk_in_line);
782 list_add_tail(&line->list, &l_mg->free_list);
783 l_mg->nr_free_lines++;
784
785 return chk_in_line;
786}
787
788static int pblk_alloc_line_meta(struct pblk *pblk, struct pblk_line *line)
789{
790 struct pblk_line_meta *lm = &pblk->lm;
791
792 line->blk_bitmap = kzalloc(lm->blk_bitmap_len, GFP_KERNEL);
793 if (!line->blk_bitmap)
794 return -ENOMEM;
795
796 line->erase_bitmap = kzalloc(lm->blk_bitmap_len, GFP_KERNEL);
797 if (!line->erase_bitmap)
798 goto free_blk_bitmap;
799
800
801 line->chks = kmalloc_array(lm->blk_per_line,
802 sizeof(struct nvm_chk_meta), GFP_KERNEL);
803 if (!line->chks)
804 goto free_erase_bitmap;
805
806 line->w_err_gc = kzalloc(sizeof(struct pblk_w_err_gc), GFP_KERNEL);
807 if (!line->w_err_gc)
808 goto free_chks;
809
810 return 0;
811
812free_chks:
813 kfree(line->chks);
814free_erase_bitmap:
815 kfree(line->erase_bitmap);
816free_blk_bitmap:
817 kfree(line->blk_bitmap);
818 return -ENOMEM;
819}
820
821static int pblk_line_mg_init(struct pblk *pblk)
822{
823 struct nvm_tgt_dev *dev = pblk->dev;
824 struct nvm_geo *geo = &dev->geo;
825 struct pblk_line_mgmt *l_mg = &pblk->l_mg;
826 struct pblk_line_meta *lm = &pblk->lm;
827 int i, bb_distance;
828
829 l_mg->nr_lines = geo->num_chk;
830 l_mg->log_line = l_mg->data_line = NULL;
831 l_mg->l_seq_nr = l_mg->d_seq_nr = 0;
832 l_mg->nr_free_lines = 0;
833 bitmap_zero(&l_mg->meta_bitmap, PBLK_DATA_LINES);
834
835 INIT_LIST_HEAD(&l_mg->free_list);
836 INIT_LIST_HEAD(&l_mg->corrupt_list);
837 INIT_LIST_HEAD(&l_mg->bad_list);
838 INIT_LIST_HEAD(&l_mg->gc_full_list);
839 INIT_LIST_HEAD(&l_mg->gc_high_list);
840 INIT_LIST_HEAD(&l_mg->gc_mid_list);
841 INIT_LIST_HEAD(&l_mg->gc_low_list);
842 INIT_LIST_HEAD(&l_mg->gc_empty_list);
843 INIT_LIST_HEAD(&l_mg->gc_werr_list);
844
845 INIT_LIST_HEAD(&l_mg->emeta_list);
846
847 l_mg->gc_lists[0] = &l_mg->gc_werr_list;
848 l_mg->gc_lists[1] = &l_mg->gc_high_list;
849 l_mg->gc_lists[2] = &l_mg->gc_mid_list;
850 l_mg->gc_lists[3] = &l_mg->gc_low_list;
851
852 spin_lock_init(&l_mg->free_lock);
853 spin_lock_init(&l_mg->close_lock);
854 spin_lock_init(&l_mg->gc_lock);
855
856 l_mg->vsc_list = kcalloc(l_mg->nr_lines, sizeof(__le32), GFP_KERNEL);
857 if (!l_mg->vsc_list)
858 goto fail;
859
860 l_mg->bb_template = kzalloc(lm->sec_bitmap_len, GFP_KERNEL);
861 if (!l_mg->bb_template)
862 goto fail_free_vsc_list;
863
864 l_mg->bb_aux = kzalloc(lm->sec_bitmap_len, GFP_KERNEL);
865 if (!l_mg->bb_aux)
866 goto fail_free_bb_template;
867
868 /* smeta is always small enough to fit on a kmalloc memory allocation,
869 * emeta depends on the number of LUNs allocated to the pblk instance
870 */
871 for (i = 0; i < PBLK_DATA_LINES; i++) {
872 l_mg->sline_meta[i] = kmalloc(lm->smeta_len, GFP_KERNEL);
873 if (!l_mg->sline_meta[i])
874 goto fail_free_smeta;
875 }
876
877 l_mg->bitmap_cache = kmem_cache_create("pblk_lm_bitmap",
878 lm->sec_bitmap_len, 0, 0, NULL);
879 if (!l_mg->bitmap_cache)
880 goto fail_free_smeta;
881
882 /* the bitmap pool is used for both valid and map bitmaps */
883 l_mg->bitmap_pool = mempool_create_slab_pool(PBLK_DATA_LINES * 2,
884 l_mg->bitmap_cache);
885 if (!l_mg->bitmap_pool)
886 goto fail_destroy_bitmap_cache;
887
888 /* emeta allocates three different buffers for managing metadata with
889 * in-memory and in-media layouts
890 */
891 for (i = 0; i < PBLK_DATA_LINES; i++) {
892 struct pblk_emeta *emeta;
893
894 emeta = kmalloc(sizeof(struct pblk_emeta), GFP_KERNEL);
895 if (!emeta)
896 goto fail_free_emeta;
897
898 emeta->buf = kvmalloc(lm->emeta_len[0], GFP_KERNEL);
899 if (!emeta->buf) {
900 kfree(emeta);
901 goto fail_free_emeta;
902 }
903
904 emeta->nr_entries = lm->emeta_sec[0];
905 l_mg->eline_meta[i] = emeta;
906 }
907
908 for (i = 0; i < l_mg->nr_lines; i++)
909 l_mg->vsc_list[i] = cpu_to_le32(EMPTY_ENTRY);
910
911 bb_distance = (geo->all_luns) * geo->ws_opt;
912 for (i = 0; i < lm->sec_per_line; i += bb_distance)
913 bitmap_set(l_mg->bb_template, i, geo->ws_opt);
914
915 return 0;
916
917fail_free_emeta:
918 while (--i >= 0) {
919 kvfree(l_mg->eline_meta[i]->buf);
920 kfree(l_mg->eline_meta[i]);
921 }
922
923 mempool_destroy(l_mg->bitmap_pool);
924fail_destroy_bitmap_cache:
925 kmem_cache_destroy(l_mg->bitmap_cache);
926fail_free_smeta:
927 for (i = 0; i < PBLK_DATA_LINES; i++)
928 kfree(l_mg->sline_meta[i]);
929 kfree(l_mg->bb_aux);
930fail_free_bb_template:
931 kfree(l_mg->bb_template);
932fail_free_vsc_list:
933 kfree(l_mg->vsc_list);
934fail:
935 return -ENOMEM;
936}
937
938static int pblk_line_meta_init(struct pblk *pblk)
939{
940 struct nvm_tgt_dev *dev = pblk->dev;
941 struct nvm_geo *geo = &dev->geo;
942 struct pblk_line_meta *lm = &pblk->lm;
943 unsigned int smeta_len, emeta_len;
944 int i;
945
946 lm->sec_per_line = geo->clba * geo->all_luns;
947 lm->blk_per_line = geo->all_luns;
948 lm->blk_bitmap_len = BITS_TO_LONGS(geo->all_luns) * sizeof(long);
949 lm->sec_bitmap_len = BITS_TO_LONGS(lm->sec_per_line) * sizeof(long);
950 lm->lun_bitmap_len = BITS_TO_LONGS(geo->all_luns) * sizeof(long);
951 lm->mid_thrs = lm->sec_per_line / 2;
952 lm->high_thrs = lm->sec_per_line / 4;
953 lm->meta_distance = (geo->all_luns / 2) * pblk->min_write_pgs;
954
955 /* Calculate necessary pages for smeta. See comment over struct
956 * line_smeta definition
957 */
958 i = 1;
959add_smeta_page:
960 lm->smeta_sec = i * geo->ws_opt;
961 lm->smeta_len = lm->smeta_sec * geo->csecs;
962
963 smeta_len = sizeof(struct line_smeta) + lm->lun_bitmap_len;
964 if (smeta_len > lm->smeta_len) {
965 i++;
966 goto add_smeta_page;
967 }
968
969 /* Calculate necessary pages for emeta. See comment over struct
970 * line_emeta definition
971 */
972 i = 1;
973add_emeta_page:
974 lm->emeta_sec[0] = i * geo->ws_opt;
975 lm->emeta_len[0] = lm->emeta_sec[0] * geo->csecs;
976
977 emeta_len = calc_emeta_len(pblk);
978 if (emeta_len > lm->emeta_len[0]) {
979 i++;
980 goto add_emeta_page;
981 }
982
983 lm->emeta_bb = geo->all_luns > i ? geo->all_luns - i : 0;
984
985 lm->min_blk_line = 1;
986 if (geo->all_luns > 1)
987 lm->min_blk_line += DIV_ROUND_UP(lm->smeta_sec +
988 lm->emeta_sec[0], geo->clba);
989
990 if (lm->min_blk_line > lm->blk_per_line) {
991 pblk_err(pblk, "config. not supported. Min. LUN in line:%d\n",
992 lm->blk_per_line);
993 return -EINVAL;
994 }
995
996 return 0;
997}
998
999static int pblk_lines_init(struct pblk *pblk)
1000{
1001 struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1002 struct pblk_line *line;
1003 void *chunk_meta;
1004 int nr_free_chks = 0;
1005 int i, ret;
1006
1007 ret = pblk_line_meta_init(pblk);
1008 if (ret)
1009 return ret;
1010
1011 ret = pblk_line_mg_init(pblk);
1012 if (ret)
1013 return ret;
1014
1015 ret = pblk_luns_init(pblk);
1016 if (ret)
1017 goto fail_free_meta;
1018
1019 chunk_meta = pblk_get_chunk_meta(pblk);
1020 if (IS_ERR(chunk_meta)) {
1021 ret = PTR_ERR(chunk_meta);
1022 goto fail_free_luns;
1023 }
1024
1025 pblk->lines = kcalloc(l_mg->nr_lines, sizeof(struct pblk_line),
1026 GFP_KERNEL);
1027 if (!pblk->lines) {
1028 ret = -ENOMEM;
1029 goto fail_free_chunk_meta;
1030 }
1031
1032 for (i = 0; i < l_mg->nr_lines; i++) {
1033 line = &pblk->lines[i];
1034
1035 ret = pblk_alloc_line_meta(pblk, line);
1036 if (ret)
1037 goto fail_free_lines;
1038
1039 nr_free_chks += pblk_setup_line_meta(pblk, line, chunk_meta, i);
1040
1041 trace_pblk_line_state(pblk_disk_name(pblk), line->id,
1042 line->state);
1043 }
1044
1045 if (!nr_free_chks) {
1046 pblk_err(pblk, "too many bad blocks prevent for sane instance\n");
1047 ret = -EINTR;
1048 goto fail_free_lines;
1049 }
1050
1051 ret = pblk_set_provision(pblk, nr_free_chks);
1052 if (ret)
1053 goto fail_free_lines;
1054
1055 vfree(chunk_meta);
1056 return 0;
1057
1058fail_free_lines:
1059 while (--i >= 0)
1060 pblk_line_meta_free(l_mg, &pblk->lines[i]);
1061 kfree(pblk->lines);
1062fail_free_chunk_meta:
1063 vfree(chunk_meta);
1064fail_free_luns:
1065 kfree(pblk->luns);
1066fail_free_meta:
1067 pblk_line_mg_free(pblk);
1068
1069 return ret;
1070}
1071
1072static int pblk_writer_init(struct pblk *pblk)
1073{
1074 pblk->writer_ts = kthread_create(pblk_write_ts, pblk, "pblk-writer-t");
1075 if (IS_ERR(pblk->writer_ts)) {
1076 int err = PTR_ERR(pblk->writer_ts);
1077
1078 if (err != -EINTR)
1079 pblk_err(pblk, "could not allocate writer kthread (%d)\n",
1080 err);
1081 return err;
1082 }
1083
1084 timer_setup(&pblk->wtimer, pblk_write_timer_fn, 0);
1085 mod_timer(&pblk->wtimer, jiffies + msecs_to_jiffies(100));
1086
1087 return 0;
1088}
1089
1090static void pblk_writer_stop(struct pblk *pblk)
1091{
1092 /* The pipeline must be stopped and the write buffer emptied before the
1093 * write thread is stopped
1094 */
1095 WARN(pblk_rb_read_count(&pblk->rwb),
1096 "Stopping not fully persisted write buffer\n");
1097
1098 WARN(pblk_rb_sync_count(&pblk->rwb),
1099 "Stopping not fully synced write buffer\n");
1100
1101 del_timer_sync(&pblk->wtimer);
1102 if (pblk->writer_ts)
1103 kthread_stop(pblk->writer_ts);
1104}
1105
1106static void pblk_free(struct pblk *pblk)
1107{
1108 pblk_lines_free(pblk);
1109 pblk_l2p_free(pblk);
1110 pblk_rwb_free(pblk);
1111 pblk_core_free(pblk);
1112
1113 kfree(pblk);
1114}
1115
1116static void pblk_tear_down(struct pblk *pblk, bool graceful)
1117{
1118 if (graceful)
1119 __pblk_pipeline_flush(pblk);
1120 __pblk_pipeline_stop(pblk);
1121 pblk_writer_stop(pblk);
1122 pblk_rb_sync_l2p(&pblk->rwb);
1123 pblk_rl_free(&pblk->rl);
1124
1125 pblk_debug(pblk, "consistent tear down (graceful:%d)\n", graceful);
1126}
1127
1128static void pblk_exit(void *private, bool graceful)
1129{
1130 struct pblk *pblk = private;
1131
1132 pblk_gc_exit(pblk, graceful);
1133 pblk_tear_down(pblk, graceful);
1134
1135#ifdef CONFIG_NVM_PBLK_DEBUG
1136 pblk_info(pblk, "exit: L2P CRC: %x\n", pblk_l2p_crc(pblk));
1137#endif
1138
1139 pblk_free(pblk);
1140}
1141
1142static sector_t pblk_capacity(void *private)
1143{
1144 struct pblk *pblk = private;
1145
1146 return pblk->capacity * NR_PHY_IN_LOG;
1147}
1148
1149static void *pblk_init(struct nvm_tgt_dev *dev, struct gendisk *tdisk,
1150 int flags)
1151{
1152 struct nvm_geo *geo = &dev->geo;
1153 struct request_queue *bqueue = dev->q;
1154 struct request_queue *tqueue = tdisk->queue;
1155 struct pblk *pblk;
1156 int ret;
1157
1158 pblk = kzalloc(sizeof(struct pblk), GFP_KERNEL);
1159 if (!pblk)
1160 return ERR_PTR(-ENOMEM);
1161
1162 pblk->dev = dev;
1163 pblk->disk = tdisk;
1164 pblk->state = PBLK_STATE_RUNNING;
1165 trace_pblk_state(pblk_disk_name(pblk), pblk->state);
1166 pblk->gc.gc_enabled = 0;
1167
1168 if (!(geo->version == NVM_OCSSD_SPEC_12 ||
1169 geo->version == NVM_OCSSD_SPEC_20)) {
1170 pblk_err(pblk, "OCSSD version not supported (%u)\n",
1171 geo->version);
1172 kfree(pblk);
1173 return ERR_PTR(-EINVAL);
1174 }
1175
1176 if (geo->ext) {
1177 pblk_err(pblk, "extended metadata not supported\n");
1178 kfree(pblk);
1179 return ERR_PTR(-EINVAL);
1180 }
1181
1182 spin_lock_init(&pblk->resubmit_lock);
1183 spin_lock_init(&pblk->trans_lock);
1184 spin_lock_init(&pblk->lock);
1185
1186#ifdef CONFIG_NVM_PBLK_DEBUG
1187 atomic_long_set(&pblk->inflight_writes, 0);
1188 atomic_long_set(&pblk->padded_writes, 0);
1189 atomic_long_set(&pblk->padded_wb, 0);
1190 atomic_long_set(&pblk->req_writes, 0);
1191 atomic_long_set(&pblk->sub_writes, 0);
1192 atomic_long_set(&pblk->sync_writes, 0);
1193 atomic_long_set(&pblk->inflight_reads, 0);
1194 atomic_long_set(&pblk->cache_reads, 0);
1195 atomic_long_set(&pblk->sync_reads, 0);
1196 atomic_long_set(&pblk->recov_writes, 0);
1197 atomic_long_set(&pblk->recov_writes, 0);
1198 atomic_long_set(&pblk->recov_gc_writes, 0);
1199 atomic_long_set(&pblk->recov_gc_reads, 0);
1200#endif
1201
1202 atomic_long_set(&pblk->read_failed, 0);
1203 atomic_long_set(&pblk->read_empty, 0);
1204 atomic_long_set(&pblk->read_high_ecc, 0);
1205 atomic_long_set(&pblk->read_failed_gc, 0);
1206 atomic_long_set(&pblk->write_failed, 0);
1207 atomic_long_set(&pblk->erase_failed, 0);
1208
1209 ret = pblk_core_init(pblk);
1210 if (ret) {
1211 pblk_err(pblk, "could not initialize core\n");
1212 goto fail;
1213 }
1214
1215 ret = pblk_lines_init(pblk);
1216 if (ret) {
1217 pblk_err(pblk, "could not initialize lines\n");
1218 goto fail_free_core;
1219 }
1220
1221 ret = pblk_rwb_init(pblk);
1222 if (ret) {
1223 pblk_err(pblk, "could not initialize write buffer\n");
1224 goto fail_free_lines;
1225 }
1226
1227 ret = pblk_l2p_init(pblk, flags & NVM_TARGET_FACTORY);
1228 if (ret) {
1229 pblk_err(pblk, "could not initialize maps\n");
1230 goto fail_free_rwb;
1231 }
1232
1233 ret = pblk_writer_init(pblk);
1234 if (ret) {
1235 if (ret != -EINTR)
1236 pblk_err(pblk, "could not initialize write thread\n");
1237 goto fail_free_l2p;
1238 }
1239
1240 ret = pblk_gc_init(pblk);
1241 if (ret) {
1242 pblk_err(pblk, "could not initialize gc\n");
1243 goto fail_stop_writer;
1244 }
1245
1246 /* inherit the size from the underlying device */
1247 blk_queue_logical_block_size(tqueue, queue_physical_block_size(bqueue));
1248 blk_queue_max_hw_sectors(tqueue, queue_max_hw_sectors(bqueue));
1249
1250 blk_queue_write_cache(tqueue, true, false);
1251
1252 tqueue->limits.discard_granularity = geo->clba * geo->csecs;
1253 tqueue->limits.discard_alignment = 0;
1254 blk_queue_max_discard_sectors(tqueue, UINT_MAX >> 9);
1255 blk_queue_flag_set(QUEUE_FLAG_DISCARD, tqueue);
1256
1257 pblk_info(pblk, "luns:%u, lines:%d, secs:%llu, buf entries:%u\n",
1258 geo->all_luns, pblk->l_mg.nr_lines,
1259 (unsigned long long)pblk->capacity,
1260 pblk->rwb.nr_entries);
1261
1262 wake_up_process(pblk->writer_ts);
1263
1264 /* Check if we need to start GC */
1265 pblk_gc_should_kick(pblk);
1266
1267 return pblk;
1268
1269fail_stop_writer:
1270 pblk_writer_stop(pblk);
1271fail_free_l2p:
1272 pblk_l2p_free(pblk);
1273fail_free_rwb:
1274 pblk_rwb_free(pblk);
1275fail_free_lines:
1276 pblk_lines_free(pblk);
1277fail_free_core:
1278 pblk_core_free(pblk);
1279fail:
1280 kfree(pblk);
1281 return ERR_PTR(ret);
1282}
1283
1284/* physical block device target */
1285static struct nvm_tgt_type tt_pblk = {
1286 .name = "pblk",
1287 .version = {1, 0, 0},
1288
1289 .bops = &pblk_bops,
1290 .capacity = pblk_capacity,
1291
1292 .init = pblk_init,
1293 .exit = pblk_exit,
1294
1295 .sysfs_init = pblk_sysfs_init,
1296 .sysfs_exit = pblk_sysfs_exit,
1297 .owner = THIS_MODULE,
1298};
1299
1300static int __init pblk_module_init(void)
1301{
1302 int ret;
1303
1304 ret = bioset_init(&pblk_bio_set, BIO_POOL_SIZE, 0, 0);
1305 if (ret)
1306 return ret;
1307 ret = nvm_register_tgt_type(&tt_pblk);
1308 if (ret)
1309 bioset_exit(&pblk_bio_set);
1310 return ret;
1311}
1312
1313static void pblk_module_exit(void)
1314{
1315 bioset_exit(&pblk_bio_set);
1316 nvm_unregister_tgt_type(&tt_pblk);
1317}
1318
1319module_init(pblk_module_init);
1320module_exit(pblk_module_exit);
1321MODULE_AUTHOR("Javier Gonzalez <javier@cnexlabs.com>");
1322MODULE_AUTHOR("Matias Bjorling <matias@cnexlabs.com>");
1323MODULE_LICENSE("GPL v2");
1324MODULE_DESCRIPTION("Physical Block-Device for Open-Channel SSDs");
1/*
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_ws_cache, *pblk_rec_cache, *pblk_g_rq_cache,
24 *pblk_w_rq_cache;
25static DECLARE_RWSEM(pblk_lock);
26struct bio_set *pblk_bio_set;
27
28static int pblk_rw_io(struct request_queue *q, struct pblk *pblk,
29 struct bio *bio)
30{
31 int ret;
32
33 /* Read requests must be <= 256kb due to NVMe's 64 bit completion bitmap
34 * constraint. Writes can be of arbitrary size.
35 */
36 if (bio_data_dir(bio) == READ) {
37 blk_queue_split(q, &bio);
38 ret = pblk_submit_read(pblk, bio);
39 if (ret == NVM_IO_DONE && bio_flagged(bio, BIO_CLONED))
40 bio_put(bio);
41
42 return ret;
43 }
44
45 /* Prevent deadlock in the case of a modest LUN configuration and large
46 * user I/Os. Unless stalled, the rate limiter leaves at least 256KB
47 * available for user I/O.
48 */
49 if (pblk_get_secs(bio) > pblk_rl_max_io(&pblk->rl))
50 blk_queue_split(q, &bio);
51
52 return pblk_write_to_cache(pblk, bio, PBLK_IOTYPE_USER);
53}
54
55static blk_qc_t pblk_make_rq(struct request_queue *q, struct bio *bio)
56{
57 struct pblk *pblk = q->queuedata;
58
59 if (bio_op(bio) == REQ_OP_DISCARD) {
60 pblk_discard(pblk, bio);
61 if (!(bio->bi_opf & REQ_PREFLUSH)) {
62 bio_endio(bio);
63 return BLK_QC_T_NONE;
64 }
65 }
66
67 switch (pblk_rw_io(q, pblk, bio)) {
68 case NVM_IO_ERR:
69 bio_io_error(bio);
70 break;
71 case NVM_IO_DONE:
72 bio_endio(bio);
73 break;
74 }
75
76 return BLK_QC_T_NONE;
77}
78
79static size_t pblk_trans_map_size(struct pblk *pblk)
80{
81 int entry_size = 8;
82
83 if (pblk->addrf_len < 32)
84 entry_size = 4;
85
86 return entry_size * pblk->rl.nr_secs;
87}
88
89#ifdef CONFIG_NVM_DEBUG
90static u32 pblk_l2p_crc(struct pblk *pblk)
91{
92 size_t map_size;
93 u32 crc = ~(u32)0;
94
95 map_size = pblk_trans_map_size(pblk);
96 crc = crc32_le(crc, pblk->trans_map, map_size);
97 return crc;
98}
99#endif
100
101static void pblk_l2p_free(struct pblk *pblk)
102{
103 vfree(pblk->trans_map);
104}
105
106static int pblk_l2p_recover(struct pblk *pblk, bool factory_init)
107{
108 struct pblk_line *line = NULL;
109
110 if (factory_init) {
111 pblk_setup_uuid(pblk);
112 } else {
113 line = pblk_recov_l2p(pblk);
114 if (IS_ERR(line)) {
115 pr_err("pblk: could not recover l2p table\n");
116 return -EFAULT;
117 }
118 }
119
120#ifdef CONFIG_NVM_DEBUG
121 pr_info("pblk init: L2P CRC: %x\n", pblk_l2p_crc(pblk));
122#endif
123
124 /* Free full lines directly as GC has not been started yet */
125 pblk_gc_free_full_lines(pblk);
126
127 if (!line) {
128 /* Configure next line for user data */
129 line = pblk_line_get_first_data(pblk);
130 if (!line) {
131 pr_err("pblk: line list corrupted\n");
132 return -EFAULT;
133 }
134 }
135
136 return 0;
137}
138
139static int pblk_l2p_init(struct pblk *pblk, bool factory_init)
140{
141 sector_t i;
142 struct ppa_addr ppa;
143 size_t map_size;
144
145 map_size = pblk_trans_map_size(pblk);
146 pblk->trans_map = vmalloc(map_size);
147 if (!pblk->trans_map)
148 return -ENOMEM;
149
150 pblk_ppa_set_empty(&ppa);
151
152 for (i = 0; i < pblk->rl.nr_secs; i++)
153 pblk_trans_map_set(pblk, i, ppa);
154
155 return pblk_l2p_recover(pblk, factory_init);
156}
157
158static void pblk_rwb_free(struct pblk *pblk)
159{
160 if (pblk_rb_tear_down_check(&pblk->rwb))
161 pr_err("pblk: write buffer error on tear down\n");
162
163 pblk_rb_data_free(&pblk->rwb);
164 vfree(pblk_rb_entries_ref(&pblk->rwb));
165}
166
167static int pblk_rwb_init(struct pblk *pblk)
168{
169 struct nvm_tgt_dev *dev = pblk->dev;
170 struct nvm_geo *geo = &dev->geo;
171 struct pblk_rb_entry *entries;
172 unsigned long nr_entries;
173 unsigned int power_size, power_seg_sz;
174
175 nr_entries = pblk_rb_calculate_size(pblk->pgs_in_buffer);
176
177 entries = vzalloc(nr_entries * sizeof(struct pblk_rb_entry));
178 if (!entries)
179 return -ENOMEM;
180
181 power_size = get_count_order(nr_entries);
182 power_seg_sz = get_count_order(geo->csecs);
183
184 return pblk_rb_init(&pblk->rwb, entries, power_size, power_seg_sz);
185}
186
187/* Minimum pages needed within a lun */
188#define ADDR_POOL_SIZE 64
189
190static int pblk_set_addrf_12(struct nvm_geo *geo, struct nvm_addrf_12 *dst)
191{
192 struct nvm_addrf_12 *src = (struct nvm_addrf_12 *)&geo->addrf;
193 int power_len;
194
195 /* Re-calculate channel and lun format to adapt to configuration */
196 power_len = get_count_order(geo->num_ch);
197 if (1 << power_len != geo->num_ch) {
198 pr_err("pblk: supports only power-of-two channel config.\n");
199 return -EINVAL;
200 }
201 dst->ch_len = power_len;
202
203 power_len = get_count_order(geo->num_lun);
204 if (1 << power_len != geo->num_lun) {
205 pr_err("pblk: supports only power-of-two LUN config.\n");
206 return -EINVAL;
207 }
208 dst->lun_len = power_len;
209
210 dst->blk_len = src->blk_len;
211 dst->pg_len = src->pg_len;
212 dst->pln_len = src->pln_len;
213 dst->sec_len = src->sec_len;
214
215 dst->sec_offset = 0;
216 dst->pln_offset = dst->sec_len;
217 dst->ch_offset = dst->pln_offset + dst->pln_len;
218 dst->lun_offset = dst->ch_offset + dst->ch_len;
219 dst->pg_offset = dst->lun_offset + dst->lun_len;
220 dst->blk_offset = dst->pg_offset + dst->pg_len;
221
222 dst->sec_mask = ((1ULL << dst->sec_len) - 1) << dst->sec_offset;
223 dst->pln_mask = ((1ULL << dst->pln_len) - 1) << dst->pln_offset;
224 dst->ch_mask = ((1ULL << dst->ch_len) - 1) << dst->ch_offset;
225 dst->lun_mask = ((1ULL << dst->lun_len) - 1) << dst->lun_offset;
226 dst->pg_mask = ((1ULL << dst->pg_len) - 1) << dst->pg_offset;
227 dst->blk_mask = ((1ULL << dst->blk_len) - 1) << dst->blk_offset;
228
229 return dst->blk_offset + src->blk_len;
230}
231
232static int pblk_set_addrf_20(struct nvm_geo *geo, struct nvm_addrf *adst,
233 struct pblk_addrf *udst)
234{
235 struct nvm_addrf *src = &geo->addrf;
236
237 adst->ch_len = get_count_order(geo->num_ch);
238 adst->lun_len = get_count_order(geo->num_lun);
239 adst->chk_len = src->chk_len;
240 adst->sec_len = src->sec_len;
241
242 adst->sec_offset = 0;
243 adst->ch_offset = adst->sec_len;
244 adst->lun_offset = adst->ch_offset + adst->ch_len;
245 adst->chk_offset = adst->lun_offset + adst->lun_len;
246
247 adst->sec_mask = ((1ULL << adst->sec_len) - 1) << adst->sec_offset;
248 adst->chk_mask = ((1ULL << adst->chk_len) - 1) << adst->chk_offset;
249 adst->lun_mask = ((1ULL << adst->lun_len) - 1) << adst->lun_offset;
250 adst->ch_mask = ((1ULL << adst->ch_len) - 1) << adst->ch_offset;
251
252 udst->sec_stripe = geo->ws_opt;
253 udst->ch_stripe = geo->num_ch;
254 udst->lun_stripe = geo->num_lun;
255
256 udst->sec_lun_stripe = udst->sec_stripe * udst->ch_stripe;
257 udst->sec_ws_stripe = udst->sec_lun_stripe * udst->lun_stripe;
258
259 return adst->chk_offset + adst->chk_len;
260}
261
262static int pblk_set_addrf(struct pblk *pblk)
263{
264 struct nvm_tgt_dev *dev = pblk->dev;
265 struct nvm_geo *geo = &dev->geo;
266 int mod;
267
268 switch (geo->version) {
269 case NVM_OCSSD_SPEC_12:
270 div_u64_rem(geo->clba, pblk->min_write_pgs, &mod);
271 if (mod) {
272 pr_err("pblk: bad configuration of sectors/pages\n");
273 return -EINVAL;
274 }
275
276 pblk->addrf_len = pblk_set_addrf_12(geo, (void *)&pblk->addrf);
277 break;
278 case NVM_OCSSD_SPEC_20:
279 pblk->addrf_len = pblk_set_addrf_20(geo, (void *)&pblk->addrf,
280 &pblk->uaddrf);
281 break;
282 default:
283 pr_err("pblk: OCSSD revision not supported (%d)\n",
284 geo->version);
285 return -EINVAL;
286 }
287
288 return 0;
289}
290
291static int pblk_init_global_caches(struct pblk *pblk)
292{
293 down_write(&pblk_lock);
294 pblk_ws_cache = kmem_cache_create("pblk_blk_ws",
295 sizeof(struct pblk_line_ws), 0, 0, NULL);
296 if (!pblk_ws_cache) {
297 up_write(&pblk_lock);
298 return -ENOMEM;
299 }
300
301 pblk_rec_cache = kmem_cache_create("pblk_rec",
302 sizeof(struct pblk_rec_ctx), 0, 0, NULL);
303 if (!pblk_rec_cache) {
304 kmem_cache_destroy(pblk_ws_cache);
305 up_write(&pblk_lock);
306 return -ENOMEM;
307 }
308
309 pblk_g_rq_cache = kmem_cache_create("pblk_g_rq", pblk_g_rq_size,
310 0, 0, NULL);
311 if (!pblk_g_rq_cache) {
312 kmem_cache_destroy(pblk_ws_cache);
313 kmem_cache_destroy(pblk_rec_cache);
314 up_write(&pblk_lock);
315 return -ENOMEM;
316 }
317
318 pblk_w_rq_cache = kmem_cache_create("pblk_w_rq", pblk_w_rq_size,
319 0, 0, NULL);
320 if (!pblk_w_rq_cache) {
321 kmem_cache_destroy(pblk_ws_cache);
322 kmem_cache_destroy(pblk_rec_cache);
323 kmem_cache_destroy(pblk_g_rq_cache);
324 up_write(&pblk_lock);
325 return -ENOMEM;
326 }
327 up_write(&pblk_lock);
328
329 return 0;
330}
331
332static void pblk_free_global_caches(struct pblk *pblk)
333{
334 kmem_cache_destroy(pblk_ws_cache);
335 kmem_cache_destroy(pblk_rec_cache);
336 kmem_cache_destroy(pblk_g_rq_cache);
337 kmem_cache_destroy(pblk_w_rq_cache);
338}
339
340static int pblk_core_init(struct pblk *pblk)
341{
342 struct nvm_tgt_dev *dev = pblk->dev;
343 struct nvm_geo *geo = &dev->geo;
344 int max_write_ppas;
345
346 atomic64_set(&pblk->user_wa, 0);
347 atomic64_set(&pblk->pad_wa, 0);
348 atomic64_set(&pblk->gc_wa, 0);
349 pblk->user_rst_wa = 0;
350 pblk->pad_rst_wa = 0;
351 pblk->gc_rst_wa = 0;
352
353 atomic64_set(&pblk->nr_flush, 0);
354 pblk->nr_flush_rst = 0;
355
356 pblk->pgs_in_buffer = geo->mw_cunits * geo->all_luns;
357
358 pblk->min_write_pgs = geo->ws_opt * (geo->csecs / PAGE_SIZE);
359 max_write_ppas = pblk->min_write_pgs * geo->all_luns;
360 pblk->max_write_pgs = min_t(int, max_write_ppas, NVM_MAX_VLBA);
361 pblk_set_sec_per_write(pblk, pblk->min_write_pgs);
362
363 if (pblk->max_write_pgs > PBLK_MAX_REQ_ADDRS) {
364 pr_err("pblk: vector list too big(%u > %u)\n",
365 pblk->max_write_pgs, PBLK_MAX_REQ_ADDRS);
366 return -EINVAL;
367 }
368
369 pblk->pad_dist = kzalloc((pblk->min_write_pgs - 1) * sizeof(atomic64_t),
370 GFP_KERNEL);
371 if (!pblk->pad_dist)
372 return -ENOMEM;
373
374 if (pblk_init_global_caches(pblk))
375 goto fail_free_pad_dist;
376
377 /* Internal bios can be at most the sectors signaled by the device. */
378 pblk->page_bio_pool = mempool_create_page_pool(NVM_MAX_VLBA, 0);
379 if (!pblk->page_bio_pool)
380 goto free_global_caches;
381
382 pblk->gen_ws_pool = mempool_create_slab_pool(PBLK_GEN_WS_POOL_SIZE,
383 pblk_ws_cache);
384 if (!pblk->gen_ws_pool)
385 goto free_page_bio_pool;
386
387 pblk->rec_pool = mempool_create_slab_pool(geo->all_luns,
388 pblk_rec_cache);
389 if (!pblk->rec_pool)
390 goto free_gen_ws_pool;
391
392 pblk->r_rq_pool = mempool_create_slab_pool(geo->all_luns,
393 pblk_g_rq_cache);
394 if (!pblk->r_rq_pool)
395 goto free_rec_pool;
396
397 pblk->e_rq_pool = mempool_create_slab_pool(geo->all_luns,
398 pblk_g_rq_cache);
399 if (!pblk->e_rq_pool)
400 goto free_r_rq_pool;
401
402 pblk->w_rq_pool = mempool_create_slab_pool(geo->all_luns,
403 pblk_w_rq_cache);
404 if (!pblk->w_rq_pool)
405 goto free_e_rq_pool;
406
407 pblk->close_wq = alloc_workqueue("pblk-close-wq",
408 WQ_MEM_RECLAIM | WQ_UNBOUND, PBLK_NR_CLOSE_JOBS);
409 if (!pblk->close_wq)
410 goto free_w_rq_pool;
411
412 pblk->bb_wq = alloc_workqueue("pblk-bb-wq",
413 WQ_MEM_RECLAIM | WQ_UNBOUND, 0);
414 if (!pblk->bb_wq)
415 goto free_close_wq;
416
417 pblk->r_end_wq = alloc_workqueue("pblk-read-end-wq",
418 WQ_MEM_RECLAIM | WQ_UNBOUND, 0);
419 if (!pblk->r_end_wq)
420 goto free_bb_wq;
421
422 if (pblk_set_addrf(pblk))
423 goto free_r_end_wq;
424
425 INIT_LIST_HEAD(&pblk->compl_list);
426
427 return 0;
428
429free_r_end_wq:
430 destroy_workqueue(pblk->r_end_wq);
431free_bb_wq:
432 destroy_workqueue(pblk->bb_wq);
433free_close_wq:
434 destroy_workqueue(pblk->close_wq);
435free_w_rq_pool:
436 mempool_destroy(pblk->w_rq_pool);
437free_e_rq_pool:
438 mempool_destroy(pblk->e_rq_pool);
439free_r_rq_pool:
440 mempool_destroy(pblk->r_rq_pool);
441free_rec_pool:
442 mempool_destroy(pblk->rec_pool);
443free_gen_ws_pool:
444 mempool_destroy(pblk->gen_ws_pool);
445free_page_bio_pool:
446 mempool_destroy(pblk->page_bio_pool);
447free_global_caches:
448 pblk_free_global_caches(pblk);
449fail_free_pad_dist:
450 kfree(pblk->pad_dist);
451 return -ENOMEM;
452}
453
454static void pblk_core_free(struct pblk *pblk)
455{
456 if (pblk->close_wq)
457 destroy_workqueue(pblk->close_wq);
458
459 if (pblk->r_end_wq)
460 destroy_workqueue(pblk->r_end_wq);
461
462 if (pblk->bb_wq)
463 destroy_workqueue(pblk->bb_wq);
464
465 mempool_destroy(pblk->page_bio_pool);
466 mempool_destroy(pblk->gen_ws_pool);
467 mempool_destroy(pblk->rec_pool);
468 mempool_destroy(pblk->r_rq_pool);
469 mempool_destroy(pblk->e_rq_pool);
470 mempool_destroy(pblk->w_rq_pool);
471
472 pblk_free_global_caches(pblk);
473 kfree(pblk->pad_dist);
474}
475
476static void pblk_line_mg_free(struct pblk *pblk)
477{
478 struct pblk_line_mgmt *l_mg = &pblk->l_mg;
479 int i;
480
481 kfree(l_mg->bb_template);
482 kfree(l_mg->bb_aux);
483 kfree(l_mg->vsc_list);
484
485 for (i = 0; i < PBLK_DATA_LINES; i++) {
486 kfree(l_mg->sline_meta[i]);
487 pblk_mfree(l_mg->eline_meta[i]->buf, l_mg->emeta_alloc_type);
488 kfree(l_mg->eline_meta[i]);
489 }
490}
491
492static void pblk_line_meta_free(struct pblk_line *line)
493{
494 kfree(line->blk_bitmap);
495 kfree(line->erase_bitmap);
496 kfree(line->chks);
497}
498
499static void pblk_lines_free(struct pblk *pblk)
500{
501 struct pblk_line_mgmt *l_mg = &pblk->l_mg;
502 struct pblk_line *line;
503 int i;
504
505 spin_lock(&l_mg->free_lock);
506 for (i = 0; i < l_mg->nr_lines; i++) {
507 line = &pblk->lines[i];
508
509 pblk_line_free(pblk, line);
510 pblk_line_meta_free(line);
511 }
512 spin_unlock(&l_mg->free_lock);
513
514 pblk_line_mg_free(pblk);
515
516 kfree(pblk->luns);
517 kfree(pblk->lines);
518}
519
520static int pblk_bb_get_tbl(struct nvm_tgt_dev *dev, struct pblk_lun *rlun,
521 u8 *blks, int nr_blks)
522{
523 struct ppa_addr ppa;
524 int ret;
525
526 ppa.ppa = 0;
527 ppa.g.ch = rlun->bppa.g.ch;
528 ppa.g.lun = rlun->bppa.g.lun;
529
530 ret = nvm_get_tgt_bb_tbl(dev, ppa, blks);
531 if (ret)
532 return ret;
533
534 nr_blks = nvm_bb_tbl_fold(dev->parent, blks, nr_blks);
535 if (nr_blks < 0)
536 return -EIO;
537
538 return 0;
539}
540
541static void *pblk_bb_get_meta(struct pblk *pblk)
542{
543 struct nvm_tgt_dev *dev = pblk->dev;
544 struct nvm_geo *geo = &dev->geo;
545 u8 *meta;
546 int i, nr_blks, blk_per_lun;
547 int ret;
548
549 blk_per_lun = geo->num_chk * geo->pln_mode;
550 nr_blks = blk_per_lun * geo->all_luns;
551
552 meta = kmalloc(nr_blks, GFP_KERNEL);
553 if (!meta)
554 return ERR_PTR(-ENOMEM);
555
556 for (i = 0; i < geo->all_luns; i++) {
557 struct pblk_lun *rlun = &pblk->luns[i];
558 u8 *meta_pos = meta + i * blk_per_lun;
559
560 ret = pblk_bb_get_tbl(dev, rlun, meta_pos, blk_per_lun);
561 if (ret) {
562 kfree(meta);
563 return ERR_PTR(-EIO);
564 }
565 }
566
567 return meta;
568}
569
570static void *pblk_chunk_get_meta(struct pblk *pblk)
571{
572 struct nvm_tgt_dev *dev = pblk->dev;
573 struct nvm_geo *geo = &dev->geo;
574
575 if (geo->version == NVM_OCSSD_SPEC_12)
576 return pblk_bb_get_meta(pblk);
577 else
578 return pblk_chunk_get_info(pblk);
579}
580
581static int pblk_luns_init(struct pblk *pblk)
582{
583 struct nvm_tgt_dev *dev = pblk->dev;
584 struct nvm_geo *geo = &dev->geo;
585 struct pblk_lun *rlun;
586 int i;
587
588 /* TODO: Implement unbalanced LUN support */
589 if (geo->num_lun < 0) {
590 pr_err("pblk: unbalanced LUN config.\n");
591 return -EINVAL;
592 }
593
594 pblk->luns = kcalloc(geo->all_luns, sizeof(struct pblk_lun),
595 GFP_KERNEL);
596 if (!pblk->luns)
597 return -ENOMEM;
598
599 for (i = 0; i < geo->all_luns; i++) {
600 /* Stripe across channels */
601 int ch = i % geo->num_ch;
602 int lun_raw = i / geo->num_ch;
603 int lunid = lun_raw + ch * geo->num_lun;
604
605 rlun = &pblk->luns[i];
606 rlun->bppa = dev->luns[lunid];
607
608 sema_init(&rlun->wr_sem, 1);
609 }
610
611 return 0;
612}
613
614/* See comment over struct line_emeta definition */
615static unsigned int calc_emeta_len(struct pblk *pblk)
616{
617 struct pblk_line_meta *lm = &pblk->lm;
618 struct pblk_line_mgmt *l_mg = &pblk->l_mg;
619 struct nvm_tgt_dev *dev = pblk->dev;
620 struct nvm_geo *geo = &dev->geo;
621
622 /* Round to sector size so that lba_list starts on its own sector */
623 lm->emeta_sec[1] = DIV_ROUND_UP(
624 sizeof(struct line_emeta) + lm->blk_bitmap_len +
625 sizeof(struct wa_counters), geo->csecs);
626 lm->emeta_len[1] = lm->emeta_sec[1] * geo->csecs;
627
628 /* Round to sector size so that vsc_list starts on its own sector */
629 lm->dsec_per_line = lm->sec_per_line - lm->emeta_sec[0];
630 lm->emeta_sec[2] = DIV_ROUND_UP(lm->dsec_per_line * sizeof(u64),
631 geo->csecs);
632 lm->emeta_len[2] = lm->emeta_sec[2] * geo->csecs;
633
634 lm->emeta_sec[3] = DIV_ROUND_UP(l_mg->nr_lines * sizeof(u32),
635 geo->csecs);
636 lm->emeta_len[3] = lm->emeta_sec[3] * geo->csecs;
637
638 lm->vsc_list_len = l_mg->nr_lines * sizeof(u32);
639
640 return (lm->emeta_len[1] + lm->emeta_len[2] + lm->emeta_len[3]);
641}
642
643static void pblk_set_provision(struct pblk *pblk, long nr_free_blks)
644{
645 struct nvm_tgt_dev *dev = pblk->dev;
646 struct pblk_line_mgmt *l_mg = &pblk->l_mg;
647 struct pblk_line_meta *lm = &pblk->lm;
648 struct nvm_geo *geo = &dev->geo;
649 sector_t provisioned;
650 int sec_meta, blk_meta;
651
652 if (geo->op == NVM_TARGET_DEFAULT_OP)
653 pblk->op = PBLK_DEFAULT_OP;
654 else
655 pblk->op = geo->op;
656
657 provisioned = nr_free_blks;
658 provisioned *= (100 - pblk->op);
659 sector_div(provisioned, 100);
660
661 pblk->op_blks = nr_free_blks - provisioned;
662
663 /* Internally pblk manages all free blocks, but all calculations based
664 * on user capacity consider only provisioned blocks
665 */
666 pblk->rl.total_blocks = nr_free_blks;
667 pblk->rl.nr_secs = nr_free_blks * geo->clba;
668
669 /* Consider sectors used for metadata */
670 sec_meta = (lm->smeta_sec + lm->emeta_sec[0]) * l_mg->nr_free_lines;
671 blk_meta = DIV_ROUND_UP(sec_meta, geo->clba);
672
673 pblk->capacity = (provisioned - blk_meta) * geo->clba;
674
675 atomic_set(&pblk->rl.free_blocks, nr_free_blks);
676 atomic_set(&pblk->rl.free_user_blocks, nr_free_blks);
677}
678
679static int pblk_setup_line_meta_12(struct pblk *pblk, struct pblk_line *line,
680 void *chunk_meta)
681{
682 struct nvm_tgt_dev *dev = pblk->dev;
683 struct nvm_geo *geo = &dev->geo;
684 struct pblk_line_meta *lm = &pblk->lm;
685 int i, chk_per_lun, nr_bad_chks = 0;
686
687 chk_per_lun = geo->num_chk * geo->pln_mode;
688
689 for (i = 0; i < lm->blk_per_line; i++) {
690 struct pblk_lun *rlun = &pblk->luns[i];
691 struct nvm_chk_meta *chunk;
692 int pos = pblk_ppa_to_pos(geo, rlun->bppa);
693 u8 *lun_bb_meta = chunk_meta + pos * chk_per_lun;
694
695 chunk = &line->chks[pos];
696
697 /*
698 * In 1.2 spec. chunk state is not persisted by the device. Thus
699 * some of the values are reset each time pblk is instantiated.
700 */
701 if (lun_bb_meta[line->id] == NVM_BLK_T_FREE)
702 chunk->state = NVM_CHK_ST_FREE;
703 else
704 chunk->state = NVM_CHK_ST_OFFLINE;
705
706 chunk->type = NVM_CHK_TP_W_SEQ;
707 chunk->wi = 0;
708 chunk->slba = -1;
709 chunk->cnlb = geo->clba;
710 chunk->wp = 0;
711
712 if (!(chunk->state & NVM_CHK_ST_OFFLINE))
713 continue;
714
715 set_bit(pos, line->blk_bitmap);
716 nr_bad_chks++;
717 }
718
719 return nr_bad_chks;
720}
721
722static int pblk_setup_line_meta_20(struct pblk *pblk, struct pblk_line *line,
723 struct nvm_chk_meta *meta)
724{
725 struct nvm_tgt_dev *dev = pblk->dev;
726 struct nvm_geo *geo = &dev->geo;
727 struct pblk_line_meta *lm = &pblk->lm;
728 int i, nr_bad_chks = 0;
729
730 for (i = 0; i < lm->blk_per_line; i++) {
731 struct pblk_lun *rlun = &pblk->luns[i];
732 struct nvm_chk_meta *chunk;
733 struct nvm_chk_meta *chunk_meta;
734 struct ppa_addr ppa;
735 int pos;
736
737 ppa = rlun->bppa;
738 pos = pblk_ppa_to_pos(geo, ppa);
739 chunk = &line->chks[pos];
740
741 ppa.m.chk = line->id;
742 chunk_meta = pblk_chunk_get_off(pblk, meta, ppa);
743
744 chunk->state = chunk_meta->state;
745 chunk->type = chunk_meta->type;
746 chunk->wi = chunk_meta->wi;
747 chunk->slba = chunk_meta->slba;
748 chunk->cnlb = chunk_meta->cnlb;
749 chunk->wp = chunk_meta->wp;
750
751 if (!(chunk->state & NVM_CHK_ST_OFFLINE))
752 continue;
753
754 if (chunk->type & NVM_CHK_TP_SZ_SPEC) {
755 WARN_ONCE(1, "pblk: custom-sized chunks unsupported\n");
756 continue;
757 }
758
759 set_bit(pos, line->blk_bitmap);
760 nr_bad_chks++;
761 }
762
763 return nr_bad_chks;
764}
765
766static long pblk_setup_line_meta(struct pblk *pblk, struct pblk_line *line,
767 void *chunk_meta, int line_id)
768{
769 struct nvm_tgt_dev *dev = pblk->dev;
770 struct nvm_geo *geo = &dev->geo;
771 struct pblk_line_mgmt *l_mg = &pblk->l_mg;
772 struct pblk_line_meta *lm = &pblk->lm;
773 long nr_bad_chks, chk_in_line;
774
775 line->pblk = pblk;
776 line->id = line_id;
777 line->type = PBLK_LINETYPE_FREE;
778 line->state = PBLK_LINESTATE_NEW;
779 line->gc_group = PBLK_LINEGC_NONE;
780 line->vsc = &l_mg->vsc_list[line_id];
781 spin_lock_init(&line->lock);
782
783 if (geo->version == NVM_OCSSD_SPEC_12)
784 nr_bad_chks = pblk_setup_line_meta_12(pblk, line, chunk_meta);
785 else
786 nr_bad_chks = pblk_setup_line_meta_20(pblk, line, chunk_meta);
787
788 chk_in_line = lm->blk_per_line - nr_bad_chks;
789 if (nr_bad_chks < 0 || nr_bad_chks > lm->blk_per_line ||
790 chk_in_line < lm->min_blk_line) {
791 line->state = PBLK_LINESTATE_BAD;
792 list_add_tail(&line->list, &l_mg->bad_list);
793 return 0;
794 }
795
796 atomic_set(&line->blk_in_line, chk_in_line);
797 list_add_tail(&line->list, &l_mg->free_list);
798 l_mg->nr_free_lines++;
799
800 return chk_in_line;
801}
802
803static int pblk_alloc_line_meta(struct pblk *pblk, struct pblk_line *line)
804{
805 struct pblk_line_meta *lm = &pblk->lm;
806
807 line->blk_bitmap = kzalloc(lm->blk_bitmap_len, GFP_KERNEL);
808 if (!line->blk_bitmap)
809 return -ENOMEM;
810
811 line->erase_bitmap = kzalloc(lm->blk_bitmap_len, GFP_KERNEL);
812 if (!line->erase_bitmap) {
813 kfree(line->blk_bitmap);
814 return -ENOMEM;
815 }
816
817 line->chks = kmalloc(lm->blk_per_line * sizeof(struct nvm_chk_meta),
818 GFP_KERNEL);
819 if (!line->chks) {
820 kfree(line->erase_bitmap);
821 kfree(line->blk_bitmap);
822 return -ENOMEM;
823 }
824
825 return 0;
826}
827
828static int pblk_line_mg_init(struct pblk *pblk)
829{
830 struct nvm_tgt_dev *dev = pblk->dev;
831 struct nvm_geo *geo = &dev->geo;
832 struct pblk_line_mgmt *l_mg = &pblk->l_mg;
833 struct pblk_line_meta *lm = &pblk->lm;
834 int i, bb_distance;
835
836 l_mg->nr_lines = geo->num_chk;
837 l_mg->log_line = l_mg->data_line = NULL;
838 l_mg->l_seq_nr = l_mg->d_seq_nr = 0;
839 l_mg->nr_free_lines = 0;
840 bitmap_zero(&l_mg->meta_bitmap, PBLK_DATA_LINES);
841
842 INIT_LIST_HEAD(&l_mg->free_list);
843 INIT_LIST_HEAD(&l_mg->corrupt_list);
844 INIT_LIST_HEAD(&l_mg->bad_list);
845 INIT_LIST_HEAD(&l_mg->gc_full_list);
846 INIT_LIST_HEAD(&l_mg->gc_high_list);
847 INIT_LIST_HEAD(&l_mg->gc_mid_list);
848 INIT_LIST_HEAD(&l_mg->gc_low_list);
849 INIT_LIST_HEAD(&l_mg->gc_empty_list);
850
851 INIT_LIST_HEAD(&l_mg->emeta_list);
852
853 l_mg->gc_lists[0] = &l_mg->gc_high_list;
854 l_mg->gc_lists[1] = &l_mg->gc_mid_list;
855 l_mg->gc_lists[2] = &l_mg->gc_low_list;
856
857 spin_lock_init(&l_mg->free_lock);
858 spin_lock_init(&l_mg->close_lock);
859 spin_lock_init(&l_mg->gc_lock);
860
861 l_mg->vsc_list = kcalloc(l_mg->nr_lines, sizeof(__le32), GFP_KERNEL);
862 if (!l_mg->vsc_list)
863 goto fail;
864
865 l_mg->bb_template = kzalloc(lm->sec_bitmap_len, GFP_KERNEL);
866 if (!l_mg->bb_template)
867 goto fail_free_vsc_list;
868
869 l_mg->bb_aux = kzalloc(lm->sec_bitmap_len, GFP_KERNEL);
870 if (!l_mg->bb_aux)
871 goto fail_free_bb_template;
872
873 /* smeta is always small enough to fit on a kmalloc memory allocation,
874 * emeta depends on the number of LUNs allocated to the pblk instance
875 */
876 for (i = 0; i < PBLK_DATA_LINES; i++) {
877 l_mg->sline_meta[i] = kmalloc(lm->smeta_len, GFP_KERNEL);
878 if (!l_mg->sline_meta[i])
879 goto fail_free_smeta;
880 }
881
882 /* emeta allocates three different buffers for managing metadata with
883 * in-memory and in-media layouts
884 */
885 for (i = 0; i < PBLK_DATA_LINES; i++) {
886 struct pblk_emeta *emeta;
887
888 emeta = kmalloc(sizeof(struct pblk_emeta), GFP_KERNEL);
889 if (!emeta)
890 goto fail_free_emeta;
891
892 if (lm->emeta_len[0] > KMALLOC_MAX_CACHE_SIZE) {
893 l_mg->emeta_alloc_type = PBLK_VMALLOC_META;
894
895 emeta->buf = vmalloc(lm->emeta_len[0]);
896 if (!emeta->buf) {
897 kfree(emeta);
898 goto fail_free_emeta;
899 }
900
901 emeta->nr_entries = lm->emeta_sec[0];
902 l_mg->eline_meta[i] = emeta;
903 } else {
904 l_mg->emeta_alloc_type = PBLK_KMALLOC_META;
905
906 emeta->buf = kmalloc(lm->emeta_len[0], GFP_KERNEL);
907 if (!emeta->buf) {
908 kfree(emeta);
909 goto fail_free_emeta;
910 }
911
912 emeta->nr_entries = lm->emeta_sec[0];
913 l_mg->eline_meta[i] = emeta;
914 }
915 }
916
917 for (i = 0; i < l_mg->nr_lines; i++)
918 l_mg->vsc_list[i] = cpu_to_le32(EMPTY_ENTRY);
919
920 bb_distance = (geo->all_luns) * geo->ws_opt;
921 for (i = 0; i < lm->sec_per_line; i += bb_distance)
922 bitmap_set(l_mg->bb_template, i, geo->ws_opt);
923
924 return 0;
925
926fail_free_emeta:
927 while (--i >= 0) {
928 if (l_mg->emeta_alloc_type == PBLK_VMALLOC_META)
929 vfree(l_mg->eline_meta[i]->buf);
930 else
931 kfree(l_mg->eline_meta[i]->buf);
932 kfree(l_mg->eline_meta[i]);
933 }
934fail_free_smeta:
935 for (i = 0; i < PBLK_DATA_LINES; i++)
936 kfree(l_mg->sline_meta[i]);
937 kfree(l_mg->bb_aux);
938fail_free_bb_template:
939 kfree(l_mg->bb_template);
940fail_free_vsc_list:
941 kfree(l_mg->vsc_list);
942fail:
943 return -ENOMEM;
944}
945
946static int pblk_line_meta_init(struct pblk *pblk)
947{
948 struct nvm_tgt_dev *dev = pblk->dev;
949 struct nvm_geo *geo = &dev->geo;
950 struct pblk_line_meta *lm = &pblk->lm;
951 unsigned int smeta_len, emeta_len;
952 int i;
953
954 lm->sec_per_line = geo->clba * geo->all_luns;
955 lm->blk_per_line = geo->all_luns;
956 lm->blk_bitmap_len = BITS_TO_LONGS(geo->all_luns) * sizeof(long);
957 lm->sec_bitmap_len = BITS_TO_LONGS(lm->sec_per_line) * sizeof(long);
958 lm->lun_bitmap_len = BITS_TO_LONGS(geo->all_luns) * sizeof(long);
959 lm->mid_thrs = lm->sec_per_line / 2;
960 lm->high_thrs = lm->sec_per_line / 4;
961 lm->meta_distance = (geo->all_luns / 2) * pblk->min_write_pgs;
962
963 /* Calculate necessary pages for smeta. See comment over struct
964 * line_smeta definition
965 */
966 i = 1;
967add_smeta_page:
968 lm->smeta_sec = i * geo->ws_opt;
969 lm->smeta_len = lm->smeta_sec * geo->csecs;
970
971 smeta_len = sizeof(struct line_smeta) + lm->lun_bitmap_len;
972 if (smeta_len > lm->smeta_len) {
973 i++;
974 goto add_smeta_page;
975 }
976
977 /* Calculate necessary pages for emeta. See comment over struct
978 * line_emeta definition
979 */
980 i = 1;
981add_emeta_page:
982 lm->emeta_sec[0] = i * geo->ws_opt;
983 lm->emeta_len[0] = lm->emeta_sec[0] * geo->csecs;
984
985 emeta_len = calc_emeta_len(pblk);
986 if (emeta_len > lm->emeta_len[0]) {
987 i++;
988 goto add_emeta_page;
989 }
990
991 lm->emeta_bb = geo->all_luns > i ? geo->all_luns - i : 0;
992
993 lm->min_blk_line = 1;
994 if (geo->all_luns > 1)
995 lm->min_blk_line += DIV_ROUND_UP(lm->smeta_sec +
996 lm->emeta_sec[0], geo->clba);
997
998 if (lm->min_blk_line > lm->blk_per_line) {
999 pr_err("pblk: config. not supported. Min. LUN in line:%d\n",
1000 lm->blk_per_line);
1001 return -EINVAL;
1002 }
1003
1004 return 0;
1005}
1006
1007static int pblk_lines_init(struct pblk *pblk)
1008{
1009 struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1010 struct pblk_line *line;
1011 void *chunk_meta;
1012 long nr_free_chks = 0;
1013 int i, ret;
1014
1015 ret = pblk_line_meta_init(pblk);
1016 if (ret)
1017 return ret;
1018
1019 ret = pblk_line_mg_init(pblk);
1020 if (ret)
1021 return ret;
1022
1023 ret = pblk_luns_init(pblk);
1024 if (ret)
1025 goto fail_free_meta;
1026
1027 chunk_meta = pblk_chunk_get_meta(pblk);
1028 if (IS_ERR(chunk_meta)) {
1029 ret = PTR_ERR(chunk_meta);
1030 goto fail_free_luns;
1031 }
1032
1033 pblk->lines = kcalloc(l_mg->nr_lines, sizeof(struct pblk_line),
1034 GFP_KERNEL);
1035 if (!pblk->lines) {
1036 ret = -ENOMEM;
1037 goto fail_free_chunk_meta;
1038 }
1039
1040 for (i = 0; i < l_mg->nr_lines; i++) {
1041 line = &pblk->lines[i];
1042
1043 ret = pblk_alloc_line_meta(pblk, line);
1044 if (ret)
1045 goto fail_free_lines;
1046
1047 nr_free_chks += pblk_setup_line_meta(pblk, line, chunk_meta, i);
1048 }
1049
1050 pblk_set_provision(pblk, nr_free_chks);
1051
1052 kfree(chunk_meta);
1053 return 0;
1054
1055fail_free_lines:
1056 while (--i >= 0)
1057 pblk_line_meta_free(&pblk->lines[i]);
1058 kfree(pblk->lines);
1059fail_free_chunk_meta:
1060 kfree(chunk_meta);
1061fail_free_luns:
1062 kfree(pblk->luns);
1063fail_free_meta:
1064 pblk_line_mg_free(pblk);
1065
1066 return ret;
1067}
1068
1069static int pblk_writer_init(struct pblk *pblk)
1070{
1071 pblk->writer_ts = kthread_create(pblk_write_ts, pblk, "pblk-writer-t");
1072 if (IS_ERR(pblk->writer_ts)) {
1073 int err = PTR_ERR(pblk->writer_ts);
1074
1075 if (err != -EINTR)
1076 pr_err("pblk: could not allocate writer kthread (%d)\n",
1077 err);
1078 return err;
1079 }
1080
1081 timer_setup(&pblk->wtimer, pblk_write_timer_fn, 0);
1082 mod_timer(&pblk->wtimer, jiffies + msecs_to_jiffies(100));
1083
1084 return 0;
1085}
1086
1087static void pblk_writer_stop(struct pblk *pblk)
1088{
1089 /* The pipeline must be stopped and the write buffer emptied before the
1090 * write thread is stopped
1091 */
1092 WARN(pblk_rb_read_count(&pblk->rwb),
1093 "Stopping not fully persisted write buffer\n");
1094
1095 WARN(pblk_rb_sync_count(&pblk->rwb),
1096 "Stopping not fully synced write buffer\n");
1097
1098 del_timer_sync(&pblk->wtimer);
1099 if (pblk->writer_ts)
1100 kthread_stop(pblk->writer_ts);
1101}
1102
1103static void pblk_free(struct pblk *pblk)
1104{
1105 pblk_lines_free(pblk);
1106 pblk_l2p_free(pblk);
1107 pblk_rwb_free(pblk);
1108 pblk_core_free(pblk);
1109
1110 kfree(pblk);
1111}
1112
1113static void pblk_tear_down(struct pblk *pblk)
1114{
1115 pblk_pipeline_stop(pblk);
1116 pblk_writer_stop(pblk);
1117 pblk_rb_sync_l2p(&pblk->rwb);
1118 pblk_rl_free(&pblk->rl);
1119
1120 pr_debug("pblk: consistent tear down\n");
1121}
1122
1123static void pblk_exit(void *private)
1124{
1125 struct pblk *pblk = private;
1126
1127 down_write(&pblk_lock);
1128 pblk_gc_exit(pblk);
1129 pblk_tear_down(pblk);
1130
1131#ifdef CONFIG_NVM_DEBUG
1132 pr_info("pblk exit: L2P CRC: %x\n", pblk_l2p_crc(pblk));
1133#endif
1134
1135 pblk_free(pblk);
1136 up_write(&pblk_lock);
1137}
1138
1139static sector_t pblk_capacity(void *private)
1140{
1141 struct pblk *pblk = private;
1142
1143 return pblk->capacity * NR_PHY_IN_LOG;
1144}
1145
1146static void *pblk_init(struct nvm_tgt_dev *dev, struct gendisk *tdisk,
1147 int flags)
1148{
1149 struct nvm_geo *geo = &dev->geo;
1150 struct request_queue *bqueue = dev->q;
1151 struct request_queue *tqueue = tdisk->queue;
1152 struct pblk *pblk;
1153 int ret;
1154
1155 /* pblk supports 1.2 and 2.0 versions */
1156 if (!(geo->version == NVM_OCSSD_SPEC_12 ||
1157 geo->version == NVM_OCSSD_SPEC_20)) {
1158 pr_err("pblk: OCSSD version not supported (%u)\n",
1159 geo->version);
1160 return ERR_PTR(-EINVAL);
1161 }
1162
1163 if (geo->version == NVM_OCSSD_SPEC_12 && geo->dom & NVM_RSP_L2P) {
1164 pr_err("pblk: host-side L2P table not supported. (%x)\n",
1165 geo->dom);
1166 return ERR_PTR(-EINVAL);
1167 }
1168
1169 pblk = kzalloc(sizeof(struct pblk), GFP_KERNEL);
1170 if (!pblk)
1171 return ERR_PTR(-ENOMEM);
1172
1173 pblk->dev = dev;
1174 pblk->disk = tdisk;
1175 pblk->state = PBLK_STATE_RUNNING;
1176 pblk->gc.gc_enabled = 0;
1177
1178 spin_lock_init(&pblk->trans_lock);
1179 spin_lock_init(&pblk->lock);
1180
1181#ifdef CONFIG_NVM_DEBUG
1182 atomic_long_set(&pblk->inflight_writes, 0);
1183 atomic_long_set(&pblk->padded_writes, 0);
1184 atomic_long_set(&pblk->padded_wb, 0);
1185 atomic_long_set(&pblk->req_writes, 0);
1186 atomic_long_set(&pblk->sub_writes, 0);
1187 atomic_long_set(&pblk->sync_writes, 0);
1188 atomic_long_set(&pblk->inflight_reads, 0);
1189 atomic_long_set(&pblk->cache_reads, 0);
1190 atomic_long_set(&pblk->sync_reads, 0);
1191 atomic_long_set(&pblk->recov_writes, 0);
1192 atomic_long_set(&pblk->recov_writes, 0);
1193 atomic_long_set(&pblk->recov_gc_writes, 0);
1194 atomic_long_set(&pblk->recov_gc_reads, 0);
1195#endif
1196
1197 atomic_long_set(&pblk->read_failed, 0);
1198 atomic_long_set(&pblk->read_empty, 0);
1199 atomic_long_set(&pblk->read_high_ecc, 0);
1200 atomic_long_set(&pblk->read_failed_gc, 0);
1201 atomic_long_set(&pblk->write_failed, 0);
1202 atomic_long_set(&pblk->erase_failed, 0);
1203
1204 ret = pblk_core_init(pblk);
1205 if (ret) {
1206 pr_err("pblk: could not initialize core\n");
1207 goto fail;
1208 }
1209
1210 ret = pblk_lines_init(pblk);
1211 if (ret) {
1212 pr_err("pblk: could not initialize lines\n");
1213 goto fail_free_core;
1214 }
1215
1216 ret = pblk_rwb_init(pblk);
1217 if (ret) {
1218 pr_err("pblk: could not initialize write buffer\n");
1219 goto fail_free_lines;
1220 }
1221
1222 ret = pblk_l2p_init(pblk, flags & NVM_TARGET_FACTORY);
1223 if (ret) {
1224 pr_err("pblk: could not initialize maps\n");
1225 goto fail_free_rwb;
1226 }
1227
1228 ret = pblk_writer_init(pblk);
1229 if (ret) {
1230 if (ret != -EINTR)
1231 pr_err("pblk: could not initialize write thread\n");
1232 goto fail_free_l2p;
1233 }
1234
1235 ret = pblk_gc_init(pblk);
1236 if (ret) {
1237 pr_err("pblk: could not initialize gc\n");
1238 goto fail_stop_writer;
1239 }
1240
1241 /* inherit the size from the underlying device */
1242 blk_queue_logical_block_size(tqueue, queue_physical_block_size(bqueue));
1243 blk_queue_max_hw_sectors(tqueue, queue_max_hw_sectors(bqueue));
1244
1245 blk_queue_write_cache(tqueue, true, false);
1246
1247 tqueue->limits.discard_granularity = geo->clba * geo->csecs;
1248 tqueue->limits.discard_alignment = 0;
1249 blk_queue_max_discard_sectors(tqueue, UINT_MAX >> 9);
1250 blk_queue_flag_set(QUEUE_FLAG_DISCARD, tqueue);
1251
1252 pr_info("pblk(%s): luns:%u, lines:%d, secs:%llu, buf entries:%u\n",
1253 tdisk->disk_name,
1254 geo->all_luns, pblk->l_mg.nr_lines,
1255 (unsigned long long)pblk->rl.nr_secs,
1256 pblk->rwb.nr_entries);
1257
1258 wake_up_process(pblk->writer_ts);
1259
1260 /* Check if we need to start GC */
1261 pblk_gc_should_kick(pblk);
1262
1263 return pblk;
1264
1265fail_stop_writer:
1266 pblk_writer_stop(pblk);
1267fail_free_l2p:
1268 pblk_l2p_free(pblk);
1269fail_free_rwb:
1270 pblk_rwb_free(pblk);
1271fail_free_lines:
1272 pblk_lines_free(pblk);
1273fail_free_core:
1274 pblk_core_free(pblk);
1275fail:
1276 kfree(pblk);
1277 return ERR_PTR(ret);
1278}
1279
1280/* physical block device target */
1281static struct nvm_tgt_type tt_pblk = {
1282 .name = "pblk",
1283 .version = {1, 0, 0},
1284
1285 .make_rq = pblk_make_rq,
1286 .capacity = pblk_capacity,
1287
1288 .init = pblk_init,
1289 .exit = pblk_exit,
1290
1291 .sysfs_init = pblk_sysfs_init,
1292 .sysfs_exit = pblk_sysfs_exit,
1293 .owner = THIS_MODULE,
1294};
1295
1296static int __init pblk_module_init(void)
1297{
1298 int ret;
1299
1300 pblk_bio_set = bioset_create(BIO_POOL_SIZE, 0, 0);
1301 if (!pblk_bio_set)
1302 return -ENOMEM;
1303 ret = nvm_register_tgt_type(&tt_pblk);
1304 if (ret)
1305 bioset_free(pblk_bio_set);
1306 return ret;
1307}
1308
1309static void pblk_module_exit(void)
1310{
1311 bioset_free(pblk_bio_set);
1312 nvm_unregister_tgt_type(&tt_pblk);
1313}
1314
1315module_init(pblk_module_init);
1316module_exit(pblk_module_exit);
1317MODULE_AUTHOR("Javier Gonzalez <javier@cnexlabs.com>");
1318MODULE_AUTHOR("Matias Bjorling <matias@cnexlabs.com>");
1319MODULE_LICENSE("GPL v2");
1320MODULE_DESCRIPTION("Physical Block-Device for Open-Channel SSDs");