drivers/lightnvm/pblk-init.c

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