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