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                 guid_gen(&pblk->instance_uuid);
 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 static int pblk_set_addrf_12(struct pblk *pblk, struct nvm_geo *geo,
 211                              struct nvm_addrf_12 *dst)
 212 {
 213         struct nvm_addrf_12 *src = (struct nvm_addrf_12 *)&geo->addrf;
 214         int power_len;
 215
 216         /* Re-calculate channel and lun format to adapt to configuration */
 217         power_len = get_count_order(geo->num_ch);
 218         if (1 << power_len != geo->num_ch) {
 219                 pblk_err(pblk, "supports only power-of-two channel config.\n");
 220                 return -EINVAL;
 221         }
 222         dst->ch_len = power_len;
 223
 224         power_len = get_count_order(geo->num_lun);
 225         if (1 << power_len != geo->num_lun) {
 226                 pblk_err(pblk, "supports only power-of-two LUN config.\n");
 227                 return -EINVAL;
 228         }
 229         dst->lun_len = power_len;
 230
 231         dst->blk_len = src->blk_len;
 232         dst->pg_len = src->pg_len;
 233         dst->pln_len = src->pln_len;
 234         dst->sec_len = src->sec_len;
 235
 236         dst->sec_offset = 0;
 237         dst->pln_offset = dst->sec_len;
 238         dst->ch_offset = dst->pln_offset + dst->pln_len;
 239         dst->lun_offset = dst->ch_offset + dst->ch_len;
 240         dst->pg_offset = dst->lun_offset + dst->lun_len;
 241         dst->blk_offset = dst->pg_offset + dst->pg_len;
 242
 243         dst->sec_mask = ((1ULL << dst->sec_len) - 1) << dst->sec_offset;
 244         dst->pln_mask = ((1ULL << dst->pln_len) - 1) << dst->pln_offset;
 245         dst->ch_mask = ((1ULL << dst->ch_len) - 1) << dst->ch_offset;
 246         dst->lun_mask = ((1ULL << dst->lun_len) - 1) << dst->lun_offset;
 247         dst->pg_mask = ((1ULL << dst->pg_len) - 1) << dst->pg_offset;
 248         dst->blk_mask = ((1ULL << dst->blk_len) - 1) << dst->blk_offset;
 249
 250         return dst->blk_offset + src->blk_len;
 251 }
 252
 253 static int pblk_set_addrf_20(struct nvm_geo *geo, struct nvm_addrf *adst,
 254                              struct pblk_addrf *udst)
 255 {
 256         struct nvm_addrf *src = &geo->addrf;
 257
 258         adst->ch_len = get_count_order(geo->num_ch);
 259         adst->lun_len = get_count_order(geo->num_lun);
 260         adst->chk_len = src->chk_len;
 261         adst->sec_len = src->sec_len;
 262
 263         adst->sec_offset = 0;
 264         adst->ch_offset = adst->sec_len;
 265         adst->lun_offset = adst->ch_offset + adst->ch_len;
 266         adst->chk_offset = adst->lun_offset + adst->lun_len;
 267
 268         adst->sec_mask = ((1ULL << adst->sec_len) - 1) << adst->sec_offset;
 269         adst->chk_mask = ((1ULL << adst->chk_len) - 1) << adst->chk_offset;
 270         adst->lun_mask = ((1ULL << adst->lun_len) - 1) << adst->lun_offset;
 271         adst->ch_mask = ((1ULL << adst->ch_len) - 1) << adst->ch_offset;
 272
 273         udst->sec_stripe = geo->ws_opt;
 274         udst->ch_stripe = geo->num_ch;
 275         udst->lun_stripe = geo->num_lun;
 276
 277         udst->sec_lun_stripe = udst->sec_stripe * udst->ch_stripe;
 278         udst->sec_ws_stripe = udst->sec_lun_stripe * udst->lun_stripe;
 279
 280         return adst->chk_offset + adst->chk_len;
 281 }
 282
 283 static int pblk_set_addrf(struct pblk *pblk)
 284 {
 285         struct nvm_tgt_dev *dev = pblk->dev;
 286         struct nvm_geo *geo = &dev->geo;
 287         int mod;
 288
 289         switch (geo->version) {
 290         case NVM_OCSSD_SPEC_12:
 291                 div_u64_rem(geo->clba, pblk->min_write_pgs, &mod);
 292                 if (mod) {
 293                         pblk_err(pblk, "bad configuration of sectors/pages\n");
 294                         return -EINVAL;
 295                 }
 296
 297                 pblk->addrf_len = pblk_set_addrf_12(pblk, geo,
 298                                                         (void *)&pblk->addrf);
 299                 break;
 300         case NVM_OCSSD_SPEC_20:
 301                 pblk->addrf_len = pblk_set_addrf_20(geo, (void *)&pblk->addrf,
 302                                                         &pblk->uaddrf);
 303                 break;
 304         default:
 305                 pblk_err(pblk, "OCSSD revision not supported (%d)\n",
 306                                                                 geo->version);
 307                 return -EINVAL;
 308         }
 309
 310         return 0;
 311 }
 312
 313 static int pblk_create_global_caches(void)
 314 {
 315
 316         pblk_caches.ws = kmem_cache_create("pblk_blk_ws",
 317                                 sizeof(struct pblk_line_ws), 0, 0, NULL);
 318         if (!pblk_caches.ws)
 319                 return -ENOMEM;
 320
 321         pblk_caches.rec = kmem_cache_create("pblk_rec",
 322                                 sizeof(struct pblk_rec_ctx), 0, 0, NULL);
 323         if (!pblk_caches.rec)
 324                 goto fail_destroy_ws;
 325
 326         pblk_caches.g_rq = kmem_cache_create("pblk_g_rq", pblk_g_rq_size,
 327                                 0, 0, NULL);
 328         if (!pblk_caches.g_rq)
 329                 goto fail_destroy_rec;
 330
 331         pblk_caches.w_rq = kmem_cache_create("pblk_w_rq", pblk_w_rq_size,
 332                                 0, 0, NULL);
 333         if (!pblk_caches.w_rq)
 334                 goto fail_destroy_g_rq;
 335
 336         return 0;
 337
 338 fail_destroy_g_rq:
 339         kmem_cache_destroy(pblk_caches.g_rq);
 340 fail_destroy_rec:
 341         kmem_cache_destroy(pblk_caches.rec);
 342 fail_destroy_ws:
 343         kmem_cache_destroy(pblk_caches.ws);
 344
 345         return -ENOMEM;
 346 }
 347
 348 static int pblk_get_global_caches(void)
 349 {
 350         int ret = 0;
 351
 352         mutex_lock(&pblk_caches.mutex);
 353
 354         if (kref_get_unless_zero(&pblk_caches.kref))
 355                 goto out;
 356
 357         ret = pblk_create_global_caches();
 358         if (!ret)
 359                 kref_init(&pblk_caches.kref);
 360
 361 out:
 362         mutex_unlock(&pblk_caches.mutex);
 363         return ret;
 364 }
 365
 366 static void pblk_destroy_global_caches(struct kref *ref)
 367 {
 368         struct pblk_global_caches *c;
 369
 370         c = container_of(ref, struct pblk_global_caches, kref);
 371
 372         kmem_cache_destroy(c->ws);
 373         kmem_cache_destroy(c->rec);
 374         kmem_cache_destroy(c->g_rq);
 375         kmem_cache_destroy(c->w_rq);
 376 }
 377
 378 static void pblk_put_global_caches(void)
 379 {
 380         mutex_lock(&pblk_caches.mutex);
 381         kref_put(&pblk_caches.kref, pblk_destroy_global_caches);
 382         mutex_unlock(&pblk_caches.mutex);
 383 }
 384
 385 static int pblk_core_init(struct pblk *pblk)
 386 {
 387         struct nvm_tgt_dev *dev = pblk->dev;
 388         struct nvm_geo *geo = &dev->geo;
 389         int ret, max_write_ppas;
 390
 391         atomic64_set(&pblk->user_wa, 0);
 392         atomic64_set(&pblk->pad_wa, 0);
 393         atomic64_set(&pblk->gc_wa, 0);
 394         pblk->user_rst_wa = 0;
 395         pblk->pad_rst_wa = 0;
 396         pblk->gc_rst_wa = 0;
 397
 398         atomic64_set(&pblk->nr_flush, 0);
 399         pblk->nr_flush_rst = 0;
 400
 401         pblk->min_write_pgs = geo->ws_opt;
 402         pblk->min_write_pgs_data = pblk->min_write_pgs;
 403         max_write_ppas = pblk->min_write_pgs * geo->all_luns;
 404         pblk->max_write_pgs = min_t(int, max_write_ppas, NVM_MAX_VLBA);
 405         pblk->max_write_pgs = min_t(int, pblk->max_write_pgs,
 406                 queue_max_hw_sectors(dev->q) / (geo->csecs >> SECTOR_SHIFT));
 407         pblk_set_sec_per_write(pblk, pblk->min_write_pgs);
 408
 409         pblk->oob_meta_size = geo->sos;
 410         if (!pblk_is_oob_meta_supported(pblk)) {
 411                 /* For drives which does not have OOB metadata feature
 412                  * in order to support recovery feature we need to use
 413                  * so called packed metadata. Packed metada will store
 414                  * the same information as OOB metadata (l2p table mapping,
 415                  * but in the form of the single page at the end of
 416                  * every write request.
 417                  */
 418                 if (pblk->min_write_pgs
 419                         * sizeof(struct pblk_sec_meta) > PAGE_SIZE) {
 420                         /* We want to keep all the packed metadata on single
 421                          * page per write requests. So we need to ensure that
 422                          * it will fit.
 423                          *
 424                          * This is more like sanity check, since there is
 425                          * no device with such a big minimal write size
 426                          * (above 1 metabytes).
 427                          */
 428                         pblk_err(pblk, "Not supported min write size\n");
 429                         return -EINVAL;
 430                 }
 431                 /* For packed meta approach we do some simplification.
 432                  * On read path we always issue requests which size
 433                  * equal to max_write_pgs, with all pages filled with
 434                  * user payload except of last one page which will be
 435                  * filled with packed metadata.
 436                  */
 437                 pblk->max_write_pgs = pblk->min_write_pgs;
 438                 pblk->min_write_pgs_data = pblk->min_write_pgs - 1;
 439         }
 440
 441         pblk->pad_dist = kcalloc(pblk->min_write_pgs - 1, sizeof(atomic64_t),
 442                                                                 GFP_KERNEL);
 443         if (!pblk->pad_dist)
 444                 return -ENOMEM;
 445
 446         if (pblk_get_global_caches())
 447                 goto fail_free_pad_dist;
 448
 449         /* Internal bios can be at most the sectors signaled by the device. */
 450         ret = mempool_init_page_pool(&pblk->page_bio_pool, NVM_MAX_VLBA, 0);
 451         if (ret)
 452                 goto free_global_caches;
 453
 454         ret = mempool_init_slab_pool(&pblk->gen_ws_pool, PBLK_GEN_WS_POOL_SIZE,
 455                                      pblk_caches.ws);
 456         if (ret)
 457                 goto free_page_bio_pool;
 458
 459         ret = mempool_init_slab_pool(&pblk->rec_pool, geo->all_luns,
 460                                      pblk_caches.rec);
 461         if (ret)
 462                 goto free_gen_ws_pool;
 463
 464         ret = mempool_init_slab_pool(&pblk->r_rq_pool, geo->all_luns,
 465                                      pblk_caches.g_rq);
 466         if (ret)
 467                 goto free_rec_pool;
 468
 469         ret = mempool_init_slab_pool(&pblk->e_rq_pool, geo->all_luns,
 470                                      pblk_caches.g_rq);
 471         if (ret)
 472                 goto free_r_rq_pool;
 473
 474         ret = mempool_init_slab_pool(&pblk->w_rq_pool, geo->all_luns,
 475                                      pblk_caches.w_rq);
 476         if (ret)
 477                 goto free_e_rq_pool;
 478
 479         pblk->close_wq = alloc_workqueue("pblk-close-wq",
 480                         WQ_MEM_RECLAIM | WQ_UNBOUND, PBLK_NR_CLOSE_JOBS);
 481         if (!pblk->close_wq)
 482                 goto free_w_rq_pool;
 483
 484         pblk->bb_wq = alloc_workqueue("pblk-bb-wq",
 485                         WQ_MEM_RECLAIM | WQ_UNBOUND, 0);
 486         if (!pblk->bb_wq)
 487                 goto free_close_wq;
 488
 489         pblk->r_end_wq = alloc_workqueue("pblk-read-end-wq",
 490                         WQ_MEM_RECLAIM | WQ_UNBOUND, 0);
 491         if (!pblk->r_end_wq)
 492                 goto free_bb_wq;
 493
 494         if (pblk_set_addrf(pblk))
 495                 goto free_r_end_wq;
 496
 497         INIT_LIST_HEAD(&pblk->compl_list);
 498         INIT_LIST_HEAD(&pblk->resubmit_list);
 499
 500         return 0;
 501
 502 free_r_end_wq:
 503         destroy_workqueue(pblk->r_end_wq);
 504 free_bb_wq:
 505         destroy_workqueue(pblk->bb_wq);
 506 free_close_wq:
 507         destroy_workqueue(pblk->close_wq);
 508 free_w_rq_pool:
 509         mempool_exit(&pblk->w_rq_pool);
 510 free_e_rq_pool:
 511         mempool_exit(&pblk->e_rq_pool);
 512 free_r_rq_pool:
 513         mempool_exit(&pblk->r_rq_pool);
 514 free_rec_pool:
 515         mempool_exit(&pblk->rec_pool);
 516 free_gen_ws_pool:
 517         mempool_exit(&pblk->gen_ws_pool);
 518 free_page_bio_pool:
 519         mempool_exit(&pblk->page_bio_pool);
 520 free_global_caches:
 521         pblk_put_global_caches();
 522 fail_free_pad_dist:
 523         kfree(pblk->pad_dist);
 524         return -ENOMEM;
 525 }
 526
 527 static void pblk_core_free(struct pblk *pblk)
 528 {
 529         if (pblk->close_wq)
 530                 destroy_workqueue(pblk->close_wq);
 531
 532         if (pblk->r_end_wq)
 533                 destroy_workqueue(pblk->r_end_wq);
 534
 535         if (pblk->bb_wq)
 536                 destroy_workqueue(pblk->bb_wq);
 537
 538         mempool_exit(&pblk->page_bio_pool);
 539         mempool_exit(&pblk->gen_ws_pool);
 540         mempool_exit(&pblk->rec_pool);
 541         mempool_exit(&pblk->r_rq_pool);
 542         mempool_exit(&pblk->e_rq_pool);
 543         mempool_exit(&pblk->w_rq_pool);
 544
 545         pblk_put_global_caches();
 546         kfree(pblk->pad_dist);
 547 }
 548
 549 static void pblk_line_mg_free(struct pblk *pblk)
 550 {
 551         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
 552         int i;
 553
 554         kfree(l_mg->bb_template);
 555         kfree(l_mg->bb_aux);
 556         kfree(l_mg->vsc_list);
 557
 558         for (i = 0; i < PBLK_DATA_LINES; i++) {
 559                 kfree(l_mg->sline_meta[i]);
 560                 pblk_mfree(l_mg->eline_meta[i]->buf, l_mg->emeta_alloc_type);
 561                 kfree(l_mg->eline_meta[i]);
 562         }
 563
 564         mempool_destroy(l_mg->bitmap_pool);
 565         kmem_cache_destroy(l_mg->bitmap_cache);
 566 }
 567
 568 static void pblk_line_meta_free(struct pblk_line_mgmt *l_mg,
 569                                 struct pblk_line *line)
 570 {
 571         struct pblk_w_err_gc *w_err_gc = line->w_err_gc;
 572
 573         kfree(line->blk_bitmap);
 574         kfree(line->erase_bitmap);
 575         kfree(line->chks);
 576
 577         pblk_mfree(w_err_gc->lba_list, l_mg->emeta_alloc_type);
 578         kfree(w_err_gc);
 579 }
 580
 581 static void pblk_lines_free(struct pblk *pblk)
 582 {
 583         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
 584         struct pblk_line *line;
 585         int i;
 586
 587         for (i = 0; i < l_mg->nr_lines; i++) {
 588                 line = &pblk->lines[i];
 589
 590                 pblk_line_free(line);
 591                 pblk_line_meta_free(l_mg, line);
 592         }
 593
 594         pblk_line_mg_free(pblk);
 595
 596         kfree(pblk->luns);
 597         kfree(pblk->lines);
 598 }
 599
 600 static int pblk_luns_init(struct pblk *pblk)
 601 {
 602         struct nvm_tgt_dev *dev = pblk->dev;
 603         struct nvm_geo *geo = &dev->geo;
 604         struct pblk_lun *rlun;
 605         int i;
 606
 607         /* TODO: Implement unbalanced LUN support */
 608         if (geo->num_lun < 0) {
 609                 pblk_err(pblk, "unbalanced LUN config.\n");
 610                 return -EINVAL;
 611         }
 612
 613         pblk->luns = kcalloc(geo->all_luns, sizeof(struct pblk_lun),
 614                                                                 GFP_KERNEL);
 615         if (!pblk->luns)
 616                 return -ENOMEM;
 617
 618         for (i = 0; i < geo->all_luns; i++) {
 619                 /* Stripe across channels */
 620                 int ch = i % geo->num_ch;
 621                 int lun_raw = i / geo->num_ch;
 622                 int lunid = lun_raw + ch * geo->num_lun;
 623
 624                 rlun = &pblk->luns[i];
 625                 rlun->bppa = dev->luns[lunid];
 626
 627                 sema_init(&rlun->wr_sem, 1);
 628         }
 629
 630         return 0;
 631 }
 632
 633 /* See comment over struct line_emeta definition */
 634 static unsigned int calc_emeta_len(struct pblk *pblk)
 635 {
 636         struct pblk_line_meta *lm = &pblk->lm;
 637         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
 638         struct nvm_tgt_dev *dev = pblk->dev;
 639         struct nvm_geo *geo = &dev->geo;
 640
 641         /* Round to sector size so that lba_list starts on its own sector */
 642         lm->emeta_sec[1] = DIV_ROUND_UP(
 643                         sizeof(struct line_emeta) + lm->blk_bitmap_len +
 644                         sizeof(struct wa_counters), geo->csecs);
 645         lm->emeta_len[1] = lm->emeta_sec[1] * geo->csecs;
 646
 647         /* Round to sector size so that vsc_list starts on its own sector */
 648         lm->dsec_per_line = lm->sec_per_line - lm->emeta_sec[0];
 649         lm->emeta_sec[2] = DIV_ROUND_UP(lm->dsec_per_line * sizeof(u64),
 650                         geo->csecs);
 651         lm->emeta_len[2] = lm->emeta_sec[2] * geo->csecs;
 652
 653         lm->emeta_sec[3] = DIV_ROUND_UP(l_mg->nr_lines * sizeof(u32),
 654                         geo->csecs);
 655         lm->emeta_len[3] = lm->emeta_sec[3] * geo->csecs;
 656
 657         lm->vsc_list_len = l_mg->nr_lines * sizeof(u32);
 658
 659         return (lm->emeta_len[1] + lm->emeta_len[2] + lm->emeta_len[3]);
 660 }
 661
 662 static int pblk_set_provision(struct pblk *pblk, int nr_free_chks)
 663 {
 664         struct nvm_tgt_dev *dev = pblk->dev;
 665         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
 666         struct pblk_line_meta *lm = &pblk->lm;
 667         struct nvm_geo *geo = &dev->geo;
 668         sector_t provisioned;
 669         int sec_meta, blk_meta, clba;
 670         int minimum;
 671
 672         if (geo->op == NVM_TARGET_DEFAULT_OP)
 673                 pblk->op = PBLK_DEFAULT_OP;
 674         else
 675                 pblk->op = geo->op;
 676
 677         minimum = pblk_get_min_chks(pblk);
 678         provisioned = nr_free_chks;
 679         provisioned *= (100 - pblk->op);
 680         sector_div(provisioned, 100);
 681
 682         if ((nr_free_chks - provisioned) < minimum) {
 683                 if (geo->op != NVM_TARGET_DEFAULT_OP) {
 684                         pblk_err(pblk, "OP too small to create a sane instance\n");
 685                         return -EINTR;
 686                 }
 687
 688                 /* If the user did not specify an OP value, and PBLK_DEFAULT_OP
 689                  * is not enough, calculate and set sane value
 690                  */
 691
 692                 provisioned = nr_free_chks - minimum;
 693                 pblk->op =  (100 * minimum) / nr_free_chks;
 694                 pblk_info(pblk, "Default OP insufficient, adjusting OP to %d\n",
 695                                 pblk->op);
 696         }
 697
 698         pblk->op_blks = nr_free_chks - provisioned;
 699
 700         /* Internally pblk manages all free blocks, but all calculations based
 701          * on user capacity consider only provisioned blocks
 702          */
 703         pblk->rl.total_blocks = nr_free_chks;
 704         pblk->rl.nr_secs = nr_free_chks * geo->clba;
 705
 706         /* Consider sectors used for metadata */
 707         sec_meta = (lm->smeta_sec + lm->emeta_sec[0]) * l_mg->nr_free_lines;
 708         blk_meta = DIV_ROUND_UP(sec_meta, geo->clba);
 709
 710         clba = (geo->clba / pblk->min_write_pgs) * pblk->min_write_pgs_data;
 711         pblk->capacity = (provisioned - blk_meta) * clba;
 712
 713         atomic_set(&pblk->rl.free_blocks, nr_free_chks);
 714         atomic_set(&pblk->rl.free_user_blocks, nr_free_chks);
 715
 716         return 0;
 717 }
 718
 719 static int pblk_setup_line_meta_chk(struct pblk *pblk, struct pblk_line *line,
 720                                    struct nvm_chk_meta *meta)
 721 {
 722         struct nvm_tgt_dev *dev = pblk->dev;
 723         struct nvm_geo *geo = &dev->geo;
 724         struct pblk_line_meta *lm = &pblk->lm;
 725         int i, nr_bad_chks = 0;
 726
 727         for (i = 0; i < lm->blk_per_line; i++) {
 728                 struct pblk_lun *rlun = &pblk->luns[i];
 729                 struct nvm_chk_meta *chunk;
 730                 struct nvm_chk_meta *chunk_meta;
 731                 struct ppa_addr ppa;
 732                 int pos;
 733
 734                 ppa = rlun->bppa;
 735                 pos = pblk_ppa_to_pos(geo, ppa);
 736                 chunk = &line->chks[pos];
 737
 738                 ppa.m.chk = line->id;
 739                 chunk_meta = pblk_chunk_get_off(pblk, meta, ppa);
 740
 741                 chunk->state = chunk_meta->state;
 742                 chunk->type = chunk_meta->type;
 743                 chunk->wi = chunk_meta->wi;
 744                 chunk->slba = chunk_meta->slba;
 745                 chunk->cnlb = chunk_meta->cnlb;
 746                 chunk->wp = chunk_meta->wp;
 747
 748                 trace_pblk_chunk_state(pblk_disk_name(pblk), &ppa,
 749                                         chunk->state);
 750
 751                 if (chunk->type & NVM_CHK_TP_SZ_SPEC) {
 752                         WARN_ONCE(1, "pblk: custom-sized chunks unsupported\n");
 753                         continue;
 754                 }
 755
 756                 if (!(chunk->state & NVM_CHK_ST_OFFLINE))
 757                         continue;
 758
 759                 set_bit(pos, line->blk_bitmap);
 760                 nr_bad_chks++;
 761         }
 762
 763         return nr_bad_chks;
 764 }
 765
 766 static long pblk_setup_line_meta(struct pblk *pblk, struct pblk_line *line,
 767                                  void *chunk_meta, int line_id)
 768 {
 769         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
 770         struct pblk_line_meta *lm = &pblk->lm;
 771         long nr_bad_chks, chk_in_line;
 772
 773         line->pblk = pblk;
 774         line->id = line_id;
 775         line->type = PBLK_LINETYPE_FREE;
 776         line->state = PBLK_LINESTATE_NEW;
 777         line->gc_group = PBLK_LINEGC_NONE;
 778         line->vsc = &l_mg->vsc_list[line_id];
 779         spin_lock_init(&line->lock);
 780
 781         nr_bad_chks = pblk_setup_line_meta_chk(pblk, line, chunk_meta);
 782
 783         chk_in_line = lm->blk_per_line - nr_bad_chks;
 784         if (nr_bad_chks < 0 || nr_bad_chks > lm->blk_per_line ||
 785                                         chk_in_line < lm->min_blk_line) {
 786                 line->state = PBLK_LINESTATE_BAD;
 787                 list_add_tail(&line->list, &l_mg->bad_list);
 788                 return 0;
 789         }
 790
 791         atomic_set(&line->blk_in_line, chk_in_line);
 792         list_add_tail(&line->list, &l_mg->free_list);
 793         l_mg->nr_free_lines++;
 794
 795         return chk_in_line;
 796 }
 797
 798 static int pblk_alloc_line_meta(struct pblk *pblk, struct pblk_line *line)
 799 {
 800         struct pblk_line_meta *lm = &pblk->lm;
 801
 802         line->blk_bitmap = kzalloc(lm->blk_bitmap_len, GFP_KERNEL);
 803         if (!line->blk_bitmap)
 804                 return -ENOMEM;
 805
 806         line->erase_bitmap = kzalloc(lm->blk_bitmap_len, GFP_KERNEL);
 807         if (!line->erase_bitmap)
 808                 goto free_blk_bitmap;
 809
 810
 811         line->chks = kmalloc_array(lm->blk_per_line,
 812                                    sizeof(struct nvm_chk_meta), GFP_KERNEL);
 813         if (!line->chks)
 814                 goto free_erase_bitmap;
 815
 816         line->w_err_gc = kzalloc(sizeof(struct pblk_w_err_gc), GFP_KERNEL);
 817         if (!line->w_err_gc)
 818                 goto free_chks;
 819
 820         return 0;
 821
 822 free_chks:
 823         kfree(line->chks);
 824 free_erase_bitmap:
 825         kfree(line->erase_bitmap);
 826 free_blk_bitmap:
 827         kfree(line->blk_bitmap);
 828         return -ENOMEM;
 829 }
 830
 831 static int pblk_line_mg_init(struct pblk *pblk)
 832 {
 833         struct nvm_tgt_dev *dev = pblk->dev;
 834         struct nvm_geo *geo = &dev->geo;
 835         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
 836         struct pblk_line_meta *lm = &pblk->lm;
 837         int i, bb_distance;
 838
 839         l_mg->nr_lines = geo->num_chk;
 840         l_mg->log_line = l_mg->data_line = NULL;
 841         l_mg->l_seq_nr = l_mg->d_seq_nr = 0;
 842         l_mg->nr_free_lines = 0;
 843         bitmap_zero(&l_mg->meta_bitmap, PBLK_DATA_LINES);
 844
 845         INIT_LIST_HEAD(&l_mg->free_list);
 846         INIT_LIST_HEAD(&l_mg->corrupt_list);
 847         INIT_LIST_HEAD(&l_mg->bad_list);
 848         INIT_LIST_HEAD(&l_mg->gc_full_list);
 849         INIT_LIST_HEAD(&l_mg->gc_high_list);
 850         INIT_LIST_HEAD(&l_mg->gc_mid_list);
 851         INIT_LIST_HEAD(&l_mg->gc_low_list);
 852         INIT_LIST_HEAD(&l_mg->gc_empty_list);
 853         INIT_LIST_HEAD(&l_mg->gc_werr_list);
 854
 855         INIT_LIST_HEAD(&l_mg->emeta_list);
 856
 857         l_mg->gc_lists[0] = &l_mg->gc_werr_list;
 858         l_mg->gc_lists[1] = &l_mg->gc_high_list;
 859         l_mg->gc_lists[2] = &l_mg->gc_mid_list;
 860         l_mg->gc_lists[3] = &l_mg->gc_low_list;
 861
 862         spin_lock_init(&l_mg->free_lock);
 863         spin_lock_init(&l_mg->close_lock);
 864         spin_lock_init(&l_mg->gc_lock);
 865
 866         l_mg->vsc_list = kcalloc(l_mg->nr_lines, sizeof(__le32), GFP_KERNEL);
 867         if (!l_mg->vsc_list)
 868                 goto fail;
 869
 870         l_mg->bb_template = kzalloc(lm->sec_bitmap_len, GFP_KERNEL);
 871         if (!l_mg->bb_template)
 872                 goto fail_free_vsc_list;
 873
 874         l_mg->bb_aux = kzalloc(lm->sec_bitmap_len, GFP_KERNEL);
 875         if (!l_mg->bb_aux)
 876                 goto fail_free_bb_template;
 877
 878         /* smeta is always small enough to fit on a kmalloc memory allocation,
 879          * emeta depends on the number of LUNs allocated to the pblk instance
 880          */
 881         for (i = 0; i < PBLK_DATA_LINES; i++) {
 882                 l_mg->sline_meta[i] = kmalloc(lm->smeta_len, GFP_KERNEL);
 883                 if (!l_mg->sline_meta[i])
 884                         goto fail_free_smeta;
 885         }
 886
 887         l_mg->bitmap_cache = kmem_cache_create("pblk_lm_bitmap",
 888                         lm->sec_bitmap_len, 0, 0, NULL);
 889         if (!l_mg->bitmap_cache)
 890                 goto fail_free_smeta;
 891
 892         /* the bitmap pool is used for both valid and map bitmaps */
 893         l_mg->bitmap_pool = mempool_create_slab_pool(PBLK_DATA_LINES * 2,
 894                                 l_mg->bitmap_cache);
 895         if (!l_mg->bitmap_pool)
 896                 goto fail_destroy_bitmap_cache;
 897
 898         /* emeta allocates three different buffers for managing metadata with
 899          * in-memory and in-media layouts
 900          */
 901         for (i = 0; i < PBLK_DATA_LINES; i++) {
 902                 struct pblk_emeta *emeta;
 903
 904                 emeta = kmalloc(sizeof(struct pblk_emeta), GFP_KERNEL);
 905                 if (!emeta)
 906                         goto fail_free_emeta;
 907
 908                 if (lm->emeta_len[0] > KMALLOC_MAX_CACHE_SIZE) {
 909                         l_mg->emeta_alloc_type = PBLK_VMALLOC_META;
 910
 911                         emeta->buf = vmalloc(lm->emeta_len[0]);
 912                         if (!emeta->buf) {
 913                                 kfree(emeta);
 914                                 goto fail_free_emeta;
 915                         }
 916
 917                         emeta->nr_entries = lm->emeta_sec[0];
 918                         l_mg->eline_meta[i] = emeta;
 919                 } else {
 920                         l_mg->emeta_alloc_type = PBLK_KMALLOC_META;
 921
 922                         emeta->buf = kmalloc(lm->emeta_len[0], GFP_KERNEL);
 923                         if (!emeta->buf) {
 924                                 kfree(emeta);
 925                                 goto fail_free_emeta;
 926                         }
 927
 928                         emeta->nr_entries = lm->emeta_sec[0];
 929                         l_mg->eline_meta[i] = emeta;
 930                 }
 931         }
 932
 933         for (i = 0; i < l_mg->nr_lines; i++)
 934                 l_mg->vsc_list[i] = cpu_to_le32(EMPTY_ENTRY);
 935
 936         bb_distance = (geo->all_luns) * geo->ws_opt;
 937         for (i = 0; i < lm->sec_per_line; i += bb_distance)
 938                 bitmap_set(l_mg->bb_template, i, geo->ws_opt);
 939
 940         return 0;
 941
 942 fail_free_emeta:
 943         while (--i >= 0) {
 944                 if (l_mg->emeta_alloc_type == PBLK_VMALLOC_META)
 945                         vfree(l_mg->eline_meta[i]->buf);
 946                 else
 947                         kfree(l_mg->eline_meta[i]->buf);
 948                 kfree(l_mg->eline_meta[i]);
 949         }
 950
 951         mempool_destroy(l_mg->bitmap_pool);
 952 fail_destroy_bitmap_cache:
 953         kmem_cache_destroy(l_mg->bitmap_cache);
 954 fail_free_smeta:
 955         for (i = 0; i < PBLK_DATA_LINES; i++)
 956                 kfree(l_mg->sline_meta[i]);
 957         kfree(l_mg->bb_aux);
 958 fail_free_bb_template:
 959         kfree(l_mg->bb_template);
 960 fail_free_vsc_list:
 961         kfree(l_mg->vsc_list);
 962 fail:
 963         return -ENOMEM;
 964 }
 965
 966 static int pblk_line_meta_init(struct pblk *pblk)
 967 {
 968         struct nvm_tgt_dev *dev = pblk->dev;
 969         struct nvm_geo *geo = &dev->geo;
 970         struct pblk_line_meta *lm = &pblk->lm;
 971         unsigned int smeta_len, emeta_len;
 972         int i;
 973
 974         lm->sec_per_line = geo->clba * geo->all_luns;
 975         lm->blk_per_line = geo->all_luns;
 976         lm->blk_bitmap_len = BITS_TO_LONGS(geo->all_luns) * sizeof(long);
 977         lm->sec_bitmap_len = BITS_TO_LONGS(lm->sec_per_line) * sizeof(long);
 978         lm->lun_bitmap_len = BITS_TO_LONGS(geo->all_luns) * sizeof(long);
 979         lm->mid_thrs = lm->sec_per_line / 2;
 980         lm->high_thrs = lm->sec_per_line / 4;
 981         lm->meta_distance = (geo->all_luns / 2) * pblk->min_write_pgs;
 982
 983         /* Calculate necessary pages for smeta. See comment over struct
 984          * line_smeta definition
 985          */
 986         i = 1;
 987 add_smeta_page:
 988         lm->smeta_sec = i * geo->ws_opt;
 989         lm->smeta_len = lm->smeta_sec * geo->csecs;
 990
 991         smeta_len = sizeof(struct line_smeta) + lm->lun_bitmap_len;
 992         if (smeta_len > lm->smeta_len) {
 993                 i++;
 994                 goto add_smeta_page;
 995         }
 996
 997         /* Calculate necessary pages for emeta. See comment over struct
 998          * line_emeta definition
 999          */
1000         i = 1;
1001 add_emeta_page:
1002         lm->emeta_sec[0] = i * geo->ws_opt;
1003         lm->emeta_len[0] = lm->emeta_sec[0] * geo->csecs;
1004
1005         emeta_len = calc_emeta_len(pblk);
1006         if (emeta_len > lm->emeta_len[0]) {
1007                 i++;
1008                 goto add_emeta_page;
1009         }
1010
1011         lm->emeta_bb = geo->all_luns > i ? geo->all_luns - i : 0;
1012
1013         lm->min_blk_line = 1;
1014         if (geo->all_luns > 1)
1015                 lm->min_blk_line += DIV_ROUND_UP(lm->smeta_sec +
1016                                         lm->emeta_sec[0], geo->clba);
1017
1018         if (lm->min_blk_line > lm->blk_per_line) {
1019                 pblk_err(pblk, "config. not supported. Min. LUN in line:%d\n",
1020                                                         lm->blk_per_line);
1021                 return -EINVAL;
1022         }
1023
1024         return 0;
1025 }
1026
1027 static int pblk_lines_init(struct pblk *pblk)
1028 {
1029         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1030         struct pblk_line *line;
1031         void *chunk_meta;
1032         int nr_free_chks = 0;
1033         int i, ret;
1034
1035         ret = pblk_line_meta_init(pblk);
1036         if (ret)
1037                 return ret;
1038
1039         ret = pblk_line_mg_init(pblk);
1040         if (ret)
1041                 return ret;
1042
1043         ret = pblk_luns_init(pblk);
1044         if (ret)
1045                 goto fail_free_meta;
1046
1047         chunk_meta = pblk_get_chunk_meta(pblk);
1048         if (IS_ERR(chunk_meta)) {
1049                 ret = PTR_ERR(chunk_meta);
1050                 goto fail_free_luns;
1051         }
1052
1053         pblk->lines = kcalloc(l_mg->nr_lines, sizeof(struct pblk_line),
1054                                                                 GFP_KERNEL);
1055         if (!pblk->lines) {
1056                 ret = -ENOMEM;
1057                 goto fail_free_chunk_meta;
1058         }
1059
1060         for (i = 0; i < l_mg->nr_lines; i++) {
1061                 line = &pblk->lines[i];
1062
1063                 ret = pblk_alloc_line_meta(pblk, line);
1064                 if (ret)
1065                         goto fail_free_lines;
1066
1067                 nr_free_chks += pblk_setup_line_meta(pblk, line, chunk_meta, i);
1068
1069                 trace_pblk_line_state(pblk_disk_name(pblk), line->id,
1070                                                                 line->state);
1071         }
1072
1073         if (!nr_free_chks) {
1074                 pblk_err(pblk, "too many bad blocks prevent for sane instance\n");
1075                 ret = -EINTR;
1076                 goto fail_free_lines;
1077         }
1078
1079         ret = pblk_set_provision(pblk, nr_free_chks);
1080         if (ret)
1081                 goto fail_free_lines;
1082
1083         vfree(chunk_meta);
1084         return 0;
1085
1086 fail_free_lines:
1087         while (--i >= 0)
1088                 pblk_line_meta_free(l_mg, &pblk->lines[i]);
1089         kfree(pblk->lines);
1090 fail_free_chunk_meta:
1091         vfree(chunk_meta);
1092 fail_free_luns:
1093         kfree(pblk->luns);
1094 fail_free_meta:
1095         pblk_line_mg_free(pblk);
1096
1097         return ret;
1098 }
1099
1100 static int pblk_writer_init(struct pblk *pblk)
1101 {
1102         pblk->writer_ts = kthread_create(pblk_write_ts, pblk, "pblk-writer-t");
1103         if (IS_ERR(pblk->writer_ts)) {
1104                 int err = PTR_ERR(pblk->writer_ts);
1105
1106                 if (err != -EINTR)
1107                         pblk_err(pblk, "could not allocate writer kthread (%d)\n",
1108                                         err);
1109                 return err;
1110         }
1111
1112         timer_setup(&pblk->wtimer, pblk_write_timer_fn, 0);
1113         mod_timer(&pblk->wtimer, jiffies + msecs_to_jiffies(100));
1114
1115         return 0;
1116 }
1117
1118 static void pblk_writer_stop(struct pblk *pblk)
1119 {
1120         /* The pipeline must be stopped and the write buffer emptied before the
1121          * write thread is stopped
1122          */
1123         WARN(pblk_rb_read_count(&pblk->rwb),
1124                         "Stopping not fully persisted write buffer\n");
1125
1126         WARN(pblk_rb_sync_count(&pblk->rwb),
1127                         "Stopping not fully synced write buffer\n");
1128
1129         del_timer_sync(&pblk->wtimer);
1130         if (pblk->writer_ts)
1131                 kthread_stop(pblk->writer_ts);
1132 }
1133
1134 static void pblk_free(struct pblk *pblk)
1135 {
1136         pblk_lines_free(pblk);
1137         pblk_l2p_free(pblk);
1138         pblk_rwb_free(pblk);
1139         pblk_core_free(pblk);
1140
1141         kfree(pblk);
1142 }
1143
1144 static void pblk_tear_down(struct pblk *pblk, bool graceful)
1145 {
1146         if (graceful)
1147                 __pblk_pipeline_flush(pblk);
1148         __pblk_pipeline_stop(pblk);
1149         pblk_writer_stop(pblk);
1150         pblk_rb_sync_l2p(&pblk->rwb);
1151         pblk_rl_free(&pblk->rl);
1152
1153         pblk_debug(pblk, "consistent tear down (graceful:%d)\n", graceful);
1154 }
1155
1156 static void pblk_exit(void *private, bool graceful)
1157 {
1158         struct pblk *pblk = private;
1159
1160         pblk_gc_exit(pblk, graceful);
1161         pblk_tear_down(pblk, graceful);
1162
1163 #ifdef CONFIG_NVM_PBLK_DEBUG
1164         pblk_info(pblk, "exit: L2P CRC: %x\n", pblk_l2p_crc(pblk));
1165 #endif
1166
1167         pblk_free(pblk);
1168 }
1169
1170 static sector_t pblk_capacity(void *private)
1171 {
1172         struct pblk *pblk = private;
1173
1174         return pblk->capacity * NR_PHY_IN_LOG;
1175 }
1176
1177 static void *pblk_init(struct nvm_tgt_dev *dev, struct gendisk *tdisk,
1178                        int flags)
1179 {
1180         struct nvm_geo *geo = &dev->geo;
1181         struct request_queue *bqueue = dev->q;
1182         struct request_queue *tqueue = tdisk->queue;
1183         struct pblk *pblk;
1184         int ret;
1185
1186         pblk = kzalloc(sizeof(struct pblk), GFP_KERNEL);
1187         if (!pblk)
1188                 return ERR_PTR(-ENOMEM);
1189
1190         pblk->dev = dev;
1191         pblk->disk = tdisk;
1192         pblk->state = PBLK_STATE_RUNNING;
1193         trace_pblk_state(pblk_disk_name(pblk), pblk->state);
1194         pblk->gc.gc_enabled = 0;
1195
1196         if (!(geo->version == NVM_OCSSD_SPEC_12 ||
1197                                         geo->version == NVM_OCSSD_SPEC_20)) {
1198                 pblk_err(pblk, "OCSSD version not supported (%u)\n",
1199                                                         geo->version);
1200                 kfree(pblk);
1201                 return ERR_PTR(-EINVAL);
1202         }
1203
1204         if (geo->ext) {
1205                 pblk_err(pblk, "extended metadata not supported\n");
1206                 kfree(pblk);
1207                 return ERR_PTR(-EINVAL);
1208         }
1209
1210         spin_lock_init(&pblk->resubmit_lock);
1211         spin_lock_init(&pblk->trans_lock);
1212         spin_lock_init(&pblk->lock);
1213
1214 #ifdef CONFIG_NVM_PBLK_DEBUG
1215         atomic_long_set(&pblk->inflight_writes, 0);
1216         atomic_long_set(&pblk->padded_writes, 0);
1217         atomic_long_set(&pblk->padded_wb, 0);
1218         atomic_long_set(&pblk->req_writes, 0);
1219         atomic_long_set(&pblk->sub_writes, 0);
1220         atomic_long_set(&pblk->sync_writes, 0);
1221         atomic_long_set(&pblk->inflight_reads, 0);
1222         atomic_long_set(&pblk->cache_reads, 0);
1223         atomic_long_set(&pblk->sync_reads, 0);
1224         atomic_long_set(&pblk->recov_writes, 0);
1225         atomic_long_set(&pblk->recov_writes, 0);
1226         atomic_long_set(&pblk->recov_gc_writes, 0);
1227         atomic_long_set(&pblk->recov_gc_reads, 0);
1228 #endif
1229
1230         atomic_long_set(&pblk->read_failed, 0);
1231         atomic_long_set(&pblk->read_empty, 0);
1232         atomic_long_set(&pblk->read_high_ecc, 0);
1233         atomic_long_set(&pblk->read_failed_gc, 0);
1234         atomic_long_set(&pblk->write_failed, 0);
1235         atomic_long_set(&pblk->erase_failed, 0);
1236
1237         ret = pblk_core_init(pblk);
1238         if (ret) {
1239                 pblk_err(pblk, "could not initialize core\n");
1240                 goto fail;
1241         }
1242
1243         ret = pblk_lines_init(pblk);
1244         if (ret) {
1245                 pblk_err(pblk, "could not initialize lines\n");
1246                 goto fail_free_core;
1247         }
1248
1249         ret = pblk_rwb_init(pblk);
1250         if (ret) {
1251                 pblk_err(pblk, "could not initialize write buffer\n");
1252                 goto fail_free_lines;
1253         }
1254
1255         ret = pblk_l2p_init(pblk, flags & NVM_TARGET_FACTORY);
1256         if (ret) {
1257                 pblk_err(pblk, "could not initialize maps\n");
1258                 goto fail_free_rwb;
1259         }
1260
1261         ret = pblk_writer_init(pblk);
1262         if (ret) {
1263                 if (ret != -EINTR)
1264                         pblk_err(pblk, "could not initialize write thread\n");
1265                 goto fail_free_l2p;
1266         }
1267
1268         ret = pblk_gc_init(pblk);
1269         if (ret) {
1270                 pblk_err(pblk, "could not initialize gc\n");
1271                 goto fail_stop_writer;
1272         }
1273
1274         /* inherit the size from the underlying device */
1275         blk_queue_logical_block_size(tqueue, queue_physical_block_size(bqueue));
1276         blk_queue_max_hw_sectors(tqueue, queue_max_hw_sectors(bqueue));
1277
1278         blk_queue_write_cache(tqueue, true, false);
1279
1280         tqueue->limits.discard_granularity = geo->clba * geo->csecs;
1281         tqueue->limits.discard_alignment = 0;
1282         blk_queue_max_discard_sectors(tqueue, UINT_MAX >> 9);
1283         blk_queue_flag_set(QUEUE_FLAG_DISCARD, tqueue);
1284
1285         pblk_info(pblk, "luns:%u, lines:%d, secs:%llu, buf entries:%u\n",
1286                         geo->all_luns, pblk->l_mg.nr_lines,
1287                         (unsigned long long)pblk->rl.nr_secs,
1288                         pblk->rwb.nr_entries);
1289
1290         wake_up_process(pblk->writer_ts);
1291
1292         /* Check if we need to start GC */
1293         pblk_gc_should_kick(pblk);
1294
1295         return pblk;
1296
1297 fail_stop_writer:
1298         pblk_writer_stop(pblk);
1299 fail_free_l2p:
1300         pblk_l2p_free(pblk);
1301 fail_free_rwb:
1302         pblk_rwb_free(pblk);
1303 fail_free_lines:
1304         pblk_lines_free(pblk);
1305 fail_free_core:
1306         pblk_core_free(pblk);
1307 fail:
1308         kfree(pblk);
1309         return ERR_PTR(ret);
1310 }
1311
1312 /* physical block device target */
1313 static struct nvm_tgt_type tt_pblk = {
1314         .name           = "pblk",
1315         .version        = {1, 0, 0},
1316
1317         .make_rq        = pblk_make_rq,
1318         .capacity       = pblk_capacity,
1319
1320         .init           = pblk_init,
1321         .exit           = pblk_exit,
1322
1323         .sysfs_init     = pblk_sysfs_init,
1324         .sysfs_exit     = pblk_sysfs_exit,
1325         .owner          = THIS_MODULE,
1326 };
1327
1328 static int __init pblk_module_init(void)
1329 {
1330         int ret;
1331
1332         ret = bioset_init(&pblk_bio_set, BIO_POOL_SIZE, 0, 0);
1333         if (ret)
1334                 return ret;
1335         ret = nvm_register_tgt_type(&tt_pblk);
1336         if (ret)
1337                 bioset_exit(&pblk_bio_set);
1338         return ret;
1339 }
1340
1341 static void pblk_module_exit(void)
1342 {
1343         bioset_exit(&pblk_bio_set);
1344         nvm_unregister_tgt_type(&tt_pblk);
1345 }
1346
1347 module_init(pblk_module_init);
1348 module_exit(pblk_module_exit);
1349 MODULE_AUTHOR("Javier Gonzalez <javier@cnexlabs.com>");
1350 MODULE_AUTHOR("Matias Bjorling <matias@cnexlabs.com>");
1351 MODULE_LICENSE("GPL v2");
1352 MODULE_DESCRIPTION("Physical Block-Device for Open-Channel SSDs");