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 <[email protected]>
   6  *                  Matias Bjorling <[email protected]>
   7  *
   8  * This program is free software; you can redistribute it and/or
   9  * modify it under the terms of the GNU General Public License version
  10  * 2 as published by the Free Software Foundation.
  11  *
  12  * This program is distributed in the hope that it will be useful, but
  13  * WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * General Public License for more details.
  16  *
  17  * Implementation of a physical block-device target for Open-channel SSDs.
  18  *
  19  * pblk-init.c - pblk's initialization.
  20  */
  21
  22 #include "pblk.h"
  23 #include "pblk-trace.h"
  24
  25 static unsigned int write_buffer_size;
  26
  27 module_param(write_buffer_size, uint, 0644);
  28 MODULE_PARM_DESC(write_buffer_size, "number of entries in a write buffer");
  29
  30 struct pblk_global_caches {
  31         struct kmem_cache       *ws;
  32         struct kmem_cache       *rec;
  33         struct kmem_cache       *g_rq;
  34         struct kmem_cache       *w_rq;
  35
  36         struct kref             kref;
  37
  38         struct mutex            mutex; /* Ensures consistency between
  39                                         * caches and kref
  40                                         */
  41 };
  42
  43 static struct pblk_global_caches pblk_caches = {
  44         .mutex = __MUTEX_INITIALIZER(pblk_caches.mutex),
  45         .kref = KREF_INIT(0),
  46 };
  47
  48 struct bio_set pblk_bio_set;
  49
  50 static int pblk_rw_io(struct request_queue *q, struct pblk *pblk,
  51                           struct bio *bio)
  52 {
  53         int ret;
  54
  55         /* Read requests must be <= 256kb due to NVMe's 64 bit completion bitmap
  56          * constraint. Writes can be of arbitrary size.
  57          */
  58         if (bio_data_dir(bio) == READ) {
  59                 blk_queue_split(q, &bio);
  60                 ret = pblk_submit_read(pblk, bio);
  61                 if (ret == NVM_IO_DONE && bio_flagged(bio, BIO_CLONED))
  62                         bio_put(bio);
  63
  64                 return ret;
  65         }
  66
  67         /* Prevent deadlock in the case of a modest LUN configuration and large
  68          * user I/Os. Unless stalled, the rate limiter leaves at least 256KB
  69          * available for user I/O.
  70          */
  71         if (pblk_get_secs(bio) > pblk_rl_max_io(&pblk->rl))
  72                 blk_queue_split(q, &bio);
  73
  74         return pblk_write_to_cache(pblk, bio, PBLK_IOTYPE_USER);
  75 }
  76
  77 static blk_qc_t pblk_make_rq(struct request_queue *q, struct bio *bio)
  78 {
  79         struct pblk *pblk = q->queuedata;
  80
  81         if (bio_op(bio) == REQ_OP_DISCARD) {
  82                 pblk_discard(pblk, bio);
  83                 if (!(bio->bi_opf & REQ_PREFLUSH)) {
  84                         bio_endio(bio);
  85                         return BLK_QC_T_NONE;
  86                 }
  87         }
  88
  89         switch (pblk_rw_io(q, pblk, bio)) {
  90         case NVM_IO_ERR:
  91                 bio_io_error(bio);
  92                 break;
  93         case NVM_IO_DONE:
  94                 bio_endio(bio);
  95                 break;
  96         }
  97
  98         return BLK_QC_T_NONE;
  99 }
 100
 101 static size_t pblk_trans_map_size(struct pblk *pblk)
 102 {
 103         int entry_size = 8;
 104
 105         if (pblk->addrf_len < 32)
 106                 entry_size = 4;
 107
 108         return entry_size * pblk->rl.nr_secs;
 109 }
 110
 111 #ifdef CONFIG_NVM_PBLK_DEBUG
 112 static u32 pblk_l2p_crc(struct pblk *pblk)
 113 {
 114         size_t map_size;
 115         u32 crc = ~(u32)0;
 116
 117         map_size = pblk_trans_map_size(pblk);
 118         crc = crc32_le(crc, pblk->trans_map, map_size);
 119         return crc;
 120 }
 121 #endif
 122
 123 static void pblk_l2p_free(struct pblk *pblk)
 124 {
 125         vfree(pblk->trans_map);
 126 }
 127
 128 static int pblk_l2p_recover(struct pblk *pblk, bool factory_init)
 129 {
 130         struct pblk_line *line = NULL;
 131
 132         if (factory_init) {
 133                 pblk_setup_uuid(pblk);
 134         } else {
 135                 line = pblk_recov_l2p(pblk);
 136                 if (IS_ERR(line)) {
 137                         pblk_err(pblk, "could not recover l2p table\n");
 138                         return -EFAULT;
 139                 }
 140         }
 141
 142 #ifdef CONFIG_NVM_PBLK_DEBUG
 143         pblk_info(pblk, "init: L2P CRC: %x\n", pblk_l2p_crc(pblk));
 144 #endif
 145
 146         /* Free full lines directly as GC has not been started yet */
 147         pblk_gc_free_full_lines(pblk);
 148
 149         if (!line) {
 150                 /* Configure next line for user data */
 151                 line = pblk_line_get_first_data(pblk);
 152                 if (!line)
 153                         return -EFAULT;
 154         }
 155
 156         return 0;
 157 }
 158
 159 static int pblk_l2p_init(struct pblk *pblk, bool factory_init)
 160 {
 161         sector_t i;
 162         struct ppa_addr ppa;
 163         size_t map_size;
 164         int ret = 0;
 165
 166         map_size = pblk_trans_map_size(pblk);
 167         pblk->trans_map = vmalloc(map_size);
 168         if (!pblk->trans_map)
 169                 return -ENOMEM;
 170
 171         pblk_ppa_set_empty(&ppa);
 172
 173         for (i = 0; i < pblk->rl.nr_secs; i++)
 174                 pblk_trans_map_set(pblk, i, ppa);
 175
 176         ret = pblk_l2p_recover(pblk, factory_init);
 177         if (ret)
 178                 vfree(pblk->trans_map);
 179
 180         return ret;
 181 }
 182
 183 static void pblk_rwb_free(struct pblk *pblk)
 184 {
 185         if (pblk_rb_tear_down_check(&pblk->rwb))
 186                 pblk_err(pblk, "write buffer error on tear down\n");
 187
 188         pblk_rb_free(&pblk->rwb);
 189 }
 190
 191 static int pblk_rwb_init(struct pblk *pblk)
 192 {
 193         struct nvm_tgt_dev *dev = pblk->dev;
 194         struct nvm_geo *geo = &dev->geo;
 195         unsigned long buffer_size;
 196         int pgs_in_buffer, threshold;
 197
 198         threshold = geo->mw_cunits * geo->all_luns;
 199         pgs_in_buffer = (max(geo->mw_cunits, geo->ws_opt) + geo->ws_opt)
 200                                                                 * geo->all_luns;
 201
 202         if (write_buffer_size && (write_buffer_size > pgs_in_buffer))
 203                 buffer_size = write_buffer_size;
 204         else
 205                 buffer_size = pgs_in_buffer;
 206
 207         return pblk_rb_init(&pblk->rwb, buffer_size, threshold, geo->csecs);
 208 }
 209
 210 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         spin_lock(&l_mg->free_lock);
 588         for (i = 0; i < l_mg->nr_lines; i++) {
 589                 line = &pblk->lines[i];
 590
 591                 pblk_line_free(line);
 592                 pblk_line_meta_free(l_mg, line);
 593         }
 594         spin_unlock(&l_mg->free_lock);
 595
 596         pblk_line_mg_free(pblk);
 597
 598         kfree(pblk->luns);
 599         kfree(pblk->lines);
 600 }
 601
 602 static int pblk_luns_init(struct pblk *pblk)
 603 {
 604         struct nvm_tgt_dev *dev = pblk->dev;
 605         struct nvm_geo *geo = &dev->geo;
 606         struct pblk_lun *rlun;
 607         int i;
 608
 609         /* TODO: Implement unbalanced LUN support */
 610         if (geo->num_lun < 0) {
 611                 pblk_err(pblk, "unbalanced LUN config.\n");
 612                 return -EINVAL;
 613         }
 614
 615         pblk->luns = kcalloc(geo->all_luns, sizeof(struct pblk_lun),
 616                                                                 GFP_KERNEL);
 617         if (!pblk->luns)
 618                 return -ENOMEM;
 619
 620         for (i = 0; i < geo->all_luns; i++) {
 621                 /* Stripe across channels */
 622                 int ch = i % geo->num_ch;
 623                 int lun_raw = i / geo->num_ch;
 624                 int lunid = lun_raw + ch * geo->num_lun;
 625
 626                 rlun = &pblk->luns[i];
 627                 rlun->bppa = dev->luns[lunid];
 628
 629                 sema_init(&rlun->wr_sem, 1);
 630         }
 631
 632         return 0;
 633 }
 634
 635 /* See comment over struct line_emeta definition */
 636 static unsigned int calc_emeta_len(struct pblk *pblk)
 637 {
 638         struct pblk_line_meta *lm = &pblk->lm;
 639         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
 640         struct nvm_tgt_dev *dev = pblk->dev;
 641         struct nvm_geo *geo = &dev->geo;
 642
 643         /* Round to sector size so that lba_list starts on its own sector */
 644         lm->emeta_sec[1] = DIV_ROUND_UP(
 645                         sizeof(struct line_emeta) + lm->blk_bitmap_len +
 646                         sizeof(struct wa_counters), geo->csecs);
 647         lm->emeta_len[1] = lm->emeta_sec[1] * geo->csecs;
 648
 649         /* Round to sector size so that vsc_list starts on its own sector */
 650         lm->dsec_per_line = lm->sec_per_line - lm->emeta_sec[0];
 651         lm->emeta_sec[2] = DIV_ROUND_UP(lm->dsec_per_line * sizeof(u64),
 652                         geo->csecs);
 653         lm->emeta_len[2] = lm->emeta_sec[2] * geo->csecs;
 654
 655         lm->emeta_sec[3] = DIV_ROUND_UP(l_mg->nr_lines * sizeof(u32),
 656                         geo->csecs);
 657         lm->emeta_len[3] = lm->emeta_sec[3] * geo->csecs;
 658
 659         lm->vsc_list_len = l_mg->nr_lines * sizeof(u32);
 660
 661         return (lm->emeta_len[1] + lm->emeta_len[2] + lm->emeta_len[3]);
 662 }
 663
 664 static int pblk_set_provision(struct pblk *pblk, int nr_free_chks)
 665 {
 666         struct nvm_tgt_dev *dev = pblk->dev;
 667         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
 668         struct pblk_line_meta *lm = &pblk->lm;
 669         struct nvm_geo *geo = &dev->geo;
 670         sector_t provisioned;
 671         int sec_meta, blk_meta, clba;
 672         int minimum;
 673
 674         if (geo->op == NVM_TARGET_DEFAULT_OP)
 675                 pblk->op = PBLK_DEFAULT_OP;
 676         else
 677                 pblk->op = geo->op;
 678
 679         minimum = pblk_get_min_chks(pblk);
 680         provisioned = nr_free_chks;
 681         provisioned *= (100 - pblk->op);
 682         sector_div(provisioned, 100);
 683
 684         if ((nr_free_chks - provisioned) < minimum) {
 685                 if (geo->op != NVM_TARGET_DEFAULT_OP) {
 686                         pblk_err(pblk, "OP too small to create a sane instance\n");
 687                         return -EINTR;
 688                 }
 689
 690                 /* If the user did not specify an OP value, and PBLK_DEFAULT_OP
 691                  * is not enough, calculate and set sane value
 692                  */
 693
 694                 provisioned = nr_free_chks - minimum;
 695                 pblk->op =  (100 * minimum) / nr_free_chks;
 696                 pblk_info(pblk, "Default OP insufficient, adjusting OP to %d\n",
 697                                 pblk->op);
 698         }
 699
 700         pblk->op_blks = nr_free_chks - provisioned;
 701
 702         /* Internally pblk manages all free blocks, but all calculations based
 703          * on user capacity consider only provisioned blocks
 704          */
 705         pblk->rl.total_blocks = nr_free_chks;
 706         pblk->rl.nr_secs = nr_free_chks * geo->clba;
 707
 708         /* Consider sectors used for metadata */
 709         sec_meta = (lm->smeta_sec + lm->emeta_sec[0]) * l_mg->nr_free_lines;
 710         blk_meta = DIV_ROUND_UP(sec_meta, geo->clba);
 711
 712         clba = (geo->clba / pblk->min_write_pgs) * pblk->min_write_pgs_data;
 713         pblk->capacity = (provisioned - blk_meta) * clba;
 714
 715         atomic_set(&pblk->rl.free_blocks, nr_free_chks);
 716         atomic_set(&pblk->rl.free_user_blocks, nr_free_chks);
 717
 718         return 0;
 719 }
 720
 721 static int pblk_setup_line_meta_chk(struct pblk *pblk, struct pblk_line *line,
 722                                    struct nvm_chk_meta *meta)
 723 {
 724         struct nvm_tgt_dev *dev = pblk->dev;
 725         struct nvm_geo *geo = &dev->geo;
 726         struct pblk_line_meta *lm = &pblk->lm;
 727         int i, nr_bad_chks = 0;
 728
 729         for (i = 0; i < lm->blk_per_line; i++) {
 730                 struct pblk_lun *rlun = &pblk->luns[i];
 731                 struct nvm_chk_meta *chunk;
 732                 struct nvm_chk_meta *chunk_meta;
 733                 struct ppa_addr ppa;
 734                 int pos;
 735
 736                 ppa = rlun->bppa;
 737                 pos = pblk_ppa_to_pos(geo, ppa);
 738                 chunk = &line->chks[pos];
 739
 740                 ppa.m.chk = line->id;
 741                 chunk_meta = pblk_chunk_get_off(pblk, meta, ppa);
 742
 743                 chunk->state = chunk_meta->state;
 744                 chunk->type = chunk_meta->type;
 745                 chunk->wi = chunk_meta->wi;
 746                 chunk->slba = chunk_meta->slba;
 747                 chunk->cnlb = chunk_meta->cnlb;
 748                 chunk->wp = chunk_meta->wp;
 749
 750                 trace_pblk_chunk_state(pblk_disk_name(pblk), &ppa,
 751                                         chunk->state);
 752
 753                 if (chunk->type & NVM_CHK_TP_SZ_SPEC) {
 754                         WARN_ONCE(1, "pblk: custom-sized chunks unsupported\n");
 755                         continue;
 756                 }
 757
 758                 if (!(chunk->state & NVM_CHK_ST_OFFLINE))
 759                         continue;
 760
 761                 set_bit(pos, line->blk_bitmap);
 762                 nr_bad_chks++;
 763         }
 764
 765         return nr_bad_chks;
 766 }
 767
 768 static long pblk_setup_line_meta(struct pblk *pblk, struct pblk_line *line,
 769                                  void *chunk_meta, int line_id)
 770 {
 771         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
 772         struct pblk_line_meta *lm = &pblk->lm;
 773         long nr_bad_chks, chk_in_line;
 774
 775         line->pblk = pblk;
 776         line->id = line_id;
 777         line->type = PBLK_LINETYPE_FREE;
 778         line->state = PBLK_LINESTATE_NEW;
 779         line->gc_group = PBLK_LINEGC_NONE;
 780         line->vsc = &l_mg->vsc_list[line_id];
 781         spin_lock_init(&line->lock);
 782
 783         nr_bad_chks = pblk_setup_line_meta_chk(pblk, line, chunk_meta);
 784
 785         chk_in_line = lm->blk_per_line - nr_bad_chks;
 786         if (nr_bad_chks < 0 || nr_bad_chks > lm->blk_per_line ||
 787                                         chk_in_line < lm->min_blk_line) {
 788                 line->state = PBLK_LINESTATE_BAD;
 789                 list_add_tail(&line->list, &l_mg->bad_list);
 790                 return 0;
 791         }
 792
 793         atomic_set(&line->blk_in_line, chk_in_line);
 794         list_add_tail(&line->list, &l_mg->free_list);
 795         l_mg->nr_free_lines++;
 796
 797         return chk_in_line;
 798 }
 799
 800 static int pblk_alloc_line_meta(struct pblk *pblk, struct pblk_line *line)
 801 {
 802         struct pblk_line_meta *lm = &pblk->lm;
 803
 804         line->blk_bitmap = kzalloc(lm->blk_bitmap_len, GFP_KERNEL);
 805         if (!line->blk_bitmap)
 806                 return -ENOMEM;
 807
 808         line->erase_bitmap = kzalloc(lm->blk_bitmap_len, GFP_KERNEL);
 809         if (!line->erase_bitmap)
 810                 goto free_blk_bitmap;
 811
 812
 813         line->chks = kmalloc_array(lm->blk_per_line,
 814                                    sizeof(struct nvm_chk_meta), GFP_KERNEL);
 815         if (!line->chks)
 816                 goto free_erase_bitmap;
 817
 818         line->w_err_gc = kzalloc(sizeof(struct pblk_w_err_gc), GFP_KERNEL);
 819         if (!line->w_err_gc)
 820                 goto free_chks;
 821
 822         return 0;
 823
 824 free_chks:
 825         kfree(line->chks);
 826 free_erase_bitmap:
 827         kfree(line->erase_bitmap);
 828 free_blk_bitmap:
 829         kfree(line->blk_bitmap);
 830         return -ENOMEM;
 831 }
 832
 833 static int pblk_line_mg_init(struct pblk *pblk)
 834 {
 835         struct nvm_tgt_dev *dev = pblk->dev;
 836         struct nvm_geo *geo = &dev->geo;
 837         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
 838         struct pblk_line_meta *lm = &pblk->lm;
 839         int i, bb_distance;
 840
 841         l_mg->nr_lines = geo->num_chk;
 842         l_mg->log_line = l_mg->data_line = NULL;
 843         l_mg->l_seq_nr = l_mg->d_seq_nr = 0;
 844         l_mg->nr_free_lines = 0;
 845         bitmap_zero(&l_mg->meta_bitmap, PBLK_DATA_LINES);
 846
 847         INIT_LIST_HEAD(&l_mg->free_list);
 848         INIT_LIST_HEAD(&l_mg->corrupt_list);
 849         INIT_LIST_HEAD(&l_mg->bad_list);
 850         INIT_LIST_HEAD(&l_mg->gc_full_list);
 851         INIT_LIST_HEAD(&l_mg->gc_high_list);
 852         INIT_LIST_HEAD(&l_mg->gc_mid_list);
 853         INIT_LIST_HEAD(&l_mg->gc_low_list);
 854         INIT_LIST_HEAD(&l_mg->gc_empty_list);
 855         INIT_LIST_HEAD(&l_mg->gc_werr_list);
 856
 857         INIT_LIST_HEAD(&l_mg->emeta_list);
 858
 859         l_mg->gc_lists[0] = &l_mg->gc_werr_list;
 860         l_mg->gc_lists[1] = &l_mg->gc_high_list;
 861         l_mg->gc_lists[2] = &l_mg->gc_mid_list;
 862         l_mg->gc_lists[3] = &l_mg->gc_low_list;
 863
 864         spin_lock_init(&l_mg->free_lock);
 865         spin_lock_init(&l_mg->close_lock);
 866         spin_lock_init(&l_mg->gc_lock);
 867
 868         l_mg->vsc_list = kcalloc(l_mg->nr_lines, sizeof(__le32), GFP_KERNEL);
 869         if (!l_mg->vsc_list)
 870                 goto fail;
 871
 872         l_mg->bb_template = kzalloc(lm->sec_bitmap_len, GFP_KERNEL);
 873         if (!l_mg->bb_template)
 874                 goto fail_free_vsc_list;
 875
 876         l_mg->bb_aux = kzalloc(lm->sec_bitmap_len, GFP_KERNEL);
 877         if (!l_mg->bb_aux)
 878                 goto fail_free_bb_template;
 879
 880         /* smeta is always small enough to fit on a kmalloc memory allocation,
 881          * emeta depends on the number of LUNs allocated to the pblk instance
 882          */
 883         for (i = 0; i < PBLK_DATA_LINES; i++) {
 884                 l_mg->sline_meta[i] = kmalloc(lm->smeta_len, GFP_KERNEL);
 885                 if (!l_mg->sline_meta[i])
 886                         goto fail_free_smeta;
 887         }
 888
 889         l_mg->bitmap_cache = kmem_cache_create("pblk_lm_bitmap",
 890                         lm->sec_bitmap_len, 0, 0, NULL);
 891         if (!l_mg->bitmap_cache)
 892                 goto fail_free_smeta;
 893
 894         /* the bitmap pool is used for both valid and map bitmaps */
 895         l_mg->bitmap_pool = mempool_create_slab_pool(PBLK_DATA_LINES * 2,
 896                                 l_mg->bitmap_cache);
 897         if (!l_mg->bitmap_pool)
 898                 goto fail_destroy_bitmap_cache;
 899
 900         /* emeta allocates three different buffers for managing metadata with
 901          * in-memory and in-media layouts
 902          */
 903         for (i = 0; i < PBLK_DATA_LINES; i++) {
 904                 struct pblk_emeta *emeta;
 905
 906                 emeta = kmalloc(sizeof(struct pblk_emeta), GFP_KERNEL);
 907                 if (!emeta)
 908                         goto fail_free_emeta;
 909
 910                 if (lm->emeta_len[0] > KMALLOC_MAX_CACHE_SIZE) {
 911                         l_mg->emeta_alloc_type = PBLK_VMALLOC_META;
 912
 913                         emeta->buf = vmalloc(lm->emeta_len[0]);
 914                         if (!emeta->buf) {
 915                                 kfree(emeta);
 916                                 goto fail_free_emeta;
 917                         }
 918
 919                         emeta->nr_entries = lm->emeta_sec[0];
 920                         l_mg->eline_meta[i] = emeta;
 921                 } else {
 922                         l_mg->emeta_alloc_type = PBLK_KMALLOC_META;
 923
 924                         emeta->buf = kmalloc(lm->emeta_len[0], GFP_KERNEL);
 925                         if (!emeta->buf) {
 926                                 kfree(emeta);
 927                                 goto fail_free_emeta;
 928                         }
 929
 930                         emeta->nr_entries = lm->emeta_sec[0];
 931                         l_mg->eline_meta[i] = emeta;
 932                 }
 933         }
 934
 935         for (i = 0; i < l_mg->nr_lines; i++)
 936                 l_mg->vsc_list[i] = cpu_to_le32(EMPTY_ENTRY);
 937
 938         bb_distance = (geo->all_luns) * geo->ws_opt;
 939         for (i = 0; i < lm->sec_per_line; i += bb_distance)
 940                 bitmap_set(l_mg->bb_template, i, geo->ws_opt);
 941
 942         return 0;
 943
 944 fail_free_emeta:
 945         while (--i >= 0) {
 946                 if (l_mg->emeta_alloc_type == PBLK_VMALLOC_META)
 947                         vfree(l_mg->eline_meta[i]->buf);
 948                 else
 949                         kfree(l_mg->eline_meta[i]->buf);
 950                 kfree(l_mg->eline_meta[i]);
 951         }
 952
 953         mempool_destroy(l_mg->bitmap_pool);
 954 fail_destroy_bitmap_cache:
 955         kmem_cache_destroy(l_mg->bitmap_cache);
 956 fail_free_smeta:
 957         for (i = 0; i < PBLK_DATA_LINES; i++)
 958                 kfree(l_mg->sline_meta[i]);
 959         kfree(l_mg->bb_aux);
 960 fail_free_bb_template:
 961         kfree(l_mg->bb_template);
 962 fail_free_vsc_list:
 963         kfree(l_mg->vsc_list);
 964 fail:
 965         return -ENOMEM;
 966 }
 967
 968 static int pblk_line_meta_init(struct pblk *pblk)
 969 {
 970         struct nvm_tgt_dev *dev = pblk->dev;
 971         struct nvm_geo *geo = &dev->geo;
 972         struct pblk_line_meta *lm = &pblk->lm;
 973         unsigned int smeta_len, emeta_len;
 974         int i;
 975
 976         lm->sec_per_line = geo->clba * geo->all_luns;
 977         lm->blk_per_line = geo->all_luns;
 978         lm->blk_bitmap_len = BITS_TO_LONGS(geo->all_luns) * sizeof(long);
 979         lm->sec_bitmap_len = BITS_TO_LONGS(lm->sec_per_line) * sizeof(long);
 980         lm->lun_bitmap_len = BITS_TO_LONGS(geo->all_luns) * sizeof(long);
 981         lm->mid_thrs = lm->sec_per_line / 2;
 982         lm->high_thrs = lm->sec_per_line / 4;
 983         lm->meta_distance = (geo->all_luns / 2) * pblk->min_write_pgs;
 984
 985         /* Calculate necessary pages for smeta. See comment over struct
 986          * line_smeta definition
 987          */
 988         i = 1;
 989 add_smeta_page:
 990         lm->smeta_sec = i * geo->ws_opt;
 991         lm->smeta_len = lm->smeta_sec * geo->csecs;
 992
 993         smeta_len = sizeof(struct line_smeta) + lm->lun_bitmap_len;
 994         if (smeta_len > lm->smeta_len) {
 995                 i++;
 996                 goto add_smeta_page;
 997         }
 998
 999         /* Calculate necessary pages for emeta. See comment over struct
1000          * line_emeta definition
1001          */
1002         i = 1;
1003 add_emeta_page:
1004         lm->emeta_sec[0] = i * geo->ws_opt;
1005         lm->emeta_len[0] = lm->emeta_sec[0] * geo->csecs;
1006
1007         emeta_len = calc_emeta_len(pblk);
1008         if (emeta_len > lm->emeta_len[0]) {
1009                 i++;
1010                 goto add_emeta_page;
1011         }
1012
1013         lm->emeta_bb = geo->all_luns > i ? geo->all_luns - i : 0;
1014
1015         lm->min_blk_line = 1;
1016         if (geo->all_luns > 1)
1017                 lm->min_blk_line += DIV_ROUND_UP(lm->smeta_sec +
1018                                         lm->emeta_sec[0], geo->clba);
1019
1020         if (lm->min_blk_line > lm->blk_per_line) {
1021                 pblk_err(pblk, "config. not supported. Min. LUN in line:%d\n",
1022                                                         lm->blk_per_line);
1023                 return -EINVAL;
1024         }
1025
1026         return 0;
1027 }
1028
1029 static int pblk_lines_init(struct pblk *pblk)
1030 {
1031         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1032         struct pblk_line *line;
1033         void *chunk_meta;
1034         int nr_free_chks = 0;
1035         int i, ret;
1036
1037         ret = pblk_line_meta_init(pblk);
1038         if (ret)
1039                 return ret;
1040
1041         ret = pblk_line_mg_init(pblk);
1042         if (ret)
1043                 return ret;
1044
1045         ret = pblk_luns_init(pblk);
1046         if (ret)
1047                 goto fail_free_meta;
1048
1049         chunk_meta = pblk_get_chunk_meta(pblk);
1050         if (IS_ERR(chunk_meta)) {
1051                 ret = PTR_ERR(chunk_meta);
1052                 goto fail_free_luns;
1053         }
1054
1055         pblk->lines = kcalloc(l_mg->nr_lines, sizeof(struct pblk_line),
1056                                                                 GFP_KERNEL);
1057         if (!pblk->lines) {
1058                 ret = -ENOMEM;
1059                 goto fail_free_chunk_meta;
1060         }
1061
1062         for (i = 0; i < l_mg->nr_lines; i++) {
1063                 line = &pblk->lines[i];
1064
1065                 ret = pblk_alloc_line_meta(pblk, line);
1066                 if (ret)
1067                         goto fail_free_lines;
1068
1069                 nr_free_chks += pblk_setup_line_meta(pblk, line, chunk_meta, i);
1070
1071                 trace_pblk_line_state(pblk_disk_name(pblk), line->id,
1072                                                                 line->state);
1073         }
1074
1075         if (!nr_free_chks) {
1076                 pblk_err(pblk, "too many bad blocks prevent for sane instance\n");
1077                 ret = -EINTR;
1078                 goto fail_free_lines;
1079         }
1080
1081         ret = pblk_set_provision(pblk, nr_free_chks);
1082         if (ret)
1083                 goto fail_free_lines;
1084
1085         vfree(chunk_meta);
1086         return 0;
1087
1088 fail_free_lines:
1089         while (--i >= 0)
1090                 pblk_line_meta_free(l_mg, &pblk->lines[i]);
1091         kfree(pblk->lines);
1092 fail_free_chunk_meta:
1093         vfree(chunk_meta);
1094 fail_free_luns:
1095         kfree(pblk->luns);
1096 fail_free_meta:
1097         pblk_line_mg_free(pblk);
1098
1099         return ret;
1100 }
1101
1102 static int pblk_writer_init(struct pblk *pblk)
1103 {
1104         pblk->writer_ts = kthread_create(pblk_write_ts, pblk, "pblk-writer-t");
1105         if (IS_ERR(pblk->writer_ts)) {
1106                 int err = PTR_ERR(pblk->writer_ts);
1107
1108                 if (err != -EINTR)
1109                         pblk_err(pblk, "could not allocate writer kthread (%d)\n",
1110                                         err);
1111                 return err;
1112         }
1113
1114         timer_setup(&pblk->wtimer, pblk_write_timer_fn, 0);
1115         mod_timer(&pblk->wtimer, jiffies + msecs_to_jiffies(100));
1116
1117         return 0;
1118 }
1119
1120 static void pblk_writer_stop(struct pblk *pblk)
1121 {
1122         /* The pipeline must be stopped and the write buffer emptied before the
1123          * write thread is stopped
1124          */
1125         WARN(pblk_rb_read_count(&pblk->rwb),
1126                         "Stopping not fully persisted write buffer\n");
1127
1128         WARN(pblk_rb_sync_count(&pblk->rwb),
1129                         "Stopping not fully synced write buffer\n");
1130
1131         del_timer_sync(&pblk->wtimer);
1132         if (pblk->writer_ts)
1133                 kthread_stop(pblk->writer_ts);
1134 }
1135
1136 static void pblk_free(struct pblk *pblk)
1137 {
1138         pblk_lines_free(pblk);
1139         pblk_l2p_free(pblk);
1140         pblk_rwb_free(pblk);
1141         pblk_core_free(pblk);
1142
1143         kfree(pblk);
1144 }
1145
1146 static void pblk_tear_down(struct pblk *pblk, bool graceful)
1147 {
1148         if (graceful)
1149                 __pblk_pipeline_flush(pblk);
1150         __pblk_pipeline_stop(pblk);
1151         pblk_writer_stop(pblk);
1152         pblk_rb_sync_l2p(&pblk->rwb);
1153         pblk_rl_free(&pblk->rl);
1154
1155         pblk_debug(pblk, "consistent tear down (graceful:%d)\n", graceful);
1156 }
1157
1158 static void pblk_exit(void *private, bool graceful)
1159 {
1160         struct pblk *pblk = private;
1161
1162         pblk_gc_exit(pblk, graceful);
1163         pblk_tear_down(pblk, graceful);
1164
1165 #ifdef CONFIG_NVM_PBLK_DEBUG
1166         pblk_info(pblk, "exit: L2P CRC: %x\n", pblk_l2p_crc(pblk));
1167 #endif
1168
1169         pblk_free(pblk);
1170 }
1171
1172 static sector_t pblk_capacity(void *private)
1173 {
1174         struct pblk *pblk = private;
1175
1176         return pblk->capacity * NR_PHY_IN_LOG;
1177 }
1178
1179 static void *pblk_init(struct nvm_tgt_dev *dev, struct gendisk *tdisk,
1180                        int flags)
1181 {
1182         struct nvm_geo *geo = &dev->geo;
1183         struct request_queue *bqueue = dev->q;
1184         struct request_queue *tqueue = tdisk->queue;
1185         struct pblk *pblk;
1186         int ret;
1187
1188         pblk = kzalloc(sizeof(struct pblk), GFP_KERNEL);
1189         if (!pblk)
1190                 return ERR_PTR(-ENOMEM);
1191
1192         pblk->dev = dev;
1193         pblk->disk = tdisk;
1194         pblk->state = PBLK_STATE_RUNNING;
1195         trace_pblk_state(pblk_disk_name(pblk), pblk->state);
1196         pblk->gc.gc_enabled = 0;
1197
1198         if (!(geo->version == NVM_OCSSD_SPEC_12 ||
1199                                         geo->version == NVM_OCSSD_SPEC_20)) {
1200                 pblk_err(pblk, "OCSSD version not supported (%u)\n",
1201                                                         geo->version);
1202                 kfree(pblk);
1203                 return ERR_PTR(-EINVAL);
1204         }
1205
1206         if (geo->ext) {
1207                 pblk_err(pblk, "extended metadata not supported\n");
1208                 kfree(pblk);
1209                 return ERR_PTR(-EINVAL);
1210         }
1211
1212         spin_lock_init(&pblk->resubmit_lock);
1213         spin_lock_init(&pblk->trans_lock);
1214         spin_lock_init(&pblk->lock);
1215
1216 #ifdef CONFIG_NVM_PBLK_DEBUG
1217         atomic_long_set(&pblk->inflight_writes, 0);
1218         atomic_long_set(&pblk->padded_writes, 0);
1219         atomic_long_set(&pblk->padded_wb, 0);
1220         atomic_long_set(&pblk->req_writes, 0);
1221         atomic_long_set(&pblk->sub_writes, 0);
1222         atomic_long_set(&pblk->sync_writes, 0);
1223         atomic_long_set(&pblk->inflight_reads, 0);
1224         atomic_long_set(&pblk->cache_reads, 0);
1225         atomic_long_set(&pblk->sync_reads, 0);
1226         atomic_long_set(&pblk->recov_writes, 0);
1227         atomic_long_set(&pblk->recov_writes, 0);
1228         atomic_long_set(&pblk->recov_gc_writes, 0);
1229         atomic_long_set(&pblk->recov_gc_reads, 0);
1230 #endif
1231
1232         atomic_long_set(&pblk->read_failed, 0);
1233         atomic_long_set(&pblk->read_empty, 0);
1234         atomic_long_set(&pblk->read_high_ecc, 0);
1235         atomic_long_set(&pblk->read_failed_gc, 0);
1236         atomic_long_set(&pblk->write_failed, 0);
1237         atomic_long_set(&pblk->erase_failed, 0);
1238
1239         ret = pblk_core_init(pblk);
1240         if (ret) {
1241                 pblk_err(pblk, "could not initialize core\n");
1242                 goto fail;
1243         }
1244
1245         ret = pblk_lines_init(pblk);
1246         if (ret) {
1247                 pblk_err(pblk, "could not initialize lines\n");
1248                 goto fail_free_core;
1249         }
1250
1251         ret = pblk_rwb_init(pblk);
1252         if (ret) {
1253                 pblk_err(pblk, "could not initialize write buffer\n");
1254                 goto fail_free_lines;
1255         }
1256
1257         ret = pblk_l2p_init(pblk, flags & NVM_TARGET_FACTORY);
1258         if (ret) {
1259                 pblk_err(pblk, "could not initialize maps\n");
1260                 goto fail_free_rwb;
1261         }
1262
1263         ret = pblk_writer_init(pblk);
1264         if (ret) {
1265                 if (ret != -EINTR)
1266                         pblk_err(pblk, "could not initialize write thread\n");
1267                 goto fail_free_l2p;
1268         }
1269
1270         ret = pblk_gc_init(pblk);
1271         if (ret) {
1272                 pblk_err(pblk, "could not initialize gc\n");
1273                 goto fail_stop_writer;
1274         }
1275
1276         /* inherit the size from the underlying device */
1277         blk_queue_logical_block_size(tqueue, queue_physical_block_size(bqueue));
1278         blk_queue_max_hw_sectors(tqueue, queue_max_hw_sectors(bqueue));
1279
1280         blk_queue_write_cache(tqueue, true, false);
1281
1282         tqueue->limits.discard_granularity = geo->clba * geo->csecs;
1283         tqueue->limits.discard_alignment = 0;
1284         blk_queue_max_discard_sectors(tqueue, UINT_MAX >> 9);
1285         blk_queue_flag_set(QUEUE_FLAG_DISCARD, tqueue);
1286
1287         pblk_info(pblk, "luns:%u, lines:%d, secs:%llu, buf entries:%u\n",
1288                         geo->all_luns, pblk->l_mg.nr_lines,
1289                         (unsigned long long)pblk->rl.nr_secs,
1290                         pblk->rwb.nr_entries);
1291
1292         wake_up_process(pblk->writer_ts);
1293
1294         /* Check if we need to start GC */
1295         pblk_gc_should_kick(pblk);
1296
1297         return pblk;
1298
1299 fail_stop_writer:
1300         pblk_writer_stop(pblk);
1301 fail_free_l2p:
1302         pblk_l2p_free(pblk);
1303 fail_free_rwb:
1304         pblk_rwb_free(pblk);
1305 fail_free_lines:
1306         pblk_lines_free(pblk);
1307 fail_free_core:
1308         pblk_core_free(pblk);
1309 fail:
1310         kfree(pblk);
1311         return ERR_PTR(ret);
1312 }
1313
1314 /* physical block device target */
1315 static struct nvm_tgt_type tt_pblk = {
1316         .name           = "pblk",
1317         .version        = {1, 0, 0},
1318
1319         .make_rq        = pblk_make_rq,
1320         .capacity       = pblk_capacity,
1321
1322         .init           = pblk_init,
1323         .exit           = pblk_exit,
1324
1325         .sysfs_init     = pblk_sysfs_init,
1326         .sysfs_exit     = pblk_sysfs_exit,
1327         .owner          = THIS_MODULE,
1328 };
1329
1330 static int __init pblk_module_init(void)
1331 {
1332         int ret;
1333
1334         ret = bioset_init(&pblk_bio_set, BIO_POOL_SIZE, 0, 0);
1335         if (ret)
1336                 return ret;
1337         ret = nvm_register_tgt_type(&tt_pblk);
1338         if (ret)
1339                 bioset_exit(&pblk_bio_set);
1340         return ret;
1341 }
1342
1343 static void pblk_module_exit(void)
1344 {
1345         bioset_exit(&pblk_bio_set);
1346         nvm_unregister_tgt_type(&tt_pblk);
1347 }
1348
1349 module_init(pblk_module_init);
1350 module_exit(pblk_module_exit);
1351 MODULE_AUTHOR("Javier Gonzalez <[email protected]>");
1352 MODULE_AUTHOR("Matias Bjorling <[email protected]>");
1353 MODULE_LICENSE("GPL v2");
1354 MODULE_DESCRIPTION("Physical Block-Device for Open-Channel SSDs");