drivers/lightnvm/pblk-init.c

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