drivers/lightnvm/pblk-init.c

   1 /*
   2  * Copyright (C) 2015 IT University of Copenhagen (rrpc.c)
   3  * Copyright (C) 2016 CNEX Labs
   4  * Initial release: Javier Gonzalez <javier@cnexlabs.com>
   5  *                  Matias Bjorling <matias@cnexlabs.com>
   6  *
   7  * This program is free software; you can redistribute it and/or
   8  * modify it under the terms of the GNU General Public License version
   9  * 2 as published by the Free Software Foundation.
  10  *
  11  * This program is distributed in the hope that it will be useful, but
  12  * WITHOUT ANY WARRANTY; without even the implied warranty of
  13  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  14  * General Public License for more details.
  15  *
  16  * Implementation of a physical block-device target for Open-channel SSDs.
  17  *
  18  * pblk-init.c - pblk's initialization.
  19  */
  20
  21 #include "pblk.h"
  22
  23 static unsigned int write_buffer_size;
  24
  25 module_param(write_buffer_size, uint, 0644);
  26 MODULE_PARM_DESC(write_buffer_size, "number of entries in a write buffer");
  27
  28 static struct kmem_cache *pblk_ws_cache, *pblk_rec_cache, *pblk_g_rq_cache,
  29                                 *pblk_w_rq_cache;
  30 static DECLARE_RWSEM(pblk_lock);
  31 struct bio_set pblk_bio_set;
  32
  33 static int pblk_rw_io(struct request_queue *q, struct pblk *pblk,
  34                           struct bio *bio)
  35 {
  36         int ret;
  37
  38         /* Read requests must be <= 256kb due to NVMe's 64 bit completion bitmap
  39          * constraint. Writes can be of arbitrary size.
  40          */
  41         if (bio_data_dir(bio) == READ) {
  42                 blk_queue_split(q, &bio);
  43                 ret = pblk_submit_read(pblk, bio);
  44                 if (ret == NVM_IO_DONE && bio_flagged(bio, BIO_CLONED))
  45                         bio_put(bio);
  46
  47                 return ret;
  48         }
  49
  50         /* Prevent deadlock in the case of a modest LUN configuration and large
  51          * user I/Os. Unless stalled, the rate limiter leaves at least 256KB
  52          * available for user I/O.
  53          */
  54         if (pblk_get_secs(bio) > pblk_rl_max_io(&pblk->rl))
  55                 blk_queue_split(q, &bio);
  56
  57         return pblk_write_to_cache(pblk, bio, PBLK_IOTYPE_USER);
  58 }
  59
  60 static blk_qc_t pblk_make_rq(struct request_queue *q, struct bio *bio)
  61 {
  62         struct pblk *pblk = q->queuedata;
  63
  64         if (bio_op(bio) == REQ_OP_DISCARD) {
  65                 pblk_discard(pblk, bio);
  66                 if (!(bio->bi_opf & REQ_PREFLUSH)) {
  67                         bio_endio(bio);
  68                         return BLK_QC_T_NONE;
  69                 }
  70         }
  71
  72         switch (pblk_rw_io(q, pblk, bio)) {
  73         case NVM_IO_ERR:
  74                 bio_io_error(bio);
  75                 break;
  76         case NVM_IO_DONE:
  77                 bio_endio(bio);
  78                 break;
  79         }
  80
  81         return BLK_QC_T_NONE;
  82 }
  83
  84 static size_t pblk_trans_map_size(struct pblk *pblk)
  85 {
  86         int entry_size = 8;
  87
  88         if (pblk->addrf_len < 32)
  89                 entry_size = 4;
  90
  91         return entry_size * pblk->rl.nr_secs;
  92 }
  93
  94 #ifdef CONFIG_NVM_PBLK_DEBUG
  95 static u32 pblk_l2p_crc(struct pblk *pblk)
  96 {
  97         size_t map_size;
  98         u32 crc = ~(u32)0;
  99
 100         map_size = pblk_trans_map_size(pblk);
 101         crc = crc32_le(crc, pblk->trans_map, map_size);
 102         return crc;
 103 }
 104 #endif
 105
 106 static void pblk_l2p_free(struct pblk *pblk)
 107 {
 108         vfree(pblk->trans_map);
 109 }
 110
 111 static int pblk_l2p_recover(struct pblk *pblk, bool factory_init)
 112 {
 113         struct pblk_line *line = NULL;
 114
 115         if (factory_init) {
 116                 pblk_setup_uuid(pblk);
 117         } else {
 118                 line = pblk_recov_l2p(pblk);
 119                 if (IS_ERR(line)) {
 120                         pblk_err(pblk, "could not recover l2p table\n");
 121                         return -EFAULT;
 122                 }
 123         }
 124
 125 #ifdef CONFIG_NVM_PBLK_DEBUG
 126         pblk_info(pblk, "init: L2P CRC: %x\n", pblk_l2p_crc(pblk));
 127 #endif
 128
 129         /* Free full lines directly as GC has not been started yet */
 130         pblk_gc_free_full_lines(pblk);
 131
 132         if (!line) {
 133                 /* Configure next line for user data */
 134                 line = pblk_line_get_first_data(pblk);
 135                 if (!line)
 136                         return -EFAULT;
 137         }
 138
 139         return 0;
 140 }
 141
 142 static int pblk_l2p_init(struct pblk *pblk, bool factory_init)
 143 {
 144         sector_t i;
 145         struct ppa_addr ppa;
 146         size_t map_size;
 147         int ret = 0;
 148
 149         map_size = pblk_trans_map_size(pblk);
 150         pblk->trans_map = vmalloc(map_size);
 151         if (!pblk->trans_map)
 152                 return -ENOMEM;
 153
 154         pblk_ppa_set_empty(&ppa);
 155
 156         for (i = 0; i < pblk->rl.nr_secs; i++)
 157                 pblk_trans_map_set(pblk, i, ppa);
 158
 159         ret = pblk_l2p_recover(pblk, factory_init);
 160         if (ret)
 161                 vfree(pblk->trans_map);
 162
 163         return ret;
 164 }
 165
 166 static void pblk_rwb_free(struct pblk *pblk)
 167 {
 168         if (pblk_rb_tear_down_check(&pblk->rwb))
 169                 pblk_err(pblk, "write buffer error on tear down\n");
 170
 171         pblk_rb_data_free(&pblk->rwb);
 172         vfree(pblk_rb_entries_ref(&pblk->rwb));
 173 }
 174
 175 static int pblk_rwb_init(struct pblk *pblk)
 176 {
 177         struct nvm_tgt_dev *dev = pblk->dev;
 178         struct nvm_geo *geo = &dev->geo;
 179         struct pblk_rb_entry *entries;
 180         unsigned long nr_entries, buffer_size;
 181         unsigned int power_size, power_seg_sz;
 182         int pgs_in_buffer;
 183
 184         pgs_in_buffer = (max(geo->mw_cunits, geo->ws_opt) + geo->ws_opt)
 185                                                                 * geo->all_luns;
 186
 187         if (write_buffer_size && (write_buffer_size > pgs_in_buffer))
 188                 buffer_size = write_buffer_size;
 189         else
 190                 buffer_size = pgs_in_buffer;
 191
 192         nr_entries = pblk_rb_calculate_size(buffer_size);
 193
 194         entries = vzalloc(array_size(nr_entries, sizeof(struct pblk_rb_entry)));
 195         if (!entries)
 196                 return -ENOMEM;
 197
 198         power_size = get_count_order(nr_entries);
 199         power_seg_sz = get_count_order(geo->csecs);
 200
 201         return pblk_rb_init(&pblk->rwb, entries, power_size, power_seg_sz);
 202 }
 203
 204 /* Minimum pages needed within a lun */
 205 #define ADDR_POOL_SIZE 64
 206
 207 static int pblk_set_addrf_12(struct pblk *pblk, struct nvm_geo *geo,
 208                              struct nvm_addrf_12 *dst)
 209 {
 210         struct nvm_addrf_12 *src = (struct nvm_addrf_12 *)&geo->addrf;
 211         int power_len;
 212
 213         /* Re-calculate channel and lun format to adapt to configuration */
 214         power_len = get_count_order(geo->num_ch);
 215         if (1 << power_len != geo->num_ch) {
 216                 pblk_err(pblk, "supports only power-of-two channel config.\n");
 217                 return -EINVAL;
 218         }
 219         dst->ch_len = power_len;
 220
 221         power_len = get_count_order(geo->num_lun);
 222         if (1 << power_len != geo->num_lun) {
 223                 pblk_err(pblk, "supports only power-of-two LUN config.\n");
 224                 return -EINVAL;
 225         }
 226         dst->lun_len = power_len;
 227
 228         dst->blk_len = src->blk_len;
 229         dst->pg_len = src->pg_len;
 230         dst->pln_len = src->pln_len;
 231         dst->sec_len = src->sec_len;
 232
 233         dst->sec_offset = 0;
 234         dst->pln_offset = dst->sec_len;
 235         dst->ch_offset = dst->pln_offset + dst->pln_len;
 236         dst->lun_offset = dst->ch_offset + dst->ch_len;
 237         dst->pg_offset = dst->lun_offset + dst->lun_len;
 238         dst->blk_offset = dst->pg_offset + dst->pg_len;
 239
 240         dst->sec_mask = ((1ULL << dst->sec_len) - 1) << dst->sec_offset;
 241         dst->pln_mask = ((1ULL << dst->pln_len) - 1) << dst->pln_offset;
 242         dst->ch_mask = ((1ULL << dst->ch_len) - 1) << dst->ch_offset;
 243         dst->lun_mask = ((1ULL << dst->lun_len) - 1) << dst->lun_offset;
 244         dst->pg_mask = ((1ULL << dst->pg_len) - 1) << dst->pg_offset;
 245         dst->blk_mask = ((1ULL << dst->blk_len) - 1) << dst->blk_offset;
 246
 247         return dst->blk_offset + src->blk_len;
 248 }
 249
 250 static int pblk_set_addrf_20(struct nvm_geo *geo, struct nvm_addrf *adst,
 251                              struct pblk_addrf *udst)
 252 {
 253         struct nvm_addrf *src = &geo->addrf;
 254
 255         adst->ch_len = get_count_order(geo->num_ch);
 256         adst->lun_len = get_count_order(geo->num_lun);
 257         adst->chk_len = src->chk_len;
 258         adst->sec_len = src->sec_len;
 259
 260         adst->sec_offset = 0;
 261         adst->ch_offset = adst->sec_len;
 262         adst->lun_offset = adst->ch_offset + adst->ch_len;
 263         adst->chk_offset = adst->lun_offset + adst->lun_len;
 264
 265         adst->sec_mask = ((1ULL << adst->sec_len) - 1) << adst->sec_offset;
 266         adst->chk_mask = ((1ULL << adst->chk_len) - 1) << adst->chk_offset;
 267         adst->lun_mask = ((1ULL << adst->lun_len) - 1) << adst->lun_offset;
 268         adst->ch_mask = ((1ULL << adst->ch_len) - 1) << adst->ch_offset;
 269
 270         udst->sec_stripe = geo->ws_opt;
 271         udst->ch_stripe = geo->num_ch;
 272         udst->lun_stripe = geo->num_lun;
 273
 274         udst->sec_lun_stripe = udst->sec_stripe * udst->ch_stripe;
 275         udst->sec_ws_stripe = udst->sec_lun_stripe * udst->lun_stripe;
 276
 277         return adst->chk_offset + adst->chk_len;
 278 }
 279
 280 static int pblk_set_addrf(struct pblk *pblk)
 281 {
 282         struct nvm_tgt_dev *dev = pblk->dev;
 283         struct nvm_geo *geo = &dev->geo;
 284         int mod;
 285
 286         switch (geo->version) {
 287         case NVM_OCSSD_SPEC_12:
 288                 div_u64_rem(geo->clba, pblk->min_write_pgs, &mod);
 289                 if (mod) {
 290                         pblk_err(pblk, "bad configuration of sectors/pages\n");
 291                         return -EINVAL;
 292                 }
 293
 294                 pblk->addrf_len = pblk_set_addrf_12(pblk, geo,
 295                                                         (void *)&pblk->addrf);
 296                 break;
 297         case NVM_OCSSD_SPEC_20:
 298                 pblk->addrf_len = pblk_set_addrf_20(geo, (void *)&pblk->addrf,
 299                                                         &pblk->uaddrf);
 300                 break;
 301         default:
 302                 pblk_err(pblk, "OCSSD revision not supported (%d)\n",
 303                                                                 geo->version);
 304                 return -EINVAL;
 305         }
 306
 307         return 0;
 308 }
 309
 310 static int pblk_init_global_caches(struct pblk *pblk)
 311 {
 312         down_write(&pblk_lock);
 313         pblk_ws_cache = kmem_cache_create("pblk_blk_ws",
 314                                 sizeof(struct pblk_line_ws), 0, 0, NULL);
 315         if (!pblk_ws_cache) {
 316                 up_write(&pblk_lock);
 317                 return -ENOMEM;
 318         }
 319
 320         pblk_rec_cache = kmem_cache_create("pblk_rec",
 321                                 sizeof(struct pblk_rec_ctx), 0, 0, NULL);
 322         if (!pblk_rec_cache) {
 323                 kmem_cache_destroy(pblk_ws_cache);
 324                 up_write(&pblk_lock);
 325                 return -ENOMEM;
 326         }
 327
 328         pblk_g_rq_cache = kmem_cache_create("pblk_g_rq", pblk_g_rq_size,
 329                                 0, 0, NULL);
 330         if (!pblk_g_rq_cache) {
 331                 kmem_cache_destroy(pblk_ws_cache);
 332                 kmem_cache_destroy(pblk_rec_cache);
 333                 up_write(&pblk_lock);
 334                 return -ENOMEM;
 335         }
 336
 337         pblk_w_rq_cache = kmem_cache_create("pblk_w_rq", pblk_w_rq_size,
 338                                 0, 0, NULL);
 339         if (!pblk_w_rq_cache) {
 340                 kmem_cache_destroy(pblk_ws_cache);
 341                 kmem_cache_destroy(pblk_rec_cache);
 342                 kmem_cache_destroy(pblk_g_rq_cache);
 343                 up_write(&pblk_lock);
 344                 return -ENOMEM;
 345         }
 346         up_write(&pblk_lock);
 347
 348         return 0;
 349 }
 350
 351 static void pblk_free_global_caches(struct pblk *pblk)
 352 {
 353         kmem_cache_destroy(pblk_ws_cache);
 354         kmem_cache_destroy(pblk_rec_cache);
 355         kmem_cache_destroy(pblk_g_rq_cache);
 356         kmem_cache_destroy(pblk_w_rq_cache);
 357 }
 358
 359 static int pblk_core_init(struct pblk *pblk)
 360 {
 361         struct nvm_tgt_dev *dev = pblk->dev;
 362         struct nvm_geo *geo = &dev->geo;
 363         int ret, max_write_ppas;
 364
 365         atomic64_set(&pblk->user_wa, 0);
 366         atomic64_set(&pblk->pad_wa, 0);
 367         atomic64_set(&pblk->gc_wa, 0);
 368         pblk->user_rst_wa = 0;
 369         pblk->pad_rst_wa = 0;
 370         pblk->gc_rst_wa = 0;
 371
 372         atomic64_set(&pblk->nr_flush, 0);
 373         pblk->nr_flush_rst = 0;
 374
 375         pblk->min_write_pgs = geo->ws_opt;
 376         max_write_ppas = pblk->min_write_pgs * geo->all_luns;
 377         pblk->max_write_pgs = min_t(int, max_write_ppas, NVM_MAX_VLBA);
 378         pblk->max_write_pgs = min_t(int, pblk->max_write_pgs,
 379                 queue_max_hw_sectors(dev->q) / (geo->csecs >> SECTOR_SHIFT));
 380         pblk_set_sec_per_write(pblk, pblk->min_write_pgs);
 381
 382         if (pblk->max_write_pgs > PBLK_MAX_REQ_ADDRS) {
 383                 pblk_err(pblk, "vector list too big(%u > %u)\n",
 384                                 pblk->max_write_pgs, PBLK_MAX_REQ_ADDRS);
 385                 return -EINVAL;
 386         }
 387
 388         pblk->pad_dist = kcalloc(pblk->min_write_pgs - 1, sizeof(atomic64_t),
 389                                                                 GFP_KERNEL);
 390         if (!pblk->pad_dist)
 391                 return -ENOMEM;
 392
 393         if (pblk_init_global_caches(pblk))
 394                 goto fail_free_pad_dist;
 395
 396         /* Internal bios can be at most the sectors signaled by the device. */
 397         ret = mempool_init_page_pool(&pblk->page_bio_pool, NVM_MAX_VLBA, 0);
 398         if (ret)
 399                 goto free_global_caches;
 400
 401         ret = mempool_init_slab_pool(&pblk->gen_ws_pool, PBLK_GEN_WS_POOL_SIZE,
 402                                      pblk_ws_cache);
 403         if (ret)
 404                 goto free_page_bio_pool;
 405
 406         ret = mempool_init_slab_pool(&pblk->rec_pool, geo->all_luns,
 407                                      pblk_rec_cache);
 408         if (ret)
 409                 goto free_gen_ws_pool;
 410
 411         ret = mempool_init_slab_pool(&pblk->r_rq_pool, geo->all_luns,
 412                                      pblk_g_rq_cache);
 413         if (ret)
 414                 goto free_rec_pool;
 415
 416         ret = mempool_init_slab_pool(&pblk->e_rq_pool, geo->all_luns,
 417                                      pblk_g_rq_cache);
 418         if (ret)
 419                 goto free_r_rq_pool;
 420
 421         ret = mempool_init_slab_pool(&pblk->w_rq_pool, geo->all_luns,
 422                                      pblk_w_rq_cache);
 423         if (ret)
 424                 goto free_e_rq_pool;
 425
 426         pblk->close_wq = alloc_workqueue("pblk-close-wq",
 427                         WQ_MEM_RECLAIM | WQ_UNBOUND, PBLK_NR_CLOSE_JOBS);
 428         if (!pblk->close_wq)
 429                 goto free_w_rq_pool;
 430
 431         pblk->bb_wq = alloc_workqueue("pblk-bb-wq",
 432                         WQ_MEM_RECLAIM | WQ_UNBOUND, 0);
 433         if (!pblk->bb_wq)
 434                 goto free_close_wq;
 435
 436         pblk->r_end_wq = alloc_workqueue("pblk-read-end-wq",
 437                         WQ_MEM_RECLAIM | WQ_UNBOUND, 0);
 438         if (!pblk->r_end_wq)
 439                 goto free_bb_wq;
 440
 441         if (pblk_set_addrf(pblk))
 442                 goto free_r_end_wq;
 443
 444         INIT_LIST_HEAD(&pblk->compl_list);
 445         INIT_LIST_HEAD(&pblk->resubmit_list);
 446
 447         return 0;
 448
 449 free_r_end_wq:
 450         destroy_workqueue(pblk->r_end_wq);
 451 free_bb_wq:
 452         destroy_workqueue(pblk->bb_wq);
 453 free_close_wq:
 454         destroy_workqueue(pblk->close_wq);
 455 free_w_rq_pool:
 456         mempool_exit(&pblk->w_rq_pool);
 457 free_e_rq_pool:
 458         mempool_exit(&pblk->e_rq_pool);
 459 free_r_rq_pool:
 460         mempool_exit(&pblk->r_rq_pool);
 461 free_rec_pool:
 462         mempool_exit(&pblk->rec_pool);
 463 free_gen_ws_pool:
 464         mempool_exit(&pblk->gen_ws_pool);
 465 free_page_bio_pool:
 466         mempool_exit(&pblk->page_bio_pool);
 467 free_global_caches:
 468         pblk_free_global_caches(pblk);
 469 fail_free_pad_dist:
 470         kfree(pblk->pad_dist);
 471         return -ENOMEM;
 472 }
 473
 474 static void pblk_core_free(struct pblk *pblk)
 475 {
 476         if (pblk->close_wq)
 477                 destroy_workqueue(pblk->close_wq);
 478
 479         if (pblk->r_end_wq)
 480                 destroy_workqueue(pblk->r_end_wq);
 481
 482         if (pblk->bb_wq)
 483                 destroy_workqueue(pblk->bb_wq);
 484
 485         mempool_exit(&pblk->page_bio_pool);
 486         mempool_exit(&pblk->gen_ws_pool);
 487         mempool_exit(&pblk->rec_pool);
 488         mempool_exit(&pblk->r_rq_pool);
 489         mempool_exit(&pblk->e_rq_pool);
 490         mempool_exit(&pblk->w_rq_pool);
 491
 492         pblk_free_global_caches(pblk);
 493         kfree(pblk->pad_dist);
 494 }
 495
 496 static void pblk_line_mg_free(struct pblk *pblk)
 497 {
 498         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
 499         int i;
 500
 501         kfree(l_mg->bb_template);
 502         kfree(l_mg->bb_aux);
 503         kfree(l_mg->vsc_list);
 504
 505         for (i = 0; i < PBLK_DATA_LINES; i++) {
 506                 kfree(l_mg->sline_meta[i]);
 507                 pblk_mfree(l_mg->eline_meta[i]->buf, l_mg->emeta_alloc_type);
 508                 kfree(l_mg->eline_meta[i]);
 509         }
 510 }
 511
 512 static void pblk_line_meta_free(struct pblk_line_mgmt *l_mg,
 513                                 struct pblk_line *line)
 514 {
 515         struct pblk_w_err_gc *w_err_gc = line->w_err_gc;
 516
 517         kfree(line->blk_bitmap);
 518         kfree(line->erase_bitmap);
 519         kfree(line->chks);
 520
 521         pblk_mfree(w_err_gc->lba_list, l_mg->emeta_alloc_type);
 522         kfree(w_err_gc);
 523 }
 524
 525 static void pblk_lines_free(struct pblk *pblk)
 526 {
 527         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
 528         struct pblk_line *line;
 529         int i;
 530
 531         spin_lock(&l_mg->free_lock);
 532         for (i = 0; i < l_mg->nr_lines; i++) {
 533                 line = &pblk->lines[i];
 534
 535                 pblk_line_free(line);
 536                 pblk_line_meta_free(l_mg, line);
 537         }
 538         spin_unlock(&l_mg->free_lock);
 539
 540         pblk_line_mg_free(pblk);
 541
 542         kfree(pblk->luns);
 543         kfree(pblk->lines);
 544 }
 545
 546 static int pblk_bb_get_tbl(struct nvm_tgt_dev *dev, struct pblk_lun *rlun,
 547                            u8 *blks, int nr_blks)
 548 {
 549         struct ppa_addr ppa;
 550         int ret;
 551
 552         ppa.ppa = 0;
 553         ppa.g.ch = rlun->bppa.g.ch;
 554         ppa.g.lun = rlun->bppa.g.lun;
 555
 556         ret = nvm_get_tgt_bb_tbl(dev, ppa, blks);
 557         if (ret)
 558                 return ret;
 559
 560         nr_blks = nvm_bb_tbl_fold(dev->parent, blks, nr_blks);
 561         if (nr_blks < 0)
 562                 return -EIO;
 563
 564         return 0;
 565 }
 566
 567 static void *pblk_bb_get_meta(struct pblk *pblk)
 568 {
 569         struct nvm_tgt_dev *dev = pblk->dev;
 570         struct nvm_geo *geo = &dev->geo;
 571         u8 *meta;
 572         int i, nr_blks, blk_per_lun;
 573         int ret;
 574
 575         blk_per_lun = geo->num_chk * geo->pln_mode;
 576         nr_blks = blk_per_lun * geo->all_luns;
 577
 578         meta = kmalloc(nr_blks, GFP_KERNEL);
 579         if (!meta)
 580                 return ERR_PTR(-ENOMEM);
 581
 582         for (i = 0; i < geo->all_luns; i++) {
 583                 struct pblk_lun *rlun = &pblk->luns[i];
 584                 u8 *meta_pos = meta + i * blk_per_lun;
 585
 586                 ret = pblk_bb_get_tbl(dev, rlun, meta_pos, blk_per_lun);
 587                 if (ret) {
 588                         kfree(meta);
 589                         return ERR_PTR(-EIO);
 590                 }
 591         }
 592
 593         return meta;
 594 }
 595
 596 static void *pblk_chunk_get_meta(struct pblk *pblk)
 597 {
 598         struct nvm_tgt_dev *dev = pblk->dev;
 599         struct nvm_geo *geo = &dev->geo;
 600
 601         if (geo->version == NVM_OCSSD_SPEC_12)
 602                 return pblk_bb_get_meta(pblk);
 603         else
 604                 return pblk_chunk_get_info(pblk);
 605 }
 606
 607 static int pblk_luns_init(struct pblk *pblk)
 608 {
 609         struct nvm_tgt_dev *dev = pblk->dev;
 610         struct nvm_geo *geo = &dev->geo;
 611         struct pblk_lun *rlun;
 612         int i;
 613
 614         /* TODO: Implement unbalanced LUN support */
 615         if (geo->num_lun < 0) {
 616                 pblk_err(pblk, "unbalanced LUN config.\n");
 617                 return -EINVAL;
 618         }
 619
 620         pblk->luns = kcalloc(geo->all_luns, sizeof(struct pblk_lun),
 621                                                                 GFP_KERNEL);
 622         if (!pblk->luns)
 623                 return -ENOMEM;
 624
 625         for (i = 0; i < geo->all_luns; i++) {
 626                 /* Stripe across channels */
 627                 int ch = i % geo->num_ch;
 628                 int lun_raw = i / geo->num_ch;
 629                 int lunid = lun_raw + ch * geo->num_lun;
 630
 631                 rlun = &pblk->luns[i];
 632                 rlun->bppa = dev->luns[lunid];
 633
 634                 sema_init(&rlun->wr_sem, 1);
 635         }
 636
 637         return 0;
 638 }
 639
 640 /* See comment over struct line_emeta definition */
 641 static unsigned int calc_emeta_len(struct pblk *pblk)
 642 {
 643         struct pblk_line_meta *lm = &pblk->lm;
 644         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
 645         struct nvm_tgt_dev *dev = pblk->dev;
 646         struct nvm_geo *geo = &dev->geo;
 647
 648         /* Round to sector size so that lba_list starts on its own sector */
 649         lm->emeta_sec[1] = DIV_ROUND_UP(
 650                         sizeof(struct line_emeta) + lm->blk_bitmap_len +
 651                         sizeof(struct wa_counters), geo->csecs);
 652         lm->emeta_len[1] = lm->emeta_sec[1] * geo->csecs;
 653
 654         /* Round to sector size so that vsc_list starts on its own sector */
 655         lm->dsec_per_line = lm->sec_per_line - lm->emeta_sec[0];
 656         lm->emeta_sec[2] = DIV_ROUND_UP(lm->dsec_per_line * sizeof(u64),
 657                         geo->csecs);
 658         lm->emeta_len[2] = lm->emeta_sec[2] * geo->csecs;
 659
 660         lm->emeta_sec[3] = DIV_ROUND_UP(l_mg->nr_lines * sizeof(u32),
 661                         geo->csecs);
 662         lm->emeta_len[3] = lm->emeta_sec[3] * geo->csecs;
 663
 664         lm->vsc_list_len = l_mg->nr_lines * sizeof(u32);
 665
 666         return (lm->emeta_len[1] + lm->emeta_len[2] + lm->emeta_len[3]);
 667 }
 668
 669 static void pblk_set_provision(struct pblk *pblk, long nr_free_blks)
 670 {
 671         struct nvm_tgt_dev *dev = pblk->dev;
 672         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
 673         struct pblk_line_meta *lm = &pblk->lm;
 674         struct nvm_geo *geo = &dev->geo;
 675         sector_t provisioned;
 676         int sec_meta, blk_meta;
 677
 678         if (geo->op == NVM_TARGET_DEFAULT_OP)
 679                 pblk->op = PBLK_DEFAULT_OP;
 680         else
 681                 pblk->op = geo->op;
 682
 683         provisioned = nr_free_blks;
 684         provisioned *= (100 - pblk->op);
 685         sector_div(provisioned, 100);
 686
 687         pblk->op_blks = nr_free_blks - provisioned;
 688
 689         /* Internally pblk manages all free blocks, but all calculations based
 690          * on user capacity consider only provisioned blocks
 691          */
 692         pblk->rl.total_blocks = nr_free_blks;
 693         pblk->rl.nr_secs = nr_free_blks * geo->clba;
 694
 695         /* Consider sectors used for metadata */
 696         sec_meta = (lm->smeta_sec + lm->emeta_sec[0]) * l_mg->nr_free_lines;
 697         blk_meta = DIV_ROUND_UP(sec_meta, geo->clba);
 698
 699         pblk->capacity = (provisioned - blk_meta) * geo->clba;
 700
 701         atomic_set(&pblk->rl.free_blocks, nr_free_blks);
 702         atomic_set(&pblk->rl.free_user_blocks, nr_free_blks);
 703 }
 704
 705 static int pblk_setup_line_meta_12(struct pblk *pblk, struct pblk_line *line,
 706                                    void *chunk_meta)
 707 {
 708         struct nvm_tgt_dev *dev = pblk->dev;
 709         struct nvm_geo *geo = &dev->geo;
 710         struct pblk_line_meta *lm = &pblk->lm;
 711         int i, chk_per_lun, nr_bad_chks = 0;
 712
 713         chk_per_lun = geo->num_chk * geo->pln_mode;
 714
 715         for (i = 0; i < lm->blk_per_line; i++) {
 716                 struct pblk_lun *rlun = &pblk->luns[i];
 717                 struct nvm_chk_meta *chunk;
 718                 int pos = pblk_ppa_to_pos(geo, rlun->bppa);
 719                 u8 *lun_bb_meta = chunk_meta + pos * chk_per_lun;
 720
 721                 chunk = &line->chks[pos];
 722
 723                 /*
 724                  * In 1.2 spec. chunk state is not persisted by the device. Thus
 725                  * some of the values are reset each time pblk is instantiated,
 726                  * so we have to assume that the block is closed.
 727                  */
 728                 if (lun_bb_meta[line->id] == NVM_BLK_T_FREE)
 729                         chunk->state =  NVM_CHK_ST_CLOSED;
 730                 else
 731                         chunk->state = NVM_CHK_ST_OFFLINE;
 732
 733                 chunk->type = NVM_CHK_TP_W_SEQ;
 734                 chunk->wi = 0;
 735                 chunk->slba = -1;
 736                 chunk->cnlb = geo->clba;
 737                 chunk->wp = 0;
 738
 739                 if (!(chunk->state & NVM_CHK_ST_OFFLINE))
 740                         continue;
 741
 742                 set_bit(pos, line->blk_bitmap);
 743                 nr_bad_chks++;
 744         }
 745
 746         return nr_bad_chks;
 747 }
 748
 749 static int pblk_setup_line_meta_20(struct pblk *pblk, struct pblk_line *line,
 750                                    struct nvm_chk_meta *meta)
 751 {
 752         struct nvm_tgt_dev *dev = pblk->dev;
 753         struct nvm_geo *geo = &dev->geo;
 754         struct pblk_line_meta *lm = &pblk->lm;
 755         int i, nr_bad_chks = 0;
 756
 757         for (i = 0; i < lm->blk_per_line; i++) {
 758                 struct pblk_lun *rlun = &pblk->luns[i];
 759                 struct nvm_chk_meta *chunk;
 760                 struct nvm_chk_meta *chunk_meta;
 761                 struct ppa_addr ppa;
 762                 int pos;
 763
 764                 ppa = rlun->bppa;
 765                 pos = pblk_ppa_to_pos(geo, ppa);
 766                 chunk = &line->chks[pos];
 767
 768                 ppa.m.chk = line->id;
 769                 chunk_meta = pblk_chunk_get_off(pblk, meta, ppa);
 770
 771                 chunk->state = chunk_meta->state;
 772                 chunk->type = chunk_meta->type;
 773                 chunk->wi = chunk_meta->wi;
 774                 chunk->slba = chunk_meta->slba;
 775                 chunk->cnlb = chunk_meta->cnlb;
 776                 chunk->wp = chunk_meta->wp;
 777
 778                 if (chunk->type & NVM_CHK_TP_SZ_SPEC) {
 779                         WARN_ONCE(1, "pblk: custom-sized chunks unsupported\n");
 780                         continue;
 781                 }
 782
 783                 if (!(chunk->state & NVM_CHK_ST_OFFLINE))
 784                         continue;
 785
 786                 set_bit(pos, line->blk_bitmap);
 787                 nr_bad_chks++;
 788         }
 789
 790         return nr_bad_chks;
 791 }
 792
 793 static long pblk_setup_line_meta(struct pblk *pblk, struct pblk_line *line,
 794                                  void *chunk_meta, int line_id)
 795 {
 796         struct nvm_tgt_dev *dev = pblk->dev;
 797         struct nvm_geo *geo = &dev->geo;
 798         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
 799         struct pblk_line_meta *lm = &pblk->lm;
 800         long nr_bad_chks, chk_in_line;
 801
 802         line->pblk = pblk;
 803         line->id = line_id;
 804         line->type = PBLK_LINETYPE_FREE;
 805         line->state = PBLK_LINESTATE_NEW;
 806         line->gc_group = PBLK_LINEGC_NONE;
 807         line->vsc = &l_mg->vsc_list[line_id];
 808         spin_lock_init(&line->lock);
 809
 810         if (geo->version == NVM_OCSSD_SPEC_12)
 811                 nr_bad_chks = pblk_setup_line_meta_12(pblk, line, chunk_meta);
 812         else
 813                 nr_bad_chks = pblk_setup_line_meta_20(pblk, line, chunk_meta);
 814
 815         chk_in_line = lm->blk_per_line - nr_bad_chks;
 816         if (nr_bad_chks < 0 || nr_bad_chks > lm->blk_per_line ||
 817                                         chk_in_line < lm->min_blk_line) {
 818                 line->state = PBLK_LINESTATE_BAD;
 819                 list_add_tail(&line->list, &l_mg->bad_list);
 820                 return 0;
 821         }
 822
 823         atomic_set(&line->blk_in_line, chk_in_line);
 824         list_add_tail(&line->list, &l_mg->free_list);
 825         l_mg->nr_free_lines++;
 826
 827         return chk_in_line;
 828 }
 829
 830 static int pblk_alloc_line_meta(struct pblk *pblk, struct pblk_line *line)
 831 {
 832         struct pblk_line_meta *lm = &pblk->lm;
 833
 834         line->blk_bitmap = kzalloc(lm->blk_bitmap_len, GFP_KERNEL);
 835         if (!line->blk_bitmap)
 836                 return -ENOMEM;
 837
 838         line->erase_bitmap = kzalloc(lm->blk_bitmap_len, GFP_KERNEL);
 839         if (!line->erase_bitmap)
 840                 goto free_blk_bitmap;
 841
 842
 843         line->chks = kmalloc_array(lm->blk_per_line,
 844                                    sizeof(struct nvm_chk_meta), GFP_KERNEL);
 845         if (!line->chks)
 846                 goto free_erase_bitmap;
 847
 848         line->w_err_gc = kzalloc(sizeof(struct pblk_w_err_gc), GFP_KERNEL);
 849         if (!line->w_err_gc)
 850                 goto free_chks;
 851
 852         return 0;
 853
 854 free_chks:
 855         kfree(line->chks);
 856 free_erase_bitmap:
 857         kfree(line->erase_bitmap);
 858 free_blk_bitmap:
 859         kfree(line->blk_bitmap);
 860         return -ENOMEM;
 861 }
 862
 863 static int pblk_line_mg_init(struct pblk *pblk)
 864 {
 865         struct nvm_tgt_dev *dev = pblk->dev;
 866         struct nvm_geo *geo = &dev->geo;
 867         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
 868         struct pblk_line_meta *lm = &pblk->lm;
 869         int i, bb_distance;
 870
 871         l_mg->nr_lines = geo->num_chk;
 872         l_mg->log_line = l_mg->data_line = NULL;
 873         l_mg->l_seq_nr = l_mg->d_seq_nr = 0;
 874         l_mg->nr_free_lines = 0;
 875         bitmap_zero(&l_mg->meta_bitmap, PBLK_DATA_LINES);
 876
 877         INIT_LIST_HEAD(&l_mg->free_list);
 878         INIT_LIST_HEAD(&l_mg->corrupt_list);
 879         INIT_LIST_HEAD(&l_mg->bad_list);
 880         INIT_LIST_HEAD(&l_mg->gc_full_list);
 881         INIT_LIST_HEAD(&l_mg->gc_high_list);
 882         INIT_LIST_HEAD(&l_mg->gc_mid_list);
 883         INIT_LIST_HEAD(&l_mg->gc_low_list);
 884         INIT_LIST_HEAD(&l_mg->gc_empty_list);
 885         INIT_LIST_HEAD(&l_mg->gc_werr_list);
 886
 887         INIT_LIST_HEAD(&l_mg->emeta_list);
 888
 889         l_mg->gc_lists[0] = &l_mg->gc_werr_list;
 890         l_mg->gc_lists[1] = &l_mg->gc_high_list;
 891         l_mg->gc_lists[2] = &l_mg->gc_mid_list;
 892         l_mg->gc_lists[3] = &l_mg->gc_low_list;
 893
 894         spin_lock_init(&l_mg->free_lock);
 895         spin_lock_init(&l_mg->close_lock);
 896         spin_lock_init(&l_mg->gc_lock);
 897
 898         l_mg->vsc_list = kcalloc(l_mg->nr_lines, sizeof(__le32), GFP_KERNEL);
 899         if (!l_mg->vsc_list)
 900                 goto fail;
 901
 902         l_mg->bb_template = kzalloc(lm->sec_bitmap_len, GFP_KERNEL);
 903         if (!l_mg->bb_template)
 904                 goto fail_free_vsc_list;
 905
 906         l_mg->bb_aux = kzalloc(lm->sec_bitmap_len, GFP_KERNEL);
 907         if (!l_mg->bb_aux)
 908                 goto fail_free_bb_template;
 909
 910         /* smeta is always small enough to fit on a kmalloc memory allocation,
 911          * emeta depends on the number of LUNs allocated to the pblk instance
 912          */
 913         for (i = 0; i < PBLK_DATA_LINES; i++) {
 914                 l_mg->sline_meta[i] = kmalloc(lm->smeta_len, GFP_KERNEL);
 915                 if (!l_mg->sline_meta[i])
 916                         goto fail_free_smeta;
 917         }
 918
 919         /* emeta allocates three different buffers for managing metadata with
 920          * in-memory and in-media layouts
 921          */
 922         for (i = 0; i < PBLK_DATA_LINES; i++) {
 923                 struct pblk_emeta *emeta;
 924
 925                 emeta = kmalloc(sizeof(struct pblk_emeta), GFP_KERNEL);
 926                 if (!emeta)
 927                         goto fail_free_emeta;
 928
 929                 if (lm->emeta_len[0] > KMALLOC_MAX_CACHE_SIZE) {
 930                         l_mg->emeta_alloc_type = PBLK_VMALLOC_META;
 931
 932                         emeta->buf = vmalloc(lm->emeta_len[0]);
 933                         if (!emeta->buf) {
 934                                 kfree(emeta);
 935                                 goto fail_free_emeta;
 936                         }
 937
 938                         emeta->nr_entries = lm->emeta_sec[0];
 939                         l_mg->eline_meta[i] = emeta;
 940                 } else {
 941                         l_mg->emeta_alloc_type = PBLK_KMALLOC_META;
 942
 943                         emeta->buf = kmalloc(lm->emeta_len[0], GFP_KERNEL);
 944                         if (!emeta->buf) {
 945                                 kfree(emeta);
 946                                 goto fail_free_emeta;
 947                         }
 948
 949                         emeta->nr_entries = lm->emeta_sec[0];
 950                         l_mg->eline_meta[i] = emeta;
 951                 }
 952         }
 953
 954         for (i = 0; i < l_mg->nr_lines; i++)
 955                 l_mg->vsc_list[i] = cpu_to_le32(EMPTY_ENTRY);
 956
 957         bb_distance = (geo->all_luns) * geo->ws_opt;
 958         for (i = 0; i < lm->sec_per_line; i += bb_distance)
 959                 bitmap_set(l_mg->bb_template, i, geo->ws_opt);
 960
 961         return 0;
 962
 963 fail_free_emeta:
 964         while (--i >= 0) {
 965                 if (l_mg->emeta_alloc_type == PBLK_VMALLOC_META)
 966                         vfree(l_mg->eline_meta[i]->buf);
 967                 else
 968                         kfree(l_mg->eline_meta[i]->buf);
 969                 kfree(l_mg->eline_meta[i]);
 970         }
 971 fail_free_smeta:
 972         for (i = 0; i < PBLK_DATA_LINES; i++)
 973                 kfree(l_mg->sline_meta[i]);
 974         kfree(l_mg->bb_aux);
 975 fail_free_bb_template:
 976         kfree(l_mg->bb_template);
 977 fail_free_vsc_list:
 978         kfree(l_mg->vsc_list);
 979 fail:
 980         return -ENOMEM;
 981 }
 982
 983 static int pblk_line_meta_init(struct pblk *pblk)
 984 {
 985         struct nvm_tgt_dev *dev = pblk->dev;
 986         struct nvm_geo *geo = &dev->geo;
 987         struct pblk_line_meta *lm = &pblk->lm;
 988         unsigned int smeta_len, emeta_len;
 989         int i;
 990
 991         lm->sec_per_line = geo->clba * geo->all_luns;
 992         lm->blk_per_line = geo->all_luns;
 993         lm->blk_bitmap_len = BITS_TO_LONGS(geo->all_luns) * sizeof(long);
 994         lm->sec_bitmap_len = BITS_TO_LONGS(lm->sec_per_line) * sizeof(long);
 995         lm->lun_bitmap_len = BITS_TO_LONGS(geo->all_luns) * sizeof(long);
 996         lm->mid_thrs = lm->sec_per_line / 2;
 997         lm->high_thrs = lm->sec_per_line / 4;
 998         lm->meta_distance = (geo->all_luns / 2) * pblk->min_write_pgs;
 999
1000         /* Calculate necessary pages for smeta. See comment over struct
1001          * line_smeta definition
1002          */
1003         i = 1;
1004 add_smeta_page:
1005         lm->smeta_sec = i * geo->ws_opt;
1006         lm->smeta_len = lm->smeta_sec * geo->csecs;
1007
1008         smeta_len = sizeof(struct line_smeta) + lm->lun_bitmap_len;
1009         if (smeta_len > lm->smeta_len) {
1010                 i++;
1011                 goto add_smeta_page;
1012         }
1013
1014         /* Calculate necessary pages for emeta. See comment over struct
1015          * line_emeta definition
1016          */
1017         i = 1;
1018 add_emeta_page:
1019         lm->emeta_sec[0] = i * geo->ws_opt;
1020         lm->emeta_len[0] = lm->emeta_sec[0] * geo->csecs;
1021
1022         emeta_len = calc_emeta_len(pblk);
1023         if (emeta_len > lm->emeta_len[0]) {
1024                 i++;
1025                 goto add_emeta_page;
1026         }
1027
1028         lm->emeta_bb = geo->all_luns > i ? geo->all_luns - i : 0;
1029
1030         lm->min_blk_line = 1;
1031         if (geo->all_luns > 1)
1032                 lm->min_blk_line += DIV_ROUND_UP(lm->smeta_sec +
1033                                         lm->emeta_sec[0], geo->clba);
1034
1035         if (lm->min_blk_line > lm->blk_per_line) {
1036                 pblk_err(pblk, "config. not supported. Min. LUN in line:%d\n",
1037                                                         lm->blk_per_line);
1038                 return -EINVAL;
1039         }
1040
1041         return 0;
1042 }
1043
1044 static int pblk_lines_init(struct pblk *pblk)
1045 {
1046         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1047         struct pblk_line *line;
1048         void *chunk_meta;
1049         long nr_free_chks = 0;
1050         int i, ret;
1051
1052         ret = pblk_line_meta_init(pblk);
1053         if (ret)
1054                 return ret;
1055
1056         ret = pblk_line_mg_init(pblk);
1057         if (ret)
1058                 return ret;
1059
1060         ret = pblk_luns_init(pblk);
1061         if (ret)
1062                 goto fail_free_meta;
1063
1064         chunk_meta = pblk_chunk_get_meta(pblk);
1065         if (IS_ERR(chunk_meta)) {
1066                 ret = PTR_ERR(chunk_meta);
1067                 goto fail_free_luns;
1068         }
1069
1070         pblk->lines = kcalloc(l_mg->nr_lines, sizeof(struct pblk_line),
1071                                                                 GFP_KERNEL);
1072         if (!pblk->lines) {
1073                 ret = -ENOMEM;
1074                 goto fail_free_chunk_meta;
1075         }
1076
1077         for (i = 0; i < l_mg->nr_lines; i++) {
1078                 line = &pblk->lines[i];
1079
1080                 ret = pblk_alloc_line_meta(pblk, line);
1081                 if (ret)
1082                         goto fail_free_lines;
1083
1084                 nr_free_chks += pblk_setup_line_meta(pblk, line, chunk_meta, i);
1085         }
1086
1087         if (!nr_free_chks) {
1088                 pblk_err(pblk, "too many bad blocks prevent for sane instance\n");
1089                 ret = -EINTR;
1090                 goto fail_free_lines;
1091         }
1092
1093         pblk_set_provision(pblk, nr_free_chks);
1094
1095         kfree(chunk_meta);
1096         return 0;
1097
1098 fail_free_lines:
1099         while (--i >= 0)
1100                 pblk_line_meta_free(l_mg, &pblk->lines[i]);
1101         kfree(pblk->lines);
1102 fail_free_chunk_meta:
1103         kfree(chunk_meta);
1104 fail_free_luns:
1105         kfree(pblk->luns);
1106 fail_free_meta:
1107         pblk_line_mg_free(pblk);
1108
1109         return ret;
1110 }
1111
1112 static int pblk_writer_init(struct pblk *pblk)
1113 {
1114         pblk->writer_ts = kthread_create(pblk_write_ts, pblk, "pblk-writer-t");
1115         if (IS_ERR(pblk->writer_ts)) {
1116                 int err = PTR_ERR(pblk->writer_ts);
1117
1118                 if (err != -EINTR)
1119                         pblk_err(pblk, "could not allocate writer kthread (%d)\n",
1120                                         err);
1121                 return err;
1122         }
1123
1124         timer_setup(&pblk->wtimer, pblk_write_timer_fn, 0);
1125         mod_timer(&pblk->wtimer, jiffies + msecs_to_jiffies(100));
1126
1127         return 0;
1128 }
1129
1130 static void pblk_writer_stop(struct pblk *pblk)
1131 {
1132         /* The pipeline must be stopped and the write buffer emptied before the
1133          * write thread is stopped
1134          */
1135         WARN(pblk_rb_read_count(&pblk->rwb),
1136                         "Stopping not fully persisted write buffer\n");
1137
1138         WARN(pblk_rb_sync_count(&pblk->rwb),
1139                         "Stopping not fully synced write buffer\n");
1140
1141         del_timer_sync(&pblk->wtimer);
1142         if (pblk->writer_ts)
1143                 kthread_stop(pblk->writer_ts);
1144 }
1145
1146 static void pblk_free(struct pblk *pblk)
1147 {
1148         pblk_lines_free(pblk);
1149         pblk_l2p_free(pblk);
1150         pblk_rwb_free(pblk);
1151         pblk_core_free(pblk);
1152
1153         kfree(pblk);
1154 }
1155
1156 static void pblk_tear_down(struct pblk *pblk, bool graceful)
1157 {
1158         if (graceful)
1159                 __pblk_pipeline_flush(pblk);
1160         __pblk_pipeline_stop(pblk);
1161         pblk_writer_stop(pblk);
1162         pblk_rb_sync_l2p(&pblk->rwb);
1163         pblk_rl_free(&pblk->rl);
1164
1165         pblk_debug(pblk, "consistent tear down (graceful:%d)\n", graceful);
1166 }
1167
1168 static void pblk_exit(void *private, bool graceful)
1169 {
1170         struct pblk *pblk = private;
1171
1172         down_write(&pblk_lock);
1173         pblk_gc_exit(pblk, graceful);
1174         pblk_tear_down(pblk, graceful);
1175
1176 #ifdef CONFIG_NVM_PBLK_DEBUG
1177         pblk_info(pblk, "exit: L2P CRC: %x\n", pblk_l2p_crc(pblk));
1178 #endif
1179
1180         pblk_free(pblk);
1181         up_write(&pblk_lock);
1182 }
1183
1184 static sector_t pblk_capacity(void *private)
1185 {
1186         struct pblk *pblk = private;
1187
1188         return pblk->capacity * NR_PHY_IN_LOG;
1189 }
1190
1191 static void *pblk_init(struct nvm_tgt_dev *dev, struct gendisk *tdisk,
1192                        int flags)
1193 {
1194         struct nvm_geo *geo = &dev->geo;
1195         struct request_queue *bqueue = dev->q;
1196         struct request_queue *tqueue = tdisk->queue;
1197         struct pblk *pblk;
1198         int ret;
1199
1200         pblk = kzalloc(sizeof(struct pblk), GFP_KERNEL);
1201         if (!pblk)
1202                 return ERR_PTR(-ENOMEM);
1203
1204         pblk->dev = dev;
1205         pblk->disk = tdisk;
1206         pblk->state = PBLK_STATE_RUNNING;
1207         pblk->gc.gc_enabled = 0;
1208
1209         if (!(geo->version == NVM_OCSSD_SPEC_12 ||
1210                                         geo->version == NVM_OCSSD_SPEC_20)) {
1211                 pblk_err(pblk, "OCSSD version not supported (%u)\n",
1212                                                         geo->version);
1213                 kfree(pblk);
1214                 return ERR_PTR(-EINVAL);
1215         }
1216
1217         if (geo->version == NVM_OCSSD_SPEC_12 && geo->dom & NVM_RSP_L2P) {
1218                 pblk_err(pblk, "host-side L2P table not supported. (%x)\n",
1219                                                         geo->dom);
1220                 kfree(pblk);
1221                 return ERR_PTR(-EINVAL);
1222         }
1223
1224         spin_lock_init(&pblk->resubmit_lock);
1225         spin_lock_init(&pblk->trans_lock);
1226         spin_lock_init(&pblk->lock);
1227
1228 #ifdef CONFIG_NVM_PBLK_DEBUG
1229         atomic_long_set(&pblk->inflight_writes, 0);
1230         atomic_long_set(&pblk->padded_writes, 0);
1231         atomic_long_set(&pblk->padded_wb, 0);
1232         atomic_long_set(&pblk->req_writes, 0);
1233         atomic_long_set(&pblk->sub_writes, 0);
1234         atomic_long_set(&pblk->sync_writes, 0);
1235         atomic_long_set(&pblk->inflight_reads, 0);
1236         atomic_long_set(&pblk->cache_reads, 0);
1237         atomic_long_set(&pblk->sync_reads, 0);
1238         atomic_long_set(&pblk->recov_writes, 0);
1239         atomic_long_set(&pblk->recov_writes, 0);
1240         atomic_long_set(&pblk->recov_gc_writes, 0);
1241         atomic_long_set(&pblk->recov_gc_reads, 0);
1242 #endif
1243
1244         atomic_long_set(&pblk->read_failed, 0);
1245         atomic_long_set(&pblk->read_empty, 0);
1246         atomic_long_set(&pblk->read_high_ecc, 0);
1247         atomic_long_set(&pblk->read_failed_gc, 0);
1248         atomic_long_set(&pblk->write_failed, 0);
1249         atomic_long_set(&pblk->erase_failed, 0);
1250
1251         ret = pblk_core_init(pblk);
1252         if (ret) {
1253                 pblk_err(pblk, "could not initialize core\n");
1254                 goto fail;
1255         }
1256
1257         ret = pblk_lines_init(pblk);
1258         if (ret) {
1259                 pblk_err(pblk, "could not initialize lines\n");
1260                 goto fail_free_core;
1261         }
1262
1263         ret = pblk_rwb_init(pblk);
1264         if (ret) {
1265                 pblk_err(pblk, "could not initialize write buffer\n");
1266                 goto fail_free_lines;
1267         }
1268
1269         ret = pblk_l2p_init(pblk, flags & NVM_TARGET_FACTORY);
1270         if (ret) {
1271                 pblk_err(pblk, "could not initialize maps\n");
1272                 goto fail_free_rwb;
1273         }
1274
1275         ret = pblk_writer_init(pblk);
1276         if (ret) {
1277                 if (ret != -EINTR)
1278                         pblk_err(pblk, "could not initialize write thread\n");
1279                 goto fail_free_l2p;
1280         }
1281
1282         ret = pblk_gc_init(pblk);
1283         if (ret) {
1284                 pblk_err(pblk, "could not initialize gc\n");
1285                 goto fail_stop_writer;
1286         }
1287
1288         /* inherit the size from the underlying device */
1289         blk_queue_logical_block_size(tqueue, queue_physical_block_size(bqueue));
1290         blk_queue_max_hw_sectors(tqueue, queue_max_hw_sectors(bqueue));
1291
1292         blk_queue_write_cache(tqueue, true, false);
1293
1294         tqueue->limits.discard_granularity = geo->clba * geo->csecs;
1295         tqueue->limits.discard_alignment = 0;
1296         blk_queue_max_discard_sectors(tqueue, UINT_MAX >> 9);
1297         blk_queue_flag_set(QUEUE_FLAG_DISCARD, tqueue);
1298
1299         pblk_info(pblk, "luns:%u, lines:%d, secs:%llu, buf entries:%u\n",
1300                         geo->all_luns, pblk->l_mg.nr_lines,
1301                         (unsigned long long)pblk->rl.nr_secs,
1302                         pblk->rwb.nr_entries);
1303
1304         wake_up_process(pblk->writer_ts);
1305
1306         /* Check if we need to start GC */
1307         pblk_gc_should_kick(pblk);
1308
1309         return pblk;
1310
1311 fail_stop_writer:
1312         pblk_writer_stop(pblk);
1313 fail_free_l2p:
1314         pblk_l2p_free(pblk);
1315 fail_free_rwb:
1316         pblk_rwb_free(pblk);
1317 fail_free_lines:
1318         pblk_lines_free(pblk);
1319 fail_free_core:
1320         pblk_core_free(pblk);
1321 fail:
1322         kfree(pblk);
1323         return ERR_PTR(ret);
1324 }
1325
1326 /* physical block device target */
1327 static struct nvm_tgt_type tt_pblk = {
1328         .name           = "pblk",
1329         .version        = {1, 0, 0},
1330
1331         .make_rq        = pblk_make_rq,
1332         .capacity       = pblk_capacity,
1333
1334         .init           = pblk_init,
1335         .exit           = pblk_exit,
1336
1337         .sysfs_init     = pblk_sysfs_init,
1338         .sysfs_exit     = pblk_sysfs_exit,
1339         .owner          = THIS_MODULE,
1340 };
1341
1342 static int __init pblk_module_init(void)
1343 {
1344         int ret;
1345
1346         ret = bioset_init(&pblk_bio_set, BIO_POOL_SIZE, 0, 0);
1347         if (ret)
1348                 return ret;
1349         ret = nvm_register_tgt_type(&tt_pblk);
1350         if (ret)
1351                 bioset_exit(&pblk_bio_set);
1352         return ret;
1353 }
1354
1355 static void pblk_module_exit(void)
1356 {
1357         bioset_exit(&pblk_bio_set);
1358         nvm_unregister_tgt_type(&tt_pblk);
1359 }
1360
1361 module_init(pblk_module_init);
1362 module_exit(pblk_module_exit);
1363 MODULE_AUTHOR("Javier Gonzalez <javier@cnexlabs.com>");
1364 MODULE_AUTHOR("Matias Bjorling <matias@cnexlabs.com>");
1365 MODULE_LICENSE("GPL v2");
1366 MODULE_DESCRIPTION("Physical Block-Device for Open-Channel SSDs");