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 struct kmem_cache *pblk_ws_cache, *pblk_rec_cache, *pblk_g_rq_cache,
  24                                 *pblk_w_rq_cache;
  25 static DECLARE_RWSEM(pblk_lock);
  26 struct bio_set *pblk_bio_set;
  27
  28 static int pblk_rw_io(struct request_queue *q, struct pblk *pblk,
  29                           struct bio *bio)
  30 {
  31         int ret;
  32
  33         /* Read requests must be <= 256kb due to NVMe's 64 bit completion bitmap
  34          * constraint. Writes can be of arbitrary size.
  35          */
  36         if (bio_data_dir(bio) == READ) {
  37                 blk_queue_split(q, &bio);
  38                 ret = pblk_submit_read(pblk, bio);
  39                 if (ret == NVM_IO_DONE && bio_flagged(bio, BIO_CLONED))
  40                         bio_put(bio);
  41
  42                 return ret;
  43         }
  44
  45         /* Prevent deadlock in the case of a modest LUN configuration and large
  46          * user I/Os. Unless stalled, the rate limiter leaves at least 256KB
  47          * available for user I/O.
  48          */
  49         if (pblk_get_secs(bio) > pblk_rl_max_io(&pblk->rl))
  50                 blk_queue_split(q, &bio);
  51
  52         return pblk_write_to_cache(pblk, bio, PBLK_IOTYPE_USER);
  53 }
  54
  55 static blk_qc_t pblk_make_rq(struct request_queue *q, struct bio *bio)
  56 {
  57         struct pblk *pblk = q->queuedata;
  58
  59         if (bio_op(bio) == REQ_OP_DISCARD) {
  60                 pblk_discard(pblk, bio);
  61                 if (!(bio->bi_opf & REQ_PREFLUSH)) {
  62                         bio_endio(bio);
  63                         return BLK_QC_T_NONE;
  64                 }
  65         }
  66
  67         switch (pblk_rw_io(q, pblk, bio)) {
  68         case NVM_IO_ERR:
  69                 bio_io_error(bio);
  70                 break;
  71         case NVM_IO_DONE:
  72                 bio_endio(bio);
  73                 break;
  74         }
  75
  76         return BLK_QC_T_NONE;
  77 }
  78
  79 static size_t pblk_trans_map_size(struct pblk *pblk)
  80 {
  81         int entry_size = 8;
  82
  83         if (pblk->ppaf_bitsize < 32)
  84                 entry_size = 4;
  85
  86         return entry_size * pblk->rl.nr_secs;
  87 }
  88
  89 #ifdef CONFIG_NVM_DEBUG
  90 static u32 pblk_l2p_crc(struct pblk *pblk)
  91 {
  92         size_t map_size;
  93         u32 crc = ~(u32)0;
  94
  95         map_size = pblk_trans_map_size(pblk);
  96         crc = crc32_le(crc, pblk->trans_map, map_size);
  97         return crc;
  98 }
  99 #endif
 100
 101 static void pblk_l2p_free(struct pblk *pblk)
 102 {
 103         vfree(pblk->trans_map);
 104 }
 105
 106 static int pblk_l2p_init(struct pblk *pblk)
 107 {
 108         sector_t i;
 109         struct ppa_addr ppa;
 110         size_t map_size;
 111
 112         map_size = pblk_trans_map_size(pblk);
 113         pblk->trans_map = vmalloc(map_size);
 114         if (!pblk->trans_map)
 115                 return -ENOMEM;
 116
 117         pblk_ppa_set_empty(&ppa);
 118
 119         for (i = 0; i < pblk->rl.nr_secs; i++)
 120                 pblk_trans_map_set(pblk, i, ppa);
 121
 122         return 0;
 123 }
 124
 125 static void pblk_rwb_free(struct pblk *pblk)
 126 {
 127         if (pblk_rb_tear_down_check(&pblk->rwb))
 128                 pr_err("pblk: write buffer error on tear down\n");
 129
 130         pblk_rb_data_free(&pblk->rwb);
 131         vfree(pblk_rb_entries_ref(&pblk->rwb));
 132 }
 133
 134 static int pblk_rwb_init(struct pblk *pblk)
 135 {
 136         struct nvm_tgt_dev *dev = pblk->dev;
 137         struct nvm_geo *geo = &dev->geo;
 138         struct pblk_rb_entry *entries;
 139         unsigned long nr_entries;
 140         unsigned int power_size, power_seg_sz;
 141
 142         nr_entries = pblk_rb_calculate_size(pblk->pgs_in_buffer);
 143
 144         entries = vzalloc(nr_entries * sizeof(struct pblk_rb_entry));
 145         if (!entries)
 146                 return -ENOMEM;
 147
 148         power_size = get_count_order(nr_entries);
 149         power_seg_sz = get_count_order(geo->sec_size);
 150
 151         return pblk_rb_init(&pblk->rwb, entries, power_size, power_seg_sz);
 152 }
 153
 154 /* Minimum pages needed within a lun */
 155 #define ADDR_POOL_SIZE 64
 156
 157 static int pblk_set_ppaf(struct pblk *pblk)
 158 {
 159         struct nvm_tgt_dev *dev = pblk->dev;
 160         struct nvm_geo *geo = &dev->geo;
 161         struct nvm_addr_format ppaf = geo->ppaf;
 162         int power_len;
 163
 164         /* Re-calculate channel and lun format to adapt to configuration */
 165         power_len = get_count_order(geo->nr_chnls);
 166         if (1 << power_len != geo->nr_chnls) {
 167                 pr_err("pblk: supports only power-of-two channel config.\n");
 168                 return -EINVAL;
 169         }
 170         ppaf.ch_len = power_len;
 171
 172         power_len = get_count_order(geo->nr_luns);
 173         if (1 << power_len != geo->nr_luns) {
 174                 pr_err("pblk: supports only power-of-two LUN config.\n");
 175                 return -EINVAL;
 176         }
 177         ppaf.lun_len = power_len;
 178
 179         pblk->ppaf.sec_offset = 0;
 180         pblk->ppaf.pln_offset = ppaf.sect_len;
 181         pblk->ppaf.ch_offset = pblk->ppaf.pln_offset + ppaf.pln_len;
 182         pblk->ppaf.lun_offset = pblk->ppaf.ch_offset + ppaf.ch_len;
 183         pblk->ppaf.pg_offset = pblk->ppaf.lun_offset + ppaf.lun_len;
 184         pblk->ppaf.blk_offset = pblk->ppaf.pg_offset + ppaf.pg_len;
 185         pblk->ppaf.sec_mask = (1ULL << ppaf.sect_len) - 1;
 186         pblk->ppaf.pln_mask = ((1ULL << ppaf.pln_len) - 1) <<
 187                                                         pblk->ppaf.pln_offset;
 188         pblk->ppaf.ch_mask = ((1ULL << ppaf.ch_len) - 1) <<
 189                                                         pblk->ppaf.ch_offset;
 190         pblk->ppaf.lun_mask = ((1ULL << ppaf.lun_len) - 1) <<
 191                                                         pblk->ppaf.lun_offset;
 192         pblk->ppaf.pg_mask = ((1ULL << ppaf.pg_len) - 1) <<
 193                                                         pblk->ppaf.pg_offset;
 194         pblk->ppaf.blk_mask = ((1ULL << ppaf.blk_len) - 1) <<
 195                                                         pblk->ppaf.blk_offset;
 196
 197         pblk->ppaf_bitsize = pblk->ppaf.blk_offset + ppaf.blk_len;
 198
 199         return 0;
 200 }
 201
 202 static int pblk_init_global_caches(struct pblk *pblk)
 203 {
 204         down_write(&pblk_lock);
 205         pblk_ws_cache = kmem_cache_create("pblk_blk_ws",
 206                                 sizeof(struct pblk_line_ws), 0, 0, NULL);
 207         if (!pblk_ws_cache) {
 208                 up_write(&pblk_lock);
 209                 return -ENOMEM;
 210         }
 211
 212         pblk_rec_cache = kmem_cache_create("pblk_rec",
 213                                 sizeof(struct pblk_rec_ctx), 0, 0, NULL);
 214         if (!pblk_rec_cache) {
 215                 kmem_cache_destroy(pblk_ws_cache);
 216                 up_write(&pblk_lock);
 217                 return -ENOMEM;
 218         }
 219
 220         pblk_g_rq_cache = kmem_cache_create("pblk_g_rq", pblk_g_rq_size,
 221                                 0, 0, NULL);
 222         if (!pblk_g_rq_cache) {
 223                 kmem_cache_destroy(pblk_ws_cache);
 224                 kmem_cache_destroy(pblk_rec_cache);
 225                 up_write(&pblk_lock);
 226                 return -ENOMEM;
 227         }
 228
 229         pblk_w_rq_cache = kmem_cache_create("pblk_w_rq", pblk_w_rq_size,
 230                                 0, 0, NULL);
 231         if (!pblk_w_rq_cache) {
 232                 kmem_cache_destroy(pblk_ws_cache);
 233                 kmem_cache_destroy(pblk_rec_cache);
 234                 kmem_cache_destroy(pblk_g_rq_cache);
 235                 up_write(&pblk_lock);
 236                 return -ENOMEM;
 237         }
 238         up_write(&pblk_lock);
 239
 240         return 0;
 241 }
 242
 243 static void pblk_free_global_caches(struct pblk *pblk)
 244 {
 245         kmem_cache_destroy(pblk_ws_cache);
 246         kmem_cache_destroy(pblk_rec_cache);
 247         kmem_cache_destroy(pblk_g_rq_cache);
 248         kmem_cache_destroy(pblk_w_rq_cache);
 249 }
 250
 251 static int pblk_core_init(struct pblk *pblk)
 252 {
 253         struct nvm_tgt_dev *dev = pblk->dev;
 254         struct nvm_geo *geo = &dev->geo;
 255
 256         pblk->pgs_in_buffer = NVM_MEM_PAGE_WRITE * geo->sec_per_pg *
 257                                                 geo->nr_planes * geo->all_luns;
 258
 259         if (pblk_init_global_caches(pblk))
 260                 return -ENOMEM;
 261
 262         /* Internal bios can be at most the sectors signaled by the device. */
 263         pblk->page_bio_pool = mempool_create_page_pool(nvm_max_phys_sects(dev),
 264                                                                         0);
 265         if (!pblk->page_bio_pool)
 266                 goto free_global_caches;
 267
 268         pblk->gen_ws_pool = mempool_create_slab_pool(PBLK_GEN_WS_POOL_SIZE,
 269                                                         pblk_ws_cache);
 270         if (!pblk->gen_ws_pool)
 271                 goto free_page_bio_pool;
 272
 273         pblk->rec_pool = mempool_create_slab_pool(geo->all_luns,
 274                                                         pblk_rec_cache);
 275         if (!pblk->rec_pool)
 276                 goto free_gen_ws_pool;
 277
 278         pblk->r_rq_pool = mempool_create_slab_pool(geo->all_luns,
 279                                                         pblk_g_rq_cache);
 280         if (!pblk->r_rq_pool)
 281                 goto free_rec_pool;
 282
 283         pblk->e_rq_pool = mempool_create_slab_pool(geo->all_luns,
 284                                                         pblk_g_rq_cache);
 285         if (!pblk->e_rq_pool)
 286                 goto free_r_rq_pool;
 287
 288         pblk->w_rq_pool = mempool_create_slab_pool(geo->all_luns,
 289                                                         pblk_w_rq_cache);
 290         if (!pblk->w_rq_pool)
 291                 goto free_e_rq_pool;
 292
 293         pblk->close_wq = alloc_workqueue("pblk-close-wq",
 294                         WQ_MEM_RECLAIM | WQ_UNBOUND, PBLK_NR_CLOSE_JOBS);
 295         if (!pblk->close_wq)
 296                 goto free_w_rq_pool;
 297
 298         pblk->bb_wq = alloc_workqueue("pblk-bb-wq",
 299                         WQ_MEM_RECLAIM | WQ_UNBOUND, 0);
 300         if (!pblk->bb_wq)
 301                 goto free_close_wq;
 302
 303         pblk->r_end_wq = alloc_workqueue("pblk-read-end-wq",
 304                         WQ_MEM_RECLAIM | WQ_UNBOUND, 0);
 305         if (!pblk->r_end_wq)
 306                 goto free_bb_wq;
 307
 308         if (pblk_set_ppaf(pblk))
 309                 goto free_r_end_wq;
 310
 311         if (pblk_rwb_init(pblk))
 312                 goto free_r_end_wq;
 313
 314         INIT_LIST_HEAD(&pblk->compl_list);
 315         return 0;
 316
 317 free_r_end_wq:
 318         destroy_workqueue(pblk->r_end_wq);
 319 free_bb_wq:
 320         destroy_workqueue(pblk->bb_wq);
 321 free_close_wq:
 322         destroy_workqueue(pblk->close_wq);
 323 free_w_rq_pool:
 324         mempool_destroy(pblk->w_rq_pool);
 325 free_e_rq_pool:
 326         mempool_destroy(pblk->e_rq_pool);
 327 free_r_rq_pool:
 328         mempool_destroy(pblk->r_rq_pool);
 329 free_rec_pool:
 330         mempool_destroy(pblk->rec_pool);
 331 free_gen_ws_pool:
 332         mempool_destroy(pblk->gen_ws_pool);
 333 free_page_bio_pool:
 334         mempool_destroy(pblk->page_bio_pool);
 335 free_global_caches:
 336         pblk_free_global_caches(pblk);
 337         return -ENOMEM;
 338 }
 339
 340 static void pblk_core_free(struct pblk *pblk)
 341 {
 342         if (pblk->close_wq)
 343                 destroy_workqueue(pblk->close_wq);
 344
 345         if (pblk->r_end_wq)
 346                 destroy_workqueue(pblk->r_end_wq);
 347
 348         if (pblk->bb_wq)
 349                 destroy_workqueue(pblk->bb_wq);
 350
 351         mempool_destroy(pblk->page_bio_pool);
 352         mempool_destroy(pblk->gen_ws_pool);
 353         mempool_destroy(pblk->rec_pool);
 354         mempool_destroy(pblk->r_rq_pool);
 355         mempool_destroy(pblk->e_rq_pool);
 356         mempool_destroy(pblk->w_rq_pool);
 357
 358         pblk_free_global_caches(pblk);
 359 }
 360
 361 static void pblk_luns_free(struct pblk *pblk)
 362 {
 363         kfree(pblk->luns);
 364 }
 365
 366 static void pblk_free_line_bitmaps(struct pblk_line *line)
 367 {
 368         kfree(line->blk_bitmap);
 369         kfree(line->erase_bitmap);
 370 }
 371
 372 static void pblk_lines_free(struct pblk *pblk)
 373 {
 374         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
 375         struct pblk_line *line;
 376         int i;
 377
 378         spin_lock(&l_mg->free_lock);
 379         for (i = 0; i < l_mg->nr_lines; i++) {
 380                 line = &pblk->lines[i];
 381
 382                 pblk_line_free(pblk, line);
 383                 pblk_free_line_bitmaps(line);
 384         }
 385         spin_unlock(&l_mg->free_lock);
 386 }
 387
 388 static void pblk_line_meta_free(struct pblk *pblk)
 389 {
 390         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
 391         int i;
 392
 393         kfree(l_mg->bb_template);
 394         kfree(l_mg->bb_aux);
 395         kfree(l_mg->vsc_list);
 396
 397         for (i = 0; i < PBLK_DATA_LINES; i++) {
 398                 kfree(l_mg->sline_meta[i]);
 399                 pblk_mfree(l_mg->eline_meta[i]->buf, l_mg->emeta_alloc_type);
 400                 kfree(l_mg->eline_meta[i]);
 401         }
 402
 403         kfree(pblk->lines);
 404 }
 405
 406 static int pblk_bb_discovery(struct nvm_tgt_dev *dev, struct pblk_lun *rlun)
 407 {
 408         struct nvm_geo *geo = &dev->geo;
 409         struct ppa_addr ppa;
 410         u8 *blks;
 411         int nr_blks, ret;
 412
 413         nr_blks = geo->nr_chks * geo->plane_mode;
 414         blks = kmalloc(nr_blks, GFP_KERNEL);
 415         if (!blks)
 416                 return -ENOMEM;
 417
 418         ppa.ppa = 0;
 419         ppa.g.ch = rlun->bppa.g.ch;
 420         ppa.g.lun = rlun->bppa.g.lun;
 421
 422         ret = nvm_get_tgt_bb_tbl(dev, ppa, blks);
 423         if (ret)
 424                 goto out;
 425
 426         nr_blks = nvm_bb_tbl_fold(dev->parent, blks, nr_blks);
 427         if (nr_blks < 0) {
 428                 ret = nr_blks;
 429                 goto out;
 430         }
 431
 432         rlun->bb_list = blks;
 433
 434         return 0;
 435 out:
 436         kfree(blks);
 437         return ret;
 438 }
 439
 440 static int pblk_bb_line(struct pblk *pblk, struct pblk_line *line,
 441                         int blk_per_line)
 442 {
 443         struct nvm_tgt_dev *dev = pblk->dev;
 444         struct nvm_geo *geo = &dev->geo;
 445         struct pblk_lun *rlun;
 446         int bb_cnt = 0;
 447         int i;
 448
 449         for (i = 0; i < blk_per_line; i++) {
 450                 rlun = &pblk->luns[i];
 451                 if (rlun->bb_list[line->id] == NVM_BLK_T_FREE)
 452                         continue;
 453
 454                 set_bit(pblk_ppa_to_pos(geo, rlun->bppa), line->blk_bitmap);
 455                 bb_cnt++;
 456         }
 457
 458         return bb_cnt;
 459 }
 460
 461 static int pblk_alloc_line_bitmaps(struct pblk *pblk, struct pblk_line *line)
 462 {
 463         struct pblk_line_meta *lm = &pblk->lm;
 464
 465         line->blk_bitmap = kzalloc(lm->blk_bitmap_len, GFP_KERNEL);
 466         if (!line->blk_bitmap)
 467                 return -ENOMEM;
 468
 469         line->erase_bitmap = kzalloc(lm->blk_bitmap_len, GFP_KERNEL);
 470         if (!line->erase_bitmap) {
 471                 kfree(line->blk_bitmap);
 472                 return -ENOMEM;
 473         }
 474
 475         return 0;
 476 }
 477
 478 static int pblk_luns_init(struct pblk *pblk, struct ppa_addr *luns)
 479 {
 480         struct nvm_tgt_dev *dev = pblk->dev;
 481         struct nvm_geo *geo = &dev->geo;
 482         struct pblk_lun *rlun;
 483         int i, ret;
 484
 485         /* TODO: Implement unbalanced LUN support */
 486         if (geo->nr_luns < 0) {
 487                 pr_err("pblk: unbalanced LUN config.\n");
 488                 return -EINVAL;
 489         }
 490
 491         pblk->luns = kcalloc(geo->all_luns, sizeof(struct pblk_lun),
 492                                                                 GFP_KERNEL);
 493         if (!pblk->luns)
 494                 return -ENOMEM;
 495
 496         for (i = 0; i < geo->all_luns; i++) {
 497                 /* Stripe across channels */
 498                 int ch = i % geo->nr_chnls;
 499                 int lun_raw = i / geo->nr_chnls;
 500                 int lunid = lun_raw + ch * geo->nr_luns;
 501
 502                 rlun = &pblk->luns[i];
 503                 rlun->bppa = luns[lunid];
 504
 505                 sema_init(&rlun->wr_sem, 1);
 506
 507                 ret = pblk_bb_discovery(dev, rlun);
 508                 if (ret) {
 509                         while (--i >= 0)
 510                                 kfree(pblk->luns[i].bb_list);
 511                         return ret;
 512                 }
 513         }
 514
 515         return 0;
 516 }
 517
 518 static int pblk_lines_configure(struct pblk *pblk, int flags)
 519 {
 520         struct pblk_line *line = NULL;
 521         int ret = 0;
 522
 523         if (!(flags & NVM_TARGET_FACTORY)) {
 524                 line = pblk_recov_l2p(pblk);
 525                 if (IS_ERR(line)) {
 526                         pr_err("pblk: could not recover l2p table\n");
 527                         ret = -EFAULT;
 528                 }
 529         }
 530
 531 #ifdef CONFIG_NVM_DEBUG
 532         pr_info("pblk init: L2P CRC: %x\n", pblk_l2p_crc(pblk));
 533 #endif
 534
 535         /* Free full lines directly as GC has not been started yet */
 536         pblk_gc_free_full_lines(pblk);
 537
 538         if (!line) {
 539                 /* Configure next line for user data */
 540                 line = pblk_line_get_first_data(pblk);
 541                 if (!line) {
 542                         pr_err("pblk: line list corrupted\n");
 543                         ret = -EFAULT;
 544                 }
 545         }
 546
 547         return ret;
 548 }
 549
 550 /* See comment over struct line_emeta definition */
 551 static unsigned int calc_emeta_len(struct pblk *pblk)
 552 {
 553         struct pblk_line_meta *lm = &pblk->lm;
 554         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
 555         struct nvm_tgt_dev *dev = pblk->dev;
 556         struct nvm_geo *geo = &dev->geo;
 557
 558         /* Round to sector size so that lba_list starts on its own sector */
 559         lm->emeta_sec[1] = DIV_ROUND_UP(
 560                         sizeof(struct line_emeta) + lm->blk_bitmap_len,
 561                         geo->sec_size);
 562         lm->emeta_len[1] = lm->emeta_sec[1] * geo->sec_size;
 563
 564         /* Round to sector size so that vsc_list starts on its own sector */
 565         lm->dsec_per_line = lm->sec_per_line - lm->emeta_sec[0];
 566         lm->emeta_sec[2] = DIV_ROUND_UP(lm->dsec_per_line * sizeof(u64),
 567                         geo->sec_size);
 568         lm->emeta_len[2] = lm->emeta_sec[2] * geo->sec_size;
 569
 570         lm->emeta_sec[3] = DIV_ROUND_UP(l_mg->nr_lines * sizeof(u32),
 571                         geo->sec_size);
 572         lm->emeta_len[3] = lm->emeta_sec[3] * geo->sec_size;
 573
 574         lm->vsc_list_len = l_mg->nr_lines * sizeof(u32);
 575
 576         return (lm->emeta_len[1] + lm->emeta_len[2] + lm->emeta_len[3]);
 577 }
 578
 579 static void pblk_set_provision(struct pblk *pblk, long nr_free_blks)
 580 {
 581         struct nvm_tgt_dev *dev = pblk->dev;
 582         struct nvm_geo *geo = &dev->geo;
 583         sector_t provisioned;
 584
 585         pblk->over_pct = 20;
 586
 587         provisioned = nr_free_blks;
 588         provisioned *= (100 - pblk->over_pct);
 589         sector_div(provisioned, 100);
 590
 591         /* Internally pblk manages all free blocks, but all calculations based
 592          * on user capacity consider only provisioned blocks
 593          */
 594         pblk->rl.total_blocks = nr_free_blks;
 595         pblk->rl.nr_secs = nr_free_blks * geo->sec_per_chk;
 596         pblk->capacity = provisioned * geo->sec_per_chk;
 597         atomic_set(&pblk->rl.free_blocks, nr_free_blks);
 598 }
 599
 600 static int pblk_lines_alloc_metadata(struct pblk *pblk)
 601 {
 602         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
 603         struct pblk_line_meta *lm = &pblk->lm;
 604         int i;
 605
 606         /* smeta is always small enough to fit on a kmalloc memory allocation,
 607          * emeta depends on the number of LUNs allocated to the pblk instance
 608          */
 609         for (i = 0; i < PBLK_DATA_LINES; i++) {
 610                 l_mg->sline_meta[i] = kmalloc(lm->smeta_len, GFP_KERNEL);
 611                 if (!l_mg->sline_meta[i])
 612                         goto fail_free_smeta;
 613         }
 614
 615         /* emeta allocates three different buffers for managing metadata with
 616          * in-memory and in-media layouts
 617          */
 618         for (i = 0; i < PBLK_DATA_LINES; i++) {
 619                 struct pblk_emeta *emeta;
 620
 621                 emeta = kmalloc(sizeof(struct pblk_emeta), GFP_KERNEL);
 622                 if (!emeta)
 623                         goto fail_free_emeta;
 624
 625                 if (lm->emeta_len[0] > KMALLOC_MAX_CACHE_SIZE) {
 626                         l_mg->emeta_alloc_type = PBLK_VMALLOC_META;
 627
 628                         emeta->buf = vmalloc(lm->emeta_len[0]);
 629                         if (!emeta->buf) {
 630                                 kfree(emeta);
 631                                 goto fail_free_emeta;
 632                         }
 633
 634                         emeta->nr_entries = lm->emeta_sec[0];
 635                         l_mg->eline_meta[i] = emeta;
 636                 } else {
 637                         l_mg->emeta_alloc_type = PBLK_KMALLOC_META;
 638
 639                         emeta->buf = kmalloc(lm->emeta_len[0], GFP_KERNEL);
 640                         if (!emeta->buf) {
 641                                 kfree(emeta);
 642                                 goto fail_free_emeta;
 643                         }
 644
 645                         emeta->nr_entries = lm->emeta_sec[0];
 646                         l_mg->eline_meta[i] = emeta;
 647                 }
 648         }
 649
 650         l_mg->vsc_list = kcalloc(l_mg->nr_lines, sizeof(__le32), GFP_KERNEL);
 651         if (!l_mg->vsc_list)
 652                 goto fail_free_emeta;
 653
 654         for (i = 0; i < l_mg->nr_lines; i++)
 655                 l_mg->vsc_list[i] = cpu_to_le32(EMPTY_ENTRY);
 656
 657         return 0;
 658
 659 fail_free_emeta:
 660         while (--i >= 0) {
 661                 if (l_mg->emeta_alloc_type == PBLK_VMALLOC_META)
 662                         vfree(l_mg->eline_meta[i]->buf);
 663                 else
 664                         kfree(l_mg->eline_meta[i]->buf);
 665                 kfree(l_mg->eline_meta[i]);
 666         }
 667
 668 fail_free_smeta:
 669         for (i = 0; i < PBLK_DATA_LINES; i++)
 670                 kfree(l_mg->sline_meta[i]);
 671
 672         return -ENOMEM;
 673 }
 674
 675 static int pblk_lines_init(struct pblk *pblk)
 676 {
 677         struct nvm_tgt_dev *dev = pblk->dev;
 678         struct nvm_geo *geo = &dev->geo;
 679         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
 680         struct pblk_line_meta *lm = &pblk->lm;
 681         struct pblk_line *line;
 682         unsigned int smeta_len, emeta_len;
 683         long nr_bad_blks, nr_free_blks;
 684         int bb_distance, max_write_ppas, mod;
 685         int i, ret;
 686
 687         pblk->min_write_pgs = geo->sec_per_pl * (geo->sec_size / PAGE_SIZE);
 688         max_write_ppas = pblk->min_write_pgs * geo->all_luns;
 689         pblk->max_write_pgs = (max_write_ppas < nvm_max_phys_sects(dev)) ?
 690                                 max_write_ppas : nvm_max_phys_sects(dev);
 691         pblk_set_sec_per_write(pblk, pblk->min_write_pgs);
 692
 693         if (pblk->max_write_pgs > PBLK_MAX_REQ_ADDRS) {
 694                 pr_err("pblk: cannot support device max_phys_sect\n");
 695                 return -EINVAL;
 696         }
 697
 698         div_u64_rem(geo->sec_per_chk, pblk->min_write_pgs, &mod);
 699         if (mod) {
 700                 pr_err("pblk: bad configuration of sectors/pages\n");
 701                 return -EINVAL;
 702         }
 703
 704         l_mg->nr_lines = geo->nr_chks;
 705         l_mg->log_line = l_mg->data_line = NULL;
 706         l_mg->l_seq_nr = l_mg->d_seq_nr = 0;
 707         l_mg->nr_free_lines = 0;
 708         bitmap_zero(&l_mg->meta_bitmap, PBLK_DATA_LINES);
 709
 710         lm->sec_per_line = geo->sec_per_chk * geo->all_luns;
 711         lm->blk_per_line = geo->all_luns;
 712         lm->blk_bitmap_len = BITS_TO_LONGS(geo->all_luns) * sizeof(long);
 713         lm->sec_bitmap_len = BITS_TO_LONGS(lm->sec_per_line) * sizeof(long);
 714         lm->lun_bitmap_len = BITS_TO_LONGS(geo->all_luns) * sizeof(long);
 715         lm->mid_thrs = lm->sec_per_line / 2;
 716         lm->high_thrs = lm->sec_per_line / 4;
 717         lm->meta_distance = (geo->all_luns / 2) * pblk->min_write_pgs;
 718
 719         /* Calculate necessary pages for smeta. See comment over struct
 720          * line_smeta definition
 721          */
 722         i = 1;
 723 add_smeta_page:
 724         lm->smeta_sec = i * geo->sec_per_pl;
 725         lm->smeta_len = lm->smeta_sec * geo->sec_size;
 726
 727         smeta_len = sizeof(struct line_smeta) + lm->lun_bitmap_len;
 728         if (smeta_len > lm->smeta_len) {
 729                 i++;
 730                 goto add_smeta_page;
 731         }
 732
 733         /* Calculate necessary pages for emeta. See comment over struct
 734          * line_emeta definition
 735          */
 736         i = 1;
 737 add_emeta_page:
 738         lm->emeta_sec[0] = i * geo->sec_per_pl;
 739         lm->emeta_len[0] = lm->emeta_sec[0] * geo->sec_size;
 740
 741         emeta_len = calc_emeta_len(pblk);
 742         if (emeta_len > lm->emeta_len[0]) {
 743                 i++;
 744                 goto add_emeta_page;
 745         }
 746
 747         lm->emeta_bb = geo->all_luns > i ? geo->all_luns - i : 0;
 748
 749         lm->min_blk_line = 1;
 750         if (geo->all_luns > 1)
 751                 lm->min_blk_line += DIV_ROUND_UP(lm->smeta_sec +
 752                                         lm->emeta_sec[0], geo->sec_per_chk);
 753
 754         if (lm->min_blk_line > lm->blk_per_line) {
 755                 pr_err("pblk: config. not supported. Min. LUN in line:%d\n",
 756                                                         lm->blk_per_line);
 757                 ret = -EINVAL;
 758                 goto fail;
 759         }
 760
 761         ret = pblk_lines_alloc_metadata(pblk);
 762         if (ret)
 763                 goto fail;
 764
 765         l_mg->bb_template = kzalloc(lm->sec_bitmap_len, GFP_KERNEL);
 766         if (!l_mg->bb_template) {
 767                 ret = -ENOMEM;
 768                 goto fail_free_meta;
 769         }
 770
 771         l_mg->bb_aux = kzalloc(lm->sec_bitmap_len, GFP_KERNEL);
 772         if (!l_mg->bb_aux) {
 773                 ret = -ENOMEM;
 774                 goto fail_free_bb_template;
 775         }
 776
 777         bb_distance = (geo->all_luns) * geo->sec_per_pl;
 778         for (i = 0; i < lm->sec_per_line; i += bb_distance)
 779                 bitmap_set(l_mg->bb_template, i, geo->sec_per_pl);
 780
 781         INIT_LIST_HEAD(&l_mg->free_list);
 782         INIT_LIST_HEAD(&l_mg->corrupt_list);
 783         INIT_LIST_HEAD(&l_mg->bad_list);
 784         INIT_LIST_HEAD(&l_mg->gc_full_list);
 785         INIT_LIST_HEAD(&l_mg->gc_high_list);
 786         INIT_LIST_HEAD(&l_mg->gc_mid_list);
 787         INIT_LIST_HEAD(&l_mg->gc_low_list);
 788         INIT_LIST_HEAD(&l_mg->gc_empty_list);
 789
 790         INIT_LIST_HEAD(&l_mg->emeta_list);
 791
 792         l_mg->gc_lists[0] = &l_mg->gc_high_list;
 793         l_mg->gc_lists[1] = &l_mg->gc_mid_list;
 794         l_mg->gc_lists[2] = &l_mg->gc_low_list;
 795
 796         spin_lock_init(&l_mg->free_lock);
 797         spin_lock_init(&l_mg->close_lock);
 798         spin_lock_init(&l_mg->gc_lock);
 799
 800         pblk->lines = kcalloc(l_mg->nr_lines, sizeof(struct pblk_line),
 801                                                                 GFP_KERNEL);
 802         if (!pblk->lines) {
 803                 ret = -ENOMEM;
 804                 goto fail_free_bb_aux;
 805         }
 806
 807         nr_free_blks = 0;
 808         for (i = 0; i < l_mg->nr_lines; i++) {
 809                 int blk_in_line;
 810
 811                 line = &pblk->lines[i];
 812
 813                 line->pblk = pblk;
 814                 line->id = i;
 815                 line->type = PBLK_LINETYPE_FREE;
 816                 line->state = PBLK_LINESTATE_FREE;
 817                 line->gc_group = PBLK_LINEGC_NONE;
 818                 line->vsc = &l_mg->vsc_list[i];
 819                 spin_lock_init(&line->lock);
 820
 821                 ret = pblk_alloc_line_bitmaps(pblk, line);
 822                 if (ret)
 823                         goto fail_free_lines;
 824
 825                 nr_bad_blks = pblk_bb_line(pblk, line, lm->blk_per_line);
 826                 if (nr_bad_blks < 0 || nr_bad_blks > lm->blk_per_line) {
 827                         pblk_free_line_bitmaps(line);
 828                         ret = -EINVAL;
 829                         goto fail_free_lines;
 830                 }
 831
 832                 blk_in_line = lm->blk_per_line - nr_bad_blks;
 833                 if (blk_in_line < lm->min_blk_line) {
 834                         line->state = PBLK_LINESTATE_BAD;
 835                         list_add_tail(&line->list, &l_mg->bad_list);
 836                         continue;
 837                 }
 838
 839                 nr_free_blks += blk_in_line;
 840                 atomic_set(&line->blk_in_line, blk_in_line);
 841
 842                 l_mg->nr_free_lines++;
 843                 list_add_tail(&line->list, &l_mg->free_list);
 844         }
 845
 846         pblk_set_provision(pblk, nr_free_blks);
 847
 848         /* Cleanup per-LUN bad block lists - managed within lines on run-time */
 849         for (i = 0; i < geo->all_luns; i++)
 850                 kfree(pblk->luns[i].bb_list);
 851
 852         return 0;
 853 fail_free_lines:
 854         while (--i >= 0)
 855                 pblk_free_line_bitmaps(&pblk->lines[i]);
 856 fail_free_bb_aux:
 857         kfree(l_mg->bb_aux);
 858 fail_free_bb_template:
 859         kfree(l_mg->bb_template);
 860 fail_free_meta:
 861         pblk_line_meta_free(pblk);
 862 fail:
 863         for (i = 0; i < geo->all_luns; i++)
 864                 kfree(pblk->luns[i].bb_list);
 865
 866         return ret;
 867 }
 868
 869 static int pblk_writer_init(struct pblk *pblk)
 870 {
 871         timer_setup(&pblk->wtimer, pblk_write_timer_fn, 0);
 872         mod_timer(&pblk->wtimer, jiffies + msecs_to_jiffies(100));
 873
 874         pblk->writer_ts = kthread_create(pblk_write_ts, pblk, "pblk-writer-t");
 875         if (IS_ERR(pblk->writer_ts)) {
 876                 pr_err("pblk: could not allocate writer kthread\n");
 877                 return PTR_ERR(pblk->writer_ts);
 878         }
 879
 880         return 0;
 881 }
 882
 883 static void pblk_writer_stop(struct pblk *pblk)
 884 {
 885         /* The pipeline must be stopped and the write buffer emptied before the
 886          * write thread is stopped
 887          */
 888         WARN(pblk_rb_read_count(&pblk->rwb),
 889                         "Stopping not fully persisted write buffer\n");
 890
 891         WARN(pblk_rb_sync_count(&pblk->rwb),
 892                         "Stopping not fully synced write buffer\n");
 893
 894         if (pblk->writer_ts)
 895                 kthread_stop(pblk->writer_ts);
 896         del_timer(&pblk->wtimer);
 897 }
 898
 899 static void pblk_free(struct pblk *pblk)
 900 {
 901         pblk_luns_free(pblk);
 902         pblk_lines_free(pblk);
 903         pblk_line_meta_free(pblk);
 904         pblk_core_free(pblk);
 905         pblk_l2p_free(pblk);
 906
 907         kfree(pblk);
 908 }
 909
 910 static void pblk_tear_down(struct pblk *pblk)
 911 {
 912         pblk_pipeline_stop(pblk);
 913         pblk_writer_stop(pblk);
 914         pblk_rb_sync_l2p(&pblk->rwb);
 915         pblk_rwb_free(pblk);
 916         pblk_rl_free(&pblk->rl);
 917
 918         pr_debug("pblk: consistent tear down\n");
 919 }
 920
 921 static void pblk_exit(void *private)
 922 {
 923         struct pblk *pblk = private;
 924
 925         down_write(&pblk_lock);
 926         pblk_gc_exit(pblk);
 927         pblk_tear_down(pblk);
 928
 929 #ifdef CONFIG_NVM_DEBUG
 930         pr_info("pblk exit: L2P CRC: %x\n", pblk_l2p_crc(pblk));
 931 #endif
 932
 933         pblk_free(pblk);
 934         up_write(&pblk_lock);
 935 }
 936
 937 static sector_t pblk_capacity(void *private)
 938 {
 939         struct pblk *pblk = private;
 940
 941         return pblk->capacity * NR_PHY_IN_LOG;
 942 }
 943
 944 static void *pblk_init(struct nvm_tgt_dev *dev, struct gendisk *tdisk,
 945                        int flags)
 946 {
 947         struct nvm_geo *geo = &dev->geo;
 948         struct request_queue *bqueue = dev->q;
 949         struct request_queue *tqueue = tdisk->queue;
 950         struct pblk *pblk;
 951         int ret;
 952
 953         if (dev->identity.dom & NVM_RSP_L2P) {
 954                 pr_err("pblk: host-side L2P table not supported. (%x)\n",
 955                                                         dev->identity.dom);
 956                 return ERR_PTR(-EINVAL);
 957         }
 958
 959         pblk = kzalloc(sizeof(struct pblk), GFP_KERNEL);
 960         if (!pblk)
 961                 return ERR_PTR(-ENOMEM);
 962
 963         pblk->dev = dev;
 964         pblk->disk = tdisk;
 965         pblk->state = PBLK_STATE_RUNNING;
 966         pblk->gc.gc_enabled = 0;
 967
 968         spin_lock_init(&pblk->trans_lock);
 969         spin_lock_init(&pblk->lock);
 970
 971         if (flags & NVM_TARGET_FACTORY)
 972                 pblk_setup_uuid(pblk);
 973
 974 #ifdef CONFIG_NVM_DEBUG
 975         atomic_long_set(&pblk->inflight_writes, 0);
 976         atomic_long_set(&pblk->padded_writes, 0);
 977         atomic_long_set(&pblk->padded_wb, 0);
 978         atomic_long_set(&pblk->nr_flush, 0);
 979         atomic_long_set(&pblk->req_writes, 0);
 980         atomic_long_set(&pblk->sub_writes, 0);
 981         atomic_long_set(&pblk->sync_writes, 0);
 982         atomic_long_set(&pblk->inflight_reads, 0);
 983         atomic_long_set(&pblk->cache_reads, 0);
 984         atomic_long_set(&pblk->sync_reads, 0);
 985         atomic_long_set(&pblk->recov_writes, 0);
 986         atomic_long_set(&pblk->recov_writes, 0);
 987         atomic_long_set(&pblk->recov_gc_writes, 0);
 988         atomic_long_set(&pblk->recov_gc_reads, 0);
 989 #endif
 990
 991         atomic_long_set(&pblk->read_failed, 0);
 992         atomic_long_set(&pblk->read_empty, 0);
 993         atomic_long_set(&pblk->read_high_ecc, 0);
 994         atomic_long_set(&pblk->read_failed_gc, 0);
 995         atomic_long_set(&pblk->write_failed, 0);
 996         atomic_long_set(&pblk->erase_failed, 0);
 997
 998         ret = pblk_luns_init(pblk, dev->luns);
 999         if (ret) {
1000                 pr_err("pblk: could not initialize luns\n");
1001                 goto fail;
1002         }
1003
1004         ret = pblk_lines_init(pblk);
1005         if (ret) {
1006                 pr_err("pblk: could not initialize lines\n");
1007                 goto fail_free_luns;
1008         }
1009
1010         ret = pblk_core_init(pblk);
1011         if (ret) {
1012                 pr_err("pblk: could not initialize core\n");
1013                 goto fail_free_line_meta;
1014         }
1015
1016         ret = pblk_l2p_init(pblk);
1017         if (ret) {
1018                 pr_err("pblk: could not initialize maps\n");
1019                 goto fail_free_core;
1020         }
1021
1022         ret = pblk_lines_configure(pblk, flags);
1023         if (ret) {
1024                 pr_err("pblk: could not configure lines\n");
1025                 goto fail_free_l2p;
1026         }
1027
1028         ret = pblk_writer_init(pblk);
1029         if (ret) {
1030                 pr_err("pblk: could not initialize write thread\n");
1031                 goto fail_free_lines;
1032         }
1033
1034         ret = pblk_gc_init(pblk);
1035         if (ret) {
1036                 pr_err("pblk: could not initialize gc\n");
1037                 goto fail_stop_writer;
1038         }
1039
1040         /* inherit the size from the underlying device */
1041         blk_queue_logical_block_size(tqueue, queue_physical_block_size(bqueue));
1042         blk_queue_max_hw_sectors(tqueue, queue_max_hw_sectors(bqueue));
1043
1044         blk_queue_write_cache(tqueue, true, false);
1045
1046         tqueue->limits.discard_granularity = geo->sec_per_chk * geo->sec_size;
1047         tqueue->limits.discard_alignment = 0;
1048         blk_queue_max_discard_sectors(tqueue, UINT_MAX >> 9);
1049         queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, tqueue);
1050
1051         pr_info("pblk init: luns:%u, lines:%d, secs:%llu, buf entries:%u\n",
1052                         geo->all_luns, pblk->l_mg.nr_lines,
1053                         (unsigned long long)pblk->rl.nr_secs,
1054                         pblk->rwb.nr_entries);
1055
1056         wake_up_process(pblk->writer_ts);
1057
1058         /* Check if we need to start GC */
1059         pblk_gc_should_kick(pblk);
1060
1061         return pblk;
1062
1063 fail_stop_writer:
1064         pblk_writer_stop(pblk);
1065 fail_free_lines:
1066         pblk_lines_free(pblk);
1067 fail_free_l2p:
1068         pblk_l2p_free(pblk);
1069 fail_free_core:
1070         pblk_core_free(pblk);
1071 fail_free_line_meta:
1072         pblk_line_meta_free(pblk);
1073 fail_free_luns:
1074         pblk_luns_free(pblk);
1075 fail:
1076         kfree(pblk);
1077         return ERR_PTR(ret);
1078 }
1079
1080 /* physical block device target */
1081 static struct nvm_tgt_type tt_pblk = {
1082         .name           = "pblk",
1083         .version        = {1, 0, 0},
1084
1085         .make_rq        = pblk_make_rq,
1086         .capacity       = pblk_capacity,
1087
1088         .init           = pblk_init,
1089         .exit           = pblk_exit,
1090
1091         .sysfs_init     = pblk_sysfs_init,
1092         .sysfs_exit     = pblk_sysfs_exit,
1093         .owner          = THIS_MODULE,
1094 };
1095
1096 static int __init pblk_module_init(void)
1097 {
1098         int ret;
1099
1100         pblk_bio_set = bioset_create(BIO_POOL_SIZE, 0, 0);
1101         if (!pblk_bio_set)
1102                 return -ENOMEM;
1103         ret = nvm_register_tgt_type(&tt_pblk);
1104         if (ret)
1105                 bioset_free(pblk_bio_set);
1106         return ret;
1107 }
1108
1109 static void pblk_module_exit(void)
1110 {
1111         bioset_free(pblk_bio_set);
1112         nvm_unregister_tgt_type(&tt_pblk);
1113 }
1114
1115 module_init(pblk_module_init);
1116 module_exit(pblk_module_exit);
1117 MODULE_AUTHOR("Javier Gonzalez <javier@cnexlabs.com>");
1118 MODULE_AUTHOR("Matias Bjorling <matias@cnexlabs.com>");
1119 MODULE_LICENSE("GPL v2");
1120 MODULE_DESCRIPTION("Physical Block-Device for Open-Channel SSDs");