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->luns_per_chnl);
 173         if (1 << power_len != geo->luns_per_chnl) {
 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->nr_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->nr_luns, pblk_rec_cache);
 274         if (!pblk->rec_pool)
 275                 goto free_gen_ws_pool;
 276
 277         pblk->r_rq_pool = mempool_create_slab_pool(geo->nr_luns,
 278                                                         pblk_g_rq_cache);
 279         if (!pblk->r_rq_pool)
 280                 goto free_rec_pool;
 281
 282         pblk->e_rq_pool = mempool_create_slab_pool(geo->nr_luns,
 283                                                         pblk_g_rq_cache);
 284         if (!pblk->e_rq_pool)
 285                 goto free_r_rq_pool;
 286
 287         pblk->w_rq_pool = mempool_create_slab_pool(geo->nr_luns,
 288                                                         pblk_w_rq_cache);
 289         if (!pblk->w_rq_pool)
 290                 goto free_e_rq_pool;
 291
 292         pblk->close_wq = alloc_workqueue("pblk-close-wq",
 293                         WQ_MEM_RECLAIM | WQ_UNBOUND, PBLK_NR_CLOSE_JOBS);
 294         if (!pblk->close_wq)
 295                 goto free_w_rq_pool;
 296
 297         pblk->bb_wq = alloc_workqueue("pblk-bb-wq",
 298                         WQ_MEM_RECLAIM | WQ_UNBOUND, 0);
 299         if (!pblk->bb_wq)
 300                 goto free_close_wq;
 301
 302         pblk->r_end_wq = alloc_workqueue("pblk-read-end-wq",
 303                         WQ_MEM_RECLAIM | WQ_UNBOUND, 0);
 304         if (!pblk->r_end_wq)
 305                 goto free_bb_wq;
 306
 307         if (pblk_set_ppaf(pblk))
 308                 goto free_r_end_wq;
 309
 310         if (pblk_rwb_init(pblk))
 311                 goto free_r_end_wq;
 312
 313         INIT_LIST_HEAD(&pblk->compl_list);
 314         return 0;
 315
 316 free_r_end_wq:
 317         destroy_workqueue(pblk->r_end_wq);
 318 free_bb_wq:
 319         destroy_workqueue(pblk->bb_wq);
 320 free_close_wq:
 321         destroy_workqueue(pblk->close_wq);
 322 free_w_rq_pool:
 323         mempool_destroy(pblk->w_rq_pool);
 324 free_e_rq_pool:
 325         mempool_destroy(pblk->e_rq_pool);
 326 free_r_rq_pool:
 327         mempool_destroy(pblk->r_rq_pool);
 328 free_rec_pool:
 329         mempool_destroy(pblk->rec_pool);
 330 free_gen_ws_pool:
 331         mempool_destroy(pblk->gen_ws_pool);
 332 free_page_bio_pool:
 333         mempool_destroy(pblk->page_bio_pool);
 334 free_global_caches:
 335         pblk_free_global_caches(pblk);
 336         return -ENOMEM;
 337 }
 338
 339 static void pblk_core_free(struct pblk *pblk)
 340 {
 341         if (pblk->close_wq)
 342                 destroy_workqueue(pblk->close_wq);
 343
 344         if (pblk->r_end_wq)
 345                 destroy_workqueue(pblk->r_end_wq);
 346
 347         if (pblk->bb_wq)
 348                 destroy_workqueue(pblk->bb_wq);
 349
 350         mempool_destroy(pblk->page_bio_pool);
 351         mempool_destroy(pblk->gen_ws_pool);
 352         mempool_destroy(pblk->rec_pool);
 353         mempool_destroy(pblk->r_rq_pool);
 354         mempool_destroy(pblk->e_rq_pool);
 355         mempool_destroy(pblk->w_rq_pool);
 356
 357         pblk_free_global_caches(pblk);
 358 }
 359
 360 static void pblk_luns_free(struct pblk *pblk)
 361 {
 362         kfree(pblk->luns);
 363 }
 364
 365 static void pblk_free_line_bitmaps(struct pblk_line *line)
 366 {
 367         kfree(line->blk_bitmap);
 368         kfree(line->erase_bitmap);
 369 }
 370
 371 static void pblk_lines_free(struct pblk *pblk)
 372 {
 373         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
 374         struct pblk_line *line;
 375         int i;
 376
 377         spin_lock(&l_mg->free_lock);
 378         for (i = 0; i < l_mg->nr_lines; i++) {
 379                 line = &pblk->lines[i];
 380
 381                 pblk_line_free(pblk, line);
 382                 pblk_free_line_bitmaps(line);
 383         }
 384         spin_unlock(&l_mg->free_lock);
 385 }
 386
 387 static void pblk_line_meta_free(struct pblk *pblk)
 388 {
 389         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
 390         int i;
 391
 392         kfree(l_mg->bb_template);
 393         kfree(l_mg->bb_aux);
 394         kfree(l_mg->vsc_list);
 395
 396         for (i = 0; i < PBLK_DATA_LINES; i++) {
 397                 kfree(l_mg->sline_meta[i]);
 398                 pblk_mfree(l_mg->eline_meta[i]->buf, l_mg->emeta_alloc_type);
 399                 kfree(l_mg->eline_meta[i]);
 400         }
 401
 402         kfree(pblk->lines);
 403 }
 404
 405 static int pblk_bb_discovery(struct nvm_tgt_dev *dev, struct pblk_lun *rlun)
 406 {
 407         struct nvm_geo *geo = &dev->geo;
 408         struct ppa_addr ppa;
 409         u8 *blks;
 410         int nr_blks, ret;
 411
 412         nr_blks = geo->blks_per_lun * geo->plane_mode;
 413         blks = kmalloc(nr_blks, GFP_KERNEL);
 414         if (!blks)
 415                 return -ENOMEM;
 416
 417         ppa.ppa = 0;
 418         ppa.g.ch = rlun->bppa.g.ch;
 419         ppa.g.lun = rlun->bppa.g.lun;
 420
 421         ret = nvm_get_tgt_bb_tbl(dev, ppa, blks);
 422         if (ret)
 423                 goto out;
 424
 425         nr_blks = nvm_bb_tbl_fold(dev->parent, blks, nr_blks);
 426         if (nr_blks < 0) {
 427                 ret = nr_blks;
 428                 goto out;
 429         }
 430
 431         rlun->bb_list = blks;
 432
 433         return 0;
 434 out:
 435         kfree(blks);
 436         return ret;
 437 }
 438
 439 static int pblk_bb_line(struct pblk *pblk, struct pblk_line *line,
 440                         int blk_per_line)
 441 {
 442         struct nvm_tgt_dev *dev = pblk->dev;
 443         struct nvm_geo *geo = &dev->geo;
 444         struct pblk_lun *rlun;
 445         int bb_cnt = 0;
 446         int i;
 447
 448         for (i = 0; i < blk_per_line; i++) {
 449                 rlun = &pblk->luns[i];
 450                 if (rlun->bb_list[line->id] == NVM_BLK_T_FREE)
 451                         continue;
 452
 453                 set_bit(pblk_ppa_to_pos(geo, rlun->bppa), line->blk_bitmap);
 454                 bb_cnt++;
 455         }
 456
 457         return bb_cnt;
 458 }
 459
 460 static int pblk_alloc_line_bitmaps(struct pblk *pblk, struct pblk_line *line)
 461 {
 462         struct pblk_line_meta *lm = &pblk->lm;
 463
 464         line->blk_bitmap = kzalloc(lm->blk_bitmap_len, GFP_KERNEL);
 465         if (!line->blk_bitmap)
 466                 return -ENOMEM;
 467
 468         line->erase_bitmap = kzalloc(lm->blk_bitmap_len, GFP_KERNEL);
 469         if (!line->erase_bitmap) {
 470                 kfree(line->blk_bitmap);
 471                 return -ENOMEM;
 472         }
 473
 474         return 0;
 475 }
 476
 477 static int pblk_luns_init(struct pblk *pblk, struct ppa_addr *luns)
 478 {
 479         struct nvm_tgt_dev *dev = pblk->dev;
 480         struct nvm_geo *geo = &dev->geo;
 481         struct pblk_lun *rlun;
 482         int i, ret;
 483
 484         /* TODO: Implement unbalanced LUN support */
 485         if (geo->luns_per_chnl < 0) {
 486                 pr_err("pblk: unbalanced LUN config.\n");
 487                 return -EINVAL;
 488         }
 489
 490         pblk->luns = kcalloc(geo->nr_luns, sizeof(struct pblk_lun), GFP_KERNEL);
 491         if (!pblk->luns)
 492                 return -ENOMEM;
 493
 494         for (i = 0; i < geo->nr_luns; i++) {
 495                 /* Stripe across channels */
 496                 int ch = i % geo->nr_chnls;
 497                 int lun_raw = i / geo->nr_chnls;
 498                 int lunid = lun_raw + ch * geo->luns_per_chnl;
 499
 500                 rlun = &pblk->luns[i];
 501                 rlun->bppa = luns[lunid];
 502
 503                 sema_init(&rlun->wr_sem, 1);
 504
 505                 ret = pblk_bb_discovery(dev, rlun);
 506                 if (ret) {
 507                         while (--i >= 0)
 508                                 kfree(pblk->luns[i].bb_list);
 509                         return ret;
 510                 }
 511         }
 512
 513         return 0;
 514 }
 515
 516 static int pblk_lines_configure(struct pblk *pblk, int flags)
 517 {
 518         struct pblk_line *line = NULL;
 519         int ret = 0;
 520
 521         if (!(flags & NVM_TARGET_FACTORY)) {
 522                 line = pblk_recov_l2p(pblk);
 523                 if (IS_ERR(line)) {
 524                         pr_err("pblk: could not recover l2p table\n");
 525                         ret = -EFAULT;
 526                 }
 527         }
 528
 529 #ifdef CONFIG_NVM_DEBUG
 530         pr_info("pblk init: L2P CRC: %x\n", pblk_l2p_crc(pblk));
 531 #endif
 532
 533         /* Free full lines directly as GC has not been started yet */
 534         pblk_gc_free_full_lines(pblk);
 535
 536         if (!line) {
 537                 /* Configure next line for user data */
 538                 line = pblk_line_get_first_data(pblk);
 539                 if (!line) {
 540                         pr_err("pblk: line list corrupted\n");
 541                         ret = -EFAULT;
 542                 }
 543         }
 544
 545         return ret;
 546 }
 547
 548 /* See comment over struct line_emeta definition */
 549 static unsigned int calc_emeta_len(struct pblk *pblk)
 550 {
 551         struct pblk_line_meta *lm = &pblk->lm;
 552         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
 553         struct nvm_tgt_dev *dev = pblk->dev;
 554         struct nvm_geo *geo = &dev->geo;
 555
 556         /* Round to sector size so that lba_list starts on its own sector */
 557         lm->emeta_sec[1] = DIV_ROUND_UP(
 558                         sizeof(struct line_emeta) + lm->blk_bitmap_len,
 559                         geo->sec_size);
 560         lm->emeta_len[1] = lm->emeta_sec[1] * geo->sec_size;
 561
 562         /* Round to sector size so that vsc_list starts on its own sector */
 563         lm->dsec_per_line = lm->sec_per_line - lm->emeta_sec[0];
 564         lm->emeta_sec[2] = DIV_ROUND_UP(lm->dsec_per_line * sizeof(u64),
 565                         geo->sec_size);
 566         lm->emeta_len[2] = lm->emeta_sec[2] * geo->sec_size;
 567
 568         lm->emeta_sec[3] = DIV_ROUND_UP(l_mg->nr_lines * sizeof(u32),
 569                         geo->sec_size);
 570         lm->emeta_len[3] = lm->emeta_sec[3] * geo->sec_size;
 571
 572         lm->vsc_list_len = l_mg->nr_lines * sizeof(u32);
 573
 574         return (lm->emeta_len[1] + lm->emeta_len[2] + lm->emeta_len[3]);
 575 }
 576
 577 static void pblk_set_provision(struct pblk *pblk, long nr_free_blks)
 578 {
 579         struct nvm_tgt_dev *dev = pblk->dev;
 580         struct nvm_geo *geo = &dev->geo;
 581         sector_t provisioned;
 582
 583         pblk->over_pct = 20;
 584
 585         provisioned = nr_free_blks;
 586         provisioned *= (100 - pblk->over_pct);
 587         sector_div(provisioned, 100);
 588
 589         /* Internally pblk manages all free blocks, but all calculations based
 590          * on user capacity consider only provisioned blocks
 591          */
 592         pblk->rl.total_blocks = nr_free_blks;
 593         pblk->rl.nr_secs = nr_free_blks * geo->sec_per_blk;
 594         pblk->capacity = provisioned * geo->sec_per_blk;
 595         atomic_set(&pblk->rl.free_blocks, nr_free_blks);
 596 }
 597
 598 static int pblk_lines_alloc_metadata(struct pblk *pblk)
 599 {
 600         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
 601         struct pblk_line_meta *lm = &pblk->lm;
 602         int i;
 603
 604         /* smeta is always small enough to fit on a kmalloc memory allocation,
 605          * emeta depends on the number of LUNs allocated to the pblk instance
 606          */
 607         for (i = 0; i < PBLK_DATA_LINES; i++) {
 608                 l_mg->sline_meta[i] = kmalloc(lm->smeta_len, GFP_KERNEL);
 609                 if (!l_mg->sline_meta[i])
 610                         goto fail_free_smeta;
 611         }
 612
 613         /* emeta allocates three different buffers for managing metadata with
 614          * in-memory and in-media layouts
 615          */
 616         for (i = 0; i < PBLK_DATA_LINES; i++) {
 617                 struct pblk_emeta *emeta;
 618
 619                 emeta = kmalloc(sizeof(struct pblk_emeta), GFP_KERNEL);
 620                 if (!emeta)
 621                         goto fail_free_emeta;
 622
 623                 if (lm->emeta_len[0] > KMALLOC_MAX_CACHE_SIZE) {
 624                         l_mg->emeta_alloc_type = PBLK_VMALLOC_META;
 625
 626                         emeta->buf = vmalloc(lm->emeta_len[0]);
 627                         if (!emeta->buf) {
 628                                 kfree(emeta);
 629                                 goto fail_free_emeta;
 630                         }
 631
 632                         emeta->nr_entries = lm->emeta_sec[0];
 633                         l_mg->eline_meta[i] = emeta;
 634                 } else {
 635                         l_mg->emeta_alloc_type = PBLK_KMALLOC_META;
 636
 637                         emeta->buf = kmalloc(lm->emeta_len[0], GFP_KERNEL);
 638                         if (!emeta->buf) {
 639                                 kfree(emeta);
 640                                 goto fail_free_emeta;
 641                         }
 642
 643                         emeta->nr_entries = lm->emeta_sec[0];
 644                         l_mg->eline_meta[i] = emeta;
 645                 }
 646         }
 647
 648         l_mg->vsc_list = kcalloc(l_mg->nr_lines, sizeof(__le32), GFP_KERNEL);
 649         if (!l_mg->vsc_list)
 650                 goto fail_free_emeta;
 651
 652         for (i = 0; i < l_mg->nr_lines; i++)
 653                 l_mg->vsc_list[i] = cpu_to_le32(EMPTY_ENTRY);
 654
 655         return 0;
 656
 657 fail_free_emeta:
 658         while (--i >= 0) {
 659                 if (l_mg->emeta_alloc_type == PBLK_VMALLOC_META)
 660                         vfree(l_mg->eline_meta[i]->buf);
 661                 else
 662                         kfree(l_mg->eline_meta[i]->buf);
 663                 kfree(l_mg->eline_meta[i]);
 664         }
 665
 666 fail_free_smeta:
 667         for (i = 0; i < PBLK_DATA_LINES; i++)
 668                 kfree(l_mg->sline_meta[i]);
 669
 670         return -ENOMEM;
 671 }
 672
 673 static int pblk_lines_init(struct pblk *pblk)
 674 {
 675         struct nvm_tgt_dev *dev = pblk->dev;
 676         struct nvm_geo *geo = &dev->geo;
 677         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
 678         struct pblk_line_meta *lm = &pblk->lm;
 679         struct pblk_line *line;
 680         unsigned int smeta_len, emeta_len;
 681         long nr_bad_blks, nr_free_blks;
 682         int bb_distance, max_write_ppas, mod;
 683         int i, ret;
 684
 685         pblk->min_write_pgs = geo->sec_per_pl * (geo->sec_size / PAGE_SIZE);
 686         max_write_ppas = pblk->min_write_pgs * geo->nr_luns;
 687         pblk->max_write_pgs = (max_write_ppas < nvm_max_phys_sects(dev)) ?
 688                                 max_write_ppas : nvm_max_phys_sects(dev);
 689         pblk_set_sec_per_write(pblk, pblk->min_write_pgs);
 690
 691         if (pblk->max_write_pgs > PBLK_MAX_REQ_ADDRS) {
 692                 pr_err("pblk: cannot support device max_phys_sect\n");
 693                 return -EINVAL;
 694         }
 695
 696         div_u64_rem(geo->sec_per_blk, pblk->min_write_pgs, &mod);
 697         if (mod) {
 698                 pr_err("pblk: bad configuration of sectors/pages\n");
 699                 return -EINVAL;
 700         }
 701
 702         l_mg->nr_lines = geo->blks_per_lun;
 703         l_mg->log_line = l_mg->data_line = NULL;
 704         l_mg->l_seq_nr = l_mg->d_seq_nr = 0;
 705         l_mg->nr_free_lines = 0;
 706         bitmap_zero(&l_mg->meta_bitmap, PBLK_DATA_LINES);
 707
 708         lm->sec_per_line = geo->sec_per_blk * geo->nr_luns;
 709         lm->blk_per_line = geo->nr_luns;
 710         lm->blk_bitmap_len = BITS_TO_LONGS(geo->nr_luns) * sizeof(long);
 711         lm->sec_bitmap_len = BITS_TO_LONGS(lm->sec_per_line) * sizeof(long);
 712         lm->lun_bitmap_len = BITS_TO_LONGS(geo->nr_luns) * sizeof(long);
 713         lm->mid_thrs = lm->sec_per_line / 2;
 714         lm->high_thrs = lm->sec_per_line / 4;
 715         lm->meta_distance = (geo->nr_luns / 2) * pblk->min_write_pgs;
 716
 717         /* Calculate necessary pages for smeta. See comment over struct
 718          * line_smeta definition
 719          */
 720         i = 1;
 721 add_smeta_page:
 722         lm->smeta_sec = i * geo->sec_per_pl;
 723         lm->smeta_len = lm->smeta_sec * geo->sec_size;
 724
 725         smeta_len = sizeof(struct line_smeta) + lm->lun_bitmap_len;
 726         if (smeta_len > lm->smeta_len) {
 727                 i++;
 728                 goto add_smeta_page;
 729         }
 730
 731         /* Calculate necessary pages for emeta. See comment over struct
 732          * line_emeta definition
 733          */
 734         i = 1;
 735 add_emeta_page:
 736         lm->emeta_sec[0] = i * geo->sec_per_pl;
 737         lm->emeta_len[0] = lm->emeta_sec[0] * geo->sec_size;
 738
 739         emeta_len = calc_emeta_len(pblk);
 740         if (emeta_len > lm->emeta_len[0]) {
 741                 i++;
 742                 goto add_emeta_page;
 743         }
 744
 745         lm->emeta_bb = geo->nr_luns > i ? geo->nr_luns - i : 0;
 746
 747         lm->min_blk_line = 1;
 748         if (geo->nr_luns > 1)
 749                 lm->min_blk_line += DIV_ROUND_UP(lm->smeta_sec +
 750                                         lm->emeta_sec[0], geo->sec_per_blk);
 751
 752         if (lm->min_blk_line > lm->blk_per_line) {
 753                 pr_err("pblk: config. not supported. Min. LUN in line:%d\n",
 754                                                         lm->blk_per_line);
 755                 ret = -EINVAL;
 756                 goto fail;
 757         }
 758
 759         ret = pblk_lines_alloc_metadata(pblk);
 760         if (ret)
 761                 goto fail;
 762
 763         l_mg->bb_template = kzalloc(lm->sec_bitmap_len, GFP_KERNEL);
 764         if (!l_mg->bb_template) {
 765                 ret = -ENOMEM;
 766                 goto fail_free_meta;
 767         }
 768
 769         l_mg->bb_aux = kzalloc(lm->sec_bitmap_len, GFP_KERNEL);
 770         if (!l_mg->bb_aux) {
 771                 ret = -ENOMEM;
 772                 goto fail_free_bb_template;
 773         }
 774
 775         bb_distance = (geo->nr_luns) * geo->sec_per_pl;
 776         for (i = 0; i < lm->sec_per_line; i += bb_distance)
 777                 bitmap_set(l_mg->bb_template, i, geo->sec_per_pl);
 778
 779         INIT_LIST_HEAD(&l_mg->free_list);
 780         INIT_LIST_HEAD(&l_mg->corrupt_list);
 781         INIT_LIST_HEAD(&l_mg->bad_list);
 782         INIT_LIST_HEAD(&l_mg->gc_full_list);
 783         INIT_LIST_HEAD(&l_mg->gc_high_list);
 784         INIT_LIST_HEAD(&l_mg->gc_mid_list);
 785         INIT_LIST_HEAD(&l_mg->gc_low_list);
 786         INIT_LIST_HEAD(&l_mg->gc_empty_list);
 787
 788         INIT_LIST_HEAD(&l_mg->emeta_list);
 789
 790         l_mg->gc_lists[0] = &l_mg->gc_high_list;
 791         l_mg->gc_lists[1] = &l_mg->gc_mid_list;
 792         l_mg->gc_lists[2] = &l_mg->gc_low_list;
 793
 794         spin_lock_init(&l_mg->free_lock);
 795         spin_lock_init(&l_mg->close_lock);
 796         spin_lock_init(&l_mg->gc_lock);
 797
 798         pblk->lines = kcalloc(l_mg->nr_lines, sizeof(struct pblk_line),
 799                                                                 GFP_KERNEL);
 800         if (!pblk->lines) {
 801                 ret = -ENOMEM;
 802                 goto fail_free_bb_aux;
 803         }
 804
 805         nr_free_blks = 0;
 806         for (i = 0; i < l_mg->nr_lines; i++) {
 807                 int blk_in_line;
 808
 809                 line = &pblk->lines[i];
 810
 811                 line->pblk = pblk;
 812                 line->id = i;
 813                 line->type = PBLK_LINETYPE_FREE;
 814                 line->state = PBLK_LINESTATE_FREE;
 815                 line->gc_group = PBLK_LINEGC_NONE;
 816                 line->vsc = &l_mg->vsc_list[i];
 817                 spin_lock_init(&line->lock);
 818
 819                 ret = pblk_alloc_line_bitmaps(pblk, line);
 820                 if (ret)
 821                         goto fail_free_lines;
 822
 823                 nr_bad_blks = pblk_bb_line(pblk, line, lm->blk_per_line);
 824                 if (nr_bad_blks < 0 || nr_bad_blks > lm->blk_per_line) {
 825                         pblk_free_line_bitmaps(line);
 826                         ret = -EINVAL;
 827                         goto fail_free_lines;
 828                 }
 829
 830                 blk_in_line = lm->blk_per_line - nr_bad_blks;
 831                 if (blk_in_line < lm->min_blk_line) {
 832                         line->state = PBLK_LINESTATE_BAD;
 833                         list_add_tail(&line->list, &l_mg->bad_list);
 834                         continue;
 835                 }
 836
 837                 nr_free_blks += blk_in_line;
 838                 atomic_set(&line->blk_in_line, blk_in_line);
 839
 840                 l_mg->nr_free_lines++;
 841                 list_add_tail(&line->list, &l_mg->free_list);
 842         }
 843
 844         pblk_set_provision(pblk, nr_free_blks);
 845
 846         /* Cleanup per-LUN bad block lists - managed within lines on run-time */
 847         for (i = 0; i < geo->nr_luns; i++)
 848                 kfree(pblk->luns[i].bb_list);
 849
 850         return 0;
 851 fail_free_lines:
 852         while (--i >= 0)
 853                 pblk_free_line_bitmaps(&pblk->lines[i]);
 854 fail_free_bb_aux:
 855         kfree(l_mg->bb_aux);
 856 fail_free_bb_template:
 857         kfree(l_mg->bb_template);
 858 fail_free_meta:
 859         pblk_line_meta_free(pblk);
 860 fail:
 861         for (i = 0; i < geo->nr_luns; i++)
 862                 kfree(pblk->luns[i].bb_list);
 863
 864         return ret;
 865 }
 866
 867 static int pblk_writer_init(struct pblk *pblk)
 868 {
 869         timer_setup(&pblk->wtimer, pblk_write_timer_fn, 0);
 870         mod_timer(&pblk->wtimer, jiffies + msecs_to_jiffies(100));
 871
 872         pblk->writer_ts = kthread_create(pblk_write_ts, pblk, "pblk-writer-t");
 873         if (IS_ERR(pblk->writer_ts)) {
 874                 pr_err("pblk: could not allocate writer kthread\n");
 875                 return PTR_ERR(pblk->writer_ts);
 876         }
 877
 878         return 0;
 879 }
 880
 881 static void pblk_writer_stop(struct pblk *pblk)
 882 {
 883         /* The pipeline must be stopped and the write buffer emptied before the
 884          * write thread is stopped
 885          */
 886         WARN(pblk_rb_read_count(&pblk->rwb),
 887                         "Stopping not fully persisted write buffer\n");
 888
 889         WARN(pblk_rb_sync_count(&pblk->rwb),
 890                         "Stopping not fully synced write buffer\n");
 891
 892         if (pblk->writer_ts)
 893                 kthread_stop(pblk->writer_ts);
 894         del_timer(&pblk->wtimer);
 895 }
 896
 897 static void pblk_free(struct pblk *pblk)
 898 {
 899         pblk_luns_free(pblk);
 900         pblk_lines_free(pblk);
 901         pblk_line_meta_free(pblk);
 902         pblk_core_free(pblk);
 903         pblk_l2p_free(pblk);
 904
 905         kfree(pblk);
 906 }
 907
 908 static void pblk_tear_down(struct pblk *pblk)
 909 {
 910         pblk_pipeline_stop(pblk);
 911         pblk_writer_stop(pblk);
 912         pblk_rb_sync_l2p(&pblk->rwb);
 913         pblk_rwb_free(pblk);
 914         pblk_rl_free(&pblk->rl);
 915
 916         pr_debug("pblk: consistent tear down\n");
 917 }
 918
 919 static void pblk_exit(void *private)
 920 {
 921         struct pblk *pblk = private;
 922
 923         down_write(&pblk_lock);
 924         pblk_gc_exit(pblk);
 925         pblk_tear_down(pblk);
 926
 927 #ifdef CONFIG_NVM_DEBUG
 928         pr_info("pblk exit: L2P CRC: %x\n", pblk_l2p_crc(pblk));
 929 #endif
 930
 931         pblk_free(pblk);
 932         up_write(&pblk_lock);
 933 }
 934
 935 static sector_t pblk_capacity(void *private)
 936 {
 937         struct pblk *pblk = private;
 938
 939         return pblk->capacity * NR_PHY_IN_LOG;
 940 }
 941
 942 static void *pblk_init(struct nvm_tgt_dev *dev, struct gendisk *tdisk,
 943                        int flags)
 944 {
 945         struct nvm_geo *geo = &dev->geo;
 946         struct request_queue *bqueue = dev->q;
 947         struct request_queue *tqueue = tdisk->queue;
 948         struct pblk *pblk;
 949         int ret;
 950
 951         if (dev->identity.dom & NVM_RSP_L2P) {
 952                 pr_err("pblk: host-side L2P table not supported. (%x)\n",
 953                                                         dev->identity.dom);
 954                 return ERR_PTR(-EINVAL);
 955         }
 956
 957         pblk = kzalloc(sizeof(struct pblk), GFP_KERNEL);
 958         if (!pblk)
 959                 return ERR_PTR(-ENOMEM);
 960
 961         pblk->dev = dev;
 962         pblk->disk = tdisk;
 963         pblk->state = PBLK_STATE_RUNNING;
 964         pblk->gc.gc_enabled = 0;
 965
 966         spin_lock_init(&pblk->trans_lock);
 967         spin_lock_init(&pblk->lock);
 968
 969         if (flags & NVM_TARGET_FACTORY)
 970                 pblk_setup_uuid(pblk);
 971
 972 #ifdef CONFIG_NVM_DEBUG
 973         atomic_long_set(&pblk->inflight_writes, 0);
 974         atomic_long_set(&pblk->padded_writes, 0);
 975         atomic_long_set(&pblk->padded_wb, 0);
 976         atomic_long_set(&pblk->nr_flush, 0);
 977         atomic_long_set(&pblk->req_writes, 0);
 978         atomic_long_set(&pblk->sub_writes, 0);
 979         atomic_long_set(&pblk->sync_writes, 0);
 980         atomic_long_set(&pblk->inflight_reads, 0);
 981         atomic_long_set(&pblk->cache_reads, 0);
 982         atomic_long_set(&pblk->sync_reads, 0);
 983         atomic_long_set(&pblk->recov_writes, 0);
 984         atomic_long_set(&pblk->recov_writes, 0);
 985         atomic_long_set(&pblk->recov_gc_writes, 0);
 986         atomic_long_set(&pblk->recov_gc_reads, 0);
 987 #endif
 988
 989         atomic_long_set(&pblk->read_failed, 0);
 990         atomic_long_set(&pblk->read_empty, 0);
 991         atomic_long_set(&pblk->read_high_ecc, 0);
 992         atomic_long_set(&pblk->read_failed_gc, 0);
 993         atomic_long_set(&pblk->write_failed, 0);
 994         atomic_long_set(&pblk->erase_failed, 0);
 995
 996         ret = pblk_luns_init(pblk, dev->luns);
 997         if (ret) {
 998                 pr_err("pblk: could not initialize luns\n");
 999                 goto fail;
1000         }
1001
1002         ret = pblk_lines_init(pblk);
1003         if (ret) {
1004                 pr_err("pblk: could not initialize lines\n");
1005                 goto fail_free_luns;
1006         }
1007
1008         ret = pblk_core_init(pblk);
1009         if (ret) {
1010                 pr_err("pblk: could not initialize core\n");
1011                 goto fail_free_line_meta;
1012         }
1013
1014         ret = pblk_l2p_init(pblk);
1015         if (ret) {
1016                 pr_err("pblk: could not initialize maps\n");
1017                 goto fail_free_core;
1018         }
1019
1020         ret = pblk_lines_configure(pblk, flags);
1021         if (ret) {
1022                 pr_err("pblk: could not configure lines\n");
1023                 goto fail_free_l2p;
1024         }
1025
1026         ret = pblk_writer_init(pblk);
1027         if (ret) {
1028                 pr_err("pblk: could not initialize write thread\n");
1029                 goto fail_free_lines;
1030         }
1031
1032         ret = pblk_gc_init(pblk);
1033         if (ret) {
1034                 pr_err("pblk: could not initialize gc\n");
1035                 goto fail_stop_writer;
1036         }
1037
1038         /* inherit the size from the underlying device */
1039         blk_queue_logical_block_size(tqueue, queue_physical_block_size(bqueue));
1040         blk_queue_max_hw_sectors(tqueue, queue_max_hw_sectors(bqueue));
1041
1042         blk_queue_write_cache(tqueue, true, false);
1043
1044         tqueue->limits.discard_granularity = geo->pgs_per_blk * geo->pfpg_size;
1045         tqueue->limits.discard_alignment = 0;
1046         blk_queue_max_discard_sectors(tqueue, UINT_MAX >> 9);
1047         queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, tqueue);
1048
1049         pr_info("pblk init: luns:%u, lines:%d, secs:%llu, buf entries:%u\n",
1050                         geo->nr_luns, pblk->l_mg.nr_lines,
1051                         (unsigned long long)pblk->rl.nr_secs,
1052                         pblk->rwb.nr_entries);
1053
1054         wake_up_process(pblk->writer_ts);
1055
1056         /* Check if we need to start GC */
1057         pblk_gc_should_kick(pblk);
1058
1059         return pblk;
1060
1061 fail_stop_writer:
1062         pblk_writer_stop(pblk);
1063 fail_free_lines:
1064         pblk_lines_free(pblk);
1065 fail_free_l2p:
1066         pblk_l2p_free(pblk);
1067 fail_free_core:
1068         pblk_core_free(pblk);
1069 fail_free_line_meta:
1070         pblk_line_meta_free(pblk);
1071 fail_free_luns:
1072         pblk_luns_free(pblk);
1073 fail:
1074         kfree(pblk);
1075         return ERR_PTR(ret);
1076 }
1077
1078 /* physical block device target */
1079 static struct nvm_tgt_type tt_pblk = {
1080         .name           = "pblk",
1081         .version        = {1, 0, 0},
1082
1083         .make_rq        = pblk_make_rq,
1084         .capacity       = pblk_capacity,
1085
1086         .init           = pblk_init,
1087         .exit           = pblk_exit,
1088
1089         .sysfs_init     = pblk_sysfs_init,
1090         .sysfs_exit     = pblk_sysfs_exit,
1091         .owner          = THIS_MODULE,
1092 };
1093
1094 static int __init pblk_module_init(void)
1095 {
1096         int ret;
1097
1098         pblk_bio_set = bioset_create(BIO_POOL_SIZE, 0, 0);
1099         if (!pblk_bio_set)
1100                 return -ENOMEM;
1101         ret = nvm_register_tgt_type(&tt_pblk);
1102         if (ret)
1103                 bioset_free(pblk_bio_set);
1104         return ret;
1105 }
1106
1107 static void pblk_module_exit(void)
1108 {
1109         bioset_free(pblk_bio_set);
1110         nvm_unregister_tgt_type(&tt_pblk);
1111 }
1112
1113 module_init(pblk_module_init);
1114 module_exit(pblk_module_exit);
1115 MODULE_AUTHOR("Javier Gonzalez <javier@cnexlabs.com>");
1116 MODULE_AUTHOR("Matias Bjorling <matias@cnexlabs.com>");
1117 MODULE_LICENSE("GPL v2");
1118 MODULE_DESCRIPTION("Physical Block-Device for Open-Channel SSDs");