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