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_r_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 PAGE_POOL_SIZE 16
 136 #define ADDR_POOL_SIZE 64
 137
 138 static int pblk_set_ppaf(struct pblk *pblk)
 139 {
 140         struct nvm_tgt_dev *dev = pblk->dev;
 141         struct nvm_geo *geo = &dev->geo;
 142         struct nvm_addr_format ppaf = geo->ppaf;
 143         int power_len;
 144
 145         /* Re-calculate channel and lun format to adapt to configuration */
 146         power_len = get_count_order(geo->nr_chnls);
 147         if (1 << power_len != geo->nr_chnls) {
 148                 pr_err("pblk: supports only power-of-two channel config.\n");
 149                 return -EINVAL;
 150         }
 151         ppaf.ch_len = power_len;
 152
 153         power_len = get_count_order(geo->luns_per_chnl);
 154         if (1 << power_len != geo->luns_per_chnl) {
 155                 pr_err("pblk: supports only power-of-two LUN config.\n");
 156                 return -EINVAL;
 157         }
 158         ppaf.lun_len = power_len;
 159
 160         pblk->ppaf.sec_offset = 0;
 161         pblk->ppaf.pln_offset = ppaf.sect_len;
 162         pblk->ppaf.ch_offset = pblk->ppaf.pln_offset + ppaf.pln_len;
 163         pblk->ppaf.lun_offset = pblk->ppaf.ch_offset + ppaf.ch_len;
 164         pblk->ppaf.pg_offset = pblk->ppaf.lun_offset + ppaf.lun_len;
 165         pblk->ppaf.blk_offset = pblk->ppaf.pg_offset + ppaf.pg_len;
 166         pblk->ppaf.sec_mask = (1ULL << ppaf.sect_len) - 1;
 167         pblk->ppaf.pln_mask = ((1ULL << ppaf.pln_len) - 1) <<
 168                                                         pblk->ppaf.pln_offset;
 169         pblk->ppaf.ch_mask = ((1ULL << ppaf.ch_len) - 1) <<
 170                                                         pblk->ppaf.ch_offset;
 171         pblk->ppaf.lun_mask = ((1ULL << ppaf.lun_len) - 1) <<
 172                                                         pblk->ppaf.lun_offset;
 173         pblk->ppaf.pg_mask = ((1ULL << ppaf.pg_len) - 1) <<
 174                                                         pblk->ppaf.pg_offset;
 175         pblk->ppaf.blk_mask = ((1ULL << ppaf.blk_len) - 1) <<
 176                                                         pblk->ppaf.blk_offset;
 177
 178         pblk->ppaf_bitsize = pblk->ppaf.blk_offset + ppaf.blk_len;
 179
 180         return 0;
 181 }
 182
 183 static int pblk_init_global_caches(struct pblk *pblk)
 184 {
 185         char cache_name[PBLK_CACHE_NAME_LEN];
 186
 187         down_write(&pblk_lock);
 188         pblk_blk_ws_cache = kmem_cache_create("pblk_blk_ws",
 189                                 sizeof(struct pblk_line_ws), 0, 0, NULL);
 190         if (!pblk_blk_ws_cache) {
 191                 up_write(&pblk_lock);
 192                 return -ENOMEM;
 193         }
 194
 195         pblk_rec_cache = kmem_cache_create("pblk_rec",
 196                                 sizeof(struct pblk_rec_ctx), 0, 0, NULL);
 197         if (!pblk_rec_cache) {
 198                 kmem_cache_destroy(pblk_blk_ws_cache);
 199                 up_write(&pblk_lock);
 200                 return -ENOMEM;
 201         }
 202
 203         pblk_r_rq_cache = kmem_cache_create("pblk_r_rq", pblk_r_rq_size,
 204                                 0, 0, NULL);
 205         if (!pblk_r_rq_cache) {
 206                 kmem_cache_destroy(pblk_blk_ws_cache);
 207                 kmem_cache_destroy(pblk_rec_cache);
 208                 up_write(&pblk_lock);
 209                 return -ENOMEM;
 210         }
 211
 212         pblk_w_rq_cache = kmem_cache_create("pblk_w_rq", pblk_w_rq_size,
 213                                 0, 0, NULL);
 214         if (!pblk_w_rq_cache) {
 215                 kmem_cache_destroy(pblk_blk_ws_cache);
 216                 kmem_cache_destroy(pblk_rec_cache);
 217                 kmem_cache_destroy(pblk_r_rq_cache);
 218                 up_write(&pblk_lock);
 219                 return -ENOMEM;
 220         }
 221
 222         snprintf(cache_name, sizeof(cache_name), "pblk_line_m_%s",
 223                                                         pblk->disk->disk_name);
 224         pblk_line_meta_cache = kmem_cache_create(cache_name,
 225                                 pblk->lm.sec_bitmap_len, 0, 0, NULL);
 226         if (!pblk_line_meta_cache) {
 227                 kmem_cache_destroy(pblk_blk_ws_cache);
 228                 kmem_cache_destroy(pblk_rec_cache);
 229                 kmem_cache_destroy(pblk_r_rq_cache);
 230                 kmem_cache_destroy(pblk_w_rq_cache);
 231                 up_write(&pblk_lock);
 232                 return -ENOMEM;
 233         }
 234         up_write(&pblk_lock);
 235
 236         return 0;
 237 }
 238
 239 static int pblk_core_init(struct pblk *pblk)
 240 {
 241         struct nvm_tgt_dev *dev = pblk->dev;
 242         struct nvm_geo *geo = &dev->geo;
 243         int max_write_ppas;
 244         int mod;
 245
 246         pblk->min_write_pgs = geo->sec_per_pl * (geo->sec_size / PAGE_SIZE);
 247         max_write_ppas = pblk->min_write_pgs * geo->nr_luns;
 248         pblk->max_write_pgs = (max_write_ppas < nvm_max_phys_sects(dev)) ?
 249                                 max_write_ppas : nvm_max_phys_sects(dev);
 250         pblk->pgs_in_buffer = NVM_MEM_PAGE_WRITE * geo->sec_per_pg *
 251                                                 geo->nr_planes * geo->nr_luns;
 252
 253         if (pblk->max_write_pgs > PBLK_MAX_REQ_ADDRS) {
 254                 pr_err("pblk: cannot support device max_phys_sect\n");
 255                 return -EINVAL;
 256         }
 257
 258         div_u64_rem(geo->sec_per_blk, pblk->min_write_pgs, &mod);
 259         if (mod) {
 260                 pr_err("pblk: bad configuration of sectors/pages\n");
 261                 return -EINVAL;
 262         }
 263
 264         if (pblk_init_global_caches(pblk))
 265                 return -ENOMEM;
 266
 267         pblk->page_pool = mempool_create_page_pool(PAGE_POOL_SIZE, 0);
 268         if (!pblk->page_pool)
 269                 return -ENOMEM;
 270
 271         pblk->line_ws_pool = mempool_create_slab_pool(geo->nr_luns,
 272                                                         pblk_blk_ws_cache);
 273         if (!pblk->line_ws_pool)
 274                 goto free_page_pool;
 275
 276         pblk->rec_pool = mempool_create_slab_pool(geo->nr_luns, pblk_rec_cache);
 277         if (!pblk->rec_pool)
 278                 goto free_blk_ws_pool;
 279
 280         pblk->r_rq_pool = mempool_create_slab_pool(64, pblk_r_rq_cache);
 281         if (!pblk->r_rq_pool)
 282                 goto free_rec_pool;
 283
 284         pblk->w_rq_pool = mempool_create_slab_pool(64, pblk_w_rq_cache);
 285         if (!pblk->w_rq_pool)
 286                 goto free_r_rq_pool;
 287
 288         pblk->line_meta_pool =
 289                         mempool_create_slab_pool(16, pblk_line_meta_cache);
 290         if (!pblk->line_meta_pool)
 291                 goto free_w_rq_pool;
 292
 293         pblk->kw_wq = alloc_workqueue("pblk-aux-wq",
 294                                         WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
 295         if (!pblk->kw_wq)
 296                 goto free_line_meta_pool;
 297
 298         if (pblk_set_ppaf(pblk))
 299                 goto free_kw_wq;
 300
 301         if (pblk_rwb_init(pblk))
 302                 goto free_kw_wq;
 303
 304         INIT_LIST_HEAD(&pblk->compl_list);
 305         return 0;
 306
 307 free_kw_wq:
 308         destroy_workqueue(pblk->kw_wq);
 309 free_line_meta_pool:
 310         mempool_destroy(pblk->line_meta_pool);
 311 free_w_rq_pool:
 312         mempool_destroy(pblk->w_rq_pool);
 313 free_r_rq_pool:
 314         mempool_destroy(pblk->r_rq_pool);
 315 free_rec_pool:
 316         mempool_destroy(pblk->rec_pool);
 317 free_blk_ws_pool:
 318         mempool_destroy(pblk->line_ws_pool);
 319 free_page_pool:
 320         mempool_destroy(pblk->page_pool);
 321         return -ENOMEM;
 322 }
 323
 324 static void pblk_core_free(struct pblk *pblk)
 325 {
 326         if (pblk->kw_wq)
 327                 destroy_workqueue(pblk->kw_wq);
 328
 329         mempool_destroy(pblk->page_pool);
 330         mempool_destroy(pblk->line_ws_pool);
 331         mempool_destroy(pblk->rec_pool);
 332         mempool_destroy(pblk->r_rq_pool);
 333         mempool_destroy(pblk->w_rq_pool);
 334         mempool_destroy(pblk->line_meta_pool);
 335
 336         kmem_cache_destroy(pblk_blk_ws_cache);
 337         kmem_cache_destroy(pblk_rec_cache);
 338         kmem_cache_destroy(pblk_r_rq_cache);
 339         kmem_cache_destroy(pblk_w_rq_cache);
 340         kmem_cache_destroy(pblk_line_meta_cache);
 341 }
 342
 343 static void pblk_luns_free(struct pblk *pblk)
 344 {
 345         kfree(pblk->luns);
 346 }
 347
 348 static void pblk_lines_free(struct pblk *pblk)
 349 {
 350         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
 351         struct pblk_line *line;
 352         int i;
 353
 354         spin_lock(&l_mg->free_lock);
 355         for (i = 0; i < l_mg->nr_lines; i++) {
 356                 line = &pblk->lines[i];
 357
 358                 pblk_line_free(pblk, line);
 359                 kfree(line->blk_bitmap);
 360                 kfree(line->erase_bitmap);
 361         }
 362         spin_unlock(&l_mg->free_lock);
 363 }
 364
 365 static void pblk_line_meta_free(struct pblk *pblk)
 366 {
 367         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
 368         int i;
 369
 370         kfree(l_mg->bb_template);
 371         kfree(l_mg->bb_aux);
 372
 373         for (i = 0; i < PBLK_DATA_LINES; i++) {
 374                 pblk_mfree(l_mg->sline_meta[i].meta, l_mg->smeta_alloc_type);
 375                 pblk_mfree(l_mg->eline_meta[i].meta, l_mg->emeta_alloc_type);
 376         }
 377
 378         kfree(pblk->lines);
 379 }
 380
 381 static int pblk_bb_discovery(struct nvm_tgt_dev *dev, struct pblk_lun *rlun)
 382 {
 383         struct nvm_geo *geo = &dev->geo;
 384         struct ppa_addr ppa;
 385         u8 *blks;
 386         int nr_blks, ret;
 387
 388         nr_blks = geo->blks_per_lun * geo->plane_mode;
 389         blks = kmalloc(nr_blks, GFP_KERNEL);
 390         if (!blks)
 391                 return -ENOMEM;
 392
 393         ppa.ppa = 0;
 394         ppa.g.ch = rlun->bppa.g.ch;
 395         ppa.g.lun = rlun->bppa.g.lun;
 396
 397         ret = nvm_get_tgt_bb_tbl(dev, ppa, blks);
 398         if (ret)
 399                 goto out;
 400
 401         nr_blks = nvm_bb_tbl_fold(dev->parent, blks, nr_blks);
 402         if (nr_blks < 0) {
 403                 ret = nr_blks;
 404                 goto out;
 405         }
 406
 407         rlun->bb_list = blks;
 408
 409         return 0;
 410 out:
 411         kfree(blks);
 412         return ret;
 413 }
 414
 415 static int pblk_bb_line(struct pblk *pblk, struct pblk_line *line)
 416 {
 417         struct pblk_line_meta *lm = &pblk->lm;
 418         struct pblk_lun *rlun;
 419         int bb_cnt = 0;
 420         int i;
 421
 422         line->blk_bitmap = kzalloc(lm->blk_bitmap_len, GFP_KERNEL);
 423         if (!line->blk_bitmap)
 424                 return -ENOMEM;
 425
 426         line->erase_bitmap = kzalloc(lm->blk_bitmap_len, GFP_KERNEL);
 427         if (!line->erase_bitmap) {
 428                 kfree(line->blk_bitmap);
 429                 return -ENOMEM;
 430         }
 431
 432         for (i = 0; i < lm->blk_per_line; i++) {
 433                 rlun = &pblk->luns[i];
 434                 if (rlun->bb_list[line->id] == NVM_BLK_T_FREE)
 435                         continue;
 436
 437                 set_bit(i, line->blk_bitmap);
 438                 bb_cnt++;
 439         }
 440
 441         return bb_cnt;
 442 }
 443
 444 static int pblk_luns_init(struct pblk *pblk, struct ppa_addr *luns)
 445 {
 446         struct nvm_tgt_dev *dev = pblk->dev;
 447         struct nvm_geo *geo = &dev->geo;
 448         struct pblk_lun *rlun;
 449         int i, ret;
 450
 451         /* TODO: Implement unbalanced LUN support */
 452         if (geo->luns_per_chnl < 0) {
 453                 pr_err("pblk: unbalanced LUN config.\n");
 454                 return -EINVAL;
 455         }
 456
 457         pblk->luns = kcalloc(geo->nr_luns, sizeof(struct pblk_lun), GFP_KERNEL);
 458         if (!pblk->luns)
 459                 return -ENOMEM;
 460
 461         for (i = 0; i < geo->nr_luns; i++) {
 462                 /* Stripe across channels */
 463                 int ch = i % geo->nr_chnls;
 464                 int lun_raw = i / geo->nr_chnls;
 465                 int lunid = lun_raw + ch * geo->luns_per_chnl;
 466
 467                 rlun = &pblk->luns[i];
 468                 rlun->bppa = luns[lunid];
 469
 470                 sema_init(&rlun->wr_sem, 1);
 471
 472                 ret = pblk_bb_discovery(dev, rlun);
 473                 if (ret) {
 474                         while (--i >= 0)
 475                                 kfree(pblk->luns[i].bb_list);
 476                         return ret;
 477                 }
 478         }
 479
 480         return 0;
 481 }
 482
 483 static int pblk_lines_configure(struct pblk *pblk, int flags)
 484 {
 485         struct pblk_line *line = NULL;
 486         int ret = 0;
 487
 488         if (!(flags & NVM_TARGET_FACTORY)) {
 489                 line = pblk_recov_l2p(pblk);
 490                 if (IS_ERR(line)) {
 491                         pr_err("pblk: could not recover l2p table\n");
 492                         ret = -EFAULT;
 493                 }
 494         }
 495
 496         if (!line) {
 497                 /* Configure next line for user data */
 498                 line = pblk_line_get_first_data(pblk);
 499                 if (!line) {
 500                         pr_err("pblk: line list corrupted\n");
 501                         ret = -EFAULT;
 502                 }
 503         }
 504
 505         return ret;
 506 }
 507
 508 /* See comment over struct line_emeta definition */
 509 static unsigned int calc_emeta_len(struct pblk *pblk, struct pblk_line_meta *lm)
 510 {
 511         return (sizeof(struct line_emeta) +
 512                         ((lm->sec_per_line - lm->emeta_sec) * sizeof(u64)) +
 513                         (pblk->l_mg.nr_lines * sizeof(u32)) +
 514                         lm->blk_bitmap_len);
 515 }
 516
 517 static void pblk_set_provision(struct pblk *pblk, long nr_free_blks)
 518 {
 519         struct nvm_tgt_dev *dev = pblk->dev;
 520         struct nvm_geo *geo = &dev->geo;
 521         sector_t provisioned;
 522
 523         pblk->over_pct = 20;
 524
 525         provisioned = nr_free_blks;
 526         provisioned *= (100 - pblk->over_pct);
 527         sector_div(provisioned, 100);
 528
 529         /* Internally pblk manages all free blocks, but all calculations based
 530          * on user capacity consider only provisioned blocks
 531          */
 532         pblk->rl.total_blocks = nr_free_blks;
 533         pblk->rl.nr_secs = nr_free_blks * geo->sec_per_blk;
 534         pblk->capacity = provisioned * geo->sec_per_blk;
 535         atomic_set(&pblk->rl.free_blocks, nr_free_blks);
 536 }
 537
 538 static int pblk_lines_init(struct pblk *pblk)
 539 {
 540         struct nvm_tgt_dev *dev = pblk->dev;
 541         struct nvm_geo *geo = &dev->geo;
 542         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
 543         struct pblk_line_meta *lm = &pblk->lm;
 544         struct pblk_line *line;
 545         unsigned int smeta_len, emeta_len;
 546         long nr_bad_blks, nr_meta_blks, nr_free_blks;
 547         int bb_distance;
 548         int i;
 549         int ret;
 550
 551         lm->sec_per_line = geo->sec_per_blk * geo->nr_luns;
 552         lm->blk_per_line = geo->nr_luns;
 553         lm->blk_bitmap_len = BITS_TO_LONGS(geo->nr_luns) * sizeof(long);
 554         lm->sec_bitmap_len = BITS_TO_LONGS(lm->sec_per_line) * sizeof(long);
 555         lm->lun_bitmap_len = BITS_TO_LONGS(geo->nr_luns) * sizeof(long);
 556         lm->high_thrs = lm->sec_per_line / 2;
 557         lm->mid_thrs = lm->sec_per_line / 4;
 558
 559         /* Calculate necessary pages for smeta. See comment over struct
 560          * line_smeta definition
 561          */
 562         lm->smeta_len = sizeof(struct line_smeta) +
 563                                 PBLK_LINE_NR_LUN_BITMAP * lm->lun_bitmap_len;
 564
 565         i = 1;
 566 add_smeta_page:
 567         lm->smeta_sec = i * geo->sec_per_pl;
 568         lm->smeta_len = lm->smeta_sec * geo->sec_size;
 569
 570         smeta_len = sizeof(struct line_smeta) +
 571                                 PBLK_LINE_NR_LUN_BITMAP * lm->lun_bitmap_len;
 572         if (smeta_len > lm->smeta_len) {
 573                 i++;
 574                 goto add_smeta_page;
 575         }
 576
 577         /* Calculate necessary pages for emeta. See comment over struct
 578          * line_emeta definition
 579          */
 580         i = 1;
 581 add_emeta_page:
 582         lm->emeta_sec = i * geo->sec_per_pl;
 583         lm->emeta_len = lm->emeta_sec * geo->sec_size;
 584
 585         emeta_len = calc_emeta_len(pblk, lm);
 586         if (emeta_len > lm->emeta_len) {
 587                 i++;
 588                 goto add_emeta_page;
 589         }
 590         lm->emeta_bb = geo->nr_luns - i;
 591
 592         nr_meta_blks = (lm->smeta_sec + lm->emeta_sec +
 593                                 (geo->sec_per_blk / 2)) / geo->sec_per_blk;
 594         lm->min_blk_line = nr_meta_blks + 1;
 595
 596         l_mg->nr_lines = geo->blks_per_lun;
 597         l_mg->log_line = l_mg->data_line = NULL;
 598         l_mg->l_seq_nr = l_mg->d_seq_nr = 0;
 599         l_mg->nr_free_lines = 0;
 600         bitmap_zero(&l_mg->meta_bitmap, PBLK_DATA_LINES);
 601
 602         /* smeta is always small enough to fit on a kmalloc memory allocation,
 603          * emeta depends on the number of LUNs allocated to the pblk instance
 604          */
 605         l_mg->smeta_alloc_type = PBLK_KMALLOC_META;
 606         for (i = 0; i < PBLK_DATA_LINES; i++) {
 607                 l_mg->sline_meta[i].meta = kmalloc(lm->smeta_len, GFP_KERNEL);
 608                 if (!l_mg->sline_meta[i].meta)
 609                         while (--i >= 0) {
 610                                 kfree(l_mg->sline_meta[i].meta);
 611                                 ret = -ENOMEM;
 612                                 goto fail;
 613                         }
 614         }
 615
 616         if (lm->emeta_len > KMALLOC_MAX_CACHE_SIZE) {
 617                 l_mg->emeta_alloc_type = PBLK_VMALLOC_META;
 618
 619                 for (i = 0; i < PBLK_DATA_LINES; i++) {
 620                         l_mg->eline_meta[i].meta = vmalloc(lm->emeta_len);
 621                         if (!l_mg->eline_meta[i].meta)
 622                                 while (--i >= 0) {
 623                                         vfree(l_mg->eline_meta[i].meta);
 624                                         ret = -ENOMEM;
 625                                         goto fail;
 626                                 }
 627                 }
 628         } else {
 629                 l_mg->emeta_alloc_type = PBLK_KMALLOC_META;
 630
 631                 for (i = 0; i < PBLK_DATA_LINES; i++) {
 632                         l_mg->eline_meta[i].meta =
 633                                         kmalloc(lm->emeta_len, GFP_KERNEL);
 634                         if (!l_mg->eline_meta[i].meta)
 635                                 while (--i >= 0) {
 636                                         kfree(l_mg->eline_meta[i].meta);
 637                                         ret = -ENOMEM;
 638                                         goto fail;
 639                                 }
 640                 }
 641         }
 642
 643         l_mg->bb_template = kzalloc(lm->sec_bitmap_len, GFP_KERNEL);
 644         if (!l_mg->bb_template) {
 645                 ret = -ENOMEM;
 646                 goto fail_free_meta;
 647         }
 648
 649         l_mg->bb_aux = kzalloc(lm->sec_bitmap_len, GFP_KERNEL);
 650         if (!l_mg->bb_aux) {
 651                 ret = -ENOMEM;
 652                 goto fail_free_bb_template;
 653         }
 654
 655         bb_distance = (geo->nr_luns) * geo->sec_per_pl;
 656         for (i = 0; i < lm->sec_per_line; i += bb_distance)
 657                 bitmap_set(l_mg->bb_template, i, geo->sec_per_pl);
 658
 659         INIT_LIST_HEAD(&l_mg->free_list);
 660         INIT_LIST_HEAD(&l_mg->corrupt_list);
 661         INIT_LIST_HEAD(&l_mg->bad_list);
 662         INIT_LIST_HEAD(&l_mg->gc_full_list);
 663         INIT_LIST_HEAD(&l_mg->gc_high_list);
 664         INIT_LIST_HEAD(&l_mg->gc_mid_list);
 665         INIT_LIST_HEAD(&l_mg->gc_low_list);
 666         INIT_LIST_HEAD(&l_mg->gc_empty_list);
 667
 668         l_mg->gc_lists[0] = &l_mg->gc_high_list;
 669         l_mg->gc_lists[1] = &l_mg->gc_mid_list;
 670         l_mg->gc_lists[2] = &l_mg->gc_low_list;
 671
 672         spin_lock_init(&l_mg->free_lock);
 673         spin_lock_init(&l_mg->gc_lock);
 674
 675         pblk->lines = kcalloc(l_mg->nr_lines, sizeof(struct pblk_line),
 676                                                                 GFP_KERNEL);
 677         if (!pblk->lines) {
 678                 ret = -ENOMEM;
 679                 goto fail_free_bb_aux;
 680         }
 681
 682         nr_free_blks = 0;
 683         for (i = 0; i < l_mg->nr_lines; i++) {
 684                 int blk_in_line;
 685
 686                 line = &pblk->lines[i];
 687
 688                 line->pblk = pblk;
 689                 line->id = i;
 690                 line->type = PBLK_LINETYPE_FREE;
 691                 line->state = PBLK_LINESTATE_FREE;
 692                 line->gc_group = PBLK_LINEGC_NONE;
 693                 spin_lock_init(&line->lock);
 694
 695                 nr_bad_blks = pblk_bb_line(pblk, line);
 696                 if (nr_bad_blks < 0 || nr_bad_blks > lm->blk_per_line) {
 697                         ret = -EINVAL;
 698                         goto fail_free_lines;
 699                 }
 700
 701                 blk_in_line = lm->blk_per_line - nr_bad_blks;
 702                 if (blk_in_line < lm->min_blk_line) {
 703                         line->state = PBLK_LINESTATE_BAD;
 704                         list_add_tail(&line->list, &l_mg->bad_list);
 705                         continue;
 706                 }
 707
 708                 nr_free_blks += blk_in_line;
 709                 atomic_set(&line->blk_in_line, blk_in_line);
 710
 711                 l_mg->nr_free_lines++;
 712                 list_add_tail(&line->list, &l_mg->free_list);
 713         }
 714
 715         pblk_set_provision(pblk, nr_free_blks);
 716
 717         sema_init(&pblk->erase_sem, 1);
 718
 719         /* Cleanup per-LUN bad block lists - managed within lines on run-time */
 720         for (i = 0; i < geo->nr_luns; i++)
 721                 kfree(pblk->luns[i].bb_list);
 722
 723         return 0;
 724 fail_free_lines:
 725         kfree(pblk->lines);
 726 fail_free_bb_aux:
 727         kfree(l_mg->bb_aux);
 728 fail_free_bb_template:
 729         kfree(l_mg->bb_template);
 730 fail_free_meta:
 731         for (i = 0; i < PBLK_DATA_LINES; i++) {
 732                 pblk_mfree(l_mg->sline_meta[i].meta, l_mg->smeta_alloc_type);
 733                 pblk_mfree(l_mg->eline_meta[i].meta, l_mg->emeta_alloc_type);
 734         }
 735 fail:
 736         for (i = 0; i < geo->nr_luns; i++)
 737                 kfree(pblk->luns[i].bb_list);
 738
 739         return ret;
 740 }
 741
 742 static int pblk_writer_init(struct pblk *pblk)
 743 {
 744         setup_timer(&pblk->wtimer, pblk_write_timer_fn, (unsigned long)pblk);
 745         mod_timer(&pblk->wtimer, jiffies + msecs_to_jiffies(100));
 746
 747         pblk->writer_ts = kthread_create(pblk_write_ts, pblk, "pblk-writer-t");
 748         if (IS_ERR(pblk->writer_ts)) {
 749                 pr_err("pblk: could not allocate writer kthread\n");
 750                 return PTR_ERR(pblk->writer_ts);
 751         }
 752
 753         return 0;
 754 }
 755
 756 static void pblk_writer_stop(struct pblk *pblk)
 757 {
 758         if (pblk->writer_ts)
 759                 kthread_stop(pblk->writer_ts);
 760         del_timer(&pblk->wtimer);
 761 }
 762
 763 static void pblk_free(struct pblk *pblk)
 764 {
 765         pblk_luns_free(pblk);
 766         pblk_lines_free(pblk);
 767         pblk_line_meta_free(pblk);
 768         pblk_core_free(pblk);
 769         pblk_l2p_free(pblk);
 770
 771         kfree(pblk);
 772 }
 773
 774 static void pblk_tear_down(struct pblk *pblk)
 775 {
 776         pblk_flush_writer(pblk);
 777         pblk_writer_stop(pblk);
 778         pblk_rb_sync_l2p(&pblk->rwb);
 779         pblk_recov_pad(pblk);
 780         pblk_rwb_free(pblk);
 781         pblk_rl_free(&pblk->rl);
 782
 783         pr_debug("pblk: consistent tear down\n");
 784 }
 785
 786 static void pblk_exit(void *private)
 787 {
 788         struct pblk *pblk = private;
 789
 790         down_write(&pblk_lock);
 791         pblk_gc_exit(pblk);
 792         pblk_tear_down(pblk);
 793         pblk_free(pblk);
 794         up_write(&pblk_lock);
 795 }
 796
 797 static sector_t pblk_capacity(void *private)
 798 {
 799         struct pblk *pblk = private;
 800
 801         return pblk->capacity * NR_PHY_IN_LOG;
 802 }
 803
 804 static void *pblk_init(struct nvm_tgt_dev *dev, struct gendisk *tdisk,
 805                        int flags)
 806 {
 807         struct nvm_geo *geo = &dev->geo;
 808         struct request_queue *bqueue = dev->q;
 809         struct request_queue *tqueue = tdisk->queue;
 810         struct pblk *pblk;
 811         int ret;
 812
 813         if (dev->identity.dom & NVM_RSP_L2P) {
 814                 pr_err("pblk: device-side L2P table not supported. (%x)\n",
 815                                                         dev->identity.dom);
 816                 return ERR_PTR(-EINVAL);
 817         }
 818
 819         pblk = kzalloc(sizeof(struct pblk), GFP_KERNEL);
 820         if (!pblk)
 821                 return ERR_PTR(-ENOMEM);
 822
 823         pblk->dev = dev;
 824         pblk->disk = tdisk;
 825
 826         spin_lock_init(&pblk->trans_lock);
 827         spin_lock_init(&pblk->lock);
 828
 829         if (flags & NVM_TARGET_FACTORY)
 830                 pblk_setup_uuid(pblk);
 831
 832 #ifdef CONFIG_NVM_DEBUG
 833         atomic_long_set(&pblk->inflight_writes, 0);
 834         atomic_long_set(&pblk->padded_writes, 0);
 835         atomic_long_set(&pblk->padded_wb, 0);
 836         atomic_long_set(&pblk->nr_flush, 0);
 837         atomic_long_set(&pblk->req_writes, 0);
 838         atomic_long_set(&pblk->sub_writes, 0);
 839         atomic_long_set(&pblk->sync_writes, 0);
 840         atomic_long_set(&pblk->compl_writes, 0);
 841         atomic_long_set(&pblk->inflight_reads, 0);
 842         atomic_long_set(&pblk->cache_reads, 0);
 843         atomic_long_set(&pblk->sync_reads, 0);
 844         atomic_long_set(&pblk->recov_writes, 0);
 845         atomic_long_set(&pblk->recov_writes, 0);
 846         atomic_long_set(&pblk->recov_gc_writes, 0);
 847 #endif
 848
 849         atomic_long_set(&pblk->read_failed, 0);
 850         atomic_long_set(&pblk->read_empty, 0);
 851         atomic_long_set(&pblk->read_high_ecc, 0);
 852         atomic_long_set(&pblk->read_failed_gc, 0);
 853         atomic_long_set(&pblk->write_failed, 0);
 854         atomic_long_set(&pblk->erase_failed, 0);
 855
 856         ret = pblk_luns_init(pblk, dev->luns);
 857         if (ret) {
 858                 pr_err("pblk: could not initialize luns\n");
 859                 goto fail;
 860         }
 861
 862         ret = pblk_lines_init(pblk);
 863         if (ret) {
 864                 pr_err("pblk: could not initialize lines\n");
 865                 goto fail_free_luns;
 866         }
 867
 868         ret = pblk_core_init(pblk);
 869         if (ret) {
 870                 pr_err("pblk: could not initialize core\n");
 871                 goto fail_free_line_meta;
 872         }
 873
 874         ret = pblk_l2p_init(pblk);
 875         if (ret) {
 876                 pr_err("pblk: could not initialize maps\n");
 877                 goto fail_free_core;
 878         }
 879
 880         ret = pblk_lines_configure(pblk, flags);
 881         if (ret) {
 882                 pr_err("pblk: could not configure lines\n");
 883                 goto fail_free_l2p;
 884         }
 885
 886         ret = pblk_writer_init(pblk);
 887         if (ret) {
 888                 pr_err("pblk: could not initialize write thread\n");
 889                 goto fail_free_lines;
 890         }
 891
 892         ret = pblk_gc_init(pblk);
 893         if (ret) {
 894                 pr_err("pblk: could not initialize gc\n");
 895                 goto fail_stop_writer;
 896         }
 897
 898         /* inherit the size from the underlying device */
 899         blk_queue_logical_block_size(tqueue, queue_physical_block_size(bqueue));
 900         blk_queue_max_hw_sectors(tqueue, queue_max_hw_sectors(bqueue));
 901
 902         blk_queue_write_cache(tqueue, true, false);
 903
 904         tqueue->limits.discard_granularity = geo->pgs_per_blk * geo->pfpg_size;
 905         tqueue->limits.discard_alignment = 0;
 906         blk_queue_max_discard_sectors(tqueue, UINT_MAX >> 9);
 907         queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, tqueue);
 908
 909         pr_info("pblk init: luns:%u, lines:%d, secs:%llu, buf entries:%u\n",
 910                         geo->nr_luns, pblk->l_mg.nr_lines,
 911                         (unsigned long long)pblk->rl.nr_secs,
 912                         pblk->rwb.nr_entries);
 913
 914         wake_up_process(pblk->writer_ts);
 915         return pblk;
 916
 917 fail_stop_writer:
 918         pblk_writer_stop(pblk);
 919 fail_free_lines:
 920         pblk_lines_free(pblk);
 921 fail_free_l2p:
 922         pblk_l2p_free(pblk);
 923 fail_free_core:
 924         pblk_core_free(pblk);
 925 fail_free_line_meta:
 926         pblk_line_meta_free(pblk);
 927 fail_free_luns:
 928         pblk_luns_free(pblk);
 929 fail:
 930         kfree(pblk);
 931         return ERR_PTR(ret);
 932 }
 933
 934 /* physical block device target */
 935 static struct nvm_tgt_type tt_pblk = {
 936         .name           = "pblk",
 937         .version        = {1, 0, 0},
 938
 939         .make_rq        = pblk_make_rq,
 940         .capacity       = pblk_capacity,
 941
 942         .init           = pblk_init,
 943         .exit           = pblk_exit,
 944
 945         .sysfs_init     = pblk_sysfs_init,
 946         .sysfs_exit     = pblk_sysfs_exit,
 947 };
 948
 949 static int __init pblk_module_init(void)
 950 {
 951         int ret;
 952
 953         pblk_bio_set = bioset_create(BIO_POOL_SIZE, 0, 0);
 954         if (!pblk_bio_set)
 955                 return -ENOMEM;
 956         ret = nvm_register_tgt_type(&tt_pblk);
 957         if (ret)
 958                 bioset_free(pblk_bio_set);
 959         return ret;
 960 }
 961
 962 static void pblk_module_exit(void)
 963 {
 964         bioset_free(pblk_bio_set);
 965         nvm_unregister_tgt_type(&tt_pblk);
 966 }
 967
 968 module_init(pblk_module_init);
 969 module_exit(pblk_module_exit);
 970 MODULE_AUTHOR("Javier Gonzalez <javier@cnexlabs.com>");
 971 MODULE_AUTHOR("Matias Bjorling <matias@cnexlabs.com>");
 972 MODULE_LICENSE("GPL v2");
 973 MODULE_DESCRIPTION("Physical Block-Device for Open-Channel SSDs");