drivers/lightnvm/pblk-init.c

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