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 <[email protected]>
   5  *                  Matias Bjorling <[email protected]>
   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_free_line_bitmaps(struct pblk_line *line)
 369 {
 370         kfree(line->blk_bitmap);
 371         kfree(line->erase_bitmap);
 372 }
 373
 374 static void pblk_lines_free(struct pblk *pblk)
 375 {
 376         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
 377         struct pblk_line *line;
 378         int i;
 379
 380         spin_lock(&l_mg->free_lock);
 381         for (i = 0; i < l_mg->nr_lines; i++) {
 382                 line = &pblk->lines[i];
 383
 384                 pblk_line_free(pblk, line);
 385                 pblk_free_line_bitmaps(line);
 386         }
 387         spin_unlock(&l_mg->free_lock);
 388 }
 389
 390 static void pblk_line_meta_free(struct pblk *pblk)
 391 {
 392         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
 393         int i;
 394
 395         kfree(l_mg->bb_template);
 396         kfree(l_mg->bb_aux);
 397         kfree(l_mg->vsc_list);
 398
 399         for (i = 0; i < PBLK_DATA_LINES; i++) {
 400                 kfree(l_mg->sline_meta[i]);
 401                 pblk_mfree(l_mg->eline_meta[i]->buf, l_mg->emeta_alloc_type);
 402                 kfree(l_mg->eline_meta[i]);
 403         }
 404
 405         kfree(pblk->lines);
 406 }
 407
 408 static int pblk_bb_discovery(struct nvm_tgt_dev *dev, struct pblk_lun *rlun)
 409 {
 410         struct nvm_geo *geo = &dev->geo;
 411         struct ppa_addr ppa;
 412         u8 *blks;
 413         int nr_blks, ret;
 414
 415         nr_blks = geo->nr_chks * geo->plane_mode;
 416         blks = kmalloc(nr_blks, GFP_KERNEL);
 417         if (!blks)
 418                 return -ENOMEM;
 419
 420         ppa.ppa = 0;
 421         ppa.g.ch = rlun->bppa.g.ch;
 422         ppa.g.lun = rlun->bppa.g.lun;
 423
 424         ret = nvm_get_tgt_bb_tbl(dev, ppa, blks);
 425         if (ret)
 426                 goto out;
 427
 428         nr_blks = nvm_bb_tbl_fold(dev->parent, blks, nr_blks);
 429         if (nr_blks < 0) {
 430                 ret = nr_blks;
 431                 goto out;
 432         }
 433
 434         rlun->bb_list = blks;
 435
 436         return 0;
 437 out:
 438         kfree(blks);
 439         return ret;
 440 }
 441
 442 static int pblk_bb_line(struct pblk *pblk, struct pblk_line *line,
 443                         int blk_per_line)
 444 {
 445         struct nvm_tgt_dev *dev = pblk->dev;
 446         struct nvm_geo *geo = &dev->geo;
 447         struct pblk_lun *rlun;
 448         int bb_cnt = 0;
 449         int i;
 450
 451         for (i = 0; i < blk_per_line; i++) {
 452                 rlun = &pblk->luns[i];
 453                 if (rlun->bb_list[line->id] == NVM_BLK_T_FREE)
 454                         continue;
 455
 456                 set_bit(pblk_ppa_to_pos(geo, rlun->bppa), line->blk_bitmap);
 457                 bb_cnt++;
 458         }
 459
 460         return bb_cnt;
 461 }
 462
 463 static int pblk_alloc_line_bitmaps(struct pblk *pblk, struct pblk_line *line)
 464 {
 465         struct pblk_line_meta *lm = &pblk->lm;
 466
 467         line->blk_bitmap = kzalloc(lm->blk_bitmap_len, GFP_KERNEL);
 468         if (!line->blk_bitmap)
 469                 return -ENOMEM;
 470
 471         line->erase_bitmap = kzalloc(lm->blk_bitmap_len, GFP_KERNEL);
 472         if (!line->erase_bitmap) {
 473                 kfree(line->blk_bitmap);
 474                 return -ENOMEM;
 475         }
 476
 477         return 0;
 478 }
 479
 480 static int pblk_luns_init(struct pblk *pblk, struct ppa_addr *luns)
 481 {
 482         struct nvm_tgt_dev *dev = pblk->dev;
 483         struct nvm_geo *geo = &dev->geo;
 484         struct pblk_lun *rlun;
 485         int i, ret;
 486
 487         /* TODO: Implement unbalanced LUN support */
 488         if (geo->nr_luns < 0) {
 489                 pr_err("pblk: unbalanced LUN config.\n");
 490                 return -EINVAL;
 491         }
 492
 493         pblk->luns = kcalloc(geo->all_luns, sizeof(struct pblk_lun),
 494                                                                 GFP_KERNEL);
 495         if (!pblk->luns)
 496                 return -ENOMEM;
 497
 498         for (i = 0; i < geo->all_luns; i++) {
 499                 /* Stripe across channels */
 500                 int ch = i % geo->nr_chnls;
 501                 int lun_raw = i / geo->nr_chnls;
 502                 int lunid = lun_raw + ch * geo->nr_luns;
 503
 504                 rlun = &pblk->luns[i];
 505                 rlun->bppa = luns[lunid];
 506
 507                 sema_init(&rlun->wr_sem, 1);
 508
 509                 ret = pblk_bb_discovery(dev, rlun);
 510                 if (ret) {
 511                         while (--i >= 0)
 512                                 kfree(pblk->luns[i].bb_list);
 513                         return ret;
 514                 }
 515         }
 516
 517         return 0;
 518 }
 519
 520 static int pblk_lines_configure(struct pblk *pblk, int flags)
 521 {
 522         struct pblk_line *line = NULL;
 523         int ret = 0;
 524
 525         if (!(flags & NVM_TARGET_FACTORY)) {
 526                 line = pblk_recov_l2p(pblk);
 527                 if (IS_ERR(line)) {
 528                         pr_err("pblk: could not recover l2p table\n");
 529                         ret = -EFAULT;
 530                 }
 531         }
 532
 533 #ifdef CONFIG_NVM_DEBUG
 534         pr_info("pblk init: L2P CRC: %x\n", pblk_l2p_crc(pblk));
 535 #endif
 536
 537         /* Free full lines directly as GC has not been started yet */
 538         pblk_gc_free_full_lines(pblk);
 539
 540         if (!line) {
 541                 /* Configure next line for user data */
 542                 line = pblk_line_get_first_data(pblk);
 543                 if (!line) {
 544                         pr_err("pblk: line list corrupted\n");
 545                         ret = -EFAULT;
 546                 }
 547         }
 548
 549         return ret;
 550 }
 551
 552 /* See comment over struct line_emeta definition */
 553 static unsigned int calc_emeta_len(struct pblk *pblk)
 554 {
 555         struct pblk_line_meta *lm = &pblk->lm;
 556         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
 557         struct nvm_tgt_dev *dev = pblk->dev;
 558         struct nvm_geo *geo = &dev->geo;
 559
 560         /* Round to sector size so that lba_list starts on its own sector */
 561         lm->emeta_sec[1] = DIV_ROUND_UP(
 562                         sizeof(struct line_emeta) + lm->blk_bitmap_len,
 563                         geo->sec_size);
 564         lm->emeta_len[1] = lm->emeta_sec[1] * geo->sec_size;
 565
 566         /* Round to sector size so that vsc_list starts on its own sector */
 567         lm->dsec_per_line = lm->sec_per_line - lm->emeta_sec[0];
 568         lm->emeta_sec[2] = DIV_ROUND_UP(lm->dsec_per_line * sizeof(u64),
 569                         geo->sec_size);
 570         lm->emeta_len[2] = lm->emeta_sec[2] * geo->sec_size;
 571
 572         lm->emeta_sec[3] = DIV_ROUND_UP(l_mg->nr_lines * sizeof(u32),
 573                         geo->sec_size);
 574         lm->emeta_len[3] = lm->emeta_sec[3] * geo->sec_size;
 575
 576         lm->vsc_list_len = l_mg->nr_lines * sizeof(u32);
 577
 578         return (lm->emeta_len[1] + lm->emeta_len[2] + lm->emeta_len[3]);
 579 }
 580
 581 static void pblk_set_provision(struct pblk *pblk, long nr_free_blks)
 582 {
 583         struct nvm_tgt_dev *dev = pblk->dev;
 584         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
 585         struct pblk_line_meta *lm = &pblk->lm;
 586         struct nvm_geo *geo = &dev->geo;
 587         sector_t provisioned;
 588         int sec_meta, blk_meta;
 589
 590         if (geo->op == NVM_TARGET_DEFAULT_OP)
 591                 pblk->op = PBLK_DEFAULT_OP;
 592         else
 593                 pblk->op = geo->op;
 594
 595         provisioned = nr_free_blks;
 596         provisioned *= (100 - pblk->op);
 597         sector_div(provisioned, 100);
 598
 599         pblk->op_blks = nr_free_blks - provisioned;
 600
 601         /* Internally pblk manages all free blocks, but all calculations based
 602          * on user capacity consider only provisioned blocks
 603          */
 604         pblk->rl.total_blocks = nr_free_blks;
 605         pblk->rl.nr_secs = nr_free_blks * geo->sec_per_chk;
 606
 607         /* Consider sectors used for metadata */
 608         sec_meta = (lm->smeta_sec + lm->emeta_sec[0]) * l_mg->nr_free_lines;
 609         blk_meta = DIV_ROUND_UP(sec_meta, geo->sec_per_chk);
 610
 611         pblk->capacity = (provisioned - blk_meta) * geo->sec_per_chk;
 612
 613         atomic_set(&pblk->rl.free_blocks, nr_free_blks);
 614         atomic_set(&pblk->rl.free_user_blocks, nr_free_blks);
 615 }
 616
 617 static int pblk_lines_alloc_metadata(struct pblk *pblk)
 618 {
 619         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
 620         struct pblk_line_meta *lm = &pblk->lm;
 621         int i;
 622
 623         /* smeta is always small enough to fit on a kmalloc memory allocation,
 624          * emeta depends on the number of LUNs allocated to the pblk instance
 625          */
 626         for (i = 0; i < PBLK_DATA_LINES; i++) {
 627                 l_mg->sline_meta[i] = kmalloc(lm->smeta_len, GFP_KERNEL);
 628                 if (!l_mg->sline_meta[i])
 629                         goto fail_free_smeta;
 630         }
 631
 632         /* emeta allocates three different buffers for managing metadata with
 633          * in-memory and in-media layouts
 634          */
 635         for (i = 0; i < PBLK_DATA_LINES; i++) {
 636                 struct pblk_emeta *emeta;
 637
 638                 emeta = kmalloc(sizeof(struct pblk_emeta), GFP_KERNEL);
 639                 if (!emeta)
 640                         goto fail_free_emeta;
 641
 642                 if (lm->emeta_len[0] > KMALLOC_MAX_CACHE_SIZE) {
 643                         l_mg->emeta_alloc_type = PBLK_VMALLOC_META;
 644
 645                         emeta->buf = vmalloc(lm->emeta_len[0]);
 646                         if (!emeta->buf) {
 647                                 kfree(emeta);
 648                                 goto fail_free_emeta;
 649                         }
 650
 651                         emeta->nr_entries = lm->emeta_sec[0];
 652                         l_mg->eline_meta[i] = emeta;
 653                 } else {
 654                         l_mg->emeta_alloc_type = PBLK_KMALLOC_META;
 655
 656                         emeta->buf = kmalloc(lm->emeta_len[0], GFP_KERNEL);
 657                         if (!emeta->buf) {
 658                                 kfree(emeta);
 659                                 goto fail_free_emeta;
 660                         }
 661
 662                         emeta->nr_entries = lm->emeta_sec[0];
 663                         l_mg->eline_meta[i] = emeta;
 664                 }
 665         }
 666
 667         l_mg->vsc_list = kcalloc(l_mg->nr_lines, sizeof(__le32), GFP_KERNEL);
 668         if (!l_mg->vsc_list)
 669                 goto fail_free_emeta;
 670
 671         for (i = 0; i < l_mg->nr_lines; i++)
 672                 l_mg->vsc_list[i] = cpu_to_le32(EMPTY_ENTRY);
 673
 674         return 0;
 675
 676 fail_free_emeta:
 677         while (--i >= 0) {
 678                 if (l_mg->emeta_alloc_type == PBLK_VMALLOC_META)
 679                         vfree(l_mg->eline_meta[i]->buf);
 680                 else
 681                         kfree(l_mg->eline_meta[i]->buf);
 682                 kfree(l_mg->eline_meta[i]);
 683         }
 684
 685 fail_free_smeta:
 686         for (i = 0; i < PBLK_DATA_LINES; i++)
 687                 kfree(l_mg->sline_meta[i]);
 688
 689         return -ENOMEM;
 690 }
 691
 692 static int pblk_lines_init(struct pblk *pblk)
 693 {
 694         struct nvm_tgt_dev *dev = pblk->dev;
 695         struct nvm_geo *geo = &dev->geo;
 696         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
 697         struct pblk_line_meta *lm = &pblk->lm;
 698         struct pblk_line *line;
 699         unsigned int smeta_len, emeta_len;
 700         long nr_bad_blks, nr_free_blks;
 701         int bb_distance, max_write_ppas, mod;
 702         int i, ret;
 703
 704         pblk->min_write_pgs = geo->sec_per_pl * (geo->sec_size / PAGE_SIZE);
 705         max_write_ppas = pblk->min_write_pgs * geo->all_luns;
 706         pblk->max_write_pgs = (max_write_ppas < nvm_max_phys_sects(dev)) ?
 707                                 max_write_ppas : nvm_max_phys_sects(dev);
 708         pblk_set_sec_per_write(pblk, pblk->min_write_pgs);
 709
 710         if (pblk->max_write_pgs > PBLK_MAX_REQ_ADDRS) {
 711                 pr_err("pblk: cannot support device max_phys_sect\n");
 712                 return -EINVAL;
 713         }
 714
 715         div_u64_rem(geo->sec_per_chk, pblk->min_write_pgs, &mod);
 716         if (mod) {
 717                 pr_err("pblk: bad configuration of sectors/pages\n");
 718                 return -EINVAL;
 719         }
 720
 721         l_mg->nr_lines = geo->nr_chks;
 722         l_mg->log_line = l_mg->data_line = NULL;
 723         l_mg->l_seq_nr = l_mg->d_seq_nr = 0;
 724         l_mg->nr_free_lines = 0;
 725         bitmap_zero(&l_mg->meta_bitmap, PBLK_DATA_LINES);
 726
 727         lm->sec_per_line = geo->sec_per_chk * geo->all_luns;
 728         lm->blk_per_line = geo->all_luns;
 729         lm->blk_bitmap_len = BITS_TO_LONGS(geo->all_luns) * sizeof(long);
 730         lm->sec_bitmap_len = BITS_TO_LONGS(lm->sec_per_line) * sizeof(long);
 731         lm->lun_bitmap_len = BITS_TO_LONGS(geo->all_luns) * sizeof(long);
 732         lm->mid_thrs = lm->sec_per_line / 2;
 733         lm->high_thrs = lm->sec_per_line / 4;
 734         lm->meta_distance = (geo->all_luns / 2) * pblk->min_write_pgs;
 735
 736         /* Calculate necessary pages for smeta. See comment over struct
 737          * line_smeta definition
 738          */
 739         i = 1;
 740 add_smeta_page:
 741         lm->smeta_sec = i * geo->sec_per_pl;
 742         lm->smeta_len = lm->smeta_sec * geo->sec_size;
 743
 744         smeta_len = sizeof(struct line_smeta) + lm->lun_bitmap_len;
 745         if (smeta_len > lm->smeta_len) {
 746                 i++;
 747                 goto add_smeta_page;
 748         }
 749
 750         /* Calculate necessary pages for emeta. See comment over struct
 751          * line_emeta definition
 752          */
 753         i = 1;
 754 add_emeta_page:
 755         lm->emeta_sec[0] = i * geo->sec_per_pl;
 756         lm->emeta_len[0] = lm->emeta_sec[0] * geo->sec_size;
 757
 758         emeta_len = calc_emeta_len(pblk);
 759         if (emeta_len > lm->emeta_len[0]) {
 760                 i++;
 761                 goto add_emeta_page;
 762         }
 763
 764         lm->emeta_bb = geo->all_luns > i ? geo->all_luns - i : 0;
 765
 766         lm->min_blk_line = 1;
 767         if (geo->all_luns > 1)
 768                 lm->min_blk_line += DIV_ROUND_UP(lm->smeta_sec +
 769                                         lm->emeta_sec[0], geo->sec_per_chk);
 770
 771         if (lm->min_blk_line > lm->blk_per_line) {
 772                 pr_err("pblk: config. not supported. Min. LUN in line:%d\n",
 773                                                         lm->blk_per_line);
 774                 ret = -EINVAL;
 775                 goto fail;
 776         }
 777
 778         ret = pblk_lines_alloc_metadata(pblk);
 779         if (ret)
 780                 goto fail;
 781
 782         l_mg->bb_template = kzalloc(lm->sec_bitmap_len, GFP_KERNEL);
 783         if (!l_mg->bb_template) {
 784                 ret = -ENOMEM;
 785                 goto fail_free_meta;
 786         }
 787
 788         l_mg->bb_aux = kzalloc(lm->sec_bitmap_len, GFP_KERNEL);
 789         if (!l_mg->bb_aux) {
 790                 ret = -ENOMEM;
 791                 goto fail_free_bb_template;
 792         }
 793
 794         bb_distance = (geo->all_luns) * geo->sec_per_pl;
 795         for (i = 0; i < lm->sec_per_line; i += bb_distance)
 796                 bitmap_set(l_mg->bb_template, i, geo->sec_per_pl);
 797
 798         INIT_LIST_HEAD(&l_mg->free_list);
 799         INIT_LIST_HEAD(&l_mg->corrupt_list);
 800         INIT_LIST_HEAD(&l_mg->bad_list);
 801         INIT_LIST_HEAD(&l_mg->gc_full_list);
 802         INIT_LIST_HEAD(&l_mg->gc_high_list);
 803         INIT_LIST_HEAD(&l_mg->gc_mid_list);
 804         INIT_LIST_HEAD(&l_mg->gc_low_list);
 805         INIT_LIST_HEAD(&l_mg->gc_empty_list);
 806
 807         INIT_LIST_HEAD(&l_mg->emeta_list);
 808
 809         l_mg->gc_lists[0] = &l_mg->gc_high_list;
 810         l_mg->gc_lists[1] = &l_mg->gc_mid_list;
 811         l_mg->gc_lists[2] = &l_mg->gc_low_list;
 812
 813         spin_lock_init(&l_mg->free_lock);
 814         spin_lock_init(&l_mg->close_lock);
 815         spin_lock_init(&l_mg->gc_lock);
 816
 817         pblk->lines = kcalloc(l_mg->nr_lines, sizeof(struct pblk_line),
 818                                                                 GFP_KERNEL);
 819         if (!pblk->lines) {
 820                 ret = -ENOMEM;
 821                 goto fail_free_bb_aux;
 822         }
 823
 824         nr_free_blks = 0;
 825         for (i = 0; i < l_mg->nr_lines; i++) {
 826                 int blk_in_line;
 827
 828                 line = &pblk->lines[i];
 829
 830                 line->pblk = pblk;
 831                 line->id = i;
 832                 line->type = PBLK_LINETYPE_FREE;
 833                 line->state = PBLK_LINESTATE_FREE;
 834                 line->gc_group = PBLK_LINEGC_NONE;
 835                 line->vsc = &l_mg->vsc_list[i];
 836                 spin_lock_init(&line->lock);
 837
 838                 ret = pblk_alloc_line_bitmaps(pblk, line);
 839                 if (ret)
 840                         goto fail_free_lines;
 841
 842                 nr_bad_blks = pblk_bb_line(pblk, line, lm->blk_per_line);
 843                 if (nr_bad_blks < 0 || nr_bad_blks > lm->blk_per_line) {
 844                         pblk_free_line_bitmaps(line);
 845                         ret = -EINVAL;
 846                         goto fail_free_lines;
 847                 }
 848
 849                 blk_in_line = lm->blk_per_line - nr_bad_blks;
 850                 if (blk_in_line < lm->min_blk_line) {
 851                         line->state = PBLK_LINESTATE_BAD;
 852                         list_add_tail(&line->list, &l_mg->bad_list);
 853                         continue;
 854                 }
 855
 856                 nr_free_blks += blk_in_line;
 857                 atomic_set(&line->blk_in_line, blk_in_line);
 858
 859                 l_mg->nr_free_lines++;
 860                 list_add_tail(&line->list, &l_mg->free_list);
 861         }
 862
 863         pblk_set_provision(pblk, nr_free_blks);
 864
 865         /* Cleanup per-LUN bad block lists - managed within lines on run-time */
 866         for (i = 0; i < geo->all_luns; i++)
 867                 kfree(pblk->luns[i].bb_list);
 868
 869         return 0;
 870 fail_free_lines:
 871         while (--i >= 0)
 872                 pblk_free_line_bitmaps(&pblk->lines[i]);
 873 fail_free_bb_aux:
 874         kfree(l_mg->bb_aux);
 875 fail_free_bb_template:
 876         kfree(l_mg->bb_template);
 877 fail_free_meta:
 878         pblk_line_meta_free(pblk);
 879 fail:
 880         for (i = 0; i < geo->all_luns; i++)
 881                 kfree(pblk->luns[i].bb_list);
 882
 883         return ret;
 884 }
 885
 886 static int pblk_writer_init(struct pblk *pblk)
 887 {
 888         pblk->writer_ts = kthread_create(pblk_write_ts, pblk, "pblk-writer-t");
 889         if (IS_ERR(pblk->writer_ts)) {
 890                 int err = PTR_ERR(pblk->writer_ts);
 891
 892                 if (err != -EINTR)
 893                         pr_err("pblk: could not allocate writer kthread (%d)\n",
 894                                         err);
 895                 return err;
 896         }
 897
 898         timer_setup(&pblk->wtimer, pblk_write_timer_fn, 0);
 899         mod_timer(&pblk->wtimer, jiffies + msecs_to_jiffies(100));
 900
 901         return 0;
 902 }
 903
 904 static void pblk_writer_stop(struct pblk *pblk)
 905 {
 906         /* The pipeline must be stopped and the write buffer emptied before the
 907          * write thread is stopped
 908          */
 909         WARN(pblk_rb_read_count(&pblk->rwb),
 910                         "Stopping not fully persisted write buffer\n");
 911
 912         WARN(pblk_rb_sync_count(&pblk->rwb),
 913                         "Stopping not fully synced write buffer\n");
 914
 915         if (pblk->writer_ts)
 916                 kthread_stop(pblk->writer_ts);
 917         del_timer(&pblk->wtimer);
 918 }
 919
 920 static void pblk_free(struct pblk *pblk)
 921 {
 922         pblk_luns_free(pblk);
 923         pblk_lines_free(pblk);
 924         pblk_line_meta_free(pblk);
 925         pblk_core_free(pblk);
 926         pblk_l2p_free(pblk);
 927
 928         kfree(pblk);
 929 }
 930
 931 static void pblk_tear_down(struct pblk *pblk)
 932 {
 933         pblk_pipeline_stop(pblk);
 934         pblk_writer_stop(pblk);
 935         pblk_rb_sync_l2p(&pblk->rwb);
 936         pblk_rl_free(&pblk->rl);
 937
 938         pr_debug("pblk: consistent tear down\n");
 939 }
 940
 941 static void pblk_exit(void *private)
 942 {
 943         struct pblk *pblk = private;
 944
 945         down_write(&pblk_lock);
 946         pblk_gc_exit(pblk);
 947         pblk_tear_down(pblk);
 948
 949 #ifdef CONFIG_NVM_DEBUG
 950         pr_info("pblk exit: L2P CRC: %x\n", pblk_l2p_crc(pblk));
 951 #endif
 952
 953         pblk_free(pblk);
 954         up_write(&pblk_lock);
 955 }
 956
 957 static sector_t pblk_capacity(void *private)
 958 {
 959         struct pblk *pblk = private;
 960
 961         return pblk->capacity * NR_PHY_IN_LOG;
 962 }
 963
 964 static void *pblk_init(struct nvm_tgt_dev *dev, struct gendisk *tdisk,
 965                        int flags)
 966 {
 967         struct nvm_geo *geo = &dev->geo;
 968         struct request_queue *bqueue = dev->q;
 969         struct request_queue *tqueue = tdisk->queue;
 970         struct pblk *pblk;
 971         int ret;
 972
 973         if (dev->identity.dom & NVM_RSP_L2P) {
 974                 pr_err("pblk: host-side L2P table not supported. (%x)\n",
 975                                                         dev->identity.dom);
 976                 return ERR_PTR(-EINVAL);
 977         }
 978
 979         pblk = kzalloc(sizeof(struct pblk), GFP_KERNEL);
 980         if (!pblk)
 981                 return ERR_PTR(-ENOMEM);
 982
 983         pblk->dev = dev;
 984         pblk->disk = tdisk;
 985         pblk->state = PBLK_STATE_RUNNING;
 986         pblk->gc.gc_enabled = 0;
 987
 988         spin_lock_init(&pblk->trans_lock);
 989         spin_lock_init(&pblk->lock);
 990
 991         if (flags & NVM_TARGET_FACTORY)
 992                 pblk_setup_uuid(pblk);
 993
 994 #ifdef CONFIG_NVM_DEBUG
 995         atomic_long_set(&pblk->inflight_writes, 0);
 996         atomic_long_set(&pblk->padded_writes, 0);
 997         atomic_long_set(&pblk->padded_wb, 0);
 998         atomic_long_set(&pblk->nr_flush, 0);
 999         atomic_long_set(&pblk->req_writes, 0);
1000         atomic_long_set(&pblk->sub_writes, 0);
1001         atomic_long_set(&pblk->sync_writes, 0);
1002         atomic_long_set(&pblk->inflight_reads, 0);
1003         atomic_long_set(&pblk->cache_reads, 0);
1004         atomic_long_set(&pblk->sync_reads, 0);
1005         atomic_long_set(&pblk->recov_writes, 0);
1006         atomic_long_set(&pblk->recov_writes, 0);
1007         atomic_long_set(&pblk->recov_gc_writes, 0);
1008         atomic_long_set(&pblk->recov_gc_reads, 0);
1009 #endif
1010
1011         atomic_long_set(&pblk->read_failed, 0);
1012         atomic_long_set(&pblk->read_empty, 0);
1013         atomic_long_set(&pblk->read_high_ecc, 0);
1014         atomic_long_set(&pblk->read_failed_gc, 0);
1015         atomic_long_set(&pblk->write_failed, 0);
1016         atomic_long_set(&pblk->erase_failed, 0);
1017
1018         ret = pblk_luns_init(pblk, dev->luns);
1019         if (ret) {
1020                 pr_err("pblk: could not initialize luns\n");
1021                 goto fail;
1022         }
1023
1024         ret = pblk_lines_init(pblk);
1025         if (ret) {
1026                 pr_err("pblk: could not initialize lines\n");
1027                 goto fail_free_luns;
1028         }
1029
1030         ret = pblk_core_init(pblk);
1031         if (ret) {
1032                 pr_err("pblk: could not initialize core\n");
1033                 goto fail_free_line_meta;
1034         }
1035
1036         ret = pblk_l2p_init(pblk);
1037         if (ret) {
1038                 pr_err("pblk: could not initialize maps\n");
1039                 goto fail_free_core;
1040         }
1041
1042         ret = pblk_lines_configure(pblk, flags);
1043         if (ret) {
1044                 pr_err("pblk: could not configure lines\n");
1045                 goto fail_free_l2p;
1046         }
1047
1048         ret = pblk_writer_init(pblk);
1049         if (ret) {
1050                 if (ret != -EINTR)
1051                         pr_err("pblk: could not initialize write thread\n");
1052                 goto fail_free_lines;
1053         }
1054
1055         ret = pblk_gc_init(pblk);
1056         if (ret) {
1057                 pr_err("pblk: could not initialize gc\n");
1058                 goto fail_stop_writer;
1059         }
1060
1061         /* inherit the size from the underlying device */
1062         blk_queue_logical_block_size(tqueue, queue_physical_block_size(bqueue));
1063         blk_queue_max_hw_sectors(tqueue, queue_max_hw_sectors(bqueue));
1064
1065         blk_queue_write_cache(tqueue, true, false);
1066
1067         tqueue->limits.discard_granularity = geo->sec_per_chk * geo->sec_size;
1068         tqueue->limits.discard_alignment = 0;
1069         blk_queue_max_discard_sectors(tqueue, UINT_MAX >> 9);
1070         queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, tqueue);
1071
1072         pr_info("pblk(%s): luns:%u, lines:%d, secs:%llu, buf entries:%u\n",
1073                         tdisk->disk_name,
1074                         geo->all_luns, pblk->l_mg.nr_lines,
1075                         (unsigned long long)pblk->rl.nr_secs,
1076                         pblk->rwb.nr_entries);
1077
1078         wake_up_process(pblk->writer_ts);
1079
1080         /* Check if we need to start GC */
1081         pblk_gc_should_kick(pblk);
1082
1083         return pblk;
1084
1085 fail_stop_writer:
1086         pblk_writer_stop(pblk);
1087 fail_free_lines:
1088         pblk_lines_free(pblk);
1089 fail_free_l2p:
1090         pblk_l2p_free(pblk);
1091 fail_free_core:
1092         pblk_core_free(pblk);
1093 fail_free_line_meta:
1094         pblk_line_meta_free(pblk);
1095 fail_free_luns:
1096         pblk_luns_free(pblk);
1097 fail:
1098         kfree(pblk);
1099         return ERR_PTR(ret);
1100 }
1101
1102 /* physical block device target */
1103 static struct nvm_tgt_type tt_pblk = {
1104         .name           = "pblk",
1105         .version        = {1, 0, 0},
1106
1107         .make_rq        = pblk_make_rq,
1108         .capacity       = pblk_capacity,
1109
1110         .init           = pblk_init,
1111         .exit           = pblk_exit,
1112
1113         .sysfs_init     = pblk_sysfs_init,
1114         .sysfs_exit     = pblk_sysfs_exit,
1115         .owner          = THIS_MODULE,
1116 };
1117
1118 static int __init pblk_module_init(void)
1119 {
1120         int ret;
1121
1122         pblk_bio_set = bioset_create(BIO_POOL_SIZE, 0, 0);
1123         if (!pblk_bio_set)
1124                 return -ENOMEM;
1125         ret = nvm_register_tgt_type(&tt_pblk);
1126         if (ret)
1127                 bioset_free(pblk_bio_set);
1128         return ret;
1129 }
1130
1131 static void pblk_module_exit(void)
1132 {
1133         bioset_free(pblk_bio_set);
1134         nvm_unregister_tgt_type(&tt_pblk);
1135 }
1136
1137 module_init(pblk_module_init);
1138 module_exit(pblk_module_exit);
1139 MODULE_AUTHOR("Javier Gonzalez <[email protected]>");
1140 MODULE_AUTHOR("Matias Bjorling <[email protected]>");
1141 MODULE_LICENSE("GPL v2");
1142 MODULE_DESCRIPTION("Physical Block-Device for Open-Channel SSDs");