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_free_global_caches(pblk);
 359 }
 360
 361 static void pblk_luns_free(struct pblk *pblk)
 362 {
 363         kfree(pblk->luns);
 364 }
 365
 366 static void pblk_free_line_bitmaps(struct pblk_line *line)
 367 {
 368         kfree(line->blk_bitmap);
 369         kfree(line->erase_bitmap);
 370 }
 371
 372 static void pblk_lines_free(struct pblk *pblk)
 373 {
 374         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
 375         struct pblk_line *line;
 376         int i;
 377
 378         spin_lock(&l_mg->free_lock);
 379         for (i = 0; i < l_mg->nr_lines; i++) {
 380                 line = &pblk->lines[i];
 381
 382                 pblk_line_free(pblk, line);
 383                 pblk_free_line_bitmaps(line);
 384         }
 385         spin_unlock(&l_mg->free_lock);
 386 }
 387
 388 static void pblk_line_meta_free(struct pblk *pblk)
 389 {
 390         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
 391         int i;
 392
 393         kfree(l_mg->bb_template);
 394         kfree(l_mg->bb_aux);
 395         kfree(l_mg->vsc_list);
 396
 397         for (i = 0; i < PBLK_DATA_LINES; i++) {
 398                 kfree(l_mg->sline_meta[i]);
 399                 pblk_mfree(l_mg->eline_meta[i]->buf, l_mg->emeta_alloc_type);
 400                 kfree(l_mg->eline_meta[i]);
 401         }
 402
 403         kfree(pblk->lines);
 404 }
 405
 406 static int pblk_bb_discovery(struct nvm_tgt_dev *dev, struct pblk_lun *rlun)
 407 {
 408         struct nvm_geo *geo = &dev->geo;
 409         struct ppa_addr ppa;
 410         u8 *blks;
 411         int nr_blks, ret;
 412
 413         nr_blks = geo->nr_chks * geo->plane_mode;
 414         blks = kmalloc(nr_blks, GFP_KERNEL);
 415         if (!blks)
 416                 return -ENOMEM;
 417
 418         ppa.ppa = 0;
 419         ppa.g.ch = rlun->bppa.g.ch;
 420         ppa.g.lun = rlun->bppa.g.lun;
 421
 422         ret = nvm_get_tgt_bb_tbl(dev, ppa, blks);
 423         if (ret)
 424                 goto out;
 425
 426         nr_blks = nvm_bb_tbl_fold(dev->parent, blks, nr_blks);
 427         if (nr_blks < 0) {
 428                 ret = nr_blks;
 429                 goto out;
 430         }
 431
 432         rlun->bb_list = blks;
 433
 434         return 0;
 435 out:
 436         kfree(blks);
 437         return ret;
 438 }
 439
 440 static int pblk_bb_line(struct pblk *pblk, struct pblk_line *line,
 441                         int blk_per_line)
 442 {
 443         struct nvm_tgt_dev *dev = pblk->dev;
 444         struct nvm_geo *geo = &dev->geo;
 445         struct pblk_lun *rlun;
 446         int bb_cnt = 0;
 447         int i;
 448
 449         for (i = 0; i < blk_per_line; i++) {
 450                 rlun = &pblk->luns[i];
 451                 if (rlun->bb_list[line->id] == NVM_BLK_T_FREE)
 452                         continue;
 453
 454                 set_bit(pblk_ppa_to_pos(geo, rlun->bppa), line->blk_bitmap);
 455                 bb_cnt++;
 456         }
 457
 458         return bb_cnt;
 459 }
 460
 461 static int pblk_alloc_line_bitmaps(struct pblk *pblk, struct pblk_line *line)
 462 {
 463         struct pblk_line_meta *lm = &pblk->lm;
 464
 465         line->blk_bitmap = kzalloc(lm->blk_bitmap_len, GFP_KERNEL);
 466         if (!line->blk_bitmap)
 467                 return -ENOMEM;
 468
 469         line->erase_bitmap = kzalloc(lm->blk_bitmap_len, GFP_KERNEL);
 470         if (!line->erase_bitmap) {
 471                 kfree(line->blk_bitmap);
 472                 return -ENOMEM;
 473         }
 474
 475         return 0;
 476 }
 477
 478 static int pblk_luns_init(struct pblk *pblk, struct ppa_addr *luns)
 479 {
 480         struct nvm_tgt_dev *dev = pblk->dev;
 481         struct nvm_geo *geo = &dev->geo;
 482         struct pblk_lun *rlun;
 483         int i, ret;
 484
 485         /* TODO: Implement unbalanced LUN support */
 486         if (geo->nr_luns < 0) {
 487                 pr_err("pblk: unbalanced LUN config.\n");
 488                 return -EINVAL;
 489         }
 490
 491         pblk->luns = kcalloc(geo->all_luns, sizeof(struct pblk_lun),
 492                                                                 GFP_KERNEL);
 493         if (!pblk->luns)
 494                 return -ENOMEM;
 495
 496         for (i = 0; i < geo->all_luns; i++) {
 497                 /* Stripe across channels */
 498                 int ch = i % geo->nr_chnls;
 499                 int lun_raw = i / geo->nr_chnls;
 500                 int lunid = lun_raw + ch * geo->nr_luns;
 501
 502                 rlun = &pblk->luns[i];
 503                 rlun->bppa = luns[lunid];
 504
 505                 sema_init(&rlun->wr_sem, 1);
 506
 507                 ret = pblk_bb_discovery(dev, rlun);
 508                 if (ret) {
 509                         while (--i >= 0)
 510                                 kfree(pblk->luns[i].bb_list);
 511                         return ret;
 512                 }
 513         }
 514
 515         return 0;
 516 }
 517
 518 static int pblk_lines_configure(struct pblk *pblk, int flags)
 519 {
 520         struct pblk_line *line = NULL;
 521         int ret = 0;
 522
 523         if (!(flags & NVM_TARGET_FACTORY)) {
 524                 line = pblk_recov_l2p(pblk);
 525                 if (IS_ERR(line)) {
 526                         pr_err("pblk: could not recover l2p table\n");
 527                         ret = -EFAULT;
 528                 }
 529         }
 530
 531 #ifdef CONFIG_NVM_DEBUG
 532         pr_info("pblk init: L2P CRC: %x\n", pblk_l2p_crc(pblk));
 533 #endif
 534
 535         /* Free full lines directly as GC has not been started yet */
 536         pblk_gc_free_full_lines(pblk);
 537
 538         if (!line) {
 539                 /* Configure next line for user data */
 540                 line = pblk_line_get_first_data(pblk);
 541                 if (!line) {
 542                         pr_err("pblk: line list corrupted\n");
 543                         ret = -EFAULT;
 544                 }
 545         }
 546
 547         return ret;
 548 }
 549
 550 /* See comment over struct line_emeta definition */
 551 static unsigned int calc_emeta_len(struct pblk *pblk)
 552 {
 553         struct pblk_line_meta *lm = &pblk->lm;
 554         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
 555         struct nvm_tgt_dev *dev = pblk->dev;
 556         struct nvm_geo *geo = &dev->geo;
 557
 558         /* Round to sector size so that lba_list starts on its own sector */
 559         lm->emeta_sec[1] = DIV_ROUND_UP(
 560                         sizeof(struct line_emeta) + lm->blk_bitmap_len,
 561                         geo->sec_size);
 562         lm->emeta_len[1] = lm->emeta_sec[1] * geo->sec_size;
 563
 564         /* Round to sector size so that vsc_list starts on its own sector */
 565         lm->dsec_per_line = lm->sec_per_line - lm->emeta_sec[0];
 566         lm->emeta_sec[2] = DIV_ROUND_UP(lm->dsec_per_line * sizeof(u64),
 567                         geo->sec_size);
 568         lm->emeta_len[2] = lm->emeta_sec[2] * geo->sec_size;
 569
 570         lm->emeta_sec[3] = DIV_ROUND_UP(l_mg->nr_lines * sizeof(u32),
 571                         geo->sec_size);
 572         lm->emeta_len[3] = lm->emeta_sec[3] * geo->sec_size;
 573
 574         lm->vsc_list_len = l_mg->nr_lines * sizeof(u32);
 575
 576         return (lm->emeta_len[1] + lm->emeta_len[2] + lm->emeta_len[3]);
 577 }
 578
 579 static void pblk_set_provision(struct pblk *pblk, long nr_free_blks)
 580 {
 581         struct nvm_tgt_dev *dev = pblk->dev;
 582         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
 583         struct pblk_line_meta *lm = &pblk->lm;
 584         struct nvm_geo *geo = &dev->geo;
 585         sector_t provisioned;
 586         int sec_meta, blk_meta;
 587
 588         pblk->op = 20;
 589
 590         provisioned = nr_free_blks;
 591         provisioned *= (100 - pblk->op);
 592         sector_div(provisioned, 100);
 593
 594         pblk->op_blks = nr_free_blks - provisioned;
 595
 596         /* Internally pblk manages all free blocks, but all calculations based
 597          * on user capacity consider only provisioned blocks
 598          */
 599         pblk->rl.total_blocks = nr_free_blks;
 600         pblk->rl.nr_secs = nr_free_blks * geo->sec_per_chk;
 601
 602         /* Consider sectors used for metadata */
 603         sec_meta = (lm->smeta_sec + lm->emeta_sec[0]) * l_mg->nr_free_lines;
 604         blk_meta = DIV_ROUND_UP(sec_meta, geo->sec_per_chk);
 605
 606         pblk->capacity = (provisioned - blk_meta) * geo->sec_per_chk;
 607
 608         atomic_set(&pblk->rl.free_blocks, nr_free_blks);
 609         atomic_set(&pblk->rl.free_user_blocks, nr_free_blks);
 610 }
 611
 612 static int pblk_lines_alloc_metadata(struct pblk *pblk)
 613 {
 614         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
 615         struct pblk_line_meta *lm = &pblk->lm;
 616         int i;
 617
 618         /* smeta is always small enough to fit on a kmalloc memory allocation,
 619          * emeta depends on the number of LUNs allocated to the pblk instance
 620          */
 621         for (i = 0; i < PBLK_DATA_LINES; i++) {
 622                 l_mg->sline_meta[i] = kmalloc(lm->smeta_len, GFP_KERNEL);
 623                 if (!l_mg->sline_meta[i])
 624                         goto fail_free_smeta;
 625         }
 626
 627         /* emeta allocates three different buffers for managing metadata with
 628          * in-memory and in-media layouts
 629          */
 630         for (i = 0; i < PBLK_DATA_LINES; i++) {
 631                 struct pblk_emeta *emeta;
 632
 633                 emeta = kmalloc(sizeof(struct pblk_emeta), GFP_KERNEL);
 634                 if (!emeta)
 635                         goto fail_free_emeta;
 636
 637                 if (lm->emeta_len[0] > KMALLOC_MAX_CACHE_SIZE) {
 638                         l_mg->emeta_alloc_type = PBLK_VMALLOC_META;
 639
 640                         emeta->buf = vmalloc(lm->emeta_len[0]);
 641                         if (!emeta->buf) {
 642                                 kfree(emeta);
 643                                 goto fail_free_emeta;
 644                         }
 645
 646                         emeta->nr_entries = lm->emeta_sec[0];
 647                         l_mg->eline_meta[i] = emeta;
 648                 } else {
 649                         l_mg->emeta_alloc_type = PBLK_KMALLOC_META;
 650
 651                         emeta->buf = kmalloc(lm->emeta_len[0], GFP_KERNEL);
 652                         if (!emeta->buf) {
 653                                 kfree(emeta);
 654                                 goto fail_free_emeta;
 655                         }
 656
 657                         emeta->nr_entries = lm->emeta_sec[0];
 658                         l_mg->eline_meta[i] = emeta;
 659                 }
 660         }
 661
 662         l_mg->vsc_list = kcalloc(l_mg->nr_lines, sizeof(__le32), GFP_KERNEL);
 663         if (!l_mg->vsc_list)
 664                 goto fail_free_emeta;
 665
 666         for (i = 0; i < l_mg->nr_lines; i++)
 667                 l_mg->vsc_list[i] = cpu_to_le32(EMPTY_ENTRY);
 668
 669         return 0;
 670
 671 fail_free_emeta:
 672         while (--i >= 0) {
 673                 if (l_mg->emeta_alloc_type == PBLK_VMALLOC_META)
 674                         vfree(l_mg->eline_meta[i]->buf);
 675                 else
 676                         kfree(l_mg->eline_meta[i]->buf);
 677                 kfree(l_mg->eline_meta[i]);
 678         }
 679
 680 fail_free_smeta:
 681         for (i = 0; i < PBLK_DATA_LINES; i++)
 682                 kfree(l_mg->sline_meta[i]);
 683
 684         return -ENOMEM;
 685 }
 686
 687 static int pblk_lines_init(struct pblk *pblk)
 688 {
 689         struct nvm_tgt_dev *dev = pblk->dev;
 690         struct nvm_geo *geo = &dev->geo;
 691         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
 692         struct pblk_line_meta *lm = &pblk->lm;
 693         struct pblk_line *line;
 694         unsigned int smeta_len, emeta_len;
 695         long nr_bad_blks, nr_free_blks;
 696         int bb_distance, max_write_ppas, mod;
 697         int i, ret;
 698
 699         pblk->min_write_pgs = geo->sec_per_pl * (geo->sec_size / PAGE_SIZE);
 700         max_write_ppas = pblk->min_write_pgs * geo->all_luns;
 701         pblk->max_write_pgs = (max_write_ppas < nvm_max_phys_sects(dev)) ?
 702                                 max_write_ppas : nvm_max_phys_sects(dev);
 703         pblk_set_sec_per_write(pblk, pblk->min_write_pgs);
 704
 705         if (pblk->max_write_pgs > PBLK_MAX_REQ_ADDRS) {
 706                 pr_err("pblk: cannot support device max_phys_sect\n");
 707                 return -EINVAL;
 708         }
 709
 710         div_u64_rem(geo->sec_per_chk, pblk->min_write_pgs, &mod);
 711         if (mod) {
 712                 pr_err("pblk: bad configuration of sectors/pages\n");
 713                 return -EINVAL;
 714         }
 715
 716         l_mg->nr_lines = geo->nr_chks;
 717         l_mg->log_line = l_mg->data_line = NULL;
 718         l_mg->l_seq_nr = l_mg->d_seq_nr = 0;
 719         l_mg->nr_free_lines = 0;
 720         bitmap_zero(&l_mg->meta_bitmap, PBLK_DATA_LINES);
 721
 722         lm->sec_per_line = geo->sec_per_chk * geo->all_luns;
 723         lm->blk_per_line = geo->all_luns;
 724         lm->blk_bitmap_len = BITS_TO_LONGS(geo->all_luns) * sizeof(long);
 725         lm->sec_bitmap_len = BITS_TO_LONGS(lm->sec_per_line) * sizeof(long);
 726         lm->lun_bitmap_len = BITS_TO_LONGS(geo->all_luns) * sizeof(long);
 727         lm->mid_thrs = lm->sec_per_line / 2;
 728         lm->high_thrs = lm->sec_per_line / 4;
 729         lm->meta_distance = (geo->all_luns / 2) * pblk->min_write_pgs;
 730
 731         /* Calculate necessary pages for smeta. See comment over struct
 732          * line_smeta definition
 733          */
 734         i = 1;
 735 add_smeta_page:
 736         lm->smeta_sec = i * geo->sec_per_pl;
 737         lm->smeta_len = lm->smeta_sec * geo->sec_size;
 738
 739         smeta_len = sizeof(struct line_smeta) + lm->lun_bitmap_len;
 740         if (smeta_len > lm->smeta_len) {
 741                 i++;
 742                 goto add_smeta_page;
 743         }
 744
 745         /* Calculate necessary pages for emeta. See comment over struct
 746          * line_emeta definition
 747          */
 748         i = 1;
 749 add_emeta_page:
 750         lm->emeta_sec[0] = i * geo->sec_per_pl;
 751         lm->emeta_len[0] = lm->emeta_sec[0] * geo->sec_size;
 752
 753         emeta_len = calc_emeta_len(pblk);
 754         if (emeta_len > lm->emeta_len[0]) {
 755                 i++;
 756                 goto add_emeta_page;
 757         }
 758
 759         lm->emeta_bb = geo->all_luns > i ? geo->all_luns - i : 0;
 760
 761         lm->min_blk_line = 1;
 762         if (geo->all_luns > 1)
 763                 lm->min_blk_line += DIV_ROUND_UP(lm->smeta_sec +
 764                                         lm->emeta_sec[0], geo->sec_per_chk);
 765
 766         if (lm->min_blk_line > lm->blk_per_line) {
 767                 pr_err("pblk: config. not supported. Min. LUN in line:%d\n",
 768                                                         lm->blk_per_line);
 769                 ret = -EINVAL;
 770                 goto fail;
 771         }
 772
 773         ret = pblk_lines_alloc_metadata(pblk);
 774         if (ret)
 775                 goto fail;
 776
 777         l_mg->bb_template = kzalloc(lm->sec_bitmap_len, GFP_KERNEL);
 778         if (!l_mg->bb_template) {
 779                 ret = -ENOMEM;
 780                 goto fail_free_meta;
 781         }
 782
 783         l_mg->bb_aux = kzalloc(lm->sec_bitmap_len, GFP_KERNEL);
 784         if (!l_mg->bb_aux) {
 785                 ret = -ENOMEM;
 786                 goto fail_free_bb_template;
 787         }
 788
 789         bb_distance = (geo->all_luns) * geo->sec_per_pl;
 790         for (i = 0; i < lm->sec_per_line; i += bb_distance)
 791                 bitmap_set(l_mg->bb_template, i, geo->sec_per_pl);
 792
 793         INIT_LIST_HEAD(&l_mg->free_list);
 794         INIT_LIST_HEAD(&l_mg->corrupt_list);
 795         INIT_LIST_HEAD(&l_mg->bad_list);
 796         INIT_LIST_HEAD(&l_mg->gc_full_list);
 797         INIT_LIST_HEAD(&l_mg->gc_high_list);
 798         INIT_LIST_HEAD(&l_mg->gc_mid_list);
 799         INIT_LIST_HEAD(&l_mg->gc_low_list);
 800         INIT_LIST_HEAD(&l_mg->gc_empty_list);
 801
 802         INIT_LIST_HEAD(&l_mg->emeta_list);
 803
 804         l_mg->gc_lists[0] = &l_mg->gc_high_list;
 805         l_mg->gc_lists[1] = &l_mg->gc_mid_list;
 806         l_mg->gc_lists[2] = &l_mg->gc_low_list;
 807
 808         spin_lock_init(&l_mg->free_lock);
 809         spin_lock_init(&l_mg->close_lock);
 810         spin_lock_init(&l_mg->gc_lock);
 811
 812         pblk->lines = kcalloc(l_mg->nr_lines, sizeof(struct pblk_line),
 813                                                                 GFP_KERNEL);
 814         if (!pblk->lines) {
 815                 ret = -ENOMEM;
 816                 goto fail_free_bb_aux;
 817         }
 818
 819         nr_free_blks = 0;
 820         for (i = 0; i < l_mg->nr_lines; i++) {
 821                 int blk_in_line;
 822
 823                 line = &pblk->lines[i];
 824
 825                 line->pblk = pblk;
 826                 line->id = i;
 827                 line->type = PBLK_LINETYPE_FREE;
 828                 line->state = PBLK_LINESTATE_FREE;
 829                 line->gc_group = PBLK_LINEGC_NONE;
 830                 line->vsc = &l_mg->vsc_list[i];
 831                 spin_lock_init(&line->lock);
 832
 833                 ret = pblk_alloc_line_bitmaps(pblk, line);
 834                 if (ret)
 835                         goto fail_free_lines;
 836
 837                 nr_bad_blks = pblk_bb_line(pblk, line, lm->blk_per_line);
 838                 if (nr_bad_blks < 0 || nr_bad_blks > lm->blk_per_line) {
 839                         pblk_free_line_bitmaps(line);
 840                         ret = -EINVAL;
 841                         goto fail_free_lines;
 842                 }
 843
 844                 blk_in_line = lm->blk_per_line - nr_bad_blks;
 845                 if (blk_in_line < lm->min_blk_line) {
 846                         line->state = PBLK_LINESTATE_BAD;
 847                         list_add_tail(&line->list, &l_mg->bad_list);
 848                         continue;
 849                 }
 850
 851                 nr_free_blks += blk_in_line;
 852                 atomic_set(&line->blk_in_line, blk_in_line);
 853
 854                 l_mg->nr_free_lines++;
 855                 list_add_tail(&line->list, &l_mg->free_list);
 856         }
 857
 858         pblk_set_provision(pblk, nr_free_blks);
 859
 860         /* Cleanup per-LUN bad block lists - managed within lines on run-time */
 861         for (i = 0; i < geo->all_luns; i++)
 862                 kfree(pblk->luns[i].bb_list);
 863
 864         return 0;
 865 fail_free_lines:
 866         while (--i >= 0)
 867                 pblk_free_line_bitmaps(&pblk->lines[i]);
 868 fail_free_bb_aux:
 869         kfree(l_mg->bb_aux);
 870 fail_free_bb_template:
 871         kfree(l_mg->bb_template);
 872 fail_free_meta:
 873         pblk_line_meta_free(pblk);
 874 fail:
 875         for (i = 0; i < geo->all_luns; i++)
 876                 kfree(pblk->luns[i].bb_list);
 877
 878         return ret;
 879 }
 880
 881 static int pblk_writer_init(struct pblk *pblk)
 882 {
 883         timer_setup(&pblk->wtimer, pblk_write_timer_fn, 0);
 884         mod_timer(&pblk->wtimer, jiffies + msecs_to_jiffies(100));
 885
 886         pblk->writer_ts = kthread_create(pblk_write_ts, pblk, "pblk-writer-t");
 887         if (IS_ERR(pblk->writer_ts)) {
 888                 pr_err("pblk: could not allocate writer kthread\n");
 889                 return PTR_ERR(pblk->writer_ts);
 890         }
 891
 892         return 0;
 893 }
 894
 895 static void pblk_writer_stop(struct pblk *pblk)
 896 {
 897         /* The pipeline must be stopped and the write buffer emptied before the
 898          * write thread is stopped
 899          */
 900         WARN(pblk_rb_read_count(&pblk->rwb),
 901                         "Stopping not fully persisted write buffer\n");
 902
 903         WARN(pblk_rb_sync_count(&pblk->rwb),
 904                         "Stopping not fully synced write buffer\n");
 905
 906         if (pblk->writer_ts)
 907                 kthread_stop(pblk->writer_ts);
 908         del_timer(&pblk->wtimer);
 909 }
 910
 911 static void pblk_free(struct pblk *pblk)
 912 {
 913         pblk_luns_free(pblk);
 914         pblk_lines_free(pblk);
 915         pblk_line_meta_free(pblk);
 916         pblk_core_free(pblk);
 917         pblk_l2p_free(pblk);
 918
 919         kfree(pblk);
 920 }
 921
 922 static void pblk_tear_down(struct pblk *pblk)
 923 {
 924         pblk_pipeline_stop(pblk);
 925         pblk_writer_stop(pblk);
 926         pblk_rb_sync_l2p(&pblk->rwb);
 927         pblk_rwb_free(pblk);
 928         pblk_rl_free(&pblk->rl);
 929
 930         pr_debug("pblk: consistent tear down\n");
 931 }
 932
 933 static void pblk_exit(void *private)
 934 {
 935         struct pblk *pblk = private;
 936
 937         down_write(&pblk_lock);
 938         pblk_gc_exit(pblk);
 939         pblk_tear_down(pblk);
 940
 941 #ifdef CONFIG_NVM_DEBUG
 942         pr_info("pblk exit: L2P CRC: %x\n", pblk_l2p_crc(pblk));
 943 #endif
 944
 945         pblk_free(pblk);
 946         up_write(&pblk_lock);
 947 }
 948
 949 static sector_t pblk_capacity(void *private)
 950 {
 951         struct pblk *pblk = private;
 952
 953         return pblk->capacity * NR_PHY_IN_LOG;
 954 }
 955
 956 static void *pblk_init(struct nvm_tgt_dev *dev, struct gendisk *tdisk,
 957                        int flags)
 958 {
 959         struct nvm_geo *geo = &dev->geo;
 960         struct request_queue *bqueue = dev->q;
 961         struct request_queue *tqueue = tdisk->queue;
 962         struct pblk *pblk;
 963         int ret;
 964
 965         if (dev->identity.dom & NVM_RSP_L2P) {
 966                 pr_err("pblk: host-side L2P table not supported. (%x)\n",
 967                                                         dev->identity.dom);
 968                 return ERR_PTR(-EINVAL);
 969         }
 970
 971         pblk = kzalloc(sizeof(struct pblk), GFP_KERNEL);
 972         if (!pblk)
 973                 return ERR_PTR(-ENOMEM);
 974
 975         pblk->dev = dev;
 976         pblk->disk = tdisk;
 977         pblk->state = PBLK_STATE_RUNNING;
 978         pblk->gc.gc_enabled = 0;
 979
 980         spin_lock_init(&pblk->trans_lock);
 981         spin_lock_init(&pblk->lock);
 982
 983         if (flags & NVM_TARGET_FACTORY)
 984                 pblk_setup_uuid(pblk);
 985
 986 #ifdef CONFIG_NVM_DEBUG
 987         atomic_long_set(&pblk->inflight_writes, 0);
 988         atomic_long_set(&pblk->padded_writes, 0);
 989         atomic_long_set(&pblk->padded_wb, 0);
 990         atomic_long_set(&pblk->nr_flush, 0);
 991         atomic_long_set(&pblk->req_writes, 0);
 992         atomic_long_set(&pblk->sub_writes, 0);
 993         atomic_long_set(&pblk->sync_writes, 0);
 994         atomic_long_set(&pblk->inflight_reads, 0);
 995         atomic_long_set(&pblk->cache_reads, 0);
 996         atomic_long_set(&pblk->sync_reads, 0);
 997         atomic_long_set(&pblk->recov_writes, 0);
 998         atomic_long_set(&pblk->recov_writes, 0);
 999         atomic_long_set(&pblk->recov_gc_writes, 0);
1000         atomic_long_set(&pblk->recov_gc_reads, 0);
1001 #endif
1002
1003         atomic_long_set(&pblk->read_failed, 0);
1004         atomic_long_set(&pblk->read_empty, 0);
1005         atomic_long_set(&pblk->read_high_ecc, 0);
1006         atomic_long_set(&pblk->read_failed_gc, 0);
1007         atomic_long_set(&pblk->write_failed, 0);
1008         atomic_long_set(&pblk->erase_failed, 0);
1009
1010         ret = pblk_luns_init(pblk, dev->luns);
1011         if (ret) {
1012                 pr_err("pblk: could not initialize luns\n");
1013                 goto fail;
1014         }
1015
1016         ret = pblk_lines_init(pblk);
1017         if (ret) {
1018                 pr_err("pblk: could not initialize lines\n");
1019                 goto fail_free_luns;
1020         }
1021
1022         ret = pblk_core_init(pblk);
1023         if (ret) {
1024                 pr_err("pblk: could not initialize core\n");
1025                 goto fail_free_line_meta;
1026         }
1027
1028         ret = pblk_l2p_init(pblk);
1029         if (ret) {
1030                 pr_err("pblk: could not initialize maps\n");
1031                 goto fail_free_core;
1032         }
1033
1034         ret = pblk_lines_configure(pblk, flags);
1035         if (ret) {
1036                 pr_err("pblk: could not configure lines\n");
1037                 goto fail_free_l2p;
1038         }
1039
1040         ret = pblk_writer_init(pblk);
1041         if (ret) {
1042                 pr_err("pblk: could not initialize write thread\n");
1043                 goto fail_free_lines;
1044         }
1045
1046         ret = pblk_gc_init(pblk);
1047         if (ret) {
1048                 pr_err("pblk: could not initialize gc\n");
1049                 goto fail_stop_writer;
1050         }
1051
1052         /* inherit the size from the underlying device */
1053         blk_queue_logical_block_size(tqueue, queue_physical_block_size(bqueue));
1054         blk_queue_max_hw_sectors(tqueue, queue_max_hw_sectors(bqueue));
1055
1056         blk_queue_write_cache(tqueue, true, false);
1057
1058         tqueue->limits.discard_granularity = geo->sec_per_chk * geo->sec_size;
1059         tqueue->limits.discard_alignment = 0;
1060         blk_queue_max_discard_sectors(tqueue, UINT_MAX >> 9);
1061         queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, tqueue);
1062
1063         pr_info("pblk init: luns:%u, lines:%d, secs:%llu, buf entries:%u\n",
1064                         geo->all_luns, pblk->l_mg.nr_lines,
1065                         (unsigned long long)pblk->rl.nr_secs,
1066                         pblk->rwb.nr_entries);
1067
1068         wake_up_process(pblk->writer_ts);
1069
1070         /* Check if we need to start GC */
1071         pblk_gc_should_kick(pblk);
1072
1073         return pblk;
1074
1075 fail_stop_writer:
1076         pblk_writer_stop(pblk);
1077 fail_free_lines:
1078         pblk_lines_free(pblk);
1079 fail_free_l2p:
1080         pblk_l2p_free(pblk);
1081 fail_free_core:
1082         pblk_core_free(pblk);
1083 fail_free_line_meta:
1084         pblk_line_meta_free(pblk);
1085 fail_free_luns:
1086         pblk_luns_free(pblk);
1087 fail:
1088         kfree(pblk);
1089         return ERR_PTR(ret);
1090 }
1091
1092 /* physical block device target */
1093 static struct nvm_tgt_type tt_pblk = {
1094         .name           = "pblk",
1095         .version        = {1, 0, 0},
1096
1097         .make_rq        = pblk_make_rq,
1098         .capacity       = pblk_capacity,
1099
1100         .init           = pblk_init,
1101         .exit           = pblk_exit,
1102
1103         .sysfs_init     = pblk_sysfs_init,
1104         .sysfs_exit     = pblk_sysfs_exit,
1105         .owner          = THIS_MODULE,
1106 };
1107
1108 static int __init pblk_module_init(void)
1109 {
1110         int ret;
1111
1112         pblk_bio_set = bioset_create(BIO_POOL_SIZE, 0, 0);
1113         if (!pblk_bio_set)
1114                 return -ENOMEM;
1115         ret = nvm_register_tgt_type(&tt_pblk);
1116         if (ret)
1117                 bioset_free(pblk_bio_set);
1118         return ret;
1119 }
1120
1121 static void pblk_module_exit(void)
1122 {
1123         bioset_free(pblk_bio_set);
1124         nvm_unregister_tgt_type(&tt_pblk);
1125 }
1126
1127 module_init(pblk_module_init);
1128 module_exit(pblk_module_exit);
1129 MODULE_AUTHOR("Javier Gonzalez <javier@cnexlabs.com>");
1130 MODULE_AUTHOR("Matias Bjorling <matias@cnexlabs.com>");
1131 MODULE_LICENSE("GPL v2");
1132 MODULE_DESCRIPTION("Physical Block-Device for Open-Channel SSDs");