drivers/lightnvm/pblk-init.c

   1 /*
   2  * Copyright (C) 2015 IT University of Copenhagen (rrpc.c)
   3  * Copyright (C) 2016 CNEX Labs
   4  * Initial release: Javier Gonzalez <javier@cnexlabs.com>
   5  *                  Matias Bjorling <matias@cnexlabs.com>
   6  *
   7  * This program is free software; you can redistribute it and/or
   8  * modify it under the terms of the GNU General Public License version
   9  * 2 as published by the Free Software Foundation.
  10  *
  11  * This program is distributed in the hope that it will be useful, but
  12  * WITHOUT ANY WARRANTY; without even the implied warranty of
  13  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  14  * General Public License for more details.
  15  *
  16  * Implementation of a physical block-device target for Open-channel SSDs.
  17  *
  18  * pblk-init.c - pblk's initialization.
  19  */
  20
  21 #include "pblk.h"
  22
  23 static struct kmem_cache *pblk_blk_ws_cache, *pblk_rec_cache, *pblk_g_rq_cache,
  24                                 *pblk_w_rq_cache, *pblk_line_meta_cache;
  25 static DECLARE_RWSEM(pblk_lock);
  26 struct bio_set *pblk_bio_set;
  27
  28 static int pblk_rw_io(struct request_queue *q, struct pblk *pblk,
  29                           struct bio *bio)
  30 {
  31         int ret;
  32
  33         /* Read requests must be <= 256kb due to NVMe's 64 bit completion bitmap
  34          * constraint. Writes can be of arbitrary size.
  35          */
  36         if (bio_data_dir(bio) == READ) {
  37                 blk_queue_split(q, &bio);
  38                 ret = pblk_submit_read(pblk, bio);
  39                 if (ret == NVM_IO_DONE && bio_flagged(bio, BIO_CLONED))
  40                         bio_put(bio);
  41
  42                 return ret;
  43         }
  44
  45         /* Prevent deadlock in the case of a modest LUN configuration and large
  46          * user I/Os. Unless stalled, the rate limiter leaves at least 256KB
  47          * available for user I/O.
  48          */
  49         if (unlikely(pblk_get_secs(bio) >= pblk_rl_sysfs_rate_show(&pblk->rl)))
  50                 blk_queue_split(q, &bio);
  51
  52         return pblk_write_to_cache(pblk, bio, PBLK_IOTYPE_USER);
  53 }
  54
  55 static blk_qc_t pblk_make_rq(struct request_queue *q, struct bio *bio)
  56 {
  57         struct pblk *pblk = q->queuedata;
  58
  59         if (bio_op(bio) == REQ_OP_DISCARD) {
  60                 pblk_discard(pblk, bio);
  61                 if (!(bio->bi_opf & REQ_PREFLUSH)) {
  62                         bio_endio(bio);
  63                         return BLK_QC_T_NONE;
  64                 }
  65         }
  66
  67         switch (pblk_rw_io(q, pblk, bio)) {
  68         case NVM_IO_ERR:
  69                 bio_io_error(bio);
  70                 break;
  71         case NVM_IO_DONE:
  72                 bio_endio(bio);
  73                 break;
  74         }
  75
  76         return BLK_QC_T_NONE;
  77 }
  78
  79 static void pblk_l2p_free(struct pblk *pblk)
  80 {
  81         vfree(pblk->trans_map);
  82 }
  83
  84 static int pblk_l2p_init(struct pblk *pblk)
  85 {
  86         sector_t i;
  87         struct ppa_addr ppa;
  88         int entry_size = 8;
  89
  90         if (pblk->ppaf_bitsize < 32)
  91                 entry_size = 4;
  92
  93         pblk->trans_map = vmalloc(entry_size * pblk->rl.nr_secs);
  94         if (!pblk->trans_map)
  95                 return -ENOMEM;
  96
  97         pblk_ppa_set_empty(&ppa);
  98
  99         for (i = 0; i < pblk->rl.nr_secs; i++)
 100                 pblk_trans_map_set(pblk, i, ppa);
 101
 102         return 0;
 103 }
 104
 105 static void pblk_rwb_free(struct pblk *pblk)
 106 {
 107         if (pblk_rb_tear_down_check(&pblk->rwb))
 108                 pr_err("pblk: write buffer error on tear down\n");
 109
 110         pblk_rb_data_free(&pblk->rwb);
 111         vfree(pblk_rb_entries_ref(&pblk->rwb));
 112 }
 113
 114 static int pblk_rwb_init(struct pblk *pblk)
 115 {
 116         struct nvm_tgt_dev *dev = pblk->dev;
 117         struct nvm_geo *geo = &dev->geo;
 118         struct pblk_rb_entry *entries;
 119         unsigned long nr_entries;
 120         unsigned int power_size, power_seg_sz;
 121
 122         nr_entries = pblk_rb_calculate_size(pblk->pgs_in_buffer);
 123
 124         entries = vzalloc(nr_entries * sizeof(struct pblk_rb_entry));
 125         if (!entries)
 126                 return -ENOMEM;
 127
 128         power_size = get_count_order(nr_entries);
 129         power_seg_sz = get_count_order(geo->sec_size);
 130
 131         return pblk_rb_init(&pblk->rwb, entries, power_size, power_seg_sz);
 132 }
 133
 134 /* Minimum pages needed within a lun */
 135 #define PAGE_POOL_SIZE 16
 136 #define ADDR_POOL_SIZE 64
 137
 138 static int pblk_set_ppaf(struct pblk *pblk)
 139 {
 140         struct nvm_tgt_dev *dev = pblk->dev;
 141         struct nvm_geo *geo = &dev->geo;
 142         struct nvm_addr_format ppaf = geo->ppaf;
 143         int power_len;
 144
 145         /* Re-calculate channel and lun format to adapt to configuration */
 146         power_len = get_count_order(geo->nr_chnls);
 147         if (1 << power_len != geo->nr_chnls) {
 148                 pr_err("pblk: supports only power-of-two channel config.\n");
 149                 return -EINVAL;
 150         }
 151         ppaf.ch_len = power_len;
 152
 153         power_len = get_count_order(geo->luns_per_chnl);
 154         if (1 << power_len != geo->luns_per_chnl) {
 155                 pr_err("pblk: supports only power-of-two LUN config.\n");
 156                 return -EINVAL;
 157         }
 158         ppaf.lun_len = power_len;
 159
 160         pblk->ppaf.sec_offset = 0;
 161         pblk->ppaf.pln_offset = ppaf.sect_len;
 162         pblk->ppaf.ch_offset = pblk->ppaf.pln_offset + ppaf.pln_len;
 163         pblk->ppaf.lun_offset = pblk->ppaf.ch_offset + ppaf.ch_len;
 164         pblk->ppaf.pg_offset = pblk->ppaf.lun_offset + ppaf.lun_len;
 165         pblk->ppaf.blk_offset = pblk->ppaf.pg_offset + ppaf.pg_len;
 166         pblk->ppaf.sec_mask = (1ULL << ppaf.sect_len) - 1;
 167         pblk->ppaf.pln_mask = ((1ULL << ppaf.pln_len) - 1) <<
 168                                                         pblk->ppaf.pln_offset;
 169         pblk->ppaf.ch_mask = ((1ULL << ppaf.ch_len) - 1) <<
 170                                                         pblk->ppaf.ch_offset;
 171         pblk->ppaf.lun_mask = ((1ULL << ppaf.lun_len) - 1) <<
 172                                                         pblk->ppaf.lun_offset;
 173         pblk->ppaf.pg_mask = ((1ULL << ppaf.pg_len) - 1) <<
 174                                                         pblk->ppaf.pg_offset;
 175         pblk->ppaf.blk_mask = ((1ULL << ppaf.blk_len) - 1) <<
 176                                                         pblk->ppaf.blk_offset;
 177
 178         pblk->ppaf_bitsize = pblk->ppaf.blk_offset + ppaf.blk_len;
 179
 180         return 0;
 181 }
 182
 183 static int pblk_init_global_caches(struct pblk *pblk)
 184 {
 185         char cache_name[PBLK_CACHE_NAME_LEN];
 186
 187         down_write(&pblk_lock);
 188         pblk_blk_ws_cache = kmem_cache_create("pblk_blk_ws",
 189                                 sizeof(struct pblk_line_ws), 0, 0, NULL);
 190         if (!pblk_blk_ws_cache) {
 191                 up_write(&pblk_lock);
 192                 return -ENOMEM;
 193         }
 194
 195         pblk_rec_cache = kmem_cache_create("pblk_rec",
 196                                 sizeof(struct pblk_rec_ctx), 0, 0, NULL);
 197         if (!pblk_rec_cache) {
 198                 kmem_cache_destroy(pblk_blk_ws_cache);
 199                 up_write(&pblk_lock);
 200                 return -ENOMEM;
 201         }
 202
 203         pblk_g_rq_cache = kmem_cache_create("pblk_g_rq", pblk_g_rq_size,
 204                                 0, 0, NULL);
 205         if (!pblk_g_rq_cache) {
 206                 kmem_cache_destroy(pblk_blk_ws_cache);
 207                 kmem_cache_destroy(pblk_rec_cache);
 208                 up_write(&pblk_lock);
 209                 return -ENOMEM;
 210         }
 211
 212         pblk_w_rq_cache = kmem_cache_create("pblk_w_rq", pblk_w_rq_size,
 213                                 0, 0, NULL);
 214         if (!pblk_w_rq_cache) {
 215                 kmem_cache_destroy(pblk_blk_ws_cache);
 216                 kmem_cache_destroy(pblk_rec_cache);
 217                 kmem_cache_destroy(pblk_g_rq_cache);
 218                 up_write(&pblk_lock);
 219                 return -ENOMEM;
 220         }
 221
 222         snprintf(cache_name, sizeof(cache_name), "pblk_line_m_%s",
 223                                                         pblk->disk->disk_name);
 224         pblk_line_meta_cache = kmem_cache_create(cache_name,
 225                                 pblk->lm.sec_bitmap_len, 0, 0, NULL);
 226         if (!pblk_line_meta_cache) {
 227                 kmem_cache_destroy(pblk_blk_ws_cache);
 228                 kmem_cache_destroy(pblk_rec_cache);
 229                 kmem_cache_destroy(pblk_g_rq_cache);
 230                 kmem_cache_destroy(pblk_w_rq_cache);
 231                 up_write(&pblk_lock);
 232                 return -ENOMEM;
 233         }
 234         up_write(&pblk_lock);
 235
 236         return 0;
 237 }
 238
 239 static int pblk_core_init(struct pblk *pblk)
 240 {
 241         struct nvm_tgt_dev *dev = pblk->dev;
 242         struct nvm_geo *geo = &dev->geo;
 243
 244         pblk->pgs_in_buffer = NVM_MEM_PAGE_WRITE * geo->sec_per_pg *
 245                                                 geo->nr_planes * geo->nr_luns;
 246
 247         if (pblk_init_global_caches(pblk))
 248                 return -ENOMEM;
 249
 250         pblk->page_pool = mempool_create_page_pool(PAGE_POOL_SIZE, 0);
 251         if (!pblk->page_pool)
 252                 return -ENOMEM;
 253
 254         pblk->line_ws_pool = mempool_create_slab_pool(geo->nr_luns,
 255                                                         pblk_blk_ws_cache);
 256         if (!pblk->line_ws_pool)
 257                 goto free_page_pool;
 258
 259         pblk->rec_pool = mempool_create_slab_pool(geo->nr_luns, pblk_rec_cache);
 260         if (!pblk->rec_pool)
 261                 goto free_blk_ws_pool;
 262
 263         pblk->g_rq_pool = mempool_create_slab_pool(64, pblk_g_rq_cache);
 264         if (!pblk->g_rq_pool)
 265                 goto free_rec_pool;
 266
 267         pblk->w_rq_pool = mempool_create_slab_pool(64, pblk_w_rq_cache);
 268         if (!pblk->w_rq_pool)
 269                 goto free_g_rq_pool;
 270
 271         pblk->line_meta_pool =
 272                         mempool_create_slab_pool(16, pblk_line_meta_cache);
 273         if (!pblk->line_meta_pool)
 274                 goto free_w_rq_pool;
 275
 276         pblk->kw_wq = alloc_workqueue("pblk-aux-wq",
 277                                         WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
 278         if (!pblk->kw_wq)
 279                 goto free_line_meta_pool;
 280
 281         if (pblk_set_ppaf(pblk))
 282                 goto free_kw_wq;
 283
 284         if (pblk_rwb_init(pblk))
 285                 goto free_kw_wq;
 286
 287         INIT_LIST_HEAD(&pblk->compl_list);
 288         return 0;
 289
 290 free_kw_wq:
 291         destroy_workqueue(pblk->kw_wq);
 292 free_line_meta_pool:
 293         mempool_destroy(pblk->line_meta_pool);
 294 free_w_rq_pool:
 295         mempool_destroy(pblk->w_rq_pool);
 296 free_g_rq_pool:
 297         mempool_destroy(pblk->g_rq_pool);
 298 free_rec_pool:
 299         mempool_destroy(pblk->rec_pool);
 300 free_blk_ws_pool:
 301         mempool_destroy(pblk->line_ws_pool);
 302 free_page_pool:
 303         mempool_destroy(pblk->page_pool);
 304         return -ENOMEM;
 305 }
 306
 307 static void pblk_core_free(struct pblk *pblk)
 308 {
 309         if (pblk->kw_wq)
 310                 destroy_workqueue(pblk->kw_wq);
 311
 312         mempool_destroy(pblk->page_pool);
 313         mempool_destroy(pblk->line_ws_pool);
 314         mempool_destroy(pblk->rec_pool);
 315         mempool_destroy(pblk->g_rq_pool);
 316         mempool_destroy(pblk->w_rq_pool);
 317         mempool_destroy(pblk->line_meta_pool);
 318
 319         kmem_cache_destroy(pblk_blk_ws_cache);
 320         kmem_cache_destroy(pblk_rec_cache);
 321         kmem_cache_destroy(pblk_g_rq_cache);
 322         kmem_cache_destroy(pblk_w_rq_cache);
 323         kmem_cache_destroy(pblk_line_meta_cache);
 324 }
 325
 326 static void pblk_luns_free(struct pblk *pblk)
 327 {
 328         kfree(pblk->luns);
 329 }
 330
 331 static void pblk_lines_free(struct pblk *pblk)
 332 {
 333         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
 334         struct pblk_line *line;
 335         int i;
 336
 337         spin_lock(&l_mg->free_lock);
 338         for (i = 0; i < l_mg->nr_lines; i++) {
 339                 line = &pblk->lines[i];
 340
 341                 pblk_line_free(pblk, line);
 342                 kfree(line->blk_bitmap);
 343                 kfree(line->erase_bitmap);
 344         }
 345         spin_unlock(&l_mg->free_lock);
 346 }
 347
 348 static void pblk_line_meta_free(struct pblk *pblk)
 349 {
 350         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
 351         int i;
 352
 353         kfree(l_mg->bb_template);
 354         kfree(l_mg->bb_aux);
 355         kfree(l_mg->vsc_list);
 356
 357         for (i = 0; i < PBLK_DATA_LINES; i++) {
 358                 pblk_mfree(&l_mg->sline_meta[i], l_mg->smeta_alloc_type);
 359                 pblk_mfree(l_mg->eline_meta[i]->buf, l_mg->emeta_alloc_type);
 360                 kfree(&l_mg->eline_meta[i]);
 361         }
 362
 363         kfree(pblk->lines);
 364 }
 365
 366 static int pblk_bb_discovery(struct nvm_tgt_dev *dev, struct pblk_lun *rlun)
 367 {
 368         struct nvm_geo *geo = &dev->geo;
 369         struct ppa_addr ppa;
 370         u8 *blks;
 371         int nr_blks, ret;
 372
 373         nr_blks = geo->blks_per_lun * geo->plane_mode;
 374         blks = kmalloc(nr_blks, GFP_KERNEL);
 375         if (!blks)
 376                 return -ENOMEM;
 377
 378         ppa.ppa = 0;
 379         ppa.g.ch = rlun->bppa.g.ch;
 380         ppa.g.lun = rlun->bppa.g.lun;
 381
 382         ret = nvm_get_tgt_bb_tbl(dev, ppa, blks);
 383         if (ret)
 384                 goto out;
 385
 386         nr_blks = nvm_bb_tbl_fold(dev->parent, blks, nr_blks);
 387         if (nr_blks < 0) {
 388                 ret = nr_blks;
 389                 goto out;
 390         }
 391
 392         rlun->bb_list = blks;
 393
 394         return 0;
 395 out:
 396         kfree(blks);
 397         return ret;
 398 }
 399
 400 static int pblk_bb_line(struct pblk *pblk, struct nvm_geo *geo,
 401                         struct pblk_line *line)
 402 {
 403         struct pblk_line_meta *lm = &pblk->lm;
 404         struct pblk_lun *rlun;
 405         int bb_cnt = 0;
 406         int i;
 407
 408         line->blk_bitmap = kzalloc(lm->blk_bitmap_len, GFP_KERNEL);
 409         if (!line->blk_bitmap)
 410                 return -ENOMEM;
 411
 412         line->erase_bitmap = kzalloc(lm->blk_bitmap_len, GFP_KERNEL);
 413         if (!line->erase_bitmap) {
 414                 kfree(line->blk_bitmap);
 415                 return -ENOMEM;
 416         }
 417
 418         for (i = 0; i < lm->blk_per_line; i++) {
 419                 rlun = &pblk->luns[i];
 420                 if (rlun->bb_list[line->id] == NVM_BLK_T_FREE)
 421                         continue;
 422
 423                 set_bit(pblk_ppa_to_pos(geo, rlun->bppa), line->blk_bitmap);
 424                 bb_cnt++;
 425         }
 426
 427         return bb_cnt;
 428 }
 429
 430 static int pblk_luns_init(struct pblk *pblk, struct ppa_addr *luns)
 431 {
 432         struct nvm_tgt_dev *dev = pblk->dev;
 433         struct nvm_geo *geo = &dev->geo;
 434         struct pblk_lun *rlun;
 435         int i, ret;
 436
 437         /* TODO: Implement unbalanced LUN support */
 438         if (geo->luns_per_chnl < 0) {
 439                 pr_err("pblk: unbalanced LUN config.\n");
 440                 return -EINVAL;
 441         }
 442
 443         pblk->luns = kcalloc(geo->nr_luns, sizeof(struct pblk_lun), GFP_KERNEL);
 444         if (!pblk->luns)
 445                 return -ENOMEM;
 446
 447         for (i = 0; i < geo->nr_luns; i++) {
 448                 /* Stripe across channels */
 449                 int ch = i % geo->nr_chnls;
 450                 int lun_raw = i / geo->nr_chnls;
 451                 int lunid = lun_raw + ch * geo->luns_per_chnl;
 452
 453                 rlun = &pblk->luns[i];
 454                 rlun->bppa = luns[lunid];
 455
 456                 sema_init(&rlun->wr_sem, 1);
 457
 458                 ret = pblk_bb_discovery(dev, rlun);
 459                 if (ret) {
 460                         while (--i >= 0)
 461                                 kfree(pblk->luns[i].bb_list);
 462                         return ret;
 463                 }
 464         }
 465
 466         return 0;
 467 }
 468
 469 static int pblk_lines_configure(struct pblk *pblk, int flags)
 470 {
 471         struct pblk_line *line = NULL;
 472         int ret = 0;
 473
 474         if (!(flags & NVM_TARGET_FACTORY)) {
 475                 line = pblk_recov_l2p(pblk);
 476                 if (IS_ERR(line)) {
 477                         pr_err("pblk: could not recover l2p table\n");
 478                         ret = -EFAULT;
 479                 }
 480         }
 481
 482         if (!line) {
 483                 /* Configure next line for user data */
 484                 line = pblk_line_get_first_data(pblk);
 485                 if (!line) {
 486                         pr_err("pblk: line list corrupted\n");
 487                         ret = -EFAULT;
 488                 }
 489         }
 490
 491         return ret;
 492 }
 493
 494 /* See comment over struct line_emeta definition */
 495 static unsigned int calc_emeta_len(struct pblk *pblk)
 496 {
 497         struct pblk_line_meta *lm = &pblk->lm;
 498         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
 499         struct nvm_tgt_dev *dev = pblk->dev;
 500         struct nvm_geo *geo = &dev->geo;
 501
 502         /* Round to sector size so that lba_list starts on its own sector */
 503         lm->emeta_sec[1] = DIV_ROUND_UP(
 504                         sizeof(struct line_emeta) + lm->blk_bitmap_len,
 505                         geo->sec_size);
 506         lm->emeta_len[1] = lm->emeta_sec[1] * geo->sec_size;
 507
 508         /* Round to sector size so that vsc_list starts on its own sector */
 509         lm->dsec_per_line = lm->sec_per_line - lm->emeta_sec[0];
 510         lm->emeta_sec[2] = DIV_ROUND_UP(lm->dsec_per_line * sizeof(u64),
 511                         geo->sec_size);
 512         lm->emeta_len[2] = lm->emeta_sec[2] * geo->sec_size;
 513
 514         lm->emeta_sec[3] = DIV_ROUND_UP(l_mg->nr_lines * sizeof(u32),
 515                         geo->sec_size);
 516         lm->emeta_len[3] = lm->emeta_sec[3] * geo->sec_size;
 517
 518         lm->vsc_list_len = l_mg->nr_lines * sizeof(u32);
 519
 520         return (lm->emeta_len[1] + lm->emeta_len[2] + lm->emeta_len[3]);
 521 }
 522
 523 static void pblk_set_provision(struct pblk *pblk, long nr_free_blks)
 524 {
 525         struct nvm_tgt_dev *dev = pblk->dev;
 526         struct nvm_geo *geo = &dev->geo;
 527         sector_t provisioned;
 528
 529         pblk->over_pct = 20;
 530
 531         provisioned = nr_free_blks;
 532         provisioned *= (100 - pblk->over_pct);
 533         sector_div(provisioned, 100);
 534
 535         /* Internally pblk manages all free blocks, but all calculations based
 536          * on user capacity consider only provisioned blocks
 537          */
 538         pblk->rl.total_blocks = nr_free_blks;
 539         pblk->rl.nr_secs = nr_free_blks * geo->sec_per_blk;
 540         pblk->capacity = provisioned * geo->sec_per_blk;
 541         atomic_set(&pblk->rl.free_blocks, nr_free_blks);
 542 }
 543
 544 static int pblk_lines_alloc_metadata(struct pblk *pblk)
 545 {
 546         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
 547         struct pblk_line_meta *lm = &pblk->lm;
 548         int i;
 549
 550         /* smeta is always small enough to fit on a kmalloc memory allocation,
 551          * emeta depends on the number of LUNs allocated to the pblk instance
 552          */
 553         l_mg->smeta_alloc_type = PBLK_KMALLOC_META;
 554         for (i = 0; i < PBLK_DATA_LINES; i++) {
 555                 l_mg->sline_meta[i] = kmalloc(lm->smeta_len, GFP_KERNEL);
 556                 if (!l_mg->sline_meta[i])
 557                         goto fail_free_smeta;
 558         }
 559
 560         /* emeta allocates three different buffers for managing metadata with
 561          * in-memory and in-media layouts
 562          */
 563         for (i = 0; i < PBLK_DATA_LINES; i++) {
 564                 struct pblk_emeta *emeta;
 565
 566                 emeta = kmalloc(sizeof(struct pblk_emeta), GFP_KERNEL);
 567                 if (!emeta)
 568                         goto fail_free_emeta;
 569
 570                 if (lm->emeta_len[0] > KMALLOC_MAX_CACHE_SIZE) {
 571                         l_mg->emeta_alloc_type = PBLK_VMALLOC_META;
 572
 573                         emeta->buf = vmalloc(lm->emeta_len[0]);
 574                         if (!emeta->buf) {
 575                                 kfree(emeta);
 576                                 goto fail_free_emeta;
 577                         }
 578
 579                         emeta->nr_entries = lm->emeta_sec[0];
 580                         l_mg->eline_meta[i] = emeta;
 581                 } else {
 582                         l_mg->emeta_alloc_type = PBLK_KMALLOC_META;
 583
 584                         emeta->buf = kmalloc(lm->emeta_len[0], GFP_KERNEL);
 585                         if (!emeta->buf) {
 586                                 kfree(emeta);
 587                                 goto fail_free_emeta;
 588                         }
 589
 590                         emeta->nr_entries = lm->emeta_sec[0];
 591                         l_mg->eline_meta[i] = emeta;
 592                 }
 593         }
 594
 595         l_mg->vsc_list = kcalloc(l_mg->nr_lines, sizeof(__le32), GFP_KERNEL);
 596         if (!l_mg->vsc_list)
 597                 goto fail_free_emeta;
 598
 599         for (i = 0; i < l_mg->nr_lines; i++)
 600                 l_mg->vsc_list[i] = cpu_to_le32(EMPTY_ENTRY);
 601
 602         return 0;
 603
 604 fail_free_emeta:
 605         while (--i >= 0) {
 606                 vfree(l_mg->eline_meta[i]->buf);
 607                 kfree(&l_mg->eline_meta[i]);
 608         }
 609
 610 fail_free_smeta:
 611         for (i = 0; i < PBLK_DATA_LINES; i++)
 612                 pblk_mfree(&l_mg->sline_meta[i], l_mg->smeta_alloc_type);
 613
 614         return -ENOMEM;
 615 }
 616
 617 static int pblk_lines_init(struct pblk *pblk)
 618 {
 619         struct nvm_tgt_dev *dev = pblk->dev;
 620         struct nvm_geo *geo = &dev->geo;
 621         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
 622         struct pblk_line_meta *lm = &pblk->lm;
 623         struct pblk_line *line;
 624         unsigned int smeta_len, emeta_len;
 625         long nr_bad_blks, nr_free_blks;
 626         int bb_distance, max_write_ppas, mod;
 627         int i, ret;
 628
 629         pblk->min_write_pgs = geo->sec_per_pl * (geo->sec_size / PAGE_SIZE);
 630         max_write_ppas = pblk->min_write_pgs * geo->nr_luns;
 631         pblk->max_write_pgs = (max_write_ppas < nvm_max_phys_sects(dev)) ?
 632                                 max_write_ppas : nvm_max_phys_sects(dev);
 633         pblk_set_sec_per_write(pblk, pblk->min_write_pgs);
 634
 635         if (pblk->max_write_pgs > PBLK_MAX_REQ_ADDRS) {
 636                 pr_err("pblk: cannot support device max_phys_sect\n");
 637                 return -EINVAL;
 638         }
 639
 640         div_u64_rem(geo->sec_per_blk, pblk->min_write_pgs, &mod);
 641         if (mod) {
 642                 pr_err("pblk: bad configuration of sectors/pages\n");
 643                 return -EINVAL;
 644         }
 645
 646         l_mg->nr_lines = geo->blks_per_lun;
 647         l_mg->log_line = l_mg->data_line = NULL;
 648         l_mg->l_seq_nr = l_mg->d_seq_nr = 0;
 649         l_mg->nr_free_lines = 0;
 650         bitmap_zero(&l_mg->meta_bitmap, PBLK_DATA_LINES);
 651
 652         lm->sec_per_line = geo->sec_per_blk * geo->nr_luns;
 653         lm->blk_per_line = geo->nr_luns;
 654         lm->blk_bitmap_len = BITS_TO_LONGS(geo->nr_luns) * sizeof(long);
 655         lm->sec_bitmap_len = BITS_TO_LONGS(lm->sec_per_line) * sizeof(long);
 656         lm->lun_bitmap_len = BITS_TO_LONGS(geo->nr_luns) * sizeof(long);
 657         lm->high_thrs = lm->sec_per_line / 2;
 658         lm->mid_thrs = lm->sec_per_line / 4;
 659         lm->meta_distance = (geo->nr_luns / 2) * pblk->min_write_pgs;
 660
 661         /* Calculate necessary pages for smeta. See comment over struct
 662          * line_smeta definition
 663          */
 664         i = 1;
 665 add_smeta_page:
 666         lm->smeta_sec = i * geo->sec_per_pl;
 667         lm->smeta_len = lm->smeta_sec * geo->sec_size;
 668
 669         smeta_len = sizeof(struct line_smeta) + lm->lun_bitmap_len;
 670         if (smeta_len > lm->smeta_len) {
 671                 i++;
 672                 goto add_smeta_page;
 673         }
 674
 675         /* Calculate necessary pages for emeta. See comment over struct
 676          * line_emeta definition
 677          */
 678         i = 1;
 679 add_emeta_page:
 680         lm->emeta_sec[0] = i * geo->sec_per_pl;
 681         lm->emeta_len[0] = lm->emeta_sec[0] * geo->sec_size;
 682
 683         emeta_len = calc_emeta_len(pblk);
 684         if (emeta_len > lm->emeta_len[0]) {
 685                 i++;
 686                 goto add_emeta_page;
 687         }
 688
 689         lm->emeta_bb = geo->nr_luns - i;
 690         lm->min_blk_line = 1 + DIV_ROUND_UP(lm->smeta_sec + lm->emeta_sec[0],
 691                                                         geo->sec_per_blk);
 692
 693         ret = pblk_lines_alloc_metadata(pblk);
 694         if (ret)
 695                 goto fail;
 696
 697         l_mg->bb_template = kzalloc(lm->sec_bitmap_len, GFP_KERNEL);
 698         if (!l_mg->bb_template) {
 699                 ret = -ENOMEM;
 700                 goto fail_free_meta;
 701         }
 702
 703         l_mg->bb_aux = kzalloc(lm->sec_bitmap_len, GFP_KERNEL);
 704         if (!l_mg->bb_aux) {
 705                 ret = -ENOMEM;
 706                 goto fail_free_bb_template;
 707         }
 708
 709         bb_distance = (geo->nr_luns) * geo->sec_per_pl;
 710         for (i = 0; i < lm->sec_per_line; i += bb_distance)
 711                 bitmap_set(l_mg->bb_template, i, geo->sec_per_pl);
 712
 713         INIT_LIST_HEAD(&l_mg->free_list);
 714         INIT_LIST_HEAD(&l_mg->corrupt_list);
 715         INIT_LIST_HEAD(&l_mg->bad_list);
 716         INIT_LIST_HEAD(&l_mg->gc_full_list);
 717         INIT_LIST_HEAD(&l_mg->gc_high_list);
 718         INIT_LIST_HEAD(&l_mg->gc_mid_list);
 719         INIT_LIST_HEAD(&l_mg->gc_low_list);
 720         INIT_LIST_HEAD(&l_mg->gc_empty_list);
 721
 722         INIT_LIST_HEAD(&l_mg->emeta_list);
 723
 724         l_mg->gc_lists[0] = &l_mg->gc_high_list;
 725         l_mg->gc_lists[1] = &l_mg->gc_mid_list;
 726         l_mg->gc_lists[2] = &l_mg->gc_low_list;
 727
 728         spin_lock_init(&l_mg->free_lock);
 729         spin_lock_init(&l_mg->close_lock);
 730         spin_lock_init(&l_mg->gc_lock);
 731
 732         pblk->lines = kcalloc(l_mg->nr_lines, sizeof(struct pblk_line),
 733                                                                 GFP_KERNEL);
 734         if (!pblk->lines) {
 735                 ret = -ENOMEM;
 736                 goto fail_free_bb_aux;
 737         }
 738
 739         nr_free_blks = 0;
 740         for (i = 0; i < l_mg->nr_lines; i++) {
 741                 int blk_in_line;
 742
 743                 line = &pblk->lines[i];
 744
 745                 line->pblk = pblk;
 746                 line->id = i;
 747                 line->type = PBLK_LINETYPE_FREE;
 748                 line->state = PBLK_LINESTATE_FREE;
 749                 line->gc_group = PBLK_LINEGC_NONE;
 750                 line->vsc = &l_mg->vsc_list[i];
 751                 spin_lock_init(&line->lock);
 752
 753                 nr_bad_blks = pblk_bb_line(pblk, geo, line);
 754                 if (nr_bad_blks < 0 || nr_bad_blks > lm->blk_per_line) {
 755                         ret = -EINVAL;
 756                         goto fail_free_lines;
 757                 }
 758
 759                 blk_in_line = lm->blk_per_line - nr_bad_blks;
 760                 if (blk_in_line < lm->min_blk_line) {
 761                         line->state = PBLK_LINESTATE_BAD;
 762                         list_add_tail(&line->list, &l_mg->bad_list);
 763                         continue;
 764                 }
 765
 766                 nr_free_blks += blk_in_line;
 767                 atomic_set(&line->blk_in_line, blk_in_line);
 768
 769                 l_mg->nr_free_lines++;
 770                 list_add_tail(&line->list, &l_mg->free_list);
 771         }
 772
 773         pblk_set_provision(pblk, nr_free_blks);
 774
 775         /* Cleanup per-LUN bad block lists - managed within lines on run-time */
 776         for (i = 0; i < geo->nr_luns; i++)
 777                 kfree(pblk->luns[i].bb_list);
 778
 779         return 0;
 780 fail_free_lines:
 781         kfree(pblk->lines);
 782 fail_free_bb_aux:
 783         kfree(l_mg->bb_aux);
 784 fail_free_bb_template:
 785         kfree(l_mg->bb_template);
 786 fail_free_meta:
 787         pblk_line_meta_free(pblk);
 788 fail:
 789         for (i = 0; i < geo->nr_luns; i++)
 790                 kfree(pblk->luns[i].bb_list);
 791
 792         return ret;
 793 }
 794
 795 static int pblk_writer_init(struct pblk *pblk)
 796 {
 797         setup_timer(&pblk->wtimer, pblk_write_timer_fn, (unsigned long)pblk);
 798         mod_timer(&pblk->wtimer, jiffies + msecs_to_jiffies(100));
 799
 800         pblk->writer_ts = kthread_create(pblk_write_ts, pblk, "pblk-writer-t");
 801         if (IS_ERR(pblk->writer_ts)) {
 802                 pr_err("pblk: could not allocate writer kthread\n");
 803                 return PTR_ERR(pblk->writer_ts);
 804         }
 805
 806         return 0;
 807 }
 808
 809 static void pblk_writer_stop(struct pblk *pblk)
 810 {
 811         if (pblk->writer_ts)
 812                 kthread_stop(pblk->writer_ts);
 813         del_timer(&pblk->wtimer);
 814 }
 815
 816 static void pblk_free(struct pblk *pblk)
 817 {
 818         pblk_luns_free(pblk);
 819         pblk_lines_free(pblk);
 820         pblk_line_meta_free(pblk);
 821         pblk_core_free(pblk);
 822         pblk_l2p_free(pblk);
 823
 824         kfree(pblk);
 825 }
 826
 827 static void pblk_tear_down(struct pblk *pblk)
 828 {
 829         pblk_flush_writer(pblk);
 830         pblk_writer_stop(pblk);
 831         pblk_rb_sync_l2p(&pblk->rwb);
 832         pblk_recov_pad(pblk);
 833         pblk_rwb_free(pblk);
 834         pblk_rl_free(&pblk->rl);
 835
 836         pr_debug("pblk: consistent tear down\n");
 837 }
 838
 839 static void pblk_exit(void *private)
 840 {
 841         struct pblk *pblk = private;
 842
 843         down_write(&pblk_lock);
 844         pblk_gc_exit(pblk);
 845         pblk_tear_down(pblk);
 846         pblk_free(pblk);
 847         up_write(&pblk_lock);
 848 }
 849
 850 static sector_t pblk_capacity(void *private)
 851 {
 852         struct pblk *pblk = private;
 853
 854         return pblk->capacity * NR_PHY_IN_LOG;
 855 }
 856
 857 static void *pblk_init(struct nvm_tgt_dev *dev, struct gendisk *tdisk,
 858                        int flags)
 859 {
 860         struct nvm_geo *geo = &dev->geo;
 861         struct request_queue *bqueue = dev->q;
 862         struct request_queue *tqueue = tdisk->queue;
 863         struct pblk *pblk;
 864         int ret;
 865
 866         if (dev->identity.dom & NVM_RSP_L2P) {
 867                 pr_err("pblk: device-side L2P table not supported. (%x)\n",
 868                                                         dev->identity.dom);
 869                 return ERR_PTR(-EINVAL);
 870         }
 871
 872         pblk = kzalloc(sizeof(struct pblk), GFP_KERNEL);
 873         if (!pblk)
 874                 return ERR_PTR(-ENOMEM);
 875
 876         pblk->dev = dev;
 877         pblk->disk = tdisk;
 878
 879         spin_lock_init(&pblk->trans_lock);
 880         spin_lock_init(&pblk->lock);
 881
 882         if (flags & NVM_TARGET_FACTORY)
 883                 pblk_setup_uuid(pblk);
 884
 885 #ifdef CONFIG_NVM_DEBUG
 886         atomic_long_set(&pblk->inflight_writes, 0);
 887         atomic_long_set(&pblk->padded_writes, 0);
 888         atomic_long_set(&pblk->padded_wb, 0);
 889         atomic_long_set(&pblk->nr_flush, 0);
 890         atomic_long_set(&pblk->req_writes, 0);
 891         atomic_long_set(&pblk->sub_writes, 0);
 892         atomic_long_set(&pblk->sync_writes, 0);
 893         atomic_long_set(&pblk->compl_writes, 0);
 894         atomic_long_set(&pblk->inflight_reads, 0);
 895         atomic_long_set(&pblk->cache_reads, 0);
 896         atomic_long_set(&pblk->sync_reads, 0);
 897         atomic_long_set(&pblk->recov_writes, 0);
 898         atomic_long_set(&pblk->recov_writes, 0);
 899         atomic_long_set(&pblk->recov_gc_writes, 0);
 900 #endif
 901
 902         atomic_long_set(&pblk->read_failed, 0);
 903         atomic_long_set(&pblk->read_empty, 0);
 904         atomic_long_set(&pblk->read_high_ecc, 0);
 905         atomic_long_set(&pblk->read_failed_gc, 0);
 906         atomic_long_set(&pblk->write_failed, 0);
 907         atomic_long_set(&pblk->erase_failed, 0);
 908
 909         ret = pblk_luns_init(pblk, dev->luns);
 910         if (ret) {
 911                 pr_err("pblk: could not initialize luns\n");
 912                 goto fail;
 913         }
 914
 915         ret = pblk_lines_init(pblk);
 916         if (ret) {
 917                 pr_err("pblk: could not initialize lines\n");
 918                 goto fail_free_luns;
 919         }
 920
 921         ret = pblk_core_init(pblk);
 922         if (ret) {
 923                 pr_err("pblk: could not initialize core\n");
 924                 goto fail_free_line_meta;
 925         }
 926
 927         ret = pblk_l2p_init(pblk);
 928         if (ret) {
 929                 pr_err("pblk: could not initialize maps\n");
 930                 goto fail_free_core;
 931         }
 932
 933         ret = pblk_lines_configure(pblk, flags);
 934         if (ret) {
 935                 pr_err("pblk: could not configure lines\n");
 936                 goto fail_free_l2p;
 937         }
 938
 939         ret = pblk_writer_init(pblk);
 940         if (ret) {
 941                 pr_err("pblk: could not initialize write thread\n");
 942                 goto fail_free_lines;
 943         }
 944
 945         ret = pblk_gc_init(pblk);
 946         if (ret) {
 947                 pr_err("pblk: could not initialize gc\n");
 948                 goto fail_stop_writer;
 949         }
 950
 951         /* inherit the size from the underlying device */
 952         blk_queue_logical_block_size(tqueue, queue_physical_block_size(bqueue));
 953         blk_queue_max_hw_sectors(tqueue, queue_max_hw_sectors(bqueue));
 954
 955         blk_queue_write_cache(tqueue, true, false);
 956
 957         tqueue->limits.discard_granularity = geo->pgs_per_blk * geo->pfpg_size;
 958         tqueue->limits.discard_alignment = 0;
 959         blk_queue_max_discard_sectors(tqueue, UINT_MAX >> 9);
 960         queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, tqueue);
 961
 962         pr_info("pblk init: luns:%u, lines:%d, secs:%llu, buf entries:%u\n",
 963                         geo->nr_luns, pblk->l_mg.nr_lines,
 964                         (unsigned long long)pblk->rl.nr_secs,
 965                         pblk->rwb.nr_entries);
 966
 967         wake_up_process(pblk->writer_ts);
 968         return pblk;
 969
 970 fail_stop_writer:
 971         pblk_writer_stop(pblk);
 972 fail_free_lines:
 973         pblk_lines_free(pblk);
 974 fail_free_l2p:
 975         pblk_l2p_free(pblk);
 976 fail_free_core:
 977         pblk_core_free(pblk);
 978 fail_free_line_meta:
 979         pblk_line_meta_free(pblk);
 980 fail_free_luns:
 981         pblk_luns_free(pblk);
 982 fail:
 983         kfree(pblk);
 984         return ERR_PTR(ret);
 985 }
 986
 987 /* physical block device target */
 988 static struct nvm_tgt_type tt_pblk = {
 989         .name           = "pblk",
 990         .version        = {1, 0, 0},
 991
 992         .make_rq        = pblk_make_rq,
 993         .capacity       = pblk_capacity,
 994
 995         .init           = pblk_init,
 996         .exit           = pblk_exit,
 997
 998         .sysfs_init     = pblk_sysfs_init,
 999         .sysfs_exit     = pblk_sysfs_exit,
1000 };
1001
1002 static int __init pblk_module_init(void)
1003 {
1004         int ret;
1005
1006         pblk_bio_set = bioset_create(BIO_POOL_SIZE, 0, 0);
1007         if (!pblk_bio_set)
1008                 return -ENOMEM;
1009         ret = nvm_register_tgt_type(&tt_pblk);
1010         if (ret)
1011                 bioset_free(pblk_bio_set);
1012         return ret;
1013 }
1014
1015 static void pblk_module_exit(void)
1016 {
1017         bioset_free(pblk_bio_set);
1018         nvm_unregister_tgt_type(&tt_pblk);
1019 }
1020
1021 module_init(pblk_module_init);
1022 module_exit(pblk_module_exit);
1023 MODULE_AUTHOR("Javier Gonzalez <javier@cnexlabs.com>");
1024 MODULE_AUTHOR("Matias Bjorling <matias@cnexlabs.com>");
1025 MODULE_LICENSE("GPL v2");
1026 MODULE_DESCRIPTION("Physical Block-Device for Open-Channel SSDs");