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(PBLK_WS_POOL_SIZE,
 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(PBLK_READ_REQ_POOL_SIZE,
 264                                                         pblk_g_rq_cache);
 265         if (!pblk->g_rq_pool)
 266                 goto free_rec_pool;
 267
 268         pblk->w_rq_pool = mempool_create_slab_pool(geo->nr_luns * 2,
 269                                                         pblk_w_rq_cache);
 270         if (!pblk->w_rq_pool)
 271                 goto free_g_rq_pool;
 272
 273         pblk->line_meta_pool =
 274                         mempool_create_slab_pool(PBLK_META_POOL_SIZE,
 275                                                         pblk_line_meta_cache);
 276         if (!pblk->line_meta_pool)
 277                 goto free_w_rq_pool;
 278
 279         pblk->close_wq = alloc_workqueue("pblk-close-wq",
 280                         WQ_MEM_RECLAIM | WQ_UNBOUND, PBLK_NR_CLOSE_JOBS);
 281         if (!pblk->close_wq)
 282                 goto free_line_meta_pool;
 283
 284         pblk->bb_wq = alloc_workqueue("pblk-bb-wq",
 285                         WQ_MEM_RECLAIM | WQ_UNBOUND, 0);
 286         if (!pblk->bb_wq)
 287                 goto free_close_wq;
 288
 289         if (pblk_set_ppaf(pblk))
 290                 goto free_bb_wq;
 291
 292         if (pblk_rwb_init(pblk))
 293                 goto free_bb_wq;
 294
 295         INIT_LIST_HEAD(&pblk->compl_list);
 296         return 0;
 297
 298 free_bb_wq:
 299         destroy_workqueue(pblk->bb_wq);
 300 free_close_wq:
 301         destroy_workqueue(pblk->close_wq);
 302 free_line_meta_pool:
 303         mempool_destroy(pblk->line_meta_pool);
 304 free_w_rq_pool:
 305         mempool_destroy(pblk->w_rq_pool);
 306 free_g_rq_pool:
 307         mempool_destroy(pblk->g_rq_pool);
 308 free_rec_pool:
 309         mempool_destroy(pblk->rec_pool);
 310 free_blk_ws_pool:
 311         mempool_destroy(pblk->line_ws_pool);
 312 free_page_pool:
 313         mempool_destroy(pblk->page_pool);
 314         return -ENOMEM;
 315 }
 316
 317 static void pblk_core_free(struct pblk *pblk)
 318 {
 319         if (pblk->close_wq)
 320                 destroy_workqueue(pblk->close_wq);
 321
 322         if (pblk->bb_wq)
 323                 destroy_workqueue(pblk->bb_wq);
 324
 325         mempool_destroy(pblk->page_pool);
 326         mempool_destroy(pblk->line_ws_pool);
 327         mempool_destroy(pblk->rec_pool);
 328         mempool_destroy(pblk->g_rq_pool);
 329         mempool_destroy(pblk->w_rq_pool);
 330         mempool_destroy(pblk->line_meta_pool);
 331
 332         kmem_cache_destroy(pblk_blk_ws_cache);
 333         kmem_cache_destroy(pblk_rec_cache);
 334         kmem_cache_destroy(pblk_g_rq_cache);
 335         kmem_cache_destroy(pblk_w_rq_cache);
 336         kmem_cache_destroy(pblk_line_meta_cache);
 337 }
 338
 339 static void pblk_luns_free(struct pblk *pblk)
 340 {
 341         kfree(pblk->luns);
 342 }
 343
 344 static void pblk_free_line_bitmaps(struct pblk_line *line)
 345 {
 346         kfree(line->blk_bitmap);
 347         kfree(line->erase_bitmap);
 348 }
 349
 350 static void pblk_lines_free(struct pblk *pblk)
 351 {
 352         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
 353         struct pblk_line *line;
 354         int i;
 355
 356         spin_lock(&l_mg->free_lock);
 357         for (i = 0; i < l_mg->nr_lines; i++) {
 358                 line = &pblk->lines[i];
 359
 360                 pblk_line_free(pblk, line);
 361                 pblk_free_line_bitmaps(line);
 362         }
 363         spin_unlock(&l_mg->free_lock);
 364 }
 365
 366 static void pblk_line_meta_free(struct pblk *pblk)
 367 {
 368         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
 369         int i;
 370
 371         kfree(l_mg->bb_template);
 372         kfree(l_mg->bb_aux);
 373         kfree(l_mg->vsc_list);
 374
 375         spin_lock(&l_mg->free_lock);
 376         for (i = 0; i < PBLK_DATA_LINES; i++) {
 377                 kfree(l_mg->sline_meta[i]);
 378                 pblk_mfree(l_mg->eline_meta[i]->buf, l_mg->emeta_alloc_type);
 379                 kfree(l_mg->eline_meta[i]);
 380         }
 381         spin_unlock(&l_mg->free_lock);
 382
 383         kfree(pblk->lines);
 384 }
 385
 386 static int pblk_bb_discovery(struct nvm_tgt_dev *dev, struct pblk_lun *rlun)
 387 {
 388         struct nvm_geo *geo = &dev->geo;
 389         struct ppa_addr ppa;
 390         u8 *blks;
 391         int nr_blks, ret;
 392
 393         nr_blks = geo->blks_per_lun * geo->plane_mode;
 394         blks = kmalloc(nr_blks, GFP_KERNEL);
 395         if (!blks)
 396                 return -ENOMEM;
 397
 398         ppa.ppa = 0;
 399         ppa.g.ch = rlun->bppa.g.ch;
 400         ppa.g.lun = rlun->bppa.g.lun;
 401
 402         ret = nvm_get_tgt_bb_tbl(dev, ppa, blks);
 403         if (ret)
 404                 goto out;
 405
 406         nr_blks = nvm_bb_tbl_fold(dev->parent, blks, nr_blks);
 407         if (nr_blks < 0) {
 408                 ret = nr_blks;
 409                 goto out;
 410         }
 411
 412         rlun->bb_list = blks;
 413
 414         return 0;
 415 out:
 416         kfree(blks);
 417         return ret;
 418 }
 419
 420 static int pblk_bb_line(struct pblk *pblk, struct pblk_line *line,
 421                         int blk_per_line)
 422 {
 423         struct nvm_tgt_dev *dev = pblk->dev;
 424         struct nvm_geo *geo = &dev->geo;
 425         struct pblk_lun *rlun;
 426         int bb_cnt = 0;
 427         int i;
 428
 429         for (i = 0; i < blk_per_line; i++) {
 430                 rlun = &pblk->luns[i];
 431                 if (rlun->bb_list[line->id] == NVM_BLK_T_FREE)
 432                         continue;
 433
 434                 set_bit(pblk_ppa_to_pos(geo, rlun->bppa), line->blk_bitmap);
 435                 bb_cnt++;
 436         }
 437
 438         return bb_cnt;
 439 }
 440
 441 static int pblk_alloc_line_bitmaps(struct pblk *pblk, struct pblk_line *line)
 442 {
 443         struct pblk_line_meta *lm = &pblk->lm;
 444
 445         line->blk_bitmap = kzalloc(lm->blk_bitmap_len, GFP_KERNEL);
 446         if (!line->blk_bitmap)
 447                 return -ENOMEM;
 448
 449         line->erase_bitmap = kzalloc(lm->blk_bitmap_len, GFP_KERNEL);
 450         if (!line->erase_bitmap) {
 451                 kfree(line->blk_bitmap);
 452                 return -ENOMEM;
 453         }
 454
 455         return 0;
 456 }
 457
 458 static int pblk_luns_init(struct pblk *pblk, struct ppa_addr *luns)
 459 {
 460         struct nvm_tgt_dev *dev = pblk->dev;
 461         struct nvm_geo *geo = &dev->geo;
 462         struct pblk_lun *rlun;
 463         int i, ret;
 464
 465         /* TODO: Implement unbalanced LUN support */
 466         if (geo->luns_per_chnl < 0) {
 467                 pr_err("pblk: unbalanced LUN config.\n");
 468                 return -EINVAL;
 469         }
 470
 471         pblk->luns = kcalloc(geo->nr_luns, sizeof(struct pblk_lun), GFP_KERNEL);
 472         if (!pblk->luns)
 473                 return -ENOMEM;
 474
 475         for (i = 0; i < geo->nr_luns; i++) {
 476                 /* Stripe across channels */
 477                 int ch = i % geo->nr_chnls;
 478                 int lun_raw = i / geo->nr_chnls;
 479                 int lunid = lun_raw + ch * geo->luns_per_chnl;
 480
 481                 rlun = &pblk->luns[i];
 482                 rlun->bppa = luns[lunid];
 483
 484                 sema_init(&rlun->wr_sem, 1);
 485
 486                 ret = pblk_bb_discovery(dev, rlun);
 487                 if (ret) {
 488                         while (--i >= 0)
 489                                 kfree(pblk->luns[i].bb_list);
 490                         return ret;
 491                 }
 492         }
 493
 494         return 0;
 495 }
 496
 497 static int pblk_lines_configure(struct pblk *pblk, int flags)
 498 {
 499         struct pblk_line *line = NULL;
 500         int ret = 0;
 501
 502         if (!(flags & NVM_TARGET_FACTORY)) {
 503                 line = pblk_recov_l2p(pblk);
 504                 if (IS_ERR(line)) {
 505                         pr_err("pblk: could not recover l2p table\n");
 506                         ret = -EFAULT;
 507                 }
 508         }
 509
 510         if (!line) {
 511                 /* Configure next line for user data */
 512                 line = pblk_line_get_first_data(pblk);
 513                 if (!line) {
 514                         pr_err("pblk: line list corrupted\n");
 515                         ret = -EFAULT;
 516                 }
 517         }
 518
 519         return ret;
 520 }
 521
 522 /* See comment over struct line_emeta definition */
 523 static unsigned int calc_emeta_len(struct pblk *pblk)
 524 {
 525         struct pblk_line_meta *lm = &pblk->lm;
 526         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
 527         struct nvm_tgt_dev *dev = pblk->dev;
 528         struct nvm_geo *geo = &dev->geo;
 529
 530         /* Round to sector size so that lba_list starts on its own sector */
 531         lm->emeta_sec[1] = DIV_ROUND_UP(
 532                         sizeof(struct line_emeta) + lm->blk_bitmap_len,
 533                         geo->sec_size);
 534         lm->emeta_len[1] = lm->emeta_sec[1] * geo->sec_size;
 535
 536         /* Round to sector size so that vsc_list starts on its own sector */
 537         lm->dsec_per_line = lm->sec_per_line - lm->emeta_sec[0];
 538         lm->emeta_sec[2] = DIV_ROUND_UP(lm->dsec_per_line * sizeof(u64),
 539                         geo->sec_size);
 540         lm->emeta_len[2] = lm->emeta_sec[2] * geo->sec_size;
 541
 542         lm->emeta_sec[3] = DIV_ROUND_UP(l_mg->nr_lines * sizeof(u32),
 543                         geo->sec_size);
 544         lm->emeta_len[3] = lm->emeta_sec[3] * geo->sec_size;
 545
 546         lm->vsc_list_len = l_mg->nr_lines * sizeof(u32);
 547
 548         return (lm->emeta_len[1] + lm->emeta_len[2] + lm->emeta_len[3]);
 549 }
 550
 551 static void pblk_set_provision(struct pblk *pblk, long nr_free_blks)
 552 {
 553         struct nvm_tgt_dev *dev = pblk->dev;
 554         struct nvm_geo *geo = &dev->geo;
 555         sector_t provisioned;
 556
 557         pblk->over_pct = 20;
 558
 559         provisioned = nr_free_blks;
 560         provisioned *= (100 - pblk->over_pct);
 561         sector_div(provisioned, 100);
 562
 563         /* Internally pblk manages all free blocks, but all calculations based
 564          * on user capacity consider only provisioned blocks
 565          */
 566         pblk->rl.total_blocks = nr_free_blks;
 567         pblk->rl.nr_secs = nr_free_blks * geo->sec_per_blk;
 568         pblk->capacity = provisioned * geo->sec_per_blk;
 569         atomic_set(&pblk->rl.free_blocks, nr_free_blks);
 570 }
 571
 572 static int pblk_lines_alloc_metadata(struct pblk *pblk)
 573 {
 574         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
 575         struct pblk_line_meta *lm = &pblk->lm;
 576         int i;
 577
 578         /* smeta is always small enough to fit on a kmalloc memory allocation,
 579          * emeta depends on the number of LUNs allocated to the pblk instance
 580          */
 581         for (i = 0; i < PBLK_DATA_LINES; i++) {
 582                 l_mg->sline_meta[i] = kmalloc(lm->smeta_len, GFP_KERNEL);
 583                 if (!l_mg->sline_meta[i])
 584                         goto fail_free_smeta;
 585         }
 586
 587         /* emeta allocates three different buffers for managing metadata with
 588          * in-memory and in-media layouts
 589          */
 590         for (i = 0; i < PBLK_DATA_LINES; i++) {
 591                 struct pblk_emeta *emeta;
 592
 593                 emeta = kmalloc(sizeof(struct pblk_emeta), GFP_KERNEL);
 594                 if (!emeta)
 595                         goto fail_free_emeta;
 596
 597                 if (lm->emeta_len[0] > KMALLOC_MAX_CACHE_SIZE) {
 598                         l_mg->emeta_alloc_type = PBLK_VMALLOC_META;
 599
 600                         emeta->buf = vmalloc(lm->emeta_len[0]);
 601                         if (!emeta->buf) {
 602                                 kfree(emeta);
 603                                 goto fail_free_emeta;
 604                         }
 605
 606                         emeta->nr_entries = lm->emeta_sec[0];
 607                         l_mg->eline_meta[i] = emeta;
 608                 } else {
 609                         l_mg->emeta_alloc_type = PBLK_KMALLOC_META;
 610
 611                         emeta->buf = kmalloc(lm->emeta_len[0], GFP_KERNEL);
 612                         if (!emeta->buf) {
 613                                 kfree(emeta);
 614                                 goto fail_free_emeta;
 615                         }
 616
 617                         emeta->nr_entries = lm->emeta_sec[0];
 618                         l_mg->eline_meta[i] = emeta;
 619                 }
 620         }
 621
 622         l_mg->vsc_list = kcalloc(l_mg->nr_lines, sizeof(__le32), GFP_KERNEL);
 623         if (!l_mg->vsc_list)
 624                 goto fail_free_emeta;
 625
 626         for (i = 0; i < l_mg->nr_lines; i++)
 627                 l_mg->vsc_list[i] = cpu_to_le32(EMPTY_ENTRY);
 628
 629         return 0;
 630
 631 fail_free_emeta:
 632         while (--i >= 0) {
 633                 vfree(l_mg->eline_meta[i]->buf);
 634                 kfree(l_mg->eline_meta[i]);
 635         }
 636
 637 fail_free_smeta:
 638         for (i = 0; i < PBLK_DATA_LINES; i++)
 639                 kfree(l_mg->sline_meta[i]);
 640
 641         return -ENOMEM;
 642 }
 643
 644 static int pblk_lines_init(struct pblk *pblk)
 645 {
 646         struct nvm_tgt_dev *dev = pblk->dev;
 647         struct nvm_geo *geo = &dev->geo;
 648         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
 649         struct pblk_line_meta *lm = &pblk->lm;
 650         struct pblk_line *line;
 651         unsigned int smeta_len, emeta_len;
 652         long nr_bad_blks, nr_free_blks;
 653         int bb_distance, max_write_ppas, mod;
 654         int i, ret;
 655
 656         pblk->min_write_pgs = geo->sec_per_pl * (geo->sec_size / PAGE_SIZE);
 657         max_write_ppas = pblk->min_write_pgs * geo->nr_luns;
 658         pblk->max_write_pgs = (max_write_ppas < nvm_max_phys_sects(dev)) ?
 659                                 max_write_ppas : nvm_max_phys_sects(dev);
 660         pblk_set_sec_per_write(pblk, pblk->min_write_pgs);
 661
 662         if (pblk->max_write_pgs > PBLK_MAX_REQ_ADDRS) {
 663                 pr_err("pblk: cannot support device max_phys_sect\n");
 664                 return -EINVAL;
 665         }
 666
 667         div_u64_rem(geo->sec_per_blk, pblk->min_write_pgs, &mod);
 668         if (mod) {
 669                 pr_err("pblk: bad configuration of sectors/pages\n");
 670                 return -EINVAL;
 671         }
 672
 673         l_mg->nr_lines = geo->blks_per_lun;
 674         l_mg->log_line = l_mg->data_line = NULL;
 675         l_mg->l_seq_nr = l_mg->d_seq_nr = 0;
 676         l_mg->nr_free_lines = 0;
 677         bitmap_zero(&l_mg->meta_bitmap, PBLK_DATA_LINES);
 678
 679         lm->sec_per_line = geo->sec_per_blk * geo->nr_luns;
 680         lm->blk_per_line = geo->nr_luns;
 681         lm->blk_bitmap_len = BITS_TO_LONGS(geo->nr_luns) * sizeof(long);
 682         lm->sec_bitmap_len = BITS_TO_LONGS(lm->sec_per_line) * sizeof(long);
 683         lm->lun_bitmap_len = BITS_TO_LONGS(geo->nr_luns) * sizeof(long);
 684         lm->high_thrs = lm->sec_per_line / 2;
 685         lm->mid_thrs = lm->sec_per_line / 4;
 686         lm->meta_distance = (geo->nr_luns / 2) * pblk->min_write_pgs;
 687
 688         /* Calculate necessary pages for smeta. See comment over struct
 689          * line_smeta definition
 690          */
 691         i = 1;
 692 add_smeta_page:
 693         lm->smeta_sec = i * geo->sec_per_pl;
 694         lm->smeta_len = lm->smeta_sec * geo->sec_size;
 695
 696         smeta_len = sizeof(struct line_smeta) + lm->lun_bitmap_len;
 697         if (smeta_len > lm->smeta_len) {
 698                 i++;
 699                 goto add_smeta_page;
 700         }
 701
 702         /* Calculate necessary pages for emeta. See comment over struct
 703          * line_emeta definition
 704          */
 705         i = 1;
 706 add_emeta_page:
 707         lm->emeta_sec[0] = i * geo->sec_per_pl;
 708         lm->emeta_len[0] = lm->emeta_sec[0] * geo->sec_size;
 709
 710         emeta_len = calc_emeta_len(pblk);
 711         if (emeta_len > lm->emeta_len[0]) {
 712                 i++;
 713                 goto add_emeta_page;
 714         }
 715
 716         lm->emeta_bb = geo->nr_luns - i;
 717         lm->min_blk_line = 1 + DIV_ROUND_UP(lm->smeta_sec + lm->emeta_sec[0],
 718                                                         geo->sec_per_blk);
 719         if (lm->min_blk_line > lm->blk_per_line) {
 720                 pr_err("pblk: config. not supported. Min. LUN in line:%d\n",
 721                                                         lm->blk_per_line);
 722                 ret = -EINVAL;
 723                 goto fail;
 724         }
 725
 726         ret = pblk_lines_alloc_metadata(pblk);
 727         if (ret)
 728                 goto fail;
 729
 730         l_mg->bb_template = kzalloc(lm->sec_bitmap_len, GFP_KERNEL);
 731         if (!l_mg->bb_template) {
 732                 ret = -ENOMEM;
 733                 goto fail_free_meta;
 734         }
 735
 736         l_mg->bb_aux = kzalloc(lm->sec_bitmap_len, GFP_KERNEL);
 737         if (!l_mg->bb_aux) {
 738                 ret = -ENOMEM;
 739                 goto fail_free_bb_template;
 740         }
 741
 742         bb_distance = (geo->nr_luns) * geo->sec_per_pl;
 743         for (i = 0; i < lm->sec_per_line; i += bb_distance)
 744                 bitmap_set(l_mg->bb_template, i, geo->sec_per_pl);
 745
 746         INIT_LIST_HEAD(&l_mg->free_list);
 747         INIT_LIST_HEAD(&l_mg->corrupt_list);
 748         INIT_LIST_HEAD(&l_mg->bad_list);
 749         INIT_LIST_HEAD(&l_mg->gc_full_list);
 750         INIT_LIST_HEAD(&l_mg->gc_high_list);
 751         INIT_LIST_HEAD(&l_mg->gc_mid_list);
 752         INIT_LIST_HEAD(&l_mg->gc_low_list);
 753         INIT_LIST_HEAD(&l_mg->gc_empty_list);
 754
 755         INIT_LIST_HEAD(&l_mg->emeta_list);
 756
 757         l_mg->gc_lists[0] = &l_mg->gc_high_list;
 758         l_mg->gc_lists[1] = &l_mg->gc_mid_list;
 759         l_mg->gc_lists[2] = &l_mg->gc_low_list;
 760
 761         spin_lock_init(&l_mg->free_lock);
 762         spin_lock_init(&l_mg->close_lock);
 763         spin_lock_init(&l_mg->gc_lock);
 764
 765         pblk->lines = kcalloc(l_mg->nr_lines, sizeof(struct pblk_line),
 766                                                                 GFP_KERNEL);
 767         if (!pblk->lines) {
 768                 ret = -ENOMEM;
 769                 goto fail_free_bb_aux;
 770         }
 771
 772         nr_free_blks = 0;
 773         for (i = 0; i < l_mg->nr_lines; i++) {
 774                 int blk_in_line;
 775
 776                 line = &pblk->lines[i];
 777
 778                 line->pblk = pblk;
 779                 line->id = i;
 780                 line->type = PBLK_LINETYPE_FREE;
 781                 line->state = PBLK_LINESTATE_FREE;
 782                 line->gc_group = PBLK_LINEGC_NONE;
 783                 line->vsc = &l_mg->vsc_list[i];
 784                 spin_lock_init(&line->lock);
 785
 786                 ret = pblk_alloc_line_bitmaps(pblk, line);
 787                 if (ret)
 788                         goto fail_free_lines;
 789
 790                 nr_bad_blks = pblk_bb_line(pblk, line, lm->blk_per_line);
 791                 if (nr_bad_blks < 0 || nr_bad_blks > lm->blk_per_line) {
 792                         pblk_free_line_bitmaps(line);
 793                         ret = -EINVAL;
 794                         goto fail_free_lines;
 795                 }
 796
 797                 blk_in_line = lm->blk_per_line - nr_bad_blks;
 798                 if (blk_in_line < lm->min_blk_line) {
 799                         line->state = PBLK_LINESTATE_BAD;
 800                         list_add_tail(&line->list, &l_mg->bad_list);
 801                         continue;
 802                 }
 803
 804                 nr_free_blks += blk_in_line;
 805                 atomic_set(&line->blk_in_line, blk_in_line);
 806
 807                 l_mg->nr_free_lines++;
 808                 list_add_tail(&line->list, &l_mg->free_list);
 809         }
 810
 811         pblk_set_provision(pblk, nr_free_blks);
 812
 813         /* Cleanup per-LUN bad block lists - managed within lines on run-time */
 814         for (i = 0; i < geo->nr_luns; i++)
 815                 kfree(pblk->luns[i].bb_list);
 816
 817         return 0;
 818 fail_free_lines:
 819         while (--i >= 0)
 820                 pblk_free_line_bitmaps(&pblk->lines[i]);
 821 fail_free_bb_aux:
 822         kfree(l_mg->bb_aux);
 823 fail_free_bb_template:
 824         kfree(l_mg->bb_template);
 825 fail_free_meta:
 826         pblk_line_meta_free(pblk);
 827 fail:
 828         for (i = 0; i < geo->nr_luns; i++)
 829                 kfree(pblk->luns[i].bb_list);
 830
 831         return ret;
 832 }
 833
 834 static int pblk_writer_init(struct pblk *pblk)
 835 {
 836         setup_timer(&pblk->wtimer, pblk_write_timer_fn, (unsigned long)pblk);
 837         mod_timer(&pblk->wtimer, jiffies + msecs_to_jiffies(100));
 838
 839         pblk->writer_ts = kthread_create(pblk_write_ts, pblk, "pblk-writer-t");
 840         if (IS_ERR(pblk->writer_ts)) {
 841                 pr_err("pblk: could not allocate writer kthread\n");
 842                 return PTR_ERR(pblk->writer_ts);
 843         }
 844
 845         return 0;
 846 }
 847
 848 static void pblk_writer_stop(struct pblk *pblk)
 849 {
 850         /* The pipeline must be stopped and the write buffer emptied before the
 851          * write thread is stopped
 852          */
 853         WARN(pblk_rb_read_count(&pblk->rwb),
 854                         "Stopping not fully persisted write buffer\n");
 855
 856         WARN(pblk_rb_sync_count(&pblk->rwb),
 857                         "Stopping not fully synced write buffer\n");
 858
 859         if (pblk->writer_ts)
 860                 kthread_stop(pblk->writer_ts);
 861         del_timer(&pblk->wtimer);
 862 }
 863
 864 static void pblk_free(struct pblk *pblk)
 865 {
 866         pblk_luns_free(pblk);
 867         pblk_lines_free(pblk);
 868         pblk_line_meta_free(pblk);
 869         pblk_core_free(pblk);
 870         pblk_l2p_free(pblk);
 871
 872         kfree(pblk);
 873 }
 874
 875 static void pblk_tear_down(struct pblk *pblk)
 876 {
 877         pblk_pipeline_stop(pblk);
 878         pblk_writer_stop(pblk);
 879         pblk_rb_sync_l2p(&pblk->rwb);
 880         pblk_rwb_free(pblk);
 881         pblk_rl_free(&pblk->rl);
 882
 883         pr_debug("pblk: consistent tear down\n");
 884 }
 885
 886 static void pblk_exit(void *private)
 887 {
 888         struct pblk *pblk = private;
 889
 890         down_write(&pblk_lock);
 891         pblk_gc_exit(pblk);
 892         pblk_tear_down(pblk);
 893         pblk_free(pblk);
 894         up_write(&pblk_lock);
 895 }
 896
 897 static sector_t pblk_capacity(void *private)
 898 {
 899         struct pblk *pblk = private;
 900
 901         return pblk->capacity * NR_PHY_IN_LOG;
 902 }
 903
 904 static void *pblk_init(struct nvm_tgt_dev *dev, struct gendisk *tdisk,
 905                        int flags)
 906 {
 907         struct nvm_geo *geo = &dev->geo;
 908         struct request_queue *bqueue = dev->q;
 909         struct request_queue *tqueue = tdisk->queue;
 910         struct pblk *pblk;
 911         int ret;
 912
 913         if (dev->identity.dom & NVM_RSP_L2P) {
 914                 pr_err("pblk: device-side L2P table not supported. (%x)\n",
 915                                                         dev->identity.dom);
 916                 return ERR_PTR(-EINVAL);
 917         }
 918
 919         pblk = kzalloc(sizeof(struct pblk), GFP_KERNEL);
 920         if (!pblk)
 921                 return ERR_PTR(-ENOMEM);
 922
 923         pblk->dev = dev;
 924         pblk->disk = tdisk;
 925         pblk->state = PBLK_STATE_RUNNING;
 926
 927         spin_lock_init(&pblk->trans_lock);
 928         spin_lock_init(&pblk->lock);
 929
 930         if (flags & NVM_TARGET_FACTORY)
 931                 pblk_setup_uuid(pblk);
 932
 933 #ifdef CONFIG_NVM_DEBUG
 934         atomic_long_set(&pblk->inflight_writes, 0);
 935         atomic_long_set(&pblk->padded_writes, 0);
 936         atomic_long_set(&pblk->padded_wb, 0);
 937         atomic_long_set(&pblk->nr_flush, 0);
 938         atomic_long_set(&pblk->req_writes, 0);
 939         atomic_long_set(&pblk->sub_writes, 0);
 940         atomic_long_set(&pblk->sync_writes, 0);
 941         atomic_long_set(&pblk->inflight_reads, 0);
 942         atomic_long_set(&pblk->cache_reads, 0);
 943         atomic_long_set(&pblk->sync_reads, 0);
 944         atomic_long_set(&pblk->recov_writes, 0);
 945         atomic_long_set(&pblk->recov_writes, 0);
 946         atomic_long_set(&pblk->recov_gc_writes, 0);
 947 #endif
 948
 949         atomic_long_set(&pblk->read_failed, 0);
 950         atomic_long_set(&pblk->read_empty, 0);
 951         atomic_long_set(&pblk->read_high_ecc, 0);
 952         atomic_long_set(&pblk->read_failed_gc, 0);
 953         atomic_long_set(&pblk->write_failed, 0);
 954         atomic_long_set(&pblk->erase_failed, 0);
 955
 956         ret = pblk_luns_init(pblk, dev->luns);
 957         if (ret) {
 958                 pr_err("pblk: could not initialize luns\n");
 959                 goto fail;
 960         }
 961
 962         ret = pblk_lines_init(pblk);
 963         if (ret) {
 964                 pr_err("pblk: could not initialize lines\n");
 965                 goto fail_free_luns;
 966         }
 967
 968         ret = pblk_core_init(pblk);
 969         if (ret) {
 970                 pr_err("pblk: could not initialize core\n");
 971                 goto fail_free_line_meta;
 972         }
 973
 974         ret = pblk_l2p_init(pblk);
 975         if (ret) {
 976                 pr_err("pblk: could not initialize maps\n");
 977                 goto fail_free_core;
 978         }
 979
 980         ret = pblk_lines_configure(pblk, flags);
 981         if (ret) {
 982                 pr_err("pblk: could not configure lines\n");
 983                 goto fail_free_l2p;
 984         }
 985
 986         ret = pblk_writer_init(pblk);
 987         if (ret) {
 988                 pr_err("pblk: could not initialize write thread\n");
 989                 goto fail_free_lines;
 990         }
 991
 992         ret = pblk_gc_init(pblk);
 993         if (ret) {
 994                 pr_err("pblk: could not initialize gc\n");
 995                 goto fail_stop_writer;
 996         }
 997
 998         /* inherit the size from the underlying device */
 999         blk_queue_logical_block_size(tqueue, queue_physical_block_size(bqueue));
1000         blk_queue_max_hw_sectors(tqueue, queue_max_hw_sectors(bqueue));
1001
1002         blk_queue_write_cache(tqueue, true, false);
1003
1004         tqueue->limits.discard_granularity = geo->pgs_per_blk * geo->pfpg_size;
1005         tqueue->limits.discard_alignment = 0;
1006         blk_queue_max_discard_sectors(tqueue, UINT_MAX >> 9);
1007         queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, tqueue);
1008
1009         pr_info("pblk init: luns:%u, lines:%d, secs:%llu, buf entries:%u\n",
1010                         geo->nr_luns, pblk->l_mg.nr_lines,
1011                         (unsigned long long)pblk->rl.nr_secs,
1012                         pblk->rwb.nr_entries);
1013
1014         wake_up_process(pblk->writer_ts);
1015         return pblk;
1016
1017 fail_stop_writer:
1018         pblk_writer_stop(pblk);
1019 fail_free_lines:
1020         pblk_lines_free(pblk);
1021 fail_free_l2p:
1022         pblk_l2p_free(pblk);
1023 fail_free_core:
1024         pblk_core_free(pblk);
1025 fail_free_line_meta:
1026         pblk_line_meta_free(pblk);
1027 fail_free_luns:
1028         pblk_luns_free(pblk);
1029 fail:
1030         kfree(pblk);
1031         return ERR_PTR(ret);
1032 }
1033
1034 /* physical block device target */
1035 static struct nvm_tgt_type tt_pblk = {
1036         .name           = "pblk",
1037         .version        = {1, 0, 0},
1038
1039         .make_rq        = pblk_make_rq,
1040         .capacity       = pblk_capacity,
1041
1042         .init           = pblk_init,
1043         .exit           = pblk_exit,
1044
1045         .sysfs_init     = pblk_sysfs_init,
1046         .sysfs_exit     = pblk_sysfs_exit,
1047 };
1048
1049 static int __init pblk_module_init(void)
1050 {
1051         int ret;
1052
1053         pblk_bio_set = bioset_create(BIO_POOL_SIZE, 0, 0);
1054         if (!pblk_bio_set)
1055                 return -ENOMEM;
1056         ret = nvm_register_tgt_type(&tt_pblk);
1057         if (ret)
1058                 bioset_free(pblk_bio_set);
1059         return ret;
1060 }
1061
1062 static void pblk_module_exit(void)
1063 {
1064         bioset_free(pblk_bio_set);
1065         nvm_unregister_tgt_type(&tt_pblk);
1066 }
1067
1068 module_init(pblk_module_init);
1069 module_exit(pblk_module_exit);
1070 MODULE_AUTHOR("Javier Gonzalez <javier@cnexlabs.com>");
1071 MODULE_AUTHOR("Matias Bjorling <matias@cnexlabs.com>");
1072 MODULE_LICENSE("GPL v2");
1073 MODULE_DESCRIPTION("Physical Block-Device for Open-Channel SSDs");