drivers/lightnvm/pblk-core.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   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  * pblk-core.c - pblk's core functionality
  17  *
  18  */
  19
  20 #define CREATE_TRACE_POINTS
  21
  22 #include "pblk.h"
  23 #include "pblk-trace.h"
  24
  25 static void pblk_line_mark_bb(struct work_struct *work)
  26 {
  27         struct pblk_line_ws *line_ws = container_of(work, struct pblk_line_ws,
  28                                                                         ws);
  29         struct pblk *pblk = line_ws->pblk;
  30         struct nvm_tgt_dev *dev = pblk->dev;
  31         struct ppa_addr *ppa = line_ws->priv;
  32         int ret;
  33
  34         ret = nvm_set_chunk_meta(dev, ppa, 1, NVM_BLK_T_GRWN_BAD);
  35         if (ret) {
  36                 struct pblk_line *line;
  37                 int pos;
  38
  39                 line = pblk_ppa_to_line(pblk, *ppa);
  40                 pos = pblk_ppa_to_pos(&dev->geo, *ppa);
  41
  42                 pblk_err(pblk, "failed to mark bb, line:%d, pos:%d\n",
  43                                 line->id, pos);
  44         }
  45
  46         kfree(ppa);
  47         mempool_free(line_ws, &pblk->gen_ws_pool);
  48 }
  49
  50 static void pblk_mark_bb(struct pblk *pblk, struct pblk_line *line,
  51                          struct ppa_addr ppa_addr)
  52 {
  53         struct nvm_tgt_dev *dev = pblk->dev;
  54         struct nvm_geo *geo = &dev->geo;
  55         struct ppa_addr *ppa;
  56         int pos = pblk_ppa_to_pos(geo, ppa_addr);
  57
  58         pblk_debug(pblk, "erase failed: line:%d, pos:%d\n", line->id, pos);
  59         atomic_long_inc(&pblk->erase_failed);
  60
  61         atomic_dec(&line->blk_in_line);
  62         if (test_and_set_bit(pos, line->blk_bitmap))
  63                 pblk_err(pblk, "attempted to erase bb: line:%d, pos:%d\n",
  64                                                         line->id, pos);
  65
  66         /* Not necessary to mark bad blocks on 2.0 spec. */
  67         if (geo->version == NVM_OCSSD_SPEC_20)
  68                 return;
  69
  70         ppa = kmalloc(sizeof(struct ppa_addr), GFP_ATOMIC);
  71         if (!ppa)
  72                 return;
  73
  74         *ppa = ppa_addr;
  75         pblk_gen_run_ws(pblk, NULL, ppa, pblk_line_mark_bb,
  76                                                 GFP_ATOMIC, pblk->bb_wq);
  77 }
  78
  79 static void __pblk_end_io_erase(struct pblk *pblk, struct nvm_rq *rqd)
  80 {
  81         struct nvm_tgt_dev *dev = pblk->dev;
  82         struct nvm_geo *geo = &dev->geo;
  83         struct nvm_chk_meta *chunk;
  84         struct pblk_line *line;
  85         int pos;
  86
  87         line = pblk_ppa_to_line(pblk, rqd->ppa_addr);
  88         pos = pblk_ppa_to_pos(geo, rqd->ppa_addr);
  89         chunk = &line->chks[pos];
  90
  91         atomic_dec(&line->left_seblks);
  92
  93         if (rqd->error) {
  94                 trace_pblk_chunk_reset(pblk_disk_name(pblk),
  95                                 &rqd->ppa_addr, PBLK_CHUNK_RESET_FAILED);
  96
  97                 chunk->state = NVM_CHK_ST_OFFLINE;
  98                 pblk_mark_bb(pblk, line, rqd->ppa_addr);
  99         } else {
 100                 trace_pblk_chunk_reset(pblk_disk_name(pblk),
 101                                 &rqd->ppa_addr, PBLK_CHUNK_RESET_DONE);
 102
 103                 chunk->state = NVM_CHK_ST_FREE;
 104         }
 105
 106         trace_pblk_chunk_state(pblk_disk_name(pblk), &rqd->ppa_addr,
 107                                 chunk->state);
 108
 109         atomic_dec(&pblk->inflight_io);
 110 }
 111
 112 /* Erase completion assumes that only one block is erased at the time */
 113 static void pblk_end_io_erase(struct nvm_rq *rqd)
 114 {
 115         struct pblk *pblk = rqd->private;
 116
 117         __pblk_end_io_erase(pblk, rqd);
 118         mempool_free(rqd, &pblk->e_rq_pool);
 119 }
 120
 121 /*
 122  * Get information for all chunks from the device.
 123  *
 124  * The caller is responsible for freeing (vmalloc) the returned structure
 125  */
 126 struct nvm_chk_meta *pblk_get_chunk_meta(struct pblk *pblk)
 127 {
 128         struct nvm_tgt_dev *dev = pblk->dev;
 129         struct nvm_geo *geo = &dev->geo;
 130         struct nvm_chk_meta *meta;
 131         struct ppa_addr ppa;
 132         unsigned long len;
 133         int ret;
 134
 135         ppa.ppa = 0;
 136
 137         len = geo->all_chunks * sizeof(*meta);
 138         meta = vzalloc(len);
 139         if (!meta)
 140                 return ERR_PTR(-ENOMEM);
 141
 142         ret = nvm_get_chunk_meta(dev, ppa, geo->all_chunks, meta);
 143         if (ret) {
 144                 vfree(meta);
 145                 return ERR_PTR(-EIO);
 146         }
 147
 148         return meta;
 149 }
 150
 151 struct nvm_chk_meta *pblk_chunk_get_off(struct pblk *pblk,
 152                                               struct nvm_chk_meta *meta,
 153                                               struct ppa_addr ppa)
 154 {
 155         struct nvm_tgt_dev *dev = pblk->dev;
 156         struct nvm_geo *geo = &dev->geo;
 157         int ch_off = ppa.m.grp * geo->num_chk * geo->num_lun;
 158         int lun_off = ppa.m.pu * geo->num_chk;
 159         int chk_off = ppa.m.chk;
 160
 161         return meta + ch_off + lun_off + chk_off;
 162 }
 163
 164 void __pblk_map_invalidate(struct pblk *pblk, struct pblk_line *line,
 165                            u64 paddr)
 166 {
 167         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
 168         struct list_head *move_list = NULL;
 169
 170         /* Lines being reclaimed (GC'ed) cannot be invalidated. Before the L2P
 171          * table is modified with reclaimed sectors, a check is done to endure
 172          * that newer updates are not overwritten.
 173          */
 174         spin_lock(&line->lock);
 175         WARN_ON(line->state == PBLK_LINESTATE_FREE);
 176
 177         if (test_and_set_bit(paddr, line->invalid_bitmap)) {
 178                 WARN_ONCE(1, "pblk: double invalidate\n");
 179                 spin_unlock(&line->lock);
 180                 return;
 181         }
 182         le32_add_cpu(line->vsc, -1);
 183
 184         if (line->state == PBLK_LINESTATE_CLOSED)
 185                 move_list = pblk_line_gc_list(pblk, line);
 186         spin_unlock(&line->lock);
 187
 188         if (move_list) {
 189                 spin_lock(&l_mg->gc_lock);
 190                 spin_lock(&line->lock);
 191                 /* Prevent moving a line that has just been chosen for GC */
 192                 if (line->state == PBLK_LINESTATE_GC) {
 193                         spin_unlock(&line->lock);
 194                         spin_unlock(&l_mg->gc_lock);
 195                         return;
 196                 }
 197                 spin_unlock(&line->lock);
 198
 199                 list_move_tail(&line->list, move_list);
 200                 spin_unlock(&l_mg->gc_lock);
 201         }
 202 }
 203
 204 void pblk_map_invalidate(struct pblk *pblk, struct ppa_addr ppa)
 205 {
 206         struct pblk_line *line;
 207         u64 paddr;
 208
 209 #ifdef CONFIG_NVM_PBLK_DEBUG
 210         /* Callers must ensure that the ppa points to a device address */
 211         BUG_ON(pblk_addr_in_cache(ppa));
 212         BUG_ON(pblk_ppa_empty(ppa));
 213 #endif
 214
 215         line = pblk_ppa_to_line(pblk, ppa);
 216         paddr = pblk_dev_ppa_to_line_addr(pblk, ppa);
 217
 218         __pblk_map_invalidate(pblk, line, paddr);
 219 }
 220
 221 static void pblk_invalidate_range(struct pblk *pblk, sector_t slba,
 222                                   unsigned int nr_secs)
 223 {
 224         sector_t lba;
 225
 226         spin_lock(&pblk->trans_lock);
 227         for (lba = slba; lba < slba + nr_secs; lba++) {
 228                 struct ppa_addr ppa;
 229
 230                 ppa = pblk_trans_map_get(pblk, lba);
 231
 232                 if (!pblk_addr_in_cache(ppa) && !pblk_ppa_empty(ppa))
 233                         pblk_map_invalidate(pblk, ppa);
 234
 235                 pblk_ppa_set_empty(&ppa);
 236                 pblk_trans_map_set(pblk, lba, ppa);
 237         }
 238         spin_unlock(&pblk->trans_lock);
 239 }
 240
 241 int pblk_alloc_rqd_meta(struct pblk *pblk, struct nvm_rq *rqd)
 242 {
 243         struct nvm_tgt_dev *dev = pblk->dev;
 244
 245         rqd->meta_list = nvm_dev_dma_alloc(dev->parent, GFP_KERNEL,
 246                                                         &rqd->dma_meta_list);
 247         if (!rqd->meta_list)
 248                 return -ENOMEM;
 249
 250         if (rqd->nr_ppas == 1)
 251                 return 0;
 252
 253         rqd->ppa_list = rqd->meta_list + pblk_dma_meta_size(pblk);
 254         rqd->dma_ppa_list = rqd->dma_meta_list + pblk_dma_meta_size(pblk);
 255
 256         return 0;
 257 }
 258
 259 void pblk_free_rqd_meta(struct pblk *pblk, struct nvm_rq *rqd)
 260 {
 261         struct nvm_tgt_dev *dev = pblk->dev;
 262
 263         if (rqd->meta_list)
 264                 nvm_dev_dma_free(dev->parent, rqd->meta_list,
 265                                 rqd->dma_meta_list);
 266 }
 267
 268 /* Caller must guarantee that the request is a valid type */
 269 struct nvm_rq *pblk_alloc_rqd(struct pblk *pblk, int type)
 270 {
 271         mempool_t *pool;
 272         struct nvm_rq *rqd;
 273         int rq_size;
 274
 275         switch (type) {
 276         case PBLK_WRITE:
 277         case PBLK_WRITE_INT:
 278                 pool = &pblk->w_rq_pool;
 279                 rq_size = pblk_w_rq_size;
 280                 break;
 281         case PBLK_READ:
 282                 pool = &pblk->r_rq_pool;
 283                 rq_size = pblk_g_rq_size;
 284                 break;
 285         default:
 286                 pool = &pblk->e_rq_pool;
 287                 rq_size = pblk_g_rq_size;
 288         }
 289
 290         rqd = mempool_alloc(pool, GFP_KERNEL);
 291         memset(rqd, 0, rq_size);
 292
 293         return rqd;
 294 }
 295
 296 /* Typically used on completion path. Cannot guarantee request consistency */
 297 void pblk_free_rqd(struct pblk *pblk, struct nvm_rq *rqd, int type)
 298 {
 299         mempool_t *pool;
 300
 301         switch (type) {
 302         case PBLK_WRITE:
 303                 kfree(((struct pblk_c_ctx *)nvm_rq_to_pdu(rqd))->lun_bitmap);
 304                 /* fall through */
 305         case PBLK_WRITE_INT:
 306                 pool = &pblk->w_rq_pool;
 307                 break;
 308         case PBLK_READ:
 309                 pool = &pblk->r_rq_pool;
 310                 break;
 311         case PBLK_ERASE:
 312                 pool = &pblk->e_rq_pool;
 313                 break;
 314         default:
 315                 pblk_err(pblk, "trying to free unknown rqd type\n");
 316                 return;
 317         }
 318
 319         pblk_free_rqd_meta(pblk, rqd);
 320         mempool_free(rqd, pool);
 321 }
 322
 323 void pblk_bio_free_pages(struct pblk *pblk, struct bio *bio, int off,
 324                          int nr_pages)
 325 {
 326         struct bio_vec *bv;
 327         struct page *page;
 328         int i, e, nbv = 0;
 329
 330         for (i = 0; i < bio->bi_vcnt; i++) {
 331                 bv = &bio->bi_io_vec[i];
 332                 page = bv->bv_page;
 333                 for (e = 0; e < bv->bv_len; e += PBLK_EXPOSED_PAGE_SIZE, nbv++)
 334                         if (nbv >= off)
 335                                 mempool_free(page++, &pblk->page_bio_pool);
 336         }
 337 }
 338
 339 int pblk_bio_add_pages(struct pblk *pblk, struct bio *bio, gfp_t flags,
 340                        int nr_pages)
 341 {
 342         struct request_queue *q = pblk->dev->q;
 343         struct page *page;
 344         int i, ret;
 345
 346         for (i = 0; i < nr_pages; i++) {
 347                 page = mempool_alloc(&pblk->page_bio_pool, flags);
 348
 349                 ret = bio_add_pc_page(q, bio, page, PBLK_EXPOSED_PAGE_SIZE, 0);
 350                 if (ret != PBLK_EXPOSED_PAGE_SIZE) {
 351                         pblk_err(pblk, "could not add page to bio\n");
 352                         mempool_free(page, &pblk->page_bio_pool);
 353                         goto err;
 354                 }
 355         }
 356
 357         return 0;
 358 err:
 359         pblk_bio_free_pages(pblk, bio, (bio->bi_vcnt - i), i);
 360         return -1;
 361 }
 362
 363 void pblk_write_kick(struct pblk *pblk)
 364 {
 365         wake_up_process(pblk->writer_ts);
 366         mod_timer(&pblk->wtimer, jiffies + msecs_to_jiffies(1000));
 367 }
 368
 369 void pblk_write_timer_fn(struct timer_list *t)
 370 {
 371         struct pblk *pblk = from_timer(pblk, t, wtimer);
 372
 373         /* kick the write thread every tick to flush outstanding data */
 374         pblk_write_kick(pblk);
 375 }
 376
 377 void pblk_write_should_kick(struct pblk *pblk)
 378 {
 379         unsigned int secs_avail = pblk_rb_read_count(&pblk->rwb);
 380
 381         if (secs_avail >= pblk->min_write_pgs_data)
 382                 pblk_write_kick(pblk);
 383 }
 384
 385 static void pblk_wait_for_meta(struct pblk *pblk)
 386 {
 387         do {
 388                 if (!atomic_read(&pblk->inflight_io))
 389                         break;
 390
 391                 schedule();
 392         } while (1);
 393 }
 394
 395 static void pblk_flush_writer(struct pblk *pblk)
 396 {
 397         pblk_rb_flush(&pblk->rwb);
 398         do {
 399                 if (!pblk_rb_sync_count(&pblk->rwb))
 400                         break;
 401
 402                 pblk_write_kick(pblk);
 403                 schedule();
 404         } while (1);
 405 }
 406
 407 struct list_head *pblk_line_gc_list(struct pblk *pblk, struct pblk_line *line)
 408 {
 409         struct pblk_line_meta *lm = &pblk->lm;
 410         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
 411         struct list_head *move_list = NULL;
 412         int packed_meta = (le32_to_cpu(*line->vsc) / pblk->min_write_pgs_data)
 413                         * (pblk->min_write_pgs - pblk->min_write_pgs_data);
 414         int vsc = le32_to_cpu(*line->vsc) + packed_meta;
 415
 416         lockdep_assert_held(&line->lock);
 417
 418         if (line->w_err_gc->has_write_err) {
 419                 if (line->gc_group != PBLK_LINEGC_WERR) {
 420                         line->gc_group = PBLK_LINEGC_WERR;
 421                         move_list = &l_mg->gc_werr_list;
 422                         pblk_rl_werr_line_in(&pblk->rl);
 423                 }
 424         } else if (!vsc) {
 425                 if (line->gc_group != PBLK_LINEGC_FULL) {
 426                         line->gc_group = PBLK_LINEGC_FULL;
 427                         move_list = &l_mg->gc_full_list;
 428                 }
 429         } else if (vsc < lm->high_thrs) {
 430                 if (line->gc_group != PBLK_LINEGC_HIGH) {
 431                         line->gc_group = PBLK_LINEGC_HIGH;
 432                         move_list = &l_mg->gc_high_list;
 433                 }
 434         } else if (vsc < lm->mid_thrs) {
 435                 if (line->gc_group != PBLK_LINEGC_MID) {
 436                         line->gc_group = PBLK_LINEGC_MID;
 437                         move_list = &l_mg->gc_mid_list;
 438                 }
 439         } else if (vsc < line->sec_in_line) {
 440                 if (line->gc_group != PBLK_LINEGC_LOW) {
 441                         line->gc_group = PBLK_LINEGC_LOW;
 442                         move_list = &l_mg->gc_low_list;
 443                 }
 444         } else if (vsc == line->sec_in_line) {
 445                 if (line->gc_group != PBLK_LINEGC_EMPTY) {
 446                         line->gc_group = PBLK_LINEGC_EMPTY;
 447                         move_list = &l_mg->gc_empty_list;
 448                 }
 449         } else {
 450                 line->state = PBLK_LINESTATE_CORRUPT;
 451                 trace_pblk_line_state(pblk_disk_name(pblk), line->id,
 452                                         line->state);
 453
 454                 line->gc_group = PBLK_LINEGC_NONE;
 455                 move_list =  &l_mg->corrupt_list;
 456                 pblk_err(pblk, "corrupted vsc for line %d, vsc:%d (%d/%d/%d)\n",
 457                                                 line->id, vsc,
 458                                                 line->sec_in_line,
 459                                                 lm->high_thrs, lm->mid_thrs);
 460         }
 461
 462         return move_list;
 463 }
 464
 465 void pblk_discard(struct pblk *pblk, struct bio *bio)
 466 {
 467         sector_t slba = pblk_get_lba(bio);
 468         sector_t nr_secs = pblk_get_secs(bio);
 469
 470         pblk_invalidate_range(pblk, slba, nr_secs);
 471 }
 472
 473 void pblk_log_write_err(struct pblk *pblk, struct nvm_rq *rqd)
 474 {
 475         atomic_long_inc(&pblk->write_failed);
 476 #ifdef CONFIG_NVM_PBLK_DEBUG
 477         pblk_print_failed_rqd(pblk, rqd, rqd->error);
 478 #endif
 479 }
 480
 481 void pblk_log_read_err(struct pblk *pblk, struct nvm_rq *rqd)
 482 {
 483         /* Empty page read is not necessarily an error (e.g., L2P recovery) */
 484         if (rqd->error == NVM_RSP_ERR_EMPTYPAGE) {
 485                 atomic_long_inc(&pblk->read_empty);
 486                 return;
 487         }
 488
 489         switch (rqd->error) {
 490         case NVM_RSP_WARN_HIGHECC:
 491                 atomic_long_inc(&pblk->read_high_ecc);
 492                 break;
 493         case NVM_RSP_ERR_FAILECC:
 494         case NVM_RSP_ERR_FAILCRC:
 495                 atomic_long_inc(&pblk->read_failed);
 496                 break;
 497         default:
 498                 pblk_err(pblk, "unknown read error:%d\n", rqd->error);
 499         }
 500 #ifdef CONFIG_NVM_PBLK_DEBUG
 501         pblk_print_failed_rqd(pblk, rqd, rqd->error);
 502 #endif
 503 }
 504
 505 void pblk_set_sec_per_write(struct pblk *pblk, int sec_per_write)
 506 {
 507         pblk->sec_per_write = sec_per_write;
 508 }
 509
 510 int pblk_submit_io(struct pblk *pblk, struct nvm_rq *rqd)
 511 {
 512         struct nvm_tgt_dev *dev = pblk->dev;
 513
 514         atomic_inc(&pblk->inflight_io);
 515
 516 #ifdef CONFIG_NVM_PBLK_DEBUG
 517         if (pblk_check_io(pblk, rqd))
 518                 return NVM_IO_ERR;
 519 #endif
 520
 521         return nvm_submit_io(dev, rqd);
 522 }
 523
 524 void pblk_check_chunk_state_update(struct pblk *pblk, struct nvm_rq *rqd)
 525 {
 526         struct ppa_addr *ppa_list = nvm_rq_to_ppa_list(rqd);
 527
 528         int i;
 529
 530         for (i = 0; i < rqd->nr_ppas; i++) {
 531                 struct ppa_addr *ppa = &ppa_list[i];
 532                 struct nvm_chk_meta *chunk = pblk_dev_ppa_to_chunk(pblk, *ppa);
 533                 u64 caddr = pblk_dev_ppa_to_chunk_addr(pblk, *ppa);
 534
 535                 if (caddr == 0)
 536                         trace_pblk_chunk_state(pblk_disk_name(pblk),
 537                                                         ppa, NVM_CHK_ST_OPEN);
 538                 else if (caddr == (chunk->cnlb - 1))
 539                         trace_pblk_chunk_state(pblk_disk_name(pblk),
 540                                                         ppa, NVM_CHK_ST_CLOSED);
 541         }
 542 }
 543
 544 int pblk_submit_io_sync(struct pblk *pblk, struct nvm_rq *rqd)
 545 {
 546         struct nvm_tgt_dev *dev = pblk->dev;
 547         int ret;
 548
 549         atomic_inc(&pblk->inflight_io);
 550
 551 #ifdef CONFIG_NVM_PBLK_DEBUG
 552         if (pblk_check_io(pblk, rqd))
 553                 return NVM_IO_ERR;
 554 #endif
 555
 556         ret = nvm_submit_io_sync(dev, rqd);
 557
 558         if (trace_pblk_chunk_state_enabled() && !ret &&
 559             rqd->opcode == NVM_OP_PWRITE)
 560                 pblk_check_chunk_state_update(pblk, rqd);
 561
 562         return ret;
 563 }
 564
 565 int pblk_submit_io_sync_sem(struct pblk *pblk, struct nvm_rq *rqd)
 566 {
 567         struct ppa_addr *ppa_list = nvm_rq_to_ppa_list(rqd);
 568         int ret;
 569
 570         pblk_down_chunk(pblk, ppa_list[0]);
 571         ret = pblk_submit_io_sync(pblk, rqd);
 572         pblk_up_chunk(pblk, ppa_list[0]);
 573
 574         return ret;
 575 }
 576
 577 static void pblk_bio_map_addr_endio(struct bio *bio)
 578 {
 579         bio_put(bio);
 580 }
 581
 582 struct bio *pblk_bio_map_addr(struct pblk *pblk, void *data,
 583                               unsigned int nr_secs, unsigned int len,
 584                               int alloc_type, gfp_t gfp_mask)
 585 {
 586         struct nvm_tgt_dev *dev = pblk->dev;
 587         void *kaddr = data;
 588         struct page *page;
 589         struct bio *bio;
 590         int i, ret;
 591
 592         if (alloc_type == PBLK_KMALLOC_META)
 593                 return bio_map_kern(dev->q, kaddr, len, gfp_mask);
 594
 595         bio = bio_kmalloc(gfp_mask, nr_secs);
 596         if (!bio)
 597                 return ERR_PTR(-ENOMEM);
 598
 599         for (i = 0; i < nr_secs; i++) {
 600                 page = vmalloc_to_page(kaddr);
 601                 if (!page) {
 602                         pblk_err(pblk, "could not map vmalloc bio\n");
 603                         bio_put(bio);
 604                         bio = ERR_PTR(-ENOMEM);
 605                         goto out;
 606                 }
 607
 608                 ret = bio_add_pc_page(dev->q, bio, page, PAGE_SIZE, 0);
 609                 if (ret != PAGE_SIZE) {
 610                         pblk_err(pblk, "could not add page to bio\n");
 611                         bio_put(bio);
 612                         bio = ERR_PTR(-ENOMEM);
 613                         goto out;
 614                 }
 615
 616                 kaddr += PAGE_SIZE;
 617         }
 618
 619         bio->bi_end_io = pblk_bio_map_addr_endio;
 620 out:
 621         return bio;
 622 }
 623
 624 int pblk_calc_secs(struct pblk *pblk, unsigned long secs_avail,
 625                    unsigned long secs_to_flush, bool skip_meta)
 626 {
 627         int max = pblk->sec_per_write;
 628         int min = pblk->min_write_pgs;
 629         int secs_to_sync = 0;
 630
 631         if (skip_meta && pblk->min_write_pgs_data != pblk->min_write_pgs)
 632                 min = max = pblk->min_write_pgs_data;
 633
 634         if (secs_avail >= max)
 635                 secs_to_sync = max;
 636         else if (secs_avail >= min)
 637                 secs_to_sync = min * (secs_avail / min);
 638         else if (secs_to_flush)
 639                 secs_to_sync = min;
 640
 641         return secs_to_sync;
 642 }
 643
 644 void pblk_dealloc_page(struct pblk *pblk, struct pblk_line *line, int nr_secs)
 645 {
 646         u64 addr;
 647         int i;
 648
 649         spin_lock(&line->lock);
 650         addr = find_next_zero_bit(line->map_bitmap,
 651                                         pblk->lm.sec_per_line, line->cur_sec);
 652         line->cur_sec = addr - nr_secs;
 653
 654         for (i = 0; i < nr_secs; i++, line->cur_sec--)
 655                 WARN_ON(!test_and_clear_bit(line->cur_sec, line->map_bitmap));
 656         spin_unlock(&line->lock);
 657 }
 658
 659 u64 __pblk_alloc_page(struct pblk *pblk, struct pblk_line *line, int nr_secs)
 660 {
 661         u64 addr;
 662         int i;
 663
 664         lockdep_assert_held(&line->lock);
 665
 666         /* logic error: ppa out-of-bounds. Prevent generating bad address */
 667         if (line->cur_sec + nr_secs > pblk->lm.sec_per_line) {
 668                 WARN(1, "pblk: page allocation out of bounds\n");
 669                 nr_secs = pblk->lm.sec_per_line - line->cur_sec;
 670         }
 671
 672         line->cur_sec = addr = find_next_zero_bit(line->map_bitmap,
 673                                         pblk->lm.sec_per_line, line->cur_sec);
 674         for (i = 0; i < nr_secs; i++, line->cur_sec++)
 675                 WARN_ON(test_and_set_bit(line->cur_sec, line->map_bitmap));
 676
 677         return addr;
 678 }
 679
 680 u64 pblk_alloc_page(struct pblk *pblk, struct pblk_line *line, int nr_secs)
 681 {
 682         u64 addr;
 683
 684         /* Lock needed in case a write fails and a recovery needs to remap
 685          * failed write buffer entries
 686          */
 687         spin_lock(&line->lock);
 688         addr = __pblk_alloc_page(pblk, line, nr_secs);
 689         line->left_msecs -= nr_secs;
 690         WARN(line->left_msecs < 0, "pblk: page allocation out of bounds\n");
 691         spin_unlock(&line->lock);
 692
 693         return addr;
 694 }
 695
 696 u64 pblk_lookup_page(struct pblk *pblk, struct pblk_line *line)
 697 {
 698         u64 paddr;
 699
 700         spin_lock(&line->lock);
 701         paddr = find_next_zero_bit(line->map_bitmap,
 702                                         pblk->lm.sec_per_line, line->cur_sec);
 703         spin_unlock(&line->lock);
 704
 705         return paddr;
 706 }
 707
 708 u64 pblk_line_smeta_start(struct pblk *pblk, struct pblk_line *line)
 709 {
 710         struct nvm_tgt_dev *dev = pblk->dev;
 711         struct nvm_geo *geo = &dev->geo;
 712         struct pblk_line_meta *lm = &pblk->lm;
 713         int bit;
 714
 715         /* This usually only happens on bad lines */
 716         bit = find_first_zero_bit(line->blk_bitmap, lm->blk_per_line);
 717         if (bit >= lm->blk_per_line)
 718                 return -1;
 719
 720         return bit * geo->ws_opt;
 721 }
 722
 723 int pblk_line_smeta_read(struct pblk *pblk, struct pblk_line *line)
 724 {
 725         struct nvm_tgt_dev *dev = pblk->dev;
 726         struct pblk_line_meta *lm = &pblk->lm;
 727         struct bio *bio;
 728         struct ppa_addr *ppa_list;
 729         struct nvm_rq rqd;
 730         u64 paddr = pblk_line_smeta_start(pblk, line);
 731         int i, ret;
 732
 733         memset(&rqd, 0, sizeof(struct nvm_rq));
 734
 735         ret = pblk_alloc_rqd_meta(pblk, &rqd);
 736         if (ret)
 737                 return ret;
 738
 739         bio = bio_map_kern(dev->q, line->smeta, lm->smeta_len, GFP_KERNEL);
 740         if (IS_ERR(bio)) {
 741                 ret = PTR_ERR(bio);
 742                 goto clear_rqd;
 743         }
 744
 745         bio->bi_iter.bi_sector = 0; /* internal bio */
 746         bio_set_op_attrs(bio, REQ_OP_READ, 0);
 747
 748         rqd.bio = bio;
 749         rqd.opcode = NVM_OP_PREAD;
 750         rqd.nr_ppas = lm->smeta_sec;
 751         rqd.is_seq = 1;
 752         ppa_list = nvm_rq_to_ppa_list(&rqd);
 753
 754         for (i = 0; i < lm->smeta_sec; i++, paddr++)
 755                 ppa_list[i] = addr_to_gen_ppa(pblk, paddr, line->id);
 756
 757         ret = pblk_submit_io_sync(pblk, &rqd);
 758         if (ret) {
 759                 pblk_err(pblk, "smeta I/O submission failed: %d\n", ret);
 760                 bio_put(bio);
 761                 goto clear_rqd;
 762         }
 763
 764         atomic_dec(&pblk->inflight_io);
 765
 766         if (rqd.error && rqd.error != NVM_RSP_WARN_HIGHECC) {
 767                 pblk_log_read_err(pblk, &rqd);
 768                 ret = -EIO;
 769         }
 770
 771 clear_rqd:
 772         pblk_free_rqd_meta(pblk, &rqd);
 773         return ret;
 774 }
 775
 776 static int pblk_line_smeta_write(struct pblk *pblk, struct pblk_line *line,
 777                                  u64 paddr)
 778 {
 779         struct nvm_tgt_dev *dev = pblk->dev;
 780         struct pblk_line_meta *lm = &pblk->lm;
 781         struct bio *bio;
 782         struct ppa_addr *ppa_list;
 783         struct nvm_rq rqd;
 784         __le64 *lba_list = emeta_to_lbas(pblk, line->emeta->buf);
 785         __le64 addr_empty = cpu_to_le64(ADDR_EMPTY);
 786         int i, ret;
 787
 788         memset(&rqd, 0, sizeof(struct nvm_rq));
 789
 790         ret = pblk_alloc_rqd_meta(pblk, &rqd);
 791         if (ret)
 792                 return ret;
 793
 794         bio = bio_map_kern(dev->q, line->smeta, lm->smeta_len, GFP_KERNEL);
 795         if (IS_ERR(bio)) {
 796                 ret = PTR_ERR(bio);
 797                 goto clear_rqd;
 798         }
 799
 800         bio->bi_iter.bi_sector = 0; /* internal bio */
 801         bio_set_op_attrs(bio, REQ_OP_WRITE, 0);
 802
 803         rqd.bio = bio;
 804         rqd.opcode = NVM_OP_PWRITE;
 805         rqd.nr_ppas = lm->smeta_sec;
 806         rqd.is_seq = 1;
 807         ppa_list = nvm_rq_to_ppa_list(&rqd);
 808
 809         for (i = 0; i < lm->smeta_sec; i++, paddr++) {
 810                 struct pblk_sec_meta *meta = pblk_get_meta(pblk,
 811                                                            rqd.meta_list, i);
 812
 813                 ppa_list[i] = addr_to_gen_ppa(pblk, paddr, line->id);
 814                 meta->lba = lba_list[paddr] = addr_empty;
 815         }
 816
 817         ret = pblk_submit_io_sync_sem(pblk, &rqd);
 818         if (ret) {
 819                 pblk_err(pblk, "smeta I/O submission failed: %d\n", ret);
 820                 bio_put(bio);
 821                 goto clear_rqd;
 822         }
 823
 824         atomic_dec(&pblk->inflight_io);
 825
 826         if (rqd.error) {
 827                 pblk_log_write_err(pblk, &rqd);
 828                 ret = -EIO;
 829         }
 830
 831 clear_rqd:
 832         pblk_free_rqd_meta(pblk, &rqd);
 833         return ret;
 834 }
 835
 836 int pblk_line_emeta_read(struct pblk *pblk, struct pblk_line *line,
 837                          void *emeta_buf)
 838 {
 839         struct nvm_tgt_dev *dev = pblk->dev;
 840         struct nvm_geo *geo = &dev->geo;
 841         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
 842         struct pblk_line_meta *lm = &pblk->lm;
 843         void *ppa_list_buf, *meta_list;
 844         struct bio *bio;
 845         struct ppa_addr *ppa_list;
 846         struct nvm_rq rqd;
 847         u64 paddr = line->emeta_ssec;
 848         dma_addr_t dma_ppa_list, dma_meta_list;
 849         int min = pblk->min_write_pgs;
 850         int left_ppas = lm->emeta_sec[0];
 851         int line_id = line->id;
 852         int rq_ppas, rq_len;
 853         int i, j;
 854         int ret;
 855
 856         meta_list = nvm_dev_dma_alloc(dev->parent, GFP_KERNEL,
 857                                                         &dma_meta_list);
 858         if (!meta_list)
 859                 return -ENOMEM;
 860
 861         ppa_list_buf = meta_list + pblk_dma_meta_size(pblk);
 862         dma_ppa_list = dma_meta_list + pblk_dma_meta_size(pblk);
 863
 864 next_rq:
 865         memset(&rqd, 0, sizeof(struct nvm_rq));
 866
 867         rq_ppas = pblk_calc_secs(pblk, left_ppas, 0, false);
 868         rq_len = rq_ppas * geo->csecs;
 869
 870         bio = pblk_bio_map_addr(pblk, emeta_buf, rq_ppas, rq_len,
 871                                         l_mg->emeta_alloc_type, GFP_KERNEL);
 872         if (IS_ERR(bio)) {
 873                 ret = PTR_ERR(bio);
 874                 goto free_rqd_dma;
 875         }
 876
 877         bio->bi_iter.bi_sector = 0; /* internal bio */
 878         bio_set_op_attrs(bio, REQ_OP_READ, 0);
 879
 880         rqd.bio = bio;
 881         rqd.meta_list = meta_list;
 882         rqd.ppa_list = ppa_list_buf;
 883         rqd.dma_meta_list = dma_meta_list;
 884         rqd.dma_ppa_list = dma_ppa_list;
 885         rqd.opcode = NVM_OP_PREAD;
 886         rqd.nr_ppas = rq_ppas;
 887         ppa_list = nvm_rq_to_ppa_list(&rqd);
 888
 889         for (i = 0; i < rqd.nr_ppas; ) {
 890                 struct ppa_addr ppa = addr_to_gen_ppa(pblk, paddr, line_id);
 891                 int pos = pblk_ppa_to_pos(geo, ppa);
 892
 893                 if (pblk_io_aligned(pblk, rq_ppas))
 894                         rqd.is_seq = 1;
 895
 896                 while (test_bit(pos, line->blk_bitmap)) {
 897                         paddr += min;
 898                         if (pblk_boundary_paddr_checks(pblk, paddr)) {
 899                                 bio_put(bio);
 900                                 ret = -EINTR;
 901                                 goto free_rqd_dma;
 902                         }
 903
 904                         ppa = addr_to_gen_ppa(pblk, paddr, line_id);
 905                         pos = pblk_ppa_to_pos(geo, ppa);
 906                 }
 907
 908                 if (pblk_boundary_paddr_checks(pblk, paddr + min)) {
 909                         bio_put(bio);
 910                         ret = -EINTR;
 911                         goto free_rqd_dma;
 912                 }
 913
 914                 for (j = 0; j < min; j++, i++, paddr++)
 915                         ppa_list[i] = addr_to_gen_ppa(pblk, paddr, line_id);
 916         }
 917
 918         ret = pblk_submit_io_sync(pblk, &rqd);
 919         if (ret) {
 920                 pblk_err(pblk, "emeta I/O submission failed: %d\n", ret);
 921                 bio_put(bio);
 922                 goto free_rqd_dma;
 923         }
 924
 925         atomic_dec(&pblk->inflight_io);
 926
 927         if (rqd.error && rqd.error != NVM_RSP_WARN_HIGHECC) {
 928                 pblk_log_read_err(pblk, &rqd);
 929                 ret = -EIO;
 930                 goto free_rqd_dma;
 931         }
 932
 933         emeta_buf += rq_len;
 934         left_ppas -= rq_ppas;
 935         if (left_ppas)
 936                 goto next_rq;
 937
 938 free_rqd_dma:
 939         nvm_dev_dma_free(dev->parent, rqd.meta_list, rqd.dma_meta_list);
 940         return ret;
 941 }
 942
 943 static void pblk_setup_e_rq(struct pblk *pblk, struct nvm_rq *rqd,
 944                             struct ppa_addr ppa)
 945 {
 946         rqd->opcode = NVM_OP_ERASE;
 947         rqd->ppa_addr = ppa;
 948         rqd->nr_ppas = 1;
 949         rqd->is_seq = 1;
 950         rqd->bio = NULL;
 951 }
 952
 953 static int pblk_blk_erase_sync(struct pblk *pblk, struct ppa_addr ppa)
 954 {
 955         struct nvm_rq rqd = {NULL};
 956         int ret;
 957
 958         trace_pblk_chunk_reset(pblk_disk_name(pblk), &ppa,
 959                                 PBLK_CHUNK_RESET_START);
 960
 961         pblk_setup_e_rq(pblk, &rqd, ppa);
 962
 963         /* The write thread schedules erases so that it minimizes disturbances
 964          * with writes. Thus, there is no need to take the LUN semaphore.
 965          */
 966         ret = pblk_submit_io_sync(pblk, &rqd);
 967         rqd.private = pblk;
 968         __pblk_end_io_erase(pblk, &rqd);
 969
 970         return ret;
 971 }
 972
 973 int pblk_line_erase(struct pblk *pblk, struct pblk_line *line)
 974 {
 975         struct pblk_line_meta *lm = &pblk->lm;
 976         struct ppa_addr ppa;
 977         int ret, bit = -1;
 978
 979         /* Erase only good blocks, one at a time */
 980         do {
 981                 spin_lock(&line->lock);
 982                 bit = find_next_zero_bit(line->erase_bitmap, lm->blk_per_line,
 983                                                                 bit + 1);
 984                 if (bit >= lm->blk_per_line) {
 985                         spin_unlock(&line->lock);
 986                         break;
 987                 }
 988
 989                 ppa = pblk->luns[bit].bppa; /* set ch and lun */
 990                 ppa.a.blk = line->id;
 991
 992                 atomic_dec(&line->left_eblks);
 993                 WARN_ON(test_and_set_bit(bit, line->erase_bitmap));
 994                 spin_unlock(&line->lock);
 995
 996                 ret = pblk_blk_erase_sync(pblk, ppa);
 997                 if (ret) {
 998                         pblk_err(pblk, "failed to erase line %d\n", line->id);
 999                         return ret;
1000                 }
1001         } while (1);
1002
1003         return 0;
1004 }
1005
1006 static void pblk_line_setup_metadata(struct pblk_line *line,
1007                                      struct pblk_line_mgmt *l_mg,
1008                                      struct pblk_line_meta *lm)
1009 {
1010         int meta_line;
1011
1012         lockdep_assert_held(&l_mg->free_lock);
1013
1014 retry_meta:
1015         meta_line = find_first_zero_bit(&l_mg->meta_bitmap, PBLK_DATA_LINES);
1016         if (meta_line == PBLK_DATA_LINES) {
1017                 spin_unlock(&l_mg->free_lock);
1018                 io_schedule();
1019                 spin_lock(&l_mg->free_lock);
1020                 goto retry_meta;
1021         }
1022
1023         set_bit(meta_line, &l_mg->meta_bitmap);
1024         line->meta_line = meta_line;
1025
1026         line->smeta = l_mg->sline_meta[meta_line];
1027         line->emeta = l_mg->eline_meta[meta_line];
1028
1029         memset(line->smeta, 0, lm->smeta_len);
1030         memset(line->emeta->buf, 0, lm->emeta_len[0]);
1031
1032         line->emeta->mem = 0;
1033         atomic_set(&line->emeta->sync, 0);
1034 }
1035
1036 /* For now lines are always assumed full lines. Thus, smeta former and current
1037  * lun bitmaps are omitted.
1038  */
1039 static int pblk_line_init_metadata(struct pblk *pblk, struct pblk_line *line,
1040                                   struct pblk_line *cur)
1041 {
1042         struct nvm_tgt_dev *dev = pblk->dev;
1043         struct nvm_geo *geo = &dev->geo;
1044         struct pblk_line_meta *lm = &pblk->lm;
1045         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1046         struct pblk_emeta *emeta = line->emeta;
1047         struct line_emeta *emeta_buf = emeta->buf;
1048         struct line_smeta *smeta_buf = (struct line_smeta *)line->smeta;
1049         int nr_blk_line;
1050
1051         /* After erasing the line, new bad blocks might appear and we risk
1052          * having an invalid line
1053          */
1054         nr_blk_line = lm->blk_per_line -
1055                         bitmap_weight(line->blk_bitmap, lm->blk_per_line);
1056         if (nr_blk_line < lm->min_blk_line) {
1057                 spin_lock(&l_mg->free_lock);
1058                 spin_lock(&line->lock);
1059                 line->state = PBLK_LINESTATE_BAD;
1060                 trace_pblk_line_state(pblk_disk_name(pblk), line->id,
1061                                         line->state);
1062                 spin_unlock(&line->lock);
1063
1064                 list_add_tail(&line->list, &l_mg->bad_list);
1065                 spin_unlock(&l_mg->free_lock);
1066
1067                 pblk_debug(pblk, "line %d is bad\n", line->id);
1068
1069                 return 0;
1070         }
1071
1072         /* Run-time metadata */
1073         line->lun_bitmap = ((void *)(smeta_buf)) + sizeof(struct line_smeta);
1074
1075         /* Mark LUNs allocated in this line (all for now) */
1076         bitmap_set(line->lun_bitmap, 0, lm->lun_bitmap_len);
1077
1078         smeta_buf->header.identifier = cpu_to_le32(PBLK_MAGIC);
1079         guid_copy((guid_t *)&smeta_buf->header.uuid, &pblk->instance_uuid);
1080         smeta_buf->header.id = cpu_to_le32(line->id);
1081         smeta_buf->header.type = cpu_to_le16(line->type);
1082         smeta_buf->header.version_major = SMETA_VERSION_MAJOR;
1083         smeta_buf->header.version_minor = SMETA_VERSION_MINOR;
1084
1085         /* Start metadata */
1086         smeta_buf->seq_nr = cpu_to_le64(line->seq_nr);
1087         smeta_buf->window_wr_lun = cpu_to_le32(geo->all_luns);
1088
1089         /* Fill metadata among lines */
1090         if (cur) {
1091                 memcpy(line->lun_bitmap, cur->lun_bitmap, lm->lun_bitmap_len);
1092                 smeta_buf->prev_id = cpu_to_le32(cur->id);
1093                 cur->emeta->buf->next_id = cpu_to_le32(line->id);
1094         } else {
1095                 smeta_buf->prev_id = cpu_to_le32(PBLK_LINE_EMPTY);
1096         }
1097
1098         /* All smeta must be set at this point */
1099         smeta_buf->header.crc = cpu_to_le32(
1100                         pblk_calc_meta_header_crc(pblk, &smeta_buf->header));
1101         smeta_buf->crc = cpu_to_le32(pblk_calc_smeta_crc(pblk, smeta_buf));
1102
1103         /* End metadata */
1104         memcpy(&emeta_buf->header, &smeta_buf->header,
1105                                                 sizeof(struct line_header));
1106
1107         emeta_buf->header.version_major = EMETA_VERSION_MAJOR;
1108         emeta_buf->header.version_minor = EMETA_VERSION_MINOR;
1109         emeta_buf->header.crc = cpu_to_le32(
1110                         pblk_calc_meta_header_crc(pblk, &emeta_buf->header));
1111
1112         emeta_buf->seq_nr = cpu_to_le64(line->seq_nr);
1113         emeta_buf->nr_lbas = cpu_to_le64(line->sec_in_line);
1114         emeta_buf->nr_valid_lbas = cpu_to_le64(0);
1115         emeta_buf->next_id = cpu_to_le32(PBLK_LINE_EMPTY);
1116         emeta_buf->crc = cpu_to_le32(0);
1117         emeta_buf->prev_id = smeta_buf->prev_id;
1118
1119         return 1;
1120 }
1121
1122 static int pblk_line_alloc_bitmaps(struct pblk *pblk, struct pblk_line *line)
1123 {
1124         struct pblk_line_meta *lm = &pblk->lm;
1125         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1126
1127         line->map_bitmap = mempool_alloc(l_mg->bitmap_pool, GFP_KERNEL);
1128         if (!line->map_bitmap)
1129                 return -ENOMEM;
1130
1131         memset(line->map_bitmap, 0, lm->sec_bitmap_len);
1132
1133         /* will be initialized using bb info from map_bitmap */
1134         line->invalid_bitmap = mempool_alloc(l_mg->bitmap_pool, GFP_KERNEL);
1135         if (!line->invalid_bitmap) {
1136                 mempool_free(line->map_bitmap, l_mg->bitmap_pool);
1137                 line->map_bitmap = NULL;
1138                 return -ENOMEM;
1139         }
1140
1141         return 0;
1142 }
1143
1144 /* For now lines are always assumed full lines. Thus, smeta former and current
1145  * lun bitmaps are omitted.
1146  */
1147 static int pblk_line_init_bb(struct pblk *pblk, struct pblk_line *line,
1148                              int init)
1149 {
1150         struct nvm_tgt_dev *dev = pblk->dev;
1151         struct nvm_geo *geo = &dev->geo;
1152         struct pblk_line_meta *lm = &pblk->lm;
1153         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1154         u64 off;
1155         int bit = -1;
1156         int emeta_secs;
1157
1158         line->sec_in_line = lm->sec_per_line;
1159
1160         /* Capture bad block information on line mapping bitmaps */
1161         while ((bit = find_next_bit(line->blk_bitmap, lm->blk_per_line,
1162                                         bit + 1)) < lm->blk_per_line) {
1163                 off = bit * geo->ws_opt;
1164                 bitmap_shift_left(l_mg->bb_aux, l_mg->bb_template, off,
1165                                                         lm->sec_per_line);
1166                 bitmap_or(line->map_bitmap, line->map_bitmap, l_mg->bb_aux,
1167                                                         lm->sec_per_line);
1168                 line->sec_in_line -= geo->clba;
1169         }
1170
1171         /* Mark smeta metadata sectors as bad sectors */
1172         bit = find_first_zero_bit(line->blk_bitmap, lm->blk_per_line);
1173         off = bit * geo->ws_opt;
1174         bitmap_set(line->map_bitmap, off, lm->smeta_sec);
1175         line->sec_in_line -= lm->smeta_sec;
1176         line->cur_sec = off + lm->smeta_sec;
1177
1178         if (init && pblk_line_smeta_write(pblk, line, off)) {
1179                 pblk_debug(pblk, "line smeta I/O failed. Retry\n");
1180                 return 0;
1181         }
1182
1183         bitmap_copy(line->invalid_bitmap, line->map_bitmap, lm->sec_per_line);
1184
1185         /* Mark emeta metadata sectors as bad sectors. We need to consider bad
1186          * blocks to make sure that there are enough sectors to store emeta
1187          */
1188         emeta_secs = lm->emeta_sec[0];
1189         off = lm->sec_per_line;
1190         while (emeta_secs) {
1191                 off -= geo->ws_opt;
1192                 if (!test_bit(off, line->invalid_bitmap)) {
1193                         bitmap_set(line->invalid_bitmap, off, geo->ws_opt);
1194                         emeta_secs -= geo->ws_opt;
1195                 }
1196         }
1197
1198         line->emeta_ssec = off;
1199         line->sec_in_line -= lm->emeta_sec[0];
1200         line->nr_valid_lbas = 0;
1201         line->left_msecs = line->sec_in_line;
1202         *line->vsc = cpu_to_le32(line->sec_in_line);
1203
1204         if (lm->sec_per_line - line->sec_in_line !=
1205                 bitmap_weight(line->invalid_bitmap, lm->sec_per_line)) {
1206                 spin_lock(&line->lock);
1207                 line->state = PBLK_LINESTATE_BAD;
1208                 trace_pblk_line_state(pblk_disk_name(pblk), line->id,
1209                                         line->state);
1210                 spin_unlock(&line->lock);
1211
1212                 list_add_tail(&line->list, &l_mg->bad_list);
1213                 pblk_err(pblk, "unexpected line %d is bad\n", line->id);
1214
1215                 return 0;
1216         }
1217
1218         return 1;
1219 }
1220
1221 static int pblk_prepare_new_line(struct pblk *pblk, struct pblk_line *line)
1222 {
1223         struct pblk_line_meta *lm = &pblk->lm;
1224         struct nvm_tgt_dev *dev = pblk->dev;
1225         struct nvm_geo *geo = &dev->geo;
1226         int blk_to_erase = atomic_read(&line->blk_in_line);
1227         int i;
1228
1229         for (i = 0; i < lm->blk_per_line; i++) {
1230                 struct pblk_lun *rlun = &pblk->luns[i];
1231                 int pos = pblk_ppa_to_pos(geo, rlun->bppa);
1232                 int state = line->chks[pos].state;
1233
1234                 /* Free chunks should not be erased */
1235                 if (state & NVM_CHK_ST_FREE) {
1236                         set_bit(pblk_ppa_to_pos(geo, rlun->bppa),
1237                                                         line->erase_bitmap);
1238                         blk_to_erase--;
1239                 }
1240         }
1241
1242         return blk_to_erase;
1243 }
1244
1245 static int pblk_line_prepare(struct pblk *pblk, struct pblk_line *line)
1246 {
1247         struct pblk_line_meta *lm = &pblk->lm;
1248         int blk_in_line = atomic_read(&line->blk_in_line);
1249         int blk_to_erase;
1250
1251         /* Bad blocks do not need to be erased */
1252         bitmap_copy(line->erase_bitmap, line->blk_bitmap, lm->blk_per_line);
1253
1254         spin_lock(&line->lock);
1255
1256         /* If we have not written to this line, we need to mark up free chunks
1257          * as already erased
1258          */
1259         if (line->state == PBLK_LINESTATE_NEW) {
1260                 blk_to_erase = pblk_prepare_new_line(pblk, line);
1261                 line->state = PBLK_LINESTATE_FREE;
1262                 trace_pblk_line_state(pblk_disk_name(pblk), line->id,
1263                                         line->state);
1264         } else {
1265                 blk_to_erase = blk_in_line;
1266         }
1267
1268         if (blk_in_line < lm->min_blk_line) {
1269                 spin_unlock(&line->lock);
1270                 return -EAGAIN;
1271         }
1272
1273         if (line->state != PBLK_LINESTATE_FREE) {
1274                 WARN(1, "pblk: corrupted line %d, state %d\n",
1275                                                         line->id, line->state);
1276                 spin_unlock(&line->lock);
1277                 return -EINTR;
1278         }
1279
1280         line->state = PBLK_LINESTATE_OPEN;
1281         trace_pblk_line_state(pblk_disk_name(pblk), line->id,
1282                                 line->state);
1283
1284         atomic_set(&line->left_eblks, blk_to_erase);
1285         atomic_set(&line->left_seblks, blk_to_erase);
1286
1287         line->meta_distance = lm->meta_distance;
1288         spin_unlock(&line->lock);
1289
1290         kref_init(&line->ref);
1291         atomic_set(&line->sec_to_update, 0);
1292
1293         return 0;
1294 }
1295
1296 /* Line allocations in the recovery path are always single threaded */
1297 int pblk_line_recov_alloc(struct pblk *pblk, struct pblk_line *line)
1298 {
1299         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1300         int ret;
1301
1302         spin_lock(&l_mg->free_lock);
1303         l_mg->data_line = line;
1304         list_del(&line->list);
1305
1306         ret = pblk_line_prepare(pblk, line);
1307         if (ret) {
1308                 list_add(&line->list, &l_mg->free_list);
1309                 spin_unlock(&l_mg->free_lock);
1310                 return ret;
1311         }
1312         spin_unlock(&l_mg->free_lock);
1313
1314         ret = pblk_line_alloc_bitmaps(pblk, line);
1315         if (ret)
1316                 goto fail;
1317
1318         if (!pblk_line_init_bb(pblk, line, 0)) {
1319                 ret = -EINTR;
1320                 goto fail;
1321         }
1322
1323         pblk_rl_free_lines_dec(&pblk->rl, line, true);
1324         return 0;
1325
1326 fail:
1327         spin_lock(&l_mg->free_lock);
1328         list_add(&line->list, &l_mg->free_list);
1329         spin_unlock(&l_mg->free_lock);
1330
1331         return ret;
1332 }
1333
1334 void pblk_line_recov_close(struct pblk *pblk, struct pblk_line *line)
1335 {
1336         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1337
1338         mempool_free(line->map_bitmap, l_mg->bitmap_pool);
1339         line->map_bitmap = NULL;
1340         line->smeta = NULL;
1341         line->emeta = NULL;
1342 }
1343
1344 static void pblk_line_reinit(struct pblk_line *line)
1345 {
1346         *line->vsc = cpu_to_le32(EMPTY_ENTRY);
1347
1348         line->map_bitmap = NULL;
1349         line->invalid_bitmap = NULL;
1350         line->smeta = NULL;
1351         line->emeta = NULL;
1352 }
1353
1354 void pblk_line_free(struct pblk_line *line)
1355 {
1356         struct pblk *pblk = line->pblk;
1357         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1358
1359         mempool_free(line->map_bitmap, l_mg->bitmap_pool);
1360         mempool_free(line->invalid_bitmap, l_mg->bitmap_pool);
1361
1362         pblk_line_reinit(line);
1363 }
1364
1365 struct pblk_line *pblk_line_get(struct pblk *pblk)
1366 {
1367         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1368         struct pblk_line_meta *lm = &pblk->lm;
1369         struct pblk_line *line;
1370         int ret, bit;
1371
1372         lockdep_assert_held(&l_mg->free_lock);
1373
1374 retry:
1375         if (list_empty(&l_mg->free_list)) {
1376                 pblk_err(pblk, "no free lines\n");
1377                 return NULL;
1378         }
1379
1380         line = list_first_entry(&l_mg->free_list, struct pblk_line, list);
1381         list_del(&line->list);
1382         l_mg->nr_free_lines--;
1383
1384         bit = find_first_zero_bit(line->blk_bitmap, lm->blk_per_line);
1385         if (unlikely(bit >= lm->blk_per_line)) {
1386                 spin_lock(&line->lock);
1387                 line->state = PBLK_LINESTATE_BAD;
1388                 trace_pblk_line_state(pblk_disk_name(pblk), line->id,
1389                                         line->state);
1390                 spin_unlock(&line->lock);
1391
1392                 list_add_tail(&line->list, &l_mg->bad_list);
1393
1394                 pblk_debug(pblk, "line %d is bad\n", line->id);
1395                 goto retry;
1396         }
1397
1398         ret = pblk_line_prepare(pblk, line);
1399         if (ret) {
1400                 switch (ret) {
1401                 case -EAGAIN:
1402                         list_add(&line->list, &l_mg->bad_list);
1403                         goto retry;
1404                 case -EINTR:
1405                         list_add(&line->list, &l_mg->corrupt_list);
1406                         goto retry;
1407                 default:
1408                         pblk_err(pblk, "failed to prepare line %d\n", line->id);
1409                         list_add(&line->list, &l_mg->free_list);
1410                         l_mg->nr_free_lines++;
1411                         return NULL;
1412                 }
1413         }
1414
1415         return line;
1416 }
1417
1418 static struct pblk_line *pblk_line_retry(struct pblk *pblk,
1419                                          struct pblk_line *line)
1420 {
1421         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1422         struct pblk_line *retry_line;
1423
1424 retry:
1425         spin_lock(&l_mg->free_lock);
1426         retry_line = pblk_line_get(pblk);
1427         if (!retry_line) {
1428                 l_mg->data_line = NULL;
1429                 spin_unlock(&l_mg->free_lock);
1430                 return NULL;
1431         }
1432
1433         retry_line->map_bitmap = line->map_bitmap;
1434         retry_line->invalid_bitmap = line->invalid_bitmap;
1435         retry_line->smeta = line->smeta;
1436         retry_line->emeta = line->emeta;
1437         retry_line->meta_line = line->meta_line;
1438
1439         pblk_line_reinit(line);
1440
1441         l_mg->data_line = retry_line;
1442         spin_unlock(&l_mg->free_lock);
1443
1444         pblk_rl_free_lines_dec(&pblk->rl, line, false);
1445
1446         if (pblk_line_erase(pblk, retry_line))
1447                 goto retry;
1448
1449         return retry_line;
1450 }
1451
1452 static void pblk_set_space_limit(struct pblk *pblk)
1453 {
1454         struct pblk_rl *rl = &pblk->rl;
1455
1456         atomic_set(&rl->rb_space, 0);
1457 }
1458
1459 struct pblk_line *pblk_line_get_first_data(struct pblk *pblk)
1460 {
1461         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1462         struct pblk_line *line;
1463
1464         spin_lock(&l_mg->free_lock);
1465         line = pblk_line_get(pblk);
1466         if (!line) {
1467                 spin_unlock(&l_mg->free_lock);
1468                 return NULL;
1469         }
1470
1471         line->seq_nr = l_mg->d_seq_nr++;
1472         line->type = PBLK_LINETYPE_DATA;
1473         l_mg->data_line = line;
1474
1475         pblk_line_setup_metadata(line, l_mg, &pblk->lm);
1476
1477         /* Allocate next line for preparation */
1478         l_mg->data_next = pblk_line_get(pblk);
1479         if (!l_mg->data_next) {
1480                 /* If we cannot get a new line, we need to stop the pipeline.
1481                  * Only allow as many writes in as we can store safely and then
1482                  * fail gracefully
1483                  */
1484                 pblk_set_space_limit(pblk);
1485
1486                 l_mg->data_next = NULL;
1487         } else {
1488                 l_mg->data_next->seq_nr = l_mg->d_seq_nr++;
1489                 l_mg->data_next->type = PBLK_LINETYPE_DATA;
1490         }
1491         spin_unlock(&l_mg->free_lock);
1492
1493         if (pblk_line_alloc_bitmaps(pblk, line))
1494                 return NULL;
1495
1496         if (pblk_line_erase(pblk, line)) {
1497                 line = pblk_line_retry(pblk, line);
1498                 if (!line)
1499                         return NULL;
1500         }
1501
1502 retry_setup:
1503         if (!pblk_line_init_metadata(pblk, line, NULL)) {
1504                 line = pblk_line_retry(pblk, line);
1505                 if (!line)
1506                         return NULL;
1507
1508                 goto retry_setup;
1509         }
1510
1511         if (!pblk_line_init_bb(pblk, line, 1)) {
1512                 line = pblk_line_retry(pblk, line);
1513                 if (!line)
1514                         return NULL;
1515
1516                 goto retry_setup;
1517         }
1518
1519         pblk_rl_free_lines_dec(&pblk->rl, line, true);
1520
1521         return line;
1522 }
1523
1524 void pblk_ppa_to_line_put(struct pblk *pblk, struct ppa_addr ppa)
1525 {
1526         struct pblk_line *line;
1527
1528         line = pblk_ppa_to_line(pblk, ppa);
1529         kref_put(&line->ref, pblk_line_put_wq);
1530 }
1531
1532 void pblk_rq_to_line_put(struct pblk *pblk, struct nvm_rq *rqd)
1533 {
1534         struct ppa_addr *ppa_list = nvm_rq_to_ppa_list(rqd);
1535         int i;
1536
1537         for (i = 0; i < rqd->nr_ppas; i++)
1538                 pblk_ppa_to_line_put(pblk, ppa_list[i]);
1539 }
1540
1541 static void pblk_stop_writes(struct pblk *pblk, struct pblk_line *line)
1542 {
1543         lockdep_assert_held(&pblk->l_mg.free_lock);
1544
1545         pblk_set_space_limit(pblk);
1546         pblk->state = PBLK_STATE_STOPPING;
1547         trace_pblk_state(pblk_disk_name(pblk), pblk->state);
1548 }
1549
1550 static void pblk_line_close_meta_sync(struct pblk *pblk)
1551 {
1552         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1553         struct pblk_line_meta *lm = &pblk->lm;
1554         struct pblk_line *line, *tline;
1555         LIST_HEAD(list);
1556
1557         spin_lock(&l_mg->close_lock);
1558         if (list_empty(&l_mg->emeta_list)) {
1559                 spin_unlock(&l_mg->close_lock);
1560                 return;
1561         }
1562
1563         list_cut_position(&list, &l_mg->emeta_list, l_mg->emeta_list.prev);
1564         spin_unlock(&l_mg->close_lock);
1565
1566         list_for_each_entry_safe(line, tline, &list, list) {
1567                 struct pblk_emeta *emeta = line->emeta;
1568
1569                 while (emeta->mem < lm->emeta_len[0]) {
1570                         int ret;
1571
1572                         ret = pblk_submit_meta_io(pblk, line);
1573                         if (ret) {
1574                                 pblk_err(pblk, "sync meta line %d failed (%d)\n",
1575                                                         line->id, ret);
1576                                 return;
1577                         }
1578                 }
1579         }
1580
1581         pblk_wait_for_meta(pblk);
1582         flush_workqueue(pblk->close_wq);
1583 }
1584
1585 void __pblk_pipeline_flush(struct pblk *pblk)
1586 {
1587         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1588         int ret;
1589
1590         spin_lock(&l_mg->free_lock);
1591         if (pblk->state == PBLK_STATE_RECOVERING ||
1592                                         pblk->state == PBLK_STATE_STOPPED) {
1593                 spin_unlock(&l_mg->free_lock);
1594                 return;
1595         }
1596         pblk->state = PBLK_STATE_RECOVERING;
1597         trace_pblk_state(pblk_disk_name(pblk), pblk->state);
1598         spin_unlock(&l_mg->free_lock);
1599
1600         pblk_flush_writer(pblk);
1601         pblk_wait_for_meta(pblk);
1602
1603         ret = pblk_recov_pad(pblk);
1604         if (ret) {
1605                 pblk_err(pblk, "could not close data on teardown(%d)\n", ret);
1606                 return;
1607         }
1608
1609         flush_workqueue(pblk->bb_wq);
1610         pblk_line_close_meta_sync(pblk);
1611 }
1612
1613 void __pblk_pipeline_stop(struct pblk *pblk)
1614 {
1615         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1616
1617         spin_lock(&l_mg->free_lock);
1618         pblk->state = PBLK_STATE_STOPPED;
1619         trace_pblk_state(pblk_disk_name(pblk), pblk->state);
1620         l_mg->data_line = NULL;
1621         l_mg->data_next = NULL;
1622         spin_unlock(&l_mg->free_lock);
1623 }
1624
1625 void pblk_pipeline_stop(struct pblk *pblk)
1626 {
1627         __pblk_pipeline_flush(pblk);
1628         __pblk_pipeline_stop(pblk);
1629 }
1630
1631 struct pblk_line *pblk_line_replace_data(struct pblk *pblk)
1632 {
1633         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1634         struct pblk_line *cur, *new = NULL;
1635         unsigned int left_seblks;
1636
1637         new = l_mg->data_next;
1638         if (!new)
1639                 goto out;
1640
1641         spin_lock(&l_mg->free_lock);
1642         cur = l_mg->data_line;
1643         l_mg->data_line = new;
1644
1645         pblk_line_setup_metadata(new, l_mg, &pblk->lm);
1646         spin_unlock(&l_mg->free_lock);
1647
1648 retry_erase:
1649         left_seblks = atomic_read(&new->left_seblks);
1650         if (left_seblks) {
1651                 /* If line is not fully erased, erase it */
1652                 if (atomic_read(&new->left_eblks)) {
1653                         if (pblk_line_erase(pblk, new))
1654                                 goto out;
1655                 } else {
1656                         io_schedule();
1657                 }
1658                 goto retry_erase;
1659         }
1660
1661         if (pblk_line_alloc_bitmaps(pblk, new))
1662                 return NULL;
1663
1664 retry_setup:
1665         if (!pblk_line_init_metadata(pblk, new, cur)) {
1666                 new = pblk_line_retry(pblk, new);
1667                 if (!new)
1668                         goto out;
1669
1670                 goto retry_setup;
1671         }
1672
1673         if (!pblk_line_init_bb(pblk, new, 1)) {
1674                 new = pblk_line_retry(pblk, new);
1675                 if (!new)
1676                         goto out;
1677
1678                 goto retry_setup;
1679         }
1680
1681         pblk_rl_free_lines_dec(&pblk->rl, new, true);
1682
1683         /* Allocate next line for preparation */
1684         spin_lock(&l_mg->free_lock);
1685         l_mg->data_next = pblk_line_get(pblk);
1686         if (!l_mg->data_next) {
1687                 /* If we cannot get a new line, we need to stop the pipeline.
1688                  * Only allow as many writes in as we can store safely and then
1689                  * fail gracefully
1690                  */
1691                 pblk_stop_writes(pblk, new);
1692                 l_mg->data_next = NULL;
1693         } else {
1694                 l_mg->data_next->seq_nr = l_mg->d_seq_nr++;
1695                 l_mg->data_next->type = PBLK_LINETYPE_DATA;
1696         }
1697         spin_unlock(&l_mg->free_lock);
1698
1699 out:
1700         return new;
1701 }
1702
1703 static void __pblk_line_put(struct pblk *pblk, struct pblk_line *line)
1704 {
1705         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1706         struct pblk_gc *gc = &pblk->gc;
1707
1708         spin_lock(&line->lock);
1709         WARN_ON(line->state != PBLK_LINESTATE_GC);
1710         if (line->w_err_gc->has_gc_err) {
1711                 spin_unlock(&line->lock);
1712                 pblk_err(pblk, "line %d had errors during GC\n", line->id);
1713                 pblk_put_line_back(pblk, line);
1714                 line->w_err_gc->has_gc_err = 0;
1715                 return;
1716         }
1717
1718         line->state = PBLK_LINESTATE_FREE;
1719         trace_pblk_line_state(pblk_disk_name(pblk), line->id,
1720                                         line->state);
1721         line->gc_group = PBLK_LINEGC_NONE;
1722         pblk_line_free(line);
1723
1724         if (line->w_err_gc->has_write_err) {
1725                 pblk_rl_werr_line_out(&pblk->rl);
1726                 line->w_err_gc->has_write_err = 0;
1727         }
1728
1729         spin_unlock(&line->lock);
1730         atomic_dec(&gc->pipeline_gc);
1731
1732         spin_lock(&l_mg->free_lock);
1733         list_add_tail(&line->list, &l_mg->free_list);
1734         l_mg->nr_free_lines++;
1735         spin_unlock(&l_mg->free_lock);
1736
1737         pblk_rl_free_lines_inc(&pblk->rl, line);
1738 }
1739
1740 static void pblk_line_put_ws(struct work_struct *work)
1741 {
1742         struct pblk_line_ws *line_put_ws = container_of(work,
1743                                                 struct pblk_line_ws, ws);
1744         struct pblk *pblk = line_put_ws->pblk;
1745         struct pblk_line *line = line_put_ws->line;
1746
1747         __pblk_line_put(pblk, line);
1748         mempool_free(line_put_ws, &pblk->gen_ws_pool);
1749 }
1750
1751 void pblk_line_put(struct kref *ref)
1752 {
1753         struct pblk_line *line = container_of(ref, struct pblk_line, ref);
1754         struct pblk *pblk = line->pblk;
1755
1756         __pblk_line_put(pblk, line);
1757 }
1758
1759 void pblk_line_put_wq(struct kref *ref)
1760 {
1761         struct pblk_line *line = container_of(ref, struct pblk_line, ref);
1762         struct pblk *pblk = line->pblk;
1763         struct pblk_line_ws *line_put_ws;
1764
1765         line_put_ws = mempool_alloc(&pblk->gen_ws_pool, GFP_ATOMIC);
1766         if (!line_put_ws)
1767                 return;
1768
1769         line_put_ws->pblk = pblk;
1770         line_put_ws->line = line;
1771         line_put_ws->priv = NULL;
1772
1773         INIT_WORK(&line_put_ws->ws, pblk_line_put_ws);
1774         queue_work(pblk->r_end_wq, &line_put_ws->ws);
1775 }
1776
1777 int pblk_blk_erase_async(struct pblk *pblk, struct ppa_addr ppa)
1778 {
1779         struct nvm_rq *rqd;
1780         int err;
1781
1782         rqd = pblk_alloc_rqd(pblk, PBLK_ERASE);
1783
1784         pblk_setup_e_rq(pblk, rqd, ppa);
1785
1786         rqd->end_io = pblk_end_io_erase;
1787         rqd->private = pblk;
1788
1789         trace_pblk_chunk_reset(pblk_disk_name(pblk),
1790                                 &ppa, PBLK_CHUNK_RESET_START);
1791
1792         /* The write thread schedules erases so that it minimizes disturbances
1793          * with writes. Thus, there is no need to take the LUN semaphore.
1794          */
1795         err = pblk_submit_io(pblk, rqd);
1796         if (err) {
1797                 struct nvm_tgt_dev *dev = pblk->dev;
1798                 struct nvm_geo *geo = &dev->geo;
1799
1800                 pblk_err(pblk, "could not async erase line:%d,blk:%d\n",
1801                                         pblk_ppa_to_line_id(ppa),
1802                                         pblk_ppa_to_pos(geo, ppa));
1803         }
1804
1805         return err;
1806 }
1807
1808 struct pblk_line *pblk_line_get_data(struct pblk *pblk)
1809 {
1810         return pblk->l_mg.data_line;
1811 }
1812
1813 /* For now, always erase next line */
1814 struct pblk_line *pblk_line_get_erase(struct pblk *pblk)
1815 {
1816         return pblk->l_mg.data_next;
1817 }
1818
1819 int pblk_line_is_full(struct pblk_line *line)
1820 {
1821         return (line->left_msecs == 0);
1822 }
1823
1824 static void pblk_line_should_sync_meta(struct pblk *pblk)
1825 {
1826         if (pblk_rl_is_limit(&pblk->rl))
1827                 pblk_line_close_meta_sync(pblk);
1828 }
1829
1830 void pblk_line_close(struct pblk *pblk, struct pblk_line *line)
1831 {
1832         struct nvm_tgt_dev *dev = pblk->dev;
1833         struct nvm_geo *geo = &dev->geo;
1834         struct pblk_line_meta *lm = &pblk->lm;
1835         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1836         struct list_head *move_list;
1837         int i;
1838
1839 #ifdef CONFIG_NVM_PBLK_DEBUG
1840         WARN(!bitmap_full(line->map_bitmap, lm->sec_per_line),
1841                                 "pblk: corrupt closed line %d\n", line->id);
1842 #endif
1843
1844         spin_lock(&l_mg->free_lock);
1845         WARN_ON(!test_and_clear_bit(line->meta_line, &l_mg->meta_bitmap));
1846         spin_unlock(&l_mg->free_lock);
1847
1848         spin_lock(&l_mg->gc_lock);
1849         spin_lock(&line->lock);
1850         WARN_ON(line->state != PBLK_LINESTATE_OPEN);
1851         line->state = PBLK_LINESTATE_CLOSED;
1852         move_list = pblk_line_gc_list(pblk, line);
1853         list_add_tail(&line->list, move_list);
1854
1855         mempool_free(line->map_bitmap, l_mg->bitmap_pool);
1856         line->map_bitmap = NULL;
1857         line->smeta = NULL;
1858         line->emeta = NULL;
1859
1860         for (i = 0; i < lm->blk_per_line; i++) {
1861                 struct pblk_lun *rlun = &pblk->luns[i];
1862                 int pos = pblk_ppa_to_pos(geo, rlun->bppa);
1863                 int state = line->chks[pos].state;
1864
1865                 if (!(state & NVM_CHK_ST_OFFLINE))
1866                         state = NVM_CHK_ST_CLOSED;
1867         }
1868
1869         spin_unlock(&line->lock);
1870         spin_unlock(&l_mg->gc_lock);
1871
1872         trace_pblk_line_state(pblk_disk_name(pblk), line->id,
1873                                         line->state);
1874 }
1875
1876 void pblk_line_close_meta(struct pblk *pblk, struct pblk_line *line)
1877 {
1878         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1879         struct pblk_line_meta *lm = &pblk->lm;
1880         struct pblk_emeta *emeta = line->emeta;
1881         struct line_emeta *emeta_buf = emeta->buf;
1882         struct wa_counters *wa = emeta_to_wa(lm, emeta_buf);
1883
1884         /* No need for exact vsc value; avoid a big line lock and take aprox. */
1885         memcpy(emeta_to_vsc(pblk, emeta_buf), l_mg->vsc_list, lm->vsc_list_len);
1886         memcpy(emeta_to_bb(emeta_buf), line->blk_bitmap, lm->blk_bitmap_len);
1887
1888         wa->user = cpu_to_le64(atomic64_read(&pblk->user_wa));
1889         wa->pad = cpu_to_le64(atomic64_read(&pblk->pad_wa));
1890         wa->gc = cpu_to_le64(atomic64_read(&pblk->gc_wa));
1891
1892         if (le32_to_cpu(emeta_buf->header.identifier) != PBLK_MAGIC) {
1893                 emeta_buf->header.identifier = cpu_to_le32(PBLK_MAGIC);
1894                 guid_copy((guid_t *)&emeta_buf->header.uuid,
1895                                                         &pblk->instance_uuid);
1896                 emeta_buf->header.id = cpu_to_le32(line->id);
1897                 emeta_buf->header.type = cpu_to_le16(line->type);
1898                 emeta_buf->header.version_major = EMETA_VERSION_MAJOR;
1899                 emeta_buf->header.version_minor = EMETA_VERSION_MINOR;
1900                 emeta_buf->header.crc = cpu_to_le32(
1901                         pblk_calc_meta_header_crc(pblk, &emeta_buf->header));
1902         }
1903
1904         emeta_buf->nr_valid_lbas = cpu_to_le64(line->nr_valid_lbas);
1905         emeta_buf->crc = cpu_to_le32(pblk_calc_emeta_crc(pblk, emeta_buf));
1906
1907         spin_lock(&l_mg->close_lock);
1908         spin_lock(&line->lock);
1909
1910         /* Update the in-memory start address for emeta, in case it has
1911          * shifted due to write errors
1912          */
1913         if (line->emeta_ssec != line->cur_sec)
1914                 line->emeta_ssec = line->cur_sec;
1915
1916         list_add_tail(&line->list, &l_mg->emeta_list);
1917         spin_unlock(&line->lock);
1918         spin_unlock(&l_mg->close_lock);
1919
1920         pblk_line_should_sync_meta(pblk);
1921 }
1922
1923 static void pblk_save_lba_list(struct pblk *pblk, struct pblk_line *line)
1924 {
1925         struct pblk_line_meta *lm = &pblk->lm;
1926         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1927         unsigned int lba_list_size = lm->emeta_len[2];
1928         struct pblk_w_err_gc *w_err_gc = line->w_err_gc;
1929         struct pblk_emeta *emeta = line->emeta;
1930
1931         w_err_gc->lba_list = pblk_malloc(lba_list_size,
1932                                          l_mg->emeta_alloc_type, GFP_KERNEL);
1933         memcpy(w_err_gc->lba_list, emeta_to_lbas(pblk, emeta->buf),
1934                                 lba_list_size);
1935 }
1936
1937 void pblk_line_close_ws(struct work_struct *work)
1938 {
1939         struct pblk_line_ws *line_ws = container_of(work, struct pblk_line_ws,
1940                                                                         ws);
1941         struct pblk *pblk = line_ws->pblk;
1942         struct pblk_line *line = line_ws->line;
1943         struct pblk_w_err_gc *w_err_gc = line->w_err_gc;
1944
1945         /* Write errors makes the emeta start address stored in smeta invalid,
1946          * so keep a copy of the lba list until we've gc'd the line
1947          */
1948         if (w_err_gc->has_write_err)
1949                 pblk_save_lba_list(pblk, line);
1950
1951         pblk_line_close(pblk, line);
1952         mempool_free(line_ws, &pblk->gen_ws_pool);
1953 }
1954
1955 void pblk_gen_run_ws(struct pblk *pblk, struct pblk_line *line, void *priv,
1956                       void (*work)(struct work_struct *), gfp_t gfp_mask,
1957                       struct workqueue_struct *wq)
1958 {
1959         struct pblk_line_ws *line_ws;
1960
1961         line_ws = mempool_alloc(&pblk->gen_ws_pool, gfp_mask);
1962
1963         line_ws->pblk = pblk;
1964         line_ws->line = line;
1965         line_ws->priv = priv;
1966
1967         INIT_WORK(&line_ws->ws, work);
1968         queue_work(wq, &line_ws->ws);
1969 }
1970
1971 static void __pblk_down_chunk(struct pblk *pblk, int pos)
1972 {
1973         struct pblk_lun *rlun = &pblk->luns[pos];
1974         int ret;
1975
1976         /*
1977          * Only send one inflight I/O per LUN. Since we map at a page
1978          * granurality, all ppas in the I/O will map to the same LUN
1979          */
1980
1981         ret = down_timeout(&rlun->wr_sem, msecs_to_jiffies(30000));
1982         if (ret == -ETIME || ret == -EINTR)
1983                 pblk_err(pblk, "taking lun semaphore timed out: err %d\n",
1984                                 -ret);
1985 }
1986
1987 void pblk_down_chunk(struct pblk *pblk, struct ppa_addr ppa)
1988 {
1989         struct nvm_tgt_dev *dev = pblk->dev;
1990         struct nvm_geo *geo = &dev->geo;
1991         int pos = pblk_ppa_to_pos(geo, ppa);
1992
1993         __pblk_down_chunk(pblk, pos);
1994 }
1995
1996 void pblk_down_rq(struct pblk *pblk, struct ppa_addr ppa,
1997                   unsigned long *lun_bitmap)
1998 {
1999         struct nvm_tgt_dev *dev = pblk->dev;
2000         struct nvm_geo *geo = &dev->geo;
2001         int pos = pblk_ppa_to_pos(geo, ppa);
2002
2003         /* If the LUN has been locked for this same request, do no attempt to
2004          * lock it again
2005          */
2006         if (test_and_set_bit(pos, lun_bitmap))
2007                 return;
2008
2009         __pblk_down_chunk(pblk, pos);
2010 }
2011
2012 void pblk_up_chunk(struct pblk *pblk, struct ppa_addr ppa)
2013 {
2014         struct nvm_tgt_dev *dev = pblk->dev;
2015         struct nvm_geo *geo = &dev->geo;
2016         struct pblk_lun *rlun;
2017         int pos = pblk_ppa_to_pos(geo, ppa);
2018
2019         rlun = &pblk->luns[pos];
2020         up(&rlun->wr_sem);
2021 }
2022
2023 void pblk_up_rq(struct pblk *pblk, unsigned long *lun_bitmap)
2024 {
2025         struct nvm_tgt_dev *dev = pblk->dev;
2026         struct nvm_geo *geo = &dev->geo;
2027         struct pblk_lun *rlun;
2028         int num_lun = geo->all_luns;
2029         int bit = -1;
2030
2031         while ((bit = find_next_bit(lun_bitmap, num_lun, bit + 1)) < num_lun) {
2032                 rlun = &pblk->luns[bit];
2033                 up(&rlun->wr_sem);
2034         }
2035 }
2036
2037 void pblk_update_map(struct pblk *pblk, sector_t lba, struct ppa_addr ppa)
2038 {
2039         struct ppa_addr ppa_l2p;
2040
2041         /* logic error: lba out-of-bounds. Ignore update */
2042         if (!(lba < pblk->capacity)) {
2043                 WARN(1, "pblk: corrupted L2P map request\n");
2044                 return;
2045         }
2046
2047         spin_lock(&pblk->trans_lock);
2048         ppa_l2p = pblk_trans_map_get(pblk, lba);
2049
2050         if (!pblk_addr_in_cache(ppa_l2p) && !pblk_ppa_empty(ppa_l2p))
2051                 pblk_map_invalidate(pblk, ppa_l2p);
2052
2053         pblk_trans_map_set(pblk, lba, ppa);
2054         spin_unlock(&pblk->trans_lock);
2055 }
2056
2057 void pblk_update_map_cache(struct pblk *pblk, sector_t lba, struct ppa_addr ppa)
2058 {
2059
2060 #ifdef CONFIG_NVM_PBLK_DEBUG
2061         /* Callers must ensure that the ppa points to a cache address */
2062         BUG_ON(!pblk_addr_in_cache(ppa));
2063         BUG_ON(pblk_rb_pos_oob(&pblk->rwb, pblk_addr_to_cacheline(ppa)));
2064 #endif
2065
2066         pblk_update_map(pblk, lba, ppa);
2067 }
2068
2069 int pblk_update_map_gc(struct pblk *pblk, sector_t lba, struct ppa_addr ppa_new,
2070                        struct pblk_line *gc_line, u64 paddr_gc)
2071 {
2072         struct ppa_addr ppa_l2p, ppa_gc;
2073         int ret = 1;
2074
2075 #ifdef CONFIG_NVM_PBLK_DEBUG
2076         /* Callers must ensure that the ppa points to a cache address */
2077         BUG_ON(!pblk_addr_in_cache(ppa_new));
2078         BUG_ON(pblk_rb_pos_oob(&pblk->rwb, pblk_addr_to_cacheline(ppa_new)));
2079 #endif
2080
2081         /* logic error: lba out-of-bounds. Ignore update */
2082         if (!(lba < pblk->capacity)) {
2083                 WARN(1, "pblk: corrupted L2P map request\n");
2084                 return 0;
2085         }
2086
2087         spin_lock(&pblk->trans_lock);
2088         ppa_l2p = pblk_trans_map_get(pblk, lba);
2089         ppa_gc = addr_to_gen_ppa(pblk, paddr_gc, gc_line->id);
2090
2091         if (!pblk_ppa_comp(ppa_l2p, ppa_gc)) {
2092                 spin_lock(&gc_line->lock);
2093                 WARN(!test_bit(paddr_gc, gc_line->invalid_bitmap),
2094                                                 "pblk: corrupted GC update");
2095                 spin_unlock(&gc_line->lock);
2096
2097                 ret = 0;
2098                 goto out;
2099         }
2100
2101         pblk_trans_map_set(pblk, lba, ppa_new);
2102 out:
2103         spin_unlock(&pblk->trans_lock);
2104         return ret;
2105 }
2106
2107 void pblk_update_map_dev(struct pblk *pblk, sector_t lba,
2108                          struct ppa_addr ppa_mapped, struct ppa_addr ppa_cache)
2109 {
2110         struct ppa_addr ppa_l2p;
2111
2112 #ifdef CONFIG_NVM_PBLK_DEBUG
2113         /* Callers must ensure that the ppa points to a device address */
2114         BUG_ON(pblk_addr_in_cache(ppa_mapped));
2115 #endif
2116         /* Invalidate and discard padded entries */
2117         if (lba == ADDR_EMPTY) {
2118                 atomic64_inc(&pblk->pad_wa);
2119 #ifdef CONFIG_NVM_PBLK_DEBUG
2120                 atomic_long_inc(&pblk->padded_wb);
2121 #endif
2122                 if (!pblk_ppa_empty(ppa_mapped))
2123                         pblk_map_invalidate(pblk, ppa_mapped);
2124                 return;
2125         }
2126
2127         /* logic error: lba out-of-bounds. Ignore update */
2128         if (!(lba < pblk->capacity)) {
2129                 WARN(1, "pblk: corrupted L2P map request\n");
2130                 return;
2131         }
2132
2133         spin_lock(&pblk->trans_lock);
2134         ppa_l2p = pblk_trans_map_get(pblk, lba);
2135
2136         /* Do not update L2P if the cacheline has been updated. In this case,
2137          * the mapped ppa must be invalidated
2138          */
2139         if (!pblk_ppa_comp(ppa_l2p, ppa_cache)) {
2140                 if (!pblk_ppa_empty(ppa_mapped))
2141                         pblk_map_invalidate(pblk, ppa_mapped);
2142                 goto out;
2143         }
2144
2145 #ifdef CONFIG_NVM_PBLK_DEBUG
2146         WARN_ON(!pblk_addr_in_cache(ppa_l2p) && !pblk_ppa_empty(ppa_l2p));
2147 #endif
2148
2149         pblk_trans_map_set(pblk, lba, ppa_mapped);
2150 out:
2151         spin_unlock(&pblk->trans_lock);
2152 }
2153
2154 int pblk_lookup_l2p_seq(struct pblk *pblk, struct ppa_addr *ppas,
2155                          sector_t blba, int nr_secs, bool *from_cache)
2156 {
2157         int i;
2158
2159         spin_lock(&pblk->trans_lock);
2160         for (i = 0; i < nr_secs; i++) {
2161                 struct ppa_addr ppa;
2162
2163                 ppa = ppas[i] = pblk_trans_map_get(pblk, blba + i);
2164
2165                 /* If the L2P entry maps to a line, the reference is valid */
2166                 if (!pblk_ppa_empty(ppa) && !pblk_addr_in_cache(ppa)) {
2167                         struct pblk_line *line = pblk_ppa_to_line(pblk, ppa);
2168
2169                         if (i > 0 && *from_cache)
2170                                 break;
2171                         *from_cache = false;
2172
2173                         kref_get(&line->ref);
2174                 } else {
2175                         if (i > 0 && !*from_cache)
2176                                 break;
2177                         *from_cache = true;
2178                 }
2179         }
2180         spin_unlock(&pblk->trans_lock);
2181         return i;
2182 }
2183
2184 void pblk_lookup_l2p_rand(struct pblk *pblk, struct ppa_addr *ppas,
2185                           u64 *lba_list, int nr_secs)
2186 {
2187         u64 lba;
2188         int i;
2189
2190         spin_lock(&pblk->trans_lock);
2191         for (i = 0; i < nr_secs; i++) {
2192                 lba = lba_list[i];
2193                 if (lba != ADDR_EMPTY) {
2194                         /* logic error: lba out-of-bounds. Ignore update */
2195                         if (!(lba < pblk->capacity)) {
2196                                 WARN(1, "pblk: corrupted L2P map request\n");
2197                                 continue;
2198                         }
2199                         ppas[i] = pblk_trans_map_get(pblk, lba);
2200                 }
2201         }
2202         spin_unlock(&pblk->trans_lock);
2203 }
2204
2205 void *pblk_get_meta_for_writes(struct pblk *pblk, struct nvm_rq *rqd)
2206 {
2207         void *buffer;
2208
2209         if (pblk_is_oob_meta_supported(pblk)) {
2210                 /* Just use OOB metadata buffer as always */
2211                 buffer = rqd->meta_list;
2212         } else {
2213                 /* We need to reuse last page of request (packed metadata)
2214                  * in similar way as traditional oob metadata
2215                  */
2216                 buffer = page_to_virt(
2217                         rqd->bio->bi_io_vec[rqd->bio->bi_vcnt - 1].bv_page);
2218         }
2219
2220         return buffer;
2221 }
2222
2223 void pblk_get_packed_meta(struct pblk *pblk, struct nvm_rq *rqd)
2224 {
2225         void *meta_list = rqd->meta_list;
2226         void *page;
2227         int i = 0;
2228
2229         if (pblk_is_oob_meta_supported(pblk))
2230                 return;
2231
2232         page = page_to_virt(rqd->bio->bi_io_vec[rqd->bio->bi_vcnt - 1].bv_page);
2233         /* We need to fill oob meta buffer with data from packed metadata */
2234         for (; i < rqd->nr_ppas; i++)
2235                 memcpy(pblk_get_meta(pblk, meta_list, i),
2236                         page + (i * sizeof(struct pblk_sec_meta)),
2237                         sizeof(struct pblk_sec_meta));
2238 }