drivers/lightnvm/pblk-core.c

   1 /*
   2  * Copyright (C) 2016 CNEX Labs
   3  * Initial release: Javier Gonzalez <javier@cnexlabs.com>
   4  *                  Matias Bjorling <matias@cnexlabs.com>
   5  *
   6  * This program is free software; you can redistribute it and/or
   7  * modify it under the terms of the GNU General Public License version
   8  * 2 as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License for more details.
  14  *
  15  * pblk-core.c - pblk's core functionality
  16  *
  17  */
  18
  19 #include "pblk.h"
  20
  21 static void pblk_line_mark_bb(struct work_struct *work)
  22 {
  23         struct pblk_line_ws *line_ws = container_of(work, struct pblk_line_ws,
  24                                                                         ws);
  25         struct pblk *pblk = line_ws->pblk;
  26         struct nvm_tgt_dev *dev = pblk->dev;
  27         struct ppa_addr *ppa = line_ws->priv;
  28         int ret;
  29
  30         ret = nvm_set_tgt_bb_tbl(dev, ppa, 1, NVM_BLK_T_GRWN_BAD);
  31         if (ret) {
  32                 struct pblk_line *line;
  33                 int pos;
  34
  35                 line = &pblk->lines[pblk_ppa_to_line(*ppa)];
  36                 pos = pblk_ppa_to_pos(&dev->geo, *ppa);
  37
  38                 pblk_err(pblk, "failed to mark bb, line:%d, pos:%d\n",
  39                                 line->id, pos);
  40         }
  41
  42         kfree(ppa);
  43         mempool_free(line_ws, &pblk->gen_ws_pool);
  44 }
  45
  46 static void pblk_mark_bb(struct pblk *pblk, struct pblk_line *line,
  47                          struct ppa_addr ppa_addr)
  48 {
  49         struct nvm_tgt_dev *dev = pblk->dev;
  50         struct nvm_geo *geo = &dev->geo;
  51         struct ppa_addr *ppa;
  52         int pos = pblk_ppa_to_pos(geo, ppa_addr);
  53
  54         pblk_debug(pblk, "erase failed: line:%d, pos:%d\n", line->id, pos);
  55         atomic_long_inc(&pblk->erase_failed);
  56
  57         atomic_dec(&line->blk_in_line);
  58         if (test_and_set_bit(pos, line->blk_bitmap))
  59                 pblk_err(pblk, "attempted to erase bb: line:%d, pos:%d\n",
  60                                                         line->id, pos);
  61
  62         /* Not necessary to mark bad blocks on 2.0 spec. */
  63         if (geo->version == NVM_OCSSD_SPEC_20)
  64                 return;
  65
  66         ppa = kmalloc(sizeof(struct ppa_addr), GFP_ATOMIC);
  67         if (!ppa)
  68                 return;
  69
  70         *ppa = ppa_addr;
  71         pblk_gen_run_ws(pblk, NULL, ppa, pblk_line_mark_bb,
  72                                                 GFP_ATOMIC, pblk->bb_wq);
  73 }
  74
  75 static void __pblk_end_io_erase(struct pblk *pblk, struct nvm_rq *rqd)
  76 {
  77         struct nvm_tgt_dev *dev = pblk->dev;
  78         struct nvm_geo *geo = &dev->geo;
  79         struct nvm_chk_meta *chunk;
  80         struct pblk_line *line;
  81         int pos;
  82
  83         line = &pblk->lines[pblk_ppa_to_line(rqd->ppa_addr)];
  84         pos = pblk_ppa_to_pos(geo, rqd->ppa_addr);
  85         chunk = &line->chks[pos];
  86
  87         atomic_dec(&line->left_seblks);
  88
  89         if (rqd->error) {
  90                 chunk->state = NVM_CHK_ST_OFFLINE;
  91                 pblk_mark_bb(pblk, line, rqd->ppa_addr);
  92         } else {
  93                 chunk->state = NVM_CHK_ST_FREE;
  94         }
  95
  96         atomic_dec(&pblk->inflight_io);
  97 }
  98
  99 /* Erase completion assumes that only one block is erased at the time */
 100 static void pblk_end_io_erase(struct nvm_rq *rqd)
 101 {
 102         struct pblk *pblk = rqd->private;
 103
 104         __pblk_end_io_erase(pblk, rqd);
 105         mempool_free(rqd, &pblk->e_rq_pool);
 106 }
 107
 108 /*
 109  * Get information for all chunks from the device.
 110  *
 111  * The caller is responsible for freeing the returned structure
 112  */
 113 struct nvm_chk_meta *pblk_chunk_get_info(struct pblk *pblk)
 114 {
 115         struct nvm_tgt_dev *dev = pblk->dev;
 116         struct nvm_geo *geo = &dev->geo;
 117         struct nvm_chk_meta *meta;
 118         struct ppa_addr ppa;
 119         unsigned long len;
 120         int ret;
 121
 122         ppa.ppa = 0;
 123
 124         len = geo->all_chunks * sizeof(*meta);
 125         meta = kzalloc(len, GFP_KERNEL);
 126         if (!meta)
 127                 return ERR_PTR(-ENOMEM);
 128
 129         ret = nvm_get_chunk_meta(dev, meta, ppa, geo->all_chunks);
 130         if (ret) {
 131                 kfree(meta);
 132                 return ERR_PTR(-EIO);
 133         }
 134
 135         return meta;
 136 }
 137
 138 struct nvm_chk_meta *pblk_chunk_get_off(struct pblk *pblk,
 139                                               struct nvm_chk_meta *meta,
 140                                               struct ppa_addr ppa)
 141 {
 142         struct nvm_tgt_dev *dev = pblk->dev;
 143         struct nvm_geo *geo = &dev->geo;
 144         int ch_off = ppa.m.grp * geo->num_chk * geo->num_lun;
 145         int lun_off = ppa.m.pu * geo->num_chk;
 146         int chk_off = ppa.m.chk;
 147
 148         return meta + ch_off + lun_off + chk_off;
 149 }
 150
 151 void __pblk_map_invalidate(struct pblk *pblk, struct pblk_line *line,
 152                            u64 paddr)
 153 {
 154         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
 155         struct list_head *move_list = NULL;
 156
 157         /* Lines being reclaimed (GC'ed) cannot be invalidated. Before the L2P
 158          * table is modified with reclaimed sectors, a check is done to endure
 159          * that newer updates are not overwritten.
 160          */
 161         spin_lock(&line->lock);
 162         WARN_ON(line->state == PBLK_LINESTATE_FREE);
 163
 164         if (test_and_set_bit(paddr, line->invalid_bitmap)) {
 165                 WARN_ONCE(1, "pblk: double invalidate\n");
 166                 spin_unlock(&line->lock);
 167                 return;
 168         }
 169         le32_add_cpu(line->vsc, -1);
 170
 171         if (line->state == PBLK_LINESTATE_CLOSED)
 172                 move_list = pblk_line_gc_list(pblk, line);
 173         spin_unlock(&line->lock);
 174
 175         if (move_list) {
 176                 spin_lock(&l_mg->gc_lock);
 177                 spin_lock(&line->lock);
 178                 /* Prevent moving a line that has just been chosen for GC */
 179                 if (line->state == PBLK_LINESTATE_GC) {
 180                         spin_unlock(&line->lock);
 181                         spin_unlock(&l_mg->gc_lock);
 182                         return;
 183                 }
 184                 spin_unlock(&line->lock);
 185
 186                 list_move_tail(&line->list, move_list);
 187                 spin_unlock(&l_mg->gc_lock);
 188         }
 189 }
 190
 191 void pblk_map_invalidate(struct pblk *pblk, struct ppa_addr ppa)
 192 {
 193         struct pblk_line *line;
 194         u64 paddr;
 195         int line_id;
 196
 197 #ifdef CONFIG_NVM_PBLK_DEBUG
 198         /* Callers must ensure that the ppa points to a device address */
 199         BUG_ON(pblk_addr_in_cache(ppa));
 200         BUG_ON(pblk_ppa_empty(ppa));
 201 #endif
 202
 203         line_id = pblk_ppa_to_line(ppa);
 204         line = &pblk->lines[line_id];
 205         paddr = pblk_dev_ppa_to_line_addr(pblk, ppa);
 206
 207         __pblk_map_invalidate(pblk, line, paddr);
 208 }
 209
 210 static void pblk_invalidate_range(struct pblk *pblk, sector_t slba,
 211                                   unsigned int nr_secs)
 212 {
 213         sector_t lba;
 214
 215         spin_lock(&pblk->trans_lock);
 216         for (lba = slba; lba < slba + nr_secs; lba++) {
 217                 struct ppa_addr ppa;
 218
 219                 ppa = pblk_trans_map_get(pblk, lba);
 220
 221                 if (!pblk_addr_in_cache(ppa) && !pblk_ppa_empty(ppa))
 222                         pblk_map_invalidate(pblk, ppa);
 223
 224                 pblk_ppa_set_empty(&ppa);
 225                 pblk_trans_map_set(pblk, lba, ppa);
 226         }
 227         spin_unlock(&pblk->trans_lock);
 228 }
 229
 230 /* Caller must guarantee that the request is a valid type */
 231 struct nvm_rq *pblk_alloc_rqd(struct pblk *pblk, int type)
 232 {
 233         mempool_t *pool;
 234         struct nvm_rq *rqd;
 235         int rq_size;
 236
 237         switch (type) {
 238         case PBLK_WRITE:
 239         case PBLK_WRITE_INT:
 240                 pool = &pblk->w_rq_pool;
 241                 rq_size = pblk_w_rq_size;
 242                 break;
 243         case PBLK_READ:
 244                 pool = &pblk->r_rq_pool;
 245                 rq_size = pblk_g_rq_size;
 246                 break;
 247         default:
 248                 pool = &pblk->e_rq_pool;
 249                 rq_size = pblk_g_rq_size;
 250         }
 251
 252         rqd = mempool_alloc(pool, GFP_KERNEL);
 253         memset(rqd, 0, rq_size);
 254
 255         return rqd;
 256 }
 257
 258 /* Typically used on completion path. Cannot guarantee request consistency */
 259 void pblk_free_rqd(struct pblk *pblk, struct nvm_rq *rqd, int type)
 260 {
 261         struct nvm_tgt_dev *dev = pblk->dev;
 262         mempool_t *pool;
 263
 264         switch (type) {
 265         case PBLK_WRITE:
 266                 kfree(((struct pblk_c_ctx *)nvm_rq_to_pdu(rqd))->lun_bitmap);
 267                 /* fall through */
 268         case PBLK_WRITE_INT:
 269                 pool = &pblk->w_rq_pool;
 270                 break;
 271         case PBLK_READ:
 272                 pool = &pblk->r_rq_pool;
 273                 break;
 274         case PBLK_ERASE:
 275                 pool = &pblk->e_rq_pool;
 276                 break;
 277         default:
 278                 pblk_err(pblk, "trying to free unknown rqd type\n");
 279                 return;
 280         }
 281
 282         if (rqd->meta_list)
 283                 nvm_dev_dma_free(dev->parent, rqd->meta_list,
 284                                 rqd->dma_meta_list);
 285         mempool_free(rqd, pool);
 286 }
 287
 288 void pblk_bio_free_pages(struct pblk *pblk, struct bio *bio, int off,
 289                          int nr_pages)
 290 {
 291         struct bio_vec bv;
 292         int i;
 293
 294         WARN_ON(off + nr_pages != bio->bi_vcnt);
 295
 296         for (i = off; i < nr_pages + off; i++) {
 297                 bv = bio->bi_io_vec[i];
 298                 mempool_free(bv.bv_page, &pblk->page_bio_pool);
 299         }
 300 }
 301
 302 int pblk_bio_add_pages(struct pblk *pblk, struct bio *bio, gfp_t flags,
 303                        int nr_pages)
 304 {
 305         struct request_queue *q = pblk->dev->q;
 306         struct page *page;
 307         int i, ret;
 308
 309         for (i = 0; i < nr_pages; i++) {
 310                 page = mempool_alloc(&pblk->page_bio_pool, flags);
 311
 312                 ret = bio_add_pc_page(q, bio, page, PBLK_EXPOSED_PAGE_SIZE, 0);
 313                 if (ret != PBLK_EXPOSED_PAGE_SIZE) {
 314                         pblk_err(pblk, "could not add page to bio\n");
 315                         mempool_free(page, &pblk->page_bio_pool);
 316                         goto err;
 317                 }
 318         }
 319
 320         return 0;
 321 err:
 322         pblk_bio_free_pages(pblk, bio, (bio->bi_vcnt - i), i);
 323         return -1;
 324 }
 325
 326 void pblk_write_kick(struct pblk *pblk)
 327 {
 328         wake_up_process(pblk->writer_ts);
 329         mod_timer(&pblk->wtimer, jiffies + msecs_to_jiffies(1000));
 330 }
 331
 332 void pblk_write_timer_fn(struct timer_list *t)
 333 {
 334         struct pblk *pblk = from_timer(pblk, t, wtimer);
 335
 336         /* kick the write thread every tick to flush outstanding data */
 337         pblk_write_kick(pblk);
 338 }
 339
 340 void pblk_write_should_kick(struct pblk *pblk)
 341 {
 342         unsigned int secs_avail = pblk_rb_read_count(&pblk->rwb);
 343
 344         if (secs_avail >= pblk->min_write_pgs)
 345                 pblk_write_kick(pblk);
 346 }
 347
 348 static void pblk_wait_for_meta(struct pblk *pblk)
 349 {
 350         do {
 351                 if (!atomic_read(&pblk->inflight_io))
 352                         break;
 353
 354                 schedule();
 355         } while (1);
 356 }
 357
 358 static void pblk_flush_writer(struct pblk *pblk)
 359 {
 360         pblk_rb_flush(&pblk->rwb);
 361         do {
 362                 if (!pblk_rb_sync_count(&pblk->rwb))
 363                         break;
 364
 365                 pblk_write_kick(pblk);
 366                 schedule();
 367         } while (1);
 368 }
 369
 370 struct list_head *pblk_line_gc_list(struct pblk *pblk, struct pblk_line *line)
 371 {
 372         struct pblk_line_meta *lm = &pblk->lm;
 373         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
 374         struct list_head *move_list = NULL;
 375         int vsc = le32_to_cpu(*line->vsc);
 376
 377         lockdep_assert_held(&line->lock);
 378
 379         if (line->w_err_gc->has_write_err) {
 380                 if (line->gc_group != PBLK_LINEGC_WERR) {
 381                         line->gc_group = PBLK_LINEGC_WERR;
 382                         move_list = &l_mg->gc_werr_list;
 383                         pblk_rl_werr_line_in(&pblk->rl);
 384                 }
 385         } else if (!vsc) {
 386                 if (line->gc_group != PBLK_LINEGC_FULL) {
 387                         line->gc_group = PBLK_LINEGC_FULL;
 388                         move_list = &l_mg->gc_full_list;
 389                 }
 390         } else if (vsc < lm->high_thrs) {
 391                 if (line->gc_group != PBLK_LINEGC_HIGH) {
 392                         line->gc_group = PBLK_LINEGC_HIGH;
 393                         move_list = &l_mg->gc_high_list;
 394                 }
 395         } else if (vsc < lm->mid_thrs) {
 396                 if (line->gc_group != PBLK_LINEGC_MID) {
 397                         line->gc_group = PBLK_LINEGC_MID;
 398                         move_list = &l_mg->gc_mid_list;
 399                 }
 400         } else if (vsc < line->sec_in_line) {
 401                 if (line->gc_group != PBLK_LINEGC_LOW) {
 402                         line->gc_group = PBLK_LINEGC_LOW;
 403                         move_list = &l_mg->gc_low_list;
 404                 }
 405         } else if (vsc == line->sec_in_line) {
 406                 if (line->gc_group != PBLK_LINEGC_EMPTY) {
 407                         line->gc_group = PBLK_LINEGC_EMPTY;
 408                         move_list = &l_mg->gc_empty_list;
 409                 }
 410         } else {
 411                 line->state = PBLK_LINESTATE_CORRUPT;
 412                 line->gc_group = PBLK_LINEGC_NONE;
 413                 move_list =  &l_mg->corrupt_list;
 414                 pblk_err(pblk, "corrupted vsc for line %d, vsc:%d (%d/%d/%d)\n",
 415                                                 line->id, vsc,
 416                                                 line->sec_in_line,
 417                                                 lm->high_thrs, lm->mid_thrs);
 418         }
 419
 420         return move_list;
 421 }
 422
 423 void pblk_discard(struct pblk *pblk, struct bio *bio)
 424 {
 425         sector_t slba = pblk_get_lba(bio);
 426         sector_t nr_secs = pblk_get_secs(bio);
 427
 428         pblk_invalidate_range(pblk, slba, nr_secs);
 429 }
 430
 431 void pblk_log_write_err(struct pblk *pblk, struct nvm_rq *rqd)
 432 {
 433         atomic_long_inc(&pblk->write_failed);
 434 #ifdef CONFIG_NVM_PBLK_DEBUG
 435         pblk_print_failed_rqd(pblk, rqd, rqd->error);
 436 #endif
 437 }
 438
 439 void pblk_log_read_err(struct pblk *pblk, struct nvm_rq *rqd)
 440 {
 441         /* Empty page read is not necessarily an error (e.g., L2P recovery) */
 442         if (rqd->error == NVM_RSP_ERR_EMPTYPAGE) {
 443                 atomic_long_inc(&pblk->read_empty);
 444                 return;
 445         }
 446
 447         switch (rqd->error) {
 448         case NVM_RSP_WARN_HIGHECC:
 449                 atomic_long_inc(&pblk->read_high_ecc);
 450                 break;
 451         case NVM_RSP_ERR_FAILECC:
 452         case NVM_RSP_ERR_FAILCRC:
 453                 atomic_long_inc(&pblk->read_failed);
 454                 break;
 455         default:
 456                 pblk_err(pblk, "unknown read error:%d\n", rqd->error);
 457         }
 458 #ifdef CONFIG_NVM_PBLK_DEBUG
 459         pblk_print_failed_rqd(pblk, rqd, rqd->error);
 460 #endif
 461 }
 462
 463 void pblk_set_sec_per_write(struct pblk *pblk, int sec_per_write)
 464 {
 465         pblk->sec_per_write = sec_per_write;
 466 }
 467
 468 int pblk_submit_io(struct pblk *pblk, struct nvm_rq *rqd)
 469 {
 470         struct nvm_tgt_dev *dev = pblk->dev;
 471
 472         atomic_inc(&pblk->inflight_io);
 473
 474 #ifdef CONFIG_NVM_PBLK_DEBUG
 475         if (pblk_check_io(pblk, rqd))
 476                 return NVM_IO_ERR;
 477 #endif
 478
 479         return nvm_submit_io(dev, rqd);
 480 }
 481
 482 int pblk_submit_io_sync(struct pblk *pblk, struct nvm_rq *rqd)
 483 {
 484         struct nvm_tgt_dev *dev = pblk->dev;
 485
 486         atomic_inc(&pblk->inflight_io);
 487
 488 #ifdef CONFIG_NVM_PBLK_DEBUG
 489         if (pblk_check_io(pblk, rqd))
 490                 return NVM_IO_ERR;
 491 #endif
 492
 493         return nvm_submit_io_sync(dev, rqd);
 494 }
 495
 496 static void pblk_bio_map_addr_endio(struct bio *bio)
 497 {
 498         bio_put(bio);
 499 }
 500
 501 struct bio *pblk_bio_map_addr(struct pblk *pblk, void *data,
 502                               unsigned int nr_secs, unsigned int len,
 503                               int alloc_type, gfp_t gfp_mask)
 504 {
 505         struct nvm_tgt_dev *dev = pblk->dev;
 506         void *kaddr = data;
 507         struct page *page;
 508         struct bio *bio;
 509         int i, ret;
 510
 511         if (alloc_type == PBLK_KMALLOC_META)
 512                 return bio_map_kern(dev->q, kaddr, len, gfp_mask);
 513
 514         bio = bio_kmalloc(gfp_mask, nr_secs);
 515         if (!bio)
 516                 return ERR_PTR(-ENOMEM);
 517
 518         for (i = 0; i < nr_secs; i++) {
 519                 page = vmalloc_to_page(kaddr);
 520                 if (!page) {
 521                         pblk_err(pblk, "could not map vmalloc bio\n");
 522                         bio_put(bio);
 523                         bio = ERR_PTR(-ENOMEM);
 524                         goto out;
 525                 }
 526
 527                 ret = bio_add_pc_page(dev->q, bio, page, PAGE_SIZE, 0);
 528                 if (ret != PAGE_SIZE) {
 529                         pblk_err(pblk, "could not add page to bio\n");
 530                         bio_put(bio);
 531                         bio = ERR_PTR(-ENOMEM);
 532                         goto out;
 533                 }
 534
 535                 kaddr += PAGE_SIZE;
 536         }
 537
 538         bio->bi_end_io = pblk_bio_map_addr_endio;
 539 out:
 540         return bio;
 541 }
 542
 543 int pblk_calc_secs(struct pblk *pblk, unsigned long secs_avail,
 544                    unsigned long secs_to_flush)
 545 {
 546         int max = pblk->sec_per_write;
 547         int min = pblk->min_write_pgs;
 548         int secs_to_sync = 0;
 549
 550         if (secs_avail >= max)
 551                 secs_to_sync = max;
 552         else if (secs_avail >= min)
 553                 secs_to_sync = min * (secs_avail / min);
 554         else if (secs_to_flush)
 555                 secs_to_sync = min;
 556
 557         return secs_to_sync;
 558 }
 559
 560 void pblk_dealloc_page(struct pblk *pblk, struct pblk_line *line, int nr_secs)
 561 {
 562         u64 addr;
 563         int i;
 564
 565         spin_lock(&line->lock);
 566         addr = find_next_zero_bit(line->map_bitmap,
 567                                         pblk->lm.sec_per_line, line->cur_sec);
 568         line->cur_sec = addr - nr_secs;
 569
 570         for (i = 0; i < nr_secs; i++, line->cur_sec--)
 571                 WARN_ON(!test_and_clear_bit(line->cur_sec, line->map_bitmap));
 572         spin_unlock(&line->lock);
 573 }
 574
 575 u64 __pblk_alloc_page(struct pblk *pblk, struct pblk_line *line, int nr_secs)
 576 {
 577         u64 addr;
 578         int i;
 579
 580         lockdep_assert_held(&line->lock);
 581
 582         /* logic error: ppa out-of-bounds. Prevent generating bad address */
 583         if (line->cur_sec + nr_secs > pblk->lm.sec_per_line) {
 584                 WARN(1, "pblk: page allocation out of bounds\n");
 585                 nr_secs = pblk->lm.sec_per_line - line->cur_sec;
 586         }
 587
 588         line->cur_sec = addr = find_next_zero_bit(line->map_bitmap,
 589                                         pblk->lm.sec_per_line, line->cur_sec);
 590         for (i = 0; i < nr_secs; i++, line->cur_sec++)
 591                 WARN_ON(test_and_set_bit(line->cur_sec, line->map_bitmap));
 592
 593         return addr;
 594 }
 595
 596 u64 pblk_alloc_page(struct pblk *pblk, struct pblk_line *line, int nr_secs)
 597 {
 598         u64 addr;
 599
 600         /* Lock needed in case a write fails and a recovery needs to remap
 601          * failed write buffer entries
 602          */
 603         spin_lock(&line->lock);
 604         addr = __pblk_alloc_page(pblk, line, nr_secs);
 605         line->left_msecs -= nr_secs;
 606         WARN(line->left_msecs < 0, "pblk: page allocation out of bounds\n");
 607         spin_unlock(&line->lock);
 608
 609         return addr;
 610 }
 611
 612 u64 pblk_lookup_page(struct pblk *pblk, struct pblk_line *line)
 613 {
 614         u64 paddr;
 615
 616         spin_lock(&line->lock);
 617         paddr = find_next_zero_bit(line->map_bitmap,
 618                                         pblk->lm.sec_per_line, line->cur_sec);
 619         spin_unlock(&line->lock);
 620
 621         return paddr;
 622 }
 623
 624 /*
 625  * Submit emeta to one LUN in the raid line at the time to avoid a deadlock when
 626  * taking the per LUN semaphore.
 627  */
 628 static int pblk_line_submit_emeta_io(struct pblk *pblk, struct pblk_line *line,
 629                                      void *emeta_buf, u64 paddr, int dir)
 630 {
 631         struct nvm_tgt_dev *dev = pblk->dev;
 632         struct nvm_geo *geo = &dev->geo;
 633         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
 634         struct pblk_line_meta *lm = &pblk->lm;
 635         void *ppa_list, *meta_list;
 636         struct bio *bio;
 637         struct nvm_rq rqd;
 638         dma_addr_t dma_ppa_list, dma_meta_list;
 639         int min = pblk->min_write_pgs;
 640         int left_ppas = lm->emeta_sec[0];
 641         int id = line->id;
 642         int rq_ppas, rq_len;
 643         int cmd_op, bio_op;
 644         int i, j;
 645         int ret;
 646
 647         if (dir == PBLK_WRITE) {
 648                 bio_op = REQ_OP_WRITE;
 649                 cmd_op = NVM_OP_PWRITE;
 650         } else if (dir == PBLK_READ) {
 651                 bio_op = REQ_OP_READ;
 652                 cmd_op = NVM_OP_PREAD;
 653         } else
 654                 return -EINVAL;
 655
 656         meta_list = nvm_dev_dma_alloc(dev->parent, GFP_KERNEL,
 657                                                         &dma_meta_list);
 658         if (!meta_list)
 659                 return -ENOMEM;
 660
 661         ppa_list = meta_list + pblk_dma_meta_size;
 662         dma_ppa_list = dma_meta_list + pblk_dma_meta_size;
 663
 664 next_rq:
 665         memset(&rqd, 0, sizeof(struct nvm_rq));
 666
 667         rq_ppas = pblk_calc_secs(pblk, left_ppas, 0);
 668         rq_len = rq_ppas * geo->csecs;
 669
 670         bio = pblk_bio_map_addr(pblk, emeta_buf, rq_ppas, rq_len,
 671                                         l_mg->emeta_alloc_type, GFP_KERNEL);
 672         if (IS_ERR(bio)) {
 673                 ret = PTR_ERR(bio);
 674                 goto free_rqd_dma;
 675         }
 676
 677         bio->bi_iter.bi_sector = 0; /* internal bio */
 678         bio_set_op_attrs(bio, bio_op, 0);
 679
 680         rqd.bio = bio;
 681         rqd.meta_list = meta_list;
 682         rqd.ppa_list = ppa_list;
 683         rqd.dma_meta_list = dma_meta_list;
 684         rqd.dma_ppa_list = dma_ppa_list;
 685         rqd.opcode = cmd_op;
 686         rqd.nr_ppas = rq_ppas;
 687
 688         if (dir == PBLK_WRITE) {
 689                 struct pblk_sec_meta *meta_list = rqd.meta_list;
 690
 691                 rqd.flags = pblk_set_progr_mode(pblk, PBLK_WRITE);
 692                 for (i = 0; i < rqd.nr_ppas; ) {
 693                         spin_lock(&line->lock);
 694                         paddr = __pblk_alloc_page(pblk, line, min);
 695                         spin_unlock(&line->lock);
 696                         for (j = 0; j < min; j++, i++, paddr++) {
 697                                 meta_list[i].lba = cpu_to_le64(ADDR_EMPTY);
 698                                 rqd.ppa_list[i] =
 699                                         addr_to_gen_ppa(pblk, paddr, id);
 700                         }
 701                 }
 702         } else {
 703                 for (i = 0; i < rqd.nr_ppas; ) {
 704                         struct ppa_addr ppa = addr_to_gen_ppa(pblk, paddr, id);
 705                         int pos = pblk_ppa_to_pos(geo, ppa);
 706                         int read_type = PBLK_READ_RANDOM;
 707
 708                         if (pblk_io_aligned(pblk, rq_ppas))
 709                                 read_type = PBLK_READ_SEQUENTIAL;
 710                         rqd.flags = pblk_set_read_mode(pblk, read_type);
 711
 712                         while (test_bit(pos, line->blk_bitmap)) {
 713                                 paddr += min;
 714                                 if (pblk_boundary_paddr_checks(pblk, paddr)) {
 715                                         pblk_err(pblk, "corrupt emeta line:%d\n",
 716                                                                 line->id);
 717                                         bio_put(bio);
 718                                         ret = -EINTR;
 719                                         goto free_rqd_dma;
 720                                 }
 721
 722                                 ppa = addr_to_gen_ppa(pblk, paddr, id);
 723                                 pos = pblk_ppa_to_pos(geo, ppa);
 724                         }
 725
 726                         if (pblk_boundary_paddr_checks(pblk, paddr + min)) {
 727                                 pblk_err(pblk, "corrupt emeta line:%d\n",
 728                                                                 line->id);
 729                                 bio_put(bio);
 730                                 ret = -EINTR;
 731                                 goto free_rqd_dma;
 732                         }
 733
 734                         for (j = 0; j < min; j++, i++, paddr++)
 735                                 rqd.ppa_list[i] =
 736                                         addr_to_gen_ppa(pblk, paddr, line->id);
 737                 }
 738         }
 739
 740         ret = pblk_submit_io_sync(pblk, &rqd);
 741         if (ret) {
 742                 pblk_err(pblk, "emeta I/O submission failed: %d\n", ret);
 743                 bio_put(bio);
 744                 goto free_rqd_dma;
 745         }
 746
 747         atomic_dec(&pblk->inflight_io);
 748
 749         if (rqd.error) {
 750                 if (dir == PBLK_WRITE)
 751                         pblk_log_write_err(pblk, &rqd);
 752                 else
 753                         pblk_log_read_err(pblk, &rqd);
 754         }
 755
 756         emeta_buf += rq_len;
 757         left_ppas -= rq_ppas;
 758         if (left_ppas)
 759                 goto next_rq;
 760 free_rqd_dma:
 761         nvm_dev_dma_free(dev->parent, rqd.meta_list, rqd.dma_meta_list);
 762         return ret;
 763 }
 764
 765 u64 pblk_line_smeta_start(struct pblk *pblk, struct pblk_line *line)
 766 {
 767         struct nvm_tgt_dev *dev = pblk->dev;
 768         struct nvm_geo *geo = &dev->geo;
 769         struct pblk_line_meta *lm = &pblk->lm;
 770         int bit;
 771
 772         /* This usually only happens on bad lines */
 773         bit = find_first_zero_bit(line->blk_bitmap, lm->blk_per_line);
 774         if (bit >= lm->blk_per_line)
 775                 return -1;
 776
 777         return bit * geo->ws_opt;
 778 }
 779
 780 static int pblk_line_submit_smeta_io(struct pblk *pblk, struct pblk_line *line,
 781                                      u64 paddr, int dir)
 782 {
 783         struct nvm_tgt_dev *dev = pblk->dev;
 784         struct pblk_line_meta *lm = &pblk->lm;
 785         struct bio *bio;
 786         struct nvm_rq rqd;
 787         __le64 *lba_list = NULL;
 788         int i, ret;
 789         int cmd_op, bio_op;
 790         int flags;
 791
 792         if (dir == PBLK_WRITE) {
 793                 bio_op = REQ_OP_WRITE;
 794                 cmd_op = NVM_OP_PWRITE;
 795                 flags = pblk_set_progr_mode(pblk, PBLK_WRITE);
 796                 lba_list = emeta_to_lbas(pblk, line->emeta->buf);
 797         } else if (dir == PBLK_READ_RECOV || dir == PBLK_READ) {
 798                 bio_op = REQ_OP_READ;
 799                 cmd_op = NVM_OP_PREAD;
 800                 flags = pblk_set_read_mode(pblk, PBLK_READ_SEQUENTIAL);
 801         } else
 802                 return -EINVAL;
 803
 804         memset(&rqd, 0, sizeof(struct nvm_rq));
 805
 806         rqd.meta_list = nvm_dev_dma_alloc(dev->parent, GFP_KERNEL,
 807                                                         &rqd.dma_meta_list);
 808         if (!rqd.meta_list)
 809                 return -ENOMEM;
 810
 811         rqd.ppa_list = rqd.meta_list + pblk_dma_meta_size;
 812         rqd.dma_ppa_list = rqd.dma_meta_list + pblk_dma_meta_size;
 813
 814         bio = bio_map_kern(dev->q, line->smeta, lm->smeta_len, GFP_KERNEL);
 815         if (IS_ERR(bio)) {
 816                 ret = PTR_ERR(bio);
 817                 goto free_ppa_list;
 818         }
 819
 820         bio->bi_iter.bi_sector = 0; /* internal bio */
 821         bio_set_op_attrs(bio, bio_op, 0);
 822
 823         rqd.bio = bio;
 824         rqd.opcode = cmd_op;
 825         rqd.flags = flags;
 826         rqd.nr_ppas = lm->smeta_sec;
 827
 828         for (i = 0; i < lm->smeta_sec; i++, paddr++) {
 829                 struct pblk_sec_meta *meta_list = rqd.meta_list;
 830
 831                 rqd.ppa_list[i] = addr_to_gen_ppa(pblk, paddr, line->id);
 832
 833                 if (dir == PBLK_WRITE) {
 834                         __le64 addr_empty = cpu_to_le64(ADDR_EMPTY);
 835
 836                         meta_list[i].lba = lba_list[paddr] = addr_empty;
 837                 }
 838         }
 839
 840         /*
 841          * This I/O is sent by the write thread when a line is replace. Since
 842          * the write thread is the only one sending write and erase commands,
 843          * there is no need to take the LUN semaphore.
 844          */
 845         ret = pblk_submit_io_sync(pblk, &rqd);
 846         if (ret) {
 847                 pblk_err(pblk, "smeta I/O submission failed: %d\n", ret);
 848                 bio_put(bio);
 849                 goto free_ppa_list;
 850         }
 851
 852         atomic_dec(&pblk->inflight_io);
 853
 854         if (rqd.error) {
 855                 if (dir == PBLK_WRITE) {
 856                         pblk_log_write_err(pblk, &rqd);
 857                         ret = 1;
 858                 } else if (dir == PBLK_READ)
 859                         pblk_log_read_err(pblk, &rqd);
 860         }
 861
 862 free_ppa_list:
 863         nvm_dev_dma_free(dev->parent, rqd.meta_list, rqd.dma_meta_list);
 864
 865         return ret;
 866 }
 867
 868 int pblk_line_read_smeta(struct pblk *pblk, struct pblk_line *line)
 869 {
 870         u64 bpaddr = pblk_line_smeta_start(pblk, line);
 871
 872         return pblk_line_submit_smeta_io(pblk, line, bpaddr, PBLK_READ_RECOV);
 873 }
 874
 875 int pblk_line_read_emeta(struct pblk *pblk, struct pblk_line *line,
 876                          void *emeta_buf)
 877 {
 878         return pblk_line_submit_emeta_io(pblk, line, emeta_buf,
 879                                                 line->emeta_ssec, PBLK_READ);
 880 }
 881
 882 static void pblk_setup_e_rq(struct pblk *pblk, struct nvm_rq *rqd,
 883                             struct ppa_addr ppa)
 884 {
 885         rqd->opcode = NVM_OP_ERASE;
 886         rqd->ppa_addr = ppa;
 887         rqd->nr_ppas = 1;
 888         rqd->flags = pblk_set_progr_mode(pblk, PBLK_ERASE);
 889         rqd->bio = NULL;
 890 }
 891
 892 static int pblk_blk_erase_sync(struct pblk *pblk, struct ppa_addr ppa)
 893 {
 894         struct nvm_rq rqd;
 895         int ret = 0;
 896
 897         memset(&rqd, 0, sizeof(struct nvm_rq));
 898
 899         pblk_setup_e_rq(pblk, &rqd, ppa);
 900
 901         /* The write thread schedules erases so that it minimizes disturbances
 902          * with writes. Thus, there is no need to take the LUN semaphore.
 903          */
 904         ret = pblk_submit_io_sync(pblk, &rqd);
 905         if (ret) {
 906                 struct nvm_tgt_dev *dev = pblk->dev;
 907                 struct nvm_geo *geo = &dev->geo;
 908
 909                 pblk_err(pblk, "could not sync erase line:%d,blk:%d\n",
 910                                         pblk_ppa_to_line(ppa),
 911                                         pblk_ppa_to_pos(geo, ppa));
 912
 913                 rqd.error = ret;
 914                 goto out;
 915         }
 916
 917 out:
 918         rqd.private = pblk;
 919         __pblk_end_io_erase(pblk, &rqd);
 920
 921         return ret;
 922 }
 923
 924 int pblk_line_erase(struct pblk *pblk, struct pblk_line *line)
 925 {
 926         struct pblk_line_meta *lm = &pblk->lm;
 927         struct ppa_addr ppa;
 928         int ret, bit = -1;
 929
 930         /* Erase only good blocks, one at a time */
 931         do {
 932                 spin_lock(&line->lock);
 933                 bit = find_next_zero_bit(line->erase_bitmap, lm->blk_per_line,
 934                                                                 bit + 1);
 935                 if (bit >= lm->blk_per_line) {
 936                         spin_unlock(&line->lock);
 937                         break;
 938                 }
 939
 940                 ppa = pblk->luns[bit].bppa; /* set ch and lun */
 941                 ppa.a.blk = line->id;
 942
 943                 atomic_dec(&line->left_eblks);
 944                 WARN_ON(test_and_set_bit(bit, line->erase_bitmap));
 945                 spin_unlock(&line->lock);
 946
 947                 ret = pblk_blk_erase_sync(pblk, ppa);
 948                 if (ret) {
 949                         pblk_err(pblk, "failed to erase line %d\n", line->id);
 950                         return ret;
 951                 }
 952         } while (1);
 953
 954         return 0;
 955 }
 956
 957 static void pblk_line_setup_metadata(struct pblk_line *line,
 958                                      struct pblk_line_mgmt *l_mg,
 959                                      struct pblk_line_meta *lm)
 960 {
 961         int meta_line;
 962
 963         lockdep_assert_held(&l_mg->free_lock);
 964
 965 retry_meta:
 966         meta_line = find_first_zero_bit(&l_mg->meta_bitmap, PBLK_DATA_LINES);
 967         if (meta_line == PBLK_DATA_LINES) {
 968                 spin_unlock(&l_mg->free_lock);
 969                 io_schedule();
 970                 spin_lock(&l_mg->free_lock);
 971                 goto retry_meta;
 972         }
 973
 974         set_bit(meta_line, &l_mg->meta_bitmap);
 975         line->meta_line = meta_line;
 976
 977         line->smeta = l_mg->sline_meta[meta_line];
 978         line->emeta = l_mg->eline_meta[meta_line];
 979
 980         memset(line->smeta, 0, lm->smeta_len);
 981         memset(line->emeta->buf, 0, lm->emeta_len[0]);
 982
 983         line->emeta->mem = 0;
 984         atomic_set(&line->emeta->sync, 0);
 985 }
 986
 987 /* For now lines are always assumed full lines. Thus, smeta former and current
 988  * lun bitmaps are omitted.
 989  */
 990 static int pblk_line_init_metadata(struct pblk *pblk, struct pblk_line *line,
 991                                   struct pblk_line *cur)
 992 {
 993         struct nvm_tgt_dev *dev = pblk->dev;
 994         struct nvm_geo *geo = &dev->geo;
 995         struct pblk_line_meta *lm = &pblk->lm;
 996         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
 997         struct pblk_emeta *emeta = line->emeta;
 998         struct line_emeta *emeta_buf = emeta->buf;
 999         struct line_smeta *smeta_buf = (struct line_smeta *)line->smeta;
1000         int nr_blk_line;
1001
1002         /* After erasing the line, new bad blocks might appear and we risk
1003          * having an invalid line
1004          */
1005         nr_blk_line = lm->blk_per_line -
1006                         bitmap_weight(line->blk_bitmap, lm->blk_per_line);
1007         if (nr_blk_line < lm->min_blk_line) {
1008                 spin_lock(&l_mg->free_lock);
1009                 spin_lock(&line->lock);
1010                 line->state = PBLK_LINESTATE_BAD;
1011                 spin_unlock(&line->lock);
1012
1013                 list_add_tail(&line->list, &l_mg->bad_list);
1014                 spin_unlock(&l_mg->free_lock);
1015
1016                 pblk_debug(pblk, "line %d is bad\n", line->id);
1017
1018                 return 0;
1019         }
1020
1021         /* Run-time metadata */
1022         line->lun_bitmap = ((void *)(smeta_buf)) + sizeof(struct line_smeta);
1023
1024         /* Mark LUNs allocated in this line (all for now) */
1025         bitmap_set(line->lun_bitmap, 0, lm->lun_bitmap_len);
1026
1027         smeta_buf->header.identifier = cpu_to_le32(PBLK_MAGIC);
1028         memcpy(smeta_buf->header.uuid, pblk->instance_uuid, 16);
1029         smeta_buf->header.id = cpu_to_le32(line->id);
1030         smeta_buf->header.type = cpu_to_le16(line->type);
1031         smeta_buf->header.version_major = SMETA_VERSION_MAJOR;
1032         smeta_buf->header.version_minor = SMETA_VERSION_MINOR;
1033
1034         /* Start metadata */
1035         smeta_buf->seq_nr = cpu_to_le64(line->seq_nr);
1036         smeta_buf->window_wr_lun = cpu_to_le32(geo->all_luns);
1037
1038         /* Fill metadata among lines */
1039         if (cur) {
1040                 memcpy(line->lun_bitmap, cur->lun_bitmap, lm->lun_bitmap_len);
1041                 smeta_buf->prev_id = cpu_to_le32(cur->id);
1042                 cur->emeta->buf->next_id = cpu_to_le32(line->id);
1043         } else {
1044                 smeta_buf->prev_id = cpu_to_le32(PBLK_LINE_EMPTY);
1045         }
1046
1047         /* All smeta must be set at this point */
1048         smeta_buf->header.crc = cpu_to_le32(
1049                         pblk_calc_meta_header_crc(pblk, &smeta_buf->header));
1050         smeta_buf->crc = cpu_to_le32(pblk_calc_smeta_crc(pblk, smeta_buf));
1051
1052         /* End metadata */
1053         memcpy(&emeta_buf->header, &smeta_buf->header,
1054                                                 sizeof(struct line_header));
1055
1056         emeta_buf->header.version_major = EMETA_VERSION_MAJOR;
1057         emeta_buf->header.version_minor = EMETA_VERSION_MINOR;
1058         emeta_buf->header.crc = cpu_to_le32(
1059                         pblk_calc_meta_header_crc(pblk, &emeta_buf->header));
1060
1061         emeta_buf->seq_nr = cpu_to_le64(line->seq_nr);
1062         emeta_buf->nr_lbas = cpu_to_le64(line->sec_in_line);
1063         emeta_buf->nr_valid_lbas = cpu_to_le64(0);
1064         emeta_buf->next_id = cpu_to_le32(PBLK_LINE_EMPTY);
1065         emeta_buf->crc = cpu_to_le32(0);
1066         emeta_buf->prev_id = smeta_buf->prev_id;
1067
1068         return 1;
1069 }
1070
1071 static int pblk_line_alloc_bitmaps(struct pblk *pblk, struct pblk_line *line)
1072 {
1073         struct pblk_line_meta *lm = &pblk->lm;
1074
1075         line->map_bitmap = kzalloc(lm->sec_bitmap_len, GFP_KERNEL);
1076         if (!line->map_bitmap)
1077                 return -ENOMEM;
1078
1079         /* will be initialized using bb info from map_bitmap */
1080         line->invalid_bitmap = kmalloc(lm->sec_bitmap_len, GFP_KERNEL);
1081         if (!line->invalid_bitmap) {
1082                 kfree(line->map_bitmap);
1083                 line->map_bitmap = NULL;
1084                 return -ENOMEM;
1085         }
1086
1087         return 0;
1088 }
1089
1090 /* For now lines are always assumed full lines. Thus, smeta former and current
1091  * lun bitmaps are omitted.
1092  */
1093 static int pblk_line_init_bb(struct pblk *pblk, struct pblk_line *line,
1094                              int init)
1095 {
1096         struct nvm_tgt_dev *dev = pblk->dev;
1097         struct nvm_geo *geo = &dev->geo;
1098         struct pblk_line_meta *lm = &pblk->lm;
1099         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1100         u64 off;
1101         int bit = -1;
1102         int emeta_secs;
1103
1104         line->sec_in_line = lm->sec_per_line;
1105
1106         /* Capture bad block information on line mapping bitmaps */
1107         while ((bit = find_next_bit(line->blk_bitmap, lm->blk_per_line,
1108                                         bit + 1)) < lm->blk_per_line) {
1109                 off = bit * geo->ws_opt;
1110                 bitmap_shift_left(l_mg->bb_aux, l_mg->bb_template, off,
1111                                                         lm->sec_per_line);
1112                 bitmap_or(line->map_bitmap, line->map_bitmap, l_mg->bb_aux,
1113                                                         lm->sec_per_line);
1114                 line->sec_in_line -= geo->clba;
1115         }
1116
1117         /* Mark smeta metadata sectors as bad sectors */
1118         bit = find_first_zero_bit(line->blk_bitmap, lm->blk_per_line);
1119         off = bit * geo->ws_opt;
1120         bitmap_set(line->map_bitmap, off, lm->smeta_sec);
1121         line->sec_in_line -= lm->smeta_sec;
1122         line->smeta_ssec = off;
1123         line->cur_sec = off + lm->smeta_sec;
1124
1125         if (init && pblk_line_submit_smeta_io(pblk, line, off, PBLK_WRITE)) {
1126                 pblk_debug(pblk, "line smeta I/O failed. Retry\n");
1127                 return 0;
1128         }
1129
1130         bitmap_copy(line->invalid_bitmap, line->map_bitmap, lm->sec_per_line);
1131
1132         /* Mark emeta metadata sectors as bad sectors. We need to consider bad
1133          * blocks to make sure that there are enough sectors to store emeta
1134          */
1135         emeta_secs = lm->emeta_sec[0];
1136         off = lm->sec_per_line;
1137         while (emeta_secs) {
1138                 off -= geo->ws_opt;
1139                 if (!test_bit(off, line->invalid_bitmap)) {
1140                         bitmap_set(line->invalid_bitmap, off, geo->ws_opt);
1141                         emeta_secs -= geo->ws_opt;
1142                 }
1143         }
1144
1145         line->emeta_ssec = off;
1146         line->sec_in_line -= lm->emeta_sec[0];
1147         line->nr_valid_lbas = 0;
1148         line->left_msecs = line->sec_in_line;
1149         *line->vsc = cpu_to_le32(line->sec_in_line);
1150
1151         if (lm->sec_per_line - line->sec_in_line !=
1152                 bitmap_weight(line->invalid_bitmap, lm->sec_per_line)) {
1153                 spin_lock(&line->lock);
1154                 line->state = PBLK_LINESTATE_BAD;
1155                 spin_unlock(&line->lock);
1156
1157                 list_add_tail(&line->list, &l_mg->bad_list);
1158                 pblk_err(pblk, "unexpected line %d is bad\n", line->id);
1159
1160                 return 0;
1161         }
1162
1163         return 1;
1164 }
1165
1166 static int pblk_prepare_new_line(struct pblk *pblk, struct pblk_line *line)
1167 {
1168         struct pblk_line_meta *lm = &pblk->lm;
1169         struct nvm_tgt_dev *dev = pblk->dev;
1170         struct nvm_geo *geo = &dev->geo;
1171         int blk_to_erase = atomic_read(&line->blk_in_line);
1172         int i;
1173
1174         for (i = 0; i < lm->blk_per_line; i++) {
1175                 struct pblk_lun *rlun = &pblk->luns[i];
1176                 int pos = pblk_ppa_to_pos(geo, rlun->bppa);
1177                 int state = line->chks[pos].state;
1178
1179                 /* Free chunks should not be erased */
1180                 if (state & NVM_CHK_ST_FREE) {
1181                         set_bit(pblk_ppa_to_pos(geo, rlun->bppa),
1182                                                         line->erase_bitmap);
1183                         blk_to_erase--;
1184                 }
1185         }
1186
1187         return blk_to_erase;
1188 }
1189
1190 static int pblk_line_prepare(struct pblk *pblk, struct pblk_line *line)
1191 {
1192         struct pblk_line_meta *lm = &pblk->lm;
1193         int blk_in_line = atomic_read(&line->blk_in_line);
1194         int blk_to_erase;
1195
1196         /* Bad blocks do not need to be erased */
1197         bitmap_copy(line->erase_bitmap, line->blk_bitmap, lm->blk_per_line);
1198
1199         spin_lock(&line->lock);
1200
1201         /* If we have not written to this line, we need to mark up free chunks
1202          * as already erased
1203          */
1204         if (line->state == PBLK_LINESTATE_NEW) {
1205                 blk_to_erase = pblk_prepare_new_line(pblk, line);
1206                 line->state = PBLK_LINESTATE_FREE;
1207         } else {
1208                 blk_to_erase = blk_in_line;
1209         }
1210
1211         if (blk_in_line < lm->min_blk_line) {
1212                 spin_unlock(&line->lock);
1213                 return -EAGAIN;
1214         }
1215
1216         if (line->state != PBLK_LINESTATE_FREE) {
1217                 WARN(1, "pblk: corrupted line %d, state %d\n",
1218                                                         line->id, line->state);
1219                 spin_unlock(&line->lock);
1220                 return -EINTR;
1221         }
1222
1223         line->state = PBLK_LINESTATE_OPEN;
1224
1225         atomic_set(&line->left_eblks, blk_to_erase);
1226         atomic_set(&line->left_seblks, blk_to_erase);
1227
1228         line->meta_distance = lm->meta_distance;
1229         spin_unlock(&line->lock);
1230
1231         kref_init(&line->ref);
1232
1233         return 0;
1234 }
1235
1236 int pblk_line_recov_alloc(struct pblk *pblk, struct pblk_line *line)
1237 {
1238         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1239         int ret;
1240
1241         spin_lock(&l_mg->free_lock);
1242         l_mg->data_line = line;
1243         list_del(&line->list);
1244
1245         ret = pblk_line_prepare(pblk, line);
1246         if (ret) {
1247                 list_add(&line->list, &l_mg->free_list);
1248                 spin_unlock(&l_mg->free_lock);
1249                 return ret;
1250         }
1251         spin_unlock(&l_mg->free_lock);
1252
1253         ret = pblk_line_alloc_bitmaps(pblk, line);
1254         if (ret)
1255                 return ret;
1256
1257         if (!pblk_line_init_bb(pblk, line, 0)) {
1258                 list_add(&line->list, &l_mg->free_list);
1259                 return -EINTR;
1260         }
1261
1262         pblk_rl_free_lines_dec(&pblk->rl, line, true);
1263         return 0;
1264 }
1265
1266 void pblk_line_recov_close(struct pblk *pblk, struct pblk_line *line)
1267 {
1268         kfree(line->map_bitmap);
1269         line->map_bitmap = NULL;
1270         line->smeta = NULL;
1271         line->emeta = NULL;
1272 }
1273
1274 static void pblk_line_reinit(struct pblk_line *line)
1275 {
1276         *line->vsc = cpu_to_le32(EMPTY_ENTRY);
1277
1278         line->map_bitmap = NULL;
1279         line->invalid_bitmap = NULL;
1280         line->smeta = NULL;
1281         line->emeta = NULL;
1282 }
1283
1284 void pblk_line_free(struct pblk_line *line)
1285 {
1286         kfree(line->map_bitmap);
1287         kfree(line->invalid_bitmap);
1288
1289         pblk_line_reinit(line);
1290 }
1291
1292 struct pblk_line *pblk_line_get(struct pblk *pblk)
1293 {
1294         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1295         struct pblk_line_meta *lm = &pblk->lm;
1296         struct pblk_line *line;
1297         int ret, bit;
1298
1299         lockdep_assert_held(&l_mg->free_lock);
1300
1301 retry:
1302         if (list_empty(&l_mg->free_list)) {
1303                 pblk_err(pblk, "no free lines\n");
1304                 return NULL;
1305         }
1306
1307         line = list_first_entry(&l_mg->free_list, struct pblk_line, list);
1308         list_del(&line->list);
1309         l_mg->nr_free_lines--;
1310
1311         bit = find_first_zero_bit(line->blk_bitmap, lm->blk_per_line);
1312         if (unlikely(bit >= lm->blk_per_line)) {
1313                 spin_lock(&line->lock);
1314                 line->state = PBLK_LINESTATE_BAD;
1315                 spin_unlock(&line->lock);
1316
1317                 list_add_tail(&line->list, &l_mg->bad_list);
1318
1319                 pblk_debug(pblk, "line %d is bad\n", line->id);
1320                 goto retry;
1321         }
1322
1323         ret = pblk_line_prepare(pblk, line);
1324         if (ret) {
1325                 switch (ret) {
1326                 case -EAGAIN:
1327                         list_add(&line->list, &l_mg->bad_list);
1328                         goto retry;
1329                 case -EINTR:
1330                         list_add(&line->list, &l_mg->corrupt_list);
1331                         goto retry;
1332                 default:
1333                         pblk_err(pblk, "failed to prepare line %d\n", line->id);
1334                         list_add(&line->list, &l_mg->free_list);
1335                         l_mg->nr_free_lines++;
1336                         return NULL;
1337                 }
1338         }
1339
1340         return line;
1341 }
1342
1343 static struct pblk_line *pblk_line_retry(struct pblk *pblk,
1344                                          struct pblk_line *line)
1345 {
1346         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1347         struct pblk_line *retry_line;
1348
1349 retry:
1350         spin_lock(&l_mg->free_lock);
1351         retry_line = pblk_line_get(pblk);
1352         if (!retry_line) {
1353                 l_mg->data_line = NULL;
1354                 spin_unlock(&l_mg->free_lock);
1355                 return NULL;
1356         }
1357
1358         retry_line->map_bitmap = line->map_bitmap;
1359         retry_line->invalid_bitmap = line->invalid_bitmap;
1360         retry_line->smeta = line->smeta;
1361         retry_line->emeta = line->emeta;
1362         retry_line->meta_line = line->meta_line;
1363
1364         pblk_line_reinit(line);
1365
1366         l_mg->data_line = retry_line;
1367         spin_unlock(&l_mg->free_lock);
1368
1369         pblk_rl_free_lines_dec(&pblk->rl, line, false);
1370
1371         if (pblk_line_erase(pblk, retry_line))
1372                 goto retry;
1373
1374         return retry_line;
1375 }
1376
1377 static void pblk_set_space_limit(struct pblk *pblk)
1378 {
1379         struct pblk_rl *rl = &pblk->rl;
1380
1381         atomic_set(&rl->rb_space, 0);
1382 }
1383
1384 struct pblk_line *pblk_line_get_first_data(struct pblk *pblk)
1385 {
1386         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1387         struct pblk_line *line;
1388
1389         spin_lock(&l_mg->free_lock);
1390         line = pblk_line_get(pblk);
1391         if (!line) {
1392                 spin_unlock(&l_mg->free_lock);
1393                 return NULL;
1394         }
1395
1396         line->seq_nr = l_mg->d_seq_nr++;
1397         line->type = PBLK_LINETYPE_DATA;
1398         l_mg->data_line = line;
1399
1400         pblk_line_setup_metadata(line, l_mg, &pblk->lm);
1401
1402         /* Allocate next line for preparation */
1403         l_mg->data_next = pblk_line_get(pblk);
1404         if (!l_mg->data_next) {
1405                 /* If we cannot get a new line, we need to stop the pipeline.
1406                  * Only allow as many writes in as we can store safely and then
1407                  * fail gracefully
1408                  */
1409                 pblk_set_space_limit(pblk);
1410
1411                 l_mg->data_next = NULL;
1412         } else {
1413                 l_mg->data_next->seq_nr = l_mg->d_seq_nr++;
1414                 l_mg->data_next->type = PBLK_LINETYPE_DATA;
1415         }
1416         spin_unlock(&l_mg->free_lock);
1417
1418         if (pblk_line_alloc_bitmaps(pblk, line))
1419                 return NULL;
1420
1421         if (pblk_line_erase(pblk, line)) {
1422                 line = pblk_line_retry(pblk, line);
1423                 if (!line)
1424                         return NULL;
1425         }
1426
1427 retry_setup:
1428         if (!pblk_line_init_metadata(pblk, line, NULL)) {
1429                 line = pblk_line_retry(pblk, line);
1430                 if (!line)
1431                         return NULL;
1432
1433                 goto retry_setup;
1434         }
1435
1436         if (!pblk_line_init_bb(pblk, line, 1)) {
1437                 line = pblk_line_retry(pblk, line);
1438                 if (!line)
1439                         return NULL;
1440
1441                 goto retry_setup;
1442         }
1443
1444         pblk_rl_free_lines_dec(&pblk->rl, line, true);
1445
1446         return line;
1447 }
1448
1449 static void pblk_stop_writes(struct pblk *pblk, struct pblk_line *line)
1450 {
1451         lockdep_assert_held(&pblk->l_mg.free_lock);
1452
1453         pblk_set_space_limit(pblk);
1454         pblk->state = PBLK_STATE_STOPPING;
1455 }
1456
1457 static void pblk_line_close_meta_sync(struct pblk *pblk)
1458 {
1459         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1460         struct pblk_line_meta *lm = &pblk->lm;
1461         struct pblk_line *line, *tline;
1462         LIST_HEAD(list);
1463
1464         spin_lock(&l_mg->close_lock);
1465         if (list_empty(&l_mg->emeta_list)) {
1466                 spin_unlock(&l_mg->close_lock);
1467                 return;
1468         }
1469
1470         list_cut_position(&list, &l_mg->emeta_list, l_mg->emeta_list.prev);
1471         spin_unlock(&l_mg->close_lock);
1472
1473         list_for_each_entry_safe(line, tline, &list, list) {
1474                 struct pblk_emeta *emeta = line->emeta;
1475
1476                 while (emeta->mem < lm->emeta_len[0]) {
1477                         int ret;
1478
1479                         ret = pblk_submit_meta_io(pblk, line);
1480                         if (ret) {
1481                                 pblk_err(pblk, "sync meta line %d failed (%d)\n",
1482                                                         line->id, ret);
1483                                 return;
1484                         }
1485                 }
1486         }
1487
1488         pblk_wait_for_meta(pblk);
1489         flush_workqueue(pblk->close_wq);
1490 }
1491
1492 void __pblk_pipeline_flush(struct pblk *pblk)
1493 {
1494         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1495         int ret;
1496
1497         spin_lock(&l_mg->free_lock);
1498         if (pblk->state == PBLK_STATE_RECOVERING ||
1499                                         pblk->state == PBLK_STATE_STOPPED) {
1500                 spin_unlock(&l_mg->free_lock);
1501                 return;
1502         }
1503         pblk->state = PBLK_STATE_RECOVERING;
1504         spin_unlock(&l_mg->free_lock);
1505
1506         pblk_flush_writer(pblk);
1507         pblk_wait_for_meta(pblk);
1508
1509         ret = pblk_recov_pad(pblk);
1510         if (ret) {
1511                 pblk_err(pblk, "could not close data on teardown(%d)\n", ret);
1512                 return;
1513         }
1514
1515         flush_workqueue(pblk->bb_wq);
1516         pblk_line_close_meta_sync(pblk);
1517 }
1518
1519 void __pblk_pipeline_stop(struct pblk *pblk)
1520 {
1521         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1522
1523         spin_lock(&l_mg->free_lock);
1524         pblk->state = PBLK_STATE_STOPPED;
1525         l_mg->data_line = NULL;
1526         l_mg->data_next = NULL;
1527         spin_unlock(&l_mg->free_lock);
1528 }
1529
1530 void pblk_pipeline_stop(struct pblk *pblk)
1531 {
1532         __pblk_pipeline_flush(pblk);
1533         __pblk_pipeline_stop(pblk);
1534 }
1535
1536 struct pblk_line *pblk_line_replace_data(struct pblk *pblk)
1537 {
1538         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1539         struct pblk_line *cur, *new = NULL;
1540         unsigned int left_seblks;
1541
1542         cur = l_mg->data_line;
1543         new = l_mg->data_next;
1544         if (!new)
1545                 goto out;
1546         l_mg->data_line = new;
1547
1548         spin_lock(&l_mg->free_lock);
1549         pblk_line_setup_metadata(new, l_mg, &pblk->lm);
1550         spin_unlock(&l_mg->free_lock);
1551
1552 retry_erase:
1553         left_seblks = atomic_read(&new->left_seblks);
1554         if (left_seblks) {
1555                 /* If line is not fully erased, erase it */
1556                 if (atomic_read(&new->left_eblks)) {
1557                         if (pblk_line_erase(pblk, new))
1558                                 goto out;
1559                 } else {
1560                         io_schedule();
1561                 }
1562                 goto retry_erase;
1563         }
1564
1565         if (pblk_line_alloc_bitmaps(pblk, new))
1566                 return NULL;
1567
1568 retry_setup:
1569         if (!pblk_line_init_metadata(pblk, new, cur)) {
1570                 new = pblk_line_retry(pblk, new);
1571                 if (!new)
1572                         goto out;
1573
1574                 goto retry_setup;
1575         }
1576
1577         if (!pblk_line_init_bb(pblk, new, 1)) {
1578                 new = pblk_line_retry(pblk, new);
1579                 if (!new)
1580                         goto out;
1581
1582                 goto retry_setup;
1583         }
1584
1585         pblk_rl_free_lines_dec(&pblk->rl, new, true);
1586
1587         /* Allocate next line for preparation */
1588         spin_lock(&l_mg->free_lock);
1589         l_mg->data_next = pblk_line_get(pblk);
1590         if (!l_mg->data_next) {
1591                 /* If we cannot get a new line, we need to stop the pipeline.
1592                  * Only allow as many writes in as we can store safely and then
1593                  * fail gracefully
1594                  */
1595                 pblk_stop_writes(pblk, new);
1596                 l_mg->data_next = NULL;
1597         } else {
1598                 l_mg->data_next->seq_nr = l_mg->d_seq_nr++;
1599                 l_mg->data_next->type = PBLK_LINETYPE_DATA;
1600         }
1601         spin_unlock(&l_mg->free_lock);
1602
1603 out:
1604         return new;
1605 }
1606
1607 static void __pblk_line_put(struct pblk *pblk, struct pblk_line *line)
1608 {
1609         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1610         struct pblk_gc *gc = &pblk->gc;
1611
1612         spin_lock(&line->lock);
1613         WARN_ON(line->state != PBLK_LINESTATE_GC);
1614         line->state = PBLK_LINESTATE_FREE;
1615         line->gc_group = PBLK_LINEGC_NONE;
1616         pblk_line_free(line);
1617
1618         if (line->w_err_gc->has_write_err) {
1619                 pblk_rl_werr_line_out(&pblk->rl);
1620                 line->w_err_gc->has_write_err = 0;
1621         }
1622
1623         spin_unlock(&line->lock);
1624         atomic_dec(&gc->pipeline_gc);
1625
1626         spin_lock(&l_mg->free_lock);
1627         list_add_tail(&line->list, &l_mg->free_list);
1628         l_mg->nr_free_lines++;
1629         spin_unlock(&l_mg->free_lock);
1630
1631         pblk_rl_free_lines_inc(&pblk->rl, line);
1632 }
1633
1634 static void pblk_line_put_ws(struct work_struct *work)
1635 {
1636         struct pblk_line_ws *line_put_ws = container_of(work,
1637                                                 struct pblk_line_ws, ws);
1638         struct pblk *pblk = line_put_ws->pblk;
1639         struct pblk_line *line = line_put_ws->line;
1640
1641         __pblk_line_put(pblk, line);
1642         mempool_free(line_put_ws, &pblk->gen_ws_pool);
1643 }
1644
1645 void pblk_line_put(struct kref *ref)
1646 {
1647         struct pblk_line *line = container_of(ref, struct pblk_line, ref);
1648         struct pblk *pblk = line->pblk;
1649
1650         __pblk_line_put(pblk, line);
1651 }
1652
1653 void pblk_line_put_wq(struct kref *ref)
1654 {
1655         struct pblk_line *line = container_of(ref, struct pblk_line, ref);
1656         struct pblk *pblk = line->pblk;
1657         struct pblk_line_ws *line_put_ws;
1658
1659         line_put_ws = mempool_alloc(&pblk->gen_ws_pool, GFP_ATOMIC);
1660         if (!line_put_ws)
1661                 return;
1662
1663         line_put_ws->pblk = pblk;
1664         line_put_ws->line = line;
1665         line_put_ws->priv = NULL;
1666
1667         INIT_WORK(&line_put_ws->ws, pblk_line_put_ws);
1668         queue_work(pblk->r_end_wq, &line_put_ws->ws);
1669 }
1670
1671 int pblk_blk_erase_async(struct pblk *pblk, struct ppa_addr ppa)
1672 {
1673         struct nvm_rq *rqd;
1674         int err;
1675
1676         rqd = pblk_alloc_rqd(pblk, PBLK_ERASE);
1677
1678         pblk_setup_e_rq(pblk, rqd, ppa);
1679
1680         rqd->end_io = pblk_end_io_erase;
1681         rqd->private = pblk;
1682
1683         /* The write thread schedules erases so that it minimizes disturbances
1684          * with writes. Thus, there is no need to take the LUN semaphore.
1685          */
1686         err = pblk_submit_io(pblk, rqd);
1687         if (err) {
1688                 struct nvm_tgt_dev *dev = pblk->dev;
1689                 struct nvm_geo *geo = &dev->geo;
1690
1691                 pblk_err(pblk, "could not async erase line:%d,blk:%d\n",
1692                                         pblk_ppa_to_line(ppa),
1693                                         pblk_ppa_to_pos(geo, ppa));
1694         }
1695
1696         return err;
1697 }
1698
1699 struct pblk_line *pblk_line_get_data(struct pblk *pblk)
1700 {
1701         return pblk->l_mg.data_line;
1702 }
1703
1704 /* For now, always erase next line */
1705 struct pblk_line *pblk_line_get_erase(struct pblk *pblk)
1706 {
1707         return pblk->l_mg.data_next;
1708 }
1709
1710 int pblk_line_is_full(struct pblk_line *line)
1711 {
1712         return (line->left_msecs == 0);
1713 }
1714
1715 static void pblk_line_should_sync_meta(struct pblk *pblk)
1716 {
1717         if (pblk_rl_is_limit(&pblk->rl))
1718                 pblk_line_close_meta_sync(pblk);
1719 }
1720
1721 void pblk_line_close(struct pblk *pblk, struct pblk_line *line)
1722 {
1723         struct nvm_tgt_dev *dev = pblk->dev;
1724         struct nvm_geo *geo = &dev->geo;
1725         struct pblk_line_meta *lm = &pblk->lm;
1726         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1727         struct list_head *move_list;
1728         int i;
1729
1730 #ifdef CONFIG_NVM_PBLK_DEBUG
1731         WARN(!bitmap_full(line->map_bitmap, lm->sec_per_line),
1732                                 "pblk: corrupt closed line %d\n", line->id);
1733 #endif
1734
1735         spin_lock(&l_mg->free_lock);
1736         WARN_ON(!test_and_clear_bit(line->meta_line, &l_mg->meta_bitmap));
1737         spin_unlock(&l_mg->free_lock);
1738
1739         spin_lock(&l_mg->gc_lock);
1740         spin_lock(&line->lock);
1741         WARN_ON(line->state != PBLK_LINESTATE_OPEN);
1742         line->state = PBLK_LINESTATE_CLOSED;
1743         move_list = pblk_line_gc_list(pblk, line);
1744
1745         list_add_tail(&line->list, move_list);
1746
1747         kfree(line->map_bitmap);
1748         line->map_bitmap = NULL;
1749         line->smeta = NULL;
1750         line->emeta = NULL;
1751
1752         for (i = 0; i < lm->blk_per_line; i++) {
1753                 struct pblk_lun *rlun = &pblk->luns[i];
1754                 int pos = pblk_ppa_to_pos(geo, rlun->bppa);
1755                 int state = line->chks[pos].state;
1756
1757                 if (!(state & NVM_CHK_ST_OFFLINE))
1758                         state = NVM_CHK_ST_CLOSED;
1759         }
1760
1761         spin_unlock(&line->lock);
1762         spin_unlock(&l_mg->gc_lock);
1763 }
1764
1765 void pblk_line_close_meta(struct pblk *pblk, struct pblk_line *line)
1766 {
1767         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1768         struct pblk_line_meta *lm = &pblk->lm;
1769         struct pblk_emeta *emeta = line->emeta;
1770         struct line_emeta *emeta_buf = emeta->buf;
1771         struct wa_counters *wa = emeta_to_wa(lm, emeta_buf);
1772
1773         /* No need for exact vsc value; avoid a big line lock and take aprox. */
1774         memcpy(emeta_to_vsc(pblk, emeta_buf), l_mg->vsc_list, lm->vsc_list_len);
1775         memcpy(emeta_to_bb(emeta_buf), line->blk_bitmap, lm->blk_bitmap_len);
1776
1777         wa->user = cpu_to_le64(atomic64_read(&pblk->user_wa));
1778         wa->pad = cpu_to_le64(atomic64_read(&pblk->pad_wa));
1779         wa->gc = cpu_to_le64(atomic64_read(&pblk->gc_wa));
1780
1781         emeta_buf->nr_valid_lbas = cpu_to_le64(line->nr_valid_lbas);
1782         emeta_buf->crc = cpu_to_le32(pblk_calc_emeta_crc(pblk, emeta_buf));
1783
1784         spin_lock(&l_mg->close_lock);
1785         spin_lock(&line->lock);
1786
1787         /* Update the in-memory start address for emeta, in case it has
1788          * shifted due to write errors
1789          */
1790         if (line->emeta_ssec != line->cur_sec)
1791                 line->emeta_ssec = line->cur_sec;
1792
1793         list_add_tail(&line->list, &l_mg->emeta_list);
1794         spin_unlock(&line->lock);
1795         spin_unlock(&l_mg->close_lock);
1796
1797         pblk_line_should_sync_meta(pblk);
1798
1799
1800 }
1801
1802 static void pblk_save_lba_list(struct pblk *pblk, struct pblk_line *line)
1803 {
1804         struct pblk_line_meta *lm = &pblk->lm;
1805         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1806         unsigned int lba_list_size = lm->emeta_len[2];
1807         struct pblk_w_err_gc *w_err_gc = line->w_err_gc;
1808         struct pblk_emeta *emeta = line->emeta;
1809
1810         w_err_gc->lba_list = pblk_malloc(lba_list_size,
1811                                          l_mg->emeta_alloc_type, GFP_KERNEL);
1812         memcpy(w_err_gc->lba_list, emeta_to_lbas(pblk, emeta->buf),
1813                                 lba_list_size);
1814 }
1815
1816 void pblk_line_close_ws(struct work_struct *work)
1817 {
1818         struct pblk_line_ws *line_ws = container_of(work, struct pblk_line_ws,
1819                                                                         ws);
1820         struct pblk *pblk = line_ws->pblk;
1821         struct pblk_line *line = line_ws->line;
1822         struct pblk_w_err_gc *w_err_gc = line->w_err_gc;
1823
1824         /* Write errors makes the emeta start address stored in smeta invalid,
1825          * so keep a copy of the lba list until we've gc'd the line
1826          */
1827         if (w_err_gc->has_write_err)
1828                 pblk_save_lba_list(pblk, line);
1829
1830         pblk_line_close(pblk, line);
1831         mempool_free(line_ws, &pblk->gen_ws_pool);
1832 }
1833
1834 void pblk_gen_run_ws(struct pblk *pblk, struct pblk_line *line, void *priv,
1835                       void (*work)(struct work_struct *), gfp_t gfp_mask,
1836                       struct workqueue_struct *wq)
1837 {
1838         struct pblk_line_ws *line_ws;
1839
1840         line_ws = mempool_alloc(&pblk->gen_ws_pool, gfp_mask);
1841
1842         line_ws->pblk = pblk;
1843         line_ws->line = line;
1844         line_ws->priv = priv;
1845
1846         INIT_WORK(&line_ws->ws, work);
1847         queue_work(wq, &line_ws->ws);
1848 }
1849
1850 static void __pblk_down_page(struct pblk *pblk, struct ppa_addr *ppa_list,
1851                              int nr_ppas, int pos)
1852 {
1853         struct pblk_lun *rlun = &pblk->luns[pos];
1854         int ret;
1855
1856         /*
1857          * Only send one inflight I/O per LUN. Since we map at a page
1858          * granurality, all ppas in the I/O will map to the same LUN
1859          */
1860 #ifdef CONFIG_NVM_PBLK_DEBUG
1861         int i;
1862
1863         for (i = 1; i < nr_ppas; i++)
1864                 WARN_ON(ppa_list[0].a.lun != ppa_list[i].a.lun ||
1865                                 ppa_list[0].a.ch != ppa_list[i].a.ch);
1866 #endif
1867
1868         ret = down_timeout(&rlun->wr_sem, msecs_to_jiffies(30000));
1869         if (ret == -ETIME || ret == -EINTR)
1870                 pblk_err(pblk, "taking lun semaphore timed out: err %d\n",
1871                                 -ret);
1872 }
1873
1874 void pblk_down_page(struct pblk *pblk, struct ppa_addr *ppa_list, int nr_ppas)
1875 {
1876         struct nvm_tgt_dev *dev = pblk->dev;
1877         struct nvm_geo *geo = &dev->geo;
1878         int pos = pblk_ppa_to_pos(geo, ppa_list[0]);
1879
1880         __pblk_down_page(pblk, ppa_list, nr_ppas, pos);
1881 }
1882
1883 void pblk_down_rq(struct pblk *pblk, struct ppa_addr *ppa_list, int nr_ppas,
1884                   unsigned long *lun_bitmap)
1885 {
1886         struct nvm_tgt_dev *dev = pblk->dev;
1887         struct nvm_geo *geo = &dev->geo;
1888         int pos = pblk_ppa_to_pos(geo, ppa_list[0]);
1889
1890         /* If the LUN has been locked for this same request, do no attempt to
1891          * lock it again
1892          */
1893         if (test_and_set_bit(pos, lun_bitmap))
1894                 return;
1895
1896         __pblk_down_page(pblk, ppa_list, nr_ppas, pos);
1897 }
1898
1899 void pblk_up_page(struct pblk *pblk, struct ppa_addr *ppa_list, int nr_ppas)
1900 {
1901         struct nvm_tgt_dev *dev = pblk->dev;
1902         struct nvm_geo *geo = &dev->geo;
1903         struct pblk_lun *rlun;
1904         int pos = pblk_ppa_to_pos(geo, ppa_list[0]);
1905
1906 #ifdef CONFIG_NVM_PBLK_DEBUG
1907         int i;
1908
1909         for (i = 1; i < nr_ppas; i++)
1910                 WARN_ON(ppa_list[0].a.lun != ppa_list[i].a.lun ||
1911                                 ppa_list[0].a.ch != ppa_list[i].a.ch);
1912 #endif
1913
1914         rlun = &pblk->luns[pos];
1915         up(&rlun->wr_sem);
1916 }
1917
1918 void pblk_up_rq(struct pblk *pblk, struct ppa_addr *ppa_list, int nr_ppas,
1919                 unsigned long *lun_bitmap)
1920 {
1921         struct nvm_tgt_dev *dev = pblk->dev;
1922         struct nvm_geo *geo = &dev->geo;
1923         struct pblk_lun *rlun;
1924         int num_lun = geo->all_luns;
1925         int bit = -1;
1926
1927         while ((bit = find_next_bit(lun_bitmap, num_lun, bit + 1)) < num_lun) {
1928                 rlun = &pblk->luns[bit];
1929                 up(&rlun->wr_sem);
1930         }
1931 }
1932
1933 void pblk_update_map(struct pblk *pblk, sector_t lba, struct ppa_addr ppa)
1934 {
1935         struct ppa_addr ppa_l2p;
1936
1937         /* logic error: lba out-of-bounds. Ignore update */
1938         if (!(lba < pblk->rl.nr_secs)) {
1939                 WARN(1, "pblk: corrupted L2P map request\n");
1940                 return;
1941         }
1942
1943         spin_lock(&pblk->trans_lock);
1944         ppa_l2p = pblk_trans_map_get(pblk, lba);
1945
1946         if (!pblk_addr_in_cache(ppa_l2p) && !pblk_ppa_empty(ppa_l2p))
1947                 pblk_map_invalidate(pblk, ppa_l2p);
1948
1949         pblk_trans_map_set(pblk, lba, ppa);
1950         spin_unlock(&pblk->trans_lock);
1951 }
1952
1953 void pblk_update_map_cache(struct pblk *pblk, sector_t lba, struct ppa_addr ppa)
1954 {
1955
1956 #ifdef CONFIG_NVM_PBLK_DEBUG
1957         /* Callers must ensure that the ppa points to a cache address */
1958         BUG_ON(!pblk_addr_in_cache(ppa));
1959         BUG_ON(pblk_rb_pos_oob(&pblk->rwb, pblk_addr_to_cacheline(ppa)));
1960 #endif
1961
1962         pblk_update_map(pblk, lba, ppa);
1963 }
1964
1965 int pblk_update_map_gc(struct pblk *pblk, sector_t lba, struct ppa_addr ppa_new,
1966                        struct pblk_line *gc_line, u64 paddr_gc)
1967 {
1968         struct ppa_addr ppa_l2p, ppa_gc;
1969         int ret = 1;
1970
1971 #ifdef CONFIG_NVM_PBLK_DEBUG
1972         /* Callers must ensure that the ppa points to a cache address */
1973         BUG_ON(!pblk_addr_in_cache(ppa_new));
1974         BUG_ON(pblk_rb_pos_oob(&pblk->rwb, pblk_addr_to_cacheline(ppa_new)));
1975 #endif
1976
1977         /* logic error: lba out-of-bounds. Ignore update */
1978         if (!(lba < pblk->rl.nr_secs)) {
1979                 WARN(1, "pblk: corrupted L2P map request\n");
1980                 return 0;
1981         }
1982
1983         spin_lock(&pblk->trans_lock);
1984         ppa_l2p = pblk_trans_map_get(pblk, lba);
1985         ppa_gc = addr_to_gen_ppa(pblk, paddr_gc, gc_line->id);
1986
1987         if (!pblk_ppa_comp(ppa_l2p, ppa_gc)) {
1988                 spin_lock(&gc_line->lock);
1989                 WARN(!test_bit(paddr_gc, gc_line->invalid_bitmap),
1990                                                 "pblk: corrupted GC update");
1991                 spin_unlock(&gc_line->lock);
1992
1993                 ret = 0;
1994                 goto out;
1995         }
1996
1997         pblk_trans_map_set(pblk, lba, ppa_new);
1998 out:
1999         spin_unlock(&pblk->trans_lock);
2000         return ret;
2001 }
2002
2003 void pblk_update_map_dev(struct pblk *pblk, sector_t lba,
2004                          struct ppa_addr ppa_mapped, struct ppa_addr ppa_cache)
2005 {
2006         struct ppa_addr ppa_l2p;
2007
2008 #ifdef CONFIG_NVM_PBLK_DEBUG
2009         /* Callers must ensure that the ppa points to a device address */
2010         BUG_ON(pblk_addr_in_cache(ppa_mapped));
2011 #endif
2012         /* Invalidate and discard padded entries */
2013         if (lba == ADDR_EMPTY) {
2014                 atomic64_inc(&pblk->pad_wa);
2015 #ifdef CONFIG_NVM_PBLK_DEBUG
2016                 atomic_long_inc(&pblk->padded_wb);
2017 #endif
2018                 if (!pblk_ppa_empty(ppa_mapped))
2019                         pblk_map_invalidate(pblk, ppa_mapped);
2020                 return;
2021         }
2022
2023         /* logic error: lba out-of-bounds. Ignore update */
2024         if (!(lba < pblk->rl.nr_secs)) {
2025                 WARN(1, "pblk: corrupted L2P map request\n");
2026                 return;
2027         }
2028
2029         spin_lock(&pblk->trans_lock);
2030         ppa_l2p = pblk_trans_map_get(pblk, lba);
2031
2032         /* Do not update L2P if the cacheline has been updated. In this case,
2033          * the mapped ppa must be invalidated
2034          */
2035         if (!pblk_ppa_comp(ppa_l2p, ppa_cache)) {
2036                 if (!pblk_ppa_empty(ppa_mapped))
2037                         pblk_map_invalidate(pblk, ppa_mapped);
2038                 goto out;
2039         }
2040
2041 #ifdef CONFIG_NVM_PBLK_DEBUG
2042         WARN_ON(!pblk_addr_in_cache(ppa_l2p) && !pblk_ppa_empty(ppa_l2p));
2043 #endif
2044
2045         pblk_trans_map_set(pblk, lba, ppa_mapped);
2046 out:
2047         spin_unlock(&pblk->trans_lock);
2048 }
2049
2050 void pblk_lookup_l2p_seq(struct pblk *pblk, struct ppa_addr *ppas,
2051                          sector_t blba, int nr_secs)
2052 {
2053         int i;
2054
2055         spin_lock(&pblk->trans_lock);
2056         for (i = 0; i < nr_secs; i++) {
2057                 struct ppa_addr ppa;
2058
2059                 ppa = ppas[i] = pblk_trans_map_get(pblk, blba + i);
2060
2061                 /* If the L2P entry maps to a line, the reference is valid */
2062                 if (!pblk_ppa_empty(ppa) && !pblk_addr_in_cache(ppa)) {
2063                         int line_id = pblk_ppa_to_line(ppa);
2064                         struct pblk_line *line = &pblk->lines[line_id];
2065
2066                         kref_get(&line->ref);
2067                 }
2068         }
2069         spin_unlock(&pblk->trans_lock);
2070 }
2071
2072 void pblk_lookup_l2p_rand(struct pblk *pblk, struct ppa_addr *ppas,
2073                           u64 *lba_list, int nr_secs)
2074 {
2075         u64 lba;
2076         int i;
2077
2078         spin_lock(&pblk->trans_lock);
2079         for (i = 0; i < nr_secs; i++) {
2080                 lba = lba_list[i];
2081                 if (lba != ADDR_EMPTY) {
2082                         /* logic error: lba out-of-bounds. Ignore update */
2083                         if (!(lba < pblk->rl.nr_secs)) {
2084                                 WARN(1, "pblk: corrupted L2P map request\n");
2085                                 continue;
2086                         }
2087                         ppas[i] = pblk_trans_map_get(pblk, lba);
2088                 }
2089         }
2090         spin_unlock(&pblk->trans_lock);
2091 }