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         struct page *page;
 293         int i, e, nbv = 0;
 294
 295         for (i = 0; i < bio->bi_vcnt; i++) {
 296                 bv = &bio->bi_io_vec[i];
 297                 page = bv->bv_page;
 298                 for (e = 0; e < bv->bv_len; e += PBLK_EXPOSED_PAGE_SIZE, nbv++)
 299                         if (nbv >= off)
 300                                 mempool_free(page++, &pblk->page_bio_pool);
 301         }
 302 }
 303
 304 int pblk_bio_add_pages(struct pblk *pblk, struct bio *bio, gfp_t flags,
 305                        int nr_pages)
 306 {
 307         struct request_queue *q = pblk->dev->q;
 308         struct page *page;
 309         int i, ret;
 310
 311         for (i = 0; i < nr_pages; i++) {
 312                 page = mempool_alloc(&pblk->page_bio_pool, flags);
 313
 314                 ret = bio_add_pc_page(q, bio, page, PBLK_EXPOSED_PAGE_SIZE, 0);
 315                 if (ret != PBLK_EXPOSED_PAGE_SIZE) {
 316                         pblk_err(pblk, "could not add page to bio\n");
 317                         mempool_free(page, &pblk->page_bio_pool);
 318                         goto err;
 319                 }
 320         }
 321
 322         return 0;
 323 err:
 324         pblk_bio_free_pages(pblk, bio, (bio->bi_vcnt - i), i);
 325         return -1;
 326 }
 327
 328 void pblk_write_kick(struct pblk *pblk)
 329 {
 330         wake_up_process(pblk->writer_ts);
 331         mod_timer(&pblk->wtimer, jiffies + msecs_to_jiffies(1000));
 332 }
 333
 334 void pblk_write_timer_fn(struct timer_list *t)
 335 {
 336         struct pblk *pblk = from_timer(pblk, t, wtimer);
 337
 338         /* kick the write thread every tick to flush outstanding data */
 339         pblk_write_kick(pblk);
 340 }
 341
 342 void pblk_write_should_kick(struct pblk *pblk)
 343 {
 344         unsigned int secs_avail = pblk_rb_read_count(&pblk->rwb);
 345
 346         if (secs_avail >= pblk->min_write_pgs)
 347                 pblk_write_kick(pblk);
 348 }
 349
 350 static void pblk_wait_for_meta(struct pblk *pblk)
 351 {
 352         do {
 353                 if (!atomic_read(&pblk->inflight_io))
 354                         break;
 355
 356                 schedule();
 357         } while (1);
 358 }
 359
 360 static void pblk_flush_writer(struct pblk *pblk)
 361 {
 362         pblk_rb_flush(&pblk->rwb);
 363         do {
 364                 if (!pblk_rb_sync_count(&pblk->rwb))
 365                         break;
 366
 367                 pblk_write_kick(pblk);
 368                 schedule();
 369         } while (1);
 370 }
 371
 372 struct list_head *pblk_line_gc_list(struct pblk *pblk, struct pblk_line *line)
 373 {
 374         struct pblk_line_meta *lm = &pblk->lm;
 375         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
 376         struct list_head *move_list = NULL;
 377         int vsc = le32_to_cpu(*line->vsc);
 378
 379         lockdep_assert_held(&line->lock);
 380
 381         if (line->w_err_gc->has_write_err) {
 382                 if (line->gc_group != PBLK_LINEGC_WERR) {
 383                         line->gc_group = PBLK_LINEGC_WERR;
 384                         move_list = &l_mg->gc_werr_list;
 385                         pblk_rl_werr_line_in(&pblk->rl);
 386                 }
 387         } else if (!vsc) {
 388                 if (line->gc_group != PBLK_LINEGC_FULL) {
 389                         line->gc_group = PBLK_LINEGC_FULL;
 390                         move_list = &l_mg->gc_full_list;
 391                 }
 392         } else if (vsc < lm->high_thrs) {
 393                 if (line->gc_group != PBLK_LINEGC_HIGH) {
 394                         line->gc_group = PBLK_LINEGC_HIGH;
 395                         move_list = &l_mg->gc_high_list;
 396                 }
 397         } else if (vsc < lm->mid_thrs) {
 398                 if (line->gc_group != PBLK_LINEGC_MID) {
 399                         line->gc_group = PBLK_LINEGC_MID;
 400                         move_list = &l_mg->gc_mid_list;
 401                 }
 402         } else if (vsc < line->sec_in_line) {
 403                 if (line->gc_group != PBLK_LINEGC_LOW) {
 404                         line->gc_group = PBLK_LINEGC_LOW;
 405                         move_list = &l_mg->gc_low_list;
 406                 }
 407         } else if (vsc == line->sec_in_line) {
 408                 if (line->gc_group != PBLK_LINEGC_EMPTY) {
 409                         line->gc_group = PBLK_LINEGC_EMPTY;
 410                         move_list = &l_mg->gc_empty_list;
 411                 }
 412         } else {
 413                 line->state = PBLK_LINESTATE_CORRUPT;
 414                 line->gc_group = PBLK_LINEGC_NONE;
 415                 move_list =  &l_mg->corrupt_list;
 416                 pblk_err(pblk, "corrupted vsc for line %d, vsc:%d (%d/%d/%d)\n",
 417                                                 line->id, vsc,
 418                                                 line->sec_in_line,
 419                                                 lm->high_thrs, lm->mid_thrs);
 420         }
 421
 422         return move_list;
 423 }
 424
 425 void pblk_discard(struct pblk *pblk, struct bio *bio)
 426 {
 427         sector_t slba = pblk_get_lba(bio);
 428         sector_t nr_secs = pblk_get_secs(bio);
 429
 430         pblk_invalidate_range(pblk, slba, nr_secs);
 431 }
 432
 433 void pblk_log_write_err(struct pblk *pblk, struct nvm_rq *rqd)
 434 {
 435         atomic_long_inc(&pblk->write_failed);
 436 #ifdef CONFIG_NVM_PBLK_DEBUG
 437         pblk_print_failed_rqd(pblk, rqd, rqd->error);
 438 #endif
 439 }
 440
 441 void pblk_log_read_err(struct pblk *pblk, struct nvm_rq *rqd)
 442 {
 443         /* Empty page read is not necessarily an error (e.g., L2P recovery) */
 444         if (rqd->error == NVM_RSP_ERR_EMPTYPAGE) {
 445                 atomic_long_inc(&pblk->read_empty);
 446                 return;
 447         }
 448
 449         switch (rqd->error) {
 450         case NVM_RSP_WARN_HIGHECC:
 451                 atomic_long_inc(&pblk->read_high_ecc);
 452                 break;
 453         case NVM_RSP_ERR_FAILECC:
 454         case NVM_RSP_ERR_FAILCRC:
 455                 atomic_long_inc(&pblk->read_failed);
 456                 break;
 457         default:
 458                 pblk_err(pblk, "unknown read error:%d\n", rqd->error);
 459         }
 460 #ifdef CONFIG_NVM_PBLK_DEBUG
 461         pblk_print_failed_rqd(pblk, rqd, rqd->error);
 462 #endif
 463 }
 464
 465 void pblk_set_sec_per_write(struct pblk *pblk, int sec_per_write)
 466 {
 467         pblk->sec_per_write = sec_per_write;
 468 }
 469
 470 int pblk_submit_io(struct pblk *pblk, struct nvm_rq *rqd)
 471 {
 472         struct nvm_tgt_dev *dev = pblk->dev;
 473
 474         atomic_inc(&pblk->inflight_io);
 475
 476 #ifdef CONFIG_NVM_PBLK_DEBUG
 477         if (pblk_check_io(pblk, rqd))
 478                 return NVM_IO_ERR;
 479 #endif
 480
 481         return nvm_submit_io(dev, rqd);
 482 }
 483
 484 int pblk_submit_io_sync(struct pblk *pblk, struct nvm_rq *rqd)
 485 {
 486         struct nvm_tgt_dev *dev = pblk->dev;
 487
 488         atomic_inc(&pblk->inflight_io);
 489
 490 #ifdef CONFIG_NVM_PBLK_DEBUG
 491         if (pblk_check_io(pblk, rqd))
 492                 return NVM_IO_ERR;
 493 #endif
 494
 495         return nvm_submit_io_sync(dev, rqd);
 496 }
 497
 498 static void pblk_bio_map_addr_endio(struct bio *bio)
 499 {
 500         bio_put(bio);
 501 }
 502
 503 struct bio *pblk_bio_map_addr(struct pblk *pblk, void *data,
 504                               unsigned int nr_secs, unsigned int len,
 505                               int alloc_type, gfp_t gfp_mask)
 506 {
 507         struct nvm_tgt_dev *dev = pblk->dev;
 508         void *kaddr = data;
 509         struct page *page;
 510         struct bio *bio;
 511         int i, ret;
 512
 513         if (alloc_type == PBLK_KMALLOC_META)
 514                 return bio_map_kern(dev->q, kaddr, len, gfp_mask);
 515
 516         bio = bio_kmalloc(gfp_mask, nr_secs);
 517         if (!bio)
 518                 return ERR_PTR(-ENOMEM);
 519
 520         for (i = 0; i < nr_secs; i++) {
 521                 page = vmalloc_to_page(kaddr);
 522                 if (!page) {
 523                         pblk_err(pblk, "could not map vmalloc bio\n");
 524                         bio_put(bio);
 525                         bio = ERR_PTR(-ENOMEM);
 526                         goto out;
 527                 }
 528
 529                 ret = bio_add_pc_page(dev->q, bio, page, PAGE_SIZE, 0);
 530                 if (ret != PAGE_SIZE) {
 531                         pblk_err(pblk, "could not add page to bio\n");
 532                         bio_put(bio);
 533                         bio = ERR_PTR(-ENOMEM);
 534                         goto out;
 535                 }
 536
 537                 kaddr += PAGE_SIZE;
 538         }
 539
 540         bio->bi_end_io = pblk_bio_map_addr_endio;
 541 out:
 542         return bio;
 543 }
 544
 545 int pblk_calc_secs(struct pblk *pblk, unsigned long secs_avail,
 546                    unsigned long secs_to_flush)
 547 {
 548         int max = pblk->sec_per_write;
 549         int min = pblk->min_write_pgs;
 550         int secs_to_sync = 0;
 551
 552         if (secs_avail >= max)
 553                 secs_to_sync = max;
 554         else if (secs_avail >= min)
 555                 secs_to_sync = min * (secs_avail / min);
 556         else if (secs_to_flush)
 557                 secs_to_sync = min;
 558
 559         return secs_to_sync;
 560 }
 561
 562 void pblk_dealloc_page(struct pblk *pblk, struct pblk_line *line, int nr_secs)
 563 {
 564         u64 addr;
 565         int i;
 566
 567         spin_lock(&line->lock);
 568         addr = find_next_zero_bit(line->map_bitmap,
 569                                         pblk->lm.sec_per_line, line->cur_sec);
 570         line->cur_sec = addr - nr_secs;
 571
 572         for (i = 0; i < nr_secs; i++, line->cur_sec--)
 573                 WARN_ON(!test_and_clear_bit(line->cur_sec, line->map_bitmap));
 574         spin_unlock(&line->lock);
 575 }
 576
 577 u64 __pblk_alloc_page(struct pblk *pblk, struct pblk_line *line, int nr_secs)
 578 {
 579         u64 addr;
 580         int i;
 581
 582         lockdep_assert_held(&line->lock);
 583
 584         /* logic error: ppa out-of-bounds. Prevent generating bad address */
 585         if (line->cur_sec + nr_secs > pblk->lm.sec_per_line) {
 586                 WARN(1, "pblk: page allocation out of bounds\n");
 587                 nr_secs = pblk->lm.sec_per_line - line->cur_sec;
 588         }
 589
 590         line->cur_sec = addr = find_next_zero_bit(line->map_bitmap,
 591                                         pblk->lm.sec_per_line, line->cur_sec);
 592         for (i = 0; i < nr_secs; i++, line->cur_sec++)
 593                 WARN_ON(test_and_set_bit(line->cur_sec, line->map_bitmap));
 594
 595         return addr;
 596 }
 597
 598 u64 pblk_alloc_page(struct pblk *pblk, struct pblk_line *line, int nr_secs)
 599 {
 600         u64 addr;
 601
 602         /* Lock needed in case a write fails and a recovery needs to remap
 603          * failed write buffer entries
 604          */
 605         spin_lock(&line->lock);
 606         addr = __pblk_alloc_page(pblk, line, nr_secs);
 607         line->left_msecs -= nr_secs;
 608         WARN(line->left_msecs < 0, "pblk: page allocation out of bounds\n");
 609         spin_unlock(&line->lock);
 610
 611         return addr;
 612 }
 613
 614 u64 pblk_lookup_page(struct pblk *pblk, struct pblk_line *line)
 615 {
 616         u64 paddr;
 617
 618         spin_lock(&line->lock);
 619         paddr = find_next_zero_bit(line->map_bitmap,
 620                                         pblk->lm.sec_per_line, line->cur_sec);
 621         spin_unlock(&line->lock);
 622
 623         return paddr;
 624 }
 625
 626 /*
 627  * Submit emeta to one LUN in the raid line at the time to avoid a deadlock when
 628  * taking the per LUN semaphore.
 629  */
 630 static int pblk_line_submit_emeta_io(struct pblk *pblk, struct pblk_line *line,
 631                                      void *emeta_buf, u64 paddr, int dir)
 632 {
 633         struct nvm_tgt_dev *dev = pblk->dev;
 634         struct nvm_geo *geo = &dev->geo;
 635         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
 636         struct pblk_line_meta *lm = &pblk->lm;
 637         void *ppa_list, *meta_list;
 638         struct bio *bio;
 639         struct nvm_rq rqd;
 640         dma_addr_t dma_ppa_list, dma_meta_list;
 641         int min = pblk->min_write_pgs;
 642         int left_ppas = lm->emeta_sec[0];
 643         int id = line->id;
 644         int rq_ppas, rq_len;
 645         int cmd_op, bio_op;
 646         int i, j;
 647         int ret;
 648
 649         if (dir == PBLK_WRITE) {
 650                 bio_op = REQ_OP_WRITE;
 651                 cmd_op = NVM_OP_PWRITE;
 652         } else if (dir == PBLK_READ) {
 653                 bio_op = REQ_OP_READ;
 654                 cmd_op = NVM_OP_PREAD;
 655         } else
 656                 return -EINVAL;
 657
 658         meta_list = nvm_dev_dma_alloc(dev->parent, GFP_KERNEL,
 659                                                         &dma_meta_list);
 660         if (!meta_list)
 661                 return -ENOMEM;
 662
 663         ppa_list = meta_list + pblk_dma_meta_size;
 664         dma_ppa_list = dma_meta_list + pblk_dma_meta_size;
 665
 666 next_rq:
 667         memset(&rqd, 0, sizeof(struct nvm_rq));
 668
 669         rq_ppas = pblk_calc_secs(pblk, left_ppas, 0);
 670         rq_len = rq_ppas * geo->csecs;
 671
 672         bio = pblk_bio_map_addr(pblk, emeta_buf, rq_ppas, rq_len,
 673                                         l_mg->emeta_alloc_type, GFP_KERNEL);
 674         if (IS_ERR(bio)) {
 675                 ret = PTR_ERR(bio);
 676                 goto free_rqd_dma;
 677         }
 678
 679         bio->bi_iter.bi_sector = 0; /* internal bio */
 680         bio_set_op_attrs(bio, bio_op, 0);
 681
 682         rqd.bio = bio;
 683         rqd.meta_list = meta_list;
 684         rqd.ppa_list = ppa_list;
 685         rqd.dma_meta_list = dma_meta_list;
 686         rqd.dma_ppa_list = dma_ppa_list;
 687         rqd.opcode = cmd_op;
 688         rqd.nr_ppas = rq_ppas;
 689
 690         if (dir == PBLK_WRITE) {
 691                 struct pblk_sec_meta *meta_list = rqd.meta_list;
 692
 693                 rqd.flags = pblk_set_progr_mode(pblk, PBLK_WRITE);
 694                 for (i = 0; i < rqd.nr_ppas; ) {
 695                         spin_lock(&line->lock);
 696                         paddr = __pblk_alloc_page(pblk, line, min);
 697                         spin_unlock(&line->lock);
 698                         for (j = 0; j < min; j++, i++, paddr++) {
 699                                 meta_list[i].lba = cpu_to_le64(ADDR_EMPTY);
 700                                 rqd.ppa_list[i] =
 701                                         addr_to_gen_ppa(pblk, paddr, id);
 702                         }
 703                 }
 704         } else {
 705                 for (i = 0; i < rqd.nr_ppas; ) {
 706                         struct ppa_addr ppa = addr_to_gen_ppa(pblk, paddr, id);
 707                         int pos = pblk_ppa_to_pos(geo, ppa);
 708                         int read_type = PBLK_READ_RANDOM;
 709
 710                         if (pblk_io_aligned(pblk, rq_ppas))
 711                                 read_type = PBLK_READ_SEQUENTIAL;
 712                         rqd.flags = pblk_set_read_mode(pblk, read_type);
 713
 714                         while (test_bit(pos, line->blk_bitmap)) {
 715                                 paddr += min;
 716                                 if (pblk_boundary_paddr_checks(pblk, paddr)) {
 717                                         pblk_err(pblk, "corrupt emeta line:%d\n",
 718                                                                 line->id);
 719                                         bio_put(bio);
 720                                         ret = -EINTR;
 721                                         goto free_rqd_dma;
 722                                 }
 723
 724                                 ppa = addr_to_gen_ppa(pblk, paddr, id);
 725                                 pos = pblk_ppa_to_pos(geo, ppa);
 726                         }
 727
 728                         if (pblk_boundary_paddr_checks(pblk, paddr + min)) {
 729                                 pblk_err(pblk, "corrupt emeta line:%d\n",
 730                                                                 line->id);
 731                                 bio_put(bio);
 732                                 ret = -EINTR;
 733                                 goto free_rqd_dma;
 734                         }
 735
 736                         for (j = 0; j < min; j++, i++, paddr++)
 737                                 rqd.ppa_list[i] =
 738                                         addr_to_gen_ppa(pblk, paddr, line->id);
 739                 }
 740         }
 741
 742         ret = pblk_submit_io_sync(pblk, &rqd);
 743         if (ret) {
 744                 pblk_err(pblk, "emeta I/O submission failed: %d\n", ret);
 745                 bio_put(bio);
 746                 goto free_rqd_dma;
 747         }
 748
 749         atomic_dec(&pblk->inflight_io);
 750
 751         if (rqd.error) {
 752                 if (dir == PBLK_WRITE)
 753                         pblk_log_write_err(pblk, &rqd);
 754                 else
 755                         pblk_log_read_err(pblk, &rqd);
 756         }
 757
 758         emeta_buf += rq_len;
 759         left_ppas -= rq_ppas;
 760         if (left_ppas)
 761                 goto next_rq;
 762 free_rqd_dma:
 763         nvm_dev_dma_free(dev->parent, rqd.meta_list, rqd.dma_meta_list);
 764         return ret;
 765 }
 766
 767 u64 pblk_line_smeta_start(struct pblk *pblk, struct pblk_line *line)
 768 {
 769         struct nvm_tgt_dev *dev = pblk->dev;
 770         struct nvm_geo *geo = &dev->geo;
 771         struct pblk_line_meta *lm = &pblk->lm;
 772         int bit;
 773
 774         /* This usually only happens on bad lines */
 775         bit = find_first_zero_bit(line->blk_bitmap, lm->blk_per_line);
 776         if (bit >= lm->blk_per_line)
 777                 return -1;
 778
 779         return bit * geo->ws_opt;
 780 }
 781
 782 static int pblk_line_submit_smeta_io(struct pblk *pblk, struct pblk_line *line,
 783                                      u64 paddr, int dir)
 784 {
 785         struct nvm_tgt_dev *dev = pblk->dev;
 786         struct pblk_line_meta *lm = &pblk->lm;
 787         struct bio *bio;
 788         struct nvm_rq rqd;
 789         __le64 *lba_list = NULL;
 790         int i, ret;
 791         int cmd_op, bio_op;
 792         int flags;
 793
 794         if (dir == PBLK_WRITE) {
 795                 bio_op = REQ_OP_WRITE;
 796                 cmd_op = NVM_OP_PWRITE;
 797                 flags = pblk_set_progr_mode(pblk, PBLK_WRITE);
 798                 lba_list = emeta_to_lbas(pblk, line->emeta->buf);
 799         } else if (dir == PBLK_READ_RECOV || dir == PBLK_READ) {
 800                 bio_op = REQ_OP_READ;
 801                 cmd_op = NVM_OP_PREAD;
 802                 flags = pblk_set_read_mode(pblk, PBLK_READ_SEQUENTIAL);
 803         } else
 804                 return -EINVAL;
 805
 806         memset(&rqd, 0, sizeof(struct nvm_rq));
 807
 808         rqd.meta_list = nvm_dev_dma_alloc(dev->parent, GFP_KERNEL,
 809                                                         &rqd.dma_meta_list);
 810         if (!rqd.meta_list)
 811                 return -ENOMEM;
 812
 813         rqd.ppa_list = rqd.meta_list + pblk_dma_meta_size;
 814         rqd.dma_ppa_list = rqd.dma_meta_list + pblk_dma_meta_size;
 815
 816         bio = bio_map_kern(dev->q, line->smeta, lm->smeta_len, GFP_KERNEL);
 817         if (IS_ERR(bio)) {
 818                 ret = PTR_ERR(bio);
 819                 goto free_ppa_list;
 820         }
 821
 822         bio->bi_iter.bi_sector = 0; /* internal bio */
 823         bio_set_op_attrs(bio, bio_op, 0);
 824
 825         rqd.bio = bio;
 826         rqd.opcode = cmd_op;
 827         rqd.flags = flags;
 828         rqd.nr_ppas = lm->smeta_sec;
 829
 830         for (i = 0; i < lm->smeta_sec; i++, paddr++) {
 831                 struct pblk_sec_meta *meta_list = rqd.meta_list;
 832
 833                 rqd.ppa_list[i] = addr_to_gen_ppa(pblk, paddr, line->id);
 834
 835                 if (dir == PBLK_WRITE) {
 836                         __le64 addr_empty = cpu_to_le64(ADDR_EMPTY);
 837
 838                         meta_list[i].lba = lba_list[paddr] = addr_empty;
 839                 }
 840         }
 841
 842         /*
 843          * This I/O is sent by the write thread when a line is replace. Since
 844          * the write thread is the only one sending write and erase commands,
 845          * there is no need to take the LUN semaphore.
 846          */
 847         ret = pblk_submit_io_sync(pblk, &rqd);
 848         if (ret) {
 849                 pblk_err(pblk, "smeta I/O submission failed: %d\n", ret);
 850                 bio_put(bio);
 851                 goto free_ppa_list;
 852         }
 853
 854         atomic_dec(&pblk->inflight_io);
 855
 856         if (rqd.error) {
 857                 if (dir == PBLK_WRITE) {
 858                         pblk_log_write_err(pblk, &rqd);
 859                         ret = 1;
 860                 } else if (dir == PBLK_READ)
 861                         pblk_log_read_err(pblk, &rqd);
 862         }
 863
 864 free_ppa_list:
 865         nvm_dev_dma_free(dev->parent, rqd.meta_list, rqd.dma_meta_list);
 866
 867         return ret;
 868 }
 869
 870 int pblk_line_read_smeta(struct pblk *pblk, struct pblk_line *line)
 871 {
 872         u64 bpaddr = pblk_line_smeta_start(pblk, line);
 873
 874         return pblk_line_submit_smeta_io(pblk, line, bpaddr, PBLK_READ_RECOV);
 875 }
 876
 877 int pblk_line_read_emeta(struct pblk *pblk, struct pblk_line *line,
 878                          void *emeta_buf)
 879 {
 880         return pblk_line_submit_emeta_io(pblk, line, emeta_buf,
 881                                                 line->emeta_ssec, PBLK_READ);
 882 }
 883
 884 static void pblk_setup_e_rq(struct pblk *pblk, struct nvm_rq *rqd,
 885                             struct ppa_addr ppa)
 886 {
 887         rqd->opcode = NVM_OP_ERASE;
 888         rqd->ppa_addr = ppa;
 889         rqd->nr_ppas = 1;
 890         rqd->flags = pblk_set_progr_mode(pblk, PBLK_ERASE);
 891         rqd->bio = NULL;
 892 }
 893
 894 static int pblk_blk_erase_sync(struct pblk *pblk, struct ppa_addr ppa)
 895 {
 896         struct nvm_rq rqd = {NULL};
 897         int ret;
 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         rqd.private = pblk;
 906         __pblk_end_io_erase(pblk, &rqd);
 907
 908         return ret;
 909 }
 910
 911 int pblk_line_erase(struct pblk *pblk, struct pblk_line *line)
 912 {
 913         struct pblk_line_meta *lm = &pblk->lm;
 914         struct ppa_addr ppa;
 915         int ret, bit = -1;
 916
 917         /* Erase only good blocks, one at a time */
 918         do {
 919                 spin_lock(&line->lock);
 920                 bit = find_next_zero_bit(line->erase_bitmap, lm->blk_per_line,
 921                                                                 bit + 1);
 922                 if (bit >= lm->blk_per_line) {
 923                         spin_unlock(&line->lock);
 924                         break;
 925                 }
 926
 927                 ppa = pblk->luns[bit].bppa; /* set ch and lun */
 928                 ppa.a.blk = line->id;
 929
 930                 atomic_dec(&line->left_eblks);
 931                 WARN_ON(test_and_set_bit(bit, line->erase_bitmap));
 932                 spin_unlock(&line->lock);
 933
 934                 ret = pblk_blk_erase_sync(pblk, ppa);
 935                 if (ret) {
 936                         pblk_err(pblk, "failed to erase line %d\n", line->id);
 937                         return ret;
 938                 }
 939         } while (1);
 940
 941         return 0;
 942 }
 943
 944 static void pblk_line_setup_metadata(struct pblk_line *line,
 945                                      struct pblk_line_mgmt *l_mg,
 946                                      struct pblk_line_meta *lm)
 947 {
 948         int meta_line;
 949
 950         lockdep_assert_held(&l_mg->free_lock);
 951
 952 retry_meta:
 953         meta_line = find_first_zero_bit(&l_mg->meta_bitmap, PBLK_DATA_LINES);
 954         if (meta_line == PBLK_DATA_LINES) {
 955                 spin_unlock(&l_mg->free_lock);
 956                 io_schedule();
 957                 spin_lock(&l_mg->free_lock);
 958                 goto retry_meta;
 959         }
 960
 961         set_bit(meta_line, &l_mg->meta_bitmap);
 962         line->meta_line = meta_line;
 963
 964         line->smeta = l_mg->sline_meta[meta_line];
 965         line->emeta = l_mg->eline_meta[meta_line];
 966
 967         memset(line->smeta, 0, lm->smeta_len);
 968         memset(line->emeta->buf, 0, lm->emeta_len[0]);
 969
 970         line->emeta->mem = 0;
 971         atomic_set(&line->emeta->sync, 0);
 972 }
 973
 974 /* For now lines are always assumed full lines. Thus, smeta former and current
 975  * lun bitmaps are omitted.
 976  */
 977 static int pblk_line_init_metadata(struct pblk *pblk, struct pblk_line *line,
 978                                   struct pblk_line *cur)
 979 {
 980         struct nvm_tgt_dev *dev = pblk->dev;
 981         struct nvm_geo *geo = &dev->geo;
 982         struct pblk_line_meta *lm = &pblk->lm;
 983         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
 984         struct pblk_emeta *emeta = line->emeta;
 985         struct line_emeta *emeta_buf = emeta->buf;
 986         struct line_smeta *smeta_buf = (struct line_smeta *)line->smeta;
 987         int nr_blk_line;
 988
 989         /* After erasing the line, new bad blocks might appear and we risk
 990          * having an invalid line
 991          */
 992         nr_blk_line = lm->blk_per_line -
 993                         bitmap_weight(line->blk_bitmap, lm->blk_per_line);
 994         if (nr_blk_line < lm->min_blk_line) {
 995                 spin_lock(&l_mg->free_lock);
 996                 spin_lock(&line->lock);
 997                 line->state = PBLK_LINESTATE_BAD;
 998                 spin_unlock(&line->lock);
 999
1000                 list_add_tail(&line->list, &l_mg->bad_list);
1001                 spin_unlock(&l_mg->free_lock);
1002
1003                 pblk_debug(pblk, "line %d is bad\n", line->id);
1004
1005                 return 0;
1006         }
1007
1008         /* Run-time metadata */
1009         line->lun_bitmap = ((void *)(smeta_buf)) + sizeof(struct line_smeta);
1010
1011         /* Mark LUNs allocated in this line (all for now) */
1012         bitmap_set(line->lun_bitmap, 0, lm->lun_bitmap_len);
1013
1014         smeta_buf->header.identifier = cpu_to_le32(PBLK_MAGIC);
1015         memcpy(smeta_buf->header.uuid, pblk->instance_uuid, 16);
1016         smeta_buf->header.id = cpu_to_le32(line->id);
1017         smeta_buf->header.type = cpu_to_le16(line->type);
1018         smeta_buf->header.version_major = SMETA_VERSION_MAJOR;
1019         smeta_buf->header.version_minor = SMETA_VERSION_MINOR;
1020
1021         /* Start metadata */
1022         smeta_buf->seq_nr = cpu_to_le64(line->seq_nr);
1023         smeta_buf->window_wr_lun = cpu_to_le32(geo->all_luns);
1024
1025         /* Fill metadata among lines */
1026         if (cur) {
1027                 memcpy(line->lun_bitmap, cur->lun_bitmap, lm->lun_bitmap_len);
1028                 smeta_buf->prev_id = cpu_to_le32(cur->id);
1029                 cur->emeta->buf->next_id = cpu_to_le32(line->id);
1030         } else {
1031                 smeta_buf->prev_id = cpu_to_le32(PBLK_LINE_EMPTY);
1032         }
1033
1034         /* All smeta must be set at this point */
1035         smeta_buf->header.crc = cpu_to_le32(
1036                         pblk_calc_meta_header_crc(pblk, &smeta_buf->header));
1037         smeta_buf->crc = cpu_to_le32(pblk_calc_smeta_crc(pblk, smeta_buf));
1038
1039         /* End metadata */
1040         memcpy(&emeta_buf->header, &smeta_buf->header,
1041                                                 sizeof(struct line_header));
1042
1043         emeta_buf->header.version_major = EMETA_VERSION_MAJOR;
1044         emeta_buf->header.version_minor = EMETA_VERSION_MINOR;
1045         emeta_buf->header.crc = cpu_to_le32(
1046                         pblk_calc_meta_header_crc(pblk, &emeta_buf->header));
1047
1048         emeta_buf->seq_nr = cpu_to_le64(line->seq_nr);
1049         emeta_buf->nr_lbas = cpu_to_le64(line->sec_in_line);
1050         emeta_buf->nr_valid_lbas = cpu_to_le64(0);
1051         emeta_buf->next_id = cpu_to_le32(PBLK_LINE_EMPTY);
1052         emeta_buf->crc = cpu_to_le32(0);
1053         emeta_buf->prev_id = smeta_buf->prev_id;
1054
1055         return 1;
1056 }
1057
1058 static int pblk_line_alloc_bitmaps(struct pblk *pblk, struct pblk_line *line)
1059 {
1060         struct pblk_line_meta *lm = &pblk->lm;
1061
1062         line->map_bitmap = kzalloc(lm->sec_bitmap_len, GFP_KERNEL);
1063         if (!line->map_bitmap)
1064                 return -ENOMEM;
1065
1066         /* will be initialized using bb info from map_bitmap */
1067         line->invalid_bitmap = kmalloc(lm->sec_bitmap_len, GFP_KERNEL);
1068         if (!line->invalid_bitmap) {
1069                 kfree(line->map_bitmap);
1070                 line->map_bitmap = NULL;
1071                 return -ENOMEM;
1072         }
1073
1074         return 0;
1075 }
1076
1077 /* For now lines are always assumed full lines. Thus, smeta former and current
1078  * lun bitmaps are omitted.
1079  */
1080 static int pblk_line_init_bb(struct pblk *pblk, struct pblk_line *line,
1081                              int init)
1082 {
1083         struct nvm_tgt_dev *dev = pblk->dev;
1084         struct nvm_geo *geo = &dev->geo;
1085         struct pblk_line_meta *lm = &pblk->lm;
1086         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1087         u64 off;
1088         int bit = -1;
1089         int emeta_secs;
1090
1091         line->sec_in_line = lm->sec_per_line;
1092
1093         /* Capture bad block information on line mapping bitmaps */
1094         while ((bit = find_next_bit(line->blk_bitmap, lm->blk_per_line,
1095                                         bit + 1)) < lm->blk_per_line) {
1096                 off = bit * geo->ws_opt;
1097                 bitmap_shift_left(l_mg->bb_aux, l_mg->bb_template, off,
1098                                                         lm->sec_per_line);
1099                 bitmap_or(line->map_bitmap, line->map_bitmap, l_mg->bb_aux,
1100                                                         lm->sec_per_line);
1101                 line->sec_in_line -= geo->clba;
1102         }
1103
1104         /* Mark smeta metadata sectors as bad sectors */
1105         bit = find_first_zero_bit(line->blk_bitmap, lm->blk_per_line);
1106         off = bit * geo->ws_opt;
1107         bitmap_set(line->map_bitmap, off, lm->smeta_sec);
1108         line->sec_in_line -= lm->smeta_sec;
1109         line->smeta_ssec = off;
1110         line->cur_sec = off + lm->smeta_sec;
1111
1112         if (init && pblk_line_submit_smeta_io(pblk, line, off, PBLK_WRITE)) {
1113                 pblk_debug(pblk, "line smeta I/O failed. Retry\n");
1114                 return 0;
1115         }
1116
1117         bitmap_copy(line->invalid_bitmap, line->map_bitmap, lm->sec_per_line);
1118
1119         /* Mark emeta metadata sectors as bad sectors. We need to consider bad
1120          * blocks to make sure that there are enough sectors to store emeta
1121          */
1122         emeta_secs = lm->emeta_sec[0];
1123         off = lm->sec_per_line;
1124         while (emeta_secs) {
1125                 off -= geo->ws_opt;
1126                 if (!test_bit(off, line->invalid_bitmap)) {
1127                         bitmap_set(line->invalid_bitmap, off, geo->ws_opt);
1128                         emeta_secs -= geo->ws_opt;
1129                 }
1130         }
1131
1132         line->emeta_ssec = off;
1133         line->sec_in_line -= lm->emeta_sec[0];
1134         line->nr_valid_lbas = 0;
1135         line->left_msecs = line->sec_in_line;
1136         *line->vsc = cpu_to_le32(line->sec_in_line);
1137
1138         if (lm->sec_per_line - line->sec_in_line !=
1139                 bitmap_weight(line->invalid_bitmap, lm->sec_per_line)) {
1140                 spin_lock(&line->lock);
1141                 line->state = PBLK_LINESTATE_BAD;
1142                 spin_unlock(&line->lock);
1143
1144                 list_add_tail(&line->list, &l_mg->bad_list);
1145                 pblk_err(pblk, "unexpected line %d is bad\n", line->id);
1146
1147                 return 0;
1148         }
1149
1150         return 1;
1151 }
1152
1153 static int pblk_prepare_new_line(struct pblk *pblk, struct pblk_line *line)
1154 {
1155         struct pblk_line_meta *lm = &pblk->lm;
1156         struct nvm_tgt_dev *dev = pblk->dev;
1157         struct nvm_geo *geo = &dev->geo;
1158         int blk_to_erase = atomic_read(&line->blk_in_line);
1159         int i;
1160
1161         for (i = 0; i < lm->blk_per_line; i++) {
1162                 struct pblk_lun *rlun = &pblk->luns[i];
1163                 int pos = pblk_ppa_to_pos(geo, rlun->bppa);
1164                 int state = line->chks[pos].state;
1165
1166                 /* Free chunks should not be erased */
1167                 if (state & NVM_CHK_ST_FREE) {
1168                         set_bit(pblk_ppa_to_pos(geo, rlun->bppa),
1169                                                         line->erase_bitmap);
1170                         blk_to_erase--;
1171                 }
1172         }
1173
1174         return blk_to_erase;
1175 }
1176
1177 static int pblk_line_prepare(struct pblk *pblk, struct pblk_line *line)
1178 {
1179         struct pblk_line_meta *lm = &pblk->lm;
1180         int blk_in_line = atomic_read(&line->blk_in_line);
1181         int blk_to_erase;
1182
1183         /* Bad blocks do not need to be erased */
1184         bitmap_copy(line->erase_bitmap, line->blk_bitmap, lm->blk_per_line);
1185
1186         spin_lock(&line->lock);
1187
1188         /* If we have not written to this line, we need to mark up free chunks
1189          * as already erased
1190          */
1191         if (line->state == PBLK_LINESTATE_NEW) {
1192                 blk_to_erase = pblk_prepare_new_line(pblk, line);
1193                 line->state = PBLK_LINESTATE_FREE;
1194         } else {
1195                 blk_to_erase = blk_in_line;
1196         }
1197
1198         if (blk_in_line < lm->min_blk_line) {
1199                 spin_unlock(&line->lock);
1200                 return -EAGAIN;
1201         }
1202
1203         if (line->state != PBLK_LINESTATE_FREE) {
1204                 WARN(1, "pblk: corrupted line %d, state %d\n",
1205                                                         line->id, line->state);
1206                 spin_unlock(&line->lock);
1207                 return -EINTR;
1208         }
1209
1210         line->state = PBLK_LINESTATE_OPEN;
1211
1212         atomic_set(&line->left_eblks, blk_to_erase);
1213         atomic_set(&line->left_seblks, blk_to_erase);
1214
1215         line->meta_distance = lm->meta_distance;
1216         spin_unlock(&line->lock);
1217
1218         kref_init(&line->ref);
1219
1220         return 0;
1221 }
1222
1223 int pblk_line_recov_alloc(struct pblk *pblk, struct pblk_line *line)
1224 {
1225         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1226         int ret;
1227
1228         spin_lock(&l_mg->free_lock);
1229         l_mg->data_line = line;
1230         list_del(&line->list);
1231
1232         ret = pblk_line_prepare(pblk, line);
1233         if (ret) {
1234                 list_add(&line->list, &l_mg->free_list);
1235                 spin_unlock(&l_mg->free_lock);
1236                 return ret;
1237         }
1238         spin_unlock(&l_mg->free_lock);
1239
1240         ret = pblk_line_alloc_bitmaps(pblk, line);
1241         if (ret)
1242                 goto fail;
1243
1244         if (!pblk_line_init_bb(pblk, line, 0)) {
1245                 ret = -EINTR;
1246                 goto fail;
1247         }
1248
1249         pblk_rl_free_lines_dec(&pblk->rl, line, true);
1250         return 0;
1251
1252 fail:
1253         spin_lock(&l_mg->free_lock);
1254         list_add(&line->list, &l_mg->free_list);
1255         spin_unlock(&l_mg->free_lock);
1256
1257         return ret;
1258 }
1259
1260 void pblk_line_recov_close(struct pblk *pblk, struct pblk_line *line)
1261 {
1262         kfree(line->map_bitmap);
1263         line->map_bitmap = NULL;
1264         line->smeta = NULL;
1265         line->emeta = NULL;
1266 }
1267
1268 static void pblk_line_reinit(struct pblk_line *line)
1269 {
1270         *line->vsc = cpu_to_le32(EMPTY_ENTRY);
1271
1272         line->map_bitmap = NULL;
1273         line->invalid_bitmap = NULL;
1274         line->smeta = NULL;
1275         line->emeta = NULL;
1276 }
1277
1278 void pblk_line_free(struct pblk_line *line)
1279 {
1280         kfree(line->map_bitmap);
1281         kfree(line->invalid_bitmap);
1282
1283         pblk_line_reinit(line);
1284 }
1285
1286 struct pblk_line *pblk_line_get(struct pblk *pblk)
1287 {
1288         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1289         struct pblk_line_meta *lm = &pblk->lm;
1290         struct pblk_line *line;
1291         int ret, bit;
1292
1293         lockdep_assert_held(&l_mg->free_lock);
1294
1295 retry:
1296         if (list_empty(&l_mg->free_list)) {
1297                 pblk_err(pblk, "no free lines\n");
1298                 return NULL;
1299         }
1300
1301         line = list_first_entry(&l_mg->free_list, struct pblk_line, list);
1302         list_del(&line->list);
1303         l_mg->nr_free_lines--;
1304
1305         bit = find_first_zero_bit(line->blk_bitmap, lm->blk_per_line);
1306         if (unlikely(bit >= lm->blk_per_line)) {
1307                 spin_lock(&line->lock);
1308                 line->state = PBLK_LINESTATE_BAD;
1309                 spin_unlock(&line->lock);
1310
1311                 list_add_tail(&line->list, &l_mg->bad_list);
1312
1313                 pblk_debug(pblk, "line %d is bad\n", line->id);
1314                 goto retry;
1315         }
1316
1317         ret = pblk_line_prepare(pblk, line);
1318         if (ret) {
1319                 switch (ret) {
1320                 case -EAGAIN:
1321                         list_add(&line->list, &l_mg->bad_list);
1322                         goto retry;
1323                 case -EINTR:
1324                         list_add(&line->list, &l_mg->corrupt_list);
1325                         goto retry;
1326                 default:
1327                         pblk_err(pblk, "failed to prepare line %d\n", line->id);
1328                         list_add(&line->list, &l_mg->free_list);
1329                         l_mg->nr_free_lines++;
1330                         return NULL;
1331                 }
1332         }
1333
1334         return line;
1335 }
1336
1337 static struct pblk_line *pblk_line_retry(struct pblk *pblk,
1338                                          struct pblk_line *line)
1339 {
1340         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1341         struct pblk_line *retry_line;
1342
1343 retry:
1344         spin_lock(&l_mg->free_lock);
1345         retry_line = pblk_line_get(pblk);
1346         if (!retry_line) {
1347                 l_mg->data_line = NULL;
1348                 spin_unlock(&l_mg->free_lock);
1349                 return NULL;
1350         }
1351
1352         retry_line->map_bitmap = line->map_bitmap;
1353         retry_line->invalid_bitmap = line->invalid_bitmap;
1354         retry_line->smeta = line->smeta;
1355         retry_line->emeta = line->emeta;
1356         retry_line->meta_line = line->meta_line;
1357
1358         pblk_line_reinit(line);
1359
1360         l_mg->data_line = retry_line;
1361         spin_unlock(&l_mg->free_lock);
1362
1363         pblk_rl_free_lines_dec(&pblk->rl, line, false);
1364
1365         if (pblk_line_erase(pblk, retry_line))
1366                 goto retry;
1367
1368         return retry_line;
1369 }
1370
1371 static void pblk_set_space_limit(struct pblk *pblk)
1372 {
1373         struct pblk_rl *rl = &pblk->rl;
1374
1375         atomic_set(&rl->rb_space, 0);
1376 }
1377
1378 struct pblk_line *pblk_line_get_first_data(struct pblk *pblk)
1379 {
1380         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1381         struct pblk_line *line;
1382
1383         spin_lock(&l_mg->free_lock);
1384         line = pblk_line_get(pblk);
1385         if (!line) {
1386                 spin_unlock(&l_mg->free_lock);
1387                 return NULL;
1388         }
1389
1390         line->seq_nr = l_mg->d_seq_nr++;
1391         line->type = PBLK_LINETYPE_DATA;
1392         l_mg->data_line = line;
1393
1394         pblk_line_setup_metadata(line, l_mg, &pblk->lm);
1395
1396         /* Allocate next line for preparation */
1397         l_mg->data_next = pblk_line_get(pblk);
1398         if (!l_mg->data_next) {
1399                 /* If we cannot get a new line, we need to stop the pipeline.
1400                  * Only allow as many writes in as we can store safely and then
1401                  * fail gracefully
1402                  */
1403                 pblk_set_space_limit(pblk);
1404
1405                 l_mg->data_next = NULL;
1406         } else {
1407                 l_mg->data_next->seq_nr = l_mg->d_seq_nr++;
1408                 l_mg->data_next->type = PBLK_LINETYPE_DATA;
1409         }
1410         spin_unlock(&l_mg->free_lock);
1411
1412         if (pblk_line_alloc_bitmaps(pblk, line))
1413                 return NULL;
1414
1415         if (pblk_line_erase(pblk, line)) {
1416                 line = pblk_line_retry(pblk, line);
1417                 if (!line)
1418                         return NULL;
1419         }
1420
1421 retry_setup:
1422         if (!pblk_line_init_metadata(pblk, line, NULL)) {
1423                 line = pblk_line_retry(pblk, line);
1424                 if (!line)
1425                         return NULL;
1426
1427                 goto retry_setup;
1428         }
1429
1430         if (!pblk_line_init_bb(pblk, line, 1)) {
1431                 line = pblk_line_retry(pblk, line);
1432                 if (!line)
1433                         return NULL;
1434
1435                 goto retry_setup;
1436         }
1437
1438         pblk_rl_free_lines_dec(&pblk->rl, line, true);
1439
1440         return line;
1441 }
1442
1443 static void pblk_stop_writes(struct pblk *pblk, struct pblk_line *line)
1444 {
1445         lockdep_assert_held(&pblk->l_mg.free_lock);
1446
1447         pblk_set_space_limit(pblk);
1448         pblk->state = PBLK_STATE_STOPPING;
1449 }
1450
1451 static void pblk_line_close_meta_sync(struct pblk *pblk)
1452 {
1453         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1454         struct pblk_line_meta *lm = &pblk->lm;
1455         struct pblk_line *line, *tline;
1456         LIST_HEAD(list);
1457
1458         spin_lock(&l_mg->close_lock);
1459         if (list_empty(&l_mg->emeta_list)) {
1460                 spin_unlock(&l_mg->close_lock);
1461                 return;
1462         }
1463
1464         list_cut_position(&list, &l_mg->emeta_list, l_mg->emeta_list.prev);
1465         spin_unlock(&l_mg->close_lock);
1466
1467         list_for_each_entry_safe(line, tline, &list, list) {
1468                 struct pblk_emeta *emeta = line->emeta;
1469
1470                 while (emeta->mem < lm->emeta_len[0]) {
1471                         int ret;
1472
1473                         ret = pblk_submit_meta_io(pblk, line);
1474                         if (ret) {
1475                                 pblk_err(pblk, "sync meta line %d failed (%d)\n",
1476                                                         line->id, ret);
1477                                 return;
1478                         }
1479                 }
1480         }
1481
1482         pblk_wait_for_meta(pblk);
1483         flush_workqueue(pblk->close_wq);
1484 }
1485
1486 void __pblk_pipeline_flush(struct pblk *pblk)
1487 {
1488         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1489         int ret;
1490
1491         spin_lock(&l_mg->free_lock);
1492         if (pblk->state == PBLK_STATE_RECOVERING ||
1493                                         pblk->state == PBLK_STATE_STOPPED) {
1494                 spin_unlock(&l_mg->free_lock);
1495                 return;
1496         }
1497         pblk->state = PBLK_STATE_RECOVERING;
1498         spin_unlock(&l_mg->free_lock);
1499
1500         pblk_flush_writer(pblk);
1501         pblk_wait_for_meta(pblk);
1502
1503         ret = pblk_recov_pad(pblk);
1504         if (ret) {
1505                 pblk_err(pblk, "could not close data on teardown(%d)\n", ret);
1506                 return;
1507         }
1508
1509         flush_workqueue(pblk->bb_wq);
1510         pblk_line_close_meta_sync(pblk);
1511 }
1512
1513 void __pblk_pipeline_stop(struct pblk *pblk)
1514 {
1515         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1516
1517         spin_lock(&l_mg->free_lock);
1518         pblk->state = PBLK_STATE_STOPPED;
1519         l_mg->data_line = NULL;
1520         l_mg->data_next = NULL;
1521         spin_unlock(&l_mg->free_lock);
1522 }
1523
1524 void pblk_pipeline_stop(struct pblk *pblk)
1525 {
1526         __pblk_pipeline_flush(pblk);
1527         __pblk_pipeline_stop(pblk);
1528 }
1529
1530 struct pblk_line *pblk_line_replace_data(struct pblk *pblk)
1531 {
1532         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1533         struct pblk_line *cur, *new = NULL;
1534         unsigned int left_seblks;
1535
1536         new = l_mg->data_next;
1537         if (!new)
1538                 goto out;
1539
1540         spin_lock(&l_mg->free_lock);
1541         cur = l_mg->data_line;
1542         l_mg->data_line = new;
1543
1544         pblk_line_setup_metadata(new, l_mg, &pblk->lm);
1545         spin_unlock(&l_mg->free_lock);
1546
1547 retry_erase:
1548         left_seblks = atomic_read(&new->left_seblks);
1549         if (left_seblks) {
1550                 /* If line is not fully erased, erase it */
1551                 if (atomic_read(&new->left_eblks)) {
1552                         if (pblk_line_erase(pblk, new))
1553                                 goto out;
1554                 } else {
1555                         io_schedule();
1556                 }
1557                 goto retry_erase;
1558         }
1559
1560         if (pblk_line_alloc_bitmaps(pblk, new))
1561                 return NULL;
1562
1563 retry_setup:
1564         if (!pblk_line_init_metadata(pblk, new, cur)) {
1565                 new = pblk_line_retry(pblk, new);
1566                 if (!new)
1567                         goto out;
1568
1569                 goto retry_setup;
1570         }
1571
1572         if (!pblk_line_init_bb(pblk, new, 1)) {
1573                 new = pblk_line_retry(pblk, new);
1574                 if (!new)
1575                         goto out;
1576
1577                 goto retry_setup;
1578         }
1579
1580         pblk_rl_free_lines_dec(&pblk->rl, new, true);
1581
1582         /* Allocate next line for preparation */
1583         spin_lock(&l_mg->free_lock);
1584         l_mg->data_next = pblk_line_get(pblk);
1585         if (!l_mg->data_next) {
1586                 /* If we cannot get a new line, we need to stop the pipeline.
1587                  * Only allow as many writes in as we can store safely and then
1588                  * fail gracefully
1589                  */
1590                 pblk_stop_writes(pblk, new);
1591                 l_mg->data_next = NULL;
1592         } else {
1593                 l_mg->data_next->seq_nr = l_mg->d_seq_nr++;
1594                 l_mg->data_next->type = PBLK_LINETYPE_DATA;
1595         }
1596         spin_unlock(&l_mg->free_lock);
1597
1598 out:
1599         return new;
1600 }
1601
1602 static void __pblk_line_put(struct pblk *pblk, struct pblk_line *line)
1603 {
1604         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1605         struct pblk_gc *gc = &pblk->gc;
1606
1607         spin_lock(&line->lock);
1608         WARN_ON(line->state != PBLK_LINESTATE_GC);
1609         line->state = PBLK_LINESTATE_FREE;
1610         line->gc_group = PBLK_LINEGC_NONE;
1611         pblk_line_free(line);
1612
1613         if (line->w_err_gc->has_write_err) {
1614                 pblk_rl_werr_line_out(&pblk->rl);
1615                 line->w_err_gc->has_write_err = 0;
1616         }
1617
1618         spin_unlock(&line->lock);
1619         atomic_dec(&gc->pipeline_gc);
1620
1621         spin_lock(&l_mg->free_lock);
1622         list_add_tail(&line->list, &l_mg->free_list);
1623         l_mg->nr_free_lines++;
1624         spin_unlock(&l_mg->free_lock);
1625
1626         pblk_rl_free_lines_inc(&pblk->rl, line);
1627 }
1628
1629 static void pblk_line_put_ws(struct work_struct *work)
1630 {
1631         struct pblk_line_ws *line_put_ws = container_of(work,
1632                                                 struct pblk_line_ws, ws);
1633         struct pblk *pblk = line_put_ws->pblk;
1634         struct pblk_line *line = line_put_ws->line;
1635
1636         __pblk_line_put(pblk, line);
1637         mempool_free(line_put_ws, &pblk->gen_ws_pool);
1638 }
1639
1640 void pblk_line_put(struct kref *ref)
1641 {
1642         struct pblk_line *line = container_of(ref, struct pblk_line, ref);
1643         struct pblk *pblk = line->pblk;
1644
1645         __pblk_line_put(pblk, line);
1646 }
1647
1648 void pblk_line_put_wq(struct kref *ref)
1649 {
1650         struct pblk_line *line = container_of(ref, struct pblk_line, ref);
1651         struct pblk *pblk = line->pblk;
1652         struct pblk_line_ws *line_put_ws;
1653
1654         line_put_ws = mempool_alloc(&pblk->gen_ws_pool, GFP_ATOMIC);
1655         if (!line_put_ws)
1656                 return;
1657
1658         line_put_ws->pblk = pblk;
1659         line_put_ws->line = line;
1660         line_put_ws->priv = NULL;
1661
1662         INIT_WORK(&line_put_ws->ws, pblk_line_put_ws);
1663         queue_work(pblk->r_end_wq, &line_put_ws->ws);
1664 }
1665
1666 int pblk_blk_erase_async(struct pblk *pblk, struct ppa_addr ppa)
1667 {
1668         struct nvm_rq *rqd;
1669         int err;
1670
1671         rqd = pblk_alloc_rqd(pblk, PBLK_ERASE);
1672
1673         pblk_setup_e_rq(pblk, rqd, ppa);
1674
1675         rqd->end_io = pblk_end_io_erase;
1676         rqd->private = pblk;
1677
1678         /* The write thread schedules erases so that it minimizes disturbances
1679          * with writes. Thus, there is no need to take the LUN semaphore.
1680          */
1681         err = pblk_submit_io(pblk, rqd);
1682         if (err) {
1683                 struct nvm_tgt_dev *dev = pblk->dev;
1684                 struct nvm_geo *geo = &dev->geo;
1685
1686                 pblk_err(pblk, "could not async erase line:%d,blk:%d\n",
1687                                         pblk_ppa_to_line(ppa),
1688                                         pblk_ppa_to_pos(geo, ppa));
1689         }
1690
1691         return err;
1692 }
1693
1694 struct pblk_line *pblk_line_get_data(struct pblk *pblk)
1695 {
1696         return pblk->l_mg.data_line;
1697 }
1698
1699 /* For now, always erase next line */
1700 struct pblk_line *pblk_line_get_erase(struct pblk *pblk)
1701 {
1702         return pblk->l_mg.data_next;
1703 }
1704
1705 int pblk_line_is_full(struct pblk_line *line)
1706 {
1707         return (line->left_msecs == 0);
1708 }
1709
1710 static void pblk_line_should_sync_meta(struct pblk *pblk)
1711 {
1712         if (pblk_rl_is_limit(&pblk->rl))
1713                 pblk_line_close_meta_sync(pblk);
1714 }
1715
1716 void pblk_line_close(struct pblk *pblk, struct pblk_line *line)
1717 {
1718         struct nvm_tgt_dev *dev = pblk->dev;
1719         struct nvm_geo *geo = &dev->geo;
1720         struct pblk_line_meta *lm = &pblk->lm;
1721         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1722         struct list_head *move_list;
1723         int i;
1724
1725 #ifdef CONFIG_NVM_PBLK_DEBUG
1726         WARN(!bitmap_full(line->map_bitmap, lm->sec_per_line),
1727                                 "pblk: corrupt closed line %d\n", line->id);
1728 #endif
1729
1730         spin_lock(&l_mg->free_lock);
1731         WARN_ON(!test_and_clear_bit(line->meta_line, &l_mg->meta_bitmap));
1732         spin_unlock(&l_mg->free_lock);
1733
1734         spin_lock(&l_mg->gc_lock);
1735         spin_lock(&line->lock);
1736         WARN_ON(line->state != PBLK_LINESTATE_OPEN);
1737         line->state = PBLK_LINESTATE_CLOSED;
1738         move_list = pblk_line_gc_list(pblk, line);
1739
1740         list_add_tail(&line->list, move_list);
1741
1742         kfree(line->map_bitmap);
1743         line->map_bitmap = NULL;
1744         line->smeta = NULL;
1745         line->emeta = NULL;
1746
1747         for (i = 0; i < lm->blk_per_line; i++) {
1748                 struct pblk_lun *rlun = &pblk->luns[i];
1749                 int pos = pblk_ppa_to_pos(geo, rlun->bppa);
1750                 int state = line->chks[pos].state;
1751
1752                 if (!(state & NVM_CHK_ST_OFFLINE))
1753                         state = NVM_CHK_ST_CLOSED;
1754         }
1755
1756         spin_unlock(&line->lock);
1757         spin_unlock(&l_mg->gc_lock);
1758 }
1759
1760 void pblk_line_close_meta(struct pblk *pblk, struct pblk_line *line)
1761 {
1762         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1763         struct pblk_line_meta *lm = &pblk->lm;
1764         struct pblk_emeta *emeta = line->emeta;
1765         struct line_emeta *emeta_buf = emeta->buf;
1766         struct wa_counters *wa = emeta_to_wa(lm, emeta_buf);
1767
1768         /* No need for exact vsc value; avoid a big line lock and take aprox. */
1769         memcpy(emeta_to_vsc(pblk, emeta_buf), l_mg->vsc_list, lm->vsc_list_len);
1770         memcpy(emeta_to_bb(emeta_buf), line->blk_bitmap, lm->blk_bitmap_len);
1771
1772         wa->user = cpu_to_le64(atomic64_read(&pblk->user_wa));
1773         wa->pad = cpu_to_le64(atomic64_read(&pblk->pad_wa));
1774         wa->gc = cpu_to_le64(atomic64_read(&pblk->gc_wa));
1775
1776         if (le32_to_cpu(emeta_buf->header.identifier) != PBLK_MAGIC) {
1777                 emeta_buf->header.identifier = cpu_to_le32(PBLK_MAGIC);
1778                 memcpy(emeta_buf->header.uuid, pblk->instance_uuid, 16);
1779                 emeta_buf->header.id = cpu_to_le32(line->id);
1780                 emeta_buf->header.type = cpu_to_le16(line->type);
1781                 emeta_buf->header.version_major = EMETA_VERSION_MAJOR;
1782                 emeta_buf->header.version_minor = EMETA_VERSION_MINOR;
1783                 emeta_buf->header.crc = cpu_to_le32(
1784                         pblk_calc_meta_header_crc(pblk, &emeta_buf->header));
1785         }
1786
1787         emeta_buf->nr_valid_lbas = cpu_to_le64(line->nr_valid_lbas);
1788         emeta_buf->crc = cpu_to_le32(pblk_calc_emeta_crc(pblk, emeta_buf));
1789
1790         spin_lock(&l_mg->close_lock);
1791         spin_lock(&line->lock);
1792
1793         /* Update the in-memory start address for emeta, in case it has
1794          * shifted due to write errors
1795          */
1796         if (line->emeta_ssec != line->cur_sec)
1797                 line->emeta_ssec = line->cur_sec;
1798
1799         list_add_tail(&line->list, &l_mg->emeta_list);
1800         spin_unlock(&line->lock);
1801         spin_unlock(&l_mg->close_lock);
1802
1803         pblk_line_should_sync_meta(pblk);
1804 }
1805
1806 static void pblk_save_lba_list(struct pblk *pblk, struct pblk_line *line)
1807 {
1808         struct pblk_line_meta *lm = &pblk->lm;
1809         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1810         unsigned int lba_list_size = lm->emeta_len[2];
1811         struct pblk_w_err_gc *w_err_gc = line->w_err_gc;
1812         struct pblk_emeta *emeta = line->emeta;
1813
1814         w_err_gc->lba_list = pblk_malloc(lba_list_size,
1815                                          l_mg->emeta_alloc_type, GFP_KERNEL);
1816         memcpy(w_err_gc->lba_list, emeta_to_lbas(pblk, emeta->buf),
1817                                 lba_list_size);
1818 }
1819
1820 void pblk_line_close_ws(struct work_struct *work)
1821 {
1822         struct pblk_line_ws *line_ws = container_of(work, struct pblk_line_ws,
1823                                                                         ws);
1824         struct pblk *pblk = line_ws->pblk;
1825         struct pblk_line *line = line_ws->line;
1826         struct pblk_w_err_gc *w_err_gc = line->w_err_gc;
1827
1828         /* Write errors makes the emeta start address stored in smeta invalid,
1829          * so keep a copy of the lba list until we've gc'd the line
1830          */
1831         if (w_err_gc->has_write_err)
1832                 pblk_save_lba_list(pblk, line);
1833
1834         pblk_line_close(pblk, line);
1835         mempool_free(line_ws, &pblk->gen_ws_pool);
1836 }
1837
1838 void pblk_gen_run_ws(struct pblk *pblk, struct pblk_line *line, void *priv,
1839                       void (*work)(struct work_struct *), gfp_t gfp_mask,
1840                       struct workqueue_struct *wq)
1841 {
1842         struct pblk_line_ws *line_ws;
1843
1844         line_ws = mempool_alloc(&pblk->gen_ws_pool, gfp_mask);
1845
1846         line_ws->pblk = pblk;
1847         line_ws->line = line;
1848         line_ws->priv = priv;
1849
1850         INIT_WORK(&line_ws->ws, work);
1851         queue_work(wq, &line_ws->ws);
1852 }
1853
1854 static void __pblk_down_page(struct pblk *pblk, struct ppa_addr *ppa_list,
1855                              int nr_ppas, int pos)
1856 {
1857         struct pblk_lun *rlun = &pblk->luns[pos];
1858         int ret;
1859
1860         /*
1861          * Only send one inflight I/O per LUN. Since we map at a page
1862          * granurality, all ppas in the I/O will map to the same LUN
1863          */
1864 #ifdef CONFIG_NVM_PBLK_DEBUG
1865         int i;
1866
1867         for (i = 1; i < nr_ppas; i++)
1868                 WARN_ON(ppa_list[0].a.lun != ppa_list[i].a.lun ||
1869                                 ppa_list[0].a.ch != ppa_list[i].a.ch);
1870 #endif
1871
1872         ret = down_timeout(&rlun->wr_sem, msecs_to_jiffies(30000));
1873         if (ret == -ETIME || ret == -EINTR)
1874                 pblk_err(pblk, "taking lun semaphore timed out: err %d\n",
1875                                 -ret);
1876 }
1877
1878 void pblk_down_page(struct pblk *pblk, struct ppa_addr *ppa_list, int nr_ppas)
1879 {
1880         struct nvm_tgt_dev *dev = pblk->dev;
1881         struct nvm_geo *geo = &dev->geo;
1882         int pos = pblk_ppa_to_pos(geo, ppa_list[0]);
1883
1884         __pblk_down_page(pblk, ppa_list, nr_ppas, pos);
1885 }
1886
1887 void pblk_down_rq(struct pblk *pblk, struct ppa_addr *ppa_list, int nr_ppas,
1888                   unsigned long *lun_bitmap)
1889 {
1890         struct nvm_tgt_dev *dev = pblk->dev;
1891         struct nvm_geo *geo = &dev->geo;
1892         int pos = pblk_ppa_to_pos(geo, ppa_list[0]);
1893
1894         /* If the LUN has been locked for this same request, do no attempt to
1895          * lock it again
1896          */
1897         if (test_and_set_bit(pos, lun_bitmap))
1898                 return;
1899
1900         __pblk_down_page(pblk, ppa_list, nr_ppas, pos);
1901 }
1902
1903 void pblk_up_page(struct pblk *pblk, struct ppa_addr *ppa_list, int nr_ppas)
1904 {
1905         struct nvm_tgt_dev *dev = pblk->dev;
1906         struct nvm_geo *geo = &dev->geo;
1907         struct pblk_lun *rlun;
1908         int pos = pblk_ppa_to_pos(geo, ppa_list[0]);
1909
1910 #ifdef CONFIG_NVM_PBLK_DEBUG
1911         int i;
1912
1913         for (i = 1; i < nr_ppas; i++)
1914                 WARN_ON(ppa_list[0].a.lun != ppa_list[i].a.lun ||
1915                                 ppa_list[0].a.ch != ppa_list[i].a.ch);
1916 #endif
1917
1918         rlun = &pblk->luns[pos];
1919         up(&rlun->wr_sem);
1920 }
1921
1922 void pblk_up_rq(struct pblk *pblk, struct ppa_addr *ppa_list, int nr_ppas,
1923                 unsigned long *lun_bitmap)
1924 {
1925         struct nvm_tgt_dev *dev = pblk->dev;
1926         struct nvm_geo *geo = &dev->geo;
1927         struct pblk_lun *rlun;
1928         int num_lun = geo->all_luns;
1929         int bit = -1;
1930
1931         while ((bit = find_next_bit(lun_bitmap, num_lun, bit + 1)) < num_lun) {
1932                 rlun = &pblk->luns[bit];
1933                 up(&rlun->wr_sem);
1934         }
1935 }
1936
1937 void pblk_update_map(struct pblk *pblk, sector_t lba, struct ppa_addr ppa)
1938 {
1939         struct ppa_addr ppa_l2p;
1940
1941         /* logic error: lba out-of-bounds. Ignore update */
1942         if (!(lba < pblk->rl.nr_secs)) {
1943                 WARN(1, "pblk: corrupted L2P map request\n");
1944                 return;
1945         }
1946
1947         spin_lock(&pblk->trans_lock);
1948         ppa_l2p = pblk_trans_map_get(pblk, lba);
1949
1950         if (!pblk_addr_in_cache(ppa_l2p) && !pblk_ppa_empty(ppa_l2p))
1951                 pblk_map_invalidate(pblk, ppa_l2p);
1952
1953         pblk_trans_map_set(pblk, lba, ppa);
1954         spin_unlock(&pblk->trans_lock);
1955 }
1956
1957 void pblk_update_map_cache(struct pblk *pblk, sector_t lba, struct ppa_addr ppa)
1958 {
1959
1960 #ifdef CONFIG_NVM_PBLK_DEBUG
1961         /* Callers must ensure that the ppa points to a cache address */
1962         BUG_ON(!pblk_addr_in_cache(ppa));
1963         BUG_ON(pblk_rb_pos_oob(&pblk->rwb, pblk_addr_to_cacheline(ppa)));
1964 #endif
1965
1966         pblk_update_map(pblk, lba, ppa);
1967 }
1968
1969 int pblk_update_map_gc(struct pblk *pblk, sector_t lba, struct ppa_addr ppa_new,
1970                        struct pblk_line *gc_line, u64 paddr_gc)
1971 {
1972         struct ppa_addr ppa_l2p, ppa_gc;
1973         int ret = 1;
1974
1975 #ifdef CONFIG_NVM_PBLK_DEBUG
1976         /* Callers must ensure that the ppa points to a cache address */
1977         BUG_ON(!pblk_addr_in_cache(ppa_new));
1978         BUG_ON(pblk_rb_pos_oob(&pblk->rwb, pblk_addr_to_cacheline(ppa_new)));
1979 #endif
1980
1981         /* logic error: lba out-of-bounds. Ignore update */
1982         if (!(lba < pblk->rl.nr_secs)) {
1983                 WARN(1, "pblk: corrupted L2P map request\n");
1984                 return 0;
1985         }
1986
1987         spin_lock(&pblk->trans_lock);
1988         ppa_l2p = pblk_trans_map_get(pblk, lba);
1989         ppa_gc = addr_to_gen_ppa(pblk, paddr_gc, gc_line->id);
1990
1991         if (!pblk_ppa_comp(ppa_l2p, ppa_gc)) {
1992                 spin_lock(&gc_line->lock);
1993                 WARN(!test_bit(paddr_gc, gc_line->invalid_bitmap),
1994                                                 "pblk: corrupted GC update");
1995                 spin_unlock(&gc_line->lock);
1996
1997                 ret = 0;
1998                 goto out;
1999         }
2000
2001         pblk_trans_map_set(pblk, lba, ppa_new);
2002 out:
2003         spin_unlock(&pblk->trans_lock);
2004         return ret;
2005 }
2006
2007 void pblk_update_map_dev(struct pblk *pblk, sector_t lba,
2008                          struct ppa_addr ppa_mapped, struct ppa_addr ppa_cache)
2009 {
2010         struct ppa_addr ppa_l2p;
2011
2012 #ifdef CONFIG_NVM_PBLK_DEBUG
2013         /* Callers must ensure that the ppa points to a device address */
2014         BUG_ON(pblk_addr_in_cache(ppa_mapped));
2015 #endif
2016         /* Invalidate and discard padded entries */
2017         if (lba == ADDR_EMPTY) {
2018                 atomic64_inc(&pblk->pad_wa);
2019 #ifdef CONFIG_NVM_PBLK_DEBUG
2020                 atomic_long_inc(&pblk->padded_wb);
2021 #endif
2022                 if (!pblk_ppa_empty(ppa_mapped))
2023                         pblk_map_invalidate(pblk, ppa_mapped);
2024                 return;
2025         }
2026
2027         /* logic error: lba out-of-bounds. Ignore update */
2028         if (!(lba < pblk->rl.nr_secs)) {
2029                 WARN(1, "pblk: corrupted L2P map request\n");
2030                 return;
2031         }
2032
2033         spin_lock(&pblk->trans_lock);
2034         ppa_l2p = pblk_trans_map_get(pblk, lba);
2035
2036         /* Do not update L2P if the cacheline has been updated. In this case,
2037          * the mapped ppa must be invalidated
2038          */
2039         if (!pblk_ppa_comp(ppa_l2p, ppa_cache)) {
2040                 if (!pblk_ppa_empty(ppa_mapped))
2041                         pblk_map_invalidate(pblk, ppa_mapped);
2042                 goto out;
2043         }
2044
2045 #ifdef CONFIG_NVM_PBLK_DEBUG
2046         WARN_ON(!pblk_addr_in_cache(ppa_l2p) && !pblk_ppa_empty(ppa_l2p));
2047 #endif
2048
2049         pblk_trans_map_set(pblk, lba, ppa_mapped);
2050 out:
2051         spin_unlock(&pblk->trans_lock);
2052 }
2053
2054 void pblk_lookup_l2p_seq(struct pblk *pblk, struct ppa_addr *ppas,
2055                          sector_t blba, int nr_secs)
2056 {
2057         int i;
2058
2059         spin_lock(&pblk->trans_lock);
2060         for (i = 0; i < nr_secs; i++) {
2061                 struct ppa_addr ppa;
2062
2063                 ppa = ppas[i] = pblk_trans_map_get(pblk, blba + i);
2064
2065                 /* If the L2P entry maps to a line, the reference is valid */
2066                 if (!pblk_ppa_empty(ppa) && !pblk_addr_in_cache(ppa)) {
2067                         int line_id = pblk_ppa_to_line(ppa);
2068                         struct pblk_line *line = &pblk->lines[line_id];
2069
2070                         kref_get(&line->ref);
2071                 }
2072         }
2073         spin_unlock(&pblk->trans_lock);
2074 }
2075
2076 void pblk_lookup_l2p_rand(struct pblk *pblk, struct ppa_addr *ppas,
2077                           u64 *lba_list, int nr_secs)
2078 {
2079         u64 lba;
2080         int i;
2081
2082         spin_lock(&pblk->trans_lock);
2083         for (i = 0; i < nr_secs; i++) {
2084                 lba = lba_list[i];
2085                 if (lba != ADDR_EMPTY) {
2086                         /* logic error: lba out-of-bounds. Ignore update */
2087                         if (!(lba < pblk->rl.nr_secs)) {
2088                                 WARN(1, "pblk: corrupted L2P map request\n");
2089                                 continue;
2090                         }
2091                         ppas[i] = pblk_trans_map_get(pblk, lba);
2092                 }
2093         }
2094         spin_unlock(&pblk->trans_lock);
2095 }