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.is_seq = 1;
 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
 707                         if (pblk_io_aligned(pblk, rq_ppas))
 708                                 rqd.is_seq = 1;
 709
 710                         while (test_bit(pos, line->blk_bitmap)) {
 711                                 paddr += min;
 712                                 if (pblk_boundary_paddr_checks(pblk, paddr)) {
 713                                         pblk_err(pblk, "corrupt emeta line:%d\n",
 714                                                                 line->id);
 715                                         bio_put(bio);
 716                                         ret = -EINTR;
 717                                         goto free_rqd_dma;
 718                                 }
 719
 720                                 ppa = addr_to_gen_ppa(pblk, paddr, id);
 721                                 pos = pblk_ppa_to_pos(geo, ppa);
 722                         }
 723
 724                         if (pblk_boundary_paddr_checks(pblk, paddr + min)) {
 725                                 pblk_err(pblk, "corrupt emeta line:%d\n",
 726                                                                 line->id);
 727                                 bio_put(bio);
 728                                 ret = -EINTR;
 729                                 goto free_rqd_dma;
 730                         }
 731
 732                         for (j = 0; j < min; j++, i++, paddr++)
 733                                 rqd.ppa_list[i] =
 734                                         addr_to_gen_ppa(pblk, paddr, line->id);
 735                 }
 736         }
 737
 738         ret = pblk_submit_io_sync(pblk, &rqd);
 739         if (ret) {
 740                 pblk_err(pblk, "emeta I/O submission failed: %d\n", ret);
 741                 bio_put(bio);
 742                 goto free_rqd_dma;
 743         }
 744
 745         atomic_dec(&pblk->inflight_io);
 746
 747         if (rqd.error) {
 748                 if (dir == PBLK_WRITE)
 749                         pblk_log_write_err(pblk, &rqd);
 750                 else
 751                         pblk_log_read_err(pblk, &rqd);
 752         }
 753
 754         emeta_buf += rq_len;
 755         left_ppas -= rq_ppas;
 756         if (left_ppas)
 757                 goto next_rq;
 758 free_rqd_dma:
 759         nvm_dev_dma_free(dev->parent, rqd.meta_list, rqd.dma_meta_list);
 760         return ret;
 761 }
 762
 763 u64 pblk_line_smeta_start(struct pblk *pblk, struct pblk_line *line)
 764 {
 765         struct nvm_tgt_dev *dev = pblk->dev;
 766         struct nvm_geo *geo = &dev->geo;
 767         struct pblk_line_meta *lm = &pblk->lm;
 768         int bit;
 769
 770         /* This usually only happens on bad lines */
 771         bit = find_first_zero_bit(line->blk_bitmap, lm->blk_per_line);
 772         if (bit >= lm->blk_per_line)
 773                 return -1;
 774
 775         return bit * geo->ws_opt;
 776 }
 777
 778 static int pblk_line_submit_smeta_io(struct pblk *pblk, struct pblk_line *line,
 779                                      u64 paddr, int dir)
 780 {
 781         struct nvm_tgt_dev *dev = pblk->dev;
 782         struct pblk_line_meta *lm = &pblk->lm;
 783         struct bio *bio;
 784         struct nvm_rq rqd;
 785         __le64 *lba_list = NULL;
 786         int i, ret;
 787         int cmd_op, bio_op;
 788
 789         if (dir == PBLK_WRITE) {
 790                 bio_op = REQ_OP_WRITE;
 791                 cmd_op = NVM_OP_PWRITE;
 792                 lba_list = emeta_to_lbas(pblk, line->emeta->buf);
 793         } else if (dir == PBLK_READ_RECOV || dir == PBLK_READ) {
 794                 bio_op = REQ_OP_READ;
 795                 cmd_op = NVM_OP_PREAD;
 796         } else
 797                 return -EINVAL;
 798
 799         memset(&rqd, 0, sizeof(struct nvm_rq));
 800
 801         rqd.meta_list = nvm_dev_dma_alloc(dev->parent, GFP_KERNEL,
 802                                                         &rqd.dma_meta_list);
 803         if (!rqd.meta_list)
 804                 return -ENOMEM;
 805
 806         rqd.ppa_list = rqd.meta_list + pblk_dma_meta_size;
 807         rqd.dma_ppa_list = rqd.dma_meta_list + pblk_dma_meta_size;
 808
 809         bio = bio_map_kern(dev->q, line->smeta, lm->smeta_len, GFP_KERNEL);
 810         if (IS_ERR(bio)) {
 811                 ret = PTR_ERR(bio);
 812                 goto free_ppa_list;
 813         }
 814
 815         bio->bi_iter.bi_sector = 0; /* internal bio */
 816         bio_set_op_attrs(bio, bio_op, 0);
 817
 818         rqd.bio = bio;
 819         rqd.opcode = cmd_op;
 820         rqd.is_seq = 1;
 821         rqd.nr_ppas = lm->smeta_sec;
 822
 823         for (i = 0; i < lm->smeta_sec; i++, paddr++) {
 824                 struct pblk_sec_meta *meta_list = rqd.meta_list;
 825
 826                 rqd.ppa_list[i] = addr_to_gen_ppa(pblk, paddr, line->id);
 827
 828                 if (dir == PBLK_WRITE) {
 829                         __le64 addr_empty = cpu_to_le64(ADDR_EMPTY);
 830
 831                         meta_list[i].lba = lba_list[paddr] = addr_empty;
 832                 }
 833         }
 834
 835         /*
 836          * This I/O is sent by the write thread when a line is replace. Since
 837          * the write thread is the only one sending write and erase commands,
 838          * there is no need to take the LUN semaphore.
 839          */
 840         ret = pblk_submit_io_sync(pblk, &rqd);
 841         if (ret) {
 842                 pblk_err(pblk, "smeta I/O submission failed: %d\n", ret);
 843                 bio_put(bio);
 844                 goto free_ppa_list;
 845         }
 846
 847         atomic_dec(&pblk->inflight_io);
 848
 849         if (rqd.error) {
 850                 if (dir == PBLK_WRITE) {
 851                         pblk_log_write_err(pblk, &rqd);
 852                         ret = 1;
 853                 } else if (dir == PBLK_READ)
 854                         pblk_log_read_err(pblk, &rqd);
 855         }
 856
 857 free_ppa_list:
 858         nvm_dev_dma_free(dev->parent, rqd.meta_list, rqd.dma_meta_list);
 859
 860         return ret;
 861 }
 862
 863 int pblk_line_read_smeta(struct pblk *pblk, struct pblk_line *line)
 864 {
 865         u64 bpaddr = pblk_line_smeta_start(pblk, line);
 866
 867         return pblk_line_submit_smeta_io(pblk, line, bpaddr, PBLK_READ_RECOV);
 868 }
 869
 870 int pblk_line_read_emeta(struct pblk *pblk, struct pblk_line *line,
 871                          void *emeta_buf)
 872 {
 873         return pblk_line_submit_emeta_io(pblk, line, emeta_buf,
 874                                                 line->emeta_ssec, PBLK_READ);
 875 }
 876
 877 static void pblk_setup_e_rq(struct pblk *pblk, struct nvm_rq *rqd,
 878                             struct ppa_addr ppa)
 879 {
 880         rqd->opcode = NVM_OP_ERASE;
 881         rqd->ppa_addr = ppa;
 882         rqd->nr_ppas = 1;
 883         rqd->is_seq = 1;
 884         rqd->bio = NULL;
 885 }
 886
 887 static int pblk_blk_erase_sync(struct pblk *pblk, struct ppa_addr ppa)
 888 {
 889         struct nvm_rq rqd = {NULL};
 890         int ret;
 891
 892         pblk_setup_e_rq(pblk, &rqd, ppa);
 893
 894         /* The write thread schedules erases so that it minimizes disturbances
 895          * with writes. Thus, there is no need to take the LUN semaphore.
 896          */
 897         ret = pblk_submit_io_sync(pblk, &rqd);
 898         rqd.private = pblk;
 899         __pblk_end_io_erase(pblk, &rqd);
 900
 901         return ret;
 902 }
 903
 904 int pblk_line_erase(struct pblk *pblk, struct pblk_line *line)
 905 {
 906         struct pblk_line_meta *lm = &pblk->lm;
 907         struct ppa_addr ppa;
 908         int ret, bit = -1;
 909
 910         /* Erase only good blocks, one at a time */
 911         do {
 912                 spin_lock(&line->lock);
 913                 bit = find_next_zero_bit(line->erase_bitmap, lm->blk_per_line,
 914                                                                 bit + 1);
 915                 if (bit >= lm->blk_per_line) {
 916                         spin_unlock(&line->lock);
 917                         break;
 918                 }
 919
 920                 ppa = pblk->luns[bit].bppa; /* set ch and lun */
 921                 ppa.a.blk = line->id;
 922
 923                 atomic_dec(&line->left_eblks);
 924                 WARN_ON(test_and_set_bit(bit, line->erase_bitmap));
 925                 spin_unlock(&line->lock);
 926
 927                 ret = pblk_blk_erase_sync(pblk, ppa);
 928                 if (ret) {
 929                         pblk_err(pblk, "failed to erase line %d\n", line->id);
 930                         return ret;
 931                 }
 932         } while (1);
 933
 934         return 0;
 935 }
 936
 937 static void pblk_line_setup_metadata(struct pblk_line *line,
 938                                      struct pblk_line_mgmt *l_mg,
 939                                      struct pblk_line_meta *lm)
 940 {
 941         int meta_line;
 942
 943         lockdep_assert_held(&l_mg->free_lock);
 944
 945 retry_meta:
 946         meta_line = find_first_zero_bit(&l_mg->meta_bitmap, PBLK_DATA_LINES);
 947         if (meta_line == PBLK_DATA_LINES) {
 948                 spin_unlock(&l_mg->free_lock);
 949                 io_schedule();
 950                 spin_lock(&l_mg->free_lock);
 951                 goto retry_meta;
 952         }
 953
 954         set_bit(meta_line, &l_mg->meta_bitmap);
 955         line->meta_line = meta_line;
 956
 957         line->smeta = l_mg->sline_meta[meta_line];
 958         line->emeta = l_mg->eline_meta[meta_line];
 959
 960         memset(line->smeta, 0, lm->smeta_len);
 961         memset(line->emeta->buf, 0, lm->emeta_len[0]);
 962
 963         line->emeta->mem = 0;
 964         atomic_set(&line->emeta->sync, 0);
 965 }
 966
 967 /* For now lines are always assumed full lines. Thus, smeta former and current
 968  * lun bitmaps are omitted.
 969  */
 970 static int pblk_line_init_metadata(struct pblk *pblk, struct pblk_line *line,
 971                                   struct pblk_line *cur)
 972 {
 973         struct nvm_tgt_dev *dev = pblk->dev;
 974         struct nvm_geo *geo = &dev->geo;
 975         struct pblk_line_meta *lm = &pblk->lm;
 976         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
 977         struct pblk_emeta *emeta = line->emeta;
 978         struct line_emeta *emeta_buf = emeta->buf;
 979         struct line_smeta *smeta_buf = (struct line_smeta *)line->smeta;
 980         int nr_blk_line;
 981
 982         /* After erasing the line, new bad blocks might appear and we risk
 983          * having an invalid line
 984          */
 985         nr_blk_line = lm->blk_per_line -
 986                         bitmap_weight(line->blk_bitmap, lm->blk_per_line);
 987         if (nr_blk_line < lm->min_blk_line) {
 988                 spin_lock(&l_mg->free_lock);
 989                 spin_lock(&line->lock);
 990                 line->state = PBLK_LINESTATE_BAD;
 991                 spin_unlock(&line->lock);
 992
 993                 list_add_tail(&line->list, &l_mg->bad_list);
 994                 spin_unlock(&l_mg->free_lock);
 995
 996                 pblk_debug(pblk, "line %d is bad\n", line->id);
 997
 998                 return 0;
 999         }
1000
1001         /* Run-time metadata */
1002         line->lun_bitmap = ((void *)(smeta_buf)) + sizeof(struct line_smeta);
1003
1004         /* Mark LUNs allocated in this line (all for now) */
1005         bitmap_set(line->lun_bitmap, 0, lm->lun_bitmap_len);
1006
1007         smeta_buf->header.identifier = cpu_to_le32(PBLK_MAGIC);
1008         memcpy(smeta_buf->header.uuid, pblk->instance_uuid, 16);
1009         smeta_buf->header.id = cpu_to_le32(line->id);
1010         smeta_buf->header.type = cpu_to_le16(line->type);
1011         smeta_buf->header.version_major = SMETA_VERSION_MAJOR;
1012         smeta_buf->header.version_minor = SMETA_VERSION_MINOR;
1013
1014         /* Start metadata */
1015         smeta_buf->seq_nr = cpu_to_le64(line->seq_nr);
1016         smeta_buf->window_wr_lun = cpu_to_le32(geo->all_luns);
1017
1018         /* Fill metadata among lines */
1019         if (cur) {
1020                 memcpy(line->lun_bitmap, cur->lun_bitmap, lm->lun_bitmap_len);
1021                 smeta_buf->prev_id = cpu_to_le32(cur->id);
1022                 cur->emeta->buf->next_id = cpu_to_le32(line->id);
1023         } else {
1024                 smeta_buf->prev_id = cpu_to_le32(PBLK_LINE_EMPTY);
1025         }
1026
1027         /* All smeta must be set at this point */
1028         smeta_buf->header.crc = cpu_to_le32(
1029                         pblk_calc_meta_header_crc(pblk, &smeta_buf->header));
1030         smeta_buf->crc = cpu_to_le32(pblk_calc_smeta_crc(pblk, smeta_buf));
1031
1032         /* End metadata */
1033         memcpy(&emeta_buf->header, &smeta_buf->header,
1034                                                 sizeof(struct line_header));
1035
1036         emeta_buf->header.version_major = EMETA_VERSION_MAJOR;
1037         emeta_buf->header.version_minor = EMETA_VERSION_MINOR;
1038         emeta_buf->header.crc = cpu_to_le32(
1039                         pblk_calc_meta_header_crc(pblk, &emeta_buf->header));
1040
1041         emeta_buf->seq_nr = cpu_to_le64(line->seq_nr);
1042         emeta_buf->nr_lbas = cpu_to_le64(line->sec_in_line);
1043         emeta_buf->nr_valid_lbas = cpu_to_le64(0);
1044         emeta_buf->next_id = cpu_to_le32(PBLK_LINE_EMPTY);
1045         emeta_buf->crc = cpu_to_le32(0);
1046         emeta_buf->prev_id = smeta_buf->prev_id;
1047
1048         return 1;
1049 }
1050
1051 static int pblk_line_alloc_bitmaps(struct pblk *pblk, struct pblk_line *line)
1052 {
1053         struct pblk_line_meta *lm = &pblk->lm;
1054
1055         line->map_bitmap = kzalloc(lm->sec_bitmap_len, GFP_KERNEL);
1056         if (!line->map_bitmap)
1057                 return -ENOMEM;
1058
1059         /* will be initialized using bb info from map_bitmap */
1060         line->invalid_bitmap = kmalloc(lm->sec_bitmap_len, GFP_KERNEL);
1061         if (!line->invalid_bitmap) {
1062                 kfree(line->map_bitmap);
1063                 line->map_bitmap = NULL;
1064                 return -ENOMEM;
1065         }
1066
1067         return 0;
1068 }
1069
1070 /* For now lines are always assumed full lines. Thus, smeta former and current
1071  * lun bitmaps are omitted.
1072  */
1073 static int pblk_line_init_bb(struct pblk *pblk, struct pblk_line *line,
1074                              int init)
1075 {
1076         struct nvm_tgt_dev *dev = pblk->dev;
1077         struct nvm_geo *geo = &dev->geo;
1078         struct pblk_line_meta *lm = &pblk->lm;
1079         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1080         u64 off;
1081         int bit = -1;
1082         int emeta_secs;
1083
1084         line->sec_in_line = lm->sec_per_line;
1085
1086         /* Capture bad block information on line mapping bitmaps */
1087         while ((bit = find_next_bit(line->blk_bitmap, lm->blk_per_line,
1088                                         bit + 1)) < lm->blk_per_line) {
1089                 off = bit * geo->ws_opt;
1090                 bitmap_shift_left(l_mg->bb_aux, l_mg->bb_template, off,
1091                                                         lm->sec_per_line);
1092                 bitmap_or(line->map_bitmap, line->map_bitmap, l_mg->bb_aux,
1093                                                         lm->sec_per_line);
1094                 line->sec_in_line -= geo->clba;
1095         }
1096
1097         /* Mark smeta metadata sectors as bad sectors */
1098         bit = find_first_zero_bit(line->blk_bitmap, lm->blk_per_line);
1099         off = bit * geo->ws_opt;
1100         bitmap_set(line->map_bitmap, off, lm->smeta_sec);
1101         line->sec_in_line -= lm->smeta_sec;
1102         line->smeta_ssec = off;
1103         line->cur_sec = off + lm->smeta_sec;
1104
1105         if (init && pblk_line_submit_smeta_io(pblk, line, off, PBLK_WRITE)) {
1106                 pblk_debug(pblk, "line smeta I/O failed. Retry\n");
1107                 return 0;
1108         }
1109
1110         bitmap_copy(line->invalid_bitmap, line->map_bitmap, lm->sec_per_line);
1111
1112         /* Mark emeta metadata sectors as bad sectors. We need to consider bad
1113          * blocks to make sure that there are enough sectors to store emeta
1114          */
1115         emeta_secs = lm->emeta_sec[0];
1116         off = lm->sec_per_line;
1117         while (emeta_secs) {
1118                 off -= geo->ws_opt;
1119                 if (!test_bit(off, line->invalid_bitmap)) {
1120                         bitmap_set(line->invalid_bitmap, off, geo->ws_opt);
1121                         emeta_secs -= geo->ws_opt;
1122                 }
1123         }
1124
1125         line->emeta_ssec = off;
1126         line->sec_in_line -= lm->emeta_sec[0];
1127         line->nr_valid_lbas = 0;
1128         line->left_msecs = line->sec_in_line;
1129         *line->vsc = cpu_to_le32(line->sec_in_line);
1130
1131         if (lm->sec_per_line - line->sec_in_line !=
1132                 bitmap_weight(line->invalid_bitmap, lm->sec_per_line)) {
1133                 spin_lock(&line->lock);
1134                 line->state = PBLK_LINESTATE_BAD;
1135                 spin_unlock(&line->lock);
1136
1137                 list_add_tail(&line->list, &l_mg->bad_list);
1138                 pblk_err(pblk, "unexpected line %d is bad\n", line->id);
1139
1140                 return 0;
1141         }
1142
1143         return 1;
1144 }
1145
1146 static int pblk_prepare_new_line(struct pblk *pblk, struct pblk_line *line)
1147 {
1148         struct pblk_line_meta *lm = &pblk->lm;
1149         struct nvm_tgt_dev *dev = pblk->dev;
1150         struct nvm_geo *geo = &dev->geo;
1151         int blk_to_erase = atomic_read(&line->blk_in_line);
1152         int i;
1153
1154         for (i = 0; i < lm->blk_per_line; i++) {
1155                 struct pblk_lun *rlun = &pblk->luns[i];
1156                 int pos = pblk_ppa_to_pos(geo, rlun->bppa);
1157                 int state = line->chks[pos].state;
1158
1159                 /* Free chunks should not be erased */
1160                 if (state & NVM_CHK_ST_FREE) {
1161                         set_bit(pblk_ppa_to_pos(geo, rlun->bppa),
1162                                                         line->erase_bitmap);
1163                         blk_to_erase--;
1164                 }
1165         }
1166
1167         return blk_to_erase;
1168 }
1169
1170 static int pblk_line_prepare(struct pblk *pblk, struct pblk_line *line)
1171 {
1172         struct pblk_line_meta *lm = &pblk->lm;
1173         int blk_in_line = atomic_read(&line->blk_in_line);
1174         int blk_to_erase;
1175
1176         /* Bad blocks do not need to be erased */
1177         bitmap_copy(line->erase_bitmap, line->blk_bitmap, lm->blk_per_line);
1178
1179         spin_lock(&line->lock);
1180
1181         /* If we have not written to this line, we need to mark up free chunks
1182          * as already erased
1183          */
1184         if (line->state == PBLK_LINESTATE_NEW) {
1185                 blk_to_erase = pblk_prepare_new_line(pblk, line);
1186                 line->state = PBLK_LINESTATE_FREE;
1187         } else {
1188                 blk_to_erase = blk_in_line;
1189         }
1190
1191         if (blk_in_line < lm->min_blk_line) {
1192                 spin_unlock(&line->lock);
1193                 return -EAGAIN;
1194         }
1195
1196         if (line->state != PBLK_LINESTATE_FREE) {
1197                 WARN(1, "pblk: corrupted line %d, state %d\n",
1198                                                         line->id, line->state);
1199                 spin_unlock(&line->lock);
1200                 return -EINTR;
1201         }
1202
1203         line->state = PBLK_LINESTATE_OPEN;
1204
1205         atomic_set(&line->left_eblks, blk_to_erase);
1206         atomic_set(&line->left_seblks, blk_to_erase);
1207
1208         line->meta_distance = lm->meta_distance;
1209         spin_unlock(&line->lock);
1210
1211         kref_init(&line->ref);
1212
1213         return 0;
1214 }
1215
1216 int pblk_line_recov_alloc(struct pblk *pblk, struct pblk_line *line)
1217 {
1218         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1219         int ret;
1220
1221         spin_lock(&l_mg->free_lock);
1222         l_mg->data_line = line;
1223         list_del(&line->list);
1224
1225         ret = pblk_line_prepare(pblk, line);
1226         if (ret) {
1227                 list_add(&line->list, &l_mg->free_list);
1228                 spin_unlock(&l_mg->free_lock);
1229                 return ret;
1230         }
1231         spin_unlock(&l_mg->free_lock);
1232
1233         ret = pblk_line_alloc_bitmaps(pblk, line);
1234         if (ret)
1235                 return ret;
1236
1237         if (!pblk_line_init_bb(pblk, line, 0)) {
1238                 list_add(&line->list, &l_mg->free_list);
1239                 return -EINTR;
1240         }
1241
1242         pblk_rl_free_lines_dec(&pblk->rl, line, true);
1243         return 0;
1244 }
1245
1246 void pblk_line_recov_close(struct pblk *pblk, struct pblk_line *line)
1247 {
1248         kfree(line->map_bitmap);
1249         line->map_bitmap = NULL;
1250         line->smeta = NULL;
1251         line->emeta = NULL;
1252 }
1253
1254 static void pblk_line_reinit(struct pblk_line *line)
1255 {
1256         *line->vsc = cpu_to_le32(EMPTY_ENTRY);
1257
1258         line->map_bitmap = NULL;
1259         line->invalid_bitmap = NULL;
1260         line->smeta = NULL;
1261         line->emeta = NULL;
1262 }
1263
1264 void pblk_line_free(struct pblk_line *line)
1265 {
1266         kfree(line->map_bitmap);
1267         kfree(line->invalid_bitmap);
1268
1269         pblk_line_reinit(line);
1270 }
1271
1272 struct pblk_line *pblk_line_get(struct pblk *pblk)
1273 {
1274         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1275         struct pblk_line_meta *lm = &pblk->lm;
1276         struct pblk_line *line;
1277         int ret, bit;
1278
1279         lockdep_assert_held(&l_mg->free_lock);
1280
1281 retry:
1282         if (list_empty(&l_mg->free_list)) {
1283                 pblk_err(pblk, "no free lines\n");
1284                 return NULL;
1285         }
1286
1287         line = list_first_entry(&l_mg->free_list, struct pblk_line, list);
1288         list_del(&line->list);
1289         l_mg->nr_free_lines--;
1290
1291         bit = find_first_zero_bit(line->blk_bitmap, lm->blk_per_line);
1292         if (unlikely(bit >= lm->blk_per_line)) {
1293                 spin_lock(&line->lock);
1294                 line->state = PBLK_LINESTATE_BAD;
1295                 spin_unlock(&line->lock);
1296
1297                 list_add_tail(&line->list, &l_mg->bad_list);
1298
1299                 pblk_debug(pblk, "line %d is bad\n", line->id);
1300                 goto retry;
1301         }
1302
1303         ret = pblk_line_prepare(pblk, line);
1304         if (ret) {
1305                 switch (ret) {
1306                 case -EAGAIN:
1307                         list_add(&line->list, &l_mg->bad_list);
1308                         goto retry;
1309                 case -EINTR:
1310                         list_add(&line->list, &l_mg->corrupt_list);
1311                         goto retry;
1312                 default:
1313                         pblk_err(pblk, "failed to prepare line %d\n", line->id);
1314                         list_add(&line->list, &l_mg->free_list);
1315                         l_mg->nr_free_lines++;
1316                         return NULL;
1317                 }
1318         }
1319
1320         return line;
1321 }
1322
1323 static struct pblk_line *pblk_line_retry(struct pblk *pblk,
1324                                          struct pblk_line *line)
1325 {
1326         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1327         struct pblk_line *retry_line;
1328
1329 retry:
1330         spin_lock(&l_mg->free_lock);
1331         retry_line = pblk_line_get(pblk);
1332         if (!retry_line) {
1333                 l_mg->data_line = NULL;
1334                 spin_unlock(&l_mg->free_lock);
1335                 return NULL;
1336         }
1337
1338         retry_line->map_bitmap = line->map_bitmap;
1339         retry_line->invalid_bitmap = line->invalid_bitmap;
1340         retry_line->smeta = line->smeta;
1341         retry_line->emeta = line->emeta;
1342         retry_line->meta_line = line->meta_line;
1343
1344         pblk_line_reinit(line);
1345
1346         l_mg->data_line = retry_line;
1347         spin_unlock(&l_mg->free_lock);
1348
1349         pblk_rl_free_lines_dec(&pblk->rl, line, false);
1350
1351         if (pblk_line_erase(pblk, retry_line))
1352                 goto retry;
1353
1354         return retry_line;
1355 }
1356
1357 static void pblk_set_space_limit(struct pblk *pblk)
1358 {
1359         struct pblk_rl *rl = &pblk->rl;
1360
1361         atomic_set(&rl->rb_space, 0);
1362 }
1363
1364 struct pblk_line *pblk_line_get_first_data(struct pblk *pblk)
1365 {
1366         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1367         struct pblk_line *line;
1368
1369         spin_lock(&l_mg->free_lock);
1370         line = pblk_line_get(pblk);
1371         if (!line) {
1372                 spin_unlock(&l_mg->free_lock);
1373                 return NULL;
1374         }
1375
1376         line->seq_nr = l_mg->d_seq_nr++;
1377         line->type = PBLK_LINETYPE_DATA;
1378         l_mg->data_line = line;
1379
1380         pblk_line_setup_metadata(line, l_mg, &pblk->lm);
1381
1382         /* Allocate next line for preparation */
1383         l_mg->data_next = pblk_line_get(pblk);
1384         if (!l_mg->data_next) {
1385                 /* If we cannot get a new line, we need to stop the pipeline.
1386                  * Only allow as many writes in as we can store safely and then
1387                  * fail gracefully
1388                  */
1389                 pblk_set_space_limit(pblk);
1390
1391                 l_mg->data_next = NULL;
1392         } else {
1393                 l_mg->data_next->seq_nr = l_mg->d_seq_nr++;
1394                 l_mg->data_next->type = PBLK_LINETYPE_DATA;
1395         }
1396         spin_unlock(&l_mg->free_lock);
1397
1398         if (pblk_line_alloc_bitmaps(pblk, line))
1399                 return NULL;
1400
1401         if (pblk_line_erase(pblk, line)) {
1402                 line = pblk_line_retry(pblk, line);
1403                 if (!line)
1404                         return NULL;
1405         }
1406
1407 retry_setup:
1408         if (!pblk_line_init_metadata(pblk, line, NULL)) {
1409                 line = pblk_line_retry(pblk, line);
1410                 if (!line)
1411                         return NULL;
1412
1413                 goto retry_setup;
1414         }
1415
1416         if (!pblk_line_init_bb(pblk, line, 1)) {
1417                 line = pblk_line_retry(pblk, line);
1418                 if (!line)
1419                         return NULL;
1420
1421                 goto retry_setup;
1422         }
1423
1424         pblk_rl_free_lines_dec(&pblk->rl, line, true);
1425
1426         return line;
1427 }
1428
1429 static void pblk_stop_writes(struct pblk *pblk, struct pblk_line *line)
1430 {
1431         lockdep_assert_held(&pblk->l_mg.free_lock);
1432
1433         pblk_set_space_limit(pblk);
1434         pblk->state = PBLK_STATE_STOPPING;
1435 }
1436
1437 static void pblk_line_close_meta_sync(struct pblk *pblk)
1438 {
1439         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1440         struct pblk_line_meta *lm = &pblk->lm;
1441         struct pblk_line *line, *tline;
1442         LIST_HEAD(list);
1443
1444         spin_lock(&l_mg->close_lock);
1445         if (list_empty(&l_mg->emeta_list)) {
1446                 spin_unlock(&l_mg->close_lock);
1447                 return;
1448         }
1449
1450         list_cut_position(&list, &l_mg->emeta_list, l_mg->emeta_list.prev);
1451         spin_unlock(&l_mg->close_lock);
1452
1453         list_for_each_entry_safe(line, tline, &list, list) {
1454                 struct pblk_emeta *emeta = line->emeta;
1455
1456                 while (emeta->mem < lm->emeta_len[0]) {
1457                         int ret;
1458
1459                         ret = pblk_submit_meta_io(pblk, line);
1460                         if (ret) {
1461                                 pblk_err(pblk, "sync meta line %d failed (%d)\n",
1462                                                         line->id, ret);
1463                                 return;
1464                         }
1465                 }
1466         }
1467
1468         pblk_wait_for_meta(pblk);
1469         flush_workqueue(pblk->close_wq);
1470 }
1471
1472 void __pblk_pipeline_flush(struct pblk *pblk)
1473 {
1474         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1475         int ret;
1476
1477         spin_lock(&l_mg->free_lock);
1478         if (pblk->state == PBLK_STATE_RECOVERING ||
1479                                         pblk->state == PBLK_STATE_STOPPED) {
1480                 spin_unlock(&l_mg->free_lock);
1481                 return;
1482         }
1483         pblk->state = PBLK_STATE_RECOVERING;
1484         spin_unlock(&l_mg->free_lock);
1485
1486         pblk_flush_writer(pblk);
1487         pblk_wait_for_meta(pblk);
1488
1489         ret = pblk_recov_pad(pblk);
1490         if (ret) {
1491                 pblk_err(pblk, "could not close data on teardown(%d)\n", ret);
1492                 return;
1493         }
1494
1495         flush_workqueue(pblk->bb_wq);
1496         pblk_line_close_meta_sync(pblk);
1497 }
1498
1499 void __pblk_pipeline_stop(struct pblk *pblk)
1500 {
1501         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1502
1503         spin_lock(&l_mg->free_lock);
1504         pblk->state = PBLK_STATE_STOPPED;
1505         l_mg->data_line = NULL;
1506         l_mg->data_next = NULL;
1507         spin_unlock(&l_mg->free_lock);
1508 }
1509
1510 void pblk_pipeline_stop(struct pblk *pblk)
1511 {
1512         __pblk_pipeline_flush(pblk);
1513         __pblk_pipeline_stop(pblk);
1514 }
1515
1516 struct pblk_line *pblk_line_replace_data(struct pblk *pblk)
1517 {
1518         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1519         struct pblk_line *cur, *new = NULL;
1520         unsigned int left_seblks;
1521
1522         cur = l_mg->data_line;
1523         new = l_mg->data_next;
1524         if (!new)
1525                 goto out;
1526         l_mg->data_line = new;
1527
1528         spin_lock(&l_mg->free_lock);
1529         pblk_line_setup_metadata(new, l_mg, &pblk->lm);
1530         spin_unlock(&l_mg->free_lock);
1531
1532 retry_erase:
1533         left_seblks = atomic_read(&new->left_seblks);
1534         if (left_seblks) {
1535                 /* If line is not fully erased, erase it */
1536                 if (atomic_read(&new->left_eblks)) {
1537                         if (pblk_line_erase(pblk, new))
1538                                 goto out;
1539                 } else {
1540                         io_schedule();
1541                 }
1542                 goto retry_erase;
1543         }
1544
1545         if (pblk_line_alloc_bitmaps(pblk, new))
1546                 return NULL;
1547
1548 retry_setup:
1549         if (!pblk_line_init_metadata(pblk, new, cur)) {
1550                 new = pblk_line_retry(pblk, new);
1551                 if (!new)
1552                         goto out;
1553
1554                 goto retry_setup;
1555         }
1556
1557         if (!pblk_line_init_bb(pblk, new, 1)) {
1558                 new = pblk_line_retry(pblk, new);
1559                 if (!new)
1560                         goto out;
1561
1562                 goto retry_setup;
1563         }
1564
1565         pblk_rl_free_lines_dec(&pblk->rl, new, true);
1566
1567         /* Allocate next line for preparation */
1568         spin_lock(&l_mg->free_lock);
1569         l_mg->data_next = pblk_line_get(pblk);
1570         if (!l_mg->data_next) {
1571                 /* If we cannot get a new line, we need to stop the pipeline.
1572                  * Only allow as many writes in as we can store safely and then
1573                  * fail gracefully
1574                  */
1575                 pblk_stop_writes(pblk, new);
1576                 l_mg->data_next = NULL;
1577         } else {
1578                 l_mg->data_next->seq_nr = l_mg->d_seq_nr++;
1579                 l_mg->data_next->type = PBLK_LINETYPE_DATA;
1580         }
1581         spin_unlock(&l_mg->free_lock);
1582
1583 out:
1584         return new;
1585 }
1586
1587 static void __pblk_line_put(struct pblk *pblk, struct pblk_line *line)
1588 {
1589         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1590         struct pblk_gc *gc = &pblk->gc;
1591
1592         spin_lock(&line->lock);
1593         WARN_ON(line->state != PBLK_LINESTATE_GC);
1594         line->state = PBLK_LINESTATE_FREE;
1595         line->gc_group = PBLK_LINEGC_NONE;
1596         pblk_line_free(line);
1597
1598         if (line->w_err_gc->has_write_err) {
1599                 pblk_rl_werr_line_out(&pblk->rl);
1600                 line->w_err_gc->has_write_err = 0;
1601         }
1602
1603         spin_unlock(&line->lock);
1604         atomic_dec(&gc->pipeline_gc);
1605
1606         spin_lock(&l_mg->free_lock);
1607         list_add_tail(&line->list, &l_mg->free_list);
1608         l_mg->nr_free_lines++;
1609         spin_unlock(&l_mg->free_lock);
1610
1611         pblk_rl_free_lines_inc(&pblk->rl, line);
1612 }
1613
1614 static void pblk_line_put_ws(struct work_struct *work)
1615 {
1616         struct pblk_line_ws *line_put_ws = container_of(work,
1617                                                 struct pblk_line_ws, ws);
1618         struct pblk *pblk = line_put_ws->pblk;
1619         struct pblk_line *line = line_put_ws->line;
1620
1621         __pblk_line_put(pblk, line);
1622         mempool_free(line_put_ws, &pblk->gen_ws_pool);
1623 }
1624
1625 void pblk_line_put(struct kref *ref)
1626 {
1627         struct pblk_line *line = container_of(ref, struct pblk_line, ref);
1628         struct pblk *pblk = line->pblk;
1629
1630         __pblk_line_put(pblk, line);
1631 }
1632
1633 void pblk_line_put_wq(struct kref *ref)
1634 {
1635         struct pblk_line *line = container_of(ref, struct pblk_line, ref);
1636         struct pblk *pblk = line->pblk;
1637         struct pblk_line_ws *line_put_ws;
1638
1639         line_put_ws = mempool_alloc(&pblk->gen_ws_pool, GFP_ATOMIC);
1640         if (!line_put_ws)
1641                 return;
1642
1643         line_put_ws->pblk = pblk;
1644         line_put_ws->line = line;
1645         line_put_ws->priv = NULL;
1646
1647         INIT_WORK(&line_put_ws->ws, pblk_line_put_ws);
1648         queue_work(pblk->r_end_wq, &line_put_ws->ws);
1649 }
1650
1651 int pblk_blk_erase_async(struct pblk *pblk, struct ppa_addr ppa)
1652 {
1653         struct nvm_rq *rqd;
1654         int err;
1655
1656         rqd = pblk_alloc_rqd(pblk, PBLK_ERASE);
1657
1658         pblk_setup_e_rq(pblk, rqd, ppa);
1659
1660         rqd->end_io = pblk_end_io_erase;
1661         rqd->private = pblk;
1662
1663         /* The write thread schedules erases so that it minimizes disturbances
1664          * with writes. Thus, there is no need to take the LUN semaphore.
1665          */
1666         err = pblk_submit_io(pblk, rqd);
1667         if (err) {
1668                 struct nvm_tgt_dev *dev = pblk->dev;
1669                 struct nvm_geo *geo = &dev->geo;
1670
1671                 pblk_err(pblk, "could not async erase line:%d,blk:%d\n",
1672                                         pblk_ppa_to_line(ppa),
1673                                         pblk_ppa_to_pos(geo, ppa));
1674         }
1675
1676         return err;
1677 }
1678
1679 struct pblk_line *pblk_line_get_data(struct pblk *pblk)
1680 {
1681         return pblk->l_mg.data_line;
1682 }
1683
1684 /* For now, always erase next line */
1685 struct pblk_line *pblk_line_get_erase(struct pblk *pblk)
1686 {
1687         return pblk->l_mg.data_next;
1688 }
1689
1690 int pblk_line_is_full(struct pblk_line *line)
1691 {
1692         return (line->left_msecs == 0);
1693 }
1694
1695 static void pblk_line_should_sync_meta(struct pblk *pblk)
1696 {
1697         if (pblk_rl_is_limit(&pblk->rl))
1698                 pblk_line_close_meta_sync(pblk);
1699 }
1700
1701 void pblk_line_close(struct pblk *pblk, struct pblk_line *line)
1702 {
1703         struct nvm_tgt_dev *dev = pblk->dev;
1704         struct nvm_geo *geo = &dev->geo;
1705         struct pblk_line_meta *lm = &pblk->lm;
1706         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1707         struct list_head *move_list;
1708         int i;
1709
1710 #ifdef CONFIG_NVM_PBLK_DEBUG
1711         WARN(!bitmap_full(line->map_bitmap, lm->sec_per_line),
1712                                 "pblk: corrupt closed line %d\n", line->id);
1713 #endif
1714
1715         spin_lock(&l_mg->free_lock);
1716         WARN_ON(!test_and_clear_bit(line->meta_line, &l_mg->meta_bitmap));
1717         spin_unlock(&l_mg->free_lock);
1718
1719         spin_lock(&l_mg->gc_lock);
1720         spin_lock(&line->lock);
1721         WARN_ON(line->state != PBLK_LINESTATE_OPEN);
1722         line->state = PBLK_LINESTATE_CLOSED;
1723         move_list = pblk_line_gc_list(pblk, line);
1724
1725         list_add_tail(&line->list, move_list);
1726
1727         kfree(line->map_bitmap);
1728         line->map_bitmap = NULL;
1729         line->smeta = NULL;
1730         line->emeta = NULL;
1731
1732         for (i = 0; i < lm->blk_per_line; i++) {
1733                 struct pblk_lun *rlun = &pblk->luns[i];
1734                 int pos = pblk_ppa_to_pos(geo, rlun->bppa);
1735                 int state = line->chks[pos].state;
1736
1737                 if (!(state & NVM_CHK_ST_OFFLINE))
1738                         state = NVM_CHK_ST_CLOSED;
1739         }
1740
1741         spin_unlock(&line->lock);
1742         spin_unlock(&l_mg->gc_lock);
1743 }
1744
1745 void pblk_line_close_meta(struct pblk *pblk, struct pblk_line *line)
1746 {
1747         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1748         struct pblk_line_meta *lm = &pblk->lm;
1749         struct pblk_emeta *emeta = line->emeta;
1750         struct line_emeta *emeta_buf = emeta->buf;
1751         struct wa_counters *wa = emeta_to_wa(lm, emeta_buf);
1752
1753         /* No need for exact vsc value; avoid a big line lock and take aprox. */
1754         memcpy(emeta_to_vsc(pblk, emeta_buf), l_mg->vsc_list, lm->vsc_list_len);
1755         memcpy(emeta_to_bb(emeta_buf), line->blk_bitmap, lm->blk_bitmap_len);
1756
1757         wa->user = cpu_to_le64(atomic64_read(&pblk->user_wa));
1758         wa->pad = cpu_to_le64(atomic64_read(&pblk->pad_wa));
1759         wa->gc = cpu_to_le64(atomic64_read(&pblk->gc_wa));
1760
1761         emeta_buf->nr_valid_lbas = cpu_to_le64(line->nr_valid_lbas);
1762         emeta_buf->crc = cpu_to_le32(pblk_calc_emeta_crc(pblk, emeta_buf));
1763
1764         spin_lock(&l_mg->close_lock);
1765         spin_lock(&line->lock);
1766
1767         /* Update the in-memory start address for emeta, in case it has
1768          * shifted due to write errors
1769          */
1770         if (line->emeta_ssec != line->cur_sec)
1771                 line->emeta_ssec = line->cur_sec;
1772
1773         list_add_tail(&line->list, &l_mg->emeta_list);
1774         spin_unlock(&line->lock);
1775         spin_unlock(&l_mg->close_lock);
1776
1777         pblk_line_should_sync_meta(pblk);
1778
1779
1780 }
1781
1782 static void pblk_save_lba_list(struct pblk *pblk, struct pblk_line *line)
1783 {
1784         struct pblk_line_meta *lm = &pblk->lm;
1785         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1786         unsigned int lba_list_size = lm->emeta_len[2];
1787         struct pblk_w_err_gc *w_err_gc = line->w_err_gc;
1788         struct pblk_emeta *emeta = line->emeta;
1789
1790         w_err_gc->lba_list = pblk_malloc(lba_list_size,
1791                                          l_mg->emeta_alloc_type, GFP_KERNEL);
1792         memcpy(w_err_gc->lba_list, emeta_to_lbas(pblk, emeta->buf),
1793                                 lba_list_size);
1794 }
1795
1796 void pblk_line_close_ws(struct work_struct *work)
1797 {
1798         struct pblk_line_ws *line_ws = container_of(work, struct pblk_line_ws,
1799                                                                         ws);
1800         struct pblk *pblk = line_ws->pblk;
1801         struct pblk_line *line = line_ws->line;
1802         struct pblk_w_err_gc *w_err_gc = line->w_err_gc;
1803
1804         /* Write errors makes the emeta start address stored in smeta invalid,
1805          * so keep a copy of the lba list until we've gc'd the line
1806          */
1807         if (w_err_gc->has_write_err)
1808                 pblk_save_lba_list(pblk, line);
1809
1810         pblk_line_close(pblk, line);
1811         mempool_free(line_ws, &pblk->gen_ws_pool);
1812 }
1813
1814 void pblk_gen_run_ws(struct pblk *pblk, struct pblk_line *line, void *priv,
1815                       void (*work)(struct work_struct *), gfp_t gfp_mask,
1816                       struct workqueue_struct *wq)
1817 {
1818         struct pblk_line_ws *line_ws;
1819
1820         line_ws = mempool_alloc(&pblk->gen_ws_pool, gfp_mask);
1821
1822         line_ws->pblk = pblk;
1823         line_ws->line = line;
1824         line_ws->priv = priv;
1825
1826         INIT_WORK(&line_ws->ws, work);
1827         queue_work(wq, &line_ws->ws);
1828 }
1829
1830 static void __pblk_down_page(struct pblk *pblk, struct ppa_addr *ppa_list,
1831                              int nr_ppas, int pos)
1832 {
1833         struct pblk_lun *rlun = &pblk->luns[pos];
1834         int ret;
1835
1836         /*
1837          * Only send one inflight I/O per LUN. Since we map at a page
1838          * granurality, all ppas in the I/O will map to the same LUN
1839          */
1840 #ifdef CONFIG_NVM_PBLK_DEBUG
1841         int i;
1842
1843         for (i = 1; i < nr_ppas; i++)
1844                 WARN_ON(ppa_list[0].a.lun != ppa_list[i].a.lun ||
1845                                 ppa_list[0].a.ch != ppa_list[i].a.ch);
1846 #endif
1847
1848         ret = down_timeout(&rlun->wr_sem, msecs_to_jiffies(30000));
1849         if (ret == -ETIME || ret == -EINTR)
1850                 pblk_err(pblk, "taking lun semaphore timed out: err %d\n",
1851                                 -ret);
1852 }
1853
1854 void pblk_down_page(struct pblk *pblk, struct ppa_addr *ppa_list, int nr_ppas)
1855 {
1856         struct nvm_tgt_dev *dev = pblk->dev;
1857         struct nvm_geo *geo = &dev->geo;
1858         int pos = pblk_ppa_to_pos(geo, ppa_list[0]);
1859
1860         __pblk_down_page(pblk, ppa_list, nr_ppas, pos);
1861 }
1862
1863 void pblk_down_rq(struct pblk *pblk, struct ppa_addr *ppa_list, int nr_ppas,
1864                   unsigned long *lun_bitmap)
1865 {
1866         struct nvm_tgt_dev *dev = pblk->dev;
1867         struct nvm_geo *geo = &dev->geo;
1868         int pos = pblk_ppa_to_pos(geo, ppa_list[0]);
1869
1870         /* If the LUN has been locked for this same request, do no attempt to
1871          * lock it again
1872          */
1873         if (test_and_set_bit(pos, lun_bitmap))
1874                 return;
1875
1876         __pblk_down_page(pblk, ppa_list, nr_ppas, pos);
1877 }
1878
1879 void pblk_up_page(struct pblk *pblk, struct ppa_addr *ppa_list, int nr_ppas)
1880 {
1881         struct nvm_tgt_dev *dev = pblk->dev;
1882         struct nvm_geo *geo = &dev->geo;
1883         struct pblk_lun *rlun;
1884         int pos = pblk_ppa_to_pos(geo, ppa_list[0]);
1885
1886 #ifdef CONFIG_NVM_PBLK_DEBUG
1887         int i;
1888
1889         for (i = 1; i < nr_ppas; i++)
1890                 WARN_ON(ppa_list[0].a.lun != ppa_list[i].a.lun ||
1891                                 ppa_list[0].a.ch != ppa_list[i].a.ch);
1892 #endif
1893
1894         rlun = &pblk->luns[pos];
1895         up(&rlun->wr_sem);
1896 }
1897
1898 void pblk_up_rq(struct pblk *pblk, struct ppa_addr *ppa_list, int nr_ppas,
1899                 unsigned long *lun_bitmap)
1900 {
1901         struct nvm_tgt_dev *dev = pblk->dev;
1902         struct nvm_geo *geo = &dev->geo;
1903         struct pblk_lun *rlun;
1904         int num_lun = geo->all_luns;
1905         int bit = -1;
1906
1907         while ((bit = find_next_bit(lun_bitmap, num_lun, bit + 1)) < num_lun) {
1908                 rlun = &pblk->luns[bit];
1909                 up(&rlun->wr_sem);
1910         }
1911 }
1912
1913 void pblk_update_map(struct pblk *pblk, sector_t lba, struct ppa_addr ppa)
1914 {
1915         struct ppa_addr ppa_l2p;
1916
1917         /* logic error: lba out-of-bounds. Ignore update */
1918         if (!(lba < pblk->rl.nr_secs)) {
1919                 WARN(1, "pblk: corrupted L2P map request\n");
1920                 return;
1921         }
1922
1923         spin_lock(&pblk->trans_lock);
1924         ppa_l2p = pblk_trans_map_get(pblk, lba);
1925
1926         if (!pblk_addr_in_cache(ppa_l2p) && !pblk_ppa_empty(ppa_l2p))
1927                 pblk_map_invalidate(pblk, ppa_l2p);
1928
1929         pblk_trans_map_set(pblk, lba, ppa);
1930         spin_unlock(&pblk->trans_lock);
1931 }
1932
1933 void pblk_update_map_cache(struct pblk *pblk, sector_t lba, struct ppa_addr ppa)
1934 {
1935
1936 #ifdef CONFIG_NVM_PBLK_DEBUG
1937         /* Callers must ensure that the ppa points to a cache address */
1938         BUG_ON(!pblk_addr_in_cache(ppa));
1939         BUG_ON(pblk_rb_pos_oob(&pblk->rwb, pblk_addr_to_cacheline(ppa)));
1940 #endif
1941
1942         pblk_update_map(pblk, lba, ppa);
1943 }
1944
1945 int pblk_update_map_gc(struct pblk *pblk, sector_t lba, struct ppa_addr ppa_new,
1946                        struct pblk_line *gc_line, u64 paddr_gc)
1947 {
1948         struct ppa_addr ppa_l2p, ppa_gc;
1949         int ret = 1;
1950
1951 #ifdef CONFIG_NVM_PBLK_DEBUG
1952         /* Callers must ensure that the ppa points to a cache address */
1953         BUG_ON(!pblk_addr_in_cache(ppa_new));
1954         BUG_ON(pblk_rb_pos_oob(&pblk->rwb, pblk_addr_to_cacheline(ppa_new)));
1955 #endif
1956
1957         /* logic error: lba out-of-bounds. Ignore update */
1958         if (!(lba < pblk->rl.nr_secs)) {
1959                 WARN(1, "pblk: corrupted L2P map request\n");
1960                 return 0;
1961         }
1962
1963         spin_lock(&pblk->trans_lock);
1964         ppa_l2p = pblk_trans_map_get(pblk, lba);
1965         ppa_gc = addr_to_gen_ppa(pblk, paddr_gc, gc_line->id);
1966
1967         if (!pblk_ppa_comp(ppa_l2p, ppa_gc)) {
1968                 spin_lock(&gc_line->lock);
1969                 WARN(!test_bit(paddr_gc, gc_line->invalid_bitmap),
1970                                                 "pblk: corrupted GC update");
1971                 spin_unlock(&gc_line->lock);
1972
1973                 ret = 0;
1974                 goto out;
1975         }
1976
1977         pblk_trans_map_set(pblk, lba, ppa_new);
1978 out:
1979         spin_unlock(&pblk->trans_lock);
1980         return ret;
1981 }
1982
1983 void pblk_update_map_dev(struct pblk *pblk, sector_t lba,
1984                          struct ppa_addr ppa_mapped, struct ppa_addr ppa_cache)
1985 {
1986         struct ppa_addr ppa_l2p;
1987
1988 #ifdef CONFIG_NVM_PBLK_DEBUG
1989         /* Callers must ensure that the ppa points to a device address */
1990         BUG_ON(pblk_addr_in_cache(ppa_mapped));
1991 #endif
1992         /* Invalidate and discard padded entries */
1993         if (lba == ADDR_EMPTY) {
1994                 atomic64_inc(&pblk->pad_wa);
1995 #ifdef CONFIG_NVM_PBLK_DEBUG
1996                 atomic_long_inc(&pblk->padded_wb);
1997 #endif
1998                 if (!pblk_ppa_empty(ppa_mapped))
1999                         pblk_map_invalidate(pblk, ppa_mapped);
2000                 return;
2001         }
2002
2003         /* logic error: lba out-of-bounds. Ignore update */
2004         if (!(lba < pblk->rl.nr_secs)) {
2005                 WARN(1, "pblk: corrupted L2P map request\n");
2006                 return;
2007         }
2008
2009         spin_lock(&pblk->trans_lock);
2010         ppa_l2p = pblk_trans_map_get(pblk, lba);
2011
2012         /* Do not update L2P if the cacheline has been updated. In this case,
2013          * the mapped ppa must be invalidated
2014          */
2015         if (!pblk_ppa_comp(ppa_l2p, ppa_cache)) {
2016                 if (!pblk_ppa_empty(ppa_mapped))
2017                         pblk_map_invalidate(pblk, ppa_mapped);
2018                 goto out;
2019         }
2020
2021 #ifdef CONFIG_NVM_PBLK_DEBUG
2022         WARN_ON(!pblk_addr_in_cache(ppa_l2p) && !pblk_ppa_empty(ppa_l2p));
2023 #endif
2024
2025         pblk_trans_map_set(pblk, lba, ppa_mapped);
2026 out:
2027         spin_unlock(&pblk->trans_lock);
2028 }
2029
2030 void pblk_lookup_l2p_seq(struct pblk *pblk, struct ppa_addr *ppas,
2031                          sector_t blba, int nr_secs)
2032 {
2033         int i;
2034
2035         spin_lock(&pblk->trans_lock);
2036         for (i = 0; i < nr_secs; i++) {
2037                 struct ppa_addr ppa;
2038
2039                 ppa = ppas[i] = pblk_trans_map_get(pblk, blba + i);
2040
2041                 /* If the L2P entry maps to a line, the reference is valid */
2042                 if (!pblk_ppa_empty(ppa) && !pblk_addr_in_cache(ppa)) {
2043                         int line_id = pblk_ppa_to_line(ppa);
2044                         struct pblk_line *line = &pblk->lines[line_id];
2045
2046                         kref_get(&line->ref);
2047                 }
2048         }
2049         spin_unlock(&pblk->trans_lock);
2050 }
2051
2052 void pblk_lookup_l2p_rand(struct pblk *pblk, struct ppa_addr *ppas,
2053                           u64 *lba_list, int nr_secs)
2054 {
2055         u64 lba;
2056         int i;
2057
2058         spin_lock(&pblk->trans_lock);
2059         for (i = 0; i < nr_secs; i++) {
2060                 lba = lba_list[i];
2061                 if (lba != ADDR_EMPTY) {
2062                         /* logic error: lba out-of-bounds. Ignore update */
2063                         if (!(lba < pblk->rl.nr_secs)) {
2064                                 WARN(1, "pblk: corrupted L2P map request\n");
2065                                 continue;
2066                         }
2067                         ppas[i] = pblk_trans_map_get(pblk, lba);
2068                 }
2069         }
2070         spin_unlock(&pblk->trans_lock);
2071 }