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;
 890         int ret = 0;
 891
 892         memset(&rqd, 0, sizeof(struct nvm_rq));
 893
 894         pblk_setup_e_rq(pblk, &rqd, ppa);
 895
 896         /* The write thread schedules erases so that it minimizes disturbances
 897          * with writes. Thus, there is no need to take the LUN semaphore.
 898          */
 899         ret = pblk_submit_io_sync(pblk, &rqd);
 900         if (ret) {
 901                 struct nvm_tgt_dev *dev = pblk->dev;
 902                 struct nvm_geo *geo = &dev->geo;
 903
 904                 pblk_err(pblk, "could not sync erase line:%d,blk:%d\n",
 905                                         pblk_ppa_to_line(ppa),
 906                                         pblk_ppa_to_pos(geo, ppa));
 907
 908                 rqd.error = ret;
 909                 goto out;
 910         }
 911
 912 out:
 913         rqd.private = pblk;
 914         __pblk_end_io_erase(pblk, &rqd);
 915
 916         return ret;
 917 }
 918
 919 int pblk_line_erase(struct pblk *pblk, struct pblk_line *line)
 920 {
 921         struct pblk_line_meta *lm = &pblk->lm;
 922         struct ppa_addr ppa;
 923         int ret, bit = -1;
 924
 925         /* Erase only good blocks, one at a time */
 926         do {
 927                 spin_lock(&line->lock);
 928                 bit = find_next_zero_bit(line->erase_bitmap, lm->blk_per_line,
 929                                                                 bit + 1);
 930                 if (bit >= lm->blk_per_line) {
 931                         spin_unlock(&line->lock);
 932                         break;
 933                 }
 934
 935                 ppa = pblk->luns[bit].bppa; /* set ch and lun */
 936                 ppa.a.blk = line->id;
 937
 938                 atomic_dec(&line->left_eblks);
 939                 WARN_ON(test_and_set_bit(bit, line->erase_bitmap));
 940                 spin_unlock(&line->lock);
 941
 942                 ret = pblk_blk_erase_sync(pblk, ppa);
 943                 if (ret) {
 944                         pblk_err(pblk, "failed to erase line %d\n", line->id);
 945                         return ret;
 946                 }
 947         } while (1);
 948
 949         return 0;
 950 }
 951
 952 static void pblk_line_setup_metadata(struct pblk_line *line,
 953                                      struct pblk_line_mgmt *l_mg,
 954                                      struct pblk_line_meta *lm)
 955 {
 956         int meta_line;
 957
 958         lockdep_assert_held(&l_mg->free_lock);
 959
 960 retry_meta:
 961         meta_line = find_first_zero_bit(&l_mg->meta_bitmap, PBLK_DATA_LINES);
 962         if (meta_line == PBLK_DATA_LINES) {
 963                 spin_unlock(&l_mg->free_lock);
 964                 io_schedule();
 965                 spin_lock(&l_mg->free_lock);
 966                 goto retry_meta;
 967         }
 968
 969         set_bit(meta_line, &l_mg->meta_bitmap);
 970         line->meta_line = meta_line;
 971
 972         line->smeta = l_mg->sline_meta[meta_line];
 973         line->emeta = l_mg->eline_meta[meta_line];
 974
 975         memset(line->smeta, 0, lm->smeta_len);
 976         memset(line->emeta->buf, 0, lm->emeta_len[0]);
 977
 978         line->emeta->mem = 0;
 979         atomic_set(&line->emeta->sync, 0);
 980 }
 981
 982 /* For now lines are always assumed full lines. Thus, smeta former and current
 983  * lun bitmaps are omitted.
 984  */
 985 static int pblk_line_init_metadata(struct pblk *pblk, struct pblk_line *line,
 986                                   struct pblk_line *cur)
 987 {
 988         struct nvm_tgt_dev *dev = pblk->dev;
 989         struct nvm_geo *geo = &dev->geo;
 990         struct pblk_line_meta *lm = &pblk->lm;
 991         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
 992         struct pblk_emeta *emeta = line->emeta;
 993         struct line_emeta *emeta_buf = emeta->buf;
 994         struct line_smeta *smeta_buf = (struct line_smeta *)line->smeta;
 995         int nr_blk_line;
 996
 997         /* After erasing the line, new bad blocks might appear and we risk
 998          * having an invalid line
 999          */
1000         nr_blk_line = lm->blk_per_line -
1001                         bitmap_weight(line->blk_bitmap, lm->blk_per_line);
1002         if (nr_blk_line < lm->min_blk_line) {
1003                 spin_lock(&l_mg->free_lock);
1004                 spin_lock(&line->lock);
1005                 line->state = PBLK_LINESTATE_BAD;
1006                 spin_unlock(&line->lock);
1007
1008                 list_add_tail(&line->list, &l_mg->bad_list);
1009                 spin_unlock(&l_mg->free_lock);
1010
1011                 pblk_debug(pblk, "line %d is bad\n", line->id);
1012
1013                 return 0;
1014         }
1015
1016         /* Run-time metadata */
1017         line->lun_bitmap = ((void *)(smeta_buf)) + sizeof(struct line_smeta);
1018
1019         /* Mark LUNs allocated in this line (all for now) */
1020         bitmap_set(line->lun_bitmap, 0, lm->lun_bitmap_len);
1021
1022         smeta_buf->header.identifier = cpu_to_le32(PBLK_MAGIC);
1023         memcpy(smeta_buf->header.uuid, pblk->instance_uuid, 16);
1024         smeta_buf->header.id = cpu_to_le32(line->id);
1025         smeta_buf->header.type = cpu_to_le16(line->type);
1026         smeta_buf->header.version_major = SMETA_VERSION_MAJOR;
1027         smeta_buf->header.version_minor = SMETA_VERSION_MINOR;
1028
1029         /* Start metadata */
1030         smeta_buf->seq_nr = cpu_to_le64(line->seq_nr);
1031         smeta_buf->window_wr_lun = cpu_to_le32(geo->all_luns);
1032
1033         /* Fill metadata among lines */
1034         if (cur) {
1035                 memcpy(line->lun_bitmap, cur->lun_bitmap, lm->lun_bitmap_len);
1036                 smeta_buf->prev_id = cpu_to_le32(cur->id);
1037                 cur->emeta->buf->next_id = cpu_to_le32(line->id);
1038         } else {
1039                 smeta_buf->prev_id = cpu_to_le32(PBLK_LINE_EMPTY);
1040         }
1041
1042         /* All smeta must be set at this point */
1043         smeta_buf->header.crc = cpu_to_le32(
1044                         pblk_calc_meta_header_crc(pblk, &smeta_buf->header));
1045         smeta_buf->crc = cpu_to_le32(pblk_calc_smeta_crc(pblk, smeta_buf));
1046
1047         /* End metadata */
1048         memcpy(&emeta_buf->header, &smeta_buf->header,
1049                                                 sizeof(struct line_header));
1050
1051         emeta_buf->header.version_major = EMETA_VERSION_MAJOR;
1052         emeta_buf->header.version_minor = EMETA_VERSION_MINOR;
1053         emeta_buf->header.crc = cpu_to_le32(
1054                         pblk_calc_meta_header_crc(pblk, &emeta_buf->header));
1055
1056         emeta_buf->seq_nr = cpu_to_le64(line->seq_nr);
1057         emeta_buf->nr_lbas = cpu_to_le64(line->sec_in_line);
1058         emeta_buf->nr_valid_lbas = cpu_to_le64(0);
1059         emeta_buf->next_id = cpu_to_le32(PBLK_LINE_EMPTY);
1060         emeta_buf->crc = cpu_to_le32(0);
1061         emeta_buf->prev_id = smeta_buf->prev_id;
1062
1063         return 1;
1064 }
1065
1066 static int pblk_line_alloc_bitmaps(struct pblk *pblk, struct pblk_line *line)
1067 {
1068         struct pblk_line_meta *lm = &pblk->lm;
1069
1070         line->map_bitmap = kzalloc(lm->sec_bitmap_len, GFP_KERNEL);
1071         if (!line->map_bitmap)
1072                 return -ENOMEM;
1073
1074         /* will be initialized using bb info from map_bitmap */
1075         line->invalid_bitmap = kmalloc(lm->sec_bitmap_len, GFP_KERNEL);
1076         if (!line->invalid_bitmap) {
1077                 kfree(line->map_bitmap);
1078                 line->map_bitmap = NULL;
1079                 return -ENOMEM;
1080         }
1081
1082         return 0;
1083 }
1084
1085 /* For now lines are always assumed full lines. Thus, smeta former and current
1086  * lun bitmaps are omitted.
1087  */
1088 static int pblk_line_init_bb(struct pblk *pblk, struct pblk_line *line,
1089                              int init)
1090 {
1091         struct nvm_tgt_dev *dev = pblk->dev;
1092         struct nvm_geo *geo = &dev->geo;
1093         struct pblk_line_meta *lm = &pblk->lm;
1094         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1095         u64 off;
1096         int bit = -1;
1097         int emeta_secs;
1098
1099         line->sec_in_line = lm->sec_per_line;
1100
1101         /* Capture bad block information on line mapping bitmaps */
1102         while ((bit = find_next_bit(line->blk_bitmap, lm->blk_per_line,
1103                                         bit + 1)) < lm->blk_per_line) {
1104                 off = bit * geo->ws_opt;
1105                 bitmap_shift_left(l_mg->bb_aux, l_mg->bb_template, off,
1106                                                         lm->sec_per_line);
1107                 bitmap_or(line->map_bitmap, line->map_bitmap, l_mg->bb_aux,
1108                                                         lm->sec_per_line);
1109                 line->sec_in_line -= geo->clba;
1110         }
1111
1112         /* Mark smeta metadata sectors as bad sectors */
1113         bit = find_first_zero_bit(line->blk_bitmap, lm->blk_per_line);
1114         off = bit * geo->ws_opt;
1115         bitmap_set(line->map_bitmap, off, lm->smeta_sec);
1116         line->sec_in_line -= lm->smeta_sec;
1117         line->smeta_ssec = off;
1118         line->cur_sec = off + lm->smeta_sec;
1119
1120         if (init && pblk_line_submit_smeta_io(pblk, line, off, PBLK_WRITE)) {
1121                 pblk_debug(pblk, "line smeta I/O failed. Retry\n");
1122                 return 0;
1123         }
1124
1125         bitmap_copy(line->invalid_bitmap, line->map_bitmap, lm->sec_per_line);
1126
1127         /* Mark emeta metadata sectors as bad sectors. We need to consider bad
1128          * blocks to make sure that there are enough sectors to store emeta
1129          */
1130         emeta_secs = lm->emeta_sec[0];
1131         off = lm->sec_per_line;
1132         while (emeta_secs) {
1133                 off -= geo->ws_opt;
1134                 if (!test_bit(off, line->invalid_bitmap)) {
1135                         bitmap_set(line->invalid_bitmap, off, geo->ws_opt);
1136                         emeta_secs -= geo->ws_opt;
1137                 }
1138         }
1139
1140         line->emeta_ssec = off;
1141         line->sec_in_line -= lm->emeta_sec[0];
1142         line->nr_valid_lbas = 0;
1143         line->left_msecs = line->sec_in_line;
1144         *line->vsc = cpu_to_le32(line->sec_in_line);
1145
1146         if (lm->sec_per_line - line->sec_in_line !=
1147                 bitmap_weight(line->invalid_bitmap, lm->sec_per_line)) {
1148                 spin_lock(&line->lock);
1149                 line->state = PBLK_LINESTATE_BAD;
1150                 spin_unlock(&line->lock);
1151
1152                 list_add_tail(&line->list, &l_mg->bad_list);
1153                 pblk_err(pblk, "unexpected line %d is bad\n", line->id);
1154
1155                 return 0;
1156         }
1157
1158         return 1;
1159 }
1160
1161 static int pblk_prepare_new_line(struct pblk *pblk, struct pblk_line *line)
1162 {
1163         struct pblk_line_meta *lm = &pblk->lm;
1164         struct nvm_tgt_dev *dev = pblk->dev;
1165         struct nvm_geo *geo = &dev->geo;
1166         int blk_to_erase = atomic_read(&line->blk_in_line);
1167         int i;
1168
1169         for (i = 0; i < lm->blk_per_line; i++) {
1170                 struct pblk_lun *rlun = &pblk->luns[i];
1171                 int pos = pblk_ppa_to_pos(geo, rlun->bppa);
1172                 int state = line->chks[pos].state;
1173
1174                 /* Free chunks should not be erased */
1175                 if (state & NVM_CHK_ST_FREE) {
1176                         set_bit(pblk_ppa_to_pos(geo, rlun->bppa),
1177                                                         line->erase_bitmap);
1178                         blk_to_erase--;
1179                 }
1180         }
1181
1182         return blk_to_erase;
1183 }
1184
1185 static int pblk_line_prepare(struct pblk *pblk, struct pblk_line *line)
1186 {
1187         struct pblk_line_meta *lm = &pblk->lm;
1188         int blk_in_line = atomic_read(&line->blk_in_line);
1189         int blk_to_erase;
1190
1191         /* Bad blocks do not need to be erased */
1192         bitmap_copy(line->erase_bitmap, line->blk_bitmap, lm->blk_per_line);
1193
1194         spin_lock(&line->lock);
1195
1196         /* If we have not written to this line, we need to mark up free chunks
1197          * as already erased
1198          */
1199         if (line->state == PBLK_LINESTATE_NEW) {
1200                 blk_to_erase = pblk_prepare_new_line(pblk, line);
1201                 line->state = PBLK_LINESTATE_FREE;
1202         } else {
1203                 blk_to_erase = blk_in_line;
1204         }
1205
1206         if (blk_in_line < lm->min_blk_line) {
1207                 spin_unlock(&line->lock);
1208                 return -EAGAIN;
1209         }
1210
1211         if (line->state != PBLK_LINESTATE_FREE) {
1212                 WARN(1, "pblk: corrupted line %d, state %d\n",
1213                                                         line->id, line->state);
1214                 spin_unlock(&line->lock);
1215                 return -EINTR;
1216         }
1217
1218         line->state = PBLK_LINESTATE_OPEN;
1219
1220         atomic_set(&line->left_eblks, blk_to_erase);
1221         atomic_set(&line->left_seblks, blk_to_erase);
1222
1223         line->meta_distance = lm->meta_distance;
1224         spin_unlock(&line->lock);
1225
1226         kref_init(&line->ref);
1227
1228         return 0;
1229 }
1230
1231 int pblk_line_recov_alloc(struct pblk *pblk, struct pblk_line *line)
1232 {
1233         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1234         int ret;
1235
1236         spin_lock(&l_mg->free_lock);
1237         l_mg->data_line = line;
1238         list_del(&line->list);
1239
1240         ret = pblk_line_prepare(pblk, line);
1241         if (ret) {
1242                 list_add(&line->list, &l_mg->free_list);
1243                 spin_unlock(&l_mg->free_lock);
1244                 return ret;
1245         }
1246         spin_unlock(&l_mg->free_lock);
1247
1248         ret = pblk_line_alloc_bitmaps(pblk, line);
1249         if (ret)
1250                 return ret;
1251
1252         if (!pblk_line_init_bb(pblk, line, 0)) {
1253                 list_add(&line->list, &l_mg->free_list);
1254                 return -EINTR;
1255         }
1256
1257         pblk_rl_free_lines_dec(&pblk->rl, line, true);
1258         return 0;
1259 }
1260
1261 void pblk_line_recov_close(struct pblk *pblk, struct pblk_line *line)
1262 {
1263         kfree(line->map_bitmap);
1264         line->map_bitmap = NULL;
1265         line->smeta = NULL;
1266         line->emeta = NULL;
1267 }
1268
1269 static void pblk_line_reinit(struct pblk_line *line)
1270 {
1271         *line->vsc = cpu_to_le32(EMPTY_ENTRY);
1272
1273         line->map_bitmap = NULL;
1274         line->invalid_bitmap = NULL;
1275         line->smeta = NULL;
1276         line->emeta = NULL;
1277 }
1278
1279 void pblk_line_free(struct pblk_line *line)
1280 {
1281         kfree(line->map_bitmap);
1282         kfree(line->invalid_bitmap);
1283
1284         pblk_line_reinit(line);
1285 }
1286
1287 struct pblk_line *pblk_line_get(struct pblk *pblk)
1288 {
1289         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1290         struct pblk_line_meta *lm = &pblk->lm;
1291         struct pblk_line *line;
1292         int ret, bit;
1293
1294         lockdep_assert_held(&l_mg->free_lock);
1295
1296 retry:
1297         if (list_empty(&l_mg->free_list)) {
1298                 pblk_err(pblk, "no free lines\n");
1299                 return NULL;
1300         }
1301
1302         line = list_first_entry(&l_mg->free_list, struct pblk_line, list);
1303         list_del(&line->list);
1304         l_mg->nr_free_lines--;
1305
1306         bit = find_first_zero_bit(line->blk_bitmap, lm->blk_per_line);
1307         if (unlikely(bit >= lm->blk_per_line)) {
1308                 spin_lock(&line->lock);
1309                 line->state = PBLK_LINESTATE_BAD;
1310                 spin_unlock(&line->lock);
1311
1312                 list_add_tail(&line->list, &l_mg->bad_list);
1313
1314                 pblk_debug(pblk, "line %d is bad\n", line->id);
1315                 goto retry;
1316         }
1317
1318         ret = pblk_line_prepare(pblk, line);
1319         if (ret) {
1320                 switch (ret) {
1321                 case -EAGAIN:
1322                         list_add(&line->list, &l_mg->bad_list);
1323                         goto retry;
1324                 case -EINTR:
1325                         list_add(&line->list, &l_mg->corrupt_list);
1326                         goto retry;
1327                 default:
1328                         pblk_err(pblk, "failed to prepare line %d\n", line->id);
1329                         list_add(&line->list, &l_mg->free_list);
1330                         l_mg->nr_free_lines++;
1331                         return NULL;
1332                 }
1333         }
1334
1335         return line;
1336 }
1337
1338 static struct pblk_line *pblk_line_retry(struct pblk *pblk,
1339                                          struct pblk_line *line)
1340 {
1341         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1342         struct pblk_line *retry_line;
1343
1344 retry:
1345         spin_lock(&l_mg->free_lock);
1346         retry_line = pblk_line_get(pblk);
1347         if (!retry_line) {
1348                 l_mg->data_line = NULL;
1349                 spin_unlock(&l_mg->free_lock);
1350                 return NULL;
1351         }
1352
1353         retry_line->map_bitmap = line->map_bitmap;
1354         retry_line->invalid_bitmap = line->invalid_bitmap;
1355         retry_line->smeta = line->smeta;
1356         retry_line->emeta = line->emeta;
1357         retry_line->meta_line = line->meta_line;
1358
1359         pblk_line_reinit(line);
1360
1361         l_mg->data_line = retry_line;
1362         spin_unlock(&l_mg->free_lock);
1363
1364         pblk_rl_free_lines_dec(&pblk->rl, line, false);
1365
1366         if (pblk_line_erase(pblk, retry_line))
1367                 goto retry;
1368
1369         return retry_line;
1370 }
1371
1372 static void pblk_set_space_limit(struct pblk *pblk)
1373 {
1374         struct pblk_rl *rl = &pblk->rl;
1375
1376         atomic_set(&rl->rb_space, 0);
1377 }
1378
1379 struct pblk_line *pblk_line_get_first_data(struct pblk *pblk)
1380 {
1381         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1382         struct pblk_line *line;
1383
1384         spin_lock(&l_mg->free_lock);
1385         line = pblk_line_get(pblk);
1386         if (!line) {
1387                 spin_unlock(&l_mg->free_lock);
1388                 return NULL;
1389         }
1390
1391         line->seq_nr = l_mg->d_seq_nr++;
1392         line->type = PBLK_LINETYPE_DATA;
1393         l_mg->data_line = line;
1394
1395         pblk_line_setup_metadata(line, l_mg, &pblk->lm);
1396
1397         /* Allocate next line for preparation */
1398         l_mg->data_next = pblk_line_get(pblk);
1399         if (!l_mg->data_next) {
1400                 /* If we cannot get a new line, we need to stop the pipeline.
1401                  * Only allow as many writes in as we can store safely and then
1402                  * fail gracefully
1403                  */
1404                 pblk_set_space_limit(pblk);
1405
1406                 l_mg->data_next = NULL;
1407         } else {
1408                 l_mg->data_next->seq_nr = l_mg->d_seq_nr++;
1409                 l_mg->data_next->type = PBLK_LINETYPE_DATA;
1410         }
1411         spin_unlock(&l_mg->free_lock);
1412
1413         if (pblk_line_alloc_bitmaps(pblk, line))
1414                 return NULL;
1415
1416         if (pblk_line_erase(pblk, line)) {
1417                 line = pblk_line_retry(pblk, line);
1418                 if (!line)
1419                         return NULL;
1420         }
1421
1422 retry_setup:
1423         if (!pblk_line_init_metadata(pblk, line, NULL)) {
1424                 line = pblk_line_retry(pblk, line);
1425                 if (!line)
1426                         return NULL;
1427
1428                 goto retry_setup;
1429         }
1430
1431         if (!pblk_line_init_bb(pblk, line, 1)) {
1432                 line = pblk_line_retry(pblk, line);
1433                 if (!line)
1434                         return NULL;
1435
1436                 goto retry_setup;
1437         }
1438
1439         pblk_rl_free_lines_dec(&pblk->rl, line, true);
1440
1441         return line;
1442 }
1443
1444 static void pblk_stop_writes(struct pblk *pblk, struct pblk_line *line)
1445 {
1446         lockdep_assert_held(&pblk->l_mg.free_lock);
1447
1448         pblk_set_space_limit(pblk);
1449         pblk->state = PBLK_STATE_STOPPING;
1450 }
1451
1452 static void pblk_line_close_meta_sync(struct pblk *pblk)
1453 {
1454         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1455         struct pblk_line_meta *lm = &pblk->lm;
1456         struct pblk_line *line, *tline;
1457         LIST_HEAD(list);
1458
1459         spin_lock(&l_mg->close_lock);
1460         if (list_empty(&l_mg->emeta_list)) {
1461                 spin_unlock(&l_mg->close_lock);
1462                 return;
1463         }
1464
1465         list_cut_position(&list, &l_mg->emeta_list, l_mg->emeta_list.prev);
1466         spin_unlock(&l_mg->close_lock);
1467
1468         list_for_each_entry_safe(line, tline, &list, list) {
1469                 struct pblk_emeta *emeta = line->emeta;
1470
1471                 while (emeta->mem < lm->emeta_len[0]) {
1472                         int ret;
1473
1474                         ret = pblk_submit_meta_io(pblk, line);
1475                         if (ret) {
1476                                 pblk_err(pblk, "sync meta line %d failed (%d)\n",
1477                                                         line->id, ret);
1478                                 return;
1479                         }
1480                 }
1481         }
1482
1483         pblk_wait_for_meta(pblk);
1484         flush_workqueue(pblk->close_wq);
1485 }
1486
1487 void __pblk_pipeline_flush(struct pblk *pblk)
1488 {
1489         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1490         int ret;
1491
1492         spin_lock(&l_mg->free_lock);
1493         if (pblk->state == PBLK_STATE_RECOVERING ||
1494                                         pblk->state == PBLK_STATE_STOPPED) {
1495                 spin_unlock(&l_mg->free_lock);
1496                 return;
1497         }
1498         pblk->state = PBLK_STATE_RECOVERING;
1499         spin_unlock(&l_mg->free_lock);
1500
1501         pblk_flush_writer(pblk);
1502         pblk_wait_for_meta(pblk);
1503
1504         ret = pblk_recov_pad(pblk);
1505         if (ret) {
1506                 pblk_err(pblk, "could not close data on teardown(%d)\n", ret);
1507                 return;
1508         }
1509
1510         flush_workqueue(pblk->bb_wq);
1511         pblk_line_close_meta_sync(pblk);
1512 }
1513
1514 void __pblk_pipeline_stop(struct pblk *pblk)
1515 {
1516         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1517
1518         spin_lock(&l_mg->free_lock);
1519         pblk->state = PBLK_STATE_STOPPED;
1520         l_mg->data_line = NULL;
1521         l_mg->data_next = NULL;
1522         spin_unlock(&l_mg->free_lock);
1523 }
1524
1525 void pblk_pipeline_stop(struct pblk *pblk)
1526 {
1527         __pblk_pipeline_flush(pblk);
1528         __pblk_pipeline_stop(pblk);
1529 }
1530
1531 struct pblk_line *pblk_line_replace_data(struct pblk *pblk)
1532 {
1533         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1534         struct pblk_line *cur, *new = NULL;
1535         unsigned int left_seblks;
1536
1537         cur = l_mg->data_line;
1538         new = l_mg->data_next;
1539         if (!new)
1540                 goto out;
1541         l_mg->data_line = new;
1542
1543         spin_lock(&l_mg->free_lock);
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         emeta_buf->nr_valid_lbas = cpu_to_le64(line->nr_valid_lbas);
1777         emeta_buf->crc = cpu_to_le32(pblk_calc_emeta_crc(pblk, emeta_buf));
1778
1779         spin_lock(&l_mg->close_lock);
1780         spin_lock(&line->lock);
1781
1782         /* Update the in-memory start address for emeta, in case it has
1783          * shifted due to write errors
1784          */
1785         if (line->emeta_ssec != line->cur_sec)
1786                 line->emeta_ssec = line->cur_sec;
1787
1788         list_add_tail(&line->list, &l_mg->emeta_list);
1789         spin_unlock(&line->lock);
1790         spin_unlock(&l_mg->close_lock);
1791
1792         pblk_line_should_sync_meta(pblk);
1793
1794
1795 }
1796
1797 static void pblk_save_lba_list(struct pblk *pblk, struct pblk_line *line)
1798 {
1799         struct pblk_line_meta *lm = &pblk->lm;
1800         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1801         unsigned int lba_list_size = lm->emeta_len[2];
1802         struct pblk_w_err_gc *w_err_gc = line->w_err_gc;
1803         struct pblk_emeta *emeta = line->emeta;
1804
1805         w_err_gc->lba_list = pblk_malloc(lba_list_size,
1806                                          l_mg->emeta_alloc_type, GFP_KERNEL);
1807         memcpy(w_err_gc->lba_list, emeta_to_lbas(pblk, emeta->buf),
1808                                 lba_list_size);
1809 }
1810
1811 void pblk_line_close_ws(struct work_struct *work)
1812 {
1813         struct pblk_line_ws *line_ws = container_of(work, struct pblk_line_ws,
1814                                                                         ws);
1815         struct pblk *pblk = line_ws->pblk;
1816         struct pblk_line *line = line_ws->line;
1817         struct pblk_w_err_gc *w_err_gc = line->w_err_gc;
1818
1819         /* Write errors makes the emeta start address stored in smeta invalid,
1820          * so keep a copy of the lba list until we've gc'd the line
1821          */
1822         if (w_err_gc->has_write_err)
1823                 pblk_save_lba_list(pblk, line);
1824
1825         pblk_line_close(pblk, line);
1826         mempool_free(line_ws, &pblk->gen_ws_pool);
1827 }
1828
1829 void pblk_gen_run_ws(struct pblk *pblk, struct pblk_line *line, void *priv,
1830                       void (*work)(struct work_struct *), gfp_t gfp_mask,
1831                       struct workqueue_struct *wq)
1832 {
1833         struct pblk_line_ws *line_ws;
1834
1835         line_ws = mempool_alloc(&pblk->gen_ws_pool, gfp_mask);
1836
1837         line_ws->pblk = pblk;
1838         line_ws->line = line;
1839         line_ws->priv = priv;
1840
1841         INIT_WORK(&line_ws->ws, work);
1842         queue_work(wq, &line_ws->ws);
1843 }
1844
1845 static void __pblk_down_page(struct pblk *pblk, struct ppa_addr *ppa_list,
1846                              int nr_ppas, int pos)
1847 {
1848         struct pblk_lun *rlun = &pblk->luns[pos];
1849         int ret;
1850
1851         /*
1852          * Only send one inflight I/O per LUN. Since we map at a page
1853          * granurality, all ppas in the I/O will map to the same LUN
1854          */
1855 #ifdef CONFIG_NVM_PBLK_DEBUG
1856         int i;
1857
1858         for (i = 1; i < nr_ppas; i++)
1859                 WARN_ON(ppa_list[0].a.lun != ppa_list[i].a.lun ||
1860                                 ppa_list[0].a.ch != ppa_list[i].a.ch);
1861 #endif
1862
1863         ret = down_timeout(&rlun->wr_sem, msecs_to_jiffies(30000));
1864         if (ret == -ETIME || ret == -EINTR)
1865                 pblk_err(pblk, "taking lun semaphore timed out: err %d\n",
1866                                 -ret);
1867 }
1868
1869 void pblk_down_page(struct pblk *pblk, struct ppa_addr *ppa_list, int nr_ppas)
1870 {
1871         struct nvm_tgt_dev *dev = pblk->dev;
1872         struct nvm_geo *geo = &dev->geo;
1873         int pos = pblk_ppa_to_pos(geo, ppa_list[0]);
1874
1875         __pblk_down_page(pblk, ppa_list, nr_ppas, pos);
1876 }
1877
1878 void pblk_down_rq(struct pblk *pblk, struct ppa_addr *ppa_list, int nr_ppas,
1879                   unsigned long *lun_bitmap)
1880 {
1881         struct nvm_tgt_dev *dev = pblk->dev;
1882         struct nvm_geo *geo = &dev->geo;
1883         int pos = pblk_ppa_to_pos(geo, ppa_list[0]);
1884
1885         /* If the LUN has been locked for this same request, do no attempt to
1886          * lock it again
1887          */
1888         if (test_and_set_bit(pos, lun_bitmap))
1889                 return;
1890
1891         __pblk_down_page(pblk, ppa_list, nr_ppas, pos);
1892 }
1893
1894 void pblk_up_page(struct pblk *pblk, struct ppa_addr *ppa_list, int nr_ppas)
1895 {
1896         struct nvm_tgt_dev *dev = pblk->dev;
1897         struct nvm_geo *geo = &dev->geo;
1898         struct pblk_lun *rlun;
1899         int pos = pblk_ppa_to_pos(geo, ppa_list[0]);
1900
1901 #ifdef CONFIG_NVM_PBLK_DEBUG
1902         int i;
1903
1904         for (i = 1; i < nr_ppas; i++)
1905                 WARN_ON(ppa_list[0].a.lun != ppa_list[i].a.lun ||
1906                                 ppa_list[0].a.ch != ppa_list[i].a.ch);
1907 #endif
1908
1909         rlun = &pblk->luns[pos];
1910         up(&rlun->wr_sem);
1911 }
1912
1913 void pblk_up_rq(struct pblk *pblk, struct ppa_addr *ppa_list, int nr_ppas,
1914                 unsigned long *lun_bitmap)
1915 {
1916         struct nvm_tgt_dev *dev = pblk->dev;
1917         struct nvm_geo *geo = &dev->geo;
1918         struct pblk_lun *rlun;
1919         int num_lun = geo->all_luns;
1920         int bit = -1;
1921
1922         while ((bit = find_next_bit(lun_bitmap, num_lun, bit + 1)) < num_lun) {
1923                 rlun = &pblk->luns[bit];
1924                 up(&rlun->wr_sem);
1925         }
1926 }
1927
1928 void pblk_update_map(struct pblk *pblk, sector_t lba, struct ppa_addr ppa)
1929 {
1930         struct ppa_addr ppa_l2p;
1931
1932         /* logic error: lba out-of-bounds. Ignore update */
1933         if (!(lba < pblk->rl.nr_secs)) {
1934                 WARN(1, "pblk: corrupted L2P map request\n");
1935                 return;
1936         }
1937
1938         spin_lock(&pblk->trans_lock);
1939         ppa_l2p = pblk_trans_map_get(pblk, lba);
1940
1941         if (!pblk_addr_in_cache(ppa_l2p) && !pblk_ppa_empty(ppa_l2p))
1942                 pblk_map_invalidate(pblk, ppa_l2p);
1943
1944         pblk_trans_map_set(pblk, lba, ppa);
1945         spin_unlock(&pblk->trans_lock);
1946 }
1947
1948 void pblk_update_map_cache(struct pblk *pblk, sector_t lba, struct ppa_addr ppa)
1949 {
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));
1954         BUG_ON(pblk_rb_pos_oob(&pblk->rwb, pblk_addr_to_cacheline(ppa)));
1955 #endif
1956
1957         pblk_update_map(pblk, lba, ppa);
1958 }
1959
1960 int pblk_update_map_gc(struct pblk *pblk, sector_t lba, struct ppa_addr ppa_new,
1961                        struct pblk_line *gc_line, u64 paddr_gc)
1962 {
1963         struct ppa_addr ppa_l2p, ppa_gc;
1964         int ret = 1;
1965
1966 #ifdef CONFIG_NVM_PBLK_DEBUG
1967         /* Callers must ensure that the ppa points to a cache address */
1968         BUG_ON(!pblk_addr_in_cache(ppa_new));
1969         BUG_ON(pblk_rb_pos_oob(&pblk->rwb, pblk_addr_to_cacheline(ppa_new)));
1970 #endif
1971
1972         /* logic error: lba out-of-bounds. Ignore update */
1973         if (!(lba < pblk->rl.nr_secs)) {
1974                 WARN(1, "pblk: corrupted L2P map request\n");
1975                 return 0;
1976         }
1977
1978         spin_lock(&pblk->trans_lock);
1979         ppa_l2p = pblk_trans_map_get(pblk, lba);
1980         ppa_gc = addr_to_gen_ppa(pblk, paddr_gc, gc_line->id);
1981
1982         if (!pblk_ppa_comp(ppa_l2p, ppa_gc)) {
1983                 spin_lock(&gc_line->lock);
1984                 WARN(!test_bit(paddr_gc, gc_line->invalid_bitmap),
1985                                                 "pblk: corrupted GC update");
1986                 spin_unlock(&gc_line->lock);
1987
1988                 ret = 0;
1989                 goto out;
1990         }
1991
1992         pblk_trans_map_set(pblk, lba, ppa_new);
1993 out:
1994         spin_unlock(&pblk->trans_lock);
1995         return ret;
1996 }
1997
1998 void pblk_update_map_dev(struct pblk *pblk, sector_t lba,
1999                          struct ppa_addr ppa_mapped, struct ppa_addr ppa_cache)
2000 {
2001         struct ppa_addr ppa_l2p;
2002
2003 #ifdef CONFIG_NVM_PBLK_DEBUG
2004         /* Callers must ensure that the ppa points to a device address */
2005         BUG_ON(pblk_addr_in_cache(ppa_mapped));
2006 #endif
2007         /* Invalidate and discard padded entries */
2008         if (lba == ADDR_EMPTY) {
2009                 atomic64_inc(&pblk->pad_wa);
2010 #ifdef CONFIG_NVM_PBLK_DEBUG
2011                 atomic_long_inc(&pblk->padded_wb);
2012 #endif
2013                 if (!pblk_ppa_empty(ppa_mapped))
2014                         pblk_map_invalidate(pblk, ppa_mapped);
2015                 return;
2016         }
2017
2018         /* logic error: lba out-of-bounds. Ignore update */
2019         if (!(lba < pblk->rl.nr_secs)) {
2020                 WARN(1, "pblk: corrupted L2P map request\n");
2021                 return;
2022         }
2023
2024         spin_lock(&pblk->trans_lock);
2025         ppa_l2p = pblk_trans_map_get(pblk, lba);
2026
2027         /* Do not update L2P if the cacheline has been updated. In this case,
2028          * the mapped ppa must be invalidated
2029          */
2030         if (!pblk_ppa_comp(ppa_l2p, ppa_cache)) {
2031                 if (!pblk_ppa_empty(ppa_mapped))
2032                         pblk_map_invalidate(pblk, ppa_mapped);
2033                 goto out;
2034         }
2035
2036 #ifdef CONFIG_NVM_PBLK_DEBUG
2037         WARN_ON(!pblk_addr_in_cache(ppa_l2p) && !pblk_ppa_empty(ppa_l2p));
2038 #endif
2039
2040         pblk_trans_map_set(pblk, lba, ppa_mapped);
2041 out:
2042         spin_unlock(&pblk->trans_lock);
2043 }
2044
2045 void pblk_lookup_l2p_seq(struct pblk *pblk, struct ppa_addr *ppas,
2046                          sector_t blba, int nr_secs)
2047 {
2048         int i;
2049
2050         spin_lock(&pblk->trans_lock);
2051         for (i = 0; i < nr_secs; i++) {
2052                 struct ppa_addr ppa;
2053
2054                 ppa = ppas[i] = pblk_trans_map_get(pblk, blba + i);
2055
2056                 /* If the L2P entry maps to a line, the reference is valid */
2057                 if (!pblk_ppa_empty(ppa) && !pblk_addr_in_cache(ppa)) {
2058                         int line_id = pblk_ppa_to_line(ppa);
2059                         struct pblk_line *line = &pblk->lines[line_id];
2060
2061                         kref_get(&line->ref);
2062                 }
2063         }
2064         spin_unlock(&pblk->trans_lock);
2065 }
2066
2067 void pblk_lookup_l2p_rand(struct pblk *pblk, struct ppa_addr *ppas,
2068                           u64 *lba_list, int nr_secs)
2069 {
2070         u64 lba;
2071         int i;
2072
2073         spin_lock(&pblk->trans_lock);
2074         for (i = 0; i < nr_secs; i++) {
2075                 lba = lba_list[i];
2076                 if (lba != ADDR_EMPTY) {
2077                         /* logic error: lba out-of-bounds. Ignore update */
2078                         if (!(lba < pblk->rl.nr_secs)) {
2079                                 WARN(1, "pblk: corrupted L2P map request\n");
2080                                 continue;
2081                         }
2082                         ppas[i] = pblk_trans_map_get(pblk, lba);
2083                 }
2084         }
2085         spin_unlock(&pblk->trans_lock);
2086 }