drivers/lightnvm/pblk-core.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * Copyright (C) 2016 CNEX Labs
   4  * Initial release: Javier Gonzalez <[email protected]>
   5  *                  Matias Bjorling <[email protected]>
   6  *
   7  * This program is free software; you can redistribute it and/or
   8  * modify it under the terms of the GNU General Public License version
   9  * 2 as published by the Free Software Foundation.
  10  *
  11  * This program is distributed in the hope that it will be useful, but
  12  * WITHOUT ANY WARRANTY; without even the implied warranty of
  13  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  14  * General Public License for more details.
  15  *
  16  * pblk-core.c - pblk's core functionality
  17  *
  18  */
  19
  20 #define CREATE_TRACE_POINTS
  21
  22 #include "pblk.h"
  23 #include "pblk-trace.h"
  24
  25 static void pblk_line_mark_bb(struct work_struct *work)
  26 {
  27         struct pblk_line_ws *line_ws = container_of(work, struct pblk_line_ws,
  28                                                                         ws);
  29         struct pblk *pblk = line_ws->pblk;
  30         struct nvm_tgt_dev *dev = pblk->dev;
  31         struct ppa_addr *ppa = line_ws->priv;
  32         int ret;
  33
  34         ret = nvm_set_chunk_meta(dev, ppa, 1, NVM_BLK_T_GRWN_BAD);
  35         if (ret) {
  36                 struct pblk_line *line;
  37                 int pos;
  38
  39                 line = pblk_ppa_to_line(pblk, *ppa);
  40                 pos = pblk_ppa_to_pos(&dev->geo, *ppa);
  41
  42                 pblk_err(pblk, "failed to mark bb, line:%d, pos:%d\n",
  43                                 line->id, pos);
  44         }
  45
  46         kfree(ppa);
  47         mempool_free(line_ws, &pblk->gen_ws_pool);
  48 }
  49
  50 static void pblk_mark_bb(struct pblk *pblk, struct pblk_line *line,
  51                          struct ppa_addr ppa_addr)
  52 {
  53         struct nvm_tgt_dev *dev = pblk->dev;
  54         struct nvm_geo *geo = &dev->geo;
  55         struct ppa_addr *ppa;
  56         int pos = pblk_ppa_to_pos(geo, ppa_addr);
  57
  58         pblk_debug(pblk, "erase failed: line:%d, pos:%d\n", line->id, pos);
  59         atomic_long_inc(&pblk->erase_failed);
  60
  61         atomic_dec(&line->blk_in_line);
  62         if (test_and_set_bit(pos, line->blk_bitmap))
  63                 pblk_err(pblk, "attempted to erase bb: line:%d, pos:%d\n",
  64                                                         line->id, pos);
  65
  66         /* Not necessary to mark bad blocks on 2.0 spec. */
  67         if (geo->version == NVM_OCSSD_SPEC_20)
  68                 return;
  69
  70         ppa = kmalloc(sizeof(struct ppa_addr), GFP_ATOMIC);
  71         if (!ppa)
  72                 return;
  73
  74         *ppa = ppa_addr;
  75         pblk_gen_run_ws(pblk, NULL, ppa, pblk_line_mark_bb,
  76                                                 GFP_ATOMIC, pblk->bb_wq);
  77 }
  78
  79 static void __pblk_end_io_erase(struct pblk *pblk, struct nvm_rq *rqd)
  80 {
  81         struct nvm_tgt_dev *dev = pblk->dev;
  82         struct nvm_geo *geo = &dev->geo;
  83         struct nvm_chk_meta *chunk;
  84         struct pblk_line *line;
  85         int pos;
  86
  87         line = pblk_ppa_to_line(pblk, rqd->ppa_addr);
  88         pos = pblk_ppa_to_pos(geo, rqd->ppa_addr);
  89         chunk = &line->chks[pos];
  90
  91         atomic_dec(&line->left_seblks);
  92
  93         if (rqd->error) {
  94                 trace_pblk_chunk_reset(pblk_disk_name(pblk),
  95                                 &rqd->ppa_addr, PBLK_CHUNK_RESET_FAILED);
  96
  97                 chunk->state = NVM_CHK_ST_OFFLINE;
  98                 pblk_mark_bb(pblk, line, rqd->ppa_addr);
  99         } else {
 100                 trace_pblk_chunk_reset(pblk_disk_name(pblk),
 101                                 &rqd->ppa_addr, PBLK_CHUNK_RESET_DONE);
 102
 103                 chunk->state = NVM_CHK_ST_FREE;
 104         }
 105
 106         trace_pblk_chunk_state(pblk_disk_name(pblk), &rqd->ppa_addr,
 107                                 chunk->state);
 108
 109         atomic_dec(&pblk->inflight_io);
 110 }
 111
 112 /* Erase completion assumes that only one block is erased at the time */
 113 static void pblk_end_io_erase(struct nvm_rq *rqd)
 114 {
 115         struct pblk *pblk = rqd->private;
 116
 117         __pblk_end_io_erase(pblk, rqd);
 118         mempool_free(rqd, &pblk->e_rq_pool);
 119 }
 120
 121 /*
 122  * Get information for all chunks from the device.
 123  *
 124  * The caller is responsible for freeing (vmalloc) the returned structure
 125  */
 126 struct nvm_chk_meta *pblk_get_chunk_meta(struct pblk *pblk)
 127 {
 128         struct nvm_tgt_dev *dev = pblk->dev;
 129         struct nvm_geo *geo = &dev->geo;
 130         struct nvm_chk_meta *meta;
 131         struct ppa_addr ppa;
 132         unsigned long len;
 133         int ret;
 134
 135         ppa.ppa = 0;
 136
 137         len = geo->all_chunks * sizeof(*meta);
 138         meta = vzalloc(len);
 139         if (!meta)
 140                 return ERR_PTR(-ENOMEM);
 141
 142         ret = nvm_get_chunk_meta(dev, ppa, geo->all_chunks, meta);
 143         if (ret) {
 144                 vfree(meta);
 145                 return ERR_PTR(-EIO);
 146         }
 147
 148         return meta;
 149 }
 150
 151 struct nvm_chk_meta *pblk_chunk_get_off(struct pblk *pblk,
 152                                               struct nvm_chk_meta *meta,
 153                                               struct ppa_addr ppa)
 154 {
 155         struct nvm_tgt_dev *dev = pblk->dev;
 156         struct nvm_geo *geo = &dev->geo;
 157         int ch_off = ppa.m.grp * geo->num_chk * geo->num_lun;
 158         int lun_off = ppa.m.pu * geo->num_chk;
 159         int chk_off = ppa.m.chk;
 160
 161         return meta + ch_off + lun_off + chk_off;
 162 }
 163
 164 void __pblk_map_invalidate(struct pblk *pblk, struct pblk_line *line,
 165                            u64 paddr)
 166 {
 167         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
 168         struct list_head *move_list = NULL;
 169
 170         /* Lines being reclaimed (GC'ed) cannot be invalidated. Before the L2P
 171          * table is modified with reclaimed sectors, a check is done to endure
 172          * that newer updates are not overwritten.
 173          */
 174         spin_lock(&line->lock);
 175         WARN_ON(line->state == PBLK_LINESTATE_FREE);
 176
 177         if (test_and_set_bit(paddr, line->invalid_bitmap)) {
 178                 WARN_ONCE(1, "pblk: double invalidate\n");
 179                 spin_unlock(&line->lock);
 180                 return;
 181         }
 182         le32_add_cpu(line->vsc, -1);
 183
 184         if (line->state == PBLK_LINESTATE_CLOSED)
 185                 move_list = pblk_line_gc_list(pblk, line);
 186         spin_unlock(&line->lock);
 187
 188         if (move_list) {
 189                 spin_lock(&l_mg->gc_lock);
 190                 spin_lock(&line->lock);
 191                 /* Prevent moving a line that has just been chosen for GC */
 192                 if (line->state == PBLK_LINESTATE_GC) {
 193                         spin_unlock(&line->lock);
 194                         spin_unlock(&l_mg->gc_lock);
 195                         return;
 196                 }
 197                 spin_unlock(&line->lock);
 198
 199                 list_move_tail(&line->list, move_list);
 200                 spin_unlock(&l_mg->gc_lock);
 201         }
 202 }
 203
 204 void pblk_map_invalidate(struct pblk *pblk, struct ppa_addr ppa)
 205 {
 206         struct pblk_line *line;
 207         u64 paddr;
 208
 209 #ifdef CONFIG_NVM_PBLK_DEBUG
 210         /* Callers must ensure that the ppa points to a device address */
 211         BUG_ON(pblk_addr_in_cache(ppa));
 212         BUG_ON(pblk_ppa_empty(ppa));
 213 #endif
 214
 215         line = pblk_ppa_to_line(pblk, ppa);
 216         paddr = pblk_dev_ppa_to_line_addr(pblk, ppa);
 217
 218         __pblk_map_invalidate(pblk, line, paddr);
 219 }
 220
 221 static void pblk_invalidate_range(struct pblk *pblk, sector_t slba,
 222                                   unsigned int nr_secs)
 223 {
 224         sector_t lba;
 225
 226         spin_lock(&pblk->trans_lock);
 227         for (lba = slba; lba < slba + nr_secs; lba++) {
 228                 struct ppa_addr ppa;
 229
 230                 ppa = pblk_trans_map_get(pblk, lba);
 231
 232                 if (!pblk_addr_in_cache(ppa) && !pblk_ppa_empty(ppa))
 233                         pblk_map_invalidate(pblk, ppa);
 234
 235                 pblk_ppa_set_empty(&ppa);
 236                 pblk_trans_map_set(pblk, lba, ppa);
 237         }
 238         spin_unlock(&pblk->trans_lock);
 239 }
 240
 241 int pblk_alloc_rqd_meta(struct pblk *pblk, struct nvm_rq *rqd)
 242 {
 243         struct nvm_tgt_dev *dev = pblk->dev;
 244
 245         rqd->meta_list = nvm_dev_dma_alloc(dev->parent, GFP_KERNEL,
 246                                                         &rqd->dma_meta_list);
 247         if (!rqd->meta_list)
 248                 return -ENOMEM;
 249
 250         if (rqd->nr_ppas == 1)
 251                 return 0;
 252
 253         rqd->ppa_list = rqd->meta_list + pblk_dma_meta_size(pblk);
 254         rqd->dma_ppa_list = rqd->dma_meta_list + pblk_dma_meta_size(pblk);
 255
 256         return 0;
 257 }
 258
 259 void pblk_free_rqd_meta(struct pblk *pblk, struct nvm_rq *rqd)
 260 {
 261         struct nvm_tgt_dev *dev = pblk->dev;
 262
 263         if (rqd->meta_list)
 264                 nvm_dev_dma_free(dev->parent, rqd->meta_list,
 265                                 rqd->dma_meta_list);
 266 }
 267
 268 /* Caller must guarantee that the request is a valid type */
 269 struct nvm_rq *pblk_alloc_rqd(struct pblk *pblk, int type)
 270 {
 271         mempool_t *pool;
 272         struct nvm_rq *rqd;
 273         int rq_size;
 274
 275         switch (type) {
 276         case PBLK_WRITE:
 277         case PBLK_WRITE_INT:
 278                 pool = &pblk->w_rq_pool;
 279                 rq_size = pblk_w_rq_size;
 280                 break;
 281         case PBLK_READ:
 282                 pool = &pblk->r_rq_pool;
 283                 rq_size = pblk_g_rq_size;
 284                 break;
 285         default:
 286                 pool = &pblk->e_rq_pool;
 287                 rq_size = pblk_g_rq_size;
 288         }
 289
 290         rqd = mempool_alloc(pool, GFP_KERNEL);
 291         memset(rqd, 0, rq_size);
 292
 293         return rqd;
 294 }
 295
 296 /* Typically used on completion path. Cannot guarantee request consistency */
 297 void pblk_free_rqd(struct pblk *pblk, struct nvm_rq *rqd, int type)
 298 {
 299         mempool_t *pool;
 300
 301         switch (type) {
 302         case PBLK_WRITE:
 303                 kfree(((struct pblk_c_ctx *)nvm_rq_to_pdu(rqd))->lun_bitmap);
 304                 /* fall through */
 305         case PBLK_WRITE_INT:
 306                 pool = &pblk->w_rq_pool;
 307                 break;
 308         case PBLK_READ:
 309                 pool = &pblk->r_rq_pool;
 310                 break;
 311         case PBLK_ERASE:
 312                 pool = &pblk->e_rq_pool;
 313                 break;
 314         default:
 315                 pblk_err(pblk, "trying to free unknown rqd type\n");
 316                 return;
 317         }
 318
 319         pblk_free_rqd_meta(pblk, rqd);
 320         mempool_free(rqd, pool);
 321 }
 322
 323 void pblk_bio_free_pages(struct pblk *pblk, struct bio *bio, int off,
 324                          int nr_pages)
 325 {
 326         struct bio_vec bv;
 327         int i;
 328
 329         WARN_ON(off + nr_pages != bio->bi_vcnt);
 330
 331         for (i = off; i < nr_pages + off; i++) {
 332                 bv = bio->bi_io_vec[i];
 333                 mempool_free(bv.bv_page, &pblk->page_bio_pool);
 334         }
 335 }
 336
 337 int pblk_bio_add_pages(struct pblk *pblk, struct bio *bio, gfp_t flags,
 338                        int nr_pages)
 339 {
 340         struct request_queue *q = pblk->dev->q;
 341         struct page *page;
 342         int i, ret;
 343
 344         for (i = 0; i < nr_pages; i++) {
 345                 page = mempool_alloc(&pblk->page_bio_pool, flags);
 346
 347                 ret = bio_add_pc_page(q, bio, page, PBLK_EXPOSED_PAGE_SIZE, 0);
 348                 if (ret != PBLK_EXPOSED_PAGE_SIZE) {
 349                         pblk_err(pblk, "could not add page to bio\n");
 350                         mempool_free(page, &pblk->page_bio_pool);
 351                         goto err;
 352                 }
 353         }
 354
 355         return 0;
 356 err:
 357         pblk_bio_free_pages(pblk, bio, (bio->bi_vcnt - i), i);
 358         return -1;
 359 }
 360
 361 void pblk_write_kick(struct pblk *pblk)
 362 {
 363         wake_up_process(pblk->writer_ts);
 364         mod_timer(&pblk->wtimer, jiffies + msecs_to_jiffies(1000));
 365 }
 366
 367 void pblk_write_timer_fn(struct timer_list *t)
 368 {
 369         struct pblk *pblk = from_timer(pblk, t, wtimer);
 370
 371         /* kick the write thread every tick to flush outstanding data */
 372         pblk_write_kick(pblk);
 373 }
 374
 375 void pblk_write_should_kick(struct pblk *pblk)
 376 {
 377         unsigned int secs_avail = pblk_rb_read_count(&pblk->rwb);
 378
 379         if (secs_avail >= pblk->min_write_pgs_data)
 380                 pblk_write_kick(pblk);
 381 }
 382
 383 static void pblk_wait_for_meta(struct pblk *pblk)
 384 {
 385         do {
 386                 if (!atomic_read(&pblk->inflight_io))
 387                         break;
 388
 389                 schedule();
 390         } while (1);
 391 }
 392
 393 static void pblk_flush_writer(struct pblk *pblk)
 394 {
 395         pblk_rb_flush(&pblk->rwb);
 396         do {
 397                 if (!pblk_rb_sync_count(&pblk->rwb))
 398                         break;
 399
 400                 pblk_write_kick(pblk);
 401                 schedule();
 402         } while (1);
 403 }
 404
 405 struct list_head *pblk_line_gc_list(struct pblk *pblk, struct pblk_line *line)
 406 {
 407         struct pblk_line_meta *lm = &pblk->lm;
 408         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
 409         struct list_head *move_list = NULL;
 410         int packed_meta = (le32_to_cpu(*line->vsc) / pblk->min_write_pgs_data)
 411                         * (pblk->min_write_pgs - pblk->min_write_pgs_data);
 412         int vsc = le32_to_cpu(*line->vsc) + packed_meta;
 413
 414         lockdep_assert_held(&line->lock);
 415
 416         if (line->w_err_gc->has_write_err) {
 417                 if (line->gc_group != PBLK_LINEGC_WERR) {
 418                         line->gc_group = PBLK_LINEGC_WERR;
 419                         move_list = &l_mg->gc_werr_list;
 420                         pblk_rl_werr_line_in(&pblk->rl);
 421                 }
 422         } else if (!vsc) {
 423                 if (line->gc_group != PBLK_LINEGC_FULL) {
 424                         line->gc_group = PBLK_LINEGC_FULL;
 425                         move_list = &l_mg->gc_full_list;
 426                 }
 427         } else if (vsc < lm->high_thrs) {
 428                 if (line->gc_group != PBLK_LINEGC_HIGH) {
 429                         line->gc_group = PBLK_LINEGC_HIGH;
 430                         move_list = &l_mg->gc_high_list;
 431                 }
 432         } else if (vsc < lm->mid_thrs) {
 433                 if (line->gc_group != PBLK_LINEGC_MID) {
 434                         line->gc_group = PBLK_LINEGC_MID;
 435                         move_list = &l_mg->gc_mid_list;
 436                 }
 437         } else if (vsc < line->sec_in_line) {
 438                 if (line->gc_group != PBLK_LINEGC_LOW) {
 439                         line->gc_group = PBLK_LINEGC_LOW;
 440                         move_list = &l_mg->gc_low_list;
 441                 }
 442         } else if (vsc == line->sec_in_line) {
 443                 if (line->gc_group != PBLK_LINEGC_EMPTY) {
 444                         line->gc_group = PBLK_LINEGC_EMPTY;
 445                         move_list = &l_mg->gc_empty_list;
 446                 }
 447         } else {
 448                 line->state = PBLK_LINESTATE_CORRUPT;
 449                 trace_pblk_line_state(pblk_disk_name(pblk), line->id,
 450                                         line->state);
 451
 452                 line->gc_group = PBLK_LINEGC_NONE;
 453                 move_list =  &l_mg->corrupt_list;
 454                 pblk_err(pblk, "corrupted vsc for line %d, vsc:%d (%d/%d/%d)\n",
 455                                                 line->id, vsc,
 456                                                 line->sec_in_line,
 457                                                 lm->high_thrs, lm->mid_thrs);
 458         }
 459
 460         return move_list;
 461 }
 462
 463 void pblk_discard(struct pblk *pblk, struct bio *bio)
 464 {
 465         sector_t slba = pblk_get_lba(bio);
 466         sector_t nr_secs = pblk_get_secs(bio);
 467
 468         pblk_invalidate_range(pblk, slba, nr_secs);
 469 }
 470
 471 void pblk_log_write_err(struct pblk *pblk, struct nvm_rq *rqd)
 472 {
 473         atomic_long_inc(&pblk->write_failed);
 474 #ifdef CONFIG_NVM_PBLK_DEBUG
 475         pblk_print_failed_rqd(pblk, rqd, rqd->error);
 476 #endif
 477 }
 478
 479 void pblk_log_read_err(struct pblk *pblk, struct nvm_rq *rqd)
 480 {
 481         /* Empty page read is not necessarily an error (e.g., L2P recovery) */
 482         if (rqd->error == NVM_RSP_ERR_EMPTYPAGE) {
 483                 atomic_long_inc(&pblk->read_empty);
 484                 return;
 485         }
 486
 487         switch (rqd->error) {
 488         case NVM_RSP_WARN_HIGHECC:
 489                 atomic_long_inc(&pblk->read_high_ecc);
 490                 break;
 491         case NVM_RSP_ERR_FAILECC:
 492         case NVM_RSP_ERR_FAILCRC:
 493                 atomic_long_inc(&pblk->read_failed);
 494                 break;
 495         default:
 496                 pblk_err(pblk, "unknown read error:%d\n", rqd->error);
 497         }
 498 #ifdef CONFIG_NVM_PBLK_DEBUG
 499         pblk_print_failed_rqd(pblk, rqd, rqd->error);
 500 #endif
 501 }
 502
 503 void pblk_set_sec_per_write(struct pblk *pblk, int sec_per_write)
 504 {
 505         pblk->sec_per_write = sec_per_write;
 506 }
 507
 508 int pblk_submit_io(struct pblk *pblk, struct nvm_rq *rqd)
 509 {
 510         struct nvm_tgt_dev *dev = pblk->dev;
 511
 512         atomic_inc(&pblk->inflight_io);
 513
 514 #ifdef CONFIG_NVM_PBLK_DEBUG
 515         if (pblk_check_io(pblk, rqd))
 516                 return NVM_IO_ERR;
 517 #endif
 518
 519         return nvm_submit_io(dev, rqd);
 520 }
 521
 522 void pblk_check_chunk_state_update(struct pblk *pblk, struct nvm_rq *rqd)
 523 {
 524         struct ppa_addr *ppa_list = nvm_rq_to_ppa_list(rqd);
 525
 526         int i;
 527
 528         for (i = 0; i < rqd->nr_ppas; i++) {
 529                 struct ppa_addr *ppa = &ppa_list[i];
 530                 struct nvm_chk_meta *chunk = pblk_dev_ppa_to_chunk(pblk, *ppa);
 531                 u64 caddr = pblk_dev_ppa_to_chunk_addr(pblk, *ppa);
 532
 533                 if (caddr == 0)
 534                         trace_pblk_chunk_state(pblk_disk_name(pblk),
 535                                                         ppa, NVM_CHK_ST_OPEN);
 536                 else if (caddr == (chunk->cnlb - 1))
 537                         trace_pblk_chunk_state(pblk_disk_name(pblk),
 538                                                         ppa, NVM_CHK_ST_CLOSED);
 539         }
 540 }
 541
 542 int pblk_submit_io_sync(struct pblk *pblk, struct nvm_rq *rqd)
 543 {
 544         struct nvm_tgt_dev *dev = pblk->dev;
 545         int ret;
 546
 547         atomic_inc(&pblk->inflight_io);
 548
 549 #ifdef CONFIG_NVM_PBLK_DEBUG
 550         if (pblk_check_io(pblk, rqd))
 551                 return NVM_IO_ERR;
 552 #endif
 553
 554         ret = nvm_submit_io_sync(dev, rqd);
 555
 556         if (trace_pblk_chunk_state_enabled() && !ret &&
 557             rqd->opcode == NVM_OP_PWRITE)
 558                 pblk_check_chunk_state_update(pblk, rqd);
 559
 560         return ret;
 561 }
 562
 563 int pblk_submit_io_sync_sem(struct pblk *pblk, struct nvm_rq *rqd)
 564 {
 565         struct ppa_addr *ppa_list;
 566         int ret;
 567
 568         ppa_list = (rqd->nr_ppas > 1) ? rqd->ppa_list : &rqd->ppa_addr;
 569
 570         pblk_down_chunk(pblk, ppa_list[0]);
 571         ret = pblk_submit_io_sync(pblk, rqd);
 572         pblk_up_chunk(pblk, ppa_list[0]);
 573
 574         return ret;
 575 }
 576
 577 static void pblk_bio_map_addr_endio(struct bio *bio)
 578 {
 579         bio_put(bio);
 580 }
 581
 582 struct bio *pblk_bio_map_addr(struct pblk *pblk, void *data,
 583                               unsigned int nr_secs, unsigned int len,
 584                               int alloc_type, gfp_t gfp_mask)
 585 {
 586         struct nvm_tgt_dev *dev = pblk->dev;
 587         void *kaddr = data;
 588         struct page *page;
 589         struct bio *bio;
 590         int i, ret;
 591
 592         if (alloc_type == PBLK_KMALLOC_META)
 593                 return bio_map_kern(dev->q, kaddr, len, gfp_mask);
 594
 595         bio = bio_kmalloc(gfp_mask, nr_secs);
 596         if (!bio)
 597                 return ERR_PTR(-ENOMEM);
 598
 599         for (i = 0; i < nr_secs; i++) {
 600                 page = vmalloc_to_page(kaddr);
 601                 if (!page) {
 602                         pblk_err(pblk, "could not map vmalloc bio\n");
 603                         bio_put(bio);
 604                         bio = ERR_PTR(-ENOMEM);
 605                         goto out;
 606                 }
 607
 608                 ret = bio_add_pc_page(dev->q, bio, page, PAGE_SIZE, 0);
 609                 if (ret != PAGE_SIZE) {
 610                         pblk_err(pblk, "could not add page to bio\n");
 611                         bio_put(bio);
 612                         bio = ERR_PTR(-ENOMEM);
 613                         goto out;
 614                 }
 615
 616                 kaddr += PAGE_SIZE;
 617         }
 618
 619         bio->bi_end_io = pblk_bio_map_addr_endio;
 620 out:
 621         return bio;
 622 }
 623
 624 int pblk_calc_secs(struct pblk *pblk, unsigned long secs_avail,
 625                    unsigned long secs_to_flush, bool skip_meta)
 626 {
 627         int max = pblk->sec_per_write;
 628         int min = pblk->min_write_pgs;
 629         int secs_to_sync = 0;
 630
 631         if (skip_meta && pblk->min_write_pgs_data != pblk->min_write_pgs)
 632                 min = max = pblk->min_write_pgs_data;
 633
 634         if (secs_avail >= max)
 635                 secs_to_sync = max;
 636         else if (secs_avail >= min)
 637                 secs_to_sync = min * (secs_avail / min);
 638         else if (secs_to_flush)
 639                 secs_to_sync = min;
 640
 641         return secs_to_sync;
 642 }
 643
 644 void pblk_dealloc_page(struct pblk *pblk, struct pblk_line *line, int nr_secs)
 645 {
 646         u64 addr;
 647         int i;
 648
 649         spin_lock(&line->lock);
 650         addr = find_next_zero_bit(line->map_bitmap,
 651                                         pblk->lm.sec_per_line, line->cur_sec);
 652         line->cur_sec = addr - nr_secs;
 653
 654         for (i = 0; i < nr_secs; i++, line->cur_sec--)
 655                 WARN_ON(!test_and_clear_bit(line->cur_sec, line->map_bitmap));
 656         spin_unlock(&line->lock);
 657 }
 658
 659 u64 __pblk_alloc_page(struct pblk *pblk, struct pblk_line *line, int nr_secs)
 660 {
 661         u64 addr;
 662         int i;
 663
 664         lockdep_assert_held(&line->lock);
 665
 666         /* logic error: ppa out-of-bounds. Prevent generating bad address */
 667         if (line->cur_sec + nr_secs > pblk->lm.sec_per_line) {
 668                 WARN(1, "pblk: page allocation out of bounds\n");
 669                 nr_secs = pblk->lm.sec_per_line - line->cur_sec;
 670         }
 671
 672         line->cur_sec = addr = find_next_zero_bit(line->map_bitmap,
 673                                         pblk->lm.sec_per_line, line->cur_sec);
 674         for (i = 0; i < nr_secs; i++, line->cur_sec++)
 675                 WARN_ON(test_and_set_bit(line->cur_sec, line->map_bitmap));
 676
 677         return addr;
 678 }
 679
 680 u64 pblk_alloc_page(struct pblk *pblk, struct pblk_line *line, int nr_secs)
 681 {
 682         u64 addr;
 683
 684         /* Lock needed in case a write fails and a recovery needs to remap
 685          * failed write buffer entries
 686          */
 687         spin_lock(&line->lock);
 688         addr = __pblk_alloc_page(pblk, line, nr_secs);
 689         line->left_msecs -= nr_secs;
 690         WARN(line->left_msecs < 0, "pblk: page allocation out of bounds\n");
 691         spin_unlock(&line->lock);
 692
 693         return addr;
 694 }
 695
 696 u64 pblk_lookup_page(struct pblk *pblk, struct pblk_line *line)
 697 {
 698         u64 paddr;
 699
 700         spin_lock(&line->lock);
 701         paddr = find_next_zero_bit(line->map_bitmap,
 702                                         pblk->lm.sec_per_line, line->cur_sec);
 703         spin_unlock(&line->lock);
 704
 705         return paddr;
 706 }
 707
 708 u64 pblk_line_smeta_start(struct pblk *pblk, struct pblk_line *line)
 709 {
 710         struct nvm_tgt_dev *dev = pblk->dev;
 711         struct nvm_geo *geo = &dev->geo;
 712         struct pblk_line_meta *lm = &pblk->lm;
 713         int bit;
 714
 715         /* This usually only happens on bad lines */
 716         bit = find_first_zero_bit(line->blk_bitmap, lm->blk_per_line);
 717         if (bit >= lm->blk_per_line)
 718                 return -1;
 719
 720         return bit * geo->ws_opt;
 721 }
 722
 723 int pblk_line_smeta_read(struct pblk *pblk, struct pblk_line *line)
 724 {
 725         struct nvm_tgt_dev *dev = pblk->dev;
 726         struct pblk_line_meta *lm = &pblk->lm;
 727         struct bio *bio;
 728         struct nvm_rq rqd;
 729         u64 paddr = pblk_line_smeta_start(pblk, line);
 730         int i, ret;
 731
 732         memset(&rqd, 0, sizeof(struct nvm_rq));
 733
 734         ret = pblk_alloc_rqd_meta(pblk, &rqd);
 735         if (ret)
 736                 return ret;
 737
 738         bio = bio_map_kern(dev->q, line->smeta, lm->smeta_len, GFP_KERNEL);
 739         if (IS_ERR(bio)) {
 740                 ret = PTR_ERR(bio);
 741                 goto clear_rqd;
 742         }
 743
 744         bio->bi_iter.bi_sector = 0; /* internal bio */
 745         bio_set_op_attrs(bio, REQ_OP_READ, 0);
 746
 747         rqd.bio = bio;
 748         rqd.opcode = NVM_OP_PREAD;
 749         rqd.nr_ppas = lm->smeta_sec;
 750         rqd.is_seq = 1;
 751
 752         for (i = 0; i < lm->smeta_sec; i++, paddr++)
 753                 rqd.ppa_list[i] = addr_to_gen_ppa(pblk, paddr, line->id);
 754
 755         ret = pblk_submit_io_sync(pblk, &rqd);
 756         if (ret) {
 757                 pblk_err(pblk, "smeta I/O submission failed: %d\n", ret);
 758                 bio_put(bio);
 759                 goto clear_rqd;
 760         }
 761
 762         atomic_dec(&pblk->inflight_io);
 763
 764         if (rqd.error)
 765                 pblk_log_read_err(pblk, &rqd);
 766
 767 clear_rqd:
 768         pblk_free_rqd_meta(pblk, &rqd);
 769         return ret;
 770 }
 771
 772 static int pblk_line_smeta_write(struct pblk *pblk, struct pblk_line *line,
 773                                  u64 paddr)
 774 {
 775         struct nvm_tgt_dev *dev = pblk->dev;
 776         struct pblk_line_meta *lm = &pblk->lm;
 777         struct bio *bio;
 778         struct nvm_rq rqd;
 779         __le64 *lba_list = emeta_to_lbas(pblk, line->emeta->buf);
 780         __le64 addr_empty = cpu_to_le64(ADDR_EMPTY);
 781         int i, ret;
 782
 783         memset(&rqd, 0, sizeof(struct nvm_rq));
 784
 785         ret = pblk_alloc_rqd_meta(pblk, &rqd);
 786         if (ret)
 787                 return ret;
 788
 789         bio = bio_map_kern(dev->q, line->smeta, lm->smeta_len, GFP_KERNEL);
 790         if (IS_ERR(bio)) {
 791                 ret = PTR_ERR(bio);
 792                 goto clear_rqd;
 793         }
 794
 795         bio->bi_iter.bi_sector = 0; /* internal bio */
 796         bio_set_op_attrs(bio, REQ_OP_WRITE, 0);
 797
 798         rqd.bio = bio;
 799         rqd.opcode = NVM_OP_PWRITE;
 800         rqd.nr_ppas = lm->smeta_sec;
 801         rqd.is_seq = 1;
 802
 803         for (i = 0; i < lm->smeta_sec; i++, paddr++) {
 804                 struct pblk_sec_meta *meta = pblk_get_meta(pblk,
 805                                                            rqd.meta_list, i);
 806
 807                 rqd.ppa_list[i] = addr_to_gen_ppa(pblk, paddr, line->id);
 808                 meta->lba = lba_list[paddr] = addr_empty;
 809         }
 810
 811         ret = pblk_submit_io_sync_sem(pblk, &rqd);
 812         if (ret) {
 813                 pblk_err(pblk, "smeta I/O submission failed: %d\n", ret);
 814                 bio_put(bio);
 815                 goto clear_rqd;
 816         }
 817
 818         atomic_dec(&pblk->inflight_io);
 819
 820         if (rqd.error) {
 821                 pblk_log_write_err(pblk, &rqd);
 822                 ret = -EIO;
 823         }
 824
 825 clear_rqd:
 826         pblk_free_rqd_meta(pblk, &rqd);
 827         return ret;
 828 }
 829
 830 int pblk_line_emeta_read(struct pblk *pblk, struct pblk_line *line,
 831                          void *emeta_buf)
 832 {
 833         struct nvm_tgt_dev *dev = pblk->dev;
 834         struct nvm_geo *geo = &dev->geo;
 835         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
 836         struct pblk_line_meta *lm = &pblk->lm;
 837         void *ppa_list, *meta_list;
 838         struct bio *bio;
 839         struct nvm_rq rqd;
 840         u64 paddr = line->emeta_ssec;
 841         dma_addr_t dma_ppa_list, dma_meta_list;
 842         int min = pblk->min_write_pgs;
 843         int left_ppas = lm->emeta_sec[0];
 844         int line_id = line->id;
 845         int rq_ppas, rq_len;
 846         int i, j;
 847         int ret;
 848
 849         meta_list = nvm_dev_dma_alloc(dev->parent, GFP_KERNEL,
 850                                                         &dma_meta_list);
 851         if (!meta_list)
 852                 return -ENOMEM;
 853
 854         ppa_list = meta_list + pblk_dma_meta_size(pblk);
 855         dma_ppa_list = dma_meta_list + pblk_dma_meta_size(pblk);
 856
 857 next_rq:
 858         memset(&rqd, 0, sizeof(struct nvm_rq));
 859
 860         rq_ppas = pblk_calc_secs(pblk, left_ppas, 0, false);
 861         rq_len = rq_ppas * geo->csecs;
 862
 863         bio = pblk_bio_map_addr(pblk, emeta_buf, rq_ppas, rq_len,
 864                                         l_mg->emeta_alloc_type, GFP_KERNEL);
 865         if (IS_ERR(bio)) {
 866                 ret = PTR_ERR(bio);
 867                 goto free_rqd_dma;
 868         }
 869
 870         bio->bi_iter.bi_sector = 0; /* internal bio */
 871         bio_set_op_attrs(bio, REQ_OP_READ, 0);
 872
 873         rqd.bio = bio;
 874         rqd.meta_list = meta_list;
 875         rqd.ppa_list = ppa_list;
 876         rqd.dma_meta_list = dma_meta_list;
 877         rqd.dma_ppa_list = dma_ppa_list;
 878         rqd.opcode = NVM_OP_PREAD;
 879         rqd.nr_ppas = rq_ppas;
 880
 881         for (i = 0; i < rqd.nr_ppas; ) {
 882                 struct ppa_addr ppa = addr_to_gen_ppa(pblk, paddr, line_id);
 883                 int pos = pblk_ppa_to_pos(geo, ppa);
 884
 885                 if (pblk_io_aligned(pblk, rq_ppas))
 886                         rqd.is_seq = 1;
 887
 888                 while (test_bit(pos, line->blk_bitmap)) {
 889                         paddr += min;
 890                         if (pblk_boundary_paddr_checks(pblk, paddr)) {
 891                                 bio_put(bio);
 892                                 ret = -EINTR;
 893                                 goto free_rqd_dma;
 894                         }
 895
 896                         ppa = addr_to_gen_ppa(pblk, paddr, line_id);
 897                         pos = pblk_ppa_to_pos(geo, ppa);
 898                 }
 899
 900                 if (pblk_boundary_paddr_checks(pblk, paddr + min)) {
 901                         bio_put(bio);
 902                         ret = -EINTR;
 903                         goto free_rqd_dma;
 904                 }
 905
 906                 for (j = 0; j < min; j++, i++, paddr++)
 907                         rqd.ppa_list[i] = addr_to_gen_ppa(pblk, paddr, line_id);
 908         }
 909
 910         ret = pblk_submit_io_sync(pblk, &rqd);
 911         if (ret) {
 912                 pblk_err(pblk, "emeta I/O submission failed: %d\n", ret);
 913                 bio_put(bio);
 914                 goto free_rqd_dma;
 915         }
 916
 917         atomic_dec(&pblk->inflight_io);
 918
 919         if (rqd.error)
 920                 pblk_log_read_err(pblk, &rqd);
 921
 922         emeta_buf += rq_len;
 923         left_ppas -= rq_ppas;
 924         if (left_ppas)
 925                 goto next_rq;
 926
 927 free_rqd_dma:
 928         nvm_dev_dma_free(dev->parent, rqd.meta_list, rqd.dma_meta_list);
 929         return ret;
 930 }
 931
 932 static void pblk_setup_e_rq(struct pblk *pblk, struct nvm_rq *rqd,
 933                             struct ppa_addr ppa)
 934 {
 935         rqd->opcode = NVM_OP_ERASE;
 936         rqd->ppa_addr = ppa;
 937         rqd->nr_ppas = 1;
 938         rqd->is_seq = 1;
 939         rqd->bio = NULL;
 940 }
 941
 942 static int pblk_blk_erase_sync(struct pblk *pblk, struct ppa_addr ppa)
 943 {
 944         struct nvm_rq rqd = {NULL};
 945         int ret;
 946
 947         trace_pblk_chunk_reset(pblk_disk_name(pblk), &ppa,
 948                                 PBLK_CHUNK_RESET_START);
 949
 950         pblk_setup_e_rq(pblk, &rqd, ppa);
 951
 952         /* The write thread schedules erases so that it minimizes disturbances
 953          * with writes. Thus, there is no need to take the LUN semaphore.
 954          */
 955         ret = pblk_submit_io_sync(pblk, &rqd);
 956         rqd.private = pblk;
 957         __pblk_end_io_erase(pblk, &rqd);
 958
 959         return ret;
 960 }
 961
 962 int pblk_line_erase(struct pblk *pblk, struct pblk_line *line)
 963 {
 964         struct pblk_line_meta *lm = &pblk->lm;
 965         struct ppa_addr ppa;
 966         int ret, bit = -1;
 967
 968         /* Erase only good blocks, one at a time */
 969         do {
 970                 spin_lock(&line->lock);
 971                 bit = find_next_zero_bit(line->erase_bitmap, lm->blk_per_line,
 972                                                                 bit + 1);
 973                 if (bit >= lm->blk_per_line) {
 974                         spin_unlock(&line->lock);
 975                         break;
 976                 }
 977
 978                 ppa = pblk->luns[bit].bppa; /* set ch and lun */
 979                 ppa.a.blk = line->id;
 980
 981                 atomic_dec(&line->left_eblks);
 982                 WARN_ON(test_and_set_bit(bit, line->erase_bitmap));
 983                 spin_unlock(&line->lock);
 984
 985                 ret = pblk_blk_erase_sync(pblk, ppa);
 986                 if (ret) {
 987                         pblk_err(pblk, "failed to erase line %d\n", line->id);
 988                         return ret;
 989                 }
 990         } while (1);
 991
 992         return 0;
 993 }
 994
 995 static void pblk_line_setup_metadata(struct pblk_line *line,
 996                                      struct pblk_line_mgmt *l_mg,
 997                                      struct pblk_line_meta *lm)
 998 {
 999         int meta_line;
1000
1001         lockdep_assert_held(&l_mg->free_lock);
1002
1003 retry_meta:
1004         meta_line = find_first_zero_bit(&l_mg->meta_bitmap, PBLK_DATA_LINES);
1005         if (meta_line == PBLK_DATA_LINES) {
1006                 spin_unlock(&l_mg->free_lock);
1007                 io_schedule();
1008                 spin_lock(&l_mg->free_lock);
1009                 goto retry_meta;
1010         }
1011
1012         set_bit(meta_line, &l_mg->meta_bitmap);
1013         line->meta_line = meta_line;
1014
1015         line->smeta = l_mg->sline_meta[meta_line];
1016         line->emeta = l_mg->eline_meta[meta_line];
1017
1018         memset(line->smeta, 0, lm->smeta_len);
1019         memset(line->emeta->buf, 0, lm->emeta_len[0]);
1020
1021         line->emeta->mem = 0;
1022         atomic_set(&line->emeta->sync, 0);
1023 }
1024
1025 /* For now lines are always assumed full lines. Thus, smeta former and current
1026  * lun bitmaps are omitted.
1027  */
1028 static int pblk_line_init_metadata(struct pblk *pblk, struct pblk_line *line,
1029                                   struct pblk_line *cur)
1030 {
1031         struct nvm_tgt_dev *dev = pblk->dev;
1032         struct nvm_geo *geo = &dev->geo;
1033         struct pblk_line_meta *lm = &pblk->lm;
1034         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1035         struct pblk_emeta *emeta = line->emeta;
1036         struct line_emeta *emeta_buf = emeta->buf;
1037         struct line_smeta *smeta_buf = (struct line_smeta *)line->smeta;
1038         int nr_blk_line;
1039
1040         /* After erasing the line, new bad blocks might appear and we risk
1041          * having an invalid line
1042          */
1043         nr_blk_line = lm->blk_per_line -
1044                         bitmap_weight(line->blk_bitmap, lm->blk_per_line);
1045         if (nr_blk_line < lm->min_blk_line) {
1046                 spin_lock(&l_mg->free_lock);
1047                 spin_lock(&line->lock);
1048                 line->state = PBLK_LINESTATE_BAD;
1049                 trace_pblk_line_state(pblk_disk_name(pblk), line->id,
1050                                         line->state);
1051                 spin_unlock(&line->lock);
1052
1053                 list_add_tail(&line->list, &l_mg->bad_list);
1054                 spin_unlock(&l_mg->free_lock);
1055
1056                 pblk_debug(pblk, "line %d is bad\n", line->id);
1057
1058                 return 0;
1059         }
1060
1061         /* Run-time metadata */
1062         line->lun_bitmap = ((void *)(smeta_buf)) + sizeof(struct line_smeta);
1063
1064         /* Mark LUNs allocated in this line (all for now) */
1065         bitmap_set(line->lun_bitmap, 0, lm->lun_bitmap_len);
1066
1067         smeta_buf->header.identifier = cpu_to_le32(PBLK_MAGIC);
1068         guid_copy((guid_t *)&smeta_buf->header.uuid, &pblk->instance_uuid);
1069         smeta_buf->header.id = cpu_to_le32(line->id);
1070         smeta_buf->header.type = cpu_to_le16(line->type);
1071         smeta_buf->header.version_major = SMETA_VERSION_MAJOR;
1072         smeta_buf->header.version_minor = SMETA_VERSION_MINOR;
1073
1074         /* Start metadata */
1075         smeta_buf->seq_nr = cpu_to_le64(line->seq_nr);
1076         smeta_buf->window_wr_lun = cpu_to_le32(geo->all_luns);
1077
1078         /* Fill metadata among lines */
1079         if (cur) {
1080                 memcpy(line->lun_bitmap, cur->lun_bitmap, lm->lun_bitmap_len);
1081                 smeta_buf->prev_id = cpu_to_le32(cur->id);
1082                 cur->emeta->buf->next_id = cpu_to_le32(line->id);
1083         } else {
1084                 smeta_buf->prev_id = cpu_to_le32(PBLK_LINE_EMPTY);
1085         }
1086
1087         /* All smeta must be set at this point */
1088         smeta_buf->header.crc = cpu_to_le32(
1089                         pblk_calc_meta_header_crc(pblk, &smeta_buf->header));
1090         smeta_buf->crc = cpu_to_le32(pblk_calc_smeta_crc(pblk, smeta_buf));
1091
1092         /* End metadata */
1093         memcpy(&emeta_buf->header, &smeta_buf->header,
1094                                                 sizeof(struct line_header));
1095
1096         emeta_buf->header.version_major = EMETA_VERSION_MAJOR;
1097         emeta_buf->header.version_minor = EMETA_VERSION_MINOR;
1098         emeta_buf->header.crc = cpu_to_le32(
1099                         pblk_calc_meta_header_crc(pblk, &emeta_buf->header));
1100
1101         emeta_buf->seq_nr = cpu_to_le64(line->seq_nr);
1102         emeta_buf->nr_lbas = cpu_to_le64(line->sec_in_line);
1103         emeta_buf->nr_valid_lbas = cpu_to_le64(0);
1104         emeta_buf->next_id = cpu_to_le32(PBLK_LINE_EMPTY);
1105         emeta_buf->crc = cpu_to_le32(0);
1106         emeta_buf->prev_id = smeta_buf->prev_id;
1107
1108         return 1;
1109 }
1110
1111 static int pblk_line_alloc_bitmaps(struct pblk *pblk, struct pblk_line *line)
1112 {
1113         struct pblk_line_meta *lm = &pblk->lm;
1114         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1115
1116         line->map_bitmap = mempool_alloc(l_mg->bitmap_pool, GFP_KERNEL);
1117         if (!line->map_bitmap)
1118                 return -ENOMEM;
1119
1120         memset(line->map_bitmap, 0, lm->sec_bitmap_len);
1121
1122         /* will be initialized using bb info from map_bitmap */
1123         line->invalid_bitmap = mempool_alloc(l_mg->bitmap_pool, GFP_KERNEL);
1124         if (!line->invalid_bitmap) {
1125                 mempool_free(line->map_bitmap, l_mg->bitmap_pool);
1126                 line->map_bitmap = NULL;
1127                 return -ENOMEM;
1128         }
1129
1130         return 0;
1131 }
1132
1133 /* For now lines are always assumed full lines. Thus, smeta former and current
1134  * lun bitmaps are omitted.
1135  */
1136 static int pblk_line_init_bb(struct pblk *pblk, struct pblk_line *line,
1137                              int init)
1138 {
1139         struct nvm_tgt_dev *dev = pblk->dev;
1140         struct nvm_geo *geo = &dev->geo;
1141         struct pblk_line_meta *lm = &pblk->lm;
1142         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1143         u64 off;
1144         int bit = -1;
1145         int emeta_secs;
1146
1147         line->sec_in_line = lm->sec_per_line;
1148
1149         /* Capture bad block information on line mapping bitmaps */
1150         while ((bit = find_next_bit(line->blk_bitmap, lm->blk_per_line,
1151                                         bit + 1)) < lm->blk_per_line) {
1152                 off = bit * geo->ws_opt;
1153                 bitmap_shift_left(l_mg->bb_aux, l_mg->bb_template, off,
1154                                                         lm->sec_per_line);
1155                 bitmap_or(line->map_bitmap, line->map_bitmap, l_mg->bb_aux,
1156                                                         lm->sec_per_line);
1157                 line->sec_in_line -= geo->clba;
1158         }
1159
1160         /* Mark smeta metadata sectors as bad sectors */
1161         bit = find_first_zero_bit(line->blk_bitmap, lm->blk_per_line);
1162         off = bit * geo->ws_opt;
1163         bitmap_set(line->map_bitmap, off, lm->smeta_sec);
1164         line->sec_in_line -= lm->smeta_sec;
1165         line->smeta_ssec = off;
1166         line->cur_sec = off + lm->smeta_sec;
1167
1168         if (init && pblk_line_smeta_write(pblk, line, off)) {
1169                 pblk_debug(pblk, "line smeta I/O failed. Retry\n");
1170                 return 0;
1171         }
1172
1173         bitmap_copy(line->invalid_bitmap, line->map_bitmap, lm->sec_per_line);
1174
1175         /* Mark emeta metadata sectors as bad sectors. We need to consider bad
1176          * blocks to make sure that there are enough sectors to store emeta
1177          */
1178         emeta_secs = lm->emeta_sec[0];
1179         off = lm->sec_per_line;
1180         while (emeta_secs) {
1181                 off -= geo->ws_opt;
1182                 if (!test_bit(off, line->invalid_bitmap)) {
1183                         bitmap_set(line->invalid_bitmap, off, geo->ws_opt);
1184                         emeta_secs -= geo->ws_opt;
1185                 }
1186         }
1187
1188         line->emeta_ssec = off;
1189         line->sec_in_line -= lm->emeta_sec[0];
1190         line->nr_valid_lbas = 0;
1191         line->left_msecs = line->sec_in_line;
1192         *line->vsc = cpu_to_le32(line->sec_in_line);
1193
1194         if (lm->sec_per_line - line->sec_in_line !=
1195                 bitmap_weight(line->invalid_bitmap, lm->sec_per_line)) {
1196                 spin_lock(&line->lock);
1197                 line->state = PBLK_LINESTATE_BAD;
1198                 trace_pblk_line_state(pblk_disk_name(pblk), line->id,
1199                                         line->state);
1200                 spin_unlock(&line->lock);
1201
1202                 list_add_tail(&line->list, &l_mg->bad_list);
1203                 pblk_err(pblk, "unexpected line %d is bad\n", line->id);
1204
1205                 return 0;
1206         }
1207
1208         return 1;
1209 }
1210
1211 static int pblk_prepare_new_line(struct pblk *pblk, struct pblk_line *line)
1212 {
1213         struct pblk_line_meta *lm = &pblk->lm;
1214         struct nvm_tgt_dev *dev = pblk->dev;
1215         struct nvm_geo *geo = &dev->geo;
1216         int blk_to_erase = atomic_read(&line->blk_in_line);
1217         int i;
1218
1219         for (i = 0; i < lm->blk_per_line; i++) {
1220                 struct pblk_lun *rlun = &pblk->luns[i];
1221                 int pos = pblk_ppa_to_pos(geo, rlun->bppa);
1222                 int state = line->chks[pos].state;
1223
1224                 /* Free chunks should not be erased */
1225                 if (state & NVM_CHK_ST_FREE) {
1226                         set_bit(pblk_ppa_to_pos(geo, rlun->bppa),
1227                                                         line->erase_bitmap);
1228                         blk_to_erase--;
1229                 }
1230         }
1231
1232         return blk_to_erase;
1233 }
1234
1235 static int pblk_line_prepare(struct pblk *pblk, struct pblk_line *line)
1236 {
1237         struct pblk_line_meta *lm = &pblk->lm;
1238         int blk_in_line = atomic_read(&line->blk_in_line);
1239         int blk_to_erase;
1240
1241         /* Bad blocks do not need to be erased */
1242         bitmap_copy(line->erase_bitmap, line->blk_bitmap, lm->blk_per_line);
1243
1244         spin_lock(&line->lock);
1245
1246         /* If we have not written to this line, we need to mark up free chunks
1247          * as already erased
1248          */
1249         if (line->state == PBLK_LINESTATE_NEW) {
1250                 blk_to_erase = pblk_prepare_new_line(pblk, line);
1251                 line->state = PBLK_LINESTATE_FREE;
1252                 trace_pblk_line_state(pblk_disk_name(pblk), line->id,
1253                                         line->state);
1254         } else {
1255                 blk_to_erase = blk_in_line;
1256         }
1257
1258         if (blk_in_line < lm->min_blk_line) {
1259                 spin_unlock(&line->lock);
1260                 return -EAGAIN;
1261         }
1262
1263         if (line->state != PBLK_LINESTATE_FREE) {
1264                 WARN(1, "pblk: corrupted line %d, state %d\n",
1265                                                         line->id, line->state);
1266                 spin_unlock(&line->lock);
1267                 return -EINTR;
1268         }
1269
1270         line->state = PBLK_LINESTATE_OPEN;
1271         trace_pblk_line_state(pblk_disk_name(pblk), line->id,
1272                                 line->state);
1273
1274         atomic_set(&line->left_eblks, blk_to_erase);
1275         atomic_set(&line->left_seblks, blk_to_erase);
1276
1277         line->meta_distance = lm->meta_distance;
1278         spin_unlock(&line->lock);
1279
1280         kref_init(&line->ref);
1281         atomic_set(&line->sec_to_update, 0);
1282
1283         return 0;
1284 }
1285
1286 /* Line allocations in the recovery path are always single threaded */
1287 int pblk_line_recov_alloc(struct pblk *pblk, struct pblk_line *line)
1288 {
1289         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1290         int ret;
1291
1292         spin_lock(&l_mg->free_lock);
1293         l_mg->data_line = line;
1294         list_del(&line->list);
1295
1296         ret = pblk_line_prepare(pblk, line);
1297         if (ret) {
1298                 list_add(&line->list, &l_mg->free_list);
1299                 spin_unlock(&l_mg->free_lock);
1300                 return ret;
1301         }
1302         spin_unlock(&l_mg->free_lock);
1303
1304         ret = pblk_line_alloc_bitmaps(pblk, line);
1305         if (ret)
1306                 goto fail;
1307
1308         if (!pblk_line_init_bb(pblk, line, 0)) {
1309                 ret = -EINTR;
1310                 goto fail;
1311         }
1312
1313         pblk_rl_free_lines_dec(&pblk->rl, line, true);
1314         return 0;
1315
1316 fail:
1317         spin_lock(&l_mg->free_lock);
1318         list_add(&line->list, &l_mg->free_list);
1319         spin_unlock(&l_mg->free_lock);
1320
1321         return ret;
1322 }
1323
1324 void pblk_line_recov_close(struct pblk *pblk, struct pblk_line *line)
1325 {
1326         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1327
1328         mempool_free(line->map_bitmap, l_mg->bitmap_pool);
1329         line->map_bitmap = NULL;
1330         line->smeta = NULL;
1331         line->emeta = NULL;
1332 }
1333
1334 static void pblk_line_reinit(struct pblk_line *line)
1335 {
1336         *line->vsc = cpu_to_le32(EMPTY_ENTRY);
1337
1338         line->map_bitmap = NULL;
1339         line->invalid_bitmap = NULL;
1340         line->smeta = NULL;
1341         line->emeta = NULL;
1342 }
1343
1344 void pblk_line_free(struct pblk_line *line)
1345 {
1346         struct pblk *pblk = line->pblk;
1347         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1348
1349         mempool_free(line->map_bitmap, l_mg->bitmap_pool);
1350         mempool_free(line->invalid_bitmap, l_mg->bitmap_pool);
1351
1352         pblk_line_reinit(line);
1353 }
1354
1355 struct pblk_line *pblk_line_get(struct pblk *pblk)
1356 {
1357         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1358         struct pblk_line_meta *lm = &pblk->lm;
1359         struct pblk_line *line;
1360         int ret, bit;
1361
1362         lockdep_assert_held(&l_mg->free_lock);
1363
1364 retry:
1365         if (list_empty(&l_mg->free_list)) {
1366                 pblk_err(pblk, "no free lines\n");
1367                 return NULL;
1368         }
1369
1370         line = list_first_entry(&l_mg->free_list, struct pblk_line, list);
1371         list_del(&line->list);
1372         l_mg->nr_free_lines--;
1373
1374         bit = find_first_zero_bit(line->blk_bitmap, lm->blk_per_line);
1375         if (unlikely(bit >= lm->blk_per_line)) {
1376                 spin_lock(&line->lock);
1377                 line->state = PBLK_LINESTATE_BAD;
1378                 trace_pblk_line_state(pblk_disk_name(pblk), line->id,
1379                                         line->state);
1380                 spin_unlock(&line->lock);
1381
1382                 list_add_tail(&line->list, &l_mg->bad_list);
1383
1384                 pblk_debug(pblk, "line %d is bad\n", line->id);
1385                 goto retry;
1386         }
1387
1388         ret = pblk_line_prepare(pblk, line);
1389         if (ret) {
1390                 switch (ret) {
1391                 case -EAGAIN:
1392                         list_add(&line->list, &l_mg->bad_list);
1393                         goto retry;
1394                 case -EINTR:
1395                         list_add(&line->list, &l_mg->corrupt_list);
1396                         goto retry;
1397                 default:
1398                         pblk_err(pblk, "failed to prepare line %d\n", line->id);
1399                         list_add(&line->list, &l_mg->free_list);
1400                         l_mg->nr_free_lines++;
1401                         return NULL;
1402                 }
1403         }
1404
1405         return line;
1406 }
1407
1408 static struct pblk_line *pblk_line_retry(struct pblk *pblk,
1409                                          struct pblk_line *line)
1410 {
1411         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1412         struct pblk_line *retry_line;
1413
1414 retry:
1415         spin_lock(&l_mg->free_lock);
1416         retry_line = pblk_line_get(pblk);
1417         if (!retry_line) {
1418                 l_mg->data_line = NULL;
1419                 spin_unlock(&l_mg->free_lock);
1420                 return NULL;
1421         }
1422
1423         retry_line->map_bitmap = line->map_bitmap;
1424         retry_line->invalid_bitmap = line->invalid_bitmap;
1425         retry_line->smeta = line->smeta;
1426         retry_line->emeta = line->emeta;
1427         retry_line->meta_line = line->meta_line;
1428
1429         pblk_line_reinit(line);
1430
1431         l_mg->data_line = retry_line;
1432         spin_unlock(&l_mg->free_lock);
1433
1434         pblk_rl_free_lines_dec(&pblk->rl, line, false);
1435
1436         if (pblk_line_erase(pblk, retry_line))
1437                 goto retry;
1438
1439         return retry_line;
1440 }
1441
1442 static void pblk_set_space_limit(struct pblk *pblk)
1443 {
1444         struct pblk_rl *rl = &pblk->rl;
1445
1446         atomic_set(&rl->rb_space, 0);
1447 }
1448
1449 struct pblk_line *pblk_line_get_first_data(struct pblk *pblk)
1450 {
1451         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1452         struct pblk_line *line;
1453
1454         spin_lock(&l_mg->free_lock);
1455         line = pblk_line_get(pblk);
1456         if (!line) {
1457                 spin_unlock(&l_mg->free_lock);
1458                 return NULL;
1459         }
1460
1461         line->seq_nr = l_mg->d_seq_nr++;
1462         line->type = PBLK_LINETYPE_DATA;
1463         l_mg->data_line = line;
1464
1465         pblk_line_setup_metadata(line, l_mg, &pblk->lm);
1466
1467         /* Allocate next line for preparation */
1468         l_mg->data_next = pblk_line_get(pblk);
1469         if (!l_mg->data_next) {
1470                 /* If we cannot get a new line, we need to stop the pipeline.
1471                  * Only allow as many writes in as we can store safely and then
1472                  * fail gracefully
1473                  */
1474                 pblk_set_space_limit(pblk);
1475
1476                 l_mg->data_next = NULL;
1477         } else {
1478                 l_mg->data_next->seq_nr = l_mg->d_seq_nr++;
1479                 l_mg->data_next->type = PBLK_LINETYPE_DATA;
1480         }
1481         spin_unlock(&l_mg->free_lock);
1482
1483         if (pblk_line_alloc_bitmaps(pblk, line))
1484                 return NULL;
1485
1486         if (pblk_line_erase(pblk, line)) {
1487                 line = pblk_line_retry(pblk, line);
1488                 if (!line)
1489                         return NULL;
1490         }
1491
1492 retry_setup:
1493         if (!pblk_line_init_metadata(pblk, line, NULL)) {
1494                 line = pblk_line_retry(pblk, line);
1495                 if (!line)
1496                         return NULL;
1497
1498                 goto retry_setup;
1499         }
1500
1501         if (!pblk_line_init_bb(pblk, line, 1)) {
1502                 line = pblk_line_retry(pblk, line);
1503                 if (!line)
1504                         return NULL;
1505
1506                 goto retry_setup;
1507         }
1508
1509         pblk_rl_free_lines_dec(&pblk->rl, line, true);
1510
1511         return line;
1512 }
1513
1514 void pblk_ppa_to_line_put(struct pblk *pblk, struct ppa_addr ppa)
1515 {
1516         struct pblk_line *line;
1517
1518         line = pblk_ppa_to_line(pblk, ppa);
1519         kref_put(&line->ref, pblk_line_put_wq);
1520 }
1521
1522 void pblk_rq_to_line_put(struct pblk *pblk, struct nvm_rq *rqd)
1523 {
1524         struct ppa_addr *ppa_list;
1525         int i;
1526
1527         ppa_list = (rqd->nr_ppas > 1) ? rqd->ppa_list : &rqd->ppa_addr;
1528
1529         for (i = 0; i < rqd->nr_ppas; i++)
1530                 pblk_ppa_to_line_put(pblk, ppa_list[i]);
1531 }
1532
1533 static void pblk_stop_writes(struct pblk *pblk, struct pblk_line *line)
1534 {
1535         lockdep_assert_held(&pblk->l_mg.free_lock);
1536
1537         pblk_set_space_limit(pblk);
1538         pblk->state = PBLK_STATE_STOPPING;
1539         trace_pblk_state(pblk_disk_name(pblk), pblk->state);
1540 }
1541
1542 static void pblk_line_close_meta_sync(struct pblk *pblk)
1543 {
1544         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1545         struct pblk_line_meta *lm = &pblk->lm;
1546         struct pblk_line *line, *tline;
1547         LIST_HEAD(list);
1548
1549         spin_lock(&l_mg->close_lock);
1550         if (list_empty(&l_mg->emeta_list)) {
1551                 spin_unlock(&l_mg->close_lock);
1552                 return;
1553         }
1554
1555         list_cut_position(&list, &l_mg->emeta_list, l_mg->emeta_list.prev);
1556         spin_unlock(&l_mg->close_lock);
1557
1558         list_for_each_entry_safe(line, tline, &list, list) {
1559                 struct pblk_emeta *emeta = line->emeta;
1560
1561                 while (emeta->mem < lm->emeta_len[0]) {
1562                         int ret;
1563
1564                         ret = pblk_submit_meta_io(pblk, line);
1565                         if (ret) {
1566                                 pblk_err(pblk, "sync meta line %d failed (%d)\n",
1567                                                         line->id, ret);
1568                                 return;
1569                         }
1570                 }
1571         }
1572
1573         pblk_wait_for_meta(pblk);
1574         flush_workqueue(pblk->close_wq);
1575 }
1576
1577 void __pblk_pipeline_flush(struct pblk *pblk)
1578 {
1579         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1580         int ret;
1581
1582         spin_lock(&l_mg->free_lock);
1583         if (pblk->state == PBLK_STATE_RECOVERING ||
1584                                         pblk->state == PBLK_STATE_STOPPED) {
1585                 spin_unlock(&l_mg->free_lock);
1586                 return;
1587         }
1588         pblk->state = PBLK_STATE_RECOVERING;
1589         trace_pblk_state(pblk_disk_name(pblk), pblk->state);
1590         spin_unlock(&l_mg->free_lock);
1591
1592         pblk_flush_writer(pblk);
1593         pblk_wait_for_meta(pblk);
1594
1595         ret = pblk_recov_pad(pblk);
1596         if (ret) {
1597                 pblk_err(pblk, "could not close data on teardown(%d)\n", ret);
1598                 return;
1599         }
1600
1601         flush_workqueue(pblk->bb_wq);
1602         pblk_line_close_meta_sync(pblk);
1603 }
1604
1605 void __pblk_pipeline_stop(struct pblk *pblk)
1606 {
1607         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1608
1609         spin_lock(&l_mg->free_lock);
1610         pblk->state = PBLK_STATE_STOPPED;
1611         trace_pblk_state(pblk_disk_name(pblk), pblk->state);
1612         l_mg->data_line = NULL;
1613         l_mg->data_next = NULL;
1614         spin_unlock(&l_mg->free_lock);
1615 }
1616
1617 void pblk_pipeline_stop(struct pblk *pblk)
1618 {
1619         __pblk_pipeline_flush(pblk);
1620         __pblk_pipeline_stop(pblk);
1621 }
1622
1623 struct pblk_line *pblk_line_replace_data(struct pblk *pblk)
1624 {
1625         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1626         struct pblk_line *cur, *new = NULL;
1627         unsigned int left_seblks;
1628
1629         new = l_mg->data_next;
1630         if (!new)
1631                 goto out;
1632
1633         spin_lock(&l_mg->free_lock);
1634         cur = l_mg->data_line;
1635         l_mg->data_line = new;
1636
1637         pblk_line_setup_metadata(new, l_mg, &pblk->lm);
1638         spin_unlock(&l_mg->free_lock);
1639
1640 retry_erase:
1641         left_seblks = atomic_read(&new->left_seblks);
1642         if (left_seblks) {
1643                 /* If line is not fully erased, erase it */
1644                 if (atomic_read(&new->left_eblks)) {
1645                         if (pblk_line_erase(pblk, new))
1646                                 goto out;
1647                 } else {
1648                         io_schedule();
1649                 }
1650                 goto retry_erase;
1651         }
1652
1653         if (pblk_line_alloc_bitmaps(pblk, new))
1654                 return NULL;
1655
1656 retry_setup:
1657         if (!pblk_line_init_metadata(pblk, new, cur)) {
1658                 new = pblk_line_retry(pblk, new);
1659                 if (!new)
1660                         goto out;
1661
1662                 goto retry_setup;
1663         }
1664
1665         if (!pblk_line_init_bb(pblk, new, 1)) {
1666                 new = pblk_line_retry(pblk, new);
1667                 if (!new)
1668                         goto out;
1669
1670                 goto retry_setup;
1671         }
1672
1673         pblk_rl_free_lines_dec(&pblk->rl, new, true);
1674
1675         /* Allocate next line for preparation */
1676         spin_lock(&l_mg->free_lock);
1677         l_mg->data_next = pblk_line_get(pblk);
1678         if (!l_mg->data_next) {
1679                 /* If we cannot get a new line, we need to stop the pipeline.
1680                  * Only allow as many writes in as we can store safely and then
1681                  * fail gracefully
1682                  */
1683                 pblk_stop_writes(pblk, new);
1684                 l_mg->data_next = NULL;
1685         } else {
1686                 l_mg->data_next->seq_nr = l_mg->d_seq_nr++;
1687                 l_mg->data_next->type = PBLK_LINETYPE_DATA;
1688         }
1689         spin_unlock(&l_mg->free_lock);
1690
1691 out:
1692         return new;
1693 }
1694
1695 static void __pblk_line_put(struct pblk *pblk, struct pblk_line *line)
1696 {
1697         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1698         struct pblk_gc *gc = &pblk->gc;
1699
1700         spin_lock(&line->lock);
1701         WARN_ON(line->state != PBLK_LINESTATE_GC);
1702         line->state = PBLK_LINESTATE_FREE;
1703         trace_pblk_line_state(pblk_disk_name(pblk), line->id,
1704                                         line->state);
1705         line->gc_group = PBLK_LINEGC_NONE;
1706         pblk_line_free(line);
1707
1708         if (line->w_err_gc->has_write_err) {
1709                 pblk_rl_werr_line_out(&pblk->rl);
1710                 line->w_err_gc->has_write_err = 0;
1711         }
1712
1713         spin_unlock(&line->lock);
1714         atomic_dec(&gc->pipeline_gc);
1715
1716         spin_lock(&l_mg->free_lock);
1717         list_add_tail(&line->list, &l_mg->free_list);
1718         l_mg->nr_free_lines++;
1719         spin_unlock(&l_mg->free_lock);
1720
1721         pblk_rl_free_lines_inc(&pblk->rl, line);
1722 }
1723
1724 static void pblk_line_put_ws(struct work_struct *work)
1725 {
1726         struct pblk_line_ws *line_put_ws = container_of(work,
1727                                                 struct pblk_line_ws, ws);
1728         struct pblk *pblk = line_put_ws->pblk;
1729         struct pblk_line *line = line_put_ws->line;
1730
1731         __pblk_line_put(pblk, line);
1732         mempool_free(line_put_ws, &pblk->gen_ws_pool);
1733 }
1734
1735 void pblk_line_put(struct kref *ref)
1736 {
1737         struct pblk_line *line = container_of(ref, struct pblk_line, ref);
1738         struct pblk *pblk = line->pblk;
1739
1740         __pblk_line_put(pblk, line);
1741 }
1742
1743 void pblk_line_put_wq(struct kref *ref)
1744 {
1745         struct pblk_line *line = container_of(ref, struct pblk_line, ref);
1746         struct pblk *pblk = line->pblk;
1747         struct pblk_line_ws *line_put_ws;
1748
1749         line_put_ws = mempool_alloc(&pblk->gen_ws_pool, GFP_ATOMIC);
1750         if (!line_put_ws)
1751                 return;
1752
1753         line_put_ws->pblk = pblk;
1754         line_put_ws->line = line;
1755         line_put_ws->priv = NULL;
1756
1757         INIT_WORK(&line_put_ws->ws, pblk_line_put_ws);
1758         queue_work(pblk->r_end_wq, &line_put_ws->ws);
1759 }
1760
1761 int pblk_blk_erase_async(struct pblk *pblk, struct ppa_addr ppa)
1762 {
1763         struct nvm_rq *rqd;
1764         int err;
1765
1766         rqd = pblk_alloc_rqd(pblk, PBLK_ERASE);
1767
1768         pblk_setup_e_rq(pblk, rqd, ppa);
1769
1770         rqd->end_io = pblk_end_io_erase;
1771         rqd->private = pblk;
1772
1773         trace_pblk_chunk_reset(pblk_disk_name(pblk),
1774                                 &ppa, PBLK_CHUNK_RESET_START);
1775
1776         /* The write thread schedules erases so that it minimizes disturbances
1777          * with writes. Thus, there is no need to take the LUN semaphore.
1778          */
1779         err = pblk_submit_io(pblk, rqd);
1780         if (err) {
1781                 struct nvm_tgt_dev *dev = pblk->dev;
1782                 struct nvm_geo *geo = &dev->geo;
1783
1784                 pblk_err(pblk, "could not async erase line:%d,blk:%d\n",
1785                                         pblk_ppa_to_line_id(ppa),
1786                                         pblk_ppa_to_pos(geo, ppa));
1787         }
1788
1789         return err;
1790 }
1791
1792 struct pblk_line *pblk_line_get_data(struct pblk *pblk)
1793 {
1794         return pblk->l_mg.data_line;
1795 }
1796
1797 /* For now, always erase next line */
1798 struct pblk_line *pblk_line_get_erase(struct pblk *pblk)
1799 {
1800         return pblk->l_mg.data_next;
1801 }
1802
1803 int pblk_line_is_full(struct pblk_line *line)
1804 {
1805         return (line->left_msecs == 0);
1806 }
1807
1808 static void pblk_line_should_sync_meta(struct pblk *pblk)
1809 {
1810         if (pblk_rl_is_limit(&pblk->rl))
1811                 pblk_line_close_meta_sync(pblk);
1812 }
1813
1814 void pblk_line_close(struct pblk *pblk, struct pblk_line *line)
1815 {
1816         struct nvm_tgt_dev *dev = pblk->dev;
1817         struct nvm_geo *geo = &dev->geo;
1818         struct pblk_line_meta *lm = &pblk->lm;
1819         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1820         struct list_head *move_list;
1821         int i;
1822
1823 #ifdef CONFIG_NVM_PBLK_DEBUG
1824         WARN(!bitmap_full(line->map_bitmap, lm->sec_per_line),
1825                                 "pblk: corrupt closed line %d\n", line->id);
1826 #endif
1827
1828         spin_lock(&l_mg->free_lock);
1829         WARN_ON(!test_and_clear_bit(line->meta_line, &l_mg->meta_bitmap));
1830         spin_unlock(&l_mg->free_lock);
1831
1832         spin_lock(&l_mg->gc_lock);
1833         spin_lock(&line->lock);
1834         WARN_ON(line->state != PBLK_LINESTATE_OPEN);
1835         line->state = PBLK_LINESTATE_CLOSED;
1836         move_list = pblk_line_gc_list(pblk, line);
1837         list_add_tail(&line->list, move_list);
1838
1839         mempool_free(line->map_bitmap, l_mg->bitmap_pool);
1840         line->map_bitmap = NULL;
1841         line->smeta = NULL;
1842         line->emeta = NULL;
1843
1844         for (i = 0; i < lm->blk_per_line; i++) {
1845                 struct pblk_lun *rlun = &pblk->luns[i];
1846                 int pos = pblk_ppa_to_pos(geo, rlun->bppa);
1847                 int state = line->chks[pos].state;
1848
1849                 if (!(state & NVM_CHK_ST_OFFLINE))
1850                         state = NVM_CHK_ST_CLOSED;
1851         }
1852
1853         spin_unlock(&line->lock);
1854         spin_unlock(&l_mg->gc_lock);
1855
1856         trace_pblk_line_state(pblk_disk_name(pblk), line->id,
1857                                         line->state);
1858 }
1859
1860 void pblk_line_close_meta(struct pblk *pblk, struct pblk_line *line)
1861 {
1862         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1863         struct pblk_line_meta *lm = &pblk->lm;
1864         struct pblk_emeta *emeta = line->emeta;
1865         struct line_emeta *emeta_buf = emeta->buf;
1866         struct wa_counters *wa = emeta_to_wa(lm, emeta_buf);
1867
1868         /* No need for exact vsc value; avoid a big line lock and take aprox. */
1869         memcpy(emeta_to_vsc(pblk, emeta_buf), l_mg->vsc_list, lm->vsc_list_len);
1870         memcpy(emeta_to_bb(emeta_buf), line->blk_bitmap, lm->blk_bitmap_len);
1871
1872         wa->user = cpu_to_le64(atomic64_read(&pblk->user_wa));
1873         wa->pad = cpu_to_le64(atomic64_read(&pblk->pad_wa));
1874         wa->gc = cpu_to_le64(atomic64_read(&pblk->gc_wa));
1875
1876         if (le32_to_cpu(emeta_buf->header.identifier) != PBLK_MAGIC) {
1877                 emeta_buf->header.identifier = cpu_to_le32(PBLK_MAGIC);
1878                 guid_copy((guid_t *)&emeta_buf->header.uuid,
1879                                                         &pblk->instance_uuid);
1880                 emeta_buf->header.id = cpu_to_le32(line->id);
1881                 emeta_buf->header.type = cpu_to_le16(line->type);
1882                 emeta_buf->header.version_major = EMETA_VERSION_MAJOR;
1883                 emeta_buf->header.version_minor = EMETA_VERSION_MINOR;
1884                 emeta_buf->header.crc = cpu_to_le32(
1885                         pblk_calc_meta_header_crc(pblk, &emeta_buf->header));
1886         }
1887
1888         emeta_buf->nr_valid_lbas = cpu_to_le64(line->nr_valid_lbas);
1889         emeta_buf->crc = cpu_to_le32(pblk_calc_emeta_crc(pblk, emeta_buf));
1890
1891         spin_lock(&l_mg->close_lock);
1892         spin_lock(&line->lock);
1893
1894         /* Update the in-memory start address for emeta, in case it has
1895          * shifted due to write errors
1896          */
1897         if (line->emeta_ssec != line->cur_sec)
1898                 line->emeta_ssec = line->cur_sec;
1899
1900         list_add_tail(&line->list, &l_mg->emeta_list);
1901         spin_unlock(&line->lock);
1902         spin_unlock(&l_mg->close_lock);
1903
1904         pblk_line_should_sync_meta(pblk);
1905 }
1906
1907 static void pblk_save_lba_list(struct pblk *pblk, struct pblk_line *line)
1908 {
1909         struct pblk_line_meta *lm = &pblk->lm;
1910         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1911         unsigned int lba_list_size = lm->emeta_len[2];
1912         struct pblk_w_err_gc *w_err_gc = line->w_err_gc;
1913         struct pblk_emeta *emeta = line->emeta;
1914
1915         w_err_gc->lba_list = pblk_malloc(lba_list_size,
1916                                          l_mg->emeta_alloc_type, GFP_KERNEL);
1917         memcpy(w_err_gc->lba_list, emeta_to_lbas(pblk, emeta->buf),
1918                                 lba_list_size);
1919 }
1920
1921 void pblk_line_close_ws(struct work_struct *work)
1922 {
1923         struct pblk_line_ws *line_ws = container_of(work, struct pblk_line_ws,
1924                                                                         ws);
1925         struct pblk *pblk = line_ws->pblk;
1926         struct pblk_line *line = line_ws->line;
1927         struct pblk_w_err_gc *w_err_gc = line->w_err_gc;
1928
1929         /* Write errors makes the emeta start address stored in smeta invalid,
1930          * so keep a copy of the lba list until we've gc'd the line
1931          */
1932         if (w_err_gc->has_write_err)
1933                 pblk_save_lba_list(pblk, line);
1934
1935         pblk_line_close(pblk, line);
1936         mempool_free(line_ws, &pblk->gen_ws_pool);
1937 }
1938
1939 void pblk_gen_run_ws(struct pblk *pblk, struct pblk_line *line, void *priv,
1940                       void (*work)(struct work_struct *), gfp_t gfp_mask,
1941                       struct workqueue_struct *wq)
1942 {
1943         struct pblk_line_ws *line_ws;
1944
1945         line_ws = mempool_alloc(&pblk->gen_ws_pool, gfp_mask);
1946
1947         line_ws->pblk = pblk;
1948         line_ws->line = line;
1949         line_ws->priv = priv;
1950
1951         INIT_WORK(&line_ws->ws, work);
1952         queue_work(wq, &line_ws->ws);
1953 }
1954
1955 static void __pblk_down_chunk(struct pblk *pblk, int pos)
1956 {
1957         struct pblk_lun *rlun = &pblk->luns[pos];
1958         int ret;
1959
1960         /*
1961          * Only send one inflight I/O per LUN. Since we map at a page
1962          * granurality, all ppas in the I/O will map to the same LUN
1963          */
1964
1965         ret = down_timeout(&rlun->wr_sem, msecs_to_jiffies(30000));
1966         if (ret == -ETIME || ret == -EINTR)
1967                 pblk_err(pblk, "taking lun semaphore timed out: err %d\n",
1968                                 -ret);
1969 }
1970
1971 void pblk_down_chunk(struct pblk *pblk, struct ppa_addr ppa)
1972 {
1973         struct nvm_tgt_dev *dev = pblk->dev;
1974         struct nvm_geo *geo = &dev->geo;
1975         int pos = pblk_ppa_to_pos(geo, ppa);
1976
1977         __pblk_down_chunk(pblk, pos);
1978 }
1979
1980 void pblk_down_rq(struct pblk *pblk, struct ppa_addr ppa,
1981                   unsigned long *lun_bitmap)
1982 {
1983         struct nvm_tgt_dev *dev = pblk->dev;
1984         struct nvm_geo *geo = &dev->geo;
1985         int pos = pblk_ppa_to_pos(geo, ppa);
1986
1987         /* If the LUN has been locked for this same request, do no attempt to
1988          * lock it again
1989          */
1990         if (test_and_set_bit(pos, lun_bitmap))
1991                 return;
1992
1993         __pblk_down_chunk(pblk, pos);
1994 }
1995
1996 void pblk_up_chunk(struct pblk *pblk, struct ppa_addr ppa)
1997 {
1998         struct nvm_tgt_dev *dev = pblk->dev;
1999         struct nvm_geo *geo = &dev->geo;
2000         struct pblk_lun *rlun;
2001         int pos = pblk_ppa_to_pos(geo, ppa);
2002
2003         rlun = &pblk->luns[pos];
2004         up(&rlun->wr_sem);
2005 }
2006
2007 void pblk_up_rq(struct pblk *pblk, unsigned long *lun_bitmap)
2008 {
2009         struct nvm_tgt_dev *dev = pblk->dev;
2010         struct nvm_geo *geo = &dev->geo;
2011         struct pblk_lun *rlun;
2012         int num_lun = geo->all_luns;
2013         int bit = -1;
2014
2015         while ((bit = find_next_bit(lun_bitmap, num_lun, bit + 1)) < num_lun) {
2016                 rlun = &pblk->luns[bit];
2017                 up(&rlun->wr_sem);
2018         }
2019 }
2020
2021 void pblk_update_map(struct pblk *pblk, sector_t lba, struct ppa_addr ppa)
2022 {
2023         struct ppa_addr ppa_l2p;
2024
2025         /* logic error: lba out-of-bounds. Ignore update */
2026         if (!(lba < pblk->rl.nr_secs)) {
2027                 WARN(1, "pblk: corrupted L2P map request\n");
2028                 return;
2029         }
2030
2031         spin_lock(&pblk->trans_lock);
2032         ppa_l2p = pblk_trans_map_get(pblk, lba);
2033
2034         if (!pblk_addr_in_cache(ppa_l2p) && !pblk_ppa_empty(ppa_l2p))
2035                 pblk_map_invalidate(pblk, ppa_l2p);
2036
2037         pblk_trans_map_set(pblk, lba, ppa);
2038         spin_unlock(&pblk->trans_lock);
2039 }
2040
2041 void pblk_update_map_cache(struct pblk *pblk, sector_t lba, struct ppa_addr ppa)
2042 {
2043
2044 #ifdef CONFIG_NVM_PBLK_DEBUG
2045         /* Callers must ensure that the ppa points to a cache address */
2046         BUG_ON(!pblk_addr_in_cache(ppa));
2047         BUG_ON(pblk_rb_pos_oob(&pblk->rwb, pblk_addr_to_cacheline(ppa)));
2048 #endif
2049
2050         pblk_update_map(pblk, lba, ppa);
2051 }
2052
2053 int pblk_update_map_gc(struct pblk *pblk, sector_t lba, struct ppa_addr ppa_new,
2054                        struct pblk_line *gc_line, u64 paddr_gc)
2055 {
2056         struct ppa_addr ppa_l2p, ppa_gc;
2057         int ret = 1;
2058
2059 #ifdef CONFIG_NVM_PBLK_DEBUG
2060         /* Callers must ensure that the ppa points to a cache address */
2061         BUG_ON(!pblk_addr_in_cache(ppa_new));
2062         BUG_ON(pblk_rb_pos_oob(&pblk->rwb, pblk_addr_to_cacheline(ppa_new)));
2063 #endif
2064
2065         /* logic error: lba out-of-bounds. Ignore update */
2066         if (!(lba < pblk->rl.nr_secs)) {
2067                 WARN(1, "pblk: corrupted L2P map request\n");
2068                 return 0;
2069         }
2070
2071         spin_lock(&pblk->trans_lock);
2072         ppa_l2p = pblk_trans_map_get(pblk, lba);
2073         ppa_gc = addr_to_gen_ppa(pblk, paddr_gc, gc_line->id);
2074
2075         if (!pblk_ppa_comp(ppa_l2p, ppa_gc)) {
2076                 spin_lock(&gc_line->lock);
2077                 WARN(!test_bit(paddr_gc, gc_line->invalid_bitmap),
2078                                                 "pblk: corrupted GC update");
2079                 spin_unlock(&gc_line->lock);
2080
2081                 ret = 0;
2082                 goto out;
2083         }
2084
2085         pblk_trans_map_set(pblk, lba, ppa_new);
2086 out:
2087         spin_unlock(&pblk->trans_lock);
2088         return ret;
2089 }
2090
2091 void pblk_update_map_dev(struct pblk *pblk, sector_t lba,
2092                          struct ppa_addr ppa_mapped, struct ppa_addr ppa_cache)
2093 {
2094         struct ppa_addr ppa_l2p;
2095
2096 #ifdef CONFIG_NVM_PBLK_DEBUG
2097         /* Callers must ensure that the ppa points to a device address */
2098         BUG_ON(pblk_addr_in_cache(ppa_mapped));
2099 #endif
2100         /* Invalidate and discard padded entries */
2101         if (lba == ADDR_EMPTY) {
2102                 atomic64_inc(&pblk->pad_wa);
2103 #ifdef CONFIG_NVM_PBLK_DEBUG
2104                 atomic_long_inc(&pblk->padded_wb);
2105 #endif
2106                 if (!pblk_ppa_empty(ppa_mapped))
2107                         pblk_map_invalidate(pblk, ppa_mapped);
2108                 return;
2109         }
2110
2111         /* logic error: lba out-of-bounds. Ignore update */
2112         if (!(lba < pblk->rl.nr_secs)) {
2113                 WARN(1, "pblk: corrupted L2P map request\n");
2114                 return;
2115         }
2116
2117         spin_lock(&pblk->trans_lock);
2118         ppa_l2p = pblk_trans_map_get(pblk, lba);
2119
2120         /* Do not update L2P if the cacheline has been updated. In this case,
2121          * the mapped ppa must be invalidated
2122          */
2123         if (!pblk_ppa_comp(ppa_l2p, ppa_cache)) {
2124                 if (!pblk_ppa_empty(ppa_mapped))
2125                         pblk_map_invalidate(pblk, ppa_mapped);
2126                 goto out;
2127         }
2128
2129 #ifdef CONFIG_NVM_PBLK_DEBUG
2130         WARN_ON(!pblk_addr_in_cache(ppa_l2p) && !pblk_ppa_empty(ppa_l2p));
2131 #endif
2132
2133         pblk_trans_map_set(pblk, lba, ppa_mapped);
2134 out:
2135         spin_unlock(&pblk->trans_lock);
2136 }
2137
2138 void pblk_lookup_l2p_seq(struct pblk *pblk, struct ppa_addr *ppas,
2139                          sector_t blba, int nr_secs)
2140 {
2141         int i;
2142
2143         spin_lock(&pblk->trans_lock);
2144         for (i = 0; i < nr_secs; i++) {
2145                 struct ppa_addr ppa;
2146
2147                 ppa = ppas[i] = pblk_trans_map_get(pblk, blba + i);
2148
2149                 /* If the L2P entry maps to a line, the reference is valid */
2150                 if (!pblk_ppa_empty(ppa) && !pblk_addr_in_cache(ppa)) {
2151                         struct pblk_line *line = pblk_ppa_to_line(pblk, ppa);
2152
2153                         kref_get(&line->ref);
2154                 }
2155         }
2156         spin_unlock(&pblk->trans_lock);
2157 }
2158
2159 void pblk_lookup_l2p_rand(struct pblk *pblk, struct ppa_addr *ppas,
2160                           u64 *lba_list, int nr_secs)
2161 {
2162         u64 lba;
2163         int i;
2164
2165         spin_lock(&pblk->trans_lock);
2166         for (i = 0; i < nr_secs; i++) {
2167                 lba = lba_list[i];
2168                 if (lba != ADDR_EMPTY) {
2169                         /* logic error: lba out-of-bounds. Ignore update */
2170                         if (!(lba < pblk->rl.nr_secs)) {
2171                                 WARN(1, "pblk: corrupted L2P map request\n");
2172                                 continue;
2173                         }
2174                         ppas[i] = pblk_trans_map_get(pblk, lba);
2175                 }
2176         }
2177         spin_unlock(&pblk->trans_lock);
2178 }
2179
2180 void *pblk_get_meta_for_writes(struct pblk *pblk, struct nvm_rq *rqd)
2181 {
2182         void *buffer;
2183
2184         if (pblk_is_oob_meta_supported(pblk)) {
2185                 /* Just use OOB metadata buffer as always */
2186                 buffer = rqd->meta_list;
2187         } else {
2188                 /* We need to reuse last page of request (packed metadata)
2189                  * in similar way as traditional oob metadata
2190                  */
2191                 buffer = page_to_virt(
2192                         rqd->bio->bi_io_vec[rqd->bio->bi_vcnt - 1].bv_page);
2193         }
2194
2195         return buffer;
2196 }
2197
2198 void pblk_get_packed_meta(struct pblk *pblk, struct nvm_rq *rqd)
2199 {
2200         void *meta_list = rqd->meta_list;
2201         void *page;
2202         int i = 0;
2203
2204         if (pblk_is_oob_meta_supported(pblk))
2205                 return;
2206
2207         page = page_to_virt(rqd->bio->bi_io_vec[rqd->bio->bi_vcnt - 1].bv_page);
2208         /* We need to fill oob meta buffer with data from packed metadata */
2209         for (; i < rqd->nr_ppas; i++)
2210                 memcpy(pblk_get_meta(pblk, meta_list, i),
2211                         page + (i * sizeof(struct pblk_sec_meta)),
2212                         sizeof(struct pblk_sec_meta));
2213 }