drivers/lightnvm/pblk-core.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * Copyright (C) 2016 CNEX Labs
   4  * Initial release: Javier Gonzalez <javier@cnexlabs.com>
   5  *                  Matias Bjorling <matias@cnexlabs.com>
   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 && rqd.error != NVM_RSP_WARN_HIGHECC) {
 765                 pblk_log_read_err(pblk, &rqd);
 766                 ret = -EIO;
 767         }
 768
 769 clear_rqd:
 770         pblk_free_rqd_meta(pblk, &rqd);
 771         return ret;
 772 }
 773
 774 static int pblk_line_smeta_write(struct pblk *pblk, struct pblk_line *line,
 775                                  u64 paddr)
 776 {
 777         struct nvm_tgt_dev *dev = pblk->dev;
 778         struct pblk_line_meta *lm = &pblk->lm;
 779         struct bio *bio;
 780         struct nvm_rq rqd;
 781         __le64 *lba_list = emeta_to_lbas(pblk, line->emeta->buf);
 782         __le64 addr_empty = cpu_to_le64(ADDR_EMPTY);
 783         int i, ret;
 784
 785         memset(&rqd, 0, sizeof(struct nvm_rq));
 786
 787         ret = pblk_alloc_rqd_meta(pblk, &rqd);
 788         if (ret)
 789                 return ret;
 790
 791         bio = bio_map_kern(dev->q, line->smeta, lm->smeta_len, GFP_KERNEL);
 792         if (IS_ERR(bio)) {
 793                 ret = PTR_ERR(bio);
 794                 goto clear_rqd;
 795         }
 796
 797         bio->bi_iter.bi_sector = 0; /* internal bio */
 798         bio_set_op_attrs(bio, REQ_OP_WRITE, 0);
 799
 800         rqd.bio = bio;
 801         rqd.opcode = NVM_OP_PWRITE;
 802         rqd.nr_ppas = lm->smeta_sec;
 803         rqd.is_seq = 1;
 804
 805         for (i = 0; i < lm->smeta_sec; i++, paddr++) {
 806                 struct pblk_sec_meta *meta = pblk_get_meta(pblk,
 807                                                            rqd.meta_list, i);
 808
 809                 rqd.ppa_list[i] = addr_to_gen_ppa(pblk, paddr, line->id);
 810                 meta->lba = lba_list[paddr] = addr_empty;
 811         }
 812
 813         ret = pblk_submit_io_sync_sem(pblk, &rqd);
 814         if (ret) {
 815                 pblk_err(pblk, "smeta I/O submission failed: %d\n", ret);
 816                 bio_put(bio);
 817                 goto clear_rqd;
 818         }
 819
 820         atomic_dec(&pblk->inflight_io);
 821
 822         if (rqd.error) {
 823                 pblk_log_write_err(pblk, &rqd);
 824                 ret = -EIO;
 825         }
 826
 827 clear_rqd:
 828         pblk_free_rqd_meta(pblk, &rqd);
 829         return ret;
 830 }
 831
 832 int pblk_line_emeta_read(struct pblk *pblk, struct pblk_line *line,
 833                          void *emeta_buf)
 834 {
 835         struct nvm_tgt_dev *dev = pblk->dev;
 836         struct nvm_geo *geo = &dev->geo;
 837         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
 838         struct pblk_line_meta *lm = &pblk->lm;
 839         void *ppa_list, *meta_list;
 840         struct bio *bio;
 841         struct nvm_rq rqd;
 842         u64 paddr = line->emeta_ssec;
 843         dma_addr_t dma_ppa_list, dma_meta_list;
 844         int min = pblk->min_write_pgs;
 845         int left_ppas = lm->emeta_sec[0];
 846         int line_id = line->id;
 847         int rq_ppas, rq_len;
 848         int i, j;
 849         int ret;
 850
 851         meta_list = nvm_dev_dma_alloc(dev->parent, GFP_KERNEL,
 852                                                         &dma_meta_list);
 853         if (!meta_list)
 854                 return -ENOMEM;
 855
 856         ppa_list = meta_list + pblk_dma_meta_size(pblk);
 857         dma_ppa_list = dma_meta_list + pblk_dma_meta_size(pblk);
 858
 859 next_rq:
 860         memset(&rqd, 0, sizeof(struct nvm_rq));
 861
 862         rq_ppas = pblk_calc_secs(pblk, left_ppas, 0, false);
 863         rq_len = rq_ppas * geo->csecs;
 864
 865         bio = pblk_bio_map_addr(pblk, emeta_buf, rq_ppas, rq_len,
 866                                         l_mg->emeta_alloc_type, GFP_KERNEL);
 867         if (IS_ERR(bio)) {
 868                 ret = PTR_ERR(bio);
 869                 goto free_rqd_dma;
 870         }
 871
 872         bio->bi_iter.bi_sector = 0; /* internal bio */
 873         bio_set_op_attrs(bio, REQ_OP_READ, 0);
 874
 875         rqd.bio = bio;
 876         rqd.meta_list = meta_list;
 877         rqd.ppa_list = ppa_list;
 878         rqd.dma_meta_list = dma_meta_list;
 879         rqd.dma_ppa_list = dma_ppa_list;
 880         rqd.opcode = NVM_OP_PREAD;
 881         rqd.nr_ppas = rq_ppas;
 882
 883         for (i = 0; i < rqd.nr_ppas; ) {
 884                 struct ppa_addr ppa = addr_to_gen_ppa(pblk, paddr, line_id);
 885                 int pos = pblk_ppa_to_pos(geo, ppa);
 886
 887                 if (pblk_io_aligned(pblk, rq_ppas))
 888                         rqd.is_seq = 1;
 889
 890                 while (test_bit(pos, line->blk_bitmap)) {
 891                         paddr += min;
 892                         if (pblk_boundary_paddr_checks(pblk, paddr)) {
 893                                 bio_put(bio);
 894                                 ret = -EINTR;
 895                                 goto free_rqd_dma;
 896                         }
 897
 898                         ppa = addr_to_gen_ppa(pblk, paddr, line_id);
 899                         pos = pblk_ppa_to_pos(geo, ppa);
 900                 }
 901
 902                 if (pblk_boundary_paddr_checks(pblk, paddr + min)) {
 903                         bio_put(bio);
 904                         ret = -EINTR;
 905                         goto free_rqd_dma;
 906                 }
 907
 908                 for (j = 0; j < min; j++, i++, paddr++)
 909                         rqd.ppa_list[i] = addr_to_gen_ppa(pblk, paddr, line_id);
 910         }
 911
 912         ret = pblk_submit_io_sync(pblk, &rqd);
 913         if (ret) {
 914                 pblk_err(pblk, "emeta I/O submission failed: %d\n", ret);
 915                 bio_put(bio);
 916                 goto free_rqd_dma;
 917         }
 918
 919         atomic_dec(&pblk->inflight_io);
 920
 921         if (rqd.error && rqd.error != NVM_RSP_WARN_HIGHECC) {
 922                 pblk_log_read_err(pblk, &rqd);
 923                 ret = -EIO;
 924                 goto free_rqd_dma;
 925         }
 926
 927         emeta_buf += rq_len;
 928         left_ppas -= rq_ppas;
 929         if (left_ppas)
 930                 goto next_rq;
 931
 932 free_rqd_dma:
 933         nvm_dev_dma_free(dev->parent, rqd.meta_list, rqd.dma_meta_list);
 934         return ret;
 935 }
 936
 937 static void pblk_setup_e_rq(struct pblk *pblk, struct nvm_rq *rqd,
 938                             struct ppa_addr ppa)
 939 {
 940         rqd->opcode = NVM_OP_ERASE;
 941         rqd->ppa_addr = ppa;
 942         rqd->nr_ppas = 1;
 943         rqd->is_seq = 1;
 944         rqd->bio = NULL;
 945 }
 946
 947 static int pblk_blk_erase_sync(struct pblk *pblk, struct ppa_addr ppa)
 948 {
 949         struct nvm_rq rqd = {NULL};
 950         int ret;
 951
 952         trace_pblk_chunk_reset(pblk_disk_name(pblk), &ppa,
 953                                 PBLK_CHUNK_RESET_START);
 954
 955         pblk_setup_e_rq(pblk, &rqd, ppa);
 956
 957         /* The write thread schedules erases so that it minimizes disturbances
 958          * with writes. Thus, there is no need to take the LUN semaphore.
 959          */
 960         ret = pblk_submit_io_sync(pblk, &rqd);
 961         rqd.private = pblk;
 962         __pblk_end_io_erase(pblk, &rqd);
 963
 964         return ret;
 965 }
 966
 967 int pblk_line_erase(struct pblk *pblk, struct pblk_line *line)
 968 {
 969         struct pblk_line_meta *lm = &pblk->lm;
 970         struct ppa_addr ppa;
 971         int ret, bit = -1;
 972
 973         /* Erase only good blocks, one at a time */
 974         do {
 975                 spin_lock(&line->lock);
 976                 bit = find_next_zero_bit(line->erase_bitmap, lm->blk_per_line,
 977                                                                 bit + 1);
 978                 if (bit >= lm->blk_per_line) {
 979                         spin_unlock(&line->lock);
 980                         break;
 981                 }
 982
 983                 ppa = pblk->luns[bit].bppa; /* set ch and lun */
 984                 ppa.a.blk = line->id;
 985
 986                 atomic_dec(&line->left_eblks);
 987                 WARN_ON(test_and_set_bit(bit, line->erase_bitmap));
 988                 spin_unlock(&line->lock);
 989
 990                 ret = pblk_blk_erase_sync(pblk, ppa);
 991                 if (ret) {
 992                         pblk_err(pblk, "failed to erase line %d\n", line->id);
 993                         return ret;
 994                 }
 995         } while (1);
 996
 997         return 0;
 998 }
 999
1000 static void pblk_line_setup_metadata(struct pblk_line *line,
1001                                      struct pblk_line_mgmt *l_mg,
1002                                      struct pblk_line_meta *lm)
1003 {
1004         int meta_line;
1005
1006         lockdep_assert_held(&l_mg->free_lock);
1007
1008 retry_meta:
1009         meta_line = find_first_zero_bit(&l_mg->meta_bitmap, PBLK_DATA_LINES);
1010         if (meta_line == PBLK_DATA_LINES) {
1011                 spin_unlock(&l_mg->free_lock);
1012                 io_schedule();
1013                 spin_lock(&l_mg->free_lock);
1014                 goto retry_meta;
1015         }
1016
1017         set_bit(meta_line, &l_mg->meta_bitmap);
1018         line->meta_line = meta_line;
1019
1020         line->smeta = l_mg->sline_meta[meta_line];
1021         line->emeta = l_mg->eline_meta[meta_line];
1022
1023         memset(line->smeta, 0, lm->smeta_len);
1024         memset(line->emeta->buf, 0, lm->emeta_len[0]);
1025
1026         line->emeta->mem = 0;
1027         atomic_set(&line->emeta->sync, 0);
1028 }
1029
1030 /* For now lines are always assumed full lines. Thus, smeta former and current
1031  * lun bitmaps are omitted.
1032  */
1033 static int pblk_line_init_metadata(struct pblk *pblk, struct pblk_line *line,
1034                                   struct pblk_line *cur)
1035 {
1036         struct nvm_tgt_dev *dev = pblk->dev;
1037         struct nvm_geo *geo = &dev->geo;
1038         struct pblk_line_meta *lm = &pblk->lm;
1039         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1040         struct pblk_emeta *emeta = line->emeta;
1041         struct line_emeta *emeta_buf = emeta->buf;
1042         struct line_smeta *smeta_buf = (struct line_smeta *)line->smeta;
1043         int nr_blk_line;
1044
1045         /* After erasing the line, new bad blocks might appear and we risk
1046          * having an invalid line
1047          */
1048         nr_blk_line = lm->blk_per_line -
1049                         bitmap_weight(line->blk_bitmap, lm->blk_per_line);
1050         if (nr_blk_line < lm->min_blk_line) {
1051                 spin_lock(&l_mg->free_lock);
1052                 spin_lock(&line->lock);
1053                 line->state = PBLK_LINESTATE_BAD;
1054                 trace_pblk_line_state(pblk_disk_name(pblk), line->id,
1055                                         line->state);
1056                 spin_unlock(&line->lock);
1057
1058                 list_add_tail(&line->list, &l_mg->bad_list);
1059                 spin_unlock(&l_mg->free_lock);
1060
1061                 pblk_debug(pblk, "line %d is bad\n", line->id);
1062
1063                 return 0;
1064         }
1065
1066         /* Run-time metadata */
1067         line->lun_bitmap = ((void *)(smeta_buf)) + sizeof(struct line_smeta);
1068
1069         /* Mark LUNs allocated in this line (all for now) */
1070         bitmap_set(line->lun_bitmap, 0, lm->lun_bitmap_len);
1071
1072         smeta_buf->header.identifier = cpu_to_le32(PBLK_MAGIC);
1073         guid_copy((guid_t *)&smeta_buf->header.uuid, &pblk->instance_uuid);
1074         smeta_buf->header.id = cpu_to_le32(line->id);
1075         smeta_buf->header.type = cpu_to_le16(line->type);
1076         smeta_buf->header.version_major = SMETA_VERSION_MAJOR;
1077         smeta_buf->header.version_minor = SMETA_VERSION_MINOR;
1078
1079         /* Start metadata */
1080         smeta_buf->seq_nr = cpu_to_le64(line->seq_nr);
1081         smeta_buf->window_wr_lun = cpu_to_le32(geo->all_luns);
1082
1083         /* Fill metadata among lines */
1084         if (cur) {
1085                 memcpy(line->lun_bitmap, cur->lun_bitmap, lm->lun_bitmap_len);
1086                 smeta_buf->prev_id = cpu_to_le32(cur->id);
1087                 cur->emeta->buf->next_id = cpu_to_le32(line->id);
1088         } else {
1089                 smeta_buf->prev_id = cpu_to_le32(PBLK_LINE_EMPTY);
1090         }
1091
1092         /* All smeta must be set at this point */
1093         smeta_buf->header.crc = cpu_to_le32(
1094                         pblk_calc_meta_header_crc(pblk, &smeta_buf->header));
1095         smeta_buf->crc = cpu_to_le32(pblk_calc_smeta_crc(pblk, smeta_buf));
1096
1097         /* End metadata */
1098         memcpy(&emeta_buf->header, &smeta_buf->header,
1099                                                 sizeof(struct line_header));
1100
1101         emeta_buf->header.version_major = EMETA_VERSION_MAJOR;
1102         emeta_buf->header.version_minor = EMETA_VERSION_MINOR;
1103         emeta_buf->header.crc = cpu_to_le32(
1104                         pblk_calc_meta_header_crc(pblk, &emeta_buf->header));
1105
1106         emeta_buf->seq_nr = cpu_to_le64(line->seq_nr);
1107         emeta_buf->nr_lbas = cpu_to_le64(line->sec_in_line);
1108         emeta_buf->nr_valid_lbas = cpu_to_le64(0);
1109         emeta_buf->next_id = cpu_to_le32(PBLK_LINE_EMPTY);
1110         emeta_buf->crc = cpu_to_le32(0);
1111         emeta_buf->prev_id = smeta_buf->prev_id;
1112
1113         return 1;
1114 }
1115
1116 static int pblk_line_alloc_bitmaps(struct pblk *pblk, struct pblk_line *line)
1117 {
1118         struct pblk_line_meta *lm = &pblk->lm;
1119         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1120
1121         line->map_bitmap = mempool_alloc(l_mg->bitmap_pool, GFP_KERNEL);
1122         if (!line->map_bitmap)
1123                 return -ENOMEM;
1124
1125         memset(line->map_bitmap, 0, lm->sec_bitmap_len);
1126
1127         /* will be initialized using bb info from map_bitmap */
1128         line->invalid_bitmap = mempool_alloc(l_mg->bitmap_pool, GFP_KERNEL);
1129         if (!line->invalid_bitmap) {
1130                 mempool_free(line->map_bitmap, l_mg->bitmap_pool);
1131                 line->map_bitmap = NULL;
1132                 return -ENOMEM;
1133         }
1134
1135         return 0;
1136 }
1137
1138 /* For now lines are always assumed full lines. Thus, smeta former and current
1139  * lun bitmaps are omitted.
1140  */
1141 static int pblk_line_init_bb(struct pblk *pblk, struct pblk_line *line,
1142                              int init)
1143 {
1144         struct nvm_tgt_dev *dev = pblk->dev;
1145         struct nvm_geo *geo = &dev->geo;
1146         struct pblk_line_meta *lm = &pblk->lm;
1147         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1148         u64 off;
1149         int bit = -1;
1150         int emeta_secs;
1151
1152         line->sec_in_line = lm->sec_per_line;
1153
1154         /* Capture bad block information on line mapping bitmaps */
1155         while ((bit = find_next_bit(line->blk_bitmap, lm->blk_per_line,
1156                                         bit + 1)) < lm->blk_per_line) {
1157                 off = bit * geo->ws_opt;
1158                 bitmap_shift_left(l_mg->bb_aux, l_mg->bb_template, off,
1159                                                         lm->sec_per_line);
1160                 bitmap_or(line->map_bitmap, line->map_bitmap, l_mg->bb_aux,
1161                                                         lm->sec_per_line);
1162                 line->sec_in_line -= geo->clba;
1163         }
1164
1165         /* Mark smeta metadata sectors as bad sectors */
1166         bit = find_first_zero_bit(line->blk_bitmap, lm->blk_per_line);
1167         off = bit * geo->ws_opt;
1168         bitmap_set(line->map_bitmap, off, lm->smeta_sec);
1169         line->sec_in_line -= lm->smeta_sec;
1170         line->cur_sec = off + lm->smeta_sec;
1171
1172         if (init && pblk_line_smeta_write(pblk, line, off)) {
1173                 pblk_debug(pblk, "line smeta I/O failed. Retry\n");
1174                 return 0;
1175         }
1176
1177         bitmap_copy(line->invalid_bitmap, line->map_bitmap, lm->sec_per_line);
1178
1179         /* Mark emeta metadata sectors as bad sectors. We need to consider bad
1180          * blocks to make sure that there are enough sectors to store emeta
1181          */
1182         emeta_secs = lm->emeta_sec[0];
1183         off = lm->sec_per_line;
1184         while (emeta_secs) {
1185                 off -= geo->ws_opt;
1186                 if (!test_bit(off, line->invalid_bitmap)) {
1187                         bitmap_set(line->invalid_bitmap, off, geo->ws_opt);
1188                         emeta_secs -= geo->ws_opt;
1189                 }
1190         }
1191
1192         line->emeta_ssec = off;
1193         line->sec_in_line -= lm->emeta_sec[0];
1194         line->nr_valid_lbas = 0;
1195         line->left_msecs = line->sec_in_line;
1196         *line->vsc = cpu_to_le32(line->sec_in_line);
1197
1198         if (lm->sec_per_line - line->sec_in_line !=
1199                 bitmap_weight(line->invalid_bitmap, lm->sec_per_line)) {
1200                 spin_lock(&line->lock);
1201                 line->state = PBLK_LINESTATE_BAD;
1202                 trace_pblk_line_state(pblk_disk_name(pblk), line->id,
1203                                         line->state);
1204                 spin_unlock(&line->lock);
1205
1206                 list_add_tail(&line->list, &l_mg->bad_list);
1207                 pblk_err(pblk, "unexpected line %d is bad\n", line->id);
1208
1209                 return 0;
1210         }
1211
1212         return 1;
1213 }
1214
1215 static int pblk_prepare_new_line(struct pblk *pblk, struct pblk_line *line)
1216 {
1217         struct pblk_line_meta *lm = &pblk->lm;
1218         struct nvm_tgt_dev *dev = pblk->dev;
1219         struct nvm_geo *geo = &dev->geo;
1220         int blk_to_erase = atomic_read(&line->blk_in_line);
1221         int i;
1222
1223         for (i = 0; i < lm->blk_per_line; i++) {
1224                 struct pblk_lun *rlun = &pblk->luns[i];
1225                 int pos = pblk_ppa_to_pos(geo, rlun->bppa);
1226                 int state = line->chks[pos].state;
1227
1228                 /* Free chunks should not be erased */
1229                 if (state & NVM_CHK_ST_FREE) {
1230                         set_bit(pblk_ppa_to_pos(geo, rlun->bppa),
1231                                                         line->erase_bitmap);
1232                         blk_to_erase--;
1233                 }
1234         }
1235
1236         return blk_to_erase;
1237 }
1238
1239 static int pblk_line_prepare(struct pblk *pblk, struct pblk_line *line)
1240 {
1241         struct pblk_line_meta *lm = &pblk->lm;
1242         int blk_in_line = atomic_read(&line->blk_in_line);
1243         int blk_to_erase;
1244
1245         /* Bad blocks do not need to be erased */
1246         bitmap_copy(line->erase_bitmap, line->blk_bitmap, lm->blk_per_line);
1247
1248         spin_lock(&line->lock);
1249
1250         /* If we have not written to this line, we need to mark up free chunks
1251          * as already erased
1252          */
1253         if (line->state == PBLK_LINESTATE_NEW) {
1254                 blk_to_erase = pblk_prepare_new_line(pblk, line);
1255                 line->state = PBLK_LINESTATE_FREE;
1256                 trace_pblk_line_state(pblk_disk_name(pblk), line->id,
1257                                         line->state);
1258         } else {
1259                 blk_to_erase = blk_in_line;
1260         }
1261
1262         if (blk_in_line < lm->min_blk_line) {
1263                 spin_unlock(&line->lock);
1264                 return -EAGAIN;
1265         }
1266
1267         if (line->state != PBLK_LINESTATE_FREE) {
1268                 WARN(1, "pblk: corrupted line %d, state %d\n",
1269                                                         line->id, line->state);
1270                 spin_unlock(&line->lock);
1271                 return -EINTR;
1272         }
1273
1274         line->state = PBLK_LINESTATE_OPEN;
1275         trace_pblk_line_state(pblk_disk_name(pblk), line->id,
1276                                 line->state);
1277
1278         atomic_set(&line->left_eblks, blk_to_erase);
1279         atomic_set(&line->left_seblks, blk_to_erase);
1280
1281         line->meta_distance = lm->meta_distance;
1282         spin_unlock(&line->lock);
1283
1284         kref_init(&line->ref);
1285         atomic_set(&line->sec_to_update, 0);
1286
1287         return 0;
1288 }
1289
1290 /* Line allocations in the recovery path are always single threaded */
1291 int pblk_line_recov_alloc(struct pblk *pblk, struct pblk_line *line)
1292 {
1293         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1294         int ret;
1295
1296         spin_lock(&l_mg->free_lock);
1297         l_mg->data_line = line;
1298         list_del(&line->list);
1299
1300         ret = pblk_line_prepare(pblk, line);
1301         if (ret) {
1302                 list_add(&line->list, &l_mg->free_list);
1303                 spin_unlock(&l_mg->free_lock);
1304                 return ret;
1305         }
1306         spin_unlock(&l_mg->free_lock);
1307
1308         ret = pblk_line_alloc_bitmaps(pblk, line);
1309         if (ret)
1310                 goto fail;
1311
1312         if (!pblk_line_init_bb(pblk, line, 0)) {
1313                 ret = -EINTR;
1314                 goto fail;
1315         }
1316
1317         pblk_rl_free_lines_dec(&pblk->rl, line, true);
1318         return 0;
1319
1320 fail:
1321         spin_lock(&l_mg->free_lock);
1322         list_add(&line->list, &l_mg->free_list);
1323         spin_unlock(&l_mg->free_lock);
1324
1325         return ret;
1326 }
1327
1328 void pblk_line_recov_close(struct pblk *pblk, struct pblk_line *line)
1329 {
1330         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1331
1332         mempool_free(line->map_bitmap, l_mg->bitmap_pool);
1333         line->map_bitmap = NULL;
1334         line->smeta = NULL;
1335         line->emeta = NULL;
1336 }
1337
1338 static void pblk_line_reinit(struct pblk_line *line)
1339 {
1340         *line->vsc = cpu_to_le32(EMPTY_ENTRY);
1341
1342         line->map_bitmap = NULL;
1343         line->invalid_bitmap = NULL;
1344         line->smeta = NULL;
1345         line->emeta = NULL;
1346 }
1347
1348 void pblk_line_free(struct pblk_line *line)
1349 {
1350         struct pblk *pblk = line->pblk;
1351         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1352
1353         mempool_free(line->map_bitmap, l_mg->bitmap_pool);
1354         mempool_free(line->invalid_bitmap, l_mg->bitmap_pool);
1355
1356         pblk_line_reinit(line);
1357 }
1358
1359 struct pblk_line *pblk_line_get(struct pblk *pblk)
1360 {
1361         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1362         struct pblk_line_meta *lm = &pblk->lm;
1363         struct pblk_line *line;
1364         int ret, bit;
1365
1366         lockdep_assert_held(&l_mg->free_lock);
1367
1368 retry:
1369         if (list_empty(&l_mg->free_list)) {
1370                 pblk_err(pblk, "no free lines\n");
1371                 return NULL;
1372         }
1373
1374         line = list_first_entry(&l_mg->free_list, struct pblk_line, list);
1375         list_del(&line->list);
1376         l_mg->nr_free_lines--;
1377
1378         bit = find_first_zero_bit(line->blk_bitmap, lm->blk_per_line);
1379         if (unlikely(bit >= lm->blk_per_line)) {
1380                 spin_lock(&line->lock);
1381                 line->state = PBLK_LINESTATE_BAD;
1382                 trace_pblk_line_state(pblk_disk_name(pblk), line->id,
1383                                         line->state);
1384                 spin_unlock(&line->lock);
1385
1386                 list_add_tail(&line->list, &l_mg->bad_list);
1387
1388                 pblk_debug(pblk, "line %d is bad\n", line->id);
1389                 goto retry;
1390         }
1391
1392         ret = pblk_line_prepare(pblk, line);
1393         if (ret) {
1394                 switch (ret) {
1395                 case -EAGAIN:
1396                         list_add(&line->list, &l_mg->bad_list);
1397                         goto retry;
1398                 case -EINTR:
1399                         list_add(&line->list, &l_mg->corrupt_list);
1400                         goto retry;
1401                 default:
1402                         pblk_err(pblk, "failed to prepare line %d\n", line->id);
1403                         list_add(&line->list, &l_mg->free_list);
1404                         l_mg->nr_free_lines++;
1405                         return NULL;
1406                 }
1407         }
1408
1409         return line;
1410 }
1411
1412 static struct pblk_line *pblk_line_retry(struct pblk *pblk,
1413                                          struct pblk_line *line)
1414 {
1415         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1416         struct pblk_line *retry_line;
1417
1418 retry:
1419         spin_lock(&l_mg->free_lock);
1420         retry_line = pblk_line_get(pblk);
1421         if (!retry_line) {
1422                 l_mg->data_line = NULL;
1423                 spin_unlock(&l_mg->free_lock);
1424                 return NULL;
1425         }
1426
1427         retry_line->map_bitmap = line->map_bitmap;
1428         retry_line->invalid_bitmap = line->invalid_bitmap;
1429         retry_line->smeta = line->smeta;
1430         retry_line->emeta = line->emeta;
1431         retry_line->meta_line = line->meta_line;
1432
1433         pblk_line_reinit(line);
1434
1435         l_mg->data_line = retry_line;
1436         spin_unlock(&l_mg->free_lock);
1437
1438         pblk_rl_free_lines_dec(&pblk->rl, line, false);
1439
1440         if (pblk_line_erase(pblk, retry_line))
1441                 goto retry;
1442
1443         return retry_line;
1444 }
1445
1446 static void pblk_set_space_limit(struct pblk *pblk)
1447 {
1448         struct pblk_rl *rl = &pblk->rl;
1449
1450         atomic_set(&rl->rb_space, 0);
1451 }
1452
1453 struct pblk_line *pblk_line_get_first_data(struct pblk *pblk)
1454 {
1455         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1456         struct pblk_line *line;
1457
1458         spin_lock(&l_mg->free_lock);
1459         line = pblk_line_get(pblk);
1460         if (!line) {
1461                 spin_unlock(&l_mg->free_lock);
1462                 return NULL;
1463         }
1464
1465         line->seq_nr = l_mg->d_seq_nr++;
1466         line->type = PBLK_LINETYPE_DATA;
1467         l_mg->data_line = line;
1468
1469         pblk_line_setup_metadata(line, l_mg, &pblk->lm);
1470
1471         /* Allocate next line for preparation */
1472         l_mg->data_next = pblk_line_get(pblk);
1473         if (!l_mg->data_next) {
1474                 /* If we cannot get a new line, we need to stop the pipeline.
1475                  * Only allow as many writes in as we can store safely and then
1476                  * fail gracefully
1477                  */
1478                 pblk_set_space_limit(pblk);
1479
1480                 l_mg->data_next = NULL;
1481         } else {
1482                 l_mg->data_next->seq_nr = l_mg->d_seq_nr++;
1483                 l_mg->data_next->type = PBLK_LINETYPE_DATA;
1484         }
1485         spin_unlock(&l_mg->free_lock);
1486
1487         if (pblk_line_alloc_bitmaps(pblk, line))
1488                 return NULL;
1489
1490         if (pblk_line_erase(pblk, line)) {
1491                 line = pblk_line_retry(pblk, line);
1492                 if (!line)
1493                         return NULL;
1494         }
1495
1496 retry_setup:
1497         if (!pblk_line_init_metadata(pblk, line, NULL)) {
1498                 line = pblk_line_retry(pblk, line);
1499                 if (!line)
1500                         return NULL;
1501
1502                 goto retry_setup;
1503         }
1504
1505         if (!pblk_line_init_bb(pblk, line, 1)) {
1506                 line = pblk_line_retry(pblk, line);
1507                 if (!line)
1508                         return NULL;
1509
1510                 goto retry_setup;
1511         }
1512
1513         pblk_rl_free_lines_dec(&pblk->rl, line, true);
1514
1515         return line;
1516 }
1517
1518 void pblk_ppa_to_line_put(struct pblk *pblk, struct ppa_addr ppa)
1519 {
1520         struct pblk_line *line;
1521
1522         line = pblk_ppa_to_line(pblk, ppa);
1523         kref_put(&line->ref, pblk_line_put_wq);
1524 }
1525
1526 void pblk_rq_to_line_put(struct pblk *pblk, struct nvm_rq *rqd)
1527 {
1528         struct ppa_addr *ppa_list;
1529         int i;
1530
1531         ppa_list = (rqd->nr_ppas > 1) ? rqd->ppa_list : &rqd->ppa_addr;
1532
1533         for (i = 0; i < rqd->nr_ppas; i++)
1534                 pblk_ppa_to_line_put(pblk, ppa_list[i]);
1535 }
1536
1537 static void pblk_stop_writes(struct pblk *pblk, struct pblk_line *line)
1538 {
1539         lockdep_assert_held(&pblk->l_mg.free_lock);
1540
1541         pblk_set_space_limit(pblk);
1542         pblk->state = PBLK_STATE_STOPPING;
1543         trace_pblk_state(pblk_disk_name(pblk), pblk->state);
1544 }
1545
1546 static void pblk_line_close_meta_sync(struct pblk *pblk)
1547 {
1548         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1549         struct pblk_line_meta *lm = &pblk->lm;
1550         struct pblk_line *line, *tline;
1551         LIST_HEAD(list);
1552
1553         spin_lock(&l_mg->close_lock);
1554         if (list_empty(&l_mg->emeta_list)) {
1555                 spin_unlock(&l_mg->close_lock);
1556                 return;
1557         }
1558
1559         list_cut_position(&list, &l_mg->emeta_list, l_mg->emeta_list.prev);
1560         spin_unlock(&l_mg->close_lock);
1561
1562         list_for_each_entry_safe(line, tline, &list, list) {
1563                 struct pblk_emeta *emeta = line->emeta;
1564
1565                 while (emeta->mem < lm->emeta_len[0]) {
1566                         int ret;
1567
1568                         ret = pblk_submit_meta_io(pblk, line);
1569                         if (ret) {
1570                                 pblk_err(pblk, "sync meta line %d failed (%d)\n",
1571                                                         line->id, ret);
1572                                 return;
1573                         }
1574                 }
1575         }
1576
1577         pblk_wait_for_meta(pblk);
1578         flush_workqueue(pblk->close_wq);
1579 }
1580
1581 void __pblk_pipeline_flush(struct pblk *pblk)
1582 {
1583         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1584         int ret;
1585
1586         spin_lock(&l_mg->free_lock);
1587         if (pblk->state == PBLK_STATE_RECOVERING ||
1588                                         pblk->state == PBLK_STATE_STOPPED) {
1589                 spin_unlock(&l_mg->free_lock);
1590                 return;
1591         }
1592         pblk->state = PBLK_STATE_RECOVERING;
1593         trace_pblk_state(pblk_disk_name(pblk), pblk->state);
1594         spin_unlock(&l_mg->free_lock);
1595
1596         pblk_flush_writer(pblk);
1597         pblk_wait_for_meta(pblk);
1598
1599         ret = pblk_recov_pad(pblk);
1600         if (ret) {
1601                 pblk_err(pblk, "could not close data on teardown(%d)\n", ret);
1602                 return;
1603         }
1604
1605         flush_workqueue(pblk->bb_wq);
1606         pblk_line_close_meta_sync(pblk);
1607 }
1608
1609 void __pblk_pipeline_stop(struct pblk *pblk)
1610 {
1611         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1612
1613         spin_lock(&l_mg->free_lock);
1614         pblk->state = PBLK_STATE_STOPPED;
1615         trace_pblk_state(pblk_disk_name(pblk), pblk->state);
1616         l_mg->data_line = NULL;
1617         l_mg->data_next = NULL;
1618         spin_unlock(&l_mg->free_lock);
1619 }
1620
1621 void pblk_pipeline_stop(struct pblk *pblk)
1622 {
1623         __pblk_pipeline_flush(pblk);
1624         __pblk_pipeline_stop(pblk);
1625 }
1626
1627 struct pblk_line *pblk_line_replace_data(struct pblk *pblk)
1628 {
1629         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1630         struct pblk_line *cur, *new = NULL;
1631         unsigned int left_seblks;
1632
1633         new = l_mg->data_next;
1634         if (!new)
1635                 goto out;
1636
1637         spin_lock(&l_mg->free_lock);
1638         cur = l_mg->data_line;
1639         l_mg->data_line = new;
1640
1641         pblk_line_setup_metadata(new, l_mg, &pblk->lm);
1642         spin_unlock(&l_mg->free_lock);
1643
1644 retry_erase:
1645         left_seblks = atomic_read(&new->left_seblks);
1646         if (left_seblks) {
1647                 /* If line is not fully erased, erase it */
1648                 if (atomic_read(&new->left_eblks)) {
1649                         if (pblk_line_erase(pblk, new))
1650                                 goto out;
1651                 } else {
1652                         io_schedule();
1653                 }
1654                 goto retry_erase;
1655         }
1656
1657         if (pblk_line_alloc_bitmaps(pblk, new))
1658                 return NULL;
1659
1660 retry_setup:
1661         if (!pblk_line_init_metadata(pblk, new, cur)) {
1662                 new = pblk_line_retry(pblk, new);
1663                 if (!new)
1664                         goto out;
1665
1666                 goto retry_setup;
1667         }
1668
1669         if (!pblk_line_init_bb(pblk, new, 1)) {
1670                 new = pblk_line_retry(pblk, new);
1671                 if (!new)
1672                         goto out;
1673
1674                 goto retry_setup;
1675         }
1676
1677         pblk_rl_free_lines_dec(&pblk->rl, new, true);
1678
1679         /* Allocate next line for preparation */
1680         spin_lock(&l_mg->free_lock);
1681         l_mg->data_next = pblk_line_get(pblk);
1682         if (!l_mg->data_next) {
1683                 /* If we cannot get a new line, we need to stop the pipeline.
1684                  * Only allow as many writes in as we can store safely and then
1685                  * fail gracefully
1686                  */
1687                 pblk_stop_writes(pblk, new);
1688                 l_mg->data_next = NULL;
1689         } else {
1690                 l_mg->data_next->seq_nr = l_mg->d_seq_nr++;
1691                 l_mg->data_next->type = PBLK_LINETYPE_DATA;
1692         }
1693         spin_unlock(&l_mg->free_lock);
1694
1695 out:
1696         return new;
1697 }
1698
1699 static void __pblk_line_put(struct pblk *pblk, struct pblk_line *line)
1700 {
1701         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1702         struct pblk_gc *gc = &pblk->gc;
1703
1704         spin_lock(&line->lock);
1705         WARN_ON(line->state != PBLK_LINESTATE_GC);
1706         if (line->w_err_gc->has_gc_err) {
1707                 spin_unlock(&line->lock);
1708                 pblk_err(pblk, "line %d had errors during GC\n", line->id);
1709                 pblk_put_line_back(pblk, line);
1710                 line->w_err_gc->has_gc_err = 0;
1711                 return;
1712         }
1713
1714         line->state = PBLK_LINESTATE_FREE;
1715         trace_pblk_line_state(pblk_disk_name(pblk), line->id,
1716                                         line->state);
1717         line->gc_group = PBLK_LINEGC_NONE;
1718         pblk_line_free(line);
1719
1720         if (line->w_err_gc->has_write_err) {
1721                 pblk_rl_werr_line_out(&pblk->rl);
1722                 line->w_err_gc->has_write_err = 0;
1723         }
1724
1725         spin_unlock(&line->lock);
1726         atomic_dec(&gc->pipeline_gc);
1727
1728         spin_lock(&l_mg->free_lock);
1729         list_add_tail(&line->list, &l_mg->free_list);
1730         l_mg->nr_free_lines++;
1731         spin_unlock(&l_mg->free_lock);
1732
1733         pblk_rl_free_lines_inc(&pblk->rl, line);
1734 }
1735
1736 static void pblk_line_put_ws(struct work_struct *work)
1737 {
1738         struct pblk_line_ws *line_put_ws = container_of(work,
1739                                                 struct pblk_line_ws, ws);
1740         struct pblk *pblk = line_put_ws->pblk;
1741         struct pblk_line *line = line_put_ws->line;
1742
1743         __pblk_line_put(pblk, line);
1744         mempool_free(line_put_ws, &pblk->gen_ws_pool);
1745 }
1746
1747 void pblk_line_put(struct kref *ref)
1748 {
1749         struct pblk_line *line = container_of(ref, struct pblk_line, ref);
1750         struct pblk *pblk = line->pblk;
1751
1752         __pblk_line_put(pblk, line);
1753 }
1754
1755 void pblk_line_put_wq(struct kref *ref)
1756 {
1757         struct pblk_line *line = container_of(ref, struct pblk_line, ref);
1758         struct pblk *pblk = line->pblk;
1759         struct pblk_line_ws *line_put_ws;
1760
1761         line_put_ws = mempool_alloc(&pblk->gen_ws_pool, GFP_ATOMIC);
1762         if (!line_put_ws)
1763                 return;
1764
1765         line_put_ws->pblk = pblk;
1766         line_put_ws->line = line;
1767         line_put_ws->priv = NULL;
1768
1769         INIT_WORK(&line_put_ws->ws, pblk_line_put_ws);
1770         queue_work(pblk->r_end_wq, &line_put_ws->ws);
1771 }
1772
1773 int pblk_blk_erase_async(struct pblk *pblk, struct ppa_addr ppa)
1774 {
1775         struct nvm_rq *rqd;
1776         int err;
1777
1778         rqd = pblk_alloc_rqd(pblk, PBLK_ERASE);
1779
1780         pblk_setup_e_rq(pblk, rqd, ppa);
1781
1782         rqd->end_io = pblk_end_io_erase;
1783         rqd->private = pblk;
1784
1785         trace_pblk_chunk_reset(pblk_disk_name(pblk),
1786                                 &ppa, PBLK_CHUNK_RESET_START);
1787
1788         /* The write thread schedules erases so that it minimizes disturbances
1789          * with writes. Thus, there is no need to take the LUN semaphore.
1790          */
1791         err = pblk_submit_io(pblk, rqd);
1792         if (err) {
1793                 struct nvm_tgt_dev *dev = pblk->dev;
1794                 struct nvm_geo *geo = &dev->geo;
1795
1796                 pblk_err(pblk, "could not async erase line:%d,blk:%d\n",
1797                                         pblk_ppa_to_line_id(ppa),
1798                                         pblk_ppa_to_pos(geo, ppa));
1799         }
1800
1801         return err;
1802 }
1803
1804 struct pblk_line *pblk_line_get_data(struct pblk *pblk)
1805 {
1806         return pblk->l_mg.data_line;
1807 }
1808
1809 /* For now, always erase next line */
1810 struct pblk_line *pblk_line_get_erase(struct pblk *pblk)
1811 {
1812         return pblk->l_mg.data_next;
1813 }
1814
1815 int pblk_line_is_full(struct pblk_line *line)
1816 {
1817         return (line->left_msecs == 0);
1818 }
1819
1820 static void pblk_line_should_sync_meta(struct pblk *pblk)
1821 {
1822         if (pblk_rl_is_limit(&pblk->rl))
1823                 pblk_line_close_meta_sync(pblk);
1824 }
1825
1826 void pblk_line_close(struct pblk *pblk, struct pblk_line *line)
1827 {
1828         struct nvm_tgt_dev *dev = pblk->dev;
1829         struct nvm_geo *geo = &dev->geo;
1830         struct pblk_line_meta *lm = &pblk->lm;
1831         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1832         struct list_head *move_list;
1833         int i;
1834
1835 #ifdef CONFIG_NVM_PBLK_DEBUG
1836         WARN(!bitmap_full(line->map_bitmap, lm->sec_per_line),
1837                                 "pblk: corrupt closed line %d\n", line->id);
1838 #endif
1839
1840         spin_lock(&l_mg->free_lock);
1841         WARN_ON(!test_and_clear_bit(line->meta_line, &l_mg->meta_bitmap));
1842         spin_unlock(&l_mg->free_lock);
1843
1844         spin_lock(&l_mg->gc_lock);
1845         spin_lock(&line->lock);
1846         WARN_ON(line->state != PBLK_LINESTATE_OPEN);
1847         line->state = PBLK_LINESTATE_CLOSED;
1848         move_list = pblk_line_gc_list(pblk, line);
1849         list_add_tail(&line->list, move_list);
1850
1851         mempool_free(line->map_bitmap, l_mg->bitmap_pool);
1852         line->map_bitmap = NULL;
1853         line->smeta = NULL;
1854         line->emeta = NULL;
1855
1856         for (i = 0; i < lm->blk_per_line; i++) {
1857                 struct pblk_lun *rlun = &pblk->luns[i];
1858                 int pos = pblk_ppa_to_pos(geo, rlun->bppa);
1859                 int state = line->chks[pos].state;
1860
1861                 if (!(state & NVM_CHK_ST_OFFLINE))
1862                         state = NVM_CHK_ST_CLOSED;
1863         }
1864
1865         spin_unlock(&line->lock);
1866         spin_unlock(&l_mg->gc_lock);
1867
1868         trace_pblk_line_state(pblk_disk_name(pblk), line->id,
1869                                         line->state);
1870 }
1871
1872 void pblk_line_close_meta(struct pblk *pblk, struct pblk_line *line)
1873 {
1874         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1875         struct pblk_line_meta *lm = &pblk->lm;
1876         struct pblk_emeta *emeta = line->emeta;
1877         struct line_emeta *emeta_buf = emeta->buf;
1878         struct wa_counters *wa = emeta_to_wa(lm, emeta_buf);
1879
1880         /* No need for exact vsc value; avoid a big line lock and take aprox. */
1881         memcpy(emeta_to_vsc(pblk, emeta_buf), l_mg->vsc_list, lm->vsc_list_len);
1882         memcpy(emeta_to_bb(emeta_buf), line->blk_bitmap, lm->blk_bitmap_len);
1883
1884         wa->user = cpu_to_le64(atomic64_read(&pblk->user_wa));
1885         wa->pad = cpu_to_le64(atomic64_read(&pblk->pad_wa));
1886         wa->gc = cpu_to_le64(atomic64_read(&pblk->gc_wa));
1887
1888         if (le32_to_cpu(emeta_buf->header.identifier) != PBLK_MAGIC) {
1889                 emeta_buf->header.identifier = cpu_to_le32(PBLK_MAGIC);
1890                 guid_copy((guid_t *)&emeta_buf->header.uuid,
1891                                                         &pblk->instance_uuid);
1892                 emeta_buf->header.id = cpu_to_le32(line->id);
1893                 emeta_buf->header.type = cpu_to_le16(line->type);
1894                 emeta_buf->header.version_major = EMETA_VERSION_MAJOR;
1895                 emeta_buf->header.version_minor = EMETA_VERSION_MINOR;
1896                 emeta_buf->header.crc = cpu_to_le32(
1897                         pblk_calc_meta_header_crc(pblk, &emeta_buf->header));
1898         }
1899
1900         emeta_buf->nr_valid_lbas = cpu_to_le64(line->nr_valid_lbas);
1901         emeta_buf->crc = cpu_to_le32(pblk_calc_emeta_crc(pblk, emeta_buf));
1902
1903         spin_lock(&l_mg->close_lock);
1904         spin_lock(&line->lock);
1905
1906         /* Update the in-memory start address for emeta, in case it has
1907          * shifted due to write errors
1908          */
1909         if (line->emeta_ssec != line->cur_sec)
1910                 line->emeta_ssec = line->cur_sec;
1911
1912         list_add_tail(&line->list, &l_mg->emeta_list);
1913         spin_unlock(&line->lock);
1914         spin_unlock(&l_mg->close_lock);
1915
1916         pblk_line_should_sync_meta(pblk);
1917 }
1918
1919 static void pblk_save_lba_list(struct pblk *pblk, struct pblk_line *line)
1920 {
1921         struct pblk_line_meta *lm = &pblk->lm;
1922         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1923         unsigned int lba_list_size = lm->emeta_len[2];
1924         struct pblk_w_err_gc *w_err_gc = line->w_err_gc;
1925         struct pblk_emeta *emeta = line->emeta;
1926
1927         w_err_gc->lba_list = pblk_malloc(lba_list_size,
1928                                          l_mg->emeta_alloc_type, GFP_KERNEL);
1929         memcpy(w_err_gc->lba_list, emeta_to_lbas(pblk, emeta->buf),
1930                                 lba_list_size);
1931 }
1932
1933 void pblk_line_close_ws(struct work_struct *work)
1934 {
1935         struct pblk_line_ws *line_ws = container_of(work, struct pblk_line_ws,
1936                                                                         ws);
1937         struct pblk *pblk = line_ws->pblk;
1938         struct pblk_line *line = line_ws->line;
1939         struct pblk_w_err_gc *w_err_gc = line->w_err_gc;
1940
1941         /* Write errors makes the emeta start address stored in smeta invalid,
1942          * so keep a copy of the lba list until we've gc'd the line
1943          */
1944         if (w_err_gc->has_write_err)
1945                 pblk_save_lba_list(pblk, line);
1946
1947         pblk_line_close(pblk, line);
1948         mempool_free(line_ws, &pblk->gen_ws_pool);
1949 }
1950
1951 void pblk_gen_run_ws(struct pblk *pblk, struct pblk_line *line, void *priv,
1952                       void (*work)(struct work_struct *), gfp_t gfp_mask,
1953                       struct workqueue_struct *wq)
1954 {
1955         struct pblk_line_ws *line_ws;
1956
1957         line_ws = mempool_alloc(&pblk->gen_ws_pool, gfp_mask);
1958
1959         line_ws->pblk = pblk;
1960         line_ws->line = line;
1961         line_ws->priv = priv;
1962
1963         INIT_WORK(&line_ws->ws, work);
1964         queue_work(wq, &line_ws->ws);
1965 }
1966
1967 static void __pblk_down_chunk(struct pblk *pblk, int pos)
1968 {
1969         struct pblk_lun *rlun = &pblk->luns[pos];
1970         int ret;
1971
1972         /*
1973          * Only send one inflight I/O per LUN. Since we map at a page
1974          * granurality, all ppas in the I/O will map to the same LUN
1975          */
1976
1977         ret = down_timeout(&rlun->wr_sem, msecs_to_jiffies(30000));
1978         if (ret == -ETIME || ret == -EINTR)
1979                 pblk_err(pblk, "taking lun semaphore timed out: err %d\n",
1980                                 -ret);
1981 }
1982
1983 void pblk_down_chunk(struct pblk *pblk, struct ppa_addr ppa)
1984 {
1985         struct nvm_tgt_dev *dev = pblk->dev;
1986         struct nvm_geo *geo = &dev->geo;
1987         int pos = pblk_ppa_to_pos(geo, ppa);
1988
1989         __pblk_down_chunk(pblk, pos);
1990 }
1991
1992 void pblk_down_rq(struct pblk *pblk, struct ppa_addr ppa,
1993                   unsigned long *lun_bitmap)
1994 {
1995         struct nvm_tgt_dev *dev = pblk->dev;
1996         struct nvm_geo *geo = &dev->geo;
1997         int pos = pblk_ppa_to_pos(geo, ppa);
1998
1999         /* If the LUN has been locked for this same request, do no attempt to
2000          * lock it again
2001          */
2002         if (test_and_set_bit(pos, lun_bitmap))
2003                 return;
2004
2005         __pblk_down_chunk(pblk, pos);
2006 }
2007
2008 void pblk_up_chunk(struct pblk *pblk, struct ppa_addr ppa)
2009 {
2010         struct nvm_tgt_dev *dev = pblk->dev;
2011         struct nvm_geo *geo = &dev->geo;
2012         struct pblk_lun *rlun;
2013         int pos = pblk_ppa_to_pos(geo, ppa);
2014
2015         rlun = &pblk->luns[pos];
2016         up(&rlun->wr_sem);
2017 }
2018
2019 void pblk_up_rq(struct pblk *pblk, unsigned long *lun_bitmap)
2020 {
2021         struct nvm_tgt_dev *dev = pblk->dev;
2022         struct nvm_geo *geo = &dev->geo;
2023         struct pblk_lun *rlun;
2024         int num_lun = geo->all_luns;
2025         int bit = -1;
2026
2027         while ((bit = find_next_bit(lun_bitmap, num_lun, bit + 1)) < num_lun) {
2028                 rlun = &pblk->luns[bit];
2029                 up(&rlun->wr_sem);
2030         }
2031 }
2032
2033 void pblk_update_map(struct pblk *pblk, sector_t lba, struct ppa_addr ppa)
2034 {
2035         struct ppa_addr ppa_l2p;
2036
2037         /* logic error: lba out-of-bounds. Ignore update */
2038         if (!(lba < pblk->capacity)) {
2039                 WARN(1, "pblk: corrupted L2P map request\n");
2040                 return;
2041         }
2042
2043         spin_lock(&pblk->trans_lock);
2044         ppa_l2p = pblk_trans_map_get(pblk, lba);
2045
2046         if (!pblk_addr_in_cache(ppa_l2p) && !pblk_ppa_empty(ppa_l2p))
2047                 pblk_map_invalidate(pblk, ppa_l2p);
2048
2049         pblk_trans_map_set(pblk, lba, ppa);
2050         spin_unlock(&pblk->trans_lock);
2051 }
2052
2053 void pblk_update_map_cache(struct pblk *pblk, sector_t lba, struct ppa_addr ppa)
2054 {
2055
2056 #ifdef CONFIG_NVM_PBLK_DEBUG
2057         /* Callers must ensure that the ppa points to a cache address */
2058         BUG_ON(!pblk_addr_in_cache(ppa));
2059         BUG_ON(pblk_rb_pos_oob(&pblk->rwb, pblk_addr_to_cacheline(ppa)));
2060 #endif
2061
2062         pblk_update_map(pblk, lba, ppa);
2063 }
2064
2065 int pblk_update_map_gc(struct pblk *pblk, sector_t lba, struct ppa_addr ppa_new,
2066                        struct pblk_line *gc_line, u64 paddr_gc)
2067 {
2068         struct ppa_addr ppa_l2p, ppa_gc;
2069         int ret = 1;
2070
2071 #ifdef CONFIG_NVM_PBLK_DEBUG
2072         /* Callers must ensure that the ppa points to a cache address */
2073         BUG_ON(!pblk_addr_in_cache(ppa_new));
2074         BUG_ON(pblk_rb_pos_oob(&pblk->rwb, pblk_addr_to_cacheline(ppa_new)));
2075 #endif
2076
2077         /* logic error: lba out-of-bounds. Ignore update */
2078         if (!(lba < pblk->capacity)) {
2079                 WARN(1, "pblk: corrupted L2P map request\n");
2080                 return 0;
2081         }
2082
2083         spin_lock(&pblk->trans_lock);
2084         ppa_l2p = pblk_trans_map_get(pblk, lba);
2085         ppa_gc = addr_to_gen_ppa(pblk, paddr_gc, gc_line->id);
2086
2087         if (!pblk_ppa_comp(ppa_l2p, ppa_gc)) {
2088                 spin_lock(&gc_line->lock);
2089                 WARN(!test_bit(paddr_gc, gc_line->invalid_bitmap),
2090                                                 "pblk: corrupted GC update");
2091                 spin_unlock(&gc_line->lock);
2092
2093                 ret = 0;
2094                 goto out;
2095         }
2096
2097         pblk_trans_map_set(pblk, lba, ppa_new);
2098 out:
2099         spin_unlock(&pblk->trans_lock);
2100         return ret;
2101 }
2102
2103 void pblk_update_map_dev(struct pblk *pblk, sector_t lba,
2104                          struct ppa_addr ppa_mapped, struct ppa_addr ppa_cache)
2105 {
2106         struct ppa_addr ppa_l2p;
2107
2108 #ifdef CONFIG_NVM_PBLK_DEBUG
2109         /* Callers must ensure that the ppa points to a device address */
2110         BUG_ON(pblk_addr_in_cache(ppa_mapped));
2111 #endif
2112         /* Invalidate and discard padded entries */
2113         if (lba == ADDR_EMPTY) {
2114                 atomic64_inc(&pblk->pad_wa);
2115 #ifdef CONFIG_NVM_PBLK_DEBUG
2116                 atomic_long_inc(&pblk->padded_wb);
2117 #endif
2118                 if (!pblk_ppa_empty(ppa_mapped))
2119                         pblk_map_invalidate(pblk, ppa_mapped);
2120                 return;
2121         }
2122
2123         /* logic error: lba out-of-bounds. Ignore update */
2124         if (!(lba < pblk->capacity)) {
2125                 WARN(1, "pblk: corrupted L2P map request\n");
2126                 return;
2127         }
2128
2129         spin_lock(&pblk->trans_lock);
2130         ppa_l2p = pblk_trans_map_get(pblk, lba);
2131
2132         /* Do not update L2P if the cacheline has been updated. In this case,
2133          * the mapped ppa must be invalidated
2134          */
2135         if (!pblk_ppa_comp(ppa_l2p, ppa_cache)) {
2136                 if (!pblk_ppa_empty(ppa_mapped))
2137                         pblk_map_invalidate(pblk, ppa_mapped);
2138                 goto out;
2139         }
2140
2141 #ifdef CONFIG_NVM_PBLK_DEBUG
2142         WARN_ON(!pblk_addr_in_cache(ppa_l2p) && !pblk_ppa_empty(ppa_l2p));
2143 #endif
2144
2145         pblk_trans_map_set(pblk, lba, ppa_mapped);
2146 out:
2147         spin_unlock(&pblk->trans_lock);
2148 }
2149
2150 void pblk_lookup_l2p_seq(struct pblk *pblk, struct ppa_addr *ppas,
2151                          sector_t blba, int nr_secs)
2152 {
2153         int i;
2154
2155         spin_lock(&pblk->trans_lock);
2156         for (i = 0; i < nr_secs; i++) {
2157                 struct ppa_addr ppa;
2158
2159                 ppa = ppas[i] = pblk_trans_map_get(pblk, blba + i);
2160
2161                 /* If the L2P entry maps to a line, the reference is valid */
2162                 if (!pblk_ppa_empty(ppa) && !pblk_addr_in_cache(ppa)) {
2163                         struct pblk_line *line = pblk_ppa_to_line(pblk, ppa);
2164
2165                         kref_get(&line->ref);
2166                 }
2167         }
2168         spin_unlock(&pblk->trans_lock);
2169 }
2170
2171 void pblk_lookup_l2p_rand(struct pblk *pblk, struct ppa_addr *ppas,
2172                           u64 *lba_list, int nr_secs)
2173 {
2174         u64 lba;
2175         int i;
2176
2177         spin_lock(&pblk->trans_lock);
2178         for (i = 0; i < nr_secs; i++) {
2179                 lba = lba_list[i];
2180                 if (lba != ADDR_EMPTY) {
2181                         /* logic error: lba out-of-bounds. Ignore update */
2182                         if (!(lba < pblk->capacity)) {
2183                                 WARN(1, "pblk: corrupted L2P map request\n");
2184                                 continue;
2185                         }
2186                         ppas[i] = pblk_trans_map_get(pblk, lba);
2187                 }
2188         }
2189         spin_unlock(&pblk->trans_lock);
2190 }
2191
2192 void *pblk_get_meta_for_writes(struct pblk *pblk, struct nvm_rq *rqd)
2193 {
2194         void *buffer;
2195
2196         if (pblk_is_oob_meta_supported(pblk)) {
2197                 /* Just use OOB metadata buffer as always */
2198                 buffer = rqd->meta_list;
2199         } else {
2200                 /* We need to reuse last page of request (packed metadata)
2201                  * in similar way as traditional oob metadata
2202                  */
2203                 buffer = page_to_virt(
2204                         rqd->bio->bi_io_vec[rqd->bio->bi_vcnt - 1].bv_page);
2205         }
2206
2207         return buffer;
2208 }
2209
2210 void pblk_get_packed_meta(struct pblk *pblk, struct nvm_rq *rqd)
2211 {
2212         void *meta_list = rqd->meta_list;
2213         void *page;
2214         int i = 0;
2215
2216         if (pblk_is_oob_meta_supported(pblk))
2217                 return;
2218
2219         page = page_to_virt(rqd->bio->bi_io_vec[rqd->bio->bi_vcnt - 1].bv_page);
2220         /* We need to fill oob meta buffer with data from packed metadata */
2221         for (; i < rqd->nr_ppas; i++)
2222                 memcpy(pblk_get_meta(pblk, meta_list, i),
2223                         page + (i * sizeof(struct pblk_sec_meta)),
2224                         sizeof(struct pblk_sec_meta));
2225 }