drivers/lightnvm/pblk-core.c

   1 /*
   2  * Copyright (C) 2016 CNEX Labs
   3  * Initial release: Javier Gonzalez <javier@cnexlabs.com>
   4  *                  Matias Bjorling <matias@cnexlabs.com>
   5  *
   6  * This program is free software; you can redistribute it and/or
   7  * modify it under the terms of the GNU General Public License version
   8  * 2 as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License for more details.
  14  *
  15  * pblk-core.c - pblk's core functionality
  16  *
  17  */
  18
  19 #include "pblk.h"
  20 #include <linux/time.h>
  21
  22 static void pblk_mark_bb(struct pblk *pblk, struct pblk_line *line,
  23                          struct ppa_addr *ppa)
  24 {
  25         struct nvm_tgt_dev *dev = pblk->dev;
  26         struct nvm_geo *geo = &dev->geo;
  27         int pos = pblk_dev_ppa_to_pos(geo, *ppa);
  28
  29         pr_debug("pblk: erase failed: line:%d, pos:%d\n", line->id, pos);
  30         atomic_long_inc(&pblk->erase_failed);
  31
  32         atomic_dec(&line->blk_in_line);
  33         if (test_and_set_bit(pos, line->blk_bitmap))
  34                 pr_err("pblk: attempted to erase bb: line:%d, pos:%d\n",
  35                                                         line->id, pos);
  36
  37         pblk_line_run_ws(pblk, NULL, ppa, pblk_line_mark_bb);
  38 }
  39
  40 static void __pblk_end_io_erase(struct pblk *pblk, struct nvm_rq *rqd)
  41 {
  42         struct pblk_line *line;
  43
  44         line = &pblk->lines[pblk_dev_ppa_to_line(rqd->ppa_addr)];
  45         atomic_dec(&line->left_seblks);
  46
  47         if (rqd->error) {
  48                 struct ppa_addr *ppa;
  49
  50                 ppa = kmalloc(sizeof(struct ppa_addr), GFP_ATOMIC);
  51                 if (!ppa)
  52                         return;
  53
  54                 *ppa = rqd->ppa_addr;
  55                 pblk_mark_bb(pblk, line, ppa);
  56         }
  57 }
  58
  59 /* Erase completion assumes that only one block is erased at the time */
  60 static void pblk_end_io_erase(struct nvm_rq *rqd)
  61 {
  62         struct pblk *pblk = rqd->private;
  63
  64         up(&pblk->erase_sem);
  65         __pblk_end_io_erase(pblk, rqd);
  66         mempool_free(rqd, pblk->r_rq_pool);
  67 }
  68
  69 static void __pblk_map_invalidate(struct pblk *pblk, struct pblk_line *line,
  70                                   u64 paddr)
  71 {
  72         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
  73         struct list_head *move_list = NULL;
  74
  75         /* Lines being reclaimed (GC'ed) cannot be invalidated. Before the L2P
  76          * table is modified with reclaimed sectors, a check is done to endure
  77          * that newer updates are not overwritten.
  78          */
  79         spin_lock(&line->lock);
  80         if (line->state == PBLK_LINESTATE_GC ||
  81                                         line->state == PBLK_LINESTATE_FREE) {
  82                 spin_unlock(&line->lock);
  83                 return;
  84         }
  85
  86         if (test_and_set_bit(paddr, line->invalid_bitmap)) {
  87                 WARN_ONCE(1, "pblk: double invalidate\n");
  88                 spin_unlock(&line->lock);
  89                 return;
  90         }
  91         line->vsc--;
  92
  93         if (line->state == PBLK_LINESTATE_CLOSED)
  94                 move_list = pblk_line_gc_list(pblk, line);
  95         spin_unlock(&line->lock);
  96
  97         if (move_list) {
  98                 spin_lock(&l_mg->gc_lock);
  99                 spin_lock(&line->lock);
 100                 /* Prevent moving a line that has just been chosen for GC */
 101                 if (line->state == PBLK_LINESTATE_GC ||
 102                                         line->state == PBLK_LINESTATE_FREE) {
 103                         spin_unlock(&line->lock);
 104                         spin_unlock(&l_mg->gc_lock);
 105                         return;
 106                 }
 107                 spin_unlock(&line->lock);
 108
 109                 list_move_tail(&line->list, move_list);
 110                 spin_unlock(&l_mg->gc_lock);
 111         }
 112 }
 113
 114 void pblk_map_invalidate(struct pblk *pblk, struct ppa_addr ppa)
 115 {
 116         struct pblk_line *line;
 117         u64 paddr;
 118         int line_id;
 119
 120 #ifdef CONFIG_NVM_DEBUG
 121         /* Callers must ensure that the ppa points to a device address */
 122         BUG_ON(pblk_addr_in_cache(ppa));
 123         BUG_ON(pblk_ppa_empty(ppa));
 124 #endif
 125
 126         line_id = pblk_tgt_ppa_to_line(ppa);
 127         line = &pblk->lines[line_id];
 128         paddr = pblk_dev_ppa_to_line_addr(pblk, ppa);
 129
 130         __pblk_map_invalidate(pblk, line, paddr);
 131 }
 132
 133 void pblk_map_pad_invalidate(struct pblk *pblk, struct pblk_line *line,
 134                              u64 paddr)
 135 {
 136         __pblk_map_invalidate(pblk, line, paddr);
 137
 138         pblk_rb_sync_init(&pblk->rwb, NULL);
 139         line->left_ssecs--;
 140         if (!line->left_ssecs)
 141                 pblk_line_run_ws(pblk, line, NULL, pblk_line_close_ws);
 142         pblk_rb_sync_end(&pblk->rwb, NULL);
 143 }
 144
 145 static void pblk_invalidate_range(struct pblk *pblk, sector_t slba,
 146                                   unsigned int nr_secs)
 147 {
 148         sector_t lba;
 149
 150         spin_lock(&pblk->trans_lock);
 151         for (lba = slba; lba < slba + nr_secs; lba++) {
 152                 struct ppa_addr ppa;
 153
 154                 ppa = pblk_trans_map_get(pblk, lba);
 155
 156                 if (!pblk_addr_in_cache(ppa) && !pblk_ppa_empty(ppa))
 157                         pblk_map_invalidate(pblk, ppa);
 158
 159                 pblk_ppa_set_empty(&ppa);
 160                 pblk_trans_map_set(pblk, lba, ppa);
 161         }
 162         spin_unlock(&pblk->trans_lock);
 163 }
 164
 165 struct nvm_rq *pblk_alloc_rqd(struct pblk *pblk, int rw)
 166 {
 167         mempool_t *pool;
 168         struct nvm_rq *rqd;
 169         int rq_size;
 170
 171         if (rw == WRITE) {
 172                 pool = pblk->w_rq_pool;
 173                 rq_size = pblk_w_rq_size;
 174         } else {
 175                 pool = pblk->r_rq_pool;
 176                 rq_size = pblk_r_rq_size;
 177         }
 178
 179         rqd = mempool_alloc(pool, GFP_KERNEL);
 180         memset(rqd, 0, rq_size);
 181
 182         return rqd;
 183 }
 184
 185 void pblk_free_rqd(struct pblk *pblk, struct nvm_rq *rqd, int rw)
 186 {
 187         mempool_t *pool;
 188
 189         if (rw == WRITE)
 190                 pool = pblk->w_rq_pool;
 191         else
 192                 pool = pblk->r_rq_pool;
 193
 194         mempool_free(rqd, pool);
 195 }
 196
 197 void pblk_bio_free_pages(struct pblk *pblk, struct bio *bio, int off,
 198                          int nr_pages)
 199 {
 200         struct bio_vec bv;
 201         int i;
 202
 203         WARN_ON(off + nr_pages != bio->bi_vcnt);
 204
 205         bio_advance(bio, off * PBLK_EXPOSED_PAGE_SIZE);
 206         for (i = off; i < nr_pages + off; i++) {
 207                 bv = bio->bi_io_vec[i];
 208                 mempool_free(bv.bv_page, pblk->page_pool);
 209         }
 210 }
 211
 212 int pblk_bio_add_pages(struct pblk *pblk, struct bio *bio, gfp_t flags,
 213                        int nr_pages)
 214 {
 215         struct request_queue *q = pblk->dev->q;
 216         struct page *page;
 217         int i, ret;
 218
 219         for (i = 0; i < nr_pages; i++) {
 220                 page = mempool_alloc(pblk->page_pool, flags);
 221                 if (!page)
 222                         goto err;
 223
 224                 ret = bio_add_pc_page(q, bio, page, PBLK_EXPOSED_PAGE_SIZE, 0);
 225                 if (ret != PBLK_EXPOSED_PAGE_SIZE) {
 226                         pr_err("pblk: could not add page to bio\n");
 227                         mempool_free(page, pblk->page_pool);
 228                         goto err;
 229                 }
 230         }
 231
 232         return 0;
 233 err:
 234         pblk_bio_free_pages(pblk, bio, 0, i - 1);
 235         return -1;
 236 }
 237
 238 static void pblk_write_kick(struct pblk *pblk)
 239 {
 240         wake_up_process(pblk->writer_ts);
 241         mod_timer(&pblk->wtimer, jiffies + msecs_to_jiffies(1000));
 242 }
 243
 244 void pblk_write_timer_fn(unsigned long data)
 245 {
 246         struct pblk *pblk = (struct pblk *)data;
 247
 248         /* kick the write thread every tick to flush outstanding data */
 249         pblk_write_kick(pblk);
 250 }
 251
 252 void pblk_write_should_kick(struct pblk *pblk)
 253 {
 254         unsigned int secs_avail = pblk_rb_read_count(&pblk->rwb);
 255
 256         if (secs_avail >= pblk->min_write_pgs)
 257                 pblk_write_kick(pblk);
 258 }
 259
 260 void pblk_end_bio_sync(struct bio *bio)
 261 {
 262         struct completion *waiting = bio->bi_private;
 263
 264         complete(waiting);
 265 }
 266
 267 void pblk_end_io_sync(struct nvm_rq *rqd)
 268 {
 269         struct completion *waiting = rqd->private;
 270
 271         complete(waiting);
 272 }
 273
 274 void pblk_flush_writer(struct pblk *pblk)
 275 {
 276         struct bio *bio;
 277         int ret;
 278         DECLARE_COMPLETION_ONSTACK(wait);
 279
 280         bio = bio_alloc(GFP_KERNEL, 1);
 281         if (!bio)
 282                 return;
 283
 284         bio->bi_iter.bi_sector = 0; /* internal bio */
 285         bio_set_op_attrs(bio, REQ_OP_WRITE, REQ_OP_FLUSH);
 286         bio->bi_private = &wait;
 287         bio->bi_end_io = pblk_end_bio_sync;
 288
 289         ret = pblk_write_to_cache(pblk, bio, 0);
 290         if (ret == NVM_IO_OK) {
 291                 if (!wait_for_completion_io_timeout(&wait,
 292                                 msecs_to_jiffies(PBLK_COMMAND_TIMEOUT_MS))) {
 293                         pr_err("pblk: flush cache timed out\n");
 294                 }
 295         } else if (ret != NVM_IO_DONE) {
 296                 pr_err("pblk: tear down bio failed\n");
 297         }
 298
 299         if (bio->bi_status)
 300                 pr_err("pblk: flush sync write failed (%u)\n", bio->bi_status);
 301
 302         bio_put(bio);
 303 }
 304
 305 struct list_head *pblk_line_gc_list(struct pblk *pblk, struct pblk_line *line)
 306 {
 307         struct pblk_line_meta *lm = &pblk->lm;
 308         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
 309         struct list_head *move_list = NULL;
 310
 311         if (!line->vsc) {
 312                 if (line->gc_group != PBLK_LINEGC_FULL) {
 313                         line->gc_group = PBLK_LINEGC_FULL;
 314                         move_list = &l_mg->gc_full_list;
 315                 }
 316         } else if (line->vsc < lm->mid_thrs) {
 317                 if (line->gc_group != PBLK_LINEGC_HIGH) {
 318                         line->gc_group = PBLK_LINEGC_HIGH;
 319                         move_list = &l_mg->gc_high_list;
 320                 }
 321         } else if (line->vsc < lm->high_thrs) {
 322                 if (line->gc_group != PBLK_LINEGC_MID) {
 323                         line->gc_group = PBLK_LINEGC_MID;
 324                         move_list = &l_mg->gc_mid_list;
 325                 }
 326         } else if (line->vsc < line->sec_in_line) {
 327                 if (line->gc_group != PBLK_LINEGC_LOW) {
 328                         line->gc_group = PBLK_LINEGC_LOW;
 329                         move_list = &l_mg->gc_low_list;
 330                 }
 331         } else if (line->vsc == line->sec_in_line) {
 332                 if (line->gc_group != PBLK_LINEGC_EMPTY) {
 333                         line->gc_group = PBLK_LINEGC_EMPTY;
 334                         move_list = &l_mg->gc_empty_list;
 335                 }
 336         } else {
 337                 line->state = PBLK_LINESTATE_CORRUPT;
 338                 line->gc_group = PBLK_LINEGC_NONE;
 339                 move_list =  &l_mg->corrupt_list;
 340                 pr_err("pblk: corrupted vsc for line %d, vsc:%d (%d/%d/%d)\n",
 341                                                 line->id, line->vsc,
 342                                                 line->sec_in_line,
 343                                                 lm->high_thrs, lm->mid_thrs);
 344         }
 345
 346         return move_list;
 347 }
 348
 349 void pblk_discard(struct pblk *pblk, struct bio *bio)
 350 {
 351         sector_t slba = pblk_get_lba(bio);
 352         sector_t nr_secs = pblk_get_secs(bio);
 353
 354         pblk_invalidate_range(pblk, slba, nr_secs);
 355 }
 356
 357 struct ppa_addr pblk_get_lba_map(struct pblk *pblk, sector_t lba)
 358 {
 359         struct ppa_addr ppa;
 360
 361         spin_lock(&pblk->trans_lock);
 362         ppa = pblk_trans_map_get(pblk, lba);
 363         spin_unlock(&pblk->trans_lock);
 364
 365         return ppa;
 366 }
 367
 368 void pblk_log_write_err(struct pblk *pblk, struct nvm_rq *rqd)
 369 {
 370         atomic_long_inc(&pblk->write_failed);
 371 #ifdef CONFIG_NVM_DEBUG
 372         pblk_print_failed_rqd(pblk, rqd, rqd->error);
 373 #endif
 374 }
 375
 376 void pblk_log_read_err(struct pblk *pblk, struct nvm_rq *rqd)
 377 {
 378         /* Empty page read is not necessarily an error (e.g., L2P recovery) */
 379         if (rqd->error == NVM_RSP_ERR_EMPTYPAGE) {
 380                 atomic_long_inc(&pblk->read_empty);
 381                 return;
 382         }
 383
 384         switch (rqd->error) {
 385         case NVM_RSP_WARN_HIGHECC:
 386                 atomic_long_inc(&pblk->read_high_ecc);
 387                 break;
 388         case NVM_RSP_ERR_FAILECC:
 389         case NVM_RSP_ERR_FAILCRC:
 390                 atomic_long_inc(&pblk->read_failed);
 391                 break;
 392         default:
 393                 pr_err("pblk: unknown read error:%d\n", rqd->error);
 394         }
 395 #ifdef CONFIG_NVM_DEBUG
 396         pblk_print_failed_rqd(pblk, rqd, rqd->error);
 397 #endif
 398 }
 399
 400 void pblk_set_sec_per_write(struct pblk *pblk, int sec_per_write)
 401 {
 402         pblk->sec_per_write = sec_per_write;
 403 }
 404
 405 int pblk_submit_io(struct pblk *pblk, struct nvm_rq *rqd)
 406 {
 407         struct nvm_tgt_dev *dev = pblk->dev;
 408
 409 #ifdef CONFIG_NVM_DEBUG
 410         struct ppa_addr *ppa_list;
 411
 412         ppa_list = (rqd->nr_ppas > 1) ? rqd->ppa_list : &rqd->ppa_addr;
 413         if (pblk_boundary_ppa_checks(dev, ppa_list, rqd->nr_ppas)) {
 414                 WARN_ON(1);
 415                 return -EINVAL;
 416         }
 417
 418         if (rqd->opcode == NVM_OP_PWRITE) {
 419                 struct pblk_line *line;
 420                 struct ppa_addr ppa;
 421                 int i;
 422
 423                 for (i = 0; i < rqd->nr_ppas; i++) {
 424                         ppa = ppa_list[i];
 425                         line = &pblk->lines[pblk_dev_ppa_to_line(ppa)];
 426
 427                         spin_lock(&line->lock);
 428                         if (line->state != PBLK_LINESTATE_OPEN) {
 429                                 pr_err("pblk: bad ppa: line:%d,state:%d\n",
 430                                                         line->id, line->state);
 431                                 WARN_ON(1);
 432                                 spin_unlock(&line->lock);
 433                                 return -EINVAL;
 434                         }
 435                         spin_unlock(&line->lock);
 436                 }
 437         }
 438 #endif
 439         return nvm_submit_io(dev, rqd);
 440 }
 441
 442 struct bio *pblk_bio_map_addr(struct pblk *pblk, void *data,
 443                               unsigned int nr_secs, unsigned int len,
 444                               gfp_t gfp_mask)
 445 {
 446         struct nvm_tgt_dev *dev = pblk->dev;
 447         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
 448         void *kaddr = data;
 449         struct page *page;
 450         struct bio *bio;
 451         int i, ret;
 452
 453         if (l_mg->emeta_alloc_type == PBLK_KMALLOC_META)
 454                 return bio_map_kern(dev->q, kaddr, len, gfp_mask);
 455
 456         bio = bio_kmalloc(gfp_mask, nr_secs);
 457         if (!bio)
 458                 return ERR_PTR(-ENOMEM);
 459
 460         for (i = 0; i < nr_secs; i++) {
 461                 page = vmalloc_to_page(kaddr);
 462                 if (!page) {
 463                         pr_err("pblk: could not map vmalloc bio\n");
 464                         bio_put(bio);
 465                         bio = ERR_PTR(-ENOMEM);
 466                         goto out;
 467                 }
 468
 469                 ret = bio_add_pc_page(dev->q, bio, page, PAGE_SIZE, 0);
 470                 if (ret != PAGE_SIZE) {
 471                         pr_err("pblk: could not add page to bio\n");
 472                         bio_put(bio);
 473                         bio = ERR_PTR(-ENOMEM);
 474                         goto out;
 475                 }
 476
 477                 kaddr += PAGE_SIZE;
 478         }
 479 out:
 480         return bio;
 481 }
 482
 483 int pblk_calc_secs(struct pblk *pblk, unsigned long secs_avail,
 484                    unsigned long secs_to_flush)
 485 {
 486         int max = pblk->sec_per_write;
 487         int min = pblk->min_write_pgs;
 488         int secs_to_sync = 0;
 489
 490         if (secs_avail >= max)
 491                 secs_to_sync = max;
 492         else if (secs_avail >= min)
 493                 secs_to_sync = min * (secs_avail / min);
 494         else if (secs_to_flush)
 495                 secs_to_sync = min;
 496
 497         return secs_to_sync;
 498 }
 499
 500 static u64 __pblk_alloc_page(struct pblk *pblk, struct pblk_line *line,
 501                              int nr_secs)
 502 {
 503         u64 addr;
 504         int i;
 505
 506         /* logic error: ppa out-of-bounds. Prevent generating bad address */
 507         if (line->cur_sec + nr_secs > pblk->lm.sec_per_line) {
 508                 WARN(1, "pblk: page allocation out of bounds\n");
 509                 nr_secs = pblk->lm.sec_per_line - line->cur_sec;
 510         }
 511
 512         line->cur_sec = addr = find_next_zero_bit(line->map_bitmap,
 513                                         pblk->lm.sec_per_line, line->cur_sec);
 514         for (i = 0; i < nr_secs; i++, line->cur_sec++)
 515                 WARN_ON(test_and_set_bit(line->cur_sec, line->map_bitmap));
 516
 517         return addr;
 518 }
 519
 520 u64 pblk_alloc_page(struct pblk *pblk, struct pblk_line *line, int nr_secs)
 521 {
 522         u64 addr;
 523
 524         /* Lock needed in case a write fails and a recovery needs to remap
 525          * failed write buffer entries
 526          */
 527         spin_lock(&line->lock);
 528         addr = __pblk_alloc_page(pblk, line, nr_secs);
 529         line->left_msecs -= nr_secs;
 530         WARN(line->left_msecs < 0, "pblk: page allocation out of bounds\n");
 531         spin_unlock(&line->lock);
 532
 533         return addr;
 534 }
 535
 536 /*
 537  * Submit emeta to one LUN in the raid line at the time to avoid a deadlock when
 538  * taking the per LUN semaphore.
 539  */
 540 static int pblk_line_submit_emeta_io(struct pblk *pblk, struct pblk_line *line,
 541                                      u64 paddr, int dir)
 542 {
 543         struct nvm_tgt_dev *dev = pblk->dev;
 544         struct nvm_geo *geo = &dev->geo;
 545         struct pblk_line_meta *lm = &pblk->lm;
 546         struct bio *bio;
 547         struct nvm_rq rqd;
 548         struct ppa_addr *ppa_list;
 549         dma_addr_t dma_ppa_list;
 550         void *emeta = line->emeta;
 551         int min = pblk->min_write_pgs;
 552         int left_ppas = lm->emeta_sec;
 553         int id = line->id;
 554         int rq_ppas, rq_len;
 555         int cmd_op, bio_op;
 556         int flags;
 557         int i, j;
 558         int ret;
 559         DECLARE_COMPLETION_ONSTACK(wait);
 560
 561         if (dir == WRITE) {
 562                 bio_op = REQ_OP_WRITE;
 563                 cmd_op = NVM_OP_PWRITE;
 564                 flags = pblk_set_progr_mode(pblk, WRITE);
 565         } else if (dir == READ) {
 566                 bio_op = REQ_OP_READ;
 567                 cmd_op = NVM_OP_PREAD;
 568                 flags = pblk_set_read_mode(pblk);
 569         } else
 570                 return -EINVAL;
 571
 572         ppa_list = nvm_dev_dma_alloc(dev->parent, GFP_KERNEL, &dma_ppa_list);
 573         if (!ppa_list)
 574                 return -ENOMEM;
 575
 576 next_rq:
 577         memset(&rqd, 0, sizeof(struct nvm_rq));
 578
 579         rq_ppas = pblk_calc_secs(pblk, left_ppas, 0);
 580         rq_len = rq_ppas * geo->sec_size;
 581
 582         bio = pblk_bio_map_addr(pblk, emeta, rq_ppas, rq_len, GFP_KERNEL);
 583         if (IS_ERR(bio)) {
 584                 ret = PTR_ERR(bio);
 585                 goto free_rqd_dma;
 586         }
 587
 588         bio->bi_iter.bi_sector = 0; /* internal bio */
 589         bio_set_op_attrs(bio, bio_op, 0);
 590
 591         rqd.bio = bio;
 592         rqd.opcode = cmd_op;
 593         rqd.flags = flags;
 594         rqd.nr_ppas = rq_ppas;
 595         rqd.ppa_list = ppa_list;
 596         rqd.dma_ppa_list = dma_ppa_list;
 597         rqd.end_io = pblk_end_io_sync;
 598         rqd.private = &wait;
 599
 600         if (dir == WRITE) {
 601                 for (i = 0; i < rqd.nr_ppas; ) {
 602                         spin_lock(&line->lock);
 603                         paddr = __pblk_alloc_page(pblk, line, min);
 604                         spin_unlock(&line->lock);
 605                         for (j = 0; j < min; j++, i++, paddr++)
 606                                 rqd.ppa_list[i] =
 607                                         addr_to_gen_ppa(pblk, paddr, id);
 608                 }
 609         } else {
 610                 for (i = 0; i < rqd.nr_ppas; ) {
 611                         struct ppa_addr ppa = addr_to_gen_ppa(pblk, paddr, id);
 612                         int pos = pblk_dev_ppa_to_pos(geo, ppa);
 613
 614                         while (test_bit(pos, line->blk_bitmap)) {
 615                                 paddr += min;
 616                                 if (pblk_boundary_paddr_checks(pblk, paddr)) {
 617                                         pr_err("pblk: corrupt emeta line:%d\n",
 618                                                                 line->id);
 619                                         bio_put(bio);
 620                                         ret = -EINTR;
 621                                         goto free_rqd_dma;
 622                                 }
 623
 624                                 ppa = addr_to_gen_ppa(pblk, paddr, id);
 625                                 pos = pblk_dev_ppa_to_pos(geo, ppa);
 626                         }
 627
 628                         if (pblk_boundary_paddr_checks(pblk, paddr + min)) {
 629                                 pr_err("pblk: corrupt emeta line:%d\n",
 630                                                                 line->id);
 631                                 bio_put(bio);
 632                                 ret = -EINTR;
 633                                 goto free_rqd_dma;
 634                         }
 635
 636                         for (j = 0; j < min; j++, i++, paddr++)
 637                                 rqd.ppa_list[i] =
 638                                         addr_to_gen_ppa(pblk, paddr, line->id);
 639                 }
 640         }
 641
 642         ret = pblk_submit_io(pblk, &rqd);
 643         if (ret) {
 644                 pr_err("pblk: emeta I/O submission failed: %d\n", ret);
 645                 bio_put(bio);
 646                 goto free_rqd_dma;
 647         }
 648
 649         if (!wait_for_completion_io_timeout(&wait,
 650                                 msecs_to_jiffies(PBLK_COMMAND_TIMEOUT_MS))) {
 651                 pr_err("pblk: emeta I/O timed out\n");
 652         }
 653         reinit_completion(&wait);
 654
 655         bio_put(bio);
 656
 657         if (rqd.error) {
 658                 if (dir == WRITE)
 659                         pblk_log_write_err(pblk, &rqd);
 660                 else
 661                         pblk_log_read_err(pblk, &rqd);
 662         }
 663
 664         emeta += rq_len;
 665         left_ppas -= rq_ppas;
 666         if (left_ppas)
 667                 goto next_rq;
 668 free_rqd_dma:
 669         nvm_dev_dma_free(dev->parent, ppa_list, dma_ppa_list);
 670         return ret;
 671 }
 672
 673 u64 pblk_line_smeta_start(struct pblk *pblk, struct pblk_line *line)
 674 {
 675         struct nvm_tgt_dev *dev = pblk->dev;
 676         struct nvm_geo *geo = &dev->geo;
 677         struct pblk_line_meta *lm = &pblk->lm;
 678         int bit;
 679
 680         /* This usually only happens on bad lines */
 681         bit = find_first_zero_bit(line->blk_bitmap, lm->blk_per_line);
 682         if (bit >= lm->blk_per_line)
 683                 return -1;
 684
 685         return bit * geo->sec_per_pl;
 686 }
 687
 688 static int pblk_line_submit_smeta_io(struct pblk *pblk, struct pblk_line *line,
 689                                      u64 paddr, int dir)
 690 {
 691         struct nvm_tgt_dev *dev = pblk->dev;
 692         struct pblk_line_meta *lm = &pblk->lm;
 693         struct bio *bio;
 694         struct nvm_rq rqd;
 695         __le64 *lba_list = NULL;
 696         int i, ret;
 697         int cmd_op, bio_op;
 698         int flags;
 699         DECLARE_COMPLETION_ONSTACK(wait);
 700
 701         if (dir == WRITE) {
 702                 bio_op = REQ_OP_WRITE;
 703                 cmd_op = NVM_OP_PWRITE;
 704                 flags = pblk_set_progr_mode(pblk, WRITE);
 705                 lba_list = pblk_line_emeta_to_lbas(line->emeta);
 706         } else if (dir == READ) {
 707                 bio_op = REQ_OP_READ;
 708                 cmd_op = NVM_OP_PREAD;
 709                 flags = pblk_set_read_mode(pblk);
 710         } else
 711                 return -EINVAL;
 712
 713         memset(&rqd, 0, sizeof(struct nvm_rq));
 714
 715         rqd.ppa_list = nvm_dev_dma_alloc(dev->parent, GFP_KERNEL,
 716                                                         &rqd.dma_ppa_list);
 717         if (!rqd.ppa_list)
 718                 return -ENOMEM;
 719
 720         bio = bio_map_kern(dev->q, line->smeta, lm->smeta_len, GFP_KERNEL);
 721         if (IS_ERR(bio)) {
 722                 ret = PTR_ERR(bio);
 723                 goto free_ppa_list;
 724         }
 725
 726         bio->bi_iter.bi_sector = 0; /* internal bio */
 727         bio_set_op_attrs(bio, bio_op, 0);
 728
 729         rqd.bio = bio;
 730         rqd.opcode = cmd_op;
 731         rqd.flags = flags;
 732         rqd.nr_ppas = lm->smeta_sec;
 733         rqd.end_io = pblk_end_io_sync;
 734         rqd.private = &wait;
 735
 736         for (i = 0; i < lm->smeta_sec; i++, paddr++) {
 737                 rqd.ppa_list[i] = addr_to_gen_ppa(pblk, paddr, line->id);
 738                 if (dir == WRITE)
 739                         lba_list[paddr] = cpu_to_le64(ADDR_EMPTY);
 740         }
 741
 742         /*
 743          * This I/O is sent by the write thread when a line is replace. Since
 744          * the write thread is the only one sending write and erase commands,
 745          * there is no need to take the LUN semaphore.
 746          */
 747         ret = pblk_submit_io(pblk, &rqd);
 748         if (ret) {
 749                 pr_err("pblk: smeta I/O submission failed: %d\n", ret);
 750                 bio_put(bio);
 751                 goto free_ppa_list;
 752         }
 753
 754         if (!wait_for_completion_io_timeout(&wait,
 755                                 msecs_to_jiffies(PBLK_COMMAND_TIMEOUT_MS))) {
 756                 pr_err("pblk: smeta I/O timed out\n");
 757         }
 758
 759         if (rqd.error) {
 760                 if (dir == WRITE)
 761                         pblk_log_write_err(pblk, &rqd);
 762                 else
 763                         pblk_log_read_err(pblk, &rqd);
 764         }
 765
 766 free_ppa_list:
 767         nvm_dev_dma_free(dev->parent, rqd.ppa_list, rqd.dma_ppa_list);
 768
 769         return ret;
 770 }
 771
 772 int pblk_line_read_smeta(struct pblk *pblk, struct pblk_line *line)
 773 {
 774         u64 bpaddr = pblk_line_smeta_start(pblk, line);
 775
 776         return pblk_line_submit_smeta_io(pblk, line, bpaddr, READ);
 777 }
 778
 779 int pblk_line_read_emeta(struct pblk *pblk, struct pblk_line *line)
 780 {
 781         return pblk_line_submit_emeta_io(pblk, line, line->emeta_ssec, READ);
 782 }
 783
 784 static void pblk_setup_e_rq(struct pblk *pblk, struct nvm_rq *rqd,
 785                             struct ppa_addr ppa)
 786 {
 787         rqd->opcode = NVM_OP_ERASE;
 788         rqd->ppa_addr = ppa;
 789         rqd->nr_ppas = 1;
 790         rqd->flags = pblk_set_progr_mode(pblk, ERASE);
 791         rqd->bio = NULL;
 792 }
 793
 794 static int pblk_blk_erase_sync(struct pblk *pblk, struct ppa_addr ppa)
 795 {
 796         struct nvm_rq rqd;
 797         int ret;
 798         DECLARE_COMPLETION_ONSTACK(wait);
 799
 800         memset(&rqd, 0, sizeof(struct nvm_rq));
 801
 802         pblk_setup_e_rq(pblk, &rqd, ppa);
 803
 804         rqd.end_io = pblk_end_io_sync;
 805         rqd.private = &wait;
 806
 807         /* The write thread schedules erases so that it minimizes disturbances
 808          * with writes. Thus, there is no need to take the LUN semaphore.
 809          */
 810         ret = pblk_submit_io(pblk, &rqd);
 811         if (ret) {
 812                 struct nvm_tgt_dev *dev = pblk->dev;
 813                 struct nvm_geo *geo = &dev->geo;
 814
 815                 pr_err("pblk: could not sync erase line:%d,blk:%d\n",
 816                                         pblk_dev_ppa_to_line(ppa),
 817                                         pblk_dev_ppa_to_pos(geo, ppa));
 818
 819                 rqd.error = ret;
 820                 goto out;
 821         }
 822
 823         if (!wait_for_completion_io_timeout(&wait,
 824                                 msecs_to_jiffies(PBLK_COMMAND_TIMEOUT_MS))) {
 825                 pr_err("pblk: sync erase timed out\n");
 826         }
 827
 828 out:
 829         rqd.private = pblk;
 830         __pblk_end_io_erase(pblk, &rqd);
 831
 832         return 0;
 833 }
 834
 835 int pblk_line_erase(struct pblk *pblk, struct pblk_line *line)
 836 {
 837         struct pblk_line_meta *lm = &pblk->lm;
 838         struct ppa_addr ppa;
 839         int bit = -1;
 840
 841         /* Erase only good blocks, one at a time */
 842         do {
 843                 spin_lock(&line->lock);
 844                 bit = find_next_zero_bit(line->erase_bitmap, lm->blk_per_line,
 845                                                                 bit + 1);
 846                 if (bit >= lm->blk_per_line) {
 847                         spin_unlock(&line->lock);
 848                         break;
 849                 }
 850
 851                 ppa = pblk->luns[bit].bppa; /* set ch and lun */
 852                 ppa.g.blk = line->id;
 853
 854                 atomic_dec(&line->left_eblks);
 855                 WARN_ON(test_and_set_bit(bit, line->erase_bitmap));
 856                 spin_unlock(&line->lock);
 857
 858                 if (pblk_blk_erase_sync(pblk, ppa)) {
 859                         pr_err("pblk: failed to erase line %d\n", line->id);
 860                         return -ENOMEM;
 861                 }
 862         } while (1);
 863
 864         return 0;
 865 }
 866
 867 /* For now lines are always assumed full lines. Thus, smeta former and current
 868  * lun bitmaps are omitted.
 869  */
 870 static int pblk_line_set_metadata(struct pblk *pblk, struct pblk_line *line,
 871                                   struct pblk_line *cur)
 872 {
 873         struct nvm_tgt_dev *dev = pblk->dev;
 874         struct nvm_geo *geo = &dev->geo;
 875         struct pblk_line_meta *lm = &pblk->lm;
 876         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
 877         struct line_smeta *smeta = line->smeta;
 878         struct line_emeta *emeta = line->emeta;
 879         int nr_blk_line;
 880
 881         /* After erasing the line, new bad blocks might appear and we risk
 882          * having an invalid line
 883          */
 884         nr_blk_line = lm->blk_per_line -
 885                         bitmap_weight(line->blk_bitmap, lm->blk_per_line);
 886         if (nr_blk_line < lm->min_blk_line) {
 887                 spin_lock(&l_mg->free_lock);
 888                 spin_lock(&line->lock);
 889                 line->state = PBLK_LINESTATE_BAD;
 890                 spin_unlock(&line->lock);
 891
 892                 list_add_tail(&line->list, &l_mg->bad_list);
 893                 spin_unlock(&l_mg->free_lock);
 894
 895                 pr_debug("pblk: line %d is bad\n", line->id);
 896
 897                 return 0;
 898         }
 899
 900         /* Run-time metadata */
 901         line->lun_bitmap = ((void *)(smeta)) + sizeof(struct line_smeta);
 902
 903         /* Mark LUNs allocated in this line (all for now) */
 904         bitmap_set(line->lun_bitmap, 0, lm->lun_bitmap_len);
 905
 906         smeta->header.identifier = cpu_to_le32(PBLK_MAGIC);
 907         memcpy(smeta->header.uuid, pblk->instance_uuid, 16);
 908         smeta->header.id = cpu_to_le32(line->id);
 909         smeta->header.type = cpu_to_le16(line->type);
 910         smeta->header.version = cpu_to_le16(1);
 911
 912         /* Start metadata */
 913         smeta->seq_nr = cpu_to_le64(line->seq_nr);
 914         smeta->window_wr_lun = cpu_to_le32(geo->nr_luns);
 915
 916         /* Fill metadata among lines */
 917         if (cur) {
 918                 memcpy(line->lun_bitmap, cur->lun_bitmap, lm->lun_bitmap_len);
 919                 smeta->prev_id = cpu_to_le32(cur->id);
 920                 cur->emeta->next_id = cpu_to_le32(line->id);
 921         } else {
 922                 smeta->prev_id = cpu_to_le32(PBLK_LINE_EMPTY);
 923         }
 924
 925         /* All smeta must be set at this point */
 926         smeta->header.crc = cpu_to_le32(pblk_calc_meta_header_crc(pblk, smeta));
 927         smeta->crc = cpu_to_le32(pblk_calc_smeta_crc(pblk, smeta));
 928
 929         /* End metadata */
 930         memcpy(&emeta->header, &smeta->header, sizeof(struct line_header));
 931         emeta->seq_nr = cpu_to_le64(line->seq_nr);
 932         emeta->nr_lbas = cpu_to_le64(line->sec_in_line);
 933         emeta->nr_valid_lbas = cpu_to_le64(0);
 934         emeta->next_id = cpu_to_le32(PBLK_LINE_EMPTY);
 935         emeta->crc = cpu_to_le32(0);
 936         emeta->prev_id = smeta->prev_id;
 937
 938         return 1;
 939 }
 940
 941 /* For now lines are always assumed full lines. Thus, smeta former and current
 942  * lun bitmaps are omitted.
 943  */
 944 static int pblk_line_init_bb(struct pblk *pblk, struct pblk_line *line,
 945                              int init)
 946 {
 947         struct nvm_tgt_dev *dev = pblk->dev;
 948         struct nvm_geo *geo = &dev->geo;
 949         struct pblk_line_meta *lm = &pblk->lm;
 950         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
 951         int nr_bb = 0;
 952         u64 off;
 953         int bit = -1;
 954
 955         line->sec_in_line = lm->sec_per_line;
 956
 957         /* Capture bad block information on line mapping bitmaps */
 958         while ((bit = find_next_bit(line->blk_bitmap, lm->blk_per_line,
 959                                         bit + 1)) < lm->blk_per_line) {
 960                 off = bit * geo->sec_per_pl;
 961                 bitmap_shift_left(l_mg->bb_aux, l_mg->bb_template, off,
 962                                                         lm->sec_per_line);
 963                 bitmap_or(line->map_bitmap, line->map_bitmap, l_mg->bb_aux,
 964                                                         lm->sec_per_line);
 965                 line->sec_in_line -= geo->sec_per_blk;
 966                 if (bit >= lm->emeta_bb)
 967                         nr_bb++;
 968         }
 969
 970         /* Mark smeta metadata sectors as bad sectors */
 971         bit = find_first_zero_bit(line->blk_bitmap, lm->blk_per_line);
 972         off = bit * geo->sec_per_pl;
 973 retry_smeta:
 974         bitmap_set(line->map_bitmap, off, lm->smeta_sec);
 975         line->sec_in_line -= lm->smeta_sec;
 976         line->smeta_ssec = off;
 977         line->cur_sec = off + lm->smeta_sec;
 978
 979         if (init && pblk_line_submit_smeta_io(pblk, line, off, WRITE)) {
 980                 pr_debug("pblk: line smeta I/O failed. Retry\n");
 981                 off += geo->sec_per_pl;
 982                 goto retry_smeta;
 983         }
 984
 985         bitmap_copy(line->invalid_bitmap, line->map_bitmap, lm->sec_per_line);
 986
 987         /* Mark emeta metadata sectors as bad sectors. We need to consider bad
 988          * blocks to make sure that there are enough sectors to store emeta
 989          */
 990         bit = lm->sec_per_line;
 991         off = lm->sec_per_line - lm->emeta_sec;
 992         bitmap_set(line->invalid_bitmap, off, lm->emeta_sec);
 993         while (nr_bb) {
 994                 off -= geo->sec_per_pl;
 995                 if (!test_bit(off, line->invalid_bitmap)) {
 996                         bitmap_set(line->invalid_bitmap, off, geo->sec_per_pl);
 997                         nr_bb--;
 998                 }
 999         }
1000
1001         line->sec_in_line -= lm->emeta_sec;
1002         line->emeta_ssec = off;
1003         line->vsc = line->left_ssecs = line->left_msecs = line->sec_in_line;
1004
1005         if (lm->sec_per_line - line->sec_in_line !=
1006                 bitmap_weight(line->invalid_bitmap, lm->sec_per_line)) {
1007                 spin_lock(&line->lock);
1008                 line->state = PBLK_LINESTATE_BAD;
1009                 spin_unlock(&line->lock);
1010
1011                 list_add_tail(&line->list, &l_mg->bad_list);
1012                 pr_err("pblk: unexpected line %d is bad\n", line->id);
1013
1014                 return 0;
1015         }
1016
1017         return 1;
1018 }
1019
1020 static int pblk_line_prepare(struct pblk *pblk, struct pblk_line *line)
1021 {
1022         struct pblk_line_meta *lm = &pblk->lm;
1023         int blk_in_line = atomic_read(&line->blk_in_line);
1024
1025         line->map_bitmap = mempool_alloc(pblk->line_meta_pool, GFP_ATOMIC);
1026         if (!line->map_bitmap)
1027                 return -ENOMEM;
1028         memset(line->map_bitmap, 0, lm->sec_bitmap_len);
1029
1030         /* invalid_bitmap is special since it is used when line is closed. No
1031          * need to zeroized; it will be initialized using bb info form
1032          * map_bitmap
1033          */
1034         line->invalid_bitmap = mempool_alloc(pblk->line_meta_pool, GFP_ATOMIC);
1035         if (!line->invalid_bitmap) {
1036                 mempool_free(line->map_bitmap, pblk->line_meta_pool);
1037                 return -ENOMEM;
1038         }
1039
1040         spin_lock(&line->lock);
1041         if (line->state != PBLK_LINESTATE_FREE) {
1042                 spin_unlock(&line->lock);
1043                 WARN(1, "pblk: corrupted line state\n");
1044                 return -EINTR;
1045         }
1046         line->state = PBLK_LINESTATE_OPEN;
1047
1048         atomic_set(&line->left_eblks, blk_in_line);
1049         atomic_set(&line->left_seblks, blk_in_line);
1050         spin_unlock(&line->lock);
1051
1052         /* Bad blocks do not need to be erased */
1053         bitmap_copy(line->erase_bitmap, line->blk_bitmap, lm->blk_per_line);
1054
1055         kref_init(&line->ref);
1056
1057         return 0;
1058 }
1059
1060 int pblk_line_recov_alloc(struct pblk *pblk, struct pblk_line *line)
1061 {
1062         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1063         int ret;
1064
1065         spin_lock(&l_mg->free_lock);
1066         l_mg->data_line = line;
1067         list_del(&line->list);
1068
1069         ret = pblk_line_prepare(pblk, line);
1070         if (ret) {
1071                 list_add(&line->list, &l_mg->free_list);
1072                 spin_unlock(&l_mg->free_lock);
1073                 return ret;
1074         }
1075         spin_unlock(&l_mg->free_lock);
1076
1077         pblk_rl_free_lines_dec(&pblk->rl, line);
1078
1079         if (!pblk_line_init_bb(pblk, line, 0)) {
1080                 list_add(&line->list, &l_mg->free_list);
1081                 return -EINTR;
1082         }
1083
1084         return 0;
1085 }
1086
1087 void pblk_line_recov_close(struct pblk *pblk, struct pblk_line *line)
1088 {
1089         mempool_free(line->map_bitmap, pblk->line_meta_pool);
1090         line->map_bitmap = NULL;
1091         line->smeta = NULL;
1092         line->emeta = NULL;
1093 }
1094
1095 struct pblk_line *pblk_line_get(struct pblk *pblk)
1096 {
1097         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1098         struct pblk_line_meta *lm = &pblk->lm;
1099         struct pblk_line *line = NULL;
1100         int bit;
1101
1102         lockdep_assert_held(&l_mg->free_lock);
1103
1104 retry_get:
1105         if (list_empty(&l_mg->free_list)) {
1106                 pr_err("pblk: no free lines\n");
1107                 goto out;
1108         }
1109
1110         line = list_first_entry(&l_mg->free_list, struct pblk_line, list);
1111         list_del(&line->list);
1112         l_mg->nr_free_lines--;
1113
1114         bit = find_first_zero_bit(line->blk_bitmap, lm->blk_per_line);
1115         if (unlikely(bit >= lm->blk_per_line)) {
1116                 spin_lock(&line->lock);
1117                 line->state = PBLK_LINESTATE_BAD;
1118                 spin_unlock(&line->lock);
1119
1120                 list_add_tail(&line->list, &l_mg->bad_list);
1121
1122                 pr_debug("pblk: line %d is bad\n", line->id);
1123                 goto retry_get;
1124         }
1125
1126         if (pblk_line_prepare(pblk, line)) {
1127                 pr_err("pblk: failed to prepare line %d\n", line->id);
1128                 list_add(&line->list, &l_mg->free_list);
1129                 return NULL;
1130         }
1131
1132 out:
1133         return line;
1134 }
1135
1136 static struct pblk_line *pblk_line_retry(struct pblk *pblk,
1137                                          struct pblk_line *line)
1138 {
1139         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1140         struct pblk_line *retry_line;
1141
1142         spin_lock(&l_mg->free_lock);
1143         retry_line = pblk_line_get(pblk);
1144         if (!retry_line) {
1145                 l_mg->data_line = NULL;
1146                 spin_unlock(&l_mg->free_lock);
1147                 return NULL;
1148         }
1149
1150         retry_line->smeta = line->smeta;
1151         retry_line->emeta = line->emeta;
1152         retry_line->meta_line = line->meta_line;
1153
1154         pblk_line_free(pblk, line);
1155         l_mg->data_line = retry_line;
1156         spin_unlock(&l_mg->free_lock);
1157
1158         if (pblk_line_erase(pblk, retry_line)) {
1159                 spin_lock(&l_mg->free_lock);
1160                 l_mg->data_line = NULL;
1161                 spin_unlock(&l_mg->free_lock);
1162                 return NULL;
1163         }
1164
1165         pblk_rl_free_lines_dec(&pblk->rl, retry_line);
1166
1167         return retry_line;
1168 }
1169
1170 struct pblk_line *pblk_line_get_first_data(struct pblk *pblk)
1171 {
1172         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1173         struct pblk_line *line;
1174         int meta_line;
1175         int is_next = 0;
1176
1177         spin_lock(&l_mg->free_lock);
1178         line = pblk_line_get(pblk);
1179         if (!line) {
1180                 spin_unlock(&l_mg->free_lock);
1181                 return NULL;
1182         }
1183
1184         line->seq_nr = l_mg->d_seq_nr++;
1185         line->type = PBLK_LINETYPE_DATA;
1186         l_mg->data_line = line;
1187
1188         meta_line = find_first_zero_bit(&l_mg->meta_bitmap, PBLK_DATA_LINES);
1189         set_bit(meta_line, &l_mg->meta_bitmap);
1190         line->smeta = l_mg->sline_meta[meta_line].meta;
1191         line->emeta = l_mg->eline_meta[meta_line].meta;
1192         line->meta_line = meta_line;
1193
1194         /* Allocate next line for preparation */
1195         l_mg->data_next = pblk_line_get(pblk);
1196         if (l_mg->data_next) {
1197                 l_mg->data_next->seq_nr = l_mg->d_seq_nr++;
1198                 l_mg->data_next->type = PBLK_LINETYPE_DATA;
1199                 is_next = 1;
1200         }
1201         spin_unlock(&l_mg->free_lock);
1202
1203         pblk_rl_free_lines_dec(&pblk->rl, line);
1204         if (is_next)
1205                 pblk_rl_free_lines_dec(&pblk->rl, l_mg->data_next);
1206
1207         if (pblk_line_erase(pblk, line))
1208                 return NULL;
1209
1210 retry_setup:
1211         if (!pblk_line_set_metadata(pblk, line, NULL)) {
1212                 line = pblk_line_retry(pblk, line);
1213                 if (!line)
1214                         return NULL;
1215
1216                 goto retry_setup;
1217         }
1218
1219         if (!pblk_line_init_bb(pblk, line, 1)) {
1220                 line = pblk_line_retry(pblk, line);
1221                 if (!line)
1222                         return NULL;
1223
1224                 goto retry_setup;
1225         }
1226
1227         return line;
1228 }
1229
1230 struct pblk_line *pblk_line_replace_data(struct pblk *pblk)
1231 {
1232         struct pblk_line_meta *lm = &pblk->lm;
1233         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1234         struct pblk_line *cur, *new;
1235         unsigned int left_seblks;
1236         int meta_line;
1237         int is_next = 0;
1238
1239         cur = l_mg->data_line;
1240         new = l_mg->data_next;
1241         if (!new)
1242                 return NULL;
1243         l_mg->data_line = new;
1244
1245 retry_line:
1246         left_seblks = atomic_read(&new->left_seblks);
1247         if (left_seblks) {
1248                 /* If line is not fully erased, erase it */
1249                 if (atomic_read(&new->left_eblks)) {
1250                         if (pblk_line_erase(pblk, new))
1251                                 return NULL;
1252                 } else {
1253                         io_schedule();
1254                 }
1255                 goto retry_line;
1256         }
1257
1258         spin_lock(&l_mg->free_lock);
1259         /* Allocate next line for preparation */
1260         l_mg->data_next = pblk_line_get(pblk);
1261         if (l_mg->data_next) {
1262                 l_mg->data_next->seq_nr = l_mg->d_seq_nr++;
1263                 l_mg->data_next->type = PBLK_LINETYPE_DATA;
1264                 is_next = 1;
1265         }
1266
1267 retry_meta:
1268         meta_line = find_first_zero_bit(&l_mg->meta_bitmap, PBLK_DATA_LINES);
1269         if (meta_line == PBLK_DATA_LINES) {
1270                 spin_unlock(&l_mg->free_lock);
1271                 io_schedule();
1272                 spin_lock(&l_mg->free_lock);
1273                 goto retry_meta;
1274         }
1275
1276         set_bit(meta_line, &l_mg->meta_bitmap);
1277         new->smeta = l_mg->sline_meta[meta_line].meta;
1278         new->emeta = l_mg->eline_meta[meta_line].meta;
1279         new->meta_line = meta_line;
1280
1281         memset(new->smeta, 0, lm->smeta_len);
1282         memset(new->emeta, 0, lm->emeta_len);
1283         spin_unlock(&l_mg->free_lock);
1284
1285         if (is_next)
1286                 pblk_rl_free_lines_dec(&pblk->rl, l_mg->data_next);
1287
1288 retry_setup:
1289         if (!pblk_line_set_metadata(pblk, new, cur)) {
1290                 new = pblk_line_retry(pblk, new);
1291                 if (!new)
1292                         return NULL;
1293
1294                 goto retry_setup;
1295         }
1296
1297         if (!pblk_line_init_bb(pblk, new, 1)) {
1298                 new = pblk_line_retry(pblk, new);
1299                 if (!new)
1300                         return NULL;
1301
1302                 goto retry_setup;
1303         }
1304
1305         return new;
1306 }
1307
1308 void pblk_line_free(struct pblk *pblk, struct pblk_line *line)
1309 {
1310         if (line->map_bitmap)
1311                 mempool_free(line->map_bitmap, pblk->line_meta_pool);
1312         if (line->invalid_bitmap)
1313                 mempool_free(line->invalid_bitmap, pblk->line_meta_pool);
1314
1315         line->map_bitmap = NULL;
1316         line->invalid_bitmap = NULL;
1317         line->smeta = NULL;
1318         line->emeta = NULL;
1319 }
1320
1321 void pblk_line_put(struct kref *ref)
1322 {
1323         struct pblk_line *line = container_of(ref, struct pblk_line, ref);
1324         struct pblk *pblk = line->pblk;
1325         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1326
1327         spin_lock(&line->lock);
1328         WARN_ON(line->state != PBLK_LINESTATE_GC);
1329         line->state = PBLK_LINESTATE_FREE;
1330         line->gc_group = PBLK_LINEGC_NONE;
1331         pblk_line_free(pblk, line);
1332         spin_unlock(&line->lock);
1333
1334         spin_lock(&l_mg->free_lock);
1335         list_add_tail(&line->list, &l_mg->free_list);
1336         l_mg->nr_free_lines++;
1337         spin_unlock(&l_mg->free_lock);
1338
1339         pblk_rl_free_lines_inc(&pblk->rl, line);
1340 }
1341
1342 int pblk_blk_erase_async(struct pblk *pblk, struct ppa_addr ppa)
1343 {
1344         struct nvm_rq *rqd;
1345         int err;
1346
1347         rqd = mempool_alloc(pblk->r_rq_pool, GFP_KERNEL);
1348         memset(rqd, 0, pblk_r_rq_size);
1349
1350         pblk_setup_e_rq(pblk, rqd, ppa);
1351
1352         rqd->end_io = pblk_end_io_erase;
1353         rqd->private = pblk;
1354
1355         /* The write thread schedules erases so that it minimizes disturbances
1356          * with writes. Thus, there is no need to take the LUN semaphore.
1357          */
1358         err = pblk_submit_io(pblk, rqd);
1359         if (err) {
1360                 struct nvm_tgt_dev *dev = pblk->dev;
1361                 struct nvm_geo *geo = &dev->geo;
1362
1363                 pr_err("pblk: could not async erase line:%d,blk:%d\n",
1364                                         pblk_dev_ppa_to_line(ppa),
1365                                         pblk_dev_ppa_to_pos(geo, ppa));
1366         }
1367
1368         return err;
1369 }
1370
1371 struct pblk_line *pblk_line_get_data(struct pblk *pblk)
1372 {
1373         return pblk->l_mg.data_line;
1374 }
1375
1376 struct pblk_line *pblk_line_get_data_next(struct pblk *pblk)
1377 {
1378         return pblk->l_mg.data_next;
1379 }
1380
1381 int pblk_line_is_full(struct pblk_line *line)
1382 {
1383         return (line->left_msecs == 0);
1384 }
1385
1386 void pblk_line_close(struct pblk *pblk, struct pblk_line *line)
1387 {
1388         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1389         struct list_head *move_list;
1390
1391         line->emeta->crc = cpu_to_le32(pblk_calc_emeta_crc(pblk, line->emeta));
1392
1393         if (pblk_line_submit_emeta_io(pblk, line, line->cur_sec, WRITE))
1394                 pr_err("pblk: line %d close I/O failed\n", line->id);
1395
1396         WARN(!bitmap_full(line->map_bitmap, line->sec_in_line),
1397                                 "pblk: corrupt closed line %d\n", line->id);
1398
1399         spin_lock(&l_mg->free_lock);
1400         WARN_ON(!test_and_clear_bit(line->meta_line, &l_mg->meta_bitmap));
1401         spin_unlock(&l_mg->free_lock);
1402
1403         spin_lock(&l_mg->gc_lock);
1404         spin_lock(&line->lock);
1405         WARN_ON(line->state != PBLK_LINESTATE_OPEN);
1406         line->state = PBLK_LINESTATE_CLOSED;
1407         move_list = pblk_line_gc_list(pblk, line);
1408
1409         list_add_tail(&line->list, move_list);
1410
1411         mempool_free(line->map_bitmap, pblk->line_meta_pool);
1412         line->map_bitmap = NULL;
1413         line->smeta = NULL;
1414         line->emeta = NULL;
1415
1416         spin_unlock(&line->lock);
1417         spin_unlock(&l_mg->gc_lock);
1418 }
1419
1420 void pblk_line_close_ws(struct work_struct *work)
1421 {
1422         struct pblk_line_ws *line_ws = container_of(work, struct pblk_line_ws,
1423                                                                         ws);
1424         struct pblk *pblk = line_ws->pblk;
1425         struct pblk_line *line = line_ws->line;
1426
1427         pblk_line_close(pblk, line);
1428         mempool_free(line_ws, pblk->line_ws_pool);
1429 }
1430
1431 void pblk_line_mark_bb(struct work_struct *work)
1432 {
1433         struct pblk_line_ws *line_ws = container_of(work, struct pblk_line_ws,
1434                                                                         ws);
1435         struct pblk *pblk = line_ws->pblk;
1436         struct nvm_tgt_dev *dev = pblk->dev;
1437         struct ppa_addr *ppa = line_ws->priv;
1438         int ret;
1439
1440         ret = nvm_set_tgt_bb_tbl(dev, ppa, 1, NVM_BLK_T_GRWN_BAD);
1441         if (ret) {
1442                 struct pblk_line *line;
1443                 int pos;
1444
1445                 line = &pblk->lines[pblk_dev_ppa_to_line(*ppa)];
1446                 pos = pblk_dev_ppa_to_pos(&dev->geo, *ppa);
1447
1448                 pr_err("pblk: failed to mark bb, line:%d, pos:%d\n",
1449                                 line->id, pos);
1450         }
1451
1452         kfree(ppa);
1453         mempool_free(line_ws, pblk->line_ws_pool);
1454 }
1455
1456 void pblk_line_run_ws(struct pblk *pblk, struct pblk_line *line, void *priv,
1457                       void (*work)(struct work_struct *))
1458 {
1459         struct pblk_line_ws *line_ws;
1460
1461         line_ws = mempool_alloc(pblk->line_ws_pool, GFP_ATOMIC);
1462         if (!line_ws)
1463                 return;
1464
1465         line_ws->pblk = pblk;
1466         line_ws->line = line;
1467         line_ws->priv = priv;
1468
1469         INIT_WORK(&line_ws->ws, work);
1470         queue_work(pblk->kw_wq, &line_ws->ws);
1471 }
1472
1473 void pblk_down_rq(struct pblk *pblk, struct ppa_addr *ppa_list, int nr_ppas,
1474                   unsigned long *lun_bitmap)
1475 {
1476         struct nvm_tgt_dev *dev = pblk->dev;
1477         struct nvm_geo *geo = &dev->geo;
1478         struct pblk_lun *rlun;
1479         int lun_id = ppa_list[0].g.ch * geo->luns_per_chnl + ppa_list[0].g.lun;
1480         int ret;
1481
1482         /*
1483          * Only send one inflight I/O per LUN. Since we map at a page
1484          * granurality, all ppas in the I/O will map to the same LUN
1485          */
1486 #ifdef CONFIG_NVM_DEBUG
1487         int i;
1488
1489         for (i = 1; i < nr_ppas; i++)
1490                 WARN_ON(ppa_list[0].g.lun != ppa_list[i].g.lun ||
1491                                 ppa_list[0].g.ch != ppa_list[i].g.ch);
1492 #endif
1493         /* If the LUN has been locked for this same request, do no attempt to
1494          * lock it again
1495          */
1496         if (test_and_set_bit(lun_id, lun_bitmap))
1497                 return;
1498
1499         rlun = &pblk->luns[lun_id];
1500         ret = down_timeout(&rlun->wr_sem, msecs_to_jiffies(5000));
1501         if (ret) {
1502                 switch (ret) {
1503                 case -ETIME:
1504                         pr_err("pblk: lun semaphore timed out\n");
1505                         break;
1506                 case -EINTR:
1507                         pr_err("pblk: lun semaphore timed out\n");
1508                         break;
1509                 }
1510         }
1511 }
1512
1513 void pblk_up_rq(struct pblk *pblk, struct ppa_addr *ppa_list, int nr_ppas,
1514                 unsigned long *lun_bitmap)
1515 {
1516         struct nvm_tgt_dev *dev = pblk->dev;
1517         struct nvm_geo *geo = &dev->geo;
1518         struct pblk_lun *rlun;
1519         int nr_luns = geo->nr_luns;
1520         int bit = -1;
1521
1522         while ((bit = find_next_bit(lun_bitmap, nr_luns, bit + 1)) < nr_luns) {
1523                 rlun = &pblk->luns[bit];
1524                 up(&rlun->wr_sem);
1525         }
1526
1527         kfree(lun_bitmap);
1528 }
1529
1530 void pblk_update_map(struct pblk *pblk, sector_t lba, struct ppa_addr ppa)
1531 {
1532         struct ppa_addr l2p_ppa;
1533
1534         /* logic error: lba out-of-bounds. Ignore update */
1535         if (!(lba < pblk->rl.nr_secs)) {
1536                 WARN(1, "pblk: corrupted L2P map request\n");
1537                 return;
1538         }
1539
1540         spin_lock(&pblk->trans_lock);
1541         l2p_ppa = pblk_trans_map_get(pblk, lba);
1542
1543         if (!pblk_addr_in_cache(l2p_ppa) && !pblk_ppa_empty(l2p_ppa))
1544                 pblk_map_invalidate(pblk, l2p_ppa);
1545
1546         pblk_trans_map_set(pblk, lba, ppa);
1547         spin_unlock(&pblk->trans_lock);
1548 }
1549
1550 void pblk_update_map_cache(struct pblk *pblk, sector_t lba, struct ppa_addr ppa)
1551 {
1552 #ifdef CONFIG_NVM_DEBUG
1553         /* Callers must ensure that the ppa points to a cache address */
1554         BUG_ON(!pblk_addr_in_cache(ppa));
1555         BUG_ON(pblk_rb_pos_oob(&pblk->rwb, pblk_addr_to_cacheline(ppa)));
1556 #endif
1557
1558         pblk_update_map(pblk, lba, ppa);
1559 }
1560
1561 int pblk_update_map_gc(struct pblk *pblk, sector_t lba, struct ppa_addr ppa,
1562                        struct pblk_line *gc_line)
1563 {
1564         struct ppa_addr l2p_ppa;
1565         int ret = 1;
1566
1567 #ifdef CONFIG_NVM_DEBUG
1568         /* Callers must ensure that the ppa points to a cache address */
1569         BUG_ON(!pblk_addr_in_cache(ppa));
1570         BUG_ON(pblk_rb_pos_oob(&pblk->rwb, pblk_addr_to_cacheline(ppa)));
1571 #endif
1572
1573         /* logic error: lba out-of-bounds. Ignore update */
1574         if (!(lba < pblk->rl.nr_secs)) {
1575                 WARN(1, "pblk: corrupted L2P map request\n");
1576                 return 0;
1577         }
1578
1579         spin_lock(&pblk->trans_lock);
1580         l2p_ppa = pblk_trans_map_get(pblk, lba);
1581
1582         /* Prevent updated entries to be overwritten by GC */
1583         if (pblk_addr_in_cache(l2p_ppa) || pblk_ppa_empty(l2p_ppa) ||
1584                                 pblk_tgt_ppa_to_line(l2p_ppa) != gc_line->id) {
1585                 ret = 0;
1586                 goto out;
1587         }
1588
1589         pblk_trans_map_set(pblk, lba, ppa);
1590 out:
1591         spin_unlock(&pblk->trans_lock);
1592         return ret;
1593 }
1594
1595 void pblk_update_map_dev(struct pblk *pblk, sector_t lba, struct ppa_addr ppa,
1596                          struct ppa_addr entry_line)
1597 {
1598         struct ppa_addr l2p_line;
1599
1600 #ifdef CONFIG_NVM_DEBUG
1601         /* Callers must ensure that the ppa points to a device address */
1602         BUG_ON(pblk_addr_in_cache(ppa));
1603 #endif
1604         /* Invalidate and discard padded entries */
1605         if (lba == ADDR_EMPTY) {
1606 #ifdef CONFIG_NVM_DEBUG
1607                 atomic_long_inc(&pblk->padded_wb);
1608 #endif
1609                 pblk_map_invalidate(pblk, ppa);
1610                 return;
1611         }
1612
1613         /* logic error: lba out-of-bounds. Ignore update */
1614         if (!(lba < pblk->rl.nr_secs)) {
1615                 WARN(1, "pblk: corrupted L2P map request\n");
1616                 return;
1617         }
1618
1619         spin_lock(&pblk->trans_lock);
1620         l2p_line = pblk_trans_map_get(pblk, lba);
1621
1622         /* Do not update L2P if the cacheline has been updated. In this case,
1623          * the mapped ppa must be invalidated
1624          */
1625         if (l2p_line.ppa != entry_line.ppa) {
1626                 if (!pblk_ppa_empty(ppa))
1627                         pblk_map_invalidate(pblk, ppa);
1628                 goto out;
1629         }
1630
1631 #ifdef CONFIG_NVM_DEBUG
1632         WARN_ON(!pblk_addr_in_cache(l2p_line) && !pblk_ppa_empty(l2p_line));
1633 #endif
1634
1635         pblk_trans_map_set(pblk, lba, ppa);
1636 out:
1637         spin_unlock(&pblk->trans_lock);
1638 }
1639
1640 void pblk_lookup_l2p_seq(struct pblk *pblk, struct ppa_addr *ppas,
1641                          sector_t blba, int nr_secs)
1642 {
1643         int i;
1644
1645         spin_lock(&pblk->trans_lock);
1646         for (i = 0; i < nr_secs; i++)
1647                 ppas[i] = pblk_trans_map_get(pblk, blba + i);
1648         spin_unlock(&pblk->trans_lock);
1649 }
1650
1651 void pblk_lookup_l2p_rand(struct pblk *pblk, struct ppa_addr *ppas,
1652                           u64 *lba_list, int nr_secs)
1653 {
1654         sector_t lba;
1655         int i;
1656
1657         spin_lock(&pblk->trans_lock);
1658         for (i = 0; i < nr_secs; i++) {
1659                 lba = lba_list[i];
1660                 if (lba == ADDR_EMPTY) {
1661                         ppas[i].ppa = ADDR_EMPTY;
1662                 } else {
1663                         /* logic error: lba out-of-bounds. Ignore update */
1664                         if (!(lba < pblk->rl.nr_secs)) {
1665                                 WARN(1, "pblk: corrupted L2P map request\n");
1666                                 continue;
1667                         }
1668                         ppas[i] = pblk_trans_map_get(pblk, lba);
1669                 }
1670         }
1671         spin_unlock(&pblk->trans_lock);
1672 }