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