drivers/lightnvm/pblk-recovery.c

   1 /*
   2  * Copyright (C) 2016 CNEX Labs
   3  * Initial: Javier Gonzalez <javier@cnexlabs.com>
   4  *
   5  * This program is free software; you can redistribute it and/or
   6  * modify it under the terms of the GNU General Public License version
   7  * 2 as published by the Free Software Foundation.
   8  *
   9  * This program is distributed in the hope that it will be useful, but
  10  * WITHOUT ANY WARRANTY; without even the implied warranty of
  11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  12  * General Public License for more details.
  13  *
  14  * pblk-recovery.c - pblk's recovery path
  15  */
  16
  17 #include "pblk.h"
  18
  19 int pblk_recov_check_emeta(struct pblk *pblk, struct line_emeta *emeta_buf)
  20 {
  21         u32 crc;
  22
  23         crc = pblk_calc_emeta_crc(pblk, emeta_buf);
  24         if (le32_to_cpu(emeta_buf->crc) != crc)
  25                 return 1;
  26
  27         if (le32_to_cpu(emeta_buf->header.identifier) != PBLK_MAGIC)
  28                 return 1;
  29
  30         return 0;
  31 }
  32
  33 static int pblk_recov_l2p_from_emeta(struct pblk *pblk, struct pblk_line *line)
  34 {
  35         struct nvm_tgt_dev *dev = pblk->dev;
  36         struct nvm_geo *geo = &dev->geo;
  37         struct pblk_line_meta *lm = &pblk->lm;
  38         struct pblk_emeta *emeta = line->emeta;
  39         struct line_emeta *emeta_buf = emeta->buf;
  40         __le64 *lba_list;
  41         u64 data_start, data_end;
  42         u64 nr_valid_lbas, nr_lbas = 0;
  43         u64 i;
  44
  45         lba_list = emeta_to_lbas(pblk, emeta_buf);
  46         if (!lba_list)
  47                 return 1;
  48
  49         data_start = pblk_line_smeta_start(pblk, line) + lm->smeta_sec;
  50         data_end = line->emeta_ssec;
  51         nr_valid_lbas = le64_to_cpu(emeta_buf->nr_valid_lbas);
  52
  53         for (i = data_start; i < data_end; i++) {
  54                 struct ppa_addr ppa;
  55                 int pos;
  56
  57                 ppa = addr_to_gen_ppa(pblk, i, line->id);
  58                 pos = pblk_ppa_to_pos(geo, ppa);
  59
  60                 /* Do not update bad blocks */
  61                 if (test_bit(pos, line->blk_bitmap))
  62                         continue;
  63
  64                 if (le64_to_cpu(lba_list[i]) == ADDR_EMPTY) {
  65                         spin_lock(&line->lock);
  66                         if (test_and_set_bit(i, line->invalid_bitmap))
  67                                 WARN_ONCE(1, "pblk: rec. double invalidate:\n");
  68                         else
  69                                 le32_add_cpu(line->vsc, -1);
  70                         spin_unlock(&line->lock);
  71
  72                         continue;
  73                 }
  74
  75                 pblk_update_map(pblk, le64_to_cpu(lba_list[i]), ppa);
  76                 nr_lbas++;
  77         }
  78
  79         if (nr_valid_lbas != nr_lbas)
  80                 pblk_err(pblk, "line %d - inconsistent lba list(%llu/%llu)\n",
  81                                 line->id, nr_valid_lbas, nr_lbas);
  82
  83         line->left_msecs = 0;
  84
  85         return 0;
  86 }
  87
  88 static int pblk_calc_sec_in_line(struct pblk *pblk, struct pblk_line *line)
  89 {
  90         struct nvm_tgt_dev *dev = pblk->dev;
  91         struct nvm_geo *geo = &dev->geo;
  92         struct pblk_line_meta *lm = &pblk->lm;
  93         int nr_bb = bitmap_weight(line->blk_bitmap, lm->blk_per_line);
  94
  95         return lm->sec_per_line - lm->smeta_sec - lm->emeta_sec[0] -
  96                                 nr_bb * geo->clba;
  97 }
  98
  99 struct pblk_recov_alloc {
 100         struct ppa_addr *ppa_list;
 101         struct pblk_sec_meta *meta_list;
 102         struct nvm_rq *rqd;
 103         void *data;
 104         dma_addr_t dma_ppa_list;
 105         dma_addr_t dma_meta_list;
 106 };
 107
 108 static int pblk_recov_read_oob(struct pblk *pblk, struct pblk_line *line,
 109                                struct pblk_recov_alloc p, u64 r_ptr)
 110 {
 111         struct nvm_tgt_dev *dev = pblk->dev;
 112         struct nvm_geo *geo = &dev->geo;
 113         struct ppa_addr *ppa_list;
 114         struct pblk_sec_meta *meta_list;
 115         struct nvm_rq *rqd;
 116         struct bio *bio;
 117         void *data;
 118         dma_addr_t dma_ppa_list, dma_meta_list;
 119         u64 r_ptr_int;
 120         int left_ppas;
 121         int rq_ppas, rq_len;
 122         int i, j;
 123         int ret = 0;
 124
 125         ppa_list = p.ppa_list;
 126         meta_list = p.meta_list;
 127         rqd = p.rqd;
 128         data = p.data;
 129         dma_ppa_list = p.dma_ppa_list;
 130         dma_meta_list = p.dma_meta_list;
 131
 132         left_ppas = line->cur_sec - r_ptr;
 133         if (!left_ppas)
 134                 return 0;
 135
 136         r_ptr_int = r_ptr;
 137
 138 next_read_rq:
 139         memset(rqd, 0, pblk_g_rq_size);
 140
 141         rq_ppas = pblk_calc_secs(pblk, left_ppas, 0);
 142         if (!rq_ppas)
 143                 rq_ppas = pblk->min_write_pgs;
 144         rq_len = rq_ppas * geo->csecs;
 145
 146         bio = bio_map_kern(dev->q, data, rq_len, GFP_KERNEL);
 147         if (IS_ERR(bio))
 148                 return PTR_ERR(bio);
 149
 150         bio->bi_iter.bi_sector = 0; /* internal bio */
 151         bio_set_op_attrs(bio, REQ_OP_READ, 0);
 152
 153         rqd->bio = bio;
 154         rqd->opcode = NVM_OP_PREAD;
 155         rqd->meta_list = meta_list;
 156         rqd->nr_ppas = rq_ppas;
 157         rqd->ppa_list = ppa_list;
 158         rqd->dma_ppa_list = dma_ppa_list;
 159         rqd->dma_meta_list = dma_meta_list;
 160
 161         if (pblk_io_aligned(pblk, rq_ppas))
 162                 rqd->flags = pblk_set_read_mode(pblk, PBLK_READ_SEQUENTIAL);
 163         else
 164                 rqd->flags = pblk_set_read_mode(pblk, PBLK_READ_RANDOM);
 165
 166         for (i = 0; i < rqd->nr_ppas; ) {
 167                 struct ppa_addr ppa;
 168                 int pos;
 169
 170                 ppa = addr_to_gen_ppa(pblk, r_ptr_int, line->id);
 171                 pos = pblk_ppa_to_pos(geo, ppa);
 172
 173                 while (test_bit(pos, line->blk_bitmap)) {
 174                         r_ptr_int += pblk->min_write_pgs;
 175                         ppa = addr_to_gen_ppa(pblk, r_ptr_int, line->id);
 176                         pos = pblk_ppa_to_pos(geo, ppa);
 177                 }
 178
 179                 for (j = 0; j < pblk->min_write_pgs; j++, i++, r_ptr_int++)
 180                         rqd->ppa_list[i] =
 181                                 addr_to_gen_ppa(pblk, r_ptr_int, line->id);
 182         }
 183
 184         /* If read fails, more padding is needed */
 185         ret = pblk_submit_io_sync(pblk, rqd);
 186         if (ret) {
 187                 pblk_err(pblk, "I/O submission failed: %d\n", ret);
 188                 return ret;
 189         }
 190
 191         atomic_dec(&pblk->inflight_io);
 192
 193         /* At this point, the read should not fail. If it does, it is a problem
 194          * we cannot recover from here. Need FTL log.
 195          */
 196         if (rqd->error && rqd->error != NVM_RSP_WARN_HIGHECC) {
 197                 pblk_err(pblk, "L2P recovery failed (%d)\n", rqd->error);
 198                 return -EINTR;
 199         }
 200
 201         for (i = 0; i < rqd->nr_ppas; i++) {
 202                 u64 lba = le64_to_cpu(meta_list[i].lba);
 203
 204                 if (lba == ADDR_EMPTY || lba > pblk->rl.nr_secs)
 205                         continue;
 206
 207                 pblk_update_map(pblk, lba, rqd->ppa_list[i]);
 208         }
 209
 210         left_ppas -= rq_ppas;
 211         if (left_ppas > 0)
 212                 goto next_read_rq;
 213
 214         return 0;
 215 }
 216
 217 static void pblk_recov_complete(struct kref *ref)
 218 {
 219         struct pblk_pad_rq *pad_rq = container_of(ref, struct pblk_pad_rq, ref);
 220
 221         complete(&pad_rq->wait);
 222 }
 223
 224 static void pblk_end_io_recov(struct nvm_rq *rqd)
 225 {
 226         struct pblk_pad_rq *pad_rq = rqd->private;
 227         struct pblk *pblk = pad_rq->pblk;
 228
 229         pblk_up_page(pblk, rqd->ppa_list, rqd->nr_ppas);
 230
 231         pblk_free_rqd(pblk, rqd, PBLK_WRITE_INT);
 232
 233         atomic_dec(&pblk->inflight_io);
 234         kref_put(&pad_rq->ref, pblk_recov_complete);
 235 }
 236
 237 static int pblk_recov_pad_oob(struct pblk *pblk, struct pblk_line *line,
 238                               int left_ppas)
 239 {
 240         struct nvm_tgt_dev *dev = pblk->dev;
 241         struct nvm_geo *geo = &dev->geo;
 242         struct ppa_addr *ppa_list;
 243         struct pblk_sec_meta *meta_list;
 244         struct pblk_pad_rq *pad_rq;
 245         struct nvm_rq *rqd;
 246         struct bio *bio;
 247         void *data;
 248         dma_addr_t dma_ppa_list, dma_meta_list;
 249         __le64 *lba_list = emeta_to_lbas(pblk, line->emeta->buf);
 250         u64 w_ptr = line->cur_sec;
 251         int left_line_ppas, rq_ppas, rq_len;
 252         int i, j;
 253         int ret = 0;
 254
 255         spin_lock(&line->lock);
 256         left_line_ppas = line->left_msecs;
 257         spin_unlock(&line->lock);
 258
 259         pad_rq = kmalloc(sizeof(struct pblk_pad_rq), GFP_KERNEL);
 260         if (!pad_rq)
 261                 return -ENOMEM;
 262
 263         data = vzalloc(array_size(pblk->max_write_pgs, geo->csecs));
 264         if (!data) {
 265                 ret = -ENOMEM;
 266                 goto free_rq;
 267         }
 268
 269         pad_rq->pblk = pblk;
 270         init_completion(&pad_rq->wait);
 271         kref_init(&pad_rq->ref);
 272
 273 next_pad_rq:
 274         rq_ppas = pblk_calc_secs(pblk, left_ppas, 0);
 275         if (rq_ppas < pblk->min_write_pgs) {
 276                 pblk_err(pblk, "corrupted pad line %d\n", line->id);
 277                 goto fail_free_pad;
 278         }
 279
 280         rq_len = rq_ppas * geo->csecs;
 281
 282         meta_list = nvm_dev_dma_alloc(dev->parent, GFP_KERNEL, &dma_meta_list);
 283         if (!meta_list) {
 284                 ret = -ENOMEM;
 285                 goto fail_free_pad;
 286         }
 287
 288         ppa_list = (void *)(meta_list) + pblk_dma_meta_size;
 289         dma_ppa_list = dma_meta_list + pblk_dma_meta_size;
 290
 291         bio = pblk_bio_map_addr(pblk, data, rq_ppas, rq_len,
 292                                                 PBLK_VMALLOC_META, GFP_KERNEL);
 293         if (IS_ERR(bio)) {
 294                 ret = PTR_ERR(bio);
 295                 goto fail_free_meta;
 296         }
 297
 298         bio->bi_iter.bi_sector = 0; /* internal bio */
 299         bio_set_op_attrs(bio, REQ_OP_WRITE, 0);
 300
 301         rqd = pblk_alloc_rqd(pblk, PBLK_WRITE_INT);
 302
 303         rqd->bio = bio;
 304         rqd->opcode = NVM_OP_PWRITE;
 305         rqd->flags = pblk_set_progr_mode(pblk, PBLK_WRITE);
 306         rqd->meta_list = meta_list;
 307         rqd->nr_ppas = rq_ppas;
 308         rqd->ppa_list = ppa_list;
 309         rqd->dma_ppa_list = dma_ppa_list;
 310         rqd->dma_meta_list = dma_meta_list;
 311         rqd->end_io = pblk_end_io_recov;
 312         rqd->private = pad_rq;
 313
 314         for (i = 0; i < rqd->nr_ppas; ) {
 315                 struct ppa_addr ppa;
 316                 int pos;
 317
 318                 w_ptr = pblk_alloc_page(pblk, line, pblk->min_write_pgs);
 319                 ppa = addr_to_gen_ppa(pblk, w_ptr, line->id);
 320                 pos = pblk_ppa_to_pos(geo, ppa);
 321
 322                 while (test_bit(pos, line->blk_bitmap)) {
 323                         w_ptr += pblk->min_write_pgs;
 324                         ppa = addr_to_gen_ppa(pblk, w_ptr, line->id);
 325                         pos = pblk_ppa_to_pos(geo, ppa);
 326                 }
 327
 328                 for (j = 0; j < pblk->min_write_pgs; j++, i++, w_ptr++) {
 329                         struct ppa_addr dev_ppa;
 330                         __le64 addr_empty = cpu_to_le64(ADDR_EMPTY);
 331
 332                         dev_ppa = addr_to_gen_ppa(pblk, w_ptr, line->id);
 333
 334                         pblk_map_invalidate(pblk, dev_ppa);
 335                         lba_list[w_ptr] = meta_list[i].lba = addr_empty;
 336                         rqd->ppa_list[i] = dev_ppa;
 337                 }
 338         }
 339
 340         kref_get(&pad_rq->ref);
 341         pblk_down_page(pblk, rqd->ppa_list, rqd->nr_ppas);
 342
 343         ret = pblk_submit_io(pblk, rqd);
 344         if (ret) {
 345                 pblk_err(pblk, "I/O submission failed: %d\n", ret);
 346                 pblk_up_page(pblk, rqd->ppa_list, rqd->nr_ppas);
 347                 goto fail_free_bio;
 348         }
 349
 350         left_line_ppas -= rq_ppas;
 351         left_ppas -= rq_ppas;
 352         if (left_ppas && left_line_ppas)
 353                 goto next_pad_rq;
 354
 355         kref_put(&pad_rq->ref, pblk_recov_complete);
 356
 357         if (!wait_for_completion_io_timeout(&pad_rq->wait,
 358                                 msecs_to_jiffies(PBLK_COMMAND_TIMEOUT_MS))) {
 359                 pblk_err(pblk, "pad write timed out\n");
 360                 ret = -ETIME;
 361         }
 362
 363         if (!pblk_line_is_full(line))
 364                 pblk_err(pblk, "corrupted padded line: %d\n", line->id);
 365
 366         vfree(data);
 367 free_rq:
 368         kfree(pad_rq);
 369         return ret;
 370
 371 fail_free_bio:
 372         bio_put(bio);
 373 fail_free_meta:
 374         nvm_dev_dma_free(dev->parent, meta_list, dma_meta_list);
 375 fail_free_pad:
 376         kfree(pad_rq);
 377         vfree(data);
 378         return ret;
 379 }
 380
 381 /* When this function is called, it means that not all upper pages have been
 382  * written in a page that contains valid data. In order to recover this data, we
 383  * first find the write pointer on the device, then we pad all necessary
 384  * sectors, and finally attempt to read the valid data
 385  */
 386 static int pblk_recov_scan_all_oob(struct pblk *pblk, struct pblk_line *line,
 387                                    struct pblk_recov_alloc p)
 388 {
 389         struct nvm_tgt_dev *dev = pblk->dev;
 390         struct nvm_geo *geo = &dev->geo;
 391         struct ppa_addr *ppa_list;
 392         struct pblk_sec_meta *meta_list;
 393         struct nvm_rq *rqd;
 394         struct bio *bio;
 395         void *data;
 396         dma_addr_t dma_ppa_list, dma_meta_list;
 397         u64 w_ptr = 0, r_ptr;
 398         int rq_ppas, rq_len;
 399         int i, j;
 400         int ret = 0;
 401         int rec_round;
 402         int left_ppas = pblk_calc_sec_in_line(pblk, line) - line->cur_sec;
 403
 404         ppa_list = p.ppa_list;
 405         meta_list = p.meta_list;
 406         rqd = p.rqd;
 407         data = p.data;
 408         dma_ppa_list = p.dma_ppa_list;
 409         dma_meta_list = p.dma_meta_list;
 410
 411         /* we could recover up until the line write pointer */
 412         r_ptr = line->cur_sec;
 413         rec_round = 0;
 414
 415 next_rq:
 416         memset(rqd, 0, pblk_g_rq_size);
 417
 418         rq_ppas = pblk_calc_secs(pblk, left_ppas, 0);
 419         if (!rq_ppas)
 420                 rq_ppas = pblk->min_write_pgs;
 421         rq_len = rq_ppas * geo->csecs;
 422
 423         bio = bio_map_kern(dev->q, data, rq_len, GFP_KERNEL);
 424         if (IS_ERR(bio))
 425                 return PTR_ERR(bio);
 426
 427         bio->bi_iter.bi_sector = 0; /* internal bio */
 428         bio_set_op_attrs(bio, REQ_OP_READ, 0);
 429
 430         rqd->bio = bio;
 431         rqd->opcode = NVM_OP_PREAD;
 432         rqd->meta_list = meta_list;
 433         rqd->nr_ppas = rq_ppas;
 434         rqd->ppa_list = ppa_list;
 435         rqd->dma_ppa_list = dma_ppa_list;
 436         rqd->dma_meta_list = dma_meta_list;
 437
 438         if (pblk_io_aligned(pblk, rq_ppas))
 439                 rqd->flags = pblk_set_read_mode(pblk, PBLK_READ_SEQUENTIAL);
 440         else
 441                 rqd->flags = pblk_set_read_mode(pblk, PBLK_READ_RANDOM);
 442
 443         for (i = 0; i < rqd->nr_ppas; ) {
 444                 struct ppa_addr ppa;
 445                 int pos;
 446
 447                 w_ptr = pblk_alloc_page(pblk, line, pblk->min_write_pgs);
 448                 ppa = addr_to_gen_ppa(pblk, w_ptr, line->id);
 449                 pos = pblk_ppa_to_pos(geo, ppa);
 450
 451                 while (test_bit(pos, line->blk_bitmap)) {
 452                         w_ptr += pblk->min_write_pgs;
 453                         ppa = addr_to_gen_ppa(pblk, w_ptr, line->id);
 454                         pos = pblk_ppa_to_pos(geo, ppa);
 455                 }
 456
 457                 for (j = 0; j < pblk->min_write_pgs; j++, i++, w_ptr++)
 458                         rqd->ppa_list[i] =
 459                                 addr_to_gen_ppa(pblk, w_ptr, line->id);
 460         }
 461
 462         ret = pblk_submit_io_sync(pblk, rqd);
 463         if (ret) {
 464                 pblk_err(pblk, "I/O submission failed: %d\n", ret);
 465                 return ret;
 466         }
 467
 468         atomic_dec(&pblk->inflight_io);
 469
 470         /* This should not happen since the read failed during normal recovery,
 471          * but the media works funny sometimes...
 472          */
 473         if (!rec_round++ && !rqd->error) {
 474                 rec_round = 0;
 475                 for (i = 0; i < rqd->nr_ppas; i++, r_ptr++) {
 476                         u64 lba = le64_to_cpu(meta_list[i].lba);
 477
 478                         if (lba == ADDR_EMPTY || lba > pblk->rl.nr_secs)
 479                                 continue;
 480
 481                         pblk_update_map(pblk, lba, rqd->ppa_list[i]);
 482                 }
 483         }
 484
 485         /* Reached the end of the written line */
 486         if (rqd->error == NVM_RSP_ERR_EMPTYPAGE) {
 487                 int pad_secs, nr_error_bits, bit;
 488                 int ret;
 489
 490                 bit = find_first_bit((void *)&rqd->ppa_status, rqd->nr_ppas);
 491                 nr_error_bits = rqd->nr_ppas - bit;
 492
 493                 /* Roll back failed sectors */
 494                 line->cur_sec -= nr_error_bits;
 495                 line->left_msecs += nr_error_bits;
 496                 bitmap_clear(line->map_bitmap, line->cur_sec, nr_error_bits);
 497
 498                 pad_secs = pblk_pad_distance(pblk);
 499                 if (pad_secs > line->left_msecs)
 500                         pad_secs = line->left_msecs;
 501
 502                 ret = pblk_recov_pad_oob(pblk, line, pad_secs);
 503                 if (ret)
 504                         pblk_err(pblk, "OOB padding failed (err:%d)\n", ret);
 505
 506                 ret = pblk_recov_read_oob(pblk, line, p, r_ptr);
 507                 if (ret)
 508                         pblk_err(pblk, "OOB read failed (err:%d)\n", ret);
 509
 510                 left_ppas = 0;
 511         }
 512
 513         left_ppas -= rq_ppas;
 514         if (left_ppas > 0)
 515                 goto next_rq;
 516
 517         return ret;
 518 }
 519
 520 static int pblk_recov_scan_oob(struct pblk *pblk, struct pblk_line *line,
 521                                struct pblk_recov_alloc p, int *done)
 522 {
 523         struct nvm_tgt_dev *dev = pblk->dev;
 524         struct nvm_geo *geo = &dev->geo;
 525         struct ppa_addr *ppa_list;
 526         struct pblk_sec_meta *meta_list;
 527         struct nvm_rq *rqd;
 528         struct bio *bio;
 529         void *data;
 530         dma_addr_t dma_ppa_list, dma_meta_list;
 531         u64 paddr;
 532         int rq_ppas, rq_len;
 533         int i, j;
 534         int ret = 0;
 535         int left_ppas = pblk_calc_sec_in_line(pblk, line);
 536
 537         ppa_list = p.ppa_list;
 538         meta_list = p.meta_list;
 539         rqd = p.rqd;
 540         data = p.data;
 541         dma_ppa_list = p.dma_ppa_list;
 542         dma_meta_list = p.dma_meta_list;
 543
 544         *done = 1;
 545
 546 next_rq:
 547         memset(rqd, 0, pblk_g_rq_size);
 548
 549         rq_ppas = pblk_calc_secs(pblk, left_ppas, 0);
 550         if (!rq_ppas)
 551                 rq_ppas = pblk->min_write_pgs;
 552         rq_len = rq_ppas * geo->csecs;
 553
 554         bio = bio_map_kern(dev->q, data, rq_len, GFP_KERNEL);
 555         if (IS_ERR(bio))
 556                 return PTR_ERR(bio);
 557
 558         bio->bi_iter.bi_sector = 0; /* internal bio */
 559         bio_set_op_attrs(bio, REQ_OP_READ, 0);
 560
 561         rqd->bio = bio;
 562         rqd->opcode = NVM_OP_PREAD;
 563         rqd->meta_list = meta_list;
 564         rqd->nr_ppas = rq_ppas;
 565         rqd->ppa_list = ppa_list;
 566         rqd->dma_ppa_list = dma_ppa_list;
 567         rqd->dma_meta_list = dma_meta_list;
 568
 569         if (pblk_io_aligned(pblk, rq_ppas))
 570                 rqd->flags = pblk_set_read_mode(pblk, PBLK_READ_SEQUENTIAL);
 571         else
 572                 rqd->flags = pblk_set_read_mode(pblk, PBLK_READ_RANDOM);
 573
 574         for (i = 0; i < rqd->nr_ppas; ) {
 575                 struct ppa_addr ppa;
 576                 int pos;
 577
 578                 paddr = pblk_alloc_page(pblk, line, pblk->min_write_pgs);
 579                 ppa = addr_to_gen_ppa(pblk, paddr, line->id);
 580                 pos = pblk_ppa_to_pos(geo, ppa);
 581
 582                 while (test_bit(pos, line->blk_bitmap)) {
 583                         paddr += pblk->min_write_pgs;
 584                         ppa = addr_to_gen_ppa(pblk, paddr, line->id);
 585                         pos = pblk_ppa_to_pos(geo, ppa);
 586                 }
 587
 588                 for (j = 0; j < pblk->min_write_pgs; j++, i++, paddr++)
 589                         rqd->ppa_list[i] =
 590                                 addr_to_gen_ppa(pblk, paddr, line->id);
 591         }
 592
 593         ret = pblk_submit_io_sync(pblk, rqd);
 594         if (ret) {
 595                 pblk_err(pblk, "I/O submission failed: %d\n", ret);
 596                 bio_put(bio);
 597                 return ret;
 598         }
 599
 600         atomic_dec(&pblk->inflight_io);
 601
 602         /* Reached the end of the written line */
 603         if (rqd->error) {
 604                 int nr_error_bits, bit;
 605
 606                 bit = find_first_bit((void *)&rqd->ppa_status, rqd->nr_ppas);
 607                 nr_error_bits = rqd->nr_ppas - bit;
 608
 609                 /* Roll back failed sectors */
 610                 line->cur_sec -= nr_error_bits;
 611                 line->left_msecs += nr_error_bits;
 612                 bitmap_clear(line->map_bitmap, line->cur_sec, nr_error_bits);
 613
 614                 left_ppas = 0;
 615                 rqd->nr_ppas = bit;
 616
 617                 if (rqd->error != NVM_RSP_ERR_EMPTYPAGE)
 618                         *done = 0;
 619         }
 620
 621         for (i = 0; i < rqd->nr_ppas; i++) {
 622                 u64 lba = le64_to_cpu(meta_list[i].lba);
 623
 624                 if (lba == ADDR_EMPTY || lba > pblk->rl.nr_secs)
 625                         continue;
 626
 627                 pblk_update_map(pblk, lba, rqd->ppa_list[i]);
 628         }
 629
 630         left_ppas -= rq_ppas;
 631         if (left_ppas > 0)
 632                 goto next_rq;
 633
 634         return ret;
 635 }
 636
 637 /* Scan line for lbas on out of bound area */
 638 static int pblk_recov_l2p_from_oob(struct pblk *pblk, struct pblk_line *line)
 639 {
 640         struct nvm_tgt_dev *dev = pblk->dev;
 641         struct nvm_geo *geo = &dev->geo;
 642         struct nvm_rq *rqd;
 643         struct ppa_addr *ppa_list;
 644         struct pblk_sec_meta *meta_list;
 645         struct pblk_recov_alloc p;
 646         void *data;
 647         dma_addr_t dma_ppa_list, dma_meta_list;
 648         int done, ret = 0;
 649
 650         meta_list = nvm_dev_dma_alloc(dev->parent, GFP_KERNEL, &dma_meta_list);
 651         if (!meta_list)
 652                 return -ENOMEM;
 653
 654         ppa_list = (void *)(meta_list) + pblk_dma_meta_size;
 655         dma_ppa_list = dma_meta_list + pblk_dma_meta_size;
 656
 657         data = kcalloc(pblk->max_write_pgs, geo->csecs, GFP_KERNEL);
 658         if (!data) {
 659                 ret = -ENOMEM;
 660                 goto free_meta_list;
 661         }
 662
 663         rqd = pblk_alloc_rqd(pblk, PBLK_READ);
 664
 665         p.ppa_list = ppa_list;
 666         p.meta_list = meta_list;
 667         p.rqd = rqd;
 668         p.data = data;
 669         p.dma_ppa_list = dma_ppa_list;
 670         p.dma_meta_list = dma_meta_list;
 671
 672         ret = pblk_recov_scan_oob(pblk, line, p, &done);
 673         if (ret) {
 674                 pblk_err(pblk, "could not recover L2P from OOB\n");
 675                 goto out;
 676         }
 677
 678         if (!done) {
 679                 ret = pblk_recov_scan_all_oob(pblk, line, p);
 680                 if (ret) {
 681                         pblk_err(pblk, "could not recover L2P from OOB\n");
 682                         goto out;
 683                 }
 684         }
 685
 686         if (pblk_line_is_full(line))
 687                 pblk_line_recov_close(pblk, line);
 688
 689 out:
 690         kfree(data);
 691 free_meta_list:
 692         nvm_dev_dma_free(dev->parent, meta_list, dma_meta_list);
 693
 694         return ret;
 695 }
 696
 697 /* Insert lines ordered by sequence number (seq_num) on list */
 698 static void pblk_recov_line_add_ordered(struct list_head *head,
 699                                         struct pblk_line *line)
 700 {
 701         struct pblk_line *t = NULL;
 702
 703         list_for_each_entry(t, head, list)
 704                 if (t->seq_nr > line->seq_nr)
 705                         break;
 706
 707         __list_add(&line->list, t->list.prev, &t->list);
 708 }
 709
 710 static u64 pblk_line_emeta_start(struct pblk *pblk, struct pblk_line *line)
 711 {
 712         struct nvm_tgt_dev *dev = pblk->dev;
 713         struct nvm_geo *geo = &dev->geo;
 714         struct pblk_line_meta *lm = &pblk->lm;
 715         unsigned int emeta_secs;
 716         u64 emeta_start;
 717         struct ppa_addr ppa;
 718         int pos;
 719
 720         emeta_secs = lm->emeta_sec[0];
 721         emeta_start = lm->sec_per_line;
 722
 723         while (emeta_secs) {
 724                 emeta_start--;
 725                 ppa = addr_to_gen_ppa(pblk, emeta_start, line->id);
 726                 pos = pblk_ppa_to_pos(geo, ppa);
 727                 if (!test_bit(pos, line->blk_bitmap))
 728                         emeta_secs--;
 729         }
 730
 731         return emeta_start;
 732 }
 733
 734 static int pblk_recov_check_line_version(struct pblk *pblk,
 735                                          struct line_emeta *emeta)
 736 {
 737         struct line_header *header = &emeta->header;
 738
 739         if (header->version_major != EMETA_VERSION_MAJOR) {
 740                 pblk_err(pblk, "line major version mismatch: %d, expected: %d\n",
 741                          header->version_major, EMETA_VERSION_MAJOR);
 742                 return 1;
 743         }
 744
 745 #ifdef CONFIG_NVM_PBLK_DEBUG
 746         if (header->version_minor > EMETA_VERSION_MINOR)
 747                 pblk_info(pblk, "newer line minor version found: %d\n",
 748                                 header->version_minor);
 749 #endif
 750
 751         return 0;
 752 }
 753
 754 static void pblk_recov_wa_counters(struct pblk *pblk,
 755                                    struct line_emeta *emeta)
 756 {
 757         struct pblk_line_meta *lm = &pblk->lm;
 758         struct line_header *header = &emeta->header;
 759         struct wa_counters *wa = emeta_to_wa(lm, emeta);
 760
 761         /* WA counters were introduced in emeta version 0.2 */
 762         if (header->version_major > 0 || header->version_minor >= 2) {
 763                 u64 user = le64_to_cpu(wa->user);
 764                 u64 pad = le64_to_cpu(wa->pad);
 765                 u64 gc = le64_to_cpu(wa->gc);
 766
 767                 atomic64_set(&pblk->user_wa, user);
 768                 atomic64_set(&pblk->pad_wa, pad);
 769                 atomic64_set(&pblk->gc_wa, gc);
 770
 771                 pblk->user_rst_wa = user;
 772                 pblk->pad_rst_wa = pad;
 773                 pblk->gc_rst_wa = gc;
 774         }
 775 }
 776
 777 static int pblk_line_was_written(struct pblk_line *line,
 778                             struct pblk *pblk)
 779 {
 780
 781         struct pblk_line_meta *lm = &pblk->lm;
 782         struct nvm_tgt_dev *dev = pblk->dev;
 783         struct nvm_geo *geo = &dev->geo;
 784         struct nvm_chk_meta *chunk;
 785         struct ppa_addr bppa;
 786         int smeta_blk;
 787
 788         if (line->state == PBLK_LINESTATE_BAD)
 789                 return 0;
 790
 791         smeta_blk = find_first_zero_bit(line->blk_bitmap, lm->blk_per_line);
 792         if (smeta_blk >= lm->blk_per_line)
 793                 return 0;
 794
 795         bppa = pblk->luns[smeta_blk].bppa;
 796         chunk = &line->chks[pblk_ppa_to_pos(geo, bppa)];
 797
 798         if (chunk->state & NVM_CHK_ST_FREE)
 799                 return 0;
 800
 801         return 1;
 802 }
 803
 804 struct pblk_line *pblk_recov_l2p(struct pblk *pblk)
 805 {
 806         struct pblk_line_meta *lm = &pblk->lm;
 807         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
 808         struct pblk_line *line, *tline, *data_line = NULL;
 809         struct pblk_smeta *smeta;
 810         struct pblk_emeta *emeta;
 811         struct line_smeta *smeta_buf;
 812         int found_lines = 0, recovered_lines = 0, open_lines = 0;
 813         int is_next = 0;
 814         int meta_line;
 815         int i, valid_uuid = 0;
 816         LIST_HEAD(recov_list);
 817
 818         /* TODO: Implement FTL snapshot */
 819
 820         /* Scan recovery - takes place when FTL snapshot fails */
 821         spin_lock(&l_mg->free_lock);
 822         meta_line = find_first_zero_bit(&l_mg->meta_bitmap, PBLK_DATA_LINES);
 823         set_bit(meta_line, &l_mg->meta_bitmap);
 824         smeta = l_mg->sline_meta[meta_line];
 825         emeta = l_mg->eline_meta[meta_line];
 826         smeta_buf = (struct line_smeta *)smeta;
 827         spin_unlock(&l_mg->free_lock);
 828
 829         /* Order data lines using their sequence number */
 830         for (i = 0; i < l_mg->nr_lines; i++) {
 831                 u32 crc;
 832
 833                 line = &pblk->lines[i];
 834
 835                 memset(smeta, 0, lm->smeta_len);
 836                 line->smeta = smeta;
 837                 line->lun_bitmap = ((void *)(smeta_buf)) +
 838                                                 sizeof(struct line_smeta);
 839
 840                 if (!pblk_line_was_written(line, pblk))
 841                         continue;
 842
 843                 /* Lines that cannot be read are assumed as not written here */
 844                 if (pblk_line_read_smeta(pblk, line))
 845                         continue;
 846
 847                 crc = pblk_calc_smeta_crc(pblk, smeta_buf);
 848                 if (le32_to_cpu(smeta_buf->crc) != crc)
 849                         continue;
 850
 851                 if (le32_to_cpu(smeta_buf->header.identifier) != PBLK_MAGIC)
 852                         continue;
 853
 854                 if (smeta_buf->header.version_major != SMETA_VERSION_MAJOR) {
 855                         pblk_err(pblk, "found incompatible line version %u\n",
 856                                         smeta_buf->header.version_major);
 857                         return ERR_PTR(-EINVAL);
 858                 }
 859
 860                 /* The first valid instance uuid is used for initialization */
 861                 if (!valid_uuid) {
 862                         memcpy(pblk->instance_uuid, smeta_buf->header.uuid, 16);
 863                         valid_uuid = 1;
 864                 }
 865
 866                 if (memcmp(pblk->instance_uuid, smeta_buf->header.uuid, 16)) {
 867                         pblk_debug(pblk, "ignore line %u due to uuid mismatch\n",
 868                                         i);
 869                         continue;
 870                 }
 871
 872                 /* Update line metadata */
 873                 spin_lock(&line->lock);
 874                 line->id = le32_to_cpu(smeta_buf->header.id);
 875                 line->type = le16_to_cpu(smeta_buf->header.type);
 876                 line->seq_nr = le64_to_cpu(smeta_buf->seq_nr);
 877                 spin_unlock(&line->lock);
 878
 879                 /* Update general metadata */
 880                 spin_lock(&l_mg->free_lock);
 881                 if (line->seq_nr >= l_mg->d_seq_nr)
 882                         l_mg->d_seq_nr = line->seq_nr + 1;
 883                 l_mg->nr_free_lines--;
 884                 spin_unlock(&l_mg->free_lock);
 885
 886                 if (pblk_line_recov_alloc(pblk, line))
 887                         goto out;
 888
 889                 pblk_recov_line_add_ordered(&recov_list, line);
 890                 found_lines++;
 891                 pblk_debug(pblk, "recovering data line %d, seq:%llu\n",
 892                                                 line->id, smeta_buf->seq_nr);
 893         }
 894
 895         if (!found_lines) {
 896                 pblk_setup_uuid(pblk);
 897
 898                 spin_lock(&l_mg->free_lock);
 899                 WARN_ON_ONCE(!test_and_clear_bit(meta_line,
 900                                                         &l_mg->meta_bitmap));
 901                 spin_unlock(&l_mg->free_lock);
 902
 903                 goto out;
 904         }
 905
 906         /* Verify closed blocks and recover this portion of L2P table*/
 907         list_for_each_entry_safe(line, tline, &recov_list, list) {
 908                 recovered_lines++;
 909
 910                 line->emeta_ssec = pblk_line_emeta_start(pblk, line);
 911                 line->emeta = emeta;
 912                 memset(line->emeta->buf, 0, lm->emeta_len[0]);
 913
 914                 if (pblk_line_read_emeta(pblk, line, line->emeta->buf)) {
 915                         pblk_recov_l2p_from_oob(pblk, line);
 916                         goto next;
 917                 }
 918
 919                 if (pblk_recov_check_emeta(pblk, line->emeta->buf)) {
 920                         pblk_recov_l2p_from_oob(pblk, line);
 921                         goto next;
 922                 }
 923
 924                 if (pblk_recov_check_line_version(pblk, line->emeta->buf))
 925                         return ERR_PTR(-EINVAL);
 926
 927                 pblk_recov_wa_counters(pblk, line->emeta->buf);
 928
 929                 if (pblk_recov_l2p_from_emeta(pblk, line))
 930                         pblk_recov_l2p_from_oob(pblk, line);
 931
 932 next:
 933                 if (pblk_line_is_full(line)) {
 934                         struct list_head *move_list;
 935
 936                         spin_lock(&line->lock);
 937                         line->state = PBLK_LINESTATE_CLOSED;
 938                         move_list = pblk_line_gc_list(pblk, line);
 939                         spin_unlock(&line->lock);
 940
 941                         spin_lock(&l_mg->gc_lock);
 942                         list_move_tail(&line->list, move_list);
 943                         spin_unlock(&l_mg->gc_lock);
 944
 945                         kfree(line->map_bitmap);
 946                         line->map_bitmap = NULL;
 947                         line->smeta = NULL;
 948                         line->emeta = NULL;
 949                 } else {
 950                         if (open_lines > 1)
 951                                 pblk_err(pblk, "failed to recover L2P\n");
 952
 953                         open_lines++;
 954                         line->meta_line = meta_line;
 955                         data_line = line;
 956                 }
 957         }
 958
 959         if (!open_lines) {
 960                 spin_lock(&l_mg->free_lock);
 961                 WARN_ON_ONCE(!test_and_clear_bit(meta_line,
 962                                                         &l_mg->meta_bitmap));
 963                 spin_unlock(&l_mg->free_lock);
 964                 pblk_line_replace_data(pblk);
 965         } else {
 966                 spin_lock(&l_mg->free_lock);
 967                 /* Allocate next line for preparation */
 968                 l_mg->data_next = pblk_line_get(pblk);
 969                 if (l_mg->data_next) {
 970                         l_mg->data_next->seq_nr = l_mg->d_seq_nr++;
 971                         l_mg->data_next->type = PBLK_LINETYPE_DATA;
 972                         is_next = 1;
 973                 }
 974                 spin_unlock(&l_mg->free_lock);
 975         }
 976
 977         if (is_next)
 978                 pblk_line_erase(pblk, l_mg->data_next);
 979
 980 out:
 981         if (found_lines != recovered_lines)
 982                 pblk_err(pblk, "failed to recover all found lines %d/%d\n",
 983                                                 found_lines, recovered_lines);
 984
 985         return data_line;
 986 }
 987
 988 /*
 989  * Pad current line
 990  */
 991 int pblk_recov_pad(struct pblk *pblk)
 992 {
 993         struct pblk_line *line;
 994         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
 995         int left_msecs;
 996         int ret = 0;
 997
 998         spin_lock(&l_mg->free_lock);
 999         line = l_mg->data_line;
1000         left_msecs = line->left_msecs;
1001         spin_unlock(&l_mg->free_lock);
1002
1003         ret = pblk_recov_pad_oob(pblk, line, left_msecs);
1004         if (ret) {
1005                 pblk_err(pblk, "tear down padding failed (%d)\n", ret);
1006                 return ret;
1007         }
1008
1009         pblk_line_close_meta(pblk, line);
1010         return ret;
1011 }