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 void pblk_submit_rec(struct work_struct *work)
  20 {
  21         struct pblk_rec_ctx *recovery =
  22                         container_of(work, struct pblk_rec_ctx, ws_rec);
  23         struct pblk *pblk = recovery->pblk;
  24         struct nvm_tgt_dev *dev = pblk->dev;
  25         struct nvm_rq *rqd = recovery->rqd;
  26         struct pblk_c_ctx *c_ctx = nvm_rq_to_pdu(rqd);
  27         int max_secs = nvm_max_phys_sects(dev);
  28         struct bio *bio;
  29         unsigned int nr_rec_secs;
  30         unsigned int pgs_read;
  31         int ret;
  32
  33         nr_rec_secs = bitmap_weight((unsigned long int *)&rqd->ppa_status,
  34                                                                 max_secs);
  35
  36         bio = bio_alloc(GFP_KERNEL, nr_rec_secs);
  37
  38         bio->bi_iter.bi_sector = 0;
  39         bio_set_op_attrs(bio, REQ_OP_WRITE, 0);
  40         rqd->bio = bio;
  41         rqd->nr_ppas = nr_rec_secs;
  42
  43         pgs_read = pblk_rb_read_to_bio_list(&pblk->rwb, bio, &recovery->failed,
  44                                                                 nr_rec_secs);
  45         if (pgs_read != nr_rec_secs) {
  46                 pr_err("pblk: could not read recovery entries\n");
  47                 goto err;
  48         }
  49
  50         if (pblk_setup_w_rec_rq(pblk, rqd, c_ctx)) {
  51                 pr_err("pblk: could not setup recovery request\n");
  52                 goto err;
  53         }
  54
  55 #ifdef CONFIG_NVM_DEBUG
  56         atomic_long_add(nr_rec_secs, &pblk->recov_writes);
  57 #endif
  58
  59         ret = pblk_submit_io(pblk, rqd);
  60         if (ret) {
  61                 pr_err("pblk: I/O submission failed: %d\n", ret);
  62                 goto err;
  63         }
  64
  65         mempool_free(recovery, pblk->rec_pool);
  66         return;
  67
  68 err:
  69         bio_put(bio);
  70         pblk_free_rqd(pblk, rqd, PBLK_WRITE);
  71 }
  72
  73 int pblk_recov_setup_rq(struct pblk *pblk, struct pblk_c_ctx *c_ctx,
  74                         struct pblk_rec_ctx *recovery, u64 *comp_bits,
  75                         unsigned int comp)
  76 {
  77         struct nvm_tgt_dev *dev = pblk->dev;
  78         int max_secs = nvm_max_phys_sects(dev);
  79         struct nvm_rq *rec_rqd;
  80         struct pblk_c_ctx *rec_ctx;
  81         int nr_entries = c_ctx->nr_valid + c_ctx->nr_padded;
  82
  83         rec_rqd = pblk_alloc_rqd(pblk, PBLK_WRITE);
  84         rec_ctx = nvm_rq_to_pdu(rec_rqd);
  85
  86         /* Copy completion bitmap, but exclude the first X completed entries */
  87         bitmap_shift_right((unsigned long int *)&rec_rqd->ppa_status,
  88                                 (unsigned long int *)comp_bits,
  89                                 comp, max_secs);
  90
  91         /* Save the context for the entries that need to be re-written and
  92          * update current context with the completed entries.
  93          */
  94         rec_ctx->sentry = pblk_rb_wrap_pos(&pblk->rwb, c_ctx->sentry + comp);
  95         if (comp >= c_ctx->nr_valid) {
  96                 rec_ctx->nr_valid = 0;
  97                 rec_ctx->nr_padded = nr_entries - comp;
  98
  99                 c_ctx->nr_padded = comp - c_ctx->nr_valid;
 100         } else {
 101                 rec_ctx->nr_valid = c_ctx->nr_valid - comp;
 102                 rec_ctx->nr_padded = c_ctx->nr_padded;
 103
 104                 c_ctx->nr_valid = comp;
 105                 c_ctx->nr_padded = 0;
 106         }
 107
 108         recovery->rqd = rec_rqd;
 109         recovery->pblk = pblk;
 110
 111         return 0;
 112 }
 113
 114 int pblk_recov_check_emeta(struct pblk *pblk, struct line_emeta *emeta_buf)
 115 {
 116         u32 crc;
 117
 118         crc = pblk_calc_emeta_crc(pblk, emeta_buf);
 119         if (le32_to_cpu(emeta_buf->crc) != crc)
 120                 return 1;
 121
 122         if (le32_to_cpu(emeta_buf->header.identifier) != PBLK_MAGIC)
 123                 return 1;
 124
 125         return 0;
 126 }
 127
 128 static int pblk_recov_l2p_from_emeta(struct pblk *pblk, struct pblk_line *line)
 129 {
 130         struct nvm_tgt_dev *dev = pblk->dev;
 131         struct nvm_geo *geo = &dev->geo;
 132         struct pblk_line_meta *lm = &pblk->lm;
 133         struct pblk_emeta *emeta = line->emeta;
 134         struct line_emeta *emeta_buf = emeta->buf;
 135         __le64 *lba_list;
 136         u64 data_start, data_end;
 137         u64 nr_valid_lbas, nr_lbas = 0;
 138         u64 i;
 139
 140         lba_list = emeta_to_lbas(pblk, emeta_buf);
 141         if (!lba_list)
 142                 return 1;
 143
 144         data_start = pblk_line_smeta_start(pblk, line) + lm->smeta_sec;
 145         data_end = line->emeta_ssec;
 146         nr_valid_lbas = le64_to_cpu(emeta_buf->nr_valid_lbas);
 147
 148         for (i = data_start; i < data_end; i++) {
 149                 struct ppa_addr ppa;
 150                 int pos;
 151
 152                 ppa = addr_to_gen_ppa(pblk, i, line->id);
 153                 pos = pblk_ppa_to_pos(geo, ppa);
 154
 155                 /* Do not update bad blocks */
 156                 if (test_bit(pos, line->blk_bitmap))
 157                         continue;
 158
 159                 if (le64_to_cpu(lba_list[i]) == ADDR_EMPTY) {
 160                         spin_lock(&line->lock);
 161                         if (test_and_set_bit(i, line->invalid_bitmap))
 162                                 WARN_ONCE(1, "pblk: rec. double invalidate:\n");
 163                         else
 164                                 le32_add_cpu(line->vsc, -1);
 165                         spin_unlock(&line->lock);
 166
 167                         continue;
 168                 }
 169
 170                 pblk_update_map(pblk, le64_to_cpu(lba_list[i]), ppa);
 171                 nr_lbas++;
 172         }
 173
 174         if (nr_valid_lbas != nr_lbas)
 175                 pr_err("pblk: line %d - inconsistent lba list(%llu/%llu)\n",
 176                                 line->id, nr_valid_lbas, nr_lbas);
 177
 178         line->left_msecs = 0;
 179
 180         return 0;
 181 }
 182
 183 static int pblk_calc_sec_in_line(struct pblk *pblk, struct pblk_line *line)
 184 {
 185         struct nvm_tgt_dev *dev = pblk->dev;
 186         struct nvm_geo *geo = &dev->geo;
 187         struct pblk_line_meta *lm = &pblk->lm;
 188         int nr_bb = bitmap_weight(line->blk_bitmap, lm->blk_per_line);
 189
 190         return lm->sec_per_line - lm->smeta_sec - lm->emeta_sec[0] -
 191                                 nr_bb * geo->sec_per_chk;
 192 }
 193
 194 struct pblk_recov_alloc {
 195         struct ppa_addr *ppa_list;
 196         struct pblk_sec_meta *meta_list;
 197         struct nvm_rq *rqd;
 198         void *data;
 199         dma_addr_t dma_ppa_list;
 200         dma_addr_t dma_meta_list;
 201 };
 202
 203 static int pblk_recov_read_oob(struct pblk *pblk, struct pblk_line *line,
 204                                struct pblk_recov_alloc p, u64 r_ptr)
 205 {
 206         struct nvm_tgt_dev *dev = pblk->dev;
 207         struct nvm_geo *geo = &dev->geo;
 208         struct ppa_addr *ppa_list;
 209         struct pblk_sec_meta *meta_list;
 210         struct nvm_rq *rqd;
 211         struct bio *bio;
 212         void *data;
 213         dma_addr_t dma_ppa_list, dma_meta_list;
 214         u64 r_ptr_int;
 215         int left_ppas;
 216         int rq_ppas, rq_len;
 217         int i, j;
 218         int ret = 0;
 219
 220         ppa_list = p.ppa_list;
 221         meta_list = p.meta_list;
 222         rqd = p.rqd;
 223         data = p.data;
 224         dma_ppa_list = p.dma_ppa_list;
 225         dma_meta_list = p.dma_meta_list;
 226
 227         left_ppas = line->cur_sec - r_ptr;
 228         if (!left_ppas)
 229                 return 0;
 230
 231         r_ptr_int = r_ptr;
 232
 233 next_read_rq:
 234         memset(rqd, 0, pblk_g_rq_size);
 235
 236         rq_ppas = pblk_calc_secs(pblk, left_ppas, 0);
 237         if (!rq_ppas)
 238                 rq_ppas = pblk->min_write_pgs;
 239         rq_len = rq_ppas * geo->sec_size;
 240
 241         bio = bio_map_kern(dev->q, data, rq_len, GFP_KERNEL);
 242         if (IS_ERR(bio))
 243                 return PTR_ERR(bio);
 244
 245         bio->bi_iter.bi_sector = 0; /* internal bio */
 246         bio_set_op_attrs(bio, REQ_OP_READ, 0);
 247
 248         rqd->bio = bio;
 249         rqd->opcode = NVM_OP_PREAD;
 250         rqd->meta_list = meta_list;
 251         rqd->nr_ppas = rq_ppas;
 252         rqd->ppa_list = ppa_list;
 253         rqd->dma_ppa_list = dma_ppa_list;
 254         rqd->dma_meta_list = dma_meta_list;
 255
 256         if (pblk_io_aligned(pblk, rq_ppas))
 257                 rqd->flags = pblk_set_read_mode(pblk, PBLK_READ_SEQUENTIAL);
 258         else
 259                 rqd->flags = pblk_set_read_mode(pblk, PBLK_READ_RANDOM);
 260
 261         for (i = 0; i < rqd->nr_ppas; ) {
 262                 struct ppa_addr ppa;
 263                 int pos;
 264
 265                 ppa = addr_to_gen_ppa(pblk, r_ptr_int, line->id);
 266                 pos = pblk_ppa_to_pos(geo, ppa);
 267
 268                 while (test_bit(pos, line->blk_bitmap)) {
 269                         r_ptr_int += pblk->min_write_pgs;
 270                         ppa = addr_to_gen_ppa(pblk, r_ptr_int, line->id);
 271                         pos = pblk_ppa_to_pos(geo, ppa);
 272                 }
 273
 274                 for (j = 0; j < pblk->min_write_pgs; j++, i++, r_ptr_int++)
 275                         rqd->ppa_list[i] =
 276                                 addr_to_gen_ppa(pblk, r_ptr_int, line->id);
 277         }
 278
 279         /* If read fails, more padding is needed */
 280         ret = pblk_submit_io_sync(pblk, rqd);
 281         if (ret) {
 282                 pr_err("pblk: I/O submission failed: %d\n", ret);
 283                 return ret;
 284         }
 285
 286         atomic_dec(&pblk->inflight_io);
 287
 288         /* At this point, the read should not fail. If it does, it is a problem
 289          * we cannot recover from here. Need FTL log.
 290          */
 291         if (rqd->error && rqd->error != NVM_RSP_WARN_HIGHECC) {
 292                 pr_err("pblk: L2P recovery failed (%d)\n", rqd->error);
 293                 return -EINTR;
 294         }
 295
 296         for (i = 0; i < rqd->nr_ppas; i++) {
 297                 u64 lba = le64_to_cpu(meta_list[i].lba);
 298
 299                 if (lba == ADDR_EMPTY || lba > pblk->rl.nr_secs)
 300                         continue;
 301
 302                 pblk_update_map(pblk, lba, rqd->ppa_list[i]);
 303         }
 304
 305         left_ppas -= rq_ppas;
 306         if (left_ppas > 0)
 307                 goto next_read_rq;
 308
 309         return 0;
 310 }
 311
 312 static void pblk_recov_complete(struct kref *ref)
 313 {
 314         struct pblk_pad_rq *pad_rq = container_of(ref, struct pblk_pad_rq, ref);
 315
 316         complete(&pad_rq->wait);
 317 }
 318
 319 static void pblk_end_io_recov(struct nvm_rq *rqd)
 320 {
 321         struct pblk_pad_rq *pad_rq = rqd->private;
 322         struct pblk *pblk = pad_rq->pblk;
 323
 324         pblk_up_page(pblk, rqd->ppa_list, rqd->nr_ppas);
 325
 326         pblk_free_rqd(pblk, rqd, PBLK_WRITE_INT);
 327
 328         atomic_dec(&pblk->inflight_io);
 329         kref_put(&pad_rq->ref, pblk_recov_complete);
 330 }
 331
 332 static int pblk_recov_pad_oob(struct pblk *pblk, struct pblk_line *line,
 333                               int left_ppas)
 334 {
 335         struct nvm_tgt_dev *dev = pblk->dev;
 336         struct nvm_geo *geo = &dev->geo;
 337         struct ppa_addr *ppa_list;
 338         struct pblk_sec_meta *meta_list;
 339         struct pblk_pad_rq *pad_rq;
 340         struct nvm_rq *rqd;
 341         struct bio *bio;
 342         void *data;
 343         dma_addr_t dma_ppa_list, dma_meta_list;
 344         __le64 *lba_list = emeta_to_lbas(pblk, line->emeta->buf);
 345         u64 w_ptr = line->cur_sec;
 346         int left_line_ppas, rq_ppas, rq_len;
 347         int i, j;
 348         int ret = 0;
 349
 350         spin_lock(&line->lock);
 351         left_line_ppas = line->left_msecs;
 352         spin_unlock(&line->lock);
 353
 354         pad_rq = kmalloc(sizeof(struct pblk_pad_rq), GFP_KERNEL);
 355         if (!pad_rq)
 356                 return -ENOMEM;
 357
 358         data = vzalloc(pblk->max_write_pgs * geo->sec_size);
 359         if (!data) {
 360                 ret = -ENOMEM;
 361                 goto free_rq;
 362         }
 363
 364         pad_rq->pblk = pblk;
 365         init_completion(&pad_rq->wait);
 366         kref_init(&pad_rq->ref);
 367
 368 next_pad_rq:
 369         rq_ppas = pblk_calc_secs(pblk, left_ppas, 0);
 370         if (rq_ppas < pblk->min_write_pgs) {
 371                 pr_err("pblk: corrupted pad line %d\n", line->id);
 372                 goto fail_free_pad;
 373         }
 374
 375         rq_len = rq_ppas * geo->sec_size;
 376
 377         meta_list = nvm_dev_dma_alloc(dev->parent, GFP_KERNEL, &dma_meta_list);
 378         if (!meta_list) {
 379                 ret = -ENOMEM;
 380                 goto fail_free_pad;
 381         }
 382
 383         ppa_list = (void *)(meta_list) + pblk_dma_meta_size;
 384         dma_ppa_list = dma_meta_list + pblk_dma_meta_size;
 385
 386         bio = pblk_bio_map_addr(pblk, data, rq_ppas, rq_len,
 387                                                 PBLK_VMALLOC_META, GFP_KERNEL);
 388         if (IS_ERR(bio)) {
 389                 ret = PTR_ERR(bio);
 390                 goto fail_free_meta;
 391         }
 392
 393         bio->bi_iter.bi_sector = 0; /* internal bio */
 394         bio_set_op_attrs(bio, REQ_OP_WRITE, 0);
 395
 396         rqd = pblk_alloc_rqd(pblk, PBLK_WRITE_INT);
 397
 398         rqd->bio = bio;
 399         rqd->opcode = NVM_OP_PWRITE;
 400         rqd->flags = pblk_set_progr_mode(pblk, PBLK_WRITE);
 401         rqd->meta_list = meta_list;
 402         rqd->nr_ppas = rq_ppas;
 403         rqd->ppa_list = ppa_list;
 404         rqd->dma_ppa_list = dma_ppa_list;
 405         rqd->dma_meta_list = dma_meta_list;
 406         rqd->end_io = pblk_end_io_recov;
 407         rqd->private = pad_rq;
 408
 409         for (i = 0; i < rqd->nr_ppas; ) {
 410                 struct ppa_addr ppa;
 411                 int pos;
 412
 413                 w_ptr = pblk_alloc_page(pblk, line, pblk->min_write_pgs);
 414                 ppa = addr_to_gen_ppa(pblk, w_ptr, line->id);
 415                 pos = pblk_ppa_to_pos(geo, ppa);
 416
 417                 while (test_bit(pos, line->blk_bitmap)) {
 418                         w_ptr += pblk->min_write_pgs;
 419                         ppa = addr_to_gen_ppa(pblk, w_ptr, line->id);
 420                         pos = pblk_ppa_to_pos(geo, ppa);
 421                 }
 422
 423                 for (j = 0; j < pblk->min_write_pgs; j++, i++, w_ptr++) {
 424                         struct ppa_addr dev_ppa;
 425                         __le64 addr_empty = cpu_to_le64(ADDR_EMPTY);
 426
 427                         dev_ppa = addr_to_gen_ppa(pblk, w_ptr, line->id);
 428
 429                         pblk_map_invalidate(pblk, dev_ppa);
 430                         lba_list[w_ptr] = meta_list[i].lba = addr_empty;
 431                         rqd->ppa_list[i] = dev_ppa;
 432                 }
 433         }
 434
 435         kref_get(&pad_rq->ref);
 436         pblk_down_page(pblk, rqd->ppa_list, rqd->nr_ppas);
 437
 438         ret = pblk_submit_io(pblk, rqd);
 439         if (ret) {
 440                 pr_err("pblk: I/O submission failed: %d\n", ret);
 441                 pblk_up_page(pblk, rqd->ppa_list, rqd->nr_ppas);
 442                 goto fail_free_bio;
 443         }
 444
 445         left_line_ppas -= rq_ppas;
 446         left_ppas -= rq_ppas;
 447         if (left_ppas && left_line_ppas)
 448                 goto next_pad_rq;
 449
 450         kref_put(&pad_rq->ref, pblk_recov_complete);
 451
 452         if (!wait_for_completion_io_timeout(&pad_rq->wait,
 453                                 msecs_to_jiffies(PBLK_COMMAND_TIMEOUT_MS))) {
 454                 pr_err("pblk: pad write timed out\n");
 455                 ret = -ETIME;
 456         }
 457
 458         if (!pblk_line_is_full(line))
 459                 pr_err("pblk: corrupted padded line: %d\n", line->id);
 460
 461         vfree(data);
 462 free_rq:
 463         kfree(pad_rq);
 464         return ret;
 465
 466 fail_free_bio:
 467         bio_put(bio);
 468 fail_free_meta:
 469         nvm_dev_dma_free(dev->parent, meta_list, dma_meta_list);
 470 fail_free_pad:
 471         kfree(pad_rq);
 472         vfree(data);
 473         return ret;
 474 }
 475
 476 /* When this function is called, it means that not all upper pages have been
 477  * written in a page that contains valid data. In order to recover this data, we
 478  * first find the write pointer on the device, then we pad all necessary
 479  * sectors, and finally attempt to read the valid data
 480  */
 481 static int pblk_recov_scan_all_oob(struct pblk *pblk, struct pblk_line *line,
 482                                    struct pblk_recov_alloc p)
 483 {
 484         struct nvm_tgt_dev *dev = pblk->dev;
 485         struct nvm_geo *geo = &dev->geo;
 486         struct ppa_addr *ppa_list;
 487         struct pblk_sec_meta *meta_list;
 488         struct nvm_rq *rqd;
 489         struct bio *bio;
 490         void *data;
 491         dma_addr_t dma_ppa_list, dma_meta_list;
 492         u64 w_ptr = 0, r_ptr;
 493         int rq_ppas, rq_len;
 494         int i, j;
 495         int ret = 0;
 496         int rec_round;
 497         int left_ppas = pblk_calc_sec_in_line(pblk, line) - line->cur_sec;
 498
 499         ppa_list = p.ppa_list;
 500         meta_list = p.meta_list;
 501         rqd = p.rqd;
 502         data = p.data;
 503         dma_ppa_list = p.dma_ppa_list;
 504         dma_meta_list = p.dma_meta_list;
 505
 506         /* we could recover up until the line write pointer */
 507         r_ptr = line->cur_sec;
 508         rec_round = 0;
 509
 510 next_rq:
 511         memset(rqd, 0, pblk_g_rq_size);
 512
 513         rq_ppas = pblk_calc_secs(pblk, left_ppas, 0);
 514         if (!rq_ppas)
 515                 rq_ppas = pblk->min_write_pgs;
 516         rq_len = rq_ppas * geo->sec_size;
 517
 518         bio = bio_map_kern(dev->q, data, rq_len, GFP_KERNEL);
 519         if (IS_ERR(bio))
 520                 return PTR_ERR(bio);
 521
 522         bio->bi_iter.bi_sector = 0; /* internal bio */
 523         bio_set_op_attrs(bio, REQ_OP_READ, 0);
 524
 525         rqd->bio = bio;
 526         rqd->opcode = NVM_OP_PREAD;
 527         rqd->meta_list = meta_list;
 528         rqd->nr_ppas = rq_ppas;
 529         rqd->ppa_list = ppa_list;
 530         rqd->dma_ppa_list = dma_ppa_list;
 531         rqd->dma_meta_list = dma_meta_list;
 532
 533         if (pblk_io_aligned(pblk, rq_ppas))
 534                 rqd->flags = pblk_set_read_mode(pblk, PBLK_READ_SEQUENTIAL);
 535         else
 536                 rqd->flags = pblk_set_read_mode(pblk, PBLK_READ_RANDOM);
 537
 538         for (i = 0; i < rqd->nr_ppas; ) {
 539                 struct ppa_addr ppa;
 540                 int pos;
 541
 542                 w_ptr = pblk_alloc_page(pblk, line, pblk->min_write_pgs);
 543                 ppa = addr_to_gen_ppa(pblk, w_ptr, line->id);
 544                 pos = pblk_ppa_to_pos(geo, ppa);
 545
 546                 while (test_bit(pos, line->blk_bitmap)) {
 547                         w_ptr += pblk->min_write_pgs;
 548                         ppa = addr_to_gen_ppa(pblk, w_ptr, line->id);
 549                         pos = pblk_ppa_to_pos(geo, ppa);
 550                 }
 551
 552                 for (j = 0; j < pblk->min_write_pgs; j++, i++, w_ptr++)
 553                         rqd->ppa_list[i] =
 554                                 addr_to_gen_ppa(pblk, w_ptr, line->id);
 555         }
 556
 557         ret = pblk_submit_io_sync(pblk, rqd);
 558         if (ret) {
 559                 pr_err("pblk: I/O submission failed: %d\n", ret);
 560                 return ret;
 561         }
 562
 563         atomic_dec(&pblk->inflight_io);
 564
 565         /* This should not happen since the read failed during normal recovery,
 566          * but the media works funny sometimes...
 567          */
 568         if (!rec_round++ && !rqd->error) {
 569                 rec_round = 0;
 570                 for (i = 0; i < rqd->nr_ppas; i++, r_ptr++) {
 571                         u64 lba = le64_to_cpu(meta_list[i].lba);
 572
 573                         if (lba == ADDR_EMPTY || lba > pblk->rl.nr_secs)
 574                                 continue;
 575
 576                         pblk_update_map(pblk, lba, rqd->ppa_list[i]);
 577                 }
 578         }
 579
 580         /* Reached the end of the written line */
 581         if (rqd->error == NVM_RSP_ERR_EMPTYPAGE) {
 582                 int pad_secs, nr_error_bits, bit;
 583                 int ret;
 584
 585                 bit = find_first_bit((void *)&rqd->ppa_status, rqd->nr_ppas);
 586                 nr_error_bits = rqd->nr_ppas - bit;
 587
 588                 /* Roll back failed sectors */
 589                 line->cur_sec -= nr_error_bits;
 590                 line->left_msecs += nr_error_bits;
 591                 bitmap_clear(line->map_bitmap, line->cur_sec, nr_error_bits);
 592
 593                 pad_secs = pblk_pad_distance(pblk);
 594                 if (pad_secs > line->left_msecs)
 595                         pad_secs = line->left_msecs;
 596
 597                 ret = pblk_recov_pad_oob(pblk, line, pad_secs);
 598                 if (ret)
 599                         pr_err("pblk: OOB padding failed (err:%d)\n", ret);
 600
 601                 ret = pblk_recov_read_oob(pblk, line, p, r_ptr);
 602                 if (ret)
 603                         pr_err("pblk: OOB read failed (err:%d)\n", ret);
 604
 605                 left_ppas = 0;
 606         }
 607
 608         left_ppas -= rq_ppas;
 609         if (left_ppas > 0)
 610                 goto next_rq;
 611
 612         return ret;
 613 }
 614
 615 static int pblk_recov_scan_oob(struct pblk *pblk, struct pblk_line *line,
 616                                struct pblk_recov_alloc p, int *done)
 617 {
 618         struct nvm_tgt_dev *dev = pblk->dev;
 619         struct nvm_geo *geo = &dev->geo;
 620         struct ppa_addr *ppa_list;
 621         struct pblk_sec_meta *meta_list;
 622         struct nvm_rq *rqd;
 623         struct bio *bio;
 624         void *data;
 625         dma_addr_t dma_ppa_list, dma_meta_list;
 626         u64 paddr;
 627         int rq_ppas, rq_len;
 628         int i, j;
 629         int ret = 0;
 630         int left_ppas = pblk_calc_sec_in_line(pblk, line);
 631
 632         ppa_list = p.ppa_list;
 633         meta_list = p.meta_list;
 634         rqd = p.rqd;
 635         data = p.data;
 636         dma_ppa_list = p.dma_ppa_list;
 637         dma_meta_list = p.dma_meta_list;
 638
 639         *done = 1;
 640
 641 next_rq:
 642         memset(rqd, 0, pblk_g_rq_size);
 643
 644         rq_ppas = pblk_calc_secs(pblk, left_ppas, 0);
 645         if (!rq_ppas)
 646                 rq_ppas = pblk->min_write_pgs;
 647         rq_len = rq_ppas * geo->sec_size;
 648
 649         bio = bio_map_kern(dev->q, data, rq_len, GFP_KERNEL);
 650         if (IS_ERR(bio))
 651                 return PTR_ERR(bio);
 652
 653         bio->bi_iter.bi_sector = 0; /* internal bio */
 654         bio_set_op_attrs(bio, REQ_OP_READ, 0);
 655
 656         rqd->bio = bio;
 657         rqd->opcode = NVM_OP_PREAD;
 658         rqd->meta_list = meta_list;
 659         rqd->nr_ppas = rq_ppas;
 660         rqd->ppa_list = ppa_list;
 661         rqd->dma_ppa_list = dma_ppa_list;
 662         rqd->dma_meta_list = dma_meta_list;
 663
 664         if (pblk_io_aligned(pblk, rq_ppas))
 665                 rqd->flags = pblk_set_read_mode(pblk, PBLK_READ_SEQUENTIAL);
 666         else
 667                 rqd->flags = pblk_set_read_mode(pblk, PBLK_READ_RANDOM);
 668
 669         for (i = 0; i < rqd->nr_ppas; ) {
 670                 struct ppa_addr ppa;
 671                 int pos;
 672
 673                 paddr = pblk_alloc_page(pblk, line, pblk->min_write_pgs);
 674                 ppa = addr_to_gen_ppa(pblk, paddr, line->id);
 675                 pos = pblk_ppa_to_pos(geo, ppa);
 676
 677                 while (test_bit(pos, line->blk_bitmap)) {
 678                         paddr += pblk->min_write_pgs;
 679                         ppa = addr_to_gen_ppa(pblk, paddr, line->id);
 680                         pos = pblk_ppa_to_pos(geo, ppa);
 681                 }
 682
 683                 for (j = 0; j < pblk->min_write_pgs; j++, i++, paddr++)
 684                         rqd->ppa_list[i] =
 685                                 addr_to_gen_ppa(pblk, paddr, line->id);
 686         }
 687
 688         ret = pblk_submit_io_sync(pblk, rqd);
 689         if (ret) {
 690                 pr_err("pblk: I/O submission failed: %d\n", ret);
 691                 bio_put(bio);
 692                 return ret;
 693         }
 694
 695         atomic_dec(&pblk->inflight_io);
 696
 697         /* Reached the end of the written line */
 698         if (rqd->error) {
 699                 int nr_error_bits, bit;
 700
 701                 bit = find_first_bit((void *)&rqd->ppa_status, rqd->nr_ppas);
 702                 nr_error_bits = rqd->nr_ppas - bit;
 703
 704                 /* Roll back failed sectors */
 705                 line->cur_sec -= nr_error_bits;
 706                 line->left_msecs += nr_error_bits;
 707                 bitmap_clear(line->map_bitmap, line->cur_sec, nr_error_bits);
 708
 709                 left_ppas = 0;
 710                 rqd->nr_ppas = bit;
 711
 712                 if (rqd->error != NVM_RSP_ERR_EMPTYPAGE)
 713                         *done = 0;
 714         }
 715
 716         for (i = 0; i < rqd->nr_ppas; i++) {
 717                 u64 lba = le64_to_cpu(meta_list[i].lba);
 718
 719                 if (lba == ADDR_EMPTY || lba > pblk->rl.nr_secs)
 720                         continue;
 721
 722                 pblk_update_map(pblk, lba, rqd->ppa_list[i]);
 723         }
 724
 725         left_ppas -= rq_ppas;
 726         if (left_ppas > 0)
 727                 goto next_rq;
 728
 729         return ret;
 730 }
 731
 732 /* Scan line for lbas on out of bound area */
 733 static int pblk_recov_l2p_from_oob(struct pblk *pblk, struct pblk_line *line)
 734 {
 735         struct nvm_tgt_dev *dev = pblk->dev;
 736         struct nvm_geo *geo = &dev->geo;
 737         struct nvm_rq *rqd;
 738         struct ppa_addr *ppa_list;
 739         struct pblk_sec_meta *meta_list;
 740         struct pblk_recov_alloc p;
 741         void *data;
 742         dma_addr_t dma_ppa_list, dma_meta_list;
 743         int done, ret = 0;
 744
 745         meta_list = nvm_dev_dma_alloc(dev->parent, GFP_KERNEL, &dma_meta_list);
 746         if (!meta_list)
 747                 return -ENOMEM;
 748
 749         ppa_list = (void *)(meta_list) + pblk_dma_meta_size;
 750         dma_ppa_list = dma_meta_list + pblk_dma_meta_size;
 751
 752         data = kcalloc(pblk->max_write_pgs, geo->sec_size, GFP_KERNEL);
 753         if (!data) {
 754                 ret = -ENOMEM;
 755                 goto free_meta_list;
 756         }
 757
 758         rqd = pblk_alloc_rqd(pblk, PBLK_READ);
 759
 760         p.ppa_list = ppa_list;
 761         p.meta_list = meta_list;
 762         p.rqd = rqd;
 763         p.data = data;
 764         p.dma_ppa_list = dma_ppa_list;
 765         p.dma_meta_list = dma_meta_list;
 766
 767         ret = pblk_recov_scan_oob(pblk, line, p, &done);
 768         if (ret) {
 769                 pr_err("pblk: could not recover L2P from OOB\n");
 770                 goto out;
 771         }
 772
 773         if (!done) {
 774                 ret = pblk_recov_scan_all_oob(pblk, line, p);
 775                 if (ret) {
 776                         pr_err("pblk: could not recover L2P from OOB\n");
 777                         goto out;
 778                 }
 779         }
 780
 781         if (pblk_line_is_full(line))
 782                 pblk_line_recov_close(pblk, line);
 783
 784 out:
 785         kfree(data);
 786 free_meta_list:
 787         nvm_dev_dma_free(dev->parent, meta_list, dma_meta_list);
 788
 789         return ret;
 790 }
 791
 792 /* Insert lines ordered by sequence number (seq_num) on list */
 793 static void pblk_recov_line_add_ordered(struct list_head *head,
 794                                         struct pblk_line *line)
 795 {
 796         struct pblk_line *t = NULL;
 797
 798         list_for_each_entry(t, head, list)
 799                 if (t->seq_nr > line->seq_nr)
 800                         break;
 801
 802         __list_add(&line->list, t->list.prev, &t->list);
 803 }
 804
 805 static u64 pblk_line_emeta_start(struct pblk *pblk, struct pblk_line *line)
 806 {
 807         struct nvm_tgt_dev *dev = pblk->dev;
 808         struct nvm_geo *geo = &dev->geo;
 809         struct pblk_line_meta *lm = &pblk->lm;
 810         unsigned int emeta_secs;
 811         u64 emeta_start;
 812         struct ppa_addr ppa;
 813         int pos;
 814
 815         emeta_secs = lm->emeta_sec[0];
 816         emeta_start = lm->sec_per_line;
 817
 818         while (emeta_secs) {
 819                 emeta_start--;
 820                 ppa = addr_to_gen_ppa(pblk, emeta_start, line->id);
 821                 pos = pblk_ppa_to_pos(geo, ppa);
 822                 if (!test_bit(pos, line->blk_bitmap))
 823                         emeta_secs--;
 824         }
 825
 826         return emeta_start;
 827 }
 828
 829 struct pblk_line *pblk_recov_l2p(struct pblk *pblk)
 830 {
 831         struct pblk_line_meta *lm = &pblk->lm;
 832         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
 833         struct pblk_line *line, *tline, *data_line = NULL;
 834         struct pblk_smeta *smeta;
 835         struct pblk_emeta *emeta;
 836         struct line_smeta *smeta_buf;
 837         int found_lines = 0, recovered_lines = 0, open_lines = 0;
 838         int is_next = 0;
 839         int meta_line;
 840         int i, valid_uuid = 0;
 841         LIST_HEAD(recov_list);
 842
 843         /* TODO: Implement FTL snapshot */
 844
 845         /* Scan recovery - takes place when FTL snapshot fails */
 846         spin_lock(&l_mg->free_lock);
 847         meta_line = find_first_zero_bit(&l_mg->meta_bitmap, PBLK_DATA_LINES);
 848         set_bit(meta_line, &l_mg->meta_bitmap);
 849         smeta = l_mg->sline_meta[meta_line];
 850         emeta = l_mg->eline_meta[meta_line];
 851         smeta_buf = (struct line_smeta *)smeta;
 852         spin_unlock(&l_mg->free_lock);
 853
 854         /* Order data lines using their sequence number */
 855         for (i = 0; i < l_mg->nr_lines; i++) {
 856                 u32 crc;
 857
 858                 line = &pblk->lines[i];
 859
 860                 memset(smeta, 0, lm->smeta_len);
 861                 line->smeta = smeta;
 862                 line->lun_bitmap = ((void *)(smeta_buf)) +
 863                                                 sizeof(struct line_smeta);
 864
 865                 /* Lines that cannot be read are assumed as not written here */
 866                 if (pblk_line_read_smeta(pblk, line))
 867                         continue;
 868
 869                 crc = pblk_calc_smeta_crc(pblk, smeta_buf);
 870                 if (le32_to_cpu(smeta_buf->crc) != crc)
 871                         continue;
 872
 873                 if (le32_to_cpu(smeta_buf->header.identifier) != PBLK_MAGIC)
 874                         continue;
 875
 876                 if (smeta_buf->header.version != SMETA_VERSION) {
 877                         pr_err("pblk: found incompatible line version %u\n",
 878                                         le16_to_cpu(smeta_buf->header.version));
 879                         return ERR_PTR(-EINVAL);
 880                 }
 881
 882                 /* The first valid instance uuid is used for initialization */
 883                 if (!valid_uuid) {
 884                         memcpy(pblk->instance_uuid, smeta_buf->header.uuid, 16);
 885                         valid_uuid = 1;
 886                 }
 887
 888                 if (memcmp(pblk->instance_uuid, smeta_buf->header.uuid, 16)) {
 889                         pr_debug("pblk: ignore line %u due to uuid mismatch\n",
 890                                         i);
 891                         continue;
 892                 }
 893
 894                 /* Update line metadata */
 895                 spin_lock(&line->lock);
 896                 line->id = le32_to_cpu(smeta_buf->header.id);
 897                 line->type = le16_to_cpu(smeta_buf->header.type);
 898                 line->seq_nr = le64_to_cpu(smeta_buf->seq_nr);
 899                 spin_unlock(&line->lock);
 900
 901                 /* Update general metadata */
 902                 spin_lock(&l_mg->free_lock);
 903                 if (line->seq_nr >= l_mg->d_seq_nr)
 904                         l_mg->d_seq_nr = line->seq_nr + 1;
 905                 l_mg->nr_free_lines--;
 906                 spin_unlock(&l_mg->free_lock);
 907
 908                 if (pblk_line_recov_alloc(pblk, line))
 909                         goto out;
 910
 911                 pblk_recov_line_add_ordered(&recov_list, line);
 912                 found_lines++;
 913                 pr_debug("pblk: recovering data line %d, seq:%llu\n",
 914                                                 line->id, smeta_buf->seq_nr);
 915         }
 916
 917         if (!found_lines) {
 918                 pblk_setup_uuid(pblk);
 919
 920                 spin_lock(&l_mg->free_lock);
 921                 WARN_ON_ONCE(!test_and_clear_bit(meta_line,
 922                                                         &l_mg->meta_bitmap));
 923                 spin_unlock(&l_mg->free_lock);
 924
 925                 goto out;
 926         }
 927
 928         /* Verify closed blocks and recover this portion of L2P table*/
 929         list_for_each_entry_safe(line, tline, &recov_list, list) {
 930                 recovered_lines++;
 931
 932                 line->emeta_ssec = pblk_line_emeta_start(pblk, line);
 933                 line->emeta = emeta;
 934                 memset(line->emeta->buf, 0, lm->emeta_len[0]);
 935
 936                 if (pblk_line_read_emeta(pblk, line, line->emeta->buf)) {
 937                         pblk_recov_l2p_from_oob(pblk, line);
 938                         goto next;
 939                 }
 940
 941                 if (pblk_recov_check_emeta(pblk, line->emeta->buf)) {
 942                         pblk_recov_l2p_from_oob(pblk, line);
 943                         goto next;
 944                 }
 945
 946                 if (pblk_recov_l2p_from_emeta(pblk, line))
 947                         pblk_recov_l2p_from_oob(pblk, line);
 948
 949 next:
 950                 if (pblk_line_is_full(line)) {
 951                         struct list_head *move_list;
 952
 953                         spin_lock(&line->lock);
 954                         line->state = PBLK_LINESTATE_CLOSED;
 955                         move_list = pblk_line_gc_list(pblk, line);
 956                         spin_unlock(&line->lock);
 957
 958                         spin_lock(&l_mg->gc_lock);
 959                         list_move_tail(&line->list, move_list);
 960                         spin_unlock(&l_mg->gc_lock);
 961
 962                         kfree(line->map_bitmap);
 963                         line->map_bitmap = NULL;
 964                         line->smeta = NULL;
 965                         line->emeta = NULL;
 966                 } else {
 967                         if (open_lines > 1)
 968                                 pr_err("pblk: failed to recover L2P\n");
 969
 970                         open_lines++;
 971                         line->meta_line = meta_line;
 972                         data_line = line;
 973                 }
 974         }
 975
 976         spin_lock(&l_mg->free_lock);
 977         if (!open_lines) {
 978                 WARN_ON_ONCE(!test_and_clear_bit(meta_line,
 979                                                         &l_mg->meta_bitmap));
 980                 pblk_line_replace_data(pblk);
 981         } else {
 982                 /* Allocate next line for preparation */
 983                 l_mg->data_next = pblk_line_get(pblk);
 984                 if (l_mg->data_next) {
 985                         l_mg->data_next->seq_nr = l_mg->d_seq_nr++;
 986                         l_mg->data_next->type = PBLK_LINETYPE_DATA;
 987                         is_next = 1;
 988                 }
 989         }
 990         spin_unlock(&l_mg->free_lock);
 991
 992         if (is_next)
 993                 pblk_line_erase(pblk, l_mg->data_next);
 994
 995 out:
 996         if (found_lines != recovered_lines)
 997                 pr_err("pblk: failed to recover all found lines %d/%d\n",
 998                                                 found_lines, recovered_lines);
 999
1000         return data_line;
1001 }
1002
1003 /*
1004  * Pad current line
1005  */
1006 int pblk_recov_pad(struct pblk *pblk)
1007 {
1008         struct pblk_line *line;
1009         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1010         int left_msecs;
1011         int ret = 0;
1012
1013         spin_lock(&l_mg->free_lock);
1014         line = l_mg->data_line;
1015         left_msecs = line->left_msecs;
1016         spin_unlock(&l_mg->free_lock);
1017
1018         ret = pblk_recov_pad_oob(pblk, line, left_msecs);
1019         if (ret) {
1020                 pr_err("pblk: Tear down padding failed (%d)\n", ret);
1021                 return ret;
1022         }
1023
1024         pblk_line_close_meta(pblk, line);
1025         return ret;
1026 }