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         if (!bio) {
  38                 pr_err("pblk: not able to create recovery bio\n");
  39                 return;
  40         }
  41
  42         bio->bi_iter.bi_sector = 0;
  43         bio_set_op_attrs(bio, REQ_OP_WRITE, 0);
  44         rqd->bio = bio;
  45         rqd->nr_ppas = nr_rec_secs;
  46
  47         pgs_read = pblk_rb_read_to_bio_list(&pblk->rwb, bio, &recovery->failed,
  48                                                                 nr_rec_secs);
  49         if (pgs_read != nr_rec_secs) {
  50                 pr_err("pblk: could not read recovery entries\n");
  51                 goto err;
  52         }
  53
  54         if (pblk_setup_w_rec_rq(pblk, rqd, c_ctx)) {
  55                 pr_err("pblk: could not setup recovery request\n");
  56                 goto err;
  57         }
  58
  59 #ifdef CONFIG_NVM_DEBUG
  60         atomic_long_add(nr_rec_secs, &pblk->recov_writes);
  61 #endif
  62
  63         ret = pblk_submit_io(pblk, rqd);
  64         if (ret) {
  65                 pr_err("pblk: I/O submission failed: %d\n", ret);
  66                 goto err;
  67         }
  68
  69         mempool_free(recovery, pblk->rec_pool);
  70         return;
  71
  72 err:
  73         bio_put(bio);
  74         pblk_free_rqd(pblk, rqd, WRITE);
  75 }
  76
  77 int pblk_recov_setup_rq(struct pblk *pblk, struct pblk_c_ctx *c_ctx,
  78                         struct pblk_rec_ctx *recovery, u64 *comp_bits,
  79                         unsigned int comp)
  80 {
  81         struct nvm_tgt_dev *dev = pblk->dev;
  82         int max_secs = nvm_max_phys_sects(dev);
  83         struct nvm_rq *rec_rqd;
  84         struct pblk_c_ctx *rec_ctx;
  85         int nr_entries = c_ctx->nr_valid + c_ctx->nr_padded;
  86
  87         rec_rqd = pblk_alloc_rqd(pblk, WRITE);
  88         if (IS_ERR(rec_rqd)) {
  89                 pr_err("pblk: could not create recovery req.\n");
  90                 return -ENOMEM;
  91         }
  92
  93         rec_ctx = nvm_rq_to_pdu(rec_rqd);
  94
  95         /* Copy completion bitmap, but exclude the first X completed entries */
  96         bitmap_shift_right((unsigned long int *)&rec_rqd->ppa_status,
  97                                 (unsigned long int *)comp_bits,
  98                                 comp, max_secs);
  99
 100         /* Save the context for the entries that need to be re-written and
 101          * update current context with the completed entries.
 102          */
 103         rec_ctx->sentry = pblk_rb_wrap_pos(&pblk->rwb, c_ctx->sentry + comp);
 104         if (comp >= c_ctx->nr_valid) {
 105                 rec_ctx->nr_valid = 0;
 106                 rec_ctx->nr_padded = nr_entries - comp;
 107
 108                 c_ctx->nr_padded = comp - c_ctx->nr_valid;
 109         } else {
 110                 rec_ctx->nr_valid = c_ctx->nr_valid - comp;
 111                 rec_ctx->nr_padded = c_ctx->nr_padded;
 112
 113                 c_ctx->nr_valid = comp;
 114                 c_ctx->nr_padded = 0;
 115         }
 116
 117         recovery->rqd = rec_rqd;
 118         recovery->pblk = pblk;
 119
 120         return 0;
 121 }
 122
 123 __le64 *pblk_recov_get_lba_list(struct pblk *pblk, struct line_emeta *emeta_buf)
 124 {
 125         u32 crc;
 126
 127         crc = pblk_calc_emeta_crc(pblk, emeta_buf);
 128         if (le32_to_cpu(emeta_buf->crc) != crc)
 129                 return NULL;
 130
 131         if (le32_to_cpu(emeta_buf->header.identifier) != PBLK_MAGIC)
 132                 return NULL;
 133
 134         return emeta_to_lbas(pblk, emeta_buf);
 135 }
 136
 137 static int pblk_recov_l2p_from_emeta(struct pblk *pblk, struct pblk_line *line)
 138 {
 139         struct nvm_tgt_dev *dev = pblk->dev;
 140         struct nvm_geo *geo = &dev->geo;
 141         struct pblk_line_meta *lm = &pblk->lm;
 142         struct pblk_emeta *emeta = line->emeta;
 143         struct line_emeta *emeta_buf = emeta->buf;
 144         __le64 *lba_list;
 145         int data_start;
 146         int nr_data_lbas, nr_valid_lbas, nr_lbas = 0;
 147         int i;
 148
 149         lba_list = pblk_recov_get_lba_list(pblk, emeta_buf);
 150         if (!lba_list)
 151                 return 1;
 152
 153         data_start = pblk_line_smeta_start(pblk, line) + lm->smeta_sec;
 154         nr_data_lbas = lm->sec_per_line - lm->emeta_sec[0];
 155         nr_valid_lbas = le64_to_cpu(emeta_buf->nr_valid_lbas);
 156
 157         for (i = data_start; i < nr_data_lbas && nr_lbas < nr_valid_lbas; i++) {
 158                 struct ppa_addr ppa;
 159                 int pos;
 160
 161                 ppa = addr_to_pblk_ppa(pblk, i, line->id);
 162                 pos = pblk_ppa_to_pos(geo, ppa);
 163
 164                 /* Do not update bad blocks */
 165                 if (test_bit(pos, line->blk_bitmap))
 166                         continue;
 167
 168                 if (le64_to_cpu(lba_list[i]) == ADDR_EMPTY) {
 169                         spin_lock(&line->lock);
 170                         if (test_and_set_bit(i, line->invalid_bitmap))
 171                                 WARN_ONCE(1, "pblk: rec. double invalidate:\n");
 172                         else
 173                                 le32_add_cpu(line->vsc, -1);
 174                         spin_unlock(&line->lock);
 175
 176                         continue;
 177                 }
 178
 179                 pblk_update_map(pblk, le64_to_cpu(lba_list[i]), ppa);
 180                 nr_lbas++;
 181         }
 182
 183         if (nr_valid_lbas != nr_lbas)
 184                 pr_err("pblk: line %d - inconsistent lba list(%llu/%d)\n",
 185                                 line->id, emeta_buf->nr_valid_lbas, nr_lbas);
 186
 187         line->left_msecs = 0;
 188
 189         return 0;
 190 }
 191
 192 static int pblk_calc_sec_in_line(struct pblk *pblk, struct pblk_line *line)
 193 {
 194         struct nvm_tgt_dev *dev = pblk->dev;
 195         struct nvm_geo *geo = &dev->geo;
 196         struct pblk_line_meta *lm = &pblk->lm;
 197         int nr_bb = bitmap_weight(line->blk_bitmap, lm->blk_per_line);
 198
 199         return lm->sec_per_line - lm->smeta_sec - lm->emeta_sec[0] -
 200                                 nr_bb * geo->sec_per_blk;
 201 }
 202
 203 struct pblk_recov_alloc {
 204         struct ppa_addr *ppa_list;
 205         struct pblk_sec_meta *meta_list;
 206         struct nvm_rq *rqd;
 207         void *data;
 208         dma_addr_t dma_ppa_list;
 209         dma_addr_t dma_meta_list;
 210 };
 211
 212 static int pblk_recov_read_oob(struct pblk *pblk, struct pblk_line *line,
 213                                struct pblk_recov_alloc p, u64 r_ptr)
 214 {
 215         struct nvm_tgt_dev *dev = pblk->dev;
 216         struct nvm_geo *geo = &dev->geo;
 217         struct ppa_addr *ppa_list;
 218         struct pblk_sec_meta *meta_list;
 219         struct nvm_rq *rqd;
 220         struct bio *bio;
 221         void *data;
 222         dma_addr_t dma_ppa_list, dma_meta_list;
 223         u64 r_ptr_int;
 224         int left_ppas;
 225         int rq_ppas, rq_len;
 226         int i, j;
 227         int ret = 0;
 228         DECLARE_COMPLETION_ONSTACK(wait);
 229
 230         ppa_list = p.ppa_list;
 231         meta_list = p.meta_list;
 232         rqd = p.rqd;
 233         data = p.data;
 234         dma_ppa_list = p.dma_ppa_list;
 235         dma_meta_list = p.dma_meta_list;
 236
 237         left_ppas = line->cur_sec - r_ptr;
 238         if (!left_ppas)
 239                 return 0;
 240
 241         r_ptr_int = r_ptr;
 242
 243 next_read_rq:
 244         memset(rqd, 0, pblk_g_rq_size);
 245
 246         rq_ppas = pblk_calc_secs(pblk, left_ppas, 0);
 247         if (!rq_ppas)
 248                 rq_ppas = pblk->min_write_pgs;
 249         rq_len = rq_ppas * geo->sec_size;
 250
 251         bio = bio_map_kern(dev->q, data, rq_len, GFP_KERNEL);
 252         if (IS_ERR(bio))
 253                 return PTR_ERR(bio);
 254
 255         bio->bi_iter.bi_sector = 0; /* internal bio */
 256         bio_set_op_attrs(bio, REQ_OP_READ, 0);
 257
 258         rqd->bio = bio;
 259         rqd->opcode = NVM_OP_PREAD;
 260         rqd->meta_list = meta_list;
 261         rqd->nr_ppas = rq_ppas;
 262         rqd->ppa_list = ppa_list;
 263         rqd->dma_ppa_list = dma_ppa_list;
 264         rqd->dma_meta_list = dma_meta_list;
 265         rqd->end_io = pblk_end_io_sync;
 266         rqd->private = &wait;
 267
 268         if (pblk_io_aligned(pblk, rq_ppas))
 269                 rqd->flags = pblk_set_read_mode(pblk, PBLK_READ_SEQUENTIAL);
 270         else
 271                 rqd->flags = pblk_set_read_mode(pblk, PBLK_READ_RANDOM);
 272
 273         for (i = 0; i < rqd->nr_ppas; ) {
 274                 struct ppa_addr ppa;
 275                 int pos;
 276
 277                 ppa = addr_to_gen_ppa(pblk, r_ptr_int, line->id);
 278                 pos = pblk_dev_ppa_to_pos(geo, ppa);
 279
 280                 while (test_bit(pos, line->blk_bitmap)) {
 281                         r_ptr_int += pblk->min_write_pgs;
 282                         ppa = addr_to_gen_ppa(pblk, r_ptr_int, line->id);
 283                         pos = pblk_dev_ppa_to_pos(geo, ppa);
 284                 }
 285
 286                 for (j = 0; j < pblk->min_write_pgs; j++, i++, r_ptr_int++)
 287                         rqd->ppa_list[i] =
 288                                 addr_to_gen_ppa(pblk, r_ptr_int, line->id);
 289         }
 290
 291         /* If read fails, more padding is needed */
 292         ret = pblk_submit_io(pblk, rqd);
 293         if (ret) {
 294                 pr_err("pblk: I/O submission failed: %d\n", ret);
 295                 return ret;
 296         }
 297
 298         if (!wait_for_completion_io_timeout(&wait,
 299                                 msecs_to_jiffies(PBLK_COMMAND_TIMEOUT_MS))) {
 300                 pr_err("pblk: L2P recovery read timed out\n");
 301                 return -EINTR;
 302         }
 303         atomic_dec(&pblk->inflight_io);
 304         reinit_completion(&wait);
 305
 306         /* At this point, the read should not fail. If it does, it is a problem
 307          * we cannot recover from here. Need FTL log.
 308          */
 309         if (rqd->error) {
 310                 pr_err("pblk: L2P recovery failed (%d)\n", rqd->error);
 311                 return -EINTR;
 312         }
 313
 314         for (i = 0; i < rqd->nr_ppas; i++) {
 315                 u64 lba = le64_to_cpu(meta_list[i].lba);
 316
 317                 if (lba == ADDR_EMPTY || lba > pblk->rl.nr_secs)
 318                         continue;
 319
 320                 pblk_update_map(pblk, lba, rqd->ppa_list[i]);
 321         }
 322
 323         left_ppas -= rq_ppas;
 324         if (left_ppas > 0)
 325                 goto next_read_rq;
 326
 327         return 0;
 328 }
 329
 330 static void pblk_recov_complete(struct kref *ref)
 331 {
 332         struct pblk_pad_rq *pad_rq = container_of(ref, struct pblk_pad_rq, ref);
 333
 334         complete(&pad_rq->wait);
 335 }
 336
 337 static void pblk_end_io_recov(struct nvm_rq *rqd)
 338 {
 339         struct pblk_pad_rq *pad_rq = rqd->private;
 340         struct pblk *pblk = pad_rq->pblk;
 341         struct nvm_tgt_dev *dev = pblk->dev;
 342
 343         kref_put(&pad_rq->ref, pblk_recov_complete);
 344         nvm_dev_dma_free(dev->parent, rqd->meta_list, rqd->dma_meta_list);
 345         pblk_free_rqd(pblk, rqd, WRITE);
 346 }
 347
 348 static int pblk_recov_pad_oob(struct pblk *pblk, struct pblk_line *line,
 349                               int left_ppas)
 350 {
 351         struct nvm_tgt_dev *dev = pblk->dev;
 352         struct nvm_geo *geo = &dev->geo;
 353         struct ppa_addr *ppa_list;
 354         struct pblk_sec_meta *meta_list;
 355         struct pblk_pad_rq *pad_rq;
 356         struct nvm_rq *rqd;
 357         struct bio *bio;
 358         void *data;
 359         dma_addr_t dma_ppa_list, dma_meta_list;
 360         __le64 *lba_list = emeta_to_lbas(pblk, line->emeta->buf);
 361         u64 w_ptr = line->cur_sec;
 362         int left_line_ppas, rq_ppas, rq_len;
 363         int i, j;
 364         int ret = 0;
 365
 366         spin_lock(&line->lock);
 367         left_line_ppas = line->left_msecs;
 368         spin_unlock(&line->lock);
 369
 370         pad_rq = kmalloc(sizeof(struct pblk_pad_rq), GFP_KERNEL);
 371         if (!pad_rq)
 372                 return -ENOMEM;
 373
 374         data = vzalloc(pblk->max_write_pgs * geo->sec_size);
 375         if (!data) {
 376                 ret = -ENOMEM;
 377                 goto free_rq;
 378         }
 379
 380         pad_rq->pblk = pblk;
 381         init_completion(&pad_rq->wait);
 382         kref_init(&pad_rq->ref);
 383
 384 next_pad_rq:
 385         rq_ppas = pblk_calc_secs(pblk, left_ppas, 0);
 386         if (rq_ppas < pblk->min_write_pgs) {
 387                 pr_err("pblk: corrupted pad line %d\n", line->id);
 388                 goto free_rq;
 389         }
 390
 391         rq_len = rq_ppas * geo->sec_size;
 392
 393         meta_list = nvm_dev_dma_alloc(dev->parent, GFP_KERNEL, &dma_meta_list);
 394         if (!meta_list) {
 395                 ret = -ENOMEM;
 396                 goto free_data;
 397         }
 398
 399         ppa_list = (void *)(meta_list) + pblk_dma_meta_size;
 400         dma_ppa_list = dma_meta_list + pblk_dma_meta_size;
 401
 402         rqd = pblk_alloc_rqd(pblk, WRITE);
 403         if (IS_ERR(rqd)) {
 404                 ret = PTR_ERR(rqd);
 405                 goto fail_free_meta;
 406         }
 407         memset(rqd, 0, pblk_w_rq_size);
 408
 409         bio = bio_map_kern(dev->q, data, rq_len, GFP_KERNEL);
 410         if (IS_ERR(bio)) {
 411                 ret = PTR_ERR(bio);
 412                 goto fail_free_rqd;
 413         }
 414
 415         bio->bi_iter.bi_sector = 0; /* internal bio */
 416         bio_set_op_attrs(bio, REQ_OP_WRITE, 0);
 417
 418         rqd->bio = bio;
 419         rqd->opcode = NVM_OP_PWRITE;
 420         rqd->flags = pblk_set_progr_mode(pblk, WRITE);
 421         rqd->meta_list = meta_list;
 422         rqd->nr_ppas = rq_ppas;
 423         rqd->ppa_list = ppa_list;
 424         rqd->dma_ppa_list = dma_ppa_list;
 425         rqd->dma_meta_list = dma_meta_list;
 426         rqd->end_io = pblk_end_io_recov;
 427         rqd->private = pad_rq;
 428
 429         for (i = 0; i < rqd->nr_ppas; ) {
 430                 struct ppa_addr ppa;
 431                 int pos;
 432
 433                 w_ptr = pblk_alloc_page(pblk, line, pblk->min_write_pgs);
 434                 ppa = addr_to_pblk_ppa(pblk, w_ptr, line->id);
 435                 pos = pblk_ppa_to_pos(geo, ppa);
 436
 437                 while (test_bit(pos, line->blk_bitmap)) {
 438                         w_ptr += pblk->min_write_pgs;
 439                         ppa = addr_to_pblk_ppa(pblk, w_ptr, line->id);
 440                         pos = pblk_ppa_to_pos(geo, ppa);
 441                 }
 442
 443                 for (j = 0; j < pblk->min_write_pgs; j++, i++, w_ptr++) {
 444                         struct ppa_addr dev_ppa;
 445                         __le64 addr_empty = cpu_to_le64(ADDR_EMPTY);
 446
 447                         dev_ppa = addr_to_gen_ppa(pblk, w_ptr, line->id);
 448
 449                         pblk_map_invalidate(pblk, dev_ppa);
 450                         lba_list[w_ptr] = meta_list[i].lba = addr_empty;
 451                         rqd->ppa_list[i] = dev_ppa;
 452                 }
 453         }
 454
 455         kref_get(&pad_rq->ref);
 456
 457         ret = pblk_submit_io(pblk, rqd);
 458         if (ret) {
 459                 pr_err("pblk: I/O submission failed: %d\n", ret);
 460                 goto free_data;
 461         }
 462
 463         atomic_dec(&pblk->inflight_io);
 464
 465         left_line_ppas -= rq_ppas;
 466         left_ppas -= rq_ppas;
 467         if (left_ppas && left_line_ppas)
 468                 goto next_pad_rq;
 469
 470         kref_put(&pad_rq->ref, pblk_recov_complete);
 471
 472         if (!wait_for_completion_io_timeout(&pad_rq->wait,
 473                                 msecs_to_jiffies(PBLK_COMMAND_TIMEOUT_MS))) {
 474                 pr_err("pblk: pad write timed out\n");
 475                 ret = -ETIME;
 476         }
 477
 478 free_rq:
 479         kfree(pad_rq);
 480 free_data:
 481         vfree(data);
 482         return ret;
 483
 484 fail_free_rqd:
 485         pblk_free_rqd(pblk, rqd, WRITE);
 486 fail_free_meta:
 487         nvm_dev_dma_free(dev->parent, meta_list, dma_meta_list);
 488         kfree(pad_rq);
 489         return ret;
 490 }
 491
 492 /* When this function is called, it means that not all upper pages have been
 493  * written in a page that contains valid data. In order to recover this data, we
 494  * first find the write pointer on the device, then we pad all necessary
 495  * sectors, and finally attempt to read the valid data
 496  */
 497 static int pblk_recov_scan_all_oob(struct pblk *pblk, struct pblk_line *line,
 498                                    struct pblk_recov_alloc p)
 499 {
 500         struct nvm_tgt_dev *dev = pblk->dev;
 501         struct nvm_geo *geo = &dev->geo;
 502         struct ppa_addr *ppa_list;
 503         struct pblk_sec_meta *meta_list;
 504         struct nvm_rq *rqd;
 505         struct bio *bio;
 506         void *data;
 507         dma_addr_t dma_ppa_list, dma_meta_list;
 508         u64 w_ptr = 0, r_ptr;
 509         int rq_ppas, rq_len;
 510         int i, j;
 511         int ret = 0;
 512         int rec_round;
 513         int left_ppas = pblk_calc_sec_in_line(pblk, line) - line->cur_sec;
 514         DECLARE_COMPLETION_ONSTACK(wait);
 515
 516         ppa_list = p.ppa_list;
 517         meta_list = p.meta_list;
 518         rqd = p.rqd;
 519         data = p.data;
 520         dma_ppa_list = p.dma_ppa_list;
 521         dma_meta_list = p.dma_meta_list;
 522
 523         /* we could recover up until the line write pointer */
 524         r_ptr = line->cur_sec;
 525         rec_round = 0;
 526
 527 next_rq:
 528         memset(rqd, 0, pblk_g_rq_size);
 529
 530         rq_ppas = pblk_calc_secs(pblk, left_ppas, 0);
 531         if (!rq_ppas)
 532                 rq_ppas = pblk->min_write_pgs;
 533         rq_len = rq_ppas * geo->sec_size;
 534
 535         bio = bio_map_kern(dev->q, data, rq_len, GFP_KERNEL);
 536         if (IS_ERR(bio))
 537                 return PTR_ERR(bio);
 538
 539         bio->bi_iter.bi_sector = 0; /* internal bio */
 540         bio_set_op_attrs(bio, REQ_OP_READ, 0);
 541
 542         rqd->bio = bio;
 543         rqd->opcode = NVM_OP_PREAD;
 544         rqd->meta_list = meta_list;
 545         rqd->nr_ppas = rq_ppas;
 546         rqd->ppa_list = ppa_list;
 547         rqd->dma_ppa_list = dma_ppa_list;
 548         rqd->dma_meta_list = dma_meta_list;
 549         rqd->end_io = pblk_end_io_sync;
 550         rqd->private = &wait;
 551
 552         if (pblk_io_aligned(pblk, rq_ppas))
 553                 rqd->flags = pblk_set_read_mode(pblk, PBLK_READ_SEQUENTIAL);
 554         else
 555                 rqd->flags = pblk_set_read_mode(pblk, PBLK_READ_RANDOM);
 556
 557         for (i = 0; i < rqd->nr_ppas; ) {
 558                 struct ppa_addr ppa;
 559                 int pos;
 560
 561                 w_ptr = pblk_alloc_page(pblk, line, pblk->min_write_pgs);
 562                 ppa = addr_to_gen_ppa(pblk, w_ptr, line->id);
 563                 pos = pblk_dev_ppa_to_pos(geo, ppa);
 564
 565                 while (test_bit(pos, line->blk_bitmap)) {
 566                         w_ptr += pblk->min_write_pgs;
 567                         ppa = addr_to_gen_ppa(pblk, w_ptr, line->id);
 568                         pos = pblk_dev_ppa_to_pos(geo, ppa);
 569                 }
 570
 571                 for (j = 0; j < pblk->min_write_pgs; j++, i++, w_ptr++)
 572                         rqd->ppa_list[i] =
 573                                 addr_to_gen_ppa(pblk, w_ptr, line->id);
 574         }
 575
 576         ret = pblk_submit_io(pblk, rqd);
 577         if (ret) {
 578                 pr_err("pblk: I/O submission failed: %d\n", ret);
 579                 return ret;
 580         }
 581
 582         if (!wait_for_completion_io_timeout(&wait,
 583                                 msecs_to_jiffies(PBLK_COMMAND_TIMEOUT_MS))) {
 584                 pr_err("pblk: L2P recovery read timed out\n");
 585         }
 586         atomic_dec(&pblk->inflight_io);
 587         reinit_completion(&wait);
 588
 589         /* This should not happen since the read failed during normal recovery,
 590          * but the media works funny sometimes...
 591          */
 592         if (!rec_round++ && !rqd->error) {
 593                 rec_round = 0;
 594                 for (i = 0; i < rqd->nr_ppas; i++, r_ptr++) {
 595                         u64 lba = le64_to_cpu(meta_list[i].lba);
 596
 597                         if (lba == ADDR_EMPTY || lba > pblk->rl.nr_secs)
 598                                 continue;
 599
 600                         pblk_update_map(pblk, lba, rqd->ppa_list[i]);
 601                 }
 602         }
 603
 604         /* Reached the end of the written line */
 605         if (rqd->error == NVM_RSP_ERR_EMPTYPAGE) {
 606                 int pad_secs, nr_error_bits, bit;
 607                 int ret;
 608
 609                 bit = find_first_bit((void *)&rqd->ppa_status, rqd->nr_ppas);
 610                 nr_error_bits = rqd->nr_ppas - bit;
 611
 612                 /* Roll back failed sectors */
 613                 line->cur_sec -= nr_error_bits;
 614                 line->left_msecs += nr_error_bits;
 615                 bitmap_clear(line->map_bitmap, line->cur_sec, nr_error_bits);
 616
 617                 pad_secs = pblk_pad_distance(pblk);
 618                 if (pad_secs > line->left_msecs)
 619                         pad_secs = line->left_msecs;
 620
 621                 ret = pblk_recov_pad_oob(pblk, line, pad_secs);
 622                 if (ret)
 623                         pr_err("pblk: OOB padding failed (err:%d)\n", ret);
 624
 625                 ret = pblk_recov_read_oob(pblk, line, p, r_ptr);
 626                 if (ret)
 627                         pr_err("pblk: OOB read failed (err:%d)\n", ret);
 628
 629                 left_ppas = 0;
 630         }
 631
 632         left_ppas -= rq_ppas;
 633         if (left_ppas > 0)
 634                 goto next_rq;
 635
 636         return ret;
 637 }
 638
 639 static int pblk_recov_scan_oob(struct pblk *pblk, struct pblk_line *line,
 640                                struct pblk_recov_alloc p, int *done)
 641 {
 642         struct nvm_tgt_dev *dev = pblk->dev;
 643         struct nvm_geo *geo = &dev->geo;
 644         struct ppa_addr *ppa_list;
 645         struct pblk_sec_meta *meta_list;
 646         struct nvm_rq *rqd;
 647         struct bio *bio;
 648         void *data;
 649         dma_addr_t dma_ppa_list, dma_meta_list;
 650         u64 paddr;
 651         int rq_ppas, rq_len;
 652         int i, j;
 653         int ret = 0;
 654         int left_ppas = pblk_calc_sec_in_line(pblk, line);
 655         DECLARE_COMPLETION_ONSTACK(wait);
 656
 657         ppa_list = p.ppa_list;
 658         meta_list = p.meta_list;
 659         rqd = p.rqd;
 660         data = p.data;
 661         dma_ppa_list = p.dma_ppa_list;
 662         dma_meta_list = p.dma_meta_list;
 663
 664         *done = 1;
 665
 666 next_rq:
 667         memset(rqd, 0, pblk_g_rq_size);
 668
 669         rq_ppas = pblk_calc_secs(pblk, left_ppas, 0);
 670         if (!rq_ppas)
 671                 rq_ppas = pblk->min_write_pgs;
 672         rq_len = rq_ppas * geo->sec_size;
 673
 674         bio = bio_map_kern(dev->q, data, rq_len, GFP_KERNEL);
 675         if (IS_ERR(bio))
 676                 return PTR_ERR(bio);
 677
 678         bio->bi_iter.bi_sector = 0; /* internal bio */
 679         bio_set_op_attrs(bio, REQ_OP_READ, 0);
 680
 681         rqd->bio = bio;
 682         rqd->opcode = NVM_OP_PREAD;
 683         rqd->meta_list = meta_list;
 684         rqd->nr_ppas = rq_ppas;
 685         rqd->ppa_list = ppa_list;
 686         rqd->dma_ppa_list = dma_ppa_list;
 687         rqd->dma_meta_list = dma_meta_list;
 688         rqd->end_io = pblk_end_io_sync;
 689         rqd->private = &wait;
 690
 691         if (pblk_io_aligned(pblk, rq_ppas))
 692                 rqd->flags = pblk_set_read_mode(pblk, PBLK_READ_SEQUENTIAL);
 693         else
 694                 rqd->flags = pblk_set_read_mode(pblk, PBLK_READ_RANDOM);
 695
 696         for (i = 0; i < rqd->nr_ppas; ) {
 697                 struct ppa_addr ppa;
 698                 int pos;
 699
 700                 paddr = pblk_alloc_page(pblk, line, pblk->min_write_pgs);
 701                 ppa = addr_to_gen_ppa(pblk, paddr, line->id);
 702                 pos = pblk_dev_ppa_to_pos(geo, ppa);
 703
 704                 while (test_bit(pos, line->blk_bitmap)) {
 705                         paddr += pblk->min_write_pgs;
 706                         ppa = addr_to_gen_ppa(pblk, paddr, line->id);
 707                         pos = pblk_dev_ppa_to_pos(geo, ppa);
 708                 }
 709
 710                 for (j = 0; j < pblk->min_write_pgs; j++, i++, paddr++)
 711                         rqd->ppa_list[i] =
 712                                 addr_to_gen_ppa(pblk, paddr, line->id);
 713         }
 714
 715         ret = pblk_submit_io(pblk, rqd);
 716         if (ret) {
 717                 pr_err("pblk: I/O submission failed: %d\n", ret);
 718                 bio_put(bio);
 719                 return ret;
 720         }
 721
 722         if (!wait_for_completion_io_timeout(&wait,
 723                                 msecs_to_jiffies(PBLK_COMMAND_TIMEOUT_MS))) {
 724                 pr_err("pblk: L2P recovery read timed out\n");
 725         }
 726         atomic_dec(&pblk->inflight_io);
 727         reinit_completion(&wait);
 728
 729         /* Reached the end of the written line */
 730         if (rqd->error) {
 731                 int nr_error_bits, bit;
 732
 733                 bit = find_first_bit((void *)&rqd->ppa_status, rqd->nr_ppas);
 734                 nr_error_bits = rqd->nr_ppas - bit;
 735
 736                 /* Roll back failed sectors */
 737                 line->cur_sec -= nr_error_bits;
 738                 line->left_msecs += nr_error_bits;
 739                 bitmap_clear(line->map_bitmap, line->cur_sec, nr_error_bits);
 740
 741                 left_ppas = 0;
 742                 rqd->nr_ppas = bit;
 743
 744                 if (rqd->error != NVM_RSP_ERR_EMPTYPAGE)
 745                         *done = 0;
 746         }
 747
 748         for (i = 0; i < rqd->nr_ppas; i++) {
 749                 u64 lba = le64_to_cpu(meta_list[i].lba);
 750
 751                 if (lba == ADDR_EMPTY || lba > pblk->rl.nr_secs)
 752                         continue;
 753
 754                 pblk_update_map(pblk, lba, rqd->ppa_list[i]);
 755         }
 756
 757         left_ppas -= rq_ppas;
 758         if (left_ppas > 0)
 759                 goto next_rq;
 760
 761         return ret;
 762 }
 763
 764 /* Scan line for lbas on out of bound area */
 765 static int pblk_recov_l2p_from_oob(struct pblk *pblk, struct pblk_line *line)
 766 {
 767         struct nvm_tgt_dev *dev = pblk->dev;
 768         struct nvm_geo *geo = &dev->geo;
 769         struct nvm_rq *rqd;
 770         struct ppa_addr *ppa_list;
 771         struct pblk_sec_meta *meta_list;
 772         struct pblk_recov_alloc p;
 773         void *data;
 774         dma_addr_t dma_ppa_list, dma_meta_list;
 775         int done, ret = 0;
 776
 777         rqd = pblk_alloc_rqd(pblk, READ);
 778         if (IS_ERR(rqd))
 779                 return PTR_ERR(rqd);
 780
 781         meta_list = nvm_dev_dma_alloc(dev->parent, GFP_KERNEL, &dma_meta_list);
 782         if (!meta_list) {
 783                 ret = -ENOMEM;
 784                 goto free_rqd;
 785         }
 786
 787         ppa_list = (void *)(meta_list) + pblk_dma_meta_size;
 788         dma_ppa_list = dma_meta_list + pblk_dma_meta_size;
 789
 790         data = kcalloc(pblk->max_write_pgs, geo->sec_size, GFP_KERNEL);
 791         if (!data) {
 792                 ret = -ENOMEM;
 793                 goto free_meta_list;
 794         }
 795
 796         p.ppa_list = ppa_list;
 797         p.meta_list = meta_list;
 798         p.rqd = rqd;
 799         p.data = data;
 800         p.dma_ppa_list = dma_ppa_list;
 801         p.dma_meta_list = dma_meta_list;
 802
 803         ret = pblk_recov_scan_oob(pblk, line, p, &done);
 804         if (ret) {
 805                 pr_err("pblk: could not recover L2P from OOB\n");
 806                 goto out;
 807         }
 808
 809         if (!done) {
 810                 ret = pblk_recov_scan_all_oob(pblk, line, p);
 811                 if (ret) {
 812                         pr_err("pblk: could not recover L2P from OOB\n");
 813                         goto out;
 814                 }
 815         }
 816
 817         if (pblk_line_is_full(line))
 818                 pblk_line_recov_close(pblk, line);
 819
 820 out:
 821         kfree(data);
 822 free_meta_list:
 823         nvm_dev_dma_free(dev->parent, meta_list, dma_meta_list);
 824 free_rqd:
 825         pblk_free_rqd(pblk, rqd, READ);
 826
 827         return ret;
 828 }
 829
 830 /* Insert lines ordered by sequence number (seq_num) on list */
 831 static void pblk_recov_line_add_ordered(struct list_head *head,
 832                                         struct pblk_line *line)
 833 {
 834         struct pblk_line *t = NULL;
 835
 836         list_for_each_entry(t, head, list)
 837                 if (t->seq_nr > line->seq_nr)
 838                         break;
 839
 840         __list_add(&line->list, t->list.prev, &t->list);
 841 }
 842
 843 struct pblk_line *pblk_recov_l2p(struct pblk *pblk)
 844 {
 845         struct nvm_tgt_dev *dev = pblk->dev;
 846         struct nvm_geo *geo = &dev->geo;
 847         struct pblk_line_meta *lm = &pblk->lm;
 848         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
 849         struct pblk_line *line, *tline, *data_line = NULL;
 850         struct pblk_smeta *smeta;
 851         struct pblk_emeta *emeta;
 852         struct line_smeta *smeta_buf;
 853         int found_lines = 0, recovered_lines = 0, open_lines = 0;
 854         int is_next = 0;
 855         int meta_line;
 856         int i, valid_uuid = 0;
 857         LIST_HEAD(recov_list);
 858
 859         /* TODO: Implement FTL snapshot */
 860
 861         /* Scan recovery - takes place when FTL snapshot fails */
 862         spin_lock(&l_mg->free_lock);
 863         meta_line = find_first_zero_bit(&l_mg->meta_bitmap, PBLK_DATA_LINES);
 864         set_bit(meta_line, &l_mg->meta_bitmap);
 865         smeta = l_mg->sline_meta[meta_line];
 866         emeta = l_mg->eline_meta[meta_line];
 867         smeta_buf = (struct line_smeta *)smeta;
 868         spin_unlock(&l_mg->free_lock);
 869
 870         /* Order data lines using their sequence number */
 871         for (i = 0; i < l_mg->nr_lines; i++) {
 872                 u32 crc;
 873
 874                 line = &pblk->lines[i];
 875
 876                 memset(smeta, 0, lm->smeta_len);
 877                 line->smeta = smeta;
 878                 line->lun_bitmap = ((void *)(smeta_buf)) +
 879                                                 sizeof(struct line_smeta);
 880
 881                 /* Lines that cannot be read are assumed as not written here */
 882                 if (pblk_line_read_smeta(pblk, line))
 883                         continue;
 884
 885                 crc = pblk_calc_smeta_crc(pblk, smeta_buf);
 886                 if (le32_to_cpu(smeta_buf->crc) != crc)
 887                         continue;
 888
 889                 if (le32_to_cpu(smeta_buf->header.identifier) != PBLK_MAGIC)
 890                         continue;
 891
 892                 if (le16_to_cpu(smeta_buf->header.version) != 1) {
 893                         pr_err("pblk: found incompatible line version %u\n",
 894                                         smeta_buf->header.version);
 895                         return ERR_PTR(-EINVAL);
 896                 }
 897
 898                 /* The first valid instance uuid is used for initialization */
 899                 if (!valid_uuid) {
 900                         memcpy(pblk->instance_uuid, smeta_buf->header.uuid, 16);
 901                         valid_uuid = 1;
 902                 }
 903
 904                 if (memcmp(pblk->instance_uuid, smeta_buf->header.uuid, 16)) {
 905                         pr_debug("pblk: ignore line %u due to uuid mismatch\n",
 906                                         i);
 907                         continue;
 908                 }
 909
 910                 /* Update line metadata */
 911                 spin_lock(&line->lock);
 912                 line->id = le32_to_cpu(smeta_buf->header.id);
 913                 line->type = le16_to_cpu(smeta_buf->header.type);
 914                 line->seq_nr = le64_to_cpu(smeta_buf->seq_nr);
 915                 spin_unlock(&line->lock);
 916
 917                 /* Update general metadata */
 918                 spin_lock(&l_mg->free_lock);
 919                 if (line->seq_nr >= l_mg->d_seq_nr)
 920                         l_mg->d_seq_nr = line->seq_nr + 1;
 921                 l_mg->nr_free_lines--;
 922                 spin_unlock(&l_mg->free_lock);
 923
 924                 if (pblk_line_recov_alloc(pblk, line))
 925                         goto out;
 926
 927                 pblk_recov_line_add_ordered(&recov_list, line);
 928                 found_lines++;
 929                 pr_debug("pblk: recovering data line %d, seq:%llu\n",
 930                                                 line->id, smeta_buf->seq_nr);
 931         }
 932
 933         if (!found_lines) {
 934                 pblk_setup_uuid(pblk);
 935
 936                 spin_lock(&l_mg->free_lock);
 937                 WARN_ON_ONCE(!test_and_clear_bit(meta_line,
 938                                                         &l_mg->meta_bitmap));
 939                 spin_unlock(&l_mg->free_lock);
 940
 941                 goto out;
 942         }
 943
 944         /* Verify closed blocks and recover this portion of L2P table*/
 945         list_for_each_entry_safe(line, tline, &recov_list, list) {
 946                 int off, nr_bb;
 947
 948                 recovered_lines++;
 949                 /* Calculate where emeta starts based on the line bb */
 950                 off = lm->sec_per_line - lm->emeta_sec[0];
 951                 nr_bb = bitmap_weight(line->blk_bitmap, lm->blk_per_line);
 952                 off -= nr_bb * geo->sec_per_pl;
 953
 954                 line->emeta_ssec = off;
 955                 line->emeta = emeta;
 956                 memset(line->emeta->buf, 0, lm->emeta_len[0]);
 957
 958                 if (pblk_line_read_emeta(pblk, line, line->emeta->buf)) {
 959                         pblk_recov_l2p_from_oob(pblk, line);
 960                         goto next;
 961                 }
 962
 963                 if (pblk_recov_l2p_from_emeta(pblk, line))
 964                         pblk_recov_l2p_from_oob(pblk, line);
 965
 966 next:
 967                 if (pblk_line_is_full(line)) {
 968                         struct list_head *move_list;
 969
 970                         spin_lock(&line->lock);
 971                         line->state = PBLK_LINESTATE_CLOSED;
 972                         move_list = pblk_line_gc_list(pblk, line);
 973                         spin_unlock(&line->lock);
 974
 975                         spin_lock(&l_mg->gc_lock);
 976                         list_move_tail(&line->list, move_list);
 977                         spin_unlock(&l_mg->gc_lock);
 978
 979                         mempool_free(line->map_bitmap, pblk->line_meta_pool);
 980                         line->map_bitmap = NULL;
 981                         line->smeta = NULL;
 982                         line->emeta = NULL;
 983                 } else {
 984                         if (open_lines > 1)
 985                                 pr_err("pblk: failed to recover L2P\n");
 986
 987                         open_lines++;
 988                         line->meta_line = meta_line;
 989                         data_line = line;
 990                 }
 991         }
 992
 993         spin_lock(&l_mg->free_lock);
 994         if (!open_lines) {
 995                 WARN_ON_ONCE(!test_and_clear_bit(meta_line,
 996                                                         &l_mg->meta_bitmap));
 997                 pblk_line_replace_data(pblk);
 998         } else {
 999                 /* Allocate next line for preparation */
1000                 l_mg->data_next = pblk_line_get(pblk);
1001                 if (l_mg->data_next) {
1002                         l_mg->data_next->seq_nr = l_mg->d_seq_nr++;
1003                         l_mg->data_next->type = PBLK_LINETYPE_DATA;
1004                         is_next = 1;
1005                 }
1006         }
1007         spin_unlock(&l_mg->free_lock);
1008
1009         if (is_next) {
1010                 pblk_line_erase(pblk, l_mg->data_next);
1011                 pblk_rl_free_lines_dec(&pblk->rl, l_mg->data_next);
1012         }
1013
1014 out:
1015         if (found_lines != recovered_lines)
1016                 pr_err("pblk: failed to recover all found lines %d/%d\n",
1017                                                 found_lines, recovered_lines);
1018
1019         return data_line;
1020 }
1021
1022 /*
1023  * Pad current line
1024  */
1025 int pblk_recov_pad(struct pblk *pblk)
1026 {
1027         struct pblk_line *line;
1028         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
1029         int left_msecs;
1030         int ret = 0;
1031
1032         spin_lock(&l_mg->free_lock);
1033         line = l_mg->data_line;
1034         left_msecs = line->left_msecs;
1035         spin_unlock(&l_mg->free_lock);
1036
1037         ret = pblk_recov_pad_oob(pblk, line, left_msecs);
1038         if (ret) {
1039                 pr_err("pblk: Tear down padding failed (%d)\n", ret);
1040                 return ret;
1041         }
1042
1043         pblk_line_close_meta(pblk, line);
1044         return ret;
1045 }