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 int pblk_recov_pad_oob(struct pblk *pblk, struct pblk_line *line,
 331                               struct pblk_recov_alloc p, int left_ppas)
 332 {
 333         struct nvm_tgt_dev *dev = pblk->dev;
 334         struct nvm_geo *geo = &dev->geo;
 335         struct ppa_addr *ppa_list;
 336         struct pblk_sec_meta *meta_list;
 337         struct nvm_rq *rqd;
 338         struct bio *bio;
 339         void *data;
 340         dma_addr_t dma_ppa_list, dma_meta_list;
 341         __le64 *lba_list = emeta_to_lbas(pblk, line->emeta->buf);
 342         u64 w_ptr = line->cur_sec;
 343         int left_line_ppas = line->left_msecs;
 344         int rq_ppas, rq_len;
 345         int i, j;
 346         int ret = 0;
 347         DECLARE_COMPLETION_ONSTACK(wait);
 348
 349         ppa_list = p.ppa_list;
 350         meta_list = p.meta_list;
 351         rqd = p.rqd;
 352         data = p.data;
 353         dma_ppa_list = p.dma_ppa_list;
 354         dma_meta_list = p.dma_meta_list;
 355
 356 next_pad_rq:
 357         rq_ppas = pblk_calc_secs(pblk, left_ppas, 0);
 358         if (!rq_ppas)
 359                 rq_ppas = pblk->min_write_pgs;
 360         rq_len = rq_ppas * geo->sec_size;
 361
 362         bio = bio_map_kern(dev->q, data, rq_len, GFP_KERNEL);
 363         if (IS_ERR(bio))
 364                 return PTR_ERR(bio);
 365
 366         bio->bi_iter.bi_sector = 0; /* internal bio */
 367         bio_set_op_attrs(bio, REQ_OP_WRITE, 0);
 368
 369         memset(rqd, 0, pblk_g_rq_size);
 370
 371         rqd->bio = bio;
 372         rqd->opcode = NVM_OP_PWRITE;
 373         rqd->flags = pblk_set_progr_mode(pblk, WRITE);
 374         rqd->meta_list = meta_list;
 375         rqd->nr_ppas = rq_ppas;
 376         rqd->ppa_list = ppa_list;
 377         rqd->dma_ppa_list = dma_ppa_list;
 378         rqd->dma_meta_list = dma_meta_list;
 379         rqd->end_io = pblk_end_io_sync;
 380         rqd->private = &wait;
 381
 382         for (i = 0; i < rqd->nr_ppas; ) {
 383                 struct ppa_addr ppa;
 384                 int pos;
 385
 386                 w_ptr = pblk_alloc_page(pblk, line, pblk->min_write_pgs);
 387                 ppa = addr_to_pblk_ppa(pblk, w_ptr, line->id);
 388                 pos = pblk_ppa_to_pos(geo, ppa);
 389
 390                 while (test_bit(pos, line->blk_bitmap)) {
 391                         w_ptr += pblk->min_write_pgs;
 392                         ppa = addr_to_pblk_ppa(pblk, w_ptr, line->id);
 393                         pos = pblk_ppa_to_pos(geo, ppa);
 394                 }
 395
 396                 for (j = 0; j < pblk->min_write_pgs; j++, i++, w_ptr++) {
 397                         struct ppa_addr dev_ppa;
 398                         u64 addr_empty = cpu_to_le64(ADDR_EMPTY);
 399
 400                         dev_ppa = addr_to_gen_ppa(pblk, w_ptr, line->id);
 401
 402                         pblk_map_invalidate(pblk, dev_ppa);
 403                         lba_list[w_ptr] = meta_list[i].lba = addr_empty;
 404                         rqd->ppa_list[i] = dev_ppa;
 405                 }
 406         }
 407
 408         ret = pblk_submit_io(pblk, rqd);
 409         if (ret) {
 410                 pr_err("pblk: I/O submission failed: %d\n", ret);
 411                 return ret;
 412         }
 413
 414         if (!wait_for_completion_io_timeout(&wait,
 415                                 msecs_to_jiffies(PBLK_COMMAND_TIMEOUT_MS))) {
 416                 pr_err("pblk: L2P recovery write timed out\n");
 417         }
 418         atomic_dec(&pblk->inflight_io);
 419         reinit_completion(&wait);
 420
 421         left_line_ppas -= rq_ppas;
 422         left_ppas -= rq_ppas;
 423         if (left_ppas > 0 && left_line_ppas)
 424                 goto next_pad_rq;
 425
 426         return 0;
 427 }
 428
 429 /* When this function is called, it means that not all upper pages have been
 430  * written in a page that contains valid data. In order to recover this data, we
 431  * first find the write pointer on the device, then we pad all necessary
 432  * sectors, and finally attempt to read the valid data
 433  */
 434 static int pblk_recov_scan_all_oob(struct pblk *pblk, struct pblk_line *line,
 435                                    struct pblk_recov_alloc p)
 436 {
 437         struct nvm_tgt_dev *dev = pblk->dev;
 438         struct nvm_geo *geo = &dev->geo;
 439         struct ppa_addr *ppa_list;
 440         struct pblk_sec_meta *meta_list;
 441         struct nvm_rq *rqd;
 442         struct bio *bio;
 443         void *data;
 444         dma_addr_t dma_ppa_list, dma_meta_list;
 445         u64 w_ptr = 0, r_ptr;
 446         int rq_ppas, rq_len;
 447         int i, j;
 448         int ret = 0;
 449         int rec_round;
 450         int left_ppas = pblk_calc_sec_in_line(pblk, line) - line->cur_sec;
 451         DECLARE_COMPLETION_ONSTACK(wait);
 452
 453         ppa_list = p.ppa_list;
 454         meta_list = p.meta_list;
 455         rqd = p.rqd;
 456         data = p.data;
 457         dma_ppa_list = p.dma_ppa_list;
 458         dma_meta_list = p.dma_meta_list;
 459
 460         /* we could recover up until the line write pointer */
 461         r_ptr = line->cur_sec;
 462         rec_round = 0;
 463
 464 next_rq:
 465         memset(rqd, 0, pblk_g_rq_size);
 466
 467         rq_ppas = pblk_calc_secs(pblk, left_ppas, 0);
 468         if (!rq_ppas)
 469                 rq_ppas = pblk->min_write_pgs;
 470         rq_len = rq_ppas * geo->sec_size;
 471
 472         bio = bio_map_kern(dev->q, data, rq_len, GFP_KERNEL);
 473         if (IS_ERR(bio))
 474                 return PTR_ERR(bio);
 475
 476         bio->bi_iter.bi_sector = 0; /* internal bio */
 477         bio_set_op_attrs(bio, REQ_OP_READ, 0);
 478
 479         rqd->bio = bio;
 480         rqd->opcode = NVM_OP_PREAD;
 481         rqd->meta_list = meta_list;
 482         rqd->nr_ppas = rq_ppas;
 483         rqd->ppa_list = ppa_list;
 484         rqd->dma_ppa_list = dma_ppa_list;
 485         rqd->dma_meta_list = dma_meta_list;
 486         rqd->end_io = pblk_end_io_sync;
 487         rqd->private = &wait;
 488
 489         if (pblk_io_aligned(pblk, rq_ppas))
 490                 rqd->flags = pblk_set_read_mode(pblk, PBLK_READ_SEQUENTIAL);
 491         else
 492                 rqd->flags = pblk_set_read_mode(pblk, PBLK_READ_RANDOM);
 493
 494         for (i = 0; i < rqd->nr_ppas; ) {
 495                 struct ppa_addr ppa;
 496                 int pos;
 497
 498                 w_ptr = pblk_alloc_page(pblk, line, pblk->min_write_pgs);
 499                 ppa = addr_to_gen_ppa(pblk, w_ptr, line->id);
 500                 pos = pblk_dev_ppa_to_pos(geo, ppa);
 501
 502                 while (test_bit(pos, line->blk_bitmap)) {
 503                         w_ptr += pblk->min_write_pgs;
 504                         ppa = addr_to_gen_ppa(pblk, w_ptr, line->id);
 505                         pos = pblk_dev_ppa_to_pos(geo, ppa);
 506                 }
 507
 508                 for (j = 0; j < pblk->min_write_pgs; j++, i++, w_ptr++)
 509                         rqd->ppa_list[i] =
 510                                 addr_to_gen_ppa(pblk, w_ptr, line->id);
 511         }
 512
 513         ret = pblk_submit_io(pblk, rqd);
 514         if (ret) {
 515                 pr_err("pblk: I/O submission failed: %d\n", ret);
 516                 return ret;
 517         }
 518
 519         if (!wait_for_completion_io_timeout(&wait,
 520                                 msecs_to_jiffies(PBLK_COMMAND_TIMEOUT_MS))) {
 521                 pr_err("pblk: L2P recovery read timed out\n");
 522         }
 523         atomic_dec(&pblk->inflight_io);
 524         reinit_completion(&wait);
 525
 526         /* This should not happen since the read failed during normal recovery,
 527          * but the media works funny sometimes...
 528          */
 529         if (!rec_round++ && !rqd->error) {
 530                 rec_round = 0;
 531                 for (i = 0; i < rqd->nr_ppas; i++, r_ptr++) {
 532                         u64 lba = le64_to_cpu(meta_list[i].lba);
 533
 534                         if (lba == ADDR_EMPTY || lba > pblk->rl.nr_secs)
 535                                 continue;
 536
 537                         pblk_update_map(pblk, lba, rqd->ppa_list[i]);
 538                 }
 539         }
 540
 541         /* Reached the end of the written line */
 542         if (rqd->error == NVM_RSP_ERR_EMPTYPAGE) {
 543                 int pad_secs, nr_error_bits, bit;
 544                 int ret;
 545
 546                 bit = find_first_bit((void *)&rqd->ppa_status, rqd->nr_ppas);
 547                 nr_error_bits = rqd->nr_ppas - bit;
 548
 549                 /* Roll back failed sectors */
 550                 line->cur_sec -= nr_error_bits;
 551                 line->left_msecs += nr_error_bits;
 552                 bitmap_clear(line->map_bitmap, line->cur_sec, nr_error_bits);
 553
 554                 pad_secs = pblk_pad_distance(pblk);
 555                 if (pad_secs > line->left_msecs)
 556                         pad_secs = line->left_msecs;
 557
 558                 ret = pblk_recov_pad_oob(pblk, line, p, pad_secs);
 559                 if (ret)
 560                         pr_err("pblk: OOB padding failed (err:%d)\n", ret);
 561
 562                 ret = pblk_recov_read_oob(pblk, line, p, r_ptr);
 563                 if (ret)
 564                         pr_err("pblk: OOB read failed (err:%d)\n", ret);
 565
 566                 left_ppas = 0;
 567         }
 568
 569         left_ppas -= rq_ppas;
 570         if (left_ppas > 0)
 571                 goto next_rq;
 572
 573         return ret;
 574 }
 575
 576 static int pblk_recov_scan_oob(struct pblk *pblk, struct pblk_line *line,
 577                                struct pblk_recov_alloc p, int *done)
 578 {
 579         struct nvm_tgt_dev *dev = pblk->dev;
 580         struct nvm_geo *geo = &dev->geo;
 581         struct ppa_addr *ppa_list;
 582         struct pblk_sec_meta *meta_list;
 583         struct nvm_rq *rqd;
 584         struct bio *bio;
 585         void *data;
 586         dma_addr_t dma_ppa_list, dma_meta_list;
 587         u64 paddr;
 588         int rq_ppas, rq_len;
 589         int i, j;
 590         int ret = 0;
 591         int left_ppas = pblk_calc_sec_in_line(pblk, line);
 592         DECLARE_COMPLETION_ONSTACK(wait);
 593
 594         ppa_list = p.ppa_list;
 595         meta_list = p.meta_list;
 596         rqd = p.rqd;
 597         data = p.data;
 598         dma_ppa_list = p.dma_ppa_list;
 599         dma_meta_list = p.dma_meta_list;
 600
 601         *done = 1;
 602
 603 next_rq:
 604         memset(rqd, 0, pblk_g_rq_size);
 605
 606         rq_ppas = pblk_calc_secs(pblk, left_ppas, 0);
 607         if (!rq_ppas)
 608                 rq_ppas = pblk->min_write_pgs;
 609         rq_len = rq_ppas * geo->sec_size;
 610
 611         bio = bio_map_kern(dev->q, data, rq_len, GFP_KERNEL);
 612         if (IS_ERR(bio))
 613                 return PTR_ERR(bio);
 614
 615         bio->bi_iter.bi_sector = 0; /* internal bio */
 616         bio_set_op_attrs(bio, REQ_OP_READ, 0);
 617
 618         rqd->bio = bio;
 619         rqd->opcode = NVM_OP_PREAD;
 620         rqd->meta_list = meta_list;
 621         rqd->nr_ppas = rq_ppas;
 622         rqd->ppa_list = ppa_list;
 623         rqd->dma_ppa_list = dma_ppa_list;
 624         rqd->dma_meta_list = dma_meta_list;
 625         rqd->end_io = pblk_end_io_sync;
 626         rqd->private = &wait;
 627
 628         if (pblk_io_aligned(pblk, rq_ppas))
 629                 rqd->flags = pblk_set_read_mode(pblk, PBLK_READ_SEQUENTIAL);
 630         else
 631                 rqd->flags = pblk_set_read_mode(pblk, PBLK_READ_RANDOM);
 632
 633         for (i = 0; i < rqd->nr_ppas; ) {
 634                 struct ppa_addr ppa;
 635                 int pos;
 636
 637                 paddr = pblk_alloc_page(pblk, line, pblk->min_write_pgs);
 638                 ppa = addr_to_gen_ppa(pblk, paddr, line->id);
 639                 pos = pblk_dev_ppa_to_pos(geo, ppa);
 640
 641                 while (test_bit(pos, line->blk_bitmap)) {
 642                         paddr += pblk->min_write_pgs;
 643                         ppa = addr_to_gen_ppa(pblk, paddr, line->id);
 644                         pos = pblk_dev_ppa_to_pos(geo, ppa);
 645                 }
 646
 647                 for (j = 0; j < pblk->min_write_pgs; j++, i++, paddr++)
 648                         rqd->ppa_list[i] =
 649                                 addr_to_gen_ppa(pblk, paddr, line->id);
 650         }
 651
 652         ret = pblk_submit_io(pblk, rqd);
 653         if (ret) {
 654                 pr_err("pblk: I/O submission failed: %d\n", ret);
 655                 bio_put(bio);
 656                 return ret;
 657         }
 658
 659         if (!wait_for_completion_io_timeout(&wait,
 660                                 msecs_to_jiffies(PBLK_COMMAND_TIMEOUT_MS))) {
 661                 pr_err("pblk: L2P recovery read timed out\n");
 662         }
 663         atomic_dec(&pblk->inflight_io);
 664         reinit_completion(&wait);
 665
 666         /* Reached the end of the written line */
 667         if (rqd->error) {
 668                 int nr_error_bits, bit;
 669
 670                 bit = find_first_bit((void *)&rqd->ppa_status, rqd->nr_ppas);
 671                 nr_error_bits = rqd->nr_ppas - bit;
 672
 673                 /* Roll back failed sectors */
 674                 line->cur_sec -= nr_error_bits;
 675                 line->left_msecs += nr_error_bits;
 676                 bitmap_clear(line->map_bitmap, line->cur_sec, nr_error_bits);
 677
 678                 left_ppas = 0;
 679                 rqd->nr_ppas = bit;
 680
 681                 if (rqd->error != NVM_RSP_ERR_EMPTYPAGE)
 682                         *done = 0;
 683         }
 684
 685         for (i = 0; i < rqd->nr_ppas; i++) {
 686                 u64 lba = le64_to_cpu(meta_list[i].lba);
 687
 688                 if (lba == ADDR_EMPTY || lba > pblk->rl.nr_secs)
 689                         continue;
 690
 691                 pblk_update_map(pblk, lba, rqd->ppa_list[i]);
 692         }
 693
 694         left_ppas -= rq_ppas;
 695         if (left_ppas > 0)
 696                 goto next_rq;
 697
 698         return ret;
 699 }
 700
 701 /* Scan line for lbas on out of bound area */
 702 static int pblk_recov_l2p_from_oob(struct pblk *pblk, struct pblk_line *line)
 703 {
 704         struct nvm_tgt_dev *dev = pblk->dev;
 705         struct nvm_geo *geo = &dev->geo;
 706         struct nvm_rq *rqd;
 707         struct ppa_addr *ppa_list;
 708         struct pblk_sec_meta *meta_list;
 709         struct pblk_recov_alloc p;
 710         void *data;
 711         dma_addr_t dma_ppa_list, dma_meta_list;
 712         int done, ret = 0;
 713
 714         rqd = pblk_alloc_rqd(pblk, READ);
 715         if (IS_ERR(rqd))
 716                 return PTR_ERR(rqd);
 717
 718         meta_list = nvm_dev_dma_alloc(dev->parent, GFP_KERNEL, &dma_meta_list);
 719         if (!meta_list) {
 720                 ret = -ENOMEM;
 721                 goto free_rqd;
 722         }
 723
 724         ppa_list = (void *)(meta_list) + pblk_dma_meta_size;
 725         dma_ppa_list = dma_meta_list + pblk_dma_meta_size;
 726
 727         data = kcalloc(pblk->max_write_pgs, geo->sec_size, GFP_KERNEL);
 728         if (!data) {
 729                 ret = -ENOMEM;
 730                 goto free_meta_list;
 731         }
 732
 733         p.ppa_list = ppa_list;
 734         p.meta_list = meta_list;
 735         p.rqd = rqd;
 736         p.data = data;
 737         p.dma_ppa_list = dma_ppa_list;
 738         p.dma_meta_list = dma_meta_list;
 739
 740         ret = pblk_recov_scan_oob(pblk, line, p, &done);
 741         if (ret) {
 742                 pr_err("pblk: could not recover L2P from OOB\n");
 743                 goto out;
 744         }
 745
 746         if (!done) {
 747                 ret = pblk_recov_scan_all_oob(pblk, line, p);
 748                 if (ret) {
 749                         pr_err("pblk: could not recover L2P from OOB\n");
 750                         goto out;
 751                 }
 752         }
 753
 754         if (pblk_line_is_full(line))
 755                 pblk_line_recov_close(pblk, line);
 756
 757 out:
 758         kfree(data);
 759 free_meta_list:
 760         nvm_dev_dma_free(dev->parent, meta_list, dma_meta_list);
 761 free_rqd:
 762         pblk_free_rqd(pblk, rqd, READ);
 763
 764         return ret;
 765 }
 766
 767 /* Insert lines ordered by sequence number (seq_num) on list */
 768 static void pblk_recov_line_add_ordered(struct list_head *head,
 769                                         struct pblk_line *line)
 770 {
 771         struct pblk_line *t = NULL;
 772
 773         list_for_each_entry(t, head, list)
 774                 if (t->seq_nr > line->seq_nr)
 775                         break;
 776
 777         __list_add(&line->list, t->list.prev, &t->list);
 778 }
 779
 780 struct pblk_line *pblk_recov_l2p(struct pblk *pblk)
 781 {
 782         struct nvm_tgt_dev *dev = pblk->dev;
 783         struct nvm_geo *geo = &dev->geo;
 784         struct pblk_line_meta *lm = &pblk->lm;
 785         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
 786         struct pblk_line *line, *tline, *data_line = NULL;
 787         struct pblk_smeta *smeta;
 788         struct pblk_emeta *emeta;
 789         struct line_smeta *smeta_buf;
 790         int found_lines = 0, recovered_lines = 0, open_lines = 0;
 791         int is_next = 0;
 792         int meta_line;
 793         int i, valid_uuid = 0;
 794         LIST_HEAD(recov_list);
 795
 796         /* TODO: Implement FTL snapshot */
 797
 798         /* Scan recovery - takes place when FTL snapshot fails */
 799         spin_lock(&l_mg->free_lock);
 800         meta_line = find_first_zero_bit(&l_mg->meta_bitmap, PBLK_DATA_LINES);
 801         set_bit(meta_line, &l_mg->meta_bitmap);
 802         smeta = l_mg->sline_meta[meta_line];
 803         emeta = l_mg->eline_meta[meta_line];
 804         smeta_buf = smeta->buf;
 805         spin_unlock(&l_mg->free_lock);
 806
 807         /* Order data lines using their sequence number */
 808         for (i = 0; i < l_mg->nr_lines; i++) {
 809                 u32 crc;
 810
 811                 line = &pblk->lines[i];
 812
 813                 memset(smeta, 0, lm->smeta_len);
 814                 line->smeta = smeta;
 815                 line->lun_bitmap = ((void *)(smeta_buf)) +
 816                                                 sizeof(struct line_smeta);
 817
 818                 /* Lines that cannot be read are assumed as not written here */
 819                 if (pblk_line_read_smeta(pblk, line))
 820                         continue;
 821
 822                 crc = pblk_calc_smeta_crc(pblk, smeta_buf);
 823                 if (le32_to_cpu(smeta_buf->crc) != crc)
 824                         continue;
 825
 826                 if (le32_to_cpu(smeta_buf->header.identifier) != PBLK_MAGIC)
 827                         continue;
 828
 829                 if (le16_to_cpu(smeta_buf->header.version) != 1) {
 830                         pr_err("pblk: found incompatible line version %u\n",
 831                                         smeta_buf->header.version);
 832                         return ERR_PTR(-EINVAL);
 833                 }
 834
 835                 /* The first valid instance uuid is used for initialization */
 836                 if (!valid_uuid) {
 837                         memcpy(pblk->instance_uuid, smeta_buf->header.uuid, 16);
 838                         valid_uuid = 1;
 839                 }
 840
 841                 if (memcmp(pblk->instance_uuid, smeta_buf->header.uuid, 16)) {
 842                         pr_debug("pblk: ignore line %u due to uuid mismatch\n",
 843                                         i);
 844                         continue;
 845                 }
 846
 847                 /* Update line metadata */
 848                 spin_lock(&line->lock);
 849                 line->id = le32_to_cpu(smeta_buf->header.id);
 850                 line->type = le16_to_cpu(smeta_buf->header.type);
 851                 line->seq_nr = le64_to_cpu(smeta_buf->seq_nr);
 852                 spin_unlock(&line->lock);
 853
 854                 /* Update general metadata */
 855                 spin_lock(&l_mg->free_lock);
 856                 if (line->seq_nr >= l_mg->d_seq_nr)
 857                         l_mg->d_seq_nr = line->seq_nr + 1;
 858                 l_mg->nr_free_lines--;
 859                 spin_unlock(&l_mg->free_lock);
 860
 861                 if (pblk_line_recov_alloc(pblk, line))
 862                         goto out;
 863
 864                 pblk_recov_line_add_ordered(&recov_list, line);
 865                 found_lines++;
 866                 pr_debug("pblk: recovering data line %d, seq:%llu\n",
 867                                                 line->id, smeta_buf->seq_nr);
 868         }
 869
 870         if (!found_lines) {
 871                 pblk_setup_uuid(pblk);
 872
 873                 spin_lock(&l_mg->free_lock);
 874                 WARN_ON_ONCE(!test_and_clear_bit(meta_line,
 875                                                         &l_mg->meta_bitmap));
 876                 spin_unlock(&l_mg->free_lock);
 877
 878                 goto out;
 879         }
 880
 881         /* Verify closed blocks and recover this portion of L2P table*/
 882         list_for_each_entry_safe(line, tline, &recov_list, list) {
 883                 int off, nr_bb;
 884
 885                 recovered_lines++;
 886                 /* Calculate where emeta starts based on the line bb */
 887                 off = lm->sec_per_line - lm->emeta_sec[0];
 888                 nr_bb = bitmap_weight(line->blk_bitmap, lm->blk_per_line);
 889                 off -= nr_bb * geo->sec_per_pl;
 890
 891                 memset(&emeta->buf, 0, lm->emeta_len[0]);
 892                 line->emeta = emeta;
 893                 line->emeta_ssec = off;
 894
 895                 if (pblk_line_read_emeta(pblk, line, line->emeta->buf)) {
 896                         pblk_recov_l2p_from_oob(pblk, line);
 897                         goto next;
 898                 }
 899
 900                 if (pblk_recov_l2p_from_emeta(pblk, line))
 901                         pblk_recov_l2p_from_oob(pblk, line);
 902
 903 next:
 904                 if (pblk_line_is_full(line)) {
 905                         struct list_head *move_list;
 906
 907                         spin_lock(&line->lock);
 908                         line->state = PBLK_LINESTATE_CLOSED;
 909                         move_list = pblk_line_gc_list(pblk, line);
 910                         spin_unlock(&line->lock);
 911
 912                         spin_lock(&l_mg->gc_lock);
 913                         list_move_tail(&line->list, move_list);
 914                         spin_unlock(&l_mg->gc_lock);
 915
 916                         mempool_free(line->map_bitmap, pblk->line_meta_pool);
 917                         line->map_bitmap = NULL;
 918                         line->smeta = NULL;
 919                         line->emeta = NULL;
 920                 } else {
 921                         if (open_lines > 1)
 922                                 pr_err("pblk: failed to recover L2P\n");
 923
 924                         open_lines++;
 925                         line->meta_line = meta_line;
 926                         data_line = line;
 927                 }
 928         }
 929
 930         spin_lock(&l_mg->free_lock);
 931         if (!open_lines) {
 932                 WARN_ON_ONCE(!test_and_clear_bit(meta_line,
 933                                                         &l_mg->meta_bitmap));
 934                 pblk_line_replace_data(pblk);
 935         } else {
 936                 /* Allocate next line for preparation */
 937                 l_mg->data_next = pblk_line_get(pblk);
 938                 if (l_mg->data_next) {
 939                         l_mg->data_next->seq_nr = l_mg->d_seq_nr++;
 940                         l_mg->data_next->type = PBLK_LINETYPE_DATA;
 941                         is_next = 1;
 942                 }
 943         }
 944         spin_unlock(&l_mg->free_lock);
 945
 946         if (is_next) {
 947                 pblk_line_erase(pblk, l_mg->data_next);
 948                 pblk_rl_free_lines_dec(&pblk->rl, l_mg->data_next);
 949         }
 950
 951 out:
 952         if (found_lines != recovered_lines)
 953                 pr_err("pblk: failed to recover all found lines %d/%d\n",
 954                                                 found_lines, recovered_lines);
 955
 956         return data_line;
 957 }
 958
 959 /*
 960  * Pad current line
 961  */
 962 int pblk_recov_pad(struct pblk *pblk)
 963 {
 964         struct nvm_tgt_dev *dev = pblk->dev;
 965         struct nvm_geo *geo = &dev->geo;
 966         struct pblk_line *line;
 967         struct pblk_line_mgmt *l_mg = &pblk->l_mg;
 968         struct nvm_rq *rqd;
 969         struct pblk_recov_alloc p;
 970         struct ppa_addr *ppa_list;
 971         struct pblk_sec_meta *meta_list;
 972         void *data;
 973         int left_msecs;
 974         int ret = 0;
 975         dma_addr_t dma_ppa_list, dma_meta_list;
 976
 977         spin_lock(&l_mg->free_lock);
 978         line = l_mg->data_line;
 979         left_msecs = line->left_msecs;
 980         spin_unlock(&l_mg->free_lock);
 981
 982         rqd = pblk_alloc_rqd(pblk, READ);
 983         if (IS_ERR(rqd))
 984                 return PTR_ERR(rqd);
 985
 986         meta_list = nvm_dev_dma_alloc(dev->parent, GFP_KERNEL, &dma_meta_list);
 987         if (!meta_list) {
 988                 ret = -ENOMEM;
 989                 goto free_rqd;
 990         }
 991
 992         ppa_list = (void *)(meta_list) + pblk_dma_meta_size;
 993         dma_ppa_list = dma_meta_list + pblk_dma_meta_size;
 994
 995         data = kcalloc(pblk->max_write_pgs, geo->sec_size, GFP_KERNEL);
 996         if (!data) {
 997                 ret = -ENOMEM;
 998                 goto free_meta_list;
 999         }
1000
1001         p.ppa_list = ppa_list;
1002         p.meta_list = meta_list;
1003         p.rqd = rqd;
1004         p.data = data;
1005         p.dma_ppa_list = dma_ppa_list;
1006         p.dma_meta_list = dma_meta_list;
1007
1008         ret = pblk_recov_pad_oob(pblk, line, p, left_msecs);
1009         if (ret) {
1010                 pr_err("pblk: Tear down padding failed (%d)\n", ret);
1011                 goto free_data;
1012         }
1013
1014         pblk_line_close_meta(pblk, line);
1015
1016 free_data:
1017         kfree(data);
1018 free_meta_list:
1019         nvm_dev_dma_free(dev->parent, meta_list, dma_meta_list);
1020 free_rqd:
1021         pblk_free_rqd(pblk, rqd, READ);
1022
1023         return ret;
1024 }