drivers/lightnvm/rrpc.c

   1 /*
   2  * Copyright (C) 2015 IT University of Copenhagen
   3  * Initial release: Matias Bjorling <m@bjorling.me>
   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  * Implementation of a Round-robin page-based Hybrid FTL for Open-channel SSDs.
  15  */
  16
  17 #include "rrpc.h"
  18
  19 static struct kmem_cache *rrpc_gcb_cache, *rrpc_rq_cache;
  20 static DECLARE_RWSEM(rrpc_lock);
  21
  22 static int rrpc_submit_io(struct rrpc *rrpc, struct bio *bio,
  23                                 struct nvm_rq *rqd, unsigned long flags);
  24
  25 #define rrpc_for_each_lun(rrpc, rlun, i) \
  26                 for ((i) = 0, rlun = &(rrpc)->luns[0]; \
  27                         (i) < (rrpc)->nr_luns; (i)++, rlun = &(rrpc)->luns[(i)])
  28
  29 static void rrpc_page_invalidate(struct rrpc *rrpc, struct rrpc_addr *a)
  30 {
  31         struct nvm_tgt_dev *dev = rrpc->dev;
  32         struct rrpc_block *rblk = a->rblk;
  33         unsigned int pg_offset;
  34
  35         lockdep_assert_held(&rrpc->rev_lock);
  36
  37         if (a->addr == ADDR_EMPTY || !rblk)
  38                 return;
  39
  40         spin_lock(&rblk->lock);
  41
  42         div_u64_rem(a->addr, dev->geo.sec_per_blk, &pg_offset);
  43         WARN_ON(test_and_set_bit(pg_offset, rblk->invalid_pages));
  44         rblk->nr_invalid_pages++;
  45
  46         spin_unlock(&rblk->lock);
  47
  48         rrpc->rev_trans_map[a->addr - rrpc->poffset].addr = ADDR_EMPTY;
  49 }
  50
  51 static void rrpc_invalidate_range(struct rrpc *rrpc, sector_t slba,
  52                                                         unsigned int len)
  53 {
  54         sector_t i;
  55
  56         spin_lock(&rrpc->rev_lock);
  57         for (i = slba; i < slba + len; i++) {
  58                 struct rrpc_addr *gp = &rrpc->trans_map[i];
  59
  60                 rrpc_page_invalidate(rrpc, gp);
  61                 gp->rblk = NULL;
  62         }
  63         spin_unlock(&rrpc->rev_lock);
  64 }
  65
  66 static struct nvm_rq *rrpc_inflight_laddr_acquire(struct rrpc *rrpc,
  67                                         sector_t laddr, unsigned int pages)
  68 {
  69         struct nvm_rq *rqd;
  70         struct rrpc_inflight_rq *inf;
  71
  72         rqd = mempool_alloc(rrpc->rq_pool, GFP_ATOMIC);
  73         if (!rqd)
  74                 return ERR_PTR(-ENOMEM);
  75
  76         inf = rrpc_get_inflight_rq(rqd);
  77         if (rrpc_lock_laddr(rrpc, laddr, pages, inf)) {
  78                 mempool_free(rqd, rrpc->rq_pool);
  79                 return NULL;
  80         }
  81
  82         return rqd;
  83 }
  84
  85 static void rrpc_inflight_laddr_release(struct rrpc *rrpc, struct nvm_rq *rqd)
  86 {
  87         struct rrpc_inflight_rq *inf = rrpc_get_inflight_rq(rqd);
  88
  89         rrpc_unlock_laddr(rrpc, inf);
  90
  91         mempool_free(rqd, rrpc->rq_pool);
  92 }
  93
  94 static void rrpc_discard(struct rrpc *rrpc, struct bio *bio)
  95 {
  96         sector_t slba = bio->bi_iter.bi_sector / NR_PHY_IN_LOG;
  97         sector_t len = bio->bi_iter.bi_size / RRPC_EXPOSED_PAGE_SIZE;
  98         struct nvm_rq *rqd;
  99
 100         while (1) {
 101                 rqd = rrpc_inflight_laddr_acquire(rrpc, slba, len);
 102                 if (rqd)
 103                         break;
 104
 105                 schedule();
 106         }
 107
 108         if (IS_ERR(rqd)) {
 109                 pr_err("rrpc: unable to acquire inflight IO\n");
 110                 bio_io_error(bio);
 111                 return;
 112         }
 113
 114         rrpc_invalidate_range(rrpc, slba, len);
 115         rrpc_inflight_laddr_release(rrpc, rqd);
 116 }
 117
 118 static int block_is_full(struct rrpc *rrpc, struct rrpc_block *rblk)
 119 {
 120         struct nvm_tgt_dev *dev = rrpc->dev;
 121
 122         return (rblk->next_page == dev->geo.sec_per_blk);
 123 }
 124
 125 /* Calculate relative addr for the given block, considering instantiated LUNs */
 126 static u64 block_to_rel_addr(struct rrpc *rrpc, struct rrpc_block *rblk)
 127 {
 128         struct nvm_tgt_dev *dev = rrpc->dev;
 129         struct nvm_block *blk = rblk->parent;
 130         int lun_blk = blk->id % (dev->geo.blks_per_lun * rrpc->nr_luns);
 131
 132         return lun_blk * dev->geo.sec_per_blk;
 133 }
 134
 135 /* Calculate global addr for the given block */
 136 static u64 block_to_addr(struct rrpc *rrpc, struct rrpc_block *rblk)
 137 {
 138         struct nvm_tgt_dev *dev = rrpc->dev;
 139         struct nvm_block *blk = rblk->parent;
 140
 141         return blk->id * dev->geo.sec_per_blk;
 142 }
 143
 144 static struct ppa_addr rrpc_ppa_to_gaddr(struct nvm_tgt_dev *dev, u64 addr)
 145 {
 146         struct ppa_addr paddr;
 147
 148         paddr.ppa = addr;
 149         return linear_to_generic_addr(&dev->geo, paddr);
 150 }
 151
 152 /* requires lun->lock taken */
 153 static void rrpc_set_lun_cur(struct rrpc_lun *rlun, struct rrpc_block *new_rblk,
 154                                                 struct rrpc_block **cur_rblk)
 155 {
 156         struct rrpc *rrpc = rlun->rrpc;
 157
 158         if (*cur_rblk) {
 159                 spin_lock(&(*cur_rblk)->lock);
 160                 WARN_ON(!block_is_full(rrpc, *cur_rblk));
 161                 spin_unlock(&(*cur_rblk)->lock);
 162         }
 163         *cur_rblk = new_rblk;
 164 }
 165
 166 static struct nvm_block *__rrpc_get_blk(struct rrpc *rrpc,
 167                                                         struct rrpc_lun *rlun)
 168 {
 169         struct nvm_lun *lun = rlun->parent;
 170         struct nvm_block *blk = NULL;
 171
 172         if (list_empty(&lun->free_list))
 173                 goto out;
 174
 175         blk = list_first_entry(&lun->free_list, struct nvm_block, list);
 176
 177         list_move_tail(&blk->list, &lun->used_list);
 178         blk->state = NVM_BLK_ST_TGT;
 179         lun->nr_free_blocks--;
 180
 181 out:
 182         return blk;
 183 }
 184
 185 static struct rrpc_block *rrpc_get_blk(struct rrpc *rrpc, struct rrpc_lun *rlun,
 186                                                         unsigned long flags)
 187 {
 188         struct nvm_tgt_dev *dev = rrpc->dev;
 189         struct nvm_lun *lun = rlun->parent;
 190         struct nvm_block *blk;
 191         struct rrpc_block *rblk;
 192         int is_gc = flags & NVM_IOTYPE_GC;
 193
 194         spin_lock(&rlun->lock);
 195         if (!is_gc && lun->nr_free_blocks < rlun->reserved_blocks) {
 196                 pr_err("nvm: rrpc: cannot give block to non GC request\n");
 197                 spin_unlock(&rlun->lock);
 198                 return NULL;
 199         }
 200
 201         blk = __rrpc_get_blk(rrpc, rlun);
 202         if (!blk) {
 203                 pr_err("nvm: rrpc: cannot get new block\n");
 204                 spin_unlock(&rlun->lock);
 205                 return NULL;
 206         }
 207         spin_unlock(&rlun->lock);
 208
 209         rblk = rrpc_get_rblk(rlun, blk->id);
 210         blk->priv = rblk;
 211         bitmap_zero(rblk->invalid_pages, dev->geo.sec_per_blk);
 212         rblk->next_page = 0;
 213         rblk->nr_invalid_pages = 0;
 214         atomic_set(&rblk->data_cmnt_size, 0);
 215
 216         return rblk;
 217 }
 218
 219 static void rrpc_put_blk(struct rrpc *rrpc, struct rrpc_block *rblk)
 220 {
 221         struct nvm_block *blk = rblk->parent;
 222         struct rrpc_lun *rlun = rblk->rlun;
 223         struct nvm_lun *lun = rlun->parent;
 224
 225         spin_lock(&rlun->lock);
 226         if (blk->state & NVM_BLK_ST_TGT) {
 227                 list_move_tail(&blk->list, &lun->free_list);
 228                 lun->nr_free_blocks++;
 229                 blk->state = NVM_BLK_ST_FREE;
 230         } else if (blk->state & NVM_BLK_ST_BAD) {
 231                 list_move_tail(&blk->list, &lun->bb_list);
 232                 blk->state = NVM_BLK_ST_BAD;
 233         } else {
 234                 WARN_ON_ONCE(1);
 235                 pr_err("rrpc: erroneous block type (%lu -> %u)\n",
 236                                                         blk->id, blk->state);
 237                 list_move_tail(&blk->list, &lun->bb_list);
 238         }
 239         spin_unlock(&rlun->lock);
 240 }
 241
 242 static void rrpc_put_blks(struct rrpc *rrpc)
 243 {
 244         struct rrpc_lun *rlun;
 245         int i;
 246
 247         for (i = 0; i < rrpc->nr_luns; i++) {
 248                 rlun = &rrpc->luns[i];
 249                 if (rlun->cur)
 250                         rrpc_put_blk(rrpc, rlun->cur);
 251                 if (rlun->gc_cur)
 252                         rrpc_put_blk(rrpc, rlun->gc_cur);
 253         }
 254 }
 255
 256 static struct rrpc_lun *get_next_lun(struct rrpc *rrpc)
 257 {
 258         int next = atomic_inc_return(&rrpc->next_lun);
 259
 260         return &rrpc->luns[next % rrpc->nr_luns];
 261 }
 262
 263 static void rrpc_gc_kick(struct rrpc *rrpc)
 264 {
 265         struct rrpc_lun *rlun;
 266         unsigned int i;
 267
 268         for (i = 0; i < rrpc->nr_luns; i++) {
 269                 rlun = &rrpc->luns[i];
 270                 queue_work(rrpc->krqd_wq, &rlun->ws_gc);
 271         }
 272 }
 273
 274 /*
 275  * timed GC every interval.
 276  */
 277 static void rrpc_gc_timer(unsigned long data)
 278 {
 279         struct rrpc *rrpc = (struct rrpc *)data;
 280
 281         rrpc_gc_kick(rrpc);
 282         mod_timer(&rrpc->gc_timer, jiffies + msecs_to_jiffies(10));
 283 }
 284
 285 static void rrpc_end_sync_bio(struct bio *bio)
 286 {
 287         struct completion *waiting = bio->bi_private;
 288
 289         if (bio->bi_error)
 290                 pr_err("nvm: gc request failed (%u).\n", bio->bi_error);
 291
 292         complete(waiting);
 293 }
 294
 295 /*
 296  * rrpc_move_valid_pages -- migrate live data off the block
 297  * @rrpc: the 'rrpc' structure
 298  * @block: the block from which to migrate live pages
 299  *
 300  * Description:
 301  *   GC algorithms may call this function to migrate remaining live
 302  *   pages off the block prior to erasing it. This function blocks
 303  *   further execution until the operation is complete.
 304  */
 305 static int rrpc_move_valid_pages(struct rrpc *rrpc, struct rrpc_block *rblk)
 306 {
 307         struct nvm_tgt_dev *dev = rrpc->dev;
 308         struct request_queue *q = dev->q;
 309         struct rrpc_rev_addr *rev;
 310         struct nvm_rq *rqd;
 311         struct bio *bio;
 312         struct page *page;
 313         int slot;
 314         int nr_sec_per_blk = dev->geo.sec_per_blk;
 315         u64 phys_addr;
 316         DECLARE_COMPLETION_ONSTACK(wait);
 317
 318         if (bitmap_full(rblk->invalid_pages, nr_sec_per_blk))
 319                 return 0;
 320
 321         bio = bio_alloc(GFP_NOIO, 1);
 322         if (!bio) {
 323                 pr_err("nvm: could not alloc bio to gc\n");
 324                 return -ENOMEM;
 325         }
 326
 327         page = mempool_alloc(rrpc->page_pool, GFP_NOIO);
 328         if (!page) {
 329                 bio_put(bio);
 330                 return -ENOMEM;
 331         }
 332
 333         while ((slot = find_first_zero_bit(rblk->invalid_pages,
 334                                             nr_sec_per_blk)) < nr_sec_per_blk) {
 335
 336                 /* Lock laddr */
 337                 phys_addr = rblk->parent->id * nr_sec_per_blk + slot;
 338
 339 try:
 340                 spin_lock(&rrpc->rev_lock);
 341                 /* Get logical address from physical to logical table */
 342                 rev = &rrpc->rev_trans_map[phys_addr - rrpc->poffset];
 343                 /* already updated by previous regular write */
 344                 if (rev->addr == ADDR_EMPTY) {
 345                         spin_unlock(&rrpc->rev_lock);
 346                         continue;
 347                 }
 348
 349                 rqd = rrpc_inflight_laddr_acquire(rrpc, rev->addr, 1);
 350                 if (IS_ERR_OR_NULL(rqd)) {
 351                         spin_unlock(&rrpc->rev_lock);
 352                         schedule();
 353                         goto try;
 354                 }
 355
 356                 spin_unlock(&rrpc->rev_lock);
 357
 358                 /* Perform read to do GC */
 359                 bio->bi_iter.bi_sector = rrpc_get_sector(rev->addr);
 360                 bio_set_op_attrs(bio,  REQ_OP_READ, 0);
 361                 bio->bi_private = &wait;
 362                 bio->bi_end_io = rrpc_end_sync_bio;
 363
 364                 /* TODO: may fail when EXP_PG_SIZE > PAGE_SIZE */
 365                 bio_add_pc_page(q, bio, page, RRPC_EXPOSED_PAGE_SIZE, 0);
 366
 367                 if (rrpc_submit_io(rrpc, bio, rqd, NVM_IOTYPE_GC)) {
 368                         pr_err("rrpc: gc read failed.\n");
 369                         rrpc_inflight_laddr_release(rrpc, rqd);
 370                         goto finished;
 371                 }
 372                 wait_for_completion_io(&wait);
 373                 if (bio->bi_error) {
 374                         rrpc_inflight_laddr_release(rrpc, rqd);
 375                         goto finished;
 376                 }
 377
 378                 bio_reset(bio);
 379                 reinit_completion(&wait);
 380
 381                 bio->bi_iter.bi_sector = rrpc_get_sector(rev->addr);
 382                 bio_set_op_attrs(bio, REQ_OP_WRITE, 0);
 383                 bio->bi_private = &wait;
 384                 bio->bi_end_io = rrpc_end_sync_bio;
 385
 386                 bio_add_pc_page(q, bio, page, RRPC_EXPOSED_PAGE_SIZE, 0);
 387
 388                 /* turn the command around and write the data back to a new
 389                  * address
 390                  */
 391                 if (rrpc_submit_io(rrpc, bio, rqd, NVM_IOTYPE_GC)) {
 392                         pr_err("rrpc: gc write failed.\n");
 393                         rrpc_inflight_laddr_release(rrpc, rqd);
 394                         goto finished;
 395                 }
 396                 wait_for_completion_io(&wait);
 397
 398                 rrpc_inflight_laddr_release(rrpc, rqd);
 399                 if (bio->bi_error)
 400                         goto finished;
 401
 402                 bio_reset(bio);
 403         }
 404
 405 finished:
 406         mempool_free(page, rrpc->page_pool);
 407         bio_put(bio);
 408
 409         if (!bitmap_full(rblk->invalid_pages, nr_sec_per_blk)) {
 410                 pr_err("nvm: failed to garbage collect block\n");
 411                 return -EIO;
 412         }
 413
 414         return 0;
 415 }
 416
 417 static void rrpc_block_gc(struct work_struct *work)
 418 {
 419         struct rrpc_block_gc *gcb = container_of(work, struct rrpc_block_gc,
 420                                                                         ws_gc);
 421         struct rrpc *rrpc = gcb->rrpc;
 422         struct rrpc_block *rblk = gcb->rblk;
 423         struct rrpc_lun *rlun = rblk->rlun;
 424         struct nvm_tgt_dev *dev = rrpc->dev;
 425
 426         mempool_free(gcb, rrpc->gcb_pool);
 427         pr_debug("nvm: block '%lu' being reclaimed\n", rblk->parent->id);
 428
 429         if (rrpc_move_valid_pages(rrpc, rblk))
 430                 goto put_back;
 431
 432         if (nvm_erase_blk(dev->parent, rblk->parent, 0))
 433                 goto put_back;
 434
 435         rrpc_put_blk(rrpc, rblk);
 436
 437         return;
 438
 439 put_back:
 440         spin_lock(&rlun->lock);
 441         list_add_tail(&rblk->prio, &rlun->prio_list);
 442         spin_unlock(&rlun->lock);
 443 }
 444
 445 /* the block with highest number of invalid pages, will be in the beginning
 446  * of the list
 447  */
 448 static struct rrpc_block *rblock_max_invalid(struct rrpc_block *ra,
 449                                                         struct rrpc_block *rb)
 450 {
 451         if (ra->nr_invalid_pages == rb->nr_invalid_pages)
 452                 return ra;
 453
 454         return (ra->nr_invalid_pages < rb->nr_invalid_pages) ? rb : ra;
 455 }
 456
 457 /* linearly find the block with highest number of invalid pages
 458  * requires lun->lock
 459  */
 460 static struct rrpc_block *block_prio_find_max(struct rrpc_lun *rlun)
 461 {
 462         struct list_head *prio_list = &rlun->prio_list;
 463         struct rrpc_block *rblock, *max;
 464
 465         BUG_ON(list_empty(prio_list));
 466
 467         max = list_first_entry(prio_list, struct rrpc_block, prio);
 468         list_for_each_entry(rblock, prio_list, prio)
 469                 max = rblock_max_invalid(max, rblock);
 470
 471         return max;
 472 }
 473
 474 static void rrpc_lun_gc(struct work_struct *work)
 475 {
 476         struct rrpc_lun *rlun = container_of(work, struct rrpc_lun, ws_gc);
 477         struct rrpc *rrpc = rlun->rrpc;
 478         struct nvm_tgt_dev *dev = rrpc->dev;
 479         struct nvm_lun *lun = rlun->parent;
 480         struct rrpc_block_gc *gcb;
 481         unsigned int nr_blocks_need;
 482
 483         nr_blocks_need = dev->geo.blks_per_lun / GC_LIMIT_INVERSE;
 484
 485         if (nr_blocks_need < rrpc->nr_luns)
 486                 nr_blocks_need = rrpc->nr_luns;
 487
 488         spin_lock(&rlun->lock);
 489         while (nr_blocks_need > lun->nr_free_blocks &&
 490                                         !list_empty(&rlun->prio_list)) {
 491                 struct rrpc_block *rblock = block_prio_find_max(rlun);
 492                 struct nvm_block *block = rblock->parent;
 493
 494                 if (!rblock->nr_invalid_pages)
 495                         break;
 496
 497                 gcb = mempool_alloc(rrpc->gcb_pool, GFP_ATOMIC);
 498                 if (!gcb)
 499                         break;
 500
 501                 list_del_init(&rblock->prio);
 502
 503                 BUG_ON(!block_is_full(rrpc, rblock));
 504
 505                 pr_debug("rrpc: selected block '%lu' for GC\n", block->id);
 506
 507                 gcb->rrpc = rrpc;
 508                 gcb->rblk = rblock;
 509                 INIT_WORK(&gcb->ws_gc, rrpc_block_gc);
 510
 511                 queue_work(rrpc->kgc_wq, &gcb->ws_gc);
 512
 513                 nr_blocks_need--;
 514         }
 515         spin_unlock(&rlun->lock);
 516
 517         /* TODO: Hint that request queue can be started again */
 518 }
 519
 520 static void rrpc_gc_queue(struct work_struct *work)
 521 {
 522         struct rrpc_block_gc *gcb = container_of(work, struct rrpc_block_gc,
 523                                                                         ws_gc);
 524         struct rrpc *rrpc = gcb->rrpc;
 525         struct rrpc_block *rblk = gcb->rblk;
 526         struct rrpc_lun *rlun = rblk->rlun;
 527
 528         spin_lock(&rlun->lock);
 529         list_add_tail(&rblk->prio, &rlun->prio_list);
 530         spin_unlock(&rlun->lock);
 531
 532         mempool_free(gcb, rrpc->gcb_pool);
 533         pr_debug("nvm: block '%lu' is full, allow GC (sched)\n",
 534                                                         rblk->parent->id);
 535 }
 536
 537 static const struct block_device_operations rrpc_fops = {
 538         .owner          = THIS_MODULE,
 539 };
 540
 541 static struct rrpc_lun *rrpc_get_lun_rr(struct rrpc *rrpc, int is_gc)
 542 {
 543         unsigned int i;
 544         struct rrpc_lun *rlun, *max_free;
 545
 546         if (!is_gc)
 547                 return get_next_lun(rrpc);
 548
 549         /* during GC, we don't care about RR, instead we want to make
 550          * sure that we maintain evenness between the block luns.
 551          */
 552         max_free = &rrpc->luns[0];
 553         /* prevent GC-ing lun from devouring pages of a lun with
 554          * little free blocks. We don't take the lock as we only need an
 555          * estimate.
 556          */
 557         rrpc_for_each_lun(rrpc, rlun, i) {
 558                 if (rlun->parent->nr_free_blocks >
 559                                         max_free->parent->nr_free_blocks)
 560                         max_free = rlun;
 561         }
 562
 563         return max_free;
 564 }
 565
 566 static struct rrpc_addr *rrpc_update_map(struct rrpc *rrpc, sector_t laddr,
 567                                         struct rrpc_block *rblk, u64 paddr)
 568 {
 569         struct rrpc_addr *gp;
 570         struct rrpc_rev_addr *rev;
 571
 572         BUG_ON(laddr >= rrpc->nr_sects);
 573
 574         gp = &rrpc->trans_map[laddr];
 575         spin_lock(&rrpc->rev_lock);
 576         if (gp->rblk)
 577                 rrpc_page_invalidate(rrpc, gp);
 578
 579         gp->addr = paddr;
 580         gp->rblk = rblk;
 581
 582         rev = &rrpc->rev_trans_map[gp->addr - rrpc->poffset];
 583         rev->addr = laddr;
 584         spin_unlock(&rrpc->rev_lock);
 585
 586         return gp;
 587 }
 588
 589 static u64 rrpc_alloc_addr(struct rrpc *rrpc, struct rrpc_block *rblk)
 590 {
 591         u64 addr = ADDR_EMPTY;
 592
 593         spin_lock(&rblk->lock);
 594         if (block_is_full(rrpc, rblk))
 595                 goto out;
 596
 597         addr = block_to_addr(rrpc, rblk) + rblk->next_page;
 598
 599         rblk->next_page++;
 600 out:
 601         spin_unlock(&rblk->lock);
 602         return addr;
 603 }
 604
 605 /* Map logical address to a physical page. The mapping implements a round robin
 606  * approach and allocates a page from the next lun available.
 607  *
 608  * Returns rrpc_addr with the physical address and block. Returns NULL if no
 609  * blocks in the next rlun are available.
 610  */
 611 static struct rrpc_addr *rrpc_map_page(struct rrpc *rrpc, sector_t laddr,
 612                                                                 int is_gc)
 613 {
 614         struct rrpc_lun *rlun;
 615         struct rrpc_block *rblk, **cur_rblk;
 616         struct nvm_lun *lun;
 617         u64 paddr;
 618         int gc_force = 0;
 619
 620         rlun = rrpc_get_lun_rr(rrpc, is_gc);
 621         lun = rlun->parent;
 622
 623         if (!is_gc && lun->nr_free_blocks < rrpc->nr_luns * 4)
 624                 return NULL;
 625
 626         /*
 627          * page allocation steps:
 628          * 1. Try to allocate new page from current rblk
 629          * 2a. If succeed, proceed to map it in and return
 630          * 2b. If fail, first try to allocate a new block from media manger,
 631          *     and then retry step 1. Retry until the normal block pool is
 632          *     exhausted.
 633          * 3. If exhausted, and garbage collector is requesting the block,
 634          *    go to the reserved block and retry step 1.
 635          *    In the case that this fails as well, or it is not GC
 636          *    requesting, report not able to retrieve a block and let the
 637          *    caller handle further processing.
 638          */
 639
 640         spin_lock(&rlun->lock);
 641         cur_rblk = &rlun->cur;
 642         rblk = rlun->cur;
 643 retry:
 644         paddr = rrpc_alloc_addr(rrpc, rblk);
 645
 646         if (paddr != ADDR_EMPTY)
 647                 goto done;
 648
 649         if (!list_empty(&rlun->wblk_list)) {
 650 new_blk:
 651                 rblk = list_first_entry(&rlun->wblk_list, struct rrpc_block,
 652                                                                         prio);
 653                 rrpc_set_lun_cur(rlun, rblk, cur_rblk);
 654                 list_del(&rblk->prio);
 655                 goto retry;
 656         }
 657         spin_unlock(&rlun->lock);
 658
 659         rblk = rrpc_get_blk(rrpc, rlun, gc_force);
 660         if (rblk) {
 661                 spin_lock(&rlun->lock);
 662                 list_add_tail(&rblk->prio, &rlun->wblk_list);
 663                 /*
 664                  * another thread might already have added a new block,
 665                  * Therefore, make sure that one is used, instead of the
 666                  * one just added.
 667                  */
 668                 goto new_blk;
 669         }
 670
 671         if (unlikely(is_gc) && !gc_force) {
 672                 /* retry from emergency gc block */
 673                 cur_rblk = &rlun->gc_cur;
 674                 rblk = rlun->gc_cur;
 675                 gc_force = 1;
 676                 spin_lock(&rlun->lock);
 677                 goto retry;
 678         }
 679
 680         pr_err("rrpc: failed to allocate new block\n");
 681         return NULL;
 682 done:
 683         spin_unlock(&rlun->lock);
 684         return rrpc_update_map(rrpc, laddr, rblk, paddr);
 685 }
 686
 687 static void rrpc_run_gc(struct rrpc *rrpc, struct rrpc_block *rblk)
 688 {
 689         struct rrpc_block_gc *gcb;
 690
 691         gcb = mempool_alloc(rrpc->gcb_pool, GFP_ATOMIC);
 692         if (!gcb) {
 693                 pr_err("rrpc: unable to queue block for gc.");
 694                 return;
 695         }
 696
 697         gcb->rrpc = rrpc;
 698         gcb->rblk = rblk;
 699
 700         INIT_WORK(&gcb->ws_gc, rrpc_gc_queue);
 701         queue_work(rrpc->kgc_wq, &gcb->ws_gc);
 702 }
 703
 704 static void __rrpc_mark_bad_block(struct nvm_tgt_dev *dev, struct ppa_addr *ppa)
 705 {
 706                 nvm_mark_blk(dev->parent, *ppa, NVM_BLK_ST_BAD);
 707                 nvm_set_bb_tbl(dev->parent, ppa, 1, NVM_BLK_T_GRWN_BAD);
 708 }
 709
 710 static void rrpc_mark_bad_block(struct rrpc *rrpc, struct nvm_rq *rqd)
 711 {
 712         struct nvm_tgt_dev *dev = rrpc->dev;
 713         void *comp_bits = &rqd->ppa_status;
 714         struct ppa_addr ppa, prev_ppa;
 715         int nr_ppas = rqd->nr_ppas;
 716         int bit;
 717
 718         if (rqd->nr_ppas == 1)
 719                 __rrpc_mark_bad_block(dev, &rqd->ppa_addr);
 720
 721         ppa_set_empty(&prev_ppa);
 722         bit = -1;
 723         while ((bit = find_next_bit(comp_bits, nr_ppas, bit + 1)) < nr_ppas) {
 724                 ppa = rqd->ppa_list[bit];
 725                 if (ppa_cmp_blk(ppa, prev_ppa))
 726                         continue;
 727
 728                 __rrpc_mark_bad_block(dev, &ppa);
 729         }
 730 }
 731
 732 static void rrpc_end_io_write(struct rrpc *rrpc, struct rrpc_rq *rrqd,
 733                                                 sector_t laddr, uint8_t npages)
 734 {
 735         struct nvm_tgt_dev *dev = rrpc->dev;
 736         struct rrpc_addr *p;
 737         struct rrpc_block *rblk;
 738         struct nvm_lun *lun;
 739         int cmnt_size, i;
 740
 741         for (i = 0; i < npages; i++) {
 742                 p = &rrpc->trans_map[laddr + i];
 743                 rblk = p->rblk;
 744                 lun = rblk->parent->lun;
 745
 746                 cmnt_size = atomic_inc_return(&rblk->data_cmnt_size);
 747                 if (unlikely(cmnt_size == dev->geo.sec_per_blk))
 748                         rrpc_run_gc(rrpc, rblk);
 749         }
 750 }
 751
 752 static void rrpc_end_io(struct nvm_rq *rqd)
 753 {
 754         struct rrpc *rrpc = container_of(rqd->ins, struct rrpc, instance);
 755         struct nvm_tgt_dev *dev = rrpc->dev;
 756         struct rrpc_rq *rrqd = nvm_rq_to_pdu(rqd);
 757         uint8_t npages = rqd->nr_ppas;
 758         sector_t laddr = rrpc_get_laddr(rqd->bio) - npages;
 759
 760         if (bio_data_dir(rqd->bio) == WRITE) {
 761                 if (rqd->error == NVM_RSP_ERR_FAILWRITE)
 762                         rrpc_mark_bad_block(rrpc, rqd);
 763
 764                 rrpc_end_io_write(rrpc, rrqd, laddr, npages);
 765         }
 766
 767         bio_put(rqd->bio);
 768
 769         if (rrqd->flags & NVM_IOTYPE_GC)
 770                 return;
 771
 772         rrpc_unlock_rq(rrpc, rqd);
 773
 774         if (npages > 1)
 775                 nvm_dev_dma_free(dev->parent, rqd->ppa_list, rqd->dma_ppa_list);
 776
 777         mempool_free(rqd, rrpc->rq_pool);
 778 }
 779
 780 static int rrpc_read_ppalist_rq(struct rrpc *rrpc, struct bio *bio,
 781                         struct nvm_rq *rqd, unsigned long flags, int npages)
 782 {
 783         struct nvm_tgt_dev *dev = rrpc->dev;
 784         struct rrpc_inflight_rq *r = rrpc_get_inflight_rq(rqd);
 785         struct rrpc_addr *gp;
 786         sector_t laddr = rrpc_get_laddr(bio);
 787         int is_gc = flags & NVM_IOTYPE_GC;
 788         int i;
 789
 790         if (!is_gc && rrpc_lock_rq(rrpc, bio, rqd)) {
 791                 nvm_dev_dma_free(dev->parent, rqd->ppa_list, rqd->dma_ppa_list);
 792                 return NVM_IO_REQUEUE;
 793         }
 794
 795         for (i = 0; i < npages; i++) {
 796                 /* We assume that mapping occurs at 4KB granularity */
 797                 BUG_ON(!(laddr + i >= 0 && laddr + i < rrpc->nr_sects));
 798                 gp = &rrpc->trans_map[laddr + i];
 799
 800                 if (gp->rblk) {
 801                         rqd->ppa_list[i] = rrpc_ppa_to_gaddr(dev, gp->addr);
 802                 } else {
 803                         BUG_ON(is_gc);
 804                         rrpc_unlock_laddr(rrpc, r);
 805                         nvm_dev_dma_free(dev->parent, rqd->ppa_list,
 806                                                         rqd->dma_ppa_list);
 807                         return NVM_IO_DONE;
 808                 }
 809         }
 810
 811         rqd->opcode = NVM_OP_HBREAD;
 812
 813         return NVM_IO_OK;
 814 }
 815
 816 static int rrpc_read_rq(struct rrpc *rrpc, struct bio *bio, struct nvm_rq *rqd,
 817                                                         unsigned long flags)
 818 {
 819         int is_gc = flags & NVM_IOTYPE_GC;
 820         sector_t laddr = rrpc_get_laddr(bio);
 821         struct rrpc_addr *gp;
 822
 823         if (!is_gc && rrpc_lock_rq(rrpc, bio, rqd))
 824                 return NVM_IO_REQUEUE;
 825
 826         BUG_ON(!(laddr >= 0 && laddr < rrpc->nr_sects));
 827         gp = &rrpc->trans_map[laddr];
 828
 829         if (gp->rblk) {
 830                 rqd->ppa_addr = rrpc_ppa_to_gaddr(rrpc->dev, gp->addr);
 831         } else {
 832                 BUG_ON(is_gc);
 833                 rrpc_unlock_rq(rrpc, rqd);
 834                 return NVM_IO_DONE;
 835         }
 836
 837         rqd->opcode = NVM_OP_HBREAD;
 838
 839         return NVM_IO_OK;
 840 }
 841
 842 static int rrpc_write_ppalist_rq(struct rrpc *rrpc, struct bio *bio,
 843                         struct nvm_rq *rqd, unsigned long flags, int npages)
 844 {
 845         struct nvm_tgt_dev *dev = rrpc->dev;
 846         struct rrpc_inflight_rq *r = rrpc_get_inflight_rq(rqd);
 847         struct rrpc_addr *p;
 848         sector_t laddr = rrpc_get_laddr(bio);
 849         int is_gc = flags & NVM_IOTYPE_GC;
 850         int i;
 851
 852         if (!is_gc && rrpc_lock_rq(rrpc, bio, rqd)) {
 853                 nvm_dev_dma_free(dev->parent, rqd->ppa_list, rqd->dma_ppa_list);
 854                 return NVM_IO_REQUEUE;
 855         }
 856
 857         for (i = 0; i < npages; i++) {
 858                 /* We assume that mapping occurs at 4KB granularity */
 859                 p = rrpc_map_page(rrpc, laddr + i, is_gc);
 860                 if (!p) {
 861                         BUG_ON(is_gc);
 862                         rrpc_unlock_laddr(rrpc, r);
 863                         nvm_dev_dma_free(dev->parent, rqd->ppa_list,
 864                                                         rqd->dma_ppa_list);
 865                         rrpc_gc_kick(rrpc);
 866                         return NVM_IO_REQUEUE;
 867                 }
 868
 869                 rqd->ppa_list[i] = rrpc_ppa_to_gaddr(dev, p->addr);
 870         }
 871
 872         rqd->opcode = NVM_OP_HBWRITE;
 873
 874         return NVM_IO_OK;
 875 }
 876
 877 static int rrpc_write_rq(struct rrpc *rrpc, struct bio *bio,
 878                                 struct nvm_rq *rqd, unsigned long flags)
 879 {
 880         struct rrpc_addr *p;
 881         int is_gc = flags & NVM_IOTYPE_GC;
 882         sector_t laddr = rrpc_get_laddr(bio);
 883
 884         if (!is_gc && rrpc_lock_rq(rrpc, bio, rqd))
 885                 return NVM_IO_REQUEUE;
 886
 887         p = rrpc_map_page(rrpc, laddr, is_gc);
 888         if (!p) {
 889                 BUG_ON(is_gc);
 890                 rrpc_unlock_rq(rrpc, rqd);
 891                 rrpc_gc_kick(rrpc);
 892                 return NVM_IO_REQUEUE;
 893         }
 894
 895         rqd->ppa_addr = rrpc_ppa_to_gaddr(rrpc->dev, p->addr);
 896         rqd->opcode = NVM_OP_HBWRITE;
 897
 898         return NVM_IO_OK;
 899 }
 900
 901 static int rrpc_setup_rq(struct rrpc *rrpc, struct bio *bio,
 902                         struct nvm_rq *rqd, unsigned long flags, uint8_t npages)
 903 {
 904         struct nvm_tgt_dev *dev = rrpc->dev;
 905
 906         if (npages > 1) {
 907                 rqd->ppa_list = nvm_dev_dma_alloc(dev->parent, GFP_KERNEL,
 908                                                         &rqd->dma_ppa_list);
 909                 if (!rqd->ppa_list) {
 910                         pr_err("rrpc: not able to allocate ppa list\n");
 911                         return NVM_IO_ERR;
 912                 }
 913
 914                 if (bio_op(bio) == REQ_OP_WRITE)
 915                         return rrpc_write_ppalist_rq(rrpc, bio, rqd, flags,
 916                                                                         npages);
 917
 918                 return rrpc_read_ppalist_rq(rrpc, bio, rqd, flags, npages);
 919         }
 920
 921         if (bio_op(bio) == REQ_OP_WRITE)
 922                 return rrpc_write_rq(rrpc, bio, rqd, flags);
 923
 924         return rrpc_read_rq(rrpc, bio, rqd, flags);
 925 }
 926
 927 static int rrpc_submit_io(struct rrpc *rrpc, struct bio *bio,
 928                                 struct nvm_rq *rqd, unsigned long flags)
 929 {
 930         struct nvm_tgt_dev *dev = rrpc->dev;
 931         struct rrpc_rq *rrq = nvm_rq_to_pdu(rqd);
 932         uint8_t nr_pages = rrpc_get_pages(bio);
 933         int bio_size = bio_sectors(bio) << 9;
 934         int err;
 935
 936         if (bio_size < dev->geo.sec_size)
 937                 return NVM_IO_ERR;
 938         else if (bio_size > dev->geo.max_rq_size)
 939                 return NVM_IO_ERR;
 940
 941         err = rrpc_setup_rq(rrpc, bio, rqd, flags, nr_pages);
 942         if (err)
 943                 return err;
 944
 945         bio_get(bio);
 946         rqd->bio = bio;
 947         rqd->ins = &rrpc->instance;
 948         rqd->nr_ppas = nr_pages;
 949         rrq->flags = flags;
 950
 951         err = nvm_submit_io(dev->parent, rqd);
 952         if (err) {
 953                 pr_err("rrpc: I/O submission failed: %d\n", err);
 954                 bio_put(bio);
 955                 if (!(flags & NVM_IOTYPE_GC)) {
 956                         rrpc_unlock_rq(rrpc, rqd);
 957                         if (rqd->nr_ppas > 1)
 958                                 nvm_dev_dma_free(dev->parent,
 959                                         rqd->ppa_list, rqd->dma_ppa_list);
 960                 }
 961                 return NVM_IO_ERR;
 962         }
 963
 964         return NVM_IO_OK;
 965 }
 966
 967 static blk_qc_t rrpc_make_rq(struct request_queue *q, struct bio *bio)
 968 {
 969         struct rrpc *rrpc = q->queuedata;
 970         struct nvm_rq *rqd;
 971         int err;
 972
 973         blk_queue_split(q, &bio, q->bio_split);
 974
 975         if (bio_op(bio) == REQ_OP_DISCARD) {
 976                 rrpc_discard(rrpc, bio);
 977                 return BLK_QC_T_NONE;
 978         }
 979
 980         rqd = mempool_alloc(rrpc->rq_pool, GFP_KERNEL);
 981         if (!rqd) {
 982                 pr_err_ratelimited("rrpc: not able to queue bio.");
 983                 bio_io_error(bio);
 984                 return BLK_QC_T_NONE;
 985         }
 986         memset(rqd, 0, sizeof(struct nvm_rq));
 987
 988         err = rrpc_submit_io(rrpc, bio, rqd, NVM_IOTYPE_NONE);
 989         switch (err) {
 990         case NVM_IO_OK:
 991                 return BLK_QC_T_NONE;
 992         case NVM_IO_ERR:
 993                 bio_io_error(bio);
 994                 break;
 995         case NVM_IO_DONE:
 996                 bio_endio(bio);
 997                 break;
 998         case NVM_IO_REQUEUE:
 999                 spin_lock(&rrpc->bio_lock);
1000                 bio_list_add(&rrpc->requeue_bios, bio);
1001                 spin_unlock(&rrpc->bio_lock);
1002                 queue_work(rrpc->kgc_wq, &rrpc->ws_requeue);
1003                 break;
1004         }
1005
1006         mempool_free(rqd, rrpc->rq_pool);
1007         return BLK_QC_T_NONE;
1008 }
1009
1010 static void rrpc_requeue(struct work_struct *work)
1011 {
1012         struct rrpc *rrpc = container_of(work, struct rrpc, ws_requeue);
1013         struct bio_list bios;
1014         struct bio *bio;
1015
1016         bio_list_init(&bios);
1017
1018         spin_lock(&rrpc->bio_lock);
1019         bio_list_merge(&bios, &rrpc->requeue_bios);
1020         bio_list_init(&rrpc->requeue_bios);
1021         spin_unlock(&rrpc->bio_lock);
1022
1023         while ((bio = bio_list_pop(&bios)))
1024                 rrpc_make_rq(rrpc->disk->queue, bio);
1025 }
1026
1027 static void rrpc_gc_free(struct rrpc *rrpc)
1028 {
1029         if (rrpc->krqd_wq)
1030                 destroy_workqueue(rrpc->krqd_wq);
1031
1032         if (rrpc->kgc_wq)
1033                 destroy_workqueue(rrpc->kgc_wq);
1034 }
1035
1036 static int rrpc_gc_init(struct rrpc *rrpc)
1037 {
1038         rrpc->krqd_wq = alloc_workqueue("rrpc-lun", WQ_MEM_RECLAIM|WQ_UNBOUND,
1039                                                                 rrpc->nr_luns);
1040         if (!rrpc->krqd_wq)
1041                 return -ENOMEM;
1042
1043         rrpc->kgc_wq = alloc_workqueue("rrpc-bg", WQ_MEM_RECLAIM, 1);
1044         if (!rrpc->kgc_wq)
1045                 return -ENOMEM;
1046
1047         setup_timer(&rrpc->gc_timer, rrpc_gc_timer, (unsigned long)rrpc);
1048
1049         return 0;
1050 }
1051
1052 static void rrpc_map_free(struct rrpc *rrpc)
1053 {
1054         vfree(rrpc->rev_trans_map);
1055         vfree(rrpc->trans_map);
1056 }
1057
1058 static int rrpc_l2p_update(u64 slba, u32 nlb, __le64 *entries, void *private)
1059 {
1060         struct rrpc *rrpc = (struct rrpc *)private;
1061         struct nvm_tgt_dev *dev = rrpc->dev;
1062         struct rrpc_addr *addr = rrpc->trans_map + slba;
1063         struct rrpc_rev_addr *raddr = rrpc->rev_trans_map;
1064         u64 i;
1065
1066         for (i = 0; i < nlb; i++) {
1067                 u64 pba = le64_to_cpu(entries[i]);
1068                 unsigned int mod;
1069                 /* LNVM treats address-spaces as silos, LBA and PBA are
1070                  * equally large and zero-indexed.
1071                  */
1072                 if (unlikely(pba >= dev->total_secs && pba != U64_MAX)) {
1073                         pr_err("nvm: L2P data entry is out of bounds!\n");
1074                         return -EINVAL;
1075                 }
1076
1077                 /* Address zero is a special one. The first page on a disk is
1078                  * protected. As it often holds internal device boot
1079                  * information.
1080                  */
1081                 if (!pba)
1082                         continue;
1083
1084                 div_u64_rem(pba, rrpc->nr_sects, &mod);
1085
1086                 addr[i].addr = pba;
1087                 raddr[mod].addr = slba + i;
1088         }
1089
1090         return 0;
1091 }
1092
1093 static int rrpc_map_init(struct rrpc *rrpc)
1094 {
1095         struct nvm_tgt_dev *dev = rrpc->dev;
1096         sector_t i;
1097         int ret;
1098
1099         rrpc->trans_map = vzalloc(sizeof(struct rrpc_addr) * rrpc->nr_sects);
1100         if (!rrpc->trans_map)
1101                 return -ENOMEM;
1102
1103         rrpc->rev_trans_map = vmalloc(sizeof(struct rrpc_rev_addr)
1104                                                         * rrpc->nr_sects);
1105         if (!rrpc->rev_trans_map)
1106                 return -ENOMEM;
1107
1108         for (i = 0; i < rrpc->nr_sects; i++) {
1109                 struct rrpc_addr *p = &rrpc->trans_map[i];
1110                 struct rrpc_rev_addr *r = &rrpc->rev_trans_map[i];
1111
1112                 p->addr = ADDR_EMPTY;
1113                 r->addr = ADDR_EMPTY;
1114         }
1115
1116         if (!dev->ops->get_l2p_tbl)
1117                 return 0;
1118
1119         /* Bring up the mapping table from device */
1120         ret = dev->ops->get_l2p_tbl(dev->parent, rrpc->soffset, rrpc->nr_sects,
1121                                         rrpc_l2p_update, rrpc);
1122         if (ret) {
1123                 pr_err("nvm: rrpc: could not read L2P table.\n");
1124                 return -EINVAL;
1125         }
1126
1127         return 0;
1128 }
1129
1130 /* Minimum pages needed within a lun */
1131 #define PAGE_POOL_SIZE 16
1132 #define ADDR_POOL_SIZE 64
1133
1134 static int rrpc_core_init(struct rrpc *rrpc)
1135 {
1136         down_write(&rrpc_lock);
1137         if (!rrpc_gcb_cache) {
1138                 rrpc_gcb_cache = kmem_cache_create("rrpc_gcb",
1139                                 sizeof(struct rrpc_block_gc), 0, 0, NULL);
1140                 if (!rrpc_gcb_cache) {
1141                         up_write(&rrpc_lock);
1142                         return -ENOMEM;
1143                 }
1144
1145                 rrpc_rq_cache = kmem_cache_create("rrpc_rq",
1146                                 sizeof(struct nvm_rq) + sizeof(struct rrpc_rq),
1147                                 0, 0, NULL);
1148                 if (!rrpc_rq_cache) {
1149                         kmem_cache_destroy(rrpc_gcb_cache);
1150                         up_write(&rrpc_lock);
1151                         return -ENOMEM;
1152                 }
1153         }
1154         up_write(&rrpc_lock);
1155
1156         rrpc->page_pool = mempool_create_page_pool(PAGE_POOL_SIZE, 0);
1157         if (!rrpc->page_pool)
1158                 return -ENOMEM;
1159
1160         rrpc->gcb_pool = mempool_create_slab_pool(rrpc->dev->geo.nr_luns,
1161                                                                 rrpc_gcb_cache);
1162         if (!rrpc->gcb_pool)
1163                 return -ENOMEM;
1164
1165         rrpc->rq_pool = mempool_create_slab_pool(64, rrpc_rq_cache);
1166         if (!rrpc->rq_pool)
1167                 return -ENOMEM;
1168
1169         spin_lock_init(&rrpc->inflights.lock);
1170         INIT_LIST_HEAD(&rrpc->inflights.reqs);
1171
1172         return 0;
1173 }
1174
1175 static void rrpc_core_free(struct rrpc *rrpc)
1176 {
1177         mempool_destroy(rrpc->page_pool);
1178         mempool_destroy(rrpc->gcb_pool);
1179         mempool_destroy(rrpc->rq_pool);
1180 }
1181
1182 static void rrpc_luns_free(struct rrpc *rrpc)
1183 {
1184         struct nvm_lun *lun;
1185         struct rrpc_lun *rlun;
1186         int i;
1187
1188         if (!rrpc->luns)
1189                 return;
1190
1191         for (i = 0; i < rrpc->nr_luns; i++) {
1192                 rlun = &rrpc->luns[i];
1193                 lun = rlun->parent;
1194                 if (!lun)
1195                         break;
1196                 vfree(rlun->blocks);
1197         }
1198
1199         kfree(rrpc->luns);
1200 }
1201
1202 static int rrpc_luns_init(struct rrpc *rrpc, int lun_begin, int lun_end)
1203 {
1204         struct nvm_tgt_dev *dev = rrpc->dev;
1205         struct nvm_geo *geo = &dev->geo;
1206         struct rrpc_lun *rlun;
1207         int i, j, ret = -EINVAL;
1208
1209         if (geo->sec_per_blk > MAX_INVALID_PAGES_STORAGE * BITS_PER_LONG) {
1210                 pr_err("rrpc: number of pages per block too high.");
1211                 return -EINVAL;
1212         }
1213
1214         spin_lock_init(&rrpc->rev_lock);
1215
1216         rrpc->luns = kcalloc(rrpc->nr_luns, sizeof(struct rrpc_lun),
1217                                                                 GFP_KERNEL);
1218         if (!rrpc->luns)
1219                 return -ENOMEM;
1220
1221         /* 1:1 mapping */
1222         for (i = 0; i < rrpc->nr_luns; i++) {
1223                 int lunid = lun_begin + i;
1224                 struct nvm_lun *lun;
1225
1226                 lun = dev->mt->get_lun(dev->parent, lunid);
1227                 if (!lun)
1228                         goto err;
1229
1230                 rlun = &rrpc->luns[i];
1231                 rlun->parent = lun;
1232                 rlun->blocks = vzalloc(sizeof(struct rrpc_block) *
1233                                                         geo->blks_per_lun);
1234                 if (!rlun->blocks) {
1235                         ret = -ENOMEM;
1236                         goto err;
1237                 }
1238
1239                 for (j = 0; j < geo->blks_per_lun; j++) {
1240                         struct rrpc_block *rblk = &rlun->blocks[j];
1241                         struct nvm_block *blk = &lun->blocks[j];
1242
1243                         rblk->parent = blk;
1244                         rblk->rlun = rlun;
1245                         INIT_LIST_HEAD(&rblk->prio);
1246                         spin_lock_init(&rblk->lock);
1247                 }
1248
1249                 rlun->reserved_blocks = 2; /* for GC only */
1250
1251                 rlun->rrpc = rrpc;
1252                 INIT_LIST_HEAD(&rlun->prio_list);
1253                 INIT_LIST_HEAD(&rlun->wblk_list);
1254
1255                 INIT_WORK(&rlun->ws_gc, rrpc_lun_gc);
1256                 spin_lock_init(&rlun->lock);
1257         }
1258
1259         return 0;
1260 err:
1261         return ret;
1262 }
1263
1264 /* returns 0 on success and stores the beginning address in *begin */
1265 static int rrpc_area_init(struct rrpc *rrpc, sector_t *begin)
1266 {
1267         struct nvm_tgt_dev *dev = rrpc->dev;
1268         struct nvmm_type *mt = dev->mt;
1269         sector_t size = rrpc->nr_sects * dev->geo.sec_size;
1270         int ret;
1271
1272         size >>= 9;
1273
1274         ret = mt->get_area(dev->parent, begin, size);
1275         if (!ret)
1276                 *begin >>= (ilog2(dev->geo.sec_size) - 9);
1277
1278         return ret;
1279 }
1280
1281 static void rrpc_area_free(struct rrpc *rrpc)
1282 {
1283         struct nvm_tgt_dev *dev = rrpc->dev;
1284         struct nvmm_type *mt = dev->mt;
1285         sector_t begin = rrpc->soffset << (ilog2(dev->geo.sec_size) - 9);
1286
1287         mt->put_area(dev->parent, begin);
1288 }
1289
1290 static void rrpc_free(struct rrpc *rrpc)
1291 {
1292         rrpc_gc_free(rrpc);
1293         rrpc_map_free(rrpc);
1294         rrpc_core_free(rrpc);
1295         rrpc_luns_free(rrpc);
1296         rrpc_area_free(rrpc);
1297
1298         kfree(rrpc);
1299 }
1300
1301 static void rrpc_exit(void *private)
1302 {
1303         struct rrpc *rrpc = private;
1304
1305         del_timer(&rrpc->gc_timer);
1306
1307         flush_workqueue(rrpc->krqd_wq);
1308         flush_workqueue(rrpc->kgc_wq);
1309
1310         rrpc_free(rrpc);
1311 }
1312
1313 static sector_t rrpc_capacity(void *private)
1314 {
1315         struct rrpc *rrpc = private;
1316         struct nvm_tgt_dev *dev = rrpc->dev;
1317         sector_t reserved, provisioned;
1318
1319         /* cur, gc, and two emergency blocks for each lun */
1320         reserved = rrpc->nr_luns * dev->geo.sec_per_blk * 4;
1321         provisioned = rrpc->nr_sects - reserved;
1322
1323         if (reserved > rrpc->nr_sects) {
1324                 pr_err("rrpc: not enough space available to expose storage.\n");
1325                 return 0;
1326         }
1327
1328         sector_div(provisioned, 10);
1329         return provisioned * 9 * NR_PHY_IN_LOG;
1330 }
1331
1332 /*
1333  * Looks up the logical address from reverse trans map and check if its valid by
1334  * comparing the logical to physical address with the physical address.
1335  * Returns 0 on free, otherwise 1 if in use
1336  */
1337 static void rrpc_block_map_update(struct rrpc *rrpc, struct rrpc_block *rblk)
1338 {
1339         struct nvm_tgt_dev *dev = rrpc->dev;
1340         int offset;
1341         struct rrpc_addr *laddr;
1342         u64 bpaddr, paddr, pladdr;
1343
1344         bpaddr = block_to_rel_addr(rrpc, rblk);
1345         for (offset = 0; offset < dev->geo.sec_per_blk; offset++) {
1346                 paddr = bpaddr + offset;
1347
1348                 pladdr = rrpc->rev_trans_map[paddr].addr;
1349                 if (pladdr == ADDR_EMPTY)
1350                         continue;
1351
1352                 laddr = &rrpc->trans_map[pladdr];
1353
1354                 if (paddr == laddr->addr) {
1355                         laddr->rblk = rblk;
1356                 } else {
1357                         set_bit(offset, rblk->invalid_pages);
1358                         rblk->nr_invalid_pages++;
1359                 }
1360         }
1361 }
1362
1363 static int rrpc_blocks_init(struct rrpc *rrpc)
1364 {
1365         struct nvm_tgt_dev *dev = rrpc->dev;
1366         struct rrpc_lun *rlun;
1367         struct rrpc_block *rblk;
1368         int lun_iter, blk_iter;
1369
1370         for (lun_iter = 0; lun_iter < rrpc->nr_luns; lun_iter++) {
1371                 rlun = &rrpc->luns[lun_iter];
1372
1373                 for (blk_iter = 0; blk_iter < dev->geo.blks_per_lun;
1374                                                                 blk_iter++) {
1375                         rblk = &rlun->blocks[blk_iter];
1376                         rrpc_block_map_update(rrpc, rblk);
1377                 }
1378         }
1379
1380         return 0;
1381 }
1382
1383 static int rrpc_luns_configure(struct rrpc *rrpc)
1384 {
1385         struct rrpc_lun *rlun;
1386         struct rrpc_block *rblk;
1387         int i;
1388
1389         for (i = 0; i < rrpc->nr_luns; i++) {
1390                 rlun = &rrpc->luns[i];
1391
1392                 rblk = rrpc_get_blk(rrpc, rlun, 0);
1393                 if (!rblk)
1394                         goto err;
1395                 rrpc_set_lun_cur(rlun, rblk, &rlun->cur);
1396
1397                 /* Emergency gc block */
1398                 rblk = rrpc_get_blk(rrpc, rlun, 1);
1399                 if (!rblk)
1400                         goto err;
1401                 rrpc_set_lun_cur(rlun, rblk, &rlun->gc_cur);
1402         }
1403
1404         return 0;
1405 err:
1406         rrpc_put_blks(rrpc);
1407         return -EINVAL;
1408 }
1409
1410 static struct nvm_tgt_type tt_rrpc;
1411
1412 static void *rrpc_init(struct nvm_tgt_dev *dev, struct gendisk *tdisk,
1413                                                 int lun_begin, int lun_end)
1414 {
1415         struct request_queue *bqueue = dev->q;
1416         struct request_queue *tqueue = tdisk->queue;
1417         struct nvm_geo *geo = &dev->geo;
1418         struct rrpc *rrpc;
1419         sector_t soffset;
1420         int ret;
1421
1422         if (!(dev->identity.dom & NVM_RSP_L2P)) {
1423                 pr_err("nvm: rrpc: device does not support l2p (%x)\n",
1424                                                         dev->identity.dom);
1425                 return ERR_PTR(-EINVAL);
1426         }
1427
1428         rrpc = kzalloc(sizeof(struct rrpc), GFP_KERNEL);
1429         if (!rrpc)
1430                 return ERR_PTR(-ENOMEM);
1431
1432         rrpc->instance.tt = &tt_rrpc;
1433         rrpc->dev = dev;
1434         rrpc->disk = tdisk;
1435
1436         bio_list_init(&rrpc->requeue_bios);
1437         spin_lock_init(&rrpc->bio_lock);
1438         INIT_WORK(&rrpc->ws_requeue, rrpc_requeue);
1439
1440         rrpc->nr_luns = geo->nr_luns;
1441         rrpc->nr_sects = (unsigned long long)geo->sec_per_lun * rrpc->nr_luns;
1442
1443         /* simple round-robin strategy */
1444         atomic_set(&rrpc->next_lun, -1);
1445
1446         ret = rrpc_area_init(rrpc, &soffset);
1447         if (ret < 0) {
1448                 pr_err("nvm: rrpc: could not initialize area\n");
1449                 return ERR_PTR(ret);
1450         }
1451         rrpc->soffset = soffset;
1452
1453         ret = rrpc_luns_init(rrpc, lun_begin, lun_end);
1454         if (ret) {
1455                 pr_err("nvm: rrpc: could not initialize luns\n");
1456                 goto err;
1457         }
1458
1459         rrpc->poffset = geo->sec_per_lun * lun_begin;
1460
1461         ret = rrpc_core_init(rrpc);
1462         if (ret) {
1463                 pr_err("nvm: rrpc: could not initialize core\n");
1464                 goto err;
1465         }
1466
1467         ret = rrpc_map_init(rrpc);
1468         if (ret) {
1469                 pr_err("nvm: rrpc: could not initialize maps\n");
1470                 goto err;
1471         }
1472
1473         ret = rrpc_blocks_init(rrpc);
1474         if (ret) {
1475                 pr_err("nvm: rrpc: could not initialize state for blocks\n");
1476                 goto err;
1477         }
1478
1479         ret = rrpc_luns_configure(rrpc);
1480         if (ret) {
1481                 pr_err("nvm: rrpc: not enough blocks available in LUNs.\n");
1482                 goto err;
1483         }
1484
1485         ret = rrpc_gc_init(rrpc);
1486         if (ret) {
1487                 pr_err("nvm: rrpc: could not initialize gc\n");
1488                 goto err;
1489         }
1490
1491         /* inherit the size from the underlying device */
1492         blk_queue_logical_block_size(tqueue, queue_physical_block_size(bqueue));
1493         blk_queue_max_hw_sectors(tqueue, queue_max_hw_sectors(bqueue));
1494
1495         pr_info("nvm: rrpc initialized with %u luns and %llu pages.\n",
1496                         rrpc->nr_luns, (unsigned long long)rrpc->nr_sects);
1497
1498         mod_timer(&rrpc->gc_timer, jiffies + msecs_to_jiffies(10));
1499
1500         return rrpc;
1501 err:
1502         rrpc_free(rrpc);
1503         return ERR_PTR(ret);
1504 }
1505
1506 /* round robin, page-based FTL, and cost-based GC */
1507 static struct nvm_tgt_type tt_rrpc = {
1508         .name           = "rrpc",
1509         .version        = {1, 0, 0},
1510
1511         .make_rq        = rrpc_make_rq,
1512         .capacity       = rrpc_capacity,
1513         .end_io         = rrpc_end_io,
1514
1515         .init           = rrpc_init,
1516         .exit           = rrpc_exit,
1517 };
1518
1519 static int __init rrpc_module_init(void)
1520 {
1521         return nvm_register_tgt_type(&tt_rrpc);
1522 }
1523
1524 static void rrpc_module_exit(void)
1525 {
1526         nvm_unregister_tgt_type(&tt_rrpc);
1527 }
1528
1529 module_init(rrpc_module_init);
1530 module_exit(rrpc_module_exit);
1531 MODULE_LICENSE("GPL v2");
1532 MODULE_DESCRIPTION("Block-Device Target for Open-Channel SSDs");