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