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