drivers/nvme/host/lightnvm.c

   1 /*
   2  * nvme-lightnvm.c - LightNVM NVMe device
   3  *
   4  * Copyright (C) 2014-2015 IT University of Copenhagen
   5  * Initial release: Matias Bjorling <mb@lightnvm.io>
   6  *
   7  * This program is free software; you can redistribute it and/or
   8  * modify it under the terms of the GNU General Public License version
   9  * 2 as published by the Free Software Foundation.
  10  *
  11  * This program is distributed in the hope that it will be useful, but
  12  * WITHOUT ANY WARRANTY; without even the implied warranty of
  13  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  14  * General Public License for more details.
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * along with this program; see the file COPYING.  If not, write to
  18  * the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139,
  19  * USA.
  20  *
  21  */
  22
  23 #include "nvme.h"
  24
  25 #ifdef CONFIG_NVM
  26
  27 #include <linux/nvme.h>
  28 #include <linux/bitops.h>
  29 #include <linux/lightnvm.h>
  30 #include <linux/vmalloc.h>
  31
  32 enum nvme_nvm_admin_opcode {
  33         nvme_nvm_admin_identity         = 0xe2,
  34         nvme_nvm_admin_get_l2p_tbl      = 0xea,
  35         nvme_nvm_admin_get_bb_tbl       = 0xf2,
  36         nvme_nvm_admin_set_bb_tbl       = 0xf1,
  37 };
  38
  39 struct nvme_nvm_hb_rw {
  40         __u8                    opcode;
  41         __u8                    flags;
  42         __u16                   command_id;
  43         __le32                  nsid;
  44         __u64                   rsvd2;
  45         __le64                  metadata;
  46         __le64                  prp1;
  47         __le64                  prp2;
  48         __le64                  spba;
  49         __le16                  length;
  50         __le16                  control;
  51         __le32                  dsmgmt;
  52         __le64                  slba;
  53 };
  54
  55 struct nvme_nvm_ph_rw {
  56         __u8                    opcode;
  57         __u8                    flags;
  58         __u16                   command_id;
  59         __le32                  nsid;
  60         __u64                   rsvd2;
  61         __le64                  metadata;
  62         __le64                  prp1;
  63         __le64                  prp2;
  64         __le64                  spba;
  65         __le16                  length;
  66         __le16                  control;
  67         __le32                  dsmgmt;
  68         __le64                  resv;
  69 };
  70
  71 struct nvme_nvm_identity {
  72         __u8                    opcode;
  73         __u8                    flags;
  74         __u16                   command_id;
  75         __le32                  nsid;
  76         __u64                   rsvd[2];
  77         __le64                  prp1;
  78         __le64                  prp2;
  79         __le32                  chnl_off;
  80         __u32                   rsvd11[5];
  81 };
  82
  83 struct nvme_nvm_l2ptbl {
  84         __u8                    opcode;
  85         __u8                    flags;
  86         __u16                   command_id;
  87         __le32                  nsid;
  88         __le32                  cdw2[4];
  89         __le64                  prp1;
  90         __le64                  prp2;
  91         __le64                  slba;
  92         __le32                  nlb;
  93         __le16                  cdw14[6];
  94 };
  95
  96 struct nvme_nvm_getbbtbl {
  97         __u8                    opcode;
  98         __u8                    flags;
  99         __u16                   command_id;
 100         __le32                  nsid;
 101         __u64                   rsvd[2];
 102         __le64                  prp1;
 103         __le64                  prp2;
 104         __le64                  spba;
 105         __u32                   rsvd4[4];
 106 };
 107
 108 struct nvme_nvm_setbbtbl {
 109         __u8                    opcode;
 110         __u8                    flags;
 111         __u16                   command_id;
 112         __le32                  nsid;
 113         __le64                  rsvd[2];
 114         __le64                  prp1;
 115         __le64                  prp2;
 116         __le64                  spba;
 117         __le16                  nlb;
 118         __u8                    value;
 119         __u8                    rsvd3;
 120         __u32                   rsvd4[3];
 121 };
 122
 123 struct nvme_nvm_erase_blk {
 124         __u8                    opcode;
 125         __u8                    flags;
 126         __u16                   command_id;
 127         __le32                  nsid;
 128         __u64                   rsvd[2];
 129         __le64                  prp1;
 130         __le64                  prp2;
 131         __le64                  spba;
 132         __le16                  length;
 133         __le16                  control;
 134         __le32                  dsmgmt;
 135         __le64                  resv;
 136 };
 137
 138 struct nvme_nvm_command {
 139         union {
 140                 struct nvme_common_command common;
 141                 struct nvme_nvm_identity identity;
 142                 struct nvme_nvm_hb_rw hb_rw;
 143                 struct nvme_nvm_ph_rw ph_rw;
 144                 struct nvme_nvm_l2ptbl l2p;
 145                 struct nvme_nvm_getbbtbl get_bb;
 146                 struct nvme_nvm_setbbtbl set_bb;
 147                 struct nvme_nvm_erase_blk erase;
 148         };
 149 };
 150
 151 struct nvme_nvm_id_group {
 152         __u8                    mtype;
 153         __u8                    fmtype;
 154         __le16                  res16;
 155         __u8                    num_ch;
 156         __u8                    num_lun;
 157         __u8                    num_pln;
 158         __u8                    rsvd1;
 159         __le16                  num_blk;
 160         __le16                  num_pg;
 161         __le16                  fpg_sz;
 162         __le16                  csecs;
 163         __le16                  sos;
 164         __le16                  rsvd2;
 165         __le32                  trdt;
 166         __le32                  trdm;
 167         __le32                  tprt;
 168         __le32                  tprm;
 169         __le32                  tbet;
 170         __le32                  tbem;
 171         __le32                  mpos;
 172         __le32                  mccap;
 173         __le16                  cpar;
 174         __u8                    reserved[906];
 175 } __packed;
 176
 177 struct nvme_nvm_addr_format {
 178         __u8                    ch_offset;
 179         __u8                    ch_len;
 180         __u8                    lun_offset;
 181         __u8                    lun_len;
 182         __u8                    pln_offset;
 183         __u8                    pln_len;
 184         __u8                    blk_offset;
 185         __u8                    blk_len;
 186         __u8                    pg_offset;
 187         __u8                    pg_len;
 188         __u8                    sect_offset;
 189         __u8                    sect_len;
 190         __u8                    res[4];
 191 } __packed;
 192
 193 struct nvme_nvm_id {
 194         __u8                    ver_id;
 195         __u8                    vmnt;
 196         __u8                    cgrps;
 197         __u8                    res;
 198         __le32                  cap;
 199         __le32                  dom;
 200         struct nvme_nvm_addr_format ppaf;
 201         __u8                    resv[228];
 202         struct nvme_nvm_id_group groups[4];
 203 } __packed;
 204
 205 struct nvme_nvm_bb_tbl {
 206         __u8    tblid[4];
 207         __le16  verid;
 208         __le16  revid;
 209         __le32  rvsd1;
 210         __le32  tblks;
 211         __le32  tfact;
 212         __le32  tgrown;
 213         __le32  tdresv;
 214         __le32  thresv;
 215         __le32  rsvd2[8];
 216         __u8    blk[0];
 217 };
 218
 219 /*
 220  * Check we didn't inadvertently grow the command struct
 221  */
 222 static inline void _nvme_nvm_check_size(void)
 223 {
 224         BUILD_BUG_ON(sizeof(struct nvme_nvm_identity) != 64);
 225         BUILD_BUG_ON(sizeof(struct nvme_nvm_hb_rw) != 64);
 226         BUILD_BUG_ON(sizeof(struct nvme_nvm_ph_rw) != 64);
 227         BUILD_BUG_ON(sizeof(struct nvme_nvm_getbbtbl) != 64);
 228         BUILD_BUG_ON(sizeof(struct nvme_nvm_setbbtbl) != 64);
 229         BUILD_BUG_ON(sizeof(struct nvme_nvm_l2ptbl) != 64);
 230         BUILD_BUG_ON(sizeof(struct nvme_nvm_erase_blk) != 64);
 231         BUILD_BUG_ON(sizeof(struct nvme_nvm_id_group) != 960);
 232         BUILD_BUG_ON(sizeof(struct nvme_nvm_addr_format) != 128);
 233         BUILD_BUG_ON(sizeof(struct nvme_nvm_id) != 4096);
 234         BUILD_BUG_ON(sizeof(struct nvme_nvm_bb_tbl) != 512);
 235 }
 236
 237 static int init_grps(struct nvm_id *nvm_id, struct nvme_nvm_id *nvme_nvm_id)
 238 {
 239         struct nvme_nvm_id_group *src;
 240         struct nvm_id_group *dst;
 241         int i, end;
 242
 243         end = min_t(u32, 4, nvm_id->cgrps);
 244
 245         for (i = 0; i < end; i++) {
 246                 src = &nvme_nvm_id->groups[i];
 247                 dst = &nvm_id->groups[i];
 248
 249                 dst->mtype = src->mtype;
 250                 dst->fmtype = src->fmtype;
 251                 dst->num_ch = src->num_ch;
 252                 dst->num_lun = src->num_lun;
 253                 dst->num_pln = src->num_pln;
 254
 255                 dst->num_pg = le16_to_cpu(src->num_pg);
 256                 dst->num_blk = le16_to_cpu(src->num_blk);
 257                 dst->fpg_sz = le16_to_cpu(src->fpg_sz);
 258                 dst->csecs = le16_to_cpu(src->csecs);
 259                 dst->sos = le16_to_cpu(src->sos);
 260
 261                 dst->trdt = le32_to_cpu(src->trdt);
 262                 dst->trdm = le32_to_cpu(src->trdm);
 263                 dst->tprt = le32_to_cpu(src->tprt);
 264                 dst->tprm = le32_to_cpu(src->tprm);
 265                 dst->tbet = le32_to_cpu(src->tbet);
 266                 dst->tbem = le32_to_cpu(src->tbem);
 267                 dst->mpos = le32_to_cpu(src->mpos);
 268                 dst->mccap = le32_to_cpu(src->mccap);
 269
 270                 dst->cpar = le16_to_cpu(src->cpar);
 271         }
 272
 273         return 0;
 274 }
 275
 276 static int nvme_nvm_identity(struct request_queue *q, struct nvm_id *nvm_id)
 277 {
 278         struct nvme_ns *ns = q->queuedata;
 279         struct nvme_nvm_id *nvme_nvm_id;
 280         struct nvme_nvm_command c = {};
 281         int ret;
 282
 283         c.identity.opcode = nvme_nvm_admin_identity;
 284         c.identity.nsid = cpu_to_le32(ns->ns_id);
 285         c.identity.chnl_off = 0;
 286
 287         nvme_nvm_id = kmalloc(sizeof(struct nvme_nvm_id), GFP_KERNEL);
 288         if (!nvme_nvm_id)
 289                 return -ENOMEM;
 290
 291         ret = nvme_submit_sync_cmd(ns->ctrl->admin_q, (struct nvme_command *)&c,
 292                                 nvme_nvm_id, sizeof(struct nvme_nvm_id));
 293         if (ret) {
 294                 ret = -EIO;
 295                 goto out;
 296         }
 297
 298         nvm_id->ver_id = nvme_nvm_id->ver_id;
 299         nvm_id->vmnt = nvme_nvm_id->vmnt;
 300         nvm_id->cgrps = nvme_nvm_id->cgrps;
 301         nvm_id->cap = le32_to_cpu(nvme_nvm_id->cap);
 302         nvm_id->dom = le32_to_cpu(nvme_nvm_id->dom);
 303         memcpy(&nvm_id->ppaf, &nvme_nvm_id->ppaf,
 304                                         sizeof(struct nvme_nvm_addr_format));
 305
 306         ret = init_grps(nvm_id, nvme_nvm_id);
 307 out:
 308         kfree(nvme_nvm_id);
 309         return ret;
 310 }
 311
 312 static int nvme_nvm_get_l2p_tbl(struct request_queue *q, u64 slba, u32 nlb,
 313                                 nvm_l2p_update_fn *update_l2p, void *priv)
 314 {
 315         struct nvme_ns *ns = q->queuedata;
 316         struct nvme_nvm_command c = {};
 317         u32 len = queue_max_hw_sectors(ns->ctrl->admin_q) << 9;
 318         u32 nlb_pr_rq = len / sizeof(u64);
 319         u64 cmd_slba = slba;
 320         void *entries;
 321         int ret = 0;
 322
 323         c.l2p.opcode = nvme_nvm_admin_get_l2p_tbl;
 324         c.l2p.nsid = cpu_to_le32(ns->ns_id);
 325         entries = kmalloc(len, GFP_KERNEL);
 326         if (!entries)
 327                 return -ENOMEM;
 328
 329         while (nlb) {
 330                 u32 cmd_nlb = min(nlb_pr_rq, nlb);
 331
 332                 c.l2p.slba = cpu_to_le64(cmd_slba);
 333                 c.l2p.nlb = cpu_to_le32(cmd_nlb);
 334
 335                 ret = nvme_submit_sync_cmd(ns->ctrl->admin_q,
 336                                 (struct nvme_command *)&c, entries, len);
 337                 if (ret) {
 338                         dev_err(ns->ctrl->dev, "L2P table transfer failed (%d)\n",
 339                                                                         ret);
 340                         ret = -EIO;
 341                         goto out;
 342                 }
 343
 344                 if (update_l2p(cmd_slba, cmd_nlb, entries, priv)) {
 345                         ret = -EINTR;
 346                         goto out;
 347                 }
 348
 349                 cmd_slba += cmd_nlb;
 350                 nlb -= cmd_nlb;
 351         }
 352
 353 out:
 354         kfree(entries);
 355         return ret;
 356 }
 357
 358 static int nvme_nvm_get_bb_tbl(struct request_queue *q, struct ppa_addr ppa,
 359                                 int nr_blocks, nvm_bb_update_fn *update_bbtbl,
 360                                 void *priv)
 361 {
 362         struct nvme_ns *ns = q->queuedata;
 363         struct nvme_ctrl *ctrl = ns->ctrl;
 364         struct nvme_nvm_command c = {};
 365         struct nvme_nvm_bb_tbl *bb_tbl;
 366         int tblsz = sizeof(struct nvme_nvm_bb_tbl) + nr_blocks;
 367         int ret = 0;
 368
 369         c.get_bb.opcode = nvme_nvm_admin_get_bb_tbl;
 370         c.get_bb.nsid = cpu_to_le32(ns->ns_id);
 371         c.get_bb.spba = cpu_to_le64(ppa.ppa);
 372
 373         bb_tbl = kzalloc(tblsz, GFP_KERNEL);
 374         if (!bb_tbl)
 375                 return -ENOMEM;
 376
 377         ret = nvme_submit_sync_cmd(ctrl->admin_q, (struct nvme_command *)&c,
 378                                                                 bb_tbl, tblsz);
 379         if (ret) {
 380                 dev_err(ctrl->dev, "get bad block table failed (%d)\n", ret);
 381                 ret = -EIO;
 382                 goto out;
 383         }
 384
 385         if (bb_tbl->tblid[0] != 'B' || bb_tbl->tblid[1] != 'B' ||
 386                 bb_tbl->tblid[2] != 'L' || bb_tbl->tblid[3] != 'T') {
 387                 dev_err(ctrl->dev, "bbt format mismatch\n");
 388                 ret = -EINVAL;
 389                 goto out;
 390         }
 391
 392         if (le16_to_cpu(bb_tbl->verid) != 1) {
 393                 ret = -EINVAL;
 394                 dev_err(ctrl->dev, "bbt version not supported\n");
 395                 goto out;
 396         }
 397
 398         if (le32_to_cpu(bb_tbl->tblks) != nr_blocks) {
 399                 ret = -EINVAL;
 400                 dev_err(ctrl->dev, "bbt unsuspected blocks returned (%u!=%u)",
 401                                         le32_to_cpu(bb_tbl->tblks), nr_blocks);
 402                 goto out;
 403         }
 404
 405         ret = update_bbtbl(ppa, nr_blocks, bb_tbl->blk, priv);
 406         if (ret) {
 407                 ret = -EINTR;
 408                 goto out;
 409         }
 410
 411 out:
 412         kfree(bb_tbl);
 413         return ret;
 414 }
 415
 416 static int nvme_nvm_set_bb_tbl(struct request_queue *q, struct nvm_rq *rqd,
 417                                                                 int type)
 418 {
 419         struct nvme_ns *ns = q->queuedata;
 420         struct nvme_nvm_command c = {};
 421         int ret = 0;
 422
 423         c.set_bb.opcode = nvme_nvm_admin_set_bb_tbl;
 424         c.set_bb.nsid = cpu_to_le32(ns->ns_id);
 425         c.set_bb.spba = cpu_to_le64(rqd->ppa_addr.ppa);
 426         c.set_bb.nlb = cpu_to_le16(rqd->nr_pages - 1);
 427         c.set_bb.value = type;
 428
 429         ret = nvme_submit_sync_cmd(ns->ctrl->admin_q, (struct nvme_command *)&c,
 430                                                                 NULL, 0);
 431         if (ret)
 432                 dev_err(ns->ctrl->dev, "set bad block table failed (%d)\n", ret);
 433         return ret;
 434 }
 435
 436 static inline void nvme_nvm_rqtocmd(struct request *rq, struct nvm_rq *rqd,
 437                                 struct nvme_ns *ns, struct nvme_nvm_command *c)
 438 {
 439         c->ph_rw.opcode = rqd->opcode;
 440         c->ph_rw.nsid = cpu_to_le32(ns->ns_id);
 441         c->ph_rw.spba = cpu_to_le64(rqd->ppa_addr.ppa);
 442         c->ph_rw.control = cpu_to_le16(rqd->flags);
 443         c->ph_rw.length = cpu_to_le16(rqd->nr_pages - 1);
 444
 445         if (rqd->opcode == NVM_OP_HBWRITE || rqd->opcode == NVM_OP_HBREAD)
 446                 c->hb_rw.slba = cpu_to_le64(nvme_block_nr(ns,
 447                                                 rqd->bio->bi_iter.bi_sector));
 448 }
 449
 450 static void nvme_nvm_end_io(struct request *rq, int error)
 451 {
 452         struct nvm_rq *rqd = rq->end_io_data;
 453         struct nvm_dev *dev = rqd->dev;
 454
 455         if (dev->mt->end_io(rqd, error))
 456                 pr_err("nvme: err status: %x result: %lx\n",
 457                                 rq->errors, (unsigned long)rq->special);
 458
 459         kfree(rq->cmd);
 460         blk_mq_free_request(rq);
 461 }
 462
 463 static int nvme_nvm_submit_io(struct request_queue *q, struct nvm_rq *rqd)
 464 {
 465         struct nvme_ns *ns = q->queuedata;
 466         struct request *rq;
 467         struct bio *bio = rqd->bio;
 468         struct nvme_nvm_command *cmd;
 469
 470         rq = blk_mq_alloc_request(q, bio_rw(bio), 0);
 471         if (IS_ERR(rq))
 472                 return -ENOMEM;
 473
 474         cmd = kzalloc(sizeof(struct nvme_nvm_command), GFP_KERNEL);
 475         if (!cmd) {
 476                 blk_mq_free_request(rq);
 477                 return -ENOMEM;
 478         }
 479
 480         rq->cmd_type = REQ_TYPE_DRV_PRIV;
 481         rq->ioprio = bio_prio(bio);
 482
 483         if (bio_has_data(bio))
 484                 rq->nr_phys_segments = bio_phys_segments(q, bio);
 485
 486         rq->__data_len = bio->bi_iter.bi_size;
 487         rq->bio = rq->biotail = bio;
 488
 489         nvme_nvm_rqtocmd(rq, rqd, ns, cmd);
 490
 491         rq->cmd = (unsigned char *)cmd;
 492         rq->cmd_len = sizeof(struct nvme_nvm_command);
 493         rq->special = (void *)0;
 494
 495         rq->end_io_data = rqd;
 496
 497         blk_execute_rq_nowait(q, NULL, rq, 0, nvme_nvm_end_io);
 498
 499         return 0;
 500 }
 501
 502 static int nvme_nvm_erase_block(struct request_queue *q, struct nvm_rq *rqd)
 503 {
 504         struct nvme_ns *ns = q->queuedata;
 505         struct nvme_nvm_command c = {};
 506
 507         c.erase.opcode = NVM_OP_ERASE;
 508         c.erase.nsid = cpu_to_le32(ns->ns_id);
 509         c.erase.spba = cpu_to_le64(rqd->ppa_addr.ppa);
 510         c.erase.length = cpu_to_le16(rqd->nr_pages - 1);
 511
 512         return nvme_submit_sync_cmd(q, (struct nvme_command *)&c, NULL, 0);
 513 }
 514
 515 static void *nvme_nvm_create_dma_pool(struct request_queue *q, char *name)
 516 {
 517         struct nvme_ns *ns = q->queuedata;
 518
 519         return dma_pool_create(name, ns->ctrl->dev, PAGE_SIZE, PAGE_SIZE, 0);
 520 }
 521
 522 static void nvme_nvm_destroy_dma_pool(void *pool)
 523 {
 524         struct dma_pool *dma_pool = pool;
 525
 526         dma_pool_destroy(dma_pool);
 527 }
 528
 529 static void *nvme_nvm_dev_dma_alloc(struct request_queue *q, void *pool,
 530                                     gfp_t mem_flags, dma_addr_t *dma_handler)
 531 {
 532         return dma_pool_alloc(pool, mem_flags, dma_handler);
 533 }
 534
 535 static void nvme_nvm_dev_dma_free(void *pool, void *ppa_list,
 536                                                         dma_addr_t dma_handler)
 537 {
 538         dma_pool_free(pool, ppa_list, dma_handler);
 539 }
 540
 541 static struct nvm_dev_ops nvme_nvm_dev_ops = {
 542         .identity               = nvme_nvm_identity,
 543
 544         .get_l2p_tbl            = nvme_nvm_get_l2p_tbl,
 545
 546         .get_bb_tbl             = nvme_nvm_get_bb_tbl,
 547         .set_bb_tbl             = nvme_nvm_set_bb_tbl,
 548
 549         .submit_io              = nvme_nvm_submit_io,
 550         .erase_block            = nvme_nvm_erase_block,
 551
 552         .create_dma_pool        = nvme_nvm_create_dma_pool,
 553         .destroy_dma_pool       = nvme_nvm_destroy_dma_pool,
 554         .dev_dma_alloc          = nvme_nvm_dev_dma_alloc,
 555         .dev_dma_free           = nvme_nvm_dev_dma_free,
 556
 557         .max_phys_sect          = 64,
 558 };
 559
 560 int nvme_nvm_register(struct request_queue *q, char *disk_name)
 561 {
 562         return nvm_register(q, disk_name, &nvme_nvm_dev_ops);
 563 }
 564
 565 void nvme_nvm_unregister(struct request_queue *q, char *disk_name)
 566 {
 567         nvm_unregister(disk_name);
 568 }
 569
 570 int nvme_nvm_ns_supported(struct nvme_ns *ns, struct nvme_id_ns *id)
 571 {
 572         struct nvme_ctrl *ctrl = ns->ctrl;
 573         /* XXX: this is poking into PCI structures from generic code! */
 574         struct pci_dev *pdev = to_pci_dev(ctrl->dev);
 575
 576         /* QEMU NVMe simulator - PCI ID + Vendor specific bit */
 577         if (pdev->vendor == PCI_VENDOR_ID_INTEL && pdev->device == 0x5845 &&
 578                                                         id->vs[0] == 0x1)
 579                 return 1;
 580
 581         /* CNEX Labs - PCI ID + Vendor specific bit */
 582         if (pdev->vendor == 0x1d1d && pdev->device == 0x2807 &&
 583                                                         id->vs[0] == 0x1)
 584                 return 1;
 585
 586         return 0;
 587 }
 588 #else
 589 int nvme_nvm_register(struct request_queue *q, char *disk_name)
 590 {
 591         return 0;
 592 }
 593 void nvme_nvm_unregister(struct request_queue *q, char *disk_name) {};
 594 int nvme_nvm_ns_supported(struct nvme_ns *ns, struct nvme_id_ns *id)
 595 {
 596         return 0;
 597 }
 598 #endif /* CONFIG_NVM */