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 #include <linux/nvme.h>
  26 #include <linux/bitops.h>
  27 #include <linux/lightnvm.h>
  28 #include <linux/vmalloc.h>
  29 #include <linux/sched/sysctl.h>
  30 #include <uapi/linux/lightnvm.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 #define NVME_NVM_LP_MLC_PAIRS 886
 152 struct nvme_nvm_lp_mlc {
 153         __le16                  num_pairs;
 154         __u8                    pairs[NVME_NVM_LP_MLC_PAIRS];
 155 };
 156
 157 struct nvme_nvm_lp_tbl {
 158         __u8                    id[8];
 159         struct nvme_nvm_lp_mlc  mlc;
 160 };
 161
 162 struct nvme_nvm_id_group {
 163         __u8                    mtype;
 164         __u8                    fmtype;
 165         __le16                  res16;
 166         __u8                    num_ch;
 167         __u8                    num_lun;
 168         __u8                    num_pln;
 169         __u8                    rsvd1;
 170         __le16                  num_blk;
 171         __le16                  num_pg;
 172         __le16                  fpg_sz;
 173         __le16                  csecs;
 174         __le16                  sos;
 175         __le16                  rsvd2;
 176         __le32                  trdt;
 177         __le32                  trdm;
 178         __le32                  tprt;
 179         __le32                  tprm;
 180         __le32                  tbet;
 181         __le32                  tbem;
 182         __le32                  mpos;
 183         __le32                  mccap;
 184         __le16                  cpar;
 185         __u8                    reserved[10];
 186         struct nvme_nvm_lp_tbl lptbl;
 187 } __packed;
 188
 189 struct nvme_nvm_addr_format {
 190         __u8                    ch_offset;
 191         __u8                    ch_len;
 192         __u8                    lun_offset;
 193         __u8                    lun_len;
 194         __u8                    pln_offset;
 195         __u8                    pln_len;
 196         __u8                    blk_offset;
 197         __u8                    blk_len;
 198         __u8                    pg_offset;
 199         __u8                    pg_len;
 200         __u8                    sect_offset;
 201         __u8                    sect_len;
 202         __u8                    res[4];
 203 } __packed;
 204
 205 struct nvme_nvm_id {
 206         __u8                    ver_id;
 207         __u8                    vmnt;
 208         __u8                    cgrps;
 209         __u8                    res;
 210         __le32                  cap;
 211         __le32                  dom;
 212         struct nvme_nvm_addr_format ppaf;
 213         __u8                    resv[228];
 214         struct nvme_nvm_id_group groups[4];
 215 } __packed;
 216
 217 struct nvme_nvm_bb_tbl {
 218         __u8    tblid[4];
 219         __le16  verid;
 220         __le16  revid;
 221         __le32  rvsd1;
 222         __le32  tblks;
 223         __le32  tfact;
 224         __le32  tgrown;
 225         __le32  tdresv;
 226         __le32  thresv;
 227         __le32  rsvd2[8];
 228         __u8    blk[0];
 229 };
 230
 231 /*
 232  * Check we didn't inadvertently grow the command struct
 233  */
 234 static inline void _nvme_nvm_check_size(void)
 235 {
 236         BUILD_BUG_ON(sizeof(struct nvme_nvm_identity) != 64);
 237         BUILD_BUG_ON(sizeof(struct nvme_nvm_hb_rw) != 64);
 238         BUILD_BUG_ON(sizeof(struct nvme_nvm_ph_rw) != 64);
 239         BUILD_BUG_ON(sizeof(struct nvme_nvm_getbbtbl) != 64);
 240         BUILD_BUG_ON(sizeof(struct nvme_nvm_setbbtbl) != 64);
 241         BUILD_BUG_ON(sizeof(struct nvme_nvm_l2ptbl) != 64);
 242         BUILD_BUG_ON(sizeof(struct nvme_nvm_erase_blk) != 64);
 243         BUILD_BUG_ON(sizeof(struct nvme_nvm_id_group) != 960);
 244         BUILD_BUG_ON(sizeof(struct nvme_nvm_addr_format) != 128);
 245         BUILD_BUG_ON(sizeof(struct nvme_nvm_id) != 4096);
 246         BUILD_BUG_ON(sizeof(struct nvme_nvm_bb_tbl) != 512);
 247 }
 248
 249 static int init_grps(struct nvm_id *nvm_id, struct nvme_nvm_id *nvme_nvm_id)
 250 {
 251         struct nvme_nvm_id_group *src;
 252         struct nvm_id_group *dst;
 253
 254         if (nvme_nvm_id->cgrps != 1)
 255                 return -EINVAL;
 256
 257         src = &nvme_nvm_id->groups[0];
 258         dst = &nvm_id->grp;
 259
 260         dst->mtype = src->mtype;
 261         dst->fmtype = src->fmtype;
 262         dst->num_ch = src->num_ch;
 263         dst->num_lun = src->num_lun;
 264         dst->num_pln = src->num_pln;
 265
 266         dst->num_pg = le16_to_cpu(src->num_pg);
 267         dst->num_blk = le16_to_cpu(src->num_blk);
 268         dst->fpg_sz = le16_to_cpu(src->fpg_sz);
 269         dst->csecs = le16_to_cpu(src->csecs);
 270         dst->sos = le16_to_cpu(src->sos);
 271
 272         dst->trdt = le32_to_cpu(src->trdt);
 273         dst->trdm = le32_to_cpu(src->trdm);
 274         dst->tprt = le32_to_cpu(src->tprt);
 275         dst->tprm = le32_to_cpu(src->tprm);
 276         dst->tbet = le32_to_cpu(src->tbet);
 277         dst->tbem = le32_to_cpu(src->tbem);
 278         dst->mpos = le32_to_cpu(src->mpos);
 279         dst->mccap = le32_to_cpu(src->mccap);
 280
 281         dst->cpar = le16_to_cpu(src->cpar);
 282
 283         if (dst->fmtype == NVM_ID_FMTYPE_MLC) {
 284                 memcpy(dst->lptbl.id, src->lptbl.id, 8);
 285                 dst->lptbl.mlc.num_pairs =
 286                                 le16_to_cpu(src->lptbl.mlc.num_pairs);
 287
 288                 if (dst->lptbl.mlc.num_pairs > NVME_NVM_LP_MLC_PAIRS) {
 289                         pr_err("nvm: number of MLC pairs not supported\n");
 290                         return -EINVAL;
 291                 }
 292
 293                 memcpy(dst->lptbl.mlc.pairs, src->lptbl.mlc.pairs,
 294                                         dst->lptbl.mlc.num_pairs);
 295         }
 296
 297         return 0;
 298 }
 299
 300 static int nvme_nvm_identity(struct nvm_dev *nvmdev, struct nvm_id *nvm_id)
 301 {
 302         struct nvme_ns *ns = nvmdev->q->queuedata;
 303         struct nvme_nvm_id *nvme_nvm_id;
 304         struct nvme_nvm_command c = {};
 305         int ret;
 306
 307         c.identity.opcode = nvme_nvm_admin_identity;
 308         c.identity.nsid = cpu_to_le32(ns->ns_id);
 309         c.identity.chnl_off = 0;
 310
 311         nvme_nvm_id = kmalloc(sizeof(struct nvme_nvm_id), GFP_KERNEL);
 312         if (!nvme_nvm_id)
 313                 return -ENOMEM;
 314
 315         ret = nvme_submit_sync_cmd(ns->ctrl->admin_q, (struct nvme_command *)&c,
 316                                 nvme_nvm_id, sizeof(struct nvme_nvm_id));
 317         if (ret) {
 318                 ret = -EIO;
 319                 goto out;
 320         }
 321
 322         nvm_id->ver_id = nvme_nvm_id->ver_id;
 323         nvm_id->vmnt = nvme_nvm_id->vmnt;
 324         nvm_id->cap = le32_to_cpu(nvme_nvm_id->cap);
 325         nvm_id->dom = le32_to_cpu(nvme_nvm_id->dom);
 326         memcpy(&nvm_id->ppaf, &nvme_nvm_id->ppaf,
 327                                         sizeof(struct nvme_nvm_addr_format));
 328
 329         ret = init_grps(nvm_id, nvme_nvm_id);
 330 out:
 331         kfree(nvme_nvm_id);
 332         return ret;
 333 }
 334
 335 static int nvme_nvm_get_l2p_tbl(struct nvm_dev *nvmdev, u64 slba, u32 nlb,
 336                                 nvm_l2p_update_fn *update_l2p, void *priv)
 337 {
 338         struct nvme_ns *ns = nvmdev->q->queuedata;
 339         struct nvme_nvm_command c = {};
 340         u32 len = queue_max_hw_sectors(ns->ctrl->admin_q) << 9;
 341         u32 nlb_pr_rq = len / sizeof(u64);
 342         u64 cmd_slba = slba;
 343         void *entries;
 344         int ret = 0;
 345
 346         c.l2p.opcode = nvme_nvm_admin_get_l2p_tbl;
 347         c.l2p.nsid = cpu_to_le32(ns->ns_id);
 348         entries = kmalloc(len, GFP_KERNEL);
 349         if (!entries)
 350                 return -ENOMEM;
 351
 352         while (nlb) {
 353                 u32 cmd_nlb = min(nlb_pr_rq, nlb);
 354                 u64 elba = slba + cmd_nlb;
 355
 356                 c.l2p.slba = cpu_to_le64(cmd_slba);
 357                 c.l2p.nlb = cpu_to_le32(cmd_nlb);
 358
 359                 ret = nvme_submit_sync_cmd(ns->ctrl->admin_q,
 360                                 (struct nvme_command *)&c, entries, len);
 361                 if (ret) {
 362                         dev_err(ns->ctrl->device,
 363                                 "L2P table transfer failed (%d)\n", ret);
 364                         ret = -EIO;
 365                         goto out;
 366                 }
 367
 368                 if (unlikely(elba > nvmdev->total_secs)) {
 369                         pr_err("nvm: L2P data from device is out of bounds!\n");
 370                         return -EINVAL;
 371                 }
 372
 373                 /* Transform physical address to target address space */
 374                 nvm_part_to_tgt(nvmdev, entries, cmd_nlb);
 375
 376                 if (update_l2p(cmd_slba, cmd_nlb, entries, priv)) {
 377                         ret = -EINTR;
 378                         goto out;
 379                 }
 380
 381                 cmd_slba += cmd_nlb;
 382                 nlb -= cmd_nlb;
 383         }
 384
 385 out:
 386         kfree(entries);
 387         return ret;
 388 }
 389
 390 static int nvme_nvm_get_bb_tbl(struct nvm_dev *nvmdev, struct ppa_addr ppa,
 391                                                                 u8 *blks)
 392 {
 393         struct request_queue *q = nvmdev->q;
 394         struct nvm_geo *geo = &nvmdev->geo;
 395         struct nvme_ns *ns = q->queuedata;
 396         struct nvme_ctrl *ctrl = ns->ctrl;
 397         struct nvme_nvm_command c = {};
 398         struct nvme_nvm_bb_tbl *bb_tbl;
 399         int nr_blks = geo->blks_per_lun * geo->plane_mode;
 400         int tblsz = sizeof(struct nvme_nvm_bb_tbl) + nr_blks;
 401         int ret = 0;
 402
 403         c.get_bb.opcode = nvme_nvm_admin_get_bb_tbl;
 404         c.get_bb.nsid = cpu_to_le32(ns->ns_id);
 405         c.get_bb.spba = cpu_to_le64(ppa.ppa);
 406
 407         bb_tbl = kzalloc(tblsz, GFP_KERNEL);
 408         if (!bb_tbl)
 409                 return -ENOMEM;
 410
 411         ret = nvme_submit_sync_cmd(ctrl->admin_q, (struct nvme_command *)&c,
 412                                                                 bb_tbl, tblsz);
 413         if (ret) {
 414                 dev_err(ctrl->device, "get bad block table failed (%d)\n", ret);
 415                 ret = -EIO;
 416                 goto out;
 417         }
 418
 419         if (bb_tbl->tblid[0] != 'B' || bb_tbl->tblid[1] != 'B' ||
 420                 bb_tbl->tblid[2] != 'L' || bb_tbl->tblid[3] != 'T') {
 421                 dev_err(ctrl->device, "bbt format mismatch\n");
 422                 ret = -EINVAL;
 423                 goto out;
 424         }
 425
 426         if (le16_to_cpu(bb_tbl->verid) != 1) {
 427                 ret = -EINVAL;
 428                 dev_err(ctrl->device, "bbt version not supported\n");
 429                 goto out;
 430         }
 431
 432         if (le32_to_cpu(bb_tbl->tblks) != nr_blks) {
 433                 ret = -EINVAL;
 434                 dev_err(ctrl->device,
 435                                 "bbt unsuspected blocks returned (%u!=%u)",
 436                                 le32_to_cpu(bb_tbl->tblks), nr_blks);
 437                 goto out;
 438         }
 439
 440         memcpy(blks, bb_tbl->blk, geo->blks_per_lun * geo->plane_mode);
 441 out:
 442         kfree(bb_tbl);
 443         return ret;
 444 }
 445
 446 static int nvme_nvm_set_bb_tbl(struct nvm_dev *nvmdev, struct ppa_addr *ppas,
 447                                                         int nr_ppas, int type)
 448 {
 449         struct nvme_ns *ns = nvmdev->q->queuedata;
 450         struct nvme_nvm_command c = {};
 451         int ret = 0;
 452
 453         c.set_bb.opcode = nvme_nvm_admin_set_bb_tbl;
 454         c.set_bb.nsid = cpu_to_le32(ns->ns_id);
 455         c.set_bb.spba = cpu_to_le64(ppas->ppa);
 456         c.set_bb.nlb = cpu_to_le16(nr_ppas - 1);
 457         c.set_bb.value = type;
 458
 459         ret = nvme_submit_sync_cmd(ns->ctrl->admin_q, (struct nvme_command *)&c,
 460                                                                 NULL, 0);
 461         if (ret)
 462                 dev_err(ns->ctrl->device, "set bad block table failed (%d)\n",
 463                                                                         ret);
 464         return ret;
 465 }
 466
 467 static inline void nvme_nvm_rqtocmd(struct request *rq, struct nvm_rq *rqd,
 468                                 struct nvme_ns *ns, struct nvme_nvm_command *c)
 469 {
 470         c->ph_rw.opcode = rqd->opcode;
 471         c->ph_rw.nsid = cpu_to_le32(ns->ns_id);
 472         c->ph_rw.spba = cpu_to_le64(rqd->ppa_addr.ppa);
 473         c->ph_rw.metadata = cpu_to_le64(rqd->dma_meta_list);
 474         c->ph_rw.control = cpu_to_le16(rqd->flags);
 475         c->ph_rw.length = cpu_to_le16(rqd->nr_ppas - 1);
 476
 477         if (rqd->opcode == NVM_OP_HBWRITE || rqd->opcode == NVM_OP_HBREAD)
 478                 c->hb_rw.slba = cpu_to_le64(nvme_block_nr(ns,
 479                                         rqd->bio->bi_iter.bi_sector));
 480 }
 481
 482 static void nvme_nvm_end_io(struct request *rq, int error)
 483 {
 484         struct nvm_rq *rqd = rq->end_io_data;
 485
 486         rqd->ppa_status = nvme_req(rq)->result.u64;
 487         rqd->error = error;
 488         nvm_end_io(rqd);
 489
 490         kfree(nvme_req(rq)->cmd);
 491         blk_mq_free_request(rq);
 492 }
 493
 494 static int nvme_nvm_submit_io(struct nvm_dev *dev, struct nvm_rq *rqd)
 495 {
 496         struct request_queue *q = dev->q;
 497         struct nvme_ns *ns = q->queuedata;
 498         struct request *rq;
 499         struct bio *bio = rqd->bio;
 500         struct nvme_nvm_command *cmd;
 501
 502         cmd = kzalloc(sizeof(struct nvme_nvm_command), GFP_KERNEL);
 503         if (!cmd)
 504                 return -ENOMEM;
 505
 506         rq = nvme_alloc_request(q, (struct nvme_command *)cmd, 0, NVME_QID_ANY);
 507         if (IS_ERR(rq)) {
 508                 kfree(cmd);
 509                 return -ENOMEM;
 510         }
 511         rq->cmd_flags &= ~REQ_FAILFAST_DRIVER;
 512
 513         rq->ioprio = bio_prio(bio);
 514         if (bio_has_data(bio))
 515                 rq->nr_phys_segments = bio_phys_segments(q, bio);
 516
 517         rq->__data_len = bio->bi_iter.bi_size;
 518         rq->bio = rq->biotail = bio;
 519
 520         nvme_nvm_rqtocmd(rq, rqd, ns, cmd);
 521
 522         rq->end_io_data = rqd;
 523
 524         blk_execute_rq_nowait(q, NULL, rq, 0, nvme_nvm_end_io);
 525
 526         return 0;
 527 }
 528
 529 static int nvme_nvm_erase_block(struct nvm_dev *dev, struct nvm_rq *rqd)
 530 {
 531         struct request_queue *q = dev->q;
 532         struct nvme_ns *ns = q->queuedata;
 533         struct nvme_nvm_command c = {};
 534
 535         c.erase.opcode = NVM_OP_ERASE;
 536         c.erase.nsid = cpu_to_le32(ns->ns_id);
 537         c.erase.spba = cpu_to_le64(rqd->ppa_addr.ppa);
 538         c.erase.length = cpu_to_le16(rqd->nr_ppas - 1);
 539         c.erase.control = cpu_to_le16(rqd->flags);
 540
 541         return nvme_submit_sync_cmd(q, (struct nvme_command *)&c, NULL, 0);
 542 }
 543
 544 static void *nvme_nvm_create_dma_pool(struct nvm_dev *nvmdev, char *name)
 545 {
 546         struct nvme_ns *ns = nvmdev->q->queuedata;
 547
 548         return dma_pool_create(name, ns->ctrl->dev, PAGE_SIZE, PAGE_SIZE, 0);
 549 }
 550
 551 static void nvme_nvm_destroy_dma_pool(void *pool)
 552 {
 553         struct dma_pool *dma_pool = pool;
 554
 555         dma_pool_destroy(dma_pool);
 556 }
 557
 558 static void *nvme_nvm_dev_dma_alloc(struct nvm_dev *dev, void *pool,
 559                                     gfp_t mem_flags, dma_addr_t *dma_handler)
 560 {
 561         return dma_pool_alloc(pool, mem_flags, dma_handler);
 562 }
 563
 564 static void nvme_nvm_dev_dma_free(void *pool, void *addr,
 565                                                         dma_addr_t dma_handler)
 566 {
 567         dma_pool_free(pool, addr, dma_handler);
 568 }
 569
 570 static struct nvm_dev_ops nvme_nvm_dev_ops = {
 571         .identity               = nvme_nvm_identity,
 572
 573         .get_l2p_tbl            = nvme_nvm_get_l2p_tbl,
 574
 575         .get_bb_tbl             = nvme_nvm_get_bb_tbl,
 576         .set_bb_tbl             = nvme_nvm_set_bb_tbl,
 577
 578         .submit_io              = nvme_nvm_submit_io,
 579         .erase_block            = nvme_nvm_erase_block,
 580
 581         .create_dma_pool        = nvme_nvm_create_dma_pool,
 582         .destroy_dma_pool       = nvme_nvm_destroy_dma_pool,
 583         .dev_dma_alloc          = nvme_nvm_dev_dma_alloc,
 584         .dev_dma_free           = nvme_nvm_dev_dma_free,
 585
 586         .max_phys_sect          = 64,
 587 };
 588
 589 static void nvme_nvm_end_user_vio(struct request *rq, int error)
 590 {
 591         struct completion *waiting = rq->end_io_data;
 592
 593         complete(waiting);
 594 }
 595
 596 static int nvme_nvm_submit_user_cmd(struct request_queue *q,
 597                                 struct nvme_ns *ns,
 598                                 struct nvme_nvm_command *vcmd,
 599                                 void __user *ubuf, unsigned int bufflen,
 600                                 void __user *meta_buf, unsigned int meta_len,
 601                                 void __user *ppa_buf, unsigned int ppa_len,
 602                                 u32 *result, u64 *status, unsigned int timeout)
 603 {
 604         bool write = nvme_is_write((struct nvme_command *)vcmd);
 605         struct nvm_dev *dev = ns->ndev;
 606         struct gendisk *disk = ns->disk;
 607         struct request *rq;
 608         struct bio *bio = NULL;
 609         __le64 *ppa_list = NULL;
 610         dma_addr_t ppa_dma;
 611         __le64 *metadata = NULL;
 612         dma_addr_t metadata_dma;
 613         DECLARE_COMPLETION_ONSTACK(wait);
 614         int ret;
 615
 616         rq = nvme_alloc_request(q, (struct nvme_command *)vcmd, 0,
 617                         NVME_QID_ANY);
 618         if (IS_ERR(rq)) {
 619                 ret = -ENOMEM;
 620                 goto err_cmd;
 621         }
 622
 623         rq->timeout = timeout ? timeout : ADMIN_TIMEOUT;
 624
 625         rq->cmd_flags &= ~REQ_FAILFAST_DRIVER;
 626         rq->end_io_data = &wait;
 627
 628         if (ppa_buf && ppa_len) {
 629                 ppa_list = dma_pool_alloc(dev->dma_pool, GFP_KERNEL, &ppa_dma);
 630                 if (!ppa_list) {
 631                         ret = -ENOMEM;
 632                         goto err_rq;
 633                 }
 634                 if (copy_from_user(ppa_list, (void __user *)ppa_buf,
 635                                                 sizeof(u64) * (ppa_len + 1))) {
 636                         ret = -EFAULT;
 637                         goto err_ppa;
 638                 }
 639                 vcmd->ph_rw.spba = cpu_to_le64(ppa_dma);
 640         } else {
 641                 vcmd->ph_rw.spba = cpu_to_le64((uintptr_t)ppa_buf);
 642         }
 643
 644         if (ubuf && bufflen) {
 645                 ret = blk_rq_map_user(q, rq, NULL, ubuf, bufflen, GFP_KERNEL);
 646                 if (ret)
 647                         goto err_ppa;
 648                 bio = rq->bio;
 649
 650                 if (meta_buf && meta_len) {
 651                         metadata = dma_pool_alloc(dev->dma_pool, GFP_KERNEL,
 652                                                                 &metadata_dma);
 653                         if (!metadata) {
 654                                 ret = -ENOMEM;
 655                                 goto err_map;
 656                         }
 657
 658                         if (write) {
 659                                 if (copy_from_user(metadata,
 660                                                 (void __user *)meta_buf,
 661                                                 meta_len)) {
 662                                         ret = -EFAULT;
 663                                         goto err_meta;
 664                                 }
 665                         }
 666                         vcmd->ph_rw.metadata = cpu_to_le64(metadata_dma);
 667                 }
 668
 669                 if (!disk)
 670                         goto submit;
 671
 672                 bio->bi_bdev = bdget_disk(disk, 0);
 673                 if (!bio->bi_bdev) {
 674                         ret = -ENODEV;
 675                         goto err_meta;
 676                 }
 677         }
 678
 679 submit:
 680         blk_execute_rq_nowait(q, NULL, rq, 0, nvme_nvm_end_user_vio);
 681
 682         wait_for_completion_io(&wait);
 683
 684         ret = nvme_error_status(rq->errors);
 685         if (result)
 686                 *result = rq->errors & 0x7ff;
 687         if (status)
 688                 *status = le64_to_cpu(nvme_req(rq)->result.u64);
 689
 690         if (metadata && !ret && !write) {
 691                 if (copy_to_user(meta_buf, (void *)metadata, meta_len))
 692                         ret = -EFAULT;
 693         }
 694 err_meta:
 695         if (meta_buf && meta_len)
 696                 dma_pool_free(dev->dma_pool, metadata, metadata_dma);
 697 err_map:
 698         if (bio) {
 699                 if (disk && bio->bi_bdev)
 700                         bdput(bio->bi_bdev);
 701                 blk_rq_unmap_user(bio);
 702         }
 703 err_ppa:
 704         if (ppa_buf && ppa_len)
 705                 dma_pool_free(dev->dma_pool, ppa_list, ppa_dma);
 706 err_rq:
 707         blk_mq_free_request(rq);
 708 err_cmd:
 709         return ret;
 710 }
 711
 712 static int nvme_nvm_submit_vio(struct nvme_ns *ns,
 713                                         struct nvm_user_vio __user *uvio)
 714 {
 715         struct nvm_user_vio vio;
 716         struct nvme_nvm_command c;
 717         unsigned int length;
 718         int ret;
 719
 720         if (copy_from_user(&vio, uvio, sizeof(vio)))
 721                 return -EFAULT;
 722         if (vio.flags)
 723                 return -EINVAL;
 724
 725         memset(&c, 0, sizeof(c));
 726         c.ph_rw.opcode = vio.opcode;
 727         c.ph_rw.nsid = cpu_to_le32(ns->ns_id);
 728         c.ph_rw.control = cpu_to_le16(vio.control);
 729         c.ph_rw.length = cpu_to_le16(vio.nppas);
 730
 731         length = (vio.nppas + 1) << ns->lba_shift;
 732
 733         ret = nvme_nvm_submit_user_cmd(ns->queue, ns, &c,
 734                         (void __user *)(uintptr_t)vio.addr, length,
 735                         (void __user *)(uintptr_t)vio.metadata,
 736                                                         vio.metadata_len,
 737                         (void __user *)(uintptr_t)vio.ppa_list, vio.nppas,
 738                         &vio.result, &vio.status, 0);
 739
 740         if (ret && copy_to_user(uvio, &vio, sizeof(vio)))
 741                 return -EFAULT;
 742
 743         return ret;
 744 }
 745
 746 static int nvme_nvm_user_vcmd(struct nvme_ns *ns, int admin,
 747                                         struct nvm_passthru_vio __user *uvcmd)
 748 {
 749         struct nvm_passthru_vio vcmd;
 750         struct nvme_nvm_command c;
 751         struct request_queue *q;
 752         unsigned int timeout = 0;
 753         int ret;
 754
 755         if (copy_from_user(&vcmd, uvcmd, sizeof(vcmd)))
 756                 return -EFAULT;
 757         if ((vcmd.opcode != 0xF2) && (!capable(CAP_SYS_ADMIN)))
 758                 return -EACCES;
 759         if (vcmd.flags)
 760                 return -EINVAL;
 761
 762         memset(&c, 0, sizeof(c));
 763         c.common.opcode = vcmd.opcode;
 764         c.common.nsid = cpu_to_le32(ns->ns_id);
 765         c.common.cdw2[0] = cpu_to_le32(vcmd.cdw2);
 766         c.common.cdw2[1] = cpu_to_le32(vcmd.cdw3);
 767         /* cdw11-12 */
 768         c.ph_rw.length = cpu_to_le16(vcmd.nppas);
 769         c.ph_rw.control  = cpu_to_le32(vcmd.control);
 770         c.common.cdw10[3] = cpu_to_le32(vcmd.cdw13);
 771         c.common.cdw10[4] = cpu_to_le32(vcmd.cdw14);
 772         c.common.cdw10[5] = cpu_to_le32(vcmd.cdw15);
 773
 774         if (vcmd.timeout_ms)
 775                 timeout = msecs_to_jiffies(vcmd.timeout_ms);
 776
 777         q = admin ? ns->ctrl->admin_q : ns->queue;
 778
 779         ret = nvme_nvm_submit_user_cmd(q, ns,
 780                         (struct nvme_nvm_command *)&c,
 781                         (void __user *)(uintptr_t)vcmd.addr, vcmd.data_len,
 782                         (void __user *)(uintptr_t)vcmd.metadata,
 783                                                         vcmd.metadata_len,
 784                         (void __user *)(uintptr_t)vcmd.ppa_list, vcmd.nppas,
 785                         &vcmd.result, &vcmd.status, timeout);
 786
 787         if (ret && copy_to_user(uvcmd, &vcmd, sizeof(vcmd)))
 788                 return -EFAULT;
 789
 790         return ret;
 791 }
 792
 793 int nvme_nvm_ioctl(struct nvme_ns *ns, unsigned int cmd, unsigned long arg)
 794 {
 795         switch (cmd) {
 796         case NVME_NVM_IOCTL_ADMIN_VIO:
 797                 return nvme_nvm_user_vcmd(ns, 1, (void __user *)arg);
 798         case NVME_NVM_IOCTL_IO_VIO:
 799                 return nvme_nvm_user_vcmd(ns, 0, (void __user *)arg);
 800         case NVME_NVM_IOCTL_SUBMIT_VIO:
 801                 return nvme_nvm_submit_vio(ns, (void __user *)arg);
 802         default:
 803                 return -ENOTTY;
 804         }
 805 }
 806
 807 int nvme_nvm_register(struct nvme_ns *ns, char *disk_name, int node)
 808 {
 809         struct request_queue *q = ns->queue;
 810         struct nvm_dev *dev;
 811
 812         dev = nvm_alloc_dev(node);
 813         if (!dev)
 814                 return -ENOMEM;
 815
 816         dev->q = q;
 817         memcpy(dev->name, disk_name, DISK_NAME_LEN);
 818         dev->ops = &nvme_nvm_dev_ops;
 819         dev->private_data = ns;
 820         ns->ndev = dev;
 821
 822         return nvm_register(dev);
 823 }
 824
 825 void nvme_nvm_unregister(struct nvme_ns *ns)
 826 {
 827         nvm_unregister(ns->ndev);
 828 }
 829
 830 static ssize_t nvm_dev_attr_show(struct device *dev,
 831                                  struct device_attribute *dattr, char *page)
 832 {
 833         struct nvme_ns *ns = nvme_get_ns_from_dev(dev);
 834         struct nvm_dev *ndev = ns->ndev;
 835         struct nvm_id *id;
 836         struct nvm_id_group *grp;
 837         struct attribute *attr;
 838
 839         if (!ndev)
 840                 return 0;
 841
 842         id = &ndev->identity;
 843         grp = &id->grp;
 844         attr = &dattr->attr;
 845
 846         if (strcmp(attr->name, "version") == 0) {
 847                 return scnprintf(page, PAGE_SIZE, "%u\n", id->ver_id);
 848         } else if (strcmp(attr->name, "vendor_opcode") == 0) {
 849                 return scnprintf(page, PAGE_SIZE, "%u\n", id->vmnt);
 850         } else if (strcmp(attr->name, "capabilities") == 0) {
 851                 return scnprintf(page, PAGE_SIZE, "%u\n", id->cap);
 852         } else if (strcmp(attr->name, "device_mode") == 0) {
 853                 return scnprintf(page, PAGE_SIZE, "%u\n", id->dom);
 854         /* kept for compatibility */
 855         } else if (strcmp(attr->name, "media_manager") == 0) {
 856                 return scnprintf(page, PAGE_SIZE, "%s\n", "gennvm");
 857         } else if (strcmp(attr->name, "ppa_format") == 0) {
 858                 return scnprintf(page, PAGE_SIZE,
 859                         "0x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x\n",
 860                         id->ppaf.ch_offset, id->ppaf.ch_len,
 861                         id->ppaf.lun_offset, id->ppaf.lun_len,
 862                         id->ppaf.pln_offset, id->ppaf.pln_len,
 863                         id->ppaf.blk_offset, id->ppaf.blk_len,
 864                         id->ppaf.pg_offset, id->ppaf.pg_len,
 865                         id->ppaf.sect_offset, id->ppaf.sect_len);
 866         } else if (strcmp(attr->name, "media_type") == 0) {     /* u8 */
 867                 return scnprintf(page, PAGE_SIZE, "%u\n", grp->mtype);
 868         } else if (strcmp(attr->name, "flash_media_type") == 0) {
 869                 return scnprintf(page, PAGE_SIZE, "%u\n", grp->fmtype);
 870         } else if (strcmp(attr->name, "num_channels") == 0) {
 871                 return scnprintf(page, PAGE_SIZE, "%u\n", grp->num_ch);
 872         } else if (strcmp(attr->name, "num_luns") == 0) {
 873                 return scnprintf(page, PAGE_SIZE, "%u\n", grp->num_lun);
 874         } else if (strcmp(attr->name, "num_planes") == 0) {
 875                 return scnprintf(page, PAGE_SIZE, "%u\n", grp->num_pln);
 876         } else if (strcmp(attr->name, "num_blocks") == 0) {     /* u16 */
 877                 return scnprintf(page, PAGE_SIZE, "%u\n", grp->num_blk);
 878         } else if (strcmp(attr->name, "num_pages") == 0) {
 879                 return scnprintf(page, PAGE_SIZE, "%u\n", grp->num_pg);
 880         } else if (strcmp(attr->name, "page_size") == 0) {
 881                 return scnprintf(page, PAGE_SIZE, "%u\n", grp->fpg_sz);
 882         } else if (strcmp(attr->name, "hw_sector_size") == 0) {
 883                 return scnprintf(page, PAGE_SIZE, "%u\n", grp->csecs);
 884         } else if (strcmp(attr->name, "oob_sector_size") == 0) {/* u32 */
 885                 return scnprintf(page, PAGE_SIZE, "%u\n", grp->sos);
 886         } else if (strcmp(attr->name, "read_typ") == 0) {
 887                 return scnprintf(page, PAGE_SIZE, "%u\n", grp->trdt);
 888         } else if (strcmp(attr->name, "read_max") == 0) {
 889                 return scnprintf(page, PAGE_SIZE, "%u\n", grp->trdm);
 890         } else if (strcmp(attr->name, "prog_typ") == 0) {
 891                 return scnprintf(page, PAGE_SIZE, "%u\n", grp->tprt);
 892         } else if (strcmp(attr->name, "prog_max") == 0) {
 893                 return scnprintf(page, PAGE_SIZE, "%u\n", grp->tprm);
 894         } else if (strcmp(attr->name, "erase_typ") == 0) {
 895                 return scnprintf(page, PAGE_SIZE, "%u\n", grp->tbet);
 896         } else if (strcmp(attr->name, "erase_max") == 0) {
 897                 return scnprintf(page, PAGE_SIZE, "%u\n", grp->tbem);
 898         } else if (strcmp(attr->name, "multiplane_modes") == 0) {
 899                 return scnprintf(page, PAGE_SIZE, "0x%08x\n", grp->mpos);
 900         } else if (strcmp(attr->name, "media_capabilities") == 0) {
 901                 return scnprintf(page, PAGE_SIZE, "0x%08x\n", grp->mccap);
 902         } else if (strcmp(attr->name, "max_phys_secs") == 0) {
 903                 return scnprintf(page, PAGE_SIZE, "%u\n",
 904                                 ndev->ops->max_phys_sect);
 905         } else {
 906                 return scnprintf(page,
 907                                  PAGE_SIZE,
 908                                  "Unhandled attr(%s) in `nvm_dev_attr_show`\n",
 909                                  attr->name);
 910         }
 911 }
 912
 913 #define NVM_DEV_ATTR_RO(_name)                                          \
 914         DEVICE_ATTR(_name, S_IRUGO, nvm_dev_attr_show, NULL)
 915
 916 static NVM_DEV_ATTR_RO(version);
 917 static NVM_DEV_ATTR_RO(vendor_opcode);
 918 static NVM_DEV_ATTR_RO(capabilities);
 919 static NVM_DEV_ATTR_RO(device_mode);
 920 static NVM_DEV_ATTR_RO(ppa_format);
 921 static NVM_DEV_ATTR_RO(media_manager);
 922
 923 static NVM_DEV_ATTR_RO(media_type);
 924 static NVM_DEV_ATTR_RO(flash_media_type);
 925 static NVM_DEV_ATTR_RO(num_channels);
 926 static NVM_DEV_ATTR_RO(num_luns);
 927 static NVM_DEV_ATTR_RO(num_planes);
 928 static NVM_DEV_ATTR_RO(num_blocks);
 929 static NVM_DEV_ATTR_RO(num_pages);
 930 static NVM_DEV_ATTR_RO(page_size);
 931 static NVM_DEV_ATTR_RO(hw_sector_size);
 932 static NVM_DEV_ATTR_RO(oob_sector_size);
 933 static NVM_DEV_ATTR_RO(read_typ);
 934 static NVM_DEV_ATTR_RO(read_max);
 935 static NVM_DEV_ATTR_RO(prog_typ);
 936 static NVM_DEV_ATTR_RO(prog_max);
 937 static NVM_DEV_ATTR_RO(erase_typ);
 938 static NVM_DEV_ATTR_RO(erase_max);
 939 static NVM_DEV_ATTR_RO(multiplane_modes);
 940 static NVM_DEV_ATTR_RO(media_capabilities);
 941 static NVM_DEV_ATTR_RO(max_phys_secs);
 942
 943 static struct attribute *nvm_dev_attrs[] = {
 944         &dev_attr_version.attr,
 945         &dev_attr_vendor_opcode.attr,
 946         &dev_attr_capabilities.attr,
 947         &dev_attr_device_mode.attr,
 948         &dev_attr_media_manager.attr,
 949
 950         &dev_attr_ppa_format.attr,
 951         &dev_attr_media_type.attr,
 952         &dev_attr_flash_media_type.attr,
 953         &dev_attr_num_channels.attr,
 954         &dev_attr_num_luns.attr,
 955         &dev_attr_num_planes.attr,
 956         &dev_attr_num_blocks.attr,
 957         &dev_attr_num_pages.attr,
 958         &dev_attr_page_size.attr,
 959         &dev_attr_hw_sector_size.attr,
 960         &dev_attr_oob_sector_size.attr,
 961         &dev_attr_read_typ.attr,
 962         &dev_attr_read_max.attr,
 963         &dev_attr_prog_typ.attr,
 964         &dev_attr_prog_max.attr,
 965         &dev_attr_erase_typ.attr,
 966         &dev_attr_erase_max.attr,
 967         &dev_attr_multiplane_modes.attr,
 968         &dev_attr_media_capabilities.attr,
 969         &dev_attr_max_phys_secs.attr,
 970         NULL,
 971 };
 972
 973 static const struct attribute_group nvm_dev_attr_group = {
 974         .name           = "lightnvm",
 975         .attrs          = nvm_dev_attrs,
 976 };
 977
 978 int nvme_nvm_register_sysfs(struct nvme_ns *ns)
 979 {
 980         return sysfs_create_group(&disk_to_dev(ns->disk)->kobj,
 981                                         &nvm_dev_attr_group);
 982 }
 983
 984 void nvme_nvm_unregister_sysfs(struct nvme_ns *ns)
 985 {
 986         sysfs_remove_group(&disk_to_dev(ns->disk)->kobj,
 987                                         &nvm_dev_attr_group);
 988 }
 989
 990 /* move to shared place when used in multiple places. */
 991 #define PCI_VENDOR_ID_CNEX 0x1d1d
 992 #define PCI_DEVICE_ID_CNEX_WL 0x2807
 993 #define PCI_DEVICE_ID_CNEX_QEMU 0x1f1f
 994
 995 int nvme_nvm_ns_supported(struct nvme_ns *ns, struct nvme_id_ns *id)
 996 {
 997         struct nvme_ctrl *ctrl = ns->ctrl;
 998         /* XXX: this is poking into PCI structures from generic code! */
 999         struct pci_dev *pdev = to_pci_dev(ctrl->dev);
1000
1001         /* QEMU NVMe simulator - PCI ID + Vendor specific bit */
1002         if (pdev->vendor == PCI_VENDOR_ID_CNEX &&
1003                                 pdev->device == PCI_DEVICE_ID_CNEX_QEMU &&
1004                                                         id->vs[0] == 0x1)
1005                 return 1;
1006
1007         /* CNEX Labs - PCI ID + Vendor specific bit */
1008         if (pdev->vendor == PCI_VENDOR_ID_CNEX &&
1009                                 pdev->device == PCI_DEVICE_ID_CNEX_WL &&
1010                                                         id->vs[0] == 0x1)
1011                 return 1;
1012
1013         return 0;
1014 }