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_bb_tbl       = 0xf2,
  35         nvme_nvm_admin_set_bb_tbl       = 0xf1,
  36 };
  37
  38 struct nvme_nvm_ph_rw {
  39         __u8                    opcode;
  40         __u8                    flags;
  41         __u16                   command_id;
  42         __le32                  nsid;
  43         __u64                   rsvd2;
  44         __le64                  metadata;
  45         __le64                  prp1;
  46         __le64                  prp2;
  47         __le64                  spba;
  48         __le16                  length;
  49         __le16                  control;
  50         __le32                  dsmgmt;
  51         __le64                  resv;
  52 };
  53
  54 struct nvme_nvm_identity {
  55         __u8                    opcode;
  56         __u8                    flags;
  57         __u16                   command_id;
  58         __le32                  nsid;
  59         __u64                   rsvd[2];
  60         __le64                  prp1;
  61         __le64                  prp2;
  62         __u32                   rsvd11[6];
  63 };
  64
  65 struct nvme_nvm_getbbtbl {
  66         __u8                    opcode;
  67         __u8                    flags;
  68         __u16                   command_id;
  69         __le32                  nsid;
  70         __u64                   rsvd[2];
  71         __le64                  prp1;
  72         __le64                  prp2;
  73         __le64                  spba;
  74         __u32                   rsvd4[4];
  75 };
  76
  77 struct nvme_nvm_setbbtbl {
  78         __u8                    opcode;
  79         __u8                    flags;
  80         __u16                   command_id;
  81         __le32                  nsid;
  82         __le64                  rsvd[2];
  83         __le64                  prp1;
  84         __le64                  prp2;
  85         __le64                  spba;
  86         __le16                  nlb;
  87         __u8                    value;
  88         __u8                    rsvd3;
  89         __u32                   rsvd4[3];
  90 };
  91
  92 struct nvme_nvm_erase_blk {
  93         __u8                    opcode;
  94         __u8                    flags;
  95         __u16                   command_id;
  96         __le32                  nsid;
  97         __u64                   rsvd[2];
  98         __le64                  prp1;
  99         __le64                  prp2;
 100         __le64                  spba;
 101         __le16                  length;
 102         __le16                  control;
 103         __le32                  dsmgmt;
 104         __le64                  resv;
 105 };
 106
 107 struct nvme_nvm_command {
 108         union {
 109                 struct nvme_common_command common;
 110                 struct nvme_nvm_identity identity;
 111                 struct nvme_nvm_ph_rw ph_rw;
 112                 struct nvme_nvm_getbbtbl get_bb;
 113                 struct nvme_nvm_setbbtbl set_bb;
 114                 struct nvme_nvm_erase_blk erase;
 115         };
 116 };
 117
 118 #define NVME_NVM_LP_MLC_PAIRS 886
 119 struct nvme_nvm_lp_mlc {
 120         __le16                  num_pairs;
 121         __u8                    pairs[NVME_NVM_LP_MLC_PAIRS];
 122 };
 123
 124 struct nvme_nvm_lp_tbl {
 125         __u8                    id[8];
 126         struct nvme_nvm_lp_mlc  mlc;
 127 };
 128
 129 struct nvme_nvm_id_group {
 130         __u8                    mtype;
 131         __u8                    fmtype;
 132         __le16                  res16;
 133         __u8                    num_ch;
 134         __u8                    num_lun;
 135         __u8                    num_pln;
 136         __u8                    rsvd1;
 137         __le16                  num_chk;
 138         __le16                  num_pg;
 139         __le16                  fpg_sz;
 140         __le16                  csecs;
 141         __le16                  sos;
 142         __le16                  rsvd2;
 143         __le32                  trdt;
 144         __le32                  trdm;
 145         __le32                  tprt;
 146         __le32                  tprm;
 147         __le32                  tbet;
 148         __le32                  tbem;
 149         __le32                  mpos;
 150         __le32                  mccap;
 151         __le16                  cpar;
 152         __u8                    reserved[10];
 153         struct nvme_nvm_lp_tbl lptbl;
 154 } __packed;
 155
 156 struct nvme_nvm_addr_format {
 157         __u8                    ch_offset;
 158         __u8                    ch_len;
 159         __u8                    lun_offset;
 160         __u8                    lun_len;
 161         __u8                    pln_offset;
 162         __u8                    pln_len;
 163         __u8                    blk_offset;
 164         __u8                    blk_len;
 165         __u8                    pg_offset;
 166         __u8                    pg_len;
 167         __u8                    sect_offset;
 168         __u8                    sect_len;
 169         __u8                    res[4];
 170 } __packed;
 171
 172 struct nvme_nvm_id {
 173         __u8                    ver_id;
 174         __u8                    vmnt;
 175         __u8                    cgrps;
 176         __u8                    res;
 177         __le32                  cap;
 178         __le32                  dom;
 179         struct nvme_nvm_addr_format ppaf;
 180         __u8                    resv[228];
 181         struct nvme_nvm_id_group groups[4];
 182 } __packed;
 183
 184 struct nvme_nvm_bb_tbl {
 185         __u8    tblid[4];
 186         __le16  verid;
 187         __le16  revid;
 188         __le32  rvsd1;
 189         __le32  tblks;
 190         __le32  tfact;
 191         __le32  tgrown;
 192         __le32  tdresv;
 193         __le32  thresv;
 194         __le32  rsvd2[8];
 195         __u8    blk[0];
 196 };
 197
 198 /*
 199  * Check we didn't inadvertently grow the command struct
 200  */
 201 static inline void _nvme_nvm_check_size(void)
 202 {
 203         BUILD_BUG_ON(sizeof(struct nvme_nvm_identity) != 64);
 204         BUILD_BUG_ON(sizeof(struct nvme_nvm_ph_rw) != 64);
 205         BUILD_BUG_ON(sizeof(struct nvme_nvm_getbbtbl) != 64);
 206         BUILD_BUG_ON(sizeof(struct nvme_nvm_setbbtbl) != 64);
 207         BUILD_BUG_ON(sizeof(struct nvme_nvm_erase_blk) != 64);
 208         BUILD_BUG_ON(sizeof(struct nvme_nvm_id_group) != 960);
 209         BUILD_BUG_ON(sizeof(struct nvme_nvm_addr_format) != 16);
 210         BUILD_BUG_ON(sizeof(struct nvme_nvm_id) != NVME_IDENTIFY_DATA_SIZE);
 211         BUILD_BUG_ON(sizeof(struct nvme_nvm_bb_tbl) != 64);
 212 }
 213
 214 static int init_grps(struct nvm_id *nvm_id, struct nvme_nvm_id *nvme_nvm_id)
 215 {
 216         struct nvme_nvm_id_group *src;
 217         struct nvm_id_group *grp;
 218         int sec_per_pg, sec_per_pl, pg_per_blk;
 219
 220         if (nvme_nvm_id->cgrps != 1)
 221                 return -EINVAL;
 222
 223         src = &nvme_nvm_id->groups[0];
 224         grp = &nvm_id->grp;
 225
 226         grp->mtype = src->mtype;
 227         grp->fmtype = src->fmtype;
 228
 229         grp->num_ch = src->num_ch;
 230         grp->num_lun = src->num_lun;
 231
 232         grp->num_chk = le16_to_cpu(src->num_chk);
 233         grp->csecs = le16_to_cpu(src->csecs);
 234         grp->sos = le16_to_cpu(src->sos);
 235
 236         pg_per_blk = le16_to_cpu(src->num_pg);
 237         sec_per_pg = le16_to_cpu(src->fpg_sz) / grp->csecs;
 238         sec_per_pl = sec_per_pg * src->num_pln;
 239         grp->clba = sec_per_pl * pg_per_blk;
 240         grp->ws_per_chk = pg_per_blk;
 241
 242         grp->mpos = le32_to_cpu(src->mpos);
 243         grp->cpar = le16_to_cpu(src->cpar);
 244         grp->mccap = le32_to_cpu(src->mccap);
 245
 246         grp->ws_opt = grp->ws_min = sec_per_pg;
 247         grp->ws_seq = NVM_IO_SNGL_ACCESS;
 248
 249         if (grp->mpos & 0x020202) {
 250                 grp->ws_seq = NVM_IO_DUAL_ACCESS;
 251                 grp->ws_opt <<= 1;
 252         } else if (grp->mpos & 0x040404) {
 253                 grp->ws_seq = NVM_IO_QUAD_ACCESS;
 254                 grp->ws_opt <<= 2;
 255         }
 256
 257         grp->trdt = le32_to_cpu(src->trdt);
 258         grp->trdm = le32_to_cpu(src->trdm);
 259         grp->tprt = le32_to_cpu(src->tprt);
 260         grp->tprm = le32_to_cpu(src->tprm);
 261         grp->tbet = le32_to_cpu(src->tbet);
 262         grp->tbem = le32_to_cpu(src->tbem);
 263
 264         /* 1.2 compatibility */
 265         grp->num_pln = src->num_pln;
 266         grp->num_pg = le16_to_cpu(src->num_pg);
 267         grp->fpg_sz = le16_to_cpu(src->fpg_sz);
 268
 269         return 0;
 270 }
 271
 272 static int nvme_nvm_identity(struct nvm_dev *nvmdev, struct nvm_id *nvm_id)
 273 {
 274         struct nvme_ns *ns = nvmdev->q->queuedata;
 275         struct nvme_nvm_id *nvme_nvm_id;
 276         struct nvme_nvm_command c = {};
 277         int ret;
 278
 279         c.identity.opcode = nvme_nvm_admin_identity;
 280         c.identity.nsid = cpu_to_le32(ns->head->ns_id);
 281
 282         nvme_nvm_id = kmalloc(sizeof(struct nvme_nvm_id), GFP_KERNEL);
 283         if (!nvme_nvm_id)
 284                 return -ENOMEM;
 285
 286         ret = nvme_submit_sync_cmd(ns->ctrl->admin_q, (struct nvme_command *)&c,
 287                                 nvme_nvm_id, sizeof(struct nvme_nvm_id));
 288         if (ret) {
 289                 ret = -EIO;
 290                 goto out;
 291         }
 292
 293         nvm_id->ver_id = nvme_nvm_id->ver_id;
 294         nvm_id->vmnt = nvme_nvm_id->vmnt;
 295         nvm_id->cap = le32_to_cpu(nvme_nvm_id->cap);
 296         nvm_id->dom = le32_to_cpu(nvme_nvm_id->dom);
 297         memcpy(&nvm_id->ppaf, &nvme_nvm_id->ppaf,
 298                                         sizeof(struct nvm_addr_format));
 299
 300         ret = init_grps(nvm_id, nvme_nvm_id);
 301 out:
 302         kfree(nvme_nvm_id);
 303         return ret;
 304 }
 305
 306 static int nvme_nvm_get_bb_tbl(struct nvm_dev *nvmdev, struct ppa_addr ppa,
 307                                                                 u8 *blks)
 308 {
 309         struct request_queue *q = nvmdev->q;
 310         struct nvm_geo *geo = &nvmdev->geo;
 311         struct nvme_ns *ns = q->queuedata;
 312         struct nvme_ctrl *ctrl = ns->ctrl;
 313         struct nvme_nvm_command c = {};
 314         struct nvme_nvm_bb_tbl *bb_tbl;
 315         int nr_blks = geo->nr_chks * geo->plane_mode;
 316         int tblsz = sizeof(struct nvme_nvm_bb_tbl) + nr_blks;
 317         int ret = 0;
 318
 319         c.get_bb.opcode = nvme_nvm_admin_get_bb_tbl;
 320         c.get_bb.nsid = cpu_to_le32(ns->head->ns_id);
 321         c.get_bb.spba = cpu_to_le64(ppa.ppa);
 322
 323         bb_tbl = kzalloc(tblsz, GFP_KERNEL);
 324         if (!bb_tbl)
 325                 return -ENOMEM;
 326
 327         ret = nvme_submit_sync_cmd(ctrl->admin_q, (struct nvme_command *)&c,
 328                                                                 bb_tbl, tblsz);
 329         if (ret) {
 330                 dev_err(ctrl->device, "get bad block table failed (%d)\n", ret);
 331                 ret = -EIO;
 332                 goto out;
 333         }
 334
 335         if (bb_tbl->tblid[0] != 'B' || bb_tbl->tblid[1] != 'B' ||
 336                 bb_tbl->tblid[2] != 'L' || bb_tbl->tblid[3] != 'T') {
 337                 dev_err(ctrl->device, "bbt format mismatch\n");
 338                 ret = -EINVAL;
 339                 goto out;
 340         }
 341
 342         if (le16_to_cpu(bb_tbl->verid) != 1) {
 343                 ret = -EINVAL;
 344                 dev_err(ctrl->device, "bbt version not supported\n");
 345                 goto out;
 346         }
 347
 348         if (le32_to_cpu(bb_tbl->tblks) != nr_blks) {
 349                 ret = -EINVAL;
 350                 dev_err(ctrl->device,
 351                                 "bbt unsuspected blocks returned (%u!=%u)",
 352                                 le32_to_cpu(bb_tbl->tblks), nr_blks);
 353                 goto out;
 354         }
 355
 356         memcpy(blks, bb_tbl->blk, geo->nr_chks * geo->plane_mode);
 357 out:
 358         kfree(bb_tbl);
 359         return ret;
 360 }
 361
 362 static int nvme_nvm_set_bb_tbl(struct nvm_dev *nvmdev, struct ppa_addr *ppas,
 363                                                         int nr_ppas, int type)
 364 {
 365         struct nvme_ns *ns = nvmdev->q->queuedata;
 366         struct nvme_nvm_command c = {};
 367         int ret = 0;
 368
 369         c.set_bb.opcode = nvme_nvm_admin_set_bb_tbl;
 370         c.set_bb.nsid = cpu_to_le32(ns->head->ns_id);
 371         c.set_bb.spba = cpu_to_le64(ppas->ppa);
 372         c.set_bb.nlb = cpu_to_le16(nr_ppas - 1);
 373         c.set_bb.value = type;
 374
 375         ret = nvme_submit_sync_cmd(ns->ctrl->admin_q, (struct nvme_command *)&c,
 376                                                                 NULL, 0);
 377         if (ret)
 378                 dev_err(ns->ctrl->device, "set bad block table failed (%d)\n",
 379                                                                         ret);
 380         return ret;
 381 }
 382
 383 static inline void nvme_nvm_rqtocmd(struct nvm_rq *rqd, struct nvme_ns *ns,
 384                                     struct nvme_nvm_command *c)
 385 {
 386         c->ph_rw.opcode = rqd->opcode;
 387         c->ph_rw.nsid = cpu_to_le32(ns->head->ns_id);
 388         c->ph_rw.spba = cpu_to_le64(rqd->ppa_addr.ppa);
 389         c->ph_rw.metadata = cpu_to_le64(rqd->dma_meta_list);
 390         c->ph_rw.control = cpu_to_le16(rqd->flags);
 391         c->ph_rw.length = cpu_to_le16(rqd->nr_ppas - 1);
 392 }
 393
 394 static void nvme_nvm_end_io(struct request *rq, blk_status_t status)
 395 {
 396         struct nvm_rq *rqd = rq->end_io_data;
 397
 398         rqd->ppa_status = le64_to_cpu(nvme_req(rq)->result.u64);
 399         rqd->error = nvme_req(rq)->status;
 400         nvm_end_io(rqd);
 401
 402         kfree(nvme_req(rq)->cmd);
 403         blk_mq_free_request(rq);
 404 }
 405
 406 static struct request *nvme_nvm_alloc_request(struct request_queue *q,
 407                                               struct nvm_rq *rqd,
 408                                               struct nvme_nvm_command *cmd)
 409 {
 410         struct nvme_ns *ns = q->queuedata;
 411         struct request *rq;
 412
 413         nvme_nvm_rqtocmd(rqd, ns, cmd);
 414
 415         rq = nvme_alloc_request(q, (struct nvme_command *)cmd, 0, NVME_QID_ANY);
 416         if (IS_ERR(rq))
 417                 return rq;
 418
 419         rq->cmd_flags &= ~REQ_FAILFAST_DRIVER;
 420
 421         if (rqd->bio) {
 422                 blk_init_request_from_bio(rq, rqd->bio);
 423         } else {
 424                 rq->ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, IOPRIO_NORM);
 425                 rq->__data_len = 0;
 426         }
 427
 428         return rq;
 429 }
 430
 431 static int nvme_nvm_submit_io(struct nvm_dev *dev, struct nvm_rq *rqd)
 432 {
 433         struct request_queue *q = dev->q;
 434         struct nvme_nvm_command *cmd;
 435         struct request *rq;
 436
 437         cmd = kzalloc(sizeof(struct nvme_nvm_command), GFP_KERNEL);
 438         if (!cmd)
 439                 return -ENOMEM;
 440
 441         rq = nvme_nvm_alloc_request(q, rqd, cmd);
 442         if (IS_ERR(rq)) {
 443                 kfree(cmd);
 444                 return PTR_ERR(rq);
 445         }
 446
 447         rq->end_io_data = rqd;
 448
 449         blk_execute_rq_nowait(q, NULL, rq, 0, nvme_nvm_end_io);
 450
 451         return 0;
 452 }
 453
 454 static int nvme_nvm_submit_io_sync(struct nvm_dev *dev, struct nvm_rq *rqd)
 455 {
 456         struct request_queue *q = dev->q;
 457         struct request *rq;
 458         struct nvme_nvm_command cmd;
 459         int ret = 0;
 460
 461         memset(&cmd, 0, sizeof(struct nvme_nvm_command));
 462
 463         rq = nvme_nvm_alloc_request(q, rqd, &cmd);
 464         if (IS_ERR(rq))
 465                 return PTR_ERR(rq);
 466
 467         /* I/Os can fail and the error is signaled through rqd. Callers must
 468          * handle the error accordingly.
 469          */
 470         blk_execute_rq(q, NULL, rq, 0);
 471         if (nvme_req(rq)->flags & NVME_REQ_CANCELLED)
 472                 ret = -EINTR;
 473
 474         rqd->ppa_status = le64_to_cpu(nvme_req(rq)->result.u64);
 475         rqd->error = nvme_req(rq)->status;
 476
 477         blk_mq_free_request(rq);
 478
 479         return ret;
 480 }
 481
 482 static void *nvme_nvm_create_dma_pool(struct nvm_dev *nvmdev, char *name)
 483 {
 484         struct nvme_ns *ns = nvmdev->q->queuedata;
 485
 486         return dma_pool_create(name, ns->ctrl->dev, PAGE_SIZE, PAGE_SIZE, 0);
 487 }
 488
 489 static void nvme_nvm_destroy_dma_pool(void *pool)
 490 {
 491         struct dma_pool *dma_pool = pool;
 492
 493         dma_pool_destroy(dma_pool);
 494 }
 495
 496 static void *nvme_nvm_dev_dma_alloc(struct nvm_dev *dev, void *pool,
 497                                     gfp_t mem_flags, dma_addr_t *dma_handler)
 498 {
 499         return dma_pool_alloc(pool, mem_flags, dma_handler);
 500 }
 501
 502 static void nvme_nvm_dev_dma_free(void *pool, void *addr,
 503                                                         dma_addr_t dma_handler)
 504 {
 505         dma_pool_free(pool, addr, dma_handler);
 506 }
 507
 508 static struct nvm_dev_ops nvme_nvm_dev_ops = {
 509         .identity               = nvme_nvm_identity,
 510
 511         .get_bb_tbl             = nvme_nvm_get_bb_tbl,
 512         .set_bb_tbl             = nvme_nvm_set_bb_tbl,
 513
 514         .submit_io              = nvme_nvm_submit_io,
 515         .submit_io_sync         = nvme_nvm_submit_io_sync,
 516
 517         .create_dma_pool        = nvme_nvm_create_dma_pool,
 518         .destroy_dma_pool       = nvme_nvm_destroy_dma_pool,
 519         .dev_dma_alloc          = nvme_nvm_dev_dma_alloc,
 520         .dev_dma_free           = nvme_nvm_dev_dma_free,
 521
 522         .max_phys_sect          = 64,
 523 };
 524
 525 static int nvme_nvm_submit_user_cmd(struct request_queue *q,
 526                                 struct nvme_ns *ns,
 527                                 struct nvme_nvm_command *vcmd,
 528                                 void __user *ubuf, unsigned int bufflen,
 529                                 void __user *meta_buf, unsigned int meta_len,
 530                                 void __user *ppa_buf, unsigned int ppa_len,
 531                                 u32 *result, u64 *status, unsigned int timeout)
 532 {
 533         bool write = nvme_is_write((struct nvme_command *)vcmd);
 534         struct nvm_dev *dev = ns->ndev;
 535         struct gendisk *disk = ns->disk;
 536         struct request *rq;
 537         struct bio *bio = NULL;
 538         __le64 *ppa_list = NULL;
 539         dma_addr_t ppa_dma;
 540         __le64 *metadata = NULL;
 541         dma_addr_t metadata_dma;
 542         DECLARE_COMPLETION_ONSTACK(wait);
 543         int ret = 0;
 544
 545         rq = nvme_alloc_request(q, (struct nvme_command *)vcmd, 0,
 546                         NVME_QID_ANY);
 547         if (IS_ERR(rq)) {
 548                 ret = -ENOMEM;
 549                 goto err_cmd;
 550         }
 551
 552         rq->timeout = timeout ? timeout : ADMIN_TIMEOUT;
 553
 554         if (ppa_buf && ppa_len) {
 555                 ppa_list = dma_pool_alloc(dev->dma_pool, GFP_KERNEL, &ppa_dma);
 556                 if (!ppa_list) {
 557                         ret = -ENOMEM;
 558                         goto err_rq;
 559                 }
 560                 if (copy_from_user(ppa_list, (void __user *)ppa_buf,
 561                                                 sizeof(u64) * (ppa_len + 1))) {
 562                         ret = -EFAULT;
 563                         goto err_ppa;
 564                 }
 565                 vcmd->ph_rw.spba = cpu_to_le64(ppa_dma);
 566         } else {
 567                 vcmd->ph_rw.spba = cpu_to_le64((uintptr_t)ppa_buf);
 568         }
 569
 570         if (ubuf && bufflen) {
 571                 ret = blk_rq_map_user(q, rq, NULL, ubuf, bufflen, GFP_KERNEL);
 572                 if (ret)
 573                         goto err_ppa;
 574                 bio = rq->bio;
 575
 576                 if (meta_buf && meta_len) {
 577                         metadata = dma_pool_alloc(dev->dma_pool, GFP_KERNEL,
 578                                                                 &metadata_dma);
 579                         if (!metadata) {
 580                                 ret = -ENOMEM;
 581                                 goto err_map;
 582                         }
 583
 584                         if (write) {
 585                                 if (copy_from_user(metadata,
 586                                                 (void __user *)meta_buf,
 587                                                 meta_len)) {
 588                                         ret = -EFAULT;
 589                                         goto err_meta;
 590                                 }
 591                         }
 592                         vcmd->ph_rw.metadata = cpu_to_le64(metadata_dma);
 593                 }
 594
 595                 bio->bi_disk = disk;
 596         }
 597
 598         blk_execute_rq(q, NULL, rq, 0);
 599
 600         if (nvme_req(rq)->flags & NVME_REQ_CANCELLED)
 601                 ret = -EINTR;
 602         else if (nvme_req(rq)->status & 0x7ff)
 603                 ret = -EIO;
 604         if (result)
 605                 *result = nvme_req(rq)->status & 0x7ff;
 606         if (status)
 607                 *status = le64_to_cpu(nvme_req(rq)->result.u64);
 608
 609         if (metadata && !ret && !write) {
 610                 if (copy_to_user(meta_buf, (void *)metadata, meta_len))
 611                         ret = -EFAULT;
 612         }
 613 err_meta:
 614         if (meta_buf && meta_len)
 615                 dma_pool_free(dev->dma_pool, metadata, metadata_dma);
 616 err_map:
 617         if (bio)
 618                 blk_rq_unmap_user(bio);
 619 err_ppa:
 620         if (ppa_buf && ppa_len)
 621                 dma_pool_free(dev->dma_pool, ppa_list, ppa_dma);
 622 err_rq:
 623         blk_mq_free_request(rq);
 624 err_cmd:
 625         return ret;
 626 }
 627
 628 static int nvme_nvm_submit_vio(struct nvme_ns *ns,
 629                                         struct nvm_user_vio __user *uvio)
 630 {
 631         struct nvm_user_vio vio;
 632         struct nvme_nvm_command c;
 633         unsigned int length;
 634         int ret;
 635
 636         if (copy_from_user(&vio, uvio, sizeof(vio)))
 637                 return -EFAULT;
 638         if (vio.flags)
 639                 return -EINVAL;
 640
 641         memset(&c, 0, sizeof(c));
 642         c.ph_rw.opcode = vio.opcode;
 643         c.ph_rw.nsid = cpu_to_le32(ns->head->ns_id);
 644         c.ph_rw.control = cpu_to_le16(vio.control);
 645         c.ph_rw.length = cpu_to_le16(vio.nppas);
 646
 647         length = (vio.nppas + 1) << ns->lba_shift;
 648
 649         ret = nvme_nvm_submit_user_cmd(ns->queue, ns, &c,
 650                         (void __user *)(uintptr_t)vio.addr, length,
 651                         (void __user *)(uintptr_t)vio.metadata,
 652                                                         vio.metadata_len,
 653                         (void __user *)(uintptr_t)vio.ppa_list, vio.nppas,
 654                         &vio.result, &vio.status, 0);
 655
 656         if (ret && copy_to_user(uvio, &vio, sizeof(vio)))
 657                 return -EFAULT;
 658
 659         return ret;
 660 }
 661
 662 static int nvme_nvm_user_vcmd(struct nvme_ns *ns, int admin,
 663                                         struct nvm_passthru_vio __user *uvcmd)
 664 {
 665         struct nvm_passthru_vio vcmd;
 666         struct nvme_nvm_command c;
 667         struct request_queue *q;
 668         unsigned int timeout = 0;
 669         int ret;
 670
 671         if (copy_from_user(&vcmd, uvcmd, sizeof(vcmd)))
 672                 return -EFAULT;
 673         if ((vcmd.opcode != 0xF2) && (!capable(CAP_SYS_ADMIN)))
 674                 return -EACCES;
 675         if (vcmd.flags)
 676                 return -EINVAL;
 677
 678         memset(&c, 0, sizeof(c));
 679         c.common.opcode = vcmd.opcode;
 680         c.common.nsid = cpu_to_le32(ns->head->ns_id);
 681         c.common.cdw2[0] = cpu_to_le32(vcmd.cdw2);
 682         c.common.cdw2[1] = cpu_to_le32(vcmd.cdw3);
 683         /* cdw11-12 */
 684         c.ph_rw.length = cpu_to_le16(vcmd.nppas);
 685         c.ph_rw.control  = cpu_to_le16(vcmd.control);
 686         c.common.cdw10[3] = cpu_to_le32(vcmd.cdw13);
 687         c.common.cdw10[4] = cpu_to_le32(vcmd.cdw14);
 688         c.common.cdw10[5] = cpu_to_le32(vcmd.cdw15);
 689
 690         if (vcmd.timeout_ms)
 691                 timeout = msecs_to_jiffies(vcmd.timeout_ms);
 692
 693         q = admin ? ns->ctrl->admin_q : ns->queue;
 694
 695         ret = nvme_nvm_submit_user_cmd(q, ns,
 696                         (struct nvme_nvm_command *)&c,
 697                         (void __user *)(uintptr_t)vcmd.addr, vcmd.data_len,
 698                         (void __user *)(uintptr_t)vcmd.metadata,
 699                                                         vcmd.metadata_len,
 700                         (void __user *)(uintptr_t)vcmd.ppa_list, vcmd.nppas,
 701                         &vcmd.result, &vcmd.status, timeout);
 702
 703         if (ret && copy_to_user(uvcmd, &vcmd, sizeof(vcmd)))
 704                 return -EFAULT;
 705
 706         return ret;
 707 }
 708
 709 int nvme_nvm_ioctl(struct nvme_ns *ns, unsigned int cmd, unsigned long arg)
 710 {
 711         switch (cmd) {
 712         case NVME_NVM_IOCTL_ADMIN_VIO:
 713                 return nvme_nvm_user_vcmd(ns, 1, (void __user *)arg);
 714         case NVME_NVM_IOCTL_IO_VIO:
 715                 return nvme_nvm_user_vcmd(ns, 0, (void __user *)arg);
 716         case NVME_NVM_IOCTL_SUBMIT_VIO:
 717                 return nvme_nvm_submit_vio(ns, (void __user *)arg);
 718         default:
 719                 return -ENOTTY;
 720         }
 721 }
 722
 723 int nvme_nvm_register(struct nvme_ns *ns, char *disk_name, int node)
 724 {
 725         struct request_queue *q = ns->queue;
 726         struct nvm_dev *dev;
 727
 728         _nvme_nvm_check_size();
 729
 730         dev = nvm_alloc_dev(node);
 731         if (!dev)
 732                 return -ENOMEM;
 733
 734         dev->q = q;
 735         memcpy(dev->name, disk_name, DISK_NAME_LEN);
 736         dev->ops = &nvme_nvm_dev_ops;
 737         dev->private_data = ns;
 738         ns->ndev = dev;
 739
 740         return nvm_register(dev);
 741 }
 742
 743 void nvme_nvm_unregister(struct nvme_ns *ns)
 744 {
 745         nvm_unregister(ns->ndev);
 746 }
 747
 748 static ssize_t nvm_dev_attr_show(struct device *dev,
 749                                  struct device_attribute *dattr, char *page)
 750 {
 751         struct nvme_ns *ns = nvme_get_ns_from_dev(dev);
 752         struct nvm_dev *ndev = ns->ndev;
 753         struct nvm_id *id;
 754         struct nvm_id_group *grp;
 755         struct attribute *attr;
 756
 757         if (!ndev)
 758                 return 0;
 759
 760         id = &ndev->identity;
 761         grp = &id->grp;
 762         attr = &dattr->attr;
 763
 764         if (strcmp(attr->name, "version") == 0) {
 765                 return scnprintf(page, PAGE_SIZE, "%u\n", id->ver_id);
 766         } else if (strcmp(attr->name, "vendor_opcode") == 0) {
 767                 return scnprintf(page, PAGE_SIZE, "%u\n", id->vmnt);
 768         } else if (strcmp(attr->name, "capabilities") == 0) {
 769                 return scnprintf(page, PAGE_SIZE, "%u\n", id->cap);
 770         } else if (strcmp(attr->name, "device_mode") == 0) {
 771                 return scnprintf(page, PAGE_SIZE, "%u\n", id->dom);
 772         /* kept for compatibility */
 773         } else if (strcmp(attr->name, "media_manager") == 0) {
 774                 return scnprintf(page, PAGE_SIZE, "%s\n", "gennvm");
 775         } else if (strcmp(attr->name, "ppa_format") == 0) {
 776                 return scnprintf(page, PAGE_SIZE,
 777                         "0x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x\n",
 778                         id->ppaf.ch_offset, id->ppaf.ch_len,
 779                         id->ppaf.lun_offset, id->ppaf.lun_len,
 780                         id->ppaf.pln_offset, id->ppaf.pln_len,
 781                         id->ppaf.blk_offset, id->ppaf.blk_len,
 782                         id->ppaf.pg_offset, id->ppaf.pg_len,
 783                         id->ppaf.sect_offset, id->ppaf.sect_len);
 784         } else if (strcmp(attr->name, "media_type") == 0) {     /* u8 */
 785                 return scnprintf(page, PAGE_SIZE, "%u\n", grp->mtype);
 786         } else if (strcmp(attr->name, "flash_media_type") == 0) {
 787                 return scnprintf(page, PAGE_SIZE, "%u\n", grp->fmtype);
 788         } else if (strcmp(attr->name, "num_channels") == 0) {
 789                 return scnprintf(page, PAGE_SIZE, "%u\n", grp->num_ch);
 790         } else if (strcmp(attr->name, "num_luns") == 0) {
 791                 return scnprintf(page, PAGE_SIZE, "%u\n", grp->num_lun);
 792         } else if (strcmp(attr->name, "num_planes") == 0) {
 793                 return scnprintf(page, PAGE_SIZE, "%u\n", grp->num_pln);
 794         } else if (strcmp(attr->name, "num_blocks") == 0) {     /* u16 */
 795                 return scnprintf(page, PAGE_SIZE, "%u\n", grp->num_chk);
 796         } else if (strcmp(attr->name, "num_pages") == 0) {
 797                 return scnprintf(page, PAGE_SIZE, "%u\n", grp->num_pg);
 798         } else if (strcmp(attr->name, "page_size") == 0) {
 799                 return scnprintf(page, PAGE_SIZE, "%u\n", grp->fpg_sz);
 800         } else if (strcmp(attr->name, "hw_sector_size") == 0) {
 801                 return scnprintf(page, PAGE_SIZE, "%u\n", grp->csecs);
 802         } else if (strcmp(attr->name, "oob_sector_size") == 0) {/* u32 */
 803                 return scnprintf(page, PAGE_SIZE, "%u\n", grp->sos);
 804         } else if (strcmp(attr->name, "read_typ") == 0) {
 805                 return scnprintf(page, PAGE_SIZE, "%u\n", grp->trdt);
 806         } else if (strcmp(attr->name, "read_max") == 0) {
 807                 return scnprintf(page, PAGE_SIZE, "%u\n", grp->trdm);
 808         } else if (strcmp(attr->name, "prog_typ") == 0) {
 809                 return scnprintf(page, PAGE_SIZE, "%u\n", grp->tprt);
 810         } else if (strcmp(attr->name, "prog_max") == 0) {
 811                 return scnprintf(page, PAGE_SIZE, "%u\n", grp->tprm);
 812         } else if (strcmp(attr->name, "erase_typ") == 0) {
 813                 return scnprintf(page, PAGE_SIZE, "%u\n", grp->tbet);
 814         } else if (strcmp(attr->name, "erase_max") == 0) {
 815                 return scnprintf(page, PAGE_SIZE, "%u\n", grp->tbem);
 816         } else if (strcmp(attr->name, "multiplane_modes") == 0) {
 817                 return scnprintf(page, PAGE_SIZE, "0x%08x\n", grp->mpos);
 818         } else if (strcmp(attr->name, "media_capabilities") == 0) {
 819                 return scnprintf(page, PAGE_SIZE, "0x%08x\n", grp->mccap);
 820         } else if (strcmp(attr->name, "max_phys_secs") == 0) {
 821                 return scnprintf(page, PAGE_SIZE, "%u\n",
 822                                 ndev->ops->max_phys_sect);
 823         } else {
 824                 return scnprintf(page,
 825                                  PAGE_SIZE,
 826                                  "Unhandled attr(%s) in `nvm_dev_attr_show`\n",
 827                                  attr->name);
 828         }
 829 }
 830
 831 #define NVM_DEV_ATTR_RO(_name)                                          \
 832         DEVICE_ATTR(_name, S_IRUGO, nvm_dev_attr_show, NULL)
 833
 834 static NVM_DEV_ATTR_RO(version);
 835 static NVM_DEV_ATTR_RO(vendor_opcode);
 836 static NVM_DEV_ATTR_RO(capabilities);
 837 static NVM_DEV_ATTR_RO(device_mode);
 838 static NVM_DEV_ATTR_RO(ppa_format);
 839 static NVM_DEV_ATTR_RO(media_manager);
 840
 841 static NVM_DEV_ATTR_RO(media_type);
 842 static NVM_DEV_ATTR_RO(flash_media_type);
 843 static NVM_DEV_ATTR_RO(num_channels);
 844 static NVM_DEV_ATTR_RO(num_luns);
 845 static NVM_DEV_ATTR_RO(num_planes);
 846 static NVM_DEV_ATTR_RO(num_blocks);
 847 static NVM_DEV_ATTR_RO(num_pages);
 848 static NVM_DEV_ATTR_RO(page_size);
 849 static NVM_DEV_ATTR_RO(hw_sector_size);
 850 static NVM_DEV_ATTR_RO(oob_sector_size);
 851 static NVM_DEV_ATTR_RO(read_typ);
 852 static NVM_DEV_ATTR_RO(read_max);
 853 static NVM_DEV_ATTR_RO(prog_typ);
 854 static NVM_DEV_ATTR_RO(prog_max);
 855 static NVM_DEV_ATTR_RO(erase_typ);
 856 static NVM_DEV_ATTR_RO(erase_max);
 857 static NVM_DEV_ATTR_RO(multiplane_modes);
 858 static NVM_DEV_ATTR_RO(media_capabilities);
 859 static NVM_DEV_ATTR_RO(max_phys_secs);
 860
 861 static struct attribute *nvm_dev_attrs[] = {
 862         &dev_attr_version.attr,
 863         &dev_attr_vendor_opcode.attr,
 864         &dev_attr_capabilities.attr,
 865         &dev_attr_device_mode.attr,
 866         &dev_attr_media_manager.attr,
 867
 868         &dev_attr_ppa_format.attr,
 869         &dev_attr_media_type.attr,
 870         &dev_attr_flash_media_type.attr,
 871         &dev_attr_num_channels.attr,
 872         &dev_attr_num_luns.attr,
 873         &dev_attr_num_planes.attr,
 874         &dev_attr_num_blocks.attr,
 875         &dev_attr_num_pages.attr,
 876         &dev_attr_page_size.attr,
 877         &dev_attr_hw_sector_size.attr,
 878         &dev_attr_oob_sector_size.attr,
 879         &dev_attr_read_typ.attr,
 880         &dev_attr_read_max.attr,
 881         &dev_attr_prog_typ.attr,
 882         &dev_attr_prog_max.attr,
 883         &dev_attr_erase_typ.attr,
 884         &dev_attr_erase_max.attr,
 885         &dev_attr_multiplane_modes.attr,
 886         &dev_attr_media_capabilities.attr,
 887         &dev_attr_max_phys_secs.attr,
 888         NULL,
 889 };
 890
 891 static const struct attribute_group nvm_dev_attr_group = {
 892         .name           = "lightnvm",
 893         .attrs          = nvm_dev_attrs,
 894 };
 895
 896 int nvme_nvm_register_sysfs(struct nvme_ns *ns)
 897 {
 898         return sysfs_create_group(&disk_to_dev(ns->disk)->kobj,
 899                                         &nvm_dev_attr_group);
 900 }
 901
 902 void nvme_nvm_unregister_sysfs(struct nvme_ns *ns)
 903 {
 904         sysfs_remove_group(&disk_to_dev(ns->disk)->kobj,
 905                                         &nvm_dev_attr_group);
 906 }