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_erase_blk {
  55         __u8                    opcode;
  56         __u8                    flags;
  57         __u16                   command_id;
  58         __le32                  nsid;
  59         __u64                   rsvd[2];
  60         __le64                  prp1;
  61         __le64                  prp2;
  62         __le64                  spba;
  63         __le16                  length;
  64         __le16                  control;
  65         __le32                  dsmgmt;
  66         __le64                  resv;
  67 };
  68
  69 struct nvme_nvm_identity {
  70         __u8                    opcode;
  71         __u8                    flags;
  72         __u16                   command_id;
  73         __le32                  nsid;
  74         __u64                   rsvd[2];
  75         __le64                  prp1;
  76         __le64                  prp2;
  77         __u32                   rsvd11[6];
  78 };
  79
  80 struct nvme_nvm_getbbtbl {
  81         __u8                    opcode;
  82         __u8                    flags;
  83         __u16                   command_id;
  84         __le32                  nsid;
  85         __u64                   rsvd[2];
  86         __le64                  prp1;
  87         __le64                  prp2;
  88         __le64                  spba;
  89         __u32                   rsvd4[4];
  90 };
  91
  92 struct nvme_nvm_setbbtbl {
  93         __u8                    opcode;
  94         __u8                    flags;
  95         __u16                   command_id;
  96         __le32                  nsid;
  97         __le64                  rsvd[2];
  98         __le64                  prp1;
  99         __le64                  prp2;
 100         __le64                  spba;
 101         __le16                  nlb;
 102         __u8                    value;
 103         __u8                    rsvd3;
 104         __u32                   rsvd4[3];
 105 };
 106
 107 struct nvme_nvm_command {
 108         union {
 109                 struct nvme_common_command common;
 110                 struct nvme_nvm_ph_rw ph_rw;
 111                 struct nvme_nvm_erase_blk erase;
 112                 struct nvme_nvm_identity identity;
 113                 struct nvme_nvm_getbbtbl get_bb;
 114                 struct nvme_nvm_setbbtbl set_bb;
 115         };
 116 };
 117
 118 struct nvme_nvm_id12_grp {
 119         __u8                    mtype;
 120         __u8                    fmtype;
 121         __le16                  res16;
 122         __u8                    num_ch;
 123         __u8                    num_lun;
 124         __u8                    num_pln;
 125         __u8                    rsvd1;
 126         __le16                  num_chk;
 127         __le16                  num_pg;
 128         __le16                  fpg_sz;
 129         __le16                  csecs;
 130         __le16                  sos;
 131         __le16                  rsvd2;
 132         __le32                  trdt;
 133         __le32                  trdm;
 134         __le32                  tprt;
 135         __le32                  tprm;
 136         __le32                  tbet;
 137         __le32                  tbem;
 138         __le32                  mpos;
 139         __le32                  mccap;
 140         __le16                  cpar;
 141         __u8                    reserved[906];
 142 } __packed;
 143
 144 struct nvme_nvm_id12_addrf {
 145         __u8                    ch_offset;
 146         __u8                    ch_len;
 147         __u8                    lun_offset;
 148         __u8                    lun_len;
 149         __u8                    pln_offset;
 150         __u8                    pln_len;
 151         __u8                    blk_offset;
 152         __u8                    blk_len;
 153         __u8                    pg_offset;
 154         __u8                    pg_len;
 155         __u8                    sect_offset;
 156         __u8                    sect_len;
 157         __u8                    res[4];
 158 } __packed;
 159
 160 struct nvme_nvm_id12 {
 161         __u8                    ver_id;
 162         __u8                    vmnt;
 163         __u8                    cgrps;
 164         __u8                    res;
 165         __le32                  cap;
 166         __le32                  dom;
 167         struct nvme_nvm_id12_addrf ppaf;
 168         __u8                    resv[228];
 169         struct nvme_nvm_id12_grp grp;
 170         __u8                    resv2[2880];
 171 } __packed;
 172
 173 struct nvme_nvm_bb_tbl {
 174         __u8    tblid[4];
 175         __le16  verid;
 176         __le16  revid;
 177         __le32  rvsd1;
 178         __le32  tblks;
 179         __le32  tfact;
 180         __le32  tgrown;
 181         __le32  tdresv;
 182         __le32  thresv;
 183         __le32  rsvd2[8];
 184         __u8    blk[0];
 185 };
 186
 187 /*
 188  * Check we didn't inadvertently grow the command struct
 189  */
 190 static inline void _nvme_nvm_check_size(void)
 191 {
 192         BUILD_BUG_ON(sizeof(struct nvme_nvm_identity) != 64);
 193         BUILD_BUG_ON(sizeof(struct nvme_nvm_ph_rw) != 64);
 194         BUILD_BUG_ON(sizeof(struct nvme_nvm_erase_blk) != 64);
 195         BUILD_BUG_ON(sizeof(struct nvme_nvm_getbbtbl) != 64);
 196         BUILD_BUG_ON(sizeof(struct nvme_nvm_setbbtbl) != 64);
 197         BUILD_BUG_ON(sizeof(struct nvme_nvm_id12_grp) != 960);
 198         BUILD_BUG_ON(sizeof(struct nvme_nvm_id12_addrf) != 16);
 199         BUILD_BUG_ON(sizeof(struct nvme_nvm_id12) != NVME_IDENTIFY_DATA_SIZE);
 200         BUILD_BUG_ON(sizeof(struct nvme_nvm_bb_tbl) != 64);
 201 }
 202
 203 static int init_grp(struct nvm_id *nvm_id, struct nvme_nvm_id12 *id12)
 204 {
 205         struct nvme_nvm_id12_grp *src;
 206         struct nvm_id_group *grp;
 207         int sec_per_pg, sec_per_pl, pg_per_blk;
 208
 209         if (id12->cgrps != 1)
 210                 return -EINVAL;
 211
 212         src = &id12->grp;
 213         grp = &nvm_id->grp;
 214
 215         grp->mtype = src->mtype;
 216         grp->fmtype = src->fmtype;
 217
 218         grp->num_ch = src->num_ch;
 219         grp->num_lun = src->num_lun;
 220
 221         grp->num_chk = le16_to_cpu(src->num_chk);
 222         grp->csecs = le16_to_cpu(src->csecs);
 223         grp->sos = le16_to_cpu(src->sos);
 224
 225         pg_per_blk = le16_to_cpu(src->num_pg);
 226         sec_per_pg = le16_to_cpu(src->fpg_sz) / grp->csecs;
 227         sec_per_pl = sec_per_pg * src->num_pln;
 228         grp->clba = sec_per_pl * pg_per_blk;
 229         grp->ws_per_chk = pg_per_blk;
 230
 231         grp->mpos = le32_to_cpu(src->mpos);
 232         grp->cpar = le16_to_cpu(src->cpar);
 233         grp->mccap = le32_to_cpu(src->mccap);
 234
 235         grp->ws_opt = grp->ws_min = sec_per_pg;
 236         grp->ws_seq = NVM_IO_SNGL_ACCESS;
 237
 238         if (grp->mpos & 0x020202) {
 239                 grp->ws_seq = NVM_IO_DUAL_ACCESS;
 240                 grp->ws_opt <<= 1;
 241         } else if (grp->mpos & 0x040404) {
 242                 grp->ws_seq = NVM_IO_QUAD_ACCESS;
 243                 grp->ws_opt <<= 2;
 244         }
 245
 246         grp->trdt = le32_to_cpu(src->trdt);
 247         grp->trdm = le32_to_cpu(src->trdm);
 248         grp->tprt = le32_to_cpu(src->tprt);
 249         grp->tprm = le32_to_cpu(src->tprm);
 250         grp->tbet = le32_to_cpu(src->tbet);
 251         grp->tbem = le32_to_cpu(src->tbem);
 252
 253         /* 1.2 compatibility */
 254         grp->num_pln = src->num_pln;
 255         grp->num_pg = le16_to_cpu(src->num_pg);
 256         grp->fpg_sz = le16_to_cpu(src->fpg_sz);
 257
 258         return 0;
 259 }
 260
 261 static int nvme_nvm_identity(struct nvm_dev *nvmdev, struct nvm_id *nvm_id)
 262 {
 263         struct nvme_ns *ns = nvmdev->q->queuedata;
 264         struct nvme_nvm_id12 *id;
 265         struct nvme_nvm_command c = {};
 266         int ret;
 267
 268         c.identity.opcode = nvme_nvm_admin_identity;
 269         c.identity.nsid = cpu_to_le32(ns->head->ns_id);
 270
 271         id = kmalloc(sizeof(struct nvme_nvm_id12), GFP_KERNEL);
 272         if (!id)
 273                 return -ENOMEM;
 274
 275         ret = nvme_submit_sync_cmd(ns->ctrl->admin_q, (struct nvme_command *)&c,
 276                                 id, sizeof(struct nvme_nvm_id12));
 277         if (ret) {
 278                 ret = -EIO;
 279                 goto out;
 280         }
 281
 282         nvm_id->ver_id = id->ver_id;
 283         nvm_id->vmnt = id->vmnt;
 284         nvm_id->cap = le32_to_cpu(id->cap);
 285         nvm_id->dom = le32_to_cpu(id->dom);
 286         memcpy(&nvm_id->ppaf, &id->ppaf,
 287                                         sizeof(struct nvm_addr_format));
 288
 289         ret = init_grp(nvm_id, id);
 290 out:
 291         kfree(id);
 292         return ret;
 293 }
 294
 295 static int nvme_nvm_get_bb_tbl(struct nvm_dev *nvmdev, struct ppa_addr ppa,
 296                                                                 u8 *blks)
 297 {
 298         struct request_queue *q = nvmdev->q;
 299         struct nvm_geo *geo = &nvmdev->geo;
 300         struct nvme_ns *ns = q->queuedata;
 301         struct nvme_ctrl *ctrl = ns->ctrl;
 302         struct nvme_nvm_command c = {};
 303         struct nvme_nvm_bb_tbl *bb_tbl;
 304         int nr_blks = geo->nr_chks * geo->plane_mode;
 305         int tblsz = sizeof(struct nvme_nvm_bb_tbl) + nr_blks;
 306         int ret = 0;
 307
 308         c.get_bb.opcode = nvme_nvm_admin_get_bb_tbl;
 309         c.get_bb.nsid = cpu_to_le32(ns->head->ns_id);
 310         c.get_bb.spba = cpu_to_le64(ppa.ppa);
 311
 312         bb_tbl = kzalloc(tblsz, GFP_KERNEL);
 313         if (!bb_tbl)
 314                 return -ENOMEM;
 315
 316         ret = nvme_submit_sync_cmd(ctrl->admin_q, (struct nvme_command *)&c,
 317                                                                 bb_tbl, tblsz);
 318         if (ret) {
 319                 dev_err(ctrl->device, "get bad block table failed (%d)\n", ret);
 320                 ret = -EIO;
 321                 goto out;
 322         }
 323
 324         if (bb_tbl->tblid[0] != 'B' || bb_tbl->tblid[1] != 'B' ||
 325                 bb_tbl->tblid[2] != 'L' || bb_tbl->tblid[3] != 'T') {
 326                 dev_err(ctrl->device, "bbt format mismatch\n");
 327                 ret = -EINVAL;
 328                 goto out;
 329         }
 330
 331         if (le16_to_cpu(bb_tbl->verid) != 1) {
 332                 ret = -EINVAL;
 333                 dev_err(ctrl->device, "bbt version not supported\n");
 334                 goto out;
 335         }
 336
 337         if (le32_to_cpu(bb_tbl->tblks) != nr_blks) {
 338                 ret = -EINVAL;
 339                 dev_err(ctrl->device,
 340                                 "bbt unsuspected blocks returned (%u!=%u)",
 341                                 le32_to_cpu(bb_tbl->tblks), nr_blks);
 342                 goto out;
 343         }
 344
 345         memcpy(blks, bb_tbl->blk, geo->nr_chks * geo->plane_mode);
 346 out:
 347         kfree(bb_tbl);
 348         return ret;
 349 }
 350
 351 static int nvme_nvm_set_bb_tbl(struct nvm_dev *nvmdev, struct ppa_addr *ppas,
 352                                                         int nr_ppas, int type)
 353 {
 354         struct nvme_ns *ns = nvmdev->q->queuedata;
 355         struct nvme_nvm_command c = {};
 356         int ret = 0;
 357
 358         c.set_bb.opcode = nvme_nvm_admin_set_bb_tbl;
 359         c.set_bb.nsid = cpu_to_le32(ns->head->ns_id);
 360         c.set_bb.spba = cpu_to_le64(ppas->ppa);
 361         c.set_bb.nlb = cpu_to_le16(nr_ppas - 1);
 362         c.set_bb.value = type;
 363
 364         ret = nvme_submit_sync_cmd(ns->ctrl->admin_q, (struct nvme_command *)&c,
 365                                                                 NULL, 0);
 366         if (ret)
 367                 dev_err(ns->ctrl->device, "set bad block table failed (%d)\n",
 368                                                                         ret);
 369         return ret;
 370 }
 371
 372 static inline void nvme_nvm_rqtocmd(struct nvm_rq *rqd, struct nvme_ns *ns,
 373                                     struct nvme_nvm_command *c)
 374 {
 375         c->ph_rw.opcode = rqd->opcode;
 376         c->ph_rw.nsid = cpu_to_le32(ns->head->ns_id);
 377         c->ph_rw.spba = cpu_to_le64(rqd->ppa_addr.ppa);
 378         c->ph_rw.metadata = cpu_to_le64(rqd->dma_meta_list);
 379         c->ph_rw.control = cpu_to_le16(rqd->flags);
 380         c->ph_rw.length = cpu_to_le16(rqd->nr_ppas - 1);
 381 }
 382
 383 static void nvme_nvm_end_io(struct request *rq, blk_status_t status)
 384 {
 385         struct nvm_rq *rqd = rq->end_io_data;
 386
 387         rqd->ppa_status = le64_to_cpu(nvme_req(rq)->result.u64);
 388         rqd->error = nvme_req(rq)->status;
 389         nvm_end_io(rqd);
 390
 391         kfree(nvme_req(rq)->cmd);
 392         blk_mq_free_request(rq);
 393 }
 394
 395 static struct request *nvme_nvm_alloc_request(struct request_queue *q,
 396                                               struct nvm_rq *rqd,
 397                                               struct nvme_nvm_command *cmd)
 398 {
 399         struct nvme_ns *ns = q->queuedata;
 400         struct request *rq;
 401
 402         nvme_nvm_rqtocmd(rqd, ns, cmd);
 403
 404         rq = nvme_alloc_request(q, (struct nvme_command *)cmd, 0, NVME_QID_ANY);
 405         if (IS_ERR(rq))
 406                 return rq;
 407
 408         rq->cmd_flags &= ~REQ_FAILFAST_DRIVER;
 409
 410         if (rqd->bio) {
 411                 blk_init_request_from_bio(rq, rqd->bio);
 412         } else {
 413                 rq->ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, IOPRIO_NORM);
 414                 rq->__data_len = 0;
 415         }
 416
 417         return rq;
 418 }
 419
 420 static int nvme_nvm_submit_io(struct nvm_dev *dev, struct nvm_rq *rqd)
 421 {
 422         struct request_queue *q = dev->q;
 423         struct nvme_nvm_command *cmd;
 424         struct request *rq;
 425
 426         cmd = kzalloc(sizeof(struct nvme_nvm_command), GFP_KERNEL);
 427         if (!cmd)
 428                 return -ENOMEM;
 429
 430         rq = nvme_nvm_alloc_request(q, rqd, cmd);
 431         if (IS_ERR(rq)) {
 432                 kfree(cmd);
 433                 return PTR_ERR(rq);
 434         }
 435
 436         rq->end_io_data = rqd;
 437
 438         blk_execute_rq_nowait(q, NULL, rq, 0, nvme_nvm_end_io);
 439
 440         return 0;
 441 }
 442
 443 static int nvme_nvm_submit_io_sync(struct nvm_dev *dev, struct nvm_rq *rqd)
 444 {
 445         struct request_queue *q = dev->q;
 446         struct request *rq;
 447         struct nvme_nvm_command cmd;
 448         int ret = 0;
 449
 450         memset(&cmd, 0, sizeof(struct nvme_nvm_command));
 451
 452         rq = nvme_nvm_alloc_request(q, rqd, &cmd);
 453         if (IS_ERR(rq))
 454                 return PTR_ERR(rq);
 455
 456         /* I/Os can fail and the error is signaled through rqd. Callers must
 457          * handle the error accordingly.
 458          */
 459         blk_execute_rq(q, NULL, rq, 0);
 460         if (nvme_req(rq)->flags & NVME_REQ_CANCELLED)
 461                 ret = -EINTR;
 462
 463         rqd->ppa_status = le64_to_cpu(nvme_req(rq)->result.u64);
 464         rqd->error = nvme_req(rq)->status;
 465
 466         blk_mq_free_request(rq);
 467
 468         return ret;
 469 }
 470
 471 static void *nvme_nvm_create_dma_pool(struct nvm_dev *nvmdev, char *name)
 472 {
 473         struct nvme_ns *ns = nvmdev->q->queuedata;
 474
 475         return dma_pool_create(name, ns->ctrl->dev, PAGE_SIZE, PAGE_SIZE, 0);
 476 }
 477
 478 static void nvme_nvm_destroy_dma_pool(void *pool)
 479 {
 480         struct dma_pool *dma_pool = pool;
 481
 482         dma_pool_destroy(dma_pool);
 483 }
 484
 485 static void *nvme_nvm_dev_dma_alloc(struct nvm_dev *dev, void *pool,
 486                                     gfp_t mem_flags, dma_addr_t *dma_handler)
 487 {
 488         return dma_pool_alloc(pool, mem_flags, dma_handler);
 489 }
 490
 491 static void nvme_nvm_dev_dma_free(void *pool, void *addr,
 492                                                         dma_addr_t dma_handler)
 493 {
 494         dma_pool_free(pool, addr, dma_handler);
 495 }
 496
 497 static struct nvm_dev_ops nvme_nvm_dev_ops = {
 498         .identity               = nvme_nvm_identity,
 499
 500         .get_bb_tbl             = nvme_nvm_get_bb_tbl,
 501         .set_bb_tbl             = nvme_nvm_set_bb_tbl,
 502
 503         .submit_io              = nvme_nvm_submit_io,
 504         .submit_io_sync         = nvme_nvm_submit_io_sync,
 505
 506         .create_dma_pool        = nvme_nvm_create_dma_pool,
 507         .destroy_dma_pool       = nvme_nvm_destroy_dma_pool,
 508         .dev_dma_alloc          = nvme_nvm_dev_dma_alloc,
 509         .dev_dma_free           = nvme_nvm_dev_dma_free,
 510
 511         .max_phys_sect          = 64,
 512 };
 513
 514 static int nvme_nvm_submit_user_cmd(struct request_queue *q,
 515                                 struct nvme_ns *ns,
 516                                 struct nvme_nvm_command *vcmd,
 517                                 void __user *ubuf, unsigned int bufflen,
 518                                 void __user *meta_buf, unsigned int meta_len,
 519                                 void __user *ppa_buf, unsigned int ppa_len,
 520                                 u32 *result, u64 *status, unsigned int timeout)
 521 {
 522         bool write = nvme_is_write((struct nvme_command *)vcmd);
 523         struct nvm_dev *dev = ns->ndev;
 524         struct gendisk *disk = ns->disk;
 525         struct request *rq;
 526         struct bio *bio = NULL;
 527         __le64 *ppa_list = NULL;
 528         dma_addr_t ppa_dma;
 529         __le64 *metadata = NULL;
 530         dma_addr_t metadata_dma;
 531         DECLARE_COMPLETION_ONSTACK(wait);
 532         int ret = 0;
 533
 534         rq = nvme_alloc_request(q, (struct nvme_command *)vcmd, 0,
 535                         NVME_QID_ANY);
 536         if (IS_ERR(rq)) {
 537                 ret = -ENOMEM;
 538                 goto err_cmd;
 539         }
 540
 541         rq->timeout = timeout ? timeout : ADMIN_TIMEOUT;
 542
 543         if (ppa_buf && ppa_len) {
 544                 ppa_list = dma_pool_alloc(dev->dma_pool, GFP_KERNEL, &ppa_dma);
 545                 if (!ppa_list) {
 546                         ret = -ENOMEM;
 547                         goto err_rq;
 548                 }
 549                 if (copy_from_user(ppa_list, (void __user *)ppa_buf,
 550                                                 sizeof(u64) * (ppa_len + 1))) {
 551                         ret = -EFAULT;
 552                         goto err_ppa;
 553                 }
 554                 vcmd->ph_rw.spba = cpu_to_le64(ppa_dma);
 555         } else {
 556                 vcmd->ph_rw.spba = cpu_to_le64((uintptr_t)ppa_buf);
 557         }
 558
 559         if (ubuf && bufflen) {
 560                 ret = blk_rq_map_user(q, rq, NULL, ubuf, bufflen, GFP_KERNEL);
 561                 if (ret)
 562                         goto err_ppa;
 563                 bio = rq->bio;
 564
 565                 if (meta_buf && meta_len) {
 566                         metadata = dma_pool_alloc(dev->dma_pool, GFP_KERNEL,
 567                                                                 &metadata_dma);
 568                         if (!metadata) {
 569                                 ret = -ENOMEM;
 570                                 goto err_map;
 571                         }
 572
 573                         if (write) {
 574                                 if (copy_from_user(metadata,
 575                                                 (void __user *)meta_buf,
 576                                                 meta_len)) {
 577                                         ret = -EFAULT;
 578                                         goto err_meta;
 579                                 }
 580                         }
 581                         vcmd->ph_rw.metadata = cpu_to_le64(metadata_dma);
 582                 }
 583
 584                 bio->bi_disk = disk;
 585         }
 586
 587         blk_execute_rq(q, NULL, rq, 0);
 588
 589         if (nvme_req(rq)->flags & NVME_REQ_CANCELLED)
 590                 ret = -EINTR;
 591         else if (nvme_req(rq)->status & 0x7ff)
 592                 ret = -EIO;
 593         if (result)
 594                 *result = nvme_req(rq)->status & 0x7ff;
 595         if (status)
 596                 *status = le64_to_cpu(nvme_req(rq)->result.u64);
 597
 598         if (metadata && !ret && !write) {
 599                 if (copy_to_user(meta_buf, (void *)metadata, meta_len))
 600                         ret = -EFAULT;
 601         }
 602 err_meta:
 603         if (meta_buf && meta_len)
 604                 dma_pool_free(dev->dma_pool, metadata, metadata_dma);
 605 err_map:
 606         if (bio)
 607                 blk_rq_unmap_user(bio);
 608 err_ppa:
 609         if (ppa_buf && ppa_len)
 610                 dma_pool_free(dev->dma_pool, ppa_list, ppa_dma);
 611 err_rq:
 612         blk_mq_free_request(rq);
 613 err_cmd:
 614         return ret;
 615 }
 616
 617 static int nvme_nvm_submit_vio(struct nvme_ns *ns,
 618                                         struct nvm_user_vio __user *uvio)
 619 {
 620         struct nvm_user_vio vio;
 621         struct nvme_nvm_command c;
 622         unsigned int length;
 623         int ret;
 624
 625         if (copy_from_user(&vio, uvio, sizeof(vio)))
 626                 return -EFAULT;
 627         if (vio.flags)
 628                 return -EINVAL;
 629
 630         memset(&c, 0, sizeof(c));
 631         c.ph_rw.opcode = vio.opcode;
 632         c.ph_rw.nsid = cpu_to_le32(ns->head->ns_id);
 633         c.ph_rw.control = cpu_to_le16(vio.control);
 634         c.ph_rw.length = cpu_to_le16(vio.nppas);
 635
 636         length = (vio.nppas + 1) << ns->lba_shift;
 637
 638         ret = nvme_nvm_submit_user_cmd(ns->queue, ns, &c,
 639                         (void __user *)(uintptr_t)vio.addr, length,
 640                         (void __user *)(uintptr_t)vio.metadata,
 641                                                         vio.metadata_len,
 642                         (void __user *)(uintptr_t)vio.ppa_list, vio.nppas,
 643                         &vio.result, &vio.status, 0);
 644
 645         if (ret && copy_to_user(uvio, &vio, sizeof(vio)))
 646                 return -EFAULT;
 647
 648         return ret;
 649 }
 650
 651 static int nvme_nvm_user_vcmd(struct nvme_ns *ns, int admin,
 652                                         struct nvm_passthru_vio __user *uvcmd)
 653 {
 654         struct nvm_passthru_vio vcmd;
 655         struct nvme_nvm_command c;
 656         struct request_queue *q;
 657         unsigned int timeout = 0;
 658         int ret;
 659
 660         if (copy_from_user(&vcmd, uvcmd, sizeof(vcmd)))
 661                 return -EFAULT;
 662         if ((vcmd.opcode != 0xF2) && (!capable(CAP_SYS_ADMIN)))
 663                 return -EACCES;
 664         if (vcmd.flags)
 665                 return -EINVAL;
 666
 667         memset(&c, 0, sizeof(c));
 668         c.common.opcode = vcmd.opcode;
 669         c.common.nsid = cpu_to_le32(ns->head->ns_id);
 670         c.common.cdw2[0] = cpu_to_le32(vcmd.cdw2);
 671         c.common.cdw2[1] = cpu_to_le32(vcmd.cdw3);
 672         /* cdw11-12 */
 673         c.ph_rw.length = cpu_to_le16(vcmd.nppas);
 674         c.ph_rw.control  = cpu_to_le16(vcmd.control);
 675         c.common.cdw10[3] = cpu_to_le32(vcmd.cdw13);
 676         c.common.cdw10[4] = cpu_to_le32(vcmd.cdw14);
 677         c.common.cdw10[5] = cpu_to_le32(vcmd.cdw15);
 678
 679         if (vcmd.timeout_ms)
 680                 timeout = msecs_to_jiffies(vcmd.timeout_ms);
 681
 682         q = admin ? ns->ctrl->admin_q : ns->queue;
 683
 684         ret = nvme_nvm_submit_user_cmd(q, ns,
 685                         (struct nvme_nvm_command *)&c,
 686                         (void __user *)(uintptr_t)vcmd.addr, vcmd.data_len,
 687                         (void __user *)(uintptr_t)vcmd.metadata,
 688                                                         vcmd.metadata_len,
 689                         (void __user *)(uintptr_t)vcmd.ppa_list, vcmd.nppas,
 690                         &vcmd.result, &vcmd.status, timeout);
 691
 692         if (ret && copy_to_user(uvcmd, &vcmd, sizeof(vcmd)))
 693                 return -EFAULT;
 694
 695         return ret;
 696 }
 697
 698 int nvme_nvm_ioctl(struct nvme_ns *ns, unsigned int cmd, unsigned long arg)
 699 {
 700         switch (cmd) {
 701         case NVME_NVM_IOCTL_ADMIN_VIO:
 702                 return nvme_nvm_user_vcmd(ns, 1, (void __user *)arg);
 703         case NVME_NVM_IOCTL_IO_VIO:
 704                 return nvme_nvm_user_vcmd(ns, 0, (void __user *)arg);
 705         case NVME_NVM_IOCTL_SUBMIT_VIO:
 706                 return nvme_nvm_submit_vio(ns, (void __user *)arg);
 707         default:
 708                 return -ENOTTY;
 709         }
 710 }
 711
 712 int nvme_nvm_register(struct nvme_ns *ns, char *disk_name, int node)
 713 {
 714         struct request_queue *q = ns->queue;
 715         struct nvm_dev *dev;
 716
 717         _nvme_nvm_check_size();
 718
 719         dev = nvm_alloc_dev(node);
 720         if (!dev)
 721                 return -ENOMEM;
 722
 723         dev->q = q;
 724         memcpy(dev->name, disk_name, DISK_NAME_LEN);
 725         dev->ops = &nvme_nvm_dev_ops;
 726         dev->private_data = ns;
 727         ns->ndev = dev;
 728
 729         return nvm_register(dev);
 730 }
 731
 732 void nvme_nvm_unregister(struct nvme_ns *ns)
 733 {
 734         nvm_unregister(ns->ndev);
 735 }
 736
 737 static ssize_t nvm_dev_attr_show(struct device *dev,
 738                                  struct device_attribute *dattr, char *page)
 739 {
 740         struct nvme_ns *ns = nvme_get_ns_from_dev(dev);
 741         struct nvm_dev *ndev = ns->ndev;
 742         struct nvm_id *id;
 743         struct nvm_id_group *grp;
 744         struct attribute *attr;
 745
 746         if (!ndev)
 747                 return 0;
 748
 749         id = &ndev->identity;
 750         grp = &id->grp;
 751         attr = &dattr->attr;
 752
 753         if (strcmp(attr->name, "version") == 0) {
 754                 return scnprintf(page, PAGE_SIZE, "%u\n", id->ver_id);
 755         } else if (strcmp(attr->name, "vendor_opcode") == 0) {
 756                 return scnprintf(page, PAGE_SIZE, "%u\n", id->vmnt);
 757         } else if (strcmp(attr->name, "capabilities") == 0) {
 758                 return scnprintf(page, PAGE_SIZE, "%u\n", id->cap);
 759         } else if (strcmp(attr->name, "device_mode") == 0) {
 760                 return scnprintf(page, PAGE_SIZE, "%u\n", id->dom);
 761         /* kept for compatibility */
 762         } else if (strcmp(attr->name, "media_manager") == 0) {
 763                 return scnprintf(page, PAGE_SIZE, "%s\n", "gennvm");
 764         } else if (strcmp(attr->name, "ppa_format") == 0) {
 765                 return scnprintf(page, PAGE_SIZE,
 766                         "0x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x\n",
 767                         id->ppaf.ch_offset, id->ppaf.ch_len,
 768                         id->ppaf.lun_offset, id->ppaf.lun_len,
 769                         id->ppaf.pln_offset, id->ppaf.pln_len,
 770                         id->ppaf.blk_offset, id->ppaf.blk_len,
 771                         id->ppaf.pg_offset, id->ppaf.pg_len,
 772                         id->ppaf.sect_offset, id->ppaf.sect_len);
 773         } else if (strcmp(attr->name, "media_type") == 0) {     /* u8 */
 774                 return scnprintf(page, PAGE_SIZE, "%u\n", grp->mtype);
 775         } else if (strcmp(attr->name, "flash_media_type") == 0) {
 776                 return scnprintf(page, PAGE_SIZE, "%u\n", grp->fmtype);
 777         } else if (strcmp(attr->name, "num_channels") == 0) {
 778                 return scnprintf(page, PAGE_SIZE, "%u\n", grp->num_ch);
 779         } else if (strcmp(attr->name, "num_luns") == 0) {
 780                 return scnprintf(page, PAGE_SIZE, "%u\n", grp->num_lun);
 781         } else if (strcmp(attr->name, "num_planes") == 0) {
 782                 return scnprintf(page, PAGE_SIZE, "%u\n", grp->num_pln);
 783         } else if (strcmp(attr->name, "num_blocks") == 0) {     /* u16 */
 784                 return scnprintf(page, PAGE_SIZE, "%u\n", grp->num_chk);
 785         } else if (strcmp(attr->name, "num_pages") == 0) {
 786                 return scnprintf(page, PAGE_SIZE, "%u\n", grp->num_pg);
 787         } else if (strcmp(attr->name, "page_size") == 0) {
 788                 return scnprintf(page, PAGE_SIZE, "%u\n", grp->fpg_sz);
 789         } else if (strcmp(attr->name, "hw_sector_size") == 0) {
 790                 return scnprintf(page, PAGE_SIZE, "%u\n", grp->csecs);
 791         } else if (strcmp(attr->name, "oob_sector_size") == 0) {/* u32 */
 792                 return scnprintf(page, PAGE_SIZE, "%u\n", grp->sos);
 793         } else if (strcmp(attr->name, "read_typ") == 0) {
 794                 return scnprintf(page, PAGE_SIZE, "%u\n", grp->trdt);
 795         } else if (strcmp(attr->name, "read_max") == 0) {
 796                 return scnprintf(page, PAGE_SIZE, "%u\n", grp->trdm);
 797         } else if (strcmp(attr->name, "prog_typ") == 0) {
 798                 return scnprintf(page, PAGE_SIZE, "%u\n", grp->tprt);
 799         } else if (strcmp(attr->name, "prog_max") == 0) {
 800                 return scnprintf(page, PAGE_SIZE, "%u\n", grp->tprm);
 801         } else if (strcmp(attr->name, "erase_typ") == 0) {
 802                 return scnprintf(page, PAGE_SIZE, "%u\n", grp->tbet);
 803         } else if (strcmp(attr->name, "erase_max") == 0) {
 804                 return scnprintf(page, PAGE_SIZE, "%u\n", grp->tbem);
 805         } else if (strcmp(attr->name, "multiplane_modes") == 0) {
 806                 return scnprintf(page, PAGE_SIZE, "0x%08x\n", grp->mpos);
 807         } else if (strcmp(attr->name, "media_capabilities") == 0) {
 808                 return scnprintf(page, PAGE_SIZE, "0x%08x\n", grp->mccap);
 809         } else if (strcmp(attr->name, "max_phys_secs") == 0) {
 810                 return scnprintf(page, PAGE_SIZE, "%u\n",
 811                                 ndev->ops->max_phys_sect);
 812         } else {
 813                 return scnprintf(page,
 814                                  PAGE_SIZE,
 815                                  "Unhandled attr(%s) in `nvm_dev_attr_show`\n",
 816                                  attr->name);
 817         }
 818 }
 819
 820 #define NVM_DEV_ATTR_RO(_name)                                          \
 821         DEVICE_ATTR(_name, S_IRUGO, nvm_dev_attr_show, NULL)
 822
 823 static NVM_DEV_ATTR_RO(version);
 824 static NVM_DEV_ATTR_RO(vendor_opcode);
 825 static NVM_DEV_ATTR_RO(capabilities);
 826 static NVM_DEV_ATTR_RO(device_mode);
 827 static NVM_DEV_ATTR_RO(ppa_format);
 828 static NVM_DEV_ATTR_RO(media_manager);
 829
 830 static NVM_DEV_ATTR_RO(media_type);
 831 static NVM_DEV_ATTR_RO(flash_media_type);
 832 static NVM_DEV_ATTR_RO(num_channels);
 833 static NVM_DEV_ATTR_RO(num_luns);
 834 static NVM_DEV_ATTR_RO(num_planes);
 835 static NVM_DEV_ATTR_RO(num_blocks);
 836 static NVM_DEV_ATTR_RO(num_pages);
 837 static NVM_DEV_ATTR_RO(page_size);
 838 static NVM_DEV_ATTR_RO(hw_sector_size);
 839 static NVM_DEV_ATTR_RO(oob_sector_size);
 840 static NVM_DEV_ATTR_RO(read_typ);
 841 static NVM_DEV_ATTR_RO(read_max);
 842 static NVM_DEV_ATTR_RO(prog_typ);
 843 static NVM_DEV_ATTR_RO(prog_max);
 844 static NVM_DEV_ATTR_RO(erase_typ);
 845 static NVM_DEV_ATTR_RO(erase_max);
 846 static NVM_DEV_ATTR_RO(multiplane_modes);
 847 static NVM_DEV_ATTR_RO(media_capabilities);
 848 static NVM_DEV_ATTR_RO(max_phys_secs);
 849
 850 static struct attribute *nvm_dev_attrs[] = {
 851         &dev_attr_version.attr,
 852         &dev_attr_vendor_opcode.attr,
 853         &dev_attr_capabilities.attr,
 854         &dev_attr_device_mode.attr,
 855         &dev_attr_media_manager.attr,
 856
 857         &dev_attr_ppa_format.attr,
 858         &dev_attr_media_type.attr,
 859         &dev_attr_flash_media_type.attr,
 860         &dev_attr_num_channels.attr,
 861         &dev_attr_num_luns.attr,
 862         &dev_attr_num_planes.attr,
 863         &dev_attr_num_blocks.attr,
 864         &dev_attr_num_pages.attr,
 865         &dev_attr_page_size.attr,
 866         &dev_attr_hw_sector_size.attr,
 867         &dev_attr_oob_sector_size.attr,
 868         &dev_attr_read_typ.attr,
 869         &dev_attr_read_max.attr,
 870         &dev_attr_prog_typ.attr,
 871         &dev_attr_prog_max.attr,
 872         &dev_attr_erase_typ.attr,
 873         &dev_attr_erase_max.attr,
 874         &dev_attr_multiplane_modes.attr,
 875         &dev_attr_media_capabilities.attr,
 876         &dev_attr_max_phys_secs.attr,
 877         NULL,
 878 };
 879
 880 static const struct attribute_group nvm_dev_attr_group = {
 881         .name           = "lightnvm",
 882         .attrs          = nvm_dev_attrs,
 883 };
 884
 885 int nvme_nvm_register_sysfs(struct nvme_ns *ns)
 886 {
 887         return sysfs_create_group(&disk_to_dev(ns->disk)->kobj,
 888                                         &nvm_dev_attr_group);
 889 }
 890
 891 void nvme_nvm_unregister_sysfs(struct nvme_ns *ns)
 892 {
 893         sysfs_remove_group(&disk_to_dev(ns->disk)->kobj,
 894                                         &nvm_dev_attr_group);
 895 }