drivers/nvme/host/core.c

   1 /*
   2  * NVM Express device driver
   3  * Copyright (c) 2011-2014, Intel Corporation.
   4  *
   5  * This program is free software; you can redistribute it and/or modify it
   6  * under the terms and conditions of the GNU General Public License,
   7  * version 2, as published by the Free Software Foundation.
   8  *
   9  * This program is distributed in the hope it will be useful, but WITHOUT
  10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  12  * more details.
  13  */
  14
  15 #include <linux/blkdev.h>
  16 #include <linux/blk-mq.h>
  17 #include <linux/delay.h>
  18 #include <linux/errno.h>
  19 #include <linux/hdreg.h>
  20 #include <linux/kernel.h>
  21 #include <linux/module.h>
  22 #include <linux/list_sort.h>
  23 #include <linux/slab.h>
  24 #include <linux/types.h>
  25 #include <linux/pr.h>
  26 #include <linux/ptrace.h>
  27 #include <linux/nvme_ioctl.h>
  28 #include <linux/t10-pi.h>
  29 #include <scsi/sg.h>
  30 #include <asm/unaligned.h>
  31
  32 #include "nvme.h"
  33
  34 #define NVME_MINORS             (1U << MINORBITS)
  35
  36 static int nvme_major;
  37 module_param(nvme_major, int, 0);
  38
  39 static int nvme_char_major;
  40 module_param(nvme_char_major, int, 0);
  41
  42 static LIST_HEAD(nvme_ctrl_list);
  43 DEFINE_SPINLOCK(dev_list_lock);
  44
  45 static struct class *nvme_class;
  46
  47 static void nvme_free_ns(struct kref *kref)
  48 {
  49         struct nvme_ns *ns = container_of(kref, struct nvme_ns, kref);
  50
  51         if (ns->type == NVME_NS_LIGHTNVM)
  52                 nvme_nvm_unregister(ns->queue, ns->disk->disk_name);
  53
  54         spin_lock(&dev_list_lock);
  55         ns->disk->private_data = NULL;
  56         spin_unlock(&dev_list_lock);
  57
  58         nvme_put_ctrl(ns->ctrl);
  59         put_disk(ns->disk);
  60         kfree(ns);
  61 }
  62
  63 static void nvme_put_ns(struct nvme_ns *ns)
  64 {
  65         kref_put(&ns->kref, nvme_free_ns);
  66 }
  67
  68 static struct nvme_ns *nvme_get_ns_from_disk(struct gendisk *disk)
  69 {
  70         struct nvme_ns *ns;
  71
  72         spin_lock(&dev_list_lock);
  73         ns = disk->private_data;
  74         if (ns && !kref_get_unless_zero(&ns->kref))
  75                 ns = NULL;
  76         spin_unlock(&dev_list_lock);
  77
  78         return ns;
  79 }
  80
  81 struct request *nvme_alloc_request(struct request_queue *q,
  82                 struct nvme_command *cmd, unsigned int flags)
  83 {
  84         bool write = cmd->common.opcode & 1;
  85         struct request *req;
  86
  87         req = blk_mq_alloc_request(q, write, flags);
  88         if (IS_ERR(req))
  89                 return req;
  90
  91         req->cmd_type = REQ_TYPE_DRV_PRIV;
  92         req->cmd_flags |= REQ_FAILFAST_DRIVER;
  93         req->__data_len = 0;
  94         req->__sector = (sector_t) -1;
  95         req->bio = req->biotail = NULL;
  96
  97         req->cmd = (unsigned char *)cmd;
  98         req->cmd_len = sizeof(struct nvme_command);
  99         req->special = (void *)0;
 100
 101         return req;
 102 }
 103
 104 /*
 105  * Returns 0 on success.  If the result is negative, it's a Linux error code;
 106  * if the result is positive, it's an NVM Express status code
 107  */
 108 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
 109                 void *buffer, unsigned bufflen, u32 *result, unsigned timeout)
 110 {
 111         struct request *req;
 112         int ret;
 113
 114         req = nvme_alloc_request(q, cmd, 0);
 115         if (IS_ERR(req))
 116                 return PTR_ERR(req);
 117
 118         req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
 119
 120         if (buffer && bufflen) {
 121                 ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL);
 122                 if (ret)
 123                         goto out;
 124         }
 125
 126         blk_execute_rq(req->q, NULL, req, 0);
 127         if (result)
 128                 *result = (u32)(uintptr_t)req->special;
 129         ret = req->errors;
 130  out:
 131         blk_mq_free_request(req);
 132         return ret;
 133 }
 134
 135 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
 136                 void *buffer, unsigned bufflen)
 137 {
 138         return __nvme_submit_sync_cmd(q, cmd, buffer, bufflen, NULL, 0);
 139 }
 140
 141 int __nvme_submit_user_cmd(struct request_queue *q, struct nvme_command *cmd,
 142                 void __user *ubuffer, unsigned bufflen,
 143                 void __user *meta_buffer, unsigned meta_len, u32 meta_seed,
 144                 u32 *result, unsigned timeout)
 145 {
 146         bool write = cmd->common.opcode & 1;
 147         struct nvme_ns *ns = q->queuedata;
 148         struct gendisk *disk = ns ? ns->disk : NULL;
 149         struct request *req;
 150         struct bio *bio = NULL;
 151         void *meta = NULL;
 152         int ret;
 153
 154         req = nvme_alloc_request(q, cmd, 0);
 155         if (IS_ERR(req))
 156                 return PTR_ERR(req);
 157
 158         req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
 159
 160         if (ubuffer && bufflen) {
 161                 ret = blk_rq_map_user(q, req, NULL, ubuffer, bufflen,
 162                                 GFP_KERNEL);
 163                 if (ret)
 164                         goto out;
 165                 bio = req->bio;
 166
 167                 if (!disk)
 168                         goto submit;
 169                 bio->bi_bdev = bdget_disk(disk, 0);
 170                 if (!bio->bi_bdev) {
 171                         ret = -ENODEV;
 172                         goto out_unmap;
 173                 }
 174
 175                 if (meta_buffer) {
 176                         struct bio_integrity_payload *bip;
 177
 178                         meta = kmalloc(meta_len, GFP_KERNEL);
 179                         if (!meta) {
 180                                 ret = -ENOMEM;
 181                                 goto out_unmap;
 182                         }
 183
 184                         if (write) {
 185                                 if (copy_from_user(meta, meta_buffer,
 186                                                 meta_len)) {
 187                                         ret = -EFAULT;
 188                                         goto out_free_meta;
 189                                 }
 190                         }
 191
 192                         bip = bio_integrity_alloc(bio, GFP_KERNEL, 1);
 193                         if (IS_ERR(bip)) {
 194                                 ret = PTR_ERR(bip);
 195                                 goto out_free_meta;
 196                         }
 197
 198                         bip->bip_iter.bi_size = meta_len;
 199                         bip->bip_iter.bi_sector = meta_seed;
 200
 201                         ret = bio_integrity_add_page(bio, virt_to_page(meta),
 202                                         meta_len, offset_in_page(meta));
 203                         if (ret != meta_len) {
 204                                 ret = -ENOMEM;
 205                                 goto out_free_meta;
 206                         }
 207                 }
 208         }
 209  submit:
 210         blk_execute_rq(req->q, disk, req, 0);
 211         ret = req->errors;
 212         if (result)
 213                 *result = (u32)(uintptr_t)req->special;
 214         if (meta && !ret && !write) {
 215                 if (copy_to_user(meta_buffer, meta, meta_len))
 216                         ret = -EFAULT;
 217         }
 218  out_free_meta:
 219         kfree(meta);
 220  out_unmap:
 221         if (bio) {
 222                 if (disk && bio->bi_bdev)
 223                         bdput(bio->bi_bdev);
 224                 blk_rq_unmap_user(bio);
 225         }
 226  out:
 227         blk_mq_free_request(req);
 228         return ret;
 229 }
 230
 231 int nvme_submit_user_cmd(struct request_queue *q, struct nvme_command *cmd,
 232                 void __user *ubuffer, unsigned bufflen, u32 *result,
 233                 unsigned timeout)
 234 {
 235         return __nvme_submit_user_cmd(q, cmd, ubuffer, bufflen, NULL, 0, 0,
 236                         result, timeout);
 237 }
 238
 239 int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
 240 {
 241         struct nvme_command c = { };
 242         int error;
 243
 244         /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
 245         c.identify.opcode = nvme_admin_identify;
 246         c.identify.cns = cpu_to_le32(1);
 247
 248         *id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
 249         if (!*id)
 250                 return -ENOMEM;
 251
 252         error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
 253                         sizeof(struct nvme_id_ctrl));
 254         if (error)
 255                 kfree(*id);
 256         return error;
 257 }
 258
 259 static int nvme_identify_ns_list(struct nvme_ctrl *dev, unsigned nsid, __le32 *ns_list)
 260 {
 261         struct nvme_command c = { };
 262
 263         c.identify.opcode = nvme_admin_identify;
 264         c.identify.cns = cpu_to_le32(2);
 265         c.identify.nsid = cpu_to_le32(nsid);
 266         return nvme_submit_sync_cmd(dev->admin_q, &c, ns_list, 0x1000);
 267 }
 268
 269 int nvme_identify_ns(struct nvme_ctrl *dev, unsigned nsid,
 270                 struct nvme_id_ns **id)
 271 {
 272         struct nvme_command c = { };
 273         int error;
 274
 275         /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
 276         c.identify.opcode = nvme_admin_identify,
 277         c.identify.nsid = cpu_to_le32(nsid),
 278
 279         *id = kmalloc(sizeof(struct nvme_id_ns), GFP_KERNEL);
 280         if (!*id)
 281                 return -ENOMEM;
 282
 283         error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
 284                         sizeof(struct nvme_id_ns));
 285         if (error)
 286                 kfree(*id);
 287         return error;
 288 }
 289
 290 int nvme_get_features(struct nvme_ctrl *dev, unsigned fid, unsigned nsid,
 291                                         dma_addr_t dma_addr, u32 *result)
 292 {
 293         struct nvme_command c;
 294
 295         memset(&c, 0, sizeof(c));
 296         c.features.opcode = nvme_admin_get_features;
 297         c.features.nsid = cpu_to_le32(nsid);
 298         c.features.prp1 = cpu_to_le64(dma_addr);
 299         c.features.fid = cpu_to_le32(fid);
 300
 301         return __nvme_submit_sync_cmd(dev->admin_q, &c, NULL, 0, result, 0);
 302 }
 303
 304 int nvme_set_features(struct nvme_ctrl *dev, unsigned fid, unsigned dword11,
 305                                         dma_addr_t dma_addr, u32 *result)
 306 {
 307         struct nvme_command c;
 308
 309         memset(&c, 0, sizeof(c));
 310         c.features.opcode = nvme_admin_set_features;
 311         c.features.prp1 = cpu_to_le64(dma_addr);
 312         c.features.fid = cpu_to_le32(fid);
 313         c.features.dword11 = cpu_to_le32(dword11);
 314
 315         return __nvme_submit_sync_cmd(dev->admin_q, &c, NULL, 0, result, 0);
 316 }
 317
 318 int nvme_get_log_page(struct nvme_ctrl *dev, struct nvme_smart_log **log)
 319 {
 320         struct nvme_command c = { };
 321         int error;
 322
 323         c.common.opcode = nvme_admin_get_log_page,
 324         c.common.nsid = cpu_to_le32(0xFFFFFFFF),
 325         c.common.cdw10[0] = cpu_to_le32(
 326                         (((sizeof(struct nvme_smart_log) / 4) - 1) << 16) |
 327                          NVME_LOG_SMART),
 328
 329         *log = kmalloc(sizeof(struct nvme_smart_log), GFP_KERNEL);
 330         if (!*log)
 331                 return -ENOMEM;
 332
 333         error = nvme_submit_sync_cmd(dev->admin_q, &c, *log,
 334                         sizeof(struct nvme_smart_log));
 335         if (error)
 336                 kfree(*log);
 337         return error;
 338 }
 339
 340 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count)
 341 {
 342         u32 q_count = (*count - 1) | ((*count - 1) << 16);
 343         u32 result;
 344         int status, nr_io_queues;
 345
 346         status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, 0,
 347                         &result);
 348         if (status)
 349                 return status;
 350
 351         nr_io_queues = min(result & 0xffff, result >> 16) + 1;
 352         *count = min(*count, nr_io_queues);
 353         return 0;
 354 }
 355
 356 static int nvme_submit_io(struct nvme_ns *ns, struct nvme_user_io __user *uio)
 357 {
 358         struct nvme_user_io io;
 359         struct nvme_command c;
 360         unsigned length, meta_len;
 361         void __user *metadata;
 362
 363         if (copy_from_user(&io, uio, sizeof(io)))
 364                 return -EFAULT;
 365
 366         switch (io.opcode) {
 367         case nvme_cmd_write:
 368         case nvme_cmd_read:
 369         case nvme_cmd_compare:
 370                 break;
 371         default:
 372                 return -EINVAL;
 373         }
 374
 375         length = (io.nblocks + 1) << ns->lba_shift;
 376         meta_len = (io.nblocks + 1) * ns->ms;
 377         metadata = (void __user *)(uintptr_t)io.metadata;
 378
 379         if (ns->ext) {
 380                 length += meta_len;
 381                 meta_len = 0;
 382         } else if (meta_len) {
 383                 if ((io.metadata & 3) || !io.metadata)
 384                         return -EINVAL;
 385         }
 386
 387         memset(&c, 0, sizeof(c));
 388         c.rw.opcode = io.opcode;
 389         c.rw.flags = io.flags;
 390         c.rw.nsid = cpu_to_le32(ns->ns_id);
 391         c.rw.slba = cpu_to_le64(io.slba);
 392         c.rw.length = cpu_to_le16(io.nblocks);
 393         c.rw.control = cpu_to_le16(io.control);
 394         c.rw.dsmgmt = cpu_to_le32(io.dsmgmt);
 395         c.rw.reftag = cpu_to_le32(io.reftag);
 396         c.rw.apptag = cpu_to_le16(io.apptag);
 397         c.rw.appmask = cpu_to_le16(io.appmask);
 398
 399         return __nvme_submit_user_cmd(ns->queue, &c,
 400                         (void __user *)(uintptr_t)io.addr, length,
 401                         metadata, meta_len, io.slba, NULL, 0);
 402 }
 403
 404 static int nvme_user_cmd(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
 405                         struct nvme_passthru_cmd __user *ucmd)
 406 {
 407         struct nvme_passthru_cmd cmd;
 408         struct nvme_command c;
 409         unsigned timeout = 0;
 410         int status;
 411
 412         if (!capable(CAP_SYS_ADMIN))
 413                 return -EACCES;
 414         if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
 415                 return -EFAULT;
 416
 417         memset(&c, 0, sizeof(c));
 418         c.common.opcode = cmd.opcode;
 419         c.common.flags = cmd.flags;
 420         c.common.nsid = cpu_to_le32(cmd.nsid);
 421         c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
 422         c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
 423         c.common.cdw10[0] = cpu_to_le32(cmd.cdw10);
 424         c.common.cdw10[1] = cpu_to_le32(cmd.cdw11);
 425         c.common.cdw10[2] = cpu_to_le32(cmd.cdw12);
 426         c.common.cdw10[3] = cpu_to_le32(cmd.cdw13);
 427         c.common.cdw10[4] = cpu_to_le32(cmd.cdw14);
 428         c.common.cdw10[5] = cpu_to_le32(cmd.cdw15);
 429
 430         if (cmd.timeout_ms)
 431                 timeout = msecs_to_jiffies(cmd.timeout_ms);
 432
 433         status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
 434                         (void __user *)(uintptr_t)cmd.addr, cmd.data_len,
 435                         &cmd.result, timeout);
 436         if (status >= 0) {
 437                 if (put_user(cmd.result, &ucmd->result))
 438                         return -EFAULT;
 439         }
 440
 441         return status;
 442 }
 443
 444 static int nvme_ioctl(struct block_device *bdev, fmode_t mode,
 445                 unsigned int cmd, unsigned long arg)
 446 {
 447         struct nvme_ns *ns = bdev->bd_disk->private_data;
 448
 449         switch (cmd) {
 450         case NVME_IOCTL_ID:
 451                 force_successful_syscall_return();
 452                 return ns->ns_id;
 453         case NVME_IOCTL_ADMIN_CMD:
 454                 return nvme_user_cmd(ns->ctrl, NULL, (void __user *)arg);
 455         case NVME_IOCTL_IO_CMD:
 456                 return nvme_user_cmd(ns->ctrl, ns, (void __user *)arg);
 457         case NVME_IOCTL_SUBMIT_IO:
 458                 return nvme_submit_io(ns, (void __user *)arg);
 459         case SG_GET_VERSION_NUM:
 460                 return nvme_sg_get_version_num((void __user *)arg);
 461         case SG_IO:
 462                 return nvme_sg_io(ns, (void __user *)arg);
 463         default:
 464                 return -ENOTTY;
 465         }
 466 }
 467
 468 #ifdef CONFIG_COMPAT
 469 static int nvme_compat_ioctl(struct block_device *bdev, fmode_t mode,
 470                         unsigned int cmd, unsigned long arg)
 471 {
 472         switch (cmd) {
 473         case SG_IO:
 474                 return -ENOIOCTLCMD;
 475         }
 476         return nvme_ioctl(bdev, mode, cmd, arg);
 477 }
 478 #else
 479 #define nvme_compat_ioctl       NULL
 480 #endif
 481
 482 static int nvme_open(struct block_device *bdev, fmode_t mode)
 483 {
 484         return nvme_get_ns_from_disk(bdev->bd_disk) ? 0 : -ENXIO;
 485 }
 486
 487 static void nvme_release(struct gendisk *disk, fmode_t mode)
 488 {
 489         nvme_put_ns(disk->private_data);
 490 }
 491
 492 static int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
 493 {
 494         /* some standard values */
 495         geo->heads = 1 << 6;
 496         geo->sectors = 1 << 5;
 497         geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
 498         return 0;
 499 }
 500
 501 #ifdef CONFIG_BLK_DEV_INTEGRITY
 502 static void nvme_init_integrity(struct nvme_ns *ns)
 503 {
 504         struct blk_integrity integrity;
 505
 506         switch (ns->pi_type) {
 507         case NVME_NS_DPS_PI_TYPE3:
 508                 integrity.profile = &t10_pi_type3_crc;
 509                 break;
 510         case NVME_NS_DPS_PI_TYPE1:
 511         case NVME_NS_DPS_PI_TYPE2:
 512                 integrity.profile = &t10_pi_type1_crc;
 513                 break;
 514         default:
 515                 integrity.profile = NULL;
 516                 break;
 517         }
 518         integrity.tuple_size = ns->ms;
 519         blk_integrity_register(ns->disk, &integrity);
 520         blk_queue_max_integrity_segments(ns->queue, 1);
 521 }
 522 #else
 523 static void nvme_init_integrity(struct nvme_ns *ns)
 524 {
 525 }
 526 #endif /* CONFIG_BLK_DEV_INTEGRITY */
 527
 528 static void nvme_config_discard(struct nvme_ns *ns)
 529 {
 530         u32 logical_block_size = queue_logical_block_size(ns->queue);
 531         ns->queue->limits.discard_zeroes_data = 0;
 532         ns->queue->limits.discard_alignment = logical_block_size;
 533         ns->queue->limits.discard_granularity = logical_block_size;
 534         blk_queue_max_discard_sectors(ns->queue, 0xffffffff);
 535         queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, ns->queue);
 536 }
 537
 538 static int nvme_revalidate_disk(struct gendisk *disk)
 539 {
 540         struct nvme_ns *ns = disk->private_data;
 541         struct nvme_id_ns *id;
 542         u8 lbaf, pi_type;
 543         u16 old_ms;
 544         unsigned short bs;
 545
 546         if (nvme_identify_ns(ns->ctrl, ns->ns_id, &id)) {
 547                 dev_warn(ns->ctrl->dev, "%s: Identify failure nvme%dn%d\n",
 548                                 __func__, ns->ctrl->instance, ns->ns_id);
 549                 return -ENODEV;
 550         }
 551         if (id->ncap == 0) {
 552                 kfree(id);
 553                 return -ENODEV;
 554         }
 555
 556         if (nvme_nvm_ns_supported(ns, id) && ns->type != NVME_NS_LIGHTNVM) {
 557                 if (nvme_nvm_register(ns->queue, disk->disk_name)) {
 558                         dev_warn(ns->ctrl->dev,
 559                                 "%s: LightNVM init failure\n", __func__);
 560                         kfree(id);
 561                         return -ENODEV;
 562                 }
 563                 ns->type = NVME_NS_LIGHTNVM;
 564         }
 565
 566         old_ms = ns->ms;
 567         lbaf = id->flbas & NVME_NS_FLBAS_LBA_MASK;
 568         ns->lba_shift = id->lbaf[lbaf].ds;
 569         ns->ms = le16_to_cpu(id->lbaf[lbaf].ms);
 570         ns->ext = ns->ms && (id->flbas & NVME_NS_FLBAS_META_EXT);
 571
 572         /*
 573          * If identify namespace failed, use default 512 byte block size so
 574          * block layer can use before failing read/write for 0 capacity.
 575          */
 576         if (ns->lba_shift == 0)
 577                 ns->lba_shift = 9;
 578         bs = 1 << ns->lba_shift;
 579         /* XXX: PI implementation requires metadata equal t10 pi tuple size */
 580         pi_type = ns->ms == sizeof(struct t10_pi_tuple) ?
 581                                         id->dps & NVME_NS_DPS_PI_MASK : 0;
 582
 583         blk_mq_freeze_queue(disk->queue);
 584         if (blk_get_integrity(disk) && (ns->pi_type != pi_type ||
 585                                 ns->ms != old_ms ||
 586                                 bs != queue_logical_block_size(disk->queue) ||
 587                                 (ns->ms && ns->ext)))
 588                 blk_integrity_unregister(disk);
 589
 590         ns->pi_type = pi_type;
 591         blk_queue_logical_block_size(ns->queue, bs);
 592
 593         if (ns->ms && !blk_get_integrity(disk) && !ns->ext)
 594                 nvme_init_integrity(ns);
 595         if (ns->ms && !(ns->ms == 8 && ns->pi_type) && !blk_get_integrity(disk))
 596                 set_capacity(disk, 0);
 597         else
 598                 set_capacity(disk, le64_to_cpup(&id->nsze) << (ns->lba_shift - 9));
 599
 600         if (ns->ctrl->oncs & NVME_CTRL_ONCS_DSM)
 601                 nvme_config_discard(ns);
 602         blk_mq_unfreeze_queue(disk->queue);
 603
 604         kfree(id);
 605         return 0;
 606 }
 607
 608 static char nvme_pr_type(enum pr_type type)
 609 {
 610         switch (type) {
 611         case PR_WRITE_EXCLUSIVE:
 612                 return 1;
 613         case PR_EXCLUSIVE_ACCESS:
 614                 return 2;
 615         case PR_WRITE_EXCLUSIVE_REG_ONLY:
 616                 return 3;
 617         case PR_EXCLUSIVE_ACCESS_REG_ONLY:
 618                 return 4;
 619         case PR_WRITE_EXCLUSIVE_ALL_REGS:
 620                 return 5;
 621         case PR_EXCLUSIVE_ACCESS_ALL_REGS:
 622                 return 6;
 623         default:
 624                 return 0;
 625         }
 626 };
 627
 628 static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
 629                                 u64 key, u64 sa_key, u8 op)
 630 {
 631         struct nvme_ns *ns = bdev->bd_disk->private_data;
 632         struct nvme_command c;
 633         u8 data[16] = { 0, };
 634
 635         put_unaligned_le64(key, &data[0]);
 636         put_unaligned_le64(sa_key, &data[8]);
 637
 638         memset(&c, 0, sizeof(c));
 639         c.common.opcode = op;
 640         c.common.nsid = cpu_to_le32(ns->ns_id);
 641         c.common.cdw10[0] = cpu_to_le32(cdw10);
 642
 643         return nvme_submit_sync_cmd(ns->queue, &c, data, 16);
 644 }
 645
 646 static int nvme_pr_register(struct block_device *bdev, u64 old,
 647                 u64 new, unsigned flags)
 648 {
 649         u32 cdw10;
 650
 651         if (flags & ~PR_FL_IGNORE_KEY)
 652                 return -EOPNOTSUPP;
 653
 654         cdw10 = old ? 2 : 0;
 655         cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
 656         cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
 657         return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
 658 }
 659
 660 static int nvme_pr_reserve(struct block_device *bdev, u64 key,
 661                 enum pr_type type, unsigned flags)
 662 {
 663         u32 cdw10;
 664
 665         if (flags & ~PR_FL_IGNORE_KEY)
 666                 return -EOPNOTSUPP;
 667
 668         cdw10 = nvme_pr_type(type) << 8;
 669         cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
 670         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
 671 }
 672
 673 static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
 674                 enum pr_type type, bool abort)
 675 {
 676         u32 cdw10 = nvme_pr_type(type) << 8 | abort ? 2 : 1;
 677         return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
 678 }
 679
 680 static int nvme_pr_clear(struct block_device *bdev, u64 key)
 681 {
 682         u32 cdw10 = 1 | (key ? 1 << 3 : 0);
 683         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_register);
 684 }
 685
 686 static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
 687 {
 688         u32 cdw10 = nvme_pr_type(type) << 8 | key ? 1 << 3 : 0;
 689         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
 690 }
 691
 692 static const struct pr_ops nvme_pr_ops = {
 693         .pr_register    = nvme_pr_register,
 694         .pr_reserve     = nvme_pr_reserve,
 695         .pr_release     = nvme_pr_release,
 696         .pr_preempt     = nvme_pr_preempt,
 697         .pr_clear       = nvme_pr_clear,
 698 };
 699
 700 static const struct block_device_operations nvme_fops = {
 701         .owner          = THIS_MODULE,
 702         .ioctl          = nvme_ioctl,
 703         .compat_ioctl   = nvme_compat_ioctl,
 704         .open           = nvme_open,
 705         .release        = nvme_release,
 706         .getgeo         = nvme_getgeo,
 707         .revalidate_disk= nvme_revalidate_disk,
 708         .pr_ops         = &nvme_pr_ops,
 709 };
 710
 711 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u64 cap, bool enabled)
 712 {
 713         unsigned long timeout =
 714                 ((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies;
 715         u32 csts, bit = enabled ? NVME_CSTS_RDY : 0;
 716         int ret;
 717
 718         while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
 719                 if ((csts & NVME_CSTS_RDY) == bit)
 720                         break;
 721
 722                 msleep(100);
 723                 if (fatal_signal_pending(current))
 724                         return -EINTR;
 725                 if (time_after(jiffies, timeout)) {
 726                         dev_err(ctrl->dev,
 727                                 "Device not ready; aborting %s\n", enabled ?
 728                                                 "initialisation" : "reset");
 729                         return -ENODEV;
 730                 }
 731         }
 732
 733         return ret;
 734 }
 735
 736 /*
 737  * If the device has been passed off to us in an enabled state, just clear
 738  * the enabled bit.  The spec says we should set the 'shutdown notification
 739  * bits', but doing so may cause the device to complete commands to the
 740  * admin queue ... and we don't know what memory that might be pointing at!
 741  */
 742 int nvme_disable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
 743 {
 744         int ret;
 745
 746         ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
 747         ctrl->ctrl_config &= ~NVME_CC_ENABLE;
 748
 749         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
 750         if (ret)
 751                 return ret;
 752         return nvme_wait_ready(ctrl, cap, false);
 753 }
 754
 755 int nvme_enable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
 756 {
 757         /*
 758          * Default to a 4K page size, with the intention to update this
 759          * path in the future to accomodate architectures with differing
 760          * kernel and IO page sizes.
 761          */
 762         unsigned dev_page_min = NVME_CAP_MPSMIN(cap) + 12, page_shift = 12;
 763         int ret;
 764
 765         if (page_shift < dev_page_min) {
 766                 dev_err(ctrl->dev,
 767                         "Minimum device page size %u too large for host (%u)\n",
 768                         1 << dev_page_min, 1 << page_shift);
 769                 return -ENODEV;
 770         }
 771
 772         ctrl->page_size = 1 << page_shift;
 773
 774         ctrl->ctrl_config = NVME_CC_CSS_NVM;
 775         ctrl->ctrl_config |= (page_shift - 12) << NVME_CC_MPS_SHIFT;
 776         ctrl->ctrl_config |= NVME_CC_ARB_RR | NVME_CC_SHN_NONE;
 777         ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
 778         ctrl->ctrl_config |= NVME_CC_ENABLE;
 779
 780         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
 781         if (ret)
 782                 return ret;
 783         return nvme_wait_ready(ctrl, cap, true);
 784 }
 785
 786 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl)
 787 {
 788         unsigned long timeout = SHUTDOWN_TIMEOUT + jiffies;
 789         u32 csts;
 790         int ret;
 791
 792         ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
 793         ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
 794
 795         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
 796         if (ret)
 797                 return ret;
 798
 799         while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
 800                 if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT)
 801                         break;
 802
 803                 msleep(100);
 804                 if (fatal_signal_pending(current))
 805                         return -EINTR;
 806                 if (time_after(jiffies, timeout)) {
 807                         dev_err(ctrl->dev,
 808                                 "Device shutdown incomplete; abort shutdown\n");
 809                         return -ENODEV;
 810                 }
 811         }
 812
 813         return ret;
 814 }
 815
 816 /*
 817  * Initialize the cached copies of the Identify data and various controller
 818  * register in our nvme_ctrl structure.  This should be called as soon as
 819  * the admin queue is fully up and running.
 820  */
 821 int nvme_init_identify(struct nvme_ctrl *ctrl)
 822 {
 823         struct nvme_id_ctrl *id;
 824         u64 cap;
 825         int ret, page_shift;
 826
 827         ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
 828         if (ret) {
 829                 dev_err(ctrl->dev, "Reading VS failed (%d)\n", ret);
 830                 return ret;
 831         }
 832
 833         ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &cap);
 834         if (ret) {
 835                 dev_err(ctrl->dev, "Reading CAP failed (%d)\n", ret);
 836                 return ret;
 837         }
 838         page_shift = NVME_CAP_MPSMIN(cap) + 12;
 839
 840         if (ctrl->vs >= NVME_VS(1, 1))
 841                 ctrl->subsystem = NVME_CAP_NSSRC(cap);
 842
 843         ret = nvme_identify_ctrl(ctrl, &id);
 844         if (ret) {
 845                 dev_err(ctrl->dev, "Identify Controller failed (%d)\n", ret);
 846                 return -EIO;
 847         }
 848
 849         ctrl->oncs = le16_to_cpup(&id->oncs);
 850         atomic_set(&ctrl->abort_limit, id->acl + 1);
 851         ctrl->vwc = id->vwc;
 852         memcpy(ctrl->serial, id->sn, sizeof(id->sn));
 853         memcpy(ctrl->model, id->mn, sizeof(id->mn));
 854         memcpy(ctrl->firmware_rev, id->fr, sizeof(id->fr));
 855         if (id->mdts)
 856                 ctrl->max_hw_sectors = 1 << (id->mdts + page_shift - 9);
 857         else
 858                 ctrl->max_hw_sectors = UINT_MAX;
 859
 860         if ((ctrl->quirks & NVME_QUIRK_STRIPE_SIZE) && id->vs[3]) {
 861                 unsigned int max_hw_sectors;
 862
 863                 ctrl->stripe_size = 1 << (id->vs[3] + page_shift);
 864                 max_hw_sectors = ctrl->stripe_size >> (page_shift - 9);
 865                 if (ctrl->max_hw_sectors) {
 866                         ctrl->max_hw_sectors = min(max_hw_sectors,
 867                                                         ctrl->max_hw_sectors);
 868                 } else {
 869                         ctrl->max_hw_sectors = max_hw_sectors;
 870                 }
 871         }
 872
 873         kfree(id);
 874         return 0;
 875 }
 876
 877 static int nvme_dev_open(struct inode *inode, struct file *file)
 878 {
 879         struct nvme_ctrl *ctrl;
 880         int instance = iminor(inode);
 881         int ret = -ENODEV;
 882
 883         spin_lock(&dev_list_lock);
 884         list_for_each_entry(ctrl, &nvme_ctrl_list, node) {
 885                 if (ctrl->instance != instance)
 886                         continue;
 887
 888                 if (!ctrl->admin_q) {
 889                         ret = -EWOULDBLOCK;
 890                         break;
 891                 }
 892                 if (!kref_get_unless_zero(&ctrl->kref))
 893                         break;
 894                 file->private_data = ctrl;
 895                 ret = 0;
 896                 break;
 897         }
 898         spin_unlock(&dev_list_lock);
 899
 900         return ret;
 901 }
 902
 903 static int nvme_dev_release(struct inode *inode, struct file *file)
 904 {
 905         nvme_put_ctrl(file->private_data);
 906         return 0;
 907 }
 908
 909 static long nvme_dev_ioctl(struct file *file, unsigned int cmd,
 910                 unsigned long arg)
 911 {
 912         struct nvme_ctrl *ctrl = file->private_data;
 913         void __user *argp = (void __user *)arg;
 914         struct nvme_ns *ns;
 915
 916         switch (cmd) {
 917         case NVME_IOCTL_ADMIN_CMD:
 918                 return nvme_user_cmd(ctrl, NULL, argp);
 919         case NVME_IOCTL_IO_CMD:
 920                 if (list_empty(&ctrl->namespaces))
 921                         return -ENOTTY;
 922                 ns = list_first_entry(&ctrl->namespaces, struct nvme_ns, list);
 923                 return nvme_user_cmd(ctrl, ns, argp);
 924         case NVME_IOCTL_RESET:
 925                 dev_warn(ctrl->dev, "resetting controller\n");
 926                 return ctrl->ops->reset_ctrl(ctrl);
 927         case NVME_IOCTL_SUBSYS_RESET:
 928                 return nvme_reset_subsystem(ctrl);
 929         default:
 930                 return -ENOTTY;
 931         }
 932 }
 933
 934 static const struct file_operations nvme_dev_fops = {
 935         .owner          = THIS_MODULE,
 936         .open           = nvme_dev_open,
 937         .release        = nvme_dev_release,
 938         .unlocked_ioctl = nvme_dev_ioctl,
 939         .compat_ioctl   = nvme_dev_ioctl,
 940 };
 941
 942 static ssize_t nvme_sysfs_reset(struct device *dev,
 943                                 struct device_attribute *attr, const char *buf,
 944                                 size_t count)
 945 {
 946         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
 947         int ret;
 948
 949         ret = ctrl->ops->reset_ctrl(ctrl);
 950         if (ret < 0)
 951                 return ret;
 952         return count;
 953 }
 954 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
 955
 956 static int ns_cmp(void *priv, struct list_head *a, struct list_head *b)
 957 {
 958         struct nvme_ns *nsa = container_of(a, struct nvme_ns, list);
 959         struct nvme_ns *nsb = container_of(b, struct nvme_ns, list);
 960
 961         return nsa->ns_id - nsb->ns_id;
 962 }
 963
 964 static struct nvme_ns *nvme_find_ns(struct nvme_ctrl *ctrl, unsigned nsid)
 965 {
 966         struct nvme_ns *ns;
 967
 968         list_for_each_entry(ns, &ctrl->namespaces, list) {
 969                 if (ns->ns_id == nsid)
 970                         return ns;
 971                 if (ns->ns_id > nsid)
 972                         break;
 973         }
 974         return NULL;
 975 }
 976
 977 static void nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid)
 978 {
 979         struct nvme_ns *ns;
 980         struct gendisk *disk;
 981         int node = dev_to_node(ctrl->dev);
 982
 983         ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
 984         if (!ns)
 985                 return;
 986
 987         ns->queue = blk_mq_init_queue(ctrl->tagset);
 988         if (IS_ERR(ns->queue))
 989                 goto out_free_ns;
 990         queue_flag_set_unlocked(QUEUE_FLAG_NOMERGES, ns->queue);
 991         queue_flag_set_unlocked(QUEUE_FLAG_NONROT, ns->queue);
 992         ns->queue->queuedata = ns;
 993         ns->ctrl = ctrl;
 994
 995         disk = alloc_disk_node(0, node);
 996         if (!disk)
 997                 goto out_free_queue;
 998
 999         kref_init(&ns->kref);
1000         ns->ns_id = nsid;
1001         ns->disk = disk;
1002         ns->lba_shift = 9; /* set to a default value for 512 until disk is validated */
1003
1004         blk_queue_logical_block_size(ns->queue, 1 << ns->lba_shift);
1005         if (ctrl->max_hw_sectors) {
1006                 blk_queue_max_hw_sectors(ns->queue, ctrl->max_hw_sectors);
1007                 blk_queue_max_segments(ns->queue,
1008                         (ctrl->max_hw_sectors / (ctrl->page_size >> 9)) + 1);
1009         }
1010         if (ctrl->stripe_size)
1011                 blk_queue_chunk_sectors(ns->queue, ctrl->stripe_size >> 9);
1012         if (ctrl->vwc & NVME_CTRL_VWC_PRESENT)
1013                 blk_queue_flush(ns->queue, REQ_FLUSH | REQ_FUA);
1014         blk_queue_virt_boundary(ns->queue, ctrl->page_size - 1);
1015
1016         disk->major = nvme_major;
1017         disk->first_minor = 0;
1018         disk->fops = &nvme_fops;
1019         disk->private_data = ns;
1020         disk->queue = ns->queue;
1021         disk->driverfs_dev = ctrl->device;
1022         disk->flags = GENHD_FL_EXT_DEVT;
1023         sprintf(disk->disk_name, "nvme%dn%d", ctrl->instance, nsid);
1024
1025         if (nvme_revalidate_disk(ns->disk))
1026                 goto out_free_disk;
1027
1028         list_add_tail(&ns->list, &ctrl->namespaces);
1029         kref_get(&ctrl->kref);
1030         if (ns->type != NVME_NS_LIGHTNVM)
1031                 add_disk(ns->disk);
1032
1033         return;
1034  out_free_disk:
1035         kfree(disk);
1036  out_free_queue:
1037         blk_cleanup_queue(ns->queue);
1038  out_free_ns:
1039         kfree(ns);
1040 }
1041
1042 static void nvme_ns_remove(struct nvme_ns *ns)
1043 {
1044         bool kill = nvme_io_incapable(ns->ctrl) &&
1045                         !blk_queue_dying(ns->queue);
1046
1047         if (kill)
1048                 blk_set_queue_dying(ns->queue);
1049         if (ns->disk->flags & GENHD_FL_UP) {
1050                 if (blk_get_integrity(ns->disk))
1051                         blk_integrity_unregister(ns->disk);
1052                 del_gendisk(ns->disk);
1053         }
1054         if (kill || !blk_queue_dying(ns->queue)) {
1055                 blk_mq_abort_requeue_list(ns->queue);
1056                 blk_cleanup_queue(ns->queue);
1057         }
1058         list_del_init(&ns->list);
1059         nvme_put_ns(ns);
1060 }
1061
1062 static void nvme_validate_ns(struct nvme_ctrl *ctrl, unsigned nsid)
1063 {
1064         struct nvme_ns *ns;
1065
1066         ns = nvme_find_ns(ctrl, nsid);
1067         if (ns) {
1068                 if (revalidate_disk(ns->disk))
1069                         nvme_ns_remove(ns);
1070         } else
1071                 nvme_alloc_ns(ctrl, nsid);
1072 }
1073
1074 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl, unsigned nn)
1075 {
1076         struct nvme_ns *ns;
1077         __le32 *ns_list;
1078         unsigned i, j, nsid, prev = 0, num_lists = DIV_ROUND_UP(nn, 1024);
1079         int ret = 0;
1080
1081         ns_list = kzalloc(0x1000, GFP_KERNEL);
1082         if (!ns_list)
1083                 return -ENOMEM;
1084
1085         for (i = 0; i < num_lists; i++) {
1086                 ret = nvme_identify_ns_list(ctrl, prev, ns_list);
1087                 if (ret)
1088                         goto out;
1089
1090                 for (j = 0; j < min(nn, 1024U); j++) {
1091                         nsid = le32_to_cpu(ns_list[j]);
1092                         if (!nsid)
1093                                 goto out;
1094
1095                         nvme_validate_ns(ctrl, nsid);
1096
1097                         while (++prev < nsid) {
1098                                 ns = nvme_find_ns(ctrl, prev);
1099                                 if (ns)
1100                                         nvme_ns_remove(ns);
1101                         }
1102                 }
1103                 nn -= j;
1104         }
1105  out:
1106         kfree(ns_list);
1107         return ret;
1108 }
1109
1110 static void __nvme_scan_namespaces(struct nvme_ctrl *ctrl, unsigned nn)
1111 {
1112         struct nvme_ns *ns, *next;
1113         unsigned i;
1114
1115         for (i = 1; i <= nn; i++)
1116                 nvme_validate_ns(ctrl, i);
1117
1118         list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
1119                 if (ns->ns_id > nn)
1120                         nvme_ns_remove(ns);
1121         }
1122 }
1123
1124 void nvme_scan_namespaces(struct nvme_ctrl *ctrl)
1125 {
1126         struct nvme_id_ctrl *id;
1127         unsigned nn;
1128
1129         if (nvme_identify_ctrl(ctrl, &id))
1130                 return;
1131
1132         nn = le32_to_cpu(id->nn);
1133         if (ctrl->vs >= NVME_VS(1, 1) &&
1134             !(ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)) {
1135                 if (!nvme_scan_ns_list(ctrl, nn))
1136                         goto done;
1137         }
1138         __nvme_scan_namespaces(ctrl, le32_to_cpup(&id->nn));
1139  done:
1140         list_sort(NULL, &ctrl->namespaces, ns_cmp);
1141         kfree(id);
1142 }
1143
1144 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
1145 {
1146         struct nvme_ns *ns, *next;
1147
1148         list_for_each_entry_safe(ns, next, &ctrl->namespaces, list)
1149                 nvme_ns_remove(ns);
1150 }
1151
1152 static DEFINE_IDA(nvme_instance_ida);
1153
1154 static int nvme_set_instance(struct nvme_ctrl *ctrl)
1155 {
1156         int instance, error;
1157
1158         do {
1159                 if (!ida_pre_get(&nvme_instance_ida, GFP_KERNEL))
1160                         return -ENODEV;
1161
1162                 spin_lock(&dev_list_lock);
1163                 error = ida_get_new(&nvme_instance_ida, &instance);
1164                 spin_unlock(&dev_list_lock);
1165         } while (error == -EAGAIN);
1166
1167         if (error)
1168                 return -ENODEV;
1169
1170         ctrl->instance = instance;
1171         return 0;
1172 }
1173
1174 static void nvme_release_instance(struct nvme_ctrl *ctrl)
1175 {
1176         spin_lock(&dev_list_lock);
1177         ida_remove(&nvme_instance_ida, ctrl->instance);
1178         spin_unlock(&dev_list_lock);
1179 }
1180
1181 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
1182  {
1183         device_remove_file(ctrl->device, &dev_attr_reset_controller);
1184         device_destroy(nvme_class, MKDEV(nvme_char_major, ctrl->instance));
1185
1186         spin_lock(&dev_list_lock);
1187         list_del(&ctrl->node);
1188         spin_unlock(&dev_list_lock);
1189 }
1190
1191 static void nvme_free_ctrl(struct kref *kref)
1192 {
1193         struct nvme_ctrl *ctrl = container_of(kref, struct nvme_ctrl, kref);
1194
1195         put_device(ctrl->device);
1196         nvme_release_instance(ctrl);
1197
1198         ctrl->ops->free_ctrl(ctrl);
1199 }
1200
1201 void nvme_put_ctrl(struct nvme_ctrl *ctrl)
1202 {
1203         kref_put(&ctrl->kref, nvme_free_ctrl);
1204 }
1205
1206 /*
1207  * Initialize a NVMe controller structures.  This needs to be called during
1208  * earliest initialization so that we have the initialized structured around
1209  * during probing.
1210  */
1211 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
1212                 const struct nvme_ctrl_ops *ops, unsigned long quirks)
1213 {
1214         int ret;
1215
1216         INIT_LIST_HEAD(&ctrl->namespaces);
1217         kref_init(&ctrl->kref);
1218         ctrl->dev = dev;
1219         ctrl->ops = ops;
1220         ctrl->quirks = quirks;
1221
1222         ret = nvme_set_instance(ctrl);
1223         if (ret)
1224                 goto out;
1225
1226         ctrl->device = device_create(nvme_class, ctrl->dev,
1227                                 MKDEV(nvme_char_major, ctrl->instance),
1228                                 dev, "nvme%d", ctrl->instance);
1229         if (IS_ERR(ctrl->device)) {
1230                 ret = PTR_ERR(ctrl->device);
1231                 goto out_release_instance;
1232         }
1233         get_device(ctrl->device);
1234         dev_set_drvdata(ctrl->device, ctrl);
1235
1236         ret = device_create_file(ctrl->device, &dev_attr_reset_controller);
1237         if (ret)
1238                 goto out_put_device;
1239
1240         spin_lock(&dev_list_lock);
1241         list_add_tail(&ctrl->node, &nvme_ctrl_list);
1242         spin_unlock(&dev_list_lock);
1243
1244         return 0;
1245
1246 out_put_device:
1247         put_device(ctrl->device);
1248         device_destroy(nvme_class, MKDEV(nvme_char_major, ctrl->instance));
1249 out_release_instance:
1250         nvme_release_instance(ctrl);
1251 out:
1252         return ret;
1253 }
1254
1255 int __init nvme_core_init(void)
1256 {
1257         int result;
1258
1259         result = register_blkdev(nvme_major, "nvme");
1260         if (result < 0)
1261                 return result;
1262         else if (result > 0)
1263                 nvme_major = result;
1264
1265         result = __register_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme",
1266                                                         &nvme_dev_fops);
1267         if (result < 0)
1268                 goto unregister_blkdev;
1269         else if (result > 0)
1270                 nvme_char_major = result;
1271
1272         nvme_class = class_create(THIS_MODULE, "nvme");
1273         if (IS_ERR(nvme_class)) {
1274                 result = PTR_ERR(nvme_class);
1275                 goto unregister_chrdev;
1276         }
1277
1278         return 0;
1279
1280  unregister_chrdev:
1281         __unregister_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme");
1282  unregister_blkdev:
1283         unregister_blkdev(nvme_major, "nvme");
1284         return result;
1285 }
1286
1287 void nvme_core_exit(void)
1288 {
1289         unregister_blkdev(nvme_major, "nvme");
1290         class_destroy(nvme_class);
1291         __unregister_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme");
1292 }