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 <linux/pm_qos.h>
  30 #include <scsi/sg.h>
  31 #include <asm/unaligned.h>
  32
  33 #include "nvme.h"
  34 #include "fabrics.h"
  35
  36 #define NVME_MINORS             (1U << MINORBITS)
  37
  38 unsigned char admin_timeout = 60;
  39 module_param(admin_timeout, byte, 0644);
  40 MODULE_PARM_DESC(admin_timeout, "timeout in seconds for admin commands");
  41 EXPORT_SYMBOL_GPL(admin_timeout);
  42
  43 unsigned char nvme_io_timeout = 30;
  44 module_param_named(io_timeout, nvme_io_timeout, byte, 0644);
  45 MODULE_PARM_DESC(io_timeout, "timeout in seconds for I/O");
  46 EXPORT_SYMBOL_GPL(nvme_io_timeout);
  47
  48 unsigned char shutdown_timeout = 5;
  49 module_param(shutdown_timeout, byte, 0644);
  50 MODULE_PARM_DESC(shutdown_timeout, "timeout in seconds for controller shutdown");
  51
  52 static u8 nvme_max_retries = 5;
  53 module_param_named(max_retries, nvme_max_retries, byte, 0644);
  54 MODULE_PARM_DESC(max_retries, "max number of retries a command may have");
  55
  56 static int nvme_char_major;
  57 module_param(nvme_char_major, int, 0);
  58
  59 static unsigned long default_ps_max_latency_us = 25000;
  60 module_param(default_ps_max_latency_us, ulong, 0644);
  61 MODULE_PARM_DESC(default_ps_max_latency_us,
  62                  "max power saving latency for new devices; use PM QOS to change per device");
  63
  64 static LIST_HEAD(nvme_ctrl_list);
  65 static DEFINE_SPINLOCK(dev_list_lock);
  66
  67 static struct class *nvme_class;
  68
  69 int nvme_error_status(struct request *req)
  70 {
  71         switch (nvme_req(req)->status & 0x7ff) {
  72         case NVME_SC_SUCCESS:
  73                 return 0;
  74         case NVME_SC_CAP_EXCEEDED:
  75                 return -ENOSPC;
  76         default:
  77                 return -EIO;
  78         }
  79 }
  80 EXPORT_SYMBOL_GPL(nvme_error_status);
  81
  82 static inline bool nvme_req_needs_retry(struct request *req)
  83 {
  84         if (blk_noretry_request(req))
  85                 return false;
  86         if (nvme_req(req)->status & NVME_SC_DNR)
  87                 return false;
  88         if (jiffies - req->start_time >= req->timeout)
  89                 return false;
  90         if (nvme_req(req)->retries >= nvme_max_retries)
  91                 return false;
  92         return true;
  93 }
  94
  95 void nvme_complete_rq(struct request *req)
  96 {
  97         if (unlikely(nvme_req(req)->status && nvme_req_needs_retry(req))) {
  98                 nvme_req(req)->retries++;
  99                 blk_mq_requeue_request(req, !blk_mq_queue_stopped(req->q));
 100                 return;
 101         }
 102
 103         blk_mq_end_request(req, nvme_error_status(req));
 104 }
 105 EXPORT_SYMBOL_GPL(nvme_complete_rq);
 106
 107 void nvme_cancel_request(struct request *req, void *data, bool reserved)
 108 {
 109         int status;
 110
 111         if (!blk_mq_request_started(req))
 112                 return;
 113
 114         dev_dbg_ratelimited(((struct nvme_ctrl *) data)->device,
 115                                 "Cancelling I/O %d", req->tag);
 116
 117         status = NVME_SC_ABORT_REQ;
 118         if (blk_queue_dying(req->q))
 119                 status |= NVME_SC_DNR;
 120         nvme_req(req)->status = status;
 121         blk_mq_complete_request(req, 0);
 122
 123 }
 124 EXPORT_SYMBOL_GPL(nvme_cancel_request);
 125
 126 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
 127                 enum nvme_ctrl_state new_state)
 128 {
 129         enum nvme_ctrl_state old_state;
 130         bool changed = false;
 131
 132         spin_lock_irq(&ctrl->lock);
 133
 134         old_state = ctrl->state;
 135         switch (new_state) {
 136         case NVME_CTRL_LIVE:
 137                 switch (old_state) {
 138                 case NVME_CTRL_NEW:
 139                 case NVME_CTRL_RESETTING:
 140                 case NVME_CTRL_RECONNECTING:
 141                         changed = true;
 142                         /* FALLTHRU */
 143                 default:
 144                         break;
 145                 }
 146                 break;
 147         case NVME_CTRL_RESETTING:
 148                 switch (old_state) {
 149                 case NVME_CTRL_NEW:
 150                 case NVME_CTRL_LIVE:
 151                 case NVME_CTRL_RECONNECTING:
 152                         changed = true;
 153                         /* FALLTHRU */
 154                 default:
 155                         break;
 156                 }
 157                 break;
 158         case NVME_CTRL_RECONNECTING:
 159                 switch (old_state) {
 160                 case NVME_CTRL_LIVE:
 161                         changed = true;
 162                         /* FALLTHRU */
 163                 default:
 164                         break;
 165                 }
 166                 break;
 167         case NVME_CTRL_DELETING:
 168                 switch (old_state) {
 169                 case NVME_CTRL_LIVE:
 170                 case NVME_CTRL_RESETTING:
 171                 case NVME_CTRL_RECONNECTING:
 172                         changed = true;
 173                         /* FALLTHRU */
 174                 default:
 175                         break;
 176                 }
 177                 break;
 178         case NVME_CTRL_DEAD:
 179                 switch (old_state) {
 180                 case NVME_CTRL_DELETING:
 181                         changed = true;
 182                         /* FALLTHRU */
 183                 default:
 184                         break;
 185                 }
 186                 break;
 187         default:
 188                 break;
 189         }
 190
 191         if (changed)
 192                 ctrl->state = new_state;
 193
 194         spin_unlock_irq(&ctrl->lock);
 195
 196         return changed;
 197 }
 198 EXPORT_SYMBOL_GPL(nvme_change_ctrl_state);
 199
 200 static void nvme_free_ns(struct kref *kref)
 201 {
 202         struct nvme_ns *ns = container_of(kref, struct nvme_ns, kref);
 203
 204         if (ns->ndev)
 205                 nvme_nvm_unregister(ns);
 206
 207         if (ns->disk) {
 208                 spin_lock(&dev_list_lock);
 209                 ns->disk->private_data = NULL;
 210                 spin_unlock(&dev_list_lock);
 211         }
 212
 213         put_disk(ns->disk);
 214         ida_simple_remove(&ns->ctrl->ns_ida, ns->instance);
 215         nvme_put_ctrl(ns->ctrl);
 216         kfree(ns);
 217 }
 218
 219 static void nvme_put_ns(struct nvme_ns *ns)
 220 {
 221         kref_put(&ns->kref, nvme_free_ns);
 222 }
 223
 224 static struct nvme_ns *nvme_get_ns_from_disk(struct gendisk *disk)
 225 {
 226         struct nvme_ns *ns;
 227
 228         spin_lock(&dev_list_lock);
 229         ns = disk->private_data;
 230         if (ns) {
 231                 if (!kref_get_unless_zero(&ns->kref))
 232                         goto fail;
 233                 if (!try_module_get(ns->ctrl->ops->module))
 234                         goto fail_put_ns;
 235         }
 236         spin_unlock(&dev_list_lock);
 237
 238         return ns;
 239
 240 fail_put_ns:
 241         kref_put(&ns->kref, nvme_free_ns);
 242 fail:
 243         spin_unlock(&dev_list_lock);
 244         return NULL;
 245 }
 246
 247 struct request *nvme_alloc_request(struct request_queue *q,
 248                 struct nvme_command *cmd, unsigned int flags, int qid)
 249 {
 250         unsigned op = nvme_is_write(cmd) ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN;
 251         struct request *req;
 252
 253         if (qid == NVME_QID_ANY) {
 254                 req = blk_mq_alloc_request(q, op, flags);
 255         } else {
 256                 req = blk_mq_alloc_request_hctx(q, op, flags,
 257                                 qid ? qid - 1 : 0);
 258         }
 259         if (IS_ERR(req))
 260                 return req;
 261
 262         req->cmd_flags |= REQ_FAILFAST_DRIVER;
 263         nvme_req(req)->cmd = cmd;
 264
 265         return req;
 266 }
 267 EXPORT_SYMBOL_GPL(nvme_alloc_request);
 268
 269 static inline void nvme_setup_flush(struct nvme_ns *ns,
 270                 struct nvme_command *cmnd)
 271 {
 272         memset(cmnd, 0, sizeof(*cmnd));
 273         cmnd->common.opcode = nvme_cmd_flush;
 274         cmnd->common.nsid = cpu_to_le32(ns->ns_id);
 275 }
 276
 277 static inline int nvme_setup_discard(struct nvme_ns *ns, struct request *req,
 278                 struct nvme_command *cmnd)
 279 {
 280         unsigned short segments = blk_rq_nr_discard_segments(req), n = 0;
 281         struct nvme_dsm_range *range;
 282         struct bio *bio;
 283
 284         range = kmalloc_array(segments, sizeof(*range), GFP_ATOMIC);
 285         if (!range)
 286                 return BLK_MQ_RQ_QUEUE_BUSY;
 287
 288         __rq_for_each_bio(bio, req) {
 289                 u64 slba = nvme_block_nr(ns, bio->bi_iter.bi_sector);
 290                 u32 nlb = bio->bi_iter.bi_size >> ns->lba_shift;
 291
 292                 range[n].cattr = cpu_to_le32(0);
 293                 range[n].nlb = cpu_to_le32(nlb);
 294                 range[n].slba = cpu_to_le64(slba);
 295                 n++;
 296         }
 297
 298         if (WARN_ON_ONCE(n != segments)) {
 299                 kfree(range);
 300                 return BLK_MQ_RQ_QUEUE_ERROR;
 301         }
 302
 303         memset(cmnd, 0, sizeof(*cmnd));
 304         cmnd->dsm.opcode = nvme_cmd_dsm;
 305         cmnd->dsm.nsid = cpu_to_le32(ns->ns_id);
 306         cmnd->dsm.nr = cpu_to_le32(segments - 1);
 307         cmnd->dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD);
 308
 309         req->special_vec.bv_page = virt_to_page(range);
 310         req->special_vec.bv_offset = offset_in_page(range);
 311         req->special_vec.bv_len = sizeof(*range) * segments;
 312         req->rq_flags |= RQF_SPECIAL_PAYLOAD;
 313
 314         return BLK_MQ_RQ_QUEUE_OK;
 315 }
 316
 317 static inline void nvme_setup_rw(struct nvme_ns *ns, struct request *req,
 318                 struct nvme_command *cmnd)
 319 {
 320         u16 control = 0;
 321         u32 dsmgmt = 0;
 322
 323         if (req->cmd_flags & REQ_FUA)
 324                 control |= NVME_RW_FUA;
 325         if (req->cmd_flags & (REQ_FAILFAST_DEV | REQ_RAHEAD))
 326                 control |= NVME_RW_LR;
 327
 328         if (req->cmd_flags & REQ_RAHEAD)
 329                 dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH;
 330
 331         memset(cmnd, 0, sizeof(*cmnd));
 332         cmnd->rw.opcode = (rq_data_dir(req) ? nvme_cmd_write : nvme_cmd_read);
 333         cmnd->rw.nsid = cpu_to_le32(ns->ns_id);
 334         cmnd->rw.slba = cpu_to_le64(nvme_block_nr(ns, blk_rq_pos(req)));
 335         cmnd->rw.length = cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
 336
 337         if (ns->ms) {
 338                 switch (ns->pi_type) {
 339                 case NVME_NS_DPS_PI_TYPE3:
 340                         control |= NVME_RW_PRINFO_PRCHK_GUARD;
 341                         break;
 342                 case NVME_NS_DPS_PI_TYPE1:
 343                 case NVME_NS_DPS_PI_TYPE2:
 344                         control |= NVME_RW_PRINFO_PRCHK_GUARD |
 345                                         NVME_RW_PRINFO_PRCHK_REF;
 346                         cmnd->rw.reftag = cpu_to_le32(
 347                                         nvme_block_nr(ns, blk_rq_pos(req)));
 348                         break;
 349                 }
 350                 if (!blk_integrity_rq(req))
 351                         control |= NVME_RW_PRINFO_PRACT;
 352         }
 353
 354         cmnd->rw.control = cpu_to_le16(control);
 355         cmnd->rw.dsmgmt = cpu_to_le32(dsmgmt);
 356 }
 357
 358 int nvme_setup_cmd(struct nvme_ns *ns, struct request *req,
 359                 struct nvme_command *cmd)
 360 {
 361         int ret = BLK_MQ_RQ_QUEUE_OK;
 362
 363         if (!(req->rq_flags & RQF_DONTPREP)) {
 364                 nvme_req(req)->retries = 0;
 365                 nvme_req(req)->flags = 0;
 366                 req->rq_flags |= RQF_DONTPREP;
 367         }
 368
 369         switch (req_op(req)) {
 370         case REQ_OP_DRV_IN:
 371         case REQ_OP_DRV_OUT:
 372                 memcpy(cmd, nvme_req(req)->cmd, sizeof(*cmd));
 373                 break;
 374         case REQ_OP_FLUSH:
 375                 nvme_setup_flush(ns, cmd);
 376                 break;
 377         case REQ_OP_WRITE_ZEROES:
 378                 /* currently only aliased to deallocate for a few ctrls: */
 379         case REQ_OP_DISCARD:
 380                 ret = nvme_setup_discard(ns, req, cmd);
 381                 break;
 382         case REQ_OP_READ:
 383         case REQ_OP_WRITE:
 384                 nvme_setup_rw(ns, req, cmd);
 385                 break;
 386         default:
 387                 WARN_ON_ONCE(1);
 388                 return BLK_MQ_RQ_QUEUE_ERROR;
 389         }
 390
 391         cmd->common.command_id = req->tag;
 392         return ret;
 393 }
 394 EXPORT_SYMBOL_GPL(nvme_setup_cmd);
 395
 396 /*
 397  * Returns 0 on success.  If the result is negative, it's a Linux error code;
 398  * if the result is positive, it's an NVM Express status code
 399  */
 400 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
 401                 union nvme_result *result, void *buffer, unsigned bufflen,
 402                 unsigned timeout, int qid, int at_head, int flags)
 403 {
 404         struct request *req;
 405         int ret;
 406
 407         req = nvme_alloc_request(q, cmd, flags, qid);
 408         if (IS_ERR(req))
 409                 return PTR_ERR(req);
 410
 411         req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
 412
 413         if (buffer && bufflen) {
 414                 ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL);
 415                 if (ret)
 416                         goto out;
 417         }
 418
 419         blk_execute_rq(req->q, NULL, req, at_head);
 420         if (result)
 421                 *result = nvme_req(req)->result;
 422         if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
 423                 ret = -EINTR;
 424         else
 425                 ret = nvme_req(req)->status;
 426  out:
 427         blk_mq_free_request(req);
 428         return ret;
 429 }
 430 EXPORT_SYMBOL_GPL(__nvme_submit_sync_cmd);
 431
 432 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
 433                 void *buffer, unsigned bufflen)
 434 {
 435         return __nvme_submit_sync_cmd(q, cmd, NULL, buffer, bufflen, 0,
 436                         NVME_QID_ANY, 0, 0);
 437 }
 438 EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd);
 439
 440 int __nvme_submit_user_cmd(struct request_queue *q, struct nvme_command *cmd,
 441                 void __user *ubuffer, unsigned bufflen,
 442                 void __user *meta_buffer, unsigned meta_len, u32 meta_seed,
 443                 u32 *result, unsigned timeout)
 444 {
 445         bool write = nvme_is_write(cmd);
 446         struct nvme_ns *ns = q->queuedata;
 447         struct gendisk *disk = ns ? ns->disk : NULL;
 448         struct request *req;
 449         struct bio *bio = NULL;
 450         void *meta = NULL;
 451         int ret;
 452
 453         req = nvme_alloc_request(q, cmd, 0, NVME_QID_ANY);
 454         if (IS_ERR(req))
 455                 return PTR_ERR(req);
 456
 457         req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
 458
 459         if (ubuffer && bufflen) {
 460                 ret = blk_rq_map_user(q, req, NULL, ubuffer, bufflen,
 461                                 GFP_KERNEL);
 462                 if (ret)
 463                         goto out;
 464                 bio = req->bio;
 465
 466                 if (!disk)
 467                         goto submit;
 468                 bio->bi_bdev = bdget_disk(disk, 0);
 469                 if (!bio->bi_bdev) {
 470                         ret = -ENODEV;
 471                         goto out_unmap;
 472                 }
 473
 474                 if (meta_buffer && meta_len) {
 475                         struct bio_integrity_payload *bip;
 476
 477                         meta = kmalloc(meta_len, GFP_KERNEL);
 478                         if (!meta) {
 479                                 ret = -ENOMEM;
 480                                 goto out_unmap;
 481                         }
 482
 483                         if (write) {
 484                                 if (copy_from_user(meta, meta_buffer,
 485                                                 meta_len)) {
 486                                         ret = -EFAULT;
 487                                         goto out_free_meta;
 488                                 }
 489                         }
 490
 491                         bip = bio_integrity_alloc(bio, GFP_KERNEL, 1);
 492                         if (IS_ERR(bip)) {
 493                                 ret = PTR_ERR(bip);
 494                                 goto out_free_meta;
 495                         }
 496
 497                         bip->bip_iter.bi_size = meta_len;
 498                         bip->bip_iter.bi_sector = meta_seed;
 499
 500                         ret = bio_integrity_add_page(bio, virt_to_page(meta),
 501                                         meta_len, offset_in_page(meta));
 502                         if (ret != meta_len) {
 503                                 ret = -ENOMEM;
 504                                 goto out_free_meta;
 505                         }
 506                 }
 507         }
 508  submit:
 509         blk_execute_rq(req->q, disk, req, 0);
 510         if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
 511                 ret = -EINTR;
 512         else
 513                 ret = nvme_req(req)->status;
 514         if (result)
 515                 *result = le32_to_cpu(nvme_req(req)->result.u32);
 516         if (meta && !ret && !write) {
 517                 if (copy_to_user(meta_buffer, meta, meta_len))
 518                         ret = -EFAULT;
 519         }
 520  out_free_meta:
 521         kfree(meta);
 522  out_unmap:
 523         if (bio) {
 524                 if (disk && bio->bi_bdev)
 525                         bdput(bio->bi_bdev);
 526                 blk_rq_unmap_user(bio);
 527         }
 528  out:
 529         blk_mq_free_request(req);
 530         return ret;
 531 }
 532
 533 int nvme_submit_user_cmd(struct request_queue *q, struct nvme_command *cmd,
 534                 void __user *ubuffer, unsigned bufflen, u32 *result,
 535                 unsigned timeout)
 536 {
 537         return __nvme_submit_user_cmd(q, cmd, ubuffer, bufflen, NULL, 0, 0,
 538                         result, timeout);
 539 }
 540
 541 static void nvme_keep_alive_end_io(struct request *rq, int error)
 542 {
 543         struct nvme_ctrl *ctrl = rq->end_io_data;
 544
 545         blk_mq_free_request(rq);
 546
 547         if (error) {
 548                 dev_err(ctrl->device,
 549                         "failed nvme_keep_alive_end_io error=%d\n", error);
 550                 return;
 551         }
 552
 553         schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
 554 }
 555
 556 static int nvme_keep_alive(struct nvme_ctrl *ctrl)
 557 {
 558         struct nvme_command c;
 559         struct request *rq;
 560
 561         memset(&c, 0, sizeof(c));
 562         c.common.opcode = nvme_admin_keep_alive;
 563
 564         rq = nvme_alloc_request(ctrl->admin_q, &c, BLK_MQ_REQ_RESERVED,
 565                         NVME_QID_ANY);
 566         if (IS_ERR(rq))
 567                 return PTR_ERR(rq);
 568
 569         rq->timeout = ctrl->kato * HZ;
 570         rq->end_io_data = ctrl;
 571
 572         blk_execute_rq_nowait(rq->q, NULL, rq, 0, nvme_keep_alive_end_io);
 573
 574         return 0;
 575 }
 576
 577 static void nvme_keep_alive_work(struct work_struct *work)
 578 {
 579         struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
 580                         struct nvme_ctrl, ka_work);
 581
 582         if (nvme_keep_alive(ctrl)) {
 583                 /* allocation failure, reset the controller */
 584                 dev_err(ctrl->device, "keep-alive failed\n");
 585                 ctrl->ops->reset_ctrl(ctrl);
 586                 return;
 587         }
 588 }
 589
 590 void nvme_start_keep_alive(struct nvme_ctrl *ctrl)
 591 {
 592         if (unlikely(ctrl->kato == 0))
 593                 return;
 594
 595         INIT_DELAYED_WORK(&ctrl->ka_work, nvme_keep_alive_work);
 596         schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
 597 }
 598 EXPORT_SYMBOL_GPL(nvme_start_keep_alive);
 599
 600 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl)
 601 {
 602         if (unlikely(ctrl->kato == 0))
 603                 return;
 604
 605         cancel_delayed_work_sync(&ctrl->ka_work);
 606 }
 607 EXPORT_SYMBOL_GPL(nvme_stop_keep_alive);
 608
 609 int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
 610 {
 611         struct nvme_command c = { };
 612         int error;
 613
 614         /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
 615         c.identify.opcode = nvme_admin_identify;
 616         c.identify.cns = NVME_ID_CNS_CTRL;
 617
 618         *id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
 619         if (!*id)
 620                 return -ENOMEM;
 621
 622         error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
 623                         sizeof(struct nvme_id_ctrl));
 624         if (error)
 625                 kfree(*id);
 626         return error;
 627 }
 628
 629 static int nvme_identify_ns_list(struct nvme_ctrl *dev, unsigned nsid, __le32 *ns_list)
 630 {
 631         struct nvme_command c = { };
 632
 633         c.identify.opcode = nvme_admin_identify;
 634         c.identify.cns = NVME_ID_CNS_NS_ACTIVE_LIST;
 635         c.identify.nsid = cpu_to_le32(nsid);
 636         return nvme_submit_sync_cmd(dev->admin_q, &c, ns_list, 0x1000);
 637 }
 638
 639 int nvme_identify_ns(struct nvme_ctrl *dev, unsigned nsid,
 640                 struct nvme_id_ns **id)
 641 {
 642         struct nvme_command c = { };
 643         int error;
 644
 645         /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
 646         c.identify.opcode = nvme_admin_identify;
 647         c.identify.nsid = cpu_to_le32(nsid);
 648         c.identify.cns = NVME_ID_CNS_NS;
 649
 650         *id = kmalloc(sizeof(struct nvme_id_ns), GFP_KERNEL);
 651         if (!*id)
 652                 return -ENOMEM;
 653
 654         error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
 655                         sizeof(struct nvme_id_ns));
 656         if (error)
 657                 kfree(*id);
 658         return error;
 659 }
 660
 661 int nvme_get_features(struct nvme_ctrl *dev, unsigned fid, unsigned nsid,
 662                       void *buffer, size_t buflen, u32 *result)
 663 {
 664         struct nvme_command c;
 665         union nvme_result res;
 666         int ret;
 667
 668         memset(&c, 0, sizeof(c));
 669         c.features.opcode = nvme_admin_get_features;
 670         c.features.nsid = cpu_to_le32(nsid);
 671         c.features.fid = cpu_to_le32(fid);
 672
 673         ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &res, buffer, buflen, 0,
 674                         NVME_QID_ANY, 0, 0);
 675         if (ret >= 0 && result)
 676                 *result = le32_to_cpu(res.u32);
 677         return ret;
 678 }
 679
 680 int nvme_set_features(struct nvme_ctrl *dev, unsigned fid, unsigned dword11,
 681                       void *buffer, size_t buflen, u32 *result)
 682 {
 683         struct nvme_command c;
 684         union nvme_result res;
 685         int ret;
 686
 687         memset(&c, 0, sizeof(c));
 688         c.features.opcode = nvme_admin_set_features;
 689         c.features.fid = cpu_to_le32(fid);
 690         c.features.dword11 = cpu_to_le32(dword11);
 691
 692         ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &res,
 693                         buffer, buflen, 0, NVME_QID_ANY, 0, 0);
 694         if (ret >= 0 && result)
 695                 *result = le32_to_cpu(res.u32);
 696         return ret;
 697 }
 698
 699 int nvme_get_log_page(struct nvme_ctrl *dev, struct nvme_smart_log **log)
 700 {
 701         struct nvme_command c = { };
 702         int error;
 703
 704         c.common.opcode = nvme_admin_get_log_page,
 705         c.common.nsid = cpu_to_le32(0xFFFFFFFF),
 706         c.common.cdw10[0] = cpu_to_le32(
 707                         (((sizeof(struct nvme_smart_log) / 4) - 1) << 16) |
 708                          NVME_LOG_SMART),
 709
 710         *log = kmalloc(sizeof(struct nvme_smart_log), GFP_KERNEL);
 711         if (!*log)
 712                 return -ENOMEM;
 713
 714         error = nvme_submit_sync_cmd(dev->admin_q, &c, *log,
 715                         sizeof(struct nvme_smart_log));
 716         if (error)
 717                 kfree(*log);
 718         return error;
 719 }
 720
 721 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count)
 722 {
 723         u32 q_count = (*count - 1) | ((*count - 1) << 16);
 724         u32 result;
 725         int status, nr_io_queues;
 726
 727         status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, NULL, 0,
 728                         &result);
 729         if (status < 0)
 730                 return status;
 731
 732         /*
 733          * Degraded controllers might return an error when setting the queue
 734          * count.  We still want to be able to bring them online and offer
 735          * access to the admin queue, as that might be only way to fix them up.
 736          */
 737         if (status > 0) {
 738                 dev_err(ctrl->dev, "Could not set queue count (%d)\n", status);
 739                 *count = 0;
 740         } else {
 741                 nr_io_queues = min(result & 0xffff, result >> 16) + 1;
 742                 *count = min(*count, nr_io_queues);
 743         }
 744
 745         return 0;
 746 }
 747 EXPORT_SYMBOL_GPL(nvme_set_queue_count);
 748
 749 static int nvme_submit_io(struct nvme_ns *ns, struct nvme_user_io __user *uio)
 750 {
 751         struct nvme_user_io io;
 752         struct nvme_command c;
 753         unsigned length, meta_len;
 754         void __user *metadata;
 755
 756         if (copy_from_user(&io, uio, sizeof(io)))
 757                 return -EFAULT;
 758         if (io.flags)
 759                 return -EINVAL;
 760
 761         switch (io.opcode) {
 762         case nvme_cmd_write:
 763         case nvme_cmd_read:
 764         case nvme_cmd_compare:
 765                 break;
 766         default:
 767                 return -EINVAL;
 768         }
 769
 770         length = (io.nblocks + 1) << ns->lba_shift;
 771         meta_len = (io.nblocks + 1) * ns->ms;
 772         metadata = (void __user *)(uintptr_t)io.metadata;
 773
 774         if (ns->ext) {
 775                 length += meta_len;
 776                 meta_len = 0;
 777         } else if (meta_len) {
 778                 if ((io.metadata & 3) || !io.metadata)
 779                         return -EINVAL;
 780         }
 781
 782         memset(&c, 0, sizeof(c));
 783         c.rw.opcode = io.opcode;
 784         c.rw.flags = io.flags;
 785         c.rw.nsid = cpu_to_le32(ns->ns_id);
 786         c.rw.slba = cpu_to_le64(io.slba);
 787         c.rw.length = cpu_to_le16(io.nblocks);
 788         c.rw.control = cpu_to_le16(io.control);
 789         c.rw.dsmgmt = cpu_to_le32(io.dsmgmt);
 790         c.rw.reftag = cpu_to_le32(io.reftag);
 791         c.rw.apptag = cpu_to_le16(io.apptag);
 792         c.rw.appmask = cpu_to_le16(io.appmask);
 793
 794         return __nvme_submit_user_cmd(ns->queue, &c,
 795                         (void __user *)(uintptr_t)io.addr, length,
 796                         metadata, meta_len, io.slba, NULL, 0);
 797 }
 798
 799 static int nvme_user_cmd(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
 800                         struct nvme_passthru_cmd __user *ucmd)
 801 {
 802         struct nvme_passthru_cmd cmd;
 803         struct nvme_command c;
 804         unsigned timeout = 0;
 805         int status;
 806
 807         if (!capable(CAP_SYS_ADMIN))
 808                 return -EACCES;
 809         if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
 810                 return -EFAULT;
 811         if (cmd.flags)
 812                 return -EINVAL;
 813
 814         memset(&c, 0, sizeof(c));
 815         c.common.opcode = cmd.opcode;
 816         c.common.flags = cmd.flags;
 817         c.common.nsid = cpu_to_le32(cmd.nsid);
 818         c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
 819         c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
 820         c.common.cdw10[0] = cpu_to_le32(cmd.cdw10);
 821         c.common.cdw10[1] = cpu_to_le32(cmd.cdw11);
 822         c.common.cdw10[2] = cpu_to_le32(cmd.cdw12);
 823         c.common.cdw10[3] = cpu_to_le32(cmd.cdw13);
 824         c.common.cdw10[4] = cpu_to_le32(cmd.cdw14);
 825         c.common.cdw10[5] = cpu_to_le32(cmd.cdw15);
 826
 827         if (cmd.timeout_ms)
 828                 timeout = msecs_to_jiffies(cmd.timeout_ms);
 829
 830         status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
 831                         (void __user *)(uintptr_t)cmd.addr, cmd.data_len,
 832                         &cmd.result, timeout);
 833         if (status >= 0) {
 834                 if (put_user(cmd.result, &ucmd->result))
 835                         return -EFAULT;
 836         }
 837
 838         return status;
 839 }
 840
 841 static int nvme_ioctl(struct block_device *bdev, fmode_t mode,
 842                 unsigned int cmd, unsigned long arg)
 843 {
 844         struct nvme_ns *ns = bdev->bd_disk->private_data;
 845
 846         switch (cmd) {
 847         case NVME_IOCTL_ID:
 848                 force_successful_syscall_return();
 849                 return ns->ns_id;
 850         case NVME_IOCTL_ADMIN_CMD:
 851                 return nvme_user_cmd(ns->ctrl, NULL, (void __user *)arg);
 852         case NVME_IOCTL_IO_CMD:
 853                 return nvme_user_cmd(ns->ctrl, ns, (void __user *)arg);
 854         case NVME_IOCTL_SUBMIT_IO:
 855                 return nvme_submit_io(ns, (void __user *)arg);
 856 #ifdef CONFIG_BLK_DEV_NVME_SCSI
 857         case SG_GET_VERSION_NUM:
 858                 return nvme_sg_get_version_num((void __user *)arg);
 859         case SG_IO:
 860                 return nvme_sg_io(ns, (void __user *)arg);
 861 #endif
 862         default:
 863 #ifdef CONFIG_NVM
 864                 if (ns->ndev)
 865                         return nvme_nvm_ioctl(ns, cmd, arg);
 866 #endif
 867                 if (is_sed_ioctl(cmd))
 868                         return sed_ioctl(ns->ctrl->opal_dev, cmd,
 869                                          (void __user *) arg);
 870                 return -ENOTTY;
 871         }
 872 }
 873
 874 #ifdef CONFIG_COMPAT
 875 static int nvme_compat_ioctl(struct block_device *bdev, fmode_t mode,
 876                         unsigned int cmd, unsigned long arg)
 877 {
 878         switch (cmd) {
 879         case SG_IO:
 880                 return -ENOIOCTLCMD;
 881         }
 882         return nvme_ioctl(bdev, mode, cmd, arg);
 883 }
 884 #else
 885 #define nvme_compat_ioctl       NULL
 886 #endif
 887
 888 static int nvme_open(struct block_device *bdev, fmode_t mode)
 889 {
 890         return nvme_get_ns_from_disk(bdev->bd_disk) ? 0 : -ENXIO;
 891 }
 892
 893 static void nvme_release(struct gendisk *disk, fmode_t mode)
 894 {
 895         struct nvme_ns *ns = disk->private_data;
 896
 897         module_put(ns->ctrl->ops->module);
 898         nvme_put_ns(ns);
 899 }
 900
 901 static int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
 902 {
 903         /* some standard values */
 904         geo->heads = 1 << 6;
 905         geo->sectors = 1 << 5;
 906         geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
 907         return 0;
 908 }
 909
 910 #ifdef CONFIG_BLK_DEV_INTEGRITY
 911 static void nvme_init_integrity(struct nvme_ns *ns)
 912 {
 913         struct blk_integrity integrity;
 914
 915         memset(&integrity, 0, sizeof(integrity));
 916         switch (ns->pi_type) {
 917         case NVME_NS_DPS_PI_TYPE3:
 918                 integrity.profile = &t10_pi_type3_crc;
 919                 integrity.tag_size = sizeof(u16) + sizeof(u32);
 920                 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
 921                 break;
 922         case NVME_NS_DPS_PI_TYPE1:
 923         case NVME_NS_DPS_PI_TYPE2:
 924                 integrity.profile = &t10_pi_type1_crc;
 925                 integrity.tag_size = sizeof(u16);
 926                 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
 927                 break;
 928         default:
 929                 integrity.profile = NULL;
 930                 break;
 931         }
 932         integrity.tuple_size = ns->ms;
 933         blk_integrity_register(ns->disk, &integrity);
 934         blk_queue_max_integrity_segments(ns->queue, 1);
 935 }
 936 #else
 937 static void nvme_init_integrity(struct nvme_ns *ns)
 938 {
 939 }
 940 #endif /* CONFIG_BLK_DEV_INTEGRITY */
 941
 942 static void nvme_config_discard(struct nvme_ns *ns)
 943 {
 944         struct nvme_ctrl *ctrl = ns->ctrl;
 945         u32 logical_block_size = queue_logical_block_size(ns->queue);
 946
 947         BUILD_BUG_ON(PAGE_SIZE / sizeof(struct nvme_dsm_range) <
 948                         NVME_DSM_MAX_RANGES);
 949
 950         ns->queue->limits.discard_alignment = logical_block_size;
 951         ns->queue->limits.discard_granularity = logical_block_size;
 952         blk_queue_max_discard_sectors(ns->queue, UINT_MAX);
 953         blk_queue_max_discard_segments(ns->queue, NVME_DSM_MAX_RANGES);
 954         queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, ns->queue);
 955
 956         if (ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
 957                 blk_queue_max_write_zeroes_sectors(ns->queue, UINT_MAX);
 958 }
 959
 960 static int nvme_revalidate_ns(struct nvme_ns *ns, struct nvme_id_ns **id)
 961 {
 962         if (nvme_identify_ns(ns->ctrl, ns->ns_id, id)) {
 963                 dev_warn(ns->ctrl->dev, "%s: Identify failure\n", __func__);
 964                 return -ENODEV;
 965         }
 966
 967         if ((*id)->ncap == 0) {
 968                 kfree(*id);
 969                 return -ENODEV;
 970         }
 971
 972         if (ns->ctrl->vs >= NVME_VS(1, 1, 0))
 973                 memcpy(ns->eui, (*id)->eui64, sizeof(ns->eui));
 974         if (ns->ctrl->vs >= NVME_VS(1, 2, 0))
 975                 memcpy(ns->uuid, (*id)->nguid, sizeof(ns->uuid));
 976
 977         return 0;
 978 }
 979
 980 static void __nvme_revalidate_disk(struct gendisk *disk, struct nvme_id_ns *id)
 981 {
 982         struct nvme_ns *ns = disk->private_data;
 983         u8 lbaf, pi_type;
 984         u16 old_ms;
 985         unsigned short bs;
 986
 987         old_ms = ns->ms;
 988         lbaf = id->flbas & NVME_NS_FLBAS_LBA_MASK;
 989         ns->lba_shift = id->lbaf[lbaf].ds;
 990         ns->ms = le16_to_cpu(id->lbaf[lbaf].ms);
 991         ns->ext = ns->ms && (id->flbas & NVME_NS_FLBAS_META_EXT);
 992
 993         /*
 994          * If identify namespace failed, use default 512 byte block size so
 995          * block layer can use before failing read/write for 0 capacity.
 996          */
 997         if (ns->lba_shift == 0)
 998                 ns->lba_shift = 9;
 999         bs = 1 << ns->lba_shift;
1000         /* XXX: PI implementation requires metadata equal t10 pi tuple size */
1001         pi_type = ns->ms == sizeof(struct t10_pi_tuple) ?
1002                                         id->dps & NVME_NS_DPS_PI_MASK : 0;
1003
1004         blk_mq_freeze_queue(disk->queue);
1005         if (blk_get_integrity(disk) && (ns->pi_type != pi_type ||
1006                                 ns->ms != old_ms ||
1007                                 bs != queue_logical_block_size(disk->queue) ||
1008                                 (ns->ms && ns->ext)))
1009                 blk_integrity_unregister(disk);
1010
1011         ns->pi_type = pi_type;
1012         blk_queue_logical_block_size(ns->queue, bs);
1013
1014         if (ns->ms && !blk_get_integrity(disk) && !ns->ext)
1015                 nvme_init_integrity(ns);
1016         if (ns->ms && !(ns->ms == 8 && ns->pi_type) && !blk_get_integrity(disk))
1017                 set_capacity(disk, 0);
1018         else
1019                 set_capacity(disk, le64_to_cpup(&id->nsze) << (ns->lba_shift - 9));
1020
1021         if (ns->ctrl->oncs & NVME_CTRL_ONCS_DSM)
1022                 nvme_config_discard(ns);
1023         blk_mq_unfreeze_queue(disk->queue);
1024 }
1025
1026 static int nvme_revalidate_disk(struct gendisk *disk)
1027 {
1028         struct nvme_ns *ns = disk->private_data;
1029         struct nvme_id_ns *id = NULL;
1030         int ret;
1031
1032         if (test_bit(NVME_NS_DEAD, &ns->flags)) {
1033                 set_capacity(disk, 0);
1034                 return -ENODEV;
1035         }
1036
1037         ret = nvme_revalidate_ns(ns, &id);
1038         if (ret)
1039                 return ret;
1040
1041         __nvme_revalidate_disk(disk, id);
1042         kfree(id);
1043
1044         return 0;
1045 }
1046
1047 static char nvme_pr_type(enum pr_type type)
1048 {
1049         switch (type) {
1050         case PR_WRITE_EXCLUSIVE:
1051                 return 1;
1052         case PR_EXCLUSIVE_ACCESS:
1053                 return 2;
1054         case PR_WRITE_EXCLUSIVE_REG_ONLY:
1055                 return 3;
1056         case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1057                 return 4;
1058         case PR_WRITE_EXCLUSIVE_ALL_REGS:
1059                 return 5;
1060         case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1061                 return 6;
1062         default:
1063                 return 0;
1064         }
1065 };
1066
1067 static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
1068                                 u64 key, u64 sa_key, u8 op)
1069 {
1070         struct nvme_ns *ns = bdev->bd_disk->private_data;
1071         struct nvme_command c;
1072         u8 data[16] = { 0, };
1073
1074         put_unaligned_le64(key, &data[0]);
1075         put_unaligned_le64(sa_key, &data[8]);
1076
1077         memset(&c, 0, sizeof(c));
1078         c.common.opcode = op;
1079         c.common.nsid = cpu_to_le32(ns->ns_id);
1080         c.common.cdw10[0] = cpu_to_le32(cdw10);
1081
1082         return nvme_submit_sync_cmd(ns->queue, &c, data, 16);
1083 }
1084
1085 static int nvme_pr_register(struct block_device *bdev, u64 old,
1086                 u64 new, unsigned flags)
1087 {
1088         u32 cdw10;
1089
1090         if (flags & ~PR_FL_IGNORE_KEY)
1091                 return -EOPNOTSUPP;
1092
1093         cdw10 = old ? 2 : 0;
1094         cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
1095         cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
1096         return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
1097 }
1098
1099 static int nvme_pr_reserve(struct block_device *bdev, u64 key,
1100                 enum pr_type type, unsigned flags)
1101 {
1102         u32 cdw10;
1103
1104         if (flags & ~PR_FL_IGNORE_KEY)
1105                 return -EOPNOTSUPP;
1106
1107         cdw10 = nvme_pr_type(type) << 8;
1108         cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
1109         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
1110 }
1111
1112 static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
1113                 enum pr_type type, bool abort)
1114 {
1115         u32 cdw10 = nvme_pr_type(type) << 8 | abort ? 2 : 1;
1116         return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
1117 }
1118
1119 static int nvme_pr_clear(struct block_device *bdev, u64 key)
1120 {
1121         u32 cdw10 = 1 | (key ? 1 << 3 : 0);
1122         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_register);
1123 }
1124
1125 static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1126 {
1127         u32 cdw10 = nvme_pr_type(type) << 8 | key ? 1 << 3 : 0;
1128         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
1129 }
1130
1131 static const struct pr_ops nvme_pr_ops = {
1132         .pr_register    = nvme_pr_register,
1133         .pr_reserve     = nvme_pr_reserve,
1134         .pr_release     = nvme_pr_release,
1135         .pr_preempt     = nvme_pr_preempt,
1136         .pr_clear       = nvme_pr_clear,
1137 };
1138
1139 #ifdef CONFIG_BLK_SED_OPAL
1140 int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
1141                 bool send)
1142 {
1143         struct nvme_ctrl *ctrl = data;
1144         struct nvme_command cmd;
1145
1146         memset(&cmd, 0, sizeof(cmd));
1147         if (send)
1148                 cmd.common.opcode = nvme_admin_security_send;
1149         else
1150                 cmd.common.opcode = nvme_admin_security_recv;
1151         cmd.common.nsid = 0;
1152         cmd.common.cdw10[0] = cpu_to_le32(((u32)secp) << 24 | ((u32)spsp) << 8);
1153         cmd.common.cdw10[1] = cpu_to_le32(len);
1154
1155         return __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, buffer, len,
1156                                       ADMIN_TIMEOUT, NVME_QID_ANY, 1, 0);
1157 }
1158 EXPORT_SYMBOL_GPL(nvme_sec_submit);
1159 #endif /* CONFIG_BLK_SED_OPAL */
1160
1161 static const struct block_device_operations nvme_fops = {
1162         .owner          = THIS_MODULE,
1163         .ioctl          = nvme_ioctl,
1164         .compat_ioctl   = nvme_compat_ioctl,
1165         .open           = nvme_open,
1166         .release        = nvme_release,
1167         .getgeo         = nvme_getgeo,
1168         .revalidate_disk= nvme_revalidate_disk,
1169         .pr_ops         = &nvme_pr_ops,
1170 };
1171
1172 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u64 cap, bool enabled)
1173 {
1174         unsigned long timeout =
1175                 ((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies;
1176         u32 csts, bit = enabled ? NVME_CSTS_RDY : 0;
1177         int ret;
1178
1179         while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
1180                 if (csts == ~0)
1181                         return -ENODEV;
1182                 if ((csts & NVME_CSTS_RDY) == bit)
1183                         break;
1184
1185                 msleep(100);
1186                 if (fatal_signal_pending(current))
1187                         return -EINTR;
1188                 if (time_after(jiffies, timeout)) {
1189                         dev_err(ctrl->device,
1190                                 "Device not ready; aborting %s\n", enabled ?
1191                                                 "initialisation" : "reset");
1192                         return -ENODEV;
1193                 }
1194         }
1195
1196         return ret;
1197 }
1198
1199 /*
1200  * If the device has been passed off to us in an enabled state, just clear
1201  * the enabled bit.  The spec says we should set the 'shutdown notification
1202  * bits', but doing so may cause the device to complete commands to the
1203  * admin queue ... and we don't know what memory that might be pointing at!
1204  */
1205 int nvme_disable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
1206 {
1207         int ret;
1208
1209         ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
1210         ctrl->ctrl_config &= ~NVME_CC_ENABLE;
1211
1212         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1213         if (ret)
1214                 return ret;
1215
1216         if (ctrl->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY)
1217                 msleep(NVME_QUIRK_DELAY_AMOUNT);
1218
1219         return nvme_wait_ready(ctrl, cap, false);
1220 }
1221 EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
1222
1223 int nvme_enable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
1224 {
1225         /*
1226          * Default to a 4K page size, with the intention to update this
1227          * path in the future to accomodate architectures with differing
1228          * kernel and IO page sizes.
1229          */
1230         unsigned dev_page_min = NVME_CAP_MPSMIN(cap) + 12, page_shift = 12;
1231         int ret;
1232
1233         if (page_shift < dev_page_min) {
1234                 dev_err(ctrl->device,
1235                         "Minimum device page size %u too large for host (%u)\n",
1236                         1 << dev_page_min, 1 << page_shift);
1237                 return -ENODEV;
1238         }
1239
1240         ctrl->page_size = 1 << page_shift;
1241
1242         ctrl->ctrl_config = NVME_CC_CSS_NVM;
1243         ctrl->ctrl_config |= (page_shift - 12) << NVME_CC_MPS_SHIFT;
1244         ctrl->ctrl_config |= NVME_CC_ARB_RR | NVME_CC_SHN_NONE;
1245         ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
1246         ctrl->ctrl_config |= NVME_CC_ENABLE;
1247
1248         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1249         if (ret)
1250                 return ret;
1251         return nvme_wait_ready(ctrl, cap, true);
1252 }
1253 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
1254
1255 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl)
1256 {
1257         unsigned long timeout = SHUTDOWN_TIMEOUT + jiffies;
1258         u32 csts;
1259         int ret;
1260
1261         ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
1262         ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
1263
1264         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1265         if (ret)
1266                 return ret;
1267
1268         while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
1269                 if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT)
1270                         break;
1271
1272                 msleep(100);
1273                 if (fatal_signal_pending(current))
1274                         return -EINTR;
1275                 if (time_after(jiffies, timeout)) {
1276                         dev_err(ctrl->device,
1277                                 "Device shutdown incomplete; abort shutdown\n");
1278                         return -ENODEV;
1279                 }
1280         }
1281
1282         return ret;
1283 }
1284 EXPORT_SYMBOL_GPL(nvme_shutdown_ctrl);
1285
1286 static void nvme_set_queue_limits(struct nvme_ctrl *ctrl,
1287                 struct request_queue *q)
1288 {
1289         bool vwc = false;
1290
1291         if (ctrl->max_hw_sectors) {
1292                 u32 max_segments =
1293                         (ctrl->max_hw_sectors / (ctrl->page_size >> 9)) + 1;
1294
1295                 blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors);
1296                 blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX));
1297         }
1298         if (ctrl->quirks & NVME_QUIRK_STRIPE_SIZE)
1299                 blk_queue_chunk_sectors(q, ctrl->max_hw_sectors);
1300         blk_queue_virt_boundary(q, ctrl->page_size - 1);
1301         if (ctrl->vwc & NVME_CTRL_VWC_PRESENT)
1302                 vwc = true;
1303         blk_queue_write_cache(q, vwc, vwc);
1304 }
1305
1306 static void nvme_configure_apst(struct nvme_ctrl *ctrl)
1307 {
1308         /*
1309          * APST (Autonomous Power State Transition) lets us program a
1310          * table of power state transitions that the controller will
1311          * perform automatically.  We configure it with a simple
1312          * heuristic: we are willing to spend at most 2% of the time
1313          * transitioning between power states.  Therefore, when running
1314          * in any given state, we will enter the next lower-power
1315          * non-operational state after waiting 100 * (enlat + exlat)
1316          * microseconds, as long as that state's total latency is under
1317          * the requested maximum latency.
1318          *
1319          * We will not autonomously enter any non-operational state for
1320          * which the total latency exceeds ps_max_latency_us.  Users
1321          * can set ps_max_latency_us to zero to turn off APST.
1322          */
1323
1324         unsigned apste;
1325         struct nvme_feat_auto_pst *table;
1326         int ret;
1327
1328         /*
1329          * If APST isn't supported or if we haven't been initialized yet,
1330          * then don't do anything.
1331          */
1332         if (!ctrl->apsta)
1333                 return;
1334
1335         if (ctrl->npss > 31) {
1336                 dev_warn(ctrl->device, "NPSS is invalid; not using APST\n");
1337                 return;
1338         }
1339
1340         table = kzalloc(sizeof(*table), GFP_KERNEL);
1341         if (!table)
1342                 return;
1343
1344         if (ctrl->ps_max_latency_us == 0) {
1345                 /* Turn off APST. */
1346                 apste = 0;
1347         } else {
1348                 __le64 target = cpu_to_le64(0);
1349                 int state;
1350
1351                 /*
1352                  * Walk through all states from lowest- to highest-power.
1353                  * According to the spec, lower-numbered states use more
1354                  * power.  NPSS, despite the name, is the index of the
1355                  * lowest-power state, not the number of states.
1356                  */
1357                 for (state = (int)ctrl->npss; state >= 0; state--) {
1358                         u64 total_latency_us, transition_ms;
1359
1360                         if (target)
1361                                 table->entries[state] = target;
1362
1363                         /*
1364                          * Is this state a useful non-operational state for
1365                          * higher-power states to autonomously transition to?
1366                          */
1367                         if (!(ctrl->psd[state].flags &
1368                               NVME_PS_FLAGS_NON_OP_STATE))
1369                                 continue;
1370
1371                         total_latency_us =
1372                                 (u64)le32_to_cpu(ctrl->psd[state].entry_lat) +
1373                                 + le32_to_cpu(ctrl->psd[state].exit_lat);
1374                         if (total_latency_us > ctrl->ps_max_latency_us)
1375                                 continue;
1376
1377                         /*
1378                          * This state is good.  Use it as the APST idle
1379                          * target for higher power states.
1380                          */
1381                         transition_ms = total_latency_us + 19;
1382                         do_div(transition_ms, 20);
1383                         if (transition_ms > (1 << 24) - 1)
1384                                 transition_ms = (1 << 24) - 1;
1385
1386                         target = cpu_to_le64((state << 3) |
1387                                              (transition_ms << 8));
1388                 }
1389
1390                 apste = 1;
1391         }
1392
1393         ret = nvme_set_features(ctrl, NVME_FEAT_AUTO_PST, apste,
1394                                 table, sizeof(*table), NULL);
1395         if (ret)
1396                 dev_err(ctrl->device, "failed to set APST feature (%d)\n", ret);
1397
1398         kfree(table);
1399 }
1400
1401 static void nvme_set_latency_tolerance(struct device *dev, s32 val)
1402 {
1403         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1404         u64 latency;
1405
1406         switch (val) {
1407         case PM_QOS_LATENCY_TOLERANCE_NO_CONSTRAINT:
1408         case PM_QOS_LATENCY_ANY:
1409                 latency = U64_MAX;
1410                 break;
1411
1412         default:
1413                 latency = val;
1414         }
1415
1416         if (ctrl->ps_max_latency_us != latency) {
1417                 ctrl->ps_max_latency_us = latency;
1418                 nvme_configure_apst(ctrl);
1419         }
1420 }
1421
1422 struct nvme_core_quirk_entry {
1423         /*
1424          * NVMe model and firmware strings are padded with spaces.  For
1425          * simplicity, strings in the quirk table are padded with NULLs
1426          * instead.
1427          */
1428         u16 vid;
1429         const char *mn;
1430         const char *fr;
1431         unsigned long quirks;
1432 };
1433
1434 static const struct nvme_core_quirk_entry core_quirks[] = {
1435         /*
1436          * Seen on a Samsung "SM951 NVMe SAMSUNG 256GB": using APST causes
1437          * the controller to go out to lunch.  It dies when the watchdog
1438          * timer reads CSTS and gets 0xffffffff.
1439          */
1440         {
1441                 .vid = 0x144d,
1442                 .fr = "BXW75D0Q",
1443                 .quirks = NVME_QUIRK_NO_APST,
1444         },
1445 };
1446
1447 /* match is null-terminated but idstr is space-padded. */
1448 static bool string_matches(const char *idstr, const char *match, size_t len)
1449 {
1450         size_t matchlen;
1451
1452         if (!match)
1453                 return true;
1454
1455         matchlen = strlen(match);
1456         WARN_ON_ONCE(matchlen > len);
1457
1458         if (memcmp(idstr, match, matchlen))
1459                 return false;
1460
1461         for (; matchlen < len; matchlen++)
1462                 if (idstr[matchlen] != ' ')
1463                         return false;
1464
1465         return true;
1466 }
1467
1468 static bool quirk_matches(const struct nvme_id_ctrl *id,
1469                           const struct nvme_core_quirk_entry *q)
1470 {
1471         return q->vid == le16_to_cpu(id->vid) &&
1472                 string_matches(id->mn, q->mn, sizeof(id->mn)) &&
1473                 string_matches(id->fr, q->fr, sizeof(id->fr));
1474 }
1475
1476 /*
1477  * Initialize the cached copies of the Identify data and various controller
1478  * register in our nvme_ctrl structure.  This should be called as soon as
1479  * the admin queue is fully up and running.
1480  */
1481 int nvme_init_identify(struct nvme_ctrl *ctrl)
1482 {
1483         struct nvme_id_ctrl *id;
1484         u64 cap;
1485         int ret, page_shift;
1486         u32 max_hw_sectors;
1487         u8 prev_apsta;
1488
1489         ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
1490         if (ret) {
1491                 dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
1492                 return ret;
1493         }
1494
1495         ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &cap);
1496         if (ret) {
1497                 dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
1498                 return ret;
1499         }
1500         page_shift = NVME_CAP_MPSMIN(cap) + 12;
1501
1502         if (ctrl->vs >= NVME_VS(1, 1, 0))
1503                 ctrl->subsystem = NVME_CAP_NSSRC(cap);
1504
1505         ret = nvme_identify_ctrl(ctrl, &id);
1506         if (ret) {
1507                 dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
1508                 return -EIO;
1509         }
1510
1511         if (!ctrl->identified) {
1512                 /*
1513                  * Check for quirks.  Quirk can depend on firmware version,
1514                  * so, in principle, the set of quirks present can change
1515                  * across a reset.  As a possible future enhancement, we
1516                  * could re-scan for quirks every time we reinitialize
1517                  * the device, but we'd have to make sure that the driver
1518                  * behaves intelligently if the quirks change.
1519                  */
1520
1521                 int i;
1522
1523                 for (i = 0; i < ARRAY_SIZE(core_quirks); i++) {
1524                         if (quirk_matches(id, &core_quirks[i]))
1525                                 ctrl->quirks |= core_quirks[i].quirks;
1526                 }
1527         }
1528
1529         ctrl->oacs = le16_to_cpu(id->oacs);
1530         ctrl->vid = le16_to_cpu(id->vid);
1531         ctrl->oncs = le16_to_cpup(&id->oncs);
1532         atomic_set(&ctrl->abort_limit, id->acl + 1);
1533         ctrl->vwc = id->vwc;
1534         ctrl->cntlid = le16_to_cpup(&id->cntlid);
1535         memcpy(ctrl->serial, id->sn, sizeof(id->sn));
1536         memcpy(ctrl->model, id->mn, sizeof(id->mn));
1537         memcpy(ctrl->firmware_rev, id->fr, sizeof(id->fr));
1538         if (id->mdts)
1539                 max_hw_sectors = 1 << (id->mdts + page_shift - 9);
1540         else
1541                 max_hw_sectors = UINT_MAX;
1542         ctrl->max_hw_sectors =
1543                 min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
1544
1545         nvme_set_queue_limits(ctrl, ctrl->admin_q);
1546         ctrl->sgls = le32_to_cpu(id->sgls);
1547         ctrl->kas = le16_to_cpu(id->kas);
1548
1549         ctrl->npss = id->npss;
1550         prev_apsta = ctrl->apsta;
1551         ctrl->apsta = (ctrl->quirks & NVME_QUIRK_NO_APST) ? 0 : id->apsta;
1552         memcpy(ctrl->psd, id->psd, sizeof(ctrl->psd));
1553
1554         if (ctrl->ops->is_fabrics) {
1555                 ctrl->icdoff = le16_to_cpu(id->icdoff);
1556                 ctrl->ioccsz = le32_to_cpu(id->ioccsz);
1557                 ctrl->iorcsz = le32_to_cpu(id->iorcsz);
1558                 ctrl->maxcmd = le16_to_cpu(id->maxcmd);
1559
1560                 /*
1561                  * In fabrics we need to verify the cntlid matches the
1562                  * admin connect
1563                  */
1564                 if (ctrl->cntlid != le16_to_cpu(id->cntlid))
1565                         ret = -EINVAL;
1566
1567                 if (!ctrl->opts->discovery_nqn && !ctrl->kas) {
1568                         dev_err(ctrl->dev,
1569                                 "keep-alive support is mandatory for fabrics\n");
1570                         ret = -EINVAL;
1571                 }
1572         } else {
1573                 ctrl->cntlid = le16_to_cpu(id->cntlid);
1574         }
1575
1576         kfree(id);
1577
1578         if (ctrl->apsta && !prev_apsta)
1579                 dev_pm_qos_expose_latency_tolerance(ctrl->device);
1580         else if (!ctrl->apsta && prev_apsta)
1581                 dev_pm_qos_hide_latency_tolerance(ctrl->device);
1582
1583         nvme_configure_apst(ctrl);
1584
1585         ctrl->identified = true;
1586
1587         return ret;
1588 }
1589 EXPORT_SYMBOL_GPL(nvme_init_identify);
1590
1591 static int nvme_dev_open(struct inode *inode, struct file *file)
1592 {
1593         struct nvme_ctrl *ctrl;
1594         int instance = iminor(inode);
1595         int ret = -ENODEV;
1596
1597         spin_lock(&dev_list_lock);
1598         list_for_each_entry(ctrl, &nvme_ctrl_list, node) {
1599                 if (ctrl->instance != instance)
1600                         continue;
1601
1602                 if (!ctrl->admin_q) {
1603                         ret = -EWOULDBLOCK;
1604                         break;
1605                 }
1606                 if (!kref_get_unless_zero(&ctrl->kref))
1607                         break;
1608                 file->private_data = ctrl;
1609                 ret = 0;
1610                 break;
1611         }
1612         spin_unlock(&dev_list_lock);
1613
1614         return ret;
1615 }
1616
1617 static int nvme_dev_release(struct inode *inode, struct file *file)
1618 {
1619         nvme_put_ctrl(file->private_data);
1620         return 0;
1621 }
1622
1623 static int nvme_dev_user_cmd(struct nvme_ctrl *ctrl, void __user *argp)
1624 {
1625         struct nvme_ns *ns;
1626         int ret;
1627
1628         mutex_lock(&ctrl->namespaces_mutex);
1629         if (list_empty(&ctrl->namespaces)) {
1630                 ret = -ENOTTY;
1631                 goto out_unlock;
1632         }
1633
1634         ns = list_first_entry(&ctrl->namespaces, struct nvme_ns, list);
1635         if (ns != list_last_entry(&ctrl->namespaces, struct nvme_ns, list)) {
1636                 dev_warn(ctrl->device,
1637                         "NVME_IOCTL_IO_CMD not supported when multiple namespaces present!\n");
1638                 ret = -EINVAL;
1639                 goto out_unlock;
1640         }
1641
1642         dev_warn(ctrl->device,
1643                 "using deprecated NVME_IOCTL_IO_CMD ioctl on the char device!\n");
1644         kref_get(&ns->kref);
1645         mutex_unlock(&ctrl->namespaces_mutex);
1646
1647         ret = nvme_user_cmd(ctrl, ns, argp);
1648         nvme_put_ns(ns);
1649         return ret;
1650
1651 out_unlock:
1652         mutex_unlock(&ctrl->namespaces_mutex);
1653         return ret;
1654 }
1655
1656 static long nvme_dev_ioctl(struct file *file, unsigned int cmd,
1657                 unsigned long arg)
1658 {
1659         struct nvme_ctrl *ctrl = file->private_data;
1660         void __user *argp = (void __user *)arg;
1661
1662         switch (cmd) {
1663         case NVME_IOCTL_ADMIN_CMD:
1664                 return nvme_user_cmd(ctrl, NULL, argp);
1665         case NVME_IOCTL_IO_CMD:
1666                 return nvme_dev_user_cmd(ctrl, argp);
1667         case NVME_IOCTL_RESET:
1668                 dev_warn(ctrl->device, "resetting controller\n");
1669                 return ctrl->ops->reset_ctrl(ctrl);
1670         case NVME_IOCTL_SUBSYS_RESET:
1671                 return nvme_reset_subsystem(ctrl);
1672         case NVME_IOCTL_RESCAN:
1673                 nvme_queue_scan(ctrl);
1674                 return 0;
1675         default:
1676                 return -ENOTTY;
1677         }
1678 }
1679
1680 static const struct file_operations nvme_dev_fops = {
1681         .owner          = THIS_MODULE,
1682         .open           = nvme_dev_open,
1683         .release        = nvme_dev_release,
1684         .unlocked_ioctl = nvme_dev_ioctl,
1685         .compat_ioctl   = nvme_dev_ioctl,
1686 };
1687
1688 static ssize_t nvme_sysfs_reset(struct device *dev,
1689                                 struct device_attribute *attr, const char *buf,
1690                                 size_t count)
1691 {
1692         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1693         int ret;
1694
1695         ret = ctrl->ops->reset_ctrl(ctrl);
1696         if (ret < 0)
1697                 return ret;
1698         return count;
1699 }
1700 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
1701
1702 static ssize_t nvme_sysfs_rescan(struct device *dev,
1703                                 struct device_attribute *attr, const char *buf,
1704                                 size_t count)
1705 {
1706         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1707
1708         nvme_queue_scan(ctrl);
1709         return count;
1710 }
1711 static DEVICE_ATTR(rescan_controller, S_IWUSR, NULL, nvme_sysfs_rescan);
1712
1713 static ssize_t wwid_show(struct device *dev, struct device_attribute *attr,
1714                                                                 char *buf)
1715 {
1716         struct nvme_ns *ns = nvme_get_ns_from_dev(dev);
1717         struct nvme_ctrl *ctrl = ns->ctrl;
1718         int serial_len = sizeof(ctrl->serial);
1719         int model_len = sizeof(ctrl->model);
1720
1721         if (memchr_inv(ns->uuid, 0, sizeof(ns->uuid)))
1722                 return sprintf(buf, "eui.%16phN\n", ns->uuid);
1723
1724         if (memchr_inv(ns->eui, 0, sizeof(ns->eui)))
1725                 return sprintf(buf, "eui.%8phN\n", ns->eui);
1726
1727         while (ctrl->serial[serial_len - 1] == ' ')
1728                 serial_len--;
1729         while (ctrl->model[model_len - 1] == ' ')
1730                 model_len--;
1731
1732         return sprintf(buf, "nvme.%04x-%*phN-%*phN-%08x\n", ctrl->vid,
1733                 serial_len, ctrl->serial, model_len, ctrl->model, ns->ns_id);
1734 }
1735 static DEVICE_ATTR(wwid, S_IRUGO, wwid_show, NULL);
1736
1737 static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
1738                                                                 char *buf)
1739 {
1740         struct nvme_ns *ns = nvme_get_ns_from_dev(dev);
1741         return sprintf(buf, "%pU\n", ns->uuid);
1742 }
1743 static DEVICE_ATTR(uuid, S_IRUGO, uuid_show, NULL);
1744
1745 static ssize_t eui_show(struct device *dev, struct device_attribute *attr,
1746                                                                 char *buf)
1747 {
1748         struct nvme_ns *ns = nvme_get_ns_from_dev(dev);
1749         return sprintf(buf, "%8phd\n", ns->eui);
1750 }
1751 static DEVICE_ATTR(eui, S_IRUGO, eui_show, NULL);
1752
1753 static ssize_t nsid_show(struct device *dev, struct device_attribute *attr,
1754                                                                 char *buf)
1755 {
1756         struct nvme_ns *ns = nvme_get_ns_from_dev(dev);
1757         return sprintf(buf, "%d\n", ns->ns_id);
1758 }
1759 static DEVICE_ATTR(nsid, S_IRUGO, nsid_show, NULL);
1760
1761 static struct attribute *nvme_ns_attrs[] = {
1762         &dev_attr_wwid.attr,
1763         &dev_attr_uuid.attr,
1764         &dev_attr_eui.attr,
1765         &dev_attr_nsid.attr,
1766         NULL,
1767 };
1768
1769 static umode_t nvme_ns_attrs_are_visible(struct kobject *kobj,
1770                 struct attribute *a, int n)
1771 {
1772         struct device *dev = container_of(kobj, struct device, kobj);
1773         struct nvme_ns *ns = nvme_get_ns_from_dev(dev);
1774
1775         if (a == &dev_attr_uuid.attr) {
1776                 if (!memchr_inv(ns->uuid, 0, sizeof(ns->uuid)))
1777                         return 0;
1778         }
1779         if (a == &dev_attr_eui.attr) {
1780                 if (!memchr_inv(ns->eui, 0, sizeof(ns->eui)))
1781                         return 0;
1782         }
1783         return a->mode;
1784 }
1785
1786 static const struct attribute_group nvme_ns_attr_group = {
1787         .attrs          = nvme_ns_attrs,
1788         .is_visible     = nvme_ns_attrs_are_visible,
1789 };
1790
1791 #define nvme_show_str_function(field)                                           \
1792 static ssize_t  field##_show(struct device *dev,                                \
1793                             struct device_attribute *attr, char *buf)           \
1794 {                                                                               \
1795         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);                          \
1796         return sprintf(buf, "%.*s\n", (int)sizeof(ctrl->field), ctrl->field);   \
1797 }                                                                               \
1798 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
1799
1800 #define nvme_show_int_function(field)                                           \
1801 static ssize_t  field##_show(struct device *dev,                                \
1802                             struct device_attribute *attr, char *buf)           \
1803 {                                                                               \
1804         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);                          \
1805         return sprintf(buf, "%d\n", ctrl->field);       \
1806 }                                                                               \
1807 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
1808
1809 nvme_show_str_function(model);
1810 nvme_show_str_function(serial);
1811 nvme_show_str_function(firmware_rev);
1812 nvme_show_int_function(cntlid);
1813
1814 static ssize_t nvme_sysfs_delete(struct device *dev,
1815                                 struct device_attribute *attr, const char *buf,
1816                                 size_t count)
1817 {
1818         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1819
1820         if (device_remove_file_self(dev, attr))
1821                 ctrl->ops->delete_ctrl(ctrl);
1822         return count;
1823 }
1824 static DEVICE_ATTR(delete_controller, S_IWUSR, NULL, nvme_sysfs_delete);
1825
1826 static ssize_t nvme_sysfs_show_transport(struct device *dev,
1827                                          struct device_attribute *attr,
1828                                          char *buf)
1829 {
1830         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1831
1832         return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->ops->name);
1833 }
1834 static DEVICE_ATTR(transport, S_IRUGO, nvme_sysfs_show_transport, NULL);
1835
1836 static ssize_t nvme_sysfs_show_state(struct device *dev,
1837                                      struct device_attribute *attr,
1838                                      char *buf)
1839 {
1840         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1841         static const char *const state_name[] = {
1842                 [NVME_CTRL_NEW]         = "new",
1843                 [NVME_CTRL_LIVE]        = "live",
1844                 [NVME_CTRL_RESETTING]   = "resetting",
1845                 [NVME_CTRL_RECONNECTING]= "reconnecting",
1846                 [NVME_CTRL_DELETING]    = "deleting",
1847                 [NVME_CTRL_DEAD]        = "dead",
1848         };
1849
1850         if ((unsigned)ctrl->state < ARRAY_SIZE(state_name) &&
1851             state_name[ctrl->state])
1852                 return sprintf(buf, "%s\n", state_name[ctrl->state]);
1853
1854         return sprintf(buf, "unknown state\n");
1855 }
1856
1857 static DEVICE_ATTR(state, S_IRUGO, nvme_sysfs_show_state, NULL);
1858
1859 static ssize_t nvme_sysfs_show_subsysnqn(struct device *dev,
1860                                          struct device_attribute *attr,
1861                                          char *buf)
1862 {
1863         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1864
1865         return snprintf(buf, PAGE_SIZE, "%s\n",
1866                         ctrl->ops->get_subsysnqn(ctrl));
1867 }
1868 static DEVICE_ATTR(subsysnqn, S_IRUGO, nvme_sysfs_show_subsysnqn, NULL);
1869
1870 static ssize_t nvme_sysfs_show_address(struct device *dev,
1871                                          struct device_attribute *attr,
1872                                          char *buf)
1873 {
1874         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1875
1876         return ctrl->ops->get_address(ctrl, buf, PAGE_SIZE);
1877 }
1878 static DEVICE_ATTR(address, S_IRUGO, nvme_sysfs_show_address, NULL);
1879
1880 static struct attribute *nvme_dev_attrs[] = {
1881         &dev_attr_reset_controller.attr,
1882         &dev_attr_rescan_controller.attr,
1883         &dev_attr_model.attr,
1884         &dev_attr_serial.attr,
1885         &dev_attr_firmware_rev.attr,
1886         &dev_attr_cntlid.attr,
1887         &dev_attr_delete_controller.attr,
1888         &dev_attr_transport.attr,
1889         &dev_attr_subsysnqn.attr,
1890         &dev_attr_address.attr,
1891         &dev_attr_state.attr,
1892         NULL
1893 };
1894
1895 #define CHECK_ATTR(ctrl, a, name)               \
1896         if ((a) == &dev_attr_##name.attr &&     \
1897             !(ctrl)->ops->get_##name)           \
1898                 return 0
1899
1900 static umode_t nvme_dev_attrs_are_visible(struct kobject *kobj,
1901                 struct attribute *a, int n)
1902 {
1903         struct device *dev = container_of(kobj, struct device, kobj);
1904         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1905
1906         if (a == &dev_attr_delete_controller.attr) {
1907                 if (!ctrl->ops->delete_ctrl)
1908                         return 0;
1909         }
1910
1911         CHECK_ATTR(ctrl, a, subsysnqn);
1912         CHECK_ATTR(ctrl, a, address);
1913
1914         return a->mode;
1915 }
1916
1917 static struct attribute_group nvme_dev_attrs_group = {
1918         .attrs          = nvme_dev_attrs,
1919         .is_visible     = nvme_dev_attrs_are_visible,
1920 };
1921
1922 static const struct attribute_group *nvme_dev_attr_groups[] = {
1923         &nvme_dev_attrs_group,
1924         NULL,
1925 };
1926
1927 static int ns_cmp(void *priv, struct list_head *a, struct list_head *b)
1928 {
1929         struct nvme_ns *nsa = container_of(a, struct nvme_ns, list);
1930         struct nvme_ns *nsb = container_of(b, struct nvme_ns, list);
1931
1932         return nsa->ns_id - nsb->ns_id;
1933 }
1934
1935 static struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid)
1936 {
1937         struct nvme_ns *ns, *ret = NULL;
1938
1939         mutex_lock(&ctrl->namespaces_mutex);
1940         list_for_each_entry(ns, &ctrl->namespaces, list) {
1941                 if (ns->ns_id == nsid) {
1942                         kref_get(&ns->kref);
1943                         ret = ns;
1944                         break;
1945                 }
1946                 if (ns->ns_id > nsid)
1947                         break;
1948         }
1949         mutex_unlock(&ctrl->namespaces_mutex);
1950         return ret;
1951 }
1952
1953 static void nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid)
1954 {
1955         struct nvme_ns *ns;
1956         struct gendisk *disk;
1957         struct nvme_id_ns *id;
1958         char disk_name[DISK_NAME_LEN];
1959         int node = dev_to_node(ctrl->dev);
1960
1961         ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
1962         if (!ns)
1963                 return;
1964
1965         ns->instance = ida_simple_get(&ctrl->ns_ida, 1, 0, GFP_KERNEL);
1966         if (ns->instance < 0)
1967                 goto out_free_ns;
1968
1969         ns->queue = blk_mq_init_queue(ctrl->tagset);
1970         if (IS_ERR(ns->queue))
1971                 goto out_release_instance;
1972         queue_flag_set_unlocked(QUEUE_FLAG_NONROT, ns->queue);
1973         ns->queue->queuedata = ns;
1974         ns->ctrl = ctrl;
1975
1976         kref_init(&ns->kref);
1977         ns->ns_id = nsid;
1978         ns->lba_shift = 9; /* set to a default value for 512 until disk is validated */
1979
1980         blk_queue_logical_block_size(ns->queue, 1 << ns->lba_shift);
1981         nvme_set_queue_limits(ctrl, ns->queue);
1982
1983         sprintf(disk_name, "nvme%dn%d", ctrl->instance, ns->instance);
1984
1985         if (nvme_revalidate_ns(ns, &id))
1986                 goto out_free_queue;
1987
1988         if (nvme_nvm_ns_supported(ns, id) &&
1989                                 nvme_nvm_register(ns, disk_name, node)) {
1990                 dev_warn(ctrl->dev, "%s: LightNVM init failure\n", __func__);
1991                 goto out_free_id;
1992         }
1993
1994         disk = alloc_disk_node(0, node);
1995         if (!disk)
1996                 goto out_free_id;
1997
1998         disk->fops = &nvme_fops;
1999         disk->private_data = ns;
2000         disk->queue = ns->queue;
2001         disk->flags = GENHD_FL_EXT_DEVT;
2002         memcpy(disk->disk_name, disk_name, DISK_NAME_LEN);
2003         ns->disk = disk;
2004
2005         __nvme_revalidate_disk(disk, id);
2006
2007         mutex_lock(&ctrl->namespaces_mutex);
2008         list_add_tail(&ns->list, &ctrl->namespaces);
2009         mutex_unlock(&ctrl->namespaces_mutex);
2010
2011         kref_get(&ctrl->kref);
2012
2013         kfree(id);
2014
2015         device_add_disk(ctrl->device, ns->disk);
2016         if (sysfs_create_group(&disk_to_dev(ns->disk)->kobj,
2017                                         &nvme_ns_attr_group))
2018                 pr_warn("%s: failed to create sysfs group for identification\n",
2019                         ns->disk->disk_name);
2020         if (ns->ndev && nvme_nvm_register_sysfs(ns))
2021                 pr_warn("%s: failed to register lightnvm sysfs group for identification\n",
2022                         ns->disk->disk_name);
2023         return;
2024  out_free_id:
2025         kfree(id);
2026  out_free_queue:
2027         blk_cleanup_queue(ns->queue);
2028  out_release_instance:
2029         ida_simple_remove(&ctrl->ns_ida, ns->instance);
2030  out_free_ns:
2031         kfree(ns);
2032 }
2033
2034 static void nvme_ns_remove(struct nvme_ns *ns)
2035 {
2036         if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
2037                 return;
2038
2039         if (ns->disk && ns->disk->flags & GENHD_FL_UP) {
2040                 if (blk_get_integrity(ns->disk))
2041                         blk_integrity_unregister(ns->disk);
2042                 sysfs_remove_group(&disk_to_dev(ns->disk)->kobj,
2043                                         &nvme_ns_attr_group);
2044                 if (ns->ndev)
2045                         nvme_nvm_unregister_sysfs(ns);
2046                 del_gendisk(ns->disk);
2047                 blk_mq_abort_requeue_list(ns->queue);
2048                 blk_cleanup_queue(ns->queue);
2049         }
2050
2051         mutex_lock(&ns->ctrl->namespaces_mutex);
2052         list_del_init(&ns->list);
2053         mutex_unlock(&ns->ctrl->namespaces_mutex);
2054
2055         nvme_put_ns(ns);
2056 }
2057
2058 static void nvme_validate_ns(struct nvme_ctrl *ctrl, unsigned nsid)
2059 {
2060         struct nvme_ns *ns;
2061
2062         ns = nvme_find_get_ns(ctrl, nsid);
2063         if (ns) {
2064                 if (ns->disk && revalidate_disk(ns->disk))
2065                         nvme_ns_remove(ns);
2066                 nvme_put_ns(ns);
2067         } else
2068                 nvme_alloc_ns(ctrl, nsid);
2069 }
2070
2071 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
2072                                         unsigned nsid)
2073 {
2074         struct nvme_ns *ns, *next;
2075
2076         list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
2077                 if (ns->ns_id > nsid)
2078                         nvme_ns_remove(ns);
2079         }
2080 }
2081
2082 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl, unsigned nn)
2083 {
2084         struct nvme_ns *ns;
2085         __le32 *ns_list;
2086         unsigned i, j, nsid, prev = 0, num_lists = DIV_ROUND_UP(nn, 1024);
2087         int ret = 0;
2088
2089         ns_list = kzalloc(0x1000, GFP_KERNEL);
2090         if (!ns_list)
2091                 return -ENOMEM;
2092
2093         for (i = 0; i < num_lists; i++) {
2094                 ret = nvme_identify_ns_list(ctrl, prev, ns_list);
2095                 if (ret)
2096                         goto free;
2097
2098                 for (j = 0; j < min(nn, 1024U); j++) {
2099                         nsid = le32_to_cpu(ns_list[j]);
2100                         if (!nsid)
2101                                 goto out;
2102
2103                         nvme_validate_ns(ctrl, nsid);
2104
2105                         while (++prev < nsid) {
2106                                 ns = nvme_find_get_ns(ctrl, prev);
2107                                 if (ns) {
2108                                         nvme_ns_remove(ns);
2109                                         nvme_put_ns(ns);
2110                                 }
2111                         }
2112                 }
2113                 nn -= j;
2114         }
2115  out:
2116         nvme_remove_invalid_namespaces(ctrl, prev);
2117  free:
2118         kfree(ns_list);
2119         return ret;
2120 }
2121
2122 static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl, unsigned nn)
2123 {
2124         unsigned i;
2125
2126         for (i = 1; i <= nn; i++)
2127                 nvme_validate_ns(ctrl, i);
2128
2129         nvme_remove_invalid_namespaces(ctrl, nn);
2130 }
2131
2132 static void nvme_scan_work(struct work_struct *work)
2133 {
2134         struct nvme_ctrl *ctrl =
2135                 container_of(work, struct nvme_ctrl, scan_work);
2136         struct nvme_id_ctrl *id;
2137         unsigned nn;
2138
2139         if (ctrl->state != NVME_CTRL_LIVE)
2140                 return;
2141
2142         if (nvme_identify_ctrl(ctrl, &id))
2143                 return;
2144
2145         nn = le32_to_cpu(id->nn);
2146         if (ctrl->vs >= NVME_VS(1, 1, 0) &&
2147             !(ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)) {
2148                 if (!nvme_scan_ns_list(ctrl, nn))
2149                         goto done;
2150         }
2151         nvme_scan_ns_sequential(ctrl, nn);
2152  done:
2153         mutex_lock(&ctrl->namespaces_mutex);
2154         list_sort(NULL, &ctrl->namespaces, ns_cmp);
2155         mutex_unlock(&ctrl->namespaces_mutex);
2156         kfree(id);
2157 }
2158
2159 void nvme_queue_scan(struct nvme_ctrl *ctrl)
2160 {
2161         /*
2162          * Do not queue new scan work when a controller is reset during
2163          * removal.
2164          */
2165         if (ctrl->state == NVME_CTRL_LIVE)
2166                 schedule_work(&ctrl->scan_work);
2167 }
2168 EXPORT_SYMBOL_GPL(nvme_queue_scan);
2169
2170 /*
2171  * This function iterates the namespace list unlocked to allow recovery from
2172  * controller failure. It is up to the caller to ensure the namespace list is
2173  * not modified by scan work while this function is executing.
2174  */
2175 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
2176 {
2177         struct nvme_ns *ns, *next;
2178
2179         /*
2180          * The dead states indicates the controller was not gracefully
2181          * disconnected. In that case, we won't be able to flush any data while
2182          * removing the namespaces' disks; fail all the queues now to avoid
2183          * potentially having to clean up the failed sync later.
2184          */
2185         if (ctrl->state == NVME_CTRL_DEAD)
2186                 nvme_kill_queues(ctrl);
2187
2188         list_for_each_entry_safe(ns, next, &ctrl->namespaces, list)
2189                 nvme_ns_remove(ns);
2190 }
2191 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
2192
2193 static void nvme_async_event_work(struct work_struct *work)
2194 {
2195         struct nvme_ctrl *ctrl =
2196                 container_of(work, struct nvme_ctrl, async_event_work);
2197
2198         spin_lock_irq(&ctrl->lock);
2199         while (ctrl->event_limit > 0) {
2200                 int aer_idx = --ctrl->event_limit;
2201
2202                 spin_unlock_irq(&ctrl->lock);
2203                 ctrl->ops->submit_async_event(ctrl, aer_idx);
2204                 spin_lock_irq(&ctrl->lock);
2205         }
2206         spin_unlock_irq(&ctrl->lock);
2207 }
2208
2209 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
2210                 union nvme_result *res)
2211 {
2212         u32 result = le32_to_cpu(res->u32);
2213         bool done = true;
2214
2215         switch (le16_to_cpu(status) >> 1) {
2216         case NVME_SC_SUCCESS:
2217                 done = false;
2218                 /*FALLTHRU*/
2219         case NVME_SC_ABORT_REQ:
2220                 ++ctrl->event_limit;
2221                 schedule_work(&ctrl->async_event_work);
2222                 break;
2223         default:
2224                 break;
2225         }
2226
2227         if (done)
2228                 return;
2229
2230         switch (result & 0xff07) {
2231         case NVME_AER_NOTICE_NS_CHANGED:
2232                 dev_info(ctrl->device, "rescanning\n");
2233                 nvme_queue_scan(ctrl);
2234                 break;
2235         default:
2236                 dev_warn(ctrl->device, "async event result %08x\n", result);
2237         }
2238 }
2239 EXPORT_SYMBOL_GPL(nvme_complete_async_event);
2240
2241 void nvme_queue_async_events(struct nvme_ctrl *ctrl)
2242 {
2243         ctrl->event_limit = NVME_NR_AERS;
2244         schedule_work(&ctrl->async_event_work);
2245 }
2246 EXPORT_SYMBOL_GPL(nvme_queue_async_events);
2247
2248 static DEFINE_IDA(nvme_instance_ida);
2249
2250 static int nvme_set_instance(struct nvme_ctrl *ctrl)
2251 {
2252         int instance, error;
2253
2254         do {
2255                 if (!ida_pre_get(&nvme_instance_ida, GFP_KERNEL))
2256                         return -ENODEV;
2257
2258                 spin_lock(&dev_list_lock);
2259                 error = ida_get_new(&nvme_instance_ida, &instance);
2260                 spin_unlock(&dev_list_lock);
2261         } while (error == -EAGAIN);
2262
2263         if (error)
2264                 return -ENODEV;
2265
2266         ctrl->instance = instance;
2267         return 0;
2268 }
2269
2270 static void nvme_release_instance(struct nvme_ctrl *ctrl)
2271 {
2272         spin_lock(&dev_list_lock);
2273         ida_remove(&nvme_instance_ida, ctrl->instance);
2274         spin_unlock(&dev_list_lock);
2275 }
2276
2277 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
2278 {
2279         flush_work(&ctrl->async_event_work);
2280         flush_work(&ctrl->scan_work);
2281         nvme_remove_namespaces(ctrl);
2282
2283         device_destroy(nvme_class, MKDEV(nvme_char_major, ctrl->instance));
2284
2285         spin_lock(&dev_list_lock);
2286         list_del(&ctrl->node);
2287         spin_unlock(&dev_list_lock);
2288 }
2289 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl);
2290
2291 static void nvme_free_ctrl(struct kref *kref)
2292 {
2293         struct nvme_ctrl *ctrl = container_of(kref, struct nvme_ctrl, kref);
2294
2295         put_device(ctrl->device);
2296         nvme_release_instance(ctrl);
2297         ida_destroy(&ctrl->ns_ida);
2298
2299         ctrl->ops->free_ctrl(ctrl);
2300 }
2301
2302 void nvme_put_ctrl(struct nvme_ctrl *ctrl)
2303 {
2304         kref_put(&ctrl->kref, nvme_free_ctrl);
2305 }
2306 EXPORT_SYMBOL_GPL(nvme_put_ctrl);
2307
2308 /*
2309  * Initialize a NVMe controller structures.  This needs to be called during
2310  * earliest initialization so that we have the initialized structured around
2311  * during probing.
2312  */
2313 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
2314                 const struct nvme_ctrl_ops *ops, unsigned long quirks)
2315 {
2316         int ret;
2317
2318         ctrl->state = NVME_CTRL_NEW;
2319         spin_lock_init(&ctrl->lock);
2320         INIT_LIST_HEAD(&ctrl->namespaces);
2321         mutex_init(&ctrl->namespaces_mutex);
2322         kref_init(&ctrl->kref);
2323         ctrl->dev = dev;
2324         ctrl->ops = ops;
2325         ctrl->quirks = quirks;
2326         INIT_WORK(&ctrl->scan_work, nvme_scan_work);
2327         INIT_WORK(&ctrl->async_event_work, nvme_async_event_work);
2328
2329         ret = nvme_set_instance(ctrl);
2330         if (ret)
2331                 goto out;
2332
2333         ctrl->device = device_create_with_groups(nvme_class, ctrl->dev,
2334                                 MKDEV(nvme_char_major, ctrl->instance),
2335                                 ctrl, nvme_dev_attr_groups,
2336                                 "nvme%d", ctrl->instance);
2337         if (IS_ERR(ctrl->device)) {
2338                 ret = PTR_ERR(ctrl->device);
2339                 goto out_release_instance;
2340         }
2341         get_device(ctrl->device);
2342         ida_init(&ctrl->ns_ida);
2343
2344         spin_lock(&dev_list_lock);
2345         list_add_tail(&ctrl->node, &nvme_ctrl_list);
2346         spin_unlock(&dev_list_lock);
2347
2348         /*
2349          * Initialize latency tolerance controls.  The sysfs files won't
2350          * be visible to userspace unless the device actually supports APST.
2351          */
2352         ctrl->device->power.set_latency_tolerance = nvme_set_latency_tolerance;
2353         dev_pm_qos_update_user_latency_tolerance(ctrl->device,
2354                 min(default_ps_max_latency_us, (unsigned long)S32_MAX));
2355
2356         return 0;
2357 out_release_instance:
2358         nvme_release_instance(ctrl);
2359 out:
2360         return ret;
2361 }
2362 EXPORT_SYMBOL_GPL(nvme_init_ctrl);
2363
2364 /**
2365  * nvme_kill_queues(): Ends all namespace queues
2366  * @ctrl: the dead controller that needs to end
2367  *
2368  * Call this function when the driver determines it is unable to get the
2369  * controller in a state capable of servicing IO.
2370  */
2371 void nvme_kill_queues(struct nvme_ctrl *ctrl)
2372 {
2373         struct nvme_ns *ns;
2374
2375         mutex_lock(&ctrl->namespaces_mutex);
2376         list_for_each_entry(ns, &ctrl->namespaces, list) {
2377                 /*
2378                  * Revalidating a dead namespace sets capacity to 0. This will
2379                  * end buffered writers dirtying pages that can't be synced.
2380                  */
2381                 if (!ns->disk || test_and_set_bit(NVME_NS_DEAD, &ns->flags))
2382                         continue;
2383                 revalidate_disk(ns->disk);
2384                 blk_set_queue_dying(ns->queue);
2385                 blk_mq_abort_requeue_list(ns->queue);
2386                 blk_mq_start_stopped_hw_queues(ns->queue, true);
2387         }
2388         mutex_unlock(&ctrl->namespaces_mutex);
2389 }
2390 EXPORT_SYMBOL_GPL(nvme_kill_queues);
2391
2392 void nvme_unfreeze(struct nvme_ctrl *ctrl)
2393 {
2394         struct nvme_ns *ns;
2395
2396         mutex_lock(&ctrl->namespaces_mutex);
2397         list_for_each_entry(ns, &ctrl->namespaces, list)
2398                 blk_mq_unfreeze_queue(ns->queue);
2399         mutex_unlock(&ctrl->namespaces_mutex);
2400 }
2401 EXPORT_SYMBOL_GPL(nvme_unfreeze);
2402
2403 void nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout)
2404 {
2405         struct nvme_ns *ns;
2406
2407         mutex_lock(&ctrl->namespaces_mutex);
2408         list_for_each_entry(ns, &ctrl->namespaces, list) {
2409                 timeout = blk_mq_freeze_queue_wait_timeout(ns->queue, timeout);
2410                 if (timeout <= 0)
2411                         break;
2412         }
2413         mutex_unlock(&ctrl->namespaces_mutex);
2414 }
2415 EXPORT_SYMBOL_GPL(nvme_wait_freeze_timeout);
2416
2417 void nvme_wait_freeze(struct nvme_ctrl *ctrl)
2418 {
2419         struct nvme_ns *ns;
2420
2421         mutex_lock(&ctrl->namespaces_mutex);
2422         list_for_each_entry(ns, &ctrl->namespaces, list)
2423                 blk_mq_freeze_queue_wait(ns->queue);
2424         mutex_unlock(&ctrl->namespaces_mutex);
2425 }
2426 EXPORT_SYMBOL_GPL(nvme_wait_freeze);
2427
2428 void nvme_start_freeze(struct nvme_ctrl *ctrl)
2429 {
2430         struct nvme_ns *ns;
2431
2432         mutex_lock(&ctrl->namespaces_mutex);
2433         list_for_each_entry(ns, &ctrl->namespaces, list)
2434                 blk_freeze_queue_start(ns->queue);
2435         mutex_unlock(&ctrl->namespaces_mutex);
2436 }
2437 EXPORT_SYMBOL_GPL(nvme_start_freeze);
2438
2439 void nvme_stop_queues(struct nvme_ctrl *ctrl)
2440 {
2441         struct nvme_ns *ns;
2442
2443         mutex_lock(&ctrl->namespaces_mutex);
2444         list_for_each_entry(ns, &ctrl->namespaces, list)
2445                 blk_mq_quiesce_queue(ns->queue);
2446         mutex_unlock(&ctrl->namespaces_mutex);
2447 }
2448 EXPORT_SYMBOL_GPL(nvme_stop_queues);
2449
2450 void nvme_start_queues(struct nvme_ctrl *ctrl)
2451 {
2452         struct nvme_ns *ns;
2453
2454         mutex_lock(&ctrl->namespaces_mutex);
2455         list_for_each_entry(ns, &ctrl->namespaces, list) {
2456                 blk_mq_start_stopped_hw_queues(ns->queue, true);
2457                 blk_mq_kick_requeue_list(ns->queue);
2458         }
2459         mutex_unlock(&ctrl->namespaces_mutex);
2460 }
2461 EXPORT_SYMBOL_GPL(nvme_start_queues);
2462
2463 int __init nvme_core_init(void)
2464 {
2465         int result;
2466
2467         result = __register_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme",
2468                                                         &nvme_dev_fops);
2469         if (result < 0)
2470                 return result;
2471         else if (result > 0)
2472                 nvme_char_major = result;
2473
2474         nvme_class = class_create(THIS_MODULE, "nvme");
2475         if (IS_ERR(nvme_class)) {
2476                 result = PTR_ERR(nvme_class);
2477                 goto unregister_chrdev;
2478         }
2479
2480         return 0;
2481
2482  unregister_chrdev:
2483         __unregister_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme");
2484         return result;
2485 }
2486
2487 void nvme_core_exit(void)
2488 {
2489         class_destroy(nvme_class);
2490         __unregister_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme");
2491 }
2492
2493 MODULE_LICENSE("GPL");
2494 MODULE_VERSION("1.0");
2495 module_init(nvme_core_init);
2496 module_exit(nvme_core_exit);