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 <asm/unaligned.h>
  31
  32 #define CREATE_TRACE_POINTS
  33 #include "trace.h"
  34
  35 #include "nvme.h"
  36 #include "fabrics.h"
  37
  38 #define NVME_MINORS             (1U << MINORBITS)
  39
  40 unsigned int admin_timeout = 60;
  41 module_param(admin_timeout, uint, 0644);
  42 MODULE_PARM_DESC(admin_timeout, "timeout in seconds for admin commands");
  43 EXPORT_SYMBOL_GPL(admin_timeout);
  44
  45 unsigned int nvme_io_timeout = 30;
  46 module_param_named(io_timeout, nvme_io_timeout, uint, 0644);
  47 MODULE_PARM_DESC(io_timeout, "timeout in seconds for I/O");
  48 EXPORT_SYMBOL_GPL(nvme_io_timeout);
  49
  50 static unsigned char shutdown_timeout = 5;
  51 module_param(shutdown_timeout, byte, 0644);
  52 MODULE_PARM_DESC(shutdown_timeout, "timeout in seconds for controller shutdown");
  53
  54 static u8 nvme_max_retries = 5;
  55 module_param_named(max_retries, nvme_max_retries, byte, 0644);
  56 MODULE_PARM_DESC(max_retries, "max number of retries a command may have");
  57
  58 static unsigned long default_ps_max_latency_us = 100000;
  59 module_param(default_ps_max_latency_us, ulong, 0644);
  60 MODULE_PARM_DESC(default_ps_max_latency_us,
  61                  "max power saving latency for new devices; use PM QOS to change per device");
  62
  63 static bool force_apst;
  64 module_param(force_apst, bool, 0644);
  65 MODULE_PARM_DESC(force_apst, "allow APST for newly enumerated devices even if quirked off");
  66
  67 static bool streams;
  68 module_param(streams, bool, 0644);
  69 MODULE_PARM_DESC(streams, "turn on support for Streams write directives");
  70
  71 /*
  72  * nvme_wq - hosts nvme related works that are not reset or delete
  73  * nvme_reset_wq - hosts nvme reset works
  74  * nvme_delete_wq - hosts nvme delete works
  75  *
  76  * nvme_wq will host works such are scan, aen handling, fw activation,
  77  * keep-alive error recovery, periodic reconnects etc. nvme_reset_wq
  78  * runs reset works which also flush works hosted on nvme_wq for
  79  * serialization purposes. nvme_delete_wq host controller deletion
  80  * works which flush reset works for serialization.
  81  */
  82 struct workqueue_struct *nvme_wq;
  83 EXPORT_SYMBOL_GPL(nvme_wq);
  84
  85 struct workqueue_struct *nvme_reset_wq;
  86 EXPORT_SYMBOL_GPL(nvme_reset_wq);
  87
  88 struct workqueue_struct *nvme_delete_wq;
  89 EXPORT_SYMBOL_GPL(nvme_delete_wq);
  90
  91 static DEFINE_IDA(nvme_subsystems_ida);
  92 static LIST_HEAD(nvme_subsystems);
  93 static DEFINE_MUTEX(nvme_subsystems_lock);
  94
  95 static DEFINE_IDA(nvme_instance_ida);
  96 static dev_t nvme_chr_devt;
  97 static struct class *nvme_class;
  98 static struct class *nvme_subsys_class;
  99
 100 static void nvme_ns_remove(struct nvme_ns *ns);
 101 static int nvme_revalidate_disk(struct gendisk *disk);
 102 static void nvme_put_subsystem(struct nvme_subsystem *subsys);
 103 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
 104                                            unsigned nsid);
 105
 106 static void nvme_set_queue_dying(struct nvme_ns *ns)
 107 {
 108         /*
 109          * Revalidating a dead namespace sets capacity to 0. This will end
 110          * buffered writers dirtying pages that can't be synced.
 111          */
 112         if (!ns->disk || test_and_set_bit(NVME_NS_DEAD, &ns->flags))
 113                 return;
 114         revalidate_disk(ns->disk);
 115         blk_set_queue_dying(ns->queue);
 116         /* Forcibly unquiesce queues to avoid blocking dispatch */
 117         blk_mq_unquiesce_queue(ns->queue);
 118 }
 119
 120 static void nvme_queue_scan(struct nvme_ctrl *ctrl)
 121 {
 122         /*
 123          * Only new queue scan work when admin and IO queues are both alive
 124          */
 125         if (ctrl->state == NVME_CTRL_LIVE)
 126                 queue_work(nvme_wq, &ctrl->scan_work);
 127 }
 128
 129 int nvme_reset_ctrl(struct nvme_ctrl *ctrl)
 130 {
 131         if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING))
 132                 return -EBUSY;
 133         if (!queue_work(nvme_reset_wq, &ctrl->reset_work))
 134                 return -EBUSY;
 135         return 0;
 136 }
 137 EXPORT_SYMBOL_GPL(nvme_reset_ctrl);
 138
 139 int nvme_reset_ctrl_sync(struct nvme_ctrl *ctrl)
 140 {
 141         int ret;
 142
 143         ret = nvme_reset_ctrl(ctrl);
 144         if (!ret) {
 145                 flush_work(&ctrl->reset_work);
 146                 if (ctrl->state != NVME_CTRL_LIVE &&
 147                     ctrl->state != NVME_CTRL_ADMIN_ONLY)
 148                         ret = -ENETRESET;
 149         }
 150
 151         return ret;
 152 }
 153 EXPORT_SYMBOL_GPL(nvme_reset_ctrl_sync);
 154
 155 static void nvme_delete_ctrl_work(struct work_struct *work)
 156 {
 157         struct nvme_ctrl *ctrl =
 158                 container_of(work, struct nvme_ctrl, delete_work);
 159
 160         dev_info(ctrl->device,
 161                  "Removing ctrl: NQN \"%s\"\n", ctrl->opts->subsysnqn);
 162
 163         flush_work(&ctrl->reset_work);
 164         nvme_stop_ctrl(ctrl);
 165         nvme_remove_namespaces(ctrl);
 166         ctrl->ops->delete_ctrl(ctrl);
 167         nvme_uninit_ctrl(ctrl);
 168         nvme_put_ctrl(ctrl);
 169 }
 170
 171 int nvme_delete_ctrl(struct nvme_ctrl *ctrl)
 172 {
 173         if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
 174                 return -EBUSY;
 175         if (!queue_work(nvme_delete_wq, &ctrl->delete_work))
 176                 return -EBUSY;
 177         return 0;
 178 }
 179 EXPORT_SYMBOL_GPL(nvme_delete_ctrl);
 180
 181 int nvme_delete_ctrl_sync(struct nvme_ctrl *ctrl)
 182 {
 183         int ret = 0;
 184
 185         /*
 186          * Keep a reference until the work is flushed since ->delete_ctrl
 187          * can free the controller.
 188          */
 189         nvme_get_ctrl(ctrl);
 190         ret = nvme_delete_ctrl(ctrl);
 191         if (!ret)
 192                 flush_work(&ctrl->delete_work);
 193         nvme_put_ctrl(ctrl);
 194         return ret;
 195 }
 196 EXPORT_SYMBOL_GPL(nvme_delete_ctrl_sync);
 197
 198 static inline bool nvme_ns_has_pi(struct nvme_ns *ns)
 199 {
 200         return ns->pi_type && ns->ms == sizeof(struct t10_pi_tuple);
 201 }
 202
 203 static blk_status_t nvme_error_status(struct request *req)
 204 {
 205         switch (nvme_req(req)->status & 0x7ff) {
 206         case NVME_SC_SUCCESS:
 207                 return BLK_STS_OK;
 208         case NVME_SC_CAP_EXCEEDED:
 209                 return BLK_STS_NOSPC;
 210         case NVME_SC_LBA_RANGE:
 211                 return BLK_STS_TARGET;
 212         case NVME_SC_BAD_ATTRIBUTES:
 213         case NVME_SC_ONCS_NOT_SUPPORTED:
 214         case NVME_SC_INVALID_OPCODE:
 215         case NVME_SC_INVALID_FIELD:
 216         case NVME_SC_INVALID_NS:
 217                 return BLK_STS_NOTSUPP;
 218         case NVME_SC_WRITE_FAULT:
 219         case NVME_SC_READ_ERROR:
 220         case NVME_SC_UNWRITTEN_BLOCK:
 221         case NVME_SC_ACCESS_DENIED:
 222         case NVME_SC_READ_ONLY:
 223         case NVME_SC_COMPARE_FAILED:
 224                 return BLK_STS_MEDIUM;
 225         case NVME_SC_GUARD_CHECK:
 226         case NVME_SC_APPTAG_CHECK:
 227         case NVME_SC_REFTAG_CHECK:
 228         case NVME_SC_INVALID_PI:
 229                 return BLK_STS_PROTECTION;
 230         case NVME_SC_RESERVATION_CONFLICT:
 231                 return BLK_STS_NEXUS;
 232         default:
 233                 return BLK_STS_IOERR;
 234         }
 235 }
 236
 237 static inline bool nvme_req_needs_retry(struct request *req)
 238 {
 239         if (blk_noretry_request(req))
 240                 return false;
 241         if (nvme_req(req)->status & NVME_SC_DNR)
 242                 return false;
 243         if (nvme_req(req)->retries >= nvme_max_retries)
 244                 return false;
 245         return true;
 246 }
 247
 248 void nvme_complete_rq(struct request *req)
 249 {
 250         blk_status_t status = nvme_error_status(req);
 251
 252         trace_nvme_complete_rq(req);
 253
 254         if (unlikely(status != BLK_STS_OK && nvme_req_needs_retry(req))) {
 255                 if ((req->cmd_flags & REQ_NVME_MPATH) &&
 256                     blk_path_error(status)) {
 257                         nvme_failover_req(req);
 258                         return;
 259                 }
 260
 261                 if (!blk_queue_dying(req->q)) {
 262                         nvme_req(req)->retries++;
 263                         blk_mq_requeue_request(req, true);
 264                         return;
 265                 }
 266         }
 267         blk_mq_end_request(req, status);
 268 }
 269 EXPORT_SYMBOL_GPL(nvme_complete_rq);
 270
 271 void nvme_cancel_request(struct request *req, void *data, bool reserved)
 272 {
 273         dev_dbg_ratelimited(((struct nvme_ctrl *) data)->device,
 274                                 "Cancelling I/O %d", req->tag);
 275
 276         nvme_req(req)->status = NVME_SC_ABORT_REQ;
 277         blk_mq_complete_request(req);
 278
 279 }
 280 EXPORT_SYMBOL_GPL(nvme_cancel_request);
 281
 282 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
 283                 enum nvme_ctrl_state new_state)
 284 {
 285         enum nvme_ctrl_state old_state;
 286         unsigned long flags;
 287         bool changed = false;
 288
 289         spin_lock_irqsave(&ctrl->lock, flags);
 290
 291         old_state = ctrl->state;
 292         switch (new_state) {
 293         case NVME_CTRL_ADMIN_ONLY:
 294                 switch (old_state) {
 295                 case NVME_CTRL_CONNECTING:
 296                         changed = true;
 297                         /* FALLTHRU */
 298                 default:
 299                         break;
 300                 }
 301                 break;
 302         case NVME_CTRL_LIVE:
 303                 switch (old_state) {
 304                 case NVME_CTRL_NEW:
 305                 case NVME_CTRL_RESETTING:
 306                 case NVME_CTRL_CONNECTING:
 307                         changed = true;
 308                         /* FALLTHRU */
 309                 default:
 310                         break;
 311                 }
 312                 break;
 313         case NVME_CTRL_RESETTING:
 314                 switch (old_state) {
 315                 case NVME_CTRL_NEW:
 316                 case NVME_CTRL_LIVE:
 317                 case NVME_CTRL_ADMIN_ONLY:
 318                         changed = true;
 319                         /* FALLTHRU */
 320                 default:
 321                         break;
 322                 }
 323                 break;
 324         case NVME_CTRL_CONNECTING:
 325                 switch (old_state) {
 326                 case NVME_CTRL_NEW:
 327                 case NVME_CTRL_RESETTING:
 328                         changed = true;
 329                         /* FALLTHRU */
 330                 default:
 331                         break;
 332                 }
 333                 break;
 334         case NVME_CTRL_DELETING:
 335                 switch (old_state) {
 336                 case NVME_CTRL_LIVE:
 337                 case NVME_CTRL_ADMIN_ONLY:
 338                 case NVME_CTRL_RESETTING:
 339                 case NVME_CTRL_CONNECTING:
 340                         changed = true;
 341                         /* FALLTHRU */
 342                 default:
 343                         break;
 344                 }
 345                 break;
 346         case NVME_CTRL_DEAD:
 347                 switch (old_state) {
 348                 case NVME_CTRL_DELETING:
 349                         changed = true;
 350                         /* FALLTHRU */
 351                 default:
 352                         break;
 353                 }
 354                 break;
 355         default:
 356                 break;
 357         }
 358
 359         if (changed)
 360                 ctrl->state = new_state;
 361
 362         spin_unlock_irqrestore(&ctrl->lock, flags);
 363         if (changed && ctrl->state == NVME_CTRL_LIVE)
 364                 nvme_kick_requeue_lists(ctrl);
 365         return changed;
 366 }
 367 EXPORT_SYMBOL_GPL(nvme_change_ctrl_state);
 368
 369 static void nvme_free_ns_head(struct kref *ref)
 370 {
 371         struct nvme_ns_head *head =
 372                 container_of(ref, struct nvme_ns_head, ref);
 373
 374         nvme_mpath_remove_disk(head);
 375         ida_simple_remove(&head->subsys->ns_ida, head->instance);
 376         list_del_init(&head->entry);
 377         cleanup_srcu_struct_quiesced(&head->srcu);
 378         nvme_put_subsystem(head->subsys);
 379         kfree(head);
 380 }
 381
 382 static void nvme_put_ns_head(struct nvme_ns_head *head)
 383 {
 384         kref_put(&head->ref, nvme_free_ns_head);
 385 }
 386
 387 static void nvme_free_ns(struct kref *kref)
 388 {
 389         struct nvme_ns *ns = container_of(kref, struct nvme_ns, kref);
 390
 391         if (ns->ndev)
 392                 nvme_nvm_unregister(ns);
 393
 394         put_disk(ns->disk);
 395         nvme_put_ns_head(ns->head);
 396         nvme_put_ctrl(ns->ctrl);
 397         kfree(ns);
 398 }
 399
 400 static void nvme_put_ns(struct nvme_ns *ns)
 401 {
 402         kref_put(&ns->kref, nvme_free_ns);
 403 }
 404
 405 static inline void nvme_clear_nvme_request(struct request *req)
 406 {
 407         if (!(req->rq_flags & RQF_DONTPREP)) {
 408                 nvme_req(req)->retries = 0;
 409                 nvme_req(req)->flags = 0;
 410                 req->rq_flags |= RQF_DONTPREP;
 411         }
 412 }
 413
 414 struct request *nvme_alloc_request(struct request_queue *q,
 415                 struct nvme_command *cmd, blk_mq_req_flags_t flags, int qid)
 416 {
 417         unsigned op = nvme_is_write(cmd) ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN;
 418         struct request *req;
 419
 420         if (qid == NVME_QID_ANY) {
 421                 req = blk_mq_alloc_request(q, op, flags);
 422         } else {
 423                 req = blk_mq_alloc_request_hctx(q, op, flags,
 424                                 qid ? qid - 1 : 0);
 425         }
 426         if (IS_ERR(req))
 427                 return req;
 428
 429         req->cmd_flags |= REQ_FAILFAST_DRIVER;
 430         nvme_clear_nvme_request(req);
 431         nvme_req(req)->cmd = cmd;
 432
 433         return req;
 434 }
 435 EXPORT_SYMBOL_GPL(nvme_alloc_request);
 436
 437 static int nvme_toggle_streams(struct nvme_ctrl *ctrl, bool enable)
 438 {
 439         struct nvme_command c;
 440
 441         memset(&c, 0, sizeof(c));
 442
 443         c.directive.opcode = nvme_admin_directive_send;
 444         c.directive.nsid = cpu_to_le32(NVME_NSID_ALL);
 445         c.directive.doper = NVME_DIR_SND_ID_OP_ENABLE;
 446         c.directive.dtype = NVME_DIR_IDENTIFY;
 447         c.directive.tdtype = NVME_DIR_STREAMS;
 448         c.directive.endir = enable ? NVME_DIR_ENDIR : 0;
 449
 450         return nvme_submit_sync_cmd(ctrl->admin_q, &c, NULL, 0);
 451 }
 452
 453 static int nvme_disable_streams(struct nvme_ctrl *ctrl)
 454 {
 455         return nvme_toggle_streams(ctrl, false);
 456 }
 457
 458 static int nvme_enable_streams(struct nvme_ctrl *ctrl)
 459 {
 460         return nvme_toggle_streams(ctrl, true);
 461 }
 462
 463 static int nvme_get_stream_params(struct nvme_ctrl *ctrl,
 464                                   struct streams_directive_params *s, u32 nsid)
 465 {
 466         struct nvme_command c;
 467
 468         memset(&c, 0, sizeof(c));
 469         memset(s, 0, sizeof(*s));
 470
 471         c.directive.opcode = nvme_admin_directive_recv;
 472         c.directive.nsid = cpu_to_le32(nsid);
 473         c.directive.numd = cpu_to_le32((sizeof(*s) >> 2) - 1);
 474         c.directive.doper = NVME_DIR_RCV_ST_OP_PARAM;
 475         c.directive.dtype = NVME_DIR_STREAMS;
 476
 477         return nvme_submit_sync_cmd(ctrl->admin_q, &c, s, sizeof(*s));
 478 }
 479
 480 static int nvme_configure_directives(struct nvme_ctrl *ctrl)
 481 {
 482         struct streams_directive_params s;
 483         int ret;
 484
 485         if (!(ctrl->oacs & NVME_CTRL_OACS_DIRECTIVES))
 486                 return 0;
 487         if (!streams)
 488                 return 0;
 489
 490         ret = nvme_enable_streams(ctrl);
 491         if (ret)
 492                 return ret;
 493
 494         ret = nvme_get_stream_params(ctrl, &s, NVME_NSID_ALL);
 495         if (ret)
 496                 return ret;
 497
 498         ctrl->nssa = le16_to_cpu(s.nssa);
 499         if (ctrl->nssa < BLK_MAX_WRITE_HINTS - 1) {
 500                 dev_info(ctrl->device, "too few streams (%u) available\n",
 501                                         ctrl->nssa);
 502                 nvme_disable_streams(ctrl);
 503                 return 0;
 504         }
 505
 506         ctrl->nr_streams = min_t(unsigned, ctrl->nssa, BLK_MAX_WRITE_HINTS - 1);
 507         dev_info(ctrl->device, "Using %u streams\n", ctrl->nr_streams);
 508         return 0;
 509 }
 510
 511 /*
 512  * Check if 'req' has a write hint associated with it. If it does, assign
 513  * a valid namespace stream to the write.
 514  */
 515 static void nvme_assign_write_stream(struct nvme_ctrl *ctrl,
 516                                      struct request *req, u16 *control,
 517                                      u32 *dsmgmt)
 518 {
 519         enum rw_hint streamid = req->write_hint;
 520
 521         if (streamid == WRITE_LIFE_NOT_SET || streamid == WRITE_LIFE_NONE)
 522                 streamid = 0;
 523         else {
 524                 streamid--;
 525                 if (WARN_ON_ONCE(streamid > ctrl->nr_streams))
 526                         return;
 527
 528                 *control |= NVME_RW_DTYPE_STREAMS;
 529                 *dsmgmt |= streamid << 16;
 530         }
 531
 532         if (streamid < ARRAY_SIZE(req->q->write_hints))
 533                 req->q->write_hints[streamid] += blk_rq_bytes(req) >> 9;
 534 }
 535
 536 static inline void nvme_setup_flush(struct nvme_ns *ns,
 537                 struct nvme_command *cmnd)
 538 {
 539         memset(cmnd, 0, sizeof(*cmnd));
 540         cmnd->common.opcode = nvme_cmd_flush;
 541         cmnd->common.nsid = cpu_to_le32(ns->head->ns_id);
 542 }
 543
 544 static blk_status_t nvme_setup_discard(struct nvme_ns *ns, struct request *req,
 545                 struct nvme_command *cmnd)
 546 {
 547         unsigned short segments = blk_rq_nr_discard_segments(req), n = 0;
 548         struct nvme_dsm_range *range;
 549         struct bio *bio;
 550
 551         range = kmalloc_array(segments, sizeof(*range), GFP_ATOMIC);
 552         if (!range)
 553                 return BLK_STS_RESOURCE;
 554
 555         __rq_for_each_bio(bio, req) {
 556                 u64 slba = nvme_block_nr(ns, bio->bi_iter.bi_sector);
 557                 u32 nlb = bio->bi_iter.bi_size >> ns->lba_shift;
 558
 559                 if (n < segments) {
 560                         range[n].cattr = cpu_to_le32(0);
 561                         range[n].nlb = cpu_to_le32(nlb);
 562                         range[n].slba = cpu_to_le64(slba);
 563                 }
 564                 n++;
 565         }
 566
 567         if (WARN_ON_ONCE(n != segments)) {
 568                 kfree(range);
 569                 return BLK_STS_IOERR;
 570         }
 571
 572         memset(cmnd, 0, sizeof(*cmnd));
 573         cmnd->dsm.opcode = nvme_cmd_dsm;
 574         cmnd->dsm.nsid = cpu_to_le32(ns->head->ns_id);
 575         cmnd->dsm.nr = cpu_to_le32(segments - 1);
 576         cmnd->dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD);
 577
 578         req->special_vec.bv_page = virt_to_page(range);
 579         req->special_vec.bv_offset = offset_in_page(range);
 580         req->special_vec.bv_len = sizeof(*range) * segments;
 581         req->rq_flags |= RQF_SPECIAL_PAYLOAD;
 582
 583         return BLK_STS_OK;
 584 }
 585
 586 static inline blk_status_t nvme_setup_rw(struct nvme_ns *ns,
 587                 struct request *req, struct nvme_command *cmnd)
 588 {
 589         struct nvme_ctrl *ctrl = ns->ctrl;
 590         u16 control = 0;
 591         u32 dsmgmt = 0;
 592
 593         if (req->cmd_flags & REQ_FUA)
 594                 control |= NVME_RW_FUA;
 595         if (req->cmd_flags & (REQ_FAILFAST_DEV | REQ_RAHEAD))
 596                 control |= NVME_RW_LR;
 597
 598         if (req->cmd_flags & REQ_RAHEAD)
 599                 dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH;
 600
 601         memset(cmnd, 0, sizeof(*cmnd));
 602         cmnd->rw.opcode = (rq_data_dir(req) ? nvme_cmd_write : nvme_cmd_read);
 603         cmnd->rw.nsid = cpu_to_le32(ns->head->ns_id);
 604         cmnd->rw.slba = cpu_to_le64(nvme_block_nr(ns, blk_rq_pos(req)));
 605         cmnd->rw.length = cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
 606
 607         if (req_op(req) == REQ_OP_WRITE && ctrl->nr_streams)
 608                 nvme_assign_write_stream(ctrl, req, &control, &dsmgmt);
 609
 610         if (ns->ms) {
 611                 /*
 612                  * If formated with metadata, the block layer always provides a
 613                  * metadata buffer if CONFIG_BLK_DEV_INTEGRITY is enabled.  Else
 614                  * we enable the PRACT bit for protection information or set the
 615                  * namespace capacity to zero to prevent any I/O.
 616                  */
 617                 if (!blk_integrity_rq(req)) {
 618                         if (WARN_ON_ONCE(!nvme_ns_has_pi(ns)))
 619                                 return BLK_STS_NOTSUPP;
 620                         control |= NVME_RW_PRINFO_PRACT;
 621                 } else if (req_op(req) == REQ_OP_WRITE) {
 622                         t10_pi_prepare(req, ns->pi_type);
 623                 }
 624
 625                 switch (ns->pi_type) {
 626                 case NVME_NS_DPS_PI_TYPE3:
 627                         control |= NVME_RW_PRINFO_PRCHK_GUARD;
 628                         break;
 629                 case NVME_NS_DPS_PI_TYPE1:
 630                 case NVME_NS_DPS_PI_TYPE2:
 631                         control |= NVME_RW_PRINFO_PRCHK_GUARD |
 632                                         NVME_RW_PRINFO_PRCHK_REF;
 633                         cmnd->rw.reftag = cpu_to_le32(t10_pi_ref_tag(req));
 634                         break;
 635                 }
 636         }
 637
 638         cmnd->rw.control = cpu_to_le16(control);
 639         cmnd->rw.dsmgmt = cpu_to_le32(dsmgmt);
 640         return 0;
 641 }
 642
 643 void nvme_cleanup_cmd(struct request *req)
 644 {
 645         if (blk_integrity_rq(req) && req_op(req) == REQ_OP_READ &&
 646             nvme_req(req)->status == 0) {
 647                 struct nvme_ns *ns = req->rq_disk->private_data;
 648
 649                 t10_pi_complete(req, ns->pi_type,
 650                                 blk_rq_bytes(req) >> ns->lba_shift);
 651         }
 652         if (req->rq_flags & RQF_SPECIAL_PAYLOAD) {
 653                 kfree(page_address(req->special_vec.bv_page) +
 654                       req->special_vec.bv_offset);
 655         }
 656 }
 657 EXPORT_SYMBOL_GPL(nvme_cleanup_cmd);
 658
 659 blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req,
 660                 struct nvme_command *cmd)
 661 {
 662         blk_status_t ret = BLK_STS_OK;
 663
 664         nvme_clear_nvme_request(req);
 665
 666         switch (req_op(req)) {
 667         case REQ_OP_DRV_IN:
 668         case REQ_OP_DRV_OUT:
 669                 memcpy(cmd, nvme_req(req)->cmd, sizeof(*cmd));
 670                 break;
 671         case REQ_OP_FLUSH:
 672                 nvme_setup_flush(ns, cmd);
 673                 break;
 674         case REQ_OP_WRITE_ZEROES:
 675                 /* currently only aliased to deallocate for a few ctrls: */
 676         case REQ_OP_DISCARD:
 677                 ret = nvme_setup_discard(ns, req, cmd);
 678                 break;
 679         case REQ_OP_READ:
 680         case REQ_OP_WRITE:
 681                 ret = nvme_setup_rw(ns, req, cmd);
 682                 break;
 683         default:
 684                 WARN_ON_ONCE(1);
 685                 return BLK_STS_IOERR;
 686         }
 687
 688         cmd->common.command_id = req->tag;
 689         trace_nvme_setup_cmd(req, cmd);
 690         return ret;
 691 }
 692 EXPORT_SYMBOL_GPL(nvme_setup_cmd);
 693
 694 /*
 695  * Returns 0 on success.  If the result is negative, it's a Linux error code;
 696  * if the result is positive, it's an NVM Express status code
 697  */
 698 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
 699                 union nvme_result *result, void *buffer, unsigned bufflen,
 700                 unsigned timeout, int qid, int at_head,
 701                 blk_mq_req_flags_t flags)
 702 {
 703         struct request *req;
 704         int ret;
 705
 706         req = nvme_alloc_request(q, cmd, flags, qid);
 707         if (IS_ERR(req))
 708                 return PTR_ERR(req);
 709
 710         req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
 711
 712         if (buffer && bufflen) {
 713                 ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL);
 714                 if (ret)
 715                         goto out;
 716         }
 717
 718         blk_execute_rq(req->q, NULL, req, at_head);
 719         if (result)
 720                 *result = nvme_req(req)->result;
 721         if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
 722                 ret = -EINTR;
 723         else
 724                 ret = nvme_req(req)->status;
 725  out:
 726         blk_mq_free_request(req);
 727         return ret;
 728 }
 729 EXPORT_SYMBOL_GPL(__nvme_submit_sync_cmd);
 730
 731 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
 732                 void *buffer, unsigned bufflen)
 733 {
 734         return __nvme_submit_sync_cmd(q, cmd, NULL, buffer, bufflen, 0,
 735                         NVME_QID_ANY, 0, 0);
 736 }
 737 EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd);
 738
 739 static void *nvme_add_user_metadata(struct bio *bio, void __user *ubuf,
 740                 unsigned len, u32 seed, bool write)
 741 {
 742         struct bio_integrity_payload *bip;
 743         int ret = -ENOMEM;
 744         void *buf;
 745
 746         buf = kmalloc(len, GFP_KERNEL);
 747         if (!buf)
 748                 goto out;
 749
 750         ret = -EFAULT;
 751         if (write && copy_from_user(buf, ubuf, len))
 752                 goto out_free_meta;
 753
 754         bip = bio_integrity_alloc(bio, GFP_KERNEL, 1);
 755         if (IS_ERR(bip)) {
 756                 ret = PTR_ERR(bip);
 757                 goto out_free_meta;
 758         }
 759
 760         bip->bip_iter.bi_size = len;
 761         bip->bip_iter.bi_sector = seed;
 762         ret = bio_integrity_add_page(bio, virt_to_page(buf), len,
 763                         offset_in_page(buf));
 764         if (ret == len)
 765                 return buf;
 766         ret = -ENOMEM;
 767 out_free_meta:
 768         kfree(buf);
 769 out:
 770         return ERR_PTR(ret);
 771 }
 772
 773 static int nvme_submit_user_cmd(struct request_queue *q,
 774                 struct nvme_command *cmd, void __user *ubuffer,
 775                 unsigned bufflen, void __user *meta_buffer, unsigned meta_len,
 776                 u32 meta_seed, u32 *result, unsigned timeout)
 777 {
 778         bool write = nvme_is_write(cmd);
 779         struct nvme_ns *ns = q->queuedata;
 780         struct gendisk *disk = ns ? ns->disk : NULL;
 781         struct request *req;
 782         struct bio *bio = NULL;
 783         void *meta = NULL;
 784         int ret;
 785
 786         req = nvme_alloc_request(q, cmd, 0, NVME_QID_ANY);
 787         if (IS_ERR(req))
 788                 return PTR_ERR(req);
 789
 790         req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
 791         nvme_req(req)->flags |= NVME_REQ_USERCMD;
 792
 793         if (ubuffer && bufflen) {
 794                 ret = blk_rq_map_user(q, req, NULL, ubuffer, bufflen,
 795                                 GFP_KERNEL);
 796                 if (ret)
 797                         goto out;
 798                 bio = req->bio;
 799                 bio->bi_disk = disk;
 800                 if (disk && meta_buffer && meta_len) {
 801                         meta = nvme_add_user_metadata(bio, meta_buffer, meta_len,
 802                                         meta_seed, write);
 803                         if (IS_ERR(meta)) {
 804                                 ret = PTR_ERR(meta);
 805                                 goto out_unmap;
 806                         }
 807                         req->cmd_flags |= REQ_INTEGRITY;
 808                 }
 809         }
 810
 811         blk_execute_rq(req->q, disk, req, 0);
 812         if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
 813                 ret = -EINTR;
 814         else
 815                 ret = nvme_req(req)->status;
 816         if (result)
 817                 *result = le32_to_cpu(nvme_req(req)->result.u32);
 818         if (meta && !ret && !write) {
 819                 if (copy_to_user(meta_buffer, meta, meta_len))
 820                         ret = -EFAULT;
 821         }
 822         kfree(meta);
 823  out_unmap:
 824         if (bio)
 825                 blk_rq_unmap_user(bio);
 826  out:
 827         blk_mq_free_request(req);
 828         return ret;
 829 }
 830
 831 static void nvme_keep_alive_end_io(struct request *rq, blk_status_t status)
 832 {
 833         struct nvme_ctrl *ctrl = rq->end_io_data;
 834
 835         blk_mq_free_request(rq);
 836
 837         if (status) {
 838                 dev_err(ctrl->device,
 839                         "failed nvme_keep_alive_end_io error=%d\n",
 840                                 status);
 841                 return;
 842         }
 843
 844         schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
 845 }
 846
 847 static int nvme_keep_alive(struct nvme_ctrl *ctrl)
 848 {
 849         struct request *rq;
 850
 851         rq = nvme_alloc_request(ctrl->admin_q, &ctrl->ka_cmd, BLK_MQ_REQ_RESERVED,
 852                         NVME_QID_ANY);
 853         if (IS_ERR(rq))
 854                 return PTR_ERR(rq);
 855
 856         rq->timeout = ctrl->kato * HZ;
 857         rq->end_io_data = ctrl;
 858
 859         blk_execute_rq_nowait(rq->q, NULL, rq, 0, nvme_keep_alive_end_io);
 860
 861         return 0;
 862 }
 863
 864 static void nvme_keep_alive_work(struct work_struct *work)
 865 {
 866         struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
 867                         struct nvme_ctrl, ka_work);
 868
 869         if (nvme_keep_alive(ctrl)) {
 870                 /* allocation failure, reset the controller */
 871                 dev_err(ctrl->device, "keep-alive failed\n");
 872                 nvme_reset_ctrl(ctrl);
 873                 return;
 874         }
 875 }
 876
 877 static void nvme_start_keep_alive(struct nvme_ctrl *ctrl)
 878 {
 879         if (unlikely(ctrl->kato == 0))
 880                 return;
 881
 882         schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
 883 }
 884
 885 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl)
 886 {
 887         if (unlikely(ctrl->kato == 0))
 888                 return;
 889
 890         cancel_delayed_work_sync(&ctrl->ka_work);
 891 }
 892 EXPORT_SYMBOL_GPL(nvme_stop_keep_alive);
 893
 894 static int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
 895 {
 896         struct nvme_command c = { };
 897         int error;
 898
 899         /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
 900         c.identify.opcode = nvme_admin_identify;
 901         c.identify.cns = NVME_ID_CNS_CTRL;
 902
 903         *id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
 904         if (!*id)
 905                 return -ENOMEM;
 906
 907         error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
 908                         sizeof(struct nvme_id_ctrl));
 909         if (error)
 910                 kfree(*id);
 911         return error;
 912 }
 913
 914 static int nvme_identify_ns_descs(struct nvme_ctrl *ctrl, unsigned nsid,
 915                 struct nvme_ns_ids *ids)
 916 {
 917         struct nvme_command c = { };
 918         int status;
 919         void *data;
 920         int pos;
 921         int len;
 922
 923         c.identify.opcode = nvme_admin_identify;
 924         c.identify.nsid = cpu_to_le32(nsid);
 925         c.identify.cns = NVME_ID_CNS_NS_DESC_LIST;
 926
 927         data = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
 928         if (!data)
 929                 return -ENOMEM;
 930
 931         status = nvme_submit_sync_cmd(ctrl->admin_q, &c, data,
 932                                       NVME_IDENTIFY_DATA_SIZE);
 933         if (status)
 934                 goto free_data;
 935
 936         for (pos = 0; pos < NVME_IDENTIFY_DATA_SIZE; pos += len) {
 937                 struct nvme_ns_id_desc *cur = data + pos;
 938
 939                 if (cur->nidl == 0)
 940                         break;
 941
 942                 switch (cur->nidt) {
 943                 case NVME_NIDT_EUI64:
 944                         if (cur->nidl != NVME_NIDT_EUI64_LEN) {
 945                                 dev_warn(ctrl->device,
 946                                          "ctrl returned bogus length: %d for NVME_NIDT_EUI64\n",
 947                                          cur->nidl);
 948                                 goto free_data;
 949                         }
 950                         len = NVME_NIDT_EUI64_LEN;
 951                         memcpy(ids->eui64, data + pos + sizeof(*cur), len);
 952                         break;
 953                 case NVME_NIDT_NGUID:
 954                         if (cur->nidl != NVME_NIDT_NGUID_LEN) {
 955                                 dev_warn(ctrl->device,
 956                                          "ctrl returned bogus length: %d for NVME_NIDT_NGUID\n",
 957                                          cur->nidl);
 958                                 goto free_data;
 959                         }
 960                         len = NVME_NIDT_NGUID_LEN;
 961                         memcpy(ids->nguid, data + pos + sizeof(*cur), len);
 962                         break;
 963                 case NVME_NIDT_UUID:
 964                         if (cur->nidl != NVME_NIDT_UUID_LEN) {
 965                                 dev_warn(ctrl->device,
 966                                          "ctrl returned bogus length: %d for NVME_NIDT_UUID\n",
 967                                          cur->nidl);
 968                                 goto free_data;
 969                         }
 970                         len = NVME_NIDT_UUID_LEN;
 971                         uuid_copy(&ids->uuid, data + pos + sizeof(*cur));
 972                         break;
 973                 default:
 974                         /* Skip unknown types */
 975                         len = cur->nidl;
 976                         break;
 977                 }
 978
 979                 len += sizeof(*cur);
 980         }
 981 free_data:
 982         kfree(data);
 983         return status;
 984 }
 985
 986 static int nvme_identify_ns_list(struct nvme_ctrl *dev, unsigned nsid, __le32 *ns_list)
 987 {
 988         struct nvme_command c = { };
 989
 990         c.identify.opcode = nvme_admin_identify;
 991         c.identify.cns = NVME_ID_CNS_NS_ACTIVE_LIST;
 992         c.identify.nsid = cpu_to_le32(nsid);
 993         return nvme_submit_sync_cmd(dev->admin_q, &c, ns_list,
 994                                     NVME_IDENTIFY_DATA_SIZE);
 995 }
 996
 997 static struct nvme_id_ns *nvme_identify_ns(struct nvme_ctrl *ctrl,
 998                 unsigned nsid)
 999 {
1000         struct nvme_id_ns *id;
1001         struct nvme_command c = { };
1002         int error;
1003
1004         /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1005         c.identify.opcode = nvme_admin_identify;
1006         c.identify.nsid = cpu_to_le32(nsid);
1007         c.identify.cns = NVME_ID_CNS_NS;
1008
1009         id = kmalloc(sizeof(*id), GFP_KERNEL);
1010         if (!id)
1011                 return NULL;
1012
1013         error = nvme_submit_sync_cmd(ctrl->admin_q, &c, id, sizeof(*id));
1014         if (error) {
1015                 dev_warn(ctrl->device, "Identify namespace failed\n");
1016                 kfree(id);
1017                 return NULL;
1018         }
1019
1020         return id;
1021 }
1022
1023 static int nvme_set_features(struct nvme_ctrl *dev, unsigned fid, unsigned dword11,
1024                       void *buffer, size_t buflen, u32 *result)
1025 {
1026         struct nvme_command c;
1027         union nvme_result res;
1028         int ret;
1029
1030         memset(&c, 0, sizeof(c));
1031         c.features.opcode = nvme_admin_set_features;
1032         c.features.fid = cpu_to_le32(fid);
1033         c.features.dword11 = cpu_to_le32(dword11);
1034
1035         ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &res,
1036                         buffer, buflen, 0, NVME_QID_ANY, 0, 0);
1037         if (ret >= 0 && result)
1038                 *result = le32_to_cpu(res.u32);
1039         return ret;
1040 }
1041
1042 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count)
1043 {
1044         u32 q_count = (*count - 1) | ((*count - 1) << 16);
1045         u32 result;
1046         int status, nr_io_queues;
1047
1048         status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, NULL, 0,
1049                         &result);
1050         if (status < 0)
1051                 return status;
1052
1053         /*
1054          * Degraded controllers might return an error when setting the queue
1055          * count.  We still want to be able to bring them online and offer
1056          * access to the admin queue, as that might be only way to fix them up.
1057          */
1058         if (status > 0) {
1059                 dev_err(ctrl->device, "Could not set queue count (%d)\n", status);
1060                 *count = 0;
1061         } else {
1062                 nr_io_queues = min(result & 0xffff, result >> 16) + 1;
1063                 *count = min(*count, nr_io_queues);
1064         }
1065
1066         return 0;
1067 }
1068 EXPORT_SYMBOL_GPL(nvme_set_queue_count);
1069
1070 #define NVME_AEN_SUPPORTED \
1071         (NVME_AEN_CFG_NS_ATTR | NVME_AEN_CFG_FW_ACT | NVME_AEN_CFG_ANA_CHANGE)
1072
1073 static void nvme_enable_aen(struct nvme_ctrl *ctrl)
1074 {
1075         u32 result, supported_aens = ctrl->oaes & NVME_AEN_SUPPORTED;
1076         int status;
1077
1078         if (!supported_aens)
1079                 return;
1080
1081         status = nvme_set_features(ctrl, NVME_FEAT_ASYNC_EVENT, supported_aens,
1082                         NULL, 0, &result);
1083         if (status)
1084                 dev_warn(ctrl->device, "Failed to configure AEN (cfg %x)\n",
1085                          supported_aens);
1086 }
1087
1088 static int nvme_submit_io(struct nvme_ns *ns, struct nvme_user_io __user *uio)
1089 {
1090         struct nvme_user_io io;
1091         struct nvme_command c;
1092         unsigned length, meta_len;
1093         void __user *metadata;
1094
1095         if (copy_from_user(&io, uio, sizeof(io)))
1096                 return -EFAULT;
1097         if (io.flags)
1098                 return -EINVAL;
1099
1100         switch (io.opcode) {
1101         case nvme_cmd_write:
1102         case nvme_cmd_read:
1103         case nvme_cmd_compare:
1104                 break;
1105         default:
1106                 return -EINVAL;
1107         }
1108
1109         length = (io.nblocks + 1) << ns->lba_shift;
1110         meta_len = (io.nblocks + 1) * ns->ms;
1111         metadata = (void __user *)(uintptr_t)io.metadata;
1112
1113         if (ns->ext) {
1114                 length += meta_len;
1115                 meta_len = 0;
1116         } else if (meta_len) {
1117                 if ((io.metadata & 3) || !io.metadata)
1118                         return -EINVAL;
1119         }
1120
1121         memset(&c, 0, sizeof(c));
1122         c.rw.opcode = io.opcode;
1123         c.rw.flags = io.flags;
1124         c.rw.nsid = cpu_to_le32(ns->head->ns_id);
1125         c.rw.slba = cpu_to_le64(io.slba);
1126         c.rw.length = cpu_to_le16(io.nblocks);
1127         c.rw.control = cpu_to_le16(io.control);
1128         c.rw.dsmgmt = cpu_to_le32(io.dsmgmt);
1129         c.rw.reftag = cpu_to_le32(io.reftag);
1130         c.rw.apptag = cpu_to_le16(io.apptag);
1131         c.rw.appmask = cpu_to_le16(io.appmask);
1132
1133         return nvme_submit_user_cmd(ns->queue, &c,
1134                         (void __user *)(uintptr_t)io.addr, length,
1135                         metadata, meta_len, lower_32_bits(io.slba), NULL, 0);
1136 }
1137
1138 static u32 nvme_known_admin_effects(u8 opcode)
1139 {
1140         switch (opcode) {
1141         case nvme_admin_format_nvm:
1142                 return NVME_CMD_EFFECTS_CSUPP | NVME_CMD_EFFECTS_LBCC |
1143                                         NVME_CMD_EFFECTS_CSE_MASK;
1144         case nvme_admin_sanitize_nvm:
1145                 return NVME_CMD_EFFECTS_CSE_MASK;
1146         default:
1147                 break;
1148         }
1149         return 0;
1150 }
1151
1152 static u32 nvme_passthru_start(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1153                                                                 u8 opcode)
1154 {
1155         u32 effects = 0;
1156
1157         if (ns) {
1158                 if (ctrl->effects)
1159                         effects = le32_to_cpu(ctrl->effects->iocs[opcode]);
1160                 if (effects & ~NVME_CMD_EFFECTS_CSUPP)
1161                         dev_warn(ctrl->device,
1162                                  "IO command:%02x has unhandled effects:%08x\n",
1163                                  opcode, effects);
1164                 return 0;
1165         }
1166
1167         if (ctrl->effects)
1168                 effects = le32_to_cpu(ctrl->effects->acs[opcode]);
1169         else
1170                 effects = nvme_known_admin_effects(opcode);
1171
1172         /*
1173          * For simplicity, IO to all namespaces is quiesced even if the command
1174          * effects say only one namespace is affected.
1175          */
1176         if (effects & (NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK)) {
1177                 nvme_start_freeze(ctrl);
1178                 nvme_wait_freeze(ctrl);
1179         }
1180         return effects;
1181 }
1182
1183 static void nvme_update_formats(struct nvme_ctrl *ctrl)
1184 {
1185         struct nvme_ns *ns;
1186
1187         down_read(&ctrl->namespaces_rwsem);
1188         list_for_each_entry(ns, &ctrl->namespaces, list)
1189                 if (ns->disk && nvme_revalidate_disk(ns->disk))
1190                         nvme_set_queue_dying(ns);
1191         up_read(&ctrl->namespaces_rwsem);
1192
1193         nvme_remove_invalid_namespaces(ctrl, NVME_NSID_ALL);
1194 }
1195
1196 static void nvme_passthru_end(struct nvme_ctrl *ctrl, u32 effects)
1197 {
1198         /*
1199          * Revalidate LBA changes prior to unfreezing. This is necessary to
1200          * prevent memory corruption if a logical block size was changed by
1201          * this command.
1202          */
1203         if (effects & NVME_CMD_EFFECTS_LBCC)
1204                 nvme_update_formats(ctrl);
1205         if (effects & (NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK))
1206                 nvme_unfreeze(ctrl);
1207         if (effects & NVME_CMD_EFFECTS_CCC)
1208                 nvme_init_identify(ctrl);
1209         if (effects & (NVME_CMD_EFFECTS_NIC | NVME_CMD_EFFECTS_NCC))
1210                 nvme_queue_scan(ctrl);
1211 }
1212
1213 static int nvme_user_cmd(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1214                         struct nvme_passthru_cmd __user *ucmd)
1215 {
1216         struct nvme_passthru_cmd cmd;
1217         struct nvme_command c;
1218         unsigned timeout = 0;
1219         u32 effects;
1220         int status;
1221
1222         if (!capable(CAP_SYS_ADMIN))
1223                 return -EACCES;
1224         if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
1225                 return -EFAULT;
1226         if (cmd.flags)
1227                 return -EINVAL;
1228
1229         memset(&c, 0, sizeof(c));
1230         c.common.opcode = cmd.opcode;
1231         c.common.flags = cmd.flags;
1232         c.common.nsid = cpu_to_le32(cmd.nsid);
1233         c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
1234         c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
1235         c.common.cdw10[0] = cpu_to_le32(cmd.cdw10);
1236         c.common.cdw10[1] = cpu_to_le32(cmd.cdw11);
1237         c.common.cdw10[2] = cpu_to_le32(cmd.cdw12);
1238         c.common.cdw10[3] = cpu_to_le32(cmd.cdw13);
1239         c.common.cdw10[4] = cpu_to_le32(cmd.cdw14);
1240         c.common.cdw10[5] = cpu_to_le32(cmd.cdw15);
1241
1242         if (cmd.timeout_ms)
1243                 timeout = msecs_to_jiffies(cmd.timeout_ms);
1244
1245         effects = nvme_passthru_start(ctrl, ns, cmd.opcode);
1246         status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
1247                         (void __user *)(uintptr_t)cmd.addr, cmd.data_len,
1248                         (void __user *)(uintptr_t)cmd.metadata, cmd.metadata_len,
1249                         0, &cmd.result, timeout);
1250         nvme_passthru_end(ctrl, effects);
1251
1252         if (status >= 0) {
1253                 if (put_user(cmd.result, &ucmd->result))
1254                         return -EFAULT;
1255         }
1256
1257         return status;
1258 }
1259
1260 /*
1261  * Issue ioctl requests on the first available path.  Note that unlike normal
1262  * block layer requests we will not retry failed request on another controller.
1263  */
1264 static struct nvme_ns *nvme_get_ns_from_disk(struct gendisk *disk,
1265                 struct nvme_ns_head **head, int *srcu_idx)
1266 {
1267 #ifdef CONFIG_NVME_MULTIPATH
1268         if (disk->fops == &nvme_ns_head_ops) {
1269                 *head = disk->private_data;
1270                 *srcu_idx = srcu_read_lock(&(*head)->srcu);
1271                 return nvme_find_path(*head);
1272         }
1273 #endif
1274         *head = NULL;
1275         *srcu_idx = -1;
1276         return disk->private_data;
1277 }
1278
1279 static void nvme_put_ns_from_disk(struct nvme_ns_head *head, int idx)
1280 {
1281         if (head)
1282                 srcu_read_unlock(&head->srcu, idx);
1283 }
1284
1285 static int nvme_ns_ioctl(struct nvme_ns *ns, unsigned cmd, unsigned long arg)
1286 {
1287         switch (cmd) {
1288         case NVME_IOCTL_ID:
1289                 force_successful_syscall_return();
1290                 return ns->head->ns_id;
1291         case NVME_IOCTL_ADMIN_CMD:
1292                 return nvme_user_cmd(ns->ctrl, NULL, (void __user *)arg);
1293         case NVME_IOCTL_IO_CMD:
1294                 return nvme_user_cmd(ns->ctrl, ns, (void __user *)arg);
1295         case NVME_IOCTL_SUBMIT_IO:
1296                 return nvme_submit_io(ns, (void __user *)arg);
1297         default:
1298 #ifdef CONFIG_NVM
1299                 if (ns->ndev)
1300                         return nvme_nvm_ioctl(ns, cmd, arg);
1301 #endif
1302                 if (is_sed_ioctl(cmd))
1303                         return sed_ioctl(ns->ctrl->opal_dev, cmd,
1304                                          (void __user *) arg);
1305                 return -ENOTTY;
1306         }
1307 }
1308
1309 static int nvme_ioctl(struct block_device *bdev, fmode_t mode,
1310                 unsigned int cmd, unsigned long arg)
1311 {
1312         struct nvme_ns_head *head = NULL;
1313         struct nvme_ns *ns;
1314         int srcu_idx, ret;
1315
1316         ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
1317         if (unlikely(!ns))
1318                 ret = -EWOULDBLOCK;
1319         else
1320                 ret = nvme_ns_ioctl(ns, cmd, arg);
1321         nvme_put_ns_from_disk(head, srcu_idx);
1322         return ret;
1323 }
1324
1325 static int nvme_open(struct block_device *bdev, fmode_t mode)
1326 {
1327         struct nvme_ns *ns = bdev->bd_disk->private_data;
1328
1329 #ifdef CONFIG_NVME_MULTIPATH
1330         /* should never be called due to GENHD_FL_HIDDEN */
1331         if (WARN_ON_ONCE(ns->head->disk))
1332                 goto fail;
1333 #endif
1334         if (!kref_get_unless_zero(&ns->kref))
1335                 goto fail;
1336         if (!try_module_get(ns->ctrl->ops->module))
1337                 goto fail_put_ns;
1338
1339         return 0;
1340
1341 fail_put_ns:
1342         nvme_put_ns(ns);
1343 fail:
1344         return -ENXIO;
1345 }
1346
1347 static void nvme_release(struct gendisk *disk, fmode_t mode)
1348 {
1349         struct nvme_ns *ns = disk->private_data;
1350
1351         module_put(ns->ctrl->ops->module);
1352         nvme_put_ns(ns);
1353 }
1354
1355 static int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1356 {
1357         /* some standard values */
1358         geo->heads = 1 << 6;
1359         geo->sectors = 1 << 5;
1360         geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
1361         return 0;
1362 }
1363
1364 #ifdef CONFIG_BLK_DEV_INTEGRITY
1365 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type)
1366 {
1367         struct blk_integrity integrity;
1368
1369         memset(&integrity, 0, sizeof(integrity));
1370         switch (pi_type) {
1371         case NVME_NS_DPS_PI_TYPE3:
1372                 integrity.profile = &t10_pi_type3_crc;
1373                 integrity.tag_size = sizeof(u16) + sizeof(u32);
1374                 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1375                 break;
1376         case NVME_NS_DPS_PI_TYPE1:
1377         case NVME_NS_DPS_PI_TYPE2:
1378                 integrity.profile = &t10_pi_type1_crc;
1379                 integrity.tag_size = sizeof(u16);
1380                 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1381                 break;
1382         default:
1383                 integrity.profile = NULL;
1384                 break;
1385         }
1386         integrity.tuple_size = ms;
1387         blk_integrity_register(disk, &integrity);
1388         blk_queue_max_integrity_segments(disk->queue, 1);
1389 }
1390 #else
1391 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type)
1392 {
1393 }
1394 #endif /* CONFIG_BLK_DEV_INTEGRITY */
1395
1396 static void nvme_set_chunk_size(struct nvme_ns *ns)
1397 {
1398         u32 chunk_size = (((u32)ns->noiob) << (ns->lba_shift - 9));
1399         blk_queue_chunk_sectors(ns->queue, rounddown_pow_of_two(chunk_size));
1400 }
1401
1402 static void nvme_config_discard(struct nvme_ns *ns)
1403 {
1404         struct nvme_ctrl *ctrl = ns->ctrl;
1405         struct request_queue *queue = ns->queue;
1406         u32 size = queue_logical_block_size(queue);
1407
1408         if (!(ctrl->oncs & NVME_CTRL_ONCS_DSM)) {
1409                 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, queue);
1410                 return;
1411         }
1412
1413         if (ctrl->nr_streams && ns->sws && ns->sgs)
1414                 size *= ns->sws * ns->sgs;
1415
1416         BUILD_BUG_ON(PAGE_SIZE / sizeof(struct nvme_dsm_range) <
1417                         NVME_DSM_MAX_RANGES);
1418
1419         queue->limits.discard_alignment = 0;
1420         queue->limits.discard_granularity = size;
1421
1422         /* If discard is already enabled, don't reset queue limits */
1423         if (blk_queue_flag_test_and_set(QUEUE_FLAG_DISCARD, queue))
1424                 return;
1425
1426         blk_queue_max_discard_sectors(queue, UINT_MAX);
1427         blk_queue_max_discard_segments(queue, NVME_DSM_MAX_RANGES);
1428
1429         if (ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
1430                 blk_queue_max_write_zeroes_sectors(queue, UINT_MAX);
1431 }
1432
1433 static void nvme_report_ns_ids(struct nvme_ctrl *ctrl, unsigned int nsid,
1434                 struct nvme_id_ns *id, struct nvme_ns_ids *ids)
1435 {
1436         memset(ids, 0, sizeof(*ids));
1437
1438         if (ctrl->vs >= NVME_VS(1, 1, 0))
1439                 memcpy(ids->eui64, id->eui64, sizeof(id->eui64));
1440         if (ctrl->vs >= NVME_VS(1, 2, 0))
1441                 memcpy(ids->nguid, id->nguid, sizeof(id->nguid));
1442         if (ctrl->vs >= NVME_VS(1, 3, 0)) {
1443                  /* Don't treat error as fatal we potentially
1444                   * already have a NGUID or EUI-64
1445                   */
1446                 if (nvme_identify_ns_descs(ctrl, nsid, ids))
1447                         dev_warn(ctrl->device,
1448                                  "%s: Identify Descriptors failed\n", __func__);
1449         }
1450 }
1451
1452 static bool nvme_ns_ids_valid(struct nvme_ns_ids *ids)
1453 {
1454         return !uuid_is_null(&ids->uuid) ||
1455                 memchr_inv(ids->nguid, 0, sizeof(ids->nguid)) ||
1456                 memchr_inv(ids->eui64, 0, sizeof(ids->eui64));
1457 }
1458
1459 static bool nvme_ns_ids_equal(struct nvme_ns_ids *a, struct nvme_ns_ids *b)
1460 {
1461         return uuid_equal(&a->uuid, &b->uuid) &&
1462                 memcmp(&a->nguid, &b->nguid, sizeof(a->nguid)) == 0 &&
1463                 memcmp(&a->eui64, &b->eui64, sizeof(a->eui64)) == 0;
1464 }
1465
1466 static void nvme_update_disk_info(struct gendisk *disk,
1467                 struct nvme_ns *ns, struct nvme_id_ns *id)
1468 {
1469         sector_t capacity = le64_to_cpup(&id->nsze) << (ns->lba_shift - 9);
1470         unsigned short bs = 1 << ns->lba_shift;
1471
1472         blk_mq_freeze_queue(disk->queue);
1473         blk_integrity_unregister(disk);
1474
1475         blk_queue_logical_block_size(disk->queue, bs);
1476         blk_queue_physical_block_size(disk->queue, bs);
1477         blk_queue_io_min(disk->queue, bs);
1478
1479         if (ns->ms && !ns->ext &&
1480             (ns->ctrl->ops->flags & NVME_F_METADATA_SUPPORTED))
1481                 nvme_init_integrity(disk, ns->ms, ns->pi_type);
1482         if (ns->ms && !nvme_ns_has_pi(ns) && !blk_get_integrity(disk))
1483                 capacity = 0;
1484
1485         set_capacity(disk, capacity);
1486         nvme_config_discard(ns);
1487
1488         if (id->nsattr & (1 << 0))
1489                 set_disk_ro(disk, true);
1490         else
1491                 set_disk_ro(disk, false);
1492
1493         blk_mq_unfreeze_queue(disk->queue);
1494 }
1495
1496 static void __nvme_revalidate_disk(struct gendisk *disk, struct nvme_id_ns *id)
1497 {
1498         struct nvme_ns *ns = disk->private_data;
1499
1500         /*
1501          * If identify namespace failed, use default 512 byte block size so
1502          * block layer can use before failing read/write for 0 capacity.
1503          */
1504         ns->lba_shift = id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ds;
1505         if (ns->lba_shift == 0)
1506                 ns->lba_shift = 9;
1507         ns->noiob = le16_to_cpu(id->noiob);
1508         ns->ms = le16_to_cpu(id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ms);
1509         ns->ext = ns->ms && (id->flbas & NVME_NS_FLBAS_META_EXT);
1510         /* the PI implementation requires metadata equal t10 pi tuple size */
1511         if (ns->ms == sizeof(struct t10_pi_tuple))
1512                 ns->pi_type = id->dps & NVME_NS_DPS_PI_MASK;
1513         else
1514                 ns->pi_type = 0;
1515
1516         if (ns->noiob)
1517                 nvme_set_chunk_size(ns);
1518         nvme_update_disk_info(disk, ns, id);
1519         if (ns->ndev)
1520                 nvme_nvm_update_nvm_info(ns);
1521 #ifdef CONFIG_NVME_MULTIPATH
1522         if (ns->head->disk) {
1523                 nvme_update_disk_info(ns->head->disk, ns, id);
1524                 blk_queue_stack_limits(ns->head->disk->queue, ns->queue);
1525         }
1526 #endif
1527 }
1528
1529 static int nvme_revalidate_disk(struct gendisk *disk)
1530 {
1531         struct nvme_ns *ns = disk->private_data;
1532         struct nvme_ctrl *ctrl = ns->ctrl;
1533         struct nvme_id_ns *id;
1534         struct nvme_ns_ids ids;
1535         int ret = 0;
1536
1537         if (test_bit(NVME_NS_DEAD, &ns->flags)) {
1538                 set_capacity(disk, 0);
1539                 return -ENODEV;
1540         }
1541
1542         id = nvme_identify_ns(ctrl, ns->head->ns_id);
1543         if (!id)
1544                 return -ENODEV;
1545
1546         if (id->ncap == 0) {
1547                 ret = -ENODEV;
1548                 goto out;
1549         }
1550
1551         __nvme_revalidate_disk(disk, id);
1552         nvme_report_ns_ids(ctrl, ns->head->ns_id, id, &ids);
1553         if (!nvme_ns_ids_equal(&ns->head->ids, &ids)) {
1554                 dev_err(ctrl->device,
1555                         "identifiers changed for nsid %d\n", ns->head->ns_id);
1556                 ret = -ENODEV;
1557         }
1558
1559 out:
1560         kfree(id);
1561         return ret;
1562 }
1563
1564 static char nvme_pr_type(enum pr_type type)
1565 {
1566         switch (type) {
1567         case PR_WRITE_EXCLUSIVE:
1568                 return 1;
1569         case PR_EXCLUSIVE_ACCESS:
1570                 return 2;
1571         case PR_WRITE_EXCLUSIVE_REG_ONLY:
1572                 return 3;
1573         case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1574                 return 4;
1575         case PR_WRITE_EXCLUSIVE_ALL_REGS:
1576                 return 5;
1577         case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1578                 return 6;
1579         default:
1580                 return 0;
1581         }
1582 };
1583
1584 static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
1585                                 u64 key, u64 sa_key, u8 op)
1586 {
1587         struct nvme_ns_head *head = NULL;
1588         struct nvme_ns *ns;
1589         struct nvme_command c;
1590         int srcu_idx, ret;
1591         u8 data[16] = { 0, };
1592
1593         ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
1594         if (unlikely(!ns))
1595                 return -EWOULDBLOCK;
1596
1597         put_unaligned_le64(key, &data[0]);
1598         put_unaligned_le64(sa_key, &data[8]);
1599
1600         memset(&c, 0, sizeof(c));
1601         c.common.opcode = op;
1602         c.common.nsid = cpu_to_le32(ns->head->ns_id);
1603         c.common.cdw10[0] = cpu_to_le32(cdw10);
1604
1605         ret = nvme_submit_sync_cmd(ns->queue, &c, data, 16);
1606         nvme_put_ns_from_disk(head, srcu_idx);
1607         return ret;
1608 }
1609
1610 static int nvme_pr_register(struct block_device *bdev, u64 old,
1611                 u64 new, unsigned flags)
1612 {
1613         u32 cdw10;
1614
1615         if (flags & ~PR_FL_IGNORE_KEY)
1616                 return -EOPNOTSUPP;
1617
1618         cdw10 = old ? 2 : 0;
1619         cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
1620         cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
1621         return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
1622 }
1623
1624 static int nvme_pr_reserve(struct block_device *bdev, u64 key,
1625                 enum pr_type type, unsigned flags)
1626 {
1627         u32 cdw10;
1628
1629         if (flags & ~PR_FL_IGNORE_KEY)
1630                 return -EOPNOTSUPP;
1631
1632         cdw10 = nvme_pr_type(type) << 8;
1633         cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
1634         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
1635 }
1636
1637 static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
1638                 enum pr_type type, bool abort)
1639 {
1640         u32 cdw10 = nvme_pr_type(type) << 8 | (abort ? 2 : 1);
1641         return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
1642 }
1643
1644 static int nvme_pr_clear(struct block_device *bdev, u64 key)
1645 {
1646         u32 cdw10 = 1 | (key ? 1 << 3 : 0);
1647         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_register);
1648 }
1649
1650 static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1651 {
1652         u32 cdw10 = nvme_pr_type(type) << 8 | (key ? 1 << 3 : 0);
1653         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
1654 }
1655
1656 static const struct pr_ops nvme_pr_ops = {
1657         .pr_register    = nvme_pr_register,
1658         .pr_reserve     = nvme_pr_reserve,
1659         .pr_release     = nvme_pr_release,
1660         .pr_preempt     = nvme_pr_preempt,
1661         .pr_clear       = nvme_pr_clear,
1662 };
1663
1664 #ifdef CONFIG_BLK_SED_OPAL
1665 int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
1666                 bool send)
1667 {
1668         struct nvme_ctrl *ctrl = data;
1669         struct nvme_command cmd;
1670
1671         memset(&cmd, 0, sizeof(cmd));
1672         if (send)
1673                 cmd.common.opcode = nvme_admin_security_send;
1674         else
1675                 cmd.common.opcode = nvme_admin_security_recv;
1676         cmd.common.nsid = 0;
1677         cmd.common.cdw10[0] = cpu_to_le32(((u32)secp) << 24 | ((u32)spsp) << 8);
1678         cmd.common.cdw10[1] = cpu_to_le32(len);
1679
1680         return __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, buffer, len,
1681                                       ADMIN_TIMEOUT, NVME_QID_ANY, 1, 0);
1682 }
1683 EXPORT_SYMBOL_GPL(nvme_sec_submit);
1684 #endif /* CONFIG_BLK_SED_OPAL */
1685
1686 static const struct block_device_operations nvme_fops = {
1687         .owner          = THIS_MODULE,
1688         .ioctl          = nvme_ioctl,
1689         .compat_ioctl   = nvme_ioctl,
1690         .open           = nvme_open,
1691         .release        = nvme_release,
1692         .getgeo         = nvme_getgeo,
1693         .revalidate_disk= nvme_revalidate_disk,
1694         .pr_ops         = &nvme_pr_ops,
1695 };
1696
1697 #ifdef CONFIG_NVME_MULTIPATH
1698 static int nvme_ns_head_open(struct block_device *bdev, fmode_t mode)
1699 {
1700         struct nvme_ns_head *head = bdev->bd_disk->private_data;
1701
1702         if (!kref_get_unless_zero(&head->ref))
1703                 return -ENXIO;
1704         return 0;
1705 }
1706
1707 static void nvme_ns_head_release(struct gendisk *disk, fmode_t mode)
1708 {
1709         nvme_put_ns_head(disk->private_data);
1710 }
1711
1712 const struct block_device_operations nvme_ns_head_ops = {
1713         .owner          = THIS_MODULE,
1714         .open           = nvme_ns_head_open,
1715         .release        = nvme_ns_head_release,
1716         .ioctl          = nvme_ioctl,
1717         .compat_ioctl   = nvme_ioctl,
1718         .getgeo         = nvme_getgeo,
1719         .pr_ops         = &nvme_pr_ops,
1720 };
1721 #endif /* CONFIG_NVME_MULTIPATH */
1722
1723 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u64 cap, bool enabled)
1724 {
1725         unsigned long timeout =
1726                 ((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies;
1727         u32 csts, bit = enabled ? NVME_CSTS_RDY : 0;
1728         int ret;
1729
1730         while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
1731                 if (csts == ~0)
1732                         return -ENODEV;
1733                 if ((csts & NVME_CSTS_RDY) == bit)
1734                         break;
1735
1736                 msleep(100);
1737                 if (fatal_signal_pending(current))
1738                         return -EINTR;
1739                 if (time_after(jiffies, timeout)) {
1740                         dev_err(ctrl->device,
1741                                 "Device not ready; aborting %s\n", enabled ?
1742                                                 "initialisation" : "reset");
1743                         return -ENODEV;
1744                 }
1745         }
1746
1747         return ret;
1748 }
1749
1750 /*
1751  * If the device has been passed off to us in an enabled state, just clear
1752  * the enabled bit.  The spec says we should set the 'shutdown notification
1753  * bits', but doing so may cause the device to complete commands to the
1754  * admin queue ... and we don't know what memory that might be pointing at!
1755  */
1756 int nvme_disable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
1757 {
1758         int ret;
1759
1760         ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
1761         ctrl->ctrl_config &= ~NVME_CC_ENABLE;
1762
1763         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1764         if (ret)
1765                 return ret;
1766
1767         if (ctrl->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY)
1768                 msleep(NVME_QUIRK_DELAY_AMOUNT);
1769
1770         return nvme_wait_ready(ctrl, cap, false);
1771 }
1772 EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
1773
1774 int nvme_enable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
1775 {
1776         /*
1777          * Default to a 4K page size, with the intention to update this
1778          * path in the future to accomodate architectures with differing
1779          * kernel and IO page sizes.
1780          */
1781         unsigned dev_page_min = NVME_CAP_MPSMIN(cap) + 12, page_shift = 12;
1782         int ret;
1783
1784         if (page_shift < dev_page_min) {
1785                 dev_err(ctrl->device,
1786                         "Minimum device page size %u too large for host (%u)\n",
1787                         1 << dev_page_min, 1 << page_shift);
1788                 return -ENODEV;
1789         }
1790
1791         ctrl->page_size = 1 << page_shift;
1792
1793         ctrl->ctrl_config = NVME_CC_CSS_NVM;
1794         ctrl->ctrl_config |= (page_shift - 12) << NVME_CC_MPS_SHIFT;
1795         ctrl->ctrl_config |= NVME_CC_AMS_RR | NVME_CC_SHN_NONE;
1796         ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
1797         ctrl->ctrl_config |= NVME_CC_ENABLE;
1798
1799         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1800         if (ret)
1801                 return ret;
1802         return nvme_wait_ready(ctrl, cap, true);
1803 }
1804 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
1805
1806 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl)
1807 {
1808         unsigned long timeout = jiffies + (ctrl->shutdown_timeout * HZ);
1809         u32 csts;
1810         int ret;
1811
1812         ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
1813         ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
1814
1815         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1816         if (ret)
1817                 return ret;
1818
1819         while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
1820                 if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT)
1821                         break;
1822
1823                 msleep(100);
1824                 if (fatal_signal_pending(current))
1825                         return -EINTR;
1826                 if (time_after(jiffies, timeout)) {
1827                         dev_err(ctrl->device,
1828                                 "Device shutdown incomplete; abort shutdown\n");
1829                         return -ENODEV;
1830                 }
1831         }
1832
1833         return ret;
1834 }
1835 EXPORT_SYMBOL_GPL(nvme_shutdown_ctrl);
1836
1837 static void nvme_set_queue_limits(struct nvme_ctrl *ctrl,
1838                 struct request_queue *q)
1839 {
1840         bool vwc = false;
1841
1842         if (ctrl->max_hw_sectors) {
1843                 u32 max_segments =
1844                         (ctrl->max_hw_sectors / (ctrl->page_size >> 9)) + 1;
1845
1846                 max_segments = min_not_zero(max_segments, ctrl->max_segments);
1847                 blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors);
1848                 blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX));
1849         }
1850         if ((ctrl->quirks & NVME_QUIRK_STRIPE_SIZE) &&
1851             is_power_of_2(ctrl->max_hw_sectors))
1852                 blk_queue_chunk_sectors(q, ctrl->max_hw_sectors);
1853         blk_queue_virt_boundary(q, ctrl->page_size - 1);
1854         if (ctrl->vwc & NVME_CTRL_VWC_PRESENT)
1855                 vwc = true;
1856         blk_queue_write_cache(q, vwc, vwc);
1857 }
1858
1859 static int nvme_configure_timestamp(struct nvme_ctrl *ctrl)
1860 {
1861         __le64 ts;
1862         int ret;
1863
1864         if (!(ctrl->oncs & NVME_CTRL_ONCS_TIMESTAMP))
1865                 return 0;
1866
1867         ts = cpu_to_le64(ktime_to_ms(ktime_get_real()));
1868         ret = nvme_set_features(ctrl, NVME_FEAT_TIMESTAMP, 0, &ts, sizeof(ts),
1869                         NULL);
1870         if (ret)
1871                 dev_warn_once(ctrl->device,
1872                         "could not set timestamp (%d)\n", ret);
1873         return ret;
1874 }
1875
1876 static int nvme_configure_apst(struct nvme_ctrl *ctrl)
1877 {
1878         /*
1879          * APST (Autonomous Power State Transition) lets us program a
1880          * table of power state transitions that the controller will
1881          * perform automatically.  We configure it with a simple
1882          * heuristic: we are willing to spend at most 2% of the time
1883          * transitioning between power states.  Therefore, when running
1884          * in any given state, we will enter the next lower-power
1885          * non-operational state after waiting 50 * (enlat + exlat)
1886          * microseconds, as long as that state's exit latency is under
1887          * the requested maximum latency.
1888          *
1889          * We will not autonomously enter any non-operational state for
1890          * which the total latency exceeds ps_max_latency_us.  Users
1891          * can set ps_max_latency_us to zero to turn off APST.
1892          */
1893
1894         unsigned apste;
1895         struct nvme_feat_auto_pst *table;
1896         u64 max_lat_us = 0;
1897         int max_ps = -1;
1898         int ret;
1899
1900         /*
1901          * If APST isn't supported or if we haven't been initialized yet,
1902          * then don't do anything.
1903          */
1904         if (!ctrl->apsta)
1905                 return 0;
1906
1907         if (ctrl->npss > 31) {
1908                 dev_warn(ctrl->device, "NPSS is invalid; not using APST\n");
1909                 return 0;
1910         }
1911
1912         table = kzalloc(sizeof(*table), GFP_KERNEL);
1913         if (!table)
1914                 return 0;
1915
1916         if (!ctrl->apst_enabled || ctrl->ps_max_latency_us == 0) {
1917                 /* Turn off APST. */
1918                 apste = 0;
1919                 dev_dbg(ctrl->device, "APST disabled\n");
1920         } else {
1921                 __le64 target = cpu_to_le64(0);
1922                 int state;
1923
1924                 /*
1925                  * Walk through all states from lowest- to highest-power.
1926                  * According to the spec, lower-numbered states use more
1927                  * power.  NPSS, despite the name, is the index of the
1928                  * lowest-power state, not the number of states.
1929                  */
1930                 for (state = (int)ctrl->npss; state >= 0; state--) {
1931                         u64 total_latency_us, exit_latency_us, transition_ms;
1932
1933                         if (target)
1934                                 table->entries[state] = target;
1935
1936                         /*
1937                          * Don't allow transitions to the deepest state
1938                          * if it's quirked off.
1939                          */
1940                         if (state == ctrl->npss &&
1941                             (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS))
1942                                 continue;
1943
1944                         /*
1945                          * Is this state a useful non-operational state for
1946                          * higher-power states to autonomously transition to?
1947                          */
1948                         if (!(ctrl->psd[state].flags &
1949                               NVME_PS_FLAGS_NON_OP_STATE))
1950                                 continue;
1951
1952                         exit_latency_us =
1953                                 (u64)le32_to_cpu(ctrl->psd[state].exit_lat);
1954                         if (exit_latency_us > ctrl->ps_max_latency_us)
1955                                 continue;
1956
1957                         total_latency_us =
1958                                 exit_latency_us +
1959                                 le32_to_cpu(ctrl->psd[state].entry_lat);
1960
1961                         /*
1962                          * This state is good.  Use it as the APST idle
1963                          * target for higher power states.
1964                          */
1965                         transition_ms = total_latency_us + 19;
1966                         do_div(transition_ms, 20);
1967                         if (transition_ms > (1 << 24) - 1)
1968                                 transition_ms = (1 << 24) - 1;
1969
1970                         target = cpu_to_le64((state << 3) |
1971                                              (transition_ms << 8));
1972
1973                         if (max_ps == -1)
1974                                 max_ps = state;
1975
1976                         if (total_latency_us > max_lat_us)
1977                                 max_lat_us = total_latency_us;
1978                 }
1979
1980                 apste = 1;
1981
1982                 if (max_ps == -1) {
1983                         dev_dbg(ctrl->device, "APST enabled but no non-operational states are available\n");
1984                 } else {
1985                         dev_dbg(ctrl->device, "APST enabled: max PS = %d, max round-trip latency = %lluus, table = %*phN\n",
1986                                 max_ps, max_lat_us, (int)sizeof(*table), table);
1987                 }
1988         }
1989
1990         ret = nvme_set_features(ctrl, NVME_FEAT_AUTO_PST, apste,
1991                                 table, sizeof(*table), NULL);
1992         if (ret)
1993                 dev_err(ctrl->device, "failed to set APST feature (%d)\n", ret);
1994
1995         kfree(table);
1996         return ret;
1997 }
1998
1999 static void nvme_set_latency_tolerance(struct device *dev, s32 val)
2000 {
2001         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2002         u64 latency;
2003
2004         switch (val) {
2005         case PM_QOS_LATENCY_TOLERANCE_NO_CONSTRAINT:
2006         case PM_QOS_LATENCY_ANY:
2007                 latency = U64_MAX;
2008                 break;
2009
2010         default:
2011                 latency = val;
2012         }
2013
2014         if (ctrl->ps_max_latency_us != latency) {
2015                 ctrl->ps_max_latency_us = latency;
2016                 nvme_configure_apst(ctrl);
2017         }
2018 }
2019
2020 struct nvme_core_quirk_entry {
2021         /*
2022          * NVMe model and firmware strings are padded with spaces.  For
2023          * simplicity, strings in the quirk table are padded with NULLs
2024          * instead.
2025          */
2026         u16 vid;
2027         const char *mn;
2028         const char *fr;
2029         unsigned long quirks;
2030 };
2031
2032 static const struct nvme_core_quirk_entry core_quirks[] = {
2033         {
2034                 /*
2035                  * This Toshiba device seems to die using any APST states.  See:
2036                  * https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1678184/comments/11
2037                  */
2038                 .vid = 0x1179,
2039                 .mn = "THNSF5256GPUK TOSHIBA",
2040                 .quirks = NVME_QUIRK_NO_APST,
2041         }
2042 };
2043
2044 /* match is null-terminated but idstr is space-padded. */
2045 static bool string_matches(const char *idstr, const char *match, size_t len)
2046 {
2047         size_t matchlen;
2048
2049         if (!match)
2050                 return true;
2051
2052         matchlen = strlen(match);
2053         WARN_ON_ONCE(matchlen > len);
2054
2055         if (memcmp(idstr, match, matchlen))
2056                 return false;
2057
2058         for (; matchlen < len; matchlen++)
2059                 if (idstr[matchlen] != ' ')
2060                         return false;
2061
2062         return true;
2063 }
2064
2065 static bool quirk_matches(const struct nvme_id_ctrl *id,
2066                           const struct nvme_core_quirk_entry *q)
2067 {
2068         return q->vid == le16_to_cpu(id->vid) &&
2069                 string_matches(id->mn, q->mn, sizeof(id->mn)) &&
2070                 string_matches(id->fr, q->fr, sizeof(id->fr));
2071 }
2072
2073 static void nvme_init_subnqn(struct nvme_subsystem *subsys, struct nvme_ctrl *ctrl,
2074                 struct nvme_id_ctrl *id)
2075 {
2076         size_t nqnlen;
2077         int off;
2078
2079         nqnlen = strnlen(id->subnqn, NVMF_NQN_SIZE);
2080         if (nqnlen > 0 && nqnlen < NVMF_NQN_SIZE) {
2081                 strlcpy(subsys->subnqn, id->subnqn, NVMF_NQN_SIZE);
2082                 return;
2083         }
2084
2085         if (ctrl->vs >= NVME_VS(1, 2, 1))
2086                 dev_warn(ctrl->device, "missing or invalid SUBNQN field.\n");
2087
2088         /* Generate a "fake" NQN per Figure 254 in NVMe 1.3 + ECN 001 */
2089         off = snprintf(subsys->subnqn, NVMF_NQN_SIZE,
2090                         "nqn.2014.08.org.nvmexpress:%4x%4x",
2091                         le16_to_cpu(id->vid), le16_to_cpu(id->ssvid));
2092         memcpy(subsys->subnqn + off, id->sn, sizeof(id->sn));
2093         off += sizeof(id->sn);
2094         memcpy(subsys->subnqn + off, id->mn, sizeof(id->mn));
2095         off += sizeof(id->mn);
2096         memset(subsys->subnqn + off, 0, sizeof(subsys->subnqn) - off);
2097 }
2098
2099 static void __nvme_release_subsystem(struct nvme_subsystem *subsys)
2100 {
2101         ida_simple_remove(&nvme_subsystems_ida, subsys->instance);
2102         kfree(subsys);
2103 }
2104
2105 static void nvme_release_subsystem(struct device *dev)
2106 {
2107         __nvme_release_subsystem(container_of(dev, struct nvme_subsystem, dev));
2108 }
2109
2110 static void nvme_destroy_subsystem(struct kref *ref)
2111 {
2112         struct nvme_subsystem *subsys =
2113                         container_of(ref, struct nvme_subsystem, ref);
2114
2115         mutex_lock(&nvme_subsystems_lock);
2116         list_del(&subsys->entry);
2117         mutex_unlock(&nvme_subsystems_lock);
2118
2119         ida_destroy(&subsys->ns_ida);
2120         device_del(&subsys->dev);
2121         put_device(&subsys->dev);
2122 }
2123
2124 static void nvme_put_subsystem(struct nvme_subsystem *subsys)
2125 {
2126         kref_put(&subsys->ref, nvme_destroy_subsystem);
2127 }
2128
2129 static struct nvme_subsystem *__nvme_find_get_subsystem(const char *subsysnqn)
2130 {
2131         struct nvme_subsystem *subsys;
2132
2133         lockdep_assert_held(&nvme_subsystems_lock);
2134
2135         list_for_each_entry(subsys, &nvme_subsystems, entry) {
2136                 if (strcmp(subsys->subnqn, subsysnqn))
2137                         continue;
2138                 if (!kref_get_unless_zero(&subsys->ref))
2139                         continue;
2140                 return subsys;
2141         }
2142
2143         return NULL;
2144 }
2145
2146 #define SUBSYS_ATTR_RO(_name, _mode, _show)                     \
2147         struct device_attribute subsys_attr_##_name = \
2148                 __ATTR(_name, _mode, _show, NULL)
2149
2150 static ssize_t nvme_subsys_show_nqn(struct device *dev,
2151                                     struct device_attribute *attr,
2152                                     char *buf)
2153 {
2154         struct nvme_subsystem *subsys =
2155                 container_of(dev, struct nvme_subsystem, dev);
2156
2157         return snprintf(buf, PAGE_SIZE, "%s\n", subsys->subnqn);
2158 }
2159 static SUBSYS_ATTR_RO(subsysnqn, S_IRUGO, nvme_subsys_show_nqn);
2160
2161 #define nvme_subsys_show_str_function(field)                            \
2162 static ssize_t subsys_##field##_show(struct device *dev,                \
2163                             struct device_attribute *attr, char *buf)   \
2164 {                                                                       \
2165         struct nvme_subsystem *subsys =                                 \
2166                 container_of(dev, struct nvme_subsystem, dev);          \
2167         return sprintf(buf, "%.*s\n",                                   \
2168                        (int)sizeof(subsys->field), subsys->field);      \
2169 }                                                                       \
2170 static SUBSYS_ATTR_RO(field, S_IRUGO, subsys_##field##_show);
2171
2172 nvme_subsys_show_str_function(model);
2173 nvme_subsys_show_str_function(serial);
2174 nvme_subsys_show_str_function(firmware_rev);
2175
2176 static struct attribute *nvme_subsys_attrs[] = {
2177         &subsys_attr_model.attr,
2178         &subsys_attr_serial.attr,
2179         &subsys_attr_firmware_rev.attr,
2180         &subsys_attr_subsysnqn.attr,
2181         NULL,
2182 };
2183
2184 static struct attribute_group nvme_subsys_attrs_group = {
2185         .attrs = nvme_subsys_attrs,
2186 };
2187
2188 static const struct attribute_group *nvme_subsys_attrs_groups[] = {
2189         &nvme_subsys_attrs_group,
2190         NULL,
2191 };
2192
2193 static int nvme_active_ctrls(struct nvme_subsystem *subsys)
2194 {
2195         int count = 0;
2196         struct nvme_ctrl *ctrl;
2197
2198         mutex_lock(&subsys->lock);
2199         list_for_each_entry(ctrl, &subsys->ctrls, subsys_entry) {
2200                 if (ctrl->state != NVME_CTRL_DELETING &&
2201                     ctrl->state != NVME_CTRL_DEAD)
2202                         count++;
2203         }
2204         mutex_unlock(&subsys->lock);
2205
2206         return count;
2207 }
2208
2209 static int nvme_init_subsystem(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2210 {
2211         struct nvme_subsystem *subsys, *found;
2212         int ret;
2213
2214         subsys = kzalloc(sizeof(*subsys), GFP_KERNEL);
2215         if (!subsys)
2216                 return -ENOMEM;
2217         ret = ida_simple_get(&nvme_subsystems_ida, 0, 0, GFP_KERNEL);
2218         if (ret < 0) {
2219                 kfree(subsys);
2220                 return ret;
2221         }
2222         subsys->instance = ret;
2223         mutex_init(&subsys->lock);
2224         kref_init(&subsys->ref);
2225         INIT_LIST_HEAD(&subsys->ctrls);
2226         INIT_LIST_HEAD(&subsys->nsheads);
2227         nvme_init_subnqn(subsys, ctrl, id);
2228         memcpy(subsys->serial, id->sn, sizeof(subsys->serial));
2229         memcpy(subsys->model, id->mn, sizeof(subsys->model));
2230         memcpy(subsys->firmware_rev, id->fr, sizeof(subsys->firmware_rev));
2231         subsys->vendor_id = le16_to_cpu(id->vid);
2232         subsys->cmic = id->cmic;
2233
2234         subsys->dev.class = nvme_subsys_class;
2235         subsys->dev.release = nvme_release_subsystem;
2236         subsys->dev.groups = nvme_subsys_attrs_groups;
2237         dev_set_name(&subsys->dev, "nvme-subsys%d", subsys->instance);
2238         device_initialize(&subsys->dev);
2239
2240         mutex_lock(&nvme_subsystems_lock);
2241         found = __nvme_find_get_subsystem(subsys->subnqn);
2242         if (found) {
2243                 /*
2244                  * Verify that the subsystem actually supports multiple
2245                  * controllers, else bail out.
2246                  */
2247                 if (!(ctrl->opts && ctrl->opts->discovery_nqn) &&
2248                     nvme_active_ctrls(found) && !(id->cmic & (1 << 1))) {
2249                         dev_err(ctrl->device,
2250                                 "ignoring ctrl due to duplicate subnqn (%s).\n",
2251                                 found->subnqn);
2252                         nvme_put_subsystem(found);
2253                         ret = -EINVAL;
2254                         goto out_unlock;
2255                 }
2256
2257                 __nvme_release_subsystem(subsys);
2258                 subsys = found;
2259         } else {
2260                 ret = device_add(&subsys->dev);
2261                 if (ret) {
2262                         dev_err(ctrl->device,
2263                                 "failed to register subsystem device.\n");
2264                         goto out_unlock;
2265                 }
2266                 ida_init(&subsys->ns_ida);
2267                 list_add_tail(&subsys->entry, &nvme_subsystems);
2268         }
2269
2270         ctrl->subsys = subsys;
2271         mutex_unlock(&nvme_subsystems_lock);
2272
2273         if (sysfs_create_link(&subsys->dev.kobj, &ctrl->device->kobj,
2274                         dev_name(ctrl->device))) {
2275                 dev_err(ctrl->device,
2276                         "failed to create sysfs link from subsystem.\n");
2277                 /* the transport driver will eventually put the subsystem */
2278                 return -EINVAL;
2279         }
2280
2281         mutex_lock(&subsys->lock);
2282         list_add_tail(&ctrl->subsys_entry, &subsys->ctrls);
2283         mutex_unlock(&subsys->lock);
2284
2285         return 0;
2286
2287 out_unlock:
2288         mutex_unlock(&nvme_subsystems_lock);
2289         put_device(&subsys->dev);
2290         return ret;
2291 }
2292
2293 int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp,
2294                 void *log, size_t size, u64 offset)
2295 {
2296         struct nvme_command c = { };
2297         unsigned long dwlen = size / 4 - 1;
2298
2299         c.get_log_page.opcode = nvme_admin_get_log_page;
2300         c.get_log_page.nsid = cpu_to_le32(nsid);
2301         c.get_log_page.lid = log_page;
2302         c.get_log_page.lsp = lsp;
2303         c.get_log_page.numdl = cpu_to_le16(dwlen & ((1 << 16) - 1));
2304         c.get_log_page.numdu = cpu_to_le16(dwlen >> 16);
2305         c.get_log_page.lpol = cpu_to_le32(lower_32_bits(offset));
2306         c.get_log_page.lpou = cpu_to_le32(upper_32_bits(offset));
2307
2308         return nvme_submit_sync_cmd(ctrl->admin_q, &c, log, size);
2309 }
2310
2311 static int nvme_get_effects_log(struct nvme_ctrl *ctrl)
2312 {
2313         int ret;
2314
2315         if (!ctrl->effects)
2316                 ctrl->effects = kzalloc(sizeof(*ctrl->effects), GFP_KERNEL);
2317
2318         if (!ctrl->effects)
2319                 return 0;
2320
2321         ret = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CMD_EFFECTS, 0,
2322                         ctrl->effects, sizeof(*ctrl->effects), 0);
2323         if (ret) {
2324                 kfree(ctrl->effects);
2325                 ctrl->effects = NULL;
2326         }
2327         return ret;
2328 }
2329
2330 /*
2331  * Initialize the cached copies of the Identify data and various controller
2332  * register in our nvme_ctrl structure.  This should be called as soon as
2333  * the admin queue is fully up and running.
2334  */
2335 int nvme_init_identify(struct nvme_ctrl *ctrl)
2336 {
2337         struct nvme_id_ctrl *id;
2338         u64 cap;
2339         int ret, page_shift;
2340         u32 max_hw_sectors;
2341         bool prev_apst_enabled;
2342
2343         ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
2344         if (ret) {
2345                 dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
2346                 return ret;
2347         }
2348
2349         ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &cap);
2350         if (ret) {
2351                 dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
2352                 return ret;
2353         }
2354         page_shift = NVME_CAP_MPSMIN(cap) + 12;
2355
2356         if (ctrl->vs >= NVME_VS(1, 1, 0))
2357                 ctrl->subsystem = NVME_CAP_NSSRC(cap);
2358
2359         ret = nvme_identify_ctrl(ctrl, &id);
2360         if (ret) {
2361                 dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
2362                 return -EIO;
2363         }
2364
2365         if (id->lpa & NVME_CTRL_LPA_CMD_EFFECTS_LOG) {
2366                 ret = nvme_get_effects_log(ctrl);
2367                 if (ret < 0)
2368                         goto out_free;
2369         }
2370
2371         if (!ctrl->identified) {
2372                 int i;
2373
2374                 ret = nvme_init_subsystem(ctrl, id);
2375                 if (ret)
2376                         goto out_free;
2377
2378                 /*
2379                  * Check for quirks.  Quirk can depend on firmware version,
2380                  * so, in principle, the set of quirks present can change
2381                  * across a reset.  As a possible future enhancement, we
2382                  * could re-scan for quirks every time we reinitialize
2383                  * the device, but we'd have to make sure that the driver
2384                  * behaves intelligently if the quirks change.
2385                  */
2386                 for (i = 0; i < ARRAY_SIZE(core_quirks); i++) {
2387                         if (quirk_matches(id, &core_quirks[i]))
2388                                 ctrl->quirks |= core_quirks[i].quirks;
2389                 }
2390         }
2391
2392         if (force_apst && (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS)) {
2393                 dev_warn(ctrl->device, "forcibly allowing all power states due to nvme_core.force_apst -- use at your own risk\n");
2394                 ctrl->quirks &= ~NVME_QUIRK_NO_DEEPEST_PS;
2395         }
2396
2397         ctrl->oacs = le16_to_cpu(id->oacs);
2398         ctrl->oncs = le16_to_cpup(&id->oncs);
2399         ctrl->oaes = le32_to_cpu(id->oaes);
2400         atomic_set(&ctrl->abort_limit, id->acl + 1);
2401         ctrl->vwc = id->vwc;
2402         ctrl->cntlid = le16_to_cpup(&id->cntlid);
2403         if (id->mdts)
2404                 max_hw_sectors = 1 << (id->mdts + page_shift - 9);
2405         else
2406                 max_hw_sectors = UINT_MAX;
2407         ctrl->max_hw_sectors =
2408                 min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
2409
2410         nvme_set_queue_limits(ctrl, ctrl->admin_q);
2411         ctrl->sgls = le32_to_cpu(id->sgls);
2412         ctrl->kas = le16_to_cpu(id->kas);
2413         ctrl->max_namespaces = le32_to_cpu(id->mnan);
2414
2415         if (id->rtd3e) {
2416                 /* us -> s */
2417                 u32 transition_time = le32_to_cpu(id->rtd3e) / 1000000;
2418
2419                 ctrl->shutdown_timeout = clamp_t(unsigned int, transition_time,
2420                                                  shutdown_timeout, 60);
2421
2422                 if (ctrl->shutdown_timeout != shutdown_timeout)
2423                         dev_info(ctrl->device,
2424                                  "Shutdown timeout set to %u seconds\n",
2425                                  ctrl->shutdown_timeout);
2426         } else
2427                 ctrl->shutdown_timeout = shutdown_timeout;
2428
2429         ctrl->npss = id->npss;
2430         ctrl->apsta = id->apsta;
2431         prev_apst_enabled = ctrl->apst_enabled;
2432         if (ctrl->quirks & NVME_QUIRK_NO_APST) {
2433                 if (force_apst && id->apsta) {
2434                         dev_warn(ctrl->device, "forcibly allowing APST due to nvme_core.force_apst -- use at your own risk\n");
2435                         ctrl->apst_enabled = true;
2436                 } else {
2437                         ctrl->apst_enabled = false;
2438                 }
2439         } else {
2440                 ctrl->apst_enabled = id->apsta;
2441         }
2442         memcpy(ctrl->psd, id->psd, sizeof(ctrl->psd));
2443
2444         if (ctrl->ops->flags & NVME_F_FABRICS) {
2445                 ctrl->icdoff = le16_to_cpu(id->icdoff);
2446                 ctrl->ioccsz = le32_to_cpu(id->ioccsz);
2447                 ctrl->iorcsz = le32_to_cpu(id->iorcsz);
2448                 ctrl->maxcmd = le16_to_cpu(id->maxcmd);
2449
2450                 /*
2451                  * In fabrics we need to verify the cntlid matches the
2452                  * admin connect
2453                  */
2454                 if (ctrl->cntlid != le16_to_cpu(id->cntlid)) {
2455                         ret = -EINVAL;
2456                         goto out_free;
2457                 }
2458
2459                 if (!ctrl->opts->discovery_nqn && !ctrl->kas) {
2460                         dev_err(ctrl->device,
2461                                 "keep-alive support is mandatory for fabrics\n");
2462                         ret = -EINVAL;
2463                         goto out_free;
2464                 }
2465         } else {
2466                 ctrl->cntlid = le16_to_cpu(id->cntlid);
2467                 ctrl->hmpre = le32_to_cpu(id->hmpre);
2468                 ctrl->hmmin = le32_to_cpu(id->hmmin);
2469                 ctrl->hmminds = le32_to_cpu(id->hmminds);
2470                 ctrl->hmmaxd = le16_to_cpu(id->hmmaxd);
2471         }
2472
2473         ret = nvme_mpath_init(ctrl, id);
2474         kfree(id);
2475
2476         if (ret < 0)
2477                 return ret;
2478
2479         if (ctrl->apst_enabled && !prev_apst_enabled)
2480                 dev_pm_qos_expose_latency_tolerance(ctrl->device);
2481         else if (!ctrl->apst_enabled && prev_apst_enabled)
2482                 dev_pm_qos_hide_latency_tolerance(ctrl->device);
2483
2484         ret = nvme_configure_apst(ctrl);
2485         if (ret < 0)
2486                 return ret;
2487
2488         ret = nvme_configure_timestamp(ctrl);
2489         if (ret < 0)
2490                 return ret;
2491
2492         ret = nvme_configure_directives(ctrl);
2493         if (ret < 0)
2494                 return ret;
2495
2496         ctrl->identified = true;
2497
2498         return 0;
2499
2500 out_free:
2501         kfree(id);
2502         return ret;
2503 }
2504 EXPORT_SYMBOL_GPL(nvme_init_identify);
2505
2506 static int nvme_dev_open(struct inode *inode, struct file *file)
2507 {
2508         struct nvme_ctrl *ctrl =
2509                 container_of(inode->i_cdev, struct nvme_ctrl, cdev);
2510
2511         switch (ctrl->state) {
2512         case NVME_CTRL_LIVE:
2513         case NVME_CTRL_ADMIN_ONLY:
2514                 break;
2515         default:
2516                 return -EWOULDBLOCK;
2517         }
2518
2519         file->private_data = ctrl;
2520         return 0;
2521 }
2522
2523 static int nvme_dev_user_cmd(struct nvme_ctrl *ctrl, void __user *argp)
2524 {
2525         struct nvme_ns *ns;
2526         int ret;
2527
2528         down_read(&ctrl->namespaces_rwsem);
2529         if (list_empty(&ctrl->namespaces)) {
2530                 ret = -ENOTTY;
2531                 goto out_unlock;
2532         }
2533
2534         ns = list_first_entry(&ctrl->namespaces, struct nvme_ns, list);
2535         if (ns != list_last_entry(&ctrl->namespaces, struct nvme_ns, list)) {
2536                 dev_warn(ctrl->device,
2537                         "NVME_IOCTL_IO_CMD not supported when multiple namespaces present!\n");
2538                 ret = -EINVAL;
2539                 goto out_unlock;
2540         }
2541
2542         dev_warn(ctrl->device,
2543                 "using deprecated NVME_IOCTL_IO_CMD ioctl on the char device!\n");
2544         kref_get(&ns->kref);
2545         up_read(&ctrl->namespaces_rwsem);
2546
2547         ret = nvme_user_cmd(ctrl, ns, argp);
2548         nvme_put_ns(ns);
2549         return ret;
2550
2551 out_unlock:
2552         up_read(&ctrl->namespaces_rwsem);
2553         return ret;
2554 }
2555
2556 static long nvme_dev_ioctl(struct file *file, unsigned int cmd,
2557                 unsigned long arg)
2558 {
2559         struct nvme_ctrl *ctrl = file->private_data;
2560         void __user *argp = (void __user *)arg;
2561
2562         switch (cmd) {
2563         case NVME_IOCTL_ADMIN_CMD:
2564                 return nvme_user_cmd(ctrl, NULL, argp);
2565         case NVME_IOCTL_IO_CMD:
2566                 return nvme_dev_user_cmd(ctrl, argp);
2567         case NVME_IOCTL_RESET:
2568                 dev_warn(ctrl->device, "resetting controller\n");
2569                 return nvme_reset_ctrl_sync(ctrl);
2570         case NVME_IOCTL_SUBSYS_RESET:
2571                 return nvme_reset_subsystem(ctrl);
2572         case NVME_IOCTL_RESCAN:
2573                 nvme_queue_scan(ctrl);
2574                 return 0;
2575         default:
2576                 return -ENOTTY;
2577         }
2578 }
2579
2580 static const struct file_operations nvme_dev_fops = {
2581         .owner          = THIS_MODULE,
2582         .open           = nvme_dev_open,
2583         .unlocked_ioctl = nvme_dev_ioctl,
2584         .compat_ioctl   = nvme_dev_ioctl,
2585 };
2586
2587 static ssize_t nvme_sysfs_reset(struct device *dev,
2588                                 struct device_attribute *attr, const char *buf,
2589                                 size_t count)
2590 {
2591         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2592         int ret;
2593
2594         ret = nvme_reset_ctrl_sync(ctrl);
2595         if (ret < 0)
2596                 return ret;
2597         return count;
2598 }
2599 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
2600
2601 static ssize_t nvme_sysfs_rescan(struct device *dev,
2602                                 struct device_attribute *attr, const char *buf,
2603                                 size_t count)
2604 {
2605         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2606
2607         nvme_queue_scan(ctrl);
2608         return count;
2609 }
2610 static DEVICE_ATTR(rescan_controller, S_IWUSR, NULL, nvme_sysfs_rescan);
2611
2612 static inline struct nvme_ns_head *dev_to_ns_head(struct device *dev)
2613 {
2614         struct gendisk *disk = dev_to_disk(dev);
2615
2616         if (disk->fops == &nvme_fops)
2617                 return nvme_get_ns_from_dev(dev)->head;
2618         else
2619                 return disk->private_data;
2620 }
2621
2622 static ssize_t wwid_show(struct device *dev, struct device_attribute *attr,
2623                 char *buf)
2624 {
2625         struct nvme_ns_head *head = dev_to_ns_head(dev);
2626         struct nvme_ns_ids *ids = &head->ids;
2627         struct nvme_subsystem *subsys = head->subsys;
2628         int serial_len = sizeof(subsys->serial);
2629         int model_len = sizeof(subsys->model);
2630
2631         if (!uuid_is_null(&ids->uuid))
2632                 return sprintf(buf, "uuid.%pU\n", &ids->uuid);
2633
2634         if (memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
2635                 return sprintf(buf, "eui.%16phN\n", ids->nguid);
2636
2637         if (memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
2638                 return sprintf(buf, "eui.%8phN\n", ids->eui64);
2639
2640         while (serial_len > 0 && (subsys->serial[serial_len - 1] == ' ' ||
2641                                   subsys->serial[serial_len - 1] == '\0'))
2642                 serial_len--;
2643         while (model_len > 0 && (subsys->model[model_len - 1] == ' ' ||
2644                                  subsys->model[model_len - 1] == '\0'))
2645                 model_len--;
2646
2647         return sprintf(buf, "nvme.%04x-%*phN-%*phN-%08x\n", subsys->vendor_id,
2648                 serial_len, subsys->serial, model_len, subsys->model,
2649                 head->ns_id);
2650 }
2651 static DEVICE_ATTR_RO(wwid);
2652
2653 static ssize_t nguid_show(struct device *dev, struct device_attribute *attr,
2654                 char *buf)
2655 {
2656         return sprintf(buf, "%pU\n", dev_to_ns_head(dev)->ids.nguid);
2657 }
2658 static DEVICE_ATTR_RO(nguid);
2659
2660 static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
2661                 char *buf)
2662 {
2663         struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
2664
2665         /* For backward compatibility expose the NGUID to userspace if
2666          * we have no UUID set
2667          */
2668         if (uuid_is_null(&ids->uuid)) {
2669                 printk_ratelimited(KERN_WARNING
2670                                    "No UUID available providing old NGUID\n");
2671                 return sprintf(buf, "%pU\n", ids->nguid);
2672         }
2673         return sprintf(buf, "%pU\n", &ids->uuid);
2674 }
2675 static DEVICE_ATTR_RO(uuid);
2676
2677 static ssize_t eui_show(struct device *dev, struct device_attribute *attr,
2678                 char *buf)
2679 {
2680         return sprintf(buf, "%8ph\n", dev_to_ns_head(dev)->ids.eui64);
2681 }
2682 static DEVICE_ATTR_RO(eui);
2683
2684 static ssize_t nsid_show(struct device *dev, struct device_attribute *attr,
2685                 char *buf)
2686 {
2687         return sprintf(buf, "%d\n", dev_to_ns_head(dev)->ns_id);
2688 }
2689 static DEVICE_ATTR_RO(nsid);
2690
2691 static struct attribute *nvme_ns_id_attrs[] = {
2692         &dev_attr_wwid.attr,
2693         &dev_attr_uuid.attr,
2694         &dev_attr_nguid.attr,
2695         &dev_attr_eui.attr,
2696         &dev_attr_nsid.attr,
2697 #ifdef CONFIG_NVME_MULTIPATH
2698         &dev_attr_ana_grpid.attr,
2699         &dev_attr_ana_state.attr,
2700 #endif
2701         NULL,
2702 };
2703
2704 static umode_t nvme_ns_id_attrs_are_visible(struct kobject *kobj,
2705                 struct attribute *a, int n)
2706 {
2707         struct device *dev = container_of(kobj, struct device, kobj);
2708         struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
2709
2710         if (a == &dev_attr_uuid.attr) {
2711                 if (uuid_is_null(&ids->uuid) &&
2712                     !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
2713                         return 0;
2714         }
2715         if (a == &dev_attr_nguid.attr) {
2716                 if (!memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
2717                         return 0;
2718         }
2719         if (a == &dev_attr_eui.attr) {
2720                 if (!memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
2721                         return 0;
2722         }
2723 #ifdef CONFIG_NVME_MULTIPATH
2724         if (a == &dev_attr_ana_grpid.attr || a == &dev_attr_ana_state.attr) {
2725                 if (dev_to_disk(dev)->fops != &nvme_fops) /* per-path attr */
2726                         return 0;
2727                 if (!nvme_ctrl_use_ana(nvme_get_ns_from_dev(dev)->ctrl))
2728                         return 0;
2729         }
2730 #endif
2731         return a->mode;
2732 }
2733
2734 static const struct attribute_group nvme_ns_id_attr_group = {
2735         .attrs          = nvme_ns_id_attrs,
2736         .is_visible     = nvme_ns_id_attrs_are_visible,
2737 };
2738
2739 const struct attribute_group *nvme_ns_id_attr_groups[] = {
2740         &nvme_ns_id_attr_group,
2741 #ifdef CONFIG_NVM
2742         &nvme_nvm_attr_group,
2743 #endif
2744         NULL,
2745 };
2746
2747 #define nvme_show_str_function(field)                                           \
2748 static ssize_t  field##_show(struct device *dev,                                \
2749                             struct device_attribute *attr, char *buf)           \
2750 {                                                                               \
2751         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);                          \
2752         return sprintf(buf, "%.*s\n",                                           \
2753                 (int)sizeof(ctrl->subsys->field), ctrl->subsys->field);         \
2754 }                                                                               \
2755 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
2756
2757 nvme_show_str_function(model);
2758 nvme_show_str_function(serial);
2759 nvme_show_str_function(firmware_rev);
2760
2761 #define nvme_show_int_function(field)                                           \
2762 static ssize_t  field##_show(struct device *dev,                                \
2763                             struct device_attribute *attr, char *buf)           \
2764 {                                                                               \
2765         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);                          \
2766         return sprintf(buf, "%d\n", ctrl->field);       \
2767 }                                                                               \
2768 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
2769
2770 nvme_show_int_function(cntlid);
2771
2772 static ssize_t nvme_sysfs_delete(struct device *dev,
2773                                 struct device_attribute *attr, const char *buf,
2774                                 size_t count)
2775 {
2776         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2777
2778         if (device_remove_file_self(dev, attr))
2779                 nvme_delete_ctrl_sync(ctrl);
2780         return count;
2781 }
2782 static DEVICE_ATTR(delete_controller, S_IWUSR, NULL, nvme_sysfs_delete);
2783
2784 static ssize_t nvme_sysfs_show_transport(struct device *dev,
2785                                          struct device_attribute *attr,
2786                                          char *buf)
2787 {
2788         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2789
2790         return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->ops->name);
2791 }
2792 static DEVICE_ATTR(transport, S_IRUGO, nvme_sysfs_show_transport, NULL);
2793
2794 static ssize_t nvme_sysfs_show_state(struct device *dev,
2795                                      struct device_attribute *attr,
2796                                      char *buf)
2797 {
2798         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2799         static const char *const state_name[] = {
2800                 [NVME_CTRL_NEW]         = "new",
2801                 [NVME_CTRL_LIVE]        = "live",
2802                 [NVME_CTRL_ADMIN_ONLY]  = "only-admin",
2803                 [NVME_CTRL_RESETTING]   = "resetting",
2804                 [NVME_CTRL_CONNECTING]  = "connecting",
2805                 [NVME_CTRL_DELETING]    = "deleting",
2806                 [NVME_CTRL_DEAD]        = "dead",
2807         };
2808
2809         if ((unsigned)ctrl->state < ARRAY_SIZE(state_name) &&
2810             state_name[ctrl->state])
2811                 return sprintf(buf, "%s\n", state_name[ctrl->state]);
2812
2813         return sprintf(buf, "unknown state\n");
2814 }
2815
2816 static DEVICE_ATTR(state, S_IRUGO, nvme_sysfs_show_state, NULL);
2817
2818 static ssize_t nvme_sysfs_show_subsysnqn(struct device *dev,
2819                                          struct device_attribute *attr,
2820                                          char *buf)
2821 {
2822         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2823
2824         return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->subsys->subnqn);
2825 }
2826 static DEVICE_ATTR(subsysnqn, S_IRUGO, nvme_sysfs_show_subsysnqn, NULL);
2827
2828 static ssize_t nvme_sysfs_show_address(struct device *dev,
2829                                          struct device_attribute *attr,
2830                                          char *buf)
2831 {
2832         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2833
2834         return ctrl->ops->get_address(ctrl, buf, PAGE_SIZE);
2835 }
2836 static DEVICE_ATTR(address, S_IRUGO, nvme_sysfs_show_address, NULL);
2837
2838 static struct attribute *nvme_dev_attrs[] = {
2839         &dev_attr_reset_controller.attr,
2840         &dev_attr_rescan_controller.attr,
2841         &dev_attr_model.attr,
2842         &dev_attr_serial.attr,
2843         &dev_attr_firmware_rev.attr,
2844         &dev_attr_cntlid.attr,
2845         &dev_attr_delete_controller.attr,
2846         &dev_attr_transport.attr,
2847         &dev_attr_subsysnqn.attr,
2848         &dev_attr_address.attr,
2849         &dev_attr_state.attr,
2850         NULL
2851 };
2852
2853 static umode_t nvme_dev_attrs_are_visible(struct kobject *kobj,
2854                 struct attribute *a, int n)
2855 {
2856         struct device *dev = container_of(kobj, struct device, kobj);
2857         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2858
2859         if (a == &dev_attr_delete_controller.attr && !ctrl->ops->delete_ctrl)
2860                 return 0;
2861         if (a == &dev_attr_address.attr && !ctrl->ops->get_address)
2862                 return 0;
2863
2864         return a->mode;
2865 }
2866
2867 static struct attribute_group nvme_dev_attrs_group = {
2868         .attrs          = nvme_dev_attrs,
2869         .is_visible     = nvme_dev_attrs_are_visible,
2870 };
2871
2872 static const struct attribute_group *nvme_dev_attr_groups[] = {
2873         &nvme_dev_attrs_group,
2874         NULL,
2875 };
2876
2877 static struct nvme_ns_head *__nvme_find_ns_head(struct nvme_subsystem *subsys,
2878                 unsigned nsid)
2879 {
2880         struct nvme_ns_head *h;
2881
2882         lockdep_assert_held(&subsys->lock);
2883
2884         list_for_each_entry(h, &subsys->nsheads, entry) {
2885                 if (h->ns_id == nsid && kref_get_unless_zero(&h->ref))
2886                         return h;
2887         }
2888
2889         return NULL;
2890 }
2891
2892 static int __nvme_check_ids(struct nvme_subsystem *subsys,
2893                 struct nvme_ns_head *new)
2894 {
2895         struct nvme_ns_head *h;
2896
2897         lockdep_assert_held(&subsys->lock);
2898
2899         list_for_each_entry(h, &subsys->nsheads, entry) {
2900                 if (nvme_ns_ids_valid(&new->ids) &&
2901                     !list_empty(&h->list) &&
2902                     nvme_ns_ids_equal(&new->ids, &h->ids))
2903                         return -EINVAL;
2904         }
2905
2906         return 0;
2907 }
2908
2909 static struct nvme_ns_head *nvme_alloc_ns_head(struct nvme_ctrl *ctrl,
2910                 unsigned nsid, struct nvme_id_ns *id)
2911 {
2912         struct nvme_ns_head *head;
2913         size_t size = sizeof(*head);
2914         int ret = -ENOMEM;
2915
2916 #ifdef CONFIG_NVME_MULTIPATH
2917         size += num_possible_nodes() * sizeof(struct nvme_ns *);
2918 #endif
2919
2920         head = kzalloc(size, GFP_KERNEL);
2921         if (!head)
2922                 goto out;
2923         ret = ida_simple_get(&ctrl->subsys->ns_ida, 1, 0, GFP_KERNEL);
2924         if (ret < 0)
2925                 goto out_free_head;
2926         head->instance = ret;
2927         INIT_LIST_HEAD(&head->list);
2928         ret = init_srcu_struct(&head->srcu);
2929         if (ret)
2930                 goto out_ida_remove;
2931         head->subsys = ctrl->subsys;
2932         head->ns_id = nsid;
2933         kref_init(&head->ref);
2934
2935         nvme_report_ns_ids(ctrl, nsid, id, &head->ids);
2936
2937         ret = __nvme_check_ids(ctrl->subsys, head);
2938         if (ret) {
2939                 dev_err(ctrl->device,
2940                         "duplicate IDs for nsid %d\n", nsid);
2941                 goto out_cleanup_srcu;
2942         }
2943
2944         ret = nvme_mpath_alloc_disk(ctrl, head);
2945         if (ret)
2946                 goto out_cleanup_srcu;
2947
2948         list_add_tail(&head->entry, &ctrl->subsys->nsheads);
2949
2950         kref_get(&ctrl->subsys->ref);
2951
2952         return head;
2953 out_cleanup_srcu:
2954         cleanup_srcu_struct(&head->srcu);
2955 out_ida_remove:
2956         ida_simple_remove(&ctrl->subsys->ns_ida, head->instance);
2957 out_free_head:
2958         kfree(head);
2959 out:
2960         return ERR_PTR(ret);
2961 }
2962
2963 static int nvme_init_ns_head(struct nvme_ns *ns, unsigned nsid,
2964                 struct nvme_id_ns *id)
2965 {
2966         struct nvme_ctrl *ctrl = ns->ctrl;
2967         bool is_shared = id->nmic & (1 << 0);
2968         struct nvme_ns_head *head = NULL;
2969         int ret = 0;
2970
2971         mutex_lock(&ctrl->subsys->lock);
2972         if (is_shared)
2973                 head = __nvme_find_ns_head(ctrl->subsys, nsid);
2974         if (!head) {
2975                 head = nvme_alloc_ns_head(ctrl, nsid, id);
2976                 if (IS_ERR(head)) {
2977                         ret = PTR_ERR(head);
2978                         goto out_unlock;
2979                 }
2980         } else {
2981                 struct nvme_ns_ids ids;
2982
2983                 nvme_report_ns_ids(ctrl, nsid, id, &ids);
2984                 if (!nvme_ns_ids_equal(&head->ids, &ids)) {
2985                         dev_err(ctrl->device,
2986                                 "IDs don't match for shared namespace %d\n",
2987                                         nsid);
2988                         ret = -EINVAL;
2989                         goto out_unlock;
2990                 }
2991         }
2992
2993         list_add_tail(&ns->siblings, &head->list);
2994         ns->head = head;
2995
2996 out_unlock:
2997         mutex_unlock(&ctrl->subsys->lock);
2998         return ret;
2999 }
3000
3001 static int ns_cmp(void *priv, struct list_head *a, struct list_head *b)
3002 {
3003         struct nvme_ns *nsa = container_of(a, struct nvme_ns, list);
3004         struct nvme_ns *nsb = container_of(b, struct nvme_ns, list);
3005
3006         return nsa->head->ns_id - nsb->head->ns_id;
3007 }
3008
3009 static struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3010 {
3011         struct nvme_ns *ns, *ret = NULL;
3012
3013         down_read(&ctrl->namespaces_rwsem);
3014         list_for_each_entry(ns, &ctrl->namespaces, list) {
3015                 if (ns->head->ns_id == nsid) {
3016                         if (!kref_get_unless_zero(&ns->kref))
3017                                 continue;
3018                         ret = ns;
3019                         break;
3020                 }
3021                 if (ns->head->ns_id > nsid)
3022                         break;
3023         }
3024         up_read(&ctrl->namespaces_rwsem);
3025         return ret;
3026 }
3027
3028 static int nvme_setup_streams_ns(struct nvme_ctrl *ctrl, struct nvme_ns *ns)
3029 {
3030         struct streams_directive_params s;
3031         int ret;
3032
3033         if (!ctrl->nr_streams)
3034                 return 0;
3035
3036         ret = nvme_get_stream_params(ctrl, &s, ns->head->ns_id);
3037         if (ret)
3038                 return ret;
3039
3040         ns->sws = le32_to_cpu(s.sws);
3041         ns->sgs = le16_to_cpu(s.sgs);
3042
3043         if (ns->sws) {
3044                 unsigned int bs = 1 << ns->lba_shift;
3045
3046                 blk_queue_io_min(ns->queue, bs * ns->sws);
3047                 if (ns->sgs)
3048                         blk_queue_io_opt(ns->queue, bs * ns->sws * ns->sgs);
3049         }
3050
3051         return 0;
3052 }
3053
3054 static void nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3055 {
3056         struct nvme_ns *ns;
3057         struct gendisk *disk;
3058         struct nvme_id_ns *id;
3059         char disk_name[DISK_NAME_LEN];
3060         int node = dev_to_node(ctrl->dev), flags = GENHD_FL_EXT_DEVT;
3061
3062         ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
3063         if (!ns)
3064                 return;
3065
3066         ns->queue = blk_mq_init_queue(ctrl->tagset);
3067         if (IS_ERR(ns->queue))
3068                 goto out_free_ns;
3069
3070         blk_queue_flag_set(QUEUE_FLAG_NONROT, ns->queue);
3071         if (ctrl->ops->flags & NVME_F_PCI_P2PDMA)
3072                 blk_queue_flag_set(QUEUE_FLAG_PCI_P2PDMA, ns->queue);
3073
3074         ns->queue->queuedata = ns;
3075         ns->ctrl = ctrl;
3076
3077         kref_init(&ns->kref);
3078         ns->lba_shift = 9; /* set to a default value for 512 until disk is validated */
3079
3080         blk_queue_logical_block_size(ns->queue, 1 << ns->lba_shift);
3081         nvme_set_queue_limits(ctrl, ns->queue);
3082
3083         id = nvme_identify_ns(ctrl, nsid);
3084         if (!id)
3085                 goto out_free_queue;
3086
3087         if (id->ncap == 0)
3088                 goto out_free_id;
3089
3090         if (nvme_init_ns_head(ns, nsid, id))
3091                 goto out_free_id;
3092         nvme_setup_streams_ns(ctrl, ns);
3093         nvme_set_disk_name(disk_name, ns, ctrl, &flags);
3094
3095         if ((ctrl->quirks & NVME_QUIRK_LIGHTNVM) && id->vs[0] == 0x1) {
3096                 if (nvme_nvm_register(ns, disk_name, node)) {
3097                         dev_warn(ctrl->device, "LightNVM init failure\n");
3098                         goto out_unlink_ns;
3099                 }
3100         }
3101
3102         disk = alloc_disk_node(0, node);
3103         if (!disk)
3104                 goto out_unlink_ns;
3105
3106         disk->fops = &nvme_fops;
3107         disk->private_data = ns;
3108         disk->queue = ns->queue;
3109         disk->flags = flags;
3110         memcpy(disk->disk_name, disk_name, DISK_NAME_LEN);
3111         ns->disk = disk;
3112
3113         __nvme_revalidate_disk(disk, id);
3114
3115         down_write(&ctrl->namespaces_rwsem);
3116         list_add_tail(&ns->list, &ctrl->namespaces);
3117         up_write(&ctrl->namespaces_rwsem);
3118
3119         nvme_get_ctrl(ctrl);
3120
3121         device_add_disk(ctrl->device, ns->disk, nvme_ns_id_attr_groups);
3122
3123         nvme_mpath_add_disk(ns, id);
3124         nvme_fault_inject_init(ns);
3125         kfree(id);
3126
3127         return;
3128  out_unlink_ns:
3129         mutex_lock(&ctrl->subsys->lock);
3130         list_del_rcu(&ns->siblings);
3131         mutex_unlock(&ctrl->subsys->lock);
3132  out_free_id:
3133         kfree(id);
3134  out_free_queue:
3135         blk_cleanup_queue(ns->queue);
3136  out_free_ns:
3137         kfree(ns);
3138 }
3139
3140 static void nvme_ns_remove(struct nvme_ns *ns)
3141 {
3142         if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
3143                 return;
3144
3145         nvme_fault_inject_fini(ns);
3146         if (ns->disk && ns->disk->flags & GENHD_FL_UP) {
3147                 del_gendisk(ns->disk);
3148                 blk_cleanup_queue(ns->queue);
3149                 if (blk_get_integrity(ns->disk))
3150                         blk_integrity_unregister(ns->disk);
3151         }
3152
3153         mutex_lock(&ns->ctrl->subsys->lock);
3154         list_del_rcu(&ns->siblings);
3155         nvme_mpath_clear_current_path(ns);
3156         mutex_unlock(&ns->ctrl->subsys->lock);
3157
3158         down_write(&ns->ctrl->namespaces_rwsem);
3159         list_del_init(&ns->list);
3160         up_write(&ns->ctrl->namespaces_rwsem);
3161
3162         synchronize_srcu(&ns->head->srcu);
3163         nvme_mpath_check_last_path(ns);
3164         nvme_put_ns(ns);
3165 }
3166
3167 static void nvme_validate_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3168 {
3169         struct nvme_ns *ns;
3170
3171         ns = nvme_find_get_ns(ctrl, nsid);
3172         if (ns) {
3173                 if (ns->disk && revalidate_disk(ns->disk))
3174                         nvme_ns_remove(ns);
3175                 nvme_put_ns(ns);
3176         } else
3177                 nvme_alloc_ns(ctrl, nsid);
3178 }
3179
3180 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
3181                                         unsigned nsid)
3182 {
3183         struct nvme_ns *ns, *next;
3184         LIST_HEAD(rm_list);
3185
3186         down_write(&ctrl->namespaces_rwsem);
3187         list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
3188                 if (ns->head->ns_id > nsid || test_bit(NVME_NS_DEAD, &ns->flags))
3189                         list_move_tail(&ns->list, &rm_list);
3190         }
3191         up_write(&ctrl->namespaces_rwsem);
3192
3193         list_for_each_entry_safe(ns, next, &rm_list, list)
3194                 nvme_ns_remove(ns);
3195
3196 }
3197
3198 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl, unsigned nn)
3199 {
3200         struct nvme_ns *ns;
3201         __le32 *ns_list;
3202         unsigned i, j, nsid, prev = 0, num_lists = DIV_ROUND_UP(nn, 1024);
3203         int ret = 0;
3204
3205         ns_list = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
3206         if (!ns_list)
3207                 return -ENOMEM;
3208
3209         for (i = 0; i < num_lists; i++) {
3210                 ret = nvme_identify_ns_list(ctrl, prev, ns_list);
3211                 if (ret)
3212                         goto free;
3213
3214                 for (j = 0; j < min(nn, 1024U); j++) {
3215                         nsid = le32_to_cpu(ns_list[j]);
3216                         if (!nsid)
3217                                 goto out;
3218
3219                         nvme_validate_ns(ctrl, nsid);
3220
3221                         while (++prev < nsid) {
3222                                 ns = nvme_find_get_ns(ctrl, prev);
3223                                 if (ns) {
3224                                         nvme_ns_remove(ns);
3225                                         nvme_put_ns(ns);
3226                                 }
3227                         }
3228                 }
3229                 nn -= j;
3230         }
3231  out:
3232         nvme_remove_invalid_namespaces(ctrl, prev);
3233  free:
3234         kfree(ns_list);
3235         return ret;
3236 }
3237
3238 static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl, unsigned nn)
3239 {
3240         unsigned i;
3241
3242         for (i = 1; i <= nn; i++)
3243                 nvme_validate_ns(ctrl, i);
3244
3245         nvme_remove_invalid_namespaces(ctrl, nn);
3246 }
3247
3248 static void nvme_clear_changed_ns_log(struct nvme_ctrl *ctrl)
3249 {
3250         size_t log_size = NVME_MAX_CHANGED_NAMESPACES * sizeof(__le32);
3251         __le32 *log;
3252         int error;
3253
3254         log = kzalloc(log_size, GFP_KERNEL);
3255         if (!log)
3256                 return;
3257
3258         /*
3259          * We need to read the log to clear the AEN, but we don't want to rely
3260          * on it for the changed namespace information as userspace could have
3261          * raced with us in reading the log page, which could cause us to miss
3262          * updates.
3263          */
3264         error = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CHANGED_NS, 0, log,
3265                         log_size, 0);
3266         if (error)
3267                 dev_warn(ctrl->device,
3268                         "reading changed ns log failed: %d\n", error);
3269
3270         kfree(log);
3271 }
3272
3273 static void nvme_scan_work(struct work_struct *work)
3274 {
3275         struct nvme_ctrl *ctrl =
3276                 container_of(work, struct nvme_ctrl, scan_work);
3277         struct nvme_id_ctrl *id;
3278         unsigned nn;
3279
3280         if (ctrl->state != NVME_CTRL_LIVE)
3281                 return;
3282
3283         WARN_ON_ONCE(!ctrl->tagset);
3284
3285         if (test_and_clear_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events)) {
3286                 dev_info(ctrl->device, "rescanning namespaces.\n");
3287                 nvme_clear_changed_ns_log(ctrl);
3288         }
3289
3290         if (nvme_identify_ctrl(ctrl, &id))
3291                 return;
3292
3293         nn = le32_to_cpu(id->nn);
3294         if (ctrl->vs >= NVME_VS(1, 1, 0) &&
3295             !(ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)) {
3296                 if (!nvme_scan_ns_list(ctrl, nn))
3297                         goto out_free_id;
3298         }
3299         nvme_scan_ns_sequential(ctrl, nn);
3300 out_free_id:
3301         kfree(id);
3302         down_write(&ctrl->namespaces_rwsem);
3303         list_sort(NULL, &ctrl->namespaces, ns_cmp);
3304         up_write(&ctrl->namespaces_rwsem);
3305 }
3306
3307 /*
3308  * This function iterates the namespace list unlocked to allow recovery from
3309  * controller failure. It is up to the caller to ensure the namespace list is
3310  * not modified by scan work while this function is executing.
3311  */
3312 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
3313 {
3314         struct nvme_ns *ns, *next;
3315         LIST_HEAD(ns_list);
3316
3317         /*
3318          * The dead states indicates the controller was not gracefully
3319          * disconnected. In that case, we won't be able to flush any data while
3320          * removing the namespaces' disks; fail all the queues now to avoid
3321          * potentially having to clean up the failed sync later.
3322          */
3323         if (ctrl->state == NVME_CTRL_DEAD)
3324                 nvme_kill_queues(ctrl);
3325
3326         down_write(&ctrl->namespaces_rwsem);
3327         list_splice_init(&ctrl->namespaces, &ns_list);
3328         up_write(&ctrl->namespaces_rwsem);
3329
3330         list_for_each_entry_safe(ns, next, &ns_list, list)
3331                 nvme_ns_remove(ns);
3332 }
3333 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
3334
3335 static void nvme_aen_uevent(struct nvme_ctrl *ctrl)
3336 {
3337         char *envp[2] = { NULL, NULL };
3338         u32 aen_result = ctrl->aen_result;
3339
3340         ctrl->aen_result = 0;
3341         if (!aen_result)
3342                 return;
3343
3344         envp[0] = kasprintf(GFP_KERNEL, "NVME_AEN=%#08x", aen_result);
3345         if (!envp[0])
3346                 return;
3347         kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp);
3348         kfree(envp[0]);
3349 }
3350
3351 static void nvme_async_event_work(struct work_struct *work)
3352 {
3353         struct nvme_ctrl *ctrl =
3354                 container_of(work, struct nvme_ctrl, async_event_work);
3355
3356         nvme_aen_uevent(ctrl);
3357         ctrl->ops->submit_async_event(ctrl);
3358 }
3359
3360 static bool nvme_ctrl_pp_status(struct nvme_ctrl *ctrl)
3361 {
3362
3363         u32 csts;
3364
3365         if (ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts))
3366                 return false;
3367
3368         if (csts == ~0)
3369                 return false;
3370
3371         return ((ctrl->ctrl_config & NVME_CC_ENABLE) && (csts & NVME_CSTS_PP));
3372 }
3373
3374 static void nvme_get_fw_slot_info(struct nvme_ctrl *ctrl)
3375 {
3376         struct nvme_fw_slot_info_log *log;
3377
3378         log = kmalloc(sizeof(*log), GFP_KERNEL);
3379         if (!log)
3380                 return;
3381
3382         if (nvme_get_log(ctrl, NVME_NSID_ALL, 0, NVME_LOG_FW_SLOT, log,
3383                         sizeof(*log), 0))
3384                 dev_warn(ctrl->device, "Get FW SLOT INFO log error\n");
3385         kfree(log);
3386 }
3387
3388 static void nvme_fw_act_work(struct work_struct *work)
3389 {
3390         struct nvme_ctrl *ctrl = container_of(work,
3391                                 struct nvme_ctrl, fw_act_work);
3392         unsigned long fw_act_timeout;
3393
3394         if (ctrl->mtfa)
3395                 fw_act_timeout = jiffies +
3396                                 msecs_to_jiffies(ctrl->mtfa * 100);
3397         else
3398                 fw_act_timeout = jiffies +
3399                                 msecs_to_jiffies(admin_timeout * 1000);
3400
3401         nvme_stop_queues(ctrl);
3402         while (nvme_ctrl_pp_status(ctrl)) {
3403                 if (time_after(jiffies, fw_act_timeout)) {
3404                         dev_warn(ctrl->device,
3405                                 "Fw activation timeout, reset controller\n");
3406                         nvme_reset_ctrl(ctrl);
3407                         break;
3408                 }
3409                 msleep(100);
3410         }
3411
3412         if (ctrl->state != NVME_CTRL_LIVE)
3413                 return;
3414
3415         nvme_start_queues(ctrl);
3416         /* read FW slot information to clear the AER */
3417         nvme_get_fw_slot_info(ctrl);
3418 }
3419
3420 static void nvme_handle_aen_notice(struct nvme_ctrl *ctrl, u32 result)
3421 {
3422         u32 aer_notice_type = (result & 0xff00) >> 8;
3423
3424         switch (aer_notice_type) {
3425         case NVME_AER_NOTICE_NS_CHANGED:
3426                 trace_nvme_async_event(ctrl, aer_notice_type);
3427                 set_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events);
3428                 nvme_queue_scan(ctrl);
3429                 break;
3430         case NVME_AER_NOTICE_FW_ACT_STARTING:
3431                 trace_nvme_async_event(ctrl, aer_notice_type);
3432                 queue_work(nvme_wq, &ctrl->fw_act_work);
3433                 break;
3434 #ifdef CONFIG_NVME_MULTIPATH
3435         case NVME_AER_NOTICE_ANA:
3436                 trace_nvme_async_event(ctrl, aer_notice_type);
3437                 if (!ctrl->ana_log_buf)
3438                         break;
3439                 queue_work(nvme_wq, &ctrl->ana_work);
3440                 break;
3441 #endif
3442         default:
3443                 dev_warn(ctrl->device, "async event result %08x\n", result);
3444         }
3445 }
3446
3447 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
3448                 volatile union nvme_result *res)
3449 {
3450         u32 result = le32_to_cpu(res->u32);
3451         u32 aer_type = result & 0x07;
3452
3453         if (le16_to_cpu(status) >> 1 != NVME_SC_SUCCESS)
3454                 return;
3455
3456         switch (aer_type) {
3457         case NVME_AER_NOTICE:
3458                 nvme_handle_aen_notice(ctrl, result);
3459                 break;
3460         case NVME_AER_ERROR:
3461         case NVME_AER_SMART:
3462         case NVME_AER_CSS:
3463         case NVME_AER_VS:
3464                 trace_nvme_async_event(ctrl, aer_type);
3465                 ctrl->aen_result = result;
3466                 break;
3467         default:
3468                 break;
3469         }
3470         queue_work(nvme_wq, &ctrl->async_event_work);
3471 }
3472 EXPORT_SYMBOL_GPL(nvme_complete_async_event);
3473
3474 void nvme_stop_ctrl(struct nvme_ctrl *ctrl)
3475 {
3476         nvme_mpath_stop(ctrl);
3477         nvme_stop_keep_alive(ctrl);
3478         flush_work(&ctrl->async_event_work);
3479         flush_work(&ctrl->scan_work);
3480         cancel_work_sync(&ctrl->fw_act_work);
3481         if (ctrl->ops->stop_ctrl)
3482                 ctrl->ops->stop_ctrl(ctrl);
3483 }
3484 EXPORT_SYMBOL_GPL(nvme_stop_ctrl);
3485
3486 void nvme_start_ctrl(struct nvme_ctrl *ctrl)
3487 {
3488         if (ctrl->kato)
3489                 nvme_start_keep_alive(ctrl);
3490
3491         if (ctrl->queue_count > 1) {
3492                 nvme_queue_scan(ctrl);
3493                 nvme_enable_aen(ctrl);
3494                 queue_work(nvme_wq, &ctrl->async_event_work);
3495                 nvme_start_queues(ctrl);
3496         }
3497 }
3498 EXPORT_SYMBOL_GPL(nvme_start_ctrl);
3499
3500 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
3501 {
3502         cdev_device_del(&ctrl->cdev, ctrl->device);
3503 }
3504 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl);
3505
3506 static void nvme_free_ctrl(struct device *dev)
3507 {
3508         struct nvme_ctrl *ctrl =
3509                 container_of(dev, struct nvme_ctrl, ctrl_device);
3510         struct nvme_subsystem *subsys = ctrl->subsys;
3511
3512         ida_simple_remove(&nvme_instance_ida, ctrl->instance);
3513         kfree(ctrl->effects);
3514         nvme_mpath_uninit(ctrl);
3515
3516         if (subsys) {
3517                 mutex_lock(&subsys->lock);
3518                 list_del(&ctrl->subsys_entry);
3519                 mutex_unlock(&subsys->lock);
3520                 sysfs_remove_link(&subsys->dev.kobj, dev_name(ctrl->device));
3521         }
3522
3523         ctrl->ops->free_ctrl(ctrl);
3524
3525         if (subsys)
3526                 nvme_put_subsystem(subsys);
3527 }
3528
3529 /*
3530  * Initialize a NVMe controller structures.  This needs to be called during
3531  * earliest initialization so that we have the initialized structured around
3532  * during probing.
3533  */
3534 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
3535                 const struct nvme_ctrl_ops *ops, unsigned long quirks)
3536 {
3537         int ret;
3538
3539         ctrl->state = NVME_CTRL_NEW;
3540         spin_lock_init(&ctrl->lock);
3541         INIT_LIST_HEAD(&ctrl->namespaces);
3542         init_rwsem(&ctrl->namespaces_rwsem);
3543         ctrl->dev = dev;
3544         ctrl->ops = ops;
3545         ctrl->quirks = quirks;
3546         INIT_WORK(&ctrl->scan_work, nvme_scan_work);
3547         INIT_WORK(&ctrl->async_event_work, nvme_async_event_work);
3548         INIT_WORK(&ctrl->fw_act_work, nvme_fw_act_work);
3549         INIT_WORK(&ctrl->delete_work, nvme_delete_ctrl_work);
3550
3551         INIT_DELAYED_WORK(&ctrl->ka_work, nvme_keep_alive_work);
3552         memset(&ctrl->ka_cmd, 0, sizeof(ctrl->ka_cmd));
3553         ctrl->ka_cmd.common.opcode = nvme_admin_keep_alive;
3554
3555         ret = ida_simple_get(&nvme_instance_ida, 0, 0, GFP_KERNEL);
3556         if (ret < 0)
3557                 goto out;
3558         ctrl->instance = ret;
3559
3560         device_initialize(&ctrl->ctrl_device);
3561         ctrl->device = &ctrl->ctrl_device;
3562         ctrl->device->devt = MKDEV(MAJOR(nvme_chr_devt), ctrl->instance);
3563         ctrl->device->class = nvme_class;
3564         ctrl->device->parent = ctrl->dev;
3565         ctrl->device->groups = nvme_dev_attr_groups;
3566         ctrl->device->release = nvme_free_ctrl;
3567         dev_set_drvdata(ctrl->device, ctrl);
3568         ret = dev_set_name(ctrl->device, "nvme%d", ctrl->instance);
3569         if (ret)
3570                 goto out_release_instance;
3571
3572         cdev_init(&ctrl->cdev, &nvme_dev_fops);
3573         ctrl->cdev.owner = ops->module;
3574         ret = cdev_device_add(&ctrl->cdev, ctrl->device);
3575         if (ret)
3576                 goto out_free_name;
3577
3578         /*
3579          * Initialize latency tolerance controls.  The sysfs files won't
3580          * be visible to userspace unless the device actually supports APST.
3581          */
3582         ctrl->device->power.set_latency_tolerance = nvme_set_latency_tolerance;
3583         dev_pm_qos_update_user_latency_tolerance(ctrl->device,
3584                 min(default_ps_max_latency_us, (unsigned long)S32_MAX));
3585
3586         return 0;
3587 out_free_name:
3588         kfree_const(dev->kobj.name);
3589 out_release_instance:
3590         ida_simple_remove(&nvme_instance_ida, ctrl->instance);
3591 out:
3592         return ret;
3593 }
3594 EXPORT_SYMBOL_GPL(nvme_init_ctrl);
3595
3596 /**
3597  * nvme_kill_queues(): Ends all namespace queues
3598  * @ctrl: the dead controller that needs to end
3599  *
3600  * Call this function when the driver determines it is unable to get the
3601  * controller in a state capable of servicing IO.
3602  */
3603 void nvme_kill_queues(struct nvme_ctrl *ctrl)
3604 {
3605         struct nvme_ns *ns;
3606
3607         down_read(&ctrl->namespaces_rwsem);
3608
3609         /* Forcibly unquiesce queues to avoid blocking dispatch */
3610         if (ctrl->admin_q)
3611                 blk_mq_unquiesce_queue(ctrl->admin_q);
3612
3613         list_for_each_entry(ns, &ctrl->namespaces, list)
3614                 nvme_set_queue_dying(ns);
3615
3616         up_read(&ctrl->namespaces_rwsem);
3617 }
3618 EXPORT_SYMBOL_GPL(nvme_kill_queues);
3619
3620 void nvme_unfreeze(struct nvme_ctrl *ctrl)
3621 {
3622         struct nvme_ns *ns;
3623
3624         down_read(&ctrl->namespaces_rwsem);
3625         list_for_each_entry(ns, &ctrl->namespaces, list)
3626                 blk_mq_unfreeze_queue(ns->queue);
3627         up_read(&ctrl->namespaces_rwsem);
3628 }
3629 EXPORT_SYMBOL_GPL(nvme_unfreeze);
3630
3631 void nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout)
3632 {
3633         struct nvme_ns *ns;
3634
3635         down_read(&ctrl->namespaces_rwsem);
3636         list_for_each_entry(ns, &ctrl->namespaces, list) {
3637                 timeout = blk_mq_freeze_queue_wait_timeout(ns->queue, timeout);
3638                 if (timeout <= 0)
3639                         break;
3640         }
3641         up_read(&ctrl->namespaces_rwsem);
3642 }
3643 EXPORT_SYMBOL_GPL(nvme_wait_freeze_timeout);
3644
3645 void nvme_wait_freeze(struct nvme_ctrl *ctrl)
3646 {
3647         struct nvme_ns *ns;
3648
3649         down_read(&ctrl->namespaces_rwsem);
3650         list_for_each_entry(ns, &ctrl->namespaces, list)
3651                 blk_mq_freeze_queue_wait(ns->queue);
3652         up_read(&ctrl->namespaces_rwsem);
3653 }
3654 EXPORT_SYMBOL_GPL(nvme_wait_freeze);
3655
3656 void nvme_start_freeze(struct nvme_ctrl *ctrl)
3657 {
3658         struct nvme_ns *ns;
3659
3660         down_read(&ctrl->namespaces_rwsem);
3661         list_for_each_entry(ns, &ctrl->namespaces, list)
3662                 blk_freeze_queue_start(ns->queue);
3663         up_read(&ctrl->namespaces_rwsem);
3664 }
3665 EXPORT_SYMBOL_GPL(nvme_start_freeze);
3666
3667 void nvme_stop_queues(struct nvme_ctrl *ctrl)
3668 {
3669         struct nvme_ns *ns;
3670
3671         down_read(&ctrl->namespaces_rwsem);
3672         list_for_each_entry(ns, &ctrl->namespaces, list)
3673                 blk_mq_quiesce_queue(ns->queue);
3674         up_read(&ctrl->namespaces_rwsem);
3675 }
3676 EXPORT_SYMBOL_GPL(nvme_stop_queues);
3677
3678 void nvme_start_queues(struct nvme_ctrl *ctrl)
3679 {
3680         struct nvme_ns *ns;
3681
3682         down_read(&ctrl->namespaces_rwsem);
3683         list_for_each_entry(ns, &ctrl->namespaces, list)
3684                 blk_mq_unquiesce_queue(ns->queue);
3685         up_read(&ctrl->namespaces_rwsem);
3686 }
3687 EXPORT_SYMBOL_GPL(nvme_start_queues);
3688
3689 int __init nvme_core_init(void)
3690 {
3691         int result = -ENOMEM;
3692
3693         nvme_wq = alloc_workqueue("nvme-wq",
3694                         WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
3695         if (!nvme_wq)
3696                 goto out;
3697
3698         nvme_reset_wq = alloc_workqueue("nvme-reset-wq",
3699                         WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
3700         if (!nvme_reset_wq)
3701                 goto destroy_wq;
3702
3703         nvme_delete_wq = alloc_workqueue("nvme-delete-wq",
3704                         WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
3705         if (!nvme_delete_wq)
3706                 goto destroy_reset_wq;
3707
3708         result = alloc_chrdev_region(&nvme_chr_devt, 0, NVME_MINORS, "nvme");
3709         if (result < 0)
3710                 goto destroy_delete_wq;
3711
3712         nvme_class = class_create(THIS_MODULE, "nvme");
3713         if (IS_ERR(nvme_class)) {
3714                 result = PTR_ERR(nvme_class);
3715                 goto unregister_chrdev;
3716         }
3717
3718         nvme_subsys_class = class_create(THIS_MODULE, "nvme-subsystem");
3719         if (IS_ERR(nvme_subsys_class)) {
3720                 result = PTR_ERR(nvme_subsys_class);
3721                 goto destroy_class;
3722         }
3723         return 0;
3724
3725 destroy_class:
3726         class_destroy(nvme_class);
3727 unregister_chrdev:
3728         unregister_chrdev_region(nvme_chr_devt, NVME_MINORS);
3729 destroy_delete_wq:
3730         destroy_workqueue(nvme_delete_wq);
3731 destroy_reset_wq:
3732         destroy_workqueue(nvme_reset_wq);
3733 destroy_wq:
3734         destroy_workqueue(nvme_wq);
3735 out:
3736         return result;
3737 }
3738
3739 void nvme_core_exit(void)
3740 {
3741         ida_destroy(&nvme_subsystems_ida);
3742         class_destroy(nvme_subsys_class);
3743         class_destroy(nvme_class);
3744         unregister_chrdev_region(nvme_chr_devt, NVME_MINORS);
3745         destroy_workqueue(nvme_delete_wq);
3746         destroy_workqueue(nvme_reset_wq);
3747         destroy_workqueue(nvme_wq);
3748 }
3749
3750 MODULE_LICENSE("GPL");
3751 MODULE_VERSION("1.0");
3752 module_init(nvme_core_init);
3753 module_exit(nvme_core_exit);