drivers/nvme/host/core.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * NVM Express device driver
   4  * Copyright (c) 2011-2014, Intel Corporation.
   5  */
   6
   7 #include <linux/blkdev.h>
   8 #include <linux/blk-mq.h>
   9 #include <linux/blk-integrity.h>
  10 #include <linux/compat.h>
  11 #include <linux/delay.h>
  12 #include <linux/errno.h>
  13 #include <linux/hdreg.h>
  14 #include <linux/kernel.h>
  15 #include <linux/module.h>
  16 #include <linux/backing-dev.h>
  17 #include <linux/slab.h>
  18 #include <linux/types.h>
  19 #include <linux/pr.h>
  20 #include <linux/ptrace.h>
  21 #include <linux/nvme_ioctl.h>
  22 #include <linux/pm_qos.h>
  23 #include <asm/unaligned.h>
  24
  25 #include "nvme.h"
  26 #include "fabrics.h"
  27
  28 #define CREATE_TRACE_POINTS
  29 #include "trace.h"
  30
  31 #define NVME_MINORS             (1U << MINORBITS)
  32
  33 unsigned int admin_timeout = 60;
  34 module_param(admin_timeout, uint, 0644);
  35 MODULE_PARM_DESC(admin_timeout, "timeout in seconds for admin commands");
  36 EXPORT_SYMBOL_GPL(admin_timeout);
  37
  38 unsigned int nvme_io_timeout = 30;
  39 module_param_named(io_timeout, nvme_io_timeout, uint, 0644);
  40 MODULE_PARM_DESC(io_timeout, "timeout in seconds for I/O");
  41 EXPORT_SYMBOL_GPL(nvme_io_timeout);
  42
  43 static unsigned char shutdown_timeout = 5;
  44 module_param(shutdown_timeout, byte, 0644);
  45 MODULE_PARM_DESC(shutdown_timeout, "timeout in seconds for controller shutdown");
  46
  47 static u8 nvme_max_retries = 5;
  48 module_param_named(max_retries, nvme_max_retries, byte, 0644);
  49 MODULE_PARM_DESC(max_retries, "max number of retries a command may have");
  50
  51 static unsigned long default_ps_max_latency_us = 100000;
  52 module_param(default_ps_max_latency_us, ulong, 0644);
  53 MODULE_PARM_DESC(default_ps_max_latency_us,
  54                  "max power saving latency for new devices; use PM QOS to change per device");
  55
  56 static bool force_apst;
  57 module_param(force_apst, bool, 0644);
  58 MODULE_PARM_DESC(force_apst, "allow APST for newly enumerated devices even if quirked off");
  59
  60 static unsigned long apst_primary_timeout_ms = 100;
  61 module_param(apst_primary_timeout_ms, ulong, 0644);
  62 MODULE_PARM_DESC(apst_primary_timeout_ms,
  63         "primary APST timeout in ms");
  64
  65 static unsigned long apst_secondary_timeout_ms = 2000;
  66 module_param(apst_secondary_timeout_ms, ulong, 0644);
  67 MODULE_PARM_DESC(apst_secondary_timeout_ms,
  68         "secondary APST timeout in ms");
  69
  70 static unsigned long apst_primary_latency_tol_us = 15000;
  71 module_param(apst_primary_latency_tol_us, ulong, 0644);
  72 MODULE_PARM_DESC(apst_primary_latency_tol_us,
  73         "primary APST latency tolerance in us");
  74
  75 static unsigned long apst_secondary_latency_tol_us = 100000;
  76 module_param(apst_secondary_latency_tol_us, ulong, 0644);
  77 MODULE_PARM_DESC(apst_secondary_latency_tol_us,
  78         "secondary APST latency tolerance in us");
  79
  80 static bool streams;
  81 module_param(streams, bool, 0644);
  82 MODULE_PARM_DESC(streams, "turn on support for Streams write directives");
  83
  84 /*
  85  * nvme_wq - hosts nvme related works that are not reset or delete
  86  * nvme_reset_wq - hosts nvme reset works
  87  * nvme_delete_wq - hosts nvme delete works
  88  *
  89  * nvme_wq will host works such as scan, aen handling, fw activation,
  90  * keep-alive, periodic reconnects etc. nvme_reset_wq
  91  * runs reset works which also flush works hosted on nvme_wq for
  92  * serialization purposes. nvme_delete_wq host controller deletion
  93  * works which flush reset works for serialization.
  94  */
  95 struct workqueue_struct *nvme_wq;
  96 EXPORT_SYMBOL_GPL(nvme_wq);
  97
  98 struct workqueue_struct *nvme_reset_wq;
  99 EXPORT_SYMBOL_GPL(nvme_reset_wq);
 100
 101 struct workqueue_struct *nvme_delete_wq;
 102 EXPORT_SYMBOL_GPL(nvme_delete_wq);
 103
 104 static LIST_HEAD(nvme_subsystems);
 105 static DEFINE_MUTEX(nvme_subsystems_lock);
 106
 107 static DEFINE_IDA(nvme_instance_ida);
 108 static dev_t nvme_ctrl_base_chr_devt;
 109 static struct class *nvme_class;
 110 static struct class *nvme_subsys_class;
 111
 112 static DEFINE_IDA(nvme_ns_chr_minor_ida);
 113 static dev_t nvme_ns_chr_devt;
 114 static struct class *nvme_ns_chr_class;
 115
 116 static void nvme_put_subsystem(struct nvme_subsystem *subsys);
 117 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
 118                                            unsigned nsid);
 119 static void nvme_update_keep_alive(struct nvme_ctrl *ctrl,
 120                                    struct nvme_command *cmd);
 121
 122 void nvme_queue_scan(struct nvme_ctrl *ctrl)
 123 {
 124         /*
 125          * Only new queue scan work when admin and IO queues are both alive
 126          */
 127         if (ctrl->state == NVME_CTRL_LIVE && ctrl->tagset)
 128                 queue_work(nvme_wq, &ctrl->scan_work);
 129 }
 130
 131 /*
 132  * Use this function to proceed with scheduling reset_work for a controller
 133  * that had previously been set to the resetting state. This is intended for
 134  * code paths that can't be interrupted by other reset attempts. A hot removal
 135  * may prevent this from succeeding.
 136  */
 137 int nvme_try_sched_reset(struct nvme_ctrl *ctrl)
 138 {
 139         if (ctrl->state != NVME_CTRL_RESETTING)
 140                 return -EBUSY;
 141         if (!queue_work(nvme_reset_wq, &ctrl->reset_work))
 142                 return -EBUSY;
 143         return 0;
 144 }
 145 EXPORT_SYMBOL_GPL(nvme_try_sched_reset);
 146
 147 static void nvme_failfast_work(struct work_struct *work)
 148 {
 149         struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
 150                         struct nvme_ctrl, failfast_work);
 151
 152         if (ctrl->state != NVME_CTRL_CONNECTING)
 153                 return;
 154
 155         set_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags);
 156         dev_info(ctrl->device, "failfast expired\n");
 157         nvme_kick_requeue_lists(ctrl);
 158 }
 159
 160 static inline void nvme_start_failfast_work(struct nvme_ctrl *ctrl)
 161 {
 162         if (!ctrl->opts || ctrl->opts->fast_io_fail_tmo == -1)
 163                 return;
 164
 165         schedule_delayed_work(&ctrl->failfast_work,
 166                               ctrl->opts->fast_io_fail_tmo * HZ);
 167 }
 168
 169 static inline void nvme_stop_failfast_work(struct nvme_ctrl *ctrl)
 170 {
 171         if (!ctrl->opts)
 172                 return;
 173
 174         cancel_delayed_work_sync(&ctrl->failfast_work);
 175         clear_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags);
 176 }
 177
 178
 179 int nvme_reset_ctrl(struct nvme_ctrl *ctrl)
 180 {
 181         if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING))
 182                 return -EBUSY;
 183         if (!queue_work(nvme_reset_wq, &ctrl->reset_work))
 184                 return -EBUSY;
 185         return 0;
 186 }
 187 EXPORT_SYMBOL_GPL(nvme_reset_ctrl);
 188
 189 int nvme_reset_ctrl_sync(struct nvme_ctrl *ctrl)
 190 {
 191         int ret;
 192
 193         ret = nvme_reset_ctrl(ctrl);
 194         if (!ret) {
 195                 flush_work(&ctrl->reset_work);
 196                 if (ctrl->state != NVME_CTRL_LIVE)
 197                         ret = -ENETRESET;
 198         }
 199
 200         return ret;
 201 }
 202
 203 static void nvme_do_delete_ctrl(struct nvme_ctrl *ctrl)
 204 {
 205         dev_info(ctrl->device,
 206                  "Removing ctrl: NQN \"%s\"\n", ctrl->opts->subsysnqn);
 207
 208         flush_work(&ctrl->reset_work);
 209         nvme_stop_ctrl(ctrl);
 210         nvme_remove_namespaces(ctrl);
 211         ctrl->ops->delete_ctrl(ctrl);
 212         nvme_uninit_ctrl(ctrl);
 213 }
 214
 215 static void nvme_delete_ctrl_work(struct work_struct *work)
 216 {
 217         struct nvme_ctrl *ctrl =
 218                 container_of(work, struct nvme_ctrl, delete_work);
 219
 220         nvme_do_delete_ctrl(ctrl);
 221 }
 222
 223 int nvme_delete_ctrl(struct nvme_ctrl *ctrl)
 224 {
 225         if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
 226                 return -EBUSY;
 227         if (!queue_work(nvme_delete_wq, &ctrl->delete_work))
 228                 return -EBUSY;
 229         return 0;
 230 }
 231 EXPORT_SYMBOL_GPL(nvme_delete_ctrl);
 232
 233 static void nvme_delete_ctrl_sync(struct nvme_ctrl *ctrl)
 234 {
 235         /*
 236          * Keep a reference until nvme_do_delete_ctrl() complete,
 237          * since ->delete_ctrl can free the controller.
 238          */
 239         nvme_get_ctrl(ctrl);
 240         if (nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
 241                 nvme_do_delete_ctrl(ctrl);
 242         nvme_put_ctrl(ctrl);
 243 }
 244
 245 static blk_status_t nvme_error_status(u16 status)
 246 {
 247         switch (status & 0x7ff) {
 248         case NVME_SC_SUCCESS:
 249                 return BLK_STS_OK;
 250         case NVME_SC_CAP_EXCEEDED:
 251                 return BLK_STS_NOSPC;
 252         case NVME_SC_LBA_RANGE:
 253         case NVME_SC_CMD_INTERRUPTED:
 254         case NVME_SC_NS_NOT_READY:
 255                 return BLK_STS_TARGET;
 256         case NVME_SC_BAD_ATTRIBUTES:
 257         case NVME_SC_ONCS_NOT_SUPPORTED:
 258         case NVME_SC_INVALID_OPCODE:
 259         case NVME_SC_INVALID_FIELD:
 260         case NVME_SC_INVALID_NS:
 261                 return BLK_STS_NOTSUPP;
 262         case NVME_SC_WRITE_FAULT:
 263         case NVME_SC_READ_ERROR:
 264         case NVME_SC_UNWRITTEN_BLOCK:
 265         case NVME_SC_ACCESS_DENIED:
 266         case NVME_SC_READ_ONLY:
 267         case NVME_SC_COMPARE_FAILED:
 268                 return BLK_STS_MEDIUM;
 269         case NVME_SC_GUARD_CHECK:
 270         case NVME_SC_APPTAG_CHECK:
 271         case NVME_SC_REFTAG_CHECK:
 272         case NVME_SC_INVALID_PI:
 273                 return BLK_STS_PROTECTION;
 274         case NVME_SC_RESERVATION_CONFLICT:
 275                 return BLK_STS_NEXUS;
 276         case NVME_SC_HOST_PATH_ERROR:
 277                 return BLK_STS_TRANSPORT;
 278         case NVME_SC_ZONE_TOO_MANY_ACTIVE:
 279                 return BLK_STS_ZONE_ACTIVE_RESOURCE;
 280         case NVME_SC_ZONE_TOO_MANY_OPEN:
 281                 return BLK_STS_ZONE_OPEN_RESOURCE;
 282         default:
 283                 return BLK_STS_IOERR;
 284         }
 285 }
 286
 287 static void nvme_retry_req(struct request *req)
 288 {
 289         unsigned long delay = 0;
 290         u16 crd;
 291
 292         /* The mask and shift result must be <= 3 */
 293         crd = (nvme_req(req)->status & NVME_SC_CRD) >> 11;
 294         if (crd)
 295                 delay = nvme_req(req)->ctrl->crdt[crd - 1] * 100;
 296
 297         nvme_req(req)->retries++;
 298         blk_mq_requeue_request(req, false);
 299         blk_mq_delay_kick_requeue_list(req->q, delay);
 300 }
 301
 302 enum nvme_disposition {
 303         COMPLETE,
 304         RETRY,
 305         FAILOVER,
 306 };
 307
 308 static inline enum nvme_disposition nvme_decide_disposition(struct request *req)
 309 {
 310         if (likely(nvme_req(req)->status == 0))
 311                 return COMPLETE;
 312
 313         if (blk_noretry_request(req) ||
 314             (nvme_req(req)->status & NVME_SC_DNR) ||
 315             nvme_req(req)->retries >= nvme_max_retries)
 316                 return COMPLETE;
 317
 318         if (req->cmd_flags & REQ_NVME_MPATH) {
 319                 if (nvme_is_path_error(nvme_req(req)->status) ||
 320                     blk_queue_dying(req->q))
 321                         return FAILOVER;
 322         } else {
 323                 if (blk_queue_dying(req->q))
 324                         return COMPLETE;
 325         }
 326
 327         return RETRY;
 328 }
 329
 330 static inline void nvme_end_req_zoned(struct request *req)
 331 {
 332         if (IS_ENABLED(CONFIG_BLK_DEV_ZONED) &&
 333             req_op(req) == REQ_OP_ZONE_APPEND)
 334                 req->__sector = nvme_lba_to_sect(req->q->queuedata,
 335                         le64_to_cpu(nvme_req(req)->result.u64));
 336 }
 337
 338 static inline void nvme_end_req(struct request *req)
 339 {
 340         blk_status_t status = nvme_error_status(nvme_req(req)->status);
 341
 342         nvme_end_req_zoned(req);
 343         nvme_trace_bio_complete(req);
 344         blk_mq_end_request(req, status);
 345 }
 346
 347 void nvme_complete_rq(struct request *req)
 348 {
 349         trace_nvme_complete_rq(req);
 350         nvme_cleanup_cmd(req);
 351
 352         if (nvme_req(req)->ctrl->kas)
 353                 nvme_req(req)->ctrl->comp_seen = true;
 354
 355         switch (nvme_decide_disposition(req)) {
 356         case COMPLETE:
 357                 nvme_end_req(req);
 358                 return;
 359         case RETRY:
 360                 nvme_retry_req(req);
 361                 return;
 362         case FAILOVER:
 363                 nvme_failover_req(req);
 364                 return;
 365         }
 366 }
 367 EXPORT_SYMBOL_GPL(nvme_complete_rq);
 368
 369 void nvme_complete_batch_req(struct request *req)
 370 {
 371         nvme_cleanup_cmd(req);
 372         nvme_end_req_zoned(req);
 373 }
 374 EXPORT_SYMBOL_GPL(nvme_complete_batch_req);
 375
 376 /*
 377  * Called to unwind from ->queue_rq on a failed command submission so that the
 378  * multipathing code gets called to potentially failover to another path.
 379  * The caller needs to unwind all transport specific resource allocations and
 380  * must return propagate the return value.
 381  */
 382 blk_status_t nvme_host_path_error(struct request *req)
 383 {
 384         nvme_req(req)->status = NVME_SC_HOST_PATH_ERROR;
 385         blk_mq_set_request_complete(req);
 386         nvme_complete_rq(req);
 387         return BLK_STS_OK;
 388 }
 389 EXPORT_SYMBOL_GPL(nvme_host_path_error);
 390
 391 bool nvme_cancel_request(struct request *req, void *data, bool reserved)
 392 {
 393         dev_dbg_ratelimited(((struct nvme_ctrl *) data)->device,
 394                                 "Cancelling I/O %d", req->tag);
 395
 396         /* don't abort one completed request */
 397         if (blk_mq_request_completed(req))
 398                 return true;
 399
 400         nvme_req(req)->status = NVME_SC_HOST_ABORTED_CMD;
 401         nvme_req(req)->flags |= NVME_REQ_CANCELLED;
 402         blk_mq_complete_request(req);
 403         return true;
 404 }
 405 EXPORT_SYMBOL_GPL(nvme_cancel_request);
 406
 407 void nvme_cancel_tagset(struct nvme_ctrl *ctrl)
 408 {
 409         if (ctrl->tagset) {
 410                 blk_mq_tagset_busy_iter(ctrl->tagset,
 411                                 nvme_cancel_request, ctrl);
 412                 blk_mq_tagset_wait_completed_request(ctrl->tagset);
 413         }
 414 }
 415 EXPORT_SYMBOL_GPL(nvme_cancel_tagset);
 416
 417 void nvme_cancel_admin_tagset(struct nvme_ctrl *ctrl)
 418 {
 419         if (ctrl->admin_tagset) {
 420                 blk_mq_tagset_busy_iter(ctrl->admin_tagset,
 421                                 nvme_cancel_request, ctrl);
 422                 blk_mq_tagset_wait_completed_request(ctrl->admin_tagset);
 423         }
 424 }
 425 EXPORT_SYMBOL_GPL(nvme_cancel_admin_tagset);
 426
 427 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
 428                 enum nvme_ctrl_state new_state)
 429 {
 430         enum nvme_ctrl_state old_state;
 431         unsigned long flags;
 432         bool changed = false;
 433
 434         spin_lock_irqsave(&ctrl->lock, flags);
 435
 436         old_state = ctrl->state;
 437         switch (new_state) {
 438         case NVME_CTRL_LIVE:
 439                 switch (old_state) {
 440                 case NVME_CTRL_NEW:
 441                 case NVME_CTRL_RESETTING:
 442                 case NVME_CTRL_CONNECTING:
 443                         changed = true;
 444                         fallthrough;
 445                 default:
 446                         break;
 447                 }
 448                 break;
 449         case NVME_CTRL_RESETTING:
 450                 switch (old_state) {
 451                 case NVME_CTRL_NEW:
 452                 case NVME_CTRL_LIVE:
 453                         changed = true;
 454                         fallthrough;
 455                 default:
 456                         break;
 457                 }
 458                 break;
 459         case NVME_CTRL_CONNECTING:
 460                 switch (old_state) {
 461                 case NVME_CTRL_NEW:
 462                 case NVME_CTRL_RESETTING:
 463                         changed = true;
 464                         fallthrough;
 465                 default:
 466                         break;
 467                 }
 468                 break;
 469         case NVME_CTRL_DELETING:
 470                 switch (old_state) {
 471                 case NVME_CTRL_LIVE:
 472                 case NVME_CTRL_RESETTING:
 473                 case NVME_CTRL_CONNECTING:
 474                         changed = true;
 475                         fallthrough;
 476                 default:
 477                         break;
 478                 }
 479                 break;
 480         case NVME_CTRL_DELETING_NOIO:
 481                 switch (old_state) {
 482                 case NVME_CTRL_DELETING:
 483                 case NVME_CTRL_DEAD:
 484                         changed = true;
 485                         fallthrough;
 486                 default:
 487                         break;
 488                 }
 489                 break;
 490         case NVME_CTRL_DEAD:
 491                 switch (old_state) {
 492                 case NVME_CTRL_DELETING:
 493                         changed = true;
 494                         fallthrough;
 495                 default:
 496                         break;
 497                 }
 498                 break;
 499         default:
 500                 break;
 501         }
 502
 503         if (changed) {
 504                 ctrl->state = new_state;
 505                 wake_up_all(&ctrl->state_wq);
 506         }
 507
 508         spin_unlock_irqrestore(&ctrl->lock, flags);
 509         if (!changed)
 510                 return false;
 511
 512         if (ctrl->state == NVME_CTRL_LIVE) {
 513                 if (old_state == NVME_CTRL_CONNECTING)
 514                         nvme_stop_failfast_work(ctrl);
 515                 nvme_kick_requeue_lists(ctrl);
 516         } else if (ctrl->state == NVME_CTRL_CONNECTING &&
 517                 old_state == NVME_CTRL_RESETTING) {
 518                 nvme_start_failfast_work(ctrl);
 519         }
 520         return changed;
 521 }
 522 EXPORT_SYMBOL_GPL(nvme_change_ctrl_state);
 523
 524 /*
 525  * Returns true for sink states that can't ever transition back to live.
 526  */
 527 static bool nvme_state_terminal(struct nvme_ctrl *ctrl)
 528 {
 529         switch (ctrl->state) {
 530         case NVME_CTRL_NEW:
 531         case NVME_CTRL_LIVE:
 532         case NVME_CTRL_RESETTING:
 533         case NVME_CTRL_CONNECTING:
 534                 return false;
 535         case NVME_CTRL_DELETING:
 536         case NVME_CTRL_DELETING_NOIO:
 537         case NVME_CTRL_DEAD:
 538                 return true;
 539         default:
 540                 WARN_ONCE(1, "Unhandled ctrl state:%d", ctrl->state);
 541                 return true;
 542         }
 543 }
 544
 545 /*
 546  * Waits for the controller state to be resetting, or returns false if it is
 547  * not possible to ever transition to that state.
 548  */
 549 bool nvme_wait_reset(struct nvme_ctrl *ctrl)
 550 {
 551         wait_event(ctrl->state_wq,
 552                    nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING) ||
 553                    nvme_state_terminal(ctrl));
 554         return ctrl->state == NVME_CTRL_RESETTING;
 555 }
 556 EXPORT_SYMBOL_GPL(nvme_wait_reset);
 557
 558 static void nvme_free_ns_head(struct kref *ref)
 559 {
 560         struct nvme_ns_head *head =
 561                 container_of(ref, struct nvme_ns_head, ref);
 562
 563         nvme_mpath_remove_disk(head);
 564         ida_simple_remove(&head->subsys->ns_ida, head->instance);
 565         cleanup_srcu_struct(&head->srcu);
 566         nvme_put_subsystem(head->subsys);
 567         kfree(head);
 568 }
 569
 570 bool nvme_tryget_ns_head(struct nvme_ns_head *head)
 571 {
 572         return kref_get_unless_zero(&head->ref);
 573 }
 574
 575 void nvme_put_ns_head(struct nvme_ns_head *head)
 576 {
 577         kref_put(&head->ref, nvme_free_ns_head);
 578 }
 579
 580 static void nvme_free_ns(struct kref *kref)
 581 {
 582         struct nvme_ns *ns = container_of(kref, struct nvme_ns, kref);
 583
 584         put_disk(ns->disk);
 585         nvme_put_ns_head(ns->head);
 586         nvme_put_ctrl(ns->ctrl);
 587         kfree(ns);
 588 }
 589
 590 static inline bool nvme_get_ns(struct nvme_ns *ns)
 591 {
 592         return kref_get_unless_zero(&ns->kref);
 593 }
 594
 595 void nvme_put_ns(struct nvme_ns *ns)
 596 {
 597         kref_put(&ns->kref, nvme_free_ns);
 598 }
 599 EXPORT_SYMBOL_NS_GPL(nvme_put_ns, NVME_TARGET_PASSTHRU);
 600
 601 static inline void nvme_clear_nvme_request(struct request *req)
 602 {
 603         nvme_req(req)->status = 0;
 604         nvme_req(req)->retries = 0;
 605         nvme_req(req)->flags = 0;
 606         req->rq_flags |= RQF_DONTPREP;
 607 }
 608
 609 static inline unsigned int nvme_req_op(struct nvme_command *cmd)
 610 {
 611         return nvme_is_write(cmd) ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN;
 612 }
 613
 614 static inline void nvme_init_request(struct request *req,
 615                 struct nvme_command *cmd)
 616 {
 617         if (req->q->queuedata)
 618                 req->timeout = NVME_IO_TIMEOUT;
 619         else /* no queuedata implies admin queue */
 620                 req->timeout = NVME_ADMIN_TIMEOUT;
 621
 622         /* passthru commands should let the driver set the SGL flags */
 623         cmd->common.flags &= ~NVME_CMD_SGL_ALL;
 624
 625         req->cmd_flags |= REQ_FAILFAST_DRIVER;
 626         if (req->mq_hctx->type == HCTX_TYPE_POLL)
 627                 req->cmd_flags |= REQ_POLLED;
 628         nvme_clear_nvme_request(req);
 629         memcpy(nvme_req(req)->cmd, cmd, sizeof(*cmd));
 630 }
 631
 632 struct request *nvme_alloc_request(struct request_queue *q,
 633                 struct nvme_command *cmd, blk_mq_req_flags_t flags)
 634 {
 635         struct request *req;
 636
 637         req = blk_mq_alloc_request(q, nvme_req_op(cmd), flags);
 638         if (!IS_ERR(req))
 639                 nvme_init_request(req, cmd);
 640         return req;
 641 }
 642 EXPORT_SYMBOL_GPL(nvme_alloc_request);
 643
 644 static struct request *nvme_alloc_request_qid(struct request_queue *q,
 645                 struct nvme_command *cmd, blk_mq_req_flags_t flags, int qid)
 646 {
 647         struct request *req;
 648
 649         req = blk_mq_alloc_request_hctx(q, nvme_req_op(cmd), flags,
 650                         qid ? qid - 1 : 0);
 651         if (!IS_ERR(req))
 652                 nvme_init_request(req, cmd);
 653         return req;
 654 }
 655
 656 /*
 657  * For something we're not in a state to send to the device the default action
 658  * is to busy it and retry it after the controller state is recovered.  However,
 659  * if the controller is deleting or if anything is marked for failfast or
 660  * nvme multipath it is immediately failed.
 661  *
 662  * Note: commands used to initialize the controller will be marked for failfast.
 663  * Note: nvme cli/ioctl commands are marked for failfast.
 664  */
 665 blk_status_t nvme_fail_nonready_command(struct nvme_ctrl *ctrl,
 666                 struct request *rq)
 667 {
 668         if (ctrl->state != NVME_CTRL_DELETING_NOIO &&
 669             ctrl->state != NVME_CTRL_DEAD &&
 670             !test_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags) &&
 671             !blk_noretry_request(rq) && !(rq->cmd_flags & REQ_NVME_MPATH))
 672                 return BLK_STS_RESOURCE;
 673         return nvme_host_path_error(rq);
 674 }
 675 EXPORT_SYMBOL_GPL(nvme_fail_nonready_command);
 676
 677 bool __nvme_check_ready(struct nvme_ctrl *ctrl, struct request *rq,
 678                 bool queue_live)
 679 {
 680         struct nvme_request *req = nvme_req(rq);
 681
 682         /*
 683          * currently we have a problem sending passthru commands
 684          * on the admin_q if the controller is not LIVE because we can't
 685          * make sure that they are going out after the admin connect,
 686          * controller enable and/or other commands in the initialization
 687          * sequence. until the controller will be LIVE, fail with
 688          * BLK_STS_RESOURCE so that they will be rescheduled.
 689          */
 690         if (rq->q == ctrl->admin_q && (req->flags & NVME_REQ_USERCMD))
 691                 return false;
 692
 693         if (ctrl->ops->flags & NVME_F_FABRICS) {
 694                 /*
 695                  * Only allow commands on a live queue, except for the connect
 696                  * command, which is require to set the queue live in the
 697                  * appropinquate states.
 698                  */
 699                 switch (ctrl->state) {
 700                 case NVME_CTRL_CONNECTING:
 701                         if (blk_rq_is_passthrough(rq) && nvme_is_fabrics(req->cmd) &&
 702                             req->cmd->fabrics.fctype == nvme_fabrics_type_connect)
 703                                 return true;
 704                         break;
 705                 default:
 706                         break;
 707                 case NVME_CTRL_DEAD:
 708                         return false;
 709                 }
 710         }
 711
 712         return queue_live;
 713 }
 714 EXPORT_SYMBOL_GPL(__nvme_check_ready);
 715
 716 static int nvme_toggle_streams(struct nvme_ctrl *ctrl, bool enable)
 717 {
 718         struct nvme_command c = { };
 719
 720         c.directive.opcode = nvme_admin_directive_send;
 721         c.directive.nsid = cpu_to_le32(NVME_NSID_ALL);
 722         c.directive.doper = NVME_DIR_SND_ID_OP_ENABLE;
 723         c.directive.dtype = NVME_DIR_IDENTIFY;
 724         c.directive.tdtype = NVME_DIR_STREAMS;
 725         c.directive.endir = enable ? NVME_DIR_ENDIR : 0;
 726
 727         return nvme_submit_sync_cmd(ctrl->admin_q, &c, NULL, 0);
 728 }
 729
 730 static int nvme_disable_streams(struct nvme_ctrl *ctrl)
 731 {
 732         return nvme_toggle_streams(ctrl, false);
 733 }
 734
 735 static int nvme_enable_streams(struct nvme_ctrl *ctrl)
 736 {
 737         return nvme_toggle_streams(ctrl, true);
 738 }
 739
 740 static int nvme_get_stream_params(struct nvme_ctrl *ctrl,
 741                                   struct streams_directive_params *s, u32 nsid)
 742 {
 743         struct nvme_command c = { };
 744
 745         memset(s, 0, sizeof(*s));
 746
 747         c.directive.opcode = nvme_admin_directive_recv;
 748         c.directive.nsid = cpu_to_le32(nsid);
 749         c.directive.numd = cpu_to_le32(nvme_bytes_to_numd(sizeof(*s)));
 750         c.directive.doper = NVME_DIR_RCV_ST_OP_PARAM;
 751         c.directive.dtype = NVME_DIR_STREAMS;
 752
 753         return nvme_submit_sync_cmd(ctrl->admin_q, &c, s, sizeof(*s));
 754 }
 755
 756 static int nvme_configure_directives(struct nvme_ctrl *ctrl)
 757 {
 758         struct streams_directive_params s;
 759         int ret;
 760
 761         if (!(ctrl->oacs & NVME_CTRL_OACS_DIRECTIVES))
 762                 return 0;
 763         if (!streams)
 764                 return 0;
 765
 766         ret = nvme_enable_streams(ctrl);
 767         if (ret)
 768                 return ret;
 769
 770         ret = nvme_get_stream_params(ctrl, &s, NVME_NSID_ALL);
 771         if (ret)
 772                 goto out_disable_stream;
 773
 774         ctrl->nssa = le16_to_cpu(s.nssa);
 775         if (ctrl->nssa < BLK_MAX_WRITE_HINTS - 1) {
 776                 dev_info(ctrl->device, "too few streams (%u) available\n",
 777                                         ctrl->nssa);
 778                 goto out_disable_stream;
 779         }
 780
 781         ctrl->nr_streams = min_t(u16, ctrl->nssa, BLK_MAX_WRITE_HINTS - 1);
 782         dev_info(ctrl->device, "Using %u streams\n", ctrl->nr_streams);
 783         return 0;
 784
 785 out_disable_stream:
 786         nvme_disable_streams(ctrl);
 787         return ret;
 788 }
 789
 790 /*
 791  * Check if 'req' has a write hint associated with it. If it does, assign
 792  * a valid namespace stream to the write.
 793  */
 794 static void nvme_assign_write_stream(struct nvme_ctrl *ctrl,
 795                                      struct request *req, u16 *control,
 796                                      u32 *dsmgmt)
 797 {
 798         enum rw_hint streamid = req->write_hint;
 799
 800         if (streamid == WRITE_LIFE_NOT_SET || streamid == WRITE_LIFE_NONE)
 801                 streamid = 0;
 802         else {
 803                 streamid--;
 804                 if (WARN_ON_ONCE(streamid > ctrl->nr_streams))
 805                         return;
 806
 807                 *control |= NVME_RW_DTYPE_STREAMS;
 808                 *dsmgmt |= streamid << 16;
 809         }
 810
 811         if (streamid < ARRAY_SIZE(req->q->write_hints))
 812                 req->q->write_hints[streamid] += blk_rq_bytes(req) >> 9;
 813 }
 814
 815 static inline void nvme_setup_flush(struct nvme_ns *ns,
 816                 struct nvme_command *cmnd)
 817 {
 818         cmnd->common.opcode = nvme_cmd_flush;
 819         cmnd->common.nsid = cpu_to_le32(ns->head->ns_id);
 820 }
 821
 822 static blk_status_t nvme_setup_discard(struct nvme_ns *ns, struct request *req,
 823                 struct nvme_command *cmnd)
 824 {
 825         unsigned short segments = blk_rq_nr_discard_segments(req), n = 0;
 826         struct nvme_dsm_range *range;
 827         struct bio *bio;
 828
 829         /*
 830          * Some devices do not consider the DSM 'Number of Ranges' field when
 831          * determining how much data to DMA. Always allocate memory for maximum
 832          * number of segments to prevent device reading beyond end of buffer.
 833          */
 834         static const size_t alloc_size = sizeof(*range) * NVME_DSM_MAX_RANGES;
 835
 836         range = kzalloc(alloc_size, GFP_ATOMIC | __GFP_NOWARN);
 837         if (!range) {
 838                 /*
 839                  * If we fail allocation our range, fallback to the controller
 840                  * discard page. If that's also busy, it's safe to return
 841                  * busy, as we know we can make progress once that's freed.
 842                  */
 843                 if (test_and_set_bit_lock(0, &ns->ctrl->discard_page_busy))
 844                         return BLK_STS_RESOURCE;
 845
 846                 range = page_address(ns->ctrl->discard_page);
 847         }
 848
 849         __rq_for_each_bio(bio, req) {
 850                 u64 slba = nvme_sect_to_lba(ns, bio->bi_iter.bi_sector);
 851                 u32 nlb = bio->bi_iter.bi_size >> ns->lba_shift;
 852
 853                 if (n < segments) {
 854                         range[n].cattr = cpu_to_le32(0);
 855                         range[n].nlb = cpu_to_le32(nlb);
 856                         range[n].slba = cpu_to_le64(slba);
 857                 }
 858                 n++;
 859         }
 860
 861         if (WARN_ON_ONCE(n != segments)) {
 862                 if (virt_to_page(range) == ns->ctrl->discard_page)
 863                         clear_bit_unlock(0, &ns->ctrl->discard_page_busy);
 864                 else
 865                         kfree(range);
 866                 return BLK_STS_IOERR;
 867         }
 868
 869         cmnd->dsm.opcode = nvme_cmd_dsm;
 870         cmnd->dsm.nsid = cpu_to_le32(ns->head->ns_id);
 871         cmnd->dsm.nr = cpu_to_le32(segments - 1);
 872         cmnd->dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD);
 873
 874         req->special_vec.bv_page = virt_to_page(range);
 875         req->special_vec.bv_offset = offset_in_page(range);
 876         req->special_vec.bv_len = alloc_size;
 877         req->rq_flags |= RQF_SPECIAL_PAYLOAD;
 878
 879         return BLK_STS_OK;
 880 }
 881
 882 static inline blk_status_t nvme_setup_write_zeroes(struct nvme_ns *ns,
 883                 struct request *req, struct nvme_command *cmnd)
 884 {
 885         if (ns->ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
 886                 return nvme_setup_discard(ns, req, cmnd);
 887
 888         cmnd->write_zeroes.opcode = nvme_cmd_write_zeroes;
 889         cmnd->write_zeroes.nsid = cpu_to_le32(ns->head->ns_id);
 890         cmnd->write_zeroes.slba =
 891                 cpu_to_le64(nvme_sect_to_lba(ns, blk_rq_pos(req)));
 892         cmnd->write_zeroes.length =
 893                 cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
 894         if (nvme_ns_has_pi(ns))
 895                 cmnd->write_zeroes.control = cpu_to_le16(NVME_RW_PRINFO_PRACT);
 896         else
 897                 cmnd->write_zeroes.control = 0;
 898         return BLK_STS_OK;
 899 }
 900
 901 static inline blk_status_t nvme_setup_rw(struct nvme_ns *ns,
 902                 struct request *req, struct nvme_command *cmnd,
 903                 enum nvme_opcode op)
 904 {
 905         struct nvme_ctrl *ctrl = ns->ctrl;
 906         u16 control = 0;
 907         u32 dsmgmt = 0;
 908
 909         if (req->cmd_flags & REQ_FUA)
 910                 control |= NVME_RW_FUA;
 911         if (req->cmd_flags & (REQ_FAILFAST_DEV | REQ_RAHEAD))
 912                 control |= NVME_RW_LR;
 913
 914         if (req->cmd_flags & REQ_RAHEAD)
 915                 dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH;
 916
 917         cmnd->rw.opcode = op;
 918         cmnd->rw.nsid = cpu_to_le32(ns->head->ns_id);
 919         cmnd->rw.slba = cpu_to_le64(nvme_sect_to_lba(ns, blk_rq_pos(req)));
 920         cmnd->rw.length = cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
 921
 922         if (req_op(req) == REQ_OP_WRITE && ctrl->nr_streams)
 923                 nvme_assign_write_stream(ctrl, req, &control, &dsmgmt);
 924
 925         if (ns->ms) {
 926                 /*
 927                  * If formated with metadata, the block layer always provides a
 928                  * metadata buffer if CONFIG_BLK_DEV_INTEGRITY is enabled.  Else
 929                  * we enable the PRACT bit for protection information or set the
 930                  * namespace capacity to zero to prevent any I/O.
 931                  */
 932                 if (!blk_integrity_rq(req)) {
 933                         if (WARN_ON_ONCE(!nvme_ns_has_pi(ns)))
 934                                 return BLK_STS_NOTSUPP;
 935                         control |= NVME_RW_PRINFO_PRACT;
 936                 }
 937
 938                 switch (ns->pi_type) {
 939                 case NVME_NS_DPS_PI_TYPE3:
 940                         control |= NVME_RW_PRINFO_PRCHK_GUARD;
 941                         break;
 942                 case NVME_NS_DPS_PI_TYPE1:
 943                 case NVME_NS_DPS_PI_TYPE2:
 944                         control |= NVME_RW_PRINFO_PRCHK_GUARD |
 945                                         NVME_RW_PRINFO_PRCHK_REF;
 946                         if (op == nvme_cmd_zone_append)
 947                                 control |= NVME_RW_APPEND_PIREMAP;
 948                         cmnd->rw.reftag = cpu_to_le32(t10_pi_ref_tag(req));
 949                         break;
 950                 }
 951         }
 952
 953         cmnd->rw.control = cpu_to_le16(control);
 954         cmnd->rw.dsmgmt = cpu_to_le32(dsmgmt);
 955         return 0;
 956 }
 957
 958 void nvme_cleanup_cmd(struct request *req)
 959 {
 960         if (req->rq_flags & RQF_SPECIAL_PAYLOAD) {
 961                 struct nvme_ctrl *ctrl = nvme_req(req)->ctrl;
 962
 963                 if (req->special_vec.bv_page == ctrl->discard_page)
 964                         clear_bit_unlock(0, &ctrl->discard_page_busy);
 965                 else
 966                         kfree(bvec_virt(&req->special_vec));
 967         }
 968 }
 969 EXPORT_SYMBOL_GPL(nvme_cleanup_cmd);
 970
 971 blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req)
 972 {
 973         struct nvme_command *cmd = nvme_req(req)->cmd;
 974         struct nvme_ctrl *ctrl = nvme_req(req)->ctrl;
 975         blk_status_t ret = BLK_STS_OK;
 976
 977         if (!(req->rq_flags & RQF_DONTPREP)) {
 978                 nvme_clear_nvme_request(req);
 979                 memset(cmd, 0, sizeof(*cmd));
 980         }
 981
 982         switch (req_op(req)) {
 983         case REQ_OP_DRV_IN:
 984         case REQ_OP_DRV_OUT:
 985                 /* these are setup prior to execution in nvme_init_request() */
 986                 break;
 987         case REQ_OP_FLUSH:
 988                 nvme_setup_flush(ns, cmd);
 989                 break;
 990         case REQ_OP_ZONE_RESET_ALL:
 991         case REQ_OP_ZONE_RESET:
 992                 ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_RESET);
 993                 break;
 994         case REQ_OP_ZONE_OPEN:
 995                 ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_OPEN);
 996                 break;
 997         case REQ_OP_ZONE_CLOSE:
 998                 ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_CLOSE);
 999                 break;
1000         case REQ_OP_ZONE_FINISH:
1001                 ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_FINISH);
1002                 break;
1003         case REQ_OP_WRITE_ZEROES:
1004                 ret = nvme_setup_write_zeroes(ns, req, cmd);
1005                 break;
1006         case REQ_OP_DISCARD:
1007                 ret = nvme_setup_discard(ns, req, cmd);
1008                 break;
1009         case REQ_OP_READ:
1010                 ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_read);
1011                 break;
1012         case REQ_OP_WRITE:
1013                 ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_write);
1014                 break;
1015         case REQ_OP_ZONE_APPEND:
1016                 ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_zone_append);
1017                 break;
1018         default:
1019                 WARN_ON_ONCE(1);
1020                 return BLK_STS_IOERR;
1021         }
1022
1023         if (!(ctrl->quirks & NVME_QUIRK_SKIP_CID_GEN))
1024                 nvme_req(req)->genctr++;
1025         cmd->common.command_id = nvme_cid(req);
1026         trace_nvme_setup_cmd(req, cmd);
1027         return ret;
1028 }
1029 EXPORT_SYMBOL_GPL(nvme_setup_cmd);
1030
1031 /*
1032  * Return values:
1033  * 0:  success
1034  * >0: nvme controller's cqe status response
1035  * <0: kernel error in lieu of controller response
1036  */
1037 static int nvme_execute_rq(struct gendisk *disk, struct request *rq,
1038                 bool at_head)
1039 {
1040         blk_status_t status;
1041
1042         status = blk_execute_rq(disk, rq, at_head);
1043         if (nvme_req(rq)->flags & NVME_REQ_CANCELLED)
1044                 return -EINTR;
1045         if (nvme_req(rq)->status)
1046                 return nvme_req(rq)->status;
1047         return blk_status_to_errno(status);
1048 }
1049
1050 /*
1051  * Returns 0 on success.  If the result is negative, it's a Linux error code;
1052  * if the result is positive, it's an NVM Express status code
1053  */
1054 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
1055                 union nvme_result *result, void *buffer, unsigned bufflen,
1056                 unsigned timeout, int qid, int at_head,
1057                 blk_mq_req_flags_t flags)
1058 {
1059         struct request *req;
1060         int ret;
1061
1062         if (qid == NVME_QID_ANY)
1063                 req = nvme_alloc_request(q, cmd, flags);
1064         else
1065                 req = nvme_alloc_request_qid(q, cmd, flags, qid);
1066         if (IS_ERR(req))
1067                 return PTR_ERR(req);
1068
1069         if (timeout)
1070                 req->timeout = timeout;
1071
1072         if (buffer && bufflen) {
1073                 ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL);
1074                 if (ret)
1075                         goto out;
1076         }
1077
1078         ret = nvme_execute_rq(NULL, req, at_head);
1079         if (result && ret >= 0)
1080                 *result = nvme_req(req)->result;
1081  out:
1082         blk_mq_free_request(req);
1083         return ret;
1084 }
1085 EXPORT_SYMBOL_GPL(__nvme_submit_sync_cmd);
1086
1087 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
1088                 void *buffer, unsigned bufflen)
1089 {
1090         return __nvme_submit_sync_cmd(q, cmd, NULL, buffer, bufflen, 0,
1091                         NVME_QID_ANY, 0, 0);
1092 }
1093 EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd);
1094
1095 static u32 nvme_known_admin_effects(u8 opcode)
1096 {
1097         switch (opcode) {
1098         case nvme_admin_format_nvm:
1099                 return NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_NCC |
1100                         NVME_CMD_EFFECTS_CSE_MASK;
1101         case nvme_admin_sanitize_nvm:
1102                 return NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK;
1103         default:
1104                 break;
1105         }
1106         return 0;
1107 }
1108
1109 u32 nvme_command_effects(struct nvme_ctrl *ctrl, struct nvme_ns *ns, u8 opcode)
1110 {
1111         u32 effects = 0;
1112
1113         if (ns) {
1114                 if (ns->head->effects)
1115                         effects = le32_to_cpu(ns->head->effects->iocs[opcode]);
1116                 if (effects & ~(NVME_CMD_EFFECTS_CSUPP | NVME_CMD_EFFECTS_LBCC))
1117                         dev_warn_once(ctrl->device,
1118                                 "IO command:%02x has unhandled effects:%08x\n",
1119                                 opcode, effects);
1120                 return 0;
1121         }
1122
1123         if (ctrl->effects)
1124                 effects = le32_to_cpu(ctrl->effects->acs[opcode]);
1125         effects |= nvme_known_admin_effects(opcode);
1126
1127         return effects;
1128 }
1129 EXPORT_SYMBOL_NS_GPL(nvme_command_effects, NVME_TARGET_PASSTHRU);
1130
1131 static u32 nvme_passthru_start(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1132                                u8 opcode)
1133 {
1134         u32 effects = nvme_command_effects(ctrl, ns, opcode);
1135
1136         /*
1137          * For simplicity, IO to all namespaces is quiesced even if the command
1138          * effects say only one namespace is affected.
1139          */
1140         if (effects & NVME_CMD_EFFECTS_CSE_MASK) {
1141                 mutex_lock(&ctrl->scan_lock);
1142                 mutex_lock(&ctrl->subsys->lock);
1143                 nvme_mpath_start_freeze(ctrl->subsys);
1144                 nvme_mpath_wait_freeze(ctrl->subsys);
1145                 nvme_start_freeze(ctrl);
1146                 nvme_wait_freeze(ctrl);
1147         }
1148         return effects;
1149 }
1150
1151 static void nvme_passthru_end(struct nvme_ctrl *ctrl, u32 effects,
1152                               struct nvme_command *cmd, int status)
1153 {
1154         if (effects & NVME_CMD_EFFECTS_CSE_MASK) {
1155                 nvme_unfreeze(ctrl);
1156                 nvme_mpath_unfreeze(ctrl->subsys);
1157                 mutex_unlock(&ctrl->subsys->lock);
1158                 nvme_remove_invalid_namespaces(ctrl, NVME_NSID_ALL);
1159                 mutex_unlock(&ctrl->scan_lock);
1160         }
1161         if (effects & NVME_CMD_EFFECTS_CCC)
1162                 nvme_init_ctrl_finish(ctrl);
1163         if (effects & (NVME_CMD_EFFECTS_NIC | NVME_CMD_EFFECTS_NCC)) {
1164                 nvme_queue_scan(ctrl);
1165                 flush_work(&ctrl->scan_work);
1166         }
1167
1168         switch (cmd->common.opcode) {
1169         case nvme_admin_set_features:
1170                 switch (le32_to_cpu(cmd->common.cdw10) & 0xFF) {
1171                 case NVME_FEAT_KATO:
1172                         /*
1173                          * Keep alive commands interval on the host should be
1174                          * updated when KATO is modified by Set Features
1175                          * commands.
1176                          */
1177                         if (!status)
1178                                 nvme_update_keep_alive(ctrl, cmd);
1179                         break;
1180                 default:
1181                         break;
1182                 }
1183                 break;
1184         default:
1185                 break;
1186         }
1187 }
1188
1189 int nvme_execute_passthru_rq(struct request *rq)
1190 {
1191         struct nvme_command *cmd = nvme_req(rq)->cmd;
1192         struct nvme_ctrl *ctrl = nvme_req(rq)->ctrl;
1193         struct nvme_ns *ns = rq->q->queuedata;
1194         struct gendisk *disk = ns ? ns->disk : NULL;
1195         u32 effects;
1196         int  ret;
1197
1198         effects = nvme_passthru_start(ctrl, ns, cmd->common.opcode);
1199         ret = nvme_execute_rq(disk, rq, false);
1200         if (effects) /* nothing to be done for zero cmd effects */
1201                 nvme_passthru_end(ctrl, effects, cmd, ret);
1202
1203         return ret;
1204 }
1205 EXPORT_SYMBOL_NS_GPL(nvme_execute_passthru_rq, NVME_TARGET_PASSTHRU);
1206
1207 /*
1208  * Recommended frequency for KATO commands per NVMe 1.4 section 7.12.1:
1209  *
1210  *   The host should send Keep Alive commands at half of the Keep Alive Timeout
1211  *   accounting for transport roundtrip times [..].
1212  */
1213 static void nvme_queue_keep_alive_work(struct nvme_ctrl *ctrl)
1214 {
1215         queue_delayed_work(nvme_wq, &ctrl->ka_work, ctrl->kato * HZ / 2);
1216 }
1217
1218 static void nvme_keep_alive_end_io(struct request *rq, blk_status_t status)
1219 {
1220         struct nvme_ctrl *ctrl = rq->end_io_data;
1221         unsigned long flags;
1222         bool startka = false;
1223
1224         blk_mq_free_request(rq);
1225
1226         if (status) {
1227                 dev_err(ctrl->device,
1228                         "failed nvme_keep_alive_end_io error=%d\n",
1229                                 status);
1230                 return;
1231         }
1232
1233         ctrl->comp_seen = false;
1234         spin_lock_irqsave(&ctrl->lock, flags);
1235         if (ctrl->state == NVME_CTRL_LIVE ||
1236             ctrl->state == NVME_CTRL_CONNECTING)
1237                 startka = true;
1238         spin_unlock_irqrestore(&ctrl->lock, flags);
1239         if (startka)
1240                 nvme_queue_keep_alive_work(ctrl);
1241 }
1242
1243 static void nvme_keep_alive_work(struct work_struct *work)
1244 {
1245         struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
1246                         struct nvme_ctrl, ka_work);
1247         bool comp_seen = ctrl->comp_seen;
1248         struct request *rq;
1249
1250         if ((ctrl->ctratt & NVME_CTRL_ATTR_TBKAS) && comp_seen) {
1251                 dev_dbg(ctrl->device,
1252                         "reschedule traffic based keep-alive timer\n");
1253                 ctrl->comp_seen = false;
1254                 nvme_queue_keep_alive_work(ctrl);
1255                 return;
1256         }
1257
1258         rq = nvme_alloc_request(ctrl->admin_q, &ctrl->ka_cmd,
1259                                 BLK_MQ_REQ_RESERVED | BLK_MQ_REQ_NOWAIT);
1260         if (IS_ERR(rq)) {
1261                 /* allocation failure, reset the controller */
1262                 dev_err(ctrl->device, "keep-alive failed: %ld\n", PTR_ERR(rq));
1263                 nvme_reset_ctrl(ctrl);
1264                 return;
1265         }
1266
1267         rq->timeout = ctrl->kato * HZ;
1268         rq->end_io_data = ctrl;
1269         blk_execute_rq_nowait(NULL, rq, 0, nvme_keep_alive_end_io);
1270 }
1271
1272 static void nvme_start_keep_alive(struct nvme_ctrl *ctrl)
1273 {
1274         if (unlikely(ctrl->kato == 0))
1275                 return;
1276
1277         nvme_queue_keep_alive_work(ctrl);
1278 }
1279
1280 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl)
1281 {
1282         if (unlikely(ctrl->kato == 0))
1283                 return;
1284
1285         cancel_delayed_work_sync(&ctrl->ka_work);
1286 }
1287 EXPORT_SYMBOL_GPL(nvme_stop_keep_alive);
1288
1289 static void nvme_update_keep_alive(struct nvme_ctrl *ctrl,
1290                                    struct nvme_command *cmd)
1291 {
1292         unsigned int new_kato =
1293                 DIV_ROUND_UP(le32_to_cpu(cmd->common.cdw11), 1000);
1294
1295         dev_info(ctrl->device,
1296                  "keep alive interval updated from %u ms to %u ms\n",
1297                  ctrl->kato * 1000 / 2, new_kato * 1000 / 2);
1298
1299         nvme_stop_keep_alive(ctrl);
1300         ctrl->kato = new_kato;
1301         nvme_start_keep_alive(ctrl);
1302 }
1303
1304 /*
1305  * In NVMe 1.0 the CNS field was just a binary controller or namespace
1306  * flag, thus sending any new CNS opcodes has a big chance of not working.
1307  * Qemu unfortunately had that bug after reporting a 1.1 version compliance
1308  * (but not for any later version).
1309  */
1310 static bool nvme_ctrl_limited_cns(struct nvme_ctrl *ctrl)
1311 {
1312         if (ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)
1313                 return ctrl->vs < NVME_VS(1, 2, 0);
1314         return ctrl->vs < NVME_VS(1, 1, 0);
1315 }
1316
1317 static int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
1318 {
1319         struct nvme_command c = { };
1320         int error;
1321
1322         /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1323         c.identify.opcode = nvme_admin_identify;
1324         c.identify.cns = NVME_ID_CNS_CTRL;
1325
1326         *id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
1327         if (!*id)
1328                 return -ENOMEM;
1329
1330         error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
1331                         sizeof(struct nvme_id_ctrl));
1332         if (error)
1333                 kfree(*id);
1334         return error;
1335 }
1336
1337 static int nvme_process_ns_desc(struct nvme_ctrl *ctrl, struct nvme_ns_ids *ids,
1338                 struct nvme_ns_id_desc *cur, bool *csi_seen)
1339 {
1340         const char *warn_str = "ctrl returned bogus length:";
1341         void *data = cur;
1342
1343         switch (cur->nidt) {
1344         case NVME_NIDT_EUI64:
1345                 if (cur->nidl != NVME_NIDT_EUI64_LEN) {
1346                         dev_warn(ctrl->device, "%s %d for NVME_NIDT_EUI64\n",
1347                                  warn_str, cur->nidl);
1348                         return -1;
1349                 }
1350                 memcpy(ids->eui64, data + sizeof(*cur), NVME_NIDT_EUI64_LEN);
1351                 return NVME_NIDT_EUI64_LEN;
1352         case NVME_NIDT_NGUID:
1353                 if (cur->nidl != NVME_NIDT_NGUID_LEN) {
1354                         dev_warn(ctrl->device, "%s %d for NVME_NIDT_NGUID\n",
1355                                  warn_str, cur->nidl);
1356                         return -1;
1357                 }
1358                 memcpy(ids->nguid, data + sizeof(*cur), NVME_NIDT_NGUID_LEN);
1359                 return NVME_NIDT_NGUID_LEN;
1360         case NVME_NIDT_UUID:
1361                 if (cur->nidl != NVME_NIDT_UUID_LEN) {
1362                         dev_warn(ctrl->device, "%s %d for NVME_NIDT_UUID\n",
1363                                  warn_str, cur->nidl);
1364                         return -1;
1365                 }
1366                 uuid_copy(&ids->uuid, data + sizeof(*cur));
1367                 return NVME_NIDT_UUID_LEN;
1368         case NVME_NIDT_CSI:
1369                 if (cur->nidl != NVME_NIDT_CSI_LEN) {
1370                         dev_warn(ctrl->device, "%s %d for NVME_NIDT_CSI\n",
1371                                  warn_str, cur->nidl);
1372                         return -1;
1373                 }
1374                 memcpy(&ids->csi, data + sizeof(*cur), NVME_NIDT_CSI_LEN);
1375                 *csi_seen = true;
1376                 return NVME_NIDT_CSI_LEN;
1377         default:
1378                 /* Skip unknown types */
1379                 return cur->nidl;
1380         }
1381 }
1382
1383 static int nvme_identify_ns_descs(struct nvme_ctrl *ctrl, unsigned nsid,
1384                 struct nvme_ns_ids *ids)
1385 {
1386         struct nvme_command c = { };
1387         bool csi_seen = false;
1388         int status, pos, len;
1389         void *data;
1390
1391         if (ctrl->vs < NVME_VS(1, 3, 0) && !nvme_multi_css(ctrl))
1392                 return 0;
1393         if (ctrl->quirks & NVME_QUIRK_NO_NS_DESC_LIST)
1394                 return 0;
1395
1396         c.identify.opcode = nvme_admin_identify;
1397         c.identify.nsid = cpu_to_le32(nsid);
1398         c.identify.cns = NVME_ID_CNS_NS_DESC_LIST;
1399
1400         data = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
1401         if (!data)
1402                 return -ENOMEM;
1403
1404         status = nvme_submit_sync_cmd(ctrl->admin_q, &c, data,
1405                                       NVME_IDENTIFY_DATA_SIZE);
1406         if (status) {
1407                 dev_warn(ctrl->device,
1408                         "Identify Descriptors failed (nsid=%u, status=0x%x)\n",
1409                         nsid, status);
1410                 goto free_data;
1411         }
1412
1413         for (pos = 0; pos < NVME_IDENTIFY_DATA_SIZE; pos += len) {
1414                 struct nvme_ns_id_desc *cur = data + pos;
1415
1416                 if (cur->nidl == 0)
1417                         break;
1418
1419                 len = nvme_process_ns_desc(ctrl, ids, cur, &csi_seen);
1420                 if (len < 0)
1421                         break;
1422
1423                 len += sizeof(*cur);
1424         }
1425
1426         if (nvme_multi_css(ctrl) && !csi_seen) {
1427                 dev_warn(ctrl->device, "Command set not reported for nsid:%d\n",
1428                          nsid);
1429                 status = -EINVAL;
1430         }
1431
1432 free_data:
1433         kfree(data);
1434         return status;
1435 }
1436
1437 static int nvme_identify_ns(struct nvme_ctrl *ctrl, unsigned nsid,
1438                         struct nvme_ns_ids *ids, struct nvme_id_ns **id)
1439 {
1440         struct nvme_command c = { };
1441         int error;
1442
1443         /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1444         c.identify.opcode = nvme_admin_identify;
1445         c.identify.nsid = cpu_to_le32(nsid);
1446         c.identify.cns = NVME_ID_CNS_NS;
1447
1448         *id = kmalloc(sizeof(**id), GFP_KERNEL);
1449         if (!*id)
1450                 return -ENOMEM;
1451
1452         error = nvme_submit_sync_cmd(ctrl->admin_q, &c, *id, sizeof(**id));
1453         if (error) {
1454                 dev_warn(ctrl->device, "Identify namespace failed (%d)\n", error);
1455                 goto out_free_id;
1456         }
1457
1458         error = NVME_SC_INVALID_NS | NVME_SC_DNR;
1459         if ((*id)->ncap == 0) /* namespace not allocated or attached */
1460                 goto out_free_id;
1461
1462         if (ctrl->vs >= NVME_VS(1, 1, 0) &&
1463             !memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
1464                 memcpy(ids->eui64, (*id)->eui64, sizeof(ids->eui64));
1465         if (ctrl->vs >= NVME_VS(1, 2, 0) &&
1466             !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
1467                 memcpy(ids->nguid, (*id)->nguid, sizeof(ids->nguid));
1468
1469         return 0;
1470
1471 out_free_id:
1472         kfree(*id);
1473         return error;
1474 }
1475
1476 static int nvme_features(struct nvme_ctrl *dev, u8 op, unsigned int fid,
1477                 unsigned int dword11, void *buffer, size_t buflen, u32 *result)
1478 {
1479         union nvme_result res = { 0 };
1480         struct nvme_command c = { };
1481         int ret;
1482
1483         c.features.opcode = op;
1484         c.features.fid = cpu_to_le32(fid);
1485         c.features.dword11 = cpu_to_le32(dword11);
1486
1487         ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &res,
1488                         buffer, buflen, 0, NVME_QID_ANY, 0, 0);
1489         if (ret >= 0 && result)
1490                 *result = le32_to_cpu(res.u32);
1491         return ret;
1492 }
1493
1494 int nvme_set_features(struct nvme_ctrl *dev, unsigned int fid,
1495                       unsigned int dword11, void *buffer, size_t buflen,
1496                       u32 *result)
1497 {
1498         return nvme_features(dev, nvme_admin_set_features, fid, dword11, buffer,
1499                              buflen, result);
1500 }
1501 EXPORT_SYMBOL_GPL(nvme_set_features);
1502
1503 int nvme_get_features(struct nvme_ctrl *dev, unsigned int fid,
1504                       unsigned int dword11, void *buffer, size_t buflen,
1505                       u32 *result)
1506 {
1507         return nvme_features(dev, nvme_admin_get_features, fid, dword11, buffer,
1508                              buflen, result);
1509 }
1510 EXPORT_SYMBOL_GPL(nvme_get_features);
1511
1512 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count)
1513 {
1514         u32 q_count = (*count - 1) | ((*count - 1) << 16);
1515         u32 result;
1516         int status, nr_io_queues;
1517
1518         status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, NULL, 0,
1519                         &result);
1520         if (status < 0)
1521                 return status;
1522
1523         /*
1524          * Degraded controllers might return an error when setting the queue
1525          * count.  We still want to be able to bring them online and offer
1526          * access to the admin queue, as that might be only way to fix them up.
1527          */
1528         if (status > 0) {
1529                 dev_err(ctrl->device, "Could not set queue count (%d)\n", status);
1530                 *count = 0;
1531         } else {
1532                 nr_io_queues = min(result & 0xffff, result >> 16) + 1;
1533                 *count = min(*count, nr_io_queues);
1534         }
1535
1536         return 0;
1537 }
1538 EXPORT_SYMBOL_GPL(nvme_set_queue_count);
1539
1540 #define NVME_AEN_SUPPORTED \
1541         (NVME_AEN_CFG_NS_ATTR | NVME_AEN_CFG_FW_ACT | \
1542          NVME_AEN_CFG_ANA_CHANGE | NVME_AEN_CFG_DISC_CHANGE)
1543
1544 static void nvme_enable_aen(struct nvme_ctrl *ctrl)
1545 {
1546         u32 result, supported_aens = ctrl->oaes & NVME_AEN_SUPPORTED;
1547         int status;
1548
1549         if (!supported_aens)
1550                 return;
1551
1552         status = nvme_set_features(ctrl, NVME_FEAT_ASYNC_EVENT, supported_aens,
1553                         NULL, 0, &result);
1554         if (status)
1555                 dev_warn(ctrl->device, "Failed to configure AEN (cfg %x)\n",
1556                          supported_aens);
1557
1558         queue_work(nvme_wq, &ctrl->async_event_work);
1559 }
1560
1561 static int nvme_ns_open(struct nvme_ns *ns)
1562 {
1563
1564         /* should never be called due to GENHD_FL_HIDDEN */
1565         if (WARN_ON_ONCE(nvme_ns_head_multipath(ns->head)))
1566                 goto fail;
1567         if (!nvme_get_ns(ns))
1568                 goto fail;
1569         if (!try_module_get(ns->ctrl->ops->module))
1570                 goto fail_put_ns;
1571
1572         return 0;
1573
1574 fail_put_ns:
1575         nvme_put_ns(ns);
1576 fail:
1577         return -ENXIO;
1578 }
1579
1580 static void nvme_ns_release(struct nvme_ns *ns)
1581 {
1582
1583         module_put(ns->ctrl->ops->module);
1584         nvme_put_ns(ns);
1585 }
1586
1587 static int nvme_open(struct block_device *bdev, fmode_t mode)
1588 {
1589         return nvme_ns_open(bdev->bd_disk->private_data);
1590 }
1591
1592 static void nvme_release(struct gendisk *disk, fmode_t mode)
1593 {
1594         nvme_ns_release(disk->private_data);
1595 }
1596
1597 int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1598 {
1599         /* some standard values */
1600         geo->heads = 1 << 6;
1601         geo->sectors = 1 << 5;
1602         geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
1603         return 0;
1604 }
1605
1606 #ifdef CONFIG_BLK_DEV_INTEGRITY
1607 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type,
1608                                 u32 max_integrity_segments)
1609 {
1610         struct blk_integrity integrity = { };
1611
1612         switch (pi_type) {
1613         case NVME_NS_DPS_PI_TYPE3:
1614                 integrity.profile = &t10_pi_type3_crc;
1615                 integrity.tag_size = sizeof(u16) + sizeof(u32);
1616                 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1617                 break;
1618         case NVME_NS_DPS_PI_TYPE1:
1619         case NVME_NS_DPS_PI_TYPE2:
1620                 integrity.profile = &t10_pi_type1_crc;
1621                 integrity.tag_size = sizeof(u16);
1622                 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1623                 break;
1624         default:
1625                 integrity.profile = NULL;
1626                 break;
1627         }
1628         integrity.tuple_size = ms;
1629         blk_integrity_register(disk, &integrity);
1630         blk_queue_max_integrity_segments(disk->queue, max_integrity_segments);
1631 }
1632 #else
1633 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type,
1634                                 u32 max_integrity_segments)
1635 {
1636 }
1637 #endif /* CONFIG_BLK_DEV_INTEGRITY */
1638
1639 static void nvme_config_discard(struct gendisk *disk, struct nvme_ns *ns)
1640 {
1641         struct nvme_ctrl *ctrl = ns->ctrl;
1642         struct request_queue *queue = disk->queue;
1643         u32 size = queue_logical_block_size(queue);
1644
1645         if (ctrl->max_discard_sectors == 0) {
1646                 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, queue);
1647                 return;
1648         }
1649
1650         if (ctrl->nr_streams && ns->sws && ns->sgs)
1651                 size *= ns->sws * ns->sgs;
1652
1653         BUILD_BUG_ON(PAGE_SIZE / sizeof(struct nvme_dsm_range) <
1654                         NVME_DSM_MAX_RANGES);
1655
1656         queue->limits.discard_alignment = 0;
1657         queue->limits.discard_granularity = size;
1658
1659         /* If discard is already enabled, don't reset queue limits */
1660         if (blk_queue_flag_test_and_set(QUEUE_FLAG_DISCARD, queue))
1661                 return;
1662
1663         blk_queue_max_discard_sectors(queue, ctrl->max_discard_sectors);
1664         blk_queue_max_discard_segments(queue, ctrl->max_discard_segments);
1665
1666         if (ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
1667                 blk_queue_max_write_zeroes_sectors(queue, UINT_MAX);
1668 }
1669
1670 static bool nvme_ns_ids_valid(struct nvme_ns_ids *ids)
1671 {
1672         return !uuid_is_null(&ids->uuid) ||
1673                 memchr_inv(ids->nguid, 0, sizeof(ids->nguid)) ||
1674                 memchr_inv(ids->eui64, 0, sizeof(ids->eui64));
1675 }
1676
1677 static bool nvme_ns_ids_equal(struct nvme_ns_ids *a, struct nvme_ns_ids *b)
1678 {
1679         return uuid_equal(&a->uuid, &b->uuid) &&
1680                 memcmp(&a->nguid, &b->nguid, sizeof(a->nguid)) == 0 &&
1681                 memcmp(&a->eui64, &b->eui64, sizeof(a->eui64)) == 0 &&
1682                 a->csi == b->csi;
1683 }
1684
1685 static int nvme_setup_streams_ns(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1686                                  u32 *phys_bs, u32 *io_opt)
1687 {
1688         struct streams_directive_params s;
1689         int ret;
1690
1691         if (!ctrl->nr_streams)
1692                 return 0;
1693
1694         ret = nvme_get_stream_params(ctrl, &s, ns->head->ns_id);
1695         if (ret)
1696                 return ret;
1697
1698         ns->sws = le32_to_cpu(s.sws);
1699         ns->sgs = le16_to_cpu(s.sgs);
1700
1701         if (ns->sws) {
1702                 *phys_bs = ns->sws * (1 << ns->lba_shift);
1703                 if (ns->sgs)
1704                         *io_opt = *phys_bs * ns->sgs;
1705         }
1706
1707         return 0;
1708 }
1709
1710 static int nvme_configure_metadata(struct nvme_ns *ns, struct nvme_id_ns *id)
1711 {
1712         struct nvme_ctrl *ctrl = ns->ctrl;
1713
1714         /*
1715          * The PI implementation requires the metadata size to be equal to the
1716          * t10 pi tuple size.
1717          */
1718         ns->ms = le16_to_cpu(id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ms);
1719         if (ns->ms == sizeof(struct t10_pi_tuple))
1720                 ns->pi_type = id->dps & NVME_NS_DPS_PI_MASK;
1721         else
1722                 ns->pi_type = 0;
1723
1724         ns->features &= ~(NVME_NS_METADATA_SUPPORTED | NVME_NS_EXT_LBAS);
1725         if (!ns->ms || !(ctrl->ops->flags & NVME_F_METADATA_SUPPORTED))
1726                 return 0;
1727         if (ctrl->ops->flags & NVME_F_FABRICS) {
1728                 /*
1729                  * The NVMe over Fabrics specification only supports metadata as
1730                  * part of the extended data LBA.  We rely on HCA/HBA support to
1731                  * remap the separate metadata buffer from the block layer.
1732                  */
1733                 if (WARN_ON_ONCE(!(id->flbas & NVME_NS_FLBAS_META_EXT)))
1734                         return -EINVAL;
1735                 if (ctrl->max_integrity_segments)
1736                         ns->features |=
1737                                 (NVME_NS_METADATA_SUPPORTED | NVME_NS_EXT_LBAS);
1738         } else {
1739                 /*
1740                  * For PCIe controllers, we can't easily remap the separate
1741                  * metadata buffer from the block layer and thus require a
1742                  * separate metadata buffer for block layer metadata/PI support.
1743                  * We allow extended LBAs for the passthrough interface, though.
1744                  */
1745                 if (id->flbas & NVME_NS_FLBAS_META_EXT)
1746                         ns->features |= NVME_NS_EXT_LBAS;
1747                 else
1748                         ns->features |= NVME_NS_METADATA_SUPPORTED;
1749         }
1750
1751         return 0;
1752 }
1753
1754 static void nvme_set_queue_limits(struct nvme_ctrl *ctrl,
1755                 struct request_queue *q)
1756 {
1757         bool vwc = ctrl->vwc & NVME_CTRL_VWC_PRESENT;
1758
1759         if (ctrl->max_hw_sectors) {
1760                 u32 max_segments =
1761                         (ctrl->max_hw_sectors / (NVME_CTRL_PAGE_SIZE >> 9)) + 1;
1762
1763                 max_segments = min_not_zero(max_segments, ctrl->max_segments);
1764                 blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors);
1765                 blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX));
1766         }
1767         blk_queue_virt_boundary(q, NVME_CTRL_PAGE_SIZE - 1);
1768         blk_queue_dma_alignment(q, 7);
1769         blk_queue_write_cache(q, vwc, vwc);
1770 }
1771
1772 static void nvme_update_disk_info(struct gendisk *disk,
1773                 struct nvme_ns *ns, struct nvme_id_ns *id)
1774 {
1775         sector_t capacity = nvme_lba_to_sect(ns, le64_to_cpu(id->nsze));
1776         unsigned short bs = 1 << ns->lba_shift;
1777         u32 atomic_bs, phys_bs, io_opt = 0;
1778
1779         /*
1780          * The block layer can't support LBA sizes larger than the page size
1781          * yet, so catch this early and don't allow block I/O.
1782          */
1783         if (ns->lba_shift > PAGE_SHIFT) {
1784                 capacity = 0;
1785                 bs = (1 << 9);
1786         }
1787
1788         blk_integrity_unregister(disk);
1789
1790         atomic_bs = phys_bs = bs;
1791         nvme_setup_streams_ns(ns->ctrl, ns, &phys_bs, &io_opt);
1792         if (id->nabo == 0) {
1793                 /*
1794                  * Bit 1 indicates whether NAWUPF is defined for this namespace
1795                  * and whether it should be used instead of AWUPF. If NAWUPF ==
1796                  * 0 then AWUPF must be used instead.
1797                  */
1798                 if (id->nsfeat & NVME_NS_FEAT_ATOMICS && id->nawupf)
1799                         atomic_bs = (1 + le16_to_cpu(id->nawupf)) * bs;
1800                 else
1801                         atomic_bs = (1 + ns->ctrl->subsys->awupf) * bs;
1802         }
1803
1804         if (id->nsfeat & NVME_NS_FEAT_IO_OPT) {
1805                 /* NPWG = Namespace Preferred Write Granularity */
1806                 phys_bs = bs * (1 + le16_to_cpu(id->npwg));
1807                 /* NOWS = Namespace Optimal Write Size */
1808                 io_opt = bs * (1 + le16_to_cpu(id->nows));
1809         }
1810
1811         blk_queue_logical_block_size(disk->queue, bs);
1812         /*
1813          * Linux filesystems assume writing a single physical block is
1814          * an atomic operation. Hence limit the physical block size to the
1815          * value of the Atomic Write Unit Power Fail parameter.
1816          */
1817         blk_queue_physical_block_size(disk->queue, min(phys_bs, atomic_bs));
1818         blk_queue_io_min(disk->queue, phys_bs);
1819         blk_queue_io_opt(disk->queue, io_opt);
1820
1821         /*
1822          * Register a metadata profile for PI, or the plain non-integrity NVMe
1823          * metadata masquerading as Type 0 if supported, otherwise reject block
1824          * I/O to namespaces with metadata except when the namespace supports
1825          * PI, as it can strip/insert in that case.
1826          */
1827         if (ns->ms) {
1828                 if (IS_ENABLED(CONFIG_BLK_DEV_INTEGRITY) &&
1829                     (ns->features & NVME_NS_METADATA_SUPPORTED))
1830                         nvme_init_integrity(disk, ns->ms, ns->pi_type,
1831                                             ns->ctrl->max_integrity_segments);
1832                 else if (!nvme_ns_has_pi(ns))
1833                         capacity = 0;
1834         }
1835
1836         set_capacity_and_notify(disk, capacity);
1837
1838         nvme_config_discard(disk, ns);
1839         blk_queue_max_write_zeroes_sectors(disk->queue,
1840                                            ns->ctrl->max_zeroes_sectors);
1841
1842         set_disk_ro(disk, (id->nsattr & NVME_NS_ATTR_RO) ||
1843                 test_bit(NVME_NS_FORCE_RO, &ns->flags));
1844 }
1845
1846 static inline bool nvme_first_scan(struct gendisk *disk)
1847 {
1848         /* nvme_alloc_ns() scans the disk prior to adding it */
1849         return !disk_live(disk);
1850 }
1851
1852 static void nvme_set_chunk_sectors(struct nvme_ns *ns, struct nvme_id_ns *id)
1853 {
1854         struct nvme_ctrl *ctrl = ns->ctrl;
1855         u32 iob;
1856
1857         if ((ctrl->quirks & NVME_QUIRK_STRIPE_SIZE) &&
1858             is_power_of_2(ctrl->max_hw_sectors))
1859                 iob = ctrl->max_hw_sectors;
1860         else
1861                 iob = nvme_lba_to_sect(ns, le16_to_cpu(id->noiob));
1862
1863         if (!iob)
1864                 return;
1865
1866         if (!is_power_of_2(iob)) {
1867                 if (nvme_first_scan(ns->disk))
1868                         pr_warn("%s: ignoring unaligned IO boundary:%u\n",
1869                                 ns->disk->disk_name, iob);
1870                 return;
1871         }
1872
1873         if (blk_queue_is_zoned(ns->disk->queue)) {
1874                 if (nvme_first_scan(ns->disk))
1875                         pr_warn("%s: ignoring zoned namespace IO boundary\n",
1876                                 ns->disk->disk_name);
1877                 return;
1878         }
1879
1880         blk_queue_chunk_sectors(ns->queue, iob);
1881 }
1882
1883 static int nvme_update_ns_info(struct nvme_ns *ns, struct nvme_id_ns *id)
1884 {
1885         unsigned lbaf = id->flbas & NVME_NS_FLBAS_LBA_MASK;
1886         int ret;
1887
1888         blk_mq_freeze_queue(ns->disk->queue);
1889         ns->lba_shift = id->lbaf[lbaf].ds;
1890         nvme_set_queue_limits(ns->ctrl, ns->queue);
1891
1892         ret = nvme_configure_metadata(ns, id);
1893         if (ret)
1894                 goto out_unfreeze;
1895         nvme_set_chunk_sectors(ns, id);
1896         nvme_update_disk_info(ns->disk, ns, id);
1897
1898         if (ns->head->ids.csi == NVME_CSI_ZNS) {
1899                 ret = nvme_update_zone_info(ns, lbaf);
1900                 if (ret)
1901                         goto out_unfreeze;
1902         }
1903
1904         set_bit(NVME_NS_READY, &ns->flags);
1905         blk_mq_unfreeze_queue(ns->disk->queue);
1906
1907         if (blk_queue_is_zoned(ns->queue)) {
1908                 ret = nvme_revalidate_zones(ns);
1909                 if (ret && !nvme_first_scan(ns->disk))
1910                         goto out;
1911         }
1912
1913         if (nvme_ns_head_multipath(ns->head)) {
1914                 blk_mq_freeze_queue(ns->head->disk->queue);
1915                 nvme_update_disk_info(ns->head->disk, ns, id);
1916                 nvme_mpath_revalidate_paths(ns);
1917                 blk_stack_limits(&ns->head->disk->queue->limits,
1918                                  &ns->queue->limits, 0);
1919                 disk_update_readahead(ns->head->disk);
1920                 blk_mq_unfreeze_queue(ns->head->disk->queue);
1921         }
1922         return 0;
1923
1924 out_unfreeze:
1925         blk_mq_unfreeze_queue(ns->disk->queue);
1926 out:
1927         /*
1928          * If probing fails due an unsupported feature, hide the block device,
1929          * but still allow other access.
1930          */
1931         if (ret == -ENODEV) {
1932                 ns->disk->flags |= GENHD_FL_HIDDEN;
1933                 ret = 0;
1934         }
1935         return ret;
1936 }
1937
1938 static char nvme_pr_type(enum pr_type type)
1939 {
1940         switch (type) {
1941         case PR_WRITE_EXCLUSIVE:
1942                 return 1;
1943         case PR_EXCLUSIVE_ACCESS:
1944                 return 2;
1945         case PR_WRITE_EXCLUSIVE_REG_ONLY:
1946                 return 3;
1947         case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1948                 return 4;
1949         case PR_WRITE_EXCLUSIVE_ALL_REGS:
1950                 return 5;
1951         case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1952                 return 6;
1953         default:
1954                 return 0;
1955         }
1956 };
1957
1958 static int nvme_send_ns_head_pr_command(struct block_device *bdev,
1959                 struct nvme_command *c, u8 data[16])
1960 {
1961         struct nvme_ns_head *head = bdev->bd_disk->private_data;
1962         int srcu_idx = srcu_read_lock(&head->srcu);
1963         struct nvme_ns *ns = nvme_find_path(head);
1964         int ret = -EWOULDBLOCK;
1965
1966         if (ns) {
1967                 c->common.nsid = cpu_to_le32(ns->head->ns_id);
1968                 ret = nvme_submit_sync_cmd(ns->queue, c, data, 16);
1969         }
1970         srcu_read_unlock(&head->srcu, srcu_idx);
1971         return ret;
1972 }
1973
1974 static int nvme_send_ns_pr_command(struct nvme_ns *ns, struct nvme_command *c,
1975                 u8 data[16])
1976 {
1977         c->common.nsid = cpu_to_le32(ns->head->ns_id);
1978         return nvme_submit_sync_cmd(ns->queue, c, data, 16);
1979 }
1980
1981 static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
1982                                 u64 key, u64 sa_key, u8 op)
1983 {
1984         struct nvme_command c = { };
1985         u8 data[16] = { 0, };
1986
1987         put_unaligned_le64(key, &data[0]);
1988         put_unaligned_le64(sa_key, &data[8]);
1989
1990         c.common.opcode = op;
1991         c.common.cdw10 = cpu_to_le32(cdw10);
1992
1993         if (IS_ENABLED(CONFIG_NVME_MULTIPATH) &&
1994             bdev->bd_disk->fops == &nvme_ns_head_ops)
1995                 return nvme_send_ns_head_pr_command(bdev, &c, data);
1996         return nvme_send_ns_pr_command(bdev->bd_disk->private_data, &c, data);
1997 }
1998
1999 static int nvme_pr_register(struct block_device *bdev, u64 old,
2000                 u64 new, unsigned flags)
2001 {
2002         u32 cdw10;
2003
2004         if (flags & ~PR_FL_IGNORE_KEY)
2005                 return -EOPNOTSUPP;
2006
2007         cdw10 = old ? 2 : 0;
2008         cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
2009         cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
2010         return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
2011 }
2012
2013 static int nvme_pr_reserve(struct block_device *bdev, u64 key,
2014                 enum pr_type type, unsigned flags)
2015 {
2016         u32 cdw10;
2017
2018         if (flags & ~PR_FL_IGNORE_KEY)
2019                 return -EOPNOTSUPP;
2020
2021         cdw10 = nvme_pr_type(type) << 8;
2022         cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
2023         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
2024 }
2025
2026 static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
2027                 enum pr_type type, bool abort)
2028 {
2029         u32 cdw10 = nvme_pr_type(type) << 8 | (abort ? 2 : 1);
2030
2031         return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
2032 }
2033
2034 static int nvme_pr_clear(struct block_device *bdev, u64 key)
2035 {
2036         u32 cdw10 = 1 | (key ? 1 << 3 : 0);
2037
2038         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_register);
2039 }
2040
2041 static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
2042 {
2043         u32 cdw10 = nvme_pr_type(type) << 8 | (key ? 1 << 3 : 0);
2044
2045         return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
2046 }
2047
2048 const struct pr_ops nvme_pr_ops = {
2049         .pr_register    = nvme_pr_register,
2050         .pr_reserve     = nvme_pr_reserve,
2051         .pr_release     = nvme_pr_release,
2052         .pr_preempt     = nvme_pr_preempt,
2053         .pr_clear       = nvme_pr_clear,
2054 };
2055
2056 #ifdef CONFIG_BLK_SED_OPAL
2057 int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
2058                 bool send)
2059 {
2060         struct nvme_ctrl *ctrl = data;
2061         struct nvme_command cmd = { };
2062
2063         if (send)
2064                 cmd.common.opcode = nvme_admin_security_send;
2065         else
2066                 cmd.common.opcode = nvme_admin_security_recv;
2067         cmd.common.nsid = 0;
2068         cmd.common.cdw10 = cpu_to_le32(((u32)secp) << 24 | ((u32)spsp) << 8);
2069         cmd.common.cdw11 = cpu_to_le32(len);
2070
2071         return __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, buffer, len, 0,
2072                         NVME_QID_ANY, 1, 0);
2073 }
2074 EXPORT_SYMBOL_GPL(nvme_sec_submit);
2075 #endif /* CONFIG_BLK_SED_OPAL */
2076
2077 #ifdef CONFIG_BLK_DEV_ZONED
2078 static int nvme_report_zones(struct gendisk *disk, sector_t sector,
2079                 unsigned int nr_zones, report_zones_cb cb, void *data)
2080 {
2081         return nvme_ns_report_zones(disk->private_data, sector, nr_zones, cb,
2082                         data);
2083 }
2084 #else
2085 #define nvme_report_zones       NULL
2086 #endif /* CONFIG_BLK_DEV_ZONED */
2087
2088 static const struct block_device_operations nvme_bdev_ops = {
2089         .owner          = THIS_MODULE,
2090         .ioctl          = nvme_ioctl,
2091         .open           = nvme_open,
2092         .release        = nvme_release,
2093         .getgeo         = nvme_getgeo,
2094         .report_zones   = nvme_report_zones,
2095         .pr_ops         = &nvme_pr_ops,
2096 };
2097
2098 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u64 cap, bool enabled)
2099 {
2100         unsigned long timeout =
2101                 ((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies;
2102         u32 csts, bit = enabled ? NVME_CSTS_RDY : 0;
2103         int ret;
2104
2105         while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2106                 if (csts == ~0)
2107                         return -ENODEV;
2108                 if ((csts & NVME_CSTS_RDY) == bit)
2109                         break;
2110
2111                 usleep_range(1000, 2000);
2112                 if (fatal_signal_pending(current))
2113                         return -EINTR;
2114                 if (time_after(jiffies, timeout)) {
2115                         dev_err(ctrl->device,
2116                                 "Device not ready; aborting %s, CSTS=0x%x\n",
2117                                 enabled ? "initialisation" : "reset", csts);
2118                         return -ENODEV;
2119                 }
2120         }
2121
2122         return ret;
2123 }
2124
2125 /*
2126  * If the device has been passed off to us in an enabled state, just clear
2127  * the enabled bit.  The spec says we should set the 'shutdown notification
2128  * bits', but doing so may cause the device to complete commands to the
2129  * admin queue ... and we don't know what memory that might be pointing at!
2130  */
2131 int nvme_disable_ctrl(struct nvme_ctrl *ctrl)
2132 {
2133         int ret;
2134
2135         ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2136         ctrl->ctrl_config &= ~NVME_CC_ENABLE;
2137
2138         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2139         if (ret)
2140                 return ret;
2141
2142         if (ctrl->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY)
2143                 msleep(NVME_QUIRK_DELAY_AMOUNT);
2144
2145         return nvme_wait_ready(ctrl, ctrl->cap, false);
2146 }
2147 EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
2148
2149 int nvme_enable_ctrl(struct nvme_ctrl *ctrl)
2150 {
2151         unsigned dev_page_min;
2152         int ret;
2153
2154         ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &ctrl->cap);
2155         if (ret) {
2156                 dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
2157                 return ret;
2158         }
2159         dev_page_min = NVME_CAP_MPSMIN(ctrl->cap) + 12;
2160
2161         if (NVME_CTRL_PAGE_SHIFT < dev_page_min) {
2162                 dev_err(ctrl->device,
2163                         "Minimum device page size %u too large for host (%u)\n",
2164                         1 << dev_page_min, 1 << NVME_CTRL_PAGE_SHIFT);
2165                 return -ENODEV;
2166         }
2167
2168         if (NVME_CAP_CSS(ctrl->cap) & NVME_CAP_CSS_CSI)
2169                 ctrl->ctrl_config = NVME_CC_CSS_CSI;
2170         else
2171                 ctrl->ctrl_config = NVME_CC_CSS_NVM;
2172         ctrl->ctrl_config |= (NVME_CTRL_PAGE_SHIFT - 12) << NVME_CC_MPS_SHIFT;
2173         ctrl->ctrl_config |= NVME_CC_AMS_RR | NVME_CC_SHN_NONE;
2174         ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
2175         ctrl->ctrl_config |= NVME_CC_ENABLE;
2176
2177         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2178         if (ret)
2179                 return ret;
2180         return nvme_wait_ready(ctrl, ctrl->cap, true);
2181 }
2182 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
2183
2184 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl)
2185 {
2186         unsigned long timeout = jiffies + (ctrl->shutdown_timeout * HZ);
2187         u32 csts;
2188         int ret;
2189
2190         ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2191         ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
2192
2193         ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2194         if (ret)
2195                 return ret;
2196
2197         while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2198                 if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT)
2199                         break;
2200
2201                 msleep(100);
2202                 if (fatal_signal_pending(current))
2203                         return -EINTR;
2204                 if (time_after(jiffies, timeout)) {
2205                         dev_err(ctrl->device,
2206                                 "Device shutdown incomplete; abort shutdown\n");
2207                         return -ENODEV;
2208                 }
2209         }
2210
2211         return ret;
2212 }
2213 EXPORT_SYMBOL_GPL(nvme_shutdown_ctrl);
2214
2215 static int nvme_configure_timestamp(struct nvme_ctrl *ctrl)
2216 {
2217         __le64 ts;
2218         int ret;
2219
2220         if (!(ctrl->oncs & NVME_CTRL_ONCS_TIMESTAMP))
2221                 return 0;
2222
2223         ts = cpu_to_le64(ktime_to_ms(ktime_get_real()));
2224         ret = nvme_set_features(ctrl, NVME_FEAT_TIMESTAMP, 0, &ts, sizeof(ts),
2225                         NULL);
2226         if (ret)
2227                 dev_warn_once(ctrl->device,
2228                         "could not set timestamp (%d)\n", ret);
2229         return ret;
2230 }
2231
2232 static int nvme_configure_acre(struct nvme_ctrl *ctrl)
2233 {
2234         struct nvme_feat_host_behavior *host;
2235         int ret;
2236
2237         /* Don't bother enabling the feature if retry delay is not reported */
2238         if (!ctrl->crdt[0])
2239                 return 0;
2240
2241         host = kzalloc(sizeof(*host), GFP_KERNEL);
2242         if (!host)
2243                 return 0;
2244
2245         host->acre = NVME_ENABLE_ACRE;
2246         ret = nvme_set_features(ctrl, NVME_FEAT_HOST_BEHAVIOR, 0,
2247                                 host, sizeof(*host), NULL);
2248         kfree(host);
2249         return ret;
2250 }
2251
2252 /*
2253  * The function checks whether the given total (exlat + enlat) latency of
2254  * a power state allows the latter to be used as an APST transition target.
2255  * It does so by comparing the latency to the primary and secondary latency
2256  * tolerances defined by module params. If there's a match, the corresponding
2257  * timeout value is returned and the matching tolerance index (1 or 2) is
2258  * reported.
2259  */
2260 static bool nvme_apst_get_transition_time(u64 total_latency,
2261                 u64 *transition_time, unsigned *last_index)
2262 {
2263         if (total_latency <= apst_primary_latency_tol_us) {
2264                 if (*last_index == 1)
2265                         return false;
2266                 *last_index = 1;
2267                 *transition_time = apst_primary_timeout_ms;
2268                 return true;
2269         }
2270         if (apst_secondary_timeout_ms &&
2271                 total_latency <= apst_secondary_latency_tol_us) {
2272                 if (*last_index <= 2)
2273                         return false;
2274                 *last_index = 2;
2275                 *transition_time = apst_secondary_timeout_ms;
2276                 return true;
2277         }
2278         return false;
2279 }
2280
2281 /*
2282  * APST (Autonomous Power State Transition) lets us program a table of power
2283  * state transitions that the controller will perform automatically.
2284  *
2285  * Depending on module params, one of the two supported techniques will be used:
2286  *
2287  * - If the parameters provide explicit timeouts and tolerances, they will be
2288  *   used to build a table with up to 2 non-operational states to transition to.
2289  *   The default parameter values were selected based on the values used by
2290  *   Microsoft's and Intel's NVMe drivers. Yet, since we don't implement dynamic
2291  *   regeneration of the APST table in the event of switching between external
2292  *   and battery power, the timeouts and tolerances reflect a compromise
2293  *   between values used by Microsoft for AC and battery scenarios.
2294  * - If not, we'll configure the table with a simple heuristic: we are willing
2295  *   to spend at most 2% of the time transitioning between power states.
2296  *   Therefore, when running in any given state, we will enter the next
2297  *   lower-power non-operational state after waiting 50 * (enlat + exlat)
2298  *   microseconds, as long as that state's exit latency is under the requested
2299  *   maximum latency.
2300  *
2301  * We will not autonomously enter any non-operational state for which the total
2302  * latency exceeds ps_max_latency_us.
2303  *
2304  * Users can set ps_max_latency_us to zero to turn off APST.
2305  */
2306 static int nvme_configure_apst(struct nvme_ctrl *ctrl)
2307 {
2308         struct nvme_feat_auto_pst *table;
2309         unsigned apste = 0;
2310         u64 max_lat_us = 0;
2311         __le64 target = 0;
2312         int max_ps = -1;
2313         int state;
2314         int ret;
2315         unsigned last_lt_index = UINT_MAX;
2316
2317         /*
2318          * If APST isn't supported or if we haven't been initialized yet,
2319          * then don't do anything.
2320          */
2321         if (!ctrl->apsta)
2322                 return 0;
2323
2324         if (ctrl->npss > 31) {
2325                 dev_warn(ctrl->device, "NPSS is invalid; not using APST\n");
2326                 return 0;
2327         }
2328
2329         table = kzalloc(sizeof(*table), GFP_KERNEL);
2330         if (!table)
2331                 return 0;
2332
2333         if (!ctrl->apst_enabled || ctrl->ps_max_latency_us == 0) {
2334                 /* Turn off APST. */
2335                 dev_dbg(ctrl->device, "APST disabled\n");
2336                 goto done;
2337         }
2338
2339         /*
2340          * Walk through all states from lowest- to highest-power.
2341          * According to the spec, lower-numbered states use more power.  NPSS,
2342          * despite the name, is the index of the lowest-power state, not the
2343          * number of states.
2344          */
2345         for (state = (int)ctrl->npss; state >= 0; state--) {
2346                 u64 total_latency_us, exit_latency_us, transition_ms;
2347
2348                 if (target)
2349                         table->entries[state] = target;
2350
2351                 /*
2352                  * Don't allow transitions to the deepest state if it's quirked
2353                  * off.
2354                  */
2355                 if (state == ctrl->npss &&
2356                     (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS))
2357                         continue;
2358
2359                 /*
2360                  * Is this state a useful non-operational state for higher-power
2361                  * states to autonomously transition to?
2362                  */
2363                 if (!(ctrl->psd[state].flags & NVME_PS_FLAGS_NON_OP_STATE))
2364                         continue;
2365
2366                 exit_latency_us = (u64)le32_to_cpu(ctrl->psd[state].exit_lat);
2367                 if (exit_latency_us > ctrl->ps_max_latency_us)
2368                         continue;
2369
2370                 total_latency_us = exit_latency_us +
2371                         le32_to_cpu(ctrl->psd[state].entry_lat);
2372
2373                 /*
2374                  * This state is good. It can be used as the APST idle target
2375                  * for higher power states.
2376                  */
2377                 if (apst_primary_timeout_ms && apst_primary_latency_tol_us) {
2378                         if (!nvme_apst_get_transition_time(total_latency_us,
2379                                         &transition_ms, &last_lt_index))
2380                                 continue;
2381                 } else {
2382                         transition_ms = total_latency_us + 19;
2383                         do_div(transition_ms, 20);
2384                         if (transition_ms > (1 << 24) - 1)
2385                                 transition_ms = (1 << 24) - 1;
2386                 }
2387
2388                 target = cpu_to_le64((state << 3) | (transition_ms << 8));
2389                 if (max_ps == -1)
2390                         max_ps = state;
2391                 if (total_latency_us > max_lat_us)
2392                         max_lat_us = total_latency_us;
2393         }
2394
2395         if (max_ps == -1)
2396                 dev_dbg(ctrl->device, "APST enabled but no non-operational states are available\n");
2397         else
2398                 dev_dbg(ctrl->device, "APST enabled: max PS = %d, max round-trip latency = %lluus, table = %*phN\n",
2399                         max_ps, max_lat_us, (int)sizeof(*table), table);
2400         apste = 1;
2401
2402 done:
2403         ret = nvme_set_features(ctrl, NVME_FEAT_AUTO_PST, apste,
2404                                 table, sizeof(*table), NULL);
2405         if (ret)
2406                 dev_err(ctrl->device, "failed to set APST feature (%d)\n", ret);
2407         kfree(table);
2408         return ret;
2409 }
2410
2411 static void nvme_set_latency_tolerance(struct device *dev, s32 val)
2412 {
2413         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2414         u64 latency;
2415
2416         switch (val) {
2417         case PM_QOS_LATENCY_TOLERANCE_NO_CONSTRAINT:
2418         case PM_QOS_LATENCY_ANY:
2419                 latency = U64_MAX;
2420                 break;
2421
2422         default:
2423                 latency = val;
2424         }
2425
2426         if (ctrl->ps_max_latency_us != latency) {
2427                 ctrl->ps_max_latency_us = latency;
2428                 if (ctrl->state == NVME_CTRL_LIVE)
2429                         nvme_configure_apst(ctrl);
2430         }
2431 }
2432
2433 struct nvme_core_quirk_entry {
2434         /*
2435          * NVMe model and firmware strings are padded with spaces.  For
2436          * simplicity, strings in the quirk table are padded with NULLs
2437          * instead.
2438          */
2439         u16 vid;
2440         const char *mn;
2441         const char *fr;
2442         unsigned long quirks;
2443 };
2444
2445 static const struct nvme_core_quirk_entry core_quirks[] = {
2446         {
2447                 /*
2448                  * This Toshiba device seems to die using any APST states.  See:
2449                  * https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1678184/comments/11
2450                  */
2451                 .vid = 0x1179,
2452                 .mn = "THNSF5256GPUK TOSHIBA",
2453                 .quirks = NVME_QUIRK_NO_APST,
2454         },
2455         {
2456                 /*
2457                  * This LiteON CL1-3D*-Q11 firmware version has a race
2458                  * condition associated with actions related to suspend to idle
2459                  * LiteON has resolved the problem in future firmware
2460                  */
2461                 .vid = 0x14a4,
2462                 .fr = "22301111",
2463                 .quirks = NVME_QUIRK_SIMPLE_SUSPEND,
2464         }
2465 };
2466
2467 /* match is null-terminated but idstr is space-padded. */
2468 static bool string_matches(const char *idstr, const char *match, size_t len)
2469 {
2470         size_t matchlen;
2471
2472         if (!match)
2473                 return true;
2474
2475         matchlen = strlen(match);
2476         WARN_ON_ONCE(matchlen > len);
2477
2478         if (memcmp(idstr, match, matchlen))
2479                 return false;
2480
2481         for (; matchlen < len; matchlen++)
2482                 if (idstr[matchlen] != ' ')
2483                         return false;
2484
2485         return true;
2486 }
2487
2488 static bool quirk_matches(const struct nvme_id_ctrl *id,
2489                           const struct nvme_core_quirk_entry *q)
2490 {
2491         return q->vid == le16_to_cpu(id->vid) &&
2492                 string_matches(id->mn, q->mn, sizeof(id->mn)) &&
2493                 string_matches(id->fr, q->fr, sizeof(id->fr));
2494 }
2495
2496 static void nvme_init_subnqn(struct nvme_subsystem *subsys, struct nvme_ctrl *ctrl,
2497                 struct nvme_id_ctrl *id)
2498 {
2499         size_t nqnlen;
2500         int off;
2501
2502         if(!(ctrl->quirks & NVME_QUIRK_IGNORE_DEV_SUBNQN)) {
2503                 nqnlen = strnlen(id->subnqn, NVMF_NQN_SIZE);
2504                 if (nqnlen > 0 && nqnlen < NVMF_NQN_SIZE) {
2505                         strlcpy(subsys->subnqn, id->subnqn, NVMF_NQN_SIZE);
2506                         return;
2507                 }
2508
2509                 if (ctrl->vs >= NVME_VS(1, 2, 1))
2510                         dev_warn(ctrl->device, "missing or invalid SUBNQN field.\n");
2511         }
2512
2513         /* Generate a "fake" NQN per Figure 254 in NVMe 1.3 + ECN 001 */
2514         off = snprintf(subsys->subnqn, NVMF_NQN_SIZE,
2515                         "nqn.2014.08.org.nvmexpress:%04x%04x",
2516                         le16_to_cpu(id->vid), le16_to_cpu(id->ssvid));
2517         memcpy(subsys->subnqn + off, id->sn, sizeof(id->sn));
2518         off += sizeof(id->sn);
2519         memcpy(subsys->subnqn + off, id->mn, sizeof(id->mn));
2520         off += sizeof(id->mn);
2521         memset(subsys->subnqn + off, 0, sizeof(subsys->subnqn) - off);
2522 }
2523
2524 static void nvme_release_subsystem(struct device *dev)
2525 {
2526         struct nvme_subsystem *subsys =
2527                 container_of(dev, struct nvme_subsystem, dev);
2528
2529         if (subsys->instance >= 0)
2530                 ida_simple_remove(&nvme_instance_ida, subsys->instance);
2531         kfree(subsys);
2532 }
2533
2534 static void nvme_destroy_subsystem(struct kref *ref)
2535 {
2536         struct nvme_subsystem *subsys =
2537                         container_of(ref, struct nvme_subsystem, ref);
2538
2539         mutex_lock(&nvme_subsystems_lock);
2540         list_del(&subsys->entry);
2541         mutex_unlock(&nvme_subsystems_lock);
2542
2543         ida_destroy(&subsys->ns_ida);
2544         device_del(&subsys->dev);
2545         put_device(&subsys->dev);
2546 }
2547
2548 static void nvme_put_subsystem(struct nvme_subsystem *subsys)
2549 {
2550         kref_put(&subsys->ref, nvme_destroy_subsystem);
2551 }
2552
2553 static struct nvme_subsystem *__nvme_find_get_subsystem(const char *subsysnqn)
2554 {
2555         struct nvme_subsystem *subsys;
2556
2557         lockdep_assert_held(&nvme_subsystems_lock);
2558
2559         /*
2560          * Fail matches for discovery subsystems. This results
2561          * in each discovery controller bound to a unique subsystem.
2562          * This avoids issues with validating controller values
2563          * that can only be true when there is a single unique subsystem.
2564          * There may be multiple and completely independent entities
2565          * that provide discovery controllers.
2566          */
2567         if (!strcmp(subsysnqn, NVME_DISC_SUBSYS_NAME))
2568                 return NULL;
2569
2570         list_for_each_entry(subsys, &nvme_subsystems, entry) {
2571                 if (strcmp(subsys->subnqn, subsysnqn))
2572                         continue;
2573                 if (!kref_get_unless_zero(&subsys->ref))
2574                         continue;
2575                 return subsys;
2576         }
2577
2578         return NULL;
2579 }
2580
2581 #define SUBSYS_ATTR_RO(_name, _mode, _show)                     \
2582         struct device_attribute subsys_attr_##_name = \
2583                 __ATTR(_name, _mode, _show, NULL)
2584
2585 static ssize_t nvme_subsys_show_nqn(struct device *dev,
2586                                     struct device_attribute *attr,
2587                                     char *buf)
2588 {
2589         struct nvme_subsystem *subsys =
2590                 container_of(dev, struct nvme_subsystem, dev);
2591
2592         return sysfs_emit(buf, "%s\n", subsys->subnqn);
2593 }
2594 static SUBSYS_ATTR_RO(subsysnqn, S_IRUGO, nvme_subsys_show_nqn);
2595
2596 #define nvme_subsys_show_str_function(field)                            \
2597 static ssize_t subsys_##field##_show(struct device *dev,                \
2598                             struct device_attribute *attr, char *buf)   \
2599 {                                                                       \
2600         struct nvme_subsystem *subsys =                                 \
2601                 container_of(dev, struct nvme_subsystem, dev);          \
2602         return sysfs_emit(buf, "%.*s\n",                                \
2603                            (int)sizeof(subsys->field), subsys->field);  \
2604 }                                                                       \
2605 static SUBSYS_ATTR_RO(field, S_IRUGO, subsys_##field##_show);
2606
2607 nvme_subsys_show_str_function(model);
2608 nvme_subsys_show_str_function(serial);
2609 nvme_subsys_show_str_function(firmware_rev);
2610
2611 static struct attribute *nvme_subsys_attrs[] = {
2612         &subsys_attr_model.attr,
2613         &subsys_attr_serial.attr,
2614         &subsys_attr_firmware_rev.attr,
2615         &subsys_attr_subsysnqn.attr,
2616 #ifdef CONFIG_NVME_MULTIPATH
2617         &subsys_attr_iopolicy.attr,
2618 #endif
2619         NULL,
2620 };
2621
2622 static const struct attribute_group nvme_subsys_attrs_group = {
2623         .attrs = nvme_subsys_attrs,
2624 };
2625
2626 static const struct attribute_group *nvme_subsys_attrs_groups[] = {
2627         &nvme_subsys_attrs_group,
2628         NULL,
2629 };
2630
2631 static inline bool nvme_discovery_ctrl(struct nvme_ctrl *ctrl)
2632 {
2633         return ctrl->opts && ctrl->opts->discovery_nqn;
2634 }
2635
2636 static bool nvme_validate_cntlid(struct nvme_subsystem *subsys,
2637                 struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2638 {
2639         struct nvme_ctrl *tmp;
2640
2641         lockdep_assert_held(&nvme_subsystems_lock);
2642
2643         list_for_each_entry(tmp, &subsys->ctrls, subsys_entry) {
2644                 if (nvme_state_terminal(tmp))
2645                         continue;
2646
2647                 if (tmp->cntlid == ctrl->cntlid) {
2648                         dev_err(ctrl->device,
2649                                 "Duplicate cntlid %u with %s, rejecting\n",
2650                                 ctrl->cntlid, dev_name(tmp->device));
2651                         return false;
2652                 }
2653
2654                 if ((id->cmic & NVME_CTRL_CMIC_MULTI_CTRL) ||
2655                     nvme_discovery_ctrl(ctrl))
2656                         continue;
2657
2658                 dev_err(ctrl->device,
2659                         "Subsystem does not support multiple controllers\n");
2660                 return false;
2661         }
2662
2663         return true;
2664 }
2665
2666 static int nvme_init_subsystem(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2667 {
2668         struct nvme_subsystem *subsys, *found;
2669         int ret;
2670
2671         subsys = kzalloc(sizeof(*subsys), GFP_KERNEL);
2672         if (!subsys)
2673                 return -ENOMEM;
2674
2675         subsys->instance = -1;
2676         mutex_init(&subsys->lock);
2677         kref_init(&subsys->ref);
2678         INIT_LIST_HEAD(&subsys->ctrls);
2679         INIT_LIST_HEAD(&subsys->nsheads);
2680         nvme_init_subnqn(subsys, ctrl, id);
2681         memcpy(subsys->serial, id->sn, sizeof(subsys->serial));
2682         memcpy(subsys->model, id->mn, sizeof(subsys->model));
2683         memcpy(subsys->firmware_rev, id->fr, sizeof(subsys->firmware_rev));
2684         subsys->vendor_id = le16_to_cpu(id->vid);
2685         subsys->cmic = id->cmic;
2686         subsys->awupf = le16_to_cpu(id->awupf);
2687 #ifdef CONFIG_NVME_MULTIPATH
2688         subsys->iopolicy = NVME_IOPOLICY_NUMA;
2689 #endif
2690
2691         subsys->dev.class = nvme_subsys_class;
2692         subsys->dev.release = nvme_release_subsystem;
2693         subsys->dev.groups = nvme_subsys_attrs_groups;
2694         dev_set_name(&subsys->dev, "nvme-subsys%d", ctrl->instance);
2695         device_initialize(&subsys->dev);
2696
2697         mutex_lock(&nvme_subsystems_lock);
2698         found = __nvme_find_get_subsystem(subsys->subnqn);
2699         if (found) {
2700                 put_device(&subsys->dev);
2701                 subsys = found;
2702
2703                 if (!nvme_validate_cntlid(subsys, ctrl, id)) {
2704                         ret = -EINVAL;
2705                         goto out_put_subsystem;
2706                 }
2707         } else {
2708                 ret = device_add(&subsys->dev);
2709                 if (ret) {
2710                         dev_err(ctrl->device,
2711                                 "failed to register subsystem device.\n");
2712                         put_device(&subsys->dev);
2713                         goto out_unlock;
2714                 }
2715                 ida_init(&subsys->ns_ida);
2716                 list_add_tail(&subsys->entry, &nvme_subsystems);
2717         }
2718
2719         ret = sysfs_create_link(&subsys->dev.kobj, &ctrl->device->kobj,
2720                                 dev_name(ctrl->device));
2721         if (ret) {
2722                 dev_err(ctrl->device,
2723                         "failed to create sysfs link from subsystem.\n");
2724                 goto out_put_subsystem;
2725         }
2726
2727         if (!found)
2728                 subsys->instance = ctrl->instance;
2729         ctrl->subsys = subsys;
2730         list_add_tail(&ctrl->subsys_entry, &subsys->ctrls);
2731         mutex_unlock(&nvme_subsystems_lock);
2732         return 0;
2733
2734 out_put_subsystem:
2735         nvme_put_subsystem(subsys);
2736 out_unlock:
2737         mutex_unlock(&nvme_subsystems_lock);
2738         return ret;
2739 }
2740
2741 int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp, u8 csi,
2742                 void *log, size_t size, u64 offset)
2743 {
2744         struct nvme_command c = { };
2745         u32 dwlen = nvme_bytes_to_numd(size);
2746
2747         c.get_log_page.opcode = nvme_admin_get_log_page;
2748         c.get_log_page.nsid = cpu_to_le32(nsid);
2749         c.get_log_page.lid = log_page;
2750         c.get_log_page.lsp = lsp;
2751         c.get_log_page.numdl = cpu_to_le16(dwlen & ((1 << 16) - 1));
2752         c.get_log_page.numdu = cpu_to_le16(dwlen >> 16);
2753         c.get_log_page.lpol = cpu_to_le32(lower_32_bits(offset));
2754         c.get_log_page.lpou = cpu_to_le32(upper_32_bits(offset));
2755         c.get_log_page.csi = csi;
2756
2757         return nvme_submit_sync_cmd(ctrl->admin_q, &c, log, size);
2758 }
2759
2760 static int nvme_get_effects_log(struct nvme_ctrl *ctrl, u8 csi,
2761                                 struct nvme_effects_log **log)
2762 {
2763         struct nvme_effects_log *cel = xa_load(&ctrl->cels, csi);
2764         int ret;
2765
2766         if (cel)
2767                 goto out;
2768
2769         cel = kzalloc(sizeof(*cel), GFP_KERNEL);
2770         if (!cel)
2771                 return -ENOMEM;
2772
2773         ret = nvme_get_log(ctrl, 0x00, NVME_LOG_CMD_EFFECTS, 0, csi,
2774                         cel, sizeof(*cel), 0);
2775         if (ret) {
2776                 kfree(cel);
2777                 return ret;
2778         }
2779
2780         xa_store(&ctrl->cels, csi, cel, GFP_KERNEL);
2781 out:
2782         *log = cel;
2783         return 0;
2784 }
2785
2786 static inline u32 nvme_mps_to_sectors(struct nvme_ctrl *ctrl, u32 units)
2787 {
2788         u32 page_shift = NVME_CAP_MPSMIN(ctrl->cap) + 12, val;
2789
2790         if (check_shl_overflow(1U, units + page_shift - 9, &val))
2791                 return UINT_MAX;
2792         return val;
2793 }
2794
2795 static int nvme_init_non_mdts_limits(struct nvme_ctrl *ctrl)
2796 {
2797         struct nvme_command c = { };
2798         struct nvme_id_ctrl_nvm *id;
2799         int ret;
2800
2801         if (ctrl->oncs & NVME_CTRL_ONCS_DSM) {
2802                 ctrl->max_discard_sectors = UINT_MAX;
2803                 ctrl->max_discard_segments = NVME_DSM_MAX_RANGES;
2804         } else {
2805                 ctrl->max_discard_sectors = 0;
2806                 ctrl->max_discard_segments = 0;
2807         }
2808
2809         /*
2810          * Even though NVMe spec explicitly states that MDTS is not applicable
2811          * to the write-zeroes, we are cautious and limit the size to the
2812          * controllers max_hw_sectors value, which is based on the MDTS field
2813          * and possibly other limiting factors.
2814          */
2815         if ((ctrl->oncs & NVME_CTRL_ONCS_WRITE_ZEROES) &&
2816             !(ctrl->quirks & NVME_QUIRK_DISABLE_WRITE_ZEROES))
2817                 ctrl->max_zeroes_sectors = ctrl->max_hw_sectors;
2818         else
2819                 ctrl->max_zeroes_sectors = 0;
2820
2821         if (nvme_ctrl_limited_cns(ctrl))
2822                 return 0;
2823
2824         id = kzalloc(sizeof(*id), GFP_KERNEL);
2825         if (!id)
2826                 return 0;
2827
2828         c.identify.opcode = nvme_admin_identify;
2829         c.identify.cns = NVME_ID_CNS_CS_CTRL;
2830         c.identify.csi = NVME_CSI_NVM;
2831
2832         ret = nvme_submit_sync_cmd(ctrl->admin_q, &c, id, sizeof(*id));
2833         if (ret)
2834                 goto free_data;
2835
2836         if (id->dmrl)
2837                 ctrl->max_discard_segments = id->dmrl;
2838         if (id->dmrsl)
2839                 ctrl->max_discard_sectors = le32_to_cpu(id->dmrsl);
2840         if (id->wzsl)
2841                 ctrl->max_zeroes_sectors = nvme_mps_to_sectors(ctrl, id->wzsl);
2842
2843 free_data:
2844         kfree(id);
2845         return ret;
2846 }
2847
2848 static int nvme_init_identify(struct nvme_ctrl *ctrl)
2849 {
2850         struct nvme_id_ctrl *id;
2851         u32 max_hw_sectors;
2852         bool prev_apst_enabled;
2853         int ret;
2854
2855         ret = nvme_identify_ctrl(ctrl, &id);
2856         if (ret) {
2857                 dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
2858                 return -EIO;
2859         }
2860
2861         if (id->lpa & NVME_CTRL_LPA_CMD_EFFECTS_LOG) {
2862                 ret = nvme_get_effects_log(ctrl, NVME_CSI_NVM, &ctrl->effects);
2863                 if (ret < 0)
2864                         goto out_free;
2865         }
2866
2867         if (!(ctrl->ops->flags & NVME_F_FABRICS))
2868                 ctrl->cntlid = le16_to_cpu(id->cntlid);
2869
2870         if (!ctrl->identified) {
2871                 unsigned int i;
2872
2873                 ret = nvme_init_subsystem(ctrl, id);
2874                 if (ret)
2875                         goto out_free;
2876
2877                 /*
2878                  * Check for quirks.  Quirk can depend on firmware version,
2879                  * so, in principle, the set of quirks present can change
2880                  * across a reset.  As a possible future enhancement, we
2881                  * could re-scan for quirks every time we reinitialize
2882                  * the device, but we'd have to make sure that the driver
2883                  * behaves intelligently if the quirks change.
2884                  */
2885                 for (i = 0; i < ARRAY_SIZE(core_quirks); i++) {
2886                         if (quirk_matches(id, &core_quirks[i]))
2887                                 ctrl->quirks |= core_quirks[i].quirks;
2888                 }
2889         }
2890
2891         if (force_apst && (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS)) {
2892                 dev_warn(ctrl->device, "forcibly allowing all power states due to nvme_core.force_apst -- use at your own risk\n");
2893                 ctrl->quirks &= ~NVME_QUIRK_NO_DEEPEST_PS;
2894         }
2895
2896         ctrl->crdt[0] = le16_to_cpu(id->crdt1);
2897         ctrl->crdt[1] = le16_to_cpu(id->crdt2);
2898         ctrl->crdt[2] = le16_to_cpu(id->crdt3);
2899
2900         ctrl->oacs = le16_to_cpu(id->oacs);
2901         ctrl->oncs = le16_to_cpu(id->oncs);
2902         ctrl->mtfa = le16_to_cpu(id->mtfa);
2903         ctrl->oaes = le32_to_cpu(id->oaes);
2904         ctrl->wctemp = le16_to_cpu(id->wctemp);
2905         ctrl->cctemp = le16_to_cpu(id->cctemp);
2906
2907         atomic_set(&ctrl->abort_limit, id->acl + 1);
2908         ctrl->vwc = id->vwc;
2909         if (id->mdts)
2910                 max_hw_sectors = nvme_mps_to_sectors(ctrl, id->mdts);
2911         else
2912                 max_hw_sectors = UINT_MAX;
2913         ctrl->max_hw_sectors =
2914                 min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
2915
2916         nvme_set_queue_limits(ctrl, ctrl->admin_q);
2917         ctrl->sgls = le32_to_cpu(id->sgls);
2918         ctrl->kas = le16_to_cpu(id->kas);
2919         ctrl->max_namespaces = le32_to_cpu(id->mnan);
2920         ctrl->ctratt = le32_to_cpu(id->ctratt);
2921
2922         if (id->rtd3e) {
2923                 /* us -> s */
2924                 u32 transition_time = le32_to_cpu(id->rtd3e) / USEC_PER_SEC;
2925
2926                 ctrl->shutdown_timeout = clamp_t(unsigned int, transition_time,
2927                                                  shutdown_timeout, 60);
2928
2929                 if (ctrl->shutdown_timeout != shutdown_timeout)
2930                         dev_info(ctrl->device,
2931                                  "Shutdown timeout set to %u seconds\n",
2932                                  ctrl->shutdown_timeout);
2933         } else
2934                 ctrl->shutdown_timeout = shutdown_timeout;
2935
2936         ctrl->npss = id->npss;
2937         ctrl->apsta = id->apsta;
2938         prev_apst_enabled = ctrl->apst_enabled;
2939         if (ctrl->quirks & NVME_QUIRK_NO_APST) {
2940                 if (force_apst && id->apsta) {
2941                         dev_warn(ctrl->device, "forcibly allowing APST due to nvme_core.force_apst -- use at your own risk\n");
2942                         ctrl->apst_enabled = true;
2943                 } else {
2944                         ctrl->apst_enabled = false;
2945                 }
2946         } else {
2947                 ctrl->apst_enabled = id->apsta;
2948         }
2949         memcpy(ctrl->psd, id->psd, sizeof(ctrl->psd));
2950
2951         if (ctrl->ops->flags & NVME_F_FABRICS) {
2952                 ctrl->icdoff = le16_to_cpu(id->icdoff);
2953                 ctrl->ioccsz = le32_to_cpu(id->ioccsz);
2954                 ctrl->iorcsz = le32_to_cpu(id->iorcsz);
2955                 ctrl->maxcmd = le16_to_cpu(id->maxcmd);
2956
2957                 /*
2958                  * In fabrics we need to verify the cntlid matches the
2959                  * admin connect
2960                  */
2961                 if (ctrl->cntlid != le16_to_cpu(id->cntlid)) {
2962                         dev_err(ctrl->device,
2963                                 "Mismatching cntlid: Connect %u vs Identify "
2964                                 "%u, rejecting\n",
2965                                 ctrl->cntlid, le16_to_cpu(id->cntlid));
2966                         ret = -EINVAL;
2967                         goto out_free;
2968                 }
2969
2970                 if (!nvme_discovery_ctrl(ctrl) && !ctrl->kas) {
2971                         dev_err(ctrl->device,
2972                                 "keep-alive support is mandatory for fabrics\n");
2973                         ret = -EINVAL;
2974                         goto out_free;
2975                 }
2976         } else {
2977                 ctrl->hmpre = le32_to_cpu(id->hmpre);
2978                 ctrl->hmmin = le32_to_cpu(id->hmmin);
2979                 ctrl->hmminds = le32_to_cpu(id->hmminds);
2980                 ctrl->hmmaxd = le16_to_cpu(id->hmmaxd);
2981         }
2982
2983         ret = nvme_mpath_init_identify(ctrl, id);
2984         if (ret < 0)
2985                 goto out_free;
2986
2987         if (ctrl->apst_enabled && !prev_apst_enabled)
2988                 dev_pm_qos_expose_latency_tolerance(ctrl->device);
2989         else if (!ctrl->apst_enabled && prev_apst_enabled)
2990                 dev_pm_qos_hide_latency_tolerance(ctrl->device);
2991
2992 out_free:
2993         kfree(id);
2994         return ret;
2995 }
2996
2997 /*
2998  * Initialize the cached copies of the Identify data and various controller
2999  * register in our nvme_ctrl structure.  This should be called as soon as
3000  * the admin queue is fully up and running.
3001  */
3002 int nvme_init_ctrl_finish(struct nvme_ctrl *ctrl)
3003 {
3004         int ret;
3005
3006         ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
3007         if (ret) {
3008                 dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
3009                 return ret;
3010         }
3011
3012         ctrl->sqsize = min_t(u16, NVME_CAP_MQES(ctrl->cap), ctrl->sqsize);
3013
3014         if (ctrl->vs >= NVME_VS(1, 1, 0))
3015                 ctrl->subsystem = NVME_CAP_NSSRC(ctrl->cap);
3016
3017         ret = nvme_init_identify(ctrl);
3018         if (ret)
3019                 return ret;
3020
3021         ret = nvme_init_non_mdts_limits(ctrl);
3022         if (ret < 0)
3023                 return ret;
3024
3025         ret = nvme_configure_apst(ctrl);
3026         if (ret < 0)
3027                 return ret;
3028
3029         ret = nvme_configure_timestamp(ctrl);
3030         if (ret < 0)
3031                 return ret;
3032
3033         ret = nvme_configure_directives(ctrl);
3034         if (ret < 0)
3035                 return ret;
3036
3037         ret = nvme_configure_acre(ctrl);
3038         if (ret < 0)
3039                 return ret;
3040
3041         if (!ctrl->identified && !nvme_discovery_ctrl(ctrl)) {
3042                 ret = nvme_hwmon_init(ctrl);
3043                 if (ret < 0)
3044                         return ret;
3045         }
3046
3047         ctrl->identified = true;
3048
3049         return 0;
3050 }
3051 EXPORT_SYMBOL_GPL(nvme_init_ctrl_finish);
3052
3053 static int nvme_dev_open(struct inode *inode, struct file *file)
3054 {
3055         struct nvme_ctrl *ctrl =
3056                 container_of(inode->i_cdev, struct nvme_ctrl, cdev);
3057
3058         switch (ctrl->state) {
3059         case NVME_CTRL_LIVE:
3060                 break;
3061         default:
3062                 return -EWOULDBLOCK;
3063         }
3064
3065         nvme_get_ctrl(ctrl);
3066         if (!try_module_get(ctrl->ops->module)) {
3067                 nvme_put_ctrl(ctrl);
3068                 return -EINVAL;
3069         }
3070
3071         file->private_data = ctrl;
3072         return 0;
3073 }
3074
3075 static int nvme_dev_release(struct inode *inode, struct file *file)
3076 {
3077         struct nvme_ctrl *ctrl =
3078                 container_of(inode->i_cdev, struct nvme_ctrl, cdev);
3079
3080         module_put(ctrl->ops->module);
3081         nvme_put_ctrl(ctrl);
3082         return 0;
3083 }
3084
3085 static const struct file_operations nvme_dev_fops = {
3086         .owner          = THIS_MODULE,
3087         .open           = nvme_dev_open,
3088         .release        = nvme_dev_release,
3089         .unlocked_ioctl = nvme_dev_ioctl,
3090         .compat_ioctl   = compat_ptr_ioctl,
3091 };
3092
3093 static ssize_t nvme_sysfs_reset(struct device *dev,
3094                                 struct device_attribute *attr, const char *buf,
3095                                 size_t count)
3096 {
3097         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3098         int ret;
3099
3100         ret = nvme_reset_ctrl_sync(ctrl);
3101         if (ret < 0)
3102                 return ret;
3103         return count;
3104 }
3105 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
3106
3107 static ssize_t nvme_sysfs_rescan(struct device *dev,
3108                                 struct device_attribute *attr, const char *buf,
3109                                 size_t count)
3110 {
3111         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3112
3113         nvme_queue_scan(ctrl);
3114         return count;
3115 }
3116 static DEVICE_ATTR(rescan_controller, S_IWUSR, NULL, nvme_sysfs_rescan);
3117
3118 static inline struct nvme_ns_head *dev_to_ns_head(struct device *dev)
3119 {
3120         struct gendisk *disk = dev_to_disk(dev);
3121
3122         if (disk->fops == &nvme_bdev_ops)
3123                 return nvme_get_ns_from_dev(dev)->head;
3124         else
3125                 return disk->private_data;
3126 }
3127
3128 static ssize_t wwid_show(struct device *dev, struct device_attribute *attr,
3129                 char *buf)
3130 {
3131         struct nvme_ns_head *head = dev_to_ns_head(dev);
3132         struct nvme_ns_ids *ids = &head->ids;
3133         struct nvme_subsystem *subsys = head->subsys;
3134         int serial_len = sizeof(subsys->serial);
3135         int model_len = sizeof(subsys->model);
3136
3137         if (!uuid_is_null(&ids->uuid))
3138                 return sysfs_emit(buf, "uuid.%pU\n", &ids->uuid);
3139
3140         if (memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3141                 return sysfs_emit(buf, "eui.%16phN\n", ids->nguid);
3142
3143         if (memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
3144                 return sysfs_emit(buf, "eui.%8phN\n", ids->eui64);
3145
3146         while (serial_len > 0 && (subsys->serial[serial_len - 1] == ' ' ||
3147                                   subsys->serial[serial_len - 1] == '\0'))
3148                 serial_len--;
3149         while (model_len > 0 && (subsys->model[model_len - 1] == ' ' ||
3150                                  subsys->model[model_len - 1] == '\0'))
3151                 model_len--;
3152
3153         return sysfs_emit(buf, "nvme.%04x-%*phN-%*phN-%08x\n", subsys->vendor_id,
3154                 serial_len, subsys->serial, model_len, subsys->model,
3155                 head->ns_id);
3156 }
3157 static DEVICE_ATTR_RO(wwid);
3158
3159 static ssize_t nguid_show(struct device *dev, struct device_attribute *attr,
3160                 char *buf)
3161 {
3162         return sysfs_emit(buf, "%pU\n", dev_to_ns_head(dev)->ids.nguid);
3163 }
3164 static DEVICE_ATTR_RO(nguid);
3165
3166 static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
3167                 char *buf)
3168 {
3169         struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
3170
3171         /* For backward compatibility expose the NGUID to userspace if
3172          * we have no UUID set
3173          */
3174         if (uuid_is_null(&ids->uuid)) {
3175                 printk_ratelimited(KERN_WARNING
3176                                    "No UUID available providing old NGUID\n");
3177                 return sysfs_emit(buf, "%pU\n", ids->nguid);
3178         }
3179         return sysfs_emit(buf, "%pU\n", &ids->uuid);
3180 }
3181 static DEVICE_ATTR_RO(uuid);
3182
3183 static ssize_t eui_show(struct device *dev, struct device_attribute *attr,
3184                 char *buf)
3185 {
3186         return sysfs_emit(buf, "%8ph\n", dev_to_ns_head(dev)->ids.eui64);
3187 }
3188 static DEVICE_ATTR_RO(eui);
3189
3190 static ssize_t nsid_show(struct device *dev, struct device_attribute *attr,
3191                 char *buf)
3192 {
3193         return sysfs_emit(buf, "%d\n", dev_to_ns_head(dev)->ns_id);
3194 }
3195 static DEVICE_ATTR_RO(nsid);
3196
3197 static struct attribute *nvme_ns_id_attrs[] = {
3198         &dev_attr_wwid.attr,
3199         &dev_attr_uuid.attr,
3200         &dev_attr_nguid.attr,
3201         &dev_attr_eui.attr,
3202         &dev_attr_nsid.attr,
3203 #ifdef CONFIG_NVME_MULTIPATH
3204         &dev_attr_ana_grpid.attr,
3205         &dev_attr_ana_state.attr,
3206 #endif
3207         NULL,
3208 };
3209
3210 static umode_t nvme_ns_id_attrs_are_visible(struct kobject *kobj,
3211                 struct attribute *a, int n)
3212 {
3213         struct device *dev = container_of(kobj, struct device, kobj);
3214         struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
3215
3216         if (a == &dev_attr_uuid.attr) {
3217                 if (uuid_is_null(&ids->uuid) &&
3218                     !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3219                         return 0;
3220         }
3221         if (a == &dev_attr_nguid.attr) {
3222                 if (!memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3223                         return 0;
3224         }
3225         if (a == &dev_attr_eui.attr) {
3226                 if (!memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
3227                         return 0;
3228         }
3229 #ifdef CONFIG_NVME_MULTIPATH
3230         if (a == &dev_attr_ana_grpid.attr || a == &dev_attr_ana_state.attr) {
3231                 if (dev_to_disk(dev)->fops != &nvme_bdev_ops) /* per-path attr */
3232                         return 0;
3233                 if (!nvme_ctrl_use_ana(nvme_get_ns_from_dev(dev)->ctrl))
3234                         return 0;
3235         }
3236 #endif
3237         return a->mode;
3238 }
3239
3240 static const struct attribute_group nvme_ns_id_attr_group = {
3241         .attrs          = nvme_ns_id_attrs,
3242         .is_visible     = nvme_ns_id_attrs_are_visible,
3243 };
3244
3245 const struct attribute_group *nvme_ns_id_attr_groups[] = {
3246         &nvme_ns_id_attr_group,
3247         NULL,
3248 };
3249
3250 #define nvme_show_str_function(field)                                           \
3251 static ssize_t  field##_show(struct device *dev,                                \
3252                             struct device_attribute *attr, char *buf)           \
3253 {                                                                               \
3254         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);                          \
3255         return sysfs_emit(buf, "%.*s\n",                                        \
3256                 (int)sizeof(ctrl->subsys->field), ctrl->subsys->field);         \
3257 }                                                                               \
3258 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3259
3260 nvme_show_str_function(model);
3261 nvme_show_str_function(serial);
3262 nvme_show_str_function(firmware_rev);
3263
3264 #define nvme_show_int_function(field)                                           \
3265 static ssize_t  field##_show(struct device *dev,                                \
3266                             struct device_attribute *attr, char *buf)           \
3267 {                                                                               \
3268         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);                          \
3269         return sysfs_emit(buf, "%d\n", ctrl->field);                            \
3270 }                                                                               \
3271 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3272
3273 nvme_show_int_function(cntlid);
3274 nvme_show_int_function(numa_node);
3275 nvme_show_int_function(queue_count);
3276 nvme_show_int_function(sqsize);
3277 nvme_show_int_function(kato);
3278
3279 static ssize_t nvme_sysfs_delete(struct device *dev,
3280                                 struct device_attribute *attr, const char *buf,
3281                                 size_t count)
3282 {
3283         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3284
3285         if (device_remove_file_self(dev, attr))
3286                 nvme_delete_ctrl_sync(ctrl);
3287         return count;
3288 }
3289 static DEVICE_ATTR(delete_controller, S_IWUSR, NULL, nvme_sysfs_delete);
3290
3291 static ssize_t nvme_sysfs_show_transport(struct device *dev,
3292                                          struct device_attribute *attr,
3293                                          char *buf)
3294 {
3295         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3296
3297         return sysfs_emit(buf, "%s\n", ctrl->ops->name);
3298 }
3299 static DEVICE_ATTR(transport, S_IRUGO, nvme_sysfs_show_transport, NULL);
3300
3301 static ssize_t nvme_sysfs_show_state(struct device *dev,
3302                                      struct device_attribute *attr,
3303                                      char *buf)
3304 {
3305         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3306         static const char *const state_name[] = {
3307                 [NVME_CTRL_NEW]         = "new",
3308                 [NVME_CTRL_LIVE]        = "live",
3309                 [NVME_CTRL_RESETTING]   = "resetting",
3310                 [NVME_CTRL_CONNECTING]  = "connecting",
3311                 [NVME_CTRL_DELETING]    = "deleting",
3312                 [NVME_CTRL_DELETING_NOIO]= "deleting (no IO)",
3313                 [NVME_CTRL_DEAD]        = "dead",
3314         };
3315
3316         if ((unsigned)ctrl->state < ARRAY_SIZE(state_name) &&
3317             state_name[ctrl->state])
3318                 return sysfs_emit(buf, "%s\n", state_name[ctrl->state]);
3319
3320         return sysfs_emit(buf, "unknown state\n");
3321 }
3322
3323 static DEVICE_ATTR(state, S_IRUGO, nvme_sysfs_show_state, NULL);
3324
3325 static ssize_t nvme_sysfs_show_subsysnqn(struct device *dev,
3326                                          struct device_attribute *attr,
3327                                          char *buf)
3328 {
3329         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3330
3331         return sysfs_emit(buf, "%s\n", ctrl->subsys->subnqn);
3332 }
3333 static DEVICE_ATTR(subsysnqn, S_IRUGO, nvme_sysfs_show_subsysnqn, NULL);
3334
3335 static ssize_t nvme_sysfs_show_hostnqn(struct device *dev,
3336                                         struct device_attribute *attr,
3337                                         char *buf)
3338 {
3339         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3340
3341         return sysfs_emit(buf, "%s\n", ctrl->opts->host->nqn);
3342 }
3343 static DEVICE_ATTR(hostnqn, S_IRUGO, nvme_sysfs_show_hostnqn, NULL);
3344
3345 static ssize_t nvme_sysfs_show_hostid(struct device *dev,
3346                                         struct device_attribute *attr,
3347                                         char *buf)
3348 {
3349         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3350
3351         return sysfs_emit(buf, "%pU\n", &ctrl->opts->host->id);
3352 }
3353 static DEVICE_ATTR(hostid, S_IRUGO, nvme_sysfs_show_hostid, NULL);
3354
3355 static ssize_t nvme_sysfs_show_address(struct device *dev,
3356                                          struct device_attribute *attr,
3357                                          char *buf)
3358 {
3359         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3360
3361         return ctrl->ops->get_address(ctrl, buf, PAGE_SIZE);
3362 }
3363 static DEVICE_ATTR(address, S_IRUGO, nvme_sysfs_show_address, NULL);
3364
3365 static ssize_t nvme_ctrl_loss_tmo_show(struct device *dev,
3366                 struct device_attribute *attr, char *buf)
3367 {
3368         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3369         struct nvmf_ctrl_options *opts = ctrl->opts;
3370
3371         if (ctrl->opts->max_reconnects == -1)
3372                 return sysfs_emit(buf, "off\n");
3373         return sysfs_emit(buf, "%d\n",
3374                           opts->max_reconnects * opts->reconnect_delay);
3375 }
3376
3377 static ssize_t nvme_ctrl_loss_tmo_store(struct device *dev,
3378                 struct device_attribute *attr, const char *buf, size_t count)
3379 {
3380         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3381         struct nvmf_ctrl_options *opts = ctrl->opts;
3382         int ctrl_loss_tmo, err;
3383
3384         err = kstrtoint(buf, 10, &ctrl_loss_tmo);
3385         if (err)
3386                 return -EINVAL;
3387
3388         if (ctrl_loss_tmo < 0)
3389                 opts->max_reconnects = -1;
3390         else
3391                 opts->max_reconnects = DIV_ROUND_UP(ctrl_loss_tmo,
3392                                                 opts->reconnect_delay);
3393         return count;
3394 }
3395 static DEVICE_ATTR(ctrl_loss_tmo, S_IRUGO | S_IWUSR,
3396         nvme_ctrl_loss_tmo_show, nvme_ctrl_loss_tmo_store);
3397
3398 static ssize_t nvme_ctrl_reconnect_delay_show(struct device *dev,
3399                 struct device_attribute *attr, char *buf)
3400 {
3401         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3402
3403         if (ctrl->opts->reconnect_delay == -1)
3404                 return sysfs_emit(buf, "off\n");
3405         return sysfs_emit(buf, "%d\n", ctrl->opts->reconnect_delay);
3406 }
3407
3408 static ssize_t nvme_ctrl_reconnect_delay_store(struct device *dev,
3409                 struct device_attribute *attr, const char *buf, size_t count)
3410 {
3411         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3412         unsigned int v;
3413         int err;
3414
3415         err = kstrtou32(buf, 10, &v);
3416         if (err)
3417                 return err;
3418
3419         ctrl->opts->reconnect_delay = v;
3420         return count;
3421 }
3422 static DEVICE_ATTR(reconnect_delay, S_IRUGO | S_IWUSR,
3423         nvme_ctrl_reconnect_delay_show, nvme_ctrl_reconnect_delay_store);
3424
3425 static ssize_t nvme_ctrl_fast_io_fail_tmo_show(struct device *dev,
3426                 struct device_attribute *attr, char *buf)
3427 {
3428         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3429
3430         if (ctrl->opts->fast_io_fail_tmo == -1)
3431                 return sysfs_emit(buf, "off\n");
3432         return sysfs_emit(buf, "%d\n", ctrl->opts->fast_io_fail_tmo);
3433 }
3434
3435 static ssize_t nvme_ctrl_fast_io_fail_tmo_store(struct device *dev,
3436                 struct device_attribute *attr, const char *buf, size_t count)
3437 {
3438         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3439         struct nvmf_ctrl_options *opts = ctrl->opts;
3440         int fast_io_fail_tmo, err;
3441
3442         err = kstrtoint(buf, 10, &fast_io_fail_tmo);
3443         if (err)
3444                 return -EINVAL;
3445
3446         if (fast_io_fail_tmo < 0)
3447                 opts->fast_io_fail_tmo = -1;
3448         else
3449                 opts->fast_io_fail_tmo = fast_io_fail_tmo;
3450         return count;
3451 }
3452 static DEVICE_ATTR(fast_io_fail_tmo, S_IRUGO | S_IWUSR,
3453         nvme_ctrl_fast_io_fail_tmo_show, nvme_ctrl_fast_io_fail_tmo_store);
3454
3455 static struct attribute *nvme_dev_attrs[] = {
3456         &dev_attr_reset_controller.attr,
3457         &dev_attr_rescan_controller.attr,
3458         &dev_attr_model.attr,
3459         &dev_attr_serial.attr,
3460         &dev_attr_firmware_rev.attr,
3461         &dev_attr_cntlid.attr,
3462         &dev_attr_delete_controller.attr,
3463         &dev_attr_transport.attr,
3464         &dev_attr_subsysnqn.attr,
3465         &dev_attr_address.attr,
3466         &dev_attr_state.attr,
3467         &dev_attr_numa_node.attr,
3468         &dev_attr_queue_count.attr,
3469         &dev_attr_sqsize.attr,
3470         &dev_attr_hostnqn.attr,
3471         &dev_attr_hostid.attr,
3472         &dev_attr_ctrl_loss_tmo.attr,
3473         &dev_attr_reconnect_delay.attr,
3474         &dev_attr_fast_io_fail_tmo.attr,
3475         &dev_attr_kato.attr,
3476         NULL
3477 };
3478
3479 static umode_t nvme_dev_attrs_are_visible(struct kobject *kobj,
3480                 struct attribute *a, int n)
3481 {
3482         struct device *dev = container_of(kobj, struct device, kobj);
3483         struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3484
3485         if (a == &dev_attr_delete_controller.attr && !ctrl->ops->delete_ctrl)
3486                 return 0;
3487         if (a == &dev_attr_address.attr && !ctrl->ops->get_address)
3488                 return 0;
3489         if (a == &dev_attr_hostnqn.attr && !ctrl->opts)
3490                 return 0;
3491         if (a == &dev_attr_hostid.attr && !ctrl->opts)
3492                 return 0;
3493         if (a == &dev_attr_ctrl_loss_tmo.attr && !ctrl->opts)
3494                 return 0;
3495         if (a == &dev_attr_reconnect_delay.attr && !ctrl->opts)
3496                 return 0;
3497         if (a == &dev_attr_fast_io_fail_tmo.attr && !ctrl->opts)
3498                 return 0;
3499
3500         return a->mode;
3501 }
3502
3503 static const struct attribute_group nvme_dev_attrs_group = {
3504         .attrs          = nvme_dev_attrs,
3505         .is_visible     = nvme_dev_attrs_are_visible,
3506 };
3507
3508 static const struct attribute_group *nvme_dev_attr_groups[] = {
3509         &nvme_dev_attrs_group,
3510         NULL,
3511 };
3512
3513 static struct nvme_ns_head *nvme_find_ns_head(struct nvme_subsystem *subsys,
3514                 unsigned nsid)
3515 {
3516         struct nvme_ns_head *h;
3517
3518         lockdep_assert_held(&subsys->lock);
3519
3520         list_for_each_entry(h, &subsys->nsheads, entry) {
3521                 if (h->ns_id != nsid)
3522                         continue;
3523                 if (!list_empty(&h->list) && nvme_tryget_ns_head(h))
3524                         return h;
3525         }
3526
3527         return NULL;
3528 }
3529
3530 static int __nvme_check_ids(struct nvme_subsystem *subsys,
3531                 struct nvme_ns_head *new)
3532 {
3533         struct nvme_ns_head *h;
3534
3535         lockdep_assert_held(&subsys->lock);
3536
3537         list_for_each_entry(h, &subsys->nsheads, entry) {
3538                 if (nvme_ns_ids_valid(&new->ids) &&
3539                     nvme_ns_ids_equal(&new->ids, &h->ids))
3540                         return -EINVAL;
3541         }
3542
3543         return 0;
3544 }
3545
3546 static void nvme_cdev_rel(struct device *dev)
3547 {
3548         ida_simple_remove(&nvme_ns_chr_minor_ida, MINOR(dev->devt));
3549 }
3550
3551 void nvme_cdev_del(struct cdev *cdev, struct device *cdev_device)
3552 {
3553         cdev_device_del(cdev, cdev_device);
3554         put_device(cdev_device);
3555 }
3556
3557 int nvme_cdev_add(struct cdev *cdev, struct device *cdev_device,
3558                 const struct file_operations *fops, struct module *owner)
3559 {
3560         int minor, ret;
3561
3562         minor = ida_simple_get(&nvme_ns_chr_minor_ida, 0, 0, GFP_KERNEL);
3563         if (minor < 0)
3564                 return minor;
3565         cdev_device->devt = MKDEV(MAJOR(nvme_ns_chr_devt), minor);
3566         cdev_device->class = nvme_ns_chr_class;
3567         cdev_device->release = nvme_cdev_rel;
3568         device_initialize(cdev_device);
3569         cdev_init(cdev, fops);
3570         cdev->owner = owner;
3571         ret = cdev_device_add(cdev, cdev_device);
3572         if (ret)
3573                 put_device(cdev_device);
3574
3575         return ret;
3576 }
3577
3578 static int nvme_ns_chr_open(struct inode *inode, struct file *file)
3579 {
3580         return nvme_ns_open(container_of(inode->i_cdev, struct nvme_ns, cdev));
3581 }
3582
3583 static int nvme_ns_chr_release(struct inode *inode, struct file *file)
3584 {
3585         nvme_ns_release(container_of(inode->i_cdev, struct nvme_ns, cdev));
3586         return 0;
3587 }
3588
3589 static const struct file_operations nvme_ns_chr_fops = {
3590         .owner          = THIS_MODULE,
3591         .open           = nvme_ns_chr_open,
3592         .release        = nvme_ns_chr_release,
3593         .unlocked_ioctl = nvme_ns_chr_ioctl,
3594         .compat_ioctl   = compat_ptr_ioctl,
3595 };
3596
3597 static int nvme_add_ns_cdev(struct nvme_ns *ns)
3598 {
3599         int ret;
3600
3601         ns->cdev_device.parent = ns->ctrl->device;
3602         ret = dev_set_name(&ns->cdev_device, "ng%dn%d",
3603                            ns->ctrl->instance, ns->head->instance);
3604         if (ret)
3605                 return ret;
3606
3607         return nvme_cdev_add(&ns->cdev, &ns->cdev_device, &nvme_ns_chr_fops,
3608                              ns->ctrl->ops->module);
3609 }
3610
3611 static struct nvme_ns_head *nvme_alloc_ns_head(struct nvme_ctrl *ctrl,
3612                 unsigned nsid, struct nvme_ns_ids *ids)
3613 {
3614         struct nvme_ns_head *head;
3615         size_t size = sizeof(*head);
3616         int ret = -ENOMEM;
3617
3618 #ifdef CONFIG_NVME_MULTIPATH
3619         size += num_possible_nodes() * sizeof(struct nvme_ns *);
3620 #endif
3621
3622         head = kzalloc(size, GFP_KERNEL);
3623         if (!head)
3624                 goto out;
3625         ret = ida_simple_get(&ctrl->subsys->ns_ida, 1, 0, GFP_KERNEL);
3626         if (ret < 0)
3627                 goto out_free_head;
3628         head->instance = ret;
3629         INIT_LIST_HEAD(&head->list);
3630         ret = init_srcu_struct(&head->srcu);
3631         if (ret)
3632                 goto out_ida_remove;
3633         head->subsys = ctrl->subsys;
3634         head->ns_id = nsid;
3635         head->ids = *ids;
3636         kref_init(&head->ref);
3637
3638         ret = __nvme_check_ids(ctrl->subsys, head);
3639         if (ret) {
3640                 dev_err(ctrl->device,
3641                         "duplicate IDs for nsid %d\n", nsid);
3642                 goto out_cleanup_srcu;
3643         }
3644
3645         if (head->ids.csi) {
3646                 ret = nvme_get_effects_log(ctrl, head->ids.csi, &head->effects);
3647                 if (ret)
3648                         goto out_cleanup_srcu;
3649         } else
3650                 head->effects = ctrl->effects;
3651
3652         ret = nvme_mpath_alloc_disk(ctrl, head);
3653         if (ret)
3654                 goto out_cleanup_srcu;
3655
3656         list_add_tail(&head->entry, &ctrl->subsys->nsheads);
3657
3658         kref_get(&ctrl->subsys->ref);
3659
3660         return head;
3661 out_cleanup_srcu:
3662         cleanup_srcu_struct(&head->srcu);
3663 out_ida_remove:
3664         ida_simple_remove(&ctrl->subsys->ns_ida, head->instance);
3665 out_free_head:
3666         kfree(head);
3667 out:
3668         if (ret > 0)
3669                 ret = blk_status_to_errno(nvme_error_status(ret));
3670         return ERR_PTR(ret);
3671 }
3672
3673 static int nvme_init_ns_head(struct nvme_ns *ns, unsigned nsid,
3674                 struct nvme_ns_ids *ids, bool is_shared)
3675 {
3676         struct nvme_ctrl *ctrl = ns->ctrl;
3677         struct nvme_ns_head *head = NULL;
3678         int ret = 0;
3679
3680         mutex_lock(&ctrl->subsys->lock);
3681         head = nvme_find_ns_head(ctrl->subsys, nsid);
3682         if (!head) {
3683                 head = nvme_alloc_ns_head(ctrl, nsid, ids);
3684                 if (IS_ERR(head)) {
3685                         ret = PTR_ERR(head);
3686                         goto out_unlock;
3687                 }
3688                 head->shared = is_shared;
3689         } else {
3690                 ret = -EINVAL;
3691                 if (!is_shared || !head->shared) {
3692                         dev_err(ctrl->device,
3693                                 "Duplicate unshared namespace %d\n", nsid);
3694                         goto out_put_ns_head;
3695                 }
3696                 if (!nvme_ns_ids_equal(&head->ids, ids)) {
3697                         dev_err(ctrl->device,
3698                                 "IDs don't match for shared namespace %d\n",
3699                                         nsid);
3700                         goto out_put_ns_head;
3701                 }
3702         }
3703
3704         list_add_tail_rcu(&ns->siblings, &head->list);
3705         ns->head = head;
3706         mutex_unlock(&ctrl->subsys->lock);
3707         return 0;
3708
3709 out_put_ns_head:
3710         nvme_put_ns_head(head);
3711 out_unlock:
3712         mutex_unlock(&ctrl->subsys->lock);
3713         return ret;
3714 }
3715
3716 struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3717 {
3718         struct nvme_ns *ns, *ret = NULL;
3719
3720         down_read(&ctrl->namespaces_rwsem);
3721         list_for_each_entry(ns, &ctrl->namespaces, list) {
3722                 if (ns->head->ns_id == nsid) {
3723                         if (!nvme_get_ns(ns))
3724                                 continue;
3725                         ret = ns;
3726                         break;
3727                 }
3728                 if (ns->head->ns_id > nsid)
3729                         break;
3730         }
3731         up_read(&ctrl->namespaces_rwsem);
3732         return ret;
3733 }
3734 EXPORT_SYMBOL_NS_GPL(nvme_find_get_ns, NVME_TARGET_PASSTHRU);
3735
3736 /*
3737  * Add the namespace to the controller list while keeping the list ordered.
3738  */
3739 static void nvme_ns_add_to_ctrl_list(struct nvme_ns *ns)
3740 {
3741         struct nvme_ns *tmp;
3742
3743         list_for_each_entry_reverse(tmp, &ns->ctrl->namespaces, list) {
3744                 if (tmp->head->ns_id < ns->head->ns_id) {
3745                         list_add(&ns->list, &tmp->list);
3746                         return;
3747                 }
3748         }
3749         list_add(&ns->list, &ns->ctrl->namespaces);
3750 }
3751
3752 static void nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid,
3753                 struct nvme_ns_ids *ids)
3754 {
3755         struct nvme_ns *ns;
3756         struct gendisk *disk;
3757         struct nvme_id_ns *id;
3758         int node = ctrl->numa_node;
3759
3760         if (nvme_identify_ns(ctrl, nsid, ids, &id))
3761                 return;
3762
3763         ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
3764         if (!ns)
3765                 goto out_free_id;
3766
3767         disk = blk_mq_alloc_disk(ctrl->tagset, ns);
3768         if (IS_ERR(disk))
3769                 goto out_free_ns;
3770         disk->fops = &nvme_bdev_ops;
3771         disk->private_data = ns;
3772
3773         ns->disk = disk;
3774         ns->queue = disk->queue;
3775
3776         if (ctrl->opts && ctrl->opts->data_digest)
3777                 blk_queue_flag_set(QUEUE_FLAG_STABLE_WRITES, ns->queue);
3778
3779         blk_queue_flag_set(QUEUE_FLAG_NONROT, ns->queue);
3780         if (ctrl->ops->flags & NVME_F_PCI_P2PDMA)
3781                 blk_queue_flag_set(QUEUE_FLAG_PCI_P2PDMA, ns->queue);
3782
3783         ns->ctrl = ctrl;
3784         kref_init(&ns->kref);
3785
3786         if (nvme_init_ns_head(ns, nsid, ids, id->nmic & NVME_NS_NMIC_SHARED))
3787                 goto out_cleanup_disk;
3788
3789         /*
3790          * Without the multipath code enabled, multiple controller per
3791          * subsystems are visible as devices and thus we cannot use the
3792          * subsystem instance.
3793          */
3794         if (!nvme_mpath_set_disk_name(ns, disk->disk_name, &disk->flags))
3795                 sprintf(disk->disk_name, "nvme%dn%d", ctrl->instance,
3796                         ns->head->instance);
3797
3798         if (nvme_update_ns_info(ns, id))
3799                 goto out_unlink_ns;
3800
3801         down_write(&ctrl->namespaces_rwsem);
3802         nvme_ns_add_to_ctrl_list(ns);
3803         up_write(&ctrl->namespaces_rwsem);
3804         nvme_get_ctrl(ctrl);
3805
3806         if (device_add_disk(ctrl->device, ns->disk, nvme_ns_id_attr_groups))
3807                 goto out_cleanup_ns_from_list;
3808
3809         if (!nvme_ns_head_multipath(ns->head))
3810                 nvme_add_ns_cdev(ns);
3811
3812         nvme_mpath_add_disk(ns, id);
3813         nvme_fault_inject_init(&ns->fault_inject, ns->disk->disk_name);
3814         kfree(id);
3815
3816         return;
3817
3818  out_cleanup_ns_from_list:
3819         nvme_put_ctrl(ctrl);
3820         down_write(&ctrl->namespaces_rwsem);
3821         list_del_init(&ns->list);
3822         up_write(&ctrl->namespaces_rwsem);
3823  out_unlink_ns:
3824         mutex_lock(&ctrl->subsys->lock);
3825         list_del_rcu(&ns->siblings);
3826         if (list_empty(&ns->head->list))
3827                 list_del_init(&ns->head->entry);
3828         mutex_unlock(&ctrl->subsys->lock);
3829         nvme_put_ns_head(ns->head);
3830  out_cleanup_disk:
3831         blk_cleanup_disk(disk);
3832  out_free_ns:
3833         kfree(ns);
3834  out_free_id:
3835         kfree(id);
3836 }
3837
3838 static void nvme_ns_remove(struct nvme_ns *ns)
3839 {
3840         bool last_path = false;
3841
3842         if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
3843                 return;
3844
3845         clear_bit(NVME_NS_READY, &ns->flags);
3846         set_capacity(ns->disk, 0);
3847         nvme_fault_inject_fini(&ns->fault_inject);
3848
3849         mutex_lock(&ns->ctrl->subsys->lock);
3850         list_del_rcu(&ns->siblings);
3851         if (list_empty(&ns->head->list)) {
3852                 list_del_init(&ns->head->entry);
3853                 last_path = true;
3854         }
3855         mutex_unlock(&ns->ctrl->subsys->lock);
3856
3857         /* guarantee not available in head->list */
3858         synchronize_rcu();
3859
3860         /* wait for concurrent submissions */
3861         if (nvme_mpath_clear_current_path(ns))
3862                 synchronize_srcu(&ns->head->srcu);
3863
3864         if (!nvme_ns_head_multipath(ns->head))
3865                 nvme_cdev_del(&ns->cdev, &ns->cdev_device);
3866         del_gendisk(ns->disk);
3867         blk_cleanup_queue(ns->queue);
3868
3869         down_write(&ns->ctrl->namespaces_rwsem);
3870         list_del_init(&ns->list);
3871         up_write(&ns->ctrl->namespaces_rwsem);
3872
3873         if (last_path)
3874                 nvme_mpath_shutdown_disk(ns->head);
3875         nvme_put_ns(ns);
3876 }
3877
3878 static void nvme_ns_remove_by_nsid(struct nvme_ctrl *ctrl, u32 nsid)
3879 {
3880         struct nvme_ns *ns = nvme_find_get_ns(ctrl, nsid);
3881
3882         if (ns) {
3883                 nvme_ns_remove(ns);
3884                 nvme_put_ns(ns);
3885         }
3886 }
3887
3888 static void nvme_validate_ns(struct nvme_ns *ns, struct nvme_ns_ids *ids)
3889 {
3890         struct nvme_id_ns *id;
3891         int ret = NVME_SC_INVALID_NS | NVME_SC_DNR;
3892
3893         if (test_bit(NVME_NS_DEAD, &ns->flags))
3894                 goto out;
3895
3896         ret = nvme_identify_ns(ns->ctrl, ns->head->ns_id, ids, &id);
3897         if (ret)
3898                 goto out;
3899
3900         ret = NVME_SC_INVALID_NS | NVME_SC_DNR;
3901         if (!nvme_ns_ids_equal(&ns->head->ids, ids)) {
3902                 dev_err(ns->ctrl->device,
3903                         "identifiers changed for nsid %d\n", ns->head->ns_id);
3904                 goto out_free_id;
3905         }
3906
3907         ret = nvme_update_ns_info(ns, id);
3908
3909 out_free_id:
3910         kfree(id);
3911 out:
3912         /*
3913          * Only remove the namespace if we got a fatal error back from the
3914          * device, otherwise ignore the error and just move on.
3915          *
3916          * TODO: we should probably schedule a delayed retry here.
3917          */
3918         if (ret > 0 && (ret & NVME_SC_DNR))
3919                 nvme_ns_remove(ns);
3920 }
3921
3922 static void nvme_validate_or_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3923 {
3924         struct nvme_ns_ids ids = { };
3925         struct nvme_ns *ns;
3926
3927         if (nvme_identify_ns_descs(ctrl, nsid, &ids))
3928                 return;
3929
3930         ns = nvme_find_get_ns(ctrl, nsid);
3931         if (ns) {
3932                 nvme_validate_ns(ns, &ids);
3933                 nvme_put_ns(ns);
3934                 return;
3935         }
3936
3937         switch (ids.csi) {
3938         case NVME_CSI_NVM:
3939                 nvme_alloc_ns(ctrl, nsid, &ids);
3940                 break;
3941         case NVME_CSI_ZNS:
3942                 if (!IS_ENABLED(CONFIG_BLK_DEV_ZONED)) {
3943                         dev_warn(ctrl->device,
3944                                 "nsid %u not supported without CONFIG_BLK_DEV_ZONED\n",
3945                                 nsid);
3946                         break;
3947                 }
3948                 if (!nvme_multi_css(ctrl)) {
3949                         dev_warn(ctrl->device,
3950                                 "command set not reported for nsid: %d\n",
3951                                 nsid);
3952                         break;
3953                 }
3954                 nvme_alloc_ns(ctrl, nsid, &ids);
3955                 break;
3956         default:
3957                 dev_warn(ctrl->device, "unknown csi %u for nsid %u\n",
3958                         ids.csi, nsid);
3959                 break;
3960         }
3961 }
3962
3963 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
3964                                         unsigned nsid)
3965 {
3966         struct nvme_ns *ns, *next;
3967         LIST_HEAD(rm_list);
3968
3969         down_write(&ctrl->namespaces_rwsem);
3970         list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
3971                 if (ns->head->ns_id > nsid || test_bit(NVME_NS_DEAD, &ns->flags))
3972                         list_move_tail(&ns->list, &rm_list);
3973         }
3974         up_write(&ctrl->namespaces_rwsem);
3975
3976         list_for_each_entry_safe(ns, next, &rm_list, list)
3977                 nvme_ns_remove(ns);
3978
3979 }
3980
3981 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl)
3982 {
3983         const int nr_entries = NVME_IDENTIFY_DATA_SIZE / sizeof(__le32);
3984         __le32 *ns_list;
3985         u32 prev = 0;
3986         int ret = 0, i;
3987
3988         if (nvme_ctrl_limited_cns(ctrl))
3989                 return -EOPNOTSUPP;
3990
3991         ns_list = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
3992         if (!ns_list)
3993                 return -ENOMEM;
3994
3995         for (;;) {
3996                 struct nvme_command cmd = {
3997                         .identify.opcode        = nvme_admin_identify,
3998                         .identify.cns           = NVME_ID_CNS_NS_ACTIVE_LIST,
3999                         .identify.nsid          = cpu_to_le32(prev),
4000                 };
4001
4002                 ret = nvme_submit_sync_cmd(ctrl->admin_q, &cmd, ns_list,
4003                                             NVME_IDENTIFY_DATA_SIZE);
4004                 if (ret) {
4005                         dev_warn(ctrl->device,
4006                                 "Identify NS List failed (status=0x%x)\n", ret);
4007                         goto free;
4008                 }
4009
4010                 for (i = 0; i < nr_entries; i++) {
4011                         u32 nsid = le32_to_cpu(ns_list[i]);
4012
4013                         if (!nsid)      /* end of the list? */
4014                                 goto out;
4015                         nvme_validate_or_alloc_ns(ctrl, nsid);
4016                         while (++prev < nsid)
4017                                 nvme_ns_remove_by_nsid(ctrl, prev);
4018                 }
4019         }
4020  out:
4021         nvme_remove_invalid_namespaces(ctrl, prev);
4022  free:
4023         kfree(ns_list);
4024         return ret;
4025 }
4026
4027 static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl)
4028 {
4029         struct nvme_id_ctrl *id;
4030         u32 nn, i;
4031
4032         if (nvme_identify_ctrl(ctrl, &id))
4033                 return;
4034         nn = le32_to_cpu(id->nn);
4035         kfree(id);
4036
4037         for (i = 1; i <= nn; i++)
4038                 nvme_validate_or_alloc_ns(ctrl, i);
4039
4040         nvme_remove_invalid_namespaces(ctrl, nn);
4041 }
4042
4043 static void nvme_clear_changed_ns_log(struct nvme_ctrl *ctrl)
4044 {
4045         size_t log_size = NVME_MAX_CHANGED_NAMESPACES * sizeof(__le32);
4046         __le32 *log;
4047         int error;
4048
4049         log = kzalloc(log_size, GFP_KERNEL);
4050         if (!log)
4051                 return;
4052
4053         /*
4054          * We need to read the log to clear the AEN, but we don't want to rely
4055          * on it for the changed namespace information as userspace could have
4056          * raced with us in reading the log page, which could cause us to miss
4057          * updates.
4058          */
4059         error = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CHANGED_NS, 0,
4060                         NVME_CSI_NVM, log, log_size, 0);
4061         if (error)
4062                 dev_warn(ctrl->device,
4063                         "reading changed ns log failed: %d\n", error);
4064
4065         kfree(log);
4066 }
4067
4068 static void nvme_scan_work(struct work_struct *work)
4069 {
4070         struct nvme_ctrl *ctrl =
4071                 container_of(work, struct nvme_ctrl, scan_work);
4072
4073         /* No tagset on a live ctrl means IO queues could not created */
4074         if (ctrl->state != NVME_CTRL_LIVE || !ctrl->tagset)
4075                 return;
4076
4077         if (test_and_clear_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events)) {
4078                 dev_info(ctrl->device, "rescanning namespaces.\n");
4079                 nvme_clear_changed_ns_log(ctrl);
4080         }
4081
4082         mutex_lock(&ctrl->scan_lock);
4083         if (nvme_scan_ns_list(ctrl) != 0)
4084                 nvme_scan_ns_sequential(ctrl);
4085         mutex_unlock(&ctrl->scan_lock);
4086 }
4087
4088 /*
4089  * This function iterates the namespace list unlocked to allow recovery from
4090  * controller failure. It is up to the caller to ensure the namespace list is
4091  * not modified by scan work while this function is executing.
4092  */
4093 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
4094 {
4095         struct nvme_ns *ns, *next;
4096         LIST_HEAD(ns_list);
4097
4098         /*
4099          * make sure to requeue I/O to all namespaces as these
4100          * might result from the scan itself and must complete
4101          * for the scan_work to make progress
4102          */
4103         nvme_mpath_clear_ctrl_paths(ctrl);
4104
4105         /* prevent racing with ns scanning */
4106         flush_work(&ctrl->scan_work);
4107
4108         /*
4109          * The dead states indicates the controller was not gracefully
4110          * disconnected. In that case, we won't be able to flush any data while
4111          * removing the namespaces' disks; fail all the queues now to avoid
4112          * potentially having to clean up the failed sync later.
4113          */
4114         if (ctrl->state == NVME_CTRL_DEAD)
4115                 nvme_kill_queues(ctrl);
4116
4117         /* this is a no-op when called from the controller reset handler */
4118         nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING_NOIO);
4119
4120         down_write(&ctrl->namespaces_rwsem);
4121         list_splice_init(&ctrl->namespaces, &ns_list);
4122         up_write(&ctrl->namespaces_rwsem);
4123
4124         list_for_each_entry_safe(ns, next, &ns_list, list)
4125                 nvme_ns_remove(ns);
4126 }
4127 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
4128
4129 static int nvme_class_uevent(struct device *dev, struct kobj_uevent_env *env)
4130 {
4131         struct nvme_ctrl *ctrl =
4132                 container_of(dev, struct nvme_ctrl, ctrl_device);
4133         struct nvmf_ctrl_options *opts = ctrl->opts;
4134         int ret;
4135
4136         ret = add_uevent_var(env, "NVME_TRTYPE=%s", ctrl->ops->name);
4137         if (ret)
4138                 return ret;
4139
4140         if (opts) {
4141                 ret = add_uevent_var(env, "NVME_TRADDR=%s", opts->traddr);
4142                 if (ret)
4143                         return ret;
4144
4145                 ret = add_uevent_var(env, "NVME_TRSVCID=%s",
4146                                 opts->trsvcid ?: "none");
4147                 if (ret)
4148                         return ret;
4149
4150                 ret = add_uevent_var(env, "NVME_HOST_TRADDR=%s",
4151                                 opts->host_traddr ?: "none");
4152                 if (ret)
4153                         return ret;
4154
4155                 ret = add_uevent_var(env, "NVME_HOST_IFACE=%s",
4156                                 opts->host_iface ?: "none");
4157         }
4158         return ret;
4159 }
4160
4161 static void nvme_aen_uevent(struct nvme_ctrl *ctrl)
4162 {
4163         char *envp[2] = { NULL, NULL };
4164         u32 aen_result = ctrl->aen_result;
4165
4166         ctrl->aen_result = 0;
4167         if (!aen_result)
4168                 return;
4169
4170         envp[0] = kasprintf(GFP_KERNEL, "NVME_AEN=%#08x", aen_result);
4171         if (!envp[0])
4172                 return;
4173         kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp);
4174         kfree(envp[0]);
4175 }
4176
4177 static void nvme_async_event_work(struct work_struct *work)
4178 {
4179         struct nvme_ctrl *ctrl =
4180                 container_of(work, struct nvme_ctrl, async_event_work);
4181
4182         nvme_aen_uevent(ctrl);
4183         ctrl->ops->submit_async_event(ctrl);
4184 }
4185
4186 static bool nvme_ctrl_pp_status(struct nvme_ctrl *ctrl)
4187 {
4188
4189         u32 csts;
4190
4191         if (ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts))
4192                 return false;
4193
4194         if (csts == ~0)
4195                 return false;
4196
4197         return ((ctrl->ctrl_config & NVME_CC_ENABLE) && (csts & NVME_CSTS_PP));
4198 }
4199
4200 static void nvme_get_fw_slot_info(struct nvme_ctrl *ctrl)
4201 {
4202         struct nvme_fw_slot_info_log *log;
4203
4204         log = kmalloc(sizeof(*log), GFP_KERNEL);
4205         if (!log)
4206                 return;
4207
4208         if (nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_FW_SLOT, 0, NVME_CSI_NVM,
4209                         log, sizeof(*log), 0))
4210                 dev_warn(ctrl->device, "Get FW SLOT INFO log error\n");
4211         kfree(log);
4212 }
4213
4214 static void nvme_fw_act_work(struct work_struct *work)
4215 {
4216         struct nvme_ctrl *ctrl = container_of(work,
4217                                 struct nvme_ctrl, fw_act_work);
4218         unsigned long fw_act_timeout;
4219
4220         if (ctrl->mtfa)
4221                 fw_act_timeout = jiffies +
4222                                 msecs_to_jiffies(ctrl->mtfa * 100);
4223         else
4224                 fw_act_timeout = jiffies +
4225                                 msecs_to_jiffies(admin_timeout * 1000);
4226
4227         nvme_stop_queues(ctrl);
4228         while (nvme_ctrl_pp_status(ctrl)) {
4229                 if (time_after(jiffies, fw_act_timeout)) {
4230                         dev_warn(ctrl->device,
4231                                 "Fw activation timeout, reset controller\n");
4232                         nvme_try_sched_reset(ctrl);
4233                         return;
4234                 }
4235                 msleep(100);
4236         }
4237
4238         if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_LIVE))
4239                 return;
4240
4241         nvme_start_queues(ctrl);
4242         /* read FW slot information to clear the AER */
4243         nvme_get_fw_slot_info(ctrl);
4244 }
4245
4246 static void nvme_handle_aen_notice(struct nvme_ctrl *ctrl, u32 result)
4247 {
4248         u32 aer_notice_type = (result & 0xff00) >> 8;
4249
4250         trace_nvme_async_event(ctrl, aer_notice_type);
4251
4252         switch (aer_notice_type) {
4253         case NVME_AER_NOTICE_NS_CHANGED:
4254                 set_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events);
4255                 nvme_queue_scan(ctrl);
4256                 break;
4257         case NVME_AER_NOTICE_FW_ACT_STARTING:
4258                 /*
4259                  * We are (ab)using the RESETTING state to prevent subsequent
4260                  * recovery actions from interfering with the controller's
4261                  * firmware activation.
4262                  */
4263                 if (nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING))
4264                         queue_work(nvme_wq, &ctrl->fw_act_work);
4265                 break;
4266 #ifdef CONFIG_NVME_MULTIPATH
4267         case NVME_AER_NOTICE_ANA:
4268                 if (!ctrl->ana_log_buf)
4269                         break;
4270                 queue_work(nvme_wq, &ctrl->ana_work);
4271                 break;
4272 #endif
4273         case NVME_AER_NOTICE_DISC_CHANGED:
4274                 ctrl->aen_result = result;
4275                 break;
4276         default:
4277                 dev_warn(ctrl->device, "async event result %08x\n", result);
4278         }
4279 }
4280
4281 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
4282                 volatile union nvme_result *res)
4283 {
4284         u32 result = le32_to_cpu(res->u32);
4285         u32 aer_type = result & 0x07;
4286
4287         if (le16_to_cpu(status) >> 1 != NVME_SC_SUCCESS)
4288                 return;
4289
4290         switch (aer_type) {
4291         case NVME_AER_NOTICE:
4292                 nvme_handle_aen_notice(ctrl, result);
4293                 break;
4294         case NVME_AER_ERROR:
4295         case NVME_AER_SMART:
4296         case NVME_AER_CSS:
4297         case NVME_AER_VS:
4298                 trace_nvme_async_event(ctrl, aer_type);
4299                 ctrl->aen_result = result;
4300                 break;
4301         default:
4302                 break;
4303         }
4304         queue_work(nvme_wq, &ctrl->async_event_work);
4305 }
4306 EXPORT_SYMBOL_GPL(nvme_complete_async_event);
4307
4308 void nvme_stop_ctrl(struct nvme_ctrl *ctrl)
4309 {
4310         nvme_mpath_stop(ctrl);
4311         nvme_stop_keep_alive(ctrl);
4312         nvme_stop_failfast_work(ctrl);
4313         flush_work(&ctrl->async_event_work);
4314         cancel_work_sync(&ctrl->fw_act_work);
4315 }
4316 EXPORT_SYMBOL_GPL(nvme_stop_ctrl);
4317
4318 void nvme_start_ctrl(struct nvme_ctrl *ctrl)
4319 {
4320         nvme_start_keep_alive(ctrl);
4321
4322         nvme_enable_aen(ctrl);
4323
4324         if (ctrl->queue_count > 1) {
4325                 nvme_queue_scan(ctrl);
4326                 nvme_start_queues(ctrl);
4327         }
4328 }
4329 EXPORT_SYMBOL_GPL(nvme_start_ctrl);
4330
4331 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
4332 {
4333         nvme_hwmon_exit(ctrl);
4334         nvme_fault_inject_fini(&ctrl->fault_inject);
4335         dev_pm_qos_hide_latency_tolerance(ctrl->device);
4336         cdev_device_del(&ctrl->cdev, ctrl->device);
4337         nvme_put_ctrl(ctrl);
4338 }
4339 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl);
4340
4341 static void nvme_free_cels(struct nvme_ctrl *ctrl)
4342 {
4343         struct nvme_effects_log *cel;
4344         unsigned long i;
4345
4346         xa_for_each(&ctrl->cels, i, cel) {
4347                 xa_erase(&ctrl->cels, i);
4348                 kfree(cel);
4349         }
4350
4351         xa_destroy(&ctrl->cels);
4352 }
4353
4354 static void nvme_free_ctrl(struct device *dev)
4355 {
4356         struct nvme_ctrl *ctrl =
4357                 container_of(dev, struct nvme_ctrl, ctrl_device);
4358         struct nvme_subsystem *subsys = ctrl->subsys;
4359
4360         if (!subsys || ctrl->instance != subsys->instance)
4361                 ida_simple_remove(&nvme_instance_ida, ctrl->instance);
4362
4363         nvme_free_cels(ctrl);
4364         nvme_mpath_uninit(ctrl);
4365         __free_page(ctrl->discard_page);
4366
4367         if (subsys) {
4368                 mutex_lock(&nvme_subsystems_lock);
4369                 list_del(&ctrl->subsys_entry);
4370                 sysfs_remove_link(&subsys->dev.kobj, dev_name(ctrl->device));
4371                 mutex_unlock(&nvme_subsystems_lock);
4372         }
4373
4374         ctrl->ops->free_ctrl(ctrl);
4375
4376         if (subsys)
4377                 nvme_put_subsystem(subsys);
4378 }
4379
4380 /*
4381  * Initialize a NVMe controller structures.  This needs to be called during
4382  * earliest initialization so that we have the initialized structured around
4383  * during probing.
4384  */
4385 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
4386                 const struct nvme_ctrl_ops *ops, unsigned long quirks)
4387 {
4388         int ret;
4389
4390         ctrl->state = NVME_CTRL_NEW;
4391         clear_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags);
4392         spin_lock_init(&ctrl->lock);
4393         mutex_init(&ctrl->scan_lock);
4394         INIT_LIST_HEAD(&ctrl->namespaces);
4395         xa_init(&ctrl->cels);
4396         init_rwsem(&ctrl->namespaces_rwsem);
4397         ctrl->dev = dev;
4398         ctrl->ops = ops;
4399         ctrl->quirks = quirks;
4400         ctrl->numa_node = NUMA_NO_NODE;
4401         INIT_WORK(&ctrl->scan_work, nvme_scan_work);
4402         INIT_WORK(&ctrl->async_event_work, nvme_async_event_work);
4403         INIT_WORK(&ctrl->fw_act_work, nvme_fw_act_work);
4404         INIT_WORK(&ctrl->delete_work, nvme_delete_ctrl_work);
4405         init_waitqueue_head(&ctrl->state_wq);
4406
4407         INIT_DELAYED_WORK(&ctrl->ka_work, nvme_keep_alive_work);
4408         INIT_DELAYED_WORK(&ctrl->failfast_work, nvme_failfast_work);
4409         memset(&ctrl->ka_cmd, 0, sizeof(ctrl->ka_cmd));
4410         ctrl->ka_cmd.common.opcode = nvme_admin_keep_alive;
4411
4412         BUILD_BUG_ON(NVME_DSM_MAX_RANGES * sizeof(struct nvme_dsm_range) >
4413                         PAGE_SIZE);
4414         ctrl->discard_page = alloc_page(GFP_KERNEL);
4415         if (!ctrl->discard_page) {
4416                 ret = -ENOMEM;
4417                 goto out;
4418         }
4419
4420         ret = ida_simple_get(&nvme_instance_ida, 0, 0, GFP_KERNEL);
4421         if (ret < 0)
4422                 goto out;
4423         ctrl->instance = ret;
4424
4425         device_initialize(&ctrl->ctrl_device);
4426         ctrl->device = &ctrl->ctrl_device;
4427         ctrl->device->devt = MKDEV(MAJOR(nvme_ctrl_base_chr_devt),
4428                         ctrl->instance);
4429         ctrl->device->class = nvme_class;
4430         ctrl->device->parent = ctrl->dev;
4431         ctrl->device->groups = nvme_dev_attr_groups;
4432         ctrl->device->release = nvme_free_ctrl;
4433         dev_set_drvdata(ctrl->device, ctrl);
4434         ret = dev_set_name(ctrl->device, "nvme%d", ctrl->instance);
4435         if (ret)
4436                 goto out_release_instance;
4437
4438         nvme_get_ctrl(ctrl);
4439         cdev_init(&ctrl->cdev, &nvme_dev_fops);
4440         ctrl->cdev.owner = ops->module;
4441         ret = cdev_device_add(&ctrl->cdev, ctrl->device);
4442         if (ret)
4443                 goto out_free_name;
4444
4445         /*
4446          * Initialize latency tolerance controls.  The sysfs files won't
4447          * be visible to userspace unless the device actually supports APST.
4448          */
4449         ctrl->device->power.set_latency_tolerance = nvme_set_latency_tolerance;
4450         dev_pm_qos_update_user_latency_tolerance(ctrl->device,
4451                 min(default_ps_max_latency_us, (unsigned long)S32_MAX));
4452
4453         nvme_fault_inject_init(&ctrl->fault_inject, dev_name(ctrl->device));
4454         nvme_mpath_init_ctrl(ctrl);
4455
4456         return 0;
4457 out_free_name:
4458         nvme_put_ctrl(ctrl);
4459         kfree_const(ctrl->device->kobj.name);
4460 out_release_instance:
4461         ida_simple_remove(&nvme_instance_ida, ctrl->instance);
4462 out:
4463         if (ctrl->discard_page)
4464                 __free_page(ctrl->discard_page);
4465         return ret;
4466 }
4467 EXPORT_SYMBOL_GPL(nvme_init_ctrl);
4468
4469 static void nvme_start_ns_queue(struct nvme_ns *ns)
4470 {
4471         if (test_and_clear_bit(NVME_NS_STOPPED, &ns->flags))
4472                 blk_mq_unquiesce_queue(ns->queue);
4473 }
4474
4475 static void nvme_stop_ns_queue(struct nvme_ns *ns)
4476 {
4477         if (!test_and_set_bit(NVME_NS_STOPPED, &ns->flags))
4478                 blk_mq_quiesce_queue(ns->queue);
4479         else
4480                 blk_mq_wait_quiesce_done(ns->queue);
4481 }
4482
4483 /*
4484  * Prepare a queue for teardown.
4485  *
4486  * This must forcibly unquiesce queues to avoid blocking dispatch, and only set
4487  * the capacity to 0 after that to avoid blocking dispatchers that may be
4488  * holding bd_butex.  This will end buffered writers dirtying pages that can't
4489  * be synced.
4490  */
4491 static void nvme_set_queue_dying(struct nvme_ns *ns)
4492 {
4493         if (test_and_set_bit(NVME_NS_DEAD, &ns->flags))
4494                 return;
4495
4496         blk_set_queue_dying(ns->queue);
4497         nvme_start_ns_queue(ns);
4498
4499         set_capacity_and_notify(ns->disk, 0);
4500 }
4501
4502 /**
4503  * nvme_kill_queues(): Ends all namespace queues
4504  * @ctrl: the dead controller that needs to end
4505  *
4506  * Call this function when the driver determines it is unable to get the
4507  * controller in a state capable of servicing IO.
4508  */
4509 void nvme_kill_queues(struct nvme_ctrl *ctrl)
4510 {
4511         struct nvme_ns *ns;
4512
4513         down_read(&ctrl->namespaces_rwsem);
4514
4515         /* Forcibly unquiesce queues to avoid blocking dispatch */
4516         if (ctrl->admin_q && !blk_queue_dying(ctrl->admin_q))
4517                 nvme_start_admin_queue(ctrl);
4518
4519         list_for_each_entry(ns, &ctrl->namespaces, list)
4520                 nvme_set_queue_dying(ns);
4521
4522         up_read(&ctrl->namespaces_rwsem);
4523 }
4524 EXPORT_SYMBOL_GPL(nvme_kill_queues);
4525
4526 void nvme_unfreeze(struct nvme_ctrl *ctrl)
4527 {
4528         struct nvme_ns *ns;
4529
4530         down_read(&ctrl->namespaces_rwsem);
4531         list_for_each_entry(ns, &ctrl->namespaces, list)
4532                 blk_mq_unfreeze_queue(ns->queue);
4533         up_read(&ctrl->namespaces_rwsem);
4534 }
4535 EXPORT_SYMBOL_GPL(nvme_unfreeze);
4536
4537 int nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout)
4538 {
4539         struct nvme_ns *ns;
4540
4541         down_read(&ctrl->namespaces_rwsem);
4542         list_for_each_entry(ns, &ctrl->namespaces, list) {
4543                 timeout = blk_mq_freeze_queue_wait_timeout(ns->queue, timeout);
4544                 if (timeout <= 0)
4545                         break;
4546         }
4547         up_read(&ctrl->namespaces_rwsem);
4548         return timeout;
4549 }
4550 EXPORT_SYMBOL_GPL(nvme_wait_freeze_timeout);
4551
4552 void nvme_wait_freeze(struct nvme_ctrl *ctrl)
4553 {
4554         struct nvme_ns *ns;
4555
4556         down_read(&ctrl->namespaces_rwsem);
4557         list_for_each_entry(ns, &ctrl->namespaces, list)
4558                 blk_mq_freeze_queue_wait(ns->queue);
4559         up_read(&ctrl->namespaces_rwsem);
4560 }
4561 EXPORT_SYMBOL_GPL(nvme_wait_freeze);
4562
4563 void nvme_start_freeze(struct nvme_ctrl *ctrl)
4564 {
4565         struct nvme_ns *ns;
4566
4567         down_read(&ctrl->namespaces_rwsem);
4568         list_for_each_entry(ns, &ctrl->namespaces, list)
4569                 blk_freeze_queue_start(ns->queue);
4570         up_read(&ctrl->namespaces_rwsem);
4571 }
4572 EXPORT_SYMBOL_GPL(nvme_start_freeze);
4573
4574 void nvme_stop_queues(struct nvme_ctrl *ctrl)
4575 {
4576         struct nvme_ns *ns;
4577
4578         down_read(&ctrl->namespaces_rwsem);
4579         list_for_each_entry(ns, &ctrl->namespaces, list)
4580                 nvme_stop_ns_queue(ns);
4581         up_read(&ctrl->namespaces_rwsem);
4582 }
4583 EXPORT_SYMBOL_GPL(nvme_stop_queues);
4584
4585 void nvme_start_queues(struct nvme_ctrl *ctrl)
4586 {
4587         struct nvme_ns *ns;
4588
4589         down_read(&ctrl->namespaces_rwsem);
4590         list_for_each_entry(ns, &ctrl->namespaces, list)
4591                 nvme_start_ns_queue(ns);
4592         up_read(&ctrl->namespaces_rwsem);
4593 }
4594 EXPORT_SYMBOL_GPL(nvme_start_queues);
4595
4596 void nvme_stop_admin_queue(struct nvme_ctrl *ctrl)
4597 {
4598         if (!test_and_set_bit(NVME_CTRL_ADMIN_Q_STOPPED, &ctrl->flags))
4599                 blk_mq_quiesce_queue(ctrl->admin_q);
4600         else
4601                 blk_mq_wait_quiesce_done(ctrl->admin_q);
4602 }
4603 EXPORT_SYMBOL_GPL(nvme_stop_admin_queue);
4604
4605 void nvme_start_admin_queue(struct nvme_ctrl *ctrl)
4606 {
4607         if (test_and_clear_bit(NVME_CTRL_ADMIN_Q_STOPPED, &ctrl->flags))
4608                 blk_mq_unquiesce_queue(ctrl->admin_q);
4609 }
4610 EXPORT_SYMBOL_GPL(nvme_start_admin_queue);
4611
4612 void nvme_sync_io_queues(struct nvme_ctrl *ctrl)
4613 {
4614         struct nvme_ns *ns;
4615
4616         down_read(&ctrl->namespaces_rwsem);
4617         list_for_each_entry(ns, &ctrl->namespaces, list)
4618                 blk_sync_queue(ns->queue);
4619         up_read(&ctrl->namespaces_rwsem);
4620 }
4621 EXPORT_SYMBOL_GPL(nvme_sync_io_queues);
4622
4623 void nvme_sync_queues(struct nvme_ctrl *ctrl)
4624 {
4625         nvme_sync_io_queues(ctrl);
4626         if (ctrl->admin_q)
4627                 blk_sync_queue(ctrl->admin_q);
4628 }
4629 EXPORT_SYMBOL_GPL(nvme_sync_queues);
4630
4631 struct nvme_ctrl *nvme_ctrl_from_file(struct file *file)
4632 {
4633         if (file->f_op != &nvme_dev_fops)
4634                 return NULL;
4635         return file->private_data;
4636 }
4637 EXPORT_SYMBOL_NS_GPL(nvme_ctrl_from_file, NVME_TARGET_PASSTHRU);
4638
4639 /*
4640  * Check we didn't inadvertently grow the command structure sizes:
4641  */
4642 static inline void _nvme_check_size(void)
4643 {
4644         BUILD_BUG_ON(sizeof(struct nvme_common_command) != 64);
4645         BUILD_BUG_ON(sizeof(struct nvme_rw_command) != 64);
4646         BUILD_BUG_ON(sizeof(struct nvme_identify) != 64);
4647         BUILD_BUG_ON(sizeof(struct nvme_features) != 64);
4648         BUILD_BUG_ON(sizeof(struct nvme_download_firmware) != 64);
4649         BUILD_BUG_ON(sizeof(struct nvme_format_cmd) != 64);
4650         BUILD_BUG_ON(sizeof(struct nvme_dsm_cmd) != 64);
4651         BUILD_BUG_ON(sizeof(struct nvme_write_zeroes_cmd) != 64);
4652         BUILD_BUG_ON(sizeof(struct nvme_abort_cmd) != 64);
4653         BUILD_BUG_ON(sizeof(struct nvme_get_log_page_command) != 64);
4654         BUILD_BUG_ON(sizeof(struct nvme_command) != 64);
4655         BUILD_BUG_ON(sizeof(struct nvme_id_ctrl) != NVME_IDENTIFY_DATA_SIZE);
4656         BUILD_BUG_ON(sizeof(struct nvme_id_ns) != NVME_IDENTIFY_DATA_SIZE);
4657         BUILD_BUG_ON(sizeof(struct nvme_id_ns_zns) != NVME_IDENTIFY_DATA_SIZE);
4658         BUILD_BUG_ON(sizeof(struct nvme_id_ctrl_zns) != NVME_IDENTIFY_DATA_SIZE);
4659         BUILD_BUG_ON(sizeof(struct nvme_id_ctrl_nvm) != NVME_IDENTIFY_DATA_SIZE);
4660         BUILD_BUG_ON(sizeof(struct nvme_lba_range_type) != 64);
4661         BUILD_BUG_ON(sizeof(struct nvme_smart_log) != 512);
4662         BUILD_BUG_ON(sizeof(struct nvme_dbbuf) != 64);
4663         BUILD_BUG_ON(sizeof(struct nvme_directive_cmd) != 64);
4664 }
4665
4666
4667 static int __init nvme_core_init(void)
4668 {
4669         int result = -ENOMEM;
4670
4671         _nvme_check_size();
4672
4673         nvme_wq = alloc_workqueue("nvme-wq",
4674                         WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4675         if (!nvme_wq)
4676                 goto out;
4677
4678         nvme_reset_wq = alloc_workqueue("nvme-reset-wq",
4679                         WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4680         if (!nvme_reset_wq)
4681                 goto destroy_wq;
4682
4683         nvme_delete_wq = alloc_workqueue("nvme-delete-wq",
4684                         WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4685         if (!nvme_delete_wq)
4686                 goto destroy_reset_wq;
4687
4688         result = alloc_chrdev_region(&nvme_ctrl_base_chr_devt, 0,
4689                         NVME_MINORS, "nvme");
4690         if (result < 0)
4691                 goto destroy_delete_wq;
4692
4693         nvme_class = class_create(THIS_MODULE, "nvme");
4694         if (IS_ERR(nvme_class)) {
4695                 result = PTR_ERR(nvme_class);
4696                 goto unregister_chrdev;
4697         }
4698         nvme_class->dev_uevent = nvme_class_uevent;
4699
4700         nvme_subsys_class = class_create(THIS_MODULE, "nvme-subsystem");
4701         if (IS_ERR(nvme_subsys_class)) {
4702                 result = PTR_ERR(nvme_subsys_class);
4703                 goto destroy_class;
4704         }
4705
4706         result = alloc_chrdev_region(&nvme_ns_chr_devt, 0, NVME_MINORS,
4707                                      "nvme-generic");
4708         if (result < 0)
4709                 goto destroy_subsys_class;
4710
4711         nvme_ns_chr_class = class_create(THIS_MODULE, "nvme-generic");
4712         if (IS_ERR(nvme_ns_chr_class)) {
4713                 result = PTR_ERR(nvme_ns_chr_class);
4714                 goto unregister_generic_ns;
4715         }
4716
4717         return 0;
4718
4719 unregister_generic_ns:
4720         unregister_chrdev_region(nvme_ns_chr_devt, NVME_MINORS);
4721 destroy_subsys_class:
4722         class_destroy(nvme_subsys_class);
4723 destroy_class:
4724         class_destroy(nvme_class);
4725 unregister_chrdev:
4726         unregister_chrdev_region(nvme_ctrl_base_chr_devt, NVME_MINORS);
4727 destroy_delete_wq:
4728         destroy_workqueue(nvme_delete_wq);
4729 destroy_reset_wq:
4730         destroy_workqueue(nvme_reset_wq);
4731 destroy_wq:
4732         destroy_workqueue(nvme_wq);
4733 out:
4734         return result;
4735 }
4736
4737 static void __exit nvme_core_exit(void)
4738 {
4739         class_destroy(nvme_ns_chr_class);
4740         class_destroy(nvme_subsys_class);
4741         class_destroy(nvme_class);
4742         unregister_chrdev_region(nvme_ns_chr_devt, NVME_MINORS);
4743         unregister_chrdev_region(nvme_ctrl_base_chr_devt, NVME_MINORS);
4744         destroy_workqueue(nvme_delete_wq);
4745         destroy_workqueue(nvme_reset_wq);
4746         destroy_workqueue(nvme_wq);
4747         ida_destroy(&nvme_ns_chr_minor_ida);
4748         ida_destroy(&nvme_instance_ida);
4749 }
4750
4751 MODULE_LICENSE("GPL");
4752 MODULE_VERSION("1.0");
4753 module_init(nvme_core_init);
4754 module_exit(nvme_core_exit);